This article was downloaded by: [134.117.10.200]On: 30 November 2014, At: 08:25Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

SAR and QSAR in EnvironmentalResearchPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/gsar20

Comparison of Estrogen Receptor αand β Subtypes Based on ComparativeMolecular Field Analysis (CoMFA)L. Xing a , W. J. Welsh a , W. Tong b , R. Perkins b & D. M.Sheehan ca Department of Chemistry and Center for MolecularElectronics , University of Missouri , St. Louis, MO, 63121, USAb R.O.W. Sciences , Jefferson, AR, 72079, USAc Division of Reproductive and Developmental Toxicology ,National Center for Toxicological Research (NCTR) , Jefferson,AR, 72079, USAPublished online: 05 Oct 2006.

To cite this article: L. Xing , W. J. Welsh , W. Tong , R. Perkins & D. M. Sheehan (1999)Comparison of Estrogen Receptor α and β Subtypes Based on Comparative MolecularField Analysis (CoMFA), SAR and QSAR in Environmental Research, 10:2-3, 215-237, DOI:10.1080/10629369908039177

To link to this article: http://dx.doi.org/10.1080/10629369908039177

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to orarising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms& Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

S A H undQSAK in En:nvrrotrmcnioi Rrseurch 1999, Vol 10, pp. 215 237 Reprints available directly from the publisher Photocopying permitted by license only

C! 1999 OPA (Ovcrseas Publishers Association) N.V Published by license under

the Gordon and Rreach Science Publishers imprint

Printed in Malaysia

COMPARISON OF ESTROGEN RECEPTOR (Y

AND p SUBTYPES BASED ON COMPARATIVE MOLECULAR FIELD ANALYSIS (CoMFA)*

L. XING a, w. J. WELSH^^.+, w. TONG h, R. PERK INS^ and D. M. SHEEHAN'

"Department o j Chemistry and Center ,for Moleculur Electronics, University of Missouri - St . Louis, MO 63121 ( U S A ) ;

' R . 0 . W. Sciences, Jefferson, A R 72079 ( U S A ) ; 'Division of Reproductive und Developmental Toxicology, Nutional Center

for Toxicological Research ( N C T R ) , Jeflerson, A R 72079 ( U S A )

(Rer>eivrd 9 July 1998, In final form 24 Ortoher 1998)

A substantial body of evidence indicates that both humans and wildlife suffer adverse health effects from exposure to environmental chemicals that are capable of interacting with the endocrine system. The recent cloning of the estrogen receptor P subtype (ER-,B) suggests that thc selective effects of estrogenic compounds may arise in part by the control of difl'erent subsets of estrogen-responsive promoters by the two ER subtypes, ER-a and ER-P. In order to identify the structural prerequisites for ligand-ER binding and to discriminate ER-a and ER-P in terms of their ligand-binding specificities, Comparative Molecular Field Analysis (CoMFA) was employed to construct a three-dimensional Quantitative Structure-Activity Relationship (3D- QSAR) model on a data set of 31 structurally-diverse compounds for which competitive binding allinitics have been measured against both ER-cu and ER-fl. Structural alignmcnt of the molecules in CvMFA was achieved by maximizing overlap of their steric and electrostatic fields using the Steric and Electrostatic ALignment (SEAL) algorithm. The final CoMFA models, generated by correlating the calculated 3D steric and electrostatic fields with the experimentally observed binding affinities using partial least-squares (PLS) regression, exhibited excellent self- consistency ( r2 > 0.99) as well as high internal predictive ability (q2 > 0.65) based on cross- validation. CoMFA-predicted values of RBA for a test set of cornpounds outside of the training set were consistent with experimental observations. These CoMFA models can serve as guides for the rational design of ER ligdnds that possess preferential binding affinities for either ER-a or ER-Lj. Thcse models can also prove useful in risk assessment programs to identify real or suspected EDCs.

*Presented at the workshop Compututionul Mcthods in Toxicology, April 20-22, 1998,

+Corresponding author. Dayton, OH, USA.

21 5

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

216 L. XING ef ul.

keyword.^ CoMFA; 3D-QSAR; endocrine disrupting compounds; estrogen receptor cy (ER-a); estrogen receptor /3 (ER-I?)

INTRODUCTION

The estrogen receptor (ER) is an intracellular transcriptional factor that exists in inactive apoprotein forms in the cell nucleus [I]. In the absence of a bound ligand, the ER cannot bind to DNA and thus has no transcriptional activity. Upon binding the hormone, ER switches to a transcriptionally active form by undergoing a conformational transformation. Subsequent binding of the ligand-ER complex to the estrogen response element (ERE) of DNA effectively activates transcription of a cis-linked gene [2-41. In addition to regulating transcription, steroid hormones occasionally regulate gene expression by affecting mRNA stability and translational efficiency [2] .

Because of their pervasive influence on health and reproductive function, estrogens have stimulated a substantial amount of research activity in the past two decades. It is believed that the structural changes in estrogen modify the ligand-receptor complex in a manner which either influences its interaction with different EREs or affects the receptor interaction with other proteins required for activation of specific promoters [2 -41. Depending on its chemical structure and target cells, estrogen can exert either positive or negative effects on the development and function of tissues. The beneficial effects are numerous and include bone maintenance and cardiovascular health [S , 61. The potential adverse effects are equally numerous. For example, estrogen stimu- lates the growth of breast cancer and hormone-responsive tumors as well as a decline in the quality and quantity of sperm production [7-91.

In contrast to other members of the nuclear receptor superfamily where multiple forms have been reported, ER was thought to exist in only one form. Then in 1996, a novel ER cDNA was cloned from rat prostate [lo, 1 I]. The newly cloned ER was named P subtype (ER-P) to distinguish it from the previously cloned ER cDNA (subsequently named ER-a). The protein sequences of ER-a and ER-P from rat are highly homologous, particularly in the DNA binding domain ( > 90% amino acid identity) and in the c-terminal ligand-binding domain (LBD) ( 5 5 % ) [ 101. While ER-a expresses high tissue distribution in uterus, testis, pituitary, ovary, kidney, epididymis, and adrenal, ER-/3 is highly concentrated in prostate, ovary, lung, bladder, brain, uterus, and testis [ 121. Assuming that the selective effects of estrogens reside in the control of different subsets of estrogen-responsive promoters by the two ER subtypes, differences between the ER subtypes in terms of their relative ligand-binding affinity and tissue distribution could contribute to the selective

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR a AND SUBTYPES 217

action of ER agonists and antagonists. Although many ligaiids bind with similar relative affinity to both subtypes, some synthetic and naturally occurring ligands exhibit different relative affinities between ER-(u and ER-P. As a consequence, alterations in ligand structure could produce agents that show preferential binding to either ER-a or ER-P [13,14]. Recent studies demonstrate that the signal transcription for ER-a and ER-/3 may be different depending on the identity of the ligand and the response element [14, 151. AIB I , a member of the SRC-1 family, has been described as a nuclear receptor coactivator that helps receptors like ER-(u and ER-/3 to activate genes and thereby contribute to the development of steroid-dependent cancers [ 161.

Regardless of the complex biochemical pathway associated with the physiological response to estrogens, the binding of estrogens to intracellular receptor proteins is acknowledged as the primary event that elicits physio- logical activity. Even in the absence of 3D structural information about the receptor, Comparative Molecular Field Analysis (CoMFA) [17] has become a vital tool for inferring salient features associated with the ligand-receptor binding interaction. In general, CoMFA attempts to correlate experimentally observed biological data for a set of compounds with the calculated steric- electrostatic fields surrounding each molecule. In a typical CoMFA study, specified alignment rules are used to superimpose the molecule structures of each compound in the data set.

The CoMFA approach has been widely used in a variety of applications including the development of 3D-QSAR models for real or suspected EDCs [18-28]. For example, CoMFA models have been reported by Waller et LIZ. for dioxins [20], estrogens [21], and androgens [24]. Similarly, Tong et al. [27] constructed CoMFA models based on relative binding affinity (RBA) data from an ER binding assay using calf uterine cytosol. They found a good correlation between their predicted RBA values for calf ER and the experimentally observed RBA values for human ER, thus providing some justification for combining RBA assay data from multiple species. More recently, Tong et al. [28] reported a CoMFA model based on a data set of 31 estrogenic chemicals for which both competitive binding assay data were available for both ER-a and ER-/? [28]. The same data set was later employed by Tong et al. [29] to compare the feasibility of CoMFA, Hologram QSAR (HQSAR) [30], and a traditional QSAR approach for screening large databases in risk assessment programs. In both of these previous studies by Tong et al. [28,29], structural alignment in CoMFA was achieved by manual superimposition of the molecules comprising the training set.

I t is generally agreed that the alignment step represents the most subjective and controversial component in CoMFA. To both confirm and validate the

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

218 L. XING et al.

manual alignment scheme employed by Tong et al. [28,29], we present results from a CoMFA study on the same data set [12] in which structural alignment is achieved by employing a novel alignment algorithm: Steric and Electro- static ALignment (SEAL) [31]. Incorporating a fast Monte Carlo search procedure, SEAL provides an automatic and thorough search of structural alignments to maximize the overlap of both steric and electrostatic features. As with other common alignment schemes, SEAL assumes that the active site remains largely undistorted by the binding of the various ligands. SEAL offers several advantages over manually-based alignment schemes, including: (1) Pre-assignment of atom-atom alignments is not required, thus counter- intuitive alignments are not ignored; (2) Being automatic and exhaustive, SEAL considers a wide variety of possible alignments; and (3) All alignments considered are ranked objectively according to a similarity score.

MATERIALS A N D METHODS

Data Sets

Values of the Relative Binding Affinity (RBA) were obtained for the data set of 37 estrogenic compounds from the published competitive binding assay of Kuiper et al. [ 121 using 1 7P-estradiol (E2) for both ER-a and ER-P. The RBA is given as lOOX the ratio of the molar concentrations of E2 and that of the competing species required to decrease the receptor-bound concentration by 50% (thus RBA = 100 for E2). Consistent with our earlier studies on the data set [28, 291, the RBAs of testosterone and progesterone were set to 0.003 and 0.0003, respectively, for the purpose of enhancing the structural diver- sity of the data set. The test-set compounds, which were excluded from the training set used to construct the CoMFA model, included four compounds (5a-androstanedione, SP-androstanedione, 4-androstenedione, and corticos- terone) for which the RBAs were reported as “less than” values [12]. The 31 compounds in the training set comprised 15 physiological estrogens (steroids), 3 stilbene estrogens, 4 triphenylethylene anti-estrogens, 4 andro- gens (CI 9 steroids), 3 plant-derived nonsteroidal compounds, and 2 estro- genic environmental chemicals. Their log RBAs covered six orders of magnitude, from - 3.5 to 2.67 for ER-tu and from - 3.5 to 2.61 for ER-/3.

Conformational Analysis

Prior to constructing a CoMFA model, the bioactive conformation of the subject compounds must be identified or, if not known, hypothesized. The

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR N AND /3 SUBTYPES 21 9

crystal structures of estradiol and diethylstilbestrol (DES) were retrieved from the Cambridge Structural Database (CSD). The remaining molecules were constructed using the building tools in Sybyl [32]. Low-energy con- formations for these compounds were identified by the following pro- cedures. First, the geometry of each molecule was optimized to its local minimum-energy conformation. Second, these energy-minimized structures were subjected to a systematic conformational search over all rotatable bonds in increments of 10". Finally, the molecules were re-minimized after being set to the set of torsion angles from the search that were associated with the lowest energy. In those cases where multiple conformers were identified with similar energies, trial CoMFA models were constructed for all possible combinations of these unique conformers. Those combinations that yielded CoMFA models exhibiting the highest statistical quality were retained for the present study.

Using the resident Tripos molecular mechanics force field in Sybyl, all molecules were geometry optimized to an energy gradient of 0.001 Kcal/mole- A or a maximum of 1000 steps, whichever was reached first. These calcula- tions employed a distance-dependent ( l /v ) dielectric function to scale the Coulombic term, and the atomic partial charges were assigned using the Gasteiger-Marsili formalism.

Structural Alignment Using SEAL

The SEAL alignment scheme adopted in this study considered only steric (van der Waals) and electrostatic (Coulombic) interactions, although additional factors (e.g., hydrogen bonding, hydrophobicity) can be included if desired. Satisfactory results were obtained using the following Lorentzian functional form to compute the similarity score A F for a given alignment:

A F = -CCwi , / ( I + where the subscript i runs over the atoms in the first structure a n d j runs over the atoms in the second structure, and vij is the distance between atom i of the first structure and atomjof thc second structure. The weighting factor is given as a simple function of the partial charges (4) and van der Waals radii (v;):

(2) U'. . - i j - W E q i Y , + ' " .yViVj

As applied in the CoMFA procedure, the Gasteiger-Marsili formalism in Sybyl was used to assign partial atomic charges while the van der Waals

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

220 L. XING et al.

radii were taken from Bondi [33 ] . The attenuation factor a, which controls the coarseness or fineness of molecular feature recognition, was chosen at a moderate level of 0.2. The steric and electrostatic contributions were equally weighted (i.e., w E = ws = 0.5).

After energy minimization, SEAL was employed to align each structure in the training set to 17,Bestradiol selected as the template molecule. A total of 10 possible alignments were generated and ranked according to similarity score. Alignments that scored poorly were examined visually and then archived. Various combinations of multiple alignments, if applicable, were subjected to PLS analysis in order to identify that combination which generated the best statistics and would thereby be retained as the final CoMFA model.

CoMFA Steric-electrostatic Fields

Following standard procedures in CoMFA, the molecules were aligned one at a time in a three-dimensional (3-D) cubic lattice of either 1 A or 2 A spacing. An sp3 C probe atom containing a + 1 charge was placed at each mesh point, after which the steric (van der Waals) and electrostatic (Coulombic) fields were calculated between each probe atom and the atoms of the molecule placed inside the lattice. The Coulombic energies were scaled by a 1 / r distance-dependent dielectric, and steric and electrostatic energies above 30 Kcal/mole or below - 30 Kcal/mole were truncated to those values. Upon completing this operation, a separate pair of steric and electrostatic field energies can be associated with each mesh site in the 3-D lattice. These energy terms later served as the independent variables (Le., descriptors) to generate the 3D-QSAR model from CoMFA.

Statistical Analysis

Similar to many QSAR studies today, the number of steric-electrostatic field descriptors derived from CoMFA is much larger than the number of training-set compounds with associated activity data. In such cases, partial least-squares (PLS) regression [34] is well suited to handling these so-called underdetermined problems in which there are many more columns ( i e . , descriptors) than rows (ie., compounds with activity data). Similar to principal components analysis (PCA), PLS first reduces the number of descriptors to a few principal components (PCs) which are linear com- binations of the original descriptor variables. PLS then establishes a linear relationship between the biological activities corresponding to the depen- dent variable and these PCs corresponding to the independent variables.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR ry AND SUBTYPES 22 1

An initial PLS analysis was performed using the “leave-one-out’’ cross validation procedure to determine the optimum number of principal components needed for subsequent analysis of the data [17]. In cross validation, each compound is systematically excluded from the data set and its activity predicted by a model that is derived from the remaining compounds. The PLS analysis was then repeated without cross validation using the optimum number of components corresponding to that which yielded the smallest standard error of prediction or the highest cross- validated r2(q2) value. It is common practice to describe a statistical model as internally predictive if q2 2 0.5 and internally consistent if r2 2 0.9. This final analysis yielded a predictive model and associated conventional r2 values, from which the CoMFA coefficient contour plots for the steric and electrostatic potentials were generated. Finally, the 3D-QSAR model so derived was used to predict the RBA values of the test-set compounds.

Hierarchical Cluster Analysis

Hierarchical cluster analysis was performed on the calculated CoMFA field energies using the single clustering method implemented in Sybyl6.3 [32]. The results of these analyses were displayed for graphical analysis using dendograms. The dendogram may be viewed as an inverted tree with the training-set compounds included in the analysis at the bottom. As one moves up the tree, pairs of clusters merge denoting physicochemical similarities between compounds. The more closely related the molecular structures of the compounds, the earlier they merge into a single cluster. The lengths of the vertical lines in the dendogram provide qualitative information as to the separation, or linkage distance, between various clusters.

RESULTS

Structural Alignment Using Seal

As shown in Figure 1, SEAL oriented all steroids in the same way with respect to the template molecule l7p-estradiol. The respective A , B, C, and D rings of the steroids were superimposed on each other. These alignments are chemically intuitive and tend to confirm the validity of SEAL technique. The steroid ICI-164384 contains a long side chain bonded to C7, con- sequently it was handled as a special case whereby only its steroid moiety was included in the SEAL alignment procedure. The alignment obtained

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

222 L. XING ei al.

FIGURE 1 Orientation of steroids in the training set after SEAL alignment

from this scheme exhibited good overlap of the ring systems for all of the steroids.

In Figure 2, representative molecules from the various subsets are displayed in their optimal orientation with respect to 17P-estradiol. Genistein (Fig. 2a) was aligned with the fused rings overlapping the A and B rings of E2, and the phenolic ring oriented over the D-ring of E2. Similarly, the six- membered fused rings of coumestrol (Fig. 2b) are superimposed with the A and B rings of E2. Bisphenol-A (Fig. 2c), methoxychlor (Fig. 2d), and diethylstilbestrol (DES) (Fig. 2e), along with hexestrol, dienestrol, 4-OH- tamoxifen, tamoxifen, and clomifene, all possess two phenolic rings. In most cases, these rings were superimposed with the phenolic A-ring and the D-ring of E2. Although zearanol and E2 are structurally very different, SEAL achieved optimal overlap by superimposing the phenolic moiety of zearanol on the A-ring of E2 (Fig. 2f).

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR N AND B SUBTYPES 223

FIGURE 2 Orientation of representative moleculcs (dark) with respect to estradiol (grey) achieved by SEAL alignment: (a) Genistein. (h) Coumestrol. (c) Bisphenol-A. (d) Methoxy- chlor. (c) Diethylstilbestrol (DES), and ( f ) Zearanol.

Effect of Grid Distance and Column Filtering

The Grid Distance (GD), which corresponds to the unit length of the cubic lattice for calculating the CoMFA interaction energies, is an adjustable parameter that determines the fineness or coarseness of the molecular field evaluation. Decreasing the grid distance by a factor of 2 will generally

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

224 L. XING e / al.

double the number of unit lengths in every dimension, thereby increasing the overall number of grid points by a factor of 8 ( i .e. , 23) if the volume of the lattice is kept constant. Ideally, the molecular field should be probed as finely as possible in order to minimize the possibility that important molecular features are missed. On the other hand, the calculation should be coarse enough to be computationally practical. Column filtering (CF) is another adjustable parameter that controls the signal-to-noise ratio in CoMFA analysis. Individual steric and electrostatic columns corresponding to particular grid intersections exhibiting a standard deviation less than the CF value are neglected in the subsequent PLS analysis. Consequently, only those columns exhibiting significant variability in the calculated steric- electrostatic interaction energies for the subject molecules are considered.

In order to examine the sensitivity of our CoMFA models to CD and CF, four different models were constructed using various combinations of these two parameters. The GD was probed at the default value (2A) and at half of the default value (1 A), while the CF threshold for PLS was set at the default value of 2Kcal/mole and at 1 Kcal/mole. As summarized in Table 1, all models displayed acceptable quality in terms of their statistical parameters and good stability as indicated by the invariance of these parameters to changes in GD and CF. For every combination of GD and CF considered, the value of q2 exceeded 0.60 and that of r2 exceeded 0.98 using either 5 or 6 PCs. For both the ER-tr and ER-/3 models, the steric- electrostatic contributions were similarly dominated by the electrostatic fields for all combinations of GD and CF. From the practical standpoint, these results indicate that decreasing the CD parameter to increase resolution of the steric-electrostatic field data or decreasing the CF parameter to exclude fewer field descriptors prior to PLS analysis failed to substantially affect the statistical quality of the resulting models. At the same time, decreasing the GD by a factor of 2 dramatically extended the computational time of the PLS analysis from minutes to hours. In light of the similarity in these various models, we chose one representative example (i.~., GD = 2 A and C'F = 1 Kcal/mole) for our subsequent analysis as described below.

Steric-electrostatic Fields

To explore the individual contributions from the steric and electrostatic fields, we constructed separate CoMFA models using the steric or electrostatic field descriptors alone (Tab. IT) as well as together in a combined steric-electrostatic model. Both the steric field model and the

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OI- ESTROGEN RECEPTOR a AND B SUBTYPES 225

TABLE 1 (GD) und columrtfillering (CF)

Summary of results from CoMFA using different combinations of grid disraiice

% Contribution of Model PCS q2 r2 Std. error F p steric electrostatic

2 .& G.D. and 2 Kcdl/mole C.F.: ER-a 6 0.727 0.989 ER-p 6 0.605 0.989 2A G.D. and 1 Kcal/mole C.F.: ER-a 6 0.772 0.992 ER-P 6 0.646 0.990 1 .& G.D. and 2 Kcal/mole C.F.: ER-tu 5 0.720 0.986 ER-8 6 0.714 0.992 1 A G.D. and 1 Kcal/mole C.F.: ER-tr 5 0.726 0.985 ER-/J: 6 0.724 0.991

0.195 0.183

0.166 0. I72

0.215 n. I 52

0.220 0.159

353 0.0 38.3 61.7 345 0.0 33.7 66.3

493 0.0 35.7 64.3 393 0.0 33.3 66.7

347 0.0 33.2 66.8 503 0.0 32.2 67.8

332 0.0 34.0 66.0 460 0.0 33.2 66.8

TABLE I1 and electrostatic fields considered separately and together (Le . , combined model)

Comparison of statistical parameters for CoMFA models generated from the steric

PCS

Steric Field Model ER-LY 6 ER-,8 3 Electrostatic Field Model E R - ~ Y 5 ER-P 6 Combined Model ER-a 6 ER-[j 6

a 2 Std. error

0.708 0.551

0.556 0.7 I3

0.772 0.646

0.988 0.822

0.973 0.980

0.992 0.990

0.200 0.681

0.299 0.239

0.166 0.172

F P

337 0.0 42 0.0

178 0.0 200 0.0

493 0.0 393 0.0

electrostatic field model were internally predictive ( i .~ . , y2 > 0.5) and self- consistent (r2 > 0.9) except for the steric field model of ER-P. However, none of the separate models displayed superior quality to the combined model. The inclusion of both fields was supported by a field correlation analysis which revealed that, on average, the steric and electrostatic fields for the training-set compounds are non-collinear as indicated by a corre- lation coefficient of - 0.04 for 2080 points. By virtue of its superiority over the single-field models, only the combined steric-electrostatic model was selected for further analysis.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

226 L. XING et ul

lnclusion of Hydrophobicity

Hydrophobicity is often crucial in determining many in vivo properties of a drug such as its bioavailability, transport properties, and ability to cross biological membranes [35 - 381. A convenient yet generally accurate measure of hydrophobicity is the logarithm of the octanol-water partition coefficient (log P). We obtained estimated values of logP for the present study by using an additive-constitutive procedure found in the computer algorithm Clog P P91.

The calculated ClogP values for the compounds in the training set vary from 0.98 for genistein to 7.15 for nafoxidine (Tab. TIT). These values of ClogP were added to the original steric-electrostatic descriptors as a separate column prior to PLS analysis. The CoMFA models obtained by including ClogP are summarized in Table TV. The contribution of ClogP to the subsequent CoMFA model was only 6.8% for the ER-ol model and

TABLE 111 Calculated Clog P values for compounds in the training set ~~ ~

Compounds CIog P

Bisphenol-A 3.67 /I'-Zearanol 4.37 Coumestrol 3.12 Genistein 0.98 Methoxychlor 5.17 Nafoxidine 7.15 2-OH-Estradiol 3.19 30-Androstanediol 3.95 30-Androstanediol 3.95 4-OH-Estradiol 3.19 4-Androstenediol 3.47 5-Androstenediol 3.47 Dehydroepiandrosterone 3.07 50-Dihydrotestosterone 3.55 170-Etradiol 3.78 17(%Estradiol 3.78 Estriol 2.55 Etrone 3.38 ICI-164384 8.63 Moxestrol 2.86 Nandrolone 2.70 Norethindrone 3.29 Norethynodrel 3.14 Progesterone 3.77 Testosterone 3.22 4-OH-Tamoxifen 5.97 Clomifene 7.07 Diethylstilbestrol 4.96 Dienestrol 4.50 Hexestrol 5.1 I Tamoxifen 6.64

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR a AND 0 SUBTYPES 227

TABLE IV CoMFA models generated including Clog P values ~~

YO Contribution of PCS r2 bterrc electrostatic Clng P

E R-a 6 0 992 33 2 60 0 6 8 ER-(J 6 0 973 32 6 63 6 3 8

3.8% for the ER-,b’ model using 6 PCs in each case. Because of its minor impact on the CoMFA models, Clog P was excluded from further considera- tion in the present study. Based on a similar analysis, values of the molecular dipole were also excluded from further consideration as a separate column since they failed to improve the statistical quality or predictive ability of the models.

Hierarchical Cluster Analysis

The results from the hierarchical cluster analysis on the CoMFA fields generated for the training-set compounds in conformations appropriate to ER-(r are depicted as a dendogram (Fig. 3). Because it is possible for the separate models for ER+ and ER-P to contain different conformations of the same molecule, the dendograms corresponding to the CoMFA fields for ER-a and ER-P were different but only slightly. The high-level clustering of these fields is indicative of the rich structural diversity of the compounds. The dendogram depicts a strong grouping of the five compounds with long side chains (clomifene, tamoxifen, 4-OH-tamoxifeq nafoxidine and ICI- 164384). The largest single cluster was composed of steroidal and tamoxifen-like structures without side chains. The remaining long spikes, which converged at high levels of the hierarchical tree, correspond to bisphenol-A, zearanol, coumestrol, genistein, and methoxychlor.

Final CoMFA Models

The final CoMFA models (Tab. V) for both ER-(r and ER-/3 possessed excellent predictive ability (q2 = 0.772 for ER-a; q2 = 0.646 for ER-P). As seen when comparing the values of q2, the values of r2 (0.992 vs . 0.990) and standard errors (0.20 vs. 0.68) were slightly better for the ER-(r model than for the ER-,b’ model (using 6 PCs). The steric/electrostatic field contributions were 36%/64% for the ER-(r model and 33%/67% for the ER-,b’ model, hence both models are similarly dominated by the electrostatic fields. Values

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

228 L. XING et al.

I r _[

1

I;

S

V

steroids with long side- m z c - gw e e o steroidswith t a u aliphatic A-ring n aromatic A-ring chain h r m 1 molecules

h o a e S e x n s t

e n y o t o c l r I h o n

1 1 A o

r

FIGURE 3 fields for the ER-a model.

Dendogram representing results of a hierarchical cluster analysis on the CoMFA

TABLE V Summary of statistical parameters for the final CoMFA models ~

Cross-validated Fitted Boot- ytrapped PCF 42 Y = Std F Runs r 2 Std dev

error

ER-a 6 0.772 0.992 0.166 493 10 0.994 0.004 ER-/3 6 0.646 0.990 0.172 393 10 0.992 0.004

of the CoMFA-predicted versus the experimentally observed RBAs for the ER-a and ER-P models are tabulated in Table VI and plotted in Figure 4. The largest deviation is 0.43 log units for estradiol in the ER-a model and 0.42 log units for nandrolone in the ER-P model.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR a AND 0 SUBTYPES 229

TABLE VI log( RBA)

Comparison of CoMFA-predicted vetws experimentally observed values of

ER-(I. Observed Predicted

Bisphenol-A p-Zearanol Coumestrol Genistein Methoxychlor Nafoxidine 2-OH -Estradiol 3~-Androstanediol 38-Androstanediol 4-OH-Estradiol 4-Androstenediol 5-Androstenediol Dehydroepiandrosterone 5wDihydrotestosterone 17a-Estradiol 170-Estradiol Estriol Estrone

Moxestrol Nandrolone Norethindrone Norethynodrel Progesterone Testosterone 4-OH-Tamoxifen Clomifene Diethylstilbestrol Dienestrol Hexestrol Tamoxifen

lC1-164384

- 1.30 -1.16 1.20 1.15 1.97 2.00 0.70 0.83

-2.00 -2.19 1.64 1.55 0.85 I .02

-1.15 -1.11 0.48 0.30 1.11 1.07

0.78 0.73 - 1.40 -1.19

1.76 1.58 2.00 I .57 1.15 I .20 1.78 1.72 1.93 2.01 1.63 1.59

-2.00 -2.22 -1.15 -0.87 -0.16 -0.18 -3.50 -3.57 -2.10 -2.11

2.25 2.30 1.40 I .44 2.67 2.62 2.35 2.42 2.48 2.61 0.85 0.80

-0.30 -0.03

-1.30 -1.26

ER-0 Observed Predicted

-0.48 -0.52 1.15 1.19 2.27 2.13 1.56 1.49

-0.89 - 1.08 1.20 1.35 1.04 1.10

-0.52 -0.22 0.85 0.9 1 0.85 0.95

-0.22 -0.08 1.23 1 .00

-1.15 - 1.06 -0.77 -0.71

1.04 1.29 2.00 1.94 I .32 1.31 1.57 1.33 2.22 2.24 0.70 0.54

-0.64 -1.06 -2.00 -2.02 -0.66 -0.58 -3.50 -3.48

2.53 2.75 1.08 1.06 2.47 2.46 2.61 2.62 2.37 2.28 0.78 0.70

-2.10 -1.91

In order to generate statistical confidence limits, ten bootstrap runs were carried out on the training set. In this procedure, bootstrap samplings of the same size as the original training set were obtained by randomly choosing samples from the original data set while allowing the repeated selection of the same row (Le., compound). Standard statistical calculations on each of these bootstrap samplings yielded values of r2, from which the mean and standard deviation of these r2 values were computed. The difference between the bootstrapped mean and the original mean provides a measure of the bias of the original calculation, while the variance reflects the accuracy with which any parameter can be estimated from the input data. The averaged bootstrapped Y* values were very high (0.994 for ER-a and 0.992 for ER-b) with small standard deviations, thus adding an additional degree of con- fidence in our CoMFA models.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

230 L. XING et al.

CoMFA Calculated log RBA vs. Experimental log RBA

for ER-a and p (n = 31)

ER-p x

-4 -3 -2 -1 0 1 2 -4 -3 -2 -1 0 1 2 3

Experimental log RBA

FIGURE 4 Plot of the CoMFA-calculated versus experimentally observed log RBA values for both ER-cu and ER-b.

Predictions for the Test-set Compounds

The test set was composed of 4-androstenedione, 5a-androstanedione, 5p- androstanedione and corticosterone, whose experimentally observed RBA values were reported as “less than” [12]. After aligning these molecules by SEAL atop our template molecule 17P-estradio1, their RBAs were predicted using the appropriate CoMFA model. As shown in Table VII, the predicted RBA values for these four compounds are consistent with the corresponding experimental “less than” values for both the ER-a and ER-P models. The only possible exception is corticosterone for which the predicted values of -2.68 (ER-a model) and -2.72 (ER-P model) are only slightly greater than the observed value of < -3.

CoMFA Contour Maps

One of the most attractive features of CoMFA is its ability to depict the 3D- QSAR model as separate color-coded 3-D contour maps for the steric and electrostatic fields, derived from the product of the standard deviation and the regression coefficient (STDDEVSOEFF). In the steric contour maps (Fig. 5), the green/yellow polyhedra denote regions around the ligands where an

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

TABLE V11 Cornparison of CoMFA-predicted versus experimentally observed [12] values of log(RBA) for the test-set compounds

ER-a: ER-p

Predicted Observed Predicted Observed

4-Androstenedione -2.21 < -2 -2.88 < - 2 5a- Androstanedione -2.38 < -2 -2.61 < -2 50- Androstanedione -2.02 < -2 -3.14 < -2 Cortiocosterone -2.68 < -3 -2.72 < -3

FIGURE 5 The CoMFA steric contour maps, where the estradiol structure has been inserted as a visual reference. Enhanced binding to ER is associated with increasing steric bulk in regions depicted by the green polyhedra and decreasing steric bulk in regions depicted by the yellow polyhedra. (See Color Plate I).

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

232 L. XING r t ul

increase in steric bulk is favorable/unfavorable for binding. Inspection of the steric contour maps for ER-a and ER-P reveals that steric bulk is favorable in the region corresponding to the 3 position on the A-ring of 17P-estradiol. Steric bulk is also favorable for ER-a around positions 7 and 17, whereas it is preferred above the C-ring near C12 and C13 for ER-P. On the contrary, increased steric bulk is unfavorable for both ER-a and ER-/3 along the plane of the A-ring or near position 16 on the D-ring. The ER-P model also depicts a sterically unfavorable region near C11 not appearing in the ER-n model.

FIGURE 6 The CoMFA electrostatic contour maps, where the estradiol structure has been insertcd as a visual reference. Enhanced binding to ER is associated with increasing positive charge/polarity in regions depicted by the blue polyhedra and increasing negative charge/ polarity in regions depicted by the red polyhedra. (See Color Plate 11).

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR a AND 0 SUBTYPES 233

In the corresponding electrostatic contour maps (Fig. 6), the red/blue polyhedra denote regions in which negative/positive potential is preferred for enhanced binding. The large blue polyhedron near the 3 position of the stero- id A-ring is probably due to the different orientations of the electropositive hydrogen atom on the hydroxyl group found in many of the training-set compounds. The ER-a and ER-P models both reveal that negative potential is preferred near the 17 position of the steroid nucleus. However, the ER-P model differs somewhat from the ER-a! model in depicting red polyhedra (indicative of a preference for negative potential) above the surface formed by the steroid rings around positions 12 and 17, below the phenol A-ring near positions 1 and 2, and near positions 6 and 7 of the B-ring of estradiol.

DISCUSSION

It is well established that the 3D-QSAR models derived from CoMFA are sensitive to the conformation adopted for each molecule and to the scheme chosen for structural alignment. In particular, the alignment step is generally regarded as the most subjective and controversial part of CoMFA. In earlier studies on the same data set [28, 291, we selected the following pharmacophoric elements as the basis for manual alignment: the centroids of the A- , B-, C-, and D-rings for the steroidal compounds, and the centroids of the A- and C-rings and the C-1 of the B-ring for the triphenylethylene antiestrogens. The structural alignments achieved in the present study using SEAL are nearly identical to those found earlier in most cases [28, 291, thus validating our original pharmocophore hypothesis.

In our previous studies on this data set [28, 291, a single set of conforma- tions for the training-set compounds was used to construct the CoMFA models for both ER-a! and ER-P. In the present study, we selected which- ever conformer of the same compound would maximize the statistical quality of the separate CoMFA models for ER-a! and ER-P. Consequently, it was possible for the two models to contain different conformations of the same molecule. It turned out that most of the compounds maintained the same conformations, the only exception being progesterone. In progesterone, the ketone group bonded to the D-ring is oriented such that the ketone oxygen points away from C12 of the C-ring in the ER-a model whereas it has flipped over in the ER-0 model. This distinction with respect to the orientation of ketone oxygen in progesterone partly explains differences between the ER-cr and ER-/3 models that can be seen in the CoMFA contour maps around the D-ring of estradiol.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

234 L. XING et al.

Molecular mechanics calculations using the Tripos force field revealed that the progesterone conformation adopted in the ER-cr model is about 2 Kcal/ mole more stable than that found in the ER-Pmodel. Further conformational analysis, in which the ketone group was converted from the orientation in the ER-CI model to that in the ER-/? model by rotating the torsion angle in 10" increments followed by constrained energy minimization at each step, revealed an energy barrier of 4 Kcal/mole for the conformational transition. Wiese and Brooks [40] estimated the binding free energy attributed to hydrophobic attraction of the estrogen skeleton within the cytosol binding site of MCF-7 cells to be at least 7.1 Kcal/mole at 4°C. Assuming that the binding energy in rat prostate ER is of similar magnitude, the estimated conformational energy barrier of 4 Kcal/mole between the two different pro- gesterone conformers in the ER-(Y and ER-/? models would be surmountable (perhaps while the ER-ligand complex is dimerizing or involved in trans- activation). It is conceivable that other estrogen ligands can similarly choose among different conformations in order to accommodate a binding site in- side the ER or at least to maximize ligand-ER binding. This proposal is con- sistent with other recent conformational studies on estrogenic compounds [40, 411.

QSAR and, more recently, 3D-QSAR methods continue to be employed successfully as essential tools in pharmaceutical-targeted drug design efforts. Their utility has been demonstrated in a wide range of other applications, including toxicology [18 - 25, 27 - 29, 411, chromatography [42], and even thermochemistry [43]. It is hoped that the CoMFA models developed in the present study will help guide the rational design of estrogenic compounds that provide their medicinal benefit by selectively binding to either ER-(Y or ER-P. An estrogenic compound that preferentially bound to either ER-cr or ER-P might prove useful as a chemotherapeutic agent for treating a specific disease (e .g . , breast cancer, Alzheimer's Disease) with fewer of side-effects associated with binding to the other ER subtype.

It is also hoped that the CoMFA models developed in this study will prove useful in risk assessment by predicting the estrogen binding affinities of actual and suspected endocrine disrupting compounds (EDCs) which are believed capable of causing developmental and reproductive abnormalities in wildlife and hutnans [7 ~ 91. To that end, the training set employed in the present study encompassed at least six classes of chemical compounds of natural and synthetic origins.

It was shown in the present study that the standard steric and electrostatic field descriptors in CoMFA were sufficient to yield models that are statistically self-consistent (r2 > 0.9) and internally predictive (y2 > 0.5).

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR rx: A N D B SUBTYPES 235

Excellent agreement was found between the CoMFA-predicted and experimentally observed values of RBA for the training set (Fig, 4). The CoMFA models were further validated on a test set of compounds by predicting RBA values that are consistent with experimental observations [12]. The addition of octano1:water partition coefficients (Clog P) or mole- cular dipole moments to supplement the standard steric-electrostatic fields generated by CoMFA was shown to make only a minor contribution to the quality of the models.

The electrostatic fields were found to contribute more than the steric fields to our CoMFA models, which is consistent with previous CoMFA studies [28, 291 as well as with observations that polar amino-acid residues are found in the hormone-binding region of ERs [44]. The CoMFA contours maps presented in this study (Fig. 5) visually depict many of the structural prerequisites for ligand-ER binding and, in particular, for discriminating between ER-a and ER-,f3. The contours maps confirm that the 3-OH group in estradiol plays a crucial role in ligand-receptor binding, and that con- formational changes in the region around the planar phenolic A-ring are not well tolerated [ 19, 251.

Acknowledgements

WJW wishes to acknowledge the National Center for Toxicological Research (NCTR) of the US . Food and Drug Administration (FDA) and the US. Environmental Protection Agency for generously funding this research.

References

[ I ] Evans, R. M. (1988). The steroid and thyroid hormone receptor superfamily. Science, 240, 889 - 895.

[2] Tsai, M.-J. and O’Malley, B. W. (1994). Molecular mechanisms of action of steroid/ thyroid receptor superfamily members. Annu. Rev. Biochem., 63, 45 1 -486.

[3] Gronemeyer, H. (1991). Transcription activation by estrogen and progesterone receptors, Annu. Rev. Genet., 25, 89-- 123.

[4] Truss, M . and Beato, M. (1993). Steroid hormone receptors: Interaction with de- oxyribonucleic acid and transcription factors. Endocr. Rev., 14, 459 -479.

[5] Katzenellenbogen, B. S. (1996). Estrogen receptors: Bioactivities and interactions with cell signaling pathways. Biol. Reprod., 54, 287 - 293.

[h] Sarrel, P. M., Lufkin, E. G., Oursler, M. J. and Keefe, D. (1994). Estrogen actions in arteries, bone and brain. Sci. Am. Sci. Med., 1, 44-53.

[7] Kavlock, R. J . , Daston, G. P., DeRosa, C., Fenner-Crisp, P., Gray, L. E., Kaattari, L. G., Luster, M., Mac, M. J., Maczka, C., Miller, R., Moore, J., Rolland, R., Scott, G. , Sheehan, D. M., Sinks, T. and Tilson, H. A. (1996). Research needs for the risk assessment of health and environmental effects of endocrine disruptors: A Report of the U.S.EPA- sponsored Workshop. Environ. Health Prrspect., 104, 71 5 - 740.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

236 L. XING et al

[8] Davis, D. L., Bradlow, H. L., Wolff, M., Woodruff, T., Hoel, D. G. and Anton-Culvcr, H. (1 993). Medical hypothesis: Xenoestrogens as preventable causes of breast cancer. Environ. Health Perspect., 101, 372- 377.

[9] Read, L. D. and Katzenellenbogen, B. S. (1991). Advances in cellular and molecular biology of breast cancer. In: Genes, Oncogenes, und Hormones: (Dickson, R. B. and Lippman, M. E. Eds.), Kluwer Academic Publishers, Boston, pp. 277-299.

[lo] Kuiper, G. G. J. M., Enmark, E., Pelto-huikko, M., Nilsson, S. and Gustafsson, J.-A. (1996). Cloning of a novel estrogen receptor expressed in rat prostate and ovary. Proc. Natl. Acad. Sci. USA, 93, 5925 - 5930.

[ I I ] Mosselman, S., Polman, J. and Dijkema, R. (1996). ER-P: Identification and characterization of a novel human estrogen receptor. FEBS Letters, 392, 49 -53.

1121 Kuiper, G. G. J. M., Carlsson, B. O., Grandien, K., Enmark, E., HTggblad, J., Nilsson, S. and Gustafsson, J. A. (1997). Comparison of the ligand binding specificity and transcript tissue distribution of estrogen receptors a and p. Endocrinology, 138, 863 - 870.

[I31 Katzenellenbogen, B. S. and Korach, K. S. (1997). Editorial: A new actor in the estrogen receptor drama-enter ER p. Endocrinology, 138, 861 - 862.

[I41 Pennisi, E. (1997). Differing roles found for estrogen’s two receptors. Science, 227, 1439. [I51 Paech, L., Webb, P., Kuiper, G. G. J. M., Nilsson, S., Gustafsson, J . A., Jushner, P. J. and

Scanlan, T. S. (1997). Differential ligand activation of estrogen receptors ER-cu and ER-fi at API sites. Science, 277, 1508-1510.

1161 Anzick, S. L., Kononen, J., Walker, R.L., Azorsa, D. O., Tanner, M. M., Guan, X. Y., Sauter, G., Kallioniem, 0.-P., Trent, J. M. and Meltzer, P. S. (1997). AIBI, a steroid receptor coactivator amplified in breast and ovarian cancer. Science, 277, 965.

1171 Cramer, R. D. 111, Patterson, D. E. and Bunce, J. D. (1988). Comparative molecular field analysis (CoMFA): 1 . Effect of shape on binding of steroids to carrier proteins. J . Am. Chem. Soc., 110, 5959-5967.

[ I 81 Loughney, D. A. and Scbwender, C. F. (1992). A Comparison of progestin and androgen receptor binding using the CoMFA technique. J . Comput-Aided Mol. Des., 6, 569- 581.

[I91 Gantchev, T. G., Ali, H. and van Lier, J. E. (1994). Quantitative structure-activity relationships/comparative molecular field analysis (QSARs/CoMFA) for receptor-binding properties of halogenated estradiol derivatives. J . Med. Chem., 37, 4164 -41 67.

[20] Waller, C. L. and McKinney, J . D. (1995). Three-dimensional quantitative structure- activity relationships of dioxins and dioxin-like compounds: Model validation and Ah receptor characterization. Chem. Res. Toxicol., 8, 847 - 858.

[21] Waller, C. L., Minor, D. L. and McKinney, J. D. (1995). Examination of the estrogen- receptor binding affinities of polychlorinated hydroxybiphenyls using three-dimensional quantitative structure-activity relationships. Environ. Health Persp., 103, 702- 707.

[22] Waller, C. L., Operea, T. I . , Chae, K., Park, H.-K., Korach, K. S., Laws, S. C., Wiese, T. E., Kelce, W. R. and Gray, L. E., Jr. (1996). Ligand-based identification of environ- mental estrogens. Chem. Res. Toxicol., 9, 1240- 1248.

(231 Bradbury, S. P., Mekenyan, 0. G. and Ankley, G. T. (1996). Quantitative structure- activity relationships for polychlorinated hydroxybiphenyl estrogen receptor binding affinity - An assessment of conformer flexibility. Environ. Tox. Chem., 15, 1945- 1954.

1241 Waller, C. L., Juma, B. W., Gray, L. E. and Kelce, W. R. (1996). Three-dimensional quantitative structure-activity relationships for androgen receptor ligdnds. 7oxicol. Appl. Pharmacol., 137, 219 - 227.

(251 Oprea, T. 1. and Garcia, A. E. (1996). Three-dimensional quantitative structure-activity relationship of steroid aromatase inhibitors. J . Comput-Aided Mol. Des., 10, 186-200.

1261 Tong, W., Collantes, E. R., Chen, Y. and Welsh, W. J. (1996). A comparative molecular field analysis study of N-benzylpiperindines as acetylcholinesterase inhibitors. J . Med. Chem., 39, 380-387.

[27] Tong, W., Perkins, R., Streliz, R., Collantes, E. R., Keenan, S., Welsh, W. J., Branham, W. S. and Sheehan, D. M. (1997). Quantitative structure-activity relationships (QSARs) for estrogen binding to the estrogen receptor: Predictions across species. Environ. Health Perspect., 105, 11 16- 1123.

[28] Tong, W., Perkins, R., Xing, L., Welsh, W. J. and Sheehan, D. M. (1997). QSAR models for binding of estrogenic compounds to extrogen receptor 01 and 9 subtypes. Endocrinology, 138, 4022 -4025.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COMPARISON OF ESTROGEN RECEPTOR a A N D p SUBTYPES 237

1291 Tong, W., Lowis, D. R., Perkins, R., Chen, Y . , Welsh, W. J., Goddette, D. W., Heritage, T. W. and Sheehan, D. M. (1998). J . Chem. InJ Comput. Sci., 38, 669-671.

[30] HQSAR (Hologram QSAR) is a product of Tripos, Inc., St. Louis, M O 63144. [31] S E A L Quantum Chemistry Program Exchange (QCPE), Department o f Chemistry,

Indiana University Bloomington, Indiana, 47405. [32] Sybyl Version 6.3, Tripos, lnc., St. Louis, MO. [33] Bondi, A. (1964). J . Physical Chemistry, 68, 441. [34] Lindberg, W., Persson, J.-A. and Wold, S. (1983). Anal. Chem., 55, 643. [35] Ghose, A. K . and Crippen, G . M. (1985). J . Med. Chem., 28, 333. [36] Ghose, A. K. and Crippen, G. M. (1986). Atomic physicochemical parameters for three-

dimensional structure-directed quantitative structure-activity relationships I . Partition coefficients as a Measure of Hydrophobicity. J . Med. Chrm., 7, 565-578.

[37] Martin, Y. (1978). Quuntitativr Drug Dcsign, Medicinal Reseurch Series No. 8, Marcel Dekker, New York.

[38] Hansch, C. and Leo, A. S. (1979). Subsrituent Constants for Correlation Ana1ysi.s in Chemistry and Biology, Wilry Interscience, New York.

[39] Molecular Spreadsheet Manual, Chapter 9, pp. 257-280. Syhyi Version 6.3, 1996. [40] Wiese, T. E. and Brooks, S. C. (1994). Molecular modeling of steroidal estrogens: Novel

conformations and their role in biologicdl activity. J . Steroid Biochem. Molrc. Eiol., 50, 61 -73.

[41] Bradbury, S. P., Mekenyan, 0. G. and Ankley, G. T. (1996). Quantitative structure- activity relationships for polychlorinated hydroxybiphenyl estrogen receptor binding affinity: An assessment of conformer flexibility. Environ. Toxicol. Chem., 15, 1945- 1954.

[42] Collantes, E. R., Tong, W., Welsh, W. J . and Zielinski, W. L. (1996). Use of moment of inertia in comparative molecular field analysis to model chromatographic retention of nonpolar solutes. Anal. Chrm., 68, 2038 - 2043.

[43] Welsh, W. .I.. Tong, W., Collantes, E. R., Chickos, J. S. and Gagarin, S. G . (1996). Enthalpies of sublimation and formation of polycyclic aromatic hydrocarbons (PAHs) derived from comparative molecular tield analysis (CoMFA): Application of moment of inertia for molecular alignment. 7hermochim. Actu, 290, 55-64.

[44] Ekena, K., Weis, K. E., Katzenellenbogen, J . A. and Katzenellenbogen, B. S. (1996). Identification of amino acids in the hormone binding domain of the human estrogen receptor important in estrogen binding. J . Eiol. Chem., 271, 20053-20059.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COLOR PLATE 1. See L. Xing et al., Figure 5

The CoMFA steric contour maps, where the estradiol structure has been inserted as a visual reference. Ehhanced binding to ER is associated with increasing steric bulk in regions depicted by the green polyhedra and decreasing steric bulk in regions depicted by the yellow polyhedra.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14

COLOR PLATE I I . See L. Xing et af., Figure 6

The CoMFA electrostaticcontour maps, whcre thc estradiol structure has been inserted as a visual reference. Enhanced binding to ER is associated with increasing positive charge/ polarity in regions depicted by the blue polyhedra and increasing negative charge/polarity in regions depicted by the red polyhcdra.

Dow

nloa

ded

by [

134.

117.

10.2

00]

at 0

8:25

30

Nov

embe

r 20

14