MEK-1 and MEK-2 are dual-specificity kinases and important components in the mitogen-activated protein kinase pathway. These enzymes are crucial for normal cell survival and are also expressed in several types of cancers, making them important targets for drug design. We have applied an integrated in silico approach that combines comparative molecular field analysis, comparative molecular similarity indices analysis, and molecular docking to study the structural determinants for the recognition of substituted isothiazole analogs as allosteric inhibitors against MEK-1 kinase. The best 3D-QSAR models for comparative molecular field analysis and comparative molecular similarity indices analysis were selected based on statistical parameters. 3D contour maps suggested that bulky or long-chain substitutions at the X position on the core part decrease the inhibitory activity, and the presence of a hydrogen bond donor substitution enhances the activity. The bulky and electronegative substitutions at the Y position on the core part enhance the activity of the inhibitors. Molecular docking studies reveal a large and hydrophobic pocket that accommodates the Y substitution and a polar pocket that accommodates substitutions on the X position and forms hydrogen bonding interactions with MEK-1 kinase. The results of the 3D-QSAR analysis corroborate with the molecular docking results, and our findings will serve as a basis for further development of better allosteric inhibitors of MEK-1 kinase against several cancers.
In eukaryotes, signals from outside the cell to inside the nucleus are transduced by various growth factors, hormones, adaptor proteins, and enzymes to regulate normal cell survival, proliferation, and differentiation. The mitogen-activated protein kinase (MAPK) pathway that includes the Ras/Raf/MEK/extracellular-signal regulated kinase (ERK) signaling proteins is one of the important pathways in signal transduction. Ras is an upstream activator of several signaling pathways such as MAPK pathway. Kinase suppressor of Ras is an essential scaffolding protein that co-ordinates the assembly of Raf–MEK–ERK complexes (1–3). B-Raf and other upstream activator kinases phosphorylate and activate the MAP kinase/ERK kinases (MEK-1 and MEK-2) via serine phosphorylation. The activated MEK-1 and MEK-2 phosphorylate and activate the ERK1 and ERK2 MAPKs. MEK-1 and MEK-2 are dual specificity kinases that phosporylate both threonine and tyrosine residues on ERK1 and ERK2 (43- and 41-kDa MAP kinases, respectively) (4–6). In tumor cells, aberrant activation of the MAPK pathway owing to the mutations in Ras and Raf is frequently observed. Also, activated MAPK or elevated MAPK expression has been detected in a variety of human tumors, including breast carcinoma and glioblastoma, as well as primary tumor cells derived from kidney, colon, and lung tissues (6). Considering the importance of the components in this pathway as important targets for antitumor drugs, the inhibition of Raf, Ras, MEK, and ERK represents a viable model to block uncontrolled cell growth and develop anticancer agents with therapeutic utility.
At least nine members of MEK family (MEK-1, MEK-2, MEK-3a, MEK-3b, MEK-4, MEK-5a, MEK-5b, MEK-6, and MEK-7) are present in humans and termed MEK-1 to MEK-7, of which two highly homologous proteins MEK-1 and MEK-2 are present in MAPK pathway. Because of the high sequence similarity between MEK-1 and MEK-2, the inhibitors developed so far are non-selective, and there is an evidence, however, that they are regulated differentially and may not be interchangeable in all cellular contexts (7–10). On the contrary, a unique hydrophobic pocket within the MEK enzyme allows for the interaction of highly selective allosteric inhibitors and therefore identification of MEK-specific inhibitors. Recent studies report specific non-ATP competitive inhibitors of MEK-1 and MEK-2 kinases. The first-generation MEK-1/2 inhibitors such as PD98059, U0126, and PD184352 were also found to inhibit MEK-5 and the ERK-5 MAP kinase pathway at higher concentrations (11,12). The MEK-5 shares 83% of amino acid sequence identity with the PD184352-like inhibitor-binding pocket of MEK-1 (13,14). Few drugs such as AZD6244, GDC-0973, RDEA119, GSK1120212, RO5126766, RO4987655 and AS703026 are presently in clinical trials (14). The MEK inhibitors CI1040, PD-0325901, and selumetinib tested in clinical trials have been associated with skin rash (15–17).
In this work, we performed the molecular modeling studies of the MEK-1 inhibitors using 3D-QSAR and docking approach. Ligand-based 3D-QSAR approaches have been found to be valuable in further development of novel potent inhibitors. Comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) are two 3D-QSAR methods that have been successfully employed in drug design (18,19). Cramer et al. proposed CoMFA that describes the molecular properties by 3D steric (Lennard-Jones) and electrostatic (Coulomb) fields, evaluated over a lattice of points. Partial least-squares (PLS) method is used to correlate the variation of these properties with the variation of the biological response (20). Comparative molecular similarity indices analysis uses a Gaussian-type distance-dependent function to assess five fields of different physicochemical properties like steric, electrostatic, hydrophobic, hydrogen bond donor, and hydrogen bond acceptor descriptors. Both the 3D-QSAR methods give contour maps as output that can be used to gain general insights into the topological features of the binding site (21). In the conservative ligand-based QSAR method, the active conformations are obtained through minimizing the molecules and selecting those with lower energy. The receptor-based conformation determination by molecular docking takes into account features of the binding pocket, and the information derived from the combination of both these models is thus more reliable.
Molecular docking of inhibitors into the active site of protein identifies the binding orientations and the protein–inhibitor interactions responsible for the observed activity. The crystal structure of MEK-1 (PDB ID: 3E8N) was used to dock the isothiazole analogs into the allosteric binding site of the kinase domain using Genetic Optimization of Ligand Docking (gold) software (22,23). The best docking pose was identified on the basis of the gold scores, and these conformations have been considered to analyze the binding position and interactions in the MEK-1–inhibitor complexes. Hence, the essential information gathered by performing simultaneous 3D-QSAR and molecular docking in this work would be helpful to understand the structure–activity relationships of MEK-1 inhibitor molecules and aid in the rationale design of new potent inhibitors.
Materials and Methods
Data set preparation
In the present work, a series of 69 substituted isothiazole analogs of MEK-1 kinase inhibitors have been used for the 3D-QSAR studies. All the molecules were collected from the literature along with their inhibitory activities (24,25). MEK-1 kinase in vitro activities were converted from IC50 to the corresponding pIC50 (−log IC50) values that spanned in the range of three log units. This total data set of molecules shown in Table S1 and Figure S1 was initially selected because the same biological assay was used for measuring their activity. Among the 69 molecules data set, 54 molecules were used as training set for generating QSAR model and 15 molecules were used as test set for validating the best QSAR model quality. We represented a balanced number of active and less-active molecules in both the training and test sets for uniform sampling of data.
Molecular modeling and alignment
The 3D structures of molecules were drawn using sketcher module in sybyl-8.0.1 software (Tripos Inc., St. Louis, MO, USA). Steepest descent method was used for the energy minimization of all molecules with Tripos force field and distance-dependent dielectric constant with a convergence criterion of 0.001 kcal/mol. Partial atomic charges were assigned using MMFF94 method. All the molecules were aligned on the isothiazole core part (as shown in Figure 1) using database alignment module implemented in sybyl-8.0.1.
CoMFA and CoMSIA
All the aligned molecules were placed in a 3D cubic lattice with a grid spacing of 2 Å, which was generated automatically by the program. Comparative molecular field analysis fields were generated using sp3 carbon atom as the steric probe and a single positive charge (+1) as the electrostatic probe at each grid point. The CoMFA model was generated using steric and electrostatic probes with a truncated cutoff ±30.0 kcal/mol. The attenuation factor is represented by α with a default value of 0.3 and an optimal value normally ranging from 0.2 to 0.4. The CoMFA fields were scaled by the CoMFA-STD (CoMFA standard) method in SYBYL (26–29). The CoMFA/CoMSIA results were graphically interpreted by field contribution maps (contour maps) using the STDEV*COEFF (standard deviation coefficient) field type.
PLS analysis and validation of QSAR models
The CoMFA/CoMSIA fields combined with the actual biological activities (pIC50) were included in a molecular spreadsheet, and PLS method was applied to generate 3D-QSAR model (30–32). The PLS algorithm with the leave-one-out cross-validation method (20) was employed to choose optimum number of components and assess the statistical significance of each model. The optimum number of components is the number of components resulting in the highest cross-validated correlation coefficient (). The cross-validated PLS analyses were performed with a column filter value of 2.0. The cross-validated r2 (), a squared correlation coefficient generated during a cross-validation procedure, is used as a diagnostic tool to evaluate the predictive power of the equation and is calculated according to following formula (33):
where Yav represents average activity value of the entire data set and Yobs and Ypred represent actual and predicted activity values of the target property (pIC50), respectively. The optimum number of components was chosen, which gave less standard error of prediction and high . To test the real predictive ability of the best 3D-QSAR CoMFA/CoMSIA models using the training set, the pIC50 values of the external validation set (test set) were calculated using the same CoMFA/CoMSIA options that generated the best models. The external predictability of the generated model was determined using the standard deviation of error prediction values for this test set according to the following equation (33):
where Ypred(test) and Ytest indicate the predicted and actual activity values, respectively of the test set compounds, and Ytraining indicates mean activity values of the training set. In addition, , and number of components, the conventional correlation coefficient r2 and its standard error were also computed.
Fischer statistics (F) is the ratio of explained to unexplained variance for a given number of degrees of freedom. F-test is a variance-related statistics that compares two models differing by one or more variables to see whether the model with greater complexity is more reliable than the less-complex one. The model is supposed to be good if the F-test is above a threshold value, i.e., tabulated value. A larger value of F is indicative of the greater probability that the QSAR equation is significant.
Selected molecules were docked manually into the binding site of MEK-1. The crystal structure of ligand bound to MEK-1(PDB: 3E8N) was used as the reference for docking studies. Structure-based studies are reported to be useful in validating 3D-QSAR results (34). Molecular docking is used to determine the binding orientation of inhibitors to their protein targets to predict their affinity and activity when bound to the protein. We performed molecular docking of all inhibitors using gold software (22,23). The docking studies identified the inhibitor orientation in the crystal structure of MEK-1. gold is a genetic algorithm for docking flexible ligands into protein binding sites. A 15 Å radius around the inhibitor in the binding site was targeted as the site for docking using the parameters as described earlier (35,36). After docking, the individual binding poses of each ligand were reranked according to the gold score, and the conformation with high gold score was analyzed using HERMES and Discovery studio-2.1 to understand the mode of protein–inhibitor binding.
Results and Discussion
CoMFA and CoMSIA
In CoMFA and CoMSIA QSAR models, their respective field descriptors were used as independent variables, and the pIC50 values were used as the dependent variable. Partial least-squares regression was employed for deriving the models. The actual and predicted pIC50 values and their residual values for the training set and test set compounds are given in Table S1.
The various statistical parameters calculated in CoMFA and CoMSIA models are shown in Table 1. Two different methods, MMFF94 and Gasteiger–Huckel (G–H), were employed to assign charges to the molecules for the generation of the QSAR models. The CoMFA model generated using MMFF94 charges gave a cross-validated correlation coefficient () value of 0.795 with an optimized component number 4. We obtained a high non-cross-validated correlation coefficient (r2) of 0.934 with a low standard error estimate (SEE) of 0.148. A QSAR equation is generally acceptable if the value of correlation coefficient (r) is approximately 0.9 or higher (37). The r value is a relative measure of the quality of fit of the model, and its value depends on the overall variance of the data. The F-value obtained for CoMFA was 172.6 [F05 4, 49 = 2.56(tab)], and predictive correlation coefficient () of 0.676 was obtained. The contributions of steric and electrostatic fields were 0.492 and 0.508, respectively.
Table 1. A summary of 3D-QSAR CoMFA and CoMSIA results
aOptimum number of components obtained from cross-validated partial least-squares analysis and same used in final non-cross-validated analysis.
bNon-cross-validated correlation coefficient.
cCross-validated correlation coefficient.
eStandard error of estimate.
gField contributions: steric (S) and electrostatic (E) fields from comparative molecular field analysis (CoMFA). Steric (S), electrostatic (E), hydrophobic (H), donor (D), and acceptor (A) fields from comparative molecular similarity indices analysis (CoMSIA).
The CoMFA model generated using G–H charges has the value (0.754) with an optimized component number 4. The r2 of this model was 0.927, and the SEE value was 0.155. The F value obtained for CoMFA was 154.9. Inspite of having the same component number, the MMFF94 charges–assigned CoMFA model had the statistical parameters such as and F-values higher and the SEE value lower than the corresponding values for the CoMFA model generated using G–H charges.
The CoMSIA model generated with the molecules having MMFF94 charges was selected for further analysis. The CoMSIA model statistical values from different combinations of the descriptors were observed for the analysis. The CoMSIA method defines explicit hydrophobic (H) and hydrogen bond donor (D) and acceptor (A) descriptors in addition to the steric (S) and electrostatic (E) fields used in CoMFA. With the descriptors SEHD, the CoMSIA model showed the best statistical values over other combinations such as SE, SEA, SEH, SED, SEDA, SEHA, and SEAHD. These statistical values are mentioned in Table 1 and showed that the CoMSIA with the SEHD had better value (0.730) with an optimized component number 5. A high r2 value (0.932) with a low SEE value (0.152), F value of 130.9 [F05 5, 48 = 2.40(tab)], and value (0.730) were obtained. On the other hand, the model generated with the molecules having the G–H charges has the statistical data as the value (0.699) with the optimized component number 5. The other statistical parameters were r2 value (0.918), F value (107.55), and the SEE value (0.166). We observed that the CoMSIA model generated using MMFF94 had good statistical values compared with the QSAR with G–H charges. We therefore selected the CoMFA and CoMSIA models generated with the MMFF94 charges–assigned molecules for further analysis.
CoMFA 3D-QSAR analysis
The generated CoMFA model with high r2 value (0.934), high value (0.795), and low SEE value (0.148) allowed us to choose the best QSAR model for further analysis. The steric and electrostatic field contributions of CoMFA model are almost equal (0.492 and 0.508, respectively), indicating the requirement of both these fields for MEK-1–inhibitor interactions. The contour maps of CoMFA show contribution for favorable and unfavorable interactions with the MEK-1 receptor in terms of steric (80% green and 20% yellow) and electrostatic (80% blue and 20% red) fields. The steric/bulky substituents on the core near green contour enhance the biological activity, whereas the yellow contours decrease the activity. The red contours show favorable electronegative region, and the blue contours indicate the region where electropositive groups are favorable.
All the molecules are classified based on the X and Y substitutions on the core part (shown in Figure 1). The most active molecule in the training set (molecule 21) is shown superimposed with CoMFA steric and electrostatic contour maps in Figure 2A,B, respectively. The substituted phenoxyphenyl and substituted phenyl groups are favorable near the green contour at Y substitution position on the core. The molecules with bulky or long-chain substitutions at the X position are disfavored for enhancing the activity. The X substitution region on the core maps with the blue contour of CoMFA, suggesting that the electropositive groups are required for contribution to the hydrophilicity at this region. From the electrostatic contour map of CoMFA, red contour showed the requirement of the electronegative substitutions at the Y position on the core. The molecule 21 has good activity and good correlation with the CoMFA contour map.
The dichloro-substituted phenoxy phenyl group provides the necessary steric and electronegative fields mapped with the contours at the Y position. The N-substituted 2-hydroxy-1-methylethyl group provides the necessary steric and electrostatic fields mapped with the contours at the X position of the molecule 21. The electronegative and electropositive groups observed in this region contribute to the good inhibitory activity of the molecule toward MEK-1 kinase. The other molecules such as 27, 38, 18, 15, and 7 having good activity in the Table S1 also have good correlation with the contour maps. From this table, we observed that the molecules with halogen substitutions on the phenyl or phenoxyphenyl groups have good inhibitory activity. The molecules 52, 53, 59, and 62 have neither bulky substitution nor electronegative atoms at Y position, thus making them less-active molecules. These molecules map with a small electropositive blue region at Y position that makes the molecules active but have relatively less inhibitory activity compared with the molecules with most electronegative substituents. These molecules comprise a bulky substitution such as cyclohexyl ring at the X position, which is observed in sterically disfavored region contributing to the less activity of the molecules.
CoMSIA 3D-QSAR analysis
The generated CoMSIA model (SEHD) with statistical parameters such as high r2 (0.932) and , (0.730), low SEE (0.152), and optimal number of components 5 allowed us to choose the QSAR model for further analysis. Based on the contribution of the contour maps, the electrostatic (0.336) and the hydrophobic (0.320) groups having high value show the importance of the electrostatic and hydrophobic nature of the substituents on the core of the molecule. The other descriptors steric (0.144) and hydrogen bond donor (0.195) also contribute to the activity but less compared with electrostatic and hydrophobic descriptors. The CoMSIA steric and electrostatic field contours obtained from database alignment employing S, E, H, and D fields are shown in Figure 3. These contour maps (Figure 3A,B) are more or less similar to the corresponding CoMFA contour maps (Figure 2A,B, respectively). The hydrophobic and hydrogen bond donor contour maps are shown in Figures 3C,D, respectively. The yellow hydrophobic contour near the Y position (Figure 3C) matches with the green steric contour that is shown in Figure 2A. This indicates that any bulky group with lipophilic character is preferred at this position. The molecules 21, 27, 38, 42, 15, and 7 have good inhibitory activity because of the presence of phenyl or phenoxyphenyl substituents, thus correlating with the CoMSIA hydrophobic contour map. These molecules with hydrophilic groups such as -OH and -NH that satisfy the white contour maps indicate the hydrophilic nature at the X position. From Figure 3D, the presence of cyan contour near the carboxyamidine NH at X position indicates the requirement of hydrogen bond donor substituent at this position to enhance the MEK-1 kinase inhibitory activity.
CoMFA versus CoMSIA
The predictive power of CoMFA and CoMSIA 3D-QSAR models was evaluated using the test set comprising 15 molecules. The relationship between the actual and predicted pIC50 values of the training set and test set compounds in CoMFA and CoMSIA is illustrated in Figures S2A,B, respectively. Based on the F-test value for CoMFA (172.6) and for CoMSIA (130.9) and other PLS statistical parameters, it is indicated that the CoMFA is slightly better than CoMSIA. In both models, the predictive values fall close to the actual pIC50 values, not deviating by more than 0.53 log units. For example, the molecule 51 with less activity in the test set has 0.53 residual value for CoMFA and is also observed as an outlier (residual value = 0.47) for CoMSIA. In summary, the differences between CoMFA and CoMSIA are minimal, and both models demonstrated good predictive ability.
The docking studies revealed that all inhibitors are docked into the allosteric binding site of the MEK-1 kinase crystal structure (PDB_ID: 3E8N) and are in similar orientation. The best docked conformations are superimposed with other crystal structures of MEK-1 kinase such as PDB_ID: 3MBL. The inhibitor conformations obtained from docking and crystal structure have low RMSD (root mean square deviation) (<1.7 Å), indicating a highly conserved binding mode and the reliability of gold software for docking studies. The ATP non-competitive inhibitor molecules bind to an allosteric site adjacent to the Mg-ATP binding region, the activation loop, and other surrounding protein residues through hydrogen bonding and hydrophobic interactions. The structural superposition of the six active molecules (RDEA119, 7, 14, 18, 21, 27, and 38) in their docked conformations is shown in Figure 4. The RDEA119 is cocrystallized allosteric inhibitor in the 3E8N crystal structure. Along with the Mg ion, two highly conserved water molecules are observed in few MEK-1 crystal structures (PDB_IDs: 3E8N, 1S9J, 3MBL, and 3DV3). Figure 5 indicates the docking of the most active molecule 21 from Table S1 and its overlay with RDEA119. The molecule 21 also forms several interactions with Asp208, Phe209, and Gly210 also known as the DFG motif that is shared across several families of protein kinases. The docking studies demonstrated that the QSAR contour maps corroborate with molecule 21 docking interaction with protein allosteric site. The allosteric site mainly has two grooves: one is large and hydrophobic and the other is hydrophilic near the activation loop. The molecule 21 has a bulky substitution dichlorophenoxyphenyl group at Y position on the core part, which is docked into the hydrophobic groove comprising amino acid residues Leu115, Leu118, Val127, Cys207, Met143, Phe129, Phe209, Leu141, and Val211. This substitution has hydrophobic interactions with the hydrophobic residues. The docking orientation also resembles the allosteric inhibitor in MEK-1 kinase crystal structure (PDB_ID: 3ORN), where the bulky 3,4-difluoro-2-[(2-fluoro-4-iodophenyl) amino phenyl group is in hydrophobic favorable region. The isothiazole core part is stabilized by the pocket residues comprising Ser212, Gly210, L215, Ile216, Val211, and Leu115. The X position on the core of molecule 21 is N-substituted 2-hydroxy-1-methylethyl group that has hydrophilic nature with the -OH and -NH groups surrounded by the residues Lys97, Asp208, Asp190, Asn78, Gly79, Met219, Val224, Ser222, and Arg234 during docking. The hydroxy group of molecule 21 forms hydrogen bond with the sidechain carbonyl of Asp190 (-OH....O=C, 2.67 Å). One of the conserved crystal structure water molecules adjacent to the Mg ion forms hydrogen bond with the hydroxy group of molecule 21 (Owater........HO-, 2.75 Å). This interaction explains the electrostatic contour, and the short chain length explains the bulky group disfavoring contour near the activation loop.
In this manuscript, we report the ligand-based 3D-QSAR analysis and molecular modeling of non-ATP competitive MEK-1 kinase allosteric inhibitors. The CoMFA and CoMSIA 3D-QSAR models developed using MMFF94 charges displayed good correlation and predictive abilities with statistically significant r2, , , SEE, F-test, and residual values. The predictive ability of the QSAR models was also validated on the external test set molecules. The results of CoMFA and CoMSIA models are quite similar and display that steric and electrostatic descriptors are important contributors to the recognition of MEK-1–inhibitor interactions. The molecular docking of inhibitors reveals the crucial residues for MEK-1–inhibitor interactions mediated through the DFG motif and two major pockets, one large and hydrophobic and the other hydrophilic in nature. The interactions identified by the CoMFA and CoMSIA 3D contour maps correlate well with the interactions identified between the non-ATP competitive inhibitors and the MEK-1 kinase allosteric binding site identified through the docking studies. The 3D-QSAR and molecular docking analysis corroborate each other, and these results will help to better interpret the structure–activity relationship of these MEK-1 inhibitors and provide valuable insights into rational drug design for further improvement in the biological activity of the non-ATP competitive MEK-1 kinase inhibitors.
The authors thank the University of Hyderabad for the Bioinformatics facility. KT thanks CSIR for providing research fellowship.