WO2000052050A2 - Homology models of the glucocorticoid receptor - Google Patents

Abstract

A method of designing a homology model of the ligand binding domain of a glucocorticoid receptor wherein the homology model may be displayed as a three-dimensional image, the method comprising: (i) providing an amino acid sequence and an x-ray crystallographic structure of the ligand binding domain of a thyroid, estrogen or progesterone receptor; (ii) modifying said x-ray crystallographic structure to take account of differences between the amino acid configuration of the ligand binding domains of the glucocorticoid receptor on the one hand and the thyroid, estrogen, or progesterone receptor on the other hand; (iii) verifying the accuracy of the homology model by comparing it with experimentally-determined binding properties of a number of ligands for the glucocorticoid receptor; and (iv) if required modifying the homology model for greater consistency with those binding properties.

Description

HOMOLOGY MODELS OF THE GLUCOCORTICOID RECEPTOR

INTRODUCTION

Technical Field

This invention relates to combined computational and chemical methods of obtaining improved homology models of the human glucocorticoid receptor and to computational methods using such models for the design of ligands that bind to the glucocorticoid receptor.

BACKGROUND

Nuclear Receptors

The glucocorticoid receptor is a member of a superfamily of soluble proteins, the nuclear receptors. Unlike receptors associated with or integrated in the cell membrane the nuclear receptors reside in the cytoplasm or the cell nucleus. The members of the nuclear receptor superfamily have in common the ability to bind specifically their respective cognate ligands as well as DNA-elements. The ligands include physiologically-relevant ligands that are small molecules such as steroid hormones (androgens, oestrogen, mineralocorticoids, progestagens, and glucocorticoids), vitamins A and D, as well as pharmacologically important synthetic hormone mimetics that can act as agonists and/or antagonists. Upon ligand binding the receptor-ligand complex is able to modulate the transcription of genes that are controlled by that particular receptor's physiologic ligand. Depending on the nature of a specific target gene, it can be either up- or down-regulated via a classical mechanism that involves the interaction of the receptor-ligand complex with specific DNA-sequences upstream of target genes or by non-classical mechanisms such as protein-protein interactions between the receptor-ligand complex and other proteins involved in a signal transduction. The nuclear receptors bind DNA by means of a protein fold that contains cysteine residues coordinated to zinc atoms, the so-called zinc finger. The zinc-finger motif has been used to identify and to clone members of the nuclear hormone receptor family. The nuclear receptors have 5 regions: the N-terminal A-B domain that contains an activation function; the DNA-binding domain; DBD(C); the hinge D; the ligand binding domain, LBD(E), and the F-domain (that is specific for the estrogen receptor). The 3D structures for the DNA-binding domains of the GR, oestrogen receptor (ER), thyroid hormone receptor (THR) and retinoic acid receptor (RAR), have earlier been determined by X-ray crystallography or NMR spectroscopy.

More recently, the 3D-structures of the ligand-LBD complexes of some nuclear receptors, (THR, RAR, ER and progesterone receptor (PR)) have been published. The overall fold of the LBD of these nuclear receptors is an anti-parallel alpha-helical sandwich. This fold does not occur in any other known protein. Therefore it has previously not been possible to homology-model nuclear receptor LBDs correctly. A prerequisite for homology modelling is that the target and template structures have a similar tertiary structure. The sequence homology between the LBD's of various nuclear hormone receptors is low to moderate (10-50%) which alone does not guarantee that the LBDs of other members of this family will share similar tertiary folds. However, all these nuclear hormone receptors share a common function (transcriptional regulation), ID organisation (vide ante; A-E(f) region organisation), and many but not all are activated by ligand (i.e. endogenous hormones). In addition the structures of the LBDs of all members of the nuclear hormone family that have been solved to date share a common fold (antiparallel alpha-helical sandwich). Taken together, this evidence suggests that other members of this family are also likely to share this fold and therefore it should be possible to create homology models for the LBDs of these nuclear hormone receptors.

The conformation of the C-terminal a-helix (helix- 12) in the X-ray crystallographic structure of ER-a complexed with raloxifene together with the PR-based GR-homology model was used to produce a GR homology model for the study of antagonist binding.

Glucocorticoids Glucocorticoids are steroid hormones that mediate some of the body's responses to stress. The primary function of glucocorticoids is to protect the organism from the potentially harmful defence mechanisms induced by different forms of stress. Two such potentially harmful stress reactions are the induction of hypoglycaemia by insulin and the inflammatory response. Increased levels of glucocorticoids will increase the blood levels of glucose as well as exerting and anti-inflammatory action.

The mobilisation of blood glucose by glucocorticoids results from their induction of gluconeogenesis primarily by the induction of key enzymes in intermediary metabolism. For instance tyrosine amino-transferase and other transaminases are induced which results in redirection of energy in the amino acid to carbohydrate metabolism. One of the key regulatory enzymes in the gluconeogenic pathway, phospho-e«o/-pyruvate carboxykinase (PEPCK), is also induced by the glucocorticoids. The mechanism of induction of these enzymes is mediated by the DNA-binding of the steroid-activated glucocorticoid receptor (GR) which results in increased transcription of the genes for these enzymes.

In contrast, the anti-inflammatory effects of glucocorticoids involve the inhibition of the expression of a large number of proteins induced within the inflammatory response cascade. Examples of such proteins are cytokines (e.g. IL-2, IL-8), ezymes (e.g. collagenase I, iNOS, cyclooxygenase-2) and adhesion molecules. The principle mechanism involved is the repression of the transcriptional activation of these genes induced by various intermediary transcriptional factors (e.g. AP-1, NF-xB). Glucocorticoids prevent the induction of these genes by protein-protein interaction between the steroid-activated GR and the intermediary activating transcriptional factors.

The initial step in the mechanism of action of glucocorticoids is their binding to a specific soluble cytoplasmic receptor protein, the glucocorticoid receptor (GR). Thereafter a chaperone protein, HSP 90 is released from the GR, and the hormone receptor complex translocatese to the cell nucleus. GR belongs to the superfamily of nuclear receptors that have a zinc-finger DNA-binding motif. This motif enables ligand activated GR to bind to glucocorticoid response elements (GREs) situated on DNA upstream of GR-regulated genes. The transcription of those genes is then up- or down-regulated in response to the hormone. The Kd for binding of dexamethasone to GR is ~ 7nM and the natural hormone cortisol binds with ~ 10% of that affinity. In the PR based GR model an alternative orientation of the side chain of Thr-739 is found which enables it to hydrogen bond to both the C-20 keto group and the C21b hydroxyl group. However, it is not currently known if the C-20 carbonyl contributes to affinity.

Homology Modelling

Homology modelling involves the replacement of the differing amino acids in a related template protein structure in order to produce a model of the target protein structure. The basic assumption and requirement is that the template and target have a similar three-dimensional structure. The usefulness of a homology model is to be judged on the ability of such a model to explain the biochemical data for the target structure. A homology model can never be correct in all details, but it should capture one or more of the essential characteristics of the protein. Before a determination of the three-dimensional structure of the target is available the only way to evaluate a homology model is to assess its explanatory power. That the model is reasonable from a protein structure standpoint of view is not enough, since it can be very different from the target, especially if it was made from an unsuitable template. Therefore we have in the present invention tried to validate our model by checking it and refining it against as much experimentally-derived biochemical data available for the glucororticoid receptor as possible. For the same purpose we have also docked representative glucocorticoids in the banana shaped cavity of the homology model and minimised the resultant steroid-protein complexes with molecular mechanics. The experimental binding affinities for GR obtained in this invention were correlated with the computed protein interaction energies.

The first homology model in accordance with this invention was based on the thyroid hormone receptor (THR).

TR was initially used for homology modeling of GR. Because of the higher degree of sequence homology of the LBDs of GR vs ER (26%) than with THR (13%), and because of the close similarity of the preferred ligands of ER and GR (i.e. steroids), ER will be a better template than THR and therefore ER was then used as a preferred template for GR homology modelling in the present invention. For similar reasons a progesterone receptor (PR) was also used as a preferred template for GR modelling in the present invention. Glucocorticoid Structure - Activity Relationships

Synthetic glucocorticoids were investigated at an early stage for their anti-inflammatory properties. The ranking order of GR binding affinities correlates well with both the metabolic and anti-inflammatory effects of glucocorticoids. A combination of experimental, QSAR and computational chemistry studies have produced the following concept of important features for ligand binding to the glucocorticoid receptor.

The C-3 and C-20 keto groups of glucocorticoids are regarded as important for binding since their reduction to hydroxyl reduces binding affinity. This has been interpreted as suggestive of hydrogen bond donors located at corresponding positions in the receptor structure. Certain pyrazolosteroids, such as deacycortivazol, bind with high affinity although they do not have the C-3 keto group. There are also steroids without the C-20 keto group that have high affinity for GR, so neither the C-3 nor the C-20 keto group appear to be absolute requirements of glucocorticoids for high affinity GR binding.

The C-l l and C-21 hydroxyl groups are likewise as important for binding, and therefore suggest the existence of complementary hydrogen bond acceptors-donors in the receptor.

The replacement of the C-l IB hydroxyl with a keto group is detrimental to binding whereas a chlorine substituent at this position does not lead to loss of binding affinity. A C-l 7 a hydroxyl group increases affinity of glucocorticoids for the human receptor, but decreases the affinity for the rat receptor.

Hydrophobic pockets of limited size appear to exist in GR corresponding to the C-6a and C-9α positions of glucocorticoids, since small halogen atoms and methyl substituents here increase binding affinity, whereas bromine or methoy substituents in the C-0 position decrease ligand binding affinity.

It may be concluded from thermodynamic analysis that the ligand binding cavity of GR is predominantly hydrophobic since the binding enthalpy decreases when the temperature increases, which indicates that the driving force for binding is hydrophobic in nature. Surface area calculations indicate that both faces of the steroid are in contact with the protein, i.e. it is completely enclosed by the binding cavity. An important feature of GR binding is the presence of the 4,5-diene double bond in the

A-ring of the steroid. A second double bond in the A-ring, the l-2,diene, further enhances the binding affinity. This double bond causes the A-ring to tilt downwards (toward the α-face of the steroid) from the main plane of the molecule. This downward bend of the

A-ring was parameterized as the distance of the C-3 to C-l 7 carbon atoms (A- to D-ring distance) in the QSAR study of Wolff et al. It was found in that study that the shorter this distance was, the higher the affinity of the glucocorticoid was for the glucocorticoid receptor, i.e. the more bent the steroid is out of its main plane, the higher its affinity is for the glucocorticoid receptor.

Taken together, the structure-activity relationships presented above clearly indicate that the receptor imposes strict steric requirements on ligand binding.

UTILITY OF THE INVENTION

The glucocorticoid receptor models described in this application can be used to design new glucocorticoid receptor ligand, that can be agonists and/or antagonists. Glucocorticoid receptor agonists are useful for treatment of inflammation and immunosuppressive therapy. Glucocorticoid antagonists are expected to be useful in treatment of hypertension, diabetes, obesity, glaucoma, depression, AIDS and for wound healing.

The glucocorticoid receptor models can be used in design of new glucocorticoids in various ways. De novo drug design can be carried out by identification of features in the binding site that can be important for binding with respect to shape, charge, and hydrogen bonding properties. Ligand fragments with complementary properties to receptor features can be optimised for binding, in the same manner, by replacement of ligand fragments by better ones. Both these processes can be carried out manually or with de novo drug design programs, like LUDI and LEAPFROG given the coordinates of the glucocorticoid receptor models described herein.

The models can also be used with molecular mechanics, or 3D-quantitative structure activity relationship programs to assess the protein binding affinity of virtual glucocorticoid receptor ligands in order to prioritise their synthesis. In summary, the homology models according to the invention will be useful for electronic screening of compound databases, de novo drug design and/or prediction of binding affinities of glucocorticoid receptor ligands for glucocorticoid receptor by means of molecular mechanics scoring functions.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig 1 is a structure of a glucocorticosteroid, dexamethasone, with atom numbering.

Fig 2a is the final alignment used between the rat thyroid hormone receptor a and β sequences and the glucocorticoid receptor.

Fig 2b is the final alignment used between the estrogen receptor and sequence and the glucocorticoid receptor.

Fig 3 is a ribbon drawing of the ligand binding domain of the glucocorticoid receptor with a ligand depicted as a space-filling model.

Fig 4 shows cross-sections of a glucocorticoid receptor ligand within the binding site of the glucocorticoid receptor. The van de Waals radii of the ligand atoms as well as the water-excluded surface of the glucocorticoid receptor-model is shown as dots.

Fig 5 shows a drawing of a glucocorticoid receptor ligand with its interactions, with residues in the binding site that are critical for ligand binding.

Fig 6 sequence alignment used for homology modelling of GR from ER and TR.

Fig 7 structure of eight representative glucocorticoids used for experimental GR-binding assays and correlation with their computed protein interaction energies.

Fig 8 Graphs showing the progression of the improved correlation between calculated protein-interatction energy with the experimental free binding energy; (i) molecular mechanics ligand-protein interaction energy, (ii) inclusion of terms for ligand solvation, (iii) inclusion of terms for ligand solvation and strain energy, (iv) scaling of the individual components with respect to each other by means of PLS. Triamcinolone is not included in (i); R²=0.02 if it is included. o

Fig 9 (a) RasMol representation of the main interactions with dexamethasone in the GR homology model.

(b) Sketch of the main interactions with dexamethasone in the GR homology model.

Figures 10 and 10b Orthographic views of mutations in GR LBD that affect transactivation and/or ligand binding (Table 4) displayed as balls on the α-carbons. The Figure being produced with RasMol.66.

Figure 11 Sequence alignment used for homology modelling of GR from PR.

Figure 12 The three dimentional coordinates of the GR model produced from ER using its X-ray crystallographic structure as a template.

Figure 13 The three dimensional coordinates of the GR model produced from PR using its X-ray crystallographic structure as a template.

Figure 14 Sequence alignment of ER and GR used for the conformation of the C-terminal-helix (helix 12) in a PR-based GR-model for the study of binding of antagonists. The three dimensional coordinates of the GR homology model using the rat TRa/T3 and human TRb/Triac crystallographic structures.

Figure 15 The three-dimensional coordinates of the GR model produced from PR using the conformation of the C-terminal a-helix (helix 12) such as is in the X-ray crystallographic structure of ER- a complexed with raloifene. A GR-specific antagonist is docked into the binding site.

DETAILED DESCRIPTION OF THE INVENTION

Homology Modelling Based on the Thyroid Receptor

Initial multiple sequence alignments of the ligand binder nuclear receptor sequences were obtained using the Pileup program from the Genetics Computer Group at Univ. of Wisconsin program package. For semi-automated homology modelling Modeler, as supplied with Quanta96 from Molecular Simulations Inc., was used. ⁸ The homology modelling that we performed involved the replacement of the differing amino acids in a related protein structure (the template) in order to produce a model of the target protein structure. It is essential that the template and target have a similar three-dimensional structure. In manual homology modelling the side-chain positions of the amino-acids are then refined e.g. using rotamer libraries and energy minimisation. Loops can be copied from libraries of the other protein structures, and/or simulated by molecular dynamics. If the model does not exhibit a reasonable protein structure, or if it fails to account for the available biochemical data, the alignment is revised and the model is rebuilt. This process is preferably repeated until the model cannot be further improved. Serious manual homology modelling is thus a tedious enterprise, since the placing and refinement of the side chain positions has to be redone in each modelling cycle. Semi-automated homology modelling facilitates the interactive process since it automates the manual placement and refinement of the amino-acid side-chains as well as modelling of loops. In the present invention we used the program Modeler for this purpose. ¹⁵

The initial sequence alignments were obtained from multiple sequence alignments of nuclear receptors. These alignments were used for the initial runs but they were adjusted for subsequent runs in order to produce a molecular model with a reasonable protein structure that also accounted for the available scientific data. The ligand was not included in the Modeler runs. Iterative Modeler runs, using sequence alignments taking account of the known scientific data, resulted in the final alignment (Fig. 2). The overall fold of the model is shown in Fig 3. In the water-excluded surface of the model a completely enclosed banana-shaped binding cavity can be observed. This accords with calculations that have shown that glucocorticosteroids should be completely enclosed by glucocorticoid receptor, and because it is known that the glucocorticoids with high affinity for glucocorticoid receptor are more bend out of their main-plane (have a shorter A to D-ring distance) than glucocorticoid receptor-ligands with lower affinities, cortisol was manually docked as a rigid body into the glucocorticoid receptor homology model by the best possible fits of its atomic van der Waals raadi to the water-excluded surface of the binding cavity (Fig 4). The amino acid residues in the cavity were mainly hydrophobic, except for two residues namely Arg-61 1 and Thr-739. These were located within 3 A of the 21 -OH and 19-OH of cortisol, and their side chain nitrogen atoms could constitute hydrogen bonding partners. Furthermore, the backbone carbonyl of Leu-563 was within 3 A of the 11 -OH group of cortisol, and could thus be a possible hydrogen bond acceptor (Fig 5.).

A small cavity surrounded by carbonyl oxygen atoms near the C-3 of the ligand was observed. This cavity may contain water molecules. Therefore water-mediated hydrogen bonding may be involved in the binding of the C-3 carbonyl oxygen atom. Possibly cortivazol which does not have a C-3 carbonyl oxygen may displace the water molecules by its pyrazole ring which could fill the cavity, thus explaining its high binding affinity.

That C-l 1 chloro substituted glucocorticosteroids have similar affinities for glucocorticoid receptor as C-l 1 hydroxyl substituted glucocorticosteroids has been rationalized by the assumption that there exists an accessory hydrophobic pocket for such halogen substituents. ³ However this assumption may not be necessary, because in our model both the C-l 1 chloro or hydroxyl substituents may interact with a carbonyl oxygen, and carbonyl oxygens in other X-ray structures are known to interact with halogen atoms at less than the sum of their van der Waals radii. In the THR x-ray structure the ligand iodines thus interact with backbone carbonyl oxygens.^

That the binding cavity in this model is more bent than steroid makes sense in view of the classical structure-affinity relationships for glucocorticoid-steroids since it would preserve the rank order of affinities where the most bent glucocorticoid receptor-binding steroids (that have the shortest A-D ring distances) bind with the highest affinities to glucocorticoid receptor and the more planar ligands bind with lower affinities. It is known that the receptor-bound conformations of ligands frequently differ from their minimum energy conformations. It may thus be concluded that ligand binding to a receptor is a complex process where both the ligand and the binding site has to adjust for binding.

Warriar et al, have shown by point mutations that M565 and G567 are important for ligand specificity and binding, respectively. ¹² In our glucocorticoid receptor model these residues are close to the binding site. Stromstedt et a[ have covalent affinity labelling with radiolabelled glucocorticoid receptor ligands and protein sequencing studies demonstrated three residues, namely Met-622, Cys-754 and Cys-656, to be in the vicinity of the binding site of the glucocorticoid receptor '³. In our initial alignment two of these residues were in our glucocorticoid receptor-model in the vicinity of the ligand. We therefore improved the alignment so that the third residue was nearer the binding site after homology modelling. However, this revision of the models did not affect the binding site. The final alignments are shown in Fig 2 A and 2B.

Homology Modelling Based on the Estrogen and Progesterone Receptors

Materials

[³H]TA was obtained from NEN-Dupont, unlabelled steroids from Sigma and cell culture media, fetal bovine serum and penicillin-streptomycin from Gibco-BRL.

Plasmids

The mammalian vector pCMNhGR, expressing the wild type hGR, was constructed by cutting out a BamHI-Xbal gragment from pUC18/ATG-NX and inserting it into pCMV4. This fragment contains the entire coding region of the human GR gene and about 400 bp of the 3 '-untranslated sequence.

Mammalian Cell Culture and Transfection

COS-7 cells were grown in Dulbecco's modified Eagle's medium, supplemented with 10% fetal calf serum, penicillin (100 IU/mL) and streptomycin (lOOμg/mL), at 37°C in a humidified atmosphere with 5% CO2. For ligand binding assays and competition assays 10 cm plates containing cells at 60-80% confluency, plated out 1-3 days before transfection. were transfected with 10-15μg expression vector using the calcium phosphate method. Cells were incubated 48 hours after transfection before assays were performed.

Ligand Binding and Competition Assays

Cells were washed with and scraped in PBS and spun in a microfuge. They were then resuspended in a buffer consisting of ImM EDTA, 20mM potassium phosphate p7.8, 10% glycerol, 20mM sodium molybdate and 1 mM DTT, homogenised with a glass homogenizer and the lysate was spun for 30 min at 100,000 xg at 4°C. For saturation analysis different concentrations of [³H]TA(0.2-0.7 nM) were added and for competitive binding assays lOnM [³H]TA and increasing concentrations of different non-radiolabelled steroids were added. The extracts were incubated at 4°C overnight. Bound and free [³H]TA were then separated by gel filtration on a Nick column (Pharmacia) and the amount of [³H]TA bound was measured in a scintillation counter. Free [³H]TA was calculated as total minus bound [³H]TA. The level of unspecific binding was negligible as monitored by adding 200 fold excess non-radiolabelled TA to parallel incubations with the different concentrations of [³H]TA.

Homology modelling and molecular dynamics

Initial multiple sequence alignments of the ligand binding nuclear receptor sequences were obtained using the Pileup program from the GCG program package. For semi-automated homology modelling, Modeler, as supplied with Quanta96, was run using the no optimism option, with the ER- LBD/estradiol complex X-ray crystallographic structure (Brookhaven PDB accession number 1ERE) as the template. Hydrogen atoms were added to the homology model using the HBUILD routine in CHARMm. Sodium and chloride counterions were placed at the maxima and minima of the protein electrostatic potential near charged amino acid residues so as to achieve net neutrality of the system. The C- and N-termini were made neutral. The 3D molecular editor of QUANTA96 was used to build the various glucocorticoids. The constructed glucocorticoids were minimised in vacuo using Gasteiger - Huckel charges and a dielectric constant of 78. Partial atomic charges for the resulting structures were calculated by fitting the water-accessible surfaces of the molecules to their 6-31G* electrostatic potentials according to Singh and Kollamn, as implemented in Gaussian 94. The 6-31G* ESP charges were used for the ensuing protein-ligand interaction studies. The fit of dexamethasone in the binding site with the lowest ligand-protein interaction energy after minimization of various explored alternative starting orientations was chosen as an initial conformation for subsequent molecular dynamics. The minimisation was carried out with CHARMm and started with 200 initial cycles of steepest descent and continued by the adopted-basis Newton-Raphson algorithm until the root means square energy gradient was less than 0.01 kcal/A. The default heuristic non-bonded list-update method and a distance dependent dielectric function (scaled with 1/r) were used. The protein-ligand interaction energies were when required calculated for each resulting minimised conformation. The system was subjected to molecular dynamics using the Nerlet and Shake algorithms using the same conditions as for the minimisation. The protein was surrounded by a 21 A solvent cap of TIP3 waters centred on the ligand for the dynamics simulation. The initial dynamics simulation was for 10 ps using a step-size of 0.01 followed by 60 ps with a step-size of 0.02. The solvent cap was then removed and the remaining dexamethasone-GR complex structure resulting from the final trajectory after 70 ps of dynamics, was energy-minimised using the same constraints as described above and thereafter used for energy-minimisation with other ligands instead of dexamethasone.

The Modeler program was used, and it has been shown that it produces results as good as manual modelling at different levels of homology.

The initial sequence alignments were obtained from multiple sequence alignments of nuclear receptors. These alignments were used for the initial runs but they were adjusted for subsequent runs in order to produce a molecular model with a reasonable protein structure that also accounted for the available scientific data. The ligand was not included in the Modeler runs. Iterative Modeler runs, using sequence alignments considering the scientific data, resulted in the final alignment (Fig. 6). In the water-excluded surface of the model a completely enclosed banana-shaped binding cavity could be observed. This is reassuring, since it has been calculated that glucocorticoids should be completely enclosed by GR, and because it is known that the glucocorticoids with high affinity for GR are more bent out of their main plane (have a short A- to D-ring distance) than GR ligand with lower affinities. Dexamethasone was manually docked as a rigid body into the GR receptor homology model by the best possible fits of its atomic van der Waals radii to the water-excluded surface of the binding cavity. The initially selected position of dexamethasone within the homology model correspond to that of estradiol in the ER- crystallographic complex (binding mode 1, Table 1). The dexamethasone molecule was also rotated 180 degrees about its long axis, such that the positions of the A- and D-rings were reversed (binding mode 5, Table 1) and thereafter at each of these two orientations, the molecule was rotated in steps 90 degrees about its short axis (binding modes 2-4 and 6-8, Table 1). An alternative orientation of the C-17-side chain corresponding to binding mode 1 was also investigated (binding mode 9, Table 1). The orientation with the most favourable protein ligand-interaction energy (Table 1 ) agrees with that of the ligand in the ER and PR X-ray crystallographic structures but not with the orientation of the glucocorticoid in an earlier published GR model. The fit with the lowest ligand-protein interaction energy was chosen as a starting conformation for subsequent molecular dynamics. A water molecule was placed in the cavity near the C-3 carbonyl oxygen, based on the bridging water between the glutamate, arginine and 3' -OH of estradiol in the ER-α X-ray crystallographic structure. Side-chains of residues or water molecules within 5A of the ligand were allowed to move freely, whereas the main chain C, Cα and N atoms within the same zone were restrained with a harmonic potential of 100 kcal/A. The rest of the protein was kept rigid. These restraints were imposed because too much unrestrained minimisation and molecular dynamics has in blind tests been shown to cause the homology models to be more off the target than the template. The system was equilibrated with molecular dynamics together with the explicit solvent for 70 ps. The potential energy and temperature appeared stable during the last 10 ps of the trajectory. The solvent water was then removed and the remaining dexamethasone-GR complex structure resulting from the final trajectory after 70 ps of dynamics was energy-minimised using the same constraints as described above. The other ligand structures were then fit into the binding site in this same position and orientation as the best fit obtained for dexamethasone, and energy-minimised and used for correlation with the in this study experimentally determined IC₅oS for human GR (Table 2).

Because a crystallographic structure of the progesterone receptor bound toan PR antagonist such as RU-38486 is not currently available, it is not certain how the conformation of PR changes when complexed with an PR antagonist. If RU-486 is docked into the binding cavity of the PR crystallographic structure or the GR homology model in an orientation analogous to progesterone in PR, there is insufficient space to accommodate its bulky 1 l-N,N-dimethylaniline substituent (i.e., it sterically clashes with helix- 12). Therefore it is likely that H12 of GR is displace by RU-486 in analogy to the displacement of H12 by ER antagonists in estrogen receptor complexes. This displacement of HI 2 will then allow sufficient space to accommodate the 11 -b-substituent of RU-486 and other GR antagonists with bulky 1 1-b-substituents. Therefore to produce an antagonistic GR homology model, HI 2 and the loop connect HI 1 and H12 in the PR based GR homology model was moved to where it is located in the ER -raloxifene complex by superimposition of the C-carbons according to the alignment in Fig 9. This was accomplished by splicing HI 2 and the loop between Hl l and H12 from the ER -raloxifene complex X-ray crystallographic structure into the GR-model, followed by mutation of the ER amino acid residues to the corresponding GR residues.

Table 1 Molecular Mechanics Interaction Energies For Various Biding Modes of Dexamethasone

*Binding mode 1 corresponds to that of estradiol in the ER-α crystallographic complex. Rotating the dexamethasone molecule in this binding mode by 180 degrees about its long axis, such that the positions of the A and D rings are reversed, results in binding mode 5. Binding modes 2-4 and 6-8 respectively, result from the progressive rotation of the molecule in steps of 90 degrees about its short axis at each of these two orientations. Binding modes 1 and 9 have a different orientation of the C-l 7 side.

Calculation of Solvation Energy

6-31G* electrostatic potential-fit charges and the corresponding CHARMw in vacuo optimised geometries of the steroids were used for calculation of solvent free energies with the GB/S A algorithm as implemented in the solvation module of the pseudoreceptor drug design software package PrGen. All calculations were run on Silicon Graphics R10000 workstations under IRIX 6.2/6.4.

Statistical Analysis The sigmoidal dose-response curves obtained by competitive ligand binding assays were linearized with the log-logit function and the IC50 determined as the intersection with the X-axis (where logit=0) as described by Rodbard. The statistical analyses were performed using Microsoft Excel 5.0 and the partial least squares methodology as implemented in the QSAR module of Sybyl 6.4.

Molecular Mechanics Interaction Energies

Initially, the last frame of the molecular dynamics simulation performed with dexamethasone was energy minimised as described in Methods. It was noted that Thr-739 could possibly form a hydrogen bond with the steroid's 21 -hydroxyl group by simple rotation of the amino acid's side-chain. The complex was then re-minimised following this alteration. As this lead to an improvement in the protein-ligand interaction energy, and the 21 -hydroxyl is known to contribute to the affinity of glucocorticoid steroids, this complex was then used as a template to investigate the interactions of the other ligands with the protein. A spectrum of ligands was chosen to represent typical glucocorticoids ranging from natural ligand cortisol to potent synthetic ligands such as triamcinolone acetonide. The ligands were selected to include various combinations of common substituents of pharmacologically interesting GR-ligands such as Δ-l, 9α-fluoro and 16α-, 17α— substitutions (c.f. Fig 7 for the structures of the ligands used in the present study).

Table 2

Table 2. Relative binding of glucocorticoids to human GR.

2„ n—=3 for all compounds except for Ta where n=2 As can be seen from Table 2, the modifications of the cortisol structure employed here lead to quite significant alterations in binding affinity. Thus, this series of steroid ligands was well suited to challenge the explanatory powers of the model. Different orientations of the side chains of both ligands and amino acid residues were investigated and the model yielding the best correlation between molecular mechanics interaction energy and experimental data is analysed below and its interactions discussed with respect to the points described in the introduction.

It was not possible to obtain a useful correlation between the molecular mechanics interaction energy and the experimental free binding energy for all ligands (data now shown; R²=0.10 vs. 0.45 when triamcinolone is excluded). Triamcinolone, which is the outlier, is the only ligand in this investigation having a 16α-hydroxyl substituent. In other studies, it has been shown that in order to reproduce experimental relative free energies of binding, ligand desolvation effects have to be taken into account. The GB/SA algorithm has been validated with a series of small substituted hydrocarbons. We have within the context of another study (Carlsson et al., manuscript in preparation) applied the PrGen solvation module to the Wolfenden data set which consists of the experimentally determined free energies of solvation in water of ten aromatic and cyclic molecules. Due to the greater degree of resemblance of the molecules in the Wolfenden data set to drug molecules, the solvation energies calculated by the solvation module of PrGen are scaled according to the equation resulting from the fit between that data set and the experimental data

(ΔGsolv (GB/SA)=(ΔGSO1V (PrGen)+0.25)/0.88;R²=0.91

with 6-31G* ESP charges calculated on the CHARMw in vacuo optimised geometries.

When a solvation correlation was included, the correlation between computed and experimental free energies for GR ligand binding was not improved (R² = 0.01 for all ligands vs. 0.45 when triamcinolone is excluded). The activity of triamcinolone was under-predicted in the correlation, as its C-15α-hydroxyl group was unable to make any hydrogen bonds with the protein but yet had the largest solvation penalty (Table 3). It was then noted that Gln-642 could make a hydrogen bond with the C-l 6a hydroxyl oxygen atom of triamcinolone and the C-l 6 ether oxygen atoms.

Table 3: Summary of Ligand-Protein Interaction Energies (kcal/mol), Calculated Solvation Energies (kcal/mol) Ligand Strain Energies (kcal/mol) and Experimental Binding Affinities (kcal/mol) of the Glucocorticoids.

Ligand Molecular Mechanics Scaled Calculated Ligand Strain Experimental free binding

Interaction Energy Solvation Energy Energy energy^a(kcal/mol) Λ (kcal/mol) (kcal/mol) (kcal/mol) -ΔG„ bind (exptl ) _

Desonide -78.0 -26.32 5.66 11.55 Λ 9 -F -72.4 -23.33 3.08 11.53 prednisolone

TA -78.7 -25.88 5.96 11.47

. Dexamethasone -74.3 ^•21.13 4.71 11.24

9α-F cortisol -72.1 ^•20.35 3.42 11.04

Prednisolone -73.9 ^■21.81 2.90 11.03

Cortisol -70.6 ^■17.99 2.93 10.32

Triamcinolone -82.9 ^•27.78 5.69 9.96

ΔG bind (exptl ) =RT In K*_, K* =(l/K_d for TH]TA) (IC ₀ for tested ligand /IC₅₀ unlabeled TA). K_d for TA = 0.66 nM.

of triamcinolone acetonide and desonide. Thus, the side chain of Gln-642 was rotated so as to achieve a hydrogen bond with the C16-oxygen substituent of these ligands and the complexes minimized. Although the other steroids do not possess a C16-substituent capable of acting as either a hydrogen bond donor or acceptor, Gln-642 in these complexes was also rotated to the same position as that in triamcinolone and the resulting complexes minimised. The amide nitrogen of Gln-642 formed hydrogen bonds (-3.1 A) with the S of Met-639 and also with the 16α oxygen atoms of triamcinolone, desonide and triamcinolone acetonide. A good correlation between the molecular mechanics interaction energy for all the ligands could then be achieved with this model if solvation and ligand strain energy (ligand strain energy is the difference in energy between the ligand conformation in the protein and when minimised in vacuo) corrections were included. Figure 8 shows the progressive improvement of the correlation with the addition of these corrections, the R² increases from 0.04 (0.49 without triamcinolone; Fig 8A) to 0.48 with the solvation penalty (Fig 8B) and further to 0.69 if the ligand strain energy penalty is also included (Fig 8C). When considering only molecular mechanics interaction energy together with ligand strain energy, the correlation coefficient is 0.1 (0.49 without triamcinolone). Thus it is the combination of the solvation and strain terms that results in good correlation with the experimental free energy binding data. A partial least squares analysis using the molecular mechanics interaction energy together with the solvation (not scaled according to Wolfenden data set) and strain energies was then performed, increasing the R² from 0.04 (molecular mechanics term only) to 0.80 (inclusion of solvation and strain energies, Fig. 3(iv)). The resulting equation was:

ΔGbind(calc)=0.384 -ΔGinter(MM)0.343-ΔGsolv(PrGen)-0.466 -ΔGstrain(MM)-31.058.

The main interactions of dexamethasone with the protein are shown schematically in Fig. 4. In another model the C-17-side-chain of the ligands was rotated so as to enable the C20-keto group to form a hydrogen bond with the CD2 hydrogen of Tyr-735, but at the same time maintaining the hydrogen bond interaction between the C21 -hydroxyl and Thr-739. Such hydrogen bond interactions have been reported in the literature and interestingly, Tyr-735 corresponds to His-524 in hERα, the residue which forms a hydrogen bond with the C-17-OH group of estradiol. The corresponding residue for the other members of the steroid hormone superfamily is either a phenylanaline or a tyrosine. However, this model yielded a poor correlation (R²<0.5) and was therefore discarded from further analysis.

It will be noted that whilst the description refers to a method of designing a GR homology model from rat thyroid hormone receptors and a Human Progesterone Receptor, the same process may be used to design a GR model from a Human Progesterone Receptor. Figure 11 shows the sequence alignment which would be used to design a GR from PR.

Explanation of the Glucocorticoids Structure-Activity Relationships

Reduction of the C-3 keto to a hydroxyl group reduces biding affinity

Tanenbaum et al have contrasted the hydrogen bonding interactions of the phenolic group of estradiol and the C-3 keto group of progesterone complexed with their respective receptors. Steroid receptors which bind 3-keto groups have a conserved glutamine corresponding to the sequence position of Gln-725 in hPR, but the equivalent residue in hERα is a glutamate residue (Glu-353). Hence, these residues are responsible for the discrimination by the steroid receptors of keto and hydroxy moieties as the arginine residues which form either direct (hPR) or direct and water mediated interactions (hERα) with the keto and hydroxyl functions are conserved throughout the steroid receptor family. Additionally, there is a conformational change (atomic displacement) associated with a change in hybridization from a sp²C-3 keto to a sp³ hydroxyl. Thus, as the ligand makes several hydrogen bonds with the receptor, a protein with functional groups at an optimum separation for simultaneous binding to the O-3 and O-20 keto oxygen atoms, and 0-11 , O-17 and O-21 hydroxyl oxygen atoms may bind less to a ligand which has a 3 -hydroxyl group instead of a 3-keto group.

In our homology model, the C-3 keto groups of the glucocorticoid ligands form direct hydrogen bonds with Gln-570 and Arg-611. In contrast, in hPR Arg-766 directly contacts progesterone' s C-3 -keto group, and in hERα the equivalent interaction (Arg-394) with the phenolic oxygen of estradiol is also direct.

Reduction of the C-20 keto group reduces binding affinity In the homology model, the C-20 keto group does not engage in hydrogen bonding, but does have a favourable electrostatic interaction with the sulfur of Met-560. For the eight glucocorticoids, the O-A distance is ~3A. (Many examples of both intermolecular and intramolecular nonbonded sulfur-nucleophile close contacts in which the sulfur-nucleophile distance is less than the sum of the sulfur and nucleophile van de Waals radii have been reported in the crystallographic literature).

The C-l β- and C-21/?- hydroxyl groups are important for binding

The main weakness of the present model is its inability to explain the importance of the C-l l ?-hydroxyl as the nearest potential hydrogen bond partner is the backbone carbonyl of Leu-563, with a distance of ~4A. That C-l l ?-chloro substituted glucocorticoids have similar affinities for Gr as C-l l ?-hydroxyl substituted glucocorticoids has been rationalised by the assumption that there exists an accessory hydrophobic pocket for such halogen substituents. However this assumption may not be necessary, because in our model both the C-l 1 chloro or hydroxyl substituents may interact with the Leu-563 carbonyl oxygen, and carbonyl oxygens in other X-ray structures are known to interact with halogen atoms at less than the sum of their van de Waals radii. In the thyroid hormone receptor X-ray structure the ligand iodines interact thus with backbone carbonyl oxygens.

The C-21 hydroxyl forms a hydrogen bond (2.6A) with Try-739 and also with the backbone carbonyl of Tyr-735 (2.9A). Thr-730 corresponds to Thr-894 in hPR, but in the complex with progesterone (which lacks the C-21 hydroxyl group) no hydrogen bond could be made by this residue with the ligand.

A C-17α hydroxyl group increases affinity for the hGR, but decreases it for rat GR

The C-17α hydroxyl group in both cortisol and 9α-F cortisol forms a good hydrogen bond with Met-639 (oxygen-sulfur distance = -3.2A). However, from our model it appears that compounds possessing a 16-methyl substituent are unable to form a short hydrogen bond with this residue due to a steric repulsion between Met-639 and 16-Me. By comparing the sequences of rat and human GR, there were no species differences in the vicinity of Met-639. Thus the model is unable to explain the intolerance of the rat GR for a C17α-OH group. Bromine or methoxy substituents in the C-9 position decrease ligand binding affinity

The C-9α fluorine of the ligands is in close contact with the CE2 carbon atom of Phe-623 (3.4 A). Thus, introducing a more bulky substituent would presumably result in the displacement of Phe-623 from a position in which it can form a "pi-teeing" interaction with the ligand's A-ring and/or displacement of the ligand from its preferred position, causing a disruption of its interactions with the receptor.

The ligand is completely enclosed by the hydrophobic binding cavity

In the model the ligand is completely enclosed by predominantly hydrophobic residues:

Leu-563, Leu-566, Trp-600, Met-601, Met-604, Phe-623, Leu-723 and Leu-753. The

ligands make many favourable hydrophobic contacts with the receptor, in particular the

C18- and C19- methyl groups with Met-601 and Leu-732 respectively. Significantly, these

residues are preserved in the hPR, human androgen receptor and human mineralocorticoid

receptor, whose ligands also have C18- and C19- methyl groups, but not in hER whose

natural ligands 17 -estradiol lacks the C19-methyl group. Additionally, hydrophobic

contact exist between the C16-methyl group of dexamethasone and the acetonide functions

of desonide and triamcinolone acetonide with Tyr-735.

In order to assess the utility of the PR-based GR homology model modified to accommodate GR antagonists for the design of such drugs the compound 1 Oe described Gebhard et al. Biorg. Med. Chem. Lett. 7(17) 2229, 1997 was used. The ligand was placed in the receptor with it steroid core in the same orientation as progesterone in PR and the complex minimized to gradient norm of 0.05 with the residues within 7 A of the ligand treated as flexible while the remainder of the protein was held rigid. After minimization the carbonyl oxygen atom of the 2-oxo-l-pyrrolidinyl group was within hydrogen bonding distance (2.7 A) to the hydroxyl oxygen atom of Thr-556, which corresponds to Ser-71 1 in PR. The distance from the steroid C-3 carbonyl oxygen atom to Arg-61 1 was 2.6 A. Thus the model may be used to interpret and improve the binding of GR antagonists to GR. A second double bond in the A-ring improves binding affinity

The fact that the binding cavity in this GR homology model is more bent than the steroid ligand is consistent with the classical structure-affinity relationship for glucocorticoids, since it would preserve the rank order of affinities where the most bent GR-binding steroids (those that have the shortest A- to D- ring distances) bind with the highest affinities to GR and the more planar ligands bind with lower affinities. It is known that the receptor bound conformation of ligands frequently differs from the minimum energy conformation of ligands. In the liganded RAR structure the 9-cis retinoic acid is more bent than the binding cavity whereas the all-trans retinoic acid is flatter than the binding cavity. It was thus concluded that ligand binding to a nuclear receptor is a complex process where both the ligand and the binding site have to adjust for binding. If our modelling has produced a correct impression of the ligand binding cavity, the above statement will be valid for GR as well.

Each of the pair cortisol and prednisolone, 9α-F, prednisolone represent ligands which differ only in whether they have a 1,2-diene double bond or not (Fig 7). In our model, the A-ring of the 1 ,2-unsaturated ligands adopt a lα, 2 ?-half chair conformation which represents one of the ideal forms. Relative to the docked 9α-F prednisolone, the 3-C keto oxygen of the docked 9α-F cotrisol is displaced by 0.2A. This leads to minor adjustments in its interactions with the protein so that the coordinates of the water oxygen atom and nitrogen of the Gln-570 side-chain differ by 0.1 A as compared to those in the 9α-F prednisolone protein biding site. Additionally, displacements of the l l-O, 17-O, 20-O and 21-0 oxygen atoms and 9α-F atom of between 0.1 to 0.2A occur. Atomic movements of the same magnitude also occur for cotrisol with respect to prednisolone, with the exception of the 17-0 and 20-O atoms which are displaced to a lesser extent, (0.04 A vs. 0.1 A respectively). Prednisolone has a distinctly superior molecular mechanics interaction energy as compared to cortisol (~3 kcal/mol), but 9α-F cortisol has only a marginally better molecular mechanics interaction energy than 9α-F prednisolone (Table 3). These differences are attenuated (9α-F cortisol vs 9α-F prednisolone) or at least maintained (cortisol vs prednisolone) when ligand strain energy is added to the molecular mechanics term (Table 3). Thus, the model does reflect the preference of GR for ligands with a 1,2-diene double bond relative to those that lack this feature.

The shorter the C-3 to C-l 7 distance the higher the affinity of a Glucocorticoid is for the Glucocorticoid Receptor

Although there is not a direct relationship between the C-3 to C-l 7 distance of the ligand in the homology model and ligand binding affinity (Table 4), it is notable that this distance is greatest for cortisol and its 9-fluorinated analog. When compared to the other ligands, both lack a 1 ,2-diene double bond in the A-ring and with the exception of triamcinolone, they have the highest free binding energies (together with prednisolone).

Explanation of Mutational Data

The HSP90 heat-shock protein is required for GR ligand binding, but as a part of activation of GR which is conceived as a steroid-induced conformation change of Gr necessary for DNA-binding and glucocorticoid dependent transcriptional regulation) it dissociates from the receptor. The HSP90 interaction sites on the GR surface have been mapped with peptide competition studies. It was found that the critical contact site is located in the region between residues 632-659 of mouse GR. In our model this region (626-653) in hGR) corresponds to S1-S2 hairpin -sheets and H6-H7 α-helices which constitute a part of the protein surface and which also line the binding cavity. Thus, our GR model is consistent with what is known about the GR-HSP90 interactions.

A number of mutations of GR in various species have been described. Most of the mutations in the GR LBD completely disrupt the function of the receptor, presumably by perturbation of the fold of the LBD. Those detrimental mutations are not useful for validation of a GR homology mode. On the other hand, mutations in the LBD that affect ligand binding specificity and/or GR ligand-dependent activation/repression of transcription without totally activating the protein are of the greatest interest for our invention, validation of a GR homology model, since they ought to be located near the binding site. Such residues whose mutations modulate the function of GR are (in terms of the human GR sequence): P541, M565, G567, A573, M601, C638, D641, C643, M646, L653, C665, E668. V729, C736, T774, 764, F774, and these are shown as purple balls in the receptor model in Fig. 10a and 10b. It can be seen that they seem to cluster around the ligand site and/or on HI 2 or its vicinity. Most of the mutations which are non-detrimental to protein activation are within 7A of the ligand, i.e. on the parts closest to the binding site of the helices that line the binding cavity (Table 5). Most of the mutated side-chains had direct ligand contact before the mutation (c.f. the inventory of the residues lining the binding cavity). That those mutations affect ligand biding affinity and/or ligand-dependent transactivation is thus consistent with our mode.

Table 4

Summary of Ligand Binding Affinities and C-3/C-17 Distances (A) within Homology Model

Table 5. Mutations affecting GR transactivation /ligand affinity within GR.

Mutation Decreased affinity for DEX Decreased transactivation with DEX

P541A nt > 100-fold

L563F 6-fold 15 - 60-fold

M601L 3-fold nt

C638Y Normal Normal

C638G 10-fold INCREASED 10-fold INCREASED

C638W Normal Normal

D641G Normal no activity

D641V Normal no activity

C643R nt Slightly reduced

C643G nt 2 - 3-fold

C643S 4-fold nt

M646T nt Slightly reduced

L653S Reduced Reduced

C665A nt > 100-fold

C665S nt 4-fold

M666I 4-fold 10 - 25-fold

E688K Normal no activity

V729I 2-fold 4-fold

C736S 2-fold 10-fold

C736G nt no activity

T744I nt Reduced

I747T 2-fold 50-fold

L753F Normal no activity

Y764N 2-fold 3 - 4-fold

F774A 6-fold 20-fold

DEX = dexamethasone nt = not tested Some mutated residues were remote from the ligand, or did not have direct ligand contact. These mutations were further scrutinised, in order to confirm that they do not invalidate our GR homology model. Mutations in residues <523 and residues >761 are in the hinge or on the C-terminal part of H12 that is not included in the template X-ray structure and outside our GR model so they cannot be used for model validation.

The P541A mutation in Gr results in that more than 100-fold increase in steroid concentration is needed to preserve biological activity, presumably due to decreased steroid affinity. Although this mutation is remove from the binding site it is at the very beginning of the strand between HI and H3 (at the end of HI) and it may therefore be critical for its conformation. This strand (consisting of residues 540-560) delimits the binding site with residues 542-544. Therefore, a conformational change in this strand brought about by the P541 A mutation could affect ligand binding from a distance.

The patient mutation V762I which reduces affinity for dexamethasone 2-fold and the C643S mutation, which reduces affinity 4-fold, are located in proximity of the ligand binding site but the side chains of the residues are not directed towards the steroid. Thus, these residues do not appear to be in direct contact with the steroid and secondary effects of these mutations would be required to explain the difference in ligand affinity. The L653S and F744 A mutants that were reported by Garabedian and Yamamoto were tested in a yeast expression system, in which binding receptor binding assays were not performed so their effect may be exerted on transcription only, and need not necessarily directly affect ligand binding.

The I747T mutation reported by Roux is 9-lθA from the ligand, so it is difficult to explain why it decreases binding affinity with our model. The E688K mutation has been reported by Garabedian and Yamamoto to abolish transcriptional activity of Gr expressed in yeast and COS cells and to decrease the affinity for dexamethasone. This residue is located on H9, as far from the binding site as you can get in the GR LBD, located near HI and facing outwards. It is here HI connects to the hinge region to the DBD and thus it is plausible that the effect of this mutation is caused by disturbances of the interaction with DBD, rather than an direct effect on the binding site. The C655A/S mutants required a 100-fold higher steroid concentration for biological activity of the mouse GR, but C665S in human GR has no effect on binding affinity. This residue in our model is located in H8, remove from the ligand, but since it does not affect binding affinity it does not have to be located near the binding site to validate our model. Hence, the inhibition of binding observed by Lee et al, by the mutation of the neighbouring residue, M666I, is then difficult to rationalise. However, their other mutation affecting binding affinity, L563F is near the C-ring of the steroid's α-face, and thus consistent with the model.

Warriar et al, have described that the point mutations M565R and A573Q greatly enhance the affinity of Gr for dexamethasone, whereas the G567A mutant fails to bind ligands efficiently. In our GR model these residues are relatively close to the binding site. The Cα's of M565, G567 and A573 are ~7A from Cl, 3 A from C2 to 11 A from the C3-carbonyl of dexamethasone, respectively. IN our model G567 and A574 are facing the binding cavity, in contrast to M565, which is directed from it( and constitutes part of the protein surface). Thus the M565R mutation should not affect the ligand affinity of GR to a great extent if our model is correct. In the context of another study (Lind. U., Wright, A.P.H. and Carlstedt-Duke, J., ms. in preparation) a random combination of GR to mineralocorticoid receptor mutations (between residues 565-574) was created and screened for activity with glucocorticoids and mineralocorticoids. We found in agreement with Warriar et a., that the A573Q mutant had increased activity with dexamethasone, and that the G567A mutation inactivated GR. Finally, in contrast to Warriar et al, it was also found that the M565R mutant had no effect on the activity of dexamethasone which is in better agreement with the GR model proposed.

Thus, of the mutations described in the literature affecting ligand affinity, only two (M666I, 1747T) cannot easily be accounted for by the present GR-model.

Carlstedt-Duke et al have demonstrated three residues to be in the vicinity of the binding site of GR by covalent affinity labelling with radiolabelled Gr ligands with protein sequencing studies, Cys-638 is affinity-labelled through an electrophilic group of the 21 -position in dexamethasone 21-mesylate reacting with a thiol. There is an absolute requirement for the C-20 carbonyl group for this labelling reaction. In our model the sulfur of Cyc-638 is located 7A from the hydroxyl oxygen on C-21 of dexamethasone. The Cys-736 sulfur is even closer, which may explain why this residue is preferentially labelled.

Regarding the other two affinity labels, the chemistry is not known following the photoactivation of UV-light of the bound TA ligand. IN our GR-model the C-18 and C-l 9 methyl groups of Ta are located within 4 A of Cys-736S, and 5.6 A of Met-604S, respectively. If the reaction mechanism involves direct binding to the A- or B-ring, the affinity labelling of Cys-736 would require that the ligand be flipped with respect to A- and B-ring location at the time of the reaction. These two support the orientation of the steroid. In an earlier GR model the orientation of the dexamethasone is reversed with respect to the location of the A- and D-rings. With that orientation the Cys-736 is closer to the A-ring, but the other two residues (Met-604 and Cys-638) involved in the covalent affinity labelling are much further away, as compared to our orientation of the steroid. Thus, the orientation of the steroid in our model is more congruent with the affinity labelling data previously known.

Figures 12 and 13 show the -ray crystalography data obtained from the GR models produced using the X-ray crystallographic structures of TR, ER and PR as templates.

References

1. Muck, A.; Guyre, P.M. Glucocorticoid physiology and homoestatis in relation to anti-inflammatory actions; in Anti-inflammatory steroid actin: Basic and clinical aspects; Academic Press: 1989, pp 30-47

2. Gronemeyer, H.; Luadet, V.; "Transcription factors 3:nuclear receptors"; Protein Profile 1995, 2, 1 173-1308.

3. Zeelan, F.J. In Medicinal Chemistry of Steroids. Elsevier: Amsterdam 1990.

4. Wolff M.E.; Baxter, J.D.; Kollman, P.A.; Lee, D.L.; Kuntz, I.e.; Bloom, E.; Matulich, D.T.; Morris, J.G.N; "Nature of steroid-glucocorticoid receptor interactions: thermodynamic analysis of the binding reaction."; Biochemistry. 1978, 3201-3208. 5. Wagner, R.L.; Apriletti, J.W.; McGrath, M.E.; West, B.L.; Baxter, J.D.; Fletterick, R.J.; "A structural role for hormone in the thyroid hormone receptor"; Nature 1995, 378, 690-697.

6. Brzozowski AM, Pike AC, Dauter Z, Hubbard RE, Bonn T, Engstrom O, Ohman L; Greene G L, Gustafsson J A; Carlquist M.; "Molecular basis of agonism and antagonism in the oestrogen receptor Nature. 1997, 389, 753-758.

7. Bourguet W, Ruff M, Chambon P, Gronemeyer H, Moras; "Crystal structure of the RAR-gamma ligand-binding domain bound to all-trans retinoic acid." Nature 1995, 378, 377-382.

Claims

1. A method of designing a homology model of the ligand binding domain of a

glucocorticoid receptor wherein the homology model may be displayed as a

three-dimensional image, the method comprising:

(i) providing an amino acid sequence and an x-ray crystallographic structure of

the ligand binding domain of a thyroid, estrogen or progesterone receptor,

(ii) modifying said x-ray crystallographic structure to take account of

differences between the amino acid configuration of the ligand binding domains of the

glucocorticoid receptor on the one hand and the thyroid, estrogen, or progesterone receptor

on the other hand,

(iii) verifying the accuracy of the homology model by comparing it with

experimentally-determined binding properties of a number of ligands for the glucocorticoid

receiptor, and

(iv) if required modifying the homology model for greater consistency with

those binding properties.

2. A method according to claim 1, wherein amino acids lining the binding cavity of

the glucocorticoid receptor are identified chemically and the information thus gained used

to verify and if appropriate modify the homology model.

3. A method according to claim 2, wherein (v) amino acids lining the binding cavity

of the glucocorticoid receptor are identified by chemically mutating the glucocorticoid

receptor so as to change one amino acid thereof, and experimentally determining how the mutation affects the binding properties of the receptor to one or more ligands known to

bind to the unmutated receptor.

4. A method according to claim 3, wherein step (v) is repeated one or more times,

each time changing a different amino acid of the glucocorticoid receptor.

5. A method according to claim 2, including: (vi) introducing a photolabile group into

a ligand having affinity for the glucocorticoid receptor, and forming a complex of the

modified ligand with the receptor, expose the resulting complex to light to cause it to

decompose to form reactive groups which bind to adjacent amino acids of the receptor,

break up the complex into peptide fragments, analysing fragments to identify those bound

to reactive groups from the modified ligand and thus the amino acids lining the binding

cavity of the receptor.

6. A method according to any preceding claim, wherein the homology model is

compared with the structures of other, similar, proteins.

7. A method according to any preceding claim, wherein the homology model is

checked and if necessary modified to ensure that it shows predominantly hydrophobic

amino acids lining the binding cavity and predominantly hydrophilic amino acids exposed to the outside.

8. A method according to any preceding claim, wherein the thyroid, estrogen or progesterone receptor is a human receptor.

9. A homology model of the ligand binding domain of a glucocorticoid receptor

designed in accordance with any preceding claim and having hydrogen bonding partners

for the C-l 6 and C-21 OH groups of a glucocorticosteroid.

10. A homology model according to claim 9, showing a hydrophilic cavity positioned

to interact with the C-3 carbonyl group of a glucocorticosteroid.

11. A homology model according to claims 8, 9 or 10, showing at least one small

hydrophobic pocket capable of interacting with a methyl or halogen substituent at the C-6

and/or C-9 position of a glucocorticosteroid.

12. Use of a homology model according to any of claims 9 to 11 to identify or design

ligands capable of binding to the ligand binding domain of a glucocorticoid receptor.

13. Use of a homology model according to any of claims 9 to 11 to identify or design

glucocorticoid receptor antagonists or agonists.

14. A glucocorticoid receptor antagonist or agonist identified by use of a homology

model according to any of claims 9 to 11.

15. A medicinal product comprising a glucocorticoid agonist according to claim 14 for

treatment of inflammation or for use in immunosuppressive therapy.

16. A medicinal product comprising a glucocorticoid antagonist according to claim 14

for use in the treatment of hypertension, diabetes, obesity, glaucoma, depression, AIDS,

and wounds.

17. A computer programmed with a homology model of the ligand binding domain of a

glucocorticoid receptor according to any of claims 9 to 11.

18. A machine-readable data-storage medium on which has been stored in

machine-readable form a homology model of the ligand binding domain of a glucocorticoid

receptor according to any of claims 9 to 11.

19. The use of a homology model according to any of claims 9 to 11 as input to a

computer programmed for drug design and/or database searching and/or molecular graphic

imaging in order to identify new ligands for the glucocorticoid receptor.

20. A computational and chemical method of iteratively generating a homology model

of the ligand binding domain of a glucocorticoid receptor, which homology model is

capable of being displayed as a three-dimensional image, the method comprising:

(i) entering into a computer an amino acid sequence and an x-ray

crystallographic structure of the ligand binding domain of a thyroid, estrogen or

progesterone receptor;

(ii) modifying under computer control said x-ray crystallographic structure to

take account of known differences between the amino acid configuration of the ligand binding domains of the glucocorticoid receptor on the one hand and at least one of the

thyroid, estrogen, or progesterone receptors on the other hand,

(iii) reconciling under computer control the resulting modified crystallographic

structure with the chemically-determined binding properties of a number of ligands for the glucocorticoid receptor;

(iv) identifying by chemical means the amino acids that line the binding cavity

of the ligand binding domain and reconciling under computer control the modified crystallographic structure with these;

(v) repeating steps (ii) and (iii).