Mol Divers (2008) 12:65–76 DOI 10.1007/s11030-008-9077-9
FULL LENGTH PAPER
Pharmacophore mapping of flavone derivatives for aromatase inhibition Shuchi Nagar · Md Ataul Islam · Suvadra Das · Arup Mukherjee · Achintya Saha
Received: 31 August 2007 / Accepted: 20 April 2008 / Published online: 28 May 2008 © Springer Science+Business Media B.V. 2008
Abstract Aromatase, which catalyses the final step in the steroidogenesis pathway of estrogen, has been target for the design of inhibitor in the treatment of hormone dependent breast cancer for postmenopausal women. The extensive SAR studies performed in the last 30 years to search for potent, selective and less toxic compounds, have led to the development of second and third generation of non-steroidal aromatase inhibitors (AI). Besides the development of synthetic compounds, several naturally occurring and synthetic flavonoids, which are ubiquitous natural phenolic compounds and mediate the host of biological activities, are found to demonstrate inhibitory effects on aromatase. The present study explores the pharmacophores, i.e., the structural requirements of flavones (Fig. 1) for inhibition of aromatase activity, using quantitative structure activity relationship (QSAR) and space modeling approaches. The classical QSAR studies generate the model (R 2 = 0.924, Q 2 = 0.895, s = 0.233) that shows the importance of aromatic rings A and C, along with substitutional requirements in meta and para positions of ring C for the activity. 3D QSAR of Comparative Molecular Field 2 = 0.791) and ComparAnalysis (CoMFA, R 2 = 0.996, Rcv ative Molecular Similarity Analysis (CoMSIA, R 2 = 0.992, 2 = 0.806) studies show contour maps of steric and hydroRcv phobic properties and contribution of acceptor and donor of the molecule, suggesting the presence of steric hindrance due to ring C and R -substituent, bulky hydrophobic substitution in ring A, along with acceptors at positions 11, and α and γ of imidazole ring, and donor in ring C favor the inhibitory activity. Further space modeling (CATALYST ) study (R = 0.941, cost = 96.96, r msd = 0.876) adjudge the presence of hydrogen bond acceptor (keto functional group), S. Nagar · M. A. Islam · S. Das · A. Mukherjee · A. Saha (B) Department of Chemical Technology, University of Calcutta, 92, A.P.C. Road, Kolkata 700009, India e-mail:
[email protected]
hydrophobic (ring A) and aromatic rings (steric hindrance) along with critical distance among features are important for the inhibitory activity. Keywords Aromatase inhibitor · Flavonoids · CoMFA · CoMSIA · QSAR · Space modeling
Introduction Sex hormones play a crucial role in living organisms. Apart from the physiological role, these hormones can be involved in pathophysiological processes, such as tumorgenesis and growth. About 80% of all prostate cancers and 62% of all breast cancers are sex hormone dependent [1, 2]. These allow the application of an endocrine therapy with more favorable active profile and lesser side effect compared to unspecific chemotherapy. Breast carcinoma is the most common form of female cancer and represents the leading cause of death among women [3]. In hormone-dependent breast cancer, estrogen is the proximate regulator of cell proliferation [4]. Since high serum level of estrogen favors progression of breast cancer; the two main pharmacological strategies have been employed to control or block the pathological activities of estrogen [5]: (i) development of drugs that act through the estrogen receptor (ER) antagonists—this group includes the selective estrogen receptor modulators (SERMs), and (ii) development of drugs that interfere with the synthesis of steroid hormones by inhibiting the enzymes controlling the inter-conversion from androgenic precursors, i.e., aromatase inhibitors (AIs). Today, SERMs and P450 AIs constitute important therapeutically weapons in the fight against breast cancer deaths [6]. The multienzymatic complex, formed by cytochrome P450 and NADPH-cytochrome P450 reductase, catalyses the
123
66
Fig. 1 General structure of flavonoid scaffold
conversion of androgen to estrogen through the aromatization of the ring A of androgen substrate and has been considered a particularly attractive target for the treatment of hormone-dependent breast cancer. P450 AIs including steroidal enzyme inactivators and non-steroidal inhibitors are currently in clinical use [7]. Steroidal AIs compete with the endogenous ligands, androstenedione and testosterone for the active site of intermediates that bind irreversibly to the active site; whereas non-steroidal AIs bind reversibly with high affinity to the heme region of the active site [8]. Besides the development of synthetic compounds, the potential of various classes of natural products to inhibit aromatase are evaluated in order to discover novel breast cancer chemopreventive agents. As a result, several naturally occurring and synthetic flavonoids, which are ubiquitous natural phenolic compounds and mediate a host of biological activities [9], are found to demonstrate inhibitory effects on aromatase [10, 11]. Flavonoids, popularly known as phytoestrogens, are found in more than 300 plants and are widely used in human nutrition [12]. These compounds are consumed in substantial amounts from dietary sources, food supplements and more recently as ‘nutraceuticals’. It has been demonstrated that 7-methoxyflavonone is a potent AI [13] and an effective antiproliferative agent against MCF-7 breast cancer cell [14], whereas other synthetic AIs increase the risk of coronary artery disease (CAD), vaginal dryness, dysparuenia and loss of interest in sex more frequently [15]. Hence novel approaches for targeting these considerable side effects in order to reduce cardiovascular morbidity have become extremely essential [16]. The Zutphen Elderly Study [17] found that increase in consumption of flavonoids resulted in decrease mortality from CAD and incidence of myocardial interaction, which confirm that flavonoid should represent a milestone in next generation safe aromatase inactivation therapy. QSAR studies are mathematical methodologies, statistically validated and mostly used to correlate experi-
123
Mol Divers (2008) 12:65–76
mental or calculated properties derived from chemical structures with biological activities. With the advent of 3D molecular space modeling, a pharmacophore hypothesis can visualize the potential interaction between the ligand and the receptor. A pharmacophore is a set of functional groups/fragment types in a 3D spatial arrangement that represents the interaction made in a common scaffold by a set of small molecular ligands with a protein receptor [18]. The pharmacophore concept is based on the kinds of interaction observed in molecular recognition and alternatively can be used as a query in a 3D database search to identify new structural classes of potential lead compounds; and it can serve as a template for generating alignment for 3D QSAR analysis [19]. These approaches may be applied to predict the activity values of non-synthesized compounds of the test set as well as give an idea on the interaction that would exist between the ligand and the receptor. The aim of the present study is to investigate the structural requirements of flavone derivatives for inhibition of selective aromatase activity with a view to deduce the active pharmacophore signals based on receptor-independent hypothesis, by using both QSAR and space modeling approaches.
Materials and methods In the present work, 33 compounds belonging to the category of flavonoid (Fig. 1) have been considered (Table 1) for modeling aspect through QSAR and space modeling approaches. The P450 19 (CYP19) inhibitory activity (IC50 , µM) of these compounds [20, 21] have been considered as biological activity and implemented as logarithmic function, pIC50 (log10 1000/IC50 ) for QSAR modeling purposes. The primary objective of the work undertaken was to generate relationships between structure and corresponding activity through Multiple Linear Regression (MLR) [22] and Partial Least Square (PLS) [23, 24] methods to deduce a pharmacophore map through receptor-independent modeling techniques. QSAR study 3D structure of molecules are minimized in MOPAC using Austin Model 1 (AM1) method to locate their global minima conformers and subsequent calculation of different molecular properties, such as physiochemical, electronic, topological, spatial and structural features are estimated for classical modeling. The partial charge is calculated using the Extended Huckel approach [25], and E-state indices [26] of all the atoms are generated using a JAVA-based program [27], while other descriptors are generated using Chem3D Pro 5.0 [25], CAChe [28] and TSAR 3.3 [29]. MLR is performed using standard and forward stepwise regression techniques [30].
Mol Divers (2008) 12:65–76
67
Table 1 Structural features and inhibitory activity of flavone (A) and flavonone (B) derivatives
A (Comps 1-9)
B (Comps 10-33)
Comp R1
R2
R3
R4
R5
R
SMILES
IC50 (µM) [20, 21]
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
H H H H H H H H H H OH OCH3 H H OH OH OCH3 OH OCH3 H H H OCH3 OCH3 OCH3 H OCH3 H H H H H H
H NO2 Br CN H NO2 Br CN OCH3 H H H OH OCH3 OH OCH3 OH H H OCH3 H H OCH3 H OCH3 OCH3 OCH3 Br Cl F CN N(CH3 ) H
H H H H H H H H H H H H H H H H H OH H H OCH3 H H OCH3 H OCH3 OCH3 H H H H H H
H H H H H H H H H H H H H H H H H H H H H OCH3 H H H H H H H H H H H
OCH3 OCH3 OCH3 OCH3 H H H H H OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3 OCH3
COc1ccc2c(c1)OC(=C(Cn1ccnc1)C2=O)c1ccccc1 COc1ccc2c(c1)OC(=C(Cn1ccnc1)C2=O)c1ccc(cc1)[N+]([O-])=O COc1ccc2c(c1)OC(=C(Cn1ccnc1)C2=O)c1ccc(cc1)Br COc1ccc2c(c1)OC(=C(Cn1ccnc1)C2=O)c1ccc(cc1)C#N O=C1C(=C(Oc2ccccc21)c1ccccc1)Cn1ccnc1 [O-][N+](=O)c1ccc(cc1)C1=C(Cn2ccnc2)C(=O)c2ccccc2O1 Brc1ccc(cc1)C1=C(Cn2ccnc2)C(=O)c2ccccc2O1 O=C1C(=C(Oc2ccccc21)c1ccc(cc1)C#N)Cn1ccnc1 COc1ccc(cc1)C1=C(Cn2ccnc2)C(=O)c2ccccc2O1 COc1ccc2c(c1)O[C@@H](CC2=O)c1ccccc1OC COc1ccc2c(c1)O[C@@H](CC2=O)c1cccc(c1)O COc1ccc2c(c1)O[C@@H](CC2=O)c1cccc(c1)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)O COc1ccc(cc1)[C@@H]1CC(=O)c2ccc(cc2O1)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(c(c1)O)O COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(c(c1)O)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(c(c1)OC)O COc1ccc2c(c1)O[C@@H](CC2=O)c1cc(cc(c1)O)O COc1ccc2c(c1)O[C@@H](CC2=O)c1cccc(c1OC)OC COc1ccc(c(c1)OC)[C@@H]1CC(=O)c2ccc(cc2O1)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1cc(ccc1OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1c(cccc1OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(c(c1)OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1cc(cc(c1)OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(c(c1OC)OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1cc(c(cc1OC)OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1cc(c(c(c1)OC)OC)OC COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)Br COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)Cl COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)F COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)C#N COc1ccc2c(c1)O[C@@H](CC2=O)c1ccc(cc1)N(C)C COc1ccc2c(c1)OC(=C(Cn1ccnc1)C2=O)c1ccccc1
0.550 0.470 4.100 1.800 0.071 0.045 0.440 0.069 0.080 6.600 3.500 11.468 3.700 11.415 2.500 6.200 5.400 3.500 6.600 8.000 19.011 19.455 11.710 15.674 4.800 17.064 14.749 12.136 10.570 11.820 13.054 30.120 8.000
H H H H H H H H H OCH3 H H H H H H H H OCH3 OCH3 OCH3 OCH3 H H OCH3 OCH3 H H H H H H H
QSAR model origination is accomplished by correlation analysis, and statistical parameters of the regression equation considered are: R (correlation coefficient), EV (explained variance is a statistical parameter defined as adjusted R 2 multiplied by hundred and is expressed in terms of per2 × 100 = [{(1 − R 2 )(n − 1)}/(n − k − centage, Radjusted 1)] × 100.), F (variance ratio—the ratio of mean square of regression to mean square of errors), df(degree of freedom) and s(standard error of estimate, which can be described as
standard deviation of the observed values to the predicted values). Leave-one-out (LOO) cross-validation [31] is also performed to generate statistical parameters, such as PRESS (predictive residual sum of squares, a good prediction of error of the model if the observations are independent), SDEP (standard deviation of error of predictions that portrays the uncertainty of prediction on LOO cross-validation) and Q 2 (cross-validated variance, signifies good predictive performance of the QSAR model).
123
68
The contour maps of flavone for CYP19 inhibitory activity are also derived from the Comparative Molecular Field Analysis (CoMFA) [32, 33] and Comparative Molecular Similarity Analysis (CoMSIA) [34, 35] studies. These also generate statistical parameters, such as R 2 (correlation coeffi2 (crosscient of PLS analysis without validation), Rcv 2 validated correlation coefficient by LOO method) and Rbs (bootstrapped correlation coefficient). Bootstrapping is a procedure in which n random selection out of the original set of n objects are performed 100 times to stimulate different samplings from a larger set of objects. In each run some objects may not be included in the PLS analysis (same method to determine the Q2 ), whereas some other might be included more than once. Confidence intervals for each term can be estimated from such a procedure, giving an independent measure of the stability of the PLS method [36, 37]. The 3D-QSAR models permit an understanding of steric (s) [38], electrostatic (e) [38], and lipophilic (p) requirements for ligand binding. As a consequence, the structural variation in the training set that give rise to variation in the molecular fields at particular regions of the space are correlated to the biological properties. In both cases conformers are generated by simulated annealing technique [39]. The molecules are heated at 700 K for 1000 fs and annealing is done at 200 K for 1000 fs. In the study of CoMFA, fields are generated using both steric (s) and electrostatic (e) interaction and are calculated [38] on a regular space grid of 3 Å. Values of the fields are truncated at 30.0 kcal/mol. Partial atomic charges are calculated by the Gasteiger-Huckle method [40, 41] and energy minimization are performed using the Tripos force field [41, 42] method. In case of CoMSIA, hydrophobicity (p), steric (s) and electrostatic (e) fields along with hydrogen bond acceptor (a) and donor (d) factors are considered. Both CoMFA and CoMSIA provide presentable models when database alignment of molecular conformer is done on a template. In the present study, the most active compound, comp 6 is used as template and points of alignment are indicated as shown in Fig. 2. PLS approach [23, 24], an extension of MLR is used to derive the 3D QSAR models in which field (CoMFA) and similarity (CoMSIA) factors are the independent variables and inhibitory activity (pIC50 ) being the dependent variable. Pharmacophore mapping Groups of compounds (n = 33) containing the flavonoid scaffold have been taken up for receptor-independent pharmacophore space modeling studies with regards to selective CYP19 inhibition. The compounds in the training set are subjected to internal strain energy minimization and conformational analysis (maximum number of conformers = 250, generation type: best quality, spacing = 300, energy range = 20 kcal/mol above the calculated global minimum). Using
123
Mol Divers (2008) 12:65–76
Fig. 2 Atoms of Comp 6 used for alignment, marked as *
the hypogen algorithm of CATALYST [43], the chemical features optimized for exploring the spatial pharmacophore map of this group of compounds are hydrogen bond (HB) acceptor (a) and donor (d), hydrophobic (p), and ring aromatic (r). The pharmacophore model (hypothesis), generated by CATALYST, consist of an array of features necessary for bioactivity of the ligands arranged in 3D space that can explain the variance in activity of the molecules w.r.t. geometric localization of the chemical features present in them. Hypotheses are scored based on errors in activity, estimates from regression and complexity. The optimization involves variation of features and/or locations to optimize activity prediction through simulated annealing approach [39]. To be retrieved as a hit, a candidate ligand must possess appropriate functional groups, which can simultaneously reside within the respective tolerance spheres of the pharmacophoric features. In the generated hypothesis, each feature signifies some degree of magnitude of the compound’s activity. The level to which this magnitude is explored by the hypothesis generator is controlled by the weight variation parameter [43]. This is varied from 0.302 to 2. The uncertainty parameter reflects the error of prediction and denotes the standard deviation of a prediction error factor, called the error cost. In the present work, the uncertainty value of 2 has been considered. While generating hypothesis, a total cost function is minimized comprising of three terms, viz. weight cost, error cost and configuration cost. Weight cost is a value that increases as the weight variation in the model deviates from an ideal value of 2. Error cost is value that increases as the rms (root mean square) difference between estimated and measured activities of the training set molecules increases. This cost factor is designed to favor models for which the correlation between estimated and measured activities is better. The standard deviation of this parameter is given by the uncertainty parameter. A fixed cost that depends on the complexity of the hypothesis space being optimized, is denoted as
Mol Divers (2008) 12:65–76
69
the configuration cost. The configuration cost is equal to the entropy of hypothesis space. The CATALYST program [43] also calculates the cost of a null hypothesis and fixed hypothesis. The costs of these two are represented as null cost and fixed cost. The greater the difference (cost) between the fixed and the null costs, it is more likely that the hypothesis does not reflect a chance correlation. The minimum difference between the total and null costs is taken as 60 bits for a hypothesis optimization [44]. The HypoGen algorithm is forced to find pharmacophore model that contains maximum two same feature and others from the input chemical features. The quality of the generated hypothesis is further adjudged through a cross-validation technique using CatScramble [43]. The validation procedure is based on Fischer’s randomization test [22], where the biological activity data are randomized within a fixed chemical data set and the HypoGen process initiated to explore possibilities of other hypotheses of good predictive values. By logic, the hypothesis generated prior to scrambling should be better to attest for a good pharmacophore model. IC50 to CYP19 inhibition has been considered as biological activity for this study. Result and discussion QSAR modeling A structure-activity relationship has been drawn, investigating physiochemical (partition coefficient, hydrophobicity, steric, and moments), electronic (atomic and partial charge functions and orbital energies), and electrotopological (E-state indices) features of the molecule for characterization of unique SAR model. In the regressional model, the 95% confidence intervals are shown in parentheses and the F-values are significant at the 99% confidence level. The regression constant for the relation is significant at 95%. While classical modeling on CYP19 inhibition of flavonoid derivatives (Fig. 1), the best single variate model for inhibitory activity could be developed with the Huckle charge [25] of atom C5 (H C5 ) that explained 85.2% variance in activity, and the statistical quality of the relation is estimated to be R 2 = 0.857, s = 0.228, n = 33 and in the case of a bivariate relationship, the best significant relationship has been explored with same atomic charge functions of C5 and C16 , that explained 87.2% variance in activity. The quality of this relationship has been estimated to be R 2 = 0.880, s = 0.216, n = 33. But the best significant relationship for inhibition of CYP19 has been deduced to be pIC50 = 20.328(±1.495)HC 1 − 0.774(±0.209)S12 +0.310(±0.146)I15−OH − 0.284(±0.129)Im−OMe +5.252(±0.222)
(1)
where, H C1 is Extended Huckle atomic partial charge on C1 and S12 is E-state indices [26] at atom C12 . I15−OH and Im−OMe indicate the presence of hydroxy and methoxy groups in para and meta positions in ring C respectively. The independent variables use in the model (Eq. 1) are not intercorelated (R < 0.5). Comp. 1 has been found as an outlier while generating the model and is not considered in the training set. The statistical quality of the model is R 2 = 0.924, E V = 91.30%, F = 82.350, d f = 4, 27, s = 0.233, P R E SS = 2.304, S D E P = 0.252, Q 2 = 0.895, n = 32. The model (Eq. 1) accounts for more than 90% variance in observed activity with cross-validated variance (CVV) [31] of 90%. It reveals the importance of atoms C1 and C12 , and presence of hydroxy and methoxy groups at positions C15 and C14 respectively of the molecule for inhibitory activity to the aromatase CYP19. According to the model, the presence of positive charge function at atom C1 favors inhibitory activity to CYP19. Further, the presence of aromatic ring C attach to the atom C8 of flavone scaffold imparting activity. The positive contribution of the indicator,I15−OH signifies the presence of hydrogen bond donor at C15 atom increases inhibitory activity [45], whereas negative contribution of the indicator, Im−OMe indicates that the more nucleophilic substitution at meta position in ring C decreases the activity. Thus, substitution in ring A of flavone scaffold that imparts hydrophobicity of the molecule [46], along with aromatic attachment, containing hydrogen bond donor, at C8 of scaffold provide increased CYP19 inhibitory activity. The calculated and predicted (LOO cross-validation) activities as per Eq. 1 are depicted in Table 2 and Fig. 3. CoMFA and CoMSIA techniques are also adopted to derive 3D QSAR models. The statistical results obtained from the database alignment done on template are tabulated in Tables 3 and 4 respectively. CoMFA is generated with steric (s) and electrostatic (e) factors. From Table 3 of field analysis, it is observed that alignment for steric factor produces a cross-validated R 2 of 0.785, non-cross-validated R 2 of 0.998 and bootstrap R 2 of 0.999. But there is no contribution of electrostatic field. Further it is seen that inclusion of hydrophobicity (AlogP) [47] with steric factor explain crossvalidated R 2 of 0.791, non-cross-validated R 2 of 0.996 and bootstrap R 2 of 0.998. The model shows that the contribution of steric and hydrophobicity are 93.6% and 6.4% respectively for the activity. The mapped features are illustrated in Figure 4 and activities are described in Table 2. The selection of CoMFA model is based on maximum 2 and F-values. The steric fields, the green (sterically Rcv favorable) and yellow (sterically unfavorable) contours represent 80% and 20% level of contributions. AlogP, an additional descriptor externally incorporated in 3D QSAR, is not mapped in Fig. 4, but contribute 6.4% of the activity (Table 3). These 3D contour maps show that the changes of molecular fields are associated with the differences of biological
123
70
Mol Divers (2008) 12:65–76
Table 2 Calculated and predicted inhibitory activity to CYP19 Comp
Activity (pIC50 ) Obs. (pIC50 )
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
3.260 3.328 2.387 2.745 4.149 4.347 3.356 4.161 4.097 2.180 2.456 1.940 2.432 1.943 2.602 2.208 2.268 2.456 2.180 2.097 1.721 1.711 1.931 1.805 2.319 1.768 1.831 1.916 1.976 1.927 1.884 1.521 2.097
Activity (IC50 ) Eq. 1
CoMFA model
CoMSIA model
Space modeling study
Cal.
Pred.
Cal.
Cal.
Obs.
Cal.
– 3.027 2.663 2.806 3.907 4.263 3.920 4.042 3.966 2.166 2.131 1.933 2.294 1.894 2.603 2.202 2.404 2.440 2.018 1.979 1.734 2.107 1.983 1.758 2.088 1.824 1.808 1.847 1.913 2.101 1.969 2.121 1.823
2.396 2.923 2.677 2.812 3.838 4.235 4.076 4.014 3.934 2.165 2.112 1.932 2.237 1.888 2.404 2.202 2.493 2.436 2.008 1.971 1.738 2.130 1.986 1.742 2.074 1.844 1.801 1.837 1.906 2.111 1.976 2.159 1.777
3.220 3.315 2.332 2.840 4.009 4.320 3.480 4.138 4.127 2.128 2.376 1.999 2.434 1.987 2.557 2.256 2.296 2.585 2.276 2.150 1.706 1.711 1.746 1.810 2.327 1.686 1.805 1.913 1.904 1.943 1.955 1.558 2.089
3.327 3.142 2.349 2.934 4.145 4.177 3.431 4.184 4.184 2.100 2.494 2.063 2.437 1.923 2.619 2.210 2.325 2.382 2.124 2.029 1.757 1.831 1.798 1.879 2.272 1.746 1.855 2.023 1.887 1.849 1.823 1.650 2.029
0.550 0.470 4.100 1.800 0.071 0.045 0.440 0.069 0.080 6.600 3.500 11.468 3.700 11.415 2.500 6.200 5.400 3.500 6.600 8.000 19.011 19.455 11.710 15.674 4.800 17.064 14.749 12.136 10.570 11.820 13.054 30.120 8.000
0.410 0.340 3.400 1.900 0.110 0.100 0.100 0.110 0.110 7.500 9.800 8.100 10.000 9.300 9.200 9.200 8.200 8.300 7.900 7.700 8.800 9.600 8.000 10.000 7.800 9.400 8.700 10.000 10.000 9.600 9.300 11.000 9.800
Obs. = Observed; Cal. = Calculated; Pred. = Predicted on LOO cross-validation; pIC50 = log (1000/IC50 )
Fig. 3 Observed, calculated and predicted values as per Eq. 1
123
Mol Divers (2008) 12:65–76
71
Table 3 Summary of CoMFA result
Components n R2 s F-value (df) 2 Rcv scv F-value (df) 2 Rbs Standard deviation Contribution Steric Electrostatic AlogP
Steric (s)
Steric + Electrostatic (s+e)
Steric + Hydrophobicity (s+AlogP)
3 33 0.998 0.042 1875.155 (6, 26) 0.785 0.385 15.862 (6, 26) 0.999 0.001
3 33 0.998 0.042 1875.155 (6, 26) 0.785 0.385 15.862 (6, 26) 0.999 0.001
3 33 0.996 0.056 1054.216 (6, 26) 0.791 0.380 16.450(6, 26) 0.998 0.001
1.000 – –
1.000 no contribution –
0.936 – 0.064
2 = Cross-validated R 2 by LOO method Rcv scv = Standard error on LOO validation R 2 = Non-cross validated R 2 s = Standard error of estimate 2 = Bootstrap R 2 Rbs
Table 4 Summary of CoMSIA results
Components n R2 s F-value (df) 2 Rcv scv F-value (df) 2 Rbs Standard deviation Contribution a p s d
a
p
s
a+p+s
a+p+d+s
1 33 0.959 0.177 102.610 (6, 26) 0.415 0.616 1.524 (6, 26) 0.975 0.014
3 33 0.979 0.126 205.639 (6, 26) 0.824 0.349 20.248 (6, 26) 0.990 0.005
3 33 0.957 0.161 149.048 (5, 27) 839 0.334 22.582 (6, 26) 0.973 0.007
3 33 0.992 0.074 912.938 (4, 28) 0.806 0.383 25.991 (4, 28) 0.996 0.002
6 33 0.992 0.078 552.752 (6, 26) 0.799 0.398 17.199 (6, 26) 0.997 0.049
1.000 – – –
– 1.000 – –
– – 1.000 –
0.261 0.486 0.252 –
0.181 0.469 0.233 0.117
a: Acceptor, p: Hydrophobic, s: Steric, d: Donor 2 = Cross-validated R 2 by LOO method Rcv scv = Standard error on LOO validation R 2 = Non-cross validated R 2 s = Standard error of estimate 2 = Bootstrap R 2 Rbs Fig. 4 Mapped features of CoMFA study fitted with Comp 6 (a) Steric: Green favorable, yellow unfavorable (b) Steric and AlogP: Green favorable, yellow unfavorable
123
72
Mol Divers (2008) 12:65–76
Fig. 5 Space mapped features of CoMSIA study fitted with Comp 6 (a) Acceptor: Magenta favorable, red unfavorable (b) Hydrophobic: Purple favorable, cyan unfavorable (c) Steric: Green favorable, yellow unfavorable (d) Acceptor: Magenta favorable, red unfavorable; hydrophobic: Purple favorable, cyan unfavorable; steric: Green favorable, yellow unfavorable
activity. The regions of green contour suggest that more bulky substituents in these positions will improve the biological activity, while the yellow regions indicate that an increased steric bulk is unfavorable for the inhibitory activity. The presence of imidazole ring as R substituent in flavone scaffold (Fig. 1) provides more steric influence on the molecule. Thus the compds 1–9 having greater inhibitory activity than rest of the compounds of the training set. Additionally, substitutions in ring C that impart steric hindrance on that region favor inhibitory activity. But similar substitution in ring A unfavors the activity. From 2D QSAR study (Eq. 1), it is observed that methoxy (nucleophilic) substituent in ring C decreases the activity, but the same in ring A imparts more hydrophobicity that increases inhibitory activity. Additionally the presence of heteroatoms at 7 and 11 positions fall into a positive favorable green region. The O11 may behave as hydrogen bond acceptor that possibly binds with the Ser 478 of the enzyme [48], suggesting the key role for imparting activity, adjudged by CoMSIA and space modeling studies. Again the steric (s), hydrogen bond (HB) acceptor (a) and donor (d), and hydrophobic (p) features are considered for similarity analysis studies (Table 4). Electrostatic (e) is not taken into consideration, as it has no contribution in CoMFA study. It is observed that cross-validated R 2 of 0.415, non-cross validated R 2 of 0.959 and bootstrap R 2 of 0.975 are obtained with HBa as descriptor. Cross-validated R 2 of 0.824, non-cross validated R 2 of 0.979 and bootstrap R 2 of 0.990 are observed with lipophilic factor. When steric field is considered, cross-validated R 2 of 0.839, non-cross-validated R 2 of 0.957 and bootstrap R 2 of 0.973 are the statistical out2 (Table put. In combination of a, d, p and s fields give good Rcv 4) and explained variance of 18.1, 11.7, 46.9 and 23.3% of activity respectively. But the best model has been derived
123
with a, p and s fields that contributes 26.1%, 48.6% and 25.2% respectively for the activity. The results show cross-validated R 2 of 0.806, non-cross-validated R 2 of 0.992 and bootstrap R 2 of 0.996. The mapped features are illustrated in Fig. 5 and activities are delineated in Table 2. The contour maps of CoMSIA models are depicted features in Figs. 5a–d of comp 6. In Fig. 5a, the CoMSIA model showing 20% magenta favorable region and 80% red unfavorable region for acceptor. O11 falls into the favorable region that is adjudged by the CoMFA model. Further the presence of acceptor in γ position of imidazole ring (Table 1A) favors inhibitory activity. But the presence of acceptor at C15 position in ring C unfavors for activity, which is further adjudged by Eq. 1, where presence of a donor on the same position increases activity. In Fig. 5b, the model describes 80% purple favorable region and 20% blue unfavorable region for hydrophobicity, suggesting that the bulky hydrophobic substituent in ring A is important for the activity. Additionally presence of suitable substituent at α position of imidazole ring that can impart hydrophobic region on that position may bring increased activity. But the substitution on connecting vicinity of B, C and imidazole rings may decrease activity. Figure 5c shows similar pattern to CoMFA model where 80% green favorable and 20% yellow unfavorable regions are pictured for steric factor. Imidazole and C rings are showing as promising steric regions. Additionally ring A also behaves as favorable for steric influence. But steric substituent at C2 in ring A and connecting vicinity of imidazole and B rings are unfavorable for activity. Figure 5d describes the combined contours map of a, p and s influence, suggesting the presence of (i) acceptors at positions 11, α and γ of imidazole ring, (ii) hydrophobic substituent at C2 , and (iii) steric hindrance due to imidazole and C rings increase inhibitory activity. But the
Mol Divers (2008) 12:65–76
73
Table 5 Hypothesis parameters observed in successive runs Run no
1 2 3∗ 4 5∗∗ 6∗∗ 7∗∗∗
Input features a, d, p, r a, r, pi a, p, r a, p, r a, p, r a, p, r a, p, r
Uncertainty
3 2 2 2 2 2 2
Weight variance
Hypothesis no
0.302 0.302 0.302 2.000 0.302 0.302 2.000
1 1 1 1 1 1 1
Pharmacophore features in generated hypothesis
Cost Null
Total
cost
a, 2 × r a, 2 × r a, p, 2 × r a, p, 2 × r 2 × a, p, r 2 × a, p, r a, p, r
155.11 206.613 206.613 206.613 206.613 206.613 206.613
131.113 123.348 122.161 122.454 124.868 125.746 121.704
30.041 96.742 97.873 96.963 98.001 97.056 99.003
R
rmsd
0.929 0.939 0.939 0.941 0.924 0.924 0.934
0.603 0.886 0.887 0.876 0.990 0.990 0.923
a: HB acceptor, d: HB donor; p: Hydrophobic factor, r: Ring aromatic; ∗ all features: min = 1, max = 2; ∗∗ for ring aromatic: min = 1, max = 1; ∗∗∗ all input features: min = 0, max = 1; cost = Null cost−Fixed cost; rmsd = rms deviation; Spacing = 300 pm; n = 33
Table 6 Inter feature distance obtained from generated pharmacophore hypothesis Features
a
p
r1
r2
a p r1 r2
0.000
7.394 0.000
4.119 9.259 0.000
7.938 8.070 5.474 0.000
a: Acceptor; p: Hydrophobic; r1: Aromatic ring 1; r2: Aromatic ring 2
presence of (i) acceptors at C15 and β-position of imidazole ring, (ii) hydrophobic substituent in ring C, and (iii) steric influence at C2 and ring B may decrease activity. Space modeling studies The result of the study has been depicted in Tables 5–7. Hypothesis 1 of run no. 4 (Table 5) has been adjudged to be the best hypothesis and demonstrate above 94% correlation with the inhibitory activity to CYP19. The above hypothesis is characterized on the basis of highest cost difference (cost), root mean square deviation (rmsd) and the best correlation (R). The mapped pharmacophore features for inhibitory activity to CYP19 are described in Table 6 and Fig. 6. Hydrogen bond (HB) acceptor (a), hydrophobic (p) and two aromatic rings (r) features might function as prime biophores for activity. The results demonstrate good predictive ability of the model as well. The quality of hypothesis generated (hypothesis 1 of run 4) is further adjudged through a cross-validation technique using Fischer’s randomization test [22] at 99% confidence, where the biological activity data are randomized within a fixed chemical data set and the HypoGen process is initiated to explore possibilities of other hypotheses of predictive values; but none of the hypothesis generated better parameters in comparison to original hypothesis. So, the cross-validation analyses clearly indicate the superiority of the hypothesis considered. The calculated activities of the training set are listed in Table 2 and estimated
fit scores are delineated in Fig. 7. Fit score indicates how well the training compound is mapped to the generated hypothesis. Along with mapping, it also measures the distance that separates function of the molecule from the centroid of the hypothesis. The larger the fit value indicates a better fit. Figure 7 is plotted with fit score vs estimated activity, suggesting the high degree of correlation between these two parameters, indicating a good hypothesis. The projected hypothesis is further validated with some reference compounds (standard and non-standard AIs) by measuring fit scores and estimated values of such compounds are listed in Table 7. Thus, the key pharmacophores of flavone derivatives for inhibition of CYP19 may be deduced to be comprised of HB acceptor (keto functional group) and hydrophobic (A ring) features, and the presence of two aromatic rings (C and imidazole), and the critical inter features distances (Table 6) are crucial for imparting the activity. The space modeling study can also be correlated with the QSAR studies. The ring A along with its substitutional pattern imparting steric hindrance that provide hydrophobic feature of the molecule. Presence of HB-acceptors in B and imidazole rings increasing inhibitory activity. Further substituents in C ring along with presence of imidazole ring (both the rings are aromatic) that bring more steric influences, increase activity.
Conclusion In this study, QSAR models using 2D-QSAR, CoMFA and CoMSIA, and space modeling approaches are adopted to rationalize the P450 19 (CYP 19) inhibitory activity of the compounds belonging to flavonoid. The classical QSAR model, obtained from electronic and electrotopological properties of the molecule, shows good correlation with biological activity and also has predictive ability. Similarly 3D QSAR of CoMFA and CoMSIA studies depict well predictive models with high correlation with biological activity. A high
123
74
Mol Divers (2008) 12:65–76
Table 7 Fit score and estimated activity of some reference (standard and non-standard AI) compounds Comp
Asigenin Cyclofenil Em800 [49] Ephedrine E2 Gefitinib Genistein Hydroxyflavone Hydroxytamoxifene Hydroxytrioxifene Hydroxytoremifene Idoxifene Lasofoxifene Letrozole Nafoxidine Norgestrel Qurecetin Stanozolol Simvastatin Tangeretin Tetrahydroisoqunoline Testosterone Toremifene Trioxifene Tse-424 [50]
Estimated IC50 (µM) Space modeling
CoMFA model
CoMSIA model
3.10×10−1 1.70 × 104 6.10×10−2 1.40 × 104 1.10 × 104 2.90 4.20×10−1 1.70 1.32×10−2 9.60×10−3 1.30 × 102 2.80×10−2 2.10×10−2 1.10×10−2 0.19 × 102 1.10 × 104 0.12×10−2 1.20 × 104 1.10 × 104 1.40 2.20×10−2 1.10 × 104 2.20 × 102 1.23 × 102 3.40×10−3
1.65×10−1 8.15 6.34×10−2 2.49×10−2 4.73×10−1 3.46 1.22×10−1 3.81 2.07 16.40 74.80 1.09 × 102 68.70 4.32×10−2 0.36 × 102 2.09 0.72×10−2 3.25 10.10 1.07 5.31 1.67×10−1 1.07 × 102 25.90 4.82×10−3
8.41×10−1 7.45 6.82×10−2 1.66 2.14 6.08 3.61×10−1 4.06 5.66 5.08 10.50 3.87 1.10 1.23×10−2 0.38 × 102 1.10 0.49×10−2 3.40 2.92 4.59 1.53 2.40 12.60 5.42 4.30×10−3
Fig. 6 Pharmacophore features for inhibitory activity to CYP19 (hypothesis 1 of run no 4). Mapped features are HB acceptor (a), hydrophobic (p) and ring aromatic (r). Comp 6 (most active compound in the training set) is fitted in the mapped pharmacophore features
bootstrap R 2 value and small standard deviation indicate a similar relationship exists in all compounds. Inclusion of AlogP, a lipophilic parameter improves the significance of CoMFA model indicating the role of lipophilicity in ligand– receptor interaction. 3D QSAR contour maps show good compatibility with the receptor properties even though the conformations and ligand are not based on the receptor structure. Similarly space-modeling study confirms the QSAR findings and indicates the critical distances between explored
123
features in 3D geometric space are important for inhibitory activity. After the analysis of these models, it is established that presence of imidazole ring and substitutional pattern in ring C that bring steric influences on that regions increase CYP19 aromatase inhibitory activity. Further presence of acceptor at C10 and appropriate position in imidazole ring, and donor at C15 in ring C increase activity. Bulky substituent in ring A that provides steric hindrance, imparting hydrophobicity of the molecule, increasing inactivation of the enzyme.
Mol Divers (2008) 12:65–76
75
16. 17.
18.
19. 20. Fig. 7 Estimated activity vs fit score of the training set 21. Acknowledgements Financial support from DST FAST track and University TEQIP schemes are thankfully acknowledged. One of the authors S Nagar thanks Lady Tata Memorial Trust for providing her Senior Research Fellowship.
22. 23.
24.
References
25.
1. Hartmann RW, Ehmer PB, Haidar S et al (2002) Inhibition of CYP 17, a new strategy for the treatment of prostate cancer. Arch Pharm 335:19–128 2. Fobes JF (1997) The control of breast cancer: the role of tamoxifene. Semin Oncol 24:S1 15–S1 19 3. Brueggemeier RW (2002) Aromatase inhibitors in breast cancer therapy. Anticancer Ther 2:181–191 4. Brueggemeier RW, Richards JA, Joomprabutra S et al (2001) Molecular pharmacology of aromatase and its regulation by endogenous and exogenous agents. J Steroid Biochem Mol Biol 79:75–84 5. Banting L, Nicholls PJ, Shaw MA, Smith HJ (1989) Recent developments in aromatase inhibition as a potential treatment for oestrogen-dependent breast cancer. Prog Med Chem 26:253–298 6. Pasqualini JR (2004) The selective estrogen enzyme modulators in breast cancer: A review. Biochem Biophys Acta 21:23–43 7. Miller WR (2003) Aromatase Inhibitor: mechanism of action and role in the treatment of breast cancer. Semin Oncol 30:3–11 8. Howell A (2005) New developments in the treatment of postmenopausal breast cancer. Trends Endocrin Metab 9:420–428 9. Carlo DG, Mascolo N, Izzo AA, Capasso F (1999) Flavonoids: old and new aspects of a class of natural therapeutic drugs. Life Sci 49:337–353 10. Kao YC, Zhou C, Sherman M et al (1998) Molecular basis of the inhibition of human aromatase (estrogen synthetase) by flavone and isoflavone phytoestrogens: a site-directed mutagenesis study. Environ Health Perspect 106:85–92 11. Jeong HJ, Shin YG, Kim IH, Pezzuto JM (1999) Inhibition of aromatase activity by flavonoids. Arch Pharm Res 22:309–312 12. Jordan VC, Mittal S, Gosden B et al (1985) Structure–activity relationships of estrogens. Environ Health Perspect 61:97–110 13. Le Bail JC, Laroche T, Marre-Fournier F, Habrioux G (1998) Aromatase and 17β-hydroxysteroid dehydrogenase inhibition by flavonoids. Cancer Let 133:101–106 14. Pouget C, Lauthier F, Alain Simon A et al (2001) Flavonoids: structural requirements for antiproliferative activity on breast cancer cells. Bioorg Med Chem Let 11:3095–3097 15. Fallowfield L, Cella D, Cuzick J et al (2004) Quality of life of postmenopausal women in the arimidex, tamoxifen, alone or in
26. 27. 28.
29. 30. 31.
32.
33.
34.
35.
36.
37. 38. 39. 40.
combination (ATAC) adjuvant breast cancer trial. J Clin Oncol 21:4261–4271 Montana JG, Dyke HJ (2002) Update on the therapeutic potential of PDE4 inhibitors. Expert Opin Invest Drugs 11:1–13 Hertog MGL, Feskens EJM, Kromhout D et al (1993) Dietary antioxidant flavonoids and risk of coronary heart disease: the Zutphen Elderly Study. Lancet 342:1007–1011 Kristam R, Gillet VJ, Lewis RA, Thorner D (2005) Comparison of conformational analysis techniques to generate pharmacophore hypotheses using catalyst. J Chem Inf Model 2:461–476 Guner OF (2000) In pharmacophore perception, development, and use in drug design. International University Line, California Gobbi S, Cavalli A, Rampa A et al (2006) Lead optimization providing a series of flavone derivatives as potent nonsteroidal inhibitors of the cytochrome P450 aromatase enzyme. J Med Chem 49:4777–4780 Pouget C, Fagnere C, Basly JP et al (2002) Synthesis and aromatase inhibitory activity of flavanones. Pharma Res 19:286–291 Snedecor GW, Cochran WG (1967) Statistical methods, 6th edn. Iowa State University Press, Iowa Wold S, Albano C (1984) Multivariate data analysis in chemistry. In: Kowalski B (ed) CHEMOMETRICS: mathematics and statistics in chemistry, Reidel Dordrecht Netherlands Hoskuldsson A (1987) PLS regression methods. J Chemometrics 2:211–228 CS Chem3D Pro and CS MOPAC Pro 5.0, CambridgeSoft, Cambridge. www.camsoft.com Kier LB, Hall LH (1990) An electrotopological-state index for atoms in molecules. Pharm Res 7:801–807 ESI-MS (2007) Chemical Technology, Calcutta University, Kolkata. www.caluniv.ac.in CAChe 6.1, Computer-aided chemistry for macintosh and windows, Fujitsu Systems Business of America Inc. Chiba, Japan. www.CACheSoftware.com TSAR 3D 3.3, Oxford Molecular Limited, Accelrys, San Diego. www.accelrys.com Statistica 5.0, StatSoft Inc, Tulsa. www.statsoft.com Wold S, Eriksson L (1995) Chemometric methods in molecular design. In: de Waterbeemd HV (ed) Chemometric methods, vol 2. Weinheim, VCH Verlagsgesellschaft mbh, p 312 Cramer RDIII, Patterson DE, Bunce JD (1989) Recent advances in comparative molecular field analysis (CoMFA). Prog Clin Biol Res 291:161–165 Allen MS, Tan YC, Koehler KF et al (1990) Synthetic and computer-assisted analyses of the pharmacophore for the benzodiazepine receptor inverse agonist site. J Med Chem 33: 2343–2357 Klebe G, Abraham U, Mietzner T (1994) Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J Med Chem 37:4130–4146 Cramer RDIII, Patterson DE, Bunce JD (1988) Comparative molecular field analysis (CoMFA). Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 110: 5959–5967 Cramer RDIII, Bunce JD, Patterson DE et al (1988) Cross-validation, bootstrapping, and partial least squares compared with multiple regression in conventional QSAR and QSPR studies. Quant Struct Act Relat 7:18–25 Wehrens R, van der Linden WE (1997) Bootstrapping principal component regression models. J Chemom 11:157–171 SYBYL 7.3, (2006) Tripos Inc, St. Louis. www.tripos.com Kirkpatrick S, Gelatt CD, Vecchi MPJr (1983) Optimization by simulated annealing. Science 220:671–680 Streitwieser A (1961) Molecular orbital theory for organic chemists. Wiley, New York
123
76 41. Vinter JG, Davis A, Saunders MR (1987) Strategic approaches to drug design. I. An integrated software framework for molecular modeling. J Comp-Aided Mol Design 1:31–51 42. White DNJ (1977) The principles and practice of molecular mechanics calculations. Comp Chem 1:225–233 43. Catalyst 4.11 (2005) Accelrys Inc, San Diego, CA. www.accelrys. com 44. Schuster D, Laggner C, Steindl TM (2006) Pharmacophore modeling and in-silico screening for new P450 19 (aromatase) inhibitors. J Chem Inf Model 46:1301–1311 45. Cavalli A, Bisi A, Bertucci C et al (2005) Enantioselective nonsteroidal aromatase inhibitors identified through a multidisciplinary medicinal chemistry approach. J Med Chem 48:7282–7289 46. Recanatini M, Bisi A, Cavalli A et al (2001) A new class of nonsteroidal aromatase inhibitors: Design and synthesis of chromone and xanthone derivatives and inhibition of the P450 enzymes aromatase and 17-hydroxylase/C17,20-lyase. J Med Chem 44:672–680 47. Viswanadhan VN, Ghose AK, Revankar GR, Robins RK (1989) Atomic physicochemical parameters for three dimen-
123
Mol Divers (2008) 12:65–76 sional structure directed quantitative structure–activity relationships. 4. Additional parameters for hydrophobic and dispersive interactions and their application for an automated superposition of certain naturally occurring nucleoside antibiotics. J Chem Inf Comput Sci 29:163–172 48. Favia AD, Cavalli A, Masetti M et al (2006) Three dimensional model of the human aromatase enzyme and density functional parameterization of the iron-containing protopophyrin IX for molecular dynamics study of heme-cysteinato cytochromes. Proteins 62:1074–1087 49. Gauthier S, Caron B, Cloutier J et al (1997) (S)-(+)-4-[7(2,2-Dimethyl-1-oxopropoxy)-4-methyl-2-[4-[2-(1-piperidinyl)ethoxy]phenyl]-2H-1-benzopyran-3-yl]-phenyl-2,2-dimethylpropanoate (EM-800): a highly potent, specific, and orally active nonsteroidal antiestrogen. J Med Chem 40:2117–2122 50. Miller CP, Collini MD, Tran BD et al (2001) Design, synthesis, and preclinical characterization of novel, highly selective indole estrogens. J Med Chem 44:1654–1657
