The PDB structures of both drug targets are not available. So, modeling of both the target proteins were performed using MODELLER. A template search has been performed through BLAST and PSI-BLAST programs [7]. Global alignment method was used for comparison between the target-template sequences [4]. Gaps with variable gap penalty function are included for structural loops and core regions, in order to get maximum correspondence between the sequences. Alignment file for MODELLER was prepared by CLUSTALW [8]. Fold recognization was done through mGenThreader and LOMETS server for fold assignment [9]. Energy minimization of generated 3D-model was done through GROMACS (OPLS force field) by using Steepest Descent and Conjugate Gradient Algorithms [10]. Parameters like covalent bond distances and angles, stereochemical validation, atom nomenclature were validated using PROCHECK and overall quality factor of non-bonded interactions between different atoms types were measured by ERRAT program [11]. RMSD (root-mean-square deviation) and RMSF (Root Mean Square Fluctuation) was calculated for modeled structures. Functionally important residues (Active-site) were identified through comparative result of POCKETFINDER and SURFACE RACER 4.0.

Drug analogs were created through mono- substitution in the hydrophilic region on the target molecule with other functional groups (-I Group to +I Group) through CHEMSKETCH 10.0. Generated analogs were further analyzed by comparative study of molecular descriptors, various energies and QSAR studies [12, 13, 14, 15, and 16]. Drug-binding specificity of target to analogs was also performed with docking through AUTODOCK 4.0 [17, 18]. Mono-substitution studies have more relevant results and valuable in the effective analog development. Initially 169 analogs were generated from the reference compound (Dutasteride, Drug bank accession DB01126). Prediction and identification of potential drugs and non-drugs from generated analogs is most important task [19]. Objectives (Independent variables) as well as subjective (Dependent variables) approaches were used in identification and selections of chemical descriptors [20].

Prediction and identification of potential drugs and non-drugs from generated analogs is most important task. Toxicity behavior of selected analogs was studied through PALLAS and different types of toxicity (i.e. Oncogen, Mutagen, Teratogen, Irritation, Immunotox and Neurotox) were reported in selected analogs on the basis of toxic fragments.

The Lipinski “Rule of Five” states that compounds are likely to have good absorption and permeation in biological systems. Pharmaceutical industry is following this index for evaluation of drug health. We performed the lipinski evaluation for all the selected analogs.

For modelling of both drug targets, template search were performed with the BLAST and PSI-BLAST and did not find single suitable PDB template for target protein sequence. Thermus Aquaticus Core Rna Polymerase-Rifampicin Complex (1I6V) and methanococcus Jannaschii phosphosulfolactate synthase (1QWG) are identified as templates for drug target-I. 1UTH and 2UYE were identified as templates for drug target-II along with other PDBs. Modelling of dutasteride drug targets was a tedious task due to very low sequence similarity and coverage. Three dimensional models of drug targets were generated through identified templates along with fold fitting. Fold recognization was done through mGenThreader and LOMETS server for fold assignment. Helices have dominance over other secondary structure i.e. sheet, coil in both generated model. Both modeled structure of drug target-I (Model-I) and drug target II (Model-II) is shown in figure-1(a&b).

Figure 1

Figure-1(a & b) Three dimensional visualization of the modeled structure of drug targets (Model-I & Model-II) through VMD software.

Energy minimization of generated 3D-model was done through GROMACS (OPLS force field by using steepest descent and conjugate gradient algorithms).The generated 3D model (Figure 1a and 1b) of target proteins was checked by Ramachandran plot (Figure-2a & 2b) through PROCHECK program.

Figure 2

Figure -2(a &b) shows torsion angles of and in the generated models through Ramachandran plot of generated model I & II

For model-I, 91.03% residues was lies in most favoured region whereas 88.09% residues were lies in most favoured region in model-II in Ramachandran plot .Besides very low sequence similarity and sequence coverage to PDB templates, the overall quality factor for drug target model-I (90.04%) and model-II (84.647) were reported through structure validation server ERRAT (Figure-3a & 3b).

Figure 3

Figure-3(a) shows the overall quality factor of generated model-I of drug target of dutasteride through structure validation server ERRAT.

Dark black regions were showing high error due to high structural variability (high transition of E, C and H) in generated models. Like –in model-I two dark places (100-120 and 245-259).

Figure 4

Figure 5

Figure-3(b) shows the overall quality factor of generated model-II of drug target of dutasteride through structure validation server ERRAT.

The molecular dynamics simulations have provided significant new information on the nature of proteins. RMSD measures the accuracy whereas dynamic fluctuations (RMSF) of proteins around their average conformations play an important indicator of many biological processes such as enzyme activity, macromolecular recognition, and complex formations [21].

Figure 6

Figure-4(a, b, c) Analyzing structural variability of model-I protein structures using RMSD

The changes in structural conformation were monitored in terms of RMSD and RMSF [5]. Figure -4(a, b, c) shows, for model-I RMSD after 10 ps time when calculated for Backbone-Backbone was 0.19, for Protein-Protein was 0.23. Further when the simulation cycles were increased to 50ps time RMSD for Backbone-Backbone was 0.22 and Protein-Protein was 0.26. Results clearly indicated that energy minimization should be done for few steps only. More energy minimization steps can cause more structural hindrance [22].

Figure 7

Figure-5(a, b and c) Analyzing structural variability of model-II protein structures using RMSD.

Figure -5(a, b, c) shows, model-II RMSD after 10 ps time (Backbone-Backbone (0.225), Protein-Protein (0.265) whereas RMSD after 50ps (Backbone-Backbone (0.3), Protein-Protein (0.265).

Figure 8

Figure-6 Residues wise (C) RMSF of model-I and model-II

Model-I has sequence length of 259 residues whereas model-II contains 254 residues. During the course of MD trajectory, coils are more prone to fluctuations than the alpha helices (figure-6). After modeling of target proteins, characterization of structure is another very important task for rational drug designing. Prediction of active site of target protein was done by comparative result of POCKET FINDER and SURFACE RACER-4.0 which is shown in table-1.

Figure 9

Table-1 Comparative results of softwares and analysis for Active-site prediction

Drug analogs were created through mono- substitution in the hydrophilic region on the drug molecule with other functional groups (-I Group to +I Group) through CHEMSKETCH 10.0 and compatibility of analogs was checked through AUTODOCK 4.0 against modeled structures of target proteins(Target-I + Target-II) with binding energies.

Figure 10

Table-2 shows binding energy list of selected analogs and their best binding conformations

Figure 11

Table-3 Comparative results of the molecular descriptors of selected analogs.

Comparative results of the molecular descriptors of selected analogs are given table-3. It was observed that changes in molecular descriptors of selected analogs are under considerable range.

Toxicity prediction is most important criteria for new analog selection. Toxicity of a molecule is highly dependent upon its structural elements. Certain molecule segments are characteristic of hazardeous compounds and therefore called toxic segments. Toxicity prediction is therefore based on Segment body of chemical, Positive condition, Negative condition that can be used to predict certain health hazards of organic chemicals combining toxicological knowledge and judgment with the use of leading-edge QSAR artificial intelligence techniques. Chemical segment body of analog generated from dutasteride is alpha-beta unsaturated (thio) amide, Aliphatic Polyholide-1-2, Benzyl halide/sulphate, T_Har_Hat_hat08. Segment body of analogs shown in Table [4].

Figure 12

Table-4 Showing toxicity of analogs with reference compound by using PALLAS-3.112.

Figure 13

Graph-1 showing biological toxicity effects (%) of analogs with reference compound

These analogs are showing different type of biological toxicity effects (Oncogenicity, Mutagenicity, Teratogenicity, Membrane Irritation, Sensitization, Immunotoxicity, and Neurotoxicity). Level of toxic effects is high in three analogs (-CCL3, -CH3, -CCL2OH) which is shown in Graph-1

The Lipinski “Rule of Five” states that compounds are likely to have good absorption and permeation in biological systems. Pharmaceutical industry is following this index for evaluation of drug health.

Figure 14

Table-5 Showing Lipinski Status of designed analogs with the reference compound.

The present model can be further extended with some modifications, if necessary for analysis of other drug compounds which have more side effects. All these analysis and predictions are made on the basis of bioinformatics tools & techniques by statistical analysis. Theoretical calculation may varie from software to software but the protocol which is developed is up to mark for the new drug discovery. This knowledge will contribute positively to develop new drug analogs with more efficacies as well as less side effects.

We are grateful to Department of Bioinformatics, MANIT, Bhopal, India for support and cooperation.