In silico identification of antidiabetic target for phytochemicals of A. marmelos and mechanistic insights by molecular dynamics simulations

Abstract The leaves and fruits of Aegle marmelos (L.) have antidiabetic activity. However, the mode of action and molecules having antidiabetic activity are not known. Hence, we conducted molecular docking of phytochemicals with various molecular antidiabetic targets to find the same. Docking prioritized Dipeptidyl peptidase-4 (DPP-4) as the main target for phytochemicals of Aegle marmelos. DPP-4 inactivates intestinal peptides, glucagon-like peptide-1 (GLP-1), and Gastric inhibitory polypeptide (GIP). GLP-1 and GIP stimulate a decline in blood glucose levels, but DPP-4 inhibits their functions resulting high level of glucose. Hence inhibiting the activity of DPP-4 is a well-known strategy to treat Type 2 diabetes. Therefore, to find a mechanism that may be involved to act as a natural inhibitor of DPP-4, we screened five phytochemicals out of seventy-three based on Virtual Screening, ADMET Drug-likeness analysis, and PAINS filtering. Further, all five phytochemicals, i.e. Aegeline, Citral, Marmesinin, Auraptene, β-Bisabolene, and reference compound subjected MDS for analyzing the stability of docked complexes to assess the fluctuation and conformational changes during protein-ligand interaction. The values of RMSD, RG, RMSF, SASA, and Gibbs energy revealed the good stability of these phytochemicals in the active site pocket of DPP-4 in comparison to reference. Additionally, we have done the pharmacophore analysis, which revealed many common pharmacophore features between screened phytochemicals of A. marmelos and reference molecule. Our results show that these phytochemicals are potential antidiabetic candidates and can be further modified and evaluated to develop more effective antidiabetic drugs against DPP-4 to treat Type 2 Diabetes. Communicated by Ramaswamy H. Sarma


Introduction
Type 1 diabetes is a chronic condition in which the pancreas produces little or no insulin. Type-2 Diabetes (T2D) is caused due to insufficient or ineffective insulin and is a major public health problem. International Diabetes Federation reported 425 million people affected with diabetes in 2017 worldwide and fore told that it would rise to 700 million by the year 2045 (IDF, 2017). The ultimate aim behind diabetes treatment is to lower and maintain the glycosylated hemoglobin level below 7% to prevent the risk of micro-and macro-vascular complications associated with the disease (Stein et al., 2013). To reduce blood glucose level and the risks associated with T2D, insulin sensitizers (thiazolidinediones), insulin secretagogues (sulphonylureas; meglitinides), and external insulin delivery (insulin analogs) are extensively used. Usually, the combinations of various therapeutics or mono therapeutics are used to control diabetes. However, the adverse side effects associated with various synthetic antidiabetic drugs have caused renewed interest in traditional systems of medicine where medicinal plants are being screened at a molecular level for the treatment of diabetes. Many medicinal plants as well as herbal formulations have been used in the treatment of diabetes. One such medicinal plant is A. marmelos (L.) Correa, commonly known as bael in India, and belongs to the family Rutaceae. This plant is native to India but widely found throughout the Indian Peninsula (Rahman & Parvin, 2014). It is a popular medicinal plant in the Ayurvedic and Siddha systems of medicine and folk medicines. A. marmelos is used to treat a wide variety of ailments such as dysentery, fever, asthma, hemorrhoids, ophthalmic, deafness, inflammations, diarrhea, cardiac diseases, asthmatic complications, urinary problems, and pancreatic disorder (Bansal & Bansal, 2011;Islam et al., 1995;Ponnachan et al., 1993). Moreover, the alcoholic extracts of leaves have been used as anti-ulcer, antibacterial, antifungal and antidiabetic activities (Bhatti et al., 2012;Kothari et al., 2011;Shenoy et al., 2012;Venkatesan et al., 2009). In the clinical trial, the formulation of leaves of A. marmelos has found effective for diabetic patients. The treatment, reportedly, tends to increase insulin secretion from the pancreas. The treatment of leaf extract on diabetic pancreas showed improved functional state of pancreatic beta cells. This study indicates the hypoglycemic nature of the leaf extract, helping in the regeneration of the damaged pancreas (Das et al., 1996).
In modern times, computational approaches are an integral part of drug discovery to expedite the drug discovery for reducing cost and time. The target-based drug discovery approach has been widely used due to its specific nature and precise action. Many molecular targets have been reported to develop new drugs against T2D. Currently, glucagon-like peptide-1 (GLP-1) agonists, sodium-dependent glucose transporter 2(SGLT2) inhibitors, Aldose Reductase (AR), Peroxisome proliferator-activated receptor gamma (PPAR-c), free fatty acid receptor 1, also known as GPR40, and dipeptidyl peptidase-4 (DPP4) are being clinically tested. Although A. marmelos is a potential antidiabetic plant but the specific phytochemicals of A. marmelos and their molecular targets are not explicitly discovered. Hence, to find out the specific targets and phytochemicals involved in exerting the antidiabetic effect of A. marmelos, virtual screening was carried out by molecular docking using the receptors; GLP-1 agonists, SGLT2, AR, PPAR-c GPR40, and DPP-4 against 73 phytochemicals (Table 1, supplementary material). Virtual Screening (VS) results revealed that DPP-4 might be the most prominent target on which phytochemicals of A. marmelos exert their action to reduce glucose level in blood.
Several studies also have demonstrated that the inhibition of DPP-4 increases the amount of circulating GLP-1 and GIP, which improves the secretion of insulin and decreases the release of glucagon in the pancreas (Baggio & Drucker, 2007;Lovshin & Drucker, 2009). Therefore, the DPP-4 target is recommended for use in patients with T2D ( Figure 1). Ansari et al., in 2021, reported that A. marmelos plant extract was  particularly effective in inhibiting DPP-4 in vitro with up to 44-96% inhibition and IC50 values of 754-790 lg/ml. This compares with almost total inhibition of DPP-4 by sitagliptin and vildagliptin with IC25 and IC50 values of 2.04 Â 10 À3 to 2.43 Â 10 À3 lg/ml and 2.04 Â 10 À2 to 1.70 Â 10 À2 lg/ml, respectively (Ansari et al., 2021). Hence, to gain deeper insights into the mechanism of action, we further extended our study by conducting MDS, ADMET drug-likeness, PAINS filters, and pharmacophore ( Figure 2). The present investigation identifies the new phytochemicals, which could be used as DPP-4 inhibitors. Therefore, this paper presents and discusses the results of our work to find out antidiabetic phytochemicals from A. marmelos.

Construction of phytochemical library
A library of A. marmelos phytochemicals was constructed by searching the scientific literature. The 3D structures of the phytochemicals of A. marmelos (L.) and reference molecules of each receptor were retrieved from PubChem [https://pubchem.ncbi.nlm.nih.gov] in SDF format. All phytochemicals and reference molecules were converted into PDB files using Open Babel software (O'Boyle et al., 2011). "ChemmineR" version 3.4.3 (Cao et al., 2008) software was used for structural similarity searching and functional group analysis, which play an important role in drug discovery for specific biological activity. The comparative study of the functional group of active compounds against DPP-4 enzyme and screened phytochemicals of A. marmelos was carried out to correlate functional groups against diabetes.

Retrieval of molecular targets
The potential antidiabetic targets were retrieved using the online server TarFisDock [http://www.dddc.ac.cn/tarfisdock/], which resulted in many potential targets, namely GLP-1 agonists, SGLT-2 inhibitors, AR, PPAR-c, GPR40, and DPP4. To find out active site prediction of receptors, we used an online server, Computed Atlas of Surface Topography of proteins CASTp 3.0 (http://cast.engr.uic.edu) (Tian et al., 2018). The active site of all receptors was analyzed by calculating the x, y, and z coordinates of a bounded ligand with each receptor in the PyRx software GUI version 0.8 (Trott & Olson, 2010

Target screening by molecular docking
To select the best molecular target for the phytochemicals of A. marmelos. Virtual screening of all targets was conducted by molecular docking into the active site of each receptor by AutoDockVina (Trott & Olson, 2010) using PyRx open-source software (GUI version 0.8 of AutoDock). The docking results of the best target obtained by AutoDockVina were done by redocking using iGEMDOCK. The pharmacological scoring function is given as:

Drug likeness prediction
Analysis of molecular properties and drug-likeness of the screened compound is an important step in drug discovery. Therefore, all screened ligands were evaluated for their druglike nature under different rules: Lipinski's rules of five;'RO5 (Leeson, 2012), Ghose filter and Verber filter (Sekar et al., 2011). The drug-likeness property of the hit molecules was checked by DruLiTo open-source software. In addition the PAINS filter was also used in drug-likeness analysis to filter the false positive from the screened ligands by using PAINS-remover.

ADMET analysis
The pharmacokinetic properties are very important parameters to decide the effectiveness of drugs. The drug which shows good Absorption, Distribution, Metabolism, Elimination, and Toxicity (ADMET) parameters could be approved easily by the FDA or any other regulatory authority. Therefore, we have used the admetSAR server [http:// lmmd.ecust.edu.cn:8000/] for predicting the ADMET parameters. The server uses the previously approved FDA drugs and experimental phytochemicals and can predict the pharmacokinetic parameters. The server predicted various parameters like Blood-Brain Barrier (BBB), Human Intestinal Absorption (HIA), Caco-2 cell permeability, P-GP (P-glycoprotein) substrate/inhibition, Cytochrome P450 metabolism, toxicity, carcinogenicity, and LD50 values.

Molecular dynamics simulation (MDS)
The Molecular dynamics simulation study is widely used for predicting protein-ligand stability. It provides the conformational changes with the time scale from where we can determine whether the protein-ligand complex is stable or not (Shukla et al., 2017). To understand the stability and interaction mechanism of the DPP-4 with its ligand, investigations were done not only at their static structure level but also at the dynamic level by conducting MDS using the Gromacs version 5.0.7 (Pronk et al., 2013). Therefore, we selected five phytochemicals of A. marmelos complexed with DPP-4 receptor and subjected for MDS on a workstation having ubuntu 16.04 LTS 64-bit, 4GB RAM, and IntelV R CoreTM i5-6400 CPU processor. We conducted MD simulation with 3,000,000 steps for 60 ns which usually took two weeks to finish one protein-ligand complex. To conduct MDS, ligand topology was generated by using the CGenFF server (Vanommeslaeghe et al., 2010;Jorgensen & Tirado-Rives, 1988). Then all these systems were solvated by using the SPC water model in a cubic box. Eight Cl À ions were added for the neutralization of all the systems and then energy minimization was carried out for removing the steric clashes of the system. Then the NVT and NPT simulation of 1 ns were carried out for maintaining the pressure (1 atm) and temperature (300 K) of the system. After that, 60 ns simulations were performed. The coordinates were saved in every 2 fs. After simulation, Root Mean Square Deviation (RMSD), Radius of gyration (RG), Root mean square fluctuation (RMSF) was calculated. Post-screening of MDS analyzed by calculating Solvent Accessible Surface Area (SASA), Principal component analysis (PCA), Number of hydrogen bonds, and Gibbs free energy by using Gromacs utilities. The trajectories were visualized by Visual Molecular Dynamics (VMD) software (Humphrey et al., 1996).

Power failure management during MDS
In the Gromacs protocol, there are commands mentioned to restore MDS steps in case of power failure. Therefore, to manage the electricity failure, we used the following command.
Gmxmdrun-smd 0 10:tpr cpi md 0 10 prev :cpt This command was very helpful because sometimes we faced long-time electricity failure, and we were able to restore whole MDS steps, which saved our previous MDS steps.

Pharmacophore
Common pharmacophores for the ligands were analyzed by using Pharmit server and ZINC pharma (Koes & Camacho, 2012). The reference molecule and all screened phytochemicals were further used for ligand-based pharmacophore studies

Construction of phytochemical library
A phytochemicals library of A. marmelos which consists of seventy-three phytochemicals was prepared. The compound ID and molecular formula of all phytochemicals are given in Table 1. The 3D structures of phytochemicals were generated from PubChem Server in sdf format.
The frequency of a functional group of active compounds data set, i.e. Hispidulin, Crisimaritin, Luteolin, Apigenin, Kaempferol, Flavone, Hesperetin, Naringenin, Genistein, Cyanidin, Cyanidin-3-glucoside, Malvidin, and Resveratrol were reported against inhibition of DPP-4 enzyme (Kalhotra et al., 2018) was compared with data set of phytochemicals of A. marmelos. The result of chemmineR revealed that various functional groups are common in both data set ( Figure 3). It was observed that several functional groups of active compounds (such as ROR, RCOR, RCOOR, RCHO, ROH, rings and aromatic) are common to the phytochemicals of A. marmelos and showing higher frequency. Since these functional groups of phytochemicals of A. marmelos may have antidiabetic activity by targeting the DPP-4 enzyme.

Selection of targets of type 2 diabetes
The targets of T2D were selected through the TarFis Dock server. The 3D structures of T2D receptors viz GLP-1 (PDB id-3IOL), SGLT-2 (PDB id-3DH4), AR (PDB id-1USO), PPAR-c (PDB id-3G8I), GPR-40(PDB id-4PHU) and DPP-4 (PDB id-4N8E) were retrieved from the Protein Data Bank [http://www.rcsb. org/pdb/home/home.do]. All water molecules, ions, and ligands were removed from the protein molecules using PyMol software (DeLano, 2002) and made perfect receptors for study. The 3D structures of therapeutic targets of T2D are given in Figure 4. The addition of hydrogen atoms to the receptor molecules was carried out by using MG Tools of AutoDockVina software.

Screening of targets by molecular docking
All phytochemicals of A. marmelos were docked in the active site of all the therapeutic targets of T2D viz., DPP-4, AR, GLP-1, SGLT-2 PPAR-c, and GPR-40 by using AutoDockVina for predicting the best possible binding pose of ligand and protein for higher scoring and better analysis. In the DPP-4 receptor, the binding energy of reference (2 kv) was À6.9 Kcal/mol. Twenty phytochemicals were screened out of 73 phytochemicals, which showed binding energy in the range of À6.9 kcal/mol to À9.3 kcal/mol. In the AR receptor, the binding energy of reference molecule (LDT) was À9.9 kcal/mol while eight phytochemicals that were screened by docking score showed binding energy ranging from À4.9 kcal/mol to 5.0 kcal/mol. Four phytochemicals showed the binding affinity (À9.9 kcal/mol to À10.5 kcal/mol) which were lower as compared to reference molecule. In GLP-1 receptor, the binding energy of reference (10 M) was À4.9 kcal/mol while binding energy eight phytochemicals varies from À4.9 kcal/mol to 5.0 kcal/mol.
In the GPR-40 receptor, the binding energy of reference (2YB) was À8.1 kcal/mol, while selected 11 phytochemicals showed tbinding energyfrom-8.1 kcal/mol to À9.3 kcal/mol. PPAR-c receptor showed significant binding energy with the reference molecule (RO7). No molecules were found to have a better binding affinity than the reference molecule (À9.0 kcal/mol). In the SGLT-2 receptor, the binding energy of the reference molecule (GAL) was À7.3 kcal/mol, and selected15 phytochemicalshada range of binding energy from À7.3 to À9.4 kcal/mol. The molecular docking results reflected the binding energy involved in the ligand-receptor complex formation and generated an account of all probable molecular interactions responsible for their activity. The values of the docking score of all phytochemicals with receptors have been given in Table 2.
The maximum numbers of phytochemicals of A. marmelos were found to interact with the DPP-4 receptor compared to  other therapeutic targets of T2D. The results showed that screened phytochemicals might have a similar mechanism of action as an inhibitor of DPP-4. Hence we further conducted re-docking and MDS to decipher DPP-4 ligand stability and mechanism of action.

Re-validate the docking score of screened target by iGEMDOCK
The iGEMDOCK can be used to validate the total binding energy of the protein-ligand complex. Twenty phytochemicals were screened against the DPP-4 receptor docking using PyRx. These 20-screened phytochemicals were revalidated by iGEMDOCK to identify hits. The total energy (Vander wall interactions þ hydrogen bonds þ electrostatics force) and AverConPair of screened ligands resulting from iGEMDOCK are shown in Table 3. Among all the screened ligands, 20 phytochemicals demonstrated best binding energy (À6.9 to À9.3 kcal/mol) in PyRx docking and À72.0066 to À98.1868 in iGEMDOCK which are comparably better than the reference 2 kv (À6.9 kcal/mol) and standard antidiabetic drug, Sixlapitin (À6.3 kcal/mol).

Drug-likeness prediction
Afterward, all screened phytochemicals were subjected to drug-likeness by using FAF Drug 3 server and DruLiTo software. DruLiTo gives various results based on Lipinski rule of 5, Ghose rule, and VERBER rule. According to Lipinski's rule, most "drug-like" molecules have Log p 5, molecular weight 500, number of hydrogen bond acceptors 10, and number of hydrogen bond donors 5. The Ghose Filter determines drug-likeness based on values of logP (À0.4 to 5.6); Molar Refractivity (40 to 130); M.W. (160-480); Number of atoms (20-70); and TPSA 140. As per Veber Rule, the drug-like molecules have Rotatable Bond Count 10 and TPSA 140. After re-validating of 20 phytochemicals, the drug-likeness prediction was done. Out of 20 re-validated phytochemicals, eight phytochemicals were screened by drug-likeness prediction. From these eight phytochemicals, the top five phytochemicals were selected based on docking energy and total energy of iGEMDOCK.
All the filters of drug-likeness of the top five hit phytochemicals and reference molecule (2 kv) have been compiled in Table 4. All hit phytochemicals satisfied Lipinski's rule, VERBER Filter, and Ghose rule and Citral was the only compound that follows Lipinski's rule and VERBER Filter. Pan assay interference compounds (PAINS) are chemical compounds that likely to interfere in screening technologies via several means but particularly through protein reactivity because they tend to react nonspecifically with numerous biological targets rather than specifically affecting desired targets.

ADMET profile
The admetSAR profiles for all five phytochemicals were evaluated using the admetSAR database. It generated pharmacokinetic properties of phytochemicals under different criteria: Absorption, Distribution, Metabolism, Excretion, and Toxicity. AdmetSAR server has predicted classification and regression values for the hit molecules by calculating different models, such as blood-brain barrier, human intestinal absorption, and CaCO 2 permeability properties.
The admetSAR profiles for all five phytochemicals were evaluated using the admetSAR database. It generated pharmacokinetic properties of phytochemicals under different criteria: Absorption, Distribution, Metabolism, Excretion, and Toxicity. Table 5 illustrates the relative ADMET profiles of the five phytochemicals of A. marmelos as compared to 2 kv as a standard. The LogS value refers to the solubility of the ligand that ideally ranges between À6.5 to 0.5. Among all the screened phytochemicals, Auraptene has the minimum LogS value (À5.24), and Aegeline and Citral have the maximum Log S value (À2.12).
The colorectal carcinoma (CaCO 2 ) and Blood-Brain Barrier permeability (BBB) can assess the membrane's permeability. The CaCO 2 permeability value for all the hit phytochemicals was comparable similar to the reference molecule, 2 kv (0.56). The computational BBB value corresponds to its entry into the central nervous system. The acceptable range of BBB value for an ideal drug candidate ranges between À3.0 to 1.2 (Nisha et al., 2016). All the hit phytochemicals have shown renal organic cationic transporter and P-glycoprotein non-inhibition as a reference molecule, 2 kv. The major enzyme considered to examine the metabolism parameter of ADME is Cytochrome P450 because of its role in Phase I drug metabolism. CYP450 is a group of enzymes that play a key part in drug and fatty acid metabolism (Guengerich, 2003).
The hit phytochemicals exhibited acceptable results in toxicity analysis ( Table 6). The Human ether-a-go-go-related gene (HERG) encodes a potassium ion (Kþ) channel, and its inhibition can lead to cardiac toxicity. All the phytochemicals have shown weak inhibition toward HERG, i.e. they showed a low risk of cardiac toxicity. All hit phytochemicals clear the Ames test except Marmesinin. All the hit phytochemicals clear the carcinogenicity except Citral. Rat acute toxicity LD50 value of reference molecule, all screened phytochemicals, and control drug have given in Table 5. Lethal Dose50 refers to the dose required to kill 50% of the population. Thus, from the different values observed, screened compounds fulfill all the enlisted criteria, and we can predict their candidature as a drug in terms of ADMET profile. Thus, the result was almost similar to the control drug and reference molecule.

2D Interactions of screened phytochemicals
The 2D interaction of protein and screened phytochemicals were visualized along with a reference compound using LigPlot þ v.1.4.5 software. The 2D interactions of the top five screened ligands and reference ligand (2 kv), with receptors depicting hydrophobic bond, hydrogen bonds, and the Hbond lengths, are shown in Figure 6. Glutamate residues at positions 205 and 206, are essential for the enzyme activity of human DPP-4 (Abbott et al., 1999). In our results, out of five phytochemicals, Aegeline, marmesinin, and b-Bisabolene showed hydrogen bonds with Glu205 and Glu206, similar to the reference compound. Citral and Aureptene also showed hydrophobic interaction with these residues, which reflects better stability of these ligands (Table 2, supplementary material).

Molecular dynamics simulation
The Molecular Dynamics Simulation (MDS) was carried out to determine the structural stability of the protein and proteinligand complexes under physiological conditions. Five best ligands which had good binding energy (Aegeline ¼   À8.0 kcal/mol, Citral ¼ À6.9 kcal/mol, Marmesinin ¼ À8.6 kcal/mol, Auraptene ¼ À7.8 kcal/mol, and b-Bisabolene ¼ À7.3 kcal/mol) were selected for MDS. These phytochemicals followed all the Drug likeness rules with accepted ADMET properties. At first, each docked complex was subjected to the process of energy minimization for 50,000 steps of steepest descent, followed by equilibration. The total MDS was run for the duration of 60 ns. After that, the RMSD, RG, RMSF, and SASA calculation analyzed the structural changes in complex and dynamic behavior. The average values of RMSD, RG, RMSF, and SASA have been shown in Table 6.

Root mean square deviation (RMSD)
To monitor conformational and structural changes, RMSD analysis of the backbone atoms of the native protein, DPP-4, and protein-ligand (2 kv, Aegeline, Citral, Marmesinin, Auraptene, and b-Bisabolene) complexes was carried out. Figure 7 depicts the RMSD plot for native DPP-4 and bound systems (protein-ligand complex) which were calculated for the 60 ns trajectory. It is worth noting that native, as well as complexes, achieved the equilibrium after 25 ns. A slight increment in the case of the DPP-4 was noted which remained stable throughout the simulation with an average RMSD 0.22 nm ( Table 6). The plot showed RMSD of DPP-4 protein (Magenta) is 0.22 nm throughout the interaction time, which is acceptable and stable. In the case of Ligands, RMSD of all complexes become stable after the slight fluctuation between 20-25 ns and showed the acceptable range of RMSD (below 0.22 nm) indicating good stability of the ligand (

Radius of gyration (RG)
The Radius of gyration (RG) value indicates the stability of native protein and the bound systems (protein-ligand complexes) along the MDS trajectories by calculating the structural compactness of the biomolecules. The radius of gyration was also determined to check whether the complexes were stably folded or unfolded after the MDS. The 60 ns trajectory was used for the calculation of RG. Figure 8 shows the plot of average RG as a function of time for native protein and all the protein DPP-4-ligand complexes (Aegeline, Citral, Marmesinin, Auraptene, b-Bisabolene, and reference (2 kv), for the 60 ns trajectory. The average RG value of native DPP-4 was 2.34 nm (Magenta). Among the studied complexes, DPP-4-Aegeline and DPP-4-b-Bisabolene complexes showed the average RG value of 2.33 nm (Red) and 2.34 nm (Orange) respectively, which is comparable to the RG value of the DPP-4-2 kv (Black) i.e. 2.33 nm. Three complexes viz. DPP-4-Citral (Green), DPP-4-Marmesinin (Blue), DPP-4-Auraptene (Maroon) an equal RG value of 2.32 nm, similar to the reference molecule.
As the RG had a relatively consistent value throughout the 60 ns trajectory of MDS, it was regarded as a stably folded structure; otherwise, it would be considered unfolded. As it can be seen, all the molecules exhibit relatively similar and consistent values of Rg as compared to the native and reference, they indicate perfectively superimposedon each other and have good stability (Table 6). Hence, all five complexes achieved relatively stable folded conformation during MDS. Overall, from the results it can be concluded that the complexation of protein with screened ligands increases the compactness/rigidity of the protein structure, leading to increased overall stability.

Root mean square fluctuation (RMSF)
The Root Mean Square Fluctuation (RMSF) was used to analyze local changes along with the protein chain residues, and analyze changes in the ligand atom positions at specific temperature and pressure. The higher value of RMSF indicated that the structure has more flexible regions like turns and loops. In comparison, the lower value of RMSF represents that the structure has a secondary structure like helix and sheets considered good as compared to the structure having high RMSF values. The fluctuations in the constituent residues were observed for protein DPP-4 and all the protein-ligand complexes (DPP-4-2 kv, DPP-4-Aegeline, DPP-4-Citral, DPP-4-Marmesinin, DPP-4-Auraptene, and DPP-4b-Bisabolene) were plotted as a function of time during 60 ns trajectories of MDS in Figure 9. All the complexes showed similar RMSF values and fluctuation in the same residues. The average RMSF value for native DPP-4 protein (Magenta), DPP-4-2 kv (Black), DPP-4-Aegeline (Red), DPP-4-Citral (Green), DPP-4-Marmesinin (Blue), DPP-4-Auraptene (Maroon) and DPP-4-b-Bisabolene (Orange) were 0.08 nm, 0.08 nm, 0.09 nm, 0.09 nm, 0.09 nm, 0.08 nm and 0.08 nm respectively. The results indicated that around 0.27 nm fluctuations were observed in the native DPP-4 in the Gln247 residues respectively. The Gln247 from the part of loop regions in all DPP-4-complexes. However, these protein residues are not involved in ligand interactions. In the reference complex, DPP-4-2 kv, the Gln247, and Asn487 showed Root Mean Square Fluctuations (RMSF) 0.23 nm and 0.44 nm respectively.
In complex DPP-4-Aegeline, the Glu73 and Gln247 showed RMSF of 0.26 nm and 0.44 nm respectively (Figure 9). The complex DPP-4-Citral showed relative active interactions with the Gln247 residues as indicated by the RMSF values of 0.40 nm. While in complex DPP-4-Marmesinin, the RMSF values of 0.40 nm were observed for the residues Gln247. Furthermore, in complex DPP-4-Auraptene the Gln247 residues showed active interaction with the RMSF values of 0.30 nm and in complex DPP-4-b-Bisabolene, the RMSF values of 0.30 nm and 0.23 nm were observed for the residues Gln247 and Glu677, respectively. The fluctuation during all interactions was below 0.2 nm, perfectly acceptable. Thus, all the complexes showed RMSF value similar to reference complex DPP-4-2 kv. The RMSF values of plotted data showed that the complexes of all screened phytochemicals showed active interactions with Glu73, Gln247, Asn487, and Glu677 comparable similar to reference complex, DPP-4-2 kv.

Solvent accessible surface area (SASA)
The Solvent Accessible Surface Area (SASA) is a parameter that measures the proportion of the protein surface, which can be accessible by the water solvent. The SASA calculation can be used to predict the extent of the conformational changes that occurred during the interaction (Marsh & Teichmann, 2011). We have carried out SASA analysis taken the 60 ns trajectory. Figure 10 shows the plot of SASA value vs. time for all the protein DPP-4-ligand complexes (Aegeline, Citral, Marmesinin, Auraptene, b-Bisabolene, and reference 2 kv). The average SASA value 334.16 nm 2 was calculated for DPP-4-Aegeline (Red) while average SASA values for DPP-4-Citral (Green), DPP-4-Marmesinin (Blue), DPP-4-Auraptene (Maroon), and b-Bisabolene (Orange) were 336.83 nm 2 , 332.90 nm 2 , 331.51nm 2 , and 334.24 nm 2 respectively. These calculations showed that the DPP-4-Auraptene was least exposed to the water solvent during the 60 ns trajectory of MDS followed by other DPP-4 complexes indicating these complexes are relatively more stable than the DPP-4-Citral complex. All the complexes showed a lower SASA value than the reference DPP-4-2 kv complex i.e. 334.59 nm2 (Black). From SASA analysis, we have concluded that our screened A. marmelos phytochemicals are good as compared to the reference compound.

Hydrogen bonds
Hydrogen bonds play a significant role in ligand binding with the receptor. They strongly influence drug specificity, drug affinity, metabolism, and adsorption. Therefore, the   bonding patterns were assessed by observing the fluctuation of the hydrogen bonds in all the complexes. The total number of hydrogen bonds present in the complexes has been shown in Figure 11. About two hydrogen bonds were observed in the DPP-4-2 kv complex (Black), while three hydrogen bonds were found in DPP-4-Aegeline (Red), and four hydrogen bonds in DPP-4-Auraptene (Maroon). The complex DPP-4-Citral (Green), DPP-4-Marmesinin (Blue), and b-Bisabolene (Orange) showed the minimum two hydrogen bonds equal to the reference. These observed bonding parameters indicated that all phytochemicals were bound to the DPP-4 as effectively and tightly through hydrogen bonds similar to the reference molecule (2 kv).

Principal component analysis (PCA)
To investigate axes of maximum variation of structure distribution during the molecular dynamics, PCA was carried out. PCA demonstrates a set of the small number of modes that capture the majority of fluctuations. It indicates overall motion of the protein is determined by only the first few eigenvectors, which is obtained by a linear transformation of internal coordinates of protein. In this study, we selected the first 40 eigenvectors to calculate concerted motions of the 60 ns trajectory. Figure 12(A) represents the eigenvalues, which were obtained from the diagonalization of the covariance matrix of atomic fluctuations in decreasing order versus the corresponding eigenvector for all studied complexes. The first five eigenvectors account for 55.01%, 57.02%, 58.01%, 44.67%, 60.25%, and 55.63% motions for the last 60 ns trajectory for reference compound 2 kv (Black), Aegeline (Red), Citral (Green), Marmesinin (Blue), Auraptene (Maroon), and b-Bisabolene (Orange), respectively. An increased eigenvalue was observed in all the complexes except DPP4-Marmesinin complex during 60 ns MD for reference DPP-4-2 kv complex. DPP4-Marmesinin complex reflects the differences in the dynamic behavior while all other complexes like DPP-4-Aegeline, DPP-4-Citral, DPP-4-Auraptene, and DPP-4b-Bisabolene did not cause any significant disturbance. Therefore, these ligands may act as potential phytochemicals against DPP-4.
2D projection of trajectory plot generation in PCA is another way to achieve the dynamics of complexes. In this study, we selected the first two principal components (PCs) i.e. PC1 and PC2 for clear prediction of the motions. Figure  12(B) showed the projection of two eigenvectors for a reference compound 2 kv, (Black) as well as hit phytochemicals, Aegeline (Red), Citral (Green), Marmesinin (Blue), Auraptene (Maroon), and b-Bisabolene (Orange). It was observed that the complexes that occupied less space showed as table cluster represented a more stable complex. In comparison, the complex that occupied more space showed a non-stable cluster represented a less stable complex. The plot found that the DPP-4-Citral complex occupied more space and showed a less stable cluster compared to other complexes. The complex DPP-4-Aegeline and DPP-4-b-Bisabolene showed a very stable cluster because it showed a very stable cluster and occupied less phase space than the other complexes. The reference compound, 2 kv (Black) also showed a stable cluster. The results suggested that DPP-4-Aegeline (Red), DPP-4-Auraptene (Maroon), and DPP-4-b-Bisabolene (Orange) were the best-studied complexes. Figure 13 shows the Gibbs energy plot for PC1 and PC2. The plot showed Gibbs energy value ranging from 0 to 11.1 KJ/mol, 0 to 11.5 KJ/mol, 0 to 11.5 KJ/mol, 0 to 11.4 KJ/ mol, 0 to 12 KJ/mol and 0 to 10.5 KJ/mol with reference 2 kv ( Figure 13A), Aegeline ( Figure 13B), Citral ( Figure 13C), Marmesinin ( Figure 13D), Auraptene ( Figure 13E) and b-Bisabolene ( Figure 13F) respectively. DPP-4-b-Bisabolene complexes showed lower and equal energy compared to the reference, DPP-4-2 kv, and other complexes, which suggest that these complexes can follow energetically more favorable transition from one conformation to another as compared to other complexes. The energy minima region was more in the complexes DPP-4-Aegeline, DPP-4-Citral, DPP-4-Marmesinin, and DPP-4-Auraptene suggesting that these complexes are thermodynamically more favorable.
From the overall RMSD, RG, RMSF, SASA, hydrogen bonds, PCA, and binding free energy analysis results, we concluded that DPP-4-Aegeline, DPP-4-Auraptene, and DPP-4-b-Bisabolene are the best stable complexes that showed excellent binding affinities when compared with reference compound.

Pharmacophore modeling
The pharmit server was used to elucidate pharmacophores for the 5 hit phytochemicals of A. marmelos, and reference ligand and reference drug. Pharmacophores were detected by extracting common chemical features from 3D structures of the active ligand set that is representing interactions between the ligands and DPP-4. The results predicted several physicochemical properties: hydrogen-bond acceptor, hydrogen-bond donor atom; a set of atoms of an aromatic ring, and adjacent hydrophobic atoms. The reference molecule, 2 kv showed two hydrogen bond acceptor, one hydrogen bond donor, four hydrophobic groups, one aromatic ring (Figure 14-A). In contrast, the common pharmacophore from five screened molecules showed three hydrogen-bond acceptors, two hydrogen bond donors, three hydrophobic groups, two aromatic rings (Figure 14-B), while the active compound; Silagaptin, showed three hydrogen-bond acceptors, one hydrogen bond donor, five hydrophobic groups, and two aromatic rings (Figure 14-C). Two hydrogen bond acceptors, three hydrophobic groups, one hydrogen bond donor, and one aromatic ring are common pharmacophores in all structures.
Diabetes mellitus is the most common endocrine disorder that affects more than 100 million people worldwide (6% of the population). It is caused by the deficiency or ineffective production of insulin by the pancreas which results in an increase or decrease in concentrations of glucose in the blood. Oral administration of A. marmelos fruit extract has been reported to result in a significant increase in body weight, weight of the pancreas, and insulin levels associated with a significant decrease in fasting blood glucose levels (Das et al., 1996), which indicates that A. marmelos may act like insulin in the restoration of blood sugar and body weight to normal levels. A. marmelos fruit extract treated for improving pancreatic b-cells and partially reversed the damage caused by streptozotocin to the pancreatic islets (Grover et al., 2002;Kamalakkannan & Stanely, 2003;Kar et al., 2003;Marzine & Gilbart, 2005). In pharmacological trials, both the fruit and root showed anti-amoebic and hypoglycemic activities (Kamalakkannan & Prince, 2005;Seema et al., 1996;Shnkar et al., 1980). Thus, it was recommended as a potential hypoglycemic agent. Keeping in mind these facts, we conducted in silico identification of phytochemicals from A. marmelos, which may have potential antidiabetic activity.
A. marmelos 20 phytochemicals screened out of 73 phytochemicals by molecular docking against DPP-4 inhibitor. Further, several phytochemicals were narrowly downed to 8 phytochemicals by drug-likeness analysis. Finally, the top five phytochemicals i.e. Aegeline, Citral, b-Bisabolene, Auraptene, and Marmesinin were finalized based on drug-likeness and selected for ADMET prediction in which three phytochemicals followed all ADMET rule while two compound Citral and Marmesinin showed as the carcinogenic and toxic effect. Hence, we searched about these phytochemicals in the scientific literature. According to the International Fragrance Association (IFA) report, Citral has been extensively tested, with no known genotoxicity or carcinogenic effect, and   applied as fragrances in perfume. This substance is also applied as a flavor additive in food stuffs (Bickers, 2005;WHO, 2004). Since Marmesinin showed the result as toxic but in a report of the cancer cell line, Marmesinin showed no significant cytotoxicity (IC50 > 100 mm) towards the SK-OV-3 cancer cell line (Hyung-In Moon et al., 2011).Therefore, these five phytochemicals were selected for MDS. In MDS, all protein-ligand complexes showed very good stability indicating that these phytochemicals, may be potential antidiabetic candidates. Aegeline has been reported to enhance the PI3kinase/Rac1/PAK1/co-filin pathways in the rearrangement of cytoskeleton, revealing the new molecular drug targets for the treatment of Type 2 diabetes (Gautam et al., 2015).Auraptene treatment has been found to suppress hyperlipidemia and triglyceride accumulation in the liver and skeletal muscle, and it increased the mRNA expression levels of the PPAR-a target genes involved in fatty acid oxidation in the liver and skeletal muscle. Moreover, the adipocyte size in the Auraptene-treated mice was significantly smaller than that in the control of HFD-fed mice resulting in the improvement of HFD induced hyperglycemia and abnormalities in glucose tolerance. Auraptene also activates PPAR-a in vivo and its treatment may improve abnormalities in lipid and glucose metabolisms, suggesting that Auraptene is a valuable food-derived compound for managing metabolic disorders (Takahashi et al., 2011). Although, there has been no report of using Beta-bisbolene in any disease, the parameter of molecular dynamic simulation presented with potential candidates along with all other screened compounds against diabetes. These revelations might affect and diversify the discovery and development of novel antidiabetic drugs based on A. marmelos phytochemicals.

Conclusion
In this study, we identified potential antidiabetic drug-like phytochemicals from a library of A. marmelos. From our in silico study results, it can be concluded that phytochemicals; Aegeline, Citral, Marmesinin, b-Bisabolene, and Auraptene may be the potent antidiabetic drug. The detailed computational analysis provided deeper structural insight into the interacting residues (i.e. Glu205 and Glu206) of DPP-4 that are involved in the process of inhibition. These hit molecules could be further used to develop a newer line of antidiabetic drugs against T2D. Thus, we suggest that these phytochemicals could be effectively used for the synthesis of drugs for the treatment of diabetes. This study is crucial for providing future pharmacological applications of A. marmelos phytochemicals with the prospect of developing an effective and natural drug for the treatment of diabetes. Hence as a therapeutic alternative agent in antidiabetic studies, this study could be useful for opening a gateway for further clinical investigations. Therefore, we suggest that further investigations, and in vivo studies are needed to verify their antidiabetic activity and the development of natural DPP-4 inhibitors from A. marmelos. These molecules could also be used as the combination therapy along with the synthetic molecules. Conclusively, these phytochemicals may enhance the discovery of effective inhibitors as well as a novel drug candidates for diabetes targeting DPP-4. MD and post-MD simulation. Shalini Mathpal contributed to the construction and analysis of Lig plots. Dr. Subhash Chandra is a Co-supervisor of Priyanka Sharma. He has guided in the methodology troubleshooting of computational techniques. Dr. Sushma Tamta is a supervisor and she has provided her critical analysis in this manuscript.