Understanding the binding interaction in therapeutic prevention of inflammation by controlling Hsp90 through curcumin and epigallocatechin

Inflammatory checkpoint blockade is the trend of current research targeting Hsp90 with either phytochemical or commercial drugs. It has been extensively studied globally by researchers but nevertheless has been unable to prevent inflammatory disorders successfully. These key considerations set the stage for identification of different combinatorial strategies, curcumin and epigallocatechin to target Hsp90 involved in inflammatory disorders to gain insight into the benefits of this approach. The primary objective in present study is the prior prediction of the binding interactions of curcumin and epigallocatechin with Hsp90 in the living system through computational studies. A suitable animal model sequence for Hsp90 studies was fetched through PHYLOGENY and BLAST. Hsp90 Mus musculus was obtained as a homologous sequence to Hsp90 Homo sapiens and its tertiary structure was homology modeled as its unabridged experimentally derived structure was not available. In additional tertiary model structure was verified using different tools. Single ligand docking and multiple ligand simultaneous docking of curcumin and epigallocatechin to the modeled Hsp90 structure was carried out using the AutoDock tool. In single ligand docking, curcumin had the binding affinity of − 8.3 kcal/mol, while for epigallocatechin − 5.9 kcal/mol and the multiple ligand simultaneous docking binding affinity for Hsp90 to curcumin and epigallocatechin was − 8.4 kcal/mol. Their interface residues were explored for mutational studies and were found to be in an unstabilized state, possibly corresponding to the conformational changes in Hsp90. These strategies emphasize the rationale for combining the modalities, curcumin and epigallocatechin in the successful reduction of Hsp90 expression, henceforth the inflammatory checkpoint would be controlled to augment the anti-inflammatory response.


Introduction
Harnessing the covert capacity of immune system to control inflammation, a basis for many ailments has long been coveted. Targeting inflammation will help to relieve several diseases, where many of these are not completely curable yet at a certain stage and worldwide leading to death on a large scale. Recent studies have shown that Heat shock protein 90 (Hsp90), a molecular chaperone plays a decisive role in implementing inflammation succession and has been marked as signature of inflammation (Tukaj et al. 2017;Mu et al. 2017). Hsp90 proteins dramatically increased their expression when stimulated with cellular stress followed through inflammation (Klein et al. 2017). It acts as a biochemical buffer and facilitates tolerance of inflammatory signaling molecules with the help of its client proteins (Eickhoff and Costa 2017;Yang et al. 2017). Hsp90 is a 90 kDa protein with three domains. The N-terminal domain contains an ATP-binding site which is also a drug-binding site. The central domain is highly charged and has high affinity for co-chaperones and client proteins. The C-terminal nucleotide binding pocket has been shown to not only bind ATP, but also has a second drug-binding site (Roy and Kapoor 2016). This protein maintain conformation of many client proteins involved in signaling pathways for inflammation 1 3 mainly nuclear factor kappa B, kinases, nitric oxide and steroid hormones (Wax et al. 2003).
Hsp90 has emerged as an inflammatory checkpoint for pharmacologic blockage in the development of anti-inflammatory agents (Holzbeierlein et al. 2010). Therefore, unraveling molecular events governing inflammation and potential avenues for therapeutic intervention is of enormous interest. There are various natural and synthetic molecules that rank among Hsp90 inhibitors both in monotherapy and in combination destabilizing Hsp90 by targeting either its N-terminal or C-terminal domain (Singh et al. 2015;Canella and Welker 2017). But none of the Hsp90 inhibitors have currently been approved as a standard drug for not showing clinical efficacy may be due to difference in in vitro and in vivo behavior, high cost, poor solubility, dose toxicity, etc. Combinatorial trials were found to have increased overall response when compared to monotherapies, although there is a lack of combinatorial trials against Hsp90 despite the combination trials showing higher drug efficacy against targets (Richardson et al. 2011). Therefore, we have made an attempt to combine curcumin and epigallocatechin (EGC) for targeting both N-and C-terminals of Hsp90. These are among the herbal compounds which are also a part of dietary sources; curcumin, a polyphenol from Curcuma longa (turmeric) is known to be an effective inhibitor of Hsp90 and is reported to bind the N-terminal domain but has least bioavailability (Lv et al. 2015). Curcumin suppress pro-inflammatory pathways related with most chronic diseases and block both production of TNF alpha and cell signaling mediated by TNF alpha in various types of cell (He et al. 2015). Another herbal compound, epigallocatechin (EGC), a major flavonoid of Camellia sinensis (green tea) is known to be more bioavailable with significant anti-inflammatory activities (Khan and Mukhtar 2008). It degrades protein kinase and can block the migration of cells involved in inflammation. EGC is also reported to bind the C-terminal domain of Hsp90 (Hani et al. 2014). Both curcumin and EGC are reported to follow a common pathway in Nrf2-mediated HO-1 induction regulated by PKCδ, which is involved in suppression of Hsp90 protein activity (Ogborne et al. 2008;Lee et al. 2014). Based on these, we speculate to restrain inflammation through sinking the effect of Hsp90 by accomplishing a multiple-target approach to N-and C-terminal domain of Hsp90 with the combination of curcumin and EGC.
This study provided a framework to predict the binding efficacy of curcumin and EGC addressing Hsp90 as a target prior to in vivo studies, with a potential to reduce and refine the animal experimentation through computational biology, primarily by searching the homologous sequence of animal model for Homo sapiens with respect to Hsp90 protein and molecular modeling of that fetched sequence. Single ligand docking and multi-ligand simultaneous docking (MLSD) was carried out for the homology modeled Hsp90 protein with curcumin and EGC. These key contemplations couple in singling out a novel combinatorial stratagem that

Phylogenetic tree and BLAST
The phylogenetic tree was constructed to determine the suitable animal model sequence, homologous to Hsp90 H. sapiens using CLUSTAL PHYLOGENY tool (https ://www.ebi. ac.uk/Tools /phylo geny/). The following species were considered along with H. sapiens (P07900) to build a phylogenetic tree: Saccharomyces cerevisiae (P02829), Caenorhabditis elegans (Q18688), Oryctolagus cuniculus (P30946), Rattus norvegicus (P82995) and Mus musculus (P07901). Blastp (https ://blast .ncbi.nlm.nih.gov/Blast .cgi) alignment was carried out on Hsp90 H. sapiens, M. musculus and R. norvegicus to contemplate the similarities between them. In addition, the homologous species of query protein was screened with SMART BLAST.

Tertiary structure modeling and their validation
The tertiary structure of Hsp90 (M. musculus) was homology modeled from SWISS-MODEL, as its complete PDB structure was not available. The reliability of Hsp90 was analyzed using QMEAN 6 (qualitative model energy analysis). The 3D structure of template and protein model was superimposed using the DALI server to reveal similarities between them. The modeled structure was further analyzed for structural composition using various tools; the loop regions were identified by the program ERRAT. Conformations in loops were stabilized in the ModLoop server. Energy minimization of the model structure was carried out in Swiss PDB Viewer package. Further, it was validated for accuracy in ProSA-Web, through overall and local model quality.
The l-asparagine and l-glutamine amide incorrect rotamers in the modeled protein structure were verified in NQ-FLIPPER. Further range of quality, alignment confidence, clash analysis, disorder prediction, Pro-Q2 quality assessment, Ramachandran analysis, rotamer analysis and complete structure mutation sensitivity were analyzed from Phyre 2 and were viewed in a structural context. Global structural assessment with overall quality, covalent bond quality, packing quality and torsion angle quality was obtained from PROCESS. A 3D profile of the model was measured to check compatibility with its own amino acid sequence using VERIFY3D. Packing quality and proline puckering were confirmed from WHATIF. The overall structure validation score was obtained from MolProbity.

Docking studies
Structures of curcumin and EGC ligands were obtained from PUBCHEM (https ://pubch em.ncbi.nlm.nih.gov/) and these structures were optimized using CHEMSKETCH and ProDrg server. The pharmacokinetic and pharmacodynamic properties of curcumin and EGC were evaluated to study more range of parameters regulating their absorption, distribution, metabolism, excretion and toxicity (ADMET) properties through pkCSM and Data warrior tools.
Single ligand docking and MLSD of Hsp90 with curcumin/EGC were performed using Autodock 4.2 tool to predict orientation and affinity of ligand candidates in their optimal pose within conformation of the target proteinbinding sites. Single ligand docking was executed by preparing separate coordinate files for Hsp90, curcumin and EGC in PDBQT format to get energy-minimized structures by deleting water molecules, adding polar hydrogen atoms and Gasteiger charges. For ligands, torsions were also set to define root atom and rotatable bonds in torsional degrees of freedom, providing ligands' flexibility. Grid maps were prepared by fixing grid box on Hsp90 active pockets (Morra et al. 2010;Prodromou 2012). Maps were set on a particular ligand to be docked and the grid parameter files were written to assign search space for interaction in the binding sites of Hsp90. Docking conformations were evaluated using these pre-calculated grid maps. Hsp90 was set as rigid to increase the accuracy of binding mode in a specified conformational grid envelope. The docking parameter file was built by calibrating ligand parameters, search parameters and docking parameters at their default values based on the Lamarckian genetic algorithm and empirical free energy scoring function, for effective conformational search and optimal pose estimation. Standard docking calculations were performed to create log files covering the coordinates of docked conformations of the best predicted energy in each cluster. Docked poses resulted were analyzed specifying the orders with ranked conformations from lowest to highest interaction energy.
To accomplish MLSD, Hsp90 and curcumin/EGC, 'PDBQT' input files were loaded. Docking parameter files of curcumin/EGC were individually prepared and combined 1 3 into one file to run MLSD in Autodock program. Randomly ligands created their particular set of state variables and moved independently by searching space for its local energy minimum, reaching active sites of Hsp90 (Li and Li 2010). Curcumin and EGC bind key residues in active pockets of Hsp90 with different binding modes. The key residues at the interface of Hsp90 protein and ligands, curcumin/EGC were observed in LigPlot.
The mutation in interface residues of docked structure between Hsp90 and curcumin/EGC was introduced to study the phenotypic effect of point mutations using PHYRE INVESTIGATOR and their stability was evaluated from mCSM server.

Screening model organism for Hsp90
Mus musculus was found to be the homologous relative species of H. sapiens Hsp90 and hence considered to be a suitable model organism for this study and which can also be further considered for in vivo studies.
The phylogenetic tree obtained from CLUSTAL PHY-LOGENY showed M. musculus and R. norvegicus shared a common node with the query sequence of H. sapiens under one clade (Soltis and Soltis 2003) than other species, S. cerevisiae, C. elegans, O. cuniculus (Fig. 1). Blastp alignment also showed M. musculus and R. norvegicus sharing significant alignment with 99% identity. In SMART blast, M. musculus was found to be the best hit when screened for homologous query protein (Matts et al. 2011) (Online Source 1).

Mus musculus Hsp90 structure modeling and its quality analysis
The 3D model structure for Hsp90 (M. musculus) was built using SWISS-MODEL based on their FASTA sequences. The model was built using template, '5fwk.1.A', a cryo-electron microscopy structure of 'Hsp90-Cdc37-Cdk4 kinase complex' with a resolution of 3.9 Ǻ. This template structure has closed conformation with RMSD of 1.59 Ǻ and the loops were ordered in this crystal structure as it was based on an interaction with kinase. Its structure validation has been reported to have better values, with Ramachandran outliers at 0.63% and sidechain rotamer outliers at 0.00%. Based on this template, the homology modeled structure showed a good sequence identity of 86.29% for Hsp90 M. musculus (Fig. 2a). In the QMEAN 6, graph readings were towards blue indicating expected accuracy scores for the specific features (Bienert et al. 2017). The DALI server showed almost 90% of superimposition onto the template 3D structure for the Hsp90 model (Holm and Rosenström 2010).
Model loops were refined using ERRAT and Modloop up to 95% providing an acceptable model within confidence limit. This makes the structure with minimal or nil steric bumping/hindrance (collisions between atoms) because of reduced loops (Colovos and Yeates 1993;Fiser and Sali 2003;Rahul et al. 2017). The energy minimization obtained an E value of − 34715.82 kJ/mol in the SPDBV package and the negative E value corresponds to increased binding accuracy of protein to ligand during docking (Guex et al. 2009).
The ProSA energy plot for overall model quality revealed a Z-score of − 9.05, reflecting the obtained models that are located within the space of X-ray crystallographic protein scores typically found for native proteins of similar size (Yaqoob et al. 2016). The local model quality test for Hsp90 showed that all residues possess negative energy distribution, as none of the residues fall under positive values corresponding to the appropriate input structure of the molecule. This connotes that the obtained model was reliable and very close to the experimentally determined structure comprising good quality (Wiederstein and Sippl 2007).
The stereo-chemical quality of Hsp90 was assessed using PROCHECK. In Hsp90, 90% of the residues came under the favored region of Ramachandran plot validating it to be a fine structure for docking studies (Fig. 2b). Further, the remaining postscript files with different parameters also had shown a satisfactory score representing the model structure that has good stereo-chemical properties (Laskowski et al. 2001;Morris et al. 1992).
NQ-Flipper showed 0% flips in the Hsp90 model representing the presence of correct rotamers resulting in favorable interactions. Phyre 2 also showed all qualities analyzed to be in good range for the modeled protein. Global structural assessment qualities obtained from PROCESS revealed the average score in the Hsp90 model. VERIFY 3D gave a profile score in which more than 80% of residues were ≥ 0.2 for the modeled structure representing them to be compatible with their own amino acid sequences. The packing quality and proline puckering were also normal when verified by WHATIF and the structural validation also scored the 99th percentile in MolProbity, suggesting that the model structure presented with good resolution (Weichenberger and Sippl 2007;Kelly et al. 2015;Berjanskii et al. 2010;Hekkelman et al. 2010;Eisenberg et al. 1997;Chen et al. 2010) (Online Source 2).

Docking of Hsp90 with curcumin and EGC
ADMET properties of curcumin and EGC were shown to be negative on mutagenic, tumorigenic, irritant, and developmental toxicity properties (Toepak and Tambunan 2017;Lagorce et al. 2017;Pires et al. 2014). Ligand EGC even though it increased with one hydrogen bond donar, one of the criteria in RO5 but satisfied with other criteria's (Galiano et al. 2016;Veber et al. 2002). These ligands also fit in the Lipinski's RO5 and Veber rules. (Online Source 3). The successful single ligand docking of Hsp90 ( Fig. 3) with curcumin had a binding affinity of − 8.3 kcal/mol, while with EGC it was − 5.9 kcal/mol. MLSD orchestrated the simultaneous binding of curcumin and EGC to the hot spots of Hsp90 with the affinity of − 8.4 kcal/mol. Stipulating good binding interactions in both single ligand docking and MLSD. The binding energies obtained were evaluated after continual series of calculations during the formation of protein-ligand docked complex structures with optimal conformational interactions as summarized in Table 1. Primarily unbound system's energies prior to the interaction of protein and ligands were calculated which were found to be near to zero for both single ligand docking and MLSD indicating that they are close to equilibrium. The torsional energies of rotatable bonds in the ligands were calculated and were found to be in a favorable range implying that they were relative to rotatable bonds of ligands by overcoming torsional strains. Additionally, ligands inhibition constants were also computed to predict their half maximum inhibitory efficiency towards protein. Finally, the total internal energy in protein and ligands' interactions were estimated and were found to be with exact same values as unbound system's energies, as by default the Autodock tool assumes internal energy of bound and unbound conformations of protein and ligands are equal, signifying that energy in both the systems occurring from their relative positions and interactions within them are equal. Eventually, docking energies were summed up from intermolecular energies and ligands' internal energies calculated and were found to be negative indicating stable docked complexes for both single ligand docking and MLSD. Overall, the MLSD complex was more stable than the single ligand docking complex (Wright and Usher 2001). In both single ligand docking and MLSD of Hsp90, curcumin and EGC bind to the N-terminal and C-terminal domains. Figure 4 illustrates binding pockets and hydrogen bonding of Hsp90 with curcumin and EGC for docked complexes obtained from Ligplot (Khalid and Paul 2014;Wallace et al. 1995). There was an increase in hydrogen bonding contacts at MLSD than single ligand docking as shown in Table 2. Mutation analysis of curcumin and EGC interface residues obtained from PYRE INVESTIGATOR and mCSM showed residues 138, 96, 139, 537, 534, 584 are highly destabilizing and other interface residues found to be among destabilizing, in model structure pointing them to be mainly involved in altering conformation of Hsp90 active structure (Kelly et al. 2015;Pires et al. 2014) (Online Source 4).
Through this research work, we predicted that M. musculus is a suitable animal model to scrutinize the effect of curcumin and EGC on Hsp90. This prediction will be further employed in designing the experiment for the in vivo work.

Conclusion
The present study validates molecular inhibition of Hsp90 through the combinatorial approach of curcumin and EGC. This type of interaction may increase the overall affinity of molecules at equilibrium, as the association of ligands at two different drug-binding sites of Hsp90 may prove to be a better therapeutic target. Based on the results obtained from computational approach as feedback, further animal tests will be performed to determine effects of inhibitors on a living system. This work may ameliorate in furnishing in vivo studies and can be explored thereby leading to the design of new inhibitors of Hsp90 to abrogate inflammation. Animal experiment results can mimic the outcomes comparable to the end points desired in humans.