Multiple Discontinuous Ligand-mimetic Antibody Binding Sites Define a Ligand Binding Pocket in Integrin a IIb b 3 *

Integrin a IIb b 3 , a platelet fibrinogen receptor, is crit- ically involved in thrombosis and hemostasis. However, how ligands interact with a IIb b 3 has been controversial. Ligand-mimetic anti- a IIb b 3 antibodies (PAC-1, LJ-CP3, and OP-G2) contain the RGD-like RYD sequence in their CDR3 in the heavy chain and have structural and functional similarities to native ligands. We have located binding sites for ligand-mimetic antibodies in a IIb and b 3 using human-to-mouse chimeras, which we expect to maintain functional integrity of a IIb b 3 . Here we report that these antibodies recognize several discontinuous binding sites in both the a IIb and b 3 subunits; these binding sites are located in residues 156–162 and 229– 230 of a IIb and residues 179–183 of b 3 . In contrast, sev- eral nonligand-mimetic antibodies ( e.g. 7E3) recognize single epitopes in either subunit. Thus, binding to several discontinuous sites in both subunits is unique to ligand-mimetic antibodies. Interestingly, these binding sites overlap with several (but not all) of the sequences that have been reported to be critical for fibrinogen binding ( e.g. N-terminal repeats 2–3 but not repeats 4–7, of a IIb ). These results suggest that ligand-mimetic anti- bodies and probably native ligands may make direct contact with these discontinuous binding sites in both subunits, which may constitute a ligand-binding pocket. Integrin a IIb b 3 is critically IIb and b 3 cDNAs were obtained from J. C. Loftus Scripps The mouse a IIb cDNA clone from the mouse EST data base was obtained from American Type Culture Col-lection (clone 1498358) and partially sequenced by ABI automatic se- quencer at the Scripps protein and nucleic acid core facility (GenBank™ accession number AF166384). Wild-type human a IIb and b 3 cDNAs were subcloned into pBJ-1 vector. Site-directed mutagenesis was car-ried out as described (38). The presence of the mutation was verified by DNA sequencing. Wild-type and mutant a IIb cDNA constructs in pBJ-1 vector were transfected by electroporation into CHO (Chinese hamster ovary) cells (1 3 10 7 cells) homogeneously expressing human b 3 ( b 3 CHO) or into parent untransfected CHO cells together with wild-type human b 3 cDNA in pBJ-1, as described previously (5). We obtained essentially the same results either way. Wild-type and mutant b 3 cDNAs were transfected into CHO-K1 cells together with wild-type a IIb cDNA. Cells were harvested 48 h after transfection and used for antibody and/or fluorescein isothiocyanate (FITC)-labeled fibrinogen and PAC-1 binding assays. Mouse a IIb cDNA encoding the N-terminal 443 residues was fused to the human a IIb cDNA using an Nhe I site after the Nhe I site was introduced into the human a IIb cDNA at the corresponding position. FITC-labeled mAb PAC-1 CHO Cloned cells stably expressing wild type or mutant a IIb b 3-1-3 were used. Reactivity to mAbs and binding of FITC-labeled soluble fibrinogen were assayed by flow cytometry. Data are expressed as the ratio of mean fluorescent intensity. The data suggest that the b 3-1-3 mutation blocks binding of ligand-mimetic mAbs and fibrinogen to a IIb b 3 .

Integrin ␣ IIb ␤ 3 , a platelet fibrinogen receptor, is critically involved in thrombosis and hemostasis. However, how ligands interact with ␣ IIb ␤ 3 has been controversial. Ligand-mimetic anti-␣ IIb ␤ 3 antibodies (PAC-1, LJ-CP3, and OP-G2) contain the RGD-like RYD sequence in their CDR3 in the heavy chain and have structural and functional similarities to native ligands. We have located binding sites for ligand-mimetic antibodies in ␣ IIb and ␤ 3 using human-to-mouse chimeras, which we expect to maintain functional integrity of ␣ IIb ␤ 3 . Here we report that these antibodies recognize several discontinuous binding sites in both the ␣ IIb and ␤ 3 subunits; these binding sites are located in residues 156 -162 and 229 -230 of ␣ IIb and residues 179 -183 of ␤ 3 . In contrast, several nonligand-mimetic antibodies (e.g. 7E3) recognize single epitopes in either subunit. Thus, binding to several discontinuous sites in both subunits is unique to ligand-mimetic antibodies. Interestingly, these binding sites overlap with several (but not all) of the sequences that have been reported to be critical for fibrinogen binding (e.g. N-terminal repeats 2-3 but not repeats 4 -7, of ␣ IIb ). These results suggest that ligand-mimetic antibodies and probably native ligands may make direct contact with these discontinuous binding sites in both subunits, which may constitute a ligand-binding pocket.
Integrin ␣ IIb ␤ 3 is a platelet fibrinogen receptor that is critically involved in platelet aggregation (1). Thus ␣ IIb ␤ 3 -fibrinogen interaction is a therapeutic target for thrombosis and hemostasis. However, how ligands interact with the integrin ␣ IIb and ␤ 3 subunits has been the subject of much discussion.
The ␣ IIb subunit has seven repeated sequences of 60 -70 residues each in its N-terminal portion. Two different regions of the ␣ IIb subunit have been implicated in ligand binding. The second metal binding site of ␣ IIb (residues 294 -314 in N-terminal repeat 5 of ␣ IIb ) has been identified as a ligand binding site by chemically cross-linking the ␥-peptide (HHLGGAKQ-AGDV 400 -411 ) of the fibrinogen ␥ chain C-terminal domain (2). Both the peptide derived from this ␣ IIb sequence and its antibodies have been shown to block fibrinogen binding to ␣ IIb ␤ 3 (3). Consistently, recombinant bacterial proteins that consist of repeats 4 -7 of ␣ IIb (residues 171-464) have been shown to bind to ligands in a cation-dependent manner (4). On the other hand, alanine-scanning mutagenesis (5) 1 suggests that the predicted loops in repeats 2 and 3 are critical for ligand binding. Also, function-blocking anti-␣ IIb ␤ 3 monoclonal antibodies (mAbs) 2 are mapped in repeats 2 and 3 (5). 1 It has not been established which regions of ␣ IIb actually interact with ligands.
The presence of an I-domain-like structure within the ␤ subunit has been suggested based on the similarity in hydropathy profiles between the I-domain of the ␣ M subunit and part of the ␤ subunit (6). The N-terminal region of the ␤ 3 subunit has components that are critical for ligand binding and its regulation (reviewed in Ref. 7). The RGD-containing peptide chemically cross-links to the N-terminal region (residues 109 -171) of the ␤ 3 subunit (8). Several different ␤ 3 sequences, MDLSYSMKDDLWSI (residues 118 -131) (9), DDLW (residues 126 -129 of ␤ 3 ) (10), DAPEGGFDAIMQATV (residues 217-231 of ␤ 3 ) (11,12), and VSRNRDAPEG (residues 211-221 of ␤ 3 ) (13,14) have been implicated in ligand interaction. The disulfidelinked CYDMKTTC sequence (residues 177-184 in ␤ 3 ) in a large predicted loop is critical for the ligand specificity of ␣ v ␤ 3 (15). There are several residues that are critical for ligand binding in the putative I-domain-like structure of ␤ subunits (16 -22). These critical oxygenated residues are located in several separate predicted loop structures within the I-domainlike structure of the ␤ subunit (15). It has not been established how these predicted loops, which are critical for ligand binding and specificity, are organized. Two distinct models of the putative I-domain-like structure of ␤ subunits have been published based on the structure of the ␤ M I-domain (21,23). It has also been proposed that there is no I-domain-like structure in the ␤ 3 subunit (24).
Three anti-human ␣ IIb ␤ 3 -specific mAbs, PAC-1 (25), OP-G2 (26), and LJ-CP3 (27), have the tripeptide RYD sequence that mimics the RGD sequence in the CDR3 region of the heavy chain. These mAbs inhibit both fibrinogen binding to platelets and fibrinogen-dependent aggregation of platelets. Binding of these mAbs is cation-dependent and is completely blocked by RGD-containing peptides. PAC-1 does not bind to nonactivated platelets (25). OP-G2 and LJ-CP3 can bind to nonactivated ␣ IIb ␤ 3 , but this binding increases upon activation. The ligandmimetic properties of these mAbs suggest that they have structural and functional similarities to ligands (e.g. fibrinogen). Structure-function studies of these mAbs indicate that the RYD sequence in the CDR3 in their heavy chain occupies the same space as RGD does in conformationally constrained, bioactive ␣ IIb ␤ 3 ligands (28).
In the present study, to clarify the controversies surrounding * This work was supported by National Institute of Health Grants GM47157 and GM49899 (to Y. T.) and by Department of the Army Grant DAMD17-97-1-7105 (to T. K.). This is Publication #12252-VB from The Scripps Research Institute. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM  ␣ IIb ␤ 3 -ligand interaction, we studied how ligand-mimetic mAbs as model ligands recognize ␣ IIb and ␤ 3 . We found that these mAbs uniquely recognize several discontinuous binding sites in both ␣ IIb and ␤ 3 . These binding sites overlap with several (but not all) of the sequences that have previously been reported to be critical for fibrinogen binding (e.g. N-terminal repeats 2-3, but not repeats 4 -7, of ␣ IIb ). These results strongly suggest that native ligands make direct contact with these discontinuous binding sites, which may constitute a ligand binding pocket at the ␣/␤ boundary.

Monoclonal Antibodies and cDNAs
The specificities of the mAbs used in this study are shown in Table I

Methods
Construction and Transfection of cDNAs for Human ␣ IIb and ␤ 3 Mutants-Human ␣ IIb and ␤ 3 cDNAs were obtained from J. C. Loftus (The Scripps Research Institute). The mouse ␣ IIb cDNA clone from the mouse EST data base was obtained from American Type Culture Collection (clone 1498358) and partially sequenced by ABI automatic sequencer at the Scripps protein and nucleic acid core facility (GenBank™ accession number AF166384). Wild-type human ␣ IIb and ␤ 3 cDNAs were subcloned into pBJ-1 vector. Site-directed mutagenesis was carried out as described (38). The presence of the mutation was verified by DNA sequencing. Wild-type and mutant ␣ IIb cDNA constructs in pBJ-1 vector were transfected by electroporation into CHO (Chinese hamster ovary) cells (1 ϫ 10 7 cells) homogeneously expressing human ␤ 3 (␤ 3 -CHO) or into parent untransfected CHO cells together with wild-type human ␤ 3 cDNA in pBJ-1, as described previously (5). We obtained essentially the same results either way. Wild-type and mutant ␤ 3 cDNAs were transfected into CHO-K1 cells together with wild-type ␣ IIb cDNA. Cells were harvested 48 h after transfection and used for antibody and/or fluorescein isothiocyanate (FITC)-labeled fibrinogen and PAC-1 binding assays. Mouse ␣ IIb cDNA encoding the N-terminal 443 residues was fused to the human ␣ IIb cDNA using an NheI site after the NheI site was introduced into the human ␣ IIb cDNA at the corresponding position.
Binding of FITC-labeled Fibrinogen and mAb PAC-1 to CHO Cells-Fibrinogen (Enzyme Research Laboratories, South Bend, IN) was labeled with FITC as described previously (39,40). mAb PAC-1 was labeled with FITC essentially as described (41). Fibrinogen binding to cells transiently expressing ␣ IIb ␤ 3 was determined as described previously (42) with some modifications. Briefly, cells were first incubated with PL98DF6 followed by phycoerythrin (PE)-conjugated anti-mouse IgG (BIOSOURCE, Camarillo, CA). After washing, cells were incubated with FITC-labeled fibrinogen or PAC-1 in the presence of control mouse IgG or PT25-2 in 5 mM Hepes/Tyrode, 2 mM Ca 2ϩ , and 2 mM Mg 2ϩ buffer, pH 7.4. Binding of FITC-labeled fibrinogen or PAC-1 to PElabeled cells expressing high level ␣ IIb ␤ 3 was examined in FACScan.
Flow Cytometry-Flow cytometric analysis was performed as described (43). To normalize the data for ␣ IIb ␤ 3 expression, % expression of each binding site in mutant ␣ IIb was first normalized using % ␣ IIb expression with mAb PL98DF6 for ␣ IIb ␤ 3 complex-specific mAbs. For the anti-␤ 3 mAbs 7E3 and 16N7C2, % expression of each binding site in mutant ␣ IIb was first normalized using % ␤ 3 expression with mAb 15 (normalized % expression). Data is shown as the normalized % of expression in mutants relative to wild-type. Binding sites for mAbs PL98DF6 and 15 were not found in the ␣ IIb and ␤ 3 regions tested in this study (data not shown).

Ligand-mimetic mAbs Recognize Several Discontinuous N-
terminal Regions of the ␣ IIb Subunit-We used human-tomouse ␣ IIb and ␤ 3 mutations to localize binding sites for three ligand-mimetic mAbs and several function-blocking or -activating mAbs ( Table I). The rationale behind this strategy is that all of the mAbs used in this study recognize human ␣ IIb ␤ 3 but not mouse ␣ IIb ␤ 3 . We also expect this strategy to enable for us to maintain the ligand binding function of the mutants while eliminating the potential problems associated with mutation of individual residues to Ala, which could indirectly induce conformational changes.
Since the mouse ␣ IIb sequence has not been published, we first sequenced the N-terminal region of putative mouse ␣ IIb cDNA clones from the mouse EST cDNA data base (Fig. 1a). The N-terminal seven-repeat region of human ␣ IIb is 82.5% identical to that of mouse ␣ IIb (they have 452 and 451 amino acid residues, respectively). To determine whether binding sites for ligand-mimetic and nonligand-mimetic mAbs are located in the seven-repeat region of ␣ IIb , we first replaced Nterminal 443 amino acid residues of human ␣ IIb with the corresponding sequence of mouse ␣ IIb . The human/mouse ␣ IIb chimera was transiently transfected into ␤ 3 -CHO cells that homogeneously express wild-type human ␤ 3 . Reactivity of the transfected cells to a panel of mAbs is shown in Table II. The results indicate that the anti-␣ IIb ␤ 3 mAbs tested recognize ␣ IIb ␤ 3 but not ␣ v ␤ 3 . All of the ␣ IIb ␤ 3 mAbs tested, except for AP-2 and D57, require the N-terminal 443-amino acid residues of human ␣ IIb for ␣ IIb ␤ 3 recognition. AP-2 and D57 require the ␣ IIb subunit but do not require the human N-terminal 443 residues. The anti-␣ IIb mAb PL98DF6 does not require the human N-terminal 443 residues.
We introduced human-to-mouse mutations, alone or ingroups, into the N-terminal seven-repeat region of human ␣ IIb to identify ␣ IIb sequences that are critical for antibody binding (Fig. 1a). The human-to-mouse ␣ IIb mutants were individually transfected into ␤ 3 -CHO cells. The capacity of transiently expressed ␣ IIb ␤ 3 mutants to bind to a panel of mAbs (listed in Table I) was tested using flow cytometry. Fig. 2 shows flow cytometric profiles of wild-type ␣ IIb ␤ 3 and the A16 mutant (V156A/N158S/D159 deletion/S161R/W162G) as an example. 30 -50% of cells transfected with wild-type ␣ IIb ␤ 3 are positive with LJ-CP3, OP-G2, and PT25-2. 30 -50% of cells expressing the A16 mutant are positive with PT25-2, but almost none of them are positive with LJ-CP3 and OP-G2. This indicates that the mutant is expressed on the cell surface but does not react with LJ-CP3 and OP-G2.
The results for 36 mutants are summarized in Table III (data is shown only for selected mutants). Because expression levels vary, we normalized the percentage of positive cells for each mutant. We used PL98DF6 (an anti-␣ IIb mAb) to normalize the ␣ IIb ␤ 3 expression and mAb 15 (an anti-␤ 3 mAb) to normalize the ␤ 3 expression. Data are presented as the ratio of the mutant cells that are positive to mAbs to the wild-type cells that are positive to mAbs. We found that LJ-CP3 binding is completely blocked by the A16 mutation and almost completely blocked by the A23 mutation (E229V/Y230F). OP-G2 binding is almost completely blocked by the A16 mutation. Binding of nonligand-mimetic function-blocking mAbs LJ-P9 and LJ-CP8 is completely blocked by the A9 mutation (V79L/S81F/T83I/L84P) and by the A16 mutation, respectively. Binding of activating mAb PT25-2 was completely blocked by the A29 (R335K/H338Q) mutation. 32 other mutations (listed in Fig. 1a and its legend) did not significantly affect the binding of the mAbs tested. The A16 mutation blocks binding of LJ-CP3, OP-G2, and LJ-CP8, but the nearby A15 and A17 mutations do not affect binding of these mAbs. Also, the A9, A16, A23, and A29 mutations specifically block binding of one or three mAbs, but not others, indicating that the human-to-mouse mutations do not induce global conformational changes, and their effects remain local.
Ligand-mimetic mAbs Recognize Discontinuous N-terminal Regions of the ␤ 3 Subunit-There are 17-amino acid-residue differences between human and mouse ␤ 3 in the putative I-domain-like region spanning residues 90 -328 (Fig. 1b). We introduced human-to-mouse mutations into the putative I-domainlike region of human ␤ 3 alone or in-groups. The human-tomouse ␤ 3 mutants (total 10) were transiently expressed together with wild-type ␣ IIb cDNA on CHO cells. Since CHO cells express endogenous ␣ v , but do not express ␤ 3 , the human ␤ 3 subunit is expressed as human ␣ IIb ␤ 3 and hamster ␣ v /human ␤ 3 . The ability of ␣ IIb ␤ 3 or ␣ v ␤ 3 to bind to a panel of mAbs (Table I) was tested in flow cytometry. We found that several human-to-mouse mutations affect binding of function-blocking mAbs (Table IV). Binding of mAb 16N7C2 (anti-␤ 3 , RGDϩ) is completely blocked by the B3 (W129S/N133T) mutation. Binding of 7E3 and A2A9 is completely blocked by the B6 (D179N/ T182N/T183A) mutation and partially blocked by the B3 mutation. Binding of AP-2 and LM609 is completely blocked by the B5 (E171Q/L173I/E174K) mutation. In contrast, these mutations only partially blocked the binding of ligand-mimetic mAbs. Binding of OP-G2 and LJ-CP3 is partially blocked by the B6 mutation.
To determine whether the B6 mutation really affects OP-G2 or LJ-CP3 binding, we studied the effects of combined mutations on OP-G2 and LJ-CP3 binding. The combined B3/B6 mutation completely blocked LJ-CP3 binding and almost completely blocked OP-G2 binding. However, the combined B3/B5 or B6/B9 mutation had only a modest effect on OP-G2 and LJ-CP3 binding. These results suggest that the B3 and B6 binding sites are involved in OP-G2 and LJ-CP3 binding, but the nearby B5 and B9 sites (controls) are probably much less involved (Table V). However, binding of several nonligandmimetic mAbs (e.g. 16N7C2) was not affected by the combined mutations tested. We did not find any ␤ 3 mutations that affect the binding of mAbs 2G12, LJ-CP8, and LJ-P9.
Effect of Human-to-mouse Mutation on Activation-dependent PAC-1 and Fibrinogen Binding-We studied the effect of the several mutations that affect binding of ligand-mimetic mAbs on the activation-dependent binding of PAC-1 (Fig. 3). Binding of FITC-labeled PAC-1 was determined as the difference in FITC fluorescence signal (FL-1) in the presence and absence of the activating anti-␣ IIb ␤ 3 mAb PT25-2 in the PE-positive (␣ IIb positive, FL-2) cell population (Fig. 3a). PT25-2 activates ␣ IIb ␤ 3 but does not activate ␣ v ␤ 3 (PT25-2 is specific to ␣ IIb ␤ 3 ). PAC-1 binding was completely blocked by the A16 mutation and was reduced by the A23 mutation. PAC-1 binding was partly blocked by the B6 mutation but not by the B3, B5, or control B7 mutation. The combined B3/B6 mutations did not further reduce PAC-1 binding. These results suggest that activation-dependent PAC-1 is similar to LJ-CP3 and OP-G2 in that site A16 is critical for binding, and several other sites (sites A23 and B6) are involved in binding.
Binding of soluble fibrinogen to these mutants was studied to determine whether these mutants still bind to ligands. Binding of FITC-labeled fibrinogen was determined as the difference in FITC fluorescence signal in the presence and absence of the activating mAb PT25-2 in the PE-positive (␣ IIb positive) cell population (Fig. 3b). Fibrinogen bound to the A16 and B6 mutants at a level comparable to or higher than that of wildtype and the control B7 mutant. Binding of soluble fibrinogen to the A23 mutant was lower than that of wild-type or the control mutants, but still detectable. These results suggest that the A16, A23, B3, and B6 mutants still maintain the integrity and ligand binding function of ␣ IIb ␤ 3 . Although the fibrinogen binding function appears to vary from mutant to mutant, this may reflect the reported potential species dif- ference in ␣ IIb ␤ 3 function (44). (Residues 177-184) of ␤ 3 in Binding to Ligand-mimetic mAbs and Fibrinogen-To determine whether the predicted loop of ␤ 3 (Residues 177-184) that contains site B6 is involved in the binding of ligands and ligandmimetic mAbs to ␣ IIb ␤ 3 , we swapped this predicted loop of ␤ 3 with the corresponding sequence of ␤ 1 . The resulting ␤ 3-1-3 mutant cDNA was transfected into CHO cells together with wild-type ␣ IIb cDNA and a neomycin-resistant gene. Cells stably expressing ␣ IIb ␤ 3-1-3 were cloned by sorting. Binding of mAbs and fibrinogen was studied by flow cytometry (Table VI). Expression of ␣ IIb was normalized using the anti-␣ IIb mAb PL98DF6. The results suggest that the ␤ 3-1-3 mutation significantly blocks binding of ␣ IIb ␤ 3 to OP-G2, LJ-CP3, and fibrinogen. The mutation did not affect binding to control mAb 15 (anti-␤ 3 ). This suggests that the predicted loop may be critically involved in the binding of ␣ IIb ␤ 3 to ligand-mimetic mAbs and fibrinogen.

What Do the Discontinuous Binding Sites in ␣ IIb for Ligandmimetic mAbs Tell Us about Structure and Function of ␣ IIb ?-
The present study provides evidence that two discontinuous sites in ␣ IIb (A16 and A23) are critical for binding to ligandmimetic mAbs. Several regions/residues that are critical for ligand binding in ␣ IIb have been identified by site-directed mutagenesis or by genetic analysis of natural function-defective ␣ IIb ␤ 3 mutants (see the Introduction), but whether these regions/residues directly interact with ligands has been unclear. Interestingly, the discontinuous ligand-mimetic mAb binding sites overlap with or are close to several conserved critical residues for ligand binding that have been identified by alanine-scanning mutagenesis (Fig. 4a). It should be noted that these conserved critical residues are not changed in human-tomouse mutations. Site A16 is localized in the predicted loop at the boundary between repeats 2 and 3 of ␣ IIb . We have found that mutating conserved residues Arg-147, Tyr-155, Phe-160, Asp-163, or Arg-165 to Ala blocks binding of fibrinogen and/or ligand-mimetic mAbs (Fig. 4a). 1 It has been reported that a patient with Glanzmann's thrombasthenia, who has functiondefective ␣ IIb , has a two-amino acid insertion (Arg-Thr) between residues 160 and 161 in this region of the ␣ IIb subunit (45). Mutating Asp-163 in this predicted loop of ␣ IIb to Ala has been shown to block binding of fibrinogen and ligand-mimetic mAbs (45). Site A23 is recognized by the ligand-mimetic mAb LJ-CP3, and is located in the predicted loop at the boundary between repeats 3 and 4 of the ␣ IIb subunit. The nearby D224V mutation of ␣ IIb has recently been shown to block PAC-1 and OP-G2 binding to ␣ IIb ␤ 3 (46). However, the function of this region has not been identified. We have found that mutating conserved Ser-222, Asp-224, Phe-231, or Asp-232 to Ala blocks binding to fibrinogen and ligand-mimetic mAbs (Fig. 4a). 1 The fact that sites A16 and A23 overlap with the ␣ IIb regions that are critical for ligand binding indicates that these residues/ regions make direct contact with native ligands. These results are consistent with the proposed ␤-propeller model of the ␣ subunit (47) in that sites A16 and A23, which are believed to make direct contact with ligands, are spatially close to each other, even though they are separate in the primary structure. In this model, site A9 (an epitope for mAb LJ-P9) is spatially close to sites A16 and A23 (Fig. 5). This is consistent with the function-blocking activity of mAb LJ-P9.
It is possible that site A16 is critically involved in RYD recognition and binding, since site A16 is recognized by all of the RYD-containing ligand-mimetic mAbs, and the A16 mutation completely blocks binding of these mAbs (the effect of the A23 mutation and the human-to-mouse mutations in ␤ 3 on the  2. Binding of mAbs to wild-type and mutant ␣ IIb ␤ 3 that are transiently expressed on CHO cells. Wild-type or mutant ␣ IIb cDNA in pBJ-1 vector was transfected into CHO cells together with wild-type human ␤ 3 cDNA in pBJ-1 vector. After 48 h, cells were harvested and stained first with mAbs LJ-CP3, OP-G2, PT25-2, or control mouse IgG (mIgG) and then with FITC-labeled goat anti-mouse IgG. Wt ␣ IIb ␤ 3 is recognized by all of the anti-␣ IIb ␤ 3 mAbs tested. The A16 mutant is recognized by PT25-2 but not by LJ-CP3 or OP-G2.
binding of ligand-mimetic antibodies is relatively weak and variable compared with that of the A16 mutation and may depend on antibody species.) It is interesting to note that the A16 mutant shows higher fibrinogen binding than wild-type (Fig. 3), although the A16 mutant does not bind to ligandmimetic antibodies. It is well known that RGD peptide does not effectively block fibrinogen binding to rat and rabbit (44) and mouse ␣ IIb ␤ 3 . It is possible that ␣ IIb ␤ 3 from these species may have higher affinity to fibrinogen than human ␣ IIb ␤ 3 due to species difference in site A16. Further biochemical studies of site A16 will be required to address this hypothesis.
The present results indicate that ligand-mimetic mAbs do not recognize the region close to the previously reported putative ligand binding (␥-chain peptide cross-linking) site in repeat 5 of ␣ IIb (2). This suggests that this region is not a major ligand binding site. This idea is consistent with the proposed ␤-propeller model, since the fibrinogen ␥-peptide cross-linking site in repeat 5 of ␣ IIb is located in the lower face of the domain, a predicted nonligand binding site. The regions critical for ligand binding in repeats 2 and 3 (containing sites A16 and A23) are located in the upper face of the model, a predicted ligand binding site (47). It is intriguing that the activating anti-␣ IIb ␤ 3 mAb PT25-2 recognizes site A29 in ␣ IIb , which is close to the ␥-chain peptide cross-linking site. It is possible that the ␥-chain peptide cross-linking site might be an allosteric binding site, which is consistent with the location of this site in the predicted nonligand binding site of the ␤-propeller model.

What Do the Discontinuous Binding Sites in ␤ 3 for Ligandmimetic mAbs Tell Us about Structure and Function of ␤ 3 ?-
Although no single human-to-mouse mutation in the ␤ 3 subunit strongly blocks binding of ligand-mimetic mAbs, we provided evidence that the combined B3 and B6 mutation strongly blocks binding of LJ-CP3 and OP-G2 to ␣ IIb ␤ 3 . The fact that more than two mutations are required to block binding of mAbs is not surprising if we assume that these mAbs interact with multiple sites in both subunits for binding. The effect of the combined B3 and B6 mutation is significant, since the nearby control B5 or B9 mutation does not increase the effect of the B3 or B6 mutation. These results indicate that sites B3 and B6 are involved in the binding of these ligand-mimetic mAbs, although the B3 mutation alone has no effect. Site B6 is located in the predicted loop (residues 177-184 of ␤ 3 ), which has been reported to be critical for ligand binding and specificity (15). Swapping this disulfide-linked predicted loop of ␤1 with the corresponding ␤ 3 sequence changes the ligand specificity of Human-to-mouse ␤ 3 mutants with combined ␤ 3 mutations were transiently expressed in CHO cells together with wild-type human ␣ IIb . Analysis of the reactivity of a panel of mAbs to mutant ␣ IIb ␤ 3 in flow cytometry and normalization of the results were performed as described in the legend to Table III

binding to mAbs
The human-to-mouse ␣ IIb mutants were transiently expressed in CHO cells together with wild-type human ␤ 3 . After 48 h, cells were tested for their ability to bind to a panel of anti-␣ IIb ␤ 3 mAbs in flow cytometry. Data are expressed as the ratio % positive cells with the mutant ␣ IIb ␤ 3 to % positive cells with wild type. The data were first normalized for expression, then binding of mAbs relative to wild type was calculated. The % expression of PL98DF6 (a nonfunctional anti-␣ IIb mAb) was used to normalize the expression of ␣ IIb ␤ 3 (for LJ-CP3, OP-G2, LJ-CP8, 2G12, PT25-2, LJ-P9, AP2, and A2A9). The % expression of mAb 15 (a nonfunctional anti-␤ 3 mAb) was used to normalize the expression of ␤ 3 (for D57, 7E3, and 16N7C2). Of the human-to-mouse ␣ IIb mutants tested in this study, only A9, A16, A23, and A29 had a noticeable effect on binding of mAbs tested.

binding to mAbs
The human-to-mouse ␤ 3 mutants were transiently expressed in CHO cells together with wild-type human ␣ IIb . Data are expressed as the ratio % positive cells with the mutant ␣ IIb ␤ 3 to % positive cells with wild type. Analysis of the reactivity of a panel of mAbs to mutant ␣ IIb ␤ 3 in flow cytometry and normalization of the results were performed as described in the legend to Table III (15). Also, fibrinogen C-terminal domains require this predicted loop sequence of ␤ 3 for binding to ␣ v ␤ 3 (48). In the present study, we have shown that the ␤ 3-1-3 mutant blocks binding of ␣ IIb ␤ 3 to OP-G2, LJ-CP3, and fibrinogen (Table VI). These results are consistent with the observation that site B6 is involved in the binding of ligand-mimetic mAbs. We found that two function-blocking mAbs, 7E3 and A2A9, also recognize site B6, consistent with their function-blocking activity. The B3 mutation completely blocks the binding of ␣ IIb ␤ 3 to the RGD-containing anti-␤ 3 mAb 16N7C2. It is highly likely that the RGD motif of 16N7C2 is involved in its binding to ␤ 3 , since this interaction is blocked by echistatin, an RGD-containing disintegrin (35). Intriguingly, site B3 overlaps with the DDLW sequence (residues 126 -129), a putative RGD binding site (Fig. 4). The CWDDGWLC peptide is a functional mimic of the ligand binding sites of RGD-directed integrins (10), and the structurally similar DDLW sequence in the integrin ␤ subunit has been proposed to be a putative RGD binding site. Also, this binding site is part of the MDLSYSMKDDLWSI sequence (residues 118 -131), to which the RGD-containing peptide has been reported to make a ternary complex with cations (9). This is consistent with the reports that RGD-containing peptides block binding of these ligand-mimetic mAbs to ␣ IIb ␤ 3 (25)(26)(27). Nearby Asp-119, Ser-121, and Ser-123 are critical for the binding of ligands and ligand-mimetic mAbs (19,21,49). Taken together, the present results are consistent with the idea that the DDLW sequence, which overlaps with site B3, makes direct contact FIG. 3. Effect of several human-to-mouse mutations in ␣ IIb and ␤ 3 on activation-dependent PAC-1 and fibrinogen binding. a, PAC-1 binding to wild-type and mutant ␣ IIb ␤ 3 . Cells were first incubated with PL98DF6 (anti-human ␣ IIb ) followed by PE-conjugated antimouse IgG. After washing, cells were incubated with FITC-labeled PAC-1 in the presence of control mouse IgG or PT25-2 (activating anti-␣ IIb ␤ 3 mAb) in 5 mM Hepes/Tyrode, 2 mM Ca 2ϩ , and 2 mM Mg 2ϩ buffer, pH 7.4. Binding of FITC-labeled PAC-1 to the PE-labeled ␣ IIbpositive cell population expressing high level ␣ IIb ␤ 3 (fluorescence intensity Ͼ10 2 ) was examined in FACScan. Data are expressed as a difference in median FITC fluorescent intensity in the presence and absence of PT25-2 (solid column). Median PE fluorescence intensity in the cell population used for measurement of PAC-1 binding is shown (open column). The data suggest that PAC-1 binding is completely blocked by the A16 mutation and partly blocked by the A23 and B6 mutations. b, fibrinogen binding to wild-type and mutant ␣ IIb ␤ 3 . Fibrinogen binding to the cell population expressing high level ␣ IIb ␤ 3 was determined as described in a, except that FITC-fibrinogen was used instead of FITC-PAC-1. Data are expressed as a difference in median FITC fluorescent intensity in the presence and absence of PT25-2 (solid column). Median PE fluorescence intensity in the PE-labeled ␣ IIb -positive cell population used for measurement of fibrinogen binding is shown (open column). The data suggest that the mutants tested maintain fibrinogen binding, although the A23 mutant shows relatively low fibrinogen binding, and the A16 mutant shows relatively high fibrinogen binding. Parent CHO cells do not show ␣ IIb ␤ 3 -specific fibrinogen binding, since parent CHO cells are PE-negative (FL-2 Ͻ10 2 ).

TABLE VI
Effects of the ␤ 3-1-3 mutation on ␣ IIb ␤ 3 binding to ligand-mimetic mAbs and fibrinogen Cloned cells stably expressing wild type or mutant ␣ IIb ␤ 3-1-3 were used. Reactivity to mAbs and binding of FITC-labeled soluble fibrinogen were assayed by flow cytometry. Data are expressed as the ratio of mean fluorescent intensity. The data suggest that the ␤ 3-1-3 mutation blocks binding of ligand-mimetic mAbs and fibrinogen to ␣ IIb ␤ 3 .  4. Binding sites for function-blocking and/or ligand-mimetic mAbs are close to or overlap with several sequences/ residues that are critical for ligand binding and specificity. a, sites A9, A16, and A23 for function-blocking and/or ligand-mimetic mAbs are located at the boundary (the 4-1 loop) between repeats 1 and 2, repeats 2 and 3, and repeats 3 and 4, respectively. Sites A16 and A23 are close to or overlap with conserved residues that are critical for ligand binding (boxed, see Introduction for references). b, sites B3 and B6 are close to or overlap with the sequences that are critical for ligand binding or specificity (boxed, see Introduction for references). These results suggest that the conserved critical sequences/residues for ligand binding and specificity that are close to sites A16, A23, B3, and B6 actually make direct contact with ligands.
with ligands through the RGD motif and is part of the ligand binding site.
It has been reported that ␣ IIb ␤ 3 may have two distinct ligand binding sites based on kinetic analysis of ligand binding (Refs. 50 and 51 and references therein), but positions of the proposed binding sites are unknown. The RGD-containing mAb 16N7C2 recognizes site B3, which overlaps with the putative RGD binding site in ␤ 3 , but this mAb does not require ␣ IIb subunit for binding. The three ligand-mimetic mAbs recognize site A16, which overlaps with critical residues for ligand binding in ␣ IIb .
Interestingly, human-to-mouse mutations in ␤ 3 (including B3) have relatively minor effects on binding to ligand-mimetic mAbs compared with the A16 mutation (see above). These results suggest that mAb 16N7C2 and the ligand-mimetic mAbs recognize two separate sites in ␣ IIb ␤ 3 (B3 and A16, respectively) in an RGD-or RYD-dependent manner. Thus sites A16 and B3 may represent two distinct RGD or RYD recognition and binding sites in ␣ IIb ␤ 3 .
We found that several function-blocking mAbs, including LM609 and AP-2, recognize site B5. These results suggest that site B5 is close to or within the putative ligand binding pocket. Since the noninhibitory mAb D57 recognizes site B5, it is not certain whether site B5 is directly related to ligand binding. Site B5 is located in the large predicted loop protruding from the global structure (Fig. 5). Thus it is possible that antibodies access this loop from different directions or change conformation of the predicted loop in different ways. However, we need more definitive structure of the domain to test these possibilities.
The predicted loop region containing Asn-215, Asp-217, and Glu-220 is critical for the binding of ligand-mimetic mAbs or ligands (21). We did not use human-to-mouse mutagenesis to study whether this region overlaps with the binding sites for ligand-mimetic mAbs, because this region is highly conserved between human and mouse ␤ 3 . The present results are useful for evaluating two folding models of the putative I-domain-like structure of the ␤ subunit (21,23). The present findings (that sites B3, B5, and B6 are located close to each other on the ligand binding side of the ␤ 3 subunit) are consistent with the model by Tuckwell et al. (23) (Fig. 5). In this model, the putative RGD binding site (close to site B3) and the predicted loop that is critical for ligand specificity (site B6) are surrounded by conserved oxygenated residues that are critical for binding to ligands and ligand-mimetic mAbs in the ␤ 3 subunit (Asp-119, Ser-121, Ser-123, Asp-217, and Glu-220). Consistently, these conserved residues are critical for binding to OPG2/PAC-1 (19,21,49). In the model by Tozer et al. (21), site B6 is in the apparently nonligand binding site of the domain, indicating that this model does not fit in with the present mapping results.
Localization of the Binding Sites for Ligand-mimetic mAbs and Ligands at the ␣/␤ Boundary- Fig. 5b is a model of the ␣ IIb ␤ 3 globular domain in which the proposed ␤-propeller domain and the putative I-domain are associated. The regions critical for the binding of ligand-mimetic mAbs and ligands are all in the upper face of both subunits. Interestingly, these results directly demonstrate that the predicted loop at the boundary between repeats 2 and 3 (containing sites A16 and A23) in the proposed ␤-propeller domain faces the metal iondependent adhesive site (MIDAS) region of the putative ␤ 3 I-domain (containing sites B3 and B6) in the quaternary structure (Fig. 5). These results predict that these binding sites for ligand-mimetic mAbs and native ligands constitute a ligand binding pocket at the ␣/␤ boundary. The positions of binding sites A9, A16, B3, B5, or B6 in this model are consistent with the function-blocking activity of several nonligand-mimetic mAbs (e.g. 7E3, LJ-P9). It is somewhat surprising that the binding of mAbs that are known to be complex-specific can be completely inhibited by substitutions in single subunits. One possibility is that one subunit is involved in antibody specificity by regulating the accessibility to the epitope region.
In summary, we have established that ligand-mimetic antibodies bind to several discontinuous sites in both ␣ IIb and ␤ 3 subunits. It is likely that this unique binding property is related to their ligand-mimetic property (cation and activation dependence). These binding sites are close to or overlapping  -propeller domain (b). a, the model of the ␤ 3 I-domain was taken from Takagi et al. (15) and Tuckwell et al. (23) and modified. Arrows indicate ␤-sheets, and columns indicate ␣-helices. Closed circles show residues that are critical for ligand binding in ␤ 1 or ␤ 3 (16). In this model, the diverse disulfide-linked sequence that is critical for ligand specificity (residues 177-184 in ␤ 3 ) (15) is located in the predicted loop, surrounded by conserved oxygenated residues in the upper face of the I-domain-like structure of the ␤ 3 subunit that are critical for ligand binding (e.g. Asp-119). The upper face of this domain is predicted to be a ligand binding site, based on its homology to the I-domains of ␣ M and ␣ L (23). Site B3 is located in helix 1 and overlaps the putative RGD binding site (10). Site B6 is located in the predicted disulfide-linked sequence that is critical for ligand specificity (15). It should be noted that all of the binding sites for ligand-mimetic or function-blocking mAbs are located on the same side of the model (the putative ligand site) and are predicted to be close to each other. b, the approximate positions of the ␣ IIb binding sites are shown. Sites A9, A16, and A23 are located at the boundary between repeats 1 and 2, repeats 2 and 3, and repeats 3 and 4, respectively. Site A29, which is recognized by the activating mAb PT25-2, is located in the lower face (nonligand binding site) of the domain. Also, the potential relative positions of the ␣ IIb proposed ␤-propeller domain and the putative ␤ 3 I-domain-like structure are shown. In this model, sites A16 and A23 face sites B3 and B6 in the ␤ 3 subunit, generating a ligand binding pocket at the ␣/␤ boundary.
with residues/regions that are critical for ligand binding, suggesting that native ligands (e.g. fibrinogen) make direct contact with these discontinuous binding sites in both subunits. These results are consistent with the previous assumption that ligand-mimetic mAbs may have structural and functional similarities to native ligands. It would be interesting to study whether the present model of ␣ IIb ␤ 3 may be applicable to other non-I domain integrins.