Molecular docking and molecular dynamics simulation identify a novel Radicicol derivative that predicts exclusive binding to Plasmodium falciparum Topoisomerase VIB

Abstract Plasmodium falciparum harbors a unique type II topoisomerase, Topoisomerase VIB (PfTopoVIB), expressed specifically at the actively replicating stage of the parasite. An earlier study showed that Radicicol inhibits the decatenation activity of PfTopoVIB and thereby arrests the parasites at the schizont stage. Radicicol targets a unique ATP-binding fold called the Bergerat fold, which is also present in the N-terminal domain of the heat shock protein 90 (PfHsp90). Hence, Radicicol may manifest off-target activity within the parasite. We speculate that the affinity of Radicicol towards PfTopoVIB could be enhanced by modifying its structure so that it shows preferential binding towards PfTopoVIB but not to PfHsp90. Here, we have performed the docking and affinity studies of 97 derivatives (structural analogs) of Radicicol and have identified 3 analogs that show selective binding only to PfTopoVIB and no binding with PfHsp90 at all. Molecular dynamics simulation study was performed for 50 ns in triplicate with those 3 analogs and we find that one of them shows a stable association with Radicicol. This study identifies the structural molecule which could be instrumental in blocking the function of PfTopoVIB and hence can serve as an important inhibitor for malaria pathogenesis. Communicated by Ramaswamy H. Sarma


Introduction
The apicomplexan parasite Plasmodium falciparum causes the most severe form of human malaria. According to the latest WHO report (https://www.who.int/news-room/fact-sheets/ detail/malaria), there are 228 million cases of malaria in 2018, of which 405,000 deaths are reported. Although malaria is an old-world disease only a few anti-malaria drugs are available which can kill the parasite (Tse et al., 2019). Further, parasites have developed resistance against the anti-malarial drugs Chloroquine and Artemisinin which subsequently have increased the mortality rate of malaria throughout the world. Hence, it is of utmost importance to develop new anti-malarial drugs. The approaches may include identification of new targets, rational drug design, repurposing of drugs and identification of natural products having anti-malarial activity (Newman & Cragg, 2016;Vale et al., 2020).
Topoisomerases are attractive targets as it constitutes one of the most important enzyme-family that maintains the super helicity of DNA and thereby is essential for DNA replication, transcription and DNA repair. Earlier studies have identified that Ciprofloxacin targets the Type II topoisomerase of bacteria namely DNA Gyrase and topoisomerase IV, thereby is used to treat various kinds of bacterial infections (Mitscher, 2005;Widdowson & Hennessy, 2010). A recent study has established that GSK299423 exhibits 57-fold selectivity towards the P. falciparum type II topoisomerase than the human counterpart (Mudeppa et al., 2015). Using a yeast-based assay system, we have shown that PfTopoVIA-VIB can decatenate the catenated DNA. Moreover Radicicol, an antifungal macrolactone antibiotic causes inhibition of the decatenation activity of PfTopoVIB in a dose-dependent manner (Chalapareddy, 2016).
Topoisomerase VI is a type IIB topoisomerase which was first identified in an archaeal species Sulpholobus shibatae (SsTopoVI). It has two subunits: Topoiosmerase VIA and VIB which forms a heterotetramer (A 2 B 2 ) in its active state. Structural studies revealed the striking similarities in the ATP binding motifs of TopoVIB and the members of the GHKL (Gyrase-Hsp90-Histidine Kinase-MutL) super-family. They all share a small three-dimensional ATP-binding fold known as the Bergerat fold. The core ATP binding domain (amino acids 22 to 166) of PfTopoVIB is homologous to that present in the GHKL ATPases and is composed of three small motifs (B1, B2, and B3) that constitute the Bergerat fold (Chalapareddy, 2016). It has been demonstrated that Radicicol competes with ATP for binding to the SsTopoVIB monomers, thereby inhibits the dimerization of SsTopoVIB ATPase domain and ATP hydrolysis, which subsequently hampers the strand passage reaction of the enzyme (Corbett & Berger, 2006). Plasmodium falciparum expresses both subunits of topoisomerase VI in the replicative stage of the parasite. In Plasmodium infected erythrocyte, Radicicol causes the growth inhibition of parasites with an IC 50 of 8.5 lM (Chalapareddy et al., 2014). It was observed that Radicicol induces over-expression of PfTopoVIB and results in the inhibition of schizont stage to ring stage transition in Plasmodium. This underscores the significance of PfTopoVI in Plasmodium biology. As most of the replication proteins are essential for the survivability, and generation of conditional knock-outs in Plasmodium is technically challenging, it is important to identify highly specific chemical inhibitors of PfTopoVIB in order to delineate the precise biological role of this protein in Plasmodium. However, owing to the similarity within the Bergerat ATP-binding fold which is present in both PfTopoVIB and PfHsp90, Radicicol is found to dock at both the structures (Chalapareddy et al., 2014). Our present study aims at identifying the analogs of Radicicol that could reduce its off-target activity. We plan to identify some specific analogs of Radicicol that would bind specifically to the PfTopoVIB but not to PfHsp90. Using the bioinformatics approaches, we have designed 97 analogs of Radicicol and they are allowed to dock on PfTopoVIB as well as on PfHsp90. Analysis of the results revealed three such analogs that have a very strong affinity for PfTopoVIB, and they do not exhibit any binding with PfHsp90. The validation of the docked complexes is done using molecular dynamics simulations which identifies one analog that shows stable ligandprotein interactions. This derivative of Radicicol can serve as a tool to decipher the biological function of Plasmodium topoisomerase VIB.

Comparative structure modeling and validation of PfTopoVIB and PfHsp90
The crystal or solution structure for PfTopoVIB is not solved yet. In case of PfHsp90, although two crystal structures are solved, they are not of the full-length protein. Hence, homology-based structure modeling was carried out for both of the proteins. The sequences of PfTopoVIB and PfHsp90 were retrieved from the UniProtKB/Swiss-Prot database (Bairoch & Apweiler, 1997) and suitable templates were selected from the SMTL, a large structural database of experimentally determined protein structures, derived from the PDB (Berman et al., 2000). Target-template sequence alignment was carried out using BLAST P suite (https://blast.ncbi.nlm. nih.gov/Blast.cgi?PAGE=Proteins). Near native structures of PfTopoVIB and PfHsp90 were modeled using the automated mode (ProMod3 Version 1.2) integrated in the Swiss-Model server (Guex et al., 2009). Necessary energy minimization steps were also carried out for the predicted models using the GROMOS 43B1 force field (Van Gunsteren, 1996) implemented in the Swiss-PDB viewer (http://www.expasy.org/ spdbv). The validation of both the modelled structures were carried out using the PROCHECK server (Laskowski et al., 1993) and the Ramachandran plot (http://services.mbi.ucla. edu/SAVES/Ramachandran/). PROCHECK server checks the stereochemical quality of the modelled structures and the Ramachandran plot shows the residue-by-residue quality and stability of the favored and the disallowed regions in the protein model. ERRAT scores (Colovos & Yeates, 1993) were also predicted for the homology models through the Structural Analysis and the Verification Server (SAVES) (https://servicesn.mbi.ucla.edu/SAVES) of UCLA-DOE Lab.

Design and preparation of Radicicol analogs
The 3D structure of Radicicol, downloaded from the PubChem compound database (Kim et al., 2019) was used as the reference molecular scaffold for the in-silico design of different derivatives (structural analogs). MarvinSketch tool (https://chemaxon.com/products/marvin), version 16.8.8.0 was used to draw the analogs by modifying different atoms, functional groups and/or side chains of Radicicol. Our strategy was to modify the structure of Radicicol and design novel analogs by substituting the functional groups which potentially enhance its inhibition against PfTopoVIB and reduce its inhibition of PfHsp90. The reported Structure-Activity Relationship (SAR) of Radicicol (Turbyville et al., 2006) was considered for selecting the attachment point for the modification and various functional groups were identified from relevant literature (Dutton et al., 2014;Pearl et al., 2008;Shinonaga et al., 2009;Shiotsu et al., 2000;Teo et al., 2015;Wang et al., 2008). We substituted various functional groups at specific attachment points of Radicicol, without disturbing the macrocyclic ring which is required for the bioactivity of Radicicol. The designed analogs were saved along with their 10 conformers each for further preparation. 'Prepare ligand' protocol in Biovia Discovery Studio (DS) 4.0 (BIOVIA, 2019) was used to prepare the analog dataset, which optimizes the charges of common groups, adds hydrogen atoms, generates tautomers/isomers and removes duplicate/bad conformers. 'Generate Conformations' protocol, using a quasi-exhaustive systematic search method (FAST conformation generation method) (Smellie et al., 1995) employed in DS 4.0 was used to create all the possible diverse 3D conformers of each analog and all the resulting diverse conformers were stored as a single file for further docking step.

Structure based virtual screening of Radicicol analogs against PfTopoVIB and PfHsp90
Along with Radicicol and its analogs, Adenosine triphosphate (ATP) molecule (downloaded from PubChem compound database) was also used as the positive control for docking studies. Since ATP binds to the monomeric form of TopoVIB we have used the monomers of PfTopoVIB and PfHsp90 as our template for our docking studies. Two sets of docking were carried out in a site-specific manner, where the 3D structures of PfTopoVIB and PfHsp90 served as the receptors and the Radicicol derivatives with their conformers served as the analog dataset. The docking program LibDock (Diller & Merz, 2001;Rao et al., 2007) implemented in DS 4.0 performs a high-throughput docking by aligning analog conformations to polar and apolar receptor interaction sites (hotspots). The binding site cavities were analyzed for both the target protein structures using the 'Eraser algorithm' (Venkatachalam et al., 2003) implemented in DS 4.0. A grid with coordinates of À9.153 Å, À79.659 Å and 25.38 Å for X, Y and Z respectively with spacing of 0.5 Å was used for positioning the binding site. In case of PfHsp90, the grid coordinates of 21.963, 32.991 and 14.764 for X, Y and Z respectively with spacing of 0.5 Å was used. The grid covered the Bergerat fold residues, which were specified as the target protein site features (termed as HotSpots) by the LibDock program for calculating the binding affinity of analogs with the receptors. Receptoranalog docking study was performed using LibDock protocol against PfTopoVIB as well as PfHsp90. An empirical scoring function, LigScore (Krammer et al., 2005) was employed to score the docked ligand poses, and the complexes were ranked and sorted according to the descending order of the LibDock score. The best docked pose for each analog towards each protein was identified based on the highest LibDock score and compared with the LibDock score of ATP and Radicicol. The lead molecules were selected based on higher LibDock score and specific docking with PfTopoVIB and little or no docking with PfHsp90. The interacting amino acids of the complexes were analyzed and the types of molecular interactions including conventional hydrogen bonds, carbon-hydrogen bonds, electrostatic interactions and hydrophobic interactions were noted. Molecular dynamics (MD) simulation of the best docked pose of lead ligand-protein complexes Molecular dynamics simulations were performed for the top three analogs using Gromacs software 5.1.4 (Lindahl et al., 2001;Van Der Spoel et al., 2005). The analog-protein complexes were subjected to 50 ns molecular dynamics simulation in triplicate to validate the stability of the complex and also to estimate the variation and conformational changes in the protein-analog interactions. For simulation, the standard CHARMM36 FF which includes parameters for the protein and other bio-molecules were employed (Lee et al., 2016). The parameter for the analogs was generated using CGenFF (Vanommeslaeghe et al., 2010). Protein-analog complexes were kept in a periodic rectangular box and solvated with TIP3P water molecules. In order to neutralize the charge 0.15 M KCl ions were added to the system. The system was then equilibrated with NPT ensembles and was subjected to 50,000 steps minimization using the steepest descent method for 1000 ps at 300 K. To maintain the temperature, Berendsen thermostat with tc ¼ 1.0 ps and Parrinello-Rahman barostat with tp ¼ 2.0 ps were used. The Van der Waals interactions were described using a Lennard Jones function with a cut off of 1.0 nm. Molecular dynamics simulations were performed in triplicate for each system using a GPU SERVER with IntelV R XeonV R Gold 6154 3.0 G,18 core 256 GB RAM with dual NVIDIA Tesla TM V100 GPU 32GB PCI-E.

Binding free energy calculation
We calculated the free energy of the PfTopoVIB bound to each of the three lead analogs. For that, we used the snapshots collected from trajectories, that resulted from the MD simulations. We used the Web server farPPI which employs the MM/PB (GB) SA approaches with the GAFF2 and ff14SB force field combination and the PB3 procedure (Wang et al., 2019). It uses a ready-to-dock benchmark database (2P2I database, ver. 2.0, 28-03-2018) and a ready-to-rescore benchmark dataset, which contains 900 binding poses for 184 protein-ligand complexes. The docked pose file and the receptor file are the input. The partial charge of ligand is assigned by AM1-BCC method via the antechamber module of Amber. The calculated binding free energy is based on the equation; Energy Binding ¼ Energy Complex -Energy Ligand -Energy Receptor.

Homology modeling of PfTopoVIB and PfHsp90
The sequence of Plasmodium falciparum TopoVIB retrieved from Uniprot (ID: Q8ID53) consists of 561 amino acids. The template search revealed that the most similar protein structure available for the sequence of PfTopoVIB was that of Sulpholobus shibatae Topoisomerase VIB (PDB: 2ZBK.D chain). Taking 2ZBK.D chain as the template, the target-template alignment was carried out using BLASTP suite, which predicted the local pairwise sequence identity to be 28.57% and the E-value as 1e-12. For homology modeling, the sequence identity between (20-25)% is considered as the twilight zone (Chung & Subbiah, 1996). Studies have demonstrated that proteins with pairwise sequence identity higher than 25% are similar in 3D structures and have a strong divergent evolutionary relationship (Chang et al., 2008;Doolittle, 1986;Laurents et al., 1994;Rost, 1999;Sander & Schneider, 1991;Subbiah et al., 1993;Yang & Honig, 2000). The E value of 1e-12, obtained for the template from the BLAST tool was considered in the acceptable range as per the recent study (Barghash & Helms, 2013), which suggests the acceptable threshold of E Value for BLAST as 1e-8. Hence, the 3D structure of 2ZBK.D chain with 28.57% sequence identity was considered as a suitable template for homology modeling of PfTopoVIB. 2ZBK is an X-ray diffraction structure with Radicicol as a native ligand having 3.6 Å resolution. Our study focuses on the ATP-binding domain of PfTopoVIB, known as the Bergerat fold, which spans through 22 to 166 amino acids. This region corresponds to the 11 to 192 amino acids of SsTopoVIB (2ZBK.D chain) with an alignment score range of (50-80) with 42% positives, 29% identity and 13% gaps. Figure 1(A) shows the sequence alignment between the target PfTopoVIB (Query) and template SsTopoVIB (Subject) in the Bergerat fold. As the aligned region covers the Bergerat fold, we preferred homology modeling over de novo protein structure prediction. It is observed that there is a unique highly charged region (containing a stretch of lysines and glutamic acids) spanning 61st-78th residues in the aminoterminal domain of PfTopoVIB. Plasmodium falciparum possesses this kind of low complexity charged residues and unstructured region which is a unique feature of parasite protein. However, the biological significance of such low complexity regions remains unknown. Homology based nearnative model was created for PfTopoVIB (Figure 1(B)) and subjected to 10 steps of energy minimization, using GROMOS 43b1 force field where in each step 20 cycles of steepest descent method was involved. 200 cycles of steepest descent were required to attain the lowest energy model, which was used for further analysis.
The sequence of PfHsp90 was retrieved from the Uniprot database (ID: Q8IL32). For building the near native model of PfHsp90, 3IED (PDB code) was selected as the template. The target-template alignment using blastp suite, predicted the local pairwise sequence identity to be 98% and the E-value as 5e-171 (Figure 1(C)). The ligand adenylyl phosphoramidate (AMPPN) was removed from 3IED template for building the near native model of PfHsp90 (Figure 1(D)). The model was subjected to 10 steps of energy minimization to obtain the lowest energy model for further analysis.

Modeled structure validation
The PROCHECK results (Supplementary Material) and the results of the Ramachandran plot ( Figure S1(A)) validate the stereo chemical quality and stability of the predicted PfTopoVIB model. Ramachandran plot for PfTopoVIB revealed that 80.4% of the residues (337 amino acids) were in the most favored regions, 15.5% (65 amino acids) in the additional allowed region, 2.9% (12 amino acids) in the generously allowed region and 1.2% (5 amino acids) in the disallowed region. The resultant model showed ERRAT score as 81.728 and the final energy as À21297.992 kJ/mol. The PROCHECK results (Supplementary Material) and the Ramachandran plot of PfHsp90 model ( Figure S1(B)) revealed that 91.9% of the residues (181 amino acids) were in the most favored regions, 8.1% (16 amino acids) in the additional allowed region, 0.0% (0 amino acids) in the generously allowed region and 0.0% (0 amino acids) in the disallowed region. The model showed ERRAT score as 81.878 and the final energy was À11965.740 kJ/mol.
In silico binding of Radicicol and ATP to the Bergerat fold of PfTopoVIB and PfHsp90 models The crystal structure 2ZBK.D has Sulpholobus shibatae Topoisomerase VIB (SsTopoVIB) bound to Radicicol (RDC531). We have presented this structure in Figure S2(A). To validate our docking, we have removed Radicicol from this structure and redocked Radicicol to the apo-structure represented in Figure S2(B). We compared the native pose ( Figure S2(A)) and the newly created docked pose ( Figure S2(B)) by superimposing, the RMSD value between the native pose and the docked pose was 1.86 Å ( Figure S2(C)). Our analysis showed that the amino acid Thr170 formed a hydrogen bond, while Ala46, Ile79, Val112 and Phe90 were actively involved in hydrophobic interactions with Radicicol in the 2ZBK-D structure ( Figure S2(D)). Our re-docked structure shows the conservation of the above interactions, in addition, it shows four more contacts including one hydrogen bond with Lys113 and hydrophobic interactions with Gly80, Ala89 and Val112 ( Figure S2(E)). After validating our model, next, we carried out two sets of docking studies using LibDock, one in which both ATP and Radicicol were docked to PfTopoVIB (Figure S2(F,G), respectively) and in the second set, both the ligands were docked with PfHsp90 ( Figure S2(H,I)). Our study shows that Radicicol binds to the Bergerat fold of PfTopoVIB and PfHsp90 like that of the intrinsic ligand ATP. This also ensured that the proteins were correctly modeled in silico and had near-native structural topology. The LibDock score for ATP and Radicicol docking were 143.657 and 76.4008 respectively (Table SI). While 96 different docking poses were obtained for ATP, only 2 different binding poses were obtained for Radicicol. The type of molecular interactions found between PfTopoVIB and ATP/Radicicol and their interatomic distance is tabulated (Table SII). ATP showed 14 molecular interactions in the form of hydrogen bonds and hydrophobic interactions. The key amino acids which are involved in conventional hydrogen bonds were Phe122, Asp43 and Thr22 whereas the residues Lys35, Gly120 and Glu36 showed carbon hydrogen bonds. Several hydrophobic interactions were also noted which included a Pi-Sigma bond with Lys35, a Pi-Sulfur bond with Met32 and four Pi-Alkyl bonds, two each with Lys121 and Lys35.
Radicicol showed 6 molecular interactions with PfTopoVIB, in which Leu126 formed a conventional hydrogen bond and a hydrophobic interaction through the alkyl group. Phe122 also formed a conventional hydrogen bond. While Asn40 formed a carbon hydrogen bond, Gly125 interacted with the Cl (halogen) of Radicicol and Lys104 made a hydrophobic interaction through the alkyl group. We found that many of the amino acids which interact with Radicicol were conserved between PfTopoVIB and SsTopoVIB. For example, the residues Phe105, Ala44 and Asn40 in PfTopoVIB correspond to that of Phe90, Ala46 and Asn42 in SsTopoVIB.
In the second set of docking studies, we allowed ATP and Radicicol to dock to the in-silico model of PfHsp90 N-terminal domain. Docking resulted in 81 different poses of the ligand ATP with the highest LibDock score of 129.333. 7 molecular interactions were found between ATP and PfHsp90 ( Figure  S2(H)). Residues Asn133, Gln230, Phe257 and Val255 were found to form conventional hydrogen bonds with ATP. Asn133 was also found to possess a Pi-Donor hydrogen bond, whereas Ala134 and Ala137 interacted through the alkyl group (hydrophobic interaction). Radicicol interacted with PfHsp90 in the Bergerat fold with a LibDock score of 91.791 generating 9 different docked poses. The key residues found to interact well with Radicicol were Asn224 through a conventional hydrogen bond, Asp211 through the halogen Cl and the amino acids Ile310, Ala134, Ala137, Met216, Leu225 and Leu130 all through hydrophobic interactions mostly of the alkyl type ( Figure S2(I)). Radicicol when docked with the template PDB structure 3IED, the key residues Ala137, Met207 and Asn215 were identified to interact which corresponds to the residues Ala137, Met216 and Asn224 in the PfHsp90 model. There is a difference of 9 residues in between Met207 and Met216 and also in between Asn215 and Asn224. This is because some amino acids were absent in 3IED, so the numbers given to amino acids in PDB is different when compared with our PfHsp90 model.

Designing of Radicicol analogs and docking to PfTopoVIB and PfHsp90
The 3D structure of Radicicol served as the reference molecule for the in-silico design of 97 different structural analogs using MarvinSketch tool. The analog structures were computationally drawn and designed by modifying different atoms, functional groups and/or side chains of Radicicol, based on its reported Structure-activity relationship (SAR) to alter its biological activity against PfTopoVIB and PfHsp90. We focused on altering or replacing those functional groups of Radicicol that cause higher affinity towards Hsp90. The IUPAC nomenclature, of all the 97 analogs are tabulated (Table SIII). 6498 different conformers were generated for 97 analogs using Biovia DS 4.0. Receptor-ligand docking performed by LibDock protocol against PfTopoVIB protein using 6498 analog conformers returned 3162 docked poses. In the same manner, docking against PfHsp90 protein using 6498 analog conformers returned 2512 docked poses. All the docked poses were analyzed and the best docked pose of each ligand towards each protein was identified based on the highest LibDock score and compared with the LibDock score of ATP and Radicicol (Table SI).
Selection of the Radicicol analogs that show specific docking to PfTopoVIB but not to PfHsp90 PfTopoVIB and PfHsp90 share a common structural ATP-binding domain known as the Bergerat fold (Bergerat et al., 1997) characterized by a structural motif consisting of an eightstranded mixed beta-sheet in two layersalpha/beta (Dutta & Inouye, 2000). Out of 97 analogs, 88 molecules interacted with PfTopoVIB within the same binding cavity occupied by Radicicol and ATP, whereas 9 analogs failed to dock. Depending upon the LibDock score of each analog compared to ATP and Radicicol, we classified them into two groups, one with a higher LibDock score than the cut off value 76.4008 and the other with lower scores than the cut off value. In case of PfTopoVIB, we have chosen this cut off value to select those analogs of Radicicol that have a higher chance of binding with PfTopoVIB as compared to the Radicicol. The LibDock score of Radicicol for PfTopoVIB is 76.4 hence all the values that are higher than 76.4 are shortlisted. To that end, 77 analogs were identified with LibDock scores higher than the cut off value and 11 analogs with lower LibDock scores than the cut off. The highest score was obtained for analog 91 (158.337) and the lowest score was obtained for analog 23 (76.108). So, based on docking score, 77 analogs of Radicicol performed better than Radicicol, 11 analogs were poor performers and 9 did not dock at all (Table SI).
The same classification criterion was adopted for analyzing the docking results of PfHsp90 also. In this case, the cut off value was taken as 91.7911. The LibDock score of Radicicol for PfHsp90 is 91.79; hence the values higher than that of 91.79 are shortlisted. Out of 97 analogs, 78 molecules were found to interact with a higher dock score than the cut off value. 13 analogs docked with a lower LibDock score and the remaining 6 analogs did not dock to PfHsp90 at all. The highest binding affinity was obtained for analog 80 (126.137) and the lowest for analog 78 (91.757). We focused our work only on those subsets of Radicicol analogs that display higher affinity towards PfTopoVIB and no affinity towards PfHsp90. Hence, although analog 91 shows the highest score for PfTopoVIB, it also shows higher docking score for PfHsp90. Similarly, analog 23 shows a poor score for PfTopoVIB compared to PfHsp90. The analysis of these ligands is beyond the scope of this manuscript. Our criterion for selecting the best analogs was to compare their binding affinity with both the proteins and finally choose those which docks specifically to PfTopoVIB but have little or no binding towards PfHsp90. We observed that out of the 6 analogs which did not dock to PfHsp90, 3 analogs did not dock to PfTopoVIB either, but the other 3 analogs were having higher LibDock scores towards PfTopoVIB. The structures of Radicicol and these three analogs are presented (Figure 2). These three analogs; analog 2, analog 6, and analog 7, with LibDock scores 133.823, 108.647 and 77.533 respectively were selected as the lead molecules. The number of molecules generated, docked and filtered at various step of this study has been summarized in Table 1.
Further, we compared the volume of Radicicol binding pockets between PfTopoVIB and PfHsp90 ( Figure S3) using a grid based 'Eraser' algorithm implemented in DS 4.0 and found that the binding pocket volume for PfTopoVIB is (180.625 Å 3 ), much larger than that of PfHsp90 (143 Å 3 ). Both the pockets were found to contain the residues of the alpha/beta structural motif, where the ATP and Radicicol interacted in our docking studies. We reason that as the size of analog 2 (637.162 Da), analog 6 (516.969 Da) and analog 7 (653.161 Da) are larger than that of ATP (507.181 Da) and Radicicol (364.777 Da), they are not able to fit properly to the binding site of PfHsp90 which is smaller than the binding pocket of PfTopoVIB.
The molecular interactions between the lead analogs and PfTopoVIB (Figure 3(A-C)) were analyzed and compared with that of parent molecule Radicicol. Analog 2 showed 12 interactions, analog 6 showed 8 interactions and analog 7 showed 14 (Table SII). The key amino acids in PfTopoVIB which established contacts with analog 2 were Glu36, Lys104, Phe105, Ly121, Phe122, Leu126 and Lys127. While a carbon hydrogen bond and a hydrophobic Pi-alkyl interaction were formed by Lys127, two hydrophobic interactions (Pi-Sigma and Pi-alkyl) were formed with Leu126. Glu36 showed two electrostatic (Pi-anion) interactions and Lys121 had a Pi-donor hydrogen bond and  one hydrophobic interaction of Pi-alkyl type. Lys104 showed a Pi-donor hydrogen bond and a hydrophobic Pi-alkyl interaction. Phe122 established an interaction through the halogen atom Cl of analog 2 and Phe105 formed a Pi-Pi T-shaped hydrophobic interaction. Analog 6 showed three conventional hydrogen bonds with the amino acids Asn40, Leu126 and Phe122. Lys127 formed a carbon hydrogen bond; Glu36 formed an electrostatic Pi-anion linkage; Leu126, Lys35 and Lys121 formed hydrophobic interactions mostly of pi-alkyl type. Analog 7, on the other hand, showed the highest number of hydrophobic interactions, but with a different set of residues present in PfTopoVIB namely Leu37, Asn40, Ala44, Cys90, Lys95, Leu126, Val183 and Ile185. Lys104 formed a Pi-donor hydrogen bond and a hydrophobic interaction through the alkyl group of analog 7. Asn48 also established a Pi-donor hydrogen bond and Ser41 showed a Pi-lone pair interaction with analog 7. When compared to ATP and Radicicol, the amino acids which retained their contacts with analog 2 were Phe122, Glu36, Lys121, Leu 126 and Lys104. The interactions with the residues Lys35, Glu36, Asn40, Lys121, Phe122, Leu126 and Ly127 were common in the case of analog 6. But analog 7 showed contacts with Leu126, Asn40 and Lys104 which were similar with those found in Radicicol. There were no common contacts found between ATP and analog 7. Phe122 was interacting with ATP, Radicicol, analog 2 and analog 6, but not with analog 7, whereas Leu126 had contact with all others except ATP. The binding of PfTopoVIB and the lead analogs within the Bergerat fold is illustrated ( Figure 4). The 2D interaction diagrams of the lead analogs docked to PfTopoVIB are presented ( Figure S4(A-C)).

Molecular dynamic simulation of best docked proteinanalog complexes
The stability of analog 2-PfTopoVIB, analog 6-PfTopoVIB and analog 7-PfTopoVIB complexes have been evaluated through   Figure 5(B)). After 50 ns run the analog 7-PfTopoVIB has shown fluctuation towards the end of the simulation, which shows that the protein-analog complex is not stable. In case of analog 6-PfTopoVIB, unusual fluctuations were observed between the 30 ns and 50 ns time points indicating weak stability of the complex. A steady RMSD plot is observed for the analog 2-PfTopoVIB complex, indicating that this analog-protein complex is highly stable throughout the run. We have performed three independent simulations for each of the three analogs and the RMSD of all replicates for each protein analog complex are presented ( Figure S5). In all the three runs, analog 2-PfTopoVIB was stable and uniform with an average RMSD value of 0.2 Å, whereas analog 7-PfTopoVIB and analog 6-PfTopoVIB were fluctuating, which indicates that analog 2-PfTopoVIB was more stable as compared to the other complexes.
We have calculated the binding free energy of each of the three complexes; PfTopoVIB-analog 2, PfTopoVIB-analog 6 and PfTopoVIB-analog 7 at 10 ns intervals (0 ns, 10 ns, 20 ns, 30 ns, 40 ns, and 50 ns), as described in the 'Materials & Methods' section and it has been presented in Table 2. The negative values obtained for our analogs show that they have a tight binding with the receptor. The relative binding energy of analog 2 is the lowest indicating strongest among the studied complexes. A detailed flowchart for the steps of virtual screening carried out in our study is given in Figure 6.
The post-MD intermolecular interactions were monitored for the most stable complex, PfTopoVIB-analog 2 (Figure 7). Majority of the contacts formed between the analog and protein residues in the docked complex were found to be retained after 50 ns MD simulation run. The residue Phe122 retained the interaction through halogen (Cl) and Phe105 retained the Pi-Pi T shaped hydrophobic interaction. Lys127 and Lys121 were found to maintain the carbon hydrogen bonds; Lys127 also had a hydrophobic pi-alkyl bond. Leu126 was found to interact with a pi-sigma and a pi-alkyl bond whereas Glu36 retained the two pi-anion interactions. The residues Asn40, Ala44, Lys104, Phe105 and Leu126 which showed strong interaction with Radicicol were found to interact well with analog 2 and retained their interactions post MD simulation as well. Though Lys113, Gly123, Val183 and Ile185 didn't interact, the residues Met32, Lys35, Asp39, Phe109, Ile124, Gly120, Gly125 and Lys127 retained their contacts with analog 2 during post MD simulation. The analysis of the interacting residues in post MD simulation suggests stronger binding thereby stabilizing the complex.

Discussion
Plasmodium falciparum possesses this unique type II topoisomerase TopoVIB, which acts as a novel target to treat malaria. Our previous studies have shown, Radicicol treatment inhibits the endoreduplication of the blood stage parasites, indicating a functional role of PfTopoVIB at that stage. Endoreduplication occurs in two other stages of the parasite life cycle namely liver stage and mosquito stage (sexual stage), and PfTopoVIB might be a critical determinant for those stages as well. Apart from its role in mitotic replication, it might be involved in meiotic recombination along with PfTopoVIA, an ortholog of Spo11. Hence, finding a chemical inhibitor that has specificity towards PfTopoVIB is the need of the hour. Our work shortlists one chemical structure out of 97 that predicts a stable exclusive binding with PfTopoVIB. In analog 2, the two hydroxyl groups present at the C-17 and C-19 position of Radicicol are substituted by two biphenyl groups and the resultant derivative is found to dock with PfTopoVIB with more points of contact and hence found to score better than that of Radicicol. Free energy calculation and 50 ns molecular dynamics simulations indicate

Disclosure statement
The authors of this paper declare that there is no conflict of interest.