Availableonlineatwww.sciencedirect.com EUROPEAN JOURNAL OF ° Science Direct MEDICINAL CHEMISTRY ELSEVIEI European Journal of Medicinal Chemistry 42(2007)977-984 http://www.elsevier.com/locate/ejmech Structure-based 3D-QSAR studies on heteroarylpiperazine derivatives as 5-ht3 receptor antagonists Ya-Ju Zhou, Li-Ping Zhu, Yun Tang D,, De-Yong Ye, *3 Department of Medicinal Chemistry, School of Pharmacy, Fudan University, 138 Yixueyuan Road, Shanghai 200032, China School of Pharmacy, East China University of Science and Technology, 130 Meilong Road, Shanghai 200237, China Received 16 June 2006: received in revised form 13 December 2006: accepted 19 December 2006 Available online 13 January 2007 Abstract Structure-based 3D-QSAR studies were performed on a series of novel heteroarylpiperazine derivatives as 5-HT3 receptor antagonists with comparative molecular field analysis( CoMFA)and comparative molecular similarity indices analysis( CoMSIA)methods. The compounds were initially docked into the binding pocket of the homology model of 5-HT3 receptor using GOLD program. The docked conformations with the highest score were then extracted and used to build the 3D-QSAR models, with cross-validated r values 0.716 and 0. 762 for CoMFA and CoMSIA, respectively. The CoMFA and CoMSIA contour plots were also fitted into the 3D structural model of the receptor to identify the key interactions between them, which might be helpful for designing new potent 5-HT3 receptor antagonists. C 2007 Elsevier Masson SAS. All rights reserved Keywords: Structure-based 3D-QSAR: CoMFA: CoMSIA: 5-HT3 receptor 1. Introduction intrinsic channel activity of 5-HT3A. The subunit 5-HT3A is homo-oligomeric while 5-HT3B is heteromeric [2] 5-HT3 receptor is a prototypical member of the Cys-loop lectron microscope images of the purified 5-HT3 receptor ligand-gated ion channel (LGIC) superfamily, which also in- are available in the literature [3] but its three-dimensional (3D) cludes the glycine(Gly), type A Y-amino butyric acid(GA- structure has not yet been resolved at atomic level. However, BAA) and nicotinic acetylcholine(nACh)receptors. The ion the structure of Limnaea acetylcholine binding protein channel of 5-HT3 receptor is assembled with five subunits (AChBP) has been determined by X-ray crystallography re- and each subunit is thought to possess an N-terminal ligand cently, which is also a member of lGICs, and shares 20%o ho- binding domain, four transmembrane (TM) domains and mologous sequence with the extracellular domain of 5-HT a large intracellular loop connecting the third and fourth TM receptor [4]. Therefore, a homology model of the extracellular domains [1]. There are two main subtypes of 5-HT3 receptors: domain of human 5-HT3 receptor was built based on the crys- (1)5-HT3A, a neuronal receptor directly coupled to cation- tal structure of AChBP, and some known ligands were docked lective(Na*, K, Ca2, Mg+, and other ions)channels, into the binding site to validate the model [5]. During the pro- which has structural and functional similarity with the cess of our work, a similar study using a range of antagonists nACh, GABAA, Gly and other ligand-gated ion channels, was published, too [6] and(2)5-HT3B, a regulatory subunit able to modulate the ar study has long been used in elucidating mechanism of drug action and optimizing lead compound. When the 3D struc- Corresponding author. Tel: +86 21 64251052 ure of drug target is also available, structure-based drug design * Corresponding author. Tel: +86 21 54237559: fax: +86 21 64042268 methods can be combined with QSAR method, i.estructure- E-mailaddresses.ytang234@ecust.edu.cn(Y.Tang),dyye(@shmuedu based QSAR study, which could provide more information (D.-Y. Ye) for lead optimization. In this work, a new kind of 5-HT3 0223-5234/S- see front matter 2007 Elsevier Masson SAS. All rights reserved doi:10.1016/1 erech2006.12029
Original article Structure-based 3D-QSAR studies on heteroarylpiperazine derivatives as 5-HT3 receptor antagonists Ya-Ju Zhou a , Li-Ping Zhu a , Yun Tang b, *, De-Yong Ye a, ** a Department of Medicinal Chemistry, School of Pharmacy, Fudan University, 138 Yixueyuan Road, Shanghai 200032, China b School of Pharmacy, East China University of Science and Technology, 130 Meilong Road, Shanghai 200237, China Received 16 June 2006; received in revised form 13 December 2006; accepted 19 December 2006 Available online 13 January 2007 Abstract Structure-based 3D-QSAR studies were performed on a series of novel heteroarylpiperazine derivatives as 5-HT3 receptor antagonists with comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) methods. The compounds were initially docked into the binding pocket of the homology model of 5-HT3 receptor using GOLD program. The docked conformations with the highest score were then extracted and used to build the 3D-QSAR models, with cross-validated r2 cv values 0.716 and 0.762 for CoMFA and CoMSIA, respectively. The CoMFA and CoMSIA contour plots were also fitted into the 3D structural model of the receptor to identify the key interactions between them, which might be helpful for designing new potent 5-HT3 receptor antagonists. 2007 Elsevier Masson SAS. All rights reserved. Keywords: Structure-based 3D-QSAR; CoMFA; CoMSIA; 5-HT3 receptor 1. Introduction 5-HT3 receptor is a prototypical member of the Cys-loop ligand-gated ion channel (LGIC) superfamily, which also includes the glycine (Gly), type A g-amino butyric acid (GABAA) and nicotinic acetylcholine (nACh) receptors. The ion channel of 5-HT3 receptor is assembled with five subunits and each subunit is thought to possess an N-terminal ligand binding domain, four transmembrane (TM) domains and a large intracellular loop connecting the third and fourth TM domains [1]. There are two main subtypes of 5-HT3 receptors: (1) 5-HT3A, a neuronal receptor directly coupled to cationselective (Naþ, Kþ, Ca2þ, Mg2þ, and other ions) channels, which has structural and functional similarity with the nACh, GABAA, Gly and other ligand-gated ion channels, and (2) 5-HT3B, a regulatory subunit able to modulate the intrinsic channel activity of 5-HT3A. The subunit 5-HT3A is homo-oligomeric while 5-HT3B is heteromeric [2]. Electron microscope images of the purified 5-HT3 receptor are available in the literature [3] but its three-dimensional (3D) structure has not yet been resolved at atomic level. However, the structure of Limnaea acetylcholine binding protein (AChBP) has been determined by X-ray crystallography recently, which is also a member of LGICs, and shares 20% homologous sequence with the extracellular domain of 5-HT3 receptor [4]. Therefore, a homology model of the extracellular domain of human 5-HT3 receptor was built based on the crystal structure of AChBP, and some known ligands were docked into the binding site to validate the model [5]. During the process of our work, a similar study using a range of antagonists was published, too [6]. QSAR study has long been used in elucidating mechanism of drug action and optimizing lead compound. When the 3D structure of drug target is also available, structure-based drug design methods can be combined with QSAR method, i.e. structurebased QSAR study, which could provide more information for lead optimization. In this work, a new kind of 5-HT3 * Corresponding author. Tel.: þ86 21 64251052. ** Corresponding author. Tel.: þ86 21 54237559; fax: þ86 21 64042268. E-mail addresses: ytang234@ecust.edu.cn (Y. Tang), dyye@shmu.edu.cn (D.-Y. Ye). 0223-5234/$ - see front matter 2007 Elsevier Masson SAS. All rights reserved. doi:10.1016/j.ejmech.2006.12.029 European Journal of Medicinal Chemistry 42 (2007) 977e984 http://www.elsevier.com/locate/ejmech
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-984 antagonists, arylpiperazine derivatives containing quipazine on the molecule with Gasteiger-Huckel charge and Tripos quinoline nucleus with different substituents in 3, 4 and 4 po- force field. The other 34 compounds were constructed on the sitions(Scheme 1), were collected from the literature [7]. The basis of the structure of compound 11. All molecules were compounds were docked into the binding site of the previously set in their unprotonated state. The 3D structure of extracellu reported homology model of the receptor [5] at first, the dock- lar domain of human 5-HT3 receptor was previously built on ing conformations were then performed by 3D-QSAR studies to the base of the crystal structure of AChBP [4, 5]. It was re- understand the interaction between the receptor and ligands and ported that the aromatic groups of antagonists were supposed to optimize the lead compound with comparative molecular to intercalate between aromatic side-chains of the receptor field analysis( CoMFA)[8] and comparative molecular similar- (Trp178-Tyr229, Tyr138-Tyr148): while the basic centers ity indices analysis(CoMSIA)[9] methods might interact with Glu231 or Glu124 (ionic interaction) and/or Trp85 (cation-T interaction) of the receptor [6] 2. Materials and methods Therefore, the binding site of the 5-HT3 receptor was defined as residues within a radius of 16 A from Ca atom of Trp178 in 2.. Data set the binding pocket to ensure that most of the residues critical for ligand binding verified/revealed by previous experimental Totally 35 heteroarylpiperazines were collected from data were included. All molecules were docked into the bind Ref. [7]. They were assayed for their potential ability to ing pocket with program GOLD v2.2 [11-13] place[H] granisetron specifically bound to the 5-HT3 receptor The default settings of GOLD were used, and no Hipping in rat cortical membrane [7]. The values of the receptor binding was allowed affinities(Ki) were converted to their inverse logarithms. The 35 compounds were randomly divided into training set (28 23 CoMFA and CoMSIA molecules)and test set (7 molecules)in the ratio of 4: 1 The docked conformation with the highest score of each 2. 2. Molecular modeling molecule in the training set was superimposed onto a 3D grid box with 2 A grid. CoMFA fields were formed usin All molecular modeling and statistical analyses were per- sp carbon probe atom carrying +I charge to generate steric formed on an R14000 SGI Fuel workstation using SYBYL (Lennard-Jones potential) and electrostatic(Coulomb poten- 6.9 molecular modeling software package [10]. The confor- tial)fields at each grid point. The CoMFA fields were scaled mation of compound 11, with the highest affinity value, was by the CoMFA-standard method in SYBYL. A 30 kcal/mol constructed using sketch option in SYBYL. Systematic con- energy cutoff was applied. A distance dependent Gaussian formation search and energy minimization were performed type functional form was employed. The default value of 0.3 was used as the attenuation factor. Similarly, a data table was constructed from similarity indices calculated at the inter- sections of a regularly spaced lattice(2 A grid) in CoMSIA. 2. 4. PLS analysis and validation of QSAR models The CoMFa/Comsia fields combined with observed bio- 8 CH, CH logical activities(pki) were included in a molecular spread- COOCH CH3 sheet and the partial least square(PLS)methods were used to generate 3D-QSAR models [14]. To check statistical signif icance of the models, cross-validations analysis performed by the leave-one-out (Loo)[15] procedure was done to choose components(N)[16]. subsequently CHg CH=CH2CH3 used to derive the final QSAr models. All cross-validated PLS analyses were performed with a column filter value of 2.0. The optimal numbers of components were selected on GCH, CON the basis of the highest cross-validated correlation coefficient (r2 ) which is defined as follows ∑(Yp ctual are p CooC,Hs CH, mean values of the target property(pKi), respectively Scheme 1. Structures and actual pk; values of molecules used for 3D-QSAR CoMFA/CoMSIA results were graphically interpreted by field studies [7 contribution maps using the "STDEV*COEFF' field type
antagonists, arylpiperazine derivatives containing quipazine quinoline nucleus with different substituents in 3, 4 and 40 positions (Scheme 1), were collected from the literature [7]. The compounds were docked into the binding site of the previously reported homology model of the receptor [5] at first, the docking conformations were then performed by 3D-QSAR studies to understand the interaction between the receptor and ligands and to optimize the lead compound with comparative molecular field analysis (CoMFA) [8] and comparative molecular similarity indices analysis (CoMSIA) [9] methods. 2. Materials and methods 2.1. Data set Totally 35 heteroarylpiperazines were collected from Ref. [7]. They were assayed for their potential ability to displace [3 H] granisetron specifically bound to the 5-HT3 receptor in rat cortical membrane [7]. The values of the receptor binding affinities (Ki) were converted to their inverse logarithms. The 35 compounds were randomly divided into training set (28 molecules) and test set (7 molecules) in the ratio of 4:1. 2.2. Molecular modeling All molecular modeling and statistical analyses were performed on an R14000 SGI Fuel workstation using SYBYL v6.9 molecular modeling software package [10]. The conformation of compound 11, with the highest affinity value, was constructed using sketch option in SYBYL. Systematic conformation search and energy minimization were performed on the molecule with GasteigereHu¨ckel charge and Tripos force field. The other 34 compounds were constructed on the basis of the structure of compound 11. All molecules were set in their unprotonated state. The 3D structure of extracellular domain of human 5-HT3 receptor was previously built on the base of the crystal structure of AChBP [4,5]. It was reported that the aromatic groups of antagonists were supposed to intercalate between aromatic side-chains of the receptor (Trp178-Tyr229, Tyr138-Tyr148); while the basic centers might interact with Glu231 or Glu124 (ionic interaction), and/or Trp85 (cationep interaction) of the receptor [6]. Therefore, the binding site of the 5-HT3 receptor was defined as residues within a radius of 16 A˚ from Ca atom of Trp178 in the binding pocket to ensure that most of the residues critical for ligand binding verified/revealed by previous experimental data were included. All molecules were docked into the binding pocket with program GOLD v2.2 [11e13]. The default settings of GOLD were used, and no flipping was allowed. 2.3. CoMFA and CoMSIA The docked conformation with the highest score of each molecule in the training set was superimposed onto a 3D grid box with 2 A˚ grid. CoMFA fields were formed using sp3 carbon probe atom carrying þ1 charge to generate steric (LennardeJones potential) and electrostatic (Coulomb potential) fields at each grid point. The CoMFA fields were scaled by the CoMFA-standard method in SYBYL. A 30 kcal/mol energy cutoff was applied. A distance dependent Gaussian type functional form was employed. The default value of 0.3 was used as the attenuation factor. Similarly, a data table was constructed from similarity indices calculated at the intersections of a regularly spaced lattice (2 A˚ grid) in CoMSIA. 2.4. PLS analysis and validation of QSAR models The CoMFA/CoMSIA fields combined with observed biological activities (pKi) were included in a molecular spreadsheet and the partial least square (PLS) methods were used to generate 3D-QSAR models [14]. To check statistical significance of the models, cross-validations analysis performed by the leave-one-out (LOO) [15] procedure was done to choose optimum number of components (N ) [16], subsequently used to derive the final QSAR models. All cross-validated PLS analyses were performed with a column filter value of 2.0. The optimal numbers of components were selected on the basis of the highest cross-validated correlation coefficient (r2 cv), which is defined as follows: r 2 cv ¼ 1 P Ypredicted Yactual2 PðYactual YmeanÞ 2 where Ypredicted, Yactual, and Ymean are predicted, actual, and mean values of the target property (pKi), respectively. The CoMFA/CoMSIA results were graphically interpreted by field contribution maps using the ‘STDEV*COEFF’ field type. Scheme 1. Structures and actual pKi values of molecules used for 3D-QSAR studies [7]. 978 Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-98 To assess the predictive power of the 3D-QSAR models between the actual and the predicted activities of the test derived using training set, biological activities of the test set molecules. In addition, the r2y, Ired and number of compo- molecules were predicted. The predictive r(red) value is cal- nents, the conventional correlation coefficient r 2and its stan- culated as follows: dard error were also computed for each model pred=(SD-PRESS)/SD 3. Results and discussion where SD is the sum of squared deviations between the biolog The docked conformations of the molecules in the training ical activity of the test set and the mean activity of training set set are shown in Fig. 1a. The superposition showed that the li molecules, and PRESS is the sum of squared deviations gands fit the binding pocket consisting of critical residues. The PHE221 TYR229 ASN123 TRP178 GLU124 PHE221 TYR229 PHE221 TYR229 GLU231 GL231 TRP178) ASN123 ASN123 TRP178 TRP85 GLU124 TRP85 TYR68 TYR68 icated as a solid surface. (All hydrogen atoms were omitted. )(b)The docked conformations of molecules whose piperazine ring located between pers in- Fig. 1.(a) The docked conformations of the molecules in the training set together with the binding site of 5-HT3 receptor. The solvent-accessible surface yr229 of the receptor and compound 1l was showed as a representative.(c)The docked conformations of molecules whose C-4 substituent located between Trp85 and Tyr229 of the receptor and the representative was compound 15 whose nitrogen atom in the quinoline ring formed hydrogen bond with Asn123 of the receptor. (d) The docked conformations of compound 33 whose phenyl ring was located between Trp85 and Tyr229 of the receptor(the hydrogen bonds
To assess the predictive power of the 3D-QSAR models derived using training set, biological activities of the test set molecules were predicted. The predictive r 2 (r2 pred) value is calculated as follows: r 2 pred ¼ ðSD PRESSÞ=SD where SD is the sum of squared deviations between the biological activity of the test set and the mean activity of training set molecules, and PRESS is the sum of squared deviations between the actual and the predicted activities of the test set molecules. In addition, the r2 cv, r2 pred and number of components, the conventional correlation coefficient r 2 and its standard error were also computed for each model. 3. Results and discussion The docked conformations of the molecules in the training set are shown in Fig. 1a. The superposition showed that the ligands fit the binding pocket consisting of critical residues. The Fig. 1. (a) The docked conformations of the molecules in the training set together with the binding site of 5-HT3 receptor. The solvent-accessible surface is indicated as a solid surface. (All hydrogen atoms were omitted.) (b) The docked conformations of molecules whose piperazine ring located between Trp85 and Tyr229 of the receptor and compound 11 was showed as a representative. (c) The docked conformations of molecules whose C-4 substituent located between Trp85 and Tyr229 of the receptor and the representative was compound 15 whose nitrogen atom in the quinoline ring formed hydrogen bond with Asn123 of the receptor. (d) The docked conformations of compound 33 whose phenyl ring was located between Trp85 and Tyr229 of the receptor (the hydrogen bonds are shown in yellow lines). Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984 979
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-984 piperazine ring of most ligands located between Trp85 and Table 2 Tyr229 of the receptor(Fig. Ib) among which compounds Experimental pKi, predictive pKi and residual values by and CoMSIA 29 and 30 formed hydrogen bonds with Asn123 of the recep- Compound Actual pKi Predicted pKi tor. However, some other ligands seemed to have their C-4 CoMFA COMSIA substituent located between Trp85 and Tyr229 of the receptor (Fig. Ic). And most of their nitrogen atom in the quinoline ring aining set 8.68 8667 8.727 0013-0.047 (i.e. 15, 16, etc. )formed hydrogen bond with Asn123 of the 2 8.538 8 -0.018-0.022 receptor, while in compounds 24 and 25 the hydrogen bonds 3 8.408 8.423 0.032 0.017 were formed by the carbonyl group of their C-4 substituents 8.38 8.381 8.330 -0.001 0.005 Compounds 33 and 35 whose phenyl ring located between 7 8.2158.2330.050-0023 9.029 0.149 Trp85 and Tyr229 of the receptor had different conformations 8 937 9410 0.04 with others(Fig. Id). These observations seem to be different 9 9.112 from the pharmacophore models mentioned above[6). There-11 10 9.730 9.720 fore, they might reveal a novel interaction mode between 5- 10.10 0.003 9.368 9.351 0.079 HT3 receptor and antagonists. And the results of the docking 15 8.937 0.075-0.137 tudies and 3D-QSAR would offer constructive suggestions 16 to the further rectification of the receptor model. And it might 17 9.043 -0.003-0.100 be helpful for designing new potent 5-HT3 receptor 18 8.705 -0.045-0.033 20 9.083 9.083 0.087 onists 0.087 Considering the ionization of the compounds in physiolog.21 8.94 8.94 0.017 22 8.984 8.892 -0.064 ical pH, we also did the 3D-QSAR studies on the protonated23 9.133 9.209 0.099 molecules, but worse results were obtained. Therefore, unpro-25 7.72 7435 0.285 tonated state of each molecule was used in the study 6.949 139.287 0.077 0.103 3.. CoMFA 7932 9.259 9.289 The statistical details of Comfa are summarized in Table 1 9.77 9.866 The cross-validated value, r2, was 0.716, with an optimum 35 0.028 0017 number of six components. This analysis was used for final non- Test set cross-validated run, giving a correlation coefficient of 0.9926 8.574 8.783 -0.163 showing a good linear correlation between the observed and 12 925 9.270 9.312 predicted activities of the molecules in the training set. These 13 9.342 statistical indexes were reasonably high, indicating that the 7.968 CoMFA model might have a credible predictive ability. The ex- 32 -0.272 perimental pKi, predictive pKi and residual values by CoMFA 34 9.64 9.360 9.342 0.298 and CoMSIA are given in Table 2. The introduction of lipophi- licity log P values(calculated with XLOGP 2.0.[17))did not improve the CoMFa models. The electrostatic field descriptor explained 63.4% of the variance while the proportion of steric descriptor accounted for 36.6%. Therefore, the electrostatic field had greater influ- Summary of results from the CoMFA and CoMSIA analysis ence than the steric field. These electrostatic and steric fields n r2 SEEd fe are presented as contour plots in Fig. 2a and b, respectively. CoMFA OMSIA(S+E) OMSIA (S+E+H) 09 3. 2. COMSIA OMSIA (S+E+A) OMSIA (S+E+D) The CoMSia results are also summarized in Table 1. a oMSIA (S+E+H+D) 0.104 213.510 cross-validated value, r2y of 0.762 and a non-cross-validation oMSIA (S+E+H+A) 0.109 191.578 correlation coefficient r2 of 0.982 were obtained. The Fvalue oMSIA(S+E+D+A) 0.735 6 0.972 0.138 119.469 and standard error are 191. 578 and 0. 109, respectively. These CoMSIA (S+E+H+D+A) 0.76 0.982 0.109 191.578 data also indicated that a reliable comsia model was success Leave-one-out(LoO)cross-validation correlation coefficient. fully constructed. The hydrogen bond acceptor(A)field de- Optimum number of components criptor explained 24.9% of the variance and the steric (S) Non-cross-validation correlation coefficient d standard error of estimate descriptor only contributed 11.7%o, while the proportion of Ftest value electrostatic(E) and hydrophobic(H) descriptor accounted f S, E, H, D, and A represent the steric, electrostatic, hydrophobic, hydrogen for 32. 4% and 31.0%o, respectively. The hydrogen bond donor bond donor and acceptor property fields, respectively field was ignored for all the compounds in the training set
piperazine ring of most ligands located between Trp85 and Tyr229 of the receptor (Fig. 1b) among which compounds 29 and 30 formed hydrogen bonds with Asn123 of the receptor. However, some other ligands seemed to have their C-4 substituent located between Trp85 and Tyr229 of the receptor (Fig. 1c). And most of their nitrogen atom in the quinoline ring (i.e. 15, 16, etc.) formed hydrogen bond with Asn123 of the receptor, while in compounds 24 and 25 the hydrogen bonds were formed by the carbonyl group of their C-4 substituents. Compounds 33 and 35 whose phenyl ring located between Trp85 and Tyr229 of the receptor had different conformations with others (Fig. 1d). These observations seem to be different from the pharmacophore models mentioned above [6]. Therefore, they might reveal a novel interaction mode between 5- HT3 receptor and antagonists. And the results of the docking studies and 3D-QSAR would offer constructive suggestions to the further rectification of the receptor model. And it might be helpful for designing new potent 5-HT3 receptor antagonists. Considering the ionization of the compounds in physiological pH, we also did the 3D-QSAR studies on the protonated molecules, but worse results were obtained. Therefore, unprotonated state of each molecule was used in the study. 3.1. CoMFA The statistical details of CoMFA are summarized in Table 1. The cross-validated value, r2 cv, was 0.716, with an optimum number of six components. This analysis was used for final noncross-validated run, giving a correlation coefficient of 0.992 showing a good linear correlation between the observed and predicted activities of the molecules in the training set. These statistical indexes were reasonably high, indicating that the CoMFA model might have a credible predictive ability. The experimental pKi, predictive pKi and residual values by CoMFA and CoMSIA are given in Table 2. The introduction of lipophilicity log P values (calculated with XLOGP 2.0. [17]) did not improve the CoMFA models. The electrostatic field descriptor explained 63.4% of the variance while the proportion of steric descriptor accounted for 36.6%. Therefore, the electrostatic field had greater influence than the steric field. These electrostatic and steric fields are presented as contour plots in Fig. 2a and b, respectively. 3.2. CoMSIA The CoMSIA results are also summarized in Table 1. A cross-validated value, r2 cv of 0.762 and a non-cross-validation correlation coefficient r 2 of 0.982 were obtained. The F value and standard error are 191.578 and 0.109, respectively. These data also indicated that a reliable CoMSIA model was successfully constructed. The hydrogen bond acceptor (A) field descriptor explained 24.9% of the variance and the steric (S) descriptor only contributed 11.7%, while the proportion of electrostatic (E) and hydrophobic (H) descriptor accounted for 32.4% and 31.0%, respectively. The hydrogen bond donor field was ignored for all the compounds in the training set Table 1 Summary of results from the CoMFA and CoMSIA analysis r2 cv a Nb r 2c SEEd Fe CoMFA 0.716 6 0.992 0.073 433.222 CoMSIA (S þ E) 0.789 6 0.979 0.118 162.425 CoMSIA (S þ E þ H) 0.783 6 0.984 0.104 213.510 CoMSIA (S þ E þ A) 0.735 6 0.972 0.138 119.469 CoMSIA (S þ E þ D) 0.789 6 0.979 0.118 162.425 CoMSIA (S þ E þ H þ D) 0.783 6 0.984 0.104 213.510 CoMSIA (S þ E þ H þ A) 0.762 6 0.982 0.109 191.578 CoMSIA (S þ E þ D þ A) 0.735 6 0.972 0.138 119.469 CoMSIA (S þ E þ H þ D þ A)f 0.762 6 0.982 0.109 191.578 a Leave-one-out (LOO) cross-validation correlation coefficient. b Optimum number of components. c Non-cross-validation correlation coefficient. d Standard error of estimate. e F-test value. f S, E, H, D, and A represent the steric, electrostatic, hydrophobic, hydrogen bond donor and acceptor property fields, respectively. Table 2 Experimental pKi, predictive pKi and residual values by CoMFA and CoMSIA Compound Actual pKi Predicted pKi Residual values CoMFA CoMSIA CoMFA CoMSIA Training set 1 8.68 8.667 8.727 0.013 0.047 2 8.52 8.538 8.542 0.018 0.022 3 8.44 8.408 8.423 0.032 0.017 4 8.38 8.381 8.330 0.001 0.005 5 8.21 8.215 8.233 0.050 0.023 7 9.08 9.029 8.931 0.051 0.149 8 9.37 9.410 9.411 0.040 0.041 9 9.07 9.112 9.078 0.042 0.008 10 9.68 9.730 9.720 0.050 0.040 11 10.10 10.103 10.127 0.003 0.027 14 9.43 9.368 9.351 0.062 0.079 15 8.80 8.875 8.937 0.075 0.137 16 9.29 9.248 9.222 0.042 0.068 17 9.04 9.043 9.140 0.003 0.100 18 8.66 8.705 8.693 0.045 0.033 20 9.17 9.083 9.083 0.087 0.087 21 8.96 8.944 8.943 0.016 0.017 22 8.92 8.984 8.892 0.064 0.028 23 9.11 9.133 9.209 0.023 0.099 24 7.72 7.640 7.435 0.080 0.285 25 6.73 6.891 6.949 0.161 0.219 26 9.26 9.336 9.268 0.076 0.008 27 9.39 9.313 9.287 0.077 0.103 28 8.00 7.911 7.976 0.089 0.024 29 7.85 7.787 7.932 0.063 0.082 30 9.35 9.259 9.289 0.091 0.061 33 9.77 9.866 9.841 0.096 0.071 35 9.62 9.592 9.603 0.028 0.017 Test set 6 8.62 8.574 8.783 0.046 0.163 12 9.25 9.270 9.312 0.020 0.062 13 8.80 8.970 9.036 0.170 0.236 19 9.96 9.342 9.357 0.618 0.603 31 7.89 7.968 7.998 0.078 0.108 32 8.57 8.842 8.558 0.272 0.012 34 9.64 9.360 9.342 0.280 0.298 980 Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-984 (a) PHE221 (b) PHE22 TYR229 TYR22 GLU231 GLU231 TRP178 TRP17 ASN123 ASN123 GLU124 GLU124 TYR68 Fig. 2. Mapping the CoMFA contours in the active site of 5-HT3A receptor with compound 1l as an example. The ligand was shown in orange. The contour plots (STDEV*COEFF)of the CoMFA: (a) electrostatic fields; blue contours indicated regions where electropositive groups increase activity, whereas red contours whereas yellow contours indicated regions where bulky groups decrease activity (all hydrogen atoms of the receptor were omited y groups increase activity which did not have any hydrogen bond donor. Therefore, the was close to N-l'of the quipazine quinoline nucleus corre- electrostatic, hydrophobic and acceptor fields had greater ef- sponding to Trp85 of the receptor. Presence of red contours fluence than steric. All the fields are also shown as contour emphasized that electronegative group was desirable at this plots in Fig. 3a-d, respectively. The addition of lipophilicity position. The carbonyl group of compound 33 was around log P(calculated as mentioned above) to the set of indepen- the red region which could explain better activity. And com- dent variable did not improve the correlation either pound 24 exhibited low activity as its carbonyl group was close to the blue region and its alkyl substituent in C-3 posi- 3.3. Validation of 3D-OSAR models tion was near to the red region The steric contour map(Fig 2b)showed a major yellow re- The test set was applied to evaluate the predictive power of gion near the N-4 position of the quipazine quinoline nucleus the CoMFA and CoMSIA 3D-QSAR models. Fig 4 shows the corresponding to the Trp178 of the 5-HT3 receptor, indicated plots of actual versus predicted activity for both training set that any bulky substituent decreased activity. The methyl on and test set. In almost all cases of 3D-QSAR models, the pre- the terminal piperazine nitrogen appeared in the bigger yellow dicted values fell close to the observed pKi values, deviating region resulting in that compound 28 had weak activity. Ho y not more than 0.7 logarithmic units. Finally, CoMFA and ever, there were one major green region surrounding pipera- CoMSIA showed similar predictive power with respect to zine and two green regions nearing the C-3 and C-4 these seven compounds (r2 values for CoMFA and CoMSIA positions of the quipazine quinoline nucleus, respectively are 0.808 and 0.816, respectively) (Scheme 1), indicating that a bulky substituent was preferred to produce higher activity. The conformation of the substituent 3.4. Structure-based analysis of receptor-ligand in C-4 position may affect the activity. Compound 10 whose Interactions ethyl in C-3 position was close to the green region showed higher activity than compound 8 Compound 1l had the best The data in Table 1 show that the CoMFA electrostatic field activity as the conformation occupies most of the green descriptor explained 63. 4%o of the variance, while the steric regions descriptor explained the rest 36.6%o. These electrostatic and ig. 3 depicts the CoMsIa coefficient contour maps with steric fields are presented as contour plots in Fig. 2a and b, compound 1l displayed for visualization. As shown in Fig. 3a, the electrostatic fields were represented by white- As shown in Fig 2a, there was a major blue region near the and yellow-colored contours(white, positive charge favored; C-5 position of the quipazine quinoline nucleus, one of which yellow, electronegative group favored). There was a major was also located near Trp178, indicating that substitution of white regions in the contour map located near C-5 position electropositive group at this position would increase the activ- of the quipazine quinoline nucleus, between Trp178 and For interpretation of the references to colour in text the Tyr229 of the receptor. Compound 26 which oriented its car- reader is referred to the web version of this article There bonyl group into the yellow region located near the terminal are two red contours: one was near the N-1 atom and the other piperazine nitrogen and located its nitrogen in N-l position
which did not have any hydrogen bond donor. Therefore, the electrostatic, hydrophobic and acceptor fields had greater ef- fluence than steric. All the fields are also shown as contour plots in Fig. 3aed, respectively. The addition of lipophilicity log P (calculated as mentioned above) to the set of independent variable did not improve the correlation either. 3.3. Validation of 3D-QSAR models The test set was applied to evaluate the predictive power of the CoMFA and CoMSIA 3D-QSAR models. Fig. 4 shows the plots of actual versus predicted activity for both training set and test set. In almost all cases of 3D-QSAR models, the predicted values fell close to the observed pKi values, deviating by not more than 0.7 logarithmic units. Finally, CoMFA and CoMSIA showed similar predictive power with respect to these seven compounds (r 2 values for CoMFA and CoMSIA are 0.808 and 0.816, respectively). 3.4. Structure-based analysis of receptoreligand interactions The data in Table 1 show that the CoMFA electrostatic field descriptor explained 63.4% of the variance, while the steric descriptor explained the rest 36.6%. These electrostatic and steric fields are presented as contour plots in Fig. 2a and b, respectively. As shown in Fig. 2a, there was a major blue region near the C-50 position of the quipazine quinoline nucleus, one of which was also located near Trp178, indicating that substitution of electropositive group at this position would increase the activity. (For interpretation of the references to colour in text the reader is referred to the web version of this article.) There are two red contours: one was near the N-1 atom and the other was close to N-10 of the quipazine quinoline nucleus corresponding to Trp85 of the receptor. Presence of red contours emphasized that electronegative group was desirable at this position. The carbonyl group of compound 33 was around the red region which could explain better activity. And compound 24 exhibited low activity as its carbonyl group was close to the blue region and its alkyl substituent in C-3 position was near to the red region. The steric contour map (Fig. 2b) showed a major yellow region near the N-40 position of the quipazine quinoline nucleus corresponding to the Trp178 of the 5-HT3 receptor, indicated that any bulky substituent decreased activity. The methyl on the terminal piperazine nitrogen appeared in the bigger yellow region resulting in that compound 28 had weak activity. However, there were one major green region surrounding piperazine and two green regions nearing the C-3 and C-4 positions of the quipazine quinoline nucleus, respectively (Scheme 1), indicating that a bulky substituent was preferred to produce higher activity. The conformation of the substituent in C-4 position may affect the activity. Compound 10 whose ethyl in C-3 position was close to the green region showed higher activity than compound 8. Compound 11 had the best activity as the conformation occupies most of the green regions. Fig. 3 depicts the CoMSIA coefficient contour maps with compound 11 displayed for visualization. As shown in Fig. 3a, the electrostatic fields were represented by whiteand yellow-colored contours (white, positive charge favored; yellow, electronegative group favored). There was a major white regions in the contour map located near C-50 position of the quipazine quinoline nucleus, between Trp178 and Tyr229 of the receptor. Compound 26 which oriented its carbonyl group into the yellow region located near the terminal piperazine nitrogen and located its nitrogen in N-10 position Fig. 2. Mapping the CoMFA contours in the active site of 5-HT3A receptor with compound 11 as an example. The ligand was shown in orange. The contour plots (STDEV*COEFF) of the CoMFA: (a) electrostatic fields; blue contours indicated regions where electropositive groups increase activity, whereas red contours indicated regions where electronegative groups increase activity and (b) steric fields; green contours indicated regions where bulky groups increase activity, whereas yellow contours indicated regions where bulky groups decrease activity (all hydrogen atoms of the receptor were omitted). Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984 981
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-984 (d) Fig 3. Contour plot of the CoMSIA STDEV*COEFF for(a) electrostatic properties: White isopleths encompass regions where an increase of positive ould enhance affinity, whereas in yellow contoured areas more negative charges were favorable for binding properties.(b) Hydrophobic properties: Whit pleths encompass regions favorable for hydrophobic groups, whereas in yellow contoured areas more hydrophilic groups were favorable for binding propertie solvent-accessible surface was indicated as a solid surface.(c) Steric features: White isopleths enclose areas where steric bulk would enhance affinity. Yellow contours highlight areas which should be kept unoccupied. (d) Acceptor fields(white, favored; yellow, disfavored). near the bigger yellow region around the N-I position of the the white region should correspond to the hydrophobic resi quipazine quinoline showed better binding affinity than com- dues whereas the yellow ones should be close to the polar res- pound 5. In the mean time, it was to be explained that com- idues. Actually, the yellow regions one of which was near pound 21 had worse activity for its two nitrogen atom in Trp68 and the other around Phe221. Therefore, the position piperazine which was occupied by the white region when com- of Trp68 and Phe221 might be rectified pared to compound 10. The distribution of electrostatic con- The steric field contour plots are displayed in Fig 3c. It wa tours in this model was almost consistent with that of indicated that a large group in white regions would be benefi CoMFA(Fig 2a) cial to the binding affinity. Three major white regions were The hydrophobic contours are shown in Fig. 3b. The white found: one was located between the Trp85 and Tyr229 of hydrophobic contours indicate that hydrophobic substituents the receptor, occupying the C-3 position of the quipazine would be good for increasing the potency, while hydrophilic quinoline nucleus, and the other two were around the C-3 substituents are beneficial to the activity at the regions of yel- and C-4 positions of the quipazine quinoline nucleus between low contours. There was a large white area near N-1, C-5 and the Phe221 and Glu231 of the receptor. This could explain C-6 positions of the quipazine quinoline nucleus indicating why compound 29 showed lower affinity than compound 27 that any lipophilic group was preferred at this position. Com- whose phenyl ring in the C-4 substitution came into the white pounds 24 and 25 oriented their nitrogen atoms in C-4 position region. It was observed that bulky substituent in yellow re- of quipazine quinoline nucleus into the white contoured region gions, which was located near C-5 and C-6 positions of the resulting in low activity. Compound 5 showed lower activities quipazine quinoline nucleus and 3, 4, 5 positions of the pi- than compound 2 for its bigger phenyl substitute occupied the perazine ring, respectively, would decrease the affinity. The re- yellow contoured region, which was occupying the C-6 posi- sult was in accordance with the contours of the CoMSIA tion of the quipazine quinoline nucleus. As for the receptor, hydrophobic field. Therefore, compound 5 whose phenyl
near the bigger yellow region around the N-1 position of the quipazine quinoline showed better binding affinity than compound 5. In the mean time, it was to be explained that compound 21 had worse activity for its two nitrogen atom in piperazine which was occupied by the white region when compared to compound 10. The distribution of electrostatic contours in this model was almost consistent with that of CoMFA (Fig. 2a). The hydrophobic contours are shown in Fig. 3b. The white hydrophobic contours indicate that hydrophobic substituents would be good for increasing the potency, while hydrophilic substituents are beneficial to the activity at the regions of yellow contours. There was a large white area near N-1, C-50 and C-60 positions of the quipazine quinoline nucleus indicating that any lipophilic group was preferred at this position. Compounds 24 and 25 oriented their nitrogen atoms in C-4 position of quipazine quinoline nucleus into the white contoured region resulting in low activity. Compound 5 showed lower activities than compound 2 for its bigger phenyl substitute occupied the yellow contoured region, which was occupying the C-6 position of the quipazine quinoline nucleus. As for the receptor, the white region should correspond to the hydrophobic residues whereas the yellow ones should be close to the polar residues. Actually, the yellow regions one of which was near Trp68 and the other around Phe221. Therefore, the position of Trp68 and Phe221 might be rectified. The steric field contour plots are displayed in Fig. 3c. It was indicated that a large group in white regions would be benefi- cial to the binding affinity. Three major white regions were found: one was located between the Trp85 and Tyr229 of the receptor, occupying the C-30 position of the quipazine quinoline nucleus, and the other two were around the C-3 and C-4 positions of the quipazine quinoline nucleus between the Phe221 and Glu231 of the receptor. This could explain why compound 29 showed lower affinity than compound 27 whose phenyl ring in the C-4 substitution came into the white region. It was observed that bulky substituent in yellow regions, which was located near C-5 and C-6 positions of the quipazine quinoline nucleus and 30 , 40 , 50 positions of the piperazine ring, respectively, would decrease the affinity. The result was in accordance with the contours of the CoMSIA hydrophobic field. Therefore, compound 5 whose phenyl Fig. 3. Contour plot of the CoMSIA STDEV*COEFF for (a) electrostatic properties: White isopleths encompass regions where an increase of positive charge would enhance affinity, whereas in yellow contoured areas more negative charges were favorable for binding properties. (b) Hydrophobic properties: White isopleths encompass regions favorable for hydrophobic groups, whereas in yellow contoured areas more hydrophilic groups were favorable for binding properties. The solvent-accessible surface was indicated as a solid surface. (c) Steric features: White isopleths enclose areas where steric bulk would enhance affinity. Yellow contours highlight areas which should be kept unoccupied. (d) Acceptor fields (white, favored; yellow, disfavored). 982 Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-98 CoM FA revealed higher affinities than compound 29 whose carbonyl a group occupied the yellow region The results showed that a moderately bulky substituent at the C-4 position was preferred to produce higher binding affin- ities. Introduction of some proper alkyl substituents to the C-5 and C-6 positions of piperazine may improve the binding af- finities. The bicyclic derivatives showed better affinity than the corresponding tricyclic ones (i.e. compare compound 8 with 2). The set of compounds was worthy of further studie In this study, CoMFA and CoMSIA 3D-QSAR analyses were performed on the docked conformations of 28 heteroar- ylpiperazines as 5-HT3 receptor antagonists. Both models showed good prediction capabilities in terms of r2y and r2 values, while CoMFA model showed better predictive ability ISEE (standard error of estimate)=0.073) than CoMSIA one(SEE=0.109). The good correlation between experi mental and predicted bioactivities for seven compounds in CoMSIA est set further verified the reliability of the constructed QSAR models. The CoMFA model provided the most sig- nificant correlation of steric and electrostatic fields with the biological activities. The effects of the electrostatic, hydro- phobic, steric and hydrogen bond acceptor fields around he aligned molecules on their activities were clarified by an- alyzing the CoMSIA contour maps. The original intention of designing the heteroarylpiperazines was yze the activ- ity influence of compounds with different hydrophobic group [7]. In this study, we found some implications could be drawn to improve the activity and selectivity of heteroarylpi perazines as 5-HT3 receptor antagonists. For example, mod- erately bulky hydrophobic electropositive group substituent at the C-4 position and the introduction of some proper alkyl ◆ Training Set substituents to the C-5 and C-6 positions of piperazine a ' Test Set might be preferable to produce higher activity. Therefore, the results would be very helpful for further structural mod 8.599.51010.5 Actual pK Acknowledgments Fig. 4. Actual versus predicted pk, of training and test set molecules for(a) CoMFA and(b) CoMSIA 3D-QSAR models. We are grateful to Dr. Wei Li for his support and encouragement group of the pyrrolidone ring entered the yellow region had lower activity than compound 4 References The hydrogen bond acceptor contours are displayed in Fig. 3d. The existence of one white region near carbonyl [I G F. Lopreato, P. Banerjee, MoL Brain Res 118(2003)45-51 group implied that there might exist hydrogen bond donor at (2 ZH. traill, Cur. Med. Chem. -Central Nervous System Agents I the corresponding positions of the receptor's active site (Trp85, Tyr148 and Tyr229), consisted with our homology [3] F.G. Boess, I.L. Martin, Neuropharmacology 33(1994)275-317 [4] K. Brejc, WJ. Van Dijk, R V. Klaassen, M. Schuurmans, J. Van der Oost, model. This observation clearly indicated that hydrogen A B. Smit, T K. Sixma, Nature (2001)269-411 bond acceptor near the white contours would increase the [5] L.-P. Zhu, D -Y. Ye, Y. Tang, J. MoL Model. 13(2007)121-131 activity. A yellow contour near C-3 position of the quipazine 6] G. Maksay, Z. Bikadi, M.J. Simonyi, J. Recept. Signal Transduct Res. 23 quinoline nucleus indicated that molecules with hydrogen bond acceptor at this position would be less active. Therefore [7 A Cappelli, A Gallelli, J Med. Chem. 48(2005)3564-3575 [8] R D. Cramer Ill, D E. Patterson, J D. Bunce, J. Am. Chem. Soc. 110 compound 22 whose carbonyl group closer to the white region (1998)5959-5967
group of the pyrrolidone ring entered the yellow region had lower activity than compound 4. The hydrogen bond acceptor contours are displayed in Fig. 3d. The existence of one white region near carbonyl group implied that there might exist hydrogen bond donor at the corresponding positions of the receptor’s active site (Trp85, Tyr148 and Tyr229), consisted with our homology model. This observation clearly indicated that hydrogen bond acceptor near the white contours would increase the activity. A yellow contour near C-3 position of the quipazine quinoline nucleus indicated that molecules with hydrogen bond acceptor at this position would be less active. Therefore, compound 22 whose carbonyl group closer to the white region revealed higher affinities than compound 29 whose carbonyl group occupied the yellow region. The results showed that a moderately bulky substituent at the C-4 position was preferred to produce higher binding affinities. Introduction of some proper alkyl substituents to the C-50 and C-60 positions of piperazine may improve the binding af- finities. The bicyclic derivatives showed better affinity than the corresponding tricyclic ones (i.e. compare compound 8 with 2). The set of compounds was worthy of further studies. 4. Summary In this study, CoMFA and CoMSIA 3D-QSAR analyses were performed on the docked conformations of 28 heteroarylpiperazines as 5-HT3 receptor antagonists. Both models showed good prediction capabilities in terms of r2 cv and r 2 values, while CoMFA model showed better predictive ability [SEE (standard error of estimate) ¼ 0.073)] than CoMSIA one (SEE ¼ 0.109). The good correlation between experimental and predicted bioactivities for seven compounds in test set further verified the reliability of the constructed QSAR models. The CoMFA model provided the most significant correlation of steric and electrostatic fields with the biological activities. The effects of the electrostatic, hydrophobic, steric and hydrogen bond acceptor fields around the aligned molecules on their activities were clarified by analyzing the CoMSIA contour maps. The original intention of designing the heteroarylpiperazines was to analyze the activity influence of compounds with different hydrophobic group [7]. In this study, we found some implications could be drawn to improve the activity and selectivity of heteroarylpiperazines as 5-HT3 receptor antagonists. For example, moderately bulky hydrophobic electropositive group substituent at the C-4 position and the introduction of some proper alkyl substituents to the C-50 and C-60 positions of piperazine might be preferable to produce higher activity. Therefore, the results would be very helpful for further structural modification of heteroparylpiperazines. Acknowledgments We are grateful to Dr. Wei Li for his support and encouragement. References [1] G.F. Lopreato, P. Banerjee, Mol. Brain Res. 118 (2003) 45e51. [2] Z.H. Israili, Curr. Med. Chem. e Central Nervous System Agents 1 (2001) 171e199. [3] F.G. Boess, I.L. Martin, Neuropharmacology 33 (1994) 275e317. [4] K. Brejc, W.J. Van Dijk, R.V. Klaassen, M. Schuurmans, J. Van der Oost, A.B. Smit, T.K. Sixma, Nature (2001) 269e411. [5] L.-P. Zhu, D.-Y. Ye, Y. Tang, J. Mol. Model. 13 (2007) 121e131. [6] G. Maksay, Z. Bika´di, M.J. Simonyi, J. Recept. Signal Transduct. Res. 23 (2003) 255e270. [7] A. Cappelli, A. Gallelli, J. Med. Chem. 48 (2005) 3564e3575. [8] R.D. Cramer III, D.E. Patterson, J.D. Bunce, J. Am. Chem. Soc. 110 (1998) 5959e5967. Fig. 4. Actual versus predicted pKi of training and test set molecules for (a) CoMFA and (b) CoMSIA 3D-QSAR models. Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984 983
Y.. Zhou et al. European Journal of Medicinal Chemistry 42(2007)977-984 19G. Klebe, U. Abraham, T.J. Mietzner. J. Med. Chem. 37(1994) [13 G. Jones, P. Willett, R.C. Glen, J Mol Biol. 245(1995)43-53 4130-4146 [14]S. Wold, A Ruhe, H. Wold, wJ. Dunn Il, SIAM J. Sci. Stat Comput. 5 [10] SYBYL V6.9 Molecular Modeling Software, Tripos Associates, Inc. 1984)735-74 1669, South Hanley Road, Suite 303, St. Louis, Missouri, MO 63144- [15] R D. Cramer Ill,JD Bunce, D.E. Patterson, Quant. Struct -Act. Relat. 7 913,Usa. 988)18-25 [11] GOLD V2.2 CCDC Software Ltd., 12 Union Road, Cambridge CB2 1EZ, [16] B.L. Bush, R.B. Nachbar Jr. J. Comput. Aided Mol. Des 7(1993) [12] G. Jones, P. willett, R.C. Glen, A.R. Leach, R. Taylor, J Mol Biol. 267 [17]R. Wang, Y. Fu, L. Lai, J. Chem. Inf. Comput. Sci. 37(1997) (1997)727-748 615-621
[9] G. Klebe, U. Abraham, T.J. Mietzner, J. Med. Chem. 37 (1994) 4130e4146. [10] SYBYL v6.9 Molecular Modeling Software, Tripos Associates, Inc.: 1669, South Hanley Road, Suite 303, St. Louis, Missouri, MO 63144- 2913, USA. . [11] GOLD v2.2 CCDC Software Ltd., 12 Union Road, Cambridge CB2 1EZ, United Kingdom. http://www.ccdc.cam.ac.uk. [12] G. Jones, P. Willett, R.C. Glen, A.R. Leach, R. Taylor, J. Mol. Biol. 267 (1997) 727e748. [13] G. Jones, P. Willett, R.C. Glen, J. Mol. Biol. 245 (1995) 43e53. [14] S. Wold, A. Ruhe, H. Wold, W.J. Dunn III, SIAM J. Sci. Stat. Comput. 5 (1984) 735e743. [15] R.D. Cramer III, J.D. Bunce, D.E. Patterson, Quant. Struct.eAct. Relat. 7 (1988) 18e25. [16] B.L. Bush, R.B. Nachbar Jr., J. Comput. Aided Mol. Des 7 (1993) 587e619. [17] R. Wang, Y. Fu, L. Lai, J. Chem. Inf. Comput. Sci. 37 (1997) 615e621.. 984 Y.-J. Zhou et al. / European Journal of Medicinal Chemistry 42 (2007) 977e984
