Design of novel JNK1 inhibitors using molecular modeling ...nopr.niscair.res.in/bitstream/123456789/45929/1/IJCB 58B(3) 403-41… · Therefore, with an aim to design a novel and selective

Indian Journal of Chemistry Vol. 58B, March 2019, pp. 403-415

Design of novel JNK1 inhibitors using molecular modeling technique: An in silico approach

Ashima Nagpal* & Sarvesh Paliwal

Department of Pharmacy, Banasthali Vidyapith, Banasthali 304 022, India

E-mail: [email protected]

Received 21 December 2017; accepted (revised) 20 December 2018

To address the issue of unavailability of selective JNK1 inhibitors resulting in off-target effects, leading to multiple diseases, an endeavour to discover novel and specific JNK1 inhibitors is taken up in the present study. To achieve this goal, computer-aided drug design approach has been used and a validated 2D QSAR model, of excellent statistical quality, has been developed through MLR (Multiple Linear Regression) and PLS (Partial Lest Square) method. The r2 value obtained through PLS method (0.97) corroborated the r2 value that has been obtained through MLR approach (0.97). The insights obtained through in-depth study of the developed model has capacitated us to design optimized molecules (compound A1OOP and compound A2SSR) with better selectivity profile than the most active compound of the selected set of compounds that have been employed to build the QSAR model. Additionally, molecular docking and structure based pharmacophore design have been performed to ensure that the affinity of the designed molecules towards JNK1 receptor and evaluation of their ADME properties have been done to ensure their lead likeness. Further, extremely small values for Tanimoto similarity index are obtained that clearly suggest that the designed molecules are novel.

Keywords: MLR, PLS, QSAR, molecular docking, structure-based pharmacophore, Tanimoto similarity index

The c-Jun N-terminal kinases (JNKs) are a well-known class of ‘stress-activated protein kinases’ (SAPKs) that in unification with extracellular signal-regulated kinases (ERKs) and p38 mitogen-activated protein kinases (p38 kinases) constitute the family of mitogen-activated protein kinase (MAPK). Expression of JNKs is stimulated by the distinct external stimuli including environmental stress like UV irradiation and osmotic shock1. Additionally, interleukin-1β (IL- 1β), cytokines and tumour necrosis factor-α (TNF-α) have also been reported to invoke their expression. JNKs are encoded by three diverse genes (jnk1, jnk2, and jnk3), that are known to give rise to 10 distinct splicing isoforms in mammalian cells2. Among the three genes, JNK1 and JNK2 exhibit ubiquitous expression in a human body,

whereas, JNK3 is specifically concentrated in the brain, heart, and to a lower extent in testis2−7. Individual JNK isoforms differ from one another in the extent of affinity they exhibit towards its substrates, clearly concluding the substrate specificity to be the fundamental governing element of JNK-dependent signaling pathways, in vivo2.

The data, concluding the modulatory functions of JNKs on innumerable pathways leading to distinct outcomes relating to both physiology and disease, has been usually obtained from diverse biochemical studies as well as through the experiments done on JNK gene knockout animal models8. One of the most crucial outcomes, found to be significantly modulated through JNK pathway, is the occurrence of Type 2 diabetes. Plethora of evidence is available indicating the direct correlation between activation of JNK1 and the emergence of Type 2 diabetes9-11. Activation of JNKs is also known to be directly related to discernible hike in concentrations of TNFα and IL-19,11. The elevated levels of (IL)-1 lead to JNK mediated destruction of pancreatic β-cells12. In a study, JNK1 inhibitor given to diabetic mice successfully prevented apoptosis induced by IL-113 leading to β-islets preservation to a considerable

_______ List of Abbreviations: QSAR: Quantitative Structure ActivityRelationship; JNK: Jun N-terminal Kinase; SAPK: StressActivated Protein Kinase; MAPK: Mitogen Activated ProteinKinase; ERK: Extracellular Signal Regulated Kinase; IL:Interleukin; TNF: Tumor Necrosis Factor; ADME: AbsorptionDistribution Metabolism and Excretion; MLR: Multiple LinearRegression; PLS: Partial Least Square; TSAR: Tools for StructureActivity Relationship; SBDD: Structure Based Drug Design;LOO: Leave One Out.

INDIAN J. CHEM., SEC B, MARCH 2019

404

extent and also exhibited improved release of insulin in response to stimulation upon exposure to glucose14. Also, mutated JIP-1 (scaffold protein JNK inhibitor protein-1) in Type II diabetic patients provides a genetic evidence, further proving the involvement of JNK1 in causing Type II diabetes15. Considering the evidence obtained through data of various studies, suggesting the profound involvement of JNK1 in occurrence of diabetes, JNK1 has become a crucial target for therapy of diabetes. But, most of the inhibitors designed against JNKs bind to the highly conserved ATP-binding domain and therefore have to compete ATP for binding. One of the greatest problem faced is that the JNK isoforms share a very high percentage of homology of their binding site and therefore the JNK inhibitors, that have been developed, so far lack desired level of selectivity. The dearth of selectivity of a molecule leads to major off-target effects which worsens the scenario when chronic diseases are concerned. As all the three JNK isoforms are localized in varied regions and modulate diverse pathways leading to altogether different outcomes, the selectivity of molecules is the most desirable aspect while designing a JNK inhibitor. Therefore, with an aim to design a novel and selective JNK inhibitor, in silico methods including the physico-chemical descriptor approach. Molecular docking and structure-based pharmacophore design, Tanimoto similarity index computation and forecasting of ADME properties have been utilized as tools to screen best compounds from the group of designed molecules. In silico methods being economic as well as time effective are, these days, the methods of choice in the realm of drug discovery. Furthermore, the molecular docking experiments facilitates the prediction of binding mode of the molecule, thereby, enhancing the probability of discovery of a selective drug. The present work has covered molecular docking experiment of the

designed compounds and only those molecules exhibiting excellent dock score and promising binding modes are intended to be taken further for synthesis. Results and Discussion 2D QSAR modeling

Corina option in TSAR 3.3 (Accelrys) converted the 2D structures of individual compounds, sketched using Chemdraw 8.0 (Cambridge Scientific Computing, 2004), into their respective 3D structures, which are then subjected to energy optimization by employing COSMIC option. The choice of the set of descriptors to be included in the final model was made by application of Correlation matrix Table I. Eventually, a validated QSAR (Quantitative Structure Activity Relationship) model of best possible statistical quality was developed (Table II). The quality of the developed model is usually represented in terms of statistical parameter ‘r’. Whereas ‘r2’ describes the percentage of data represented by the derived regression equation. For the developed QSAR model, its value was determined to be 0.97 that implies 97% variance in experimental activity. Predictive power of the model is explained by r2

(CV) and its high value (0.93) is indicative of excellent quality of the developed model. ‘s-value’ depicts the standard error in the derived regression equation and its extremely low value (0.18) shows that probability of error in regression equation is very less. The value of F test is the ratio of variance described by the model and variance owing to error in the regression

Table I — Correlation matrix showing correlation of biological activity with descriptors included in the model

−logIC50 −logIC50 Verloop B5

(Subst. 3) Kier Chiv4(path/cluster)

index(Subst. ) Kappa3 index

(Whole molecule) Number of H-bond Donors (Subst. 1)

1 −0.7708 0.41068 −0.49873 −0.13635 Verloop B5 (Subs. 3) −0.7708 1 −0.31411 0.16551 −0.00045899 Kier Chiv4(path) index(Subs.1 )

0.41068 −0.31411 1 0.37518 −0.024665

Kappa3 index (Whole molecule)

−0.49873 0.16551 0.37518 1 −0.16712

Number of H-bond Donors (Subs. 1)

−0.13635 −0.00045899 −0.024665 −0.16712 1

Table II — Depicting values of statistical parameters utilized to assess predictability of the developed model

Final Model Test set compounds

r r2 r2CV s value F value

9,27,29,30,32,43,49

0.98

0.97 0.93 0.18 175.47

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

405

equation. Higher the value of F test, statistically better the developed model is16. For the developed model its value was found out to be 175.47 which is excellent, hence, it can be deduced from the above results that the developed model is reliable and robust with excellent predictive power. The equation retrieved from MLR technique is as follows:

Y = −1.3958528*X1+1.0120927*X −0.53234375*X3−1.3114671*X4+ 3.2512817

Where X1 = Verloop B5 (Subs.3), X2 =Kier Chiv4 (path) index (Subs.1), X3 = Kappa3 index (Whole molecule), X4= Number of H-bond Donors (Subs.1) and Y signifies the biological activity.

The same training set was then subjected to PLS analysis. This evaluation step was performed to cross validate the results of MLR because, for a statistically superior QSAR model, results of MLR must corroborate with those obtained from PLS17. The equation derived through PLS is as follows:

Y= −1.474665*X1+0.93197149*X2 − 0.50213444*X3 −1.4019291*X4+ 3.4813688

An ideal 2D QSAR model must possess r2>0.6 and r2 (CV) >0.5 (Ref 18). The model developed in the

present study is clearly of an outstanding statistical relevance with r= 0.98 and r2= 0.93 and therefore can be considered to be highly predictable and reliable. The values of crucial statistical parameters, obtained through MLR method of model development, such as F-value, s-value, r, r2 and r2 (CV) are given in Table II.

In addition, by utilising the information encoded by derived set of descriptors, important deductions regarding the dependency of inhibitory activity on structural architecture of the molecules were made. A plot of experimental versus estimated −logIC50 of compounds in the training set and the test set is shown in Figure 1 and Figure 2. This plot is a graphical demonstration of the predictive power of the developed QSAR model. Additionally, the scatter plots can be employed to detect the presence of any outliers in the selected dataset19. In the present model, no molecule was found to be an outlier. The same tools were applied to predict the inhibitory activity of the test set compounds. The calculated experimental activity values along with predicted activities of the training set compounds and the test set are shown in Table III and Table IV, respectively. The high value

Figure 1 — Plot between experimental activities and activities predicted through the developed model using MLR

Figure 2 — Plot between experimental activities and activities predicted through the developed model using PLS

INDIAN J. CHEM., SEC B, MARCH 2019

406

Table III — Training Set Compounds Included In the Final Model With Their Experimental And Predicted Activities

Compd R1 R2 R3 −Log IC50 Predicted activity

(nM) MLR PLS 4

OMe

Me

O OMe

−2.77085 −2.60982 −2.57773

5

OMe

Et

O OMe

−4.28556 −4.30521 −4.27357

6

OMe

cPr

O OEt

−4.19257 −4.21101 −4.18432

7

OMe

CF3

O OEt

−4.27646 −4.38918 −4.34442

8

OMe

CN

O OEt

−2.54407 −2.47466 −2.43238

15

N

Me

O O

−2.45179 −2.2635 −2.26515

16 O

N

N

CH2CH2CH3

Me

O O

−2.77597 −2.9975 −2.99129

(contd.)

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

407

Table III — Training Set Compounds Included In the Final Model With Their Experimental And Predicted Activities (contd.)

Compd R1 R2 R3 −Log IC50 Predicted activity

(nM) MLR PLS

17

O

N

CH2CH2CH3

Me

O O

−2.46982 −2.47603 −2.51546

18

O

N

CH2CH2CH3

Me

O O

−2.45332 −2.36215 −2.39666

19 HN

Me

O O

−2.83632 −2.83632 −2.67322

20

N

N

Me

O O

−1.97772 −1.85465 −1.8579

21

N

Me

O O

−1.74036 −1.72652 −1.73793

22

N

Me

O O

−1.49136 −1.70431 −1.71719

23

N O

Me

O O

−1.89209 −1.43608 −1.46738

28

N

Me

N

N

NH

Me

−2.35411 −2.19538 −2.19857

(contd.)

INDIAN J. CHEM., SEC B, MARCH 2019

408

Table III — Training Set Compounds Included In the Final Model With Their Experimental And Predicted Activities (contd.)

Compd R1 R2 R3 −Log IC50 Predicted activity

(nM) MLR PLS

34

N O

Me

N

N

NH

Me

−1.50515 −1.30644 −1.34059

44

N O

Cl

N

N

NH

−0.60206 −0.97388 −1.00734

45

N O

CN

N

N

NH

−0.954243 −0.88553 −0.91586

46

N

Br

N

N

NH

−1.44716 −1.36369 −1.37943

47

N

Cl

N

N

NH

−1.14613 −1.2241 −1.23956

48

N

CN

N

N

NH

−1.04139 −1.14474 −1.15686

50

N

Cl

N

N

NH

−1.17609 −1.24695 −1.26094

51

N

CN

N

N

NH

−1.07918 −1.16758 −1.17824

(0.87) of the correlation coefficient (r2) calculated between predicted and experimental values, for the

test set, proves the robustness of the developed model20.

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

409

Eventually, after data reduction, QSAR model was built by four descriptors: Verloop B5 (Subs.3), Kier Chiv4 (path) index (Subs.1), Kappa3 index (Whole molecule) and Number of H-bond Donors (Subs.1) (Figure 3). An in-depth study of the structural

architecture, of the selected group of molecules, provided interesting evidences pertaining to influence of these descriptors on JNK1 inhibitory activity of the molecules. Interestingly, all the four descriptors included in the final model correlated well to the

Table IV — Test set compounds, excluded from final model, with their experimental and predicted activities

Compd R1 R2 R3 −LogIC50 Predicted activity MLR PLS 9

OMe

CCH

O OEt

−3.02119 −2.48156 −2.43929

27

OMe

Me

N

N

NH

Me

−2.51983 −2.50616 −2.47636

29

O

N

CH2CH2CH3

Me

N

N

NH

Me

−2.66558 −2.30092 −2.34428

30 N

N

Me

N

N

NH

Me

−2.47422 −1.83708 −1.8407

32

N

Me

N

N

NH

Me

−2.02119 −1.57925 −1.59393

43

N O

Br

N

N

NH

−1.39794 −1.11346 −1.14721

49

N

Br

N

N

NH

−1.62325 −1.38653 −1.40081

INDIAN J. CHEM., SEC B, MARCH 2019

410

change in bioactivity with change in substitution around the common nucleus. Interpretation of the included descriptors

Verloop B constitutes multidimensional group of physicochemical parameters and provide information related to shape as well as volume of the molecule under study. This is very important parameter as it determines the influence of diverse substituents on shape of the molecule as it determines the fate of the molecule when it interacts with the binding site of the protein receptor. An optimum shape plays a crucial role in determining the ability of the molecule to enter and fit closely to the walls of the receptor binding domain.

Kier ChiV4 and kappa-3 descriptive parameters come under class of topological descriptors and are known to provide information pertaining to the size, overall shape, degree of branching and flexibility of the molecule under probe. These parameters helps us to understand the characteristics that the substituents, defined around the molecule, must contain to exhibit strong and effective binding with the desired target21,22.

The number of H-bond donor parameter gives an

explicit idea about the acidic value, due to diverse substitutions, of a molecule. Higher acidic value of any substitution group indicates its greater tendency to donate hydrogen.

Structural skeleton of a molecule is an extremely crucial parameter that governs its activity as a molecule, to align in an appropriate position with the walls of receptor binding site and to establish bonds, must possess groups of desirable physicochemical properties that not only bear characteristics to establish good bonding with the receptor but provide it balanced flexibility as well as rigidity to take most beneficial conformation. Therefore, a molecule containing a group will be able to exhibit desirable activity profile only if that group is present at appropriate position and not at just any position.

As it is clear from the t-values of the descriptors (Table V), the activity is expected to improve with the increase in KierChiv4 path/cluster index, which indicates enhancing the bulkiness at R1 position will lead to increase in the activity but on the contrary activity is predicted to decrease with the increase in Verloop B5 parameter at position R3, Kappa3 index (Whole molecule) and number of hydrogen donors at R1 position in the molecule. This QSAR model was found to be successful in deciphering the significant aspects of the structure of the molecules imposing profound effect on the inhibitory activity of individual molecules. As predicted by the positive correlation of KierChiv4 descriptor, increasing the bulkiness, by replacing the single ring substituent (compound 5; mol. wt. 333.65) with fused heterocyclic component (Compound 44; mol. wt. 387.66) resulted in remarkable increase in JNK1 inhibitory activity (19300nm to 4nm). Also, a consequential hike in inhibitory activity was observed on replacing the ester group by triazole ring, which decreased the molecular volume by compressing the molecule, corroborated the negative correlation of Verloop B5 at R3 position.

S

R3

NH

O

R1

R2

Kappa 3 index (wholemolecule)

Hydrogen bond donor(subst. 1)

KierChi4(path/cluster)index(subst.1)

Verloop B5 (subst. 3)

Figure 3 — Descriptors included in the final model and theirinfluence on different substituents

Table V — Depicting values of certain statistical parameters for individual descriptor

Descriptors Coefficienta Jackknifeb Covariance SEc t-valued t-probabilitye

Verloop B5 (Subs.3) −1.3959 0.07721 0.11345 −12.304 8.7255e−011 Kier Chiv4(path) index(Subst.1) 1.0121 0.12448 0.1187 8.526 4.2944e−006 Kappa3 index (Whole molecule) −0.53234 0.047936 0.04166 −12.778 4.4442e−011 Number of H-bond Donors (Subs.1) −1.3115 9.463 0.19727 −6.648 1.7931e−006 aSignifies the regression coefficient for an individual variable in the regression equations. bSignifies an estimate of the standard error of individual regression coefficient obtained from a jack knife technique on the final model. cSignifies an approximation of the standard error on an individual regression coefficient derived from covariance matrix. d Is the estimate of the significance of an individual variable included in the final model. eSignifies statistical significance for -test values.

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

411

In addition to this, negative correlation of Kappa3 index (whole molecule) suggested that the overall increase in branching of the molecule can be detrimental to the biological activity. Furthermore, negatively correlated “number of hydrogen bond donor” parameter signifies the activity of a molecule will improve by placing a hydrogen acceptor group at R1 position thus reducing the overall hydrogen donating capacity of a molecule. Interestingly, the same was observed when the methoxy-benzyl group was replaced with fused cyclic substituent at this position. Collectively, the above observations validated the reliability of the developed model constituted by the aforementioned descriptors. Designing of optimized molecules

Structural architecture plays a pivotal role in determining pharmacokinetic profile, hence, drug like character of any molecule. It is an extremely challenging task to design a molecule, consisting of distinct features, placed at appropriate position so as to achieve the most effective binding patterns, with profound extent of selectivity, towards a particular receptor. Therefore, upon gaining valuable insights from the developed QSAR model, a set of molecules, by replacing the substituents (R1, R2, R3), in accord with the information retrieved from the above study were designed (Figure 4). The compounds were found to be following Lipinski’s rule of five and thus, were possessing lead-like characteristics. Molecular docking experiments

In the present study, molecular docking was utilized as a screening tool to obtain optimized compounds with better activity than the most active compound of the selected set of compounds that were used to build QSAR model. Interestingly, we were

able to design a set of 10 compounds (not reported here), by utilizing the extracted information. Out of those 10 molecules, two were found to exhibit higher libdock score than our reference compound (compound 44). Docking experiments revealed that compound A1OOP took its position in adenosine binding region of the receptor covering the ribose binding region and established hydrogen bonds, of the receptor with PDB name ‘4G1W’, with GLN37 and showed Vanderwaal’s interaction with ARG72, LYS55, MET77 and PHE170 (Figure 5). Whereas, compound A2SSR occupied same position on the binding domain of the receptor and formed hydrogen

Figure 5 — Molecular docking of A1OOP on receptor with PDB entry ‘4G1W’

Figure 4 — Molecules designed using the information unveiled through the developed QSAR model

INDIAN J. CHEM., SEC B, MARCH 2019

412

bond with GLN37 and exhibited Vanderwaal’s interactions with PHE170, GLY171 and ASN156 (Figure 6). Molecular docking study of compound 44 (most active molecule of selected data set, IC50 4nm) was also performed for the reference purpose (Figure 7). Intriguingly, all the three compounds got aligned on the same position of the receptor, near gatekeeper amino acid MET111, but with different affinity as indicated by their libdock scores. Among the three molecules, A1OOP was found to have highest libdock score of 120.96 followed by A2SSR, 116.73, and then compound 44, 105.2. The obtained results proved the designed molecules to be of better selectivity profile.

Mapping over SBDD pharmacophore Furthermore, mapping of the designed molecules

on the SBDD pharmacophore revealed excellent results as both the molecules exhibited an excellent five feature mapping, clearly suggesting their strong affinity towards the receptor (Figure 8).

Followed by designing of the aforementioned molecules, their Tanimoto similarity index was calculated. Value of Tanimoto similarity index for A1OOP was found out to be 0.06 and for compound A2SSR was 0.09 and such extremely low values proved that both the molecules do not bear similarity with the reported ligands and therefore are quite novel.

ADME property evaluation

In order to achieve in-depth knowledge regarding the drug-like character of the designed molecules, various parameters including Lipinski’s violations23, egan24, veber25, etc. along with their ADME properties were estimated using online available program “swissADME”26 that provided explicit detail pertaining to their pharmacokinetic profile. The ADME studies revealed that none of the compounds violated any of the rules pertaining to exhibit drug-like properties (Table VI).

Figure 6 — Molecular docking of A2SSR with JNK1 receptor with PDB entry’ 4G1W’

Figure 7 — Molecular docking of reference compound of the selected series (compound 44) on JNK1 receptor with PDB entry ‘4G1W’

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

413

Materials and Methods Data set and methodology

Onset of study involved the calculation of linear equation using Multiple Linear Regression (MLR) method which was further correlated with the results obtained from Partial least square (PLS) approach. In order to derive a linear relationship, firstly, the bioactivity data consisting of 32 compounds27, reported as IC50 values for the JNK inhibitory activity, was converted to –logIC50 and was employed as the dependent variable in the QSAR model development. Chemdraw ultra 8.0 was used for sketching of the primary two-dimensional structures of the compounds. Further, each structures was uploaded in the TSAR 3.3 (Accelrys, San Diego, CA, USA) spreadsheet followed by defining of their substituents (R1, R2, R3) around N-(3,4-dimethylthiophen-2-yl)acetamide common moiety (Figure 9). Since, certain characteristics of molecular physicochemical properties can be well understood only through the 3D orientation, conversion of those 2D structures, into high quality 3D molecular structures, was executed through Corina28 option. The structure conversion process was then followed by their energy calculation by employing Cosmic29 force field option of TSAR 3.3. Molecular descriptors for QSAR analysis

Followed by energy minimization of the molecules, the descriptor calculation was performed for each molecule of the selected data set. TSAR 3.3 is an integrated software package and was employed for the computation of wide array of descriptors of Geometric, topological, and electrostatic class of physicochemical parameters30,31 such as molecular surface area and volume, molecular mass, moments of inertia (moment 1, 2, 3 (size, length)), Verloop parameters, Dipole moment (total, bond and x, y, z components), Lipole moment (total, bond and x, y, z components), topological indices (Balaban, Wiener and Randic indices), molecular connectivity indices (Chi and ChiV indices), Molecular shape indices (Kappa and KAlpha indices), LogP, Atom counts (C, H, N and S), LUMO and HOMO eigenvalues, etc. Stepwise multiple regression

Step-wise multiple regression method was performed to obtain a linear equation for the developed model. Multiple linear regression was employed as a tool in order to build a high quality QSAR model. Multiple number of models were generated in which

experimental activity was utilized as a dependent variable and physicochemical descriptors as independent ones. The obtained model forecasted the

Figure 8 — SBDD pharmacophore mapping of the designed compounds A) A1OOP and B) A2SSR on JNK1 receptor (4G1W)

Table VI — Describing certain crucial parameters for the prediction of ADME properties of the designed molecules

Parameters A1OOP A2SSR

Molecular weight 480.51 489.47 Hydrogen bond donors 2 2 Hydrogen bond acceptors

5 8

No. of rotatable bonds 5 6 iLOGP 2.55 2.10 XLOGP 3.4 1.69 P-gp substrate YES NO GI absorption High High Water solubility Moderately soluble Moderately solubleLipinski’s violation 0 0 Ghose violations 0 0 Veber violations 0 0 Egan violations 0 0

Figure 9 — Various substituents around the N-(3,4-dimethylthiophen-2-yl)acetamide nucleus

INDIAN J. CHEM., SEC B, MARCH 2019

414

activity for each molecule from the reported experimental activity. Further, to test the statistical importance of the retrieved linear equations, standard parameters such as Fischer’s ratio (F), standard deviation (s-value) and squared regression coefficient (r2), etc. were determined. The same training set was then subjected to PLS approach that also generated a linear equation for the same set of descriptors consisted the final QSAR model. It is an alternative technique for MLR as it augments the information pool from the regression equation as well as alleviate the chances of over fitting of the built model32. Data reduction

Initially, more than 200 descriptors were obtained but the model, which is composed of a large data, is not considered to be reliable as precise predictions cannot be made through it. The r2 value of this model was found to be (0.57). The reason for obtaining such a statistically poor model was due to the redundant data present in it. Therefore, to eradicate the chances of redundancy and to enhance the reliability of the model, a pair wise data reduction, by employing correlation matrix, was carried out. In descriptors exhibiting inter-correlation, the one showing high correlation with the biological activity was retained and the one with lower correlation value was removed from the dataset33. Eventually, only four descriptors, exhibiting high correlation with the biological activity and no correlation with each other, were included in the final model. Dataset preparation

After the data reduction step, the entire data set was split into the training set and the test set. The quality of the generated model, in terms of predictability, robustness and reliability, was ascertained through the values obtained for certain statistical parameters including standard error of estimate (s), Fischer test value (F-value), correlation coefficient (r), squared correlation coefficient (r2) and cross validated squared correlation coefficient (r2 CV). The model was also observed for the presence of outliers (a molecule exhibiting 1.5 times higher residual value than the standard error of estimate in an equation). Model validation

Validation constitutes an integral part of the model development. Validation process ascertains the quality of the model by meticulously determining the statistical standards of the developed model. In the

present study, two approaches, internal test set and Leave one out (LOO) method were applied to forecast the quality of the developed 2D QSAR model.

Design of new compounds Design of novel molecules, for a desired target

receptor, entails accurate information regarding the structural requirement for a molecule to possess strong binding with receptor. Therefore, the developed 2D QSAR model was meticulously studied and the information unveiled by the descriptors, that entered the final model, and the obtained MLR equation was then utilized to carry out changes around N-(3,4-dimethylthiophen-2-yl)acetamide central nucleus, to obtain a new set of molecules.

Tanimoto similarity index The main goal of the present work was to design

novel JNK1 inhibitors with better selectivity profile. Therefore, to estimate the novelty of those molecules, a tool known as Tanimoto Similarity index was applied. Ligands from PDB were downloaded and used for similarity analysis of the designed molecules through fingerprinting technique.

Molecular docking experiments As an excellent tool for deducing the possible

interactions between the ligand and the protein receptor, molecular docking does not only makes the discovery of lead molecules more reliable, but also makes it extremely economic as irrelevant squandering of money on enzymatic assays can be prevented. For this purpose, crystal structure of a receptor with high resolution, PDB entry “4G1W”, was downloaded in PDB format and prepared according to protocol provided by Accelrys Discovery studio version 2.0.

Structure based pharmacophore design To enrich our findings, an additional approach,

utilizing pharmacophore obtained from the receptor, was used as a validating procedure. Again for this purpose, Discovery studio version 2.0 was employed and crystal structure of the receptor entry “4G1W”was subjected to removal of water molecules, addition of hydrogens and radius interaction site sphere was defined to be 9Å. This was followed by clustering of all non-features including hydrogen bond donor, hydrogen bond acceptor to 2 and lastly, exclusion constraints were clustered to 10. The file was then saved in .msv format.

NAGPAL & PALIWAL: NOVEL JNK1 INHIBITORS

415

ADME prediction studies Provided the knowledge of (Absorption, Distribution,

Metabolism, Elimination) ADME properties of any drug, a forecast regarding its bioavailability as well as toxicity profile can be conveniently made. Additionally, for any molecule to act like a drug, must possess optimal physicochemical properties that eventually decides its fate to be a drug or a poison. Therefore to confirm the drug-like nature of the optimized compounds, a software “swissADME” was used to estimate all the related parameters that are extremely crucial to be exhibited by an effective and safe drug.

Conclusion The problem emanated from lack of selective

JNK1 inhibitor invoked to exert efforts to design novel JNK1 inhibitors. To achieve this aim, a validated and statistically excellent QSAR model was developed and the knowledge, bestowed by contribution of various descriptors, capacitated us to understand the structural prerequisites of JNK1 inhibitors with better selectivity profile. The information was thus applied to design novel molecules. Interestingly, molecular docking studies as well as structure based pharmacophore model revealed two compounds to possess better affinity towards the JNK1 receptor, compared to the most active molecule of the selected dataset. Even when imposed upon pharmacophore, obtained through SBDD approach, both the compounds were found to map five out of six features. Acknowledgement

I am extremely thankful to the Vice-Chancellor, Banasthali Vidyapith, my guide, my colleagues and my lab mates for extending incomparable support to me for my research work. References 1 Barr R K & Bogoyevitch M A, Int J Biochem Cell Biol, 33

(2001) 1047. 2 Gupta S, Barette T, Whitmarsh A J, Cavanagh J, Sluss H K

& Derijard B, EMBO J, 15 (1996) 2760. 3 Derijard B, Hibi M, Wu I H, Barrett T, Su B, Deng T &

Karin M, Cell, 76 (1994) 1025. 4 Kyriakis J M, Banerjee P, Nikolakaki E, Dai T, Rubie E A,

Ahmad M F, Avruch J & Woodgett J R, Nature, 369 (1994) 156.

5 Pulverer B J, Kyriakis J M, Avruch J, Nikolakaki E & Woodgett J R, Nature, 353 (1991) 670.

6 Yang D D, Kuan C Y, Whitmarsh A J, Rincon M, Zheng T S & Davis R J, Nature, 389 (1997) 865.

7 Mohit A A, Martin J H & Miller C A, Neuron, 14 (1995) 67. 8 Kuida K & Boucher D M, J Biochem, 135 (2004) 653. 9 Hirosumi J, Tuncman G, Chang L, Gorgun C Z, Uysal K T,

Maeda K, Karin M & Hotamisligil G S, Nature, 420 (2002) 333. 10 Hotamisligil G S, Diabetes, 54 (Suppl 2) (2005) S73. 11 Musi N & Goodyear L J, Endocrine, 29 (2006) 73. 12 Ammendrup A, Maillard A, Nielsen K, Andersen N A, Serup

P, Madsen O D, Mandrup Poulsen T & Bonny C, Diabetes, 49 (2000) 1468.

13 Bonny C, Oberson A, Negri S, Sauser C & Schorderet D F, Diabetes, 50 (2001) 77.

14 Bennett B L, Satoh Y & Lewis A J, Curr Opin Pharmacol, 3 (2003) 420.

15 Waeber G, Delplanque J, Bonny C, Mooser V, Steinmann M, Widmann C, Maillard A, Miklossy J, Dina C, Hani E H, Vionnet N, Nicod P, Boutin P & Froquel P, Nature Genetics, 24 (2000) 291.

16 Prasad Y R, Kumar P R, Smiles D J & Babu P A, Arkivoc, 11 (2008) 266.

17 Cramer R D, Perspect Drug Discov Des, 1 (1993) 269. 18 Medina-Franco J L, Golbraikh A, Oloff S, Castillo R &

Tropsha A, J Comput Aided Mol Des, 19 (2005) 229. 19 Sanja O, Podunavac K & Velimirović S D, Acta Period

Technol, 41 (2010) 177. 20 Golbraikh A & Tropsha A, Journal of Molecular Graphics

and Modelling, 20 (2002) 269. 21 Hall L H, Mohney B K & Kier L B, Quant Struct Act Relat,

10 (1991) 43. 22 Hall L H & Kier L B, in Moleculer Structure Description: The

Electrotopological State (Academic Press, New York) (1999). 23 Lipinski C A, Lombardo F, Dominy D W & Feeney P J, Adv

Drug Deliv Rev, 46 (2001) 3. 24 Egan W J, Kenneth M M & Baldwin J J, J Med Chem, 43

(2000) 3867. 25 Veber D F, Stephen R, Cheng J H Y, Smith B R, Ward K W

& Kopple K W, J Med Chem, 45 (2002) 2615. 26 Daina A, Michielin O & Zoete V, Nature (2017) doi:

10.1038/srep42717. 27 Bowers S, Truong A P, Neitz R J, Neitzel M, Probst G D &

Hom R K, Bioorg Med Chem Lett, 21 (2011) 1838. 28 Dalby A, Nourse J G, Hounshell W D, Gushurst A K I, Grier

D L, Leland B A & Laufer J, J Chem Inf Comput Sci, 32 (1992) 244.

29 Molecular Modelling and Drug Design, edited by Wylie W A, Vinter J G & Gardner M (Macmillan, London) (1994).

30 Cronin M T & Schultz T W, Chem Res Toxicol, 14 (2001) 1284.

31 Dessalew N, Acta Pharm, 59 (2009) 31. 32 Kubinyi H, Drug Discov Today, 2 (1997) 457. 33 Rameshwar N, Krishna K, Kumar B A & Parthasarathy T,

Bioorg Med Chem, 14 (2006) 319.