REFERENCE TO TABLES SUBMITTED ON A COMPACT DISC This application includes a compact disc (submitted in duplicate) containing Table 1 and Table 2. The tables are identified on the compact disc as follows. File Name Date of Creation Size (bytes) Table 1 February 6, 2002 828 kB Table 2 February 6, 2002 831 kB The entire content of each of these files is hereby incorporated by reference.

BACKGROUND Integrins are adhesion receptors that mediate vital bi-directional signals during morphogenesis, tissue remodeling and repair (reviewed in M. J. Humphries, Biochem Soc Trans 28,311-39 (2000) ). Integrins are heterodimers formed by noncovalent association of an a and a (3 subunit, both type I membrane proteins with large extracellular segments. In mammals, eighteen a and eight ß subunits assemble into 24 different receptors. Integrins depend on divalent-cations to bind their extracellular ligands. Although these ligands are structurally diverse, they all employ an acidic residue during integrin recognition. Specificity for a particular ligand is then determined by additional contacts with the integrin. High affinity binding of integrins to ligands is usually not constitutive, but is elicited in response to cell"activation" signals (so-called"inside-out'signaling) that alter the tertiary and quaternary structure of the extracellular region, making the integrin ligand-competent. Ligand binding in turn induces structural rearrangements in integrins that trigger"outside-in"signaling

(reviewed in E. F. Plow, T. A. Haas, L. Zhang, J. Loftus, J. W. Smith, J Biol Chem 275,21785-8. (2000) ).

Integrins can be grouped into two classes based on the presence or absence of a-180 amino acid A-type domain (aA or I domain ; see M. Michishita, V. Videm, M.

A. Arnaout, Cell 72,857-867 (1993) ). In the nine aA-containing integrins (aA- integrins), aA is the major ligand binding site. Thus, isolated aA binds directly and in a divalent-cation-dependent manner to physiologic ligands with affinity equal to that of the respective ligand-competent heterodimer (Michishita et al., supra). The structures of isolated aA domains in"liganded" (high-affinity) and"unliganded" (low-affinity) conformations (J. -O. Lee, P. Rieu, M. A. Arnaout, R. Liddington, Cell 80,631-638. (1995); J. P. Xiong, R. Li, M. Essafi, T. Stehle, M. A. Arnaout, J Biol Chem 275,38762-7. (2000) ) have revealed how this domain interacts with ligands. A metal ion is coordinated at the ligand-binding interface of aA through a conserved five amino acid motif, the metal-ion-dependent-adhesion-site (MIDASF), and the metal coordination is completed by a glutamate from the ligand (Lee et al., supra ; J.

Emsley, C. G. Knight, R. W. Farndale, M. J. Barnes, R. C. Liddington, Cell 100,47- 56 (2000) ) or, in its absence, by a water molecule (J. -O. Lee, L. Anne-Bankston, M.

A. Arnaout, R. C. Liddington, Structure 3,1333-1340 (1995) ). In aA-lacking integrins, such as the aVß3 integrin, it has been proposed that ligand recognition is supported by an aA-like domain (aA) present in all integrin P subunits (Lee et al., supra). The a subunit is believed to participate in ligand recognition but the precise location of its ligand binding site is not known (reviewed in Humphries, supra).

SUMMARY The invention features methods for identifying molecules which will bind to the aA-lacking integrin aVß3 receptor extracellular domain and modulate function, e. g. , by acting as receptor agonists, such as agonists that induce some or all of the biological responses induced by the ligand Arg-Gly-Asp- (D-Phe)- (N-methyl-Val), although the methods are not limited to the identification of this particular class of agonist. The structural information described herein can also be used to identify various agonists and inhibitors of aVß3 activity. Preferred mimetics and agonists identified using the method of the invention act as agonists in one or more in vitro or

in vivo biological assays of the activity of a natural ligand of aVß3 receptor activity.

The methods of the invention entail identification of compounds having a particular structure. The methods rely on the use of precise structural information derived from x-ray crystallographic studies of the extracellular domain of αVß3 receptor complexed with a representative peptide ligand. This crystallographic data permit the identification of atoms in the peptide ligand mimetic that are important for receptor binding and activation. More importantly, these data define a three dimensional array of the important contact atoms. Other molecules which include a portion in which the atoms have a similar three dimensional arrangement similar to some or all of these contact atoms are likely to be capable of acting as a receptor ligand.

In one aspect, the invention provides a method for determining whether a test compound is a potential modulator of αVß3 integrin, the method comprising: a) providing a computer model of the three-dimensional structure comprising a binding site of αVß3 integrin defined by the atomic coordinates of αVß3 integrin amino acids aV : Ala215, aV: Asp218 ; aV : Aspl50 ; aV : Tyrl78; p3 : Tyrl22 ; (33 : Arg214 ; f33 : Asn215 ; and p3 : Arg216 according to Table 2; b) providing a computer model of the three dimensional structure of a test compound; c) computationally performing a fitting operation between the computer model of the binding site and the computer model of the test compound; and d) evaluating the results of the fitting operation to evaluate the ability of the test compound to bind aV (33 integrin; wherein a test compound having the ability to bind αVß3 integrin is a potential modulator of ocVß3 integrin.

In some embodiments, the computer model of the three-dimensional structure of a test compound is from a database of compounds of known structure. In some embodiments, the three-dimensional structure of the binding site of ocVß3 integrin is defined by the atomic coordinates of ocVß3 integrin amino acids according to Table 2.

In some embodiments, the fitting operation comprises determining an energy minima configuration of computer model of the three-dimensional structure of the test

compound in the computer model of the three-dimensional structure of αVß3 integrin.

In some embodiments, the test compound is: (i) computationally assembled molecular fragments; (ii) selected from a small molecule database; or (iii) computationally created by de novo ligand design.

In a second aspect, the invention provides a method for determining whether a test compound is a potential modulator of aVß3 integrin, the method comprising: a) providing a computer model of the three-dimensional structure comprising a binding site of αVß3 integrin defined by the atomic coordinates of αVß3 integrin amino acids aV: Ala215, aV: Asp218 ; aV: Aspl50 ; aV: Tyrl78 ; ß3 : Aspll9 ; ß3 : Ser121 ; ß3 : Tyrl22 ; ß3 : Serl23 ; p3 : Aspl26 ; ß3 : Aspl27 ; p3 : Aspl58 ; ß3 : Arg214 ; ß3 : Asn215 ; ß3 : Arg216 ; p3 : Asp217 ; ß3 : Ala218 ; p3 : Pro219 ; ß3 : Glu220 ; and p3 : Asp251 according to Table 2; b) providing a computer model of the three dimensional structure of a test compound; c) computationally performing a fitting operation between the computer model of the binding site and the computer model of the test compound; and d) evaluating the results of the fitting operation to evaluate the ability of the test compound to bind aVß3 integrin; wherein a test compound having the ability to bind aVß3 integrin is a potential modulator of αVß3 integrin.

In some embodiments, the computer model of the three-dimensional structure of a binding site of αVß3 is further defined by the inclusion of the atomic coordinates of one or more divalent cations according to Table 2. In some embodiments, the computer model of the three-dimensional structure of a test compound is from a database of compounds of known structure. In some embodiments, the three- dimensional structure of the binding site of αVß3 integrin is defined by the atomic coordinates of αVß3 integrin amino acids according to Table 2. In some embodiments, the fitting operation comprises determining an energy minima configuration of computer model of the three-dimensional structure of the test compound in the computer model of the three-dimensional structure of aVß3.

In a third aspect, the invention provides a method for determining whether a test compound is a potential modulator of aVß3 integrin, the method comprising: a) providing a computer model of the three-dimensional structure comprising a binding site of αVß3 defined by the atomic coordinates of αVß3 amino acids aV: Ala215, aV: Asp218 ; aV : Aspl50 ; aV: Tyrl78; j33 : Aspll9 ; 33 : Ser121 ; p3 : Tyrl22 ; (33 : Serl23 ; ß3 : Aspl26 ; ß3 : Aspl27 ; (33 : Aspl58 ; ß3 : Arg214 ; ß3 : Asn215 ; p3 : Arg216 ; ß3 : Asp217 ; ß3 : Ala218 ; f33 : Pro219 ; ß3 : Glu220 ; and ß3 : Asp251 according to Table 2; b) providing a computer model of the three dimensional structure of a test compound; c) computationally performing a fitting operation between the computer model of the binding site and the computer model of the test compound; d) evaluating the results of the fitting operation to evaluate the ability of the test compound to bind aVß3 integrin; e) electing a test compound having the ability to bind aV (33 integrin as a potential modulator of αVß3 integrin; f) obtaining or synthesizing the potential modulator; and g) evaluating the ability of the potential modulator to modulate the activity of aVß3 integrin.

In some embodiments, the computer model of the three-dimensional structure of a binding site of αVß3 integrin is further defined by the inclusion of the atomic coordinates of one or more divalent cations according to Table 2. In some embodiments, the fitting operation comprises determining an energy minima configuration of computer model of the three-dimensional structure of the test compound in the computer model of the three-dimensional structure of αVß3 integrin.

In some embodiments, the evaluating comprises determining the binding affinity of the test compound for αVß3 integrin.

In a fourth aspect, the invention provides a method for determining whether a test compound is a potential modulator of αVß3 integrin, the method comprising: a) providing a computer model of the three-dimensional structure of cyclo (RGDf-N-Me-V) according to Table 1 or Table 2;

b) providing a computer model of the three dimensional structure of a test compound; c) computationally comparing the computer model of the binding site and the computer model of the test compound; and d) evaluating the results of the comparison to evaluate the ability of the test compound to bind aVß3 integrin; wherein a test compound having a structure similar to cyclo (RGDf-N-Me-V) is a potential modulator of aVß3 integrin.

In a fifth aspect, the invention provides a method for determining whether a test compound is a potential modulator of aVß3 integrin, the method comprising: a) providing a computer model of the three-dimensional structure comprising an active site groove of aVß3 integrin defined by the atomic coordinates of aVß3 integrin amino acids aV: Ala215, aV: Asp218 ; aV : Aspl50 ; aV: Tyrl78 ; ß3 : Tyrl22 ; ß3 : Arg214 ; ß3 : Asn215 ; and ß3 : Arg216 according to Table 2; b) providing a computer model of the three dimensional structure of a test compound; c) computationally performing a fitting operation between the computer model of the active site groove and the computer model of the test compound; and d) evaluating the results of the fitting operation to evaluate the ability of the test compound to interact with the active site groove of aVß3 integrin; wherein a test compound having the ability to interact with the active site groove of aVß3 integrin is a potential modulator of aVß3 integrin.

In some embodiments, the computer model of the three-dimensional structure of a test compound is from a database of compounds of known structure. In some embodiments, the three-dimensional structure of the active site groove of aVß3 integrin is defined by the atomic coordinates of aVß3 integrin amino acids according to Table 2. In some embodiments, the fitting operation comprises determining an energy minima configuration of computer model of the three-dimensional structure of the test compound in the computer model of the three-dimensional structure of aVß3 integrin. In some embodiments, the active site groove is formed by the D3-A3, A3- B3, and D4-A4 loops (as shown in FIG. 2A).

In a sixth aspect, the invention provides a method for evaluating the potential of a chemical entity to associate with: a) a molecule or molecular complex comprising a binding pocket defined by the atomic coordinates of αVß3 integrin amino acids aV : Ala215, aV: Asp218 ; aV : Aspl50 ; aV: Tyrl78 ; ß3 : Tyrl22 ; ß3 : Arg214 ; ß3 : Asn215 ; and ß3 : Arg216 according to Table 2, or b) a homologue of the molecule or molecular complex, wherein the homologue comprises a binding pocket that has a root mean square deviation from the backbone atoms of the amino acids of not more than 1.5 A, the method comprising: i) employing computational means to perform a fitting operation between the chemical entity and a binding pocket defined by the structure coordinates of αVß3 integrin amino acids aV: Ala215, aV: Asp218 ; aV: Aspl50 ; aV: Tyrl78 ; ß3 : Tyrl22 ; (33 : Arg214 ; p3 : Asn215 ; and ß3 : Arg216 according to Table 2 a root mean square deviation from the backbone of the amino acids of not more than 1.5 A ; and ii) analyzing the results of the fitting operation to quantify the association between the chemical entity and the binding pocket.

In some embodiments, the method evaluates the potential of a chemical entity to associate with: a) a molecule or molecular complex comprising a binding pocket defined by the atomic coordinates of aVß3 integrin amino acids aV: Ala215, aV: Asp218 ; aV: Aspl50 ; aV : Tyrl78; p3 : Aspll9 ; P3 : Ser121 ; p3 : Tyrl22 ; p3 : Serl23 ; p3 : Aspl26 ; ß3 : Aspl27 ; ß3 : Aspl58 ; ß3 : Arg214 ; p3 : Asn215 ; ß3 : Arg216 ; ß3 : Asp217 ; p3 : Ala218 ; ß3 : Pro219 ; ß3 : Glu220 ; and ß3 : Asp251 according to Table 2, or b) a homologue of the molecule or molecular complex, wherein the homologue comprises a binding pocket that has a root mean square deviation from the backbone atoms of the amino acids of not more than 1.5 A.

In a seventh aspect, the invention provides a method for identifying a potential modulator of molecule or molecular complex comprising αVß3 integrin-like binding pocket, the method comprising:

a) using the atomic coordinates of aVp3 integrin amino acids aV : Ala215, aV: Asp218 ; aV: Aspl50 ; aV : Tyrl78 ; ß3 : Tyrl22 ; ß3 : Arg214 ; ß3 : Asn215 ; and ß3 : Arg216 according to Table 2 a root mean square deviation from the backbone atoms of the amino acids of not more than 1.5 A, to generate a three-dimensional structural model of a molecule or molecular complex comprising an aV (33 integrin- like binding pocket; b) employing the three-dimensional structural model to design or select said potential modulator; c) synthesizing the potential modulator; and d) contacting the potential modulator with the molecule or molecular complex to determine the ability of the potential modulator to interact with the molecule or molecular complex.

In an eighth aspect, the invention provides a method for evaluating the potential of a chemical entity to associate with a molecule or molecular complex comprising a ligand binding pocket of an aVß3 extracellular domain, the method comprising: a) employing computational means to perform a fitting operation between the chemical entity and a binding pocket defined by the structural coordinates described in Table 1 or Table 2; and b) analyzing the results of said fitting operation to quantify the association between the chemical entity and the binding pocket.

In a ninth aspect, the invention provides a computer for producing a three- dimensional representation of a molecule or molecular complex, wherein said molecule or molecular complex comprises a binding pocket defined by structure coordinates of Table 1 or Table 2 wherein said computer comprises: a) a machine-readable data storage medium comprising a data storage material encoded with machine-readable data, wherein said data comprises the structure coordinates of Table 1 or Table 2 amino acids of the aVß3 extracellular domain; b) a working memory for storing instructions for processing said machine- readable data;

c) a central-processing unit coupled to said working memory and to said machine-readable data storage medium for processing said machine readable data into said three-dimensional representation; and d) a display coupled to said central-processing unit for displaying said three- dimensional representation.

In a tenth aspect, the invention provides a crystal comprising an integrin aVß3 extracellular domain complexed with a cyclic RGD peptide. In some embodiments, the cyclic RGD peptide ligand comprises an amino acid sequence comprising Arg- Gly-Asp- (D-Phe)- (N-methyl-Val). In some embodiments, the cyclic RGD peptide ligand comprises an amino acid sequence comprising Arg-Gly-Asp- (D-Phe)- (N- methyl-Val) in the presence of a metal.

Definitions As used herein,"integrin"and"integrin receptor"are used interchangeably.

"Integrin"or"integrin receptor"refers to any of the many cell surface receptor proteins, also referred to as adhesion receptors which bind to extracellular matrix ligands or other cell adhesion protein ligands thereby mediating cell-cell and cell- matrix adhesion processes. The integrins are encoded by genes belonging to a gene superfamily and are typically composed of heterodimeric transmembrane glycoproteins containing a-and (3-subunits. Integrin subfamilies contain a (3-subunit combined with different a-subunits to form adhesion protein receptors with different specificities. The integrins are grouped into two classes, those containing the aA domain and those that do not contain the aA domain. Both classes have a PA domain."aVß3"and"aVß3 integrin"are also used interchangeably, and refer to integrins comprising the aV and ß3 subunits.

"Compounds"refer to a chemical entity that can comprise a peptide or polypeptide, including antibodies, phage display antibodies, and their biologically active fragments, a small molecules, (e. g. , chemically synthesized or of natural origin), or synthetic peptides or polypeptides (e. g. , non-naturally occurring polypeptides, e. g. , peptoids or peptidomimetics). The compound has the ability to bind the integrin at an extracellular ligand binding site domain. As such,"ligand" refers to the naturally occurring ligand that binds the integrin so that it can perform its physiological function or functions. This site is distinguishable from the ligand

binding site. Compounds can include synthetic and naturally occurring mimetics of the CD loop or fragments of the CD loop of the ß tail domain (ßTD) that contacts the strand-F/a7 loop in the PA domain of an unliganded integrin, the al/strand-A loop or fragments of that loop that contact the hybrid domain of an unliganded integrin, and the ADMIDAS coordination site in the PA domain of an unliganded integrin. A "mimetic"has structural similarity or has similar binding properties as the entity it is mimicking. Compounds can comprise or consist of the mimetic. According to the invention, "compounds"generally refers to chemical entities as described above, selected from the group consisting of a) naturally occurring or synthetic peptides; b) naturally occurring or synthetic polypeptides; c) small molecules, typically small molecules as defined below; d) antibodies, typically antibodies selected from the group consisting of monoclonal antibodies, polyclonal antibodies, chimeric antibodies, antibody fusions, and the like; e) antibody fragments, generally antibody fragments of antibodies as given under d); and f) non-peptidic organic molecules, typically having a formula weight above 150 g/mol and generally less than 1500 g/mol, more typically above 250 g/mol and preferably less than 800 g/mol.

"Modulation"refers to regulating or changing the activatability or ligand binding of an integrin. For example, a modulator can inhibit or promote the activation of the integrin or it can prevent or promote ligand binding to the integrin.

A"peptidomimetic"refers to a chemical variant of a polypeptide or a peptide in which the side chains of the polypeptide or peptide are substantially maintained in the variant, yet the chemical backbone of the peptidomimetic is altered relative to the polypeptide or peptide in at least one peptide bond.

A"peptoid"is an oligomer of N-substituted glycines. A peptoid can be synthesized from a variety of different N-alkylglycines that have side chains similar to amino acid side chains, (e. g. , as described in Simon et al. , (1992) PNAS 89: 9367- 9371). It can serve as a motif for the generation of chemically diverse libraries of novel molecules. As an alternative to natural polymers, it is a modular system that

allows one to synthesize monomers in large amounts. The monomers have a wide variety of functional groups presented as side chains off of an oligomeric backbone, the linking chemistry is high yielding and amenable to automation. The linkage in a peptoid is resistant to hydrolytic enzymes such as proteases. Another advantage is that the monomers are achiral.

A"small molecule"is a molecule of less than 32kDa, e. g., 0. 5kDa, lkDa, 5kDa, 10, kDa, 15kDa, 20kDa, 25kDa, 30kDa, or 32kDa.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Although methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, suitable methods and materials are described below. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control. In addition, the materials, methods, and examples are illustrative only and not intended to be limiting.

DESCRIPTION OF DRAWINGS FIG. 1A is a depiction of the structure of aVß3-Mn complexed with cyclo (RGDf-N-Me-V). This stereoview of a simulated-annealing omit map (cyan) of the peptide-integrin complex. Cutoff is 5.0 s for Mn2+ and 1.0 a for ligand residues.

Densities (magenta) of the adjacent metal ions at ADMIDAS, MIDAS and LIMBS (shown here and in subsequent figures as violet, cyan and gray respectively) are from the same map.

FIG. 1B is a depiction of a ribbon drawing (M. Carson, J. Mol. Graph. 5,103- 106. (1987)) ofthe aVß3-RGD-Mn structure. In this and subsequent figures, aV and , B3 are shown in blue and red, respectively. The peptide is bound at the propeller/ (3A domain interface with the ADMIDAS, MIDAS and LIMBS metal ions shown. The remaining five Mn2+ ions of the integrin (four in p-hairpin loops of blades 4-7 of the propeller and one at the aV genu) are shown in orange. The carbon, nitrogen and oxygen atoms of cyclo (RGDf-N-Me-V) are shown in yellow, blue and red

respectively. (inset) Surface representation (A. Nicholls, K. A. Sharp, B. Honig, Proteins 11,281-96 (1991)) of the integrin head, showing the location of the peptide binding site. The peptide is shown as a ball-and-stick model.

FIGS. 2A and 2B are depictions of the ligand-integrin binding site. FIG 2A depicts a surface representation of the ligand binding site, with the ligand peptide shown as ball-and stick model. Color code for the ligand and the two visible Mn2+ ions (MIDAS and ADMIDAS) is as in FIGS. 1 A and 1B. FIG 2B depicts interactions between ligand and integrin. The peptide (yellow) and residues interacting with the ligand or with Mn2+ ions are shown in ball-and-stick representation. aV and ß3 residues are labeled blue and red, respectively. Oxygen and nitrogen atoms are in red and blue, respectively. The three Mu2+ ions in (33 at MIDAS, ADMIDAS and LIMBS are also shown. Hydrogen bonds and salt bridges (distance cutoff = 3.5 A) are represented with dotted lines.

FIGS. 3A and 3B are depictions of the MIDAS motif in PA and FIGS. 3C and 3D are depictions of the aA from CDl lb. FIGS. 3A and 3B depict MIDAS residues (single letter) in unliganded and liganded PA, respectively. Coordinating side-chains are shown as ball-and-stick with oxygen atoms in red, carbon in green. The ligand aspartate is in gold. In addition to the ligand aspartate, the Mn2+ (cyan) in the PA MIDAS is coordinated directly with the hydroxyl oxygens of Serl21 and Serl23 and with one carboxylate oxygen from Glu220. The carboxyl oxygens of Asp 119 and Asp251 of PA lie within 6A of the metal ion and likely mediate additional contacts through water molecules similar to the liganded forms of aA (FIG 3D). The Mn2+ ion at ADMIDAS (magenta) is present in (A) and (B). The Mn2+ ions at MIDAS and at LIMBS (cyan and gray respectively) are only present in (B). FIGS. 3C and 3D depict MIDAS residues in unliganded and liganded aA from Cell, respectively. The metal ion (cyan) is present in both. Water molecules are labeled"co"; the pseudoligand glutamate is in gold. Hydrogen bonds and metal ion coordination are represented with dotted yellow lines.

FIGS. 4A, 4B, and 4C are depictions of the ligand-induced structural changes in PA in comparison with those of aA (from CD1 lb). FIG. 4A depicts a superposition, in stereo, of the aVß3-Mn (gray) and aVß3-RGD-Mn (red) structures.

The superposition is based on the Ca atoms of the central a-sheet (43 atoms per

structure, rmsd = 0.42 A). Residues of aVß3-RGD-Mn with a distance of more than 1.5 A to corresponding residues of aVß3-Mn are shown with thicker red lines. The major structural changes in PA involve helices al, al', a2, the Fa-7 loop and the ligand-specificity region. FIG. 4B depicts a magnified view of the rearrangements at the ligand binding site in PA. Superposition of the propeller and PA domains of aVß3-Mn (gray) and aVß3-RGD-Mn (aV, blue; ß3, red) is based on the Ca atoms of the aV propeller domain. The directions of protein movements (including the 4A displacement of Mn2+ at ADMIDAS) are indicated by red arrows. This view differs from (A) by a rotation of 180° around a vertical axis. FIG. 4C depicts a superposition, in stereo, of the"liganded" (red) and"unliganded"forms of aA from the CD1 lb integrin. The metal ion sphere at MIDAS is in cyan. The superposition is based on the Ca atoms of the central (3-sheet (43 atoms, rmsd = 0.43 A). Residues of liganded aA with a distance of more than 1.5 A to corresponding residues of unliganded aA are shown with thicker red lines. The major structural changes in aA involve helices al, a7, the F-a7 and E-a6 loops. Arrows (red) indicate the direction of the major protein movements in each case.

FIGS. 5A and 5B are schematic depictions of a model for the role of PA in ligand recognition by all integrins. FIG. 5A is a cartoon representation of the propeller (blue) and PA (pink) domains in liganded aVß3, with RGD (yellow) contacting both subunits. The ligand D contacts the metal ion in MIDAS directly.

FIG. 5B is a hypothetical model of the propeller PA domains in an aA-containing integrin. aA (light blue) projects from a loop in the propeller. We envision that an alpha-subunit derived invariant Glu at the base of the a7 helix of aA coordinates the PA MIDAS, acting as an endogenous ligand. This may occur when the a7 helix down-shifts by 10A, perhaps in response to inside-out signals, thus favoring binding of aA to ligand (E, dotted box). This model also allows for outside-in signaling by a liganded aA. The MIDAS metal ion is in cyan.

FIG. 5C depicts the conservation of the putative Glu in the a subunits of aA- integrins. Alignment of the primary sequence of all nine aA-integrin a subunits was carried out with CLUSTALW 1.8 (F. Jeanmougin, J. D. Thompson, M. Gouy, D. G.

Higgins, T. J. Gibson, Trends Biochem Sci 23,403-5. (1998) ). Only the sequence surrounding the putative endogenous Glu ligand (red) is shown. The aA c-terminal

helix residues are underlined (based on structures of CD11b, CD11a, al and a2) (Lee et al., supra ; A. Qu, D. J. Leahy, Proc. Natl. Acad. Sci. (USA) 92,10277-10281 (1995); M. Nolte et al. , FEBS Lett 452,379-85 (1999); J. Emsley, S. L. King, J. M.

Bergelson, R. C. Liddington, J Biol Chem 272,28512-7 (1997) ). Secondary structure predictions of the cup tetrapeptide and the A4 strands of the alpha subunit (J. P. Xiong et al., Science 294,339-45. (2001) ) are underlined. In the figure, amino acids are represented in single letters.

DETAILED DESCRIPTION Described below is structure of a aVß3-Mn complex and a aVß3-RGD-Mn complex. The structure was solved by x-ray crystallography. The coordinates of the atoms in the aVß3-Mn complex are presented in Table 1. The coordinates of the aVß3-RGD-Mn complex are presented in Table 2. Data collection and refinement statistics are in Table 3.

Expression, purification and crystallization of extracellular aVß3 were carried out as described (Xiong et al., supra). The integrin-ligand complex (aVß3-RGD-Mn) was generated by soaking aVß3-Ca crystals for three days at 4°C in 100 mM MES pH 6.0, 100 mM NaCI, 5mM MnCl2 and 2.4 mM cyclo (RGDf-N-Me-V). Cyclo (RGDf-N- Me-V) competes with binding of physiologic ligands to native or extracellular aVß3 (IC50 0. 5-3 nM) ( R. J. Mehta et al. , Biochem J 330,861-9. (1998); M. A.

Dechantsreiter et al. , J Med Chem 42,3033-40. (1999) ), and is in clinical trials as an anti-angiogenic therapeutic. Binding of vitronectin, fibronectin and fibrinogen to extracellular aVß3 was unaffected by the crystallization buffer used here (data not shown). Crystals of aVß3-Mn were grown by the hanging drop method as described (Xiong et al., supra), with 5 mM MnCl2 replacing CaCl2 in the crystallization buffer (Table 3). All protein crystals were cryoprotected in 24% glycerol, and data were collected at 100 K at the APS beamline ID-19. Crystals of all three structures are isomorphous (Table 3). Diffraction data were sharpened with a B-factor of-40. 0A2.

The aVß3-Mn and aVß3-RGD-Mn structures were solved by molecular replacement at 3.3 A and 3.2 A resolution respectively, using the previously reported aVß3-Ca structure as the initial model. For the aVß3-Mn structure, the original coordinates were modified by removing all six calcium ions. Coordinates were then subjected to

rigid body minimization. A Fo-Fc difference map showed positive density in all six previously determined calcium-binding sites. The eight metal ion densities in the structure were all assigned as manganese because crystals were soaked with buffer containing 5mM MnCl2. The positions of manganese ions were confirmed from the anomalous difference Fourier maps using data collected at the wavelength 1.2398 A, where Mn2+ has reasonable anomalous contribution (f"=1. 96 electrons). For the aVß3-RGD-Mn structure, a similar Fo-Fc difference density map showed clear density for all five amino acids of the cyclic peptide ligand and for eight Mn2+ ions: six at the original sites and two in the vicinity of the ligand. Each model was then modified according to the difference density features and refined using bulk solvent correction and several rounds of simulated annealing protocols in XPLOR (A. T.

Briinger et al. , Acta Crystallogr D Biol Crystallogr 54,905-21. (1998) ). About 5% of reflections were used to calculate the free R-factor in each case; the reflections included in the two"free sets"were the same as those used for the aVß3-Ca structure determination.

TABLE 3 Data collection statistics* αVß3-Mn αVß3-RGD-Mn Space group P3221 P3221 Unit cell dimensions (A) a = b = 130. 43, c = 310. 32 a=b=129. 79, c=308.78 Resolution (A) 50.0-3. 3 50.0-3. 2 Completeness 94.4 (86.1) 99.9 (99.4) Unique reflections 44739 (4011) 48911 (4853) Redundancy 4.8 6.7 9. 5 (43. 9) 16.4 (40.0) I/# 16.1 (2.5) 14.0 (3.7) Refinement statistics Resolution (A) 20.0-3. 3 20.0-3. 2 Rfactorif (%) (work set) 24.2 24.9 Rtactor (%) (free set) 32.3 33.0 Average B factor (A) 44.8 36.0 Protein atoms in the model 11651 11692 (including peptide) Number of Glc-NAc 11 11 Number of Mn2+ 6 8 Model statistics Rmsd&num from ideality Bond lengths (A) 0.01 0.009 Bond angles (°) 1.7 1.7 Dihedral angles (°) 26. 4 26.7 * Values in parentheses are for the highest resolution shell.

Rsym = I, where I is the observed intensity and <I> is the average intensity from multiple observations of symmetry-related reflections.

Rfactor = #hkl #Fobs(hkl) - Fcalc(hkl)#/ #hkl Fobs(hkl).

# root mean square deviation The crystal structure of the extracellular segment of the aA-lacking integrin

aVß3 was recently determined in the presence of Ca2+ (aVß3-Ca) (Xiong et al., supra). The aVß3-Ca structure has twelve domains assembled into an ovoid head and two"legs. "The head is primarily formed of a seven-bladed 0-propeller domain from the aV subunit [SEQ ID NO : 2] and a PA A-type domain from the ß3 subunit [SEQ ID NO : 3]. In solution, the crystallized aVß3 segment binds its cognate ligands in a manner indistinguishable from that of the native high-affinity receptor (R. J. Mehta et al. , supra), suggesting that it is ligand-competent.

Described herein is the structure of the extracellular segment of integrin aVß3 in complex with the cyclic pentapeptide ligand Arg-Gly-Asp-(D-Phe)-(N-methyl-Val) "cyclo (RGDf-N-Me-V)" [SEQ ID NO : 1] or"RGD peptide" (M. A. Dechantsreiter et al. , supra) in the presence of Mn2+ (αVß3-RGD-Mn) is described herein (Table 2).

The structure of the unliganded aVß3 protein in the presence of Mn2+ (aVß3-Mn) has also been determined for comparison and is described herein (Table 1). The two structures contain the previously reported extracellular residues of the integrin (Xiong et al., supra). In addition, aVß3-Mn contains six Mn2+ ions (replacing each Ca2+ ion in αVß3-Ca), and aVp3-RGD-Mn contains the cyclic pentapeptide plus eight Mn2+ ions. Replacement of Ca2+ with Mn2+ at all six sites in the αVß3-Mn structure did not result in significant structural rearrangements in the integrin. As with aVp3-Ca (Xiong et al., supra), no metal ion is visible at MIDAS in αVß3-Mn, although the millimolar concentration of Mn2+ used far exceeds those needed for maximal ligand binding (R. J. Mehta et al., supra). FIGS. 1A and 1B show respectively a representative electron density omit map and a ribbon diagram of the integrin- pentapeptide complex.

The αVß3-RGD-Mn structure reveals that the pentapeptide inserts into a crevice between the propeller and PA domains, on the integrin head (FIG. 1B, and inset). As the protein in the crystal binds ligand, this confirms that αVß3-Mn is in a "ligand-competent"state. The bound peptide is pentagonal as expected; its side chains radiate outwards. The RGD peptide sequence makes the main contact area with the integrin, and each residue participates extensively in the interaction, which buries 355 Å2 or 45% of the total surface area of the peptide. Arg and Asp side chains point in opposite directions, exclusively contacting the propeller and PA domains, respectively. Thus the peptide clamps the subunits together at the head. The five Ca-

atoms of the cyclic peptide form a slightly distorted pentagon, with the Ca atoms of Arg and Asp farther apart from each other (distance = 6.4 A) than the Ca atoms of Gly and D-Phe (distance = 5. 4 A). The NMR structure of this peptide in solution (M.

A. Dechantsreiter et al., supra) suggested a more regular pentagonal conformation.

Molecular dynamics simulations of the peptide in the absence of the integrin, performed using the same geometric parameters used for the crystallographic refinement, also result in a more regular pentagonal shape with roughly equal inter- Ca-atom distances (data not shown). Thus, distortion of the peptide ring apparently develops following contact with aVß3-Mn. The main chain conformation of the RGD motif in the pentapeptide is almost identical to that of the RGD tripeptide in the natural ligand Echistatin (R. A. Atkinson, V. Saudek, J. T. Pelton, Int J Pept Protein Res 43,563-72. (1994) ), suggesting that the structure presented here can serve as a basis for understanding the interaction of integrins with other RGD-containing ligands.

The Arg side-chain inserts into a narrow groove at the top of the propeller domain, formed by the D3-A3, A3-B3 and D4-A4 loops (FIG. 2A). The arginine guanidinium group is held in place by a bidentate salt bridge to aV: Asp218 at the bottom of the groove, and by an additional salt bridge to aV: Aspl 50 at the rear. The hydrophobic portion of the Arg side-chain is sandwiched between the side-chains of aV residues Tyrl78 and Ala215, which form the walls of the groove (FIG. 2B). The contacts leave most of the upper portion of the Arg side-chain exposed to solvent, while the spacious rear of the groove probably contains water molecules that may provide additional contacts to the Arg guanidinium group.

Contacts between the ligand Asp and PA primarily involve the Asp carboxylate group, which protrudes into a cleft between the PA loops A'-al and C'- a3 and forms the center of an extensive network of polar interactions (FIGS. 2A, 2B).

One of the Asp carboxylate oxygens contacts a Mn2+ ion at MIDAS in PA (FIG. 2B); it also hydrogen bonds to the carbonyl oxygen of ß3 : Tyrl22. The second Asp carboxyl oxygen forms hydrogen bonds with the backbone amides of ß3 : Tyrl22 and ß3 : Asn215 (FIG. 2B) and also contacts the hydrophobic portion of the 03 : Arg214 side-chain. Additional contacts involve the hydrophobic portion of the Asp side-chain

and the CB atom of ß3 : Asn215. Unlike the ligand Arg, the ligand Asp side-chain is completely buried in the complex.

The glycine residue, which completes the prototype RGD ligand sequence, lies at the interface between the a and (3 subunits. It makes hydrophobic interactions with Arg216, Asp217 and Ala218 of PA and with Tyrl78 of aV (FIG. 2B). The most critical of these interactions appears to be the contact with the carbonyl oxygen of ß3 : Arg216, which closely approaches the CA atom of the glycine. The remaining two residues of the pentapeptide face away from the ap interface and are not in the consensus ligand sequence. The D-Phe side-chain packs against PA residues Tyrl22 and Serl23 ; the N-methyl valine does not contact the integrin.

It is striking that the peptidal aspartate contacts PA in a manner that closely resembles the interaction of aA with its ligands (Lee et al., supra ; Emsley et al., supra) (FIG. 3): in both cases, an acidic ligand residue coordinates the receptor via a metal ion in MIDAS. Since MIDAS in the unliganded aVß3-Ca or aVß3-Mn structures does not contain a metal ion, it appears that completion of the coordination by the ligand carboxylate is required for high-affinity metal binding here. This differs from aA, which can bind a metal ion in MIDAS without a ligand (Qu et al., supra ; Lee et al., supra). Thus the MIDAS site of PA has a lower affinity for cations than the very similar site in aA. The one difference between the two sites is the replacement of a Thr, which contacts the cation in liganded aA, with Glu220 in PA.

In the unliganded aV (33-Mn structure, the Glu220 side-chain intrudes into the MIDAS site, approaching the space where a cation would bind: it thus appears to reduce the affinity for cations at MIDAS through steric hindrance. In the liganded aVß3-RGD- Mn structure, the ß3 : Glu220 side-chain occupies a different position, allowing accommodation of a cation at MIDAS.

In addition to incorporating Mn2+ at MIDAS upon ligand binding, PA also unexpectedly incorporates a second Mn2+ ion. Only 6 A from MIDAS, this ion defines a ligand-induced metal-binding-site (LIMBS) formed by the other carboxylate oxygen of ß3 : Glu220, the side chains of ß3 : Asp158, ß3 : Asn215 and p3 : Asp217 and the carbonyl oxygens of p3 : Asp217 and p3 : Pro219 (FIG. 2B). Although the LIMBS Mn2+ ion does not contact the ligand, coordination of Mn2+ nevertheless depends on it. p3 : Asp158 and ß3 : Glu220 occupy different positions in the unliganded structure, and

the coordination sphere for LIMBS does not exist without the ligand. The incoming ligand Asp displaces ß3 : Asn215 and also shifts the ß3 : Glu220 side-chain, thus inducing this metal binding site. The most likely role of LIMBS is to stabilize the reoriented ß3 : Glu220 and to add conformational stability and structural rigidity to the ligand binding surface.

The present data provide the structural basis for the RGD consensus in aVß3 ligands, where even conserved substitutions such as Arg to Lys, Gly to Ala or Asp to Glu are not tolerated (R. J. Mehta et al., supra ; M. A. Dechantsreiter et al., supra) : the shorter side chain of Lys (vs. Arg) cannot make a bidentate salt bridge to ß3 : Asp218 in aVß3 ; interestingly, an Arg to Lys substitution is accommodated in the aIlbp3 integrin (R. M. Scarborough et al. , J Biol Chem 268,1066-73. (1993) ), which lacks an aV: Asp218-corresponding residue and likely contacts the Arg side-chain in a different manner. Substitution of Gly with any other amino acid would introduce a severe clash between that residue's side-chain and the carbonyl oxygen of ß3 : Arg216 ; the longer side-chain of Glu (vs. Asp) in the context of RGD, would result in steric clashes with residues on the ligand binding interface of aVß3. The structure of the complex also explains the loss of ligand binding observed in natural or experimental mutations in ß3 integrins. p3 : Aspl 19 to Tyr (J. C. Loftus et al. , Science 249,915-8.

(1990) ) and ß3 : Arg214 to Trp or Gln (M. L. Bajt, M. H. Ginsberg, A. L. Frelinger, M.

C. Berndt, J. C. Loftus, J. Biol. Chem. 267,3789-3794 (1992) ) are naturally-occurring loss of function mutations of ß3 seen in patients with the bleeding disorder thrombasthenia: p3 : Aspl 19 is a MIDAS residue likely involved in indirect metal ion coordination. The p3 : Arg214 side chain lies within 5A from the ligand Asp, and thus a substitution with Trp or Gln will likely change the ligand binding surface. Alanine substitutions of ß3 : Glu220 or Asp217 (or their equivalents in pi and the aA- containing ß2 integrins) also abolish ligand binding (W. Puzon-McLaughlin, Y.

Takada, J. Biol. Chem. 271,20438-20443 (1996); E. C. Tozer, R. C. Liddington, M.

J. Sutcliffe, A. H. Smeeton, J. C. Loftus, J Biol Chem 271,21978-84. (1996); T. G.

Goodman, M. L. Bajt, J Biol Chem 271,23729-36. (1996) ): ß3 : Glu220 directly coordinates the metal ion at MIDAS; Asp217 is part of LIMBS, which helps position p3 : Glu220 for optimal metal ion accommodation in MIDAS.

Binding of the pentapeptide ligand induces tertiary and quaternary changes in aVp3-Mn. Changes in tertiary structure involve PA, affecting primarily its al-a2 loops/helices, a2-C', Fa7 and B-C ("ligand specificity") loops (FIGS. 4A, 4B). The observed movements appear to be causally linked to the top of helix al which approaches MIDAS, permitting contacts with both MIDAS cation and ligand through p3 : Serl21, ß3 : Tyrl22 and ß3 : Serl23. In the complex, the backbone amide and carbonyl oxygens of ß3 : Tyrl22 directly contact the ligand Asp, and both serine side chains coordinate the MIDAS cation. Thus al is fastened to the ligand/MIDAS assembly within the complex. The ADMIDAS cation moves in concert with al since it is primarily coordinated by al residues P3 : Aspl26 and ß3 : Aspl27. Most of the remaining ligand-induced structural changes can be viewed as indirectly caused by the shift of al : al'directly follows al in sequence, and a2 and the top of a7 flank al'. The ligand-specificity region also approaches the ligand. This movement may be related to a salt bridge in this region between p3 : Aspl79 and p3 : Arg214.

ß3 : Arg214 is near the ligand Asp, and it does not form a salt bridge to ß3 : Asp 179 in the unliganded structure. The functional implications of these changes are indicated by the location in the al-a2 segment of ? A ofepitopes both for activation and inhibitory monoclonal antibodies (Y. Takada, W. Puzon, J. Biol. Chem. 268,17597- 17601 (1993); A. Mould, A. Garratt, J. Askari, S. Akiyama, M. Humphries, FEBS Lett. 363,118-22 (1995); A. Andrieux, M. J. Rabiet, A. Chapel, E. Concord, G.

Marguerie, J Biol Chem 266,14202-7. (1991) ; C. Lu, M. Shimaoka, Q. Zang, J.

Takagi, T. A. Springer, Proc Natl Acad Sci U S A 98,2393-8. (2001) ).

Interestingly, the above ligand-induced movements in PA resemble the rearrangements in the al helix and the F-a7 loop in aA that accompany its transition from the"unliganded"low-affinity to the"liganded'high-affinity form (Emsley et al., supra ; Xiong et al., supra ; Lee et al., supra) (FIG. 4). In aA, a major distinguishing feature of this transition is a 10 A downward shift of the c-terminal a7 helix with realignment of its hydrophobic contacts (Xiong et al., supra) (FIG. 4C). However, the position of the a7 helix in liganded PA does not change (it already occupies an equivalent position to liganded aA when ligand is absent). One interpretation of these data is that the conformation of PA in the unliganded aVß3-Mn and aVß3-Ca structures is a ligand-competent (high-affinity) conformation of the A-type domain

when activated in the context of an integrin heterodimer. An alternative interpretation is that this ligand-competent conformation is unique to ßA, and that Glu220 (which is conserved in PA but not aA domains) ensures a low affinity state of PA until ligand becomes available.

Quaternary rearrangements in the integrin head region are also observed in the complex. The interface between PA and the aV propeller undergoes a small change, with the two domains moving slightly closer together at the peptide binding site. The net effect of peptide binding is to"close up"the binding site. In addition, the propeller undergoes a small rotation at the propeller/thigh interface, with PA moving in concert.

It is remarkable that the smallest tripeptide recognition motif in an integrin ligand produces detectable changes both quaternary and tertiary structure (noted above), even within the constrained crystal lattice of the heterodimeric receptor.

Natural integrin ligands are significantly larger, structurally diverse and often multivalent, and are therefore expected to induce conformational changes in addition to those produced here by the pentapeptide ligand. We expect that the diversity of the cellular responses will reflect the spectrum of such tertiary and quaternary changes in the integrins. These rearrangements are not restricted to aA-lacking integrins but are a general feature of all integrins. Although the isolated aA binds ligand when activated, an intact PA is also needed for ligand binding in the aA containing integrin heterodimer, albeit indirectly (W. Puzon-McLaughlin et al., supra ; C. Lu, M.

Shimaoka, supra). Therefore interactions between aA and PA may regulate ligand binding in aA-containing integrins. The location of the RGD-binding site in aVß3 suggests a tantalizing hypothesis for how such regulation might occur: in aA- integrins, aA emanates from the propeller's D3-A3 loop, which forms part of the ligand Arg binding pocket in aA-lacking integrins such as aVß3 (FIG. 2B). The termini of aA, and perhaps part of the domain itself, are expected to lie above the MIDAS motif of PA, and a potential interaction between the two domains might involve residues in the vicinity of the PA MIDAS motif. We note that immediately c- terminal to the a7 helix of aA, there is an invariant Glu existing in a different Ile-Glu- Gly-Thr consensus (FIG. 5). This Glu could act as an endogenous ligand for the PA MIDAS when aA is activated, enabling it to stably bind exogenous ligand. Liganded

aA could also use the Glu"hook"to transmit outside-in signals. The basic tertiary and quaternary changes observed in the present structure may thus be applicable to all integrins.

Computer modeling The methods of the invention employ computer-based methods for identifying compounds having a desired structure. These computer-based methods fall into two broad classes: database methods and de novo design methods. Database methods fall in two main classes, those based on a compound (i. e. , a ligand of a binding site alone) or those based on the three dimensional structure of the binding site. In the former approach, the compound of interest is compared to all compounds present in a database of chemical structures and compounds whose structure is in some way similar to the compound of interest are identified. In the latter approach, all compounds in a database are docked by appropriate computer software into the binding site, and their degree of fit is evaluated and ranked. The structures in the database are based on either experimental data, generated by NMR or x-ray crystallography, or modeled three-dimensional structures based on two-dimensional protein or DNA sequence data.

In de novo design methods, models of compounds whose structure is in some way similar to the compound of interest are generated by a computer program using information derived from known structures (e. g. , data generated by x-ray crystallography and/or theoretical rules). Such design methods can build a compound having a desired structure in either an atom-by-atom manner or by assembling stored small molecular fragments.

The success of both database and de novo methods in identifying compounds with activities similar to the compound of interest depends on the identification of the functionally relevant portion of the compound of interest. For drugs, the functionally relevant portion is referred to as a pharmacophore. A pharmacophore, then, is an arrangement of structural features and functional groups important for biological activity.

Not all identified compounds having the desired pharmacophore will act as an integrin modulator. The actual activity can be finally determined only by measuring

the activity of the compound in relevant biological assays. However, the methods of the invention are extremely valuable because they can be used to greatly reduce the number of compounds which must be tested to identify an actual mimetic.

Programs suitable for generating predicted three-dimensional structures from two-dimensional data include: Concord (Tripos Associated, St. Louis, MO), 3-D Builder (Chemical Design Ltd. , Oxford, U. K. ), Catalyst (Bio-CAD Corp. , Mountain View, CA), Daylight (Abbott Laboratories, Abbott Park, IL).

Programs suitable for searching three-dimensional databases to identify molecules bearing a desired pharmacophore include: MACCS-3D and ISIS/3D (Molecular Design Ltd. , San Leandro, CA), ChemDBS-3D (Chemical Design Ltd., Oxford, U. K. ), Sybyl/3DB Unity (Tripos Associates, St. Louis, MO). Programs suitable for pharmacophore selection and design include: DISCO (Abbott Laboratories, Abbott Park, IL), Catalyst (Bio-CAD Corp. , Mountain View, CA), and ChemDBS-3D (Chemical Design Ltd. , Oxford, U. K.).

Databases of chemical structures are available from Cambridge Crystallographic Data Centre (Cambridge, U. K. ) and Chemical Abstracts Service (Columbus, OH).

De novo design programs include Ludi (Biosym Technologies Inc. , San Diego, CA) and Aladdin (Daylight Chemical Information Systems, Irvine CA), LEGEND (Nishibata, Y. , Itai, A. , Tetrahedron, 47,8985 (1991)) (Molecular Simultations, Burlington, MA), and LeapFrog (available from Tripos associates, St.

Louis, MO).

Upon determination of the three-dimensional structure of an integrin, a potential modulator can be evaluated by any of several methods, alone or in combination. Such evaluation can utilize visual inspection of a three-dimensional representation of the binding site on the integrin, based on the x-ray coordinates of a crystal described herein, on a computer screen. Evaluation, or modeling, can be accomplished through the use of computer modeling techniques, hardware, and software known in the art. This can additionally involve model building, model docking, or other analysis of protein-ligand interactions using software including, for example, QSC, GOLD (Jones et al. , J. Mol. Biol. , 245,43-53, 1995), FlexX (Lengauer, Rarey, 1996), Autodock (Morris et. al. , 1998), GLIDE, Modeler, or Sybyl,

followed by energy minimization and molecular dynamics with standard molecular mechanics forcefields including, for example, CHARMM and AMBER. The three- dimensional structural information of an unliganded integrin, (e. g. , the CD loop of the PTD contacting the strand-F/a7 loop of the PA domain in an unliganded integrin) can also be utilized in conjunction with computer modeling to generate computer models of other unliganded integrins. Computer models of unliganded integrin structures can be created using standard methods and techniques known to those of ordinary skill in the art, including software packages described herein.

Once the three-dimensional structure of a crystal, or solution structure via NMR, comprising an unliganded integrins determined, a potential non-ligand site binder (e. g. , a binder that mimics the ßTD binding of the strand-F/a7 loop of the PA domain) is examined through the use of computer modeling using a docking program such as QSC, GOLD, FlexX, or Autodock to identify potential non-ligand binding site binders to ascertain how well the shape and the chemical structure of the potential ligand will interact with the binding site. Computer programs can also be employed to estimate the attraction, repulsion, and steric hindrance of the two binding partners (i. e. , the non-ligand binding site and the modulating binder). Generally tighter fit, lower steric hindrances, and greater attractive force between the potential ligand and the allosteric binding site are consistent with a tighter binding constant between the two. Furthermore, the more specificity in the design of a potential drug, the more likely that the drug will also not interact with other proteins. This will minimize potential side-effects due to unwanted interactions with other proteins.

A variety of methods are available to one skilled in the art for evaluating and virtually screening molecules or chemical fragments appropriate for associating with a protein, particularly an integrin. Such association can be in a variety of forms including, for example, steric interactions, van der Waals interactions, electrostatic interactions, solvation interactions, charge interactions, covalent bonding interactions, non-covalent bonding interactions (e. g. , hydrogen-bonding interactions), entropically or enthalpically favorable interactions, and the like.

Numerous computer programs are available and suitable for rational drug design and the processes of computer modeling, model building, and computationally identifying, selecting and evaluating potential modulating compounds in the methods

described herein. These include, for example, QSC (WO 01/98457), FlexX, Autodock, Glide, Accelrys'Discovery Studio, or Sybyl. Potential inhibitors can also be computationally designed"de novo"using such software packages as QSC (WO 01/98457), Accelrys'Discovery Studio, Sybyl, ISIS, ChemDraw, or Daylight.

Compound deformation energy and electrostatic repulsion, can be evaluated using programs such as GAUSSIAN 92, AMBER, QUANTA/CHARMM, AND INSIGHT II/DISCOVER.

There are a number of ways to select moieties to fill individual binding pockets. These include QUANTA [Molecular Simulations, Inc. , Burlington, Mass., 1992], SYBYL [Molecular Modeling Software, Tripos Associates, Inc. , St. Louis, Mo. , 1992], AMBER [S. J. Weiner, P. A. Kollman, D. A. Case, U. C. Singh, C. Ghio, G. Alagona, and P. Weiner, J. Am. Chem. Soc. , vol. 106, pp. 765-784 (1984) ], or CHARMM [B. R. Brooks, R. E. Bruccoleri, B. D. Olafson, D. J. States, S Swaminathan, and M. Karplus, J. Comp. Chem. vol. 4, pp. 187-217 (1983) ]. This modeling step may be followed by energy minimization with standard molecular mechanics forcefields such as CHARMM and AMBER. In addition, there are a number of more specialized computer programs to assist in the process of selecting the binding moieties of this invention. These include: 1. ) GRID (Goodford, P. J. A Computational Procedure for Determining Energetically Favorable Binding Sites on Biologically Important Macromolecules. J.

Med. Chem. , 28, pp. 849-857 (1985)). GRID is available from Oxford University, Oxford, UK.

2. ) MCSS (Miranker, A.; Karplus, M. Functionality Maps of Binding Sites: A Multiple Copy Simultaneous Search Method. Proteins: Structure, Function and Genetics, 11, pp. 29-34 (1991) ). MCSS is available from Molecular Simulations, Burlington, Mass.

3. ) AUTODOCK (Goodsell, D. S.; Olsen, A. J. Automated Docking of Substrates to Proteins by Simmulated Annealing. PROTEINS: Structure, Function and Genetics, 8, pp. 195-202 (1990) ). AUTODOCK is available from the Scripps Research Institute, La Jolla, Calif.

4. ) DOCK (Kuntz, 1. D.; Blaney, J. M.; Oatley, S. J.; Langridge, R.; Ferrin, T.

E. A Geometric Approach to Macromolecule-Ligand Interactions. J. Mol. Biol. , 161,

pp. 269-288 (1982) ). DOCK is available from the University of California, San Francisco, Calif.

5.) GOLD (Jones et al. , J. Mol. Biol. , 245,43-53, 1995). GOLD is available from the Cambridge Crystallography Data Centre, Camdridge, UK.

6. ) FlexX (T. Lengauer and M. Rarey, Computational Methods for Biomolecular Docking, Current Opinion in Structural Biology, Vol. 6, pp. 402-406, 1996). FlexX is available through Tripos Associated, St. Louis, MO.

These computer evaluation and modeling techniques can be performed on any suitable hardware including for example, workstations available from Silicon Graphics, Sun Microsystems, and the like. These techniques, methods, hardware and software packages are representative and are not intended to be comprehensive listing.

Other modeling techniques known in the art can also be employed in accordance with this invention. See for example, QSC (WO 01/98457), FlexX, Autodock, Glide, Accelrys'Discovery Studio, or Sybyl and software identified at various internet sites (e. g., netsci. org/Resources/Software/Modeling/CADD/ ch. cam. ac. uk/SGTL/software. html cmm. info. nih. gov/modeling/universal_software. html dasher. wustl. edu/tinker/ zeus. polsl. gliwice. pl/~nikodem//linux4chemistry. html nyu. edu/pages/mathmol/software. html msi. umn. edu/usersupport/software/MolecularModeling. html us. expasy. org/ sisweb. com/software/model. htm).

A potential integrin modulator can be selected by performing rational drug design with the three-dimensional structure (or structures) determined for the ligand binding site as described herein, in conjunction with or solely by computer modeling and methods described above. The potential modulator can be obtained from commercial sources or synthesized from readily available starting materials using standard synthetic techniques and methodologies known in the art. The potential inhibitor can then be assayed to determine its ability to modulate the target (e. g., integrin, e. g., aV3) and/or integrin pathway.

A potential modulator can also be selected by screening a library of compounds (e. g. , a combinatorial library, e. g. , a mass-coded combinatorial library).

The library of compounds can be screened by affinity screening in which members with the greatest affinity to a particular integrin at the new non-ligand binding site can be selected.

Once suitable binding moieties have been selected, they can be assembled into a single modulating binder. This assembly may be accomplished by connecting the various moieties to a central scaffold. The assembly process can, for example, be done by visual inspection followed by manual model building, again using software such as Quanta or Sybyl. A number of other programs may also be used to help select ways to connect the various moieties. These include: CAVEAT (Bartlett, P. A.; Shea, G. T.; Telfer, S. J.; Waterman, S. CAVEAT: A Program to Facilitate the Structure- Derived Design of Biologically Active Molecules. In"Molecular Recognition in Chemical and Biological Problems, "Special Pub. , Royal Chem. Soc. , 78, pp. 182-196 (1989) ) (available from the University of California, Berkeley, CA); 3D Database systems such as MACCS-3D (MDL Information Systems, San Leandro, CA (reviewed by Martin (Martin, Y. C. 3D Database Searching in Drug Design. J. Med.

Chem. , 35, pp. 2145-2154 (1992) )) ; and HOOK (available from Molecular Simulations, Burlington, MA).

A number of techniques commonly used for modeling drugs may be employed (for a review, see: Cohen, N. C.; Blaney, J. M.; Humblet, C.; Gund, P.; Barry, D. C., "Molecular Modeling Software and Methods for Medicinal Chemistry", J. Med.

Chem. , 33, pp. 883-894 (1990) ). There are likewise a number of examples in the chemical literature of techniques that can be applied to specific drug design projects (for a review, see: Navia, M. A. and Murcko, M. A. ,"The Use of Structural Information in Drug Design", Current Opinions in Structural Biology, 2, pp. 202-210 (1992) ). Some examples of these specific applications include: Baldwin, J. J. et al., "Thienothiopyran-2-sulfonamides : Novel Topically Active Carbonic Anhydrase Inhibitors for the Treatment of Glaucoma", J. Med. Chem. , 32, pp. 2510-2513 (1989); Appelt, K. et al. ,"Design of Enzyme Inhibitors Using Iterative Protein Crystallographic Analysis", J. Med. Chem. , 34, pp. 1925-1934 (1991); and Ealick, S.

E. et al. ,"Application of Crystallographic and Modeling Methods in the Design of

Purine Nucleotide Phosphorylase Inhibitors"Proc. Nat. Acad. Sci USA, 88, pp.

11540-11544 (1991).

A variety of conventional techniques can be used to carry out each of the above evaluations as well as the evaluations necessary in screening a candidate compound in modulation (e. g. , inhibition) of an integrin. Generally, these techniques involve determining the location and binding proximity of a given moiety, the occupied space of a bound modulator (e. g. , inhibitor), the deformation energy of binding of a given compound and electrostatic interaction energies. Examples of conventional techniques useful in the above evaluations include: quantum mechanics, molecular mechanics, molecular dynamics, Monte Carlo sampling, systematic searches and distance geometry methods (G. R. Marshall, Ann. Ref. Pharmacol.

Toxicol. , 27, p. 193 (1987) ). Computer software has been developed for use in carrying out these methods. Examples of programs designed for such uses include: Gaussian 92, revision E. 2 (M. J. Frisch, Gaussian, Inc. , Pittsburgh, PA @1993) ; AMBER, version 4.0 (P. A. Kollman, University of California at San Francisco, (C 1993); QUANTA/CHARMM [Molecular Simulations, Inc. , Burlington, Mass. (D 1992]; and Insight II/Discover (Biosysm Technologies Inc. , San Diego, Calif. t1992). These programs can be implemented, for instance, using a Silicon Graphics Indigo 2 workstation or IBM RISC/6000 workstation model 550. Other hardware systems and software packages will be known to those skilled in the art.

Conventional screening In addition to computer-based technologies, the invention includes the use of assays for conventional drug library screening to directly identify compounds capable of modulating the interaction between these two domains, and/or to evaluate whether a compound identified by a computer-based method modulates the activity of the integrin.

These assays can be based on binding and interaction assays, e. g. , assays where one partner is marked (e. g. , by biotin or fluorescent labeling) and the other partner immobilized (e. g. , on 96-well ELISA plates). Compound libraries can be screened for their ability to enhance or block the interaction between the immobilized and the added biotinylated or fluorescent partner.

Binding interaction can be measured by anti-biotin antibodies, or fluorescence spectrometry, e. g. , using an assay analogous to the method described herein for the aVß3-vitronectin binding assay. Many other labeling technologies known in the art can be used in this method (e. g. , radioactive marking, proximity assay).

Alternatively, the interaction between the domains can be generated in a yeast two-hybrid system, using the PA domain (or larger protein fragment containing that domain) as bait and the ßTD domain (or larger protein fragment containing that domain) as prey. Compound libraries can be tested for their ability to perturb the transcription of a suitable marker gene on a Gal4 promoter.

Compound synthesis Synthetic chemistry transformations and protecting group methodologies (protection and deprotection) useful in synthesizing the modulating compounds described herein are known in the art and include, for example, those such as described in R. Larock, Comprehensive Organic Transformations, VCH Publishers (1989); T. W. Greene and P. G. M. Wuts, Protective Groups in Organic Synthesis, 2nd ed. , John Wiley and Sons (1991) ; L. Fieser and M. Fieser, Fieser and Fieser's Reagents for Organic Synthesis, John Wiley and Sons (1994); and L. Paquette, ed., Encyclopedia of Reagents for Organic Synthesis, John Wiley and Sons (1995), and subsequent editions thereof.

The modulating compounds described herein can contain one or more asymmetric centers and thus occur as racemates and racemic mixtures, single enantiomers, individual diastereomers and diastereomeric mixtures. All such isomeric forms of these compounds are expressly included in the present invention.

The modulating compounds described herein can also be represented in multiple tautomeric forms, all of which are included herein. The modulating compounds can also occur in cis-or trans-or E-or Z-double bond isomeric forms. All such isomeric forms of such modulating compounds are expressly included in the present invention.

Conservative Amino Acid Substitution Peptide mimetic compounds can have a different amino acid content as the RGD peptide of SEQ ID No. 1 and serve as a useful mimetic. Substitution mutants

can include amino acid residues that represent either a conservative or non- conservative change (or, where more than one residue is varied, possibly both). A "conservative"substitution is one in which one amino acid residue is replaced with another having a similar side chain. Families of amino acid residues having similar side chains have been defined in the art. These families include amino acids with basic side chains (e. g., lysine, arginine, histidine), acidic side chains (e. g. , aspartic acid, glutamic acid), uncharged polar side chains (e. g. , glycine, asparagine, glutamine, serine, threonine, tyrosine, cysteine), nonpolar side chains (e. g. , alanine, valine, leucine, isoleucine, proline, phenylalanine, methionine, tryptophan), beta-branched side chains (e. g. , threonine, valine, isoleucine) and aromatic side chains (e. g., tyrosine, phenylalanine, tryptophan, histidine). The invention includes polypeptides that include one, two, three, five, or more conservative amino acid substitutions, where the resulting mutant polypeptide binds an extracellular ligand binding site of the integrin au 3.

Fragments or other mutant nucleic acids can be made by mutagenesis techniques well known in the art, including those applied to polynucleotides, cells, or organisms (e. g. , mutations can be introduced randomly along all or part of the nucleic acid encoding the RGD peptide of SEQ ID No. 1 by saturation mutagenesis), and the resultant proteins can be screened for ability to inhibit integrin activation as seen in one or more of the following assays.

Integrin inhibition assays The integrin-modulating ability of compounds identified by the methods of the present invention can be assessed by testing in one or more of the following assays as described in detail below and further described in U. S. Patent No. 6,489, 333: Purified aVß3 (human placenta) -Vitronectin ELISA, aVß3-Vitronectin Binding Assay, Human Aortic Smooth Muscle Cell Migration Assay, In Vivo Angiogenesis Model, Pig Restenosis Model, Mouse Retinopathy Model. The assays are assumed to be made appropriate for the integrin of interest and the following are not limiting and merely serve as examples. A compound identified by the present invention is considered to be active if it has an ICso or Ki value of less than about 10 M for the inhibition of aVß3-Vitronectin Binding Assay, with compounds preferably having Ki

values of less than about 0. 1 µM. Tested compounds of the present invention are active in the aVß3-Vitronectin Binding Assay as well as in cell-based assays of integrin adhesion mediated by the αVß3-receptor. Generally, the assays can be adopted to more appropriately apply to the particular integrin of interest. For example, use of the appropriate ligand (e. g. , RGD-containing, e. g. , fibrinogen, vitronectin, fibronectin, thrombospondin, laminin, collagen, VCAM-1, ICAM-1, ICAM-2, Factor X, osteopontin, bone sialoprotein, or vWF), use of appropriate cell types, and use of appropriate conditions which is knowledge readily available to one skilled in the art. The assays that follow are useful for detecting modulation, e. g., inhibition of aV (33, and can also be directly applied to the testing of other integrins, or may be minimally modified for appropriate use for other integrins.

Purified aVß3 (human placenta) -Vitronectin ELISA The aVß3 receptor can be isolated from human placental extracts prepared using octylglucoside. The extracts can be passed over an affinity column composed of anti-aV (33 monoclonal antibody (LM609) to Affigel. The column can subsequently be washed extensively at pH 7 and pH 4.5 followed by elution at pH 3.

The resulting sample can be concentrated by wheat germ agglutinin chromatography and can be identified by the presence of two bands on SDS gel and confirmed as aVß3 by western blotting. The receptor can also be prepared in a soluble recombinant form using baculovirus expression as described (Mehta et al. , Biochem.

J. 330 (pt. 2): 861-869 (1998) ).

Affinity purified protein can be diluted at different levels and plated to 96 well plates. ELISA can be performed using fixed concentration of biotinylated vitronectin (approximately 80 nM/well). This receptor preparation is confirmed to contain the (xVP3 with no detectable levels of ? 5 by gel (aVß3) and by testing the effects of blocking antibodies for the aVß3 or αVß5 in the ELISA.

A submaximal concentration of biotinylated vitronectin can be selected based on a concentration response curve with a fixed concentration of receptor and variable concentrations of biotinylated vitronectin. aVp3-Vitronectin Binding Assay

Integrin-ligand binding interactions can be measured as detailed previously (Mehta et al. , supra). The purified receptor can be diluted with coating buffer (20 mM Tris HCI, 150 mM NaCl, 1.0 mM CaCl2, 1.0 mM MgC126H20, 10.0 M MnC12. 4H2 O) and coated (100 µL/well) on Costar (3590) high capacity binding plates overnight at 4°C. The coating solution is discarded and the plates washed once with blocking/binding buffer (B/B buffer, 50 mM Tris HCI, 100 mM NaCI, 1.0 mM Cal2, 1. 0 mM MgCl2 6H20,10. 0 I1M MnCl2. 4H2 O). Receptor is then blocked (200 µL/well) with 3.5% BSA in B/B buffer for 2 hours at room temperature. After washing once with 1.0% BSA in B/B buffer, biotinylated vitronectin (100 1L) and either inhibitor (11 uL) or B/B buffer w/1.0% BSA (11 pL) is added to each well.

The plates are incubated 2 hours at room temperature. The plates are washed twice with B/B buffer and incubated 1 hour at room temperature with anti-biotin alkaline phosphatase (100 uL/well) in B/B buffer containing 1.0% BSA. The plates are washed twice with B/B buffer and alkaline phosphatase substrate (100 µL) is added.

Color is developed at room temperature. Color development is stopped by addition of 2N NaOH (25 pL/well) and absorbance is read at 405 nm. The ICso is the concentration of test substance needed to block 50% of the vitronectin binding to the receptor. A compound is considered to be active if it has an IC50 value of less than or equal to about 10 M in the aVp3-Vitronectin Binding Assay. Compounds with an ICso less than 100 nM for the inhibition of vitronectin are generally desirable.

Integrin Cell-Based Adhesion Assays In the adhesion assays, a 96 well plate are coated with the appropriate ligand (e. g. , fibrinogen, vitronectin, fibronectin, thrombospondin, laminin, collagen, VCAM- 1, ICAM-1, ICAM-2, Factor X, osteopontin, bone sialoprotein, or vWF) for the integrin to be tested and incubated overnight at 4°C. The following day, the cells are harvested, washed, and loaded with a fluorescent dye. Compounds and cells are added together and then are immediately added to the coated plate. After incubation, loose cells are removed from the plate, and the plate (with adherent cells) is counted on a fluorometer. The ability of test compounds to inhibit cell adhesion by 50% is given by the IC50 value and represents a measure of potency of inhibition of integrin

mediated binding. Compounds are tested for their ability to block cell adhesion using integrin interaction assays specific for the integrin of interest.

Platelet Aggregation Assay Venous blood is obtained from anesthetized mongrel dogs or from healthy human donors who are drug-and aspirin-free for at least two weeks prior to blood collection. Blood is collected into citrated Vacutainer tubes. The blood is centrifuged for 15 minutes at 150 x g (850 RPM in a Sorvall RT6000 Tabletop Centrifuge with H- 1000 B rotor) at room temperature, and platelet-rich plasma (PRP) is removed. The remaining blood is centrifuged for 15 minutes at 1500 x g (26,780 RPM) at room temperature, and platelet-poor plasma (PPP) is removed. Samples are assayed on a PAP-4 Platelet Aggregation Profiler, using PPP as the blank (100% transmittance).

200 ßL of PRP (5 x 108 platelets/mL) are added to each micro test tube, and transmittance is set to zero percent. 20 L of ADP (10 M) is added to each tube, and the aggregation profiles are plotted (percent transmittance versus time). Test agent (20 L) is added at different concentrations prior to the addition of the platelet agonist. Results are expressed as percent inhibition of agonist-induced platelet aggregation.

Human Aortic Smooth Muscle Cell Migration Assay A method for assessing aV (33-mediated smooth muscle cell migration and agents which inhibit aV (33-mediated smooth muscle cell migration is described in Liaw et al. , J. Clin. Invest. (1995) 95: 713-724).

In Vivo Angiogenesis Model A quantitative method for assessing angiogenesis and antiangiogenic agents is described in Passaniti et al., Laboratory Investigation (1992) 67: 519-528.

Pig Restenosis Model A method for assessing restenosis and agents which inhibit restenosis is described in Schwartz et al. , J. Am. College of Cardiology (1992) 19: 267-274.

Mouse Retinopathy Model A method for assessing retinopathy and agents which inhibit retinopathy is described in Smith et al., Invest. Ophthal. & Visual Science (1994) 35 : 101-111.

A number of embodiments of the invention have been described.

Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, other embodiments are within the scope of the following claims.