Exploring the viral protease inhibitor space driven by consensus scoring-based virtual screening

Abstract

Dengue fever presents a major health concern, and the lack of an effective vaccine or definite therapeutic regimen has led the research community to identify safe-by-design potential targets for drug discovery. Since the association of the NS2B co-factor with the protease domain of NS3 is imperative for the catalytic activity of the enzyme complex, inhibitors blocking their interaction could provide an alternative strategy to combat the dengue virus. In this context, the present study is aimed at exploring computer-assisted modeling of significant physicochemical features required for the inhibition of the dengue virus protease complex. First of all, alanine scanning was utilized to map hot spot residues critical for the association of the two subunits, NS2B and NS3pro, by studying their energy profiles. Then, consensus score-based virtual screening was performed to search through the commercially available chemical datasets. After screening, 1,575 small molecules were moved forward into docking studies to investigate their interactions with crucial interfacial residues (i.e., Tyr23, Lys26, Phe46, and Leu58), with only 233 molecules passing that stage. The top 30 molecules were selected based on a detailed profile of intermolecular interactions. After that, the top five molecules were selected for detailed mechanistic studies via molecular dynamics simulations followed by subsequent binding free energy calculations, principal component analysis in conjunction with free energy landscape. To the best of our knowledge, this is the first systematic and comprehensive investigation to identify protein–protein interaction blockers against the target protein at such a large scale, using integrated computational tools. Our results highlight the enhanced stability and good binding affinities towards the target protein of these compounds, which might act as new scaffolds for NS2B–NS3 protease inhibition. Future studies will be directed to explore the detailed atomistic-based structural and energetic framework of the mutation-induced affinity change between the protease domain of the DENV-2 NS3 protein and its cofactor NS2B. The detailed insight in turn might suggest precise and focused targeted points for the structure-based drug design but the computational cost may be a challenge.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s40203-023-00174-0.

Introduction

The emergence of viral diseases, in particular, vector-borne viral diseases, poses an extreme threat to mankind, causing mild to severe infections and paradoxically premature deaths (Marchi et al. 2018). Dengue virus (DENV), an arthropod-borne virus, is primarily vectored by infected Aedes mosquitoes (Aedes aegypti and Aedes albopictus). It belongs to the genus Flavivirus within the family Flaviviridae and is phylogenetically related to other notorious viruses, i.e., Zika virus (ZIKV), Japanese encephalitis virus (JEV), and West Nile virus (WNV) (St John and Rathore 2019). The four serotypes of DENV (DENV1-4), the etiological agents of dengue, infect approximately 400 million individuals per annum, with more than a quarter of the world’s population residing in DENV endemic areas (Pierson and Diamond 2020). DENV infection in humans is associated with a wide spectrum of clinical manifestations, which range from mild dengue fever (DF) to severe dengue hemorrhagic fever (DHF) and life-threatening dengue shock syndrome (DSS) (Xu et al. 2020). The epidemiological data suggest an alarming increase in dengue during the past years and levels are expected to rise many folds in the future. Population density, urbanization, habit modification, climate change, and poor vector control measures are some of the factors attributed to the geographical expansion of DENV (Ketkar et al. 2019; Messina et al. 2019).

The DENV, a spherical particle of icosahedral symmetry, is encapsulated by a membranous envelope shielding the viral genome—specifically a monopartite positive sense single-stranded RNA. The ∼11 kb genome encodes a long precursor polyprotein within the host cell utilizing both cellular and viral machinery to increase viral progeny (Lim et al. 2020). The co- and post-translational processing of the polyprotein driven by cellular and viral proteases liberates functional viral components. The three structural proteins, the capsid (C), the precursor-membrane (prM), and the envelope (E) proteins, encase the genetic material of the virion. Further, the remaining portion encodes the seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) which are crucial for the virus life cycle owing to their indispensable role in virus replication and maturation (Gotz et al. 2021).

The DENV protease complex has emerged as a potential drug target for therapeutics development. Its viability as a target has been supported by the development of marketed protease inhibitors for hepatitis C virus (HCV) and human immunodeficiency viruses (HIV) (Kuhl et al. 2020). The DENV protease complex (NS2B–NS3pro), belongs to the family of chymotrypsin-like serine-proteases and harbors a classical catalytic triad formed by His51, Asp75, and Ser135. Once activated, this complex mediates the cleavage of the viral precursor polyprotein between the junctions of NS2A–NS2B, NS2B–NS3, NS3–NS4A, and NS4B–NS5, releasing mature non-structural proteins (Erbel et al. 2006; Yao et al. 2019). Further, this enzyme conjugate is also responsible for circumventing the host’s immune response by cleaving various protein mediators implicated in defense (Majerova et al. 2019). Thus, the critical role of NS2B–NS3 protease for virus maturation and propagation has been exploited to develop drugs against DENV which resulted in several discoveries of novel inhibitors representing distinct chemotypes. However, the documented activities of those compounds are quite low, and the follow-up hit-to-lead development is often missing for these hits.

Despite its significant clinical importance and growing distribution, there is still no specific therapeutic available for the treatment of DENV infection. The prime preventive means currently available to protect people from infection are circumvention of bites from infected mosquitoes and vector control measures. However, unfortunately, presently, neither of the strategies is effective enough at controlling both the vector population and escalation in transmission, simultaneously (Overgaard et al. 2018). Moreover, the advancements in vaccine development have been hampered by antibody-dependent enhancement. Dengvaxia^® (CYD-TDV), developed by Sanofi Pasteur, is the first licensed vaccine candidate against DENV in the market (Martinez et al. 2021). Unfortunately, due to its limited efficacy against acute dengue and increased rate of severe illness and hospitalization upon vaccination in dengue-naïve individuals, the administration of Dengvaxia is restricted to seropositive individuals. Accordingly, the present treatment options just focus on alleviation of the infection symptoms. Thus, developing new prophylactic agents and effective therapeutic interventions remains an urgent need (Organization 2019; Torres et al. 2019).

The endeavor of drug discovery is extremely tedious and multifaceted, and researchers are frequently demoralized by the incessant possibilities one has to search through. Fortunately, computer-aided drug design (CADD) has certainly accelerated the process of drug discovery (Santos et al. 2019; Sethi et al. 2020). Since nuclear magnetic resonance (NMR) and molecular dynamics simulations (MD) indicated the imperative role of NS2B in providing stability and proper catalytic activity to NS3pro (Gupta et al. 2015), the NS2B–NS3pro protein–protein interaction interface is a potential target site for drug discovery. The significance of this site has been endorsed by at least two studies, conducted by Pumbudi et al. (2013) and Li et al. (2018), which resulted in the identification of SK-12 and erythrosine B, respectively, as potential candidates targeting the interfacial site.

Hence in the present study, a combination of in silico modeling techniques was employed to search for new small molecules targeting the binding pocket on NS3pro that holds the cofactor protein, NS2B. This might serve as an alternative and superior drug discovery approach, to find a small molecule that can lock the enzyme in an inactive conformation preventing the required functioning of NS3pro and rendering DENV virions unable to replicate.

Materials and methods

Structure retrieval and preparation

The atomic structure employed in the present investigation for the DENV2 protease complex was retrieved from RCSB Protein Databank (PDB) under the accession code 2FOM, an X-ray diffraction structure with a good resolution of 1.5 Å (Erbel et al. 2006). This structure contains two subunits from different DENV proteins linked via the G4SG4 linker, i.e., the central hydrophilic portion of NS2B (Chain A: 62 residues) and the protease domain from NS3 (chain B: 185 residues). At first, a preprocessing step was performed in which all molecular species other than the protein were removed. The truncated portions were modeled, where required, via Modeller version 9.22 (Fiser et al. 2000). Afterward, the target protein was modified to ensure correct bond orders and numbers of hydrogens using the structure preparation wizard of Molecular Operating Environment v2019.01 (MOE) (CCG Inc 2021). Subsequent steps involved the application of partial charges (AMBER99) (Wang et al. 2000) and the release of structural restraints to stabilize bond energies using energy minimization. The RMS gradient and the dielectric constant were kept as default.

Computational mutagenesis of NS3pro interfacial residues

In an attempt to explore mutation-induced differences in NS3pro stability and NS3pro/NS2B protein interaction free energies (ΔΔG kcal/mol), we selected nine residues (Tyr23, Arg24, Ile25, Lys26, Gln27, Phe46, Leu58, Met59, and His60) distributed at the interface region of NS3pro. It is noteworthy to mention here, that the mutations were only introduced in NS3pro (Chain B) since it is the central unit of the DENV protease, possessing the enzymatic activity. Primarily, the “protein design” menu incorporated in MOE was employed, to determine the predicted changes in stability and affinity caused by the single mutation. We used the “Alanine scan variant” which substituted the particular residues with alanine and predicted the resultant change in binding energy. The structural coordinates utilized in the mutational investigation were derived from the prepared structure of DENV protease. The mutations were only introduced in chain B and we limited the movement of residues more distant than 4.5 Å from the residue to be mutated. The threshold for the rotamer explorer deviation and energy window were 0.25 Å and 10 kcal/mol, respectively. For a systematic conformational search, the ensembles were generated under the Low Mode MD search method with search iterations up to 50 (Kalathiya et al. 2019). In parallel, to ascertain the validity of the results and to further evaluate the prediction accuracy of the energy differences, two other web servers, i.e., Mutabind2 (Zhang et al. 2020) and mCSM-PPI2 (Rodrigues et al. 2019) were taken into consideration. With these web servers, the wild-type DENV protease structure (2FOM) was loaded as input followed by the substitution of the targeted residues with alanine. For alanine scanning, the relative differences in the energies obtained reflect the contribution of the respective amino acid in deriving the association, i.e., the greater the difference, the more critical the amino acid is for the formation of an assembly. The ribbon diagram and active residues involved in the interaction are depicted in Fig. 1.

Fig. 1.

A 3-D structure of DENV protease complex (NS3pro: sky blue; NS2B: goldenrod) with catalytic triad shown in spheres (His51: green; Asp75: orange; Ser135: blue); parts of protein surface shown in white. B Structural details of the binding interface between NS3pro (sky blue ribbons; select sidechains shown with sky blue sticks) and the cofactor protein NS2B (goldenrod ribbons; select sidechains shown with goldenrod sticks)

Modeling the mutant complexes via HADDOCK

To further validate the structural determinants involved in molecular recognition and to validate the energy profiles from alanine scanning, the generated mutants of NS3pro were docked with wild-type NS2B via a well-documented open-source docking platform, HADDOCK2.2 (High Ambiguity Driven protein–protein Docking), which has an easy interface (Zundert et al. 2016). The software is tailored to model and analyze protein–protein interactions using ambiguous interaction restraints (AIRs) at the interface. Here the docking process is guided by the AIRs which in turn is dependent on the description of the active and passive residues. Arg24, Ile25, Lys26, Gln27, and Met59; and Leu51, Glu52, Leu53, and Glu54, respectively, were defined as active residues for NS3pro and NS2B. Passive residues were the ones surrounding the active residues.

The clusters generated after a three-step docking protocol mentioned elsewhere (Vries et al. 2010), are ranked according to their average HADDOCK score, which is a summation of various interaction energies (electrostatic, van der Waals, and AIRs energy terms) and the buried surface area (BSA). The clusters with the lowest energy score were kept for further analysis.

Virtual screening

In this investigation, a two-step docking strategy was employed comprising a rigid-body consensus scoring (Utilization of multiple scoring functions) docking followed by refinement using induced-fit docking.

Preparation of small molecule libraries

Two small-molecule libraries (the ZINC protease subset (Irwin and Shoichet 2005) and ICCBS in-house library) containing a total of 76,499 compounds were explored in search of potential inhibitors against DENV protease. These small molecules were subjected to optimization in MOE in which the Protonate3D module was used to adjust the hydrogens and lone pairs. Subsequently, partial charges were assigned (using MMFF94X) (Halgren and Nachbar 1996), after which all compounds were energetically minimized with an RMS gradient threshold of 0.1 kcal/mol/Å and dielectric constant of 1. To filter out the compounds with the best physicochemical properties, the Lipinski drug (Lipinski et al. 1997) and Oprea lead-like filters (Oprea et al. 2001) were applied using the descriptor calculation module. The screened libraries were then used for successive molecular docking steps. Two of the reported small molecules, erythrosin B and SK-12, were sketched in ChemDraw Ultra 9.0 (Cousins 2005) and prepared via the same protocol as mentioned above.

Molecular docking

The binding energetics of the intermolecular complexes in docking are estimated using the implemented scoring functions (SFs). Since none of the standalone scoring functions is efficient in providing an accurate ranking for all chemotypes, the known benefits of consensus scoring inspired us to utilize multiple SFs to rank the compounds from the libraries (Warren et al. 2005; Perez-Castillo et al. 2019).

Consensus docking

Rigid Receptor Docking: The first step of docking was performed with the MOE suite. The docking centroid (x; 10.09, y; − 13.51, z; 18.11) was based on the following residues from NS3pro: Tyr23, Arg24, Ile25, Lys26, Gln27, Phe46, Leu58, Met59, and His60. The docking process was started with an initial search of the conformational space of each molecule followed by the generation of docking poses from the pool of conformations using the Triangle matcher algorithm in the placement stage. Then, the determined docked poses were scored with the default SF, London dG. Afterward, the best-scored poses were carried to the refinement phase in which the rigid receptor protocol was employed. In the last stage, the GBVI/WSA dG SF score was calculated. Further considering the limitations of a single scoring function, the first step of scoring was repeated with other available SFs in MOE, for instance, Affinity dG, ASE, Alpha HB, and GBVI/WSA dG (Ye et al. 2020), while the other parameters were kept the same.

In the following step, ligands were subjected to molecular docking via AutoDock Vina (Trott and Olson 2010). The required input files in PDBQT format were generated using the Python scripts available in AutoDock Tools (Huey et al. 2012). The partial charges were assigned as per the calculation of Gasteiger–Marsili and Kollmann charges to the ligands and receptor, respectively. The search space was limited to the interface site with the size and spacing of 46 × 54 × 46 ų and 0.375 Å, respectively. Docking was performed using the search exhaustiveness of twenty. The predicted poses with the lowest affinity energy values were retained for further investigations.

Flexible Docking: The compounds ranked on the top stratum by all the tested SFs were shortlisted and subjected to the flexible docking protocol. Around 1575 small molecules were docked to the targeted pocket of NS3pro. The docking grid was placed on the NS3pro residues as mentioned in rigid receptor docking via MOE with grid dimensions of X; 10.09 Å, Y; − 13.51 Å, Z; 18.11 Å for x, y, and z coordinates respectively. The placement of the ligands in the binding pocket was done using the Triangle matcher algorithm and the generated poses were scored initially scored via London dG scoring using flexible docking. The generated poses were energy-minimized in the targeted cavity followed by scoring with GBVI/WSA dG SF.

Intermolecular interaction profiling

The top-most scoring poses were retained for protein–ligand interaction fingerprinting (PLIF), specifying the crucial interfacial residues highlighted in the mutational analysis. To further confirm the interaction pattern at the molecular level, the hits obtained were further visualized using a protein–ligand interaction profiler (PLIP) (Salentin et al. 2015).

Molecular dynamics simulation

Classical molecular dynamics simulations were run to probe the conformational dynamics and stability of the shortlisted compounds within the interfacial site of NS3pro. The ligands were described with the general Amber force field (GAFF) (Wang et al. 2004) via the ACPYPE program (Sousa da Silva and Vranken 2012). The topologies and coordinate files for the receptor were generated using the Amber ff99SB‐ILDN force field (Lindorff-Larsen et al. 2010). The complexes were initially centered in a cubic box. Solvation was implemented with TIP3P water (Jorgensen et al. 1983). To counter undesirable edge effects, periodic boundary conditions were maintained. The neutrality of the systems was conserved by the addition of counter ions utilizing 0.15 m NaCl solution. Energy minimization was performed via the steepest descent method to remove possible steric clashes, using a convergence limit of 1000 kJ∙mol^–1 nm⁻¹. The systems were equilibrated first under the NVT ensemble (isothermal-isochoric) followed by equilibration at isothermal-isobaric conditions (NPT ensemble) covering 1 ns each. The cut-offs for short-range van der Waals and electrostatic interactions were set at 1.0 nm while the long-range electrostatics interactions were computed with the PME (Particle-Mesh Ewald) method (Essmann et al. 1995). The short-range van der Waals and electrostatic interactions were truncated at 1.0 nm. The Linear Constraint Solver (LINCS) algorithm (Hess et al. 1997) was employed to constrain bond lengths. The velocity-rescale thermostat and Parrinello–Rahman pressure coupling methods were utilized to attain the desired temperature (310 K) and pressure (1 atm), respectively (Bussi et al. 2007; Martoňák et al. 2003). The equilibrated systems were submitted for production runs of 100 ns. The integration time was adjusted to 2 fs, utilizing the leap-frog integrator (Gunsteren and Berendsen 1988). The atomic coordinates were recorded every 10 ps during the MD simulations. All simulations were carried out with the CUDA version of the GROMACS 2020 package (Abraham et al. 2015) on a NVIDIA K20X GPU workstation.

The stability of the targeted complexes was evaluated via standard parameters, i.e., root mean square deviation (RMSD), root mean square fluctuation (RMSF), and radius of gyration (Rog). All the graphs were plotted using Xmgrace (Turner 2005). Chimera software (Pettersen et al. 2004) was employed for molecular interaction diagrams.

MM-PBSA binding free energy calculations

A total of 1000 snapshots were extracted from the MD trajectories and were utilized for the calculation of the free energy of the protein–ligand complexes via the Molecular Mechanics Poisson–Boltzmann Surface Area (MMPBSA) approach using the g_mmpbsa package (Kumari et al. 2014). This approach principally involves calculating the differences in free energies between the ligand, protein, and the complex in presence of a solvent. The free energy for each entity participating in the reaction is described as a sum of a gas-phase energy, polar and non-polar solvation terms, and an entropy term.

Principal component analysis and free energy landscape

Principal component analysis (PCA) was performed using the trajectory of the receptor as apo state and bound with inhouse and zinc compounds; with coordinates of backbone atoms by using the pca module in MDAnalysis tools (Gowers et al. 2016; Michaud-Agrawal et al. 2011) Additionally, the combination of free energy landscape (FEL) with PCA results may provide a useful picture to study system dynamics (Mu et al. 2005; Riccardi et al. 2009; Wang et al. 2023). The FEL are used to investigate molecular folding and aggregation with mapping of all possible conformations of a system in simulation trajectory (Frauenfelder et al. 1991; Li et al. 2022).

$$\mathrm{\Delta }G=-K_{B}TlnP(C{V}_1,V_2)$$

the FEL can be calculated using the formula given above. In this formula $K_B$ and $T$ are the Boltzmann constant and absolute temperature and $P\left(C,V_1,,,V_2\right)$ is the probability distribution of the molecular system along the reaction coordinates variables (CV₁ and CV₂), here we used principal components results as variables and it as done by gmx sham package in gromacs with assistance of mangeshdamre/Free-Energy-Landscape (https://github.com/mangeshdamre/Free-Energy-Landscape).

Results and discussion

Computational alanine scanning

Some potential residues involved in the interaction between NS3pro and NS2B of DENV have been identified by computer modeling, while some experimental investigations reported the crucial residues for catalytic activity (Li et al. 2018; Nitsche et al. 2014). Nevertheless, the actual interactions and their relative energetic importance have not been explored in detail. In this context, we propose that identifying via mutation those specific residues that drive the complexation or conjugation can be exploited in drug design. With this objective in mind, we estimated the change in stability and affinity for all single-site mutations, using MOE alanine scanning. In parallel, we considered the mCSM-PPI2 and mutabind2 web servers to assess the differences in the binding free energies among the wild-type and mutated complexes.

In general, it was observed that the mutation in NS3pro may lead to altered NS2B-binding ability associated with a decreased binding affinity. Y23A, R24A, K26A, F46A, and M59A were identified to be the mutations that caused the largest affinity change, with $\mathrm{\Delta }\mathrm{\Delta }{G}_{wild-mutation}$ > 4 kcal/mol. For Q27A and L58A, the difference in affinity observed was ≥ 2 kcal/mol. In contrast, the alanine variant of I25 had a very small impact with an affinity difference of 1.08 kcal/mol. Surprisingly, for H60A an increase in affinity was observed. Further the tested mutation rendered the enzyme complex unstable as evident from the stability differences predicted by MOE except for Q27A where a negligible impact was observed with difference of < 1 kcal/mol compared to wild type. The mutational landscape and results of in silico alanine scanning are displayed in Fig. 2 while the numeric values are presented in Table S1.

Fig. 2.

A Mutational landscape of NS3pro, for which the residues considered herein are highlighted in stick representation (blue). B Binding energy change (ΔΔG = ΔG_wild-type – ΔG_ALA) computed via MOE from the computational alanine-scanning mutagenesis for DENV NS3pro in complex with NS2B; blue bars represent the difference in affinity (dAffinity) while the differences in stability (dStability) pattern is represented as golden yellow bars. Red dashed lines highlight the largest energy differences

Likewise, Tyr23, Arg24 and Lys26 are suggested as the hot-spot residues of NS3pro, required for proper interaction with the NS2B with the binding affinity differences of − 2.74, − 2.74, − 2.45 and 2.22, 2.69, 3.71 kcal/mol predicted by mCSM-PPI2 and Mutabind2 respectively as depicted in Fig. 3 and Table S2.

Fig. 3.

The changes in binding affinity for all mutations predicted via A mCSM-PPI2 (ΔΔG_Bind = ΔG_{WT Bind} – ΔG_{Mut Bind}) and B MutaBind2 (ΔΔG_Bind = ΔG_{Mut Bind} – ΔG_{WT Bind}). Red dashed lines highlight the largest energy differences

Modeling of mutated complexes using HADDOCK

The generated NS3pro variants were subjected to protein–protein docking. The obtained results revealed a decrease in the binding score for all of the tested variants compared to the wild-type enzyme. HADDOCK score is described as the sum of various empirical and physical energy terms including van der Waals, electrostatic, and desolvation energy. Additionally, the BSA is considered while computing a cumulative score. HADDOCK generated a number of clusters (docking poses) in a single docking run, so we considered those with maximum dock scores. The predicted scores are depicted in Fig. 4 while the components of the total score are listed in Table 1. As evident from these results, the mutations resulted in altered energy profiles and thus are not favorable for the enzyme’s activity. Y23A greatly affected the complex formation between the targeted subunits. Further, significant alterations were also observed for the alanine variants of Arg24, Ile25, Lys26, Phe46, and Leu58.

Fig. 4.

HADDOCK scores generated by protein–protein docking of Wild-type (WT) NS3pro and various NS3pro variants with native cofactor protein, NS2B. Red dashed lines highlight the largest score differences

Table 1.

Profiling of various energy components revealed after protein–protein docking via HADDOCK

Comparatively, the least impact relative to WT was observed for the Gln27A and Met59A mutants. When investigated in detail, the van der Waals and electrostatic energies were found altered for all mutations, resulting in only a weak interaction between the two subunits NS3pro and NS2B. The Tyr23A mutant revealed the greatest change in van der Waals interactions while electrostatic interactions were affected the most for the Met59A mutant. The variants were found to form less well-compacted states relative to WT as suggested by their lower BSA.

Considering all the binding affinities obtained using the different computational platforms, a total of four hot-spot residues were found to impact the energy landscape the most, Tyr23, Lys26, Phe46, and Leu58. Lys26 was predicted to affect the binding affinity to a much greater extent compared to the others. The mutation in Tyr23, Phe46 or Leu58 is also reported to decrease the proteolytic activity of the enzyme (Li et al. 2018). It is important to mention here that the identified residues i.e., Tyr23 and Phe46 are conserved across the documented DENV serotypes, and hence if targeted may favor the development of pan-serotype inhibitors. We explored the small molecule libraries to attempt to identify new inhibitors, capable of antagonizing the formation of the active complex by disrupting the interaction between NS3pro and NS2B.

Descriptor-based filtration of small molecule libraries

Initially, a virtual library of 76,499 compounds (ZINC and in-house database) was screened for drug-likeness and lead-likeness, leaving 51,546 compounds. The physicochemical attributes for the filtered dataset vary between 74.13 and 635.34 for molecular weight (MW); − 2.04 and 8.69 for SlogP; 0 and 8 for H-bond donor (HBD); 0 and 11 for H-bond acceptor (HBA); and 0 and 18 rotatable bonds as listed in Table S3.

Consensus scoring-based virtual screening

The use of a single scoring function to rank compounds in virtual screening has been documented to produce erroneous results as evident from the study of Warren and colleagues. When assessed for a set of eight proteins none of the tested standalone scoring functions yielded accurate ligand binding affinities (Warren et al. 2005). The consensus scoring approach, which combines rigorous and accurate scoring functions, is a useful alternative since it combines the virtues and at the same time attenuates the limitations of standalone methods (Houston and Walkinshaw 2013). In this regard, six different scoring functions implemented in two of the well-known docking tools, MOE and Autodock Vina, were used in this work (Table S4). We advanced only compounds that were highly ranked by all the SFs. Each of the shortlisted hits exhibited good binding energies relative to those of the documented inhibitors, Erythrosin B and SK-12. The consensus scoring yielded 221 hits from the in-house database and 1,354 from the ZINC database.

Protein–ligand interaction profiling

In the succeeding step, the flexibility of the receptor was accounted for by employing the induced-fit docking protocol, where the shortlisted molecules from the first phase of docking were considered. The obtained results indicated binding scores within the range of − 6.450 and − 9.180 kcal/mol. Next, the binding mode analysis at the molecular level was investigated via protein–ligand interaction fingerprints (PLIF) and protein–ligand interaction profiling (PLIP). The compounds exhibiting the interactions with the hot-spot interfacial residues, i.e., Tyr23, Lys26, Phe46, and Leu58 were selected for further study. Consequently, 11 and 222 molecules from the in-house and ZINC database, respectively, were considered for further investigations; the compounds had predicted binding scores between − 6.653 to − 7.805 and − 7.230 to − 9.180 kcal/mol, respectively. The obtained hits from the ZINC library were further screened by applying the cut-off score of − 8.637 along with maximizing the number of interactions, leaving 19 compounds. These 30 compounds (11 from the in-house library, 19 from the ZINC database) each exhibited a good interaction profile and were capable of establishing multiple interactions with the residues lining the pocket. Five compounds (AAM282, AAO342, in-house library; ZINC000109904737, ZINC000108913259, and ZINC000031797477, ZINC library) were selected for detailed mechanistic and structural dynamics studies and binding free energy calculations based on docking scores and interaction with the maximum number of binding site residues. Furthermore, for comparison we also performed MD simulations for SK-12. The structural details of the shortlisted hits and SK-12 are reported in Table 2.

Table 2.

2-D structures of the shortlisted hits along with SK-12, a reported active compound

Since the majority of the interfacial site residues are of hydrophobic nature, the binding was mainly driven by a number of hydrophobic and π–π stacking contacts between these shortlisted compounds and NS3pro, rather than electrostatic interactions. The intermolecular interaction pattern of the reference compound used in the present study exhibited both hydrophobic and electrostatic contacts with key residues of the binding site (Fig. 5A). A total of four hydrogen bonds were observed. The terminal carboxylic acid of SK-12 accepted the hydrogen from the amide nitrogen of Arg24 at a distance of 2.68 Å. Another H-bond was between the naphthyl hydroxyl and the histidine imidazole nitrogen with a distance of 3.30 Å. Additional electrostatic interactions were observed for the central sulfonamide N–H with Lys26 and sulfone with Met59 at distances of 2.07 and 1.91 Å, respectively.

Fig. 5.

Intermolecular interaction diagrams of the shortlisted small molecules in complex with NS3pro, generated after molecular docking: A SK-12, B AAM282, C AAO342, D ZINC000109904737, E ZINC000108913259, and F ZINC000031797477

AAM282 (Fig. 5B) possesses an aromatic ring and exhibited a number of hydrophobic contacts with crucial residues at the NS3pro interface, including Tyr23, Arg24, Ile25, Lys26, Val57, Leu58, and Met59. The only electrostatic interaction observed was between the backbone amide of Lys26 and the thiazole nitrogen of AAM282 with an H-bond distance of 2.80 Å.

In comparison, multiple hydrogen bond interactions were displayed by AAO342 along with hydrophobic contacts with crucial residues (Fig. 5C). In total, four hydrogen bond interactions were observed between the ligand and the binding pocket residues. The pyrimidine ring of the ligand participated in H-bonds with Arg24’s N–H and with His60’s imidazole at a distance of 2.01 and 3.25 Å, respectively. The ligand’s amide and sulfonamide also made electrostatic contacts, with Arg24’s backbone carbonyl at a distance of 2.35 Å and with Met59’s backbone amine located at a distance of 3.04 Å. Also, the residues Tyr23, Arg24, Ile25, Val57, and Leu58 were involved in hydrophobic contacts.

ZINC000109904737 and ZINC000108913259 exhibited the greatest number of interactions from among all the ligands we studied. Hydrophobic contacts were observed with almost all of the residues in the binding pocket including Tyr23, Arg24, Ile25, Gln27, Phe46, and Leu58. Similarly, multiple hydrogen bonds were also observed with crucial residues, including Lys26, with contact distances of 2.10/2.20 Å, and 1.97/2.06 Å, and Met59, with distances of 2.29/2.28 Å and 1.93/2.83 Å for ZINC000109904737/ZINC000108913259 (Fig. 5D/E), respectively. An additional H-bond with a distance of 2.75 Å was observed with Arg24 in the case of ZINC000108913259. The detailed intermolecular interaction pattern for ZINC000031797477 revealed the formation of three hydrogen bonds: two with Lys26 having distances of 1.97 Å and 2.58 Å, along with a single H-bond with Met59 having a distance of 2.26 Å. Van der Waals interactions were observed with Tyr23, Ile25, Gln27, Tyr33, and Met59 (Fig. 5F).

Dynamic studies of the protein–ligand complexes

To further investigate the stability dynamics of the complexes with respect to time, and using a more accurate energy representation, molecular dynamics (MD) simulations were performed.

Root mean square deviation

The root mean square deviation (RMSD) is undoubtedly the most popular parameter which allows establishing the quality and equilibration of biomolecular simulations. The RMSD was calculated to investigate the structural deviations induced by the shortlisted compounds within the targeted protein, NS3pro. It is evident from the results displayed in Fig. 6 that all of the systems attained an equilibrated state at least by the terminal phase of the simulations, which displayed an average deviation ranging from 1.9 to 2.1 Å. For the reference compound SK-12, after an initial phase of instability for ~ 60 ns, a well converged state with an average deviation of 2.3 ± 0.2 Å was achieved. The shortlisted hits from the in-house library attained an equilibrated state within the initial 10 ns of the simulation period and remained converged throughout the targeted time frame, showing the consistent binding mode. The average fluctuation recorded during the course of MD was 2.2 ± 0.2 Å. However, a plateau of 2.8 Å was observed between the sampling frame of 73–75 and 67–69 ns for the NS3pro–AAM282 and –AAO342 simulations, respectively. In comparison, the compounds from the ZINC database experienced larger fluctuations, with ZINC000031797477 bound to NS3pro being the most perturbed one. For the compound ZINC000109904737, an equilibrated state was not attained till after the initial 55 ns. The average RMSD attained by the complex during the converged state was 2.4 ± 0.2 Å. The complex containing ZINC000108913259 experienced a non-uniform trend in fluctuations after the attainment of the equilibrated state. The deviation peaks intensified between 47–50 and 87–89 ns with an extended and maximum variation of 2.5 Å and an average deviation of 1.9 ± 0.2 Å. NS3pro bound to ZINC000031797477, however, attained a converged state during the initial stages of the simulation with an average RMSD of 2.0 ± 0.3 Å. However, the gradual increase in RMSD intensified during the 73–82 ns time frame during which considerable fluctuations were observed. In summary, the RMSD analysis suggests that the MD trajectories for all the studied MD systems achieved a stable state during the sampling phase. Further, the shortlisted compounds exhibited good stability profiles compared to the apoenzyme (NS2B–NS3pro), for which significant fluctuations were recorded during the simulation and the stability was not achieved till after the initial 80 ns of the simulation run. The obtained stability pattern exhibited by the shortlisted compounds is comparable to that of the reference used.

Fig. 6.

Time courses of the RMSD for NS3pro bound to SK-12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477 (various colors), in comparison to NS3pro bound to the cofactor protein NS2B (black)

Radius of gyration

The structural properties of the simulated ensembles were further assessed via the time-dependent convergence of the radius of gyration (R_g), depicted in Fig. 7. The R_g, which is related to the compactness of the protein, is the root-mean-squared distance of the components within an object with respect to its center of mass. The R_g data showed that ligand-bound complexes attained improved folding and were structurally more compact and ordered compared to the NS2B–NS3pro complex. In general, no sharp increase in fluctuation was noted after 60 ns, except for ZINC000108913259 and ZINC000031797477.

Fig. 7.

The compactness (Rg) of native NS3pro in complex with SK-12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477 (various colors), in comparison to the cofactor protein NS2B (black)

The stable conformation achieved by the target protein bound to SK-12, AAM282, AAO342, and ZINC000109904737 can be attributed to the strength of the intermolecular hydrogen bonds that formed. The number of hydrogen bonds established by these complexes was not greater compared to the most perturbed systems, i.e., ZINC000031797477 and ZINC000108913259. When we investigated in detail, we found that the associated hydrogen bonds with these two complexes did not remain persistent during the course of simulation, which in turn resulted in a non-uniform trend of R_g. The average R_g for NS3pro in complex with NS2B, SK12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477 was found to be 17.1 ± 0.2, 15.7 ± 0.1, 15.6 ± 0.1, 15.6 ± 0.1, 15.6 ± 0.1, 15.7 ± 0.1, and 15.6 ± 0.1 Å, respectively.

Dynamics of intermolecular interaction patterns

In order to investigate the potential stability of various intermolecular contacts, the interaction patterns were also analyzed for each system during the terminal stages of the simulation run. The reference compound, SK-12 (Fig. 8A) revealed the formation of new hydrogen bonds with Tyr 23 at the distance of 3.22 Å while the H-bonds with Arg24, Lys26, and His60 remained conserved during the simulation time period with H-bond distances of 3.31, 1.65 and 3.55 Å, respectively. Of the observed H-bonds, the contact with Lys26 exhibited the highest occupancy. However, the hydrogen bond with Met59 was lost during the simulation. Hydrophobic contacts were observed with Tyr23, Phe46, Leu58, and His60, with the maximum occupancy observed for Phe46 and Leu58.

Fig. 8.

Intermolecular interaction diagram of the shortlisted small molecules in complex with NS3pro generated from molecular dynamics: A SK-12, B AAM282, C AAO342, D ZINC000109904737, E ZINC000108913259, and F ZINC000031797477. (*Stick model in grey is the starting pose of the simulation while the tan color represents the poses from terminal stages of the simulation. The differently colored poses for each compound in right side of image represents the binding poses during terminal stages of MD where maximum interactions were observed while the detailed intermolecular interaction pattern is shown in respective left panels)

The detailed inter-molecular interaction pattern for AAM282 (Fig. 8B) revealed hydrophobic interactions with Tyr23, Tyr33, Phe46, and His60. Though hydrophobic interactions were also observed with Lys26 and Leu58, those interactions had lower occupancy during the terminal stages of the simulation. During the MD simulation, consistent hydrogen bonds were established between ligand and the crucial interfacial residue, Lys26, with H-bond distances of 2.60, 3.18, and 3.20 Å.

AAO342 retained hydrophobic interactions with crucial residues including Tyr23 and Leu58 with the formation of new hydrophobic contacts with Tyr41, Phe46, and His60. The highest occupancy was revealed for Tyr23 followed by Tyr41. The ligand was stabilized in the binding pocket via multiple hydrogen bonds. Two hydrogen bonds were observed with Arg24 at a distance of 2.75 and 1.91 Å. However, the hydrogen bond interactions observed with Met59 and His60 before molecular dynamics simulation were not observed at the end of the simulation run. Two additional H-bonds were established between the carbonyl oxygen of the tested ligand and nitrogen atom from Lys61 and Gly62 at the distance of 2.18 and 3.17 Å respectively (Fig. 8C).

Among all the tested compounds, ZINC000109904737 revealed the most exciting results with the highest number of interactions that remained persistent throughout the simulation. The MD interaction analysis revealed multiple hydrophobic interactions with binding site residues including the residues identified in the in silico mutagenesis i.e. Tyr23, Lys26, Phe46, Leu58. When observed in detail the hydrophobic contacts with these residues remain conserved during the targeted time frame. These results are in correlation with the observed stability patterns for NS3pro in complex with NS2B. In total four hydrogen bonds were observed i.e., with Lys26 at a distance of 2.16, 2.20 Å, and Met59 with the bond distance of 2.30 and 2.92 Å. Remarkably, these electrostatic interactions were found to be preserved during the simulation run. It is important to note here that a similar trend of electrostatic interactions was observed in docking studies thus highlighting the remarkable potential of ZINC000109904737 to maintain crucial interactions with time (Fig. 8D).

For ZINC000108913259 some of the hydrophobic contacts were lost during the simulation run however, the interactions with the identified crucial residues remained stable throughout the MD. Additionally, new hydrophobic contacts were also established with Lys61 and Ile65. During MD simulation the H-bond interactions with Arg24 and Met59 did not persist. However, two new and stable H-bonds were established between imidazole nitrogen of His60 and amide hydrogen of ligand atoms with a distance of 2.34 and 2.58 Å. This compound after simulation retained the H-bond contacts with the crucial residue Lys26 with bond distance of 2.15 and 2.50 Å (Fig. 8E).

After simulation, the intermolecular interaction analysis for ZINC000031797477 revealed the formation of six hydrogen bonds. The hydrogen atom from the terminal amine group of the ligand formed electrostatic interaction with the oxygen atom of the hydroxyl group in Tyr23 at the distance of 2.84 Å. Further, the amine hydrogen from Met59 and the oxygen atom from the terminal carboxyl group acted as H-bond donor and acceptor respectively in the establishment of H-bond contacts with the distance of 2.63 and 3.00 Å. Two hydrogens at a distance of 2.09 and 2.16 Å were donated by amide nitrogen to the carboxyl oxygen of Thr53. Lys26 was also involved in the electrostatic interaction with terminal carboxyl of ZINC000031797477 distant at 2.83 Å. Of these, the H-bonds with Thr53, followed by Met59 and Lys26 revealed the highest occupancy while for others non-uniform pattern was observed. Additionally hydrophobic interactions were observed with Thr53 and Tyr79 (Fig. 8F).

It is evident from the observed results that most of the interactions with residues, those identified in alanine scanning, remained preserved in a dynamic state thus indicating the stable and persistent binding mode. In particular the highest occupancy was recorded for Lys26 with percent occupancy of 48.23, 26.79, 23.88, 34.05, 7.55, and 53.10 for SK-12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477. Secondary to Lys26 was Met59 where the hydrogen bond occupancy was 42.52, 0.61, 8.41, 8.15, 6.72, and 24.49% respectively. The binding site residue Arg24 displayed the highest occupancy of 36.00% for AAO342 Further details are presented in Table S5. Further to compare the binding modes, the initial pose at 0 ns is compared to the two of the poses from the terminal stages of the simulation. The superimposed poses are also depicted in the left panel of Fig. 8 for each tested compound.

Root mean square fluctuation

To better understand the structural dynamics of the conjugated systems, the RMSF for the backbone atoms was calculated that describes the fluctuation at the residue level. As depicted in Fig. 9, the obtained results for SK-12 revealed an average fluctuation of 1.2 ± 0.6 Å, while when observed in detail, the associated fluctuations were more intense compared to the native enzyme complex (average RMSF: 2.0 ± 0.5 Å). For compounds retrieved from the in-house library when taken together, RMSF analysis revealed a fluctuating pattern in the binding region with a minimum RMSF of 0.5 nm to a maximum fluctuation of 3.5 Å. In correlation with RMSD, AAM282 and AAO342 stabilized the fluctuation experienced by the residues in the apo state with the average RMSF of 0.9 ± 0.5 and 1.0 ± 0.4 Å respectively. In specific, the fluctuation of binding site residues Tyr23, Arg24, Ile25, Phe46, Leu58, Met59, was greatly reduced compared to NS3pro bound to NS2B. While the stability provided to Lys26 was comparable to NS2B. However, some residues from the binding pocket i.e., Gln27 experienced an extended deviation of 2.2 and 3.1 Å for AAM282 and AAO342 respectively.

Fig. 9.

Fluctuations (RMSF) in NS3pro residues induced by interactions with SK-12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477 (various colors) in comparison to the cofactor protein NS2B (black)

In comparison, the compounds from the ZINC library displayed higher fluctuations in the binding pocket’s residues though stabilizing the overall structure of the target protein. However, most of the regions experienced lesser fluctuations in comparison to the apo state. Similar to the RMSD trend, ZINC000031797477 displayed the highest fluctuations when compared to the remaining tested compounds with a maximum peak of 0.35 nm. The recorded residue fluctuations at average were 0.9 ± 0.4, 1.0 ± 0.5, and 1.2 ± 0.6 Å for NS3pro bound to ZINC000109904737, ZINC000108913259, and ZINC000031797477 respectively.

Binding free energy calculation

Since all molecular processes, for instance chemical reaction, molecular association, protein folding, etc., are derived by the free energy, correct estimation of the free energy is of fundamental importance in biomolecular investigations. In the present study, the free energies for the binding (ΔGBind) of the shortlisted ligands to the targeted protein to form the complex, were determined via the MM/PBSA approach. The value of ΔGBind estimated via this approach is further decomposed into individual components of electrostatic van der Waals energies, nonpolar and solvation free energies. The obtained results, in general, suggested that the non-polar interactions are the prime contributors to the binding free energy.

The reference compound, SK-12 exhibited remarkable results with the lowest binding free energy of − 463.35 ± 39.16. On the other hand, the tested ligands from the ZINC database displayed the highest binding free energy in comparison to the compounds shortlisted from the in-house database. The second lowest ΔGbinding of − 165.33 ± 13.09 kcal/mol was observed for ZINC000109904737 followed by ZINC000108913259 (− 93.94 ± 26.48 kcal/mol) and ZINC000031797477 (− 93.71 ± 16.14 kcal/mol). These results are attributed to the consistent and stable electrostatic and hydrophobic contacts with the crucial residues of NS3pro protein. Compounds AAM282 and AAO342 from the in-house database displayed binding free energies of around − 72.08 ± 17.36 kcal/mol and − 86.10 ± 24.53 kcal/mol. These results are also in agreement with comparatively a smaller number of contacts observed throughout the simulation.

The total binding free energy was further decomposed into various energy components to comprehend the main divers involved in binding. A summary of binding components in the binding free energy of selected inhibitors with NS3pro is depicted in Table 3. In general, For all five complexes, the negative binding free energy was obtained from the electrostatic (ΔE_elec) and the van der Waals (ΔE_vdW) interactions thus favoring the complexation. While the complexation was dis-favored by the polar solvation free energy (ΔG_pol) as indicated by the positive values. It is important to mention here that the protein–protein interaction interfaces mainly involve hydrophobic surfaces hence there is considerable influence of van der Waals forces. Although for the reference compound the overall binding energy was much greater in comparison to the compounds shortlisted via our study. But as evident from the results here the major part was contributed by the electrostatic interactions. In contrast for the tested compounds, the van der Waals interactions were mainly the major driver in the binding however, the electrostatic contributions were also involved but to a lesser extent. The solvent-accessible surface area energy (SASA) was also calculated where lower values are indicative of contracted centers and higher thermodynamic stability. All of the systems exhibited stability while ZINC000109904737 exhibited good results with SASA energy of − 23.27 ± 1.00 kcal/mol. Moreover, the SASA (Fig. S1) calculated for the targeted system revealed an average value of 114.35 ± 2.18 nm² for ligand free state of NS3 pro. While NS3 bound to SK-12, AAM282, AAO342, ZINC000109904737, ZINC000108913259, and ZINC000031797477 featured 93.33 ± 2.15, 92.30 ± 1.72, 91.73 ± 2.07, 93.35 ± 1.73, 93.39 ± 1.96, and 92.98 ± 2.03 nm² respectively.

Table 3.

Results of MMPBSA calculations of the shortlisted compounds

Energy components (kcal/mol)	SK-12	AAM282	AAO342	ZINC000109904737	ZINC000108913259	ZINC000031797477
Van der Waals energy	− 124.01 ± 17.80	− 131.89 ± 16.07	− 162.23 ± 10.72	− 212.13 ± 11.67	− 153.24 ± 22.01	− 194.67 ± 14.21
Electrostatic energy	− 1004.17 ± 97.23	− 32.97 ± 11.54	− 46.86 ± 9.30	− 23.37 ± 5.34	− 36.53 ± 22.77	− 75.81 ± 20.26
Polar solvation energy	− 680.46 ± 94.52	107.30 ± 20.86	139.19 ± 22.78	93.44 ± 9.22	114.62 ± 31.28	197.80 ± 18.12
SASA energy	− 15.63 ± 1.04	− 14.47 ± 1.24	− 16.21 ± 1.06	− 23.27 ± 1.00	− 18.79 ± 1.81	− 21.07 ± 1.15
Binding energy	− 463.35 ± 39.16	− 72.08 ± 17.36	− 86.10 ± 24.53	− 165.33 ± 13.09	− 93.94 ± 26.48	− 93.71 ± 16.14

Principal component analysis and free energy landscape analysis

Principal component analysis (PCA) is a highly utilized method for decreasing data dimensionality or in MD simulation used to obtain the internal motion of the system after removing its translation and overall rotation from the trajectory (Amadei et al. 1993; Sittel et al. 2014; Yuan et al. 2022). In PCA, its general objective is to identify a group of orthogonal vectors, known as principal components, that effectively capture the majority of the variance existing in the initial data (Jolliffe 2002). The first principal component (PC1) represents the most substantial amount of variation in the data, while each following component represents a progressively smaller amount of variation. In the realm of trajectory analysis, the molecular structure’s fluctuation can be expressed as a linear combination of these principal components, which comprise the fundamental dynamics of the system. Herein a hybrid plot of FEL and PCA is constructed by projecting MD trajectories of the NS2B (Fig. 10) and ligand bound complexes (Fig. 11) of NS3pro. With the help of this plot the relation between energetic coupling and conformational states can easily be visualized.

Fig. 10.

Principal component analysis and free energy landscape for NS3pro bound to NS2B, explaining the conformational space and associated energies

Fig. 11.

Collective motion and potential energy surfaces of inhibitor bound NS3pro. A SK-12, B AAM282, C AAO342, D ZINC000109904737, E ZINC000108913259, and F ZINC000031797477 (* X and Y axis represent PC1 and PC2 respectively)

For the system where NS3pro is bound to NS2B (Apo state), four conformational transitions as observed with four global energy minima along PC1 axis which is indicative of larger degree of flexibility. However, in ligand bound complexes of NS3pro the structural variation was comparatively lower, particularly as observed for AAM282, ZINC000109904737, and ZINC000108913259. In reference simulation (SK-12), two conformational states feature relatively less variation than apo state. The trend of variance over time is depicted in Figure S2. In conclusion, the variations were quite low in the case of the predicted inhibitors, indicating the potential of mentioned compounds in rendering the protein stable.

Conclusion

In the present piece of work, we outlined the potential drivers responsible for the formation of enzyme complex using in silico mutagenesis and demonstrated the role of these mutations in shaping the energy landscape. The four residues identified herein, based on stability and affinity changes are Tyr23, Lys26, Phe46, and Ler58. Further, an extensive structure-based virtual screening (SBVS) was performed by utilizing the consensus docking strategy, to identify new inhibitors by targeting the identified hot-spot residues. The screening resulted in five compounds (ZINC000109904737, ZINC000108913259, ZINC000031797477, AAM282 and AAO342) with best scores and remarkable potential to establish interaction with target site residues and were subjected to detailed mechanistic studies. Their stability dynamics in combination with good binding affinities particularly for ZINC000109904737 (− 165.33 ± 13.09 kcal/mol) make them potential lead compounds that could pave the way for the development of more potent derivatives targeting the protein–protein interactions in the future. Further, targeting the protease serves as a double-pronged attack on the DENV by restoring the host’s immunity and preventing virus maturation. The unavailability of the active data set against the target protein and the utilization of rigid docking in the initial phase of molecular docking to save computational time are the limitations of the study. Additionally, while performing consensus scoring approach the different placement methods available could also be considered.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, data curation, investigation, methodology, writing—original draft: M. M., S. N., and S. A. Funding acquisition Z. U and N. A Writing, and editing: M. M., Z. U., and R. D. All authors reviewed the manuscript.

Data availability

ChemDraw was used to build 2D structures and can be purchased at https://perkinelmerinformatics.com. Molecular Docking was performed using MOE version 2018.0, a commercial software which can be purchased from https://www.chemcomp.com. AutoDock vina is freely available and can be installed from https://vina.scripps.edu/downloads/. GROMACS, an open-source software (https://manual.gromacs.org/2020/download.html), used for all MD simulation. VMD and Chimera are available to noncommercial users under a distribution-specific license. Additional data are available from the authors upon request.

Declarations

Conflict of interest

The authors have no conflicts of interest to declare.