Virtual screening of small natural compounds against NS1 protein of DENV, YFV and ZIKV

Abstract Diseases caused by viruses of the genus Flavivirus are among the main diseases that affect the world and they are a serious public health problem. Three of them stand out: Dengue, Yellow fever and Zika viruses. The non-structural protein 1 (NS1), encoded by this viral genus, in its dimeric form, plays important roles in the pathogenesis and RNA replication of these viruses. Therefore, the identification of chemicals with the potential to inhibit the formation of the NS1 protein dimer of DENV, YFV and ZIKV would enable them to act as a multi-target drug. For this, we selected conformations of the NS1 protein monomer with similar β-roll domain structure among the three virus species from conformations obtained from molecular dynamics simulations performed in GROMACS in 5 replicates of 150 ns for each species. After selecting the protein structures, a virtual screening of compounds from the natural products catalog of the ZINC database was performed using AutoDock Vina. The 100 best compounds were classified according efficiency criteria. Two compounds were observed in common to the species, with energy scores ranging from −9.2 kcal/mol to −10.1 kcal/mol. The results obtained here demonstrate the high similarity of NS1 proteins in the Flavivirus genus and high affinity for the same compounds; thus justifying the potential of these small molecules act in multitarget therapy. Communicated by Ramaswamy H. Sarma


Introduction
Viruses are classified into different families and genera. Dengue virus (DENV), Zika virus (ZIKV) and Yellow fever virus (YFV) are classified as entities of the family Flaviviridae and genus Flavivirus. Viruses belonging to this genus have a spherical shape and measure about 40 to 60 nm, have a lipoprotein envelope whose nucleic acid is the positive singlestranded RNA of approximately 11 kilobases (Chambers et al., 1990;Korsman et al., 2012;Vasconcelos, 2003).
The Flavivirus genome is initially translated into a precursor polyprotein that is further processed by viral and host proteases into three structural proteins: capsid protein (C), envelope protein (E) and pre-membrane (prM) or membrane (M) in immature and mature virion, respectively. In addition to these, there is the production of seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) that are not part of the virus structure, and the protease responsible for cleavage of the NS1/NS2 junction remains unknown (Wang et al., 2017).
All viruses that are part of the genus Flavivirus have the gene capable of producing the non-structural protein 1 (NS1), which has 352 amino acids and has a molecular weight of approximately 50 kD, this genus being the only representative of the family Flaviviridae that the encodes. This suggests, therefore, that this protein helps in the replication and transmission of two completely different hosts: mammals and insects (Yen et al., 2016).
Currently, it is considered that NS1 protein (whether in the plasma membrane or in the extracellular environment) has important functions in viral pathogenesis by modulating host immune system acting through antagonizing or modifying complement protein functions (Avirutnan et al., 2011). Inside the cell, NS1 plays an important role in modulatory events of viral RNA replication as it has the ability to remodel the structure of the endoplasmic reticulum, which is a crucial step in the formation of the Flavivirus replication compartment (Ci et al., 2020). Scaturro (Scaturro et al., 2015) in their study demonstrated that interactions of NS1 with envelope glycoproteins on the surface of the virion are essential for the efficient production of infective viral particles.
Despite the several conserved regions in the NS1 sequence between Flavivirus species, the divergences between them may be responsible for differences in the interactions between pathogen and host Xu et al., 2016).For DENV NS1, it was observed that it alone is able to induce endothelial hyperpermeability of human lung, dermal and umbilical cells in vitro, as well as vascular extravasation in the lung, liver and small intestine in mice (Beatty et al., 2015;Modhiran et al., 2015).
The NS1 of the ZIKV, which targets human brain and placental cells, was described to increase the permeability of brain endothelial cells and the umbilical vein, and the NS1 of the YFV was related to the in vitro induction of hyperpermeability only in the cells endothelial cells of the liver, and these findings are associated with diseases caused by these respective species (Glasner et al., 2018).
Although prophylactic measures are quite widespread, diseases caused by these viruses do not have pharmacological treatment with antiviral activity. Since the NS1 protein has an important viral function and is structurally similar to each other, this work aims to find small natural compounds that bind with high affinity to the NS1 protein in the monomeric state of these three viral species, in order to avoid the dimerization process. Thus, this work suggests a potential drug that can act concomitantly against DENV, YFV and ZIKV.

Molecular dynamics simulations
NS1 proteins of DENV and YFV species were retrieved from previous studies developed by Gonc¸alves et al. (2020) and Menezes et al. (Menezes et al., 2020), where both proteins were modelled using online servers. ZIKV NS1 was downloaded from Protein Data Bank under PDB ID 5K6K. For monomer study proposal only chain B was used, since it is completely solved. All proteins had sugar (NAG) covalently bonded to ASN130 and ASN208 residues.
For all three proteins, histidine protonation state in 7.4 pH was done using PropKa (Olsson et al., 2011) server. Tleap tool, from AMBER14 (Case et al., 2014) software, performed disulfides bridge bonding between cysteines residues in AMBERff12SB force field as well as it was used to generate topology for sugars using GLYCAM_06j-1 (Kirschner et al., 2008) force field.
After glycoprotein setup in tleap tool, ACPYPE (Sousa da Silva & Vranken, 2012) script was used to generate topology and parameters for GROMACS format. Hence, each NS1 protein was through molecular dynamics simulation using GROMACS 5.1.2 software (Abraham et al., 2015) under AMBER99SB-ILDN force field. Initial protein structure was inserted in a cubic box dimensioned with a minimum distance of 1.2 nm between any protein atom and box edge. This box was solvated with TIP3P water model and system was neutralized with chloride ion (Cl-) or sodium ion (Naþ).
LINCS algorithm (Hess, 2008) was used to constrain all bonds, with the exception of water bonds, which were constrained by the SETTLE algorithm (Miyamoto & Kollman, 1992). System temperature was adjusted to 310 K and pressure were adjusted to 1 atm. Both temperature and pressure were regulated by Berendsen et al. (1984) and Parrinello-Rahman (Hutter, 2012) algorithms, respectively. For nonbonded interactions a 1.0 nm cutoff was defined and the Particle Mesh Ewald summation method was used to calculate long-range electrostatic interaction. To integrate motion equations, leap-frog algorithm (Hockney et al., 1974) was applied using 2 fs time step.
Initially, system was submitted to two steps energy minimization. The first was performed in 500 steps or until maximum tolerance of 50 kJ/mol/nm using the steepest descent minimization algorithm with protein position constraint. The second step was performed through the same algorithm, but in 10000 steps or until maximum tolerance reaches 250 kJ/ mol/nm and flexible water.
Thus, after minimization steps system was submitted to equilibration step comprising to two 100 ps simulation: NVT ensemble and NPT ensemble for thermodynamics equilibration with protein position constraint. Equilibration was done with ensemble NPT of 1 ns without protein position constraint.
Production MD run was carried out at 310 K and 150 ns, without protein conformation constraint. In order to guarantee the fluctuation profile of the NS1 protein, for each species, four other replicas were submitted to molecular dynamics simulation following the same protocol described from the equilibration step. Thus, for each NS1 system (specie), quintuplicate simulations (five of each species) were performed.
Trajectory analysis was done through Root Mean Square Deviation (RMSD) calculation using the first frame of each simulation as reference and to perform cluster analysis, g_cluster package from GROMACS was used. UCSF Chimera software (Pettersen et al., 2004) was used to visualize 3 D protein behavior through all trajectory.

Binding cavity analysis and NS1 structures definition
The definition of the cavity where the potential inhibitor molecules will anchor was based on previous studies that point to the ß-roll region (residues 1-28 and 182-216) as an important interface of monomer-monomer interaction for the formation of the dimeric functional structure. The work carried out by Gonçalves et al. (2020) pointed this region as a key in the NS1 stability mechanism of ZIKV and DENV. The work carried out by Menezes et al. (2020) conferred the importance of this region for NS1 in the YFV. In this work, the mutation of two amino acids (PHE20 and PHE22) significantly altered the stability of the monomer-monomer interaction interface.
To find out cavity structure that were similar among the three species, the five replica trajectories of each virus were concatenated, and then a cluster analysis (clustering of conformations) only of the cavity region (1-28 and 182-216) was performed. For conformational clustering, a 0.15 cutoff was used which means that a structure is added to a cluster when its distance to any element of the cluster is less than 0.15 nm. The central structures of each cluster were then submitted for structural similarity analysis in TM-align (Zhang, 2005).
As the TM-align program performs the analysis in pairs of structures, the structures of the cavity of the NS1 of the ZIKV were compared with the cavities of the YFV, followed by the comparison ZIKV with the structures of the cavities of the DENV and finally, the comparison between the NS1 cavities of the YFV and the DENV. Thus, the TM-score values were obtained for each pair of analysis. This analysis was automated through a script developed in house written in Python language.
TM-score value > 0.5 suggests that proteins generally have the same folding. Thus, initially a cutoff of 0.6 was used for the TM-score value to select the pairs that resulted in at least this similarity value. As no comparative analysis generated this TM-score value, the cutoff value was reduced until finding a considerable number of comparisons with TM-score equal to or greater than it. Upon reaching the 0.57 TM-score cutoff, NS1 structural pairs ZIKV-YFV, ZIKV-DENV and YFV-DENV were filtered from the analysis. The structures that were present in the three analyzes after filtering were considered similar to each other. For compound binding purpose, additional two analysis were performed ( Figure 1). The first one was visual binding cavity viability, where it was observed presence or absence of promising pocket in the region of interest. Structures where it was not observed a cavity were excluded from the study. Finally, analysis in DoGSiteScorer (Volkamer et al., 2012) was performed to select the best two NS1 structure of each specie based on drugscore parameter which takes into account shape, size and pocket hydrophobicity. These structures were then used for the virtual screening of compounds, as this increases the probability of finding a drug common to all species.

Virtual screening of natural compounds
The natural compounds from the ZINC database were downloaded from https://zinc15.docking.org/substances/subsets/ natural-products/ in a single file in SDF format. The compounds were separated into single files and converted into PDBQT format through the automation of the OpenBabel program (O'Boyle et al., 2011) inserted into a Python script. Then, the NS1 structures were prepared using the prepare_r-ecpetor4.py script from MGLTools (Morris et al., 2009) and the obtainment of the box coordinates to delimit the anchorage region was done in UCSF Chimera.
The execution of the virtual screening of compounds was performed in two steps using the AutoDock Vina program (Trott & Olson, 2010). The first step corresponded to 01 (one) simulation per compound with generation of 10 binding modes. The second step consisted of 100 (one hundred) simulations for each compound classified among the top 100 according to the efficiency criterion. The efficiency criterion, proposed by Abad-Zapatero (Abad-Zapatero, 2007), takes into account the number of atoms in the ligand. The equation can be described by: Thus, the top 100 compounds (higher EL values) select for second step can be described as small compounds with high affinity. In this work, DG value will be replaced by AutoDock Vina scores. Discovery Studios Visualizer (Spassov & Yan, 2013) and nAPOLI server (Fassio et al., 2019) was used in order to detect conserved protein-ligand interactions.

In silico ADMET compound profile
The in silico analysis of the pharmacokinetic and toxicological properties (Absorption, Distribution, Metabolism, Excretion and Toxicity -ADMET) of the selected compounds was performed by the OSIRIS Property Explorer program (http:// www.organic-chemistry.org/prog/peo/) and SwissADME server (Daina et al., 2017). The main parameters used to analyze the compounds were toxicological risks (mutagenic, tumorigenic, irritant and reproductive), pharmacokinetics profile (GI absorption, BBB permeant, Lipinski's rule, bioavailability score) and scores (druglikeness and drug-score).

Statistical analysis
The main data evaluated in the study were the energy scores of the AutoDock Vina. Normality Anderson-Darling test was applied to that data. For all cases the data was not normally distributed.
Therefore, the Wilcoxon (Wilcoxon, 1945) non-parametric test for the comparison of two samples. The analyzes were performed in version 3.6.1 of the R program (https://www.rproject.org) and the graphs were plotted in RStudio using the ggboxplot function. Differences were considered significant when p < 0.05.

Molecular dynamics simulations
The RMSD average of replicas can be seen in Figure 2a. It is noticed that DENV NS1 has more stability from initial model than ZIKV and YFV NS1 protein. However, ZIKV and DENV NS1 average behavior are very similar with RMSD values raging for 0.45 to 0.55 nm. Interesting, YFV has higher average RMSD value after protein stabilization ($ 0.72 nm). These differences may be related to protein sequence similarity which also affects protein dynamics. Protein sequence alignment analysis shows all of them has 100% query coverage when compared to each other, unlike percent protein identity. While DENV NS1 and ZIKV NS1 has 54.26% of percentage identity (percentage of residues that match up in the alignment), DENV NS1 and YFV NS1 o has 44.76%, and ZIKV NS1 and YFV NS1 has 47.31%.
RMSF profile analysis of DENV's NS1 behavior in solution (Figure 2(b)) showed the high fluctuations in the ß-roll (red rectangle) and wing (blue rectangle) domain regions. The first domain is related to dimeric protein interface and for this reason it is expected to have high fluctuations. Moreover, N-terminal portion is more flexible in most proteins. This interestingly, a third fluctuation present in the other species (in CLoop of NS1 of ZIKV and in SLI of NS1 of YFV) was not observed for NS1 of DENV.
The RMSF analysis of YFV NS1 protein showed high fluctuations (Figure 2(c)) in the region known as Spaghetti Loop Interface (SLI) (residues 228 to 238), besides fluctuations in the ß-roll and wing domain regions observed previously in DENV NS1. As with the ß-roll, the SLI is an intermolecular interaction region, which would justify the increased fluctuation. However, the absence of this fluctuation in the NS1 of the DENV and ZIKV NS1 must be investigated. Furthermore, in replica 3 a fluctuation outside the other replicas profile was observed in the 188-192 residues region. This region comprises one of the Steps after finding similar NS1 structures. After finding similar structures in TM-Align analysis, it was performed two additional analyses. The first one was pocket viability, through visual inspection. Structures without promising binding cavity in the ß-roll residues domain (residues 1-28) and part of ß-ladder domain (residues 182-216) were removed from this study. Orange surface structure is represented as excluded structure and blue surface structure is represented as included structure. The second step was DoGSiteScorer. This server provides drugscore measure, which take into account size, shape and pocket hydrophobicity. The colors of the dotted rectangles refer to the colors of the regions pointed out in the structure (ß-rollred, CLoop -green and Wing -blue). In all RMSF graphs the x-axis is colored according to the protein domain (redß-roll, orangeconnector, yellowwing and blue -ß-ladder).
ß-ladder strands that is stabilized by hydrogen bonds. Probably, the rupture of these bridges in replica 3 (see inserted ribbon representations in Figure 2(c)) triggered instability in the region. This rupture was not observed in other YFV NS1 replicas. RMSF analysis (Figure 2(d)) shows that of all unstable regions, the only one that had the most apparent difference between the replicates was in the ß-roll region. Certainly, the need to dock this region with the corresponding one of the other monomer, to obtain the functional structure in the form of a dimer, makes this domain more flexible and unstable to facilitate the adjustment of the connection between them. The other regions that showed greater fluctuations in the RMSF comprise the wing domain and the region called the connector loop (CLoopgreen rectangle). Despite unknown roles in monomeric NS1, it is believed that these regions may play an important role in the protein's interaction with the host's membrane .
Altogether, these RMSF analysis showed a third fluctuation in YFV NS1 replicas (Spaghetti loop interface) and ZIKV NS1 replicas (CLoop), besides ß-roll and wing domain fluctuations, also observed in DENV NS1 replicas. The key to the differences observed between species may lie in the amino acid sequence that differs between them. For example, the SLI region in NS1 YFV differs from proteins in DENV and ZIKV (supplementary material, Fig. 2, delimited by red arrows). This composition may favor increased instability in this region. In the CLoop region of NS1 of ZIKV, no major changes were observed to justify the difference, except for the replacement of H164T in YFV and DENV NS1 (supplementary material, Figure 2, blue arrow). More detailed analysis must be carried out in order to confirm the interference of this substitution in NS1 of the ZIKV.

Dynamics of binding pocket formation
It was observed that not all molecular dynamics simulation of NS1 protein resulted in a stable binding pocket formation. Analysis of residues distances were performed in order to detect important intramolecular interactions resulting in viable ß-roll interaction interface.
All DENV NS1 replicas formed a stable ß-roll pocket in the majority of trajectory time. In common to all, it was noticed that this stability was favored by interaction between ASP23 Figure 3. ß-roll stability in NS1 replicas (A) Minimum distance between residues ASP23 and HIS26 in NS1 replicas from DENV. The right-upper chart shows ß-roll structure in grey ribbon highlighting when ASP23 and HIS26 are not near to each other this domain opens. The right-lower chart highlights the opposite case, when ASP23 and HIS26 are near ß-roll domain takes a more closed position in replica 5 (yellow line). (B) Minimum distance between residues ASN7/PHE8 and ASP27/TRP28 from YFV NS1 replicas. The right-upper chart shows ß-roll structure in grey ribbon highlighting when residues ASN7/PHE8 are not near to ASP27/ TRP27 ß-roll domain opens. On the other hand, the right-lower chart shows when these residues are closer ß-roll pocket forms and stabilize. and HIS26, where is observed hydrogen bond formation between them (Figure 3(a)).
In case of YFV NS1 replicas were different. Only replica 1 formed stable ß-roll pocket that could allow dimer formation. When we compared trajectories, it is observed that stable pocket is formed when residues ASN7/PHE8 and ASP27/ TRP28 are closer (Figure 3(b)) and it gets open (unstable) when they are far from each other, suggesting these residues interaction may increase pocket stability in NS1 from YFV specie.
Different from DENV and YFV NS1, ZIKV NS1 replicas do not shown pocket stability and for this reason it was not observed an interaction pattern that could be associated to ß-roll pocket formation. However, this does not mean that ZIKV NS1 does not form an interface to allow monomermonomer interaction. This may be due to simulation time had not be enough to stabilize this domain or the monomer-monomer interaction occurs by induction when domains from different monomers gets closer to each other as proposed previously by Menezes et al. (2020).

Similar ß-roll structures obtained from molecular dynamics simulation
After clustering structures from MD simulations taking into account ß-roll residues domain (residues 1-28) and part of ß-ladder domain (residues 182-216), it was observed 247, 506 and 415 clusters for DENV, YFV and ZIKV NS1 proteins, respectively. The central structure of each cluster was then submitted to TM-Align analysis and TM-Score ! 0.57 was defined as cutoff. The analyses were performed between two viral species. Between ZIKV NS1 and DENV NS1 7036 pairs of structure had TM-Score higher than cutoff, YFV NS1 and DENV NS1 were found 1338 pairs, and between ZIKV NS1 and YFV NS1 337 pairs. Notice that ZIKV and DENV NS1 comparison has more similar structures than those involving YFV NS1. This suggests that dynamic behavior observed previously in RMSD analysis also reflects the structure similarity between DENV and ZIKV NS1. In these analyses, 11 structures of DENV NS1 were found concomitantly similar (TM-score ! 0.57) with structures of YFV and ZIKV clusters, 15 conformations of YFV NS1 were classified similar compared to DENV and ZIKV, and finally 30 conformations of the NS1 of the ZIKV similar with DENV and YFV (Figure 4(a)). Thus, we can infer these structures (11, 15 All proteins are in transparent cartoon except ß-roll residues domain (residues 1-28) and part of ß-ladder domain (residues 182-216). Notice that ß-ladder domain (to where gray arrow is pointing) are very similar among NS1 proteins, unlike ß-roll (to where green arrow is pointing) region that is very flexible. and 30) are similar among themselves, once this TM-Score value is within the indicated value for proteins with similar folding (TM-score > 0.5). Finally, two other filters (visual inspection of pocket availability and DoGSiteScorer) were applied and two NS1 protein of each specie were selected for virtual screening (Figure 4(b)). NS1 protein name are here designed as specie and the number of cluster from cluster analysis (example: DENV73). Table 1 describes analysis of DoGSiteScorer for NS1 proteins with pocket availability. The two highest drugscores for NS1 structures of each specie were selected for virtual screening. Graphic details of each step can be seen in the supplementary material (Figure 2). Figure 5 shows the selected NS1 structures after DoGSiteScorer analysis. Also, it is possible to see the main residues that form the binding pocket. Some of these residues are ILE/VAL19, PHE20, ILE/VAL21, TRP201. PHE20, ILE21, TYR22, ASN23 and TRP201 residues were previously pointed out as five major interaction residues for ZIKV NS1 (Raza et al., 2020). Except for TYR22, all other residues can be seen in ZIKV NS1 pocket of our study. For others NS1 species, it can be noticed ASP/ARG23 instead of ASN23, and residue at 22 position was not observed. This analysis suggests that these pockets are very promising to act as binding site.

Small compounds with high NS1 DENV, YFV and ZIKV affinity
After NS1 structures selection, virtual screening of natural compounds from ZINC database was performed for each In mesh colored surface are represented the pocket pointed out by DoGSiteScorer server. Notice it can be seen only a part of the pocket, which does not correlate to total area and volume calculated and presented in Table 1. In sticks it can be seen main residues that are part of pocket and can interact to compound. NS1 protein (total of six). After that, the top 100 compounds for each virtual screening were classified based on efficiency criteria (Equation (1)). These 100 compounds were then compared to find intersection (in common classified compounds) in at least one NS1 structure of each specie. Two compounds were found to have high affinity for at least one NS1 structure of DENV, YFV and ZIKV (Table 2). In the boxplot of the distribution of the 250 smallest energies resulting from the anchoring of these two compounds in the NS1 structures (Figure 6(a)), it is possible to observe that the compound 271839 has a higher energy value (À7.6 kcal/mol, for the NS1 structure of the ZIKV 144) and the compound 485867 has lower energy (À10.1 kcal/ mol, in the structure of NS1 YFV 162). However, the global distribution of energies (not distinguished by NS1 structures) is quite similar for the two compounds and it is noted that there is no statistically significant difference between them (p > 0.05) (Figure 6(b)).
Interaction analysis performed by Discovery Studios Visualizer of best binding modes (lowest Vina docking scores) for each complex can be seen in Figures 7 and 8. Figure 7 shows interactions of complexes for compound 271839. For these four complexes, it was observed four interactions with LYS/ARG14 (which one is unfavorable interaction) and ILE/ VAL19. For complexes with compound 485867 (Figure 8), interactions with LYS/ARG14, ILE/VAL21 and HID/ASP/GLU26 were preserved in all four complexes.
For a wider analysis, nAPOLI server was used to see main protein-ligand interactions (1000 for each complex). Analysis of the conserved residues interactions between ligands and  protein revealed that interactions at residues ILE/VAL19 and TRP201 were maintained in all NS1 structures in the analysis (Supplementary material, Figure 4). These residues at the same position 19 are classified as conservative substitution, since amino acids have similar biochemical characteristics.
These interactions results suggest the importance of these residues for a possible increase in the affinity of these molecules, not only in relation to their position, but also their biochemical characteristics. Also, it is important to highlight conservation of interaction with ILE/VAL19, PHE20 and ILE/VAL21 in all NS1 structures, except DENV73. This amino acid triad were point out previously (Gonçalves et al., 2020) as important residues for entropic role in the interaction between monomers.
Finally, in silico result performed in the SwissADME showed promising result for drug development, specially that both compounds follow Lipinski's Rule. However, compound 485867 is blood-brain barrier permeant and it could cause neurological effects. OSIRIS program showed good results for the compound 271839, with low risk of toxicity for all analyzed parameters and drug-score of 0.87 (Table 3). The ligand 485867, on the other hand, presented a high risk of toxicity in the reproductive risk assessment and a drug-score of 0.12.

Conclusion
The MD simulations of NS1 proteins from DENV, YFV and ZIKV viruses allowed to find similar conformations of important monomer-monomer interface interaction in order to virtual screening for similar compound with high binding affinity. The main idea of this work was based that a compound affinity is related to protein structure. Since it was selected structures with similar binding cavity based on TM-Score, it was expected to find same compounds with high binding affinity to NS1 protein species. Two compounds ranked as top 100 among more than 220000 natural compounds were find as molecules with high binding affinity to DENV, YFV and DENV NS1 protein simultaneously. Intermolecular interactions between NS1 and ligands showed important interactions conserved among complexes involving residues with relevant role in monomer-monomer interaction.
As far as we know, although there are other studies targeting NS1 protein of Flavivirus (Ahmad et al., 2021; Raza Figure 8. Maps of intermolecular interaction between ligand 485867 and (A) DENV73 NS1 protein structure with the highest Vina docking score, À9.1 kcal/mol, (B) ZIKV84 NS1 protein structure with the highest binding mode Vina docking score, À9.2 kcal/mol, (C) YFV65 NS1 protein structure with the highest binding mode Vina docking score, À9.3 kcal/mol, and (D) YFV162 NS1 protein structure with the highest binding mode Vina docking score, À10.1 kcal/mol.  Songprakhon et al., 2020), this is the first study targeting NS1 of three different Flavivirus species. In this sense, the results presented here contributes to development of a common drug to treat three different virus infections when there is not any available pharmacological molecule with antiviral activity. However, in order to do so, new studies and analysis in vitro and in vivo should be performed.