Computational Screening of Isolated Compounds From Goniothalamus Species to Identify Potential Inhibitors for Dengue Virus


 In this study, a set of 234 chemical constituents reported from Goniothalamus species were docked against envelope (E), NS2B/NS3, NS5 methyltransferase, and NS5 RdRp dengue virus (DENV) protein. As the result, compounds 95, 96, 97, 100, 149, 155, and 187 were identified as potential dengue protease inhibitors based on minimal docking energy values and multiple interactions with binding sites. The results from in-silico Lipinski’ rule and ADMET analysis showed that compound 149 was predicted as the most potential compound that fulfills the drug-likeness properties. Ligand 149 was found to be able to fit in well and remain stable in the binding site of proteins envelope, NS2B/NS3, NS5 methyltransferase and NS5 RdRp. The results from molecular dynamic simulations indicate that the ligand-protein complex of 149 in NS5 methyltransferase showed the most preferable, successfully interacted within the active sites and were able to reach convergence within 100 ns.


Introduction
Dengue fever has been a major concern around the world. The World Health Organization (WHO) estimated about 100-400 million dengue infections each year and about half of the world's population is now at risk. It is caused by the dengue virus (DENV), a member of the Flavivirus genus [1]. DENV consists of four serotypes named DENV-1, DENV-2, DENV-3, and DENV-4. DENV of serotype-2 has traditionally been studied in more detail than other serotypes because it is a more virulent strain [2][3][4]. The mature proteins for the dengue virus include three structural proteins (capsid (c), pre-membrane (prM), envelope (E)), and seven nonstructural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5). The E protein is essential for initial attachment of viral particles to host cell receptors. Thus, any disturbance in the function of E protein will prevent the dengue virus from attaching to the host cell receptors [5]. Inhibition of viral replication and infectivity will take place when there is any disruption in the NS3/NS2B complex region. Hence, NS3/NS2B complex is considered a promising target protein [6]. Besides, non-structural DENV protein NS5 also plays an important role in the inhibition of viral replication because of its dual enzymatic activity. The methyltransferase (MTase) domain protects viral mRNA from degradation through RNA capping (post-transcriptional modi cation), while RNA-dependent RNA-polymerase (RdRp) domain is responsible for the replication of positive-strand RNA within the host cell [7,8].
Nowadays, there are a lot of treatments that use medicinal plants to inhibit virus replication [9]. Unfortunately, only a handful of research had been done on potential plants against the dengue virus. Goniothalamus species was reported to have secondary metabolites that are bene cial for activities like anti-tumor, anti-aging, anti-stress, and anti-malaria [10,11]. The genus Goniothalamus (Family Annonaceae) were found to grow in the rainforest of tropical southeast Asia [12]. The most dominant classes of compounds present in this species are alkaloids, styryl-lactones, acetogenins, avonoids, and fatty acids. Based on previous pharmacological studies, styryl lactones and acetogenins from Goniothalamus species are potent against anti-cancer cells such as kidney cells, breast, colon, and pancreatic carcinoma [13][14][15][16]. Previous studies showed that bio avonoid quercetin exhibited signi cant anti-DENV replication properties [17]. Quercetin and setin affect intracellular DENV replication and can be proposed as candidate for anti-DENV therapeutics [17], while acetogenins can potentially inhibit DENV replication [18]. Styrylpyrone derivatives (SPD) from G. umbrosus root were also reported to show potential activity against dengue virus type 2 [60].
In continuation of our studies on compounds isolated from the roots and stembarks of the G. lanceolatus [28,62], we aim to predict the possible compounds reported from the same genus of the plant that can interact well with target proteins for DENV.
This study focuses on targeting structural DENV proteins envelope protein (E), non-structural DENV protease (NS3/NS2B), NS5 methyltransferase (MTase), and NS5 RNA-dependent RNA-polymerase (RdRp) protein. A set of plant-derived secondary metabolites isolated from Goniothalamus species were virtually screened against targeted DENV proteins through molecular docking and molecular dynamics (MD) simulation approaches. The set of compounds comprises different classes of compounds which are styryl-lactones, alkaloids, acetogenins, and avonoids.

Experimental
Selection of target protein Selecting the potential drug target or receptor that is responsible for the dengue replication process is the most important step in the drug designing procedure. Several target proteins for dengue virus that had been selected for this study include dengue virus envelope protein (E), protease (NS2B-NS3), helicase (NS3 helicase), methyltransferase (MTase), and the dengue virus RNAdependent RNA-polymerase (RdRp). The crystal structure of DENV target proteins were retrieved from the Protein Data Bank. Molecular docking was conducted using the three-dimensional crystal structure of dengue virus envelope protein (PDB:1OKE); NS2B-NS3 (PDB:2FOM); NS5 methyltransferase (PDB:1R6A) and NS5 RdRp (PDB:3VWS).

Ligand Database Preparation
The comprehensive literature search was done to compile a set of 234 compounds phytochemicals previously isolated from Goniothalamus species and 12 reported protein inhibitors. The chemical structure of phytochemicals was taken from the online database of Dictionary of Natural Product (DNP), Pubchem, and ChEMBL. The structure was stored in .mol format. The energy minimization for each ligand was performed by using the MMFF94X force eld.

Molecular Docking
The structure of isolated compounds from Goniothalamus species and reported DENV protein inhibitors were drawn by using ChemBioDraw® Ultra 13.0. The protein-ligand docking studies of the inhibitors and proposed drugs with the receptors 1OKE, 2FOM, 1R6A, and 3VWS were performed using validated CDOCKER protocol in Discovery Studio® 3.1 (Accelrys, San Diego, USA).
This CDOCKER program generally employs CHARMm force elds as a grid-based molecular docking method. Water molecules and unwanted molecules from the crystal structure of the receptor were removed before performing docking experiments. The ligands conformations were generated through high-temperature molecular dynamics followed by random rotations and re nement by grid-based (GRID I) simulated annealing, full force eld minimization, or nal grid-based [26]. In this study, the cooling steps were set in 5000 steps to 300 K while for the heated steps, the temperature of 700 K in 2000 steps was used with the grid extension set to 10 Å. The top ten binding poses for each ligand were ranked based on their CDOCKER energies value.
To predict the best binding interaction, the top ten ranked poses per ligand were then analyzed. All the compounds were ranked based on the 1) interaction with an active residue of dengue protein; 2) binding energy value; 3) the total number of interactions and 4) bond distance. The idea was to select only those compounds that bind to the crucial amino acid of dengue protein with favorable docking energy value. To validate the docking protocol, the co-crystallized inhibitors were docked as reference ligands with non-structural E, NS3/NS2B, NS5 methyltransferase, and NS5 RdRp protein using CDOCKER to reproduce the binding interactions of ligands within the catalytic pockets.

Molecular descriptors calculation
Molecular properties and practicality of the compounds as drug candidates were determined based on the threshold set by "Lipinski's Rule of Five" through the Molinspiration server (www.molinspiration.com). The thirteen descriptors that include log-P, polar surface area, molecular weight, number of atoms, number of O or N, number of OH or NH, number of rotatable bonds, volume, G protein-coupled receptors (GPCR) ligand, ion channel modulator, a kinase inhibitor, and nuclear receptor ligand, and the number of violations to Lipinski's rule.

ADMET analysis
Pharmacokinetic studies are conventional methods to evaluate the drugs as they are being administered. Poor toxicity and clinical safety contribute to failures in drug discovery. To predict the level of clinical safety and toxicity, in-silico ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) analysis was performed using SwissADME (http://www.swissadme.ch/index.php). In-silico ADMET analysis gave information about human intestinal absorption (HIA), aqueous solubility (AS), blood-brain barrier (BBB), cytochrome P450 2D6 (CYP2D6), plasma protein binding (PPB), hepatotoxicity (HT), atom-based logP (Alog P98), and polar surface area (PSA) descriptors.

Molecular Dynamic Simulation
Molecular Dynamic simulations analysis were performed using Desmond v.2018 suite. The systems were set up using "System Builder'' in Maestro. Energy minimization of complex 149 with target protein was performed by applying an enhanced Optimized Potentials for Liquid Simulations (OPLS3e) molecular mechanics force eld, which was then placed in the cubic box with a buffer distance of 10Å to create a hydration model. Water molecules were placed between the ligand-receptor complex using a simple point charge (SPC) water model. The system was set to 2.0 fs, 9Å, 300K, and 1.01325bar for the cut-off radius for time step, van der Waals, initial temperature, and pressure, respectively. To evaluate the electrostatic force of the Desmond system, near term and far term was divided by a boundary at 9 Å. Finally, MD simulations were performed under the NPT ensemble for 100 ns. The result was recorded for every 50 ps with a total of 2000 frames.

Molecular Docking
In this work, docking results were ranked based on these factors, 1) maximum hydrogen bond interaction with active site residues; 2) minimum binding energy value; 3) the total number of interactions; 4) bond distance. A total of 234 compounds previously reported from Goniothalamus species were screened (Table S1) and the top three ranked compounds were selected. Firstly, the validation of the docking procedure was done before carrying out molecular docking for the selected ligands. A co-crystalline ligand extracted from a selected crystallographic protease structure was re-docked into the same binding pocket. Based on the validation of the docking protocol with co-crystalline ligand, the RMSD value for 1OKE (1.16 Å), 1R6A (0.72 Å), and 3VWS (1.73 Å) were below the 2.0 Å threshold. NS3/NS2B developed an interesting active site that has a highly conserved catalytic triad (His51, Asp75, and Ser135) that is of prime functional importance. Any interaction that takes place in this active site will result in the inhibition of viral replication [2].
The top-ranked compounds were ranked based on their minimum docking energy value and a maximum number of interactions with active site residues. The molecules that can disrupt virus-host binding and inhibiting the entry of virus particles into target cells can be considered as potential inhibitors. Envelope protein is a vital protease because it is crucial for the initial attachment of viral particles to host cell receptors. Thus, any compound that can disrupt E protein activity can already be considered as a lead compound towards anti-DENV drug discovery [19]. The chemical structure of top ranked compounds was drawn and listed in gure 1. The results of docking studies showed that compounds 96, 97, and 149 can bind well to the active site of DENV E protein. Among these three, xylomaticinone was selected for having a minimum number of docking energy values and a maximum number of interactions with the crucial amino acid residues. However, due to the large molecular weight (MW≤500), acetogenin compounds like 96 and 97 have poor pharmacokinetic properties and therefore it was not further pursued as potential inhibitors for envelope protein.
Compound 149 passes all the Lipinski rule's criteria and has the docking energy value of -45.9 kcal/mol (Table 1). This compound binds to the enzyme through hydrogen bonding with the active residue such as Lys128 and Thr48 while being stabilized by other residues through hydrophobic interactions. The ligand-enzyme complex was stabilized by a carbon-hydrogen bond interaction with Thr48 at 2.54 Å and 2.83 Å (Fig. 2). The compound is further stabilized through more contacts in the form of π−π interactions with the binding site amino acids, Ala50, Pro53, and Ile270. These π−π interactions are necessary to obtain stronger and more selective binding interactions. This ligand-enzyme complex also displayed the presence of electrostatic attraction in the form of a salt bridge in protein (Lys128-1.69 Å). The binding of compound 149 to envelope protein prevents protein-protein interaction between DENV-2 and host cells during infection. Hence, this compound has potential in preventing the attachment and entry process.
NS3/NS2B was considered to have a central role in enhancing viral replication in the host cells. Allosteric pockets of NS3/NS2B protease were located close to its catalytic triad (His51, Asp75, Ser135), so any close interaction with residues in an allosteric pocket will disrupt the activities of the protease [61]. Residues His51, Asp75, Gly133, Ser135, Gly151, Asn152, and Gly153 are located in the active site regions, and it is crucial for ligand interactions with NS3/NS2B protease inhibitors while residues Lys74, Leu76, Asn152, Trp83, Leu149, Gly148, and Glu88 are important residue near to allosteric pockets for NS3/NS2B [19,20]. Compounds 100 and 155 both have interactions with the catalytic triad (His51) and exhibited the minimum docking energy value, but these two compounds have several violations towards drug-likeness test due to the large molecular weight and large value of octanol-water partition coe cient so, these two acetogenins were not further investigated. The results showed that 149 produced a docking energy value of -25.9 kcal/mol. Even though 149 does not interact with the catalytic triad, this compound has several interactions with the active residues (Lys74, Leu76, and Asn152) in the allosteric pocket that is crucial for an inhibitor. The ligandenzyme complex was stabilized by a conventional hydrogen bond (N---HO) of hydrogen from the hydroxyl group of a compound with the nitrogen of amine from residue Asn152 and Lys74. Compound 149 also forms carbon-hydrogen bonding with other residues, Gly87 (2.50 Å), Gly87 (2.56 Å), and Val147 (2.50 Å).
This complex was further stabilized by π−π interactions with the binding sites involving Ala166, and Leu76 (Fig. 3). Furthermore, salt bridge interaction between lysine (Lys74-1.69 Å) residues and electrostatic charge at the oxygen of hydroxyl group from the ligand side chain also play an important role to stabilize the complex. Thus, any interaction with these residues in the allosteric pocket of the active site will enhance the inhibition activities in the processing of polyprotein during the replication of virus dengue.
Besides, both NS5 methyltransferase and NS5 RdRp are crucial for the inhibition of viral replication [7]. Compound 149 showed the best docking interaction for both proteases due to the maximum number of interactions with active residues and without having any violation for the drug-likeness test. Compound 149 showed the best docking interaction with IR6A with a docking energy value of -53.6 kcal/mol. The stability of the ligand-enzyme complex is based on the carbon-hydrogen bond with the following active residues (Ser150, Lys29, Pro152, and Lys22). The results displayed π−π interactions with the binding sites involving Phe25 and Pro152 (Fig. 4). This complex was further stabilized by two electrostatic attractions between oppositely charged residues that are su ciently close to each other (Lys22-1.63 Å and Lys29-1.69 Å). Compound 149 also showed the best docking interaction with the 3VWS protease enzyme with a docking energy value of -56.8 kcal/mol. This compound binds to the enzyme through conventional hydrogen bonding with the active residue Thr413 and Asn405 and is further stabilized by other residues through hydrophobic interactions. This compound is able to form π−π interactions with Trp795 (5.00 Å) and attractive charge interaction with Arg792 at 2.88 Å (Fig. 5). As mentioned above, compound 149 is bound to a common active site residue that is structurally and functionally important for NS5 dengue proteases.
Eighteen avones showed favorable binding energy of less than -56.8 kcal/mol. Overall, from the docking results, compound 149 was observed to have the highest a nity with E, NS3/NS2B, NS5 methyltransferase, and NS5 RdRp. This compound successfully ts in the binding site of the protein and forms the most number of interactions with the active site residues at the lowest docking energy value. Compound 149 is a quercetin derivative, which is known as 3-O-methylquercetin because of the presence of four hydroxy and one methoxy group. The presence of hydroxyl groups at C5, C7, C3' and C4' on the phenyl rings produces a resonance phenomenon. The electrons from the hydroxyl group are delocalized into the aromatic ring. Hence, this facilitates the formation of hydrogen bonding with the residues in the protein binding sites. In addition, the methoxy group at the C3 position could enhance the overall binding interactions (Fig. S1). This compound was reported to display antioxidant and free radical scavenging activities [21]. Besides, 3-O-methylquercetin also displayed several in vitro and in vivo biological activities such as anti-in ammatory, neuroprotective, bronchodilatory, vasodilatory, antinociceptive, immunomodulatory, antitumor, and antiviral [22].
The acetogenins were observed to have a minimal docking energy value lower than -59.0 kcal/mol. This is due to the long alkyl chain in the skeleton that could enhance the interactions with the protein target. However, acetogenins normally have molecular weights of more than 500, which causes di culty in terms of tting in the protein binding site and not being absorbed through the gastrointestinal tract. Styryl-lactones are the dominant class of compounds in this genus, but they were found to exhibit poor binding energy of less than -33.1 kcal/mol while alkaloids were observed to have even lower binding energy than -25.8 kcal/mol.  reported inhibitors showed different interactions with the target protein of the binding pockets. Among these reported inhibitors, compound 23i was observed to have no interaction at all with the catalytic triad but produces hydrogen bond interactions with amino acid residues Asn152, Asn167, and Ala166. Other than that, for envelope DENV protein, DO2 was found to interact with crucial amino acid residues (Lys128 and Ala50) and another active amino acid (Pro53 and Ile270). For NS5 methyltransferase DENV protein, NSC14778 exhibited interaction with three crucial amino acid residues and several interactions with other active amino acids. These inhibitors can form hydrogen bonds with Lys14, Ser214, and Lys22, π-interaction with Phe25, and salt-bridge interactions with Lys14 and Lys22.
Moreover, it also has the ability to interact with Cys331 (Fig. 8). From the ndings, most of the top-ranked phytochemicals from Goniothalamus sp. exhibited good binding interaction as compared to reported inhibitors based on the docking energy value and the maximum number with crucial amino acids. Several compounds showed a maximum interaction with crucial amino acid residues that makes them potential lead compounds for anti-dengue drug discovery.

Drug-likeness and ADMET
Drug scan tools at Molinspiration and SwissADME server predict the drug-likeness of the proposed DENV inhibitors.  Table 3). The remaining acetogenins had some violations because of the molecular weight of more than 500, Log-P value bigger than 5 and exceeding the number of hydrogen bond donors (≤5). The phytochemical that successfully passed all these parameters is predicted and considered as potential lead compounds with the good pharmacokinetic property.
Based on the results, compound 149 (C16H12O7) is the only phytochemical that passed the drug-likeness analyses without any violations. ADMET analyses were further conducted to predict the inhibitor's capability of drug-likeness. This analysis was conducted based on several parameters, physicochemical property, absorption, distribution, metabolism, and toxicity. These parameters have then been evaluated based on the number of thresholds. The results showed that all the acetogenins compounds are not quali ed for certain parameters. However, the ADME prediction result for compound 149 inhibitors were predicted to pass the ADMETsar threshold of drug ability (Table 4). The ADME prediction result for compound 149 showed some solubility (Log S) with value -3.88. The distribution coe cient P (Log P) for this compound was 1.96 (should be between -0.7 and + 5.0). The compounds were unable to permeate BBB and act as non-substrate of P-glycoprotein. However, this compound was considered as an inhibitor for enzyme CYP1A2, CYP23A4, CYP2C9 and CYP2D6 subtypes. From this in-silico ADME analysis, compound 149 also was predicted to have no hepatotoxicity and did not have the plasma protein binding (PPB) ability. The results of Lipinski's calculation and ADME analysis indicated that compound 149 is the most selected among others due to the zero violations in Lipinski's rule of ve and it was also predicted as a potential inhibitor by ADME analysis. system can be monitored using root-mean-square deviation (RMSD) that gives information equilibration on structural conformation throughout the simulation. If uctuations happen towards the end of the simulation and the changes are larger than 3 Å, it indicates that the protein undergoes a large conformational change, and the ligand is not stable in the binding pocket of the protein during the simulations.
The relative uctuation in the RMSD of Cα carbon atoms (Cα-RMSD) for 149 bound to 1OKE protein (Fig. 9A) was observed for the rst 42 ns. Between 42-90 ns, the RMSD of the 1OKE protein slightly decreases while 149 shows some marginal increase with a small uctuation at the end with an RMSD value of 0.75 Å. In the case of 2FOM (Fig. 9B), the RMSD showed a marginal decrease for 5 ns followed by a slight uctuation for 38 ns. Then, the RMSD of 149 increases to 0.8 Å and converges at the end of the simulation. Similarly, the RMSD of 1R6A (Fig. 9C) slightly uctuate for the rst 35 ns followed by uctuation between 35 to 45 ns and continue to uctuate marginally between 45 to 85 ns. The RMSD starts to converge at 85 ns to the end of the simulation (RMSD: 0.4 Å). The RMSD for 3VWS (Fig. 9D) showed the most stable ligand-protein complex due to the shorter time taken to reach convergence. At rst, the RMSD showed a marginal increase for 25 ns before it started to converge throughout the simulation (RMSD: 0.8 Å).
Root-mean-square uctuation (RMSF) is useful to indicate the local changes along with the mobility of the protein during the simulation. In the case of 1OKE (Fig. 10A), a higher uctuation was observed at the C-terminus residues up to 3.6 Å. Residues Thr48, Lys128, and Ala50 were recognized as a binding residue for 1OKE that showed lesser uctuation. Unlike 149 complexed with 2FOM (Fig. 10B), N-terminus residues showed a higher uctuation of up to 3.0 Å. Residues 10-15 and 40-45 tend to show more uctuation as compared to the binding site residues (Lys74, Leu76, Gly87, Val147, Asn152, and Ala166). In the case of NS5 methyltransferase (Fig. 10C), higher uctuation was observed at the C-terminus residues up to 2.8 Å. The RMSF for residues 40-50, 100-105 and 165-170 tend to show high uctuation as compared to active site residues of Leu20, Lys14, Asn18, Lys29, Phe25, Leu17. Compound 149 complexed with 3VWS (Fig. 10D) was observed to have a marginally higher uctuation at the N-terminus and the uctuation was minimal for binding site residue, so it showed higher stability during the simulation.
To understand the conformational evolution of the ligand, the torsional conformations of each rotatable bond in the ligand throughout the simulation run were calculated (Fig. 11). The angle of each bond was illustrated by the dartboard plots while the probability of the torsions as a function of angle were illustrated by the histograms. The dartboard plot was placed in the centre of the radial plot at the beginning of the simulation trajectories while radially outwards at the time evolution. The angular coordinate was the torsional angle while the radial coordinate was the torsion rotatable bond [63]. The relationship between the torsional angle changes and interactions are directional proportional. This relationship was discovered while combining with the results on interactions observed through the docking model ( Fig. 6 and Fig. 11). The results for compound 149 in the binding pocket of envelope dengue protein (Fig. 6A and Fig. 11A) displayed torsional angle changes in bond e and d, promoted by the hydrogen bond interaction between the hydroxyl group and residue Thr48. In a similar way, the torsional angle changes in bond c correspond to the hydrogen bond interaction between the hydroxyl group and residue Pro53. The torsional angle also changes in bond b corresponding to the π-π interaction between the aromatic ring and residue Ala50. The dartboard plots and the bar charts showed that bond c is rigid while bonds b, d, and e were more exible.
In Fig. 6B and Fig. 11B, ligand 149 in the binding cavity of NS3/NS2B dengue protein showed the torsional angle changes in bond a and e promoted by the hydrogen bond interaction between the hydroxyl group and residue Asn152, Lys74, Gly87 and Val147. The torsional angle changes in bond c were promoted by the hydrophobic (salt-bridge) interaction between the hydroxyl group and residue Lys74. The torsional angles also change in bond b corresponding to the π-π interaction between the aromatic ring and residues Ala166, Leu76, Lys74 and Gol202. The bar charts and dial plots showed that the bonds b and c were rigid while bond a and e are more exible.
The torsional angle for compound 149 in the binding site of NS5 methyltransferase was shown in combination of Fig. 6C and Fig. 11C. The Torsional angle changes in bond e promotes the hydrogen bond interaction between the hydroxyl group and residue Ser150. The torsional angle changes in bond c and d were promoted by the hydrophobic interaction between the hydroxyl group and residues Lys22 and Lys29. The torsional angles also change in bond b correspond to the π-π interaction between the aromatic ring and residue Phe25. From the bar charts and dial plots, bond b, d and e were exible compared to bond c which was more rigid.
Ligand 149 in the binding pocket of NS5 RdRp (Fig. 6D and Fig. 11D) showed torsional angle changes in bond a and c promoted by the hydrogen bond interaction between the hydroxyl group and residues Asn405 and Thr413. The torsional angles in the bond a corresponded to the hydrophobic interaction between hydroxyl group and residues Arg792. In addition, the torsional angle also changes in bond b promoted by the π-π interaction between aromatic ring and residue Trp795. The dial plots and bar charts showed that the bonds a, b, and c are more exible. It was shown that ligand 149 in the binding pocket of NS5 RdRp had more rotatable bonds which indicates that the molecule is exible, and the complex is stable throughout the simulation.
To illustrate the stability of the selected ligand in the binding pocket of the proteins during the simulation, these six properties were analyzed: 1) root mean square deviation (RMSD) of a ligand concerning the reference con rmation; 2) radius of gyration The results of changes in the rGyr of ligand bound to the active site of the proteins are shown in Fig. 13. The rGyr value for the ligand compound 149 in the binding pocket of the envelope protein (Fig. 13A) was stable at 3.65-3.80 Å from 0 to 100 ns. The rGyr value for compound 149 in the binding pocket of NS3/NS2B protein (Fig. 13B), NS5 methyltransferase protein (Fig. 13C), and NS5 RdRp protein (Fig. 13D) showed a constant value at 3.70-3.80 Å from 0 to 100 ns. These constant values predicted a steady behaviour and compactness of the structure that exerts a great effect on the quality of the substrate binding to the active site.
The consistency of the ligand during the simulation run was also indicated by intraHB, MolSA, PSA, and SASA. The properties of MolSA and PSA for compound 149 with all targeted proteins produced a consistent value throughout the simulation run. This constant value discovered the consistency of the ligand in the binding pocket throughout the simulation run. Additionally, the solvent-accessible surface area (SASA) was also calculated. This analysis measured the surface area of a molecule accessible by a water molecule. An increase in SASA values from the initial pose indicates that the ligand was exiting the binding pocket or sticking out into water while decreasing value of SASA indicates that more of the molecule was buried in the protein. To pursue the stability of the ligand in the binding pocket of protein, lower scores of SASA are priority. According to Fig. 14A, the SASA analysis for ligand 149 in envelope protein showed a constant value at 180 Å 2 for the rst 40 ns. However, the instability of the SASA value was observed from 40 ns to the end of the simulation run. The SASA analysis of compound 149 in NS3/NS2B (Fig.   14B) showed some uctuation (100-200 Å 2 ) for the rst 40 ns but then it showed a decreasing score at 50 Å 2 until the end of the simulation run. The SASA value for ligand 149 in NS5 methyltransferase (Fig. 14C) showed a marginal decrease for the rst 5 ns. Then, it showed a constant value at 5 to 37 ns. After that, the SASA value was observed to slightly uctuate throughout the simulation at 180-240 Å 2 . For the ligand 149 in NS5 RdRp protein (Fig. 14D), the SASA analysis showed some uctuations for the rst 20 ns. After 20 ns, SASA was observed to form a steady behaviour at 30 Å 2 throughout the simulation process.
The comparison of the protein-ligand contacts from docking results and molecular dynamic simulation are important to predict the stability and the interactions in the binding site. Based on the docking result of compound 149 with envelope protein, the complex was stable due to H-bond interaction with residues Thr48 and Pro53, hydrophobic interaction with residue Lys128 and ππ interaction with residues Ala50, Ile270 and Pro53. In comparison, the molecular dynamic simulation result ( Fig. 15A and   Fig.16A) showed that this complex was able to maintain H-bond interaction with Asn18, Phe25 and Glu149 at 99%, 70% and 90% of the simulation time respectively. The presence of interactions with new residues in the simulation run showed the instability of the compound 149 in the binding site of envelope protein. For the compound 149 in NS3/NS2B, the complex was stable due to Hbond with residues Asn152, Lys74, Val147 and Gly87, hydrophobic interaction with Lys74 and π-π interaction with residues Ala166, Leu76, Lys74 and Ala166, while for the molecular dynamic analysis ( Fig. 15B and Fig.16B), interaction with Lys74 was observed to maintain at 32% of the simulation run. In addition, the interaction with residues Asp75 (60%), Leu149 (100%) and Asn152 (20%) was also present during the simulation run.
The docking result of complex 149 in NS5 methyltransferase showed the H-bond interaction with Ser150 and Pro152, hydrophobic interaction with Lys22 and Lys29 and π-π interaction with Phe25 and Pro152. However, the interaction with residues Lys22 and Lys29 was only maintained for less than 10% of the simulation run ( Fig. 15C and Fig. 16C). Interestingly, the presence of new interaction with active residues Lys14 (62%), Asn18 (99%), Glu149 (30%) and Ser150 (10%) was also observed. For the NS5 RdRp, docking results showed that compound 149 forms a H-bond interaction with residues Thr413 and Asn405, hydrophobic interaction with residue Arg792 and π-π interaction with residue Trp795. In comparison with molecular dynamic simulation ( Fig. 15D and Fig. 16D), the interaction with residues Asn405, Thr413, Trp795 and Arg792 was maintained at <10%, 20%, 48%, 100% of the simulation run respectively. In addition, new interaction with active residues Val411, Arg352, Ser796 and His801 was also observed.
Based on the RMSD and RMSF value, the complex of 149 with protein 3VWS was more stable due to the shorter time to reach convergence and minor uctuation of the active residues during the simulation run. Throughout the simulation, protein-ligand interaction can be monitored. The stability of the speci c interaction of the protein-ligand complex was recorded during the simulation. Based on the result, the complex 149 with 3VWS was stable as compared to other protein targets due to the ability to maintain the most number of interactions with active residues Asn405, Thr413, Trp795 and Arg792 during the simulation run. In addition, the six properties analysis to predict the stabilities of the selected ligand in the binding pocket also showed that it takes a shorter time to reach convergence. It shows that ligand 149 can interact well and it is stable in the binding site of the NS5 RdRp proteins.

Conclusion
The ndings indicate that compound 149 has the most number of interactions and one of the best docking energy values with all the targeted proteins through the docking analysis. This compound, which belongs to the avonoid derivatives, passed the Lipinski' rule and showed potential through ADMET analysis. Overall, all the ligands t well in the binding site of the targeted protein through simulation run. Among these targeted proteins, the complex of 149 with 3VWS was considered the most preferable and the most stable due to the shorter time taken to reach convergence and minor uctuation of the active residues during the simulation. Compound 149 also interacts well with binding site residues Thr413, Asn405, Arg792, and Trp795 for more than 30% of the simulation time. This showed that the interaction of the complex was stable and could lead to prevent replication of the dengue virus.      Chemical structure of top-ranked reported protein inhibitors.