Dengue infection, also known as DENV infection, is a serious and global viral health concern because of its morbidity and mortality rates. These rates are because there are no proper treatments for severe dengue infection available at present (Kan et al., 2020). Dengue virus can be transmitted mainly through Aedes aegypti and Aedes albopictus mosquitoes. Dengue virus is a Flavivirus that belongs to the Flaviviridae family with a positive-strand RNA (Wahab et al., 2007).

The dengue genome comprises three structural proteins and seven non-structural (NS) proteins. The three structural proteins include the nucleocapsids, also known as core (C) protein, a membrane-associated (M) protein, and an envelope (E) glycoprotein. In contrast, the seven non-structural (NS) proteins include NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5 (Bell et al., 1985). There are four antigenically distinct but closely related serotypes of dengue virus, mainly DENV1, DENV2, DENV3, and DENV4, that exhibit almost 65%–70% of the sequence homology. To date, dengue fever has been reported to be triggered by these four different serotypes. The fifth variant, DENV5, was first isolated in October 2013, and it follows the sylvatic cycle, unlike the other four serotypes which follow the human cycle. The possible cause of the emergence of the new serotype could be genetic recombination, natural selection, and genetic bottlenecks (Mustafa et al., 2015).

NS3 is a trypsin-like serine protease with a catalytic triad made up of HIS51, ASP75, and SER135, and its activity is enhanced by NS2B as the cofactor. This cofactor contributes to the NS3 activity through its hydrophilic region that is responsible for holding and promoting the activation of NS3. In contrast, the hydrophobic region takes part in the membrane association during the cleavage process (Yap et al., 2019). The non-structural protein NS2B-NS3 serine protease plays an important role in the replication of Flavivirus, which is required for the synthesis of polyprotein precursors. Therefore, this protease is considered to be an essential target for the development of anti-DENV drugs.

Previous studies reported quinoline, benzimidazole, and thioguanine, as DENV2 NS2B-NS3 protease inhibitors (Deng et al., 2012; Hariono et al., 2019). It was suggested that an N-heterocycle in the compound’s structure offers a significant scaffold that might contribute to the inhibitory activities of DENV2 NS2B-NS3 protease. Deng and co-workers identified one potential hit from the ethyl 4-hydroxy-3,5-bis((2-(4-nitrophenyl)hydrazinylidene)-methyl)benzoate (22a) scaffold (Figure 1), which exhibited an IC₅₀ of 14.58 µM (Deng et al., 2012). In addition, the inhibitory activity of the N-heterocyclic dimer of compounds attached with 3-nitrophenyl demonstrated protease inhibitory activity with an IC₅₀ of 29.41 µM. Based on these findings, selected key structures containing nitrophenyl, hydrazinylidene, and the N-heterocycle were subjected to complex-based pharmacophore computational modeling. Important pharmacophoric features towards determining structural requirements for dengue protease inhibitor development and its feasibility to be synthesised on a laboratory scale were elucidated. The pharmacophore-based virtual screening obtained a nitro-benzylidene phenazine derivatives library, which was then selected to screen for lead inhibitors toward DENV2 NS2B-NS3 protease. These derivatives were then synthesised employing the Schiff base reaction, which is considered a cost-effective reaction feasible for upscaling.

Computer-aided drug design (CADD) and its application in virtual screening can be used to screen for a hit compound and further optimise it into a lead (Hariono et al., 2021). In this work, a pharmacophore model was generated based on the docked structure of 22a into the DENV2 NS2B-NS3 protease binding site. In contrast with the conventional ligand-based pharmacophore approach, which cannot provide detailed structural information during lead optimisation, the structure-based pharmacophore provides an alternative approach to address the ligand-based pharmacophore limitation. Despite the recently reported DENV protease-inhibitor complexes at the allosteric site (Yildiz et al., 2013; Yao et al., 2019), the unavailability of a crystal structure of the NS2B-NS3 protease in complex with an inhibitor at its catalytic pocket hinders the adaptation of this approach. Therefore, to obtain appropriate binding of 22a (IC₅₀ = 14.58 µM) in the active site of DENV2 NS2B-NS3 protease, molecular docking was conducted and followed by structure-based pharmacophore generation using Ligand Scout (Inte:Ligand GmbH, Vienna, Austria). Using the heuristic approach and template-based numeric analysis, the ligand is perceived and interpreted according to a probable molecular topology along with the hybridisation states and bond types from (often ambiguous) geometric information (Gaurav and Gautam, 2014). Subsequently, the generated pharmacophore was validated against active and decoy compounds, as detailed in the methodology. Upon retrospective validation, the pharmacophore model was used as queries to screen hits from the nitro-benzylidene phenazine database. These hits were then studied in terms of their binding mode, true positive potential inhibitor, drug-likeness, potential toxicity and mutagenicity, and pharmacokinetic profiles.

Recently, two novel inhibitors (DC-RA016 and DC-RA052) were successfully identified through structure-based drug design against SARS-CoV-2 S-RBD/ACE2, which then were experimentally verified (Xiong et al., 2021). In addition, MaleLiu et al., 2021 described developing a high-throughput screening model as a critical technology to screen for S protein-ACE2 blockers, focusing on virtual screening to offer concepts for rapid discovery (MaleLiu et al., 2021). Moreover, successful predictions of stability and binding affinities of the 11 (re)designed nanobody scaffolds (Nbs) based on the structures of the five Nbs in complex with protein receptor-binding domain (RBD), suggest that the appropriate computational approach can help in the design of efficient SARS-CoV-2 infection inhibitors (Yang et al., 2021). Besides, a computational study employing conformation-based virtual screening uncovered SARS-CoV-2 Mpro inhibitors based on molecular dynamics and the binding free energy analysis (Yang et al., 2021).

The docking of the reference compound 22a showed a ΔG_bind of -9.36 kcal/mol (Deng et al., 2012), which is slightly lower than compound 1855 (-8.90 kcal/mol) (Hariono et al., 2019). Considering the docking error of ±2.0 kcal/mol, these two compounds have a comparable binding affinity towards DENV2 NS2B-NS3 protease. The binding mode of 22a depicted in Figure 2, shows H-bond interactions with the S1 pocket (HIS51, SER135, and ASP129), S2 pocket (ASN152), and S4 pocket (VAL155 and ILE36). Other H-bond interactions with non-specific pockets also occur with PRO132 and ARG54. In the S4 pocket, the NO₂ group interacts with VAL155, whereas hydrazine NH interacts with ASN152 of the S2 pocket. The other hydrazine NH is exposed to the S1 pocket by interacting with SER135. In addition, the hydrazine N = interacts with HIS51, whereas the ethoxy ester O interacts with ASP129 at the S1 pocket. Although it is not explicitly exposed to the sub-pocket, another NO₂ group and the hydrazine NH interact with ARG54 and PRO132, respectively. These binding modes describe the initial pharmacophore feature of 22a that potentially inhibits the DENV2 NS2B-NS3 protease. Figure 2 illustrates the docking profile of 22a in the binding pocket of DENV2 NS2B-NS3 protease.

The pharmacophore of 22a was successfully generated using a structure-based pharmacophore method. As initially determined by the molecular docking, the common features of the pharmacophores are identified as follows: five hydrogen bond acceptors (HBA) at NO₂ (N), hydrazine (N = ), and carbonyl’s O with the number of features of 2, 2, and 1, respectively; three hydrogen bond donors (HBD) at hydrazine (NH), OH, with the number of features of 2 and 1, respectively; and one hydrophobic at ethyl group. These common pharmacophoric features strongly agree with the binding mode demonstrated by molecular docking. The 3D structure of 22a pharmacophore queries is presented in Figure 3.

The pharmacophore model of 22a forms a heptagonal shape with inter-distances from feature to feature as follows: HBA-H (4.19 Å), H-HBA (8.54 Å), HBA-HBA (7.17 Å), HBA-HBD (1.32 Å), HBA-HBD (4.07 Å), HBD-HBA (7.89 Å), HBA-HBD (6.24 Å), and lastly HBD-HBA (1.32 Å). These distances represent the relative positions of the functional groups that are essential for the binding affinity with the DENV2 NS2B-NS3 protease. Before the pharmacophore model is carried out in the screening for hits, a retrospective validation must be done to confirm the capability of the model to predict whether the compounds are true positive or false positive (Hariono et al., 2021).

A series of compounds that had been tested for their inhibitory activity against DENV2 NS2B-NS3 protease with the IC₅₀s ranging from 7.46 to 41.24 µM was categorised as an active set. In contrast, a series of compounds showing IC₅₀ > 50 µM against the corresponding enzyme was categorised as a decoy set. The result shows that the pharmacophore model can predict 10 out of 13 true positive compounds and 4 out of 11 false positive compounds. This means that the pharmacophore model has a strong predictive capability against unknown compounds regarding its activity against DENV2 NS2B-NS3 protease. The receiver operating characteristic (ROC) curve (Figure 4), showed AUC values of 1.00, 1.00, 0.79, and 0.65 at 1, 5, 10, and 100%, respectively. Since these AUC values are close to 100%, they are considered good values. In addition, the Enrichment Factors (EF) reflect that in the thresholds 1, 5, 10 and 100%, the active compound was found at 0, 1.8, 0.9 and 1.3, respectively, times greater than expected. Thus, based on the AUC and EF values, the pharmacophore model has a strong hypothesis to predict hits that act as DENV2 NS2B-NS3 protease inhibitors.

Subsequently, two hits, compounds 4 (fit score = 55.84) and compound 5 (fit score = 55.11) were identified (Figure 5). Although these two compounds could only fit in at least 50% relative position of the pharmacophore model, their potency can still be optimised as leads for DENV2 NS2B-NS3 protease inhibitors. The common pharmacophoric features of compound 4 which fit the model are 5 HBAs at NO₂ (O), hydrazine N = , and heterocyclic N. Compound 5 has slightly different common features as it can fit with 3 HBAs at OH (O), NO₂ (O), and heterocyclic N. In addition, compound 5 possesses 2 HBDs at OH (H).

The two hits were then investigated for their ΔG_bind and binding modes via molecular docking. As presented in Table 2, the ΔG_bind of compound 4 (-9.48 kcal/mol) is almost similar to the value of the reference compound (22a; -9.36 kcal/mol). Thus, it can be assumed that these two compounds might also have DENV2 NS2B-NS3 protease inhibitory activities. However, H-bond interactions are only observed with ARG54 and PRO132, and the estimated K_i value is 0.11 µM. The predicted K_i is much lower than the experimental IC₅₀ of 14.58µM, indicating that compound 4 might be more active than 22a. On the other hand, compound 5 shows higher ΔG_bind (-7.72 kcal/mol) than those of 4 and 22a. However, this energy was contributed by diverse H-bond interactions with ILE36, HIS51 and ARG54, resulting in the estimated K_i equal to 2.21 µM, which demonstrates high potency as shown by compound 4. Figure 6 illustrates the binding conformations of compounds 4 and 5 in the binding pocket of DENV-2 NS2B-NS3 protease.

The non-bonding interactions such as van der Waals (vdW), unfavourable acceptor-acceptor (UAA), pi-cation and pi-alkyl also support the binding of the compounds in the active site that lead to the stable conformation. As summarised in Table 3, vdW interactions are major non-bonding interactions in both selected hits. Other residues such as ARG54, PRO132, HIS51 and ILE36 are also involved in the non-bonding type, in addition to major contributions provided through H-bond interactions. Figure 7 illustrates the non-bonding type environment of compounds 4 and 5, which might support their docked conformations.

The selected hits also passed the PAINS filter; thus, they might not give a false positive result due to the interference such as aggregating compounds, covalent bonding, and chelate formation when tested in vitro (Baell & Holloway, 2010). The hits are said to possess a drug-like structure when they meet the requirement as postulated by Lipinski’s Rule of Five, in which drugs should have a maximum of 500 g/mol MW, 5 for LogP, 5 for HBD, and 10 for HBA (Lipinski, 2004). Other drug-like structure rules also require that the drug have a maximum of 10 rotatable bonds and 140 in the surface area. Therefore, according to Table 4, both hits are slightly over in the MW, LogP and surface area.

The mutagenic and toxic potency of compounds were predicted by their AMES toxicity (No), hMTD (>0.477), hERG 1 and hERG 2 inhibitor (No), ORAT and LOAEL (>2000 mg/kg, hepatotoxicity (No), skin sensitisation (No), TP toxicity (>0.5) and Minnow toxicity (>-0.3) (Huerta E, 2007; Barlow et al., 2009; Sorell, 2016; McCormick, 2017; Gadaleta et al., 2019; Hariyono et al., 2021). From the results presented in Table 5, both hits are likely to be potentially toxic and mutagenic, raising an alarm that the further process would need structural optimisation. On one hand, the optimisation of the structure should maintain the activity toward DENV2 NS2B-NS3 protease, but on the other hand, the toxicity and mutagenicity potential should be reduced accordingly.

Good absorption is one of the most important pharmacokinetic profiles that the hits should meet. Both hits showed similar absorption profiles (Table 6), in which they meet the requirement for water solubility (>-4), human intestinal absorption (100%), and skin permeability (<-2), with no action as the Pgp substrate. It should be noted, however, that 5 does not pass the criteria set for Pgp substrate as this compound potentially acts as a substrate for this protein. In addition, both compounds fail Caco2 permeability (<0.9) and PgpI and PgP II inhibitors criteria (Lin & Yamazaki, 2003; de Angelis and Turco, 2011; Hariyono et al., 2021).

The second pharmacokinetic profile should be passed by the distribution of the hits. Table 7 shows that 4 does not meet two criteria, i.e., VDss (human), since the criteria should be > -0.15, and CNS permeability (<-3). Moreover, 5 fails in three criteria as it does not meet the requirement of BBB permeability, which should be < -1 (Ballabh et al., 2004). Interestingly, both hits passed the fraction unbound in humans, in which the values are above 0.15. Overall, 4 shows a better distribution compared to 5.

Metabolism is another important pharmacokinetic profile, which most likely changes the active drug to its inactive metabolite (Kinirons and O’Mahony, 2004). Table 8 shows that compound 4 demonstrates a better metabolism profile than compound 5. However, both are neither CYP2D6 substrates nor inhibitors. Furthermore, 4 does not act as CYP1A2 and CYP3A4 inhibitors. In contrast, 5 acts as the substrate of CYP3A4, CYP1A2 inhibitor, CYP2C19 inhibitor, CYP2C9 inhibitor and CYP3A4 inhibitor. This indicates that both hits can induce or inhibit a certain CYP, which might result in either the under-dose or over-dose of another drug metabolised by that certain CYP.

Excretion, in which the metabolite is mostly eliminated from the body via urine, is also an important predictive pharmacokinetic profile (Rosenbaum, 2016). Compound 4 is predicted to be excreted from the body at a faster rate than 5 due to its highest total clearance. In contrast, 5 might undergo slower removal from the body due to its lowest total clearance. Despite this, both hits have low total clearance (logCl < 0.763), and it is generally desirable to develop a drug for oral administration without a high dosage regimen (Toutain et al., 2004). Interestingly, none of the hits is predicted to act as renal OCT2 substrate, which might lead to undesirable side effects (Burt et al., 2016). Table 9 presents the total clearance of all hits that reflects their speed to be eliminated from the body system.

Additionally, to assess the dynamic evolution and the stability of compounds 4 and 5 in complex with the NS2B-NS3 protease model of DENV2, molecular dynamics simulations were carried out for 50 ns. Generally, both systems demonstrated stability and consistent RMSD values with an average of 1.65 ± 0.17 and 1.71 ± 0.22 Å for 4- and 5-NS2B-NS3 protease, respectively, with a slight increment to a maximum of 3.06 Å in compound 5 complex at around 27.5 ns (Figure 8A). The slight elevations in the RMSD at 27.5 ns and the RMSF of GLU19 of compound 5 are attributed to its continuous conformation change between the S1 and S2 sub-pockets of the protease, yet it was not seen to leave the main binding pocket (Figure 9). Likewise, the RMSF profiles (Figure 8B) of NS2B-NS3 protease in complex with both compounds displayed comparable fluctuation patterns of the amino acid residues with higher fluctuation for compound 5 seen with GLU19 at around 3.7 Å. In addition, RadGyr analysis also showed a coherent rigidity and compactness of the systems (Figure 8C). The average RadGyr for compound 4’s and 5’s systems were 14.13 ± 0.20 and 14.11 ± 0.16 Å, respectively, and both systems were seen to fluctuate within less than 0.9 Å. Slight elevations were noticed between 48 and 49 ns for the system with compound 4 peaking at 14.40 Å, and for compound 5’s system to around the same value between 49 and 50 ns. The slight elevation in the RadGyr profile in the case of compound 4 is due to the phenazine ring system protruding out of the protease binding pocket (Figure 10).

The H-bond interaction analyses of both systems at the catalytic binding site of the protease demonstrated that compound 4 was able to make up to five H-bonds throughout the 50 ns simulation time (Figure 8D). However, there was no persistent H-bond formed between any of the possible HA of compound 4 in contrast to that proposed by the docking. The two H-bond interactions with SER135 (one of the catalytic residues) and ARG54 with low occupancies of 0.21 and 0.10%, respectively demonstrated in Supplementary Table S3 are attributed to the oxygen and nitrogen atoms of compound 4 acting as acceptors. On the other hand, compound 5 exhibited a maximum of four H-bond interactions throughout the 50 ns simulation, with their frequency higher than compound 4 (Figure 8E). Six H-bond interactions at the binding site of the protease with higher occupancies were observed especially with HIS51 (11.57%), which is a catalytic residue of the protease in addition to VAL52 (1.83%), GLY55 (0.76%), GLY133 (0.61%, total), and ARG54 (0.11%). These six interactions are attributed to two oxygen atoms of compound 5 acting as acceptors, and some hydrogens atoms of the 5’s oxygens and nitrogens acting as donors (Supplementary Table S3). The low occupancies of the H-bonds suggested that the H-bond interactions do not play an important role in the binding of the ligands. This is supported by the MM-PBSA energy components analyses that indicated the van der Waals interactions are the main contributors to the stability of the compounds in the protease binding site. MM-PBSA calculation is also agreeable to the other predictions that compound 4 has a better affinity towards the binding to the protease with ΔG_bind of -22.53 ± 4.21 kcal/mol than compound 5 (-17.01 ± 3.14 kcal/mol) (Supplementary Table S4).

Dengue type-2 (DENV2) is the serotype most commonly found endemic in South East Asia (Halstead, 2006). Still, no specific drug clinically inhibits viral replication with a clear understanding of its molecular mechanism. This might be due to the limited availability of a 3D crystal structure of the virus that is required for further investigations. For example, the NS2B-NS3 protease of DENV2 is only available as an apo-enzyme (Erbel et al., 2006). Therefore, there is a limitation in confirming the coordinate of the ligand binding when the structure-based drug design is implemented. Lately, a crystal structure of DENV2 NS2B-NS3 protease complex with peptide inhibitor from mung bean was proven invalid. Hence, all related publications were retracted. Fortunately, homology modelling has been performed to solve this problem by utilising the DENV2 apo-enzyme structure with the ligand from West Nile Virus (WNV) NS2B-NS3 protease crystal structure (Wichapong et al., 2010). This model was successfully used in finding hits to lead in a few projects of DENV2 NS2B-NS3 protease inhibitor discovery bearing thioguanine (Abduraman et al., 2018; Hariono et al., 2019), benzofuranone (Sulaiman et al., 2019), and malabaricone-acylphenol (Sivasothy et al., 2021). Consequently, the homology model of DENV2 NS2B-NS3 protease structure was chosen in designing the corresponding target inhibitor from the nitro-benzylidene phenazine compound.

A structure-based design strategy was adopted based on the biological potential of compound 1 bearing p-hydroxyethyl benzoate ester linked to benzimidazole rings by hydrazine bridges, albeit at low micromolar potency. Analogue 22a was obtained by modifying the nitrophenyl ring and was shown to be active against the protease enzyme. Subsequently, screening of the National Cancer Institute (NCI) compounds database identified a series of N-heterocyclic compounds linked to nitrophenyl rings with inhibitory activity against DENV2 NS2B-NS3 protease. These compounds are associated with pharmacophore features like hydrazine bridges (HBD/HBA), nitrophenyl (HBA/H), and N-heterocycle (HBA/H), up to the compound having nitro-benzylidene phenazine structure.

Compound 22a was used as a reference compound to validate the generated pharmacophore model, which corroborated the latest findings upon retrospective validation. The developed pharmacophore model allowed the identification of a true positive DENV2 NS2B-NS3 protease inhibitor than its false positive compound. A validated pharmacophore model was then used to screen the hits of DENV2 NS2B-NS3 protease inhibitor from a series of nitro-benzylidene phenazine analogues. The nitro-benzylidene phenazine was designed based on the consensus between the common pharmacophore features of DENV2 NS2B-NS3 protease inhibitor and its synthesis feasibility, resulting in 13 designed ligands. Despite the identification of only two ligands (4 and 5) as hits from the structure-based pharmacophore screening, they are predicted to become promising candidates for further optimisation.

Results from molecular docking revealed low values in their ΔG_bind for both compounds 4 and 5 at the DENV2 NS2B-NS3 protease catalytic binding site. This suggested the high affinity of the hits towards the enzyme. Furthermore, the binding mode of the two hits demonstrated similar key interactions as in the case of 22a with the amino acid residues lining the binding pocket. Designing new ligands that could occupy the catalytic region might interfere with substrate binding ability and thus could be a potential target site to design such specific ligands that can interfere with the protein function.

In general, ligands can exhibit false positive activities in an in vitro assay. Herein, prediction using the PAINS filter was applied. Undetected PAINS with false-positive activities are often due to covalent modifications, redox effects, chelation, autofluorescence, or degradation. The study of pharmacokinetics refers to the investigations of drug-likeness, mutagenic/toxicity and pharmacokinetic predictions on the potential new hits before entering the next process; lead optimisation.

One of the main factors that limit oral drug absorption might be the large structure of compounds 4 and 5. Furthermore, they show possible human intoxication and mutagenic effects. This could be due to the presence of a nitro aromatic group which had been identified to possess a mutagenic effect (Chin Chung et al., 2011). The nitro group is a unique functional group in medicinal chemistry as a strong electron-withdrawing and a relatively polar group. Although found in medicines for a long time, the nitro group has toxicity issues due to metabolic liabilities. Therefore, it is often considered a structural alert or toxicophore, leading to mutagenicity and genotoxicity.

However, numerous nitro-drugs take advantage of bio-reductive activation mechanisms, mainly in the antibacterial and antiparasitic fields. Recently, renewed interest in nitro drugs has led to the approval of fexinidazole for the treatment of human African trypanosomiasis (HAT), a neglected tropical disease. Due to its unique properties, the nitro group has been found to have great interest in the development of anticancer, antitubercular and antiparasitic agents, as well as hypoxia-activated prodrugs. In addition, nitro groups have been found in the development of self-immolative spacers (linkers) to overcome limitations for drug delivery and have attracted wide interest in medicinal chemistry. In this context, many studies have been undertaken on the structure-mutagenicity or structure-genotoxicity relationships of drugs containing nitro groups (https://www.mdpi.com/journal/pharmaceuticals/special_issues/Nitro_Drugs).

During the distribution, both hits (compounds 4 and 5) were found unable to reach the maximum dose due to their low VDss. It was observed that both hits are capable of inhibiting or inducing the CYP enzyme activities in combination medicines that include other active ingredients in the presence of the compound with no significant changes in the excretion process. As shown in Supplementary Table S5, the % efficiency of both hits in terms of drug-like structure, toxicity, absorption, distribution, metabolism and excretion revealed less than 100%, with only 52–55% efficiency for nitro-benzylidene phenazine to be considered for further lead optimisation. This result suggested substituting some functional groups with their bioisostere in future work. For example, the nitro group has bioisosteres such as trifluoromethyl (CF₃) group in CB₁ receptor positive allosteric modulation (Tseng et al., 2019). An earlier study has identified the bioisostere of a nitro-aromatic group with some groups including benzo [c][1,2,5]oxadiazole, isobenzofuran-1(3H)-one, and benzonitrile (Meanwell, 2011) (Figure 11).

As compounds 4 and 5 are interesting and suggested for further lead optimisation, the binding stability of those compounds in the DENV2 NS2B-NS3 protease active site should be investigated. Molecular dynamics simulations demonstrated that compounds 4 and 5 exhibited excellent stability when in complex with the DENV2 NS2B-NS3 protease model. Due to the continuous conformation change of compound 5 between the S1 and S2 sub-pockets of the protease that was shown earlier, compound 5 was seen to contribute through higher H-bond interaction occupancies, especially with HIS51, which constitutes one of the DENV protease catalytic residues. Moreover, although compound 4 showed an H-bond interaction with SER135 with limited occupancy, it can be concluded that both compounds have their total free energy of binding contributions through vdW and non-polar interactions with the amino acids lining the main binding pocket of the protease. All these findings confirm and support the results obtained from the molecular docking earlier, where ARG54 does not contribute mainly through H-bonds with both ligands rather than through non-bonding interactions. Moreover, the unfavourable acceptor-acceptor interaction between PRO132 and compound 5 that was seen from the docking results might have influenced its higher ΔG_bind from the MM-PBSA analysis, not to mention that both the molecular docking and dynamics studies agree that compound 4 seems to have a higher binding affinity towards the protease than compound 5.

A pharmacophore model for DENV2 NS2B-NS3 protease inhibitor has been generated from the docking pose of ethyl 4-hydroxy-3,5-bis((2-(4-nitrophenyl)hydrazinylidene)-methyl)benzoate (22a) in the binding pocket of DENV2 NS2B-NS3 protease. The pharmacophore model employing 5 HBAs, 3 HBDs and 1 H was able to identify two hits from the nitro-benzylidene phenazine series. These two hits show potential binding in the DENV2 NS2B-NS3 protease based on their molecular docking results. However, after drug-likeness, mutagenic/toxicity and pharmacokinetic prediction, these two compounds resulted in 52–55% efficiency as drug candidates while exhibiting stability at the catalytic pocket of the protease mainly through vdW and non-polar interactions. The challenge of these hits is due to their mutagenic-toxic effect on the nitro aromatic group. Therefore, further studies should consider the replacement of this group with its bioisostere to improve the bioactivity against DENV2 NS2B-NS3 protease.