Characterization of Binding Mode for Human Coagulation Factor XI (FXI) Inhibitors

Transcription

1 1212 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Jae Eun Cho et al. Characterization of Binding Mode for Human Coagulation Factor XI (FXI) Inhibitors Jae Eun Cho, Jun Tae Kim, Seo Hee Jung, and Nam Sook Kang * Graduate School of New Drug Discovery and Development, Chungnam National University, Daejeon , Korea * nskang@cnu.ac.kr Received January 14, 2012, Accepted January 28, 2013 The human coagulation factor XI (FXI) is a serine protease that plays a significant role in blocking of the blood coagulation cascade as an attractive antithrombotic target. Selective inhibition of FXIa (an activated form of factor XI) disrupts the intrinsic coagulation pathway without affecting the extrinsic pathway or other coagulation factors such as FXa, FIIa, FVIIa. Furthermore, targeting the FXIa might significantly reduce the bleeding side effects and improve the safety index. This paper reports on a docking-based three dimensional quantitative structure activity relationship (3D-QSAR) study of the potent FXIa inhibitors, the chloro-phenyl tetrazole scaffold series, using comparative molecular field analysis (CoMFA) and comparative molecular similarity analysis (CoMSIA) methods. Due to the characterization of FXIa binding site, we classified the alignment of the known FXIa inhibitors into two groups according to the docked pose: S1-S2-S4 and S1-S1'- S2'. Consequently, highly predictive 3D-QSAR models of our result will provide insight for designing new potent FXIa inhibitors. Key Words : Human coagulation factor XIa (FXIa), Three dimensional quantitative structure activity relationship (3D QSAR), Comparative molecular field analysis (CoMFA), Comparative molecular similarity indices analysis (CoMSIA), Docking Introduction Thromboembolic diseases such as acute myocardial infarction, ischemic stroke, heart attack, pulmonary embolism and deep vein thrombosis are the major causes of morbidity and mortality in developed countries. 1,2 To treat thromboembolic disease, heparin and warfarin are commonly used as anticoagulants. However, several researchers have reported that the current medications use non-specific inhibitors that cause a high risk of bleeding side effects by interfering with the activities of other blood clotting factors. 3 Many different coagulation enzymes, such as thrombin, factor Xa and factor VIIa, have been intensely studied because of the key roles they play in the blood coagulation cascade. Nevertheless, the major safety issues of anticoagulants, such as bleeding problems, are still associated with those targets. 4,5 Thus, there is an urgent need to develop new anticoagulant agents with a novel target. Human coagulation factor XIa (FXIa), a key enzyme that is specifically involved in the amplification phase of the intrinsic pathway, is a promising target for the treatment of thromboembolic disease. A zymogen form of FXI is converted to an active form (FXIa) by factor XIIa which cleaves FXI into a heavy chain and a light chain. Inhibition of the cleaved light chain (FXIa) reduces the size of the clot and may lead to fewer bleeding side effects than when other coagulation factor isozymes are targeted. 6,7 In recent years, several studies of human factor XI have been reported on FXIa-specific inhibitors in order to replace the existing therapies. Previous studies have found that the basic P1group, such as benzamidine or guanidine, interacts with Asp189 in the S1 pocket of FXIa. The 3D-QSAR study was published using the a-ketotiazole arginine analogue inhibitors in However, those basic groups were reported to have poor solubility and low bioavailability. Therefore, recent studies developed more promising compounds with nonbasic P1, the chloro-phenyl tetrazole group, as a novel anticoagulant oral drug In this study, we collected various FXIa inhibitors from several publications issued by Bristol-Myers Squibb Company (US Patent Application Publication No. US 2012/ A1, US 2010/ A1 and US 2010/ A ) and Merck & Co. 12 Based on the known FXIa crystalized-complex conformations of small molecules (pdb id: 3SOS, 3SOR), 12 peptide-mimetic molecules (pdb id: 1ZOM), 13 and substrate (pdb id: 1XX9), 14 we analyzed the possible binding mode of FXIa inhibitors. Two types of alignment were induced for the 3D-QSAR study, specifically CoMFA 15,16 and CoMSIA. 17 In addition, by the automated docking, we evaluated our binding mode and found that it enhances the validity and reliability of our models. Materials and Methods Data Sets and Alignment. The data sets of FXIa compounds were taken from the published patents and research papers reported by Corte et al. in 2012, 9 Pinto et al. in ,11 and Fradera et al. in for this study. The 42 compounds shown in Tables 1 and 3 were divided into a training set (32 compounds) and a test set (10 compounds). We generated the 3D-QSAR models using the training set and the predictive power of our model was validated by the randomly

2 Binding Mode for FXIa Bull. Korean Chem. Soc. 2013, Vol. 34, No Table 1. Structures of choloro-phenyltetrazole P1 core compounds used in training set and test set Compound Structure pic 50 Compound Structure pic

3 1214 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Jae Eun Cho et al. Table 1. Continued Compound Structure pic 50 Compound Structure pic a a a a a a a a a a 7.49 Note: a Test set molecules. selected test set compounds. Different types of biological activity values (IC 50, Ki) were reported in each publication. In order to standardize the biological activity units, we applied the Cheng-Prusoff equation which states IC 50 = 2K i. 18 The dependent variables that were used in the 3D-QSAR analysis were values converted

4 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Binding Mode for FXIa Table 2. CoMFA and CoMSIA models for S1-S1'-S2' typed conformation Compound pic a 34a 35a 36a 37a 38a 39a 40a 41a 42a CoMFA CoMSIA predicted residuals predicted residuals Table 3. Summary of CoMFA and CoMSIA models for S1-S1'-S2' typed conformation Parameters 2a q Nb r2c F-test values SEEd Probe of r2 = 0 r2prede Steric Electrostatic Hydrophobic Donor Acceptor CoMFA CoMSIA Note: aq2 = cross-validated correlation coefficient, bn = number of components, cr2 = conventional correlation coefficient, dsee = standard error of estimate, er2pred = predictive correlation coefficient. Note: atest set molecules. into the corresponding pic50 (-log IC50) and the activity values of 32 training set compounds covered 4 log units (pic50 = 4-8). The biological activity value range of the test set is similar to that of the training set. According to previously published research,12,17,18 the orientation of the FXIa crystal ligand can be categorized by two directions in the protein binding site: S1-S2-S4 and S1S1'-S2' (Figure 1). Thus, we conducted two different cate- Figure 1. Two different types of data set alignments. The alignment of S1-S1'-S2' type orientation is shown in (a) and (b) depicted the S1-S2-S4 type alignment.

5 1216 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Jae Eun Cho et al. Figure 3. CoMFA and CoMSIA field contour map for S1-S1'-S2' typed conformation. Green contour favors bulky group and yellow region disfavors bulky group depicted in (a). For electrostatic map shown in (b), the red region is a favored negative-charged group and blue region is a favored positive-charged group. CoMSIA model (c) and (d) describes hydrogen bond donor property (cyan: favored, purple: disfavored) and hydrogen bond acceptor property (magenta: favored, orange: disfavored). Compound 4 was superimposed as the reference molecule in this map. Figure 2. Correlation plot between observed activity values (pic50) and calculated activity values for S1-S1'-S2' typed conformation; CoMFA (a) and CoMSIA (b). The blue squares represent training set and red circles signify test set compounds. gories of dataset alignment in the active site of FXIa crystal structure (pdb id: 1ZOM) which was taken from the Protein Data Bank (PDB, The aligned poses of the training set compounds were directly extracted for a 3D-QSAR analysis. Generation of CoMFA and CoMSIA Models. The CoMFA study described in this research was performed using the SYBYL 2.0 molecular modeling software (Tripos Inc.). For the CoMFA analysis, the aligned training set molecules were placed into a 3D grid box to calculate both the steric (Lennard-Jones) and electrostatic (Coulombic) field energies using a sp3 carbon atom and a +1 net charge atom. The Tripos standard force field was applied with the distance dependent dielectric constant field value. The steric and electrostatic values were set to cut off to ± 30 kcal mol 1. The truncation value for column filtering was set to 2.0 kcal mol 1 to reduce the noise and speed up the analysis. In the CoMSIA model, five physicochemical properties steric, electrostatic, hydrophobic, hydrogen bond donor and hydrogen bond acceptor potential fields were calculated at each lattice of grid space (2.0 Å). To calculate the five different types of field, radius 1.0 Å and +1 charge with hydrophobicity of +1 and hydrogen bond donor and hydrogen bond acceptor properties of +1 probe atoms were used. Attenuation factor α of 0.3 for Gaussian-type distance dependence was used. PLS Regression Analysis. The partial least squares (PLS), a multiple regression analysis method was carried out using SYBYL 2.0 to evaluate the predictive power of our model using linear correlation between the inhibitory activity values and CoMFA/CoMSIA descriptors. In our research, the PLS method was performed in two steps. Initially, the cross validation method leave-one-out (LOO) was applied to determine how many components should be used to describe the complexity of our model. For the scaling variable, the CoMFA standard value was selected for both the CoMFA and CoMSIA models. Second, the non-cross validated conventional analysis resulted in the optimum number of components and the lowest standard error of prediction was considered for further investigation. Molecular Docking Study. To investigate the correlation between the formerly conducted 3D-QSAR models and the calculated docking mode, we carried out the rigid docking

6 Binding Mode for FXIa Bull. Korean Chem. Soc. 2013, Vol. 34, No method implemented in Discovery Studio 3.1 (Accelrys, Inc.) which adopts binding site features (hotspots). Initially the crystal structure of the FXIa protein (1XX9) was retrieved from Protein Data Bank. The Clean method for protein in Discovery Studio 3.1 was applied to correct minor problems of the imported structure, such as missing side chains correction, hydrogen representation correction and crystallographic disorder deletion. The CHARMM force field was assigned to the clean 1XX9 and the binding sphere was defined as volume using advanced defines and edit binding site module, provided in Discovery Studio 3.1, by using the information of the known substrate structure of 1XX9. The previously prepared ligand data sets were typed with the Merck Molecular Force Field (MMFF) force field, which provides an accurate calculation for small organic molecules. 19 Result and Discussion Four distinct CoMFA and CoMSIA models were created using two different alignment types of training sets. The statistical parameters of both the CoMFA and CoMSIA models and the biological predictions of the training set molecules are shown in Tables 2 and 4. For the S1-S1'-S2' binding type training set, PLS analysis of CoMFA and CoMSIA models showed a good cross-validated coefficient of q 2, 0.51 and 0.46 with 6 components, respectively. The non-cross validated r 2 value for both models was 0.99 for CoMFA and 0.96 for CoMSIA. F-test values of for CoMFA and for CoMSIA were calculated. Estimated standard errors of 0.16 and 0.15 were obtained for the CoMFA and CoMSIA models, respectively. The predictive correlation coefficient (r 2 pred) for the test set was obtained for the CoMFA and CoMSIA models, 0.93 and 0.83, respectively. The steric fields explain 44% of the variance while the electrostatic descriptors explain 56% for the CoMFA model. For the model CoMSIA, the contribution of steric and electrostatic fields were 9% and 23%, respectively, while the hydrophobic, hydrogen bond donor and acceptor contributions were 11%, 25% and 26%, respectively. On the other hand, the model for the S1-S2-S4 binding type training set resulted in poor prediction scores for both the CoMFA and CoMSIA models. We found the recently released crystalized ligands (compounds 1-3 and 28-30) exactly has the S1-S1'-S2' binding type. Therefore, we ignored the six compounds (compounds 1-3 and 28-30) which had a single-bonded linker moiety in the S1-S2-S4 type model as outliers. Consequently, the removal of the six single-bonded linker compounds (compound 1-3 and 28-30) led to dramatic improvement in predictive power of the S1- S2-S4 type model. The cross-validated coefficient of the q 2 value resulted in 0.51 with 6 components for the CoMFA model. For the CoMSIA model using 26 compounds of the S1-S2-S4 type training set, a cross-validated coefficient q 2 of 0.49 with 4 components was obtained. F-values of for CoMFA and for CoMSIA were calculated. Estimated standard errors of 0.11 and 0.23 were obtained for Table 4. CoMFA and CoMSIA model of S1-S2-S4 conformation CoMFA CoMSIA Compound pic 50 predicted residuals predicted residuals a a a a a a a a a a Note: a Test set compounds. the CoMFA and CoMSIA models, respectively. The noncross validated r 2 value for CoMFA and CoMSIA models were 0.99 and 0.97, respectively. The contribution of the steric and electrostatic fields for the CoMFA model was 51% and 49%, respectively. For the model CoMSIA, the contribution of the steric and electrostatic fields were 14% and 18%, respectively, while the hydrophobic, hydrogen bond donor and acceptor contributions were 22%, 25% and 21%, respectively.

7 1218 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Jae Eun Cho et al. Table 5. Summary of CoMFA and CoMSIA models for S1-S2-S4 conformation Parameters CoMFA CoMSIA q 2a N b 6 5 r 2c F-test values SEE d Probe of r 2 = r 2 pred e Steric Electrostatic Hydrophobic 0.20 Donor 0.23 Acceptor 0.22 Note: a q 2 = cross-validated correlation coefficient, b N = number of components, c r 2 = conventional correlation coefficient, d SEE = standard error of estimate, e r 2 pred = predictive correlation coefficient. However, a poor predictive correlation coefficient (r 2 pred) for the test set was obtained for the CoMFA and CoMSIA models, 0.08 and 0.36, respectively, as shown in Figure 4. Three particular compounds (33, 35 and 42) had notably low prediction results among the 10 test set compounds. Upon removing those three compounds (33, 35 and 42), the predictive correlation coefficient (r 2 pred) had significantly high values for the CoMFA and CoMSIA models, and 0.967, respectively. To sum up, our 3D-QSAR result shows that the compounds containing single-bonded linker (compounds 1-3 and 28-30) obtained considerably accurate output with the S1-S1'-S2' binding mode while the S1-S2-S4 type binding mode provided poor output. Therefore we concluded that singlebonded linker type compounds prefer to form the S1-S1'-S2' binding mode which was also proved by the known crystal ligand structures (compound 1and 2) that follows the S1-S1'- S2' binding mode. In addition, compounds 33 and 42 resulted in very poor prediction outputs with the S1-S2-S4 type alignment, whereas the output of the S1-S1'-S2' type alignment had better predictive power for both compounds. When the bulky moiety comes at the P2 site as shown in compounds 33 and 42, they prefer to form the S1-S1'-S2' binding mode because the S1' binding pocket is wider than the S2 pocket. Compound 35 also preferred the S1-S1'-S2' binding mode according to our 3D-QSAR result. The benzyl moiety attached to the nitrogen of the liker amide group, directly interacted with Ser195 of the catalytic region which does not favor the bulky moiety. Thus, this N-benzylacetamide linker compounds would prefer to follow the S1- S1'-S2' binding mode. Contour Map of CoMFA and CoMSIA Models. The numeric results of the 3D-QSAR study were illustrated with contour maps using the STDEV*COEFF field type in order to help understanding of structure-activity relationship. The scale for the contour maps for both the CoMFA and CoMSIA models was generated with 80% (favored) and Figure 4. Correlation plot between observed activity values (pic 50) and calculated activity values for S1-S2-S4 typed conformation; CoMFA (a) and CoMSIA (b). The blue squares represent training set and red circles signify test set compounds. 20% (disfavored). The contour maps of the CoMFA model showed favorable region for the steric interaction, which are indicated with the green contours at the wide pockets of the S2' and S4 area shown in Figure 3(a) and 5(a). Electrostatic interactions represented by red and blue colored contours signify that the increase in positive charge is favored in the blue regions while the increase in negative charge is favored in red regions (Figure 3(b) and Figure 5(b)). For the CoMSIA contour maps described in Figure 3(c), 3(d), 5(c) and 5(d), the hydrogen bond donor and hydrogen bond acceptor properties are highlighted in cyan and magenta, respectively. The cyan contour in the S4 pocket indicates that the hydrogen bond donor adjacent to Glu98 could significantly improve the biological activities. Molecular Docking Result. A further study of molecular

8 Binding Mode for FXIa Bull. Korean Chem. Soc. 2013, Vol. 34, No S1'-S2', S1-S2-S4) of binding modes are clearly depicted in Figure 6. For the S1-S1'-S2' orientation of the binding mode, the most active compound (compound 4) and the most inactive compound (compound 30) are described in Figure 6(a) and 6(b) as the reference molecules. The resulting binding mode indicated that a positive charged bulky group toward the S2' region increased activity by forming hydrogen boning with Tyr143; however, the negatively charged group in the S2 site is disfavored as illustrated in the steric and electrostatic map of CoMFA. The S1-S2-S4 type docking mode also validated our CoMFA and CoMSIA models. A bulky group in the S4 site is a crucial factor for the biological reactivity matters. Specifically, the hydrogen bond donor group should be placed adjacent to the Glu98 in order to form hydrogen bonding tightly. Conclusion Figure 5. CoMFA and CoMSIA field contour map for S1-S2-S4 typed conformation. Green contour favors bulky group and yellow region disfavors bulky group depicted in (a). For electrostatic map shown in (b), the red region is favored negative charged group and blue region is favored positive charged group. CoMSIA model (c) and (d) describes hydrogen bond donor property (cyan: favored, purple: disfavored) and hydrogen bond acceptor property (magenta: favored, orange: disfavored). Compound 4 was superimposed as the reference molecule in this map. In this research, docking based 3D-QSAR studies derived from two possible alignments were carried out successfully. Both of our 3D-QSAR models were highly accurate and predictive: CoMFA (q2, 0.51; r2, 0.99) CoMSIA (q2, 0.46; r2, 0.96) for S1-S1 -S2 type alignment and CoMFA (q2, 0.51; r2, 0.99) CoMSIA (q2, 0.49; r2, 0.97) for and S1-S2-S4 type. By analyzing the accurate and distinct sets of the 3DQSAR models, we concluded that the compounds containing single bonded linker (compounds 1-3 and 28-30), the compounds containing a bulky P2 moiety (compounds 33 and 42) and the compounds containing bulky moiety attached to the nitrogen of the linker amide group (compound 35) prefer to form the S1-S1'-S2' type binding mode. Additionally our docking study also demonstrated that the two types of binding modes can form and the docking results support the correlation between compound structure and activity. Thus, the information obtained from this study will provide the tools to predict the biological activity of related compounds and guild for further structural modification as well as design of new selective FXIa inhibitors. Acknowledgments. This study was financially supported by research fund of Chungnam National University in References Figure 6. The docking mode of the most active compound (compound 4: yellow) and inactive compound (compound 30: purple). S1-S2-S4 type orientation is shown (a) and (b) whereas (c) and (d) represent the S1-S1'-S2' orientation. The blue dotted line indicates hydrogen bonding. docking was conducted with rigid docking method and considerable results were obtained. Two distinct types (S1-1. Mann, K. G.; Butenas, S.; Brummel, K. Arterioscler Thromb Vasc Biol. 2003, 45, Jenny, N. S.; Mann, K. G. In Thrombosis and Hemorrhage, Loscalzo, J., Schafer, A. I., Ed.; 1998, p Hirsh, J.; Dalen, J.; Anderson, D. R.; Poller, L.; Bussey, H.; Ansellm, J.; Deykin, D. Chest. 2001, 119, 8S. 4. Hirsh, J.; O Donnell, M.; Weitz, J. I. Blood. 2005, 105, Weitz, J. I. Circulation 2004, 110, I Gailani, D.; Smith, S. B. J. Thromb Haemost 2009, 7, Schumacher, W.; Luettgen, J.; Quan, M.; Seiffert, D. Arterioscler Thromb Vasc Biol. 2010, 30, Mercado, J.; Gómez, J.; Vivas-Reyes, R. New J. Chem. 2011, 35, Corte, J. R.; Fang, T.; Decicco, C. P.; Pinto, D. J. P.; Rossi, K. A.; Hu, Z.; Jeon, Y.; Quan, M. L.; Smallheer, J. M.; Wang, Y.; Yang,

9 1220 Bull. Korean Chem. Soc. 2013, Vol. 34, No. 4 Jae Eun Cho et al. W. United States Patent US 2010/ A1, Pinto, D. J. P. United States Patent US 2010/ A1, Pinto, D. J. P.; Quan, M. L.; Smith, L. M.; Orwat, M. J.; Gilligan, P. J. United States Patent US 2010/ A1, Fradera, X.; Kazemier, B.; Carswell, E.; Cooke, A.; Oubrie, A.; Hamilton, W.; Dempster, D.; Krapp, S.; Nagelc, S.; Jeste, A. Acta Cryst. 2012, F68, Lin, J.; Deng. H.; Jin, L.; Pandey, P.; Quinn, J.; Cantin, S.; Rynkiewicz, M. J.; Gorga, J. C.; Bibbins, F.; Celatka, C. A.; Nagafuji, P.; Bannister, T. D.; Meyers, H. V.; Babine, R. E.; Hayward, N. J.; Weaver, D.; Benjamin, H.; Stassen, F.; Abdel-Meguid, S. S.; Strickler, J. E. J. Med. Chem. 2006, 49, Jin. L.; Pandey, P.; Babine, R. E.; Gorga, J. C.; Seidl, K. J.; Gelfand, E.; Weaver, D. T.; Abdel-Meguid, S. S.; Strickler, J. E. J. Biol. Chem. 2004, 250, Cramer, R. D.; Patterson, D. E.; Bunce, J. D. J. Am. Chem. Soc. 1988, 110, Cramer, R. D.; Bunce, J. D.; Patterson, D. E. Quant. Struct. Act. Relat. 1988, 7, Klebe, G.; Abraham, U.; Mietzner, T. J. Med. Chem. 1994, 37, Cheng, Y.; Prusoff, W. H. Biochem. Pharmacol. 1973, 22, Halgren, T. A. J. Comput. Chem. 1995, 17, 490.