Availableonlineatwww.sciencedirect.com BCIENCEODDIRECTO Bioorganic edicinal Chemistry ELSEVIER Bioorganic Medicinal Chemistry 14(2006)601-610 Molecular modeling and 3D-QSAR studies of indolomorphinan derivatives as kappa opioid antagonists Wei li. a yun ta You-Li Zheng and Zhui-Bai Qiu Department of Medicinal Chemistry, School of Pharmacy, Fudan University, 138 Yixueyuan Road, Shanghai 200032, China sChool of Pharmacy, East China University of Science Technology, 130 Meilong Road, Shanghai 200237, China Received 6 July 2005: revised 19 August 2005: accepted 20 August 2005 Available online 3 October 2005 Abstract-Molecular modeling and 3D-Qsar studies were performed on 31 indolomorphinan derivatives to evaluate their antag- onistic behaviors on k opioid receptor and provide information for further modification of this kind of compounds. Best predictions were obtained with CoMFA standard model =0.693, N=4, r=0.900)and CoMSIA combined model (q=0.617, N=4, r=0.607 for CoMFA and r=0.701 for CoMSIA. In addition, the 3D structure of human k opioid receptor was constructed based on the crystal structure of bovine rhodopsin, and the ComsIa contour plots were then mapped into the structural model of K opioid receptor-GNTI complex to identify key residues, which might account for K antagonist potency and selectivity. The roles of non- conserved Glu297 and conserved Lys 227 of human k opioid receptor were then discussed c 2005 Elsevier Ltd. All rights reserved 1. Introduction the 5-substitutes of naltrindole, a potent 8 selective antagonist, led to the discovery of a more potent K selec There are three well-adopted subtypes of opioid recep- tive antagonist, 5-guanidine naltrindole(GNTI, 2). 1 tors:A, K, and 8. However, to date the precise role There were also some other types of k selective antago- of k opioid receptor has not been well established yet nists such as JDTic, 3(3)and KAA-1(4) 4 recently It appears that K opioid receptor exerts its physiological reported by Carroll and co-workers, which also owned roles by participating in pain process and regulating im- a second basic group(Fig. I mune systems.4 where selective K opioid antagonists could provide powerful tools to investigate detailed In order to elucidate the mechanism of action of known interactions between the receptor and ligand. Mean- kappa antagonists and design new kappa selective while, K selective antagonists also showed some clinical antagonists, molecular modeling and 3D-QSAR studi were conducted here. At first, 39 indolomorphinan derivatives, potent kappa selective antagonists by add- Nor-BNI (norbinaltorphimine, 1)is the first highly ing a basic or neutral group in the 5-position of naltrin- selective x opioid antagonist reported by portoghese dole, were collected from the literature. They were then et al, who believed that it was the second basic group divided into two groups: 31 compounds as training set that conferred K potency and selectivity of the com- and the other eight ones as test set. The training set pound. The hypothesis was confirmed in their subse- was used to build 3D-QSAR models with CoMFA quent work by structure simplification of nor-BNI.(comparative molecular field analysis) and CoMSIA The basic group was further identified interacting with (comparative molecular similarity indices analysis) Glu297 of k opioid receptor by site-directed mutagene- methods, while the test set was used to validate the sis. Moreover, the attachment of a basic group to 3D-QSAR models further. Meanwhile, due to unavail- ability of experimental structure, the 3D structure of kappa opioid receptor was constructed based on the Opioid receptor; Indolomorphinan derivatives: 3D rystal structure of bovine rhodopsin. The interaction lecular Modeling: CoMFA; CoMSIA mode of GNTI with kappa receptor was hence obtained ding authors.Tel:+862154237595:fax:+862154237264 Finally, the contour plots of CoMfA were mapped into tel. +86 21 54237419 (YT); e-mail addresses the binding site of kappa receptor to identify key hoo. com. cn; zbqiu(@shmu educn residues that might account for ligand binding and 0968-0896 ter 2005 Elsevier Ltd. All rights reserved. doi: 10.1016j.bmc. 200
Molecular modeling and 3D-QSAR studies of indolomorphinan derivatives as kappa opioid antagonists Wei Li,a Yun Tang,a,b,* You-Li Zhenga and Zhui-Bai Qiua,* 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 & Technology, 130 Meilong Road, Shanghai 200237, China Received 6 July 2005; revised 19 August 2005; accepted 20 August 2005 Available online 3 October 2005 Abstract—Molecular modeling and 3D-QSAR studies were performed on 31 indolomorphinan derivatives to evaluate their antagonistic behaviors on j opioid receptor and provide information for further modification of this kind of compounds. Best predictions were obtained with CoMFA standard model (q2 = 0.693, N = 4, r 2 = 0.900) and CoMSIA combined model (q2 = 0.617, N = 4, r 2 = 0.904). Both models were further validated by an external test set of eight compounds with satisfactory predictions: r 2 = 0.607 for CoMFA and r 2 = 0.701 for CoMSIA. In addition, the 3D structure of human j opioid receptor was constructed based on the crystal structure of bovine rhodopsin, and the CoMSIA contour plots were then mapped into the structural model of j opioid receptor–GNTI complex to identify key residues, which might account for j antagonist potency and selectivity. The roles of nonconserved Glu297 and conserved Lys227 of human j opioid receptor were then discussed. 2005 Elsevier Ltd. All rights reserved. 1. Introduction There are three well-adopted subtypes of opioid receptors: l, j, and d. 1,2 However, to date the precise role of j opioid receptor has not been well established yet. It appears that j opioid receptor exerts its physiological roles by participating in pain process and regulating immune systems3,4 where selective j opioid antagonists could provide powerful tools to investigate detailed interactions between the receptor and ligand. Meanwhile, j selective antagonists also showed some clinical potentials.5,6 Nor-BNI (norbinaltorphimine, 1) is the first highly selective j opioid antagonist reported by Portoghese et al.,7 who believed that it was the second basic group that conferred j potency and selectivity of the compound.8 The hypothesis was confirmed in their subsequent work9 by structure simplification of nor-BNI. The basic group was further identified interacting with Glu297 of j opioid receptor by site-directed mutagenesis.10 Moreover, the attachment of a basic group to the 50 -substitutes of naltrindole, a potent d selective antagonist, led to the discovery of a more potent j selective antagonist, 50 -guanidine naltrindole (GNTI, 2).11 There were also some other types of j selective antagonists such as JDTic12,13 (3) and KAA-1(4) 14 recently reported by Carroll and co-workers, which also owned a second basic group (Fig. 1). In order to elucidate the mechanism of action of known kappa antagonists and design new kappa selective antagonists, molecular modeling and 3D-QSAR studies were conducted here. At first, 39 indolomorphinan derivatives, potent kappa selective antagonists by adding a basic or neutral group in the 50 -position of naltrindole, were collected from the literature. They were then divided into two groups: 31 compounds as training set and the other eight ones as test set. The training set was used to build 3D-QSAR models with CoMFA (comparative molecular field analysis) and CoMSIA (comparative molecular similarity indices analysis) methods, while the test set was used to validate the 3D-QSAR models further. Meanwhile, due to unavailability of experimental structure, the 3D structure of kappa opioid receptor was constructed based on the crystal structure of bovine rhodopsin. The interaction mode of GNTI with kappa receptor was hence obtained. Finally, the contour plots of CoMFA were mapped into the binding site of kappa receptor to identify key residues that might account for ligand binding and 0968-0896/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.bmc.2005.08.052 Keywords: j Opioid receptor; Indolomorphinan derivatives; 3DQSAR; Molecular Modeling; CoMFA; CoMSIA. * Corresponding authors. Tel.: +86 21 54237595; fax: +86 21 54237264 (Z.B.Q.); tel.: +86 21 54237419 (Y.T.); e-mail addresses: ytang234@yahoo.com.cn; zbqiu@shmu.edu.cn Bioorganic & Medicinal Chemistry 14 (2006) 601–610
W. Li et al./ Bioorg. Med. Chem. 14(2006)601-610 NH2 nor-BNI 1 GNTI 2 DTIc 3 (-)-KAA14 Figure 1. Some potent highly selective K antagonists reported recently selectivity. The studies provided us helpful information The most crucial step in performing CoMFA and CoM- on how to modify indolomorphinan derivatives SIA is to determine the bioactive conformations of the compounds so that all compounds could be aligned together. Though nor-BNI was not employed to estab 2. Materials and methods lish 3D-QSAR models here, its structure was selected as the template for structural alignment from the align 2.1. Data set ment facility in SYBYL, due to its selective x antagonistic potency and quite rigid structure. The final structura Thirty nine compounds were collected from several re- alignment is shown in Figure 2 ports of the Lewis group>-(Table 1). Most of these compounds were indolomorphinan analogs of GNTI 2.3. PLS analys (2)and showed significant k antagonistic potency. though they were also antagonists against H and As usual, PLS(partial least squares)method was used to subtypes, too. All compounds were evaluated in com- establish and validate comfa and comsia models petition binding assays with [H U69593 in cloned here The binding affinity K, values were converted into human k opioid receptor transfected into Chinese pK, values, to describe the biological activities. CoMFA hamster ovary (CHO)cells. Similar assays were also was set at standard values, with a sp carbon atom with intI by Jones and Portoghese. In one positive charge used to probe steric and electrostatic his study, gnti was only considered as a reference fields. The standard cutoff value was set to 30 kcal/mol structure to eliminate evaluation errors between differ- CoMSIA fields were set in their default opinions ent groups. Eight compounds were randomly selected from the 39 molecules to make a test set for further LOo (leave one out)cross-validation method was used model validation, and the rest of the 31 compounds to evaluate the initial models. The cross-validated coef. served as the training set ficient q was calculated using the following equation 2. 2. Molecular modeling and structural alignment 矿=10-20073)2 All calculations were carried out on a r14000 SGI Fuel actu workstation using molecular modeling software package where ,pred, actual, and ,mean are predicted, actual, and SYBYL V6.9.0 All compounds were constructed in sYBYL mean values of the target property (pKi), respectively, based on the crystal structure of nor-BNI- since these and press (7pred-7actua)" is the sum of predictive compounds may similarly bind to k opioid receptor. sum of squares. The optimum number of components Considering the vital role of a basic group for k selectiv- was then given, and CoMFA and CoMSIA models were ity and antagonistic potency, the basic groups of all hence derived corresponding to the optimum number compounds were fixed near the protonated nitrogen The parameters of confidence intervals were further esti atom of nor-BNI supposing they shared similar binding mated by bootstrap in 10 runs. The column filtering box models to K opioid receptor. All compounds were pre was kept unchecked during all operations. tonated and assigned with Gasteiger-Huckel charges For some more flexible compounds, systematic searches 2. 4. Validation of CoMFA and CoMSIa models were performed with an interval of 10 on rotatory bonds to ensure their low energy conformations. Final- In addition to LOO method to validate the CoMFA and ly, they were minimized with Tripos force field. CoMSIA models, a test set made up of eight compounds
selectivity. The studies provided us helpful information on how to modify indolomorphinan derivatives. 2. Materials and methods 2.1. Data set Thirty nine compounds were collected from several reports of the Lewis group15–18 (Table 1). Most of these compounds were indolomorphinan analogs of GNTI (2) and showed significant j antagonistic potency, though they were also antagonists against l and d subtypes, too. All compounds were evaluated in competition binding assays with [3 H] U69593 in cloned human j opioid receptor transfected into Chinese hamster ovary (CHO) cells. Similar assays were also performed on GNTI by Jones and Portoghese.19 In this study, GNTI was only considered as a reference structure to eliminate evaluation errors between different groups. Eight compounds were randomly selected from the 39 molecules to make a test set for further model validation, and the rest of the 31 compounds served as the training set. 2.2. Molecular modeling and structural alignment All calculations were carried out on a R14000 SGI Fuel workstation using molecular modeling software package SYBYL V6.9 SYBYL V6.9. 20 All compounds were constructed in SYBYL based on the crystal structure of nor-BNI21 since these compounds may similarly bind to j opioid receptor. Considering the vital role of a basic group for j selectivity and antagonistic potency, the basic groups of all compounds were fixed near the protonated nitrogen atom of nor-BNI supposing they shared similar binding models to j opioid receptor. All compounds were protonated and assigned with Gasteiger–Hu¨ckel charges. For some more flexible compounds, systematic searches were performed with an interval of 10 on rotatory bonds to ensure their low energy conformations. Finally, they were minimized with Tripos force field.22 The most crucial step in performing CoMFA and CoMSIA is to determine the bioactive conformations of the compounds so that all compounds could be aligned together. Though nor-BNI was not employed to establish 3D-QSAR models here, its structure was selected as the template for structural alignment from the alignment facility in SYBYL, due to its selective j antagonistic potency and quite rigid structure. The final structural alignment is shown in Figure 2. 2.3. PLS analysis As usual, PLS (partial least squares) method was used to establish and validate CoMFA and CoMSIA models here. The binding affinity Ki values were converted into pKi values, to describe the biological activities. CoMFA was set at standard values, with a sp3 carbon atom with one positive charge used to probe steric and electrostatic fields. The standard cutoff value was set to 30 kcal/mol. CoMSIA fields were set in their default opinions. LOO (leave one out) cross-validation method was used to evaluate the initial models. The cross-validated coef- ficient q2 was calculated using the following equation: q2 ¼ 1:0 P cðcpred cactualÞ 2 P cðcactual cmeanÞ 2 where cpred, cactual, and cmean are predicted, actual, and mean values of the target property (pKi), respectively, and PRESS ¼ P cðcpred cactualÞ 2 is the sum of predictive sum of squares. The optimum number of components was then given, and CoMFA and CoMSIA models were hence derived corresponding to the optimum number. The parameters of confidence intervals were further estimated by bootstrap in 10 runs. The column filtering box was kept unchecked during all operations. 2.4. Validation of CoMFA and CoMSIA models In addition to LOO method to validate the CoMFA and CoMSIA models, a test set made up of eight compounds N H O O N OH OH OH HO N nor-BNI 1 HO O N OH N H N H NH2 NH GNTI 2 N OH CH3 JDTic H3C NH O H N HO 3 (-)-KAA1 OH N H3C HN N O 4 Figure 1. Some potent highly selective j antagonists reported recently. 602 W. Li et al. / Bioorg. Med. Chem. 14 (2006) 601–610
W. Li et al./ Bioorg. Med. Chem. 14(2006)601-610 Table 1. Structures and binding affinities in the training set and test set △ Ki(nM) 122 244 十++ 234568911 HHHHHHHHH 0.95±0.04 0.86±0.20 200000200022222222211 3.26±0.12 n.OH 即P 6.96±0.85 2.72±0.39 B4518902346 PM 2.38±0.37 HoooNN L57±0.80 0.68±0.30 H 0.29±0.10 0.30±0.20 iBu L.39±0.14 6.33±040 NH-l.Hex 8.13±2.67 44±0.0 n-Hept 5.6l±0.37 1243 000 191±3.9 Test set ocHH NO 61±045 161583 0202 B1HH 430 n-Pr 1.60±0.28
Table 1. Structures and binding affinities in the training set and test set Compound n R1 R2 Ki (nM) Training set 1 2 H H 1.42 ± 0.17 2 2 H Cl 2.41 ± 0.22 3 2 H NO2 2.14 ± 0.34 4 2 H NH2 0.95 ± 0.04 5 0 H H 0.86 ± 0.20 6 0 H Cl 0.66 ± 0.05 8 0 H NH2 0.63 ± 0.10 9 0 H OH 3.26 ± 0.12 10 0 H m-OH 2.74 ± 0.74 11 2 H H 0.49 ± 0.00 13 0 n-Bu n-Bu 6.96 ± 0.85 14 0 n-Pr n-Pr 2.72 ± 0.39 15 0 n-Pr CPM 2.38 ± 0.37 17 2 O Et 1.57 ± 0.80 18 2 O Pr 0.85 ± 0.40 19 2 O Bu 0.68 ± 0.30 20 2 NH Me 0.29 ± 0.10 22 2 NH Pr 0.25 ± 0.10 23 2 NH n-Bu 0.30 ± 0.20 24 2 NH i-Bu 1.39 ± 0.14 26 2 O NH–n-Bu 6.33 ± 0.40 27 2 O NH–n-Hex 8.13 ± 2.67 29 1 NH n-Pen 1.44 ± 0.04 30 1 NH n-Hept 5.61 ± 0.37 31 0 O n-Hept 21.89 ± 7.11 32 0 O Bn 10.33 ± 0.66 34 0 O (CH2)4Ph 6.18 ± 0.47 35 0 O p-MeO–Bn 2.11 ± 0.35 36 10.1 ± 0.65 38 CH3 H 183 ± 16 39 H Pr 191 ± 3.9 Test set 7 0 H NO2 1.61 ± 0.45 12 2 n-Bu n-Bu 4.80 ± 0.02 16 0 Bn CPM 3.91 ± 0.60 21 2 NH Et 0.28 ± 0.10 25 2 O NH–Et 12.32 ± 1.29 28 1 NH n-Pr 1.60 ± 0.28 33 0 O (CH2)2Ph 2.21 ± 0.35 37 H H 120 ± 9.6 W. Li et al. / Bioorg. Med. Chem. 14 (2006) 601–610 603
W. Li et al./ Bioorg. Med. Chem. 14(2006)601-610 2.5. Homology modeling of kappa opioid receptor The sequence of human kappa opioid receptor was re trieved from the Swiss Prot database(Accession N P41 145). 3The sequences of bovine rhodopsin and hu Pror k and 8 opioid receptors were obtained from Swiss- L,too,for sequence alignments(see Fig. 3). The crystal structure of bovine rhodopsin was retrieved from Protein Data Bank(PDB entry code 1F88),24 which served as the template to generate the structural of kappa opioid receptor. At first the 7 TM fra were constructed by mutating the corresponding m in the template into target residue in kappa receptor Residue Ala106 was inserted into the target structure The extracellular loop 2(EL2), connecting TM4 and TM5, was built on the basis of EL2 of rhodopsin. The 2. Superposition of 39 molecules including compounds in the other extra- and intra-cellular loop regions were built set and test set based on the template of nor-BNI, a potent and with loop search function of sYBYL/Biopolymer module ctive antagonist. (The structure of nor-BNI was removed for The N-and C-terminal regions were extended from the purposes of clarity. ransmembrane regions for 10 residues, not completely built.a disulfide bond was formed between the side was used for model validation. Similar to cross-validat chains of residues Cys131 and Cys210. After all done, ed g values of LOO method, the predictive performance adding all side chains and hydrogen atoms and loading of models on the test set was estimated by predictive Kollman All-Atom charges, the initial structure was values, which is expressed in the following equation energy minimized for 5000 steps with Kollman All-At predictive 2 SSD-press om force field. 25 SSD The protonated GNTi was docked into the minimized where SSD is the sum of squared deviation between the structure of kappa receptor manually, by putting the pK, values of test set molecules and PRESS is the sum of protonated nitrogen toward the side chain of residue squared deviations between the observed and the pre- Asp138 and the guanidine group close to the side chain dicted pKi values of Glu297. The whole complex structure was then min- bRo 35 WQFS hKOR 57 AIPV hMOK68⊥A⊥ h DOR 47 ALAI IL2 SLHGYEVEGPTG FEATLGGEIA LAIERYVVVCKPMSNFRFG-ENH 152 hKOR ISIDYYNMETS HPVKALDERTPLK 174 hMOR ISIDYYNMETS HPVKALDERTPRN 185 LMETWPFGEL Ls工 DYYNMETS IAVCHPVKALDERTPAK 164 EL2 Hairpin bOho 153 WSRYI PEGMQC-SCGIDYYTPHEET 211 hKOR 175 KVREDVDVIECSLQFPDDDXSWWD 234 186 TTKYRQG-S-IDCT'LTESHPTW hDOR 165 MAVTRPRDGA-VV-CMLQFPSPSW 工L3 bRno 212 KEAAAQQQESATTQKAEKEVTRMVIIMVIAFT hKoR 235 RLKSVRLLSG-SREKDRN TRLVLVVVAV 293 hMOR 243 SVRMLSG-SKEKDRNLRRITRMVLVVVAVE h DOR 221 LRSVRLLSG-SKEKDRS 280 bRo 272 328 hKOR 294 ILYA RCERDEC FPLKMRM 352 hMOR 302 RCEREFC 360 RCERQLCRKPCGRP 340 Sequence alignments of three subtypes of human opioid receptors with bovine rhodopsin(N- and C-terminals omitted). The among them were highlight- b(EL), and intra-cellular loop (IL)regions were labeled correspondingly. In transmembrane regions,identical mbrane(TM)extracellular in dark blue, while residues in opioid receptors analogous to those in rhodopsin were colored in red
was used for model validation. Similar to cross-validated q2 values of LOO method, the predictive performance of models on the test set was estimated by predictive r 2 values, which is expressed in the following equation: predictive r 2 ¼ SSD PRESS SSD where SSD is the sum of squared deviation between the pKi values of test set molecules and PRESS is the sum of squared deviations between the observed and the predicted pKi values. 2.5. Homology modeling of kappa opioid receptor The sequence of human kappa opioid receptor was retrieved from the SwissProt database (Accession No. P41145).23 The sequences of bovine rhodopsin and human l and d opioid receptors were obtained from SwissProt, too, for sequence alignments (see Fig. 3). The crystal structure of bovine rhodopsin was retrieved from Protein Data Bank (PDB entry code 1F88),24 which served as the template to generate the structural model of kappa opioid receptor. At first the 7 TM fragments were constructed by mutating the corresponding residue in the template into target residue in kappa receptor. Residue Ala106 was inserted into the target structure. The extracellular loop 2 (EL2), connecting TM4 and TM5, was built on the basis of EL2 of rhodopsin. The other extra- and intra-cellular loop regions were built with loop search function of SYBYL/Biopolymer module. The N- and C-terminal regions were extended from the transmembrane regions for 10 residues, not completely built. A disulfide bond was formed between the side chains of residues Cys131 and Cys210. After all done, adding all side chains and hydrogen atoms and loading Kollman All-Atom charges, the initial structure was energy minimized for 5000 steps with Kollman All-Atom force field.25 The protonated GNTI was docked into the minimized structure of kappa receptor manually, by putting the protonated nitrogen toward the side chain of residue Asp138 and the guanidine group close to the side chain of Glu297. The whole complex structure was then minFigure 2. Superposition of 39 molecules including compounds in the training set and test set based on the template of nor-BNI, a potent and j selective antagonist. (The structure of nor-BNI was removed for purposes of clarity.) Figure 3. Sequence alignments of three subtypes of human opioid receptors with bovine rhodopsin (N- and C-terminals omitted). The transmembrane (TM) extracellular loop (EL), and intra-cellular loop (IL) regions were labeled correspondingly. In transmembrane regions, identical residues among them were highlighted in dark blue, while residues in opioid receptors analogous to those in rhodopsin were colored in red. 604 W. Li et al. / Bioorg. Med. Chem. 14 (2006) 601–610
W. Li et al. Bioorg. Med. Chem. 14(2006)601-610 imized for 5000 steps again, with Tripos force field. charged carbon atom in the center in MMFF94 and Del The minimized GNTI-receptor complex structure was Re methods. Huckel calculation was questionable be- used for further analysis cause of wrong nitrogen charge assignment And one po- sitive ch harge were distributed among three nearby groups 2.6. QSAR coefficient contour maps with the central carbon little contribution to this charge in the Pullman method. However, Gasteiger-Huickel and CoMFA and CoMSIA results were visualized by st Dev" Gasteiger calculations led to different results, in which Coeff contours. The molecule-5 was visualized as the the positive charge spread along the whole guanidine reference structure. Both CoMFA and CoMSIA plots group. The carbon atom and three nearby groups made were contoured by actual values. All the molecules used almost equal contributions to this one positive charge for QSAR analysis were aligned onto the GNTI struc- Considering that the uniform charge distribution may ture in the GNTI-receptor complex, which led to the be mostly preferable for guanidine group, Gasteiger mapping of CoMSIa plots onto the bound receptor Huckel method was finally used in this study model. Key residues, which should account for K selec mmo ty and potency were hence recognized on the receptor 3. 2. CoMFA and CoMSIA models and validation The best predictions were obtained with CoMFA stan dard model(q-=0.693, N=4)and CoMSIA combined 3. Results and discussion model with all descriptors (-=0.617. N=4)Table 2) their predictive performance on the test was r=0.607 3. 1. Charge assignment of guanidine group by CoMFA and r=0.701 by CoMSIA, which indicated that the built 3D-QSAR models were reliable and able to Charge assignment is crucial to a successful QSAR anal- predict binding affinities of new derivatives accurately S, especially when molecules under investigation were CoMSIA models with different field combinations were positive charged or negative charged. In this study, also evaluated by Loo and test set methodologies and many compounds contained a guanidine group with hese models also showed good predictive values table +l charge, which was quite different from those men- 2), which indicated that the CoMsia models were less tioned in any other 3D-QSAR studies in the literature. affected by fields employed. The conventional fit values Different from other groups like-NH3, there are four on training set and prediction values on the test set made central atoms, rather than one or two atoms, to share by the two models are shown in Table 3. The relationshi the one positive charge. How to correctly assign the curves between observed values versus conventional fit one positive charge on each atom in guanidine group values (prediction values) on the training set( test set) is quite challenging. Many methods led to different re- are also displayed in Figure 5 sults, and different charge assignments led to different or even converse 3D-QSAR contours 3. 3. CoMFA and CoMSIA contours At first compounds with amino group in the study were set CoMFA and CoMSIA contour plots showed that there in their protonated type and fixed near the protonated was a large blue contour around the second basic group nitrogen atom of nor-BNI. Various methods were used (Fig 6a), which was in good agreement with previous re- uate rges such as Gasteiger, 26 ports, 10, and indicated that the positive-charged Gasteiger-Huickel, Del Re, 27 MMFF94, 28, 29 Huckel, 0 group in this region helped to increase binding affinity and Pullman methods( Fig 4). Based on different charge However, a red contour was also observed in CoMSIA ssignments on the guanidine group by various methods, contour plots(Fig. 6b), which brought complexity to the positive charges was focused on the central carbon this region. a detailed analysis on electrostatic, steric atom, and the nearby three groups(each group composed and hydrophobic(Fig. 6c) interactions of this kind of of one nitrogen atom and two hydrogen atoms)showed a compounds is discussed below in combination with the little negative charge to neutralize the highly positive K opioid receptor model. Unlike other fields describing asteiger-Huckel Del-Re MMFF94 Huckel Pullman C+0.384 +0286 1419 1.200-0.360+0.059 H+0.194 +0.261 +0.208+0.4500.000 +0.208 -0.285 -0.555-0967+04530.102 Figure 4. Different charge distributions of guanidine group calculated by multiple methods available in SYBYL. Carbon atoms were colored in gray. nitrogen atoms in blue, and hydrogen atoms in white
imized for 5000 steps again, with Tripos force field.22 The minimized GNTI–receptor complex structure was used for further analysis. 2.6. QSAR coefficient contour maps CoMFA and CoMSIA results were visualized by stDev* Coeff contours. The molecule-5 was visualized as the reference structure. Both CoMFA and CoMSIA plots were contoured by actual values. All the molecules used for QSAR analysis were aligned onto the GNTI structure in the GNTI–receptor complex, which led to the mapping of CoMSIA plots onto the bound receptor model. Key residues, which should account for j selectivity and potency were hence recognized on the receptor model. 3. Results and discussion 3.1. Charge assignment of guanidine group Charge assignment is crucial to a successful QSAR analysis, especially when molecules under investigation were positive charged or negative charged. In this study, many compounds contained a guanidine group with +1 charge, which was quite different from those mentioned in any other 3D-QSAR studies in the literature. Different from other groups like –NH3 +, there are four central atoms, rather than one or two atoms, to share the one positive charge. How to correctly assign the one positive charge on each atom in guanidine group is quite challenging. Many methods led to different results, and different charge assignments led to different or even converse 3D-QSAR contours. At first compounds with amino group in the study were set in their protonated type and fixed near the protonated nitrogen atom of nor-BNI. Various methods were used to calculate guanidine group charges such as Gasteiger,26 Gasteiger–Hu¨ckel, Del Re,27 MMFF94,28,29 Hu¨ckel,30 and Pullman31 methods (Fig. 4). Based on different charge assignments on the guanidine group by various methods, the positive charges was focused on the central carbon atom, and the nearby three groups (each group composed of one nitrogen atom and two hydrogen atoms) showed a little negative charge to neutralize the highly positive charged carbon atom in the center in MMFF94 and Del Re methods. Hu¨ckel calculation was questionable because of wrong nitrogen charge assignment. And one positive charge were distributed among three nearby groups with the central carbon little contribution to this charge in the Pullman method. However, Gasteiger–Hu¨ckel and Gasteiger calculations led to different results, in which the positive charge spread along the whole guanidine group. The carbon atom and three nearby groups made almost equal contributions to this one positive charge. Considering that the uniform charge distribution may be mostly preferable for guanidine group, Gasteiger– Hu¨ckel method was finally used in this study. 3.2. CoMFA and CoMSIA models and validation The best predictions were obtained with CoMFA standard model (q2 = 0.693, N = 4) and CoMSIA combined model with all descriptors (q2 = 0.617. N = 4) (Table 2); their predictive performance on the test was r 2 = 0.607 by CoMFA and r 2 = 0.701 by CoMSIA, which indicated that the built 3D-QSAR models were reliable and able to predict binding affinities of new derivatives accurately. CoMSIA models with different field combinations were also evaluated by LOO and test set methodologies and these models also showed good predictive values (Table 2), which indicated that the CoMSIA models were less affected by fields employed. The conventional fit values on training set and prediction values on the test set made by the two models are shown in Table 3. The relationship curves between observed values versus conventional fit values (prediction values) on the training set (test set) are also displayed in Figure 5. 3.3. CoMFA and CoMSIA contours CoMFA and CoMSIA contour plots showed that there was a large blue contour around the second basic group (Fig. 6a), which was in good agreement with previous reports8,10,11 and indicated that the positive-charged group in this region helped to increase binding affinity. However, a red contour was also observed in CoMSIA contour plots (Fig. 6b), which brought complexity to this region. A detailed analysis on electrostatic, steric, and hydrophobic (Fig. 6c) interactions of this kind of compounds is discussed below in combination with the j opioid receptor model. Unlike other fields describing Gasteiger Gasteiger-Hückel Del-Re MMFF94 Hückel Pullman C +0.384 +0.286 +1.419 +1.200 -0.360 +0.059 H +0.194 +0.261 +0.208 +0.450 0.000 +0.208 N -0.183 -0.285 -0.555 -0.967 +0.453 -0.102 Figure 4. Different charge distributions of guanidine group calculated by multiple methods available in SYBYL. Carbon atoms were colored in gray, nitrogen atoms in blue, and hydrogen atoms in white. W. Li et al. / Bioorg. Med. Chem. 14 (2006) 601–610 605