J Mol model(2006)12:877-884 DOI10.1007/00894005-00849 ORIGINAL PAPER Wei li· Yun Tang Qiong Xie· Wei sheng Zhui-Bai Qiu 3D-QSAR studies of orvinol analogs as k-opioid agonists Received: 14 April 2005/ Accepted: 11 November 2005/ Published online: 22 March 2006 c) Springer-Verlag 2006 Abstract Orvinols are potent analgesics that target opioid 1960-1970s, and some of them were several thousand receptors. However, their analgesic mechanism remains times more potent than morphine in analgesic assays of the unclear and no significant preference for subtype opioid rat tail pressure test [1, 2]. Similar to morphine, their receptor has been achieved. In order to find new orvinols biological targets were suggested to be opioid receptors that target the K-receptor, comparative 3D-QSAR studies However, because opioid receptors had not been identified were performed on 26 orvinol analogs using comparative at that time, the detailed analgesic mechanism was difficult molecular field analysis( CoMFA)and comparative mo- to elucidate. Furthermore, only rodent antinociceptive lecular similarity indices analysis(CoMSIA). The best pre- models were available to evaluate the analgesic effects of dictions for the K-receptor were obtained with the CoMFa newly prepared compounds standard model(q-=0.686, r=0.947)and CoMSIA model Until the early 1990s as opioid receptors were cloned combined steric, electrostatic, hydrophobic, and hydrogen gradually, at least three subtypes u, 8 and K were identified, bond donor/acceptor fields ((=0.678, r2=0.914). The which greatly facilitated studies of opioid receptors and models built were further validated by a test set made up their ligands. Typical analgesics, such as morphine and 0.672 for CoMFA and 0.593 for CoMSIA. The study could ever, it has long been recognized that H-opioids have some e helpful for designing and prepare new category notorious side effects such as respiratory depression, drug K-agonists from orvinols cuon ane nd dependence. Recent studies on the k-opioid receptor revealed that k-agonists show potent analgesic Keywords k-opioid agonists. Opiate analgesics effects but lack the above-mentioned side effects which COMFA· COMSIA indicate that k-selective agonists might be developed as a new generation of analgesics without addiction [3, 4] As semi-synthetic opiates, orvinols also target multiple Introduction opioid receptors with diverse efficacy profiles [5]. Howev er, BU46, one of the orvinols, displayed some K-like Orvinols, derived from the structure of thebaine are potent induced analgesic mechanism in vivo [6]. In order to nalgesic. Their structure and activity relationships were investigate K-selective agonists, since the late 1990s thoroughly investigated by Bentley and Colman in the Lewis, successor of Bentley, has focused on orvinols and synthesized a large number of new derivatives and struc s of BU46. They then proposed a tri w.Li·Y.Tang:Q.Xie·w. Sheng:Z.-B.Qiu binding-site model [7] to explain the pharmacological Department of Medicinal Chemistry, School of Pharmac phenomena of orvinols on k-receptor, in combination with niversity other moderately k-selective compounds derived from ori vinous(Fig. 1, except for KT-95 and TRK-820)[8-10] and 32, People's Republic of China all structure-activity relationship (SAR) analyses were carried out qualitatively on these orvinols Therefore in this paper, in order to guide the synthesis of hool of Pharmacy, East China University of Science Technology new orvinol derivatives as k-specific agonists, we in- vestigated orvinol analogs with comparative 3D-QSAR Shanghai 200237, People's Republic of China methods in order to inspect the 3D-QSAR for the same e-mail: yang 234@ yahoo. series of compounds with affinities against the k-opioid Tel:+86-21-54237419 receptor, using comparative molecular field analysis
J Mol Model (2006) 12: 877–884 DOI 10.1007/s00894-005-0084-9 OR IG INAL PAPER Wei Li . Yun Tang . Qiong Xie . Wei Sheng . Zhui-Bai Qiu 3D–QSAR studies of orvinol analogs as κ-opioid agonists Received: 14 April 2005 / Accepted: 11 November 2005 / Published online: 22 March 2006 # Springer-Verlag 2006 Abstract Orvinols are potent analgesics that target opioid receptors. However, their analgesic mechanism remains unclear and no significant preference for subtype opioid receptor has been achieved. In order to find new orvinols that target the κ-receptor, comparative 3D–QSAR studies were performed on 26 orvinol analogs using comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA). The best predictions for the κ-receptor were obtained with the CoMFA standard model (q 2 =0.686, r 2 =0.947) and CoMSIA model combined steric, electrostatic, hydrophobic, and hydrogen bond donor/acceptor fields (q 2 =0.678, r 2 =0.914). The models built were further validated by a test set made up of seven compounds, leading to predictive r 2 values of 0.672 for CoMFA and 0.593 for CoMSIA. The study could be helpful for designing and prepare new category κ-agonists from orvinols. Keywords κ-opioid agonists . Opiate analgesics . CoMFA . CoMSIA Introduction Orvinols, derived from the structure of thebaine, are potent analgesics. Their structure and activity relationships were thoroughly investigated by Bentley and Colman in the 1960–1970s, and some of them were several thousand times more potent than morphine in analgesic assays of the rat tail pressure test [1, 2]. Similar to morphine, their biological targets were suggested to be opioid receptors. However, because opioid receptors had not been identified at that time, the detailed analgesic mechanism was difficult to elucidate. Furthermore, only rodent antinociceptive models were available to evaluate the analgesic effects of newly prepared compounds. Until the early 1990s as opioid receptors were cloned gradually, at least three subtypes μ, δ and κ were identified, which greatly facilitated studies of opioid receptors and their ligands. Typical analgesics, such as morphine and fentanyl, are found to be potent μ-opioid agonists. However, it has long been recognized that μ-opioids have some notorious side effects such as respiratory depression, drug addiction and dependence. Recent studies on the κ-opioid receptor revealed that κ-agonists show potent analgesic effects but lack the above-mentioned side effects, which indicate that κ-selective agonists might be developed as a new generation of analgesics without addiction [3, 4]. As semi-synthetic opiates, orvinols also target multiple opioid receptors with diverse efficacy profiles [5]. However, BU46, one of the orvinols, displayed some κ-like induced analgesic mechanism in vivo [6]. In order to investigate κ-selective agonists, since the late 1990s, Lewis, successor of Bentley, has focused on orvinols and synthesized a large number of new derivatives and structurally related analogs of BU46. They then proposed a triple binding-site model [7] to explain the pharmacological phenomena of orvinols on κ-receptor, in combination with other moderately κ-selective compounds derived from orivinols (Fig. 1, except for KT-95 and TRK-820) [8–10] and all structure-activity relationship (SAR) analyses were carried out qualitatively on these orvinols. Therefore in this paper, in order to guide the synthesis of new orvinol derivatives as κ-specific agonists, we investigated orvinol analogs with comparative 3D–QSAR methods in order to inspect the 3D–QSAR for the same series of compounds with affinities against the κ-opioid receptor, using comparative molecular field analysis W. Li . Y. Tang . Q. Xie . W. Sheng . Z.-B. Qiu Department of Medicinal Chemistry, School of Pharmacy, Fudan University, 138 Yixueyuan Road, Shanghai 200032, People’s Republic of China e-mail: zbqiu@shmu.edu.cn Y. Tang (*) School of Pharmacy, East China University of Science & Technology, 130 Meilong Road, Shanghai 200237, People’s Republic of China e-mail: ytang234@yahoo.com.cn Tel.: +86-21-54237419
878 Haco OH H3CO OI HN Y H3C BU46 Bermejo Fm. et aL. 1999 irundt P. et al. 2003 Hac O BU74 KT-95 TRK-820 CoMFA) and comparative molecular similarity indices an interval of 15% on rotatable bonds to obtain their lowest analysis( CoMSIA) energy conformations. Finally, all molecules were mini- mized using the Tripos force field [16]. The energy. minimized conformation of molecule 17. which showed Materials and methods the highest binding affinity to the k-receptor of all compounds studied in this paper, served as the template Data set A well-known opiate analgesic pharmacophore-a tyra mine fragment [17] was common among all molecules Thirty-three compounds were collected from several Ihis fragment was used as the common substructure for structural alignments. The alignment resulting from the reports by the Lewis and Husbands group(see Table 1) alignment facility in SYBYL is shown in Fig. 2 [7-9, 11-13. The binding affinities to the K-opioid recep tor were determined by displacement binding assays of cloned human opioid receptors transferred onto CHO Cells PLS anal Seven of 33 compounds from the dataset were randor selected to form the test set, with the remaining 26 com- The Ki value to the K-receptor of each molecule was ds used to make the training set. converted to pK(logKi) and set as dependent values while CoMFA and CoMSIA descriptors were set as independent variables to perform PLS regression analyses Molecular modeling and structural alignment The CoMfa cutoff values were set to 30 kcal mol for both steric and electrostatic fields and all fields were scaled All calculations were carried out on a R14000 SGI Fuel by the default options in SYBYL workstation running the molecular modeling software The initial predictive coefficient q2 values and the 9 The initial structures of the 3 3 molecules were built (leave-one-out) cross-validation method. The cross-vali- ckage SYBYL V6.9[14] optimal number of components were obtained by the Loo ased on crystal structures of their analogs etorphine, dated coefficient q was calculated using the following diprenorphine and oxymorphone[15]. All molecules were formula: set in their unprotonated states and assigned Gasteiger- Huckel charges available in SYBYL. Except for the rigid morphine-like skeleton, random searches were performed ∑(prat-7 actual) on additional ring systems to ensure that their conforma- 9=1.0 tions were energetically favorable. For molecules with more flexibility, systematic searches were carried out with
(CoMFA) and comparative molecular similarity indices analysis (CoMSIA). Materials and methods Data set Thirty-three compounds were collected from several reports by the Lewis and Husbands group (see Table 1) [7–9, 11–13]. The binding affinities to the κ-opioid receptor were determined by displacement binding assays of cloned human opioid receptors transferred onto CHO Cells with [3 H]U69593 as the label. Seven of 33 compounds from the dataset were randomly selected to form the test set, with the remaining 26 compounds used to make the training set. Molecular modeling and structural alignment All calculations were carried out on a R14000 SGI Fuel workstation running the molecular modeling software package SYBYL v6.9 [14]. The initial structures of the 33 molecules were built based on crystal structures of their analogs etorphine, diprenorphine and oxymorphone [15]. All molecules were set in their unprotonated states and assigned Gasteiger– Hückel charges available in SYBYL. Except for the rigid morphine-like skeleton, random searches were performed on additional ring systems to ensure that their conformations were energetically favorable. For molecules with more flexibility, systematic searches were carried out with an interval of 15° on rotatable bonds to obtain their lowest energy conformations. Finally, all molecules were minimized using the Tripos force field [16]. The energyminimized conformation of molecule 17, which showed the highest binding affinity to the κ-receptor of all compounds studied in this paper, served as the template. A well-known opiate analgesic pharmacophore—a tyramine fragment [17] was common among all molecules. This fragment was used as the common substructure for structural alignments. The alignment resulting from the alignment facility in SYBYL is shown in Fig. 2. PLS analysis The Ki value to the κ-receptor of each molecule was converted to pKi(−logKi) and set as dependent values, while CoMFA and CoMSIA descriptors were set as independent variables to perform PLS regression analyses. The CoMFA cutoff values were set to 30 kcal mol−1 for both steric and electrostatic fields and all fields were scaled by the default options in SYBYL. The initial predictive coefficient q 2 values and the optimal number of components were obtained by the LOO (leave-one-out) cross-validation method. The cross-validated coefficient q 2 was calculated using the following formula: q2 ¼ 1:0 P γ ðγpred γactualÞ 2 P γ γactual γmean ð Þ2 N HO O H3CO OH H 7 8 BU46 N H3CO OH O CH3 Ph H3C 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 10 19 Bermejo FM, et al. 1999 N H3CO OH CH3 Ph HN H3C Grundt P, et al. 2003 N HO H3CO CH3 HN Ph O AcO N S O CH3 O N OH HO O N H3C O O BU74 KT-95 TRK-820 Fig. 1 Some opioids with potent κ-agonist potency 878
able 1 Structures and binding affinities of molecules used in training and test sets CH3 1-4 NRy CH3 R,O R2 R4 12-17 R,O K1(nM)() Training set Me 0.84±0.04 OBn 11±0.08 H 24.1±3.1 0.87±0.13 1.04±0.01 11 3.12±0.20 2.0±1.0 15 HMHH Ph Me 0.3±0.2 16 Ph H 0.2±0.02 0.14±0.10 19 Et 555±35 0.61±0.22 HHHPHHHH 239±88 94.2±0.49 27 NHCH?-nBu 36.1+56 13.8±0.78 NHCOBn 31 CH,,Bn 63.3±16.4 CH2-iBu 457±110 Test
Table 1 Structures and binding affinities of molecules used in training and test sets Compound r1 r2 r3 r4 Ki (nM) (κ) Training set 1 Me Me OMe H 3.9±1.1 2 Me Me OBn H 0.84±0.04 4 Me H OBn Me 1.1±0.08 5 2.4±1.3 6 Me 899±150 7 H 24.1±3.1 8 Me H 0.87±0.13 10 CPM H 1.04±0.01 11 CPM Me 3.12±0.20 12 Me H Me Ph 5.3±2.7 13 H H Me Ph 2.0±1.0 15 Me Me Ph Me 0.3±0.2 16 Me H Ph H 0.2±0.02 17 Me H Ph Me 0.14±0.10 18 Me 4841±303 19 Et 555±35 20 Ph 2608±28 22 Me H Ph CH3 0.61±0.22 23 Me H Me H 239±88 24 Me H Me Me 28.4±2.6 25 H Ph 94.2±0.49 27 NHCH2-nBu H 36.1±5.6 28 NHCH2Bn H 13.8±0.78 30 NHCOBn H 148±58 31 CH2CH2Bn H 63.3±16.4 33 CH2-iBu Me 457±110 Test Set N HO O H3CO R3 R2 R4 R1 1-4 N HO O H3CO CH3 OH 5 N HO O H3CO Me CH3 R1 Ph 6-7 8-11 NR1 HO O H3CO HO H R2 R2 N R1O OR2 O CH3 R3 R4 12-17 N CH3 MeO HO R1 Me H3C HO 18-20 N R1O OR2 CH3 R3 R4 21-24 HN N HO O CH3 HN CH3 25-26 R1 R2 N HO O CH3 27-33 HN R2 R1 879
880 Tablel(continued) Compound Ki(nM)(k) H OMe Me 43.3±2. 128±0.01 H H 34.6+4.1 440±119 NHCH, CH,Bn H 779±0.34 CH,CH, Bn 6.40.1 improved to ensure the highest q values corresponded to a suitable optimum number of components. According to the optimum number of components with lowest PRESS values obtained, PLS regression models were derived. Finally, the confidence intervals for the parameters(mean and standard derivation) were estimated by bootstrap in ten runs QSAR model validation In addition to LOO method to validate QSar models, the established test set was used for further evaluation The overall predictive performance of models on the test set is often reported as a predictive r value, defined analogously to the cross-validated by comparing the accuracy of predictions with the variation in the actual data in the test set. The value of predictive r is computed by th e following Fig. 2 Structual alignment of 33 orvinols and structurally related equation: K-agonists(containing compounds in the test set) predictive r2 SSD- PRESS where pred, actual and mean are predicted, actual, and mean values of the target property(pKi), respectively. And where SSD is the sum of squared deviation between the pk; PRESS=>(pred -Actual) is the sum of predictive values of test set molecules and PRESS is the sum of sum of squares. The Column Filtering box was checked at a pk; valueviations between the observed and the predicted squared dev typical value of 2.0, to reduce analysis time with small effect on the q values. Overall fits of this alignment were ig. 3 11.00 versus pECso values from 27 orvinols derived kappa agonists 7 1100
where γpred, γactual and γmean are predicted, actual, and mean values of the target property (pKi), respectively. And PRESS ¼ P γ ðγpred γactualÞ 2 is the sum of predictive sum of squares. The Column Filtering box was checked at a typical value of 2.0, to reduce analysis time with small effect on the q 2 values. Overall fits of this alignment were improved to ensure the highest q 2 values corresponded to a suitable optimum number of components. According to the optimum number of components with lowest PRESS values obtained, PLS regression models were derived. Finally, the confidence intervals for the parameters (mean and standard derivation) were estimated by bootstrap in ten runs. QSAR model validation In addition to LOO method to validate QSAR models, the established test set was used for further evaluation. The overall predictive performance of models on the test set is often reported as a predictive r 2 value, defined analogously to the cross-validated q 2 by comparing the accuracy of predictions with the variation in the actual data in the test set. The value of predictive r 2 is computed by the following equation: predictive r2 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. 11.00 10.00 9.00 8.00 7.00 6.00 5.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 pKi pEC50 Fig. 3 Analysis on pKi versus pEC50 values from 27 orvinols derived kappa agonists Fig. 2 Structual alignment of 33 orvinols and structurally related κ-agonists (containing compounds in the test set) Compound r1 r2 r3 r4 Ki (nM) (κ) 3 Me H OMe Me 43.3±2.5 9 Me Me 1.28±0.01 14 H Me Ph Me 0.2±0.1 21 Me H Ph H 34.6±4.1 26 Me Me 440±119 29 NHCH2CH2Bn H 7.79±0.34 32 CH2CH2Bn Me 36.4±0.16 Table1 (continued) 880
Table 2 CoMFA and CoMSIA results Model SE SEE bsr 92 SD CoMFA(standard) 0.947 0.30 94.645 0.265 0.019 CoMSIA(steric+electro) 0.705 0.861 0.512 71220 0.418 0.040 CoMSIA(steric+electro+ hydrophobic) 0.713 0.874 0.488 79.74 0.427 0.015 CoMSIA(steric+electro+ 0.682 0.465 0.335 0.938 CoMSIA(steric+electro+ Hydrophobictacceptor) 0.692 0.90 0.430 70.911 0.384 0.921 CoMSIA(all descriptors) 0.678 0.421 55.97 0.94 cr: noncross-validated correlation coefficient, dsEE standard error of estimate F: F-test value of estimate for boot strapping analysis, QSAR coefficient contour maps training set and the observed values versus predicted values in the test set. All pki values in the test set were The results of all the CoMFA and CoMSIA models were among the scope of the training set and both QSAr visualized using the ' Dev*Coeff mapping option. And models here were well validated from the test set methods olecule 17 was set as the template to validate these contour maps Characteristics of orvinol analogs binding to the K-opioid receptor Resuits and discussion From K-agonists that retain the full morphine structure, as CoMFA and CoMSIA models for k-opioid receptor well as the compounds in this study, some potent and elective k-agonists found through structural mod- In order to establish reliable 3D-QSAR models, analgesic ifications on two specific regions. Since compound 17([8] activities(in vitro ECso values) were also collected from in Fig. 1)was selected as the template for QSAr contour the literature. However, there was no obviously linear demonstration. The two special regions mapped to this relationship between pECso and pk; values(shown in compound were 6-substituted groups in the parent mor 3), and QSAR models established from pECso seemed phine structure and additional 19-substituted groups(see to be poor(q-=0. 269, N-2 and r=0.530 for 27 molecules Fig. 1). The influence of these substitutent groups on thei in CoMFA, data not shown), so only pki was used as the binding to the k-receptor are discussed below from QSAR dependent value. The reason was possibly that both contour plots CoMFA and CoMSIA methods are based on the assump- From the Electrostatic Fields contours shown in Fig 5a tion that changes in binding affinities of ligands are related (CoMFA contours), it is clear that any introduction of six to changes in molecular properties represented by multiple substitution or 6, 7-ring constrained structure could facilitate fields, steric, electrostatic, etc., and other factors besides binding only if they contain electronegative groups. This is binding affinity might be involved in determining the true since some potent K-agonists such as KT-95 [18] and analgesic potency of those compounds on the k-opioid TRK-820 [19](see Fig. 1)contain highly electronegative rec oxygen in this region. Interestingly, CoMSIA contours 3D-QSAR models of orvinol analogs were investigated showed replacing C-6 with more electron-donating groups by their binding affinities to the k-opioid receptor with K may favor binding to the k-receptor. However, no pre- values ranging from 0. 14 to 4841 nM. The best dominant electrostatic contours around the C-19 region were predictions were obtained by the CoMFA standard observed in Fig 5a model (=0.686, /=0.947)and CoMSIA combined Both yellow contours found in CoMFA and CoMSIA steric, electrostatic, hydrophobic, and donor/acceptor plots( Fig. 5b)showed that large steric groups may reduce fields((=0.678, /=0.914). All CoMFA and CoMSIA compound binding affinities in six substitutions.The analysis parameters and results are shown in Table 2. In binding affinities of all compounds with 6, 7-ring con- addition, Table 3 and Fig. 4 show the table and graph of strained indole structures decreased considerably when the observed values versus conventional fit values in the compared to the compounds without such structures in thi
QSAR coefficient contour maps The results of all the CoMFA and CoMSIA models were visualized using the ‘stDev*Coeff’ mapping option. And molecule 17 was set as the template to validate these contour maps. Results and discussion CoMFA and CoMSIA models for κ-opioid receptor In order to establish reliable 3D–QSAR models, analgesic activities (in vitro EC50 values) were also collected from the literature. However, there was no obviously linear relationship between pEC50 and pKi values (shown in Fig. 3), and QSAR models established from pEC50 seemed to be poor (q 2 =0.269, N=2 and r 2 =0.530 for 27 molecules in CoMFA, data not shown), so only pKi was used as the dependent value. The reason was possibly that both CoMFA and CoMSIA methods are based on the assumption that changes in binding affinities of ligands are related to changes in molecular properties represented by multiple fields, steric, electrostatic, etc., and other factors besides binding affinity might be involved in determining the analgesic potency of those compounds on the κ-opioid receptor. 3D–QSAR models of orvinol analogs were investigated by their binding affinities to the κ-opioid receptor with Ki values ranging from 0.14 to 4841 nM. The best predictions were obtained by the CoMFA standard model (q 2 =0.686, r 2 =0.947) and CoMSIA combined steric, electrostatic, hydrophobic, and donor/acceptor fields (q 2 =0.678, r 2 =0.914). All CoMFA and CoMSIA analysis parameters and results are shown in Table 2. In addition, Table 3 and Fig. 4 show the table and graph of the observed values versus conventional fit values in the training set and the observed values versus predicted values in the test set. All pKi values in the test set were among the scope of the training set and both QSAR models here were well validated from the test set methods. Characteristics of orvinol analogs binding to the κ-opioid receptor From κ-agonists that retain the full morphine structure, as well as the compounds in this study, some potent and selective κ-agonists were found through structural modifications on two specific regions. Since compound 17 ([8] in Fig. 1) was selected as the template for QSAR contour demonstration. The two special regions mapped to this compound were 6-substituted groups in the parent morphine structure and additional 19-substituted groups (see Fig. 1). The influence of these substitutent groups on their binding to the κ-receptor are discussed below from QSAR contour plots: From the Electrostatic Fields contours shown in Fig. 5a (CoMFA contours), it is clear that any introduction of six substitution or 6, 7-ring constrained structure could facilitate binding only if they contain electronegative groups. This is true since some potent κ-agonists such as KT-95 [18] and TRK-820 [19] (see Fig. 1) contain highly electronegative oxygen in this region. Interestingly, CoMSIA contours showed replacing C-6 with more electron-donating groups may favor binding to the κ-receptor. However, no predominant electrostatic contours around the C-19 region were observed in Fig. 5a. Both yellow contours found in CoMFA and CoMSIA plots (Fig. 5b) showed that large steric groups may reduce compound binding affinities in six substitutions. The binding affinities of all compounds with 6, 7-ring constrained indole structures decreased considerably when compared to the compounds without such structures in this Table 2 CoMFA and CoMSIA results Model q2a N b r 2c SEEd F e SEE bsf q2 bsg SDh CoMFA(standard) 0.686 4 0.947 0.330 94.645 0.265 0.962 0.019 CoMSIA(steric+electro) 0.705 2 0.861 0.512 71.220 0.418 0.903 0.040 CoMSIA(steric+electro+ hydrophobic) 0.713 2 0.874 0.488 79.747 0.427 0.895 0.015 CoMSIA(steric+electro+ hydrophobic+donor) 0.682 3 0.895 0.465 62.345 0.335 0.938 0.032 CoMSIA(steric+electro+ Hydrophobic+acceptor) 0.692 3 0.906 0.430 70.911 0.384 0.921 0.030 CoMSIA(all descriptors) 0.678 4 0.914 0.421 55.975 0.337 0.946 0.019 a q2 : leave-one-out (LOO) cross-validated correlation coefficient, b N: optimum number of components, c r 2 : noncross-validated correlation coefficient, d SEE: standard error of estimate, e F: F-test value, f SEEbs: standard error of estimate for boot strapping analysis, g q2 bs: mean r squared of boot strapping analysis (ten runs), h SD: standard deviation 881
Table 3 Actual versus conventional fit values (predicted values) activities of CoMFA standard and CoMSIA combined model Compound number Actual pKi CoMFA CoMSIA Conventional fit R Conventional fit 8.409 8.639 0.230 8.698 0.289 9.076 9.039 245678 0.136 8.973 0.014 8.620 0.122 0.171 6.046 6.732 0.686 6.625 0.579 7.618 8.735 0.325 8.618 0.442 8.983 9.018 0.035 0.284 8.506 0.299 8.665 0.159 0.044 8.29 -0.017 235678 8.699 832 9.5 9.596 925 9.699 9.802 0.103 9.794 0.09 9.854 9787 0.067 9.808 0.046 5.315 5.482 0.167 5.651 0.336 6.256 5.824 5.584 5.539 0.045 22 9.215 8.748 0.467 8.432 6.622 0436 0.838 7.54 0.214 7431 0.116 7.026 6.731 6.548 0.478 7.607 0.165 7.860 30 6.830 2 0.119 6982 0.152 0.173 7332 0.133 6.340 0.312 6.588 0.248 Test set Compound number CoMF CoMSI 7364 7.871 0.507 7917 0.553 9.699 0.256 9.463 21 7461 0933 8.427 -0.966 6.357 0.212 32 7439 6.621 0.818 6.468 CoMFA standard model, CoMSIA combined(steric, electrostatic, hydrophobic, donor and acceptor) model, Conventional fitted value Difference between actual and fitted(predicted) pki values "Predicted pki value tudy. However, the introduction of large steric groups may be preferred in C-19 substitution. The contours of obviously could help to improve the binding affinities in the hydrogen bond fields(Fig. 6b)showed that there should be region of C-19 substitutions. Both geometric isomers of a hydrogen-acceptor site on the k-opioid receptor near C-6 bulky C-19 substitution contribute to binding, but the trans- that is essential for binding. The whole compound 17 isomer with large steric substitution was more active than structure was surrounded by unfavorable hydrogen-accep the cis-isomer. The bulky trans-geometric isomer in C-19 tor field contours(Plot not shown) is important for k-receptor binding In addition, we also performed QSAR studies on In CoMSIA hydrophobic fields contours(Fig. 6a), we selectivities between K and u with (pkik-pKil) as the found that the attachment of hydrophobic groups to C-6 dependent values, also considering their available binding may favor compound binding, but more hydrophilic group affinities to the u-opioid receptor. However, the poor cross-
study. However, the introduction of large steric groups obviously could help to improve the binding affinities in the region of C-19 substitutions. Both geometric isomers of bulky C-19 substitution contribute to binding, but the transisomer with large steric substitution was more active than the cis- isomer. The bulky trans-geometric isomer in C-19 is important for κ-receptor binding. In CoMSIA hydrophobic fields contours (Fig. 6a), we found that the attachment of hydrophobic groups to C-6 may favor compound binding, but more hydrophilic group may be preferred in C-19 substitution. The contours of hydrogen bond fields (Fig. 6b) showed that there should be a hydrogen-acceptor site on the κ-opioid receptor near C-6 that is essential for binding. The whole compound 17 structure was surrounded by unfavorable hydrogen-acceptor field contours (Plot not shown). In addition, we also performed QSAR studies on selectivities between κ and μ with (pKiκ−pKiμ) as the dependent values, also considering their available binding affinities to the μ-opioid receptor. However, the poor crossTable 3 Actual versus conventional fit values (predicted values) activities of CoMFA standard and CoMSIA combined model Compound number Actual pKi CoMFAa CoMSIAb Conventional fitc Res.d Conventional fitc Res.d Training set 1 8.409 8.639 −0.230 8.698 −0.289 2 9.076 9.167 −0.091 9.039 0.037 4 8.959 9.095 −0.136 8.973 −0.014 5 8.620 8.498 0.122 8.791 −0.171 6 6.046 6.732 −0.686 6.625 −0.579 7 7.618 6.895 0.723 6.762 0.856 8 9.060 8.735 0.325 8.618 0.442 10 8.983 9.018 −0.035 9.267 −0.284 11 8.506 8.805 −0.299 8.665 −0.159 12 8.276 8.320 −0.044 8.293 −0.017 13 8.699 8.527 0.172 8.612 0.087 15 9.523 9.596 −0.073 9.258 0.265 16 9.699 9.802 −0.103 9.794 −0.095 17 9.854 9.787 0.067 9.808 0.046 18 5.315 5.482 −0.167 5.651 −0.336 19 6.256 5.905 0.351 5.824 0.432 20 5.584 5.539 0.045 5.58 0.004 22 9.215 8.748 0.467 8.432 0.783 23 6.622 7.058 −0.436 7.46 −0.838 24 7.547 7.333 0.214 7.431 0.116 25 7.026 6.731 0.295 6.548 0.478 27 7.442 7.607 −0.165 7.842 −0.400 28 7.860 7.569 0.291 7.688 0.172 30 6.830 6.949 −0.119 6.982 −0.152 31 7.199 7.372 −0.173 7.332 −0.133 33 6.340 6.652 −0.312 6.588 −0.248 Test set Compound number Actual pKi CoMFAa CoMSIAb Pred.e Res.d Pred.e Res.d 3 7.364 7.871 −0.507 7.917 −0.553 9 8.893 8.294 0.599 8.481 0.412 14 9.699 9.955 −0.256 9.463 0.236 21 7.461 8.394 −0.933 8.427 −0.966 26 6.357 6.569 −0.212 6.522 −0.165 29 8.108 7.735 0.373 7.366 0.742 32 7.439 6.621 0.818 6.468 0.971 a CoMFA standard model, b CoMSIA combined (steric, electrostatic, hydrophobic, donor and acceptor) model, c Conventional fitted value, d Difference between actual and fitted (predicted) pKi values, e Predicted pKi value 882
883 Fig 4 Plot of observed ersus conventional fitting 10 lining set(a) and test set ue rhombs show redictions(conventional fit) of CoMFA standard model and 三 CoMSIA combined model 6 10.0 b 8 10.0 Fig 5 CoMFA std and CoMSIA combined stdev* coeff whereas blue contours indicate sitive charge favor binding ffinity; b green contours dicate bulk favored for favored for binding 人 CoMFA CoMSIA
10 9.5 9 8.5 8 7.5 7 6.5 6 5.5 5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 Observed Activities Conventional Fit a 10 9.5 9 8.5 8 7.5 7 6.5 6 5.5 5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 Observed Activities Predicted Activities b Fig. 4 Plot of observed versus conventional fitting predictions (predicted activities) of training set (a) and test set (b). Blue rhombs show predictions (conventional fit) of CoMFA standard model and pink triangles show those of CoMSIA combined model Fig. 5 CoMFA std and CoMSIA combined stdev* coeff contour plots: a red contours indicate negative charge increase binding affinity, whereas blue contours indicate positive charge favor binding affinity; b green contours indicate bulk favored for binding affinity, whereas yellow contour indicates less bulk favored for binding 883
884 Fig 6 CoMSIA combined stdev*coeff hydrophobic(a) H-bond donor(b): yellow contours indicate hydrophobic hereas while contours indicate tan contours indicate a H-bond acceptor groups in the recepto increase binding validation values('=0.313, N=2 for 33 compounds of this References paper) and narrow selectivity range (from-1.53 to 1.51), only lead to negative results. These may result from the low l. Bentley kw, Hardy DG(1967)J Am Chem Soc 89: 3281-3292 selectivity in these compounds for k and u To date no highly selective potent K-agonist from the 4. Husbands sM(2004) Expert Opin ther patents 14: 1725-1741 morphine structure has been revealed. These QSAR studies 5. Lewis JW, Husbands SM (2004) Curr Pharm Des 10: 717-732 also proved to be too poor to establish selective QSAR 6. Traynor JR, Guo L, Coop A, Lewis JW, Woods JH(1999) J Pharmacol Exp Ther 291: 1093-1099 models. This leads us to conclude that orvinols'parent 7 Coop A, Norton CL, Berzetei-Gurske I, Burnside J, Toll L, structure(morphine) is a non-selective pharmacophore targeting opioid receptors 8. Bermejo FM, Husbands SM, Lewis JW(1999)Hel Chim Acta 9. Grundt P, Martinez-Bermejo F, Lewis JW, Husbands SM 10. Husbands SM, Neilan CL, Broadbear J, Grundt P, Breeden S Aceto MD, Woods JH, Lewis Jw, Traynor JR(2005)Eur J In this paper CoMFA and CoMSIA 3D-QSAR models were Pharmacol 509: 117-125 ative, small steric and hydrophobic groups in the postia s- 12. Grundi ped Chem Lett 9: 831-83y as K, Lewis JW(1999) The influence of six and 19 substitutions on binding af- Bi finities were also highlighted by these QSAR studies. Ne les AR, Traynor JR, Lewis JW, Husbands SM 13. Grundt P, Martinez-Bem3-1566 (2003)J Med Chem 46: 15 Lewis JW. Husbands SM lky trans-geometric isomer is also important for K- 14. SYBYL, version 6.9(2003)1699 Hanley Road, St. Louis, receptor binding. This study could be helpful for designing 15. The crystal structures of etorphine, diprenorphine and oxymor and prepare new category K-agonists from orvinols onewereobtainedfromhttp://www.opioid.umn.edw/ 16. Clark M, Cramer RDl, van Opdenbosch N(1989)J Comput Acknowledgement We appreciated the reviewers efforts to Chem10982-1012 improve the quality of this article. 17. Hom A, Rodgers J(1977)J Pharm Pharmacol 29: 257-265 18. Sagara T, Ozawa S, Kushiyama E, Koike K, Takayanagi I, Kanematsu K(1995)Bioorg Med Chem Lett 5: 1505-1508 19 Nagase H, Hayakawa J, Kawamura K, Kawai K, Takezawa Y, Matsuura H, Tajima C, Endo T(1998)Chem Pharm Bull 46 366-369
validation values (q 2 =0.313, N=2 for 33 compounds of this paper) and narrow selectivity range (from −1.53 to 1.51), only lead to negative results. These may result from the low selectivity in these compounds for κ and μ. To date no highly selective potent κ-agonist from the morphine structure has been revealed. These QSAR studies also proved to be too poor to establish selective QSAR models. This leads us to conclude that orvinols’ parent structure (morphine) is a non-selective pharmacophore targeting opioid receptors. Conclusions In this paper CoMFA and CoMSIA 3D-QSAR models were established successfully for orvinol analogs as κ-agonists. The influence of six and 19 substitutions on binding affinities were also highlighted by these QSAR studies. Negative, small steric and hydrophobic groups in the position six and hydrophilic 19-substitutions facilitate binding. Bulky trans-geometric isomer is also important for κ- receptor binding. This study could be helpful for designing and prepare new category κ-agonists from orvinols. Acknowledgement We appreciated the reviewer’s efforts to improve the quality of this article. References 1. Bentley KW, Hardy DG (1967) J Am Chem Soc 89:3281–3292 2. Bently KW, Hardy DG (1967) J Am Chem Soc 89:3267–3273 3. Eguchi M (2004) Med Res Rev 24:182–212 4. Husbands SM (2004) Expert Opin Ther Patents 14:1725–1741 5. Lewis JW, Husbands SM (2004) Curr Pharm Des 10:717–732 6. Traynor JR, Guo L, Coop A, Lewis JW, Woods JH (1999) J Pharmacol Exp Ther 291:1093–1099 7. Coop A, Norton CL, Berzetei-Gurske I, Burnside J, Toll L, Husbands SM, Lewis JW (2000) J Med Chem 43:1852–1857 8. Bermejo FM, Husbands SM, Lewis JW (1999) Hel Chim Acta 82:1721–1727 9. Grundt P, Martinez-Bermejo F, Lewis JW, Husbands SM (2003) J Med Chem 46:3174–3177 10. Husbands SM, Neilan CL, Broadbear J, Grundt P, Breeden S, Aceto MD, Woods JH, Lewis JW, Traynor JR (2005) Eur J Pharmacol 509:117–125 11. Husbands SM, Breeden SW, Grivas K, Lewis JW (1999) Bioorg Med Chem Lett 9:831–834 12. Grundt P, Jales AR, Traynor JR, Lewis JW, Husbands SM (2003) J Med Chem 46:1563–1566 13. Grundt P, Martinez-Bermejo F, Lewis JW, Husbands SM (2003) Hel Chim Acta 86:793–798 14. SYBYL, version 6.9 (2003) 1699 Hanley Road, St. Louis, MO63144 15. The crystal structures of etorphine, diprenorphine and oxymorphone were obtained from http://www.opioid.umn.edu/ 16. Clark M, Cramer RDI, van Opdenbosch N (1989) J Comput Chem 10:982–1012 17. Hom A, Rodgers J (1977) J Pharm Pharmacol 29:257–265 18. Sagara T, Ozawa S, Kushiyama E, Koike K, Takayanagi I, Kanematsu K (1995) Bioorg Med Chem Lett 5:1505–1508 19. Nagase H, Hayakawa J, Kawamura K, Kawai K, Takezawa Y, Matsuura H, Tajima C, Endo T (1998) Chem Pharm Bull 46: 366–369 Fig. 6 CoMSIA combined stdev*coeff hydrophobic (a) and H-bond donor (b): yellow contours indicate hydrophobic groups increase binding affinity whereas white contours indicate hydrophilic groups favored; cyan contours indicate a H-bond acceptor groups in the receptor increase binding 884
