A Structural Hypothesis for the Transition between Bent and Extended Conformations of the Leukocyte β2 Integrins*

Integrins mediate cell adhesion in response to activation signals that trigger conformational changes within their ectodomain. It is thought that a compact bent conformation of the molecule represents its physiological low affinity state and extended conformations its active state. We have determined the structure of two integrin fragments of the β2 subunit. The first structure, consisting of the plexin-semaphorin-integrin domain, hybrid, integrin-epidermal growth factor 1 (I-EGF1), and I-EGF2 domains (PHE2), showed an L-shaped conformation with the bend located between the I-EGF1 and I-EGF2 domains. The second structure, which includes, in addition, the I-EGF3 domain, showed an extended conformation. The major reorientation of I-EGF2 with respect to the other domains in the two structures is accompanied by a change of torsion angle of the disulfide bond between Cys461-Cys492 by 180° and the conversion of a short α-helix (residues Ser468-Cys475) into a flexible coil. Based on the PHE2 structure, we introduced a disulfide bond between the plexin-semaphorin-integrin domain and I-EGF2 domains in the β2 subunit. The resultant αLβ2 integrin (leukocyte function-associated antigen-1) variant was locked in a bent state and could not be detected with the monoclonal antibody KIM127 in Mg2+/EGTA. However, it retained the binding activity to ICAM-1. These results provide a structural hypothesis for our understanding of the transition between the resting and active states of leukocyte function-associated antigen-1.

Integrins are cell surface receptors that play key roles in cellcell, cell-extracellular matrix, and cell-pathogen interactions (1). Moreover, their importance in tumor metastasis is increasingly appreciated (2). Integrins on circulating leukocytes are normally in a resting state of low adhesiveness, but they can rapidly become activated in response to internal "inside-out" or external "outside-in" signals. Integrin molecules are formed by two non-covalently associated ␣ and ␤ subunits, both type 1 membrane glycoproteins, with a globular ligand-binding "head" linked to two rod-like "legs" (3,4). The domain organization of a typical integrin molecule is schematically depicted in Fig. 1A. The crystal structure of the ectodomain of the ␣V␤3 integrin reveals a compact V-shaped molecule having each leg markedly bent, thus orientating the headpiece toward the plasma membrane (4). Extensive analyses of electron microscopic images of ␣V␤3 suggested that its compact form could represent a resting state, and extension of the legs may be associated with activation (5). In association with activation, it was also suggested that the hybrid domain swings out with respect to the headpiece, a hypothesis later substantiated by x-ray crystallographic studies of a fragment of the ␣IIb␤3 integrin, consisting of the ␤-propeller from the ␣ subunit and the PSI, 3 ␤I, and hybrid domains from the ␤ subunit (6). This interpretation was supported by further electron microscopic image analyses of the ␣IIb␤3, ␣5␤1, ␣L␤2, ␣X␤2, and ␣IIb␤3 integrins (5,(7)(8)(9)(10). However, it is not clear whether extension is absolutely necessary for ligand binding (11).
The structures of I-EGF1, I-EGF2, and I-EGF3 however, were missing in these studies. The structure of I-EGF3 of the ␤2 integrin subunit was determined by NMR (12), which led to the definition of a novel subset of EGF-like domains typically found in integrins, now referred to as integrin-EGF (I-EGF) domains. Using x-ray crystallography, we have determined the structure of I-EGF1 as part of a fragment of the ␤2 subunit, consisting of the PSI, hybrid, and I-EGF1 domains (13). This fragment is hereafter named PHE1 (Fig. 1B). Superposition of the PHE1 structure onto the bent ␣V␤3 ectodomain crystallographic structure (4) suggested that the bend in the ␤ subunit must lie between the I-EGF1 and I-EGF2 domains, a conclusion also supported by electron microscopic studies of the ␣L␤2 and ␣X␤2 integrins (10). However, a clear picture of the modules making up the "knee" (Fig. 1A) in the ␤ subunit has not emerged yet, and the atomic basis for its flexibility is still largely unresolved. Thus, to further understand the molecular mechanism of integrin activation, it is crucial to obtain the structure of the I-EGF2 domain, and to determine how it is connected with the neighboring integrin modules, both in their bent and extended states. Accordingly, we expressed the PHE2 and PHE3 fragments (Fig. 1B), both from the ␤2 integrin subunit, and determined their structures by x-ray crystallography. The PHE2 integrin fragment assumes a bent conformation at the junction between the I-EGF1 and I-EGF2 domains, and PHE3 is extended. We propose that structural changes within the I-EGF2 domain may act as a conformational switch associated with integrin activation. To test this hypothesis, we engineered a disulfide bond in the full-length leukocyte function-associated antigen-1 (LFA-1) molecule and showed that it can be locked in a bent conformation.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification-The construction of PHE2 and PHE3 was similar to that reported for PHE1 (13), except that the I-EGF2 domain (residues Cys 461 -Glu 513 ) is also included in PHE2, and both I-EGF2 and I-EGF3 domains (residues Cys 461 -Gln 552 ) in PHE3. The final amplified cDNA constructs of PHE2 and PHE3 were subcloned into the pIRES2-EGFP vector (BD Biosciences Clontech) and transfected using the Polyfect reagent (Qiagen) into the HEK293S GnTI Ϫ cell line that is deficient in N-acetylglucosaminyltransferase-I (GnTI) (14). Stable cell lines incorporating the expression vector were selected by culturing transfectants in media (Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum and penicillin/streptomycin) containing G418 (1 mg/ml, Invitrogen), followed by their isolation based on the EGFP expression, using the fluorescence-activated cell sorting Aria flow cytometry-based cell sorter (BD Biosciences). The clones that secreted the highest quantity of recombinant proteins, as determined by the intensity of EGFP signal and confirmed by Western blotting using an anti-His-tag antibody, were maintained in culture media in 150-mm tissue culture dishes. The spent media were harvested every 10 days, concentrated, and exchanged into buffer A (50 mM Tris-HCl, 200 mM NaCl, pH 7.5). PHE2 and PHE3 were purified using nickel-nitrilotriacetic acid-agarose beads (Qiagen). Nickel-nitrilotriacetic acid-purified PHE3 was dialyzed against 30 mM Tris-HCl at pH 8.0 overnight at 4°C and further purified by anion-exchange chromatography on a Mono Q 5/5 column (Amersham Biosciences), mounted on an Akta fast-protein liquid chromatography system (Amersham Biosciences). PHE2 and PHE3 samples were concentrated to 1 ml and subjected to size-exclusion chromatography on a Superdex-75 16/180 column (Amersham Biosciences) in buffer A. Purified PHE2 and PHE3 were concentrated to 1 mg/ml and treated with endoglycosidase H f (4000 units/ml, New England Biolabs) overnight at room temperature. After further purification by size-exclusion chromatography, the proteins were concentrated to 15 mg/ml in 20 mM Tris-HCl, pH 7.2. Molecular masses were determined by matrix-assisted laser desorption ionization time-of-flight mass spectrometry.
Crystallization and Data Collection-Crystallization was performed by vapor diffusion using the hanging drop method at 18°C. PHE2 crystals grew over a week into thin plates of dimensions ϳ0.05 ϫ 0.3 ϫ 0.4 mm 3 in a precipitating solution containing 0.2 M magnesium acetate, 0.1 M sodium acetate (pH 4.6), and 18% polyethylene glycol 3350. PHE3 crystals grow in 0.2 M ammonium sulfate, 15% polyethylene glycol 4000, and 5% isopropanol as hexagonal plates, with dimensions of ϳ0.1 ϫ 0.4 ϫ 0.5 mm 3 . Both crystals contain one monomer per asymmetric unit, with solvent contents of 34 and 48% and V m of 1.86 and 2.35, respectively. Before cooling the crystals to 100 K in a nitrogen gas stream (Oxford Cryosystems), three rounds of 12-h increases in polyethylene glycol concentrations (7% each cycle) were carried out in the mother liquors (15), and finally the polyethylene glycol concentrations were 39 and 36% for PHE2 and PHE3, respectively. PHE2 native data were collected on an R-axis IVϩϩ Image Plate detector using CuK ␣ radiation from a Micromax-007 rotating anode. PHE3 native diffraction data were recorded on an ADSC charge-coupled device detector (ADSC Corp., Powey, CA) on the ID23-1 beamline at the European Synchrotron Radiation Facility (Grenoble, France). A data set was collected at a wavelength of 1.907 Å, and anomalous difference Fourier synthesis was performed to confirm the location of the disulfide bonds. Data were processed using programs MOSFLM and SCALA (16). Crystal parameters and data collection statistics are summarized in Table 1.
Structure Determination and Refinement-The PHE2 and PHE3 structures were determined by molecular replacement using the program PHASER (17) with PHE1 (1YUK) and PHE2 lacking the I-EGF2 domain, as search models, respectively. Molecular replacement for the I-EGF3 domain was performed using its NMR structure 1L3Y (12) as a search probe. Models were improved by alternating cycles of model building with the program Coot (18) and refinement with REFMAC (16). The relatively high refinement R values for the PHE3 crystal form are probably caused by the large anisotropy observed in its unitcell dimensions (a ϭ b ϭ 52.3 Å, c ϭ 423.9 Å, see Table 1). PHE3 crystals diffract anisotropically, and refinement programs can only partially compensate for such effects. Based on the R free values, the estimated overall coordinate errors for the PHE2 and PHE3 molecules are 0.16 and 0.26 Å, respectively. Overall, the chain traces are unambiguous with clear electron density, including, for a single N-acetylglucosamine residue observed on all three potential N-glycosylation sites in the PHE2 protein, corresponding to residue Asn 28 in the PSI, Asn 94 in the hybrid, and Asn 479 in the I-EGF2 domains. For the PHE3 protein, electron density for an N-acetylglucosamine residue could be observed only on Asn 94 in the hybrid domain. Four histidine residues from the hexahistidine tag were visible at the C-terminal end. Three segments are poorly defined in PHE3: residues His 69 -Gly 72 in the hybrid domain, residues Arg 432 -Asp 433 in the I-EGF1 domain, and residues Ser 467 in the I-EGF2 domain. These regions, which are not included in the refined PHE3 structure, are exposed to the solvent and are likely to be flexible. Translation, libration, and screw-rotation displacement introduced in the last refinement step. Solvent-accessible surfaces were calculated using the program AREAIMOL (16) with a 1.7-Å radius sphere as the probe (Table 3) and values rounded to the nearest 5 Å 2 . Conformational differences were analyzed using the DynDom server. Figures were created using PyMOL (19).
Mutagenesis and Transfection of 293T Cells-Plasmids encoding the full-length ␣L and ␤2 human integrin subunits in pcDNA3 were described previously (20). Site-directed mutagenesis was performed using the QuikChange kit (Stratagene). All mutations were confirmed by DNA sequencing. 293T cells were transfected using Polyfect reagent (Qiagen), according to the manufacturer's instructions. 2.0 g each of the ␣L and ␤2 cDNA, wild type or mutant, were used to cotransfect one 6-cm dish of cells at 70 -80% confluency.
Surface Labeling and Detection-Free cysteine labeling was performed as described previously (21). Briefly, transfected cells were washed once Tris-buffered saline (TBS) (20 mM Tris-HCl, 200 mM NaCl, pH 7.4) containing 1 mM Ca 2ϩ , and labeled with 400 M of biotin-BMCC (1-biotinamido-4-(4Ј-[maleimidoethylcyclohexane]carboxamido)butane, Pierce) for 30 min at room temperature. Cells were washed three times with TBS with 1 mM Ca 2ϩ and lysed in TBS containing 1% Triton X-100 and 0.1% Nonidet P-40 at 4°C. LFA-1 was immunoprecipitated with MHM23 (2 g) and Protein A-Sepharose beads (Amersham Biosciences) at 4°C for 1 h and resolved on a non-reducing 7.5% SDS-PAGE gel. After transferring onto polyvinylidene difluoride membrane (Millipore), biotin-labeled LFA-1 bands were detected using streptavidin-horseradish peroxidase, followed by ECL using the ECL-plus Western blotting kit (Amersham Biosciences). A positive control was performed with wildtype LFA-1 transfectants. Cells were first treated with 5 mM DTT at 37°C for 30 min in TBS supplemented with 1 mM Ca 2ϩ before labeling with biotin-BMCC. The ␣L band of LFA-1 was detected on a separate blot using the anti-␣L antibody (clone 27) followed by detection with a horseradish peroxidase-conjugated sheep anti-mouse IgG.
Integrin surface labeling with biotin was performed as described before (22). Briefly, cells were washed once in phosphate-buffered saline, and treated with 0.5 mg/ml sulfo-NHSbiotin (Pierce) in phosphate-buffered saline at room temperature for 30 min. The reaction was quenched by washing the cells once in TBS. The labeled cells were incubated with 2 g of MHM23 or KIM127 in culture medium or in phosphate-buffered saline containing Mg 2ϩ /EGTA (5 mM MgCl 2 and 1.5 mM EGTA) at 37°C for 30 min. The labeled cells were lysed in TBS containing 1% Triton X-100 and 0.1% Nonidet P-40 at 4°C. LFA-1 was precipitated with Protein A-Sepharose beads (Amersham Biosciences) at 4°C for 1 h and resolved on a 7.5% SDS-PAGE gel under reducing conditions. Biotinylated LFA-1 was detected after transferring onto polyvinylidene difluoride membrane. In the attempt to reduce the introduced disulfide bond in the ␣L␤2G33C/G486C variant, the transfectants were first incubated with varying concentrations of DTT in culture medium with Mg 2ϩ /EGTA at 37°C for 15 min before the addition of the KIM127 antibody for immunoprecipitation.
Flow Cytometry-Staining of cells for flow cytometry analyses was described previously (23). Briefly, cells were incubated with the primary mAb at 10 g/ml in Dulbecco's modified Eagle's medium media for 1 h at 4°C. Cells were washed twice and incubated with fluorescein isothiocyanate-conjugated sheep F(abЈ) 2 anti-mouse antibody (Sigma) at 1:400 dilution for 45 min at 4°C. Stained cells were washed once and fixed in 1% (v/v) formaldehyde in phosphate-buffered saline before analysis on a FACSCalibur flow cytometer (BD Biosciences) using the CellQuest software.
Adhesion Assay-Adhesion of 293 transfectants to ICAM-1 was carried out as previously described (24).

RESULTS
Expression of the PHE2 and PHE3 Fragments-The PHE2 fragment comprises the PSI (residues Gln 1 -Asp 58 and Glu 424 -Arg 426 ), hybrid (residues Pro 59 -Ala 100 and Lys 340 -Cys 423 ), I-EGF1 (residues Cys 427 -Glu 460 ) and I-EGF2 domains (residues Cys 461 -Glu 513 ). In addition to the domains that make up PHE2, the PHE3 fragment also includes the I-EGF3 domain at its C-terminal end (residues Cys 514 -Gln 552 ) (Fig. 1B). Typical preparations of PHE2 and PHE3 proteins before and after treatment with endoglycosidase H f are shown in Fig. 1C. The recombinant PHE2 protein recovered from the cultured media was shown to express conformational epitopes of mAbs H52, 7E4, KIM202, and MEM148 (20,23,24) and KIM89 4 (Fig. 1D) indicating that it is in a native conformation. In addition, the PHE3 fragment, but not PHE2, is recognized by mAbs MEM48 and KIM127, in agreement with previous findings that the expression of the epitopes of these two antibodies requires both I-EGF2 and I-EGF3 domains (25,26). MHM24, whose epitope is in the ␣L subunit, is included as a negative control.
Structural Determination of PHE2 and PHE3-The structures of PHE2 and PHE3 recombinant integrin ␤2 chain fragments were determined by x-ray crystallography. A summary of the data collection and refinement statistics for crystals of the PHE2 and PHE3 is shown in Table 1. Both structures were determined by molecular replacement using individual components from the PHE1 fragment and the I-EGF3 domain as search probes (12,13). As independent confirmations of the correctness of the polypeptide chain traces, we used the anomalous signal from the sulfur atoms to confirm the positions of the 13 disulfide bridges present in the PHE2 structure, using CuK ␣ radiation (at 1.5418 Å; see Table 1 and supplementary Fig. S1A). In addition, an anomalous Fourier map was computed from a low energy data set (collected at 1.907 Å, see Table  1) using the same PHE3 crystal (Table 1), confirming the location of the disulfide bridges present in the PHE3 fragment (Fig.  S1B). The overall structures of PHE2 and PHE3 were drastically different: whereas PHE2 assumed a compact L-shaped configuration, PHE3 was extended (Fig. 2).
The Bent Structure of PHE2-Each of the three domains found in both PHE1 (13)  r.m.s. deviations in atomic positions Ͻ0.9 Å (Table 2). Moreover, their relative orientation was essentially preserved between these two fragments. The largest movement observed among these three domains occurred in a long loop (Thr 30 to Arg 39 ) in the PSI domain, which moved toward the I-EGF1 domain by ϳ2.0 Å in PHE2, thereby bringing several of its residues into contact with Lys 457 of I-EGF1 and Asp 489 of I-EGF2 ( Figs.  2A and 4A). I-EGF2 formed an acute angle with I-EGF1, giving PHE2 an L-shaped conformation (Fig. 2A). The associated interfaces, however, were modest with values between the I-EGF2 domain and the PSI and I-EGF1 domains of 405 Å 2 and 740 Å 2 , respectively ( Table 3).   a The same crystal of PHE3 was used for both data sets, which were collected successively at high (left) and low (right) energy. The second PHE3 data set was collected to maximize the anomalous signal from the sulfur atoms. An anomalous difference Fourier map was calculated to confirm the trace of the polypeptide chain (see supplementary Fig. S1). The PHE3 structure was refined using the high energy data set only. b The numbers in parentheses refer to the last (highest) resolution shell. c R merge ϭ ⌺ h ⌺ i ͉I hi Ϫ ͗I h ͉͘/⌺ h,i I hi , where I hi is the ith observation of the reflection h, whereas ͗I h ͘ is its mean intensity. d N/A, not applicable. e R work ϭ ⌺ʈF obs ͉ Ϫ ͉F calc ʈ/⌺͉F obs ͉. f R free was calculated with 5% of reflections excluded from the whole refinement procedure. OCTOBER 12, 2007 • VOLUME 282 • NUMBER 41

JOURNAL OF BIOLOGICAL CHEMISTRY 30201
Ser 468 to Cys 475 in the bent (PHE2) conformation, but becomes flexible in the elongated (PHE3) form (see below).
The Extended Conformation of PHE3-In contrast to PHE2, the PHE3 fragment adopted an extended conformation with overall dimensions of ϳ128 Å ϫ 30 Å ϫ 30 Å (Fig. 2B). Superposition of individual domains from PHE3 with the corresponding segments in the PHE1 and PHE2 showed higher r.m.s. deviations compared with those observed between PHE1 and PHE2 ( Table 2). The PSI-hybrid tandem remains invariant in all structures reported for ␤2 and ␤3 fragments (6,13,27), including PHE2 and PHE3 reported here. However, the I-EGF domains in these two structures show remarkable differences in their relative orientation to the PSI-hybrid tandem. With reference to the invariant PSIhybrid tandem, two maneuvers are required to bring the I-EGF1 and I-EGF2 domains from the PHE2 conformation to the PHE3 conformation: first a rotation of 40°about an axis passing through residues Lys 427 -Ser 431 and a 0.2-Å shift (a value that falls within the experimental errors), followed by a rotation of 69°of the I-EGF2 domain about an axis passing through residues Ser 468 -Glu 472 with a translation of 0.8 Å (supplementary Fig S2). As a result, the interface between I-EGF1 and I-EGF2 domains is reduced from 740 Å 2 in PHE2 to 535 Å 2 in PHE3, and the contact area of 405 Å 2 between the PSI and I-EGF2 domains in PHE2 no longer exists in PHE3. Accordingly, the interactions between Lys 457 of I-EGF1, Asp 489 of I-EGF2, and the main-chain amide groups projecting from the loop of the PSI domain in PHE2 were disrupted in PHE3 (Fig. 4A). Perhaps the most interesting difference between the two structures lies in the I-EGF2 domain. The disulfide bond between Cys 461 and Cys 492 (C1 to C5 within the I-EGF2 domain) assumes different conformations in PHE2 and PHE3. The dihedral angle about this S-S disulfide bond (3 torsion angle) switches from the right-handed (3 ϳ 100°) in PHE2 to the left-handed (3 ϳ Ϫ80°) stable conformer (Fig. 3B). Concomitantly, the short ␣-helix (residues Ser 468 -Cys 475 ) in PHE2 became disordered in the elongated integrin structure.
The Bent Conformation Observed in the PHE2 Crystal Structure Can Be Adopted by a Complete Integrin Heterodimer-The PHE2 and PHE3 molecules are only fragments of a complete integrin ␤2 subunit. To rule out possible artifacts introduced by crystal packing forces and to assess whether the bent conformation displayed by the PHE2 molecule could also be adopted by a complete integrin receptor, we used site-directed   mutagenesis to introduce an artificial disulfide bond that would constrain the integrin in the bent conformation observed in PHE2. In the PHE2 crystal structure, the ␣-carbon atoms of residues Gly 33 from the PSI and Gly 486 and Leu 487 from the I-EGF2 domains are separated by distances of 4.2 and 5.0 Å, respectively, which are compatible with the distance spanned by disulfide bonds (Fig. 4A). Two double substitutions were made in the full-length ␤2 integrin subunit, with cysteines replacing Gly 33 and Gly 486 in one (␤2G33C/G486C) and Gly 33 and Leu 487 in the other (␤2G33C/L487C). In addition, a single mutant having Gly 33 substituted by a cysteine residue (␤2G33C) was also constructed. The ␣L and ␤2 cDNA plasmids were transfected into 293T cells. All three ␤2 variants supported LFA-1 (leukocyte function-associated antigen-1, ␣L␤2 integrin) surface expression, as detected by flow cytometry using mAb MHM23, which is specific for  the heterodimer (Fig. 4B). The transfectants were surface-labeled with biotin-BMCC, a reagent containing maleimide, which binds to free sulfhydryl groups. LFA-1 was immunoprecipitated from solubilized membranes of the transfectants with MHM23. After SDS-PAGE and blotting, the proteins were analyzed for biotin labeling (Fig. 4C, upper panel). Minimal signal was detected from the bands of the wild-type ␣L␤2. In contrast, the ␤2G33C and ␤2G33C/L487C bands were labeled, indicating the presence of free sulfhydryl in these variant ␤2 subunits. The intensity of the ␤2G33C/ G486C band was significantly lower than those of the other two mutants, suggesting that, for the majority of the molecules, a disulfide bond was formed between the two introduced cysteine residues. The same samples were also analyzed for the presence of the ␣L subunit by Western blot using the anti-␣L mAb clone 27. The results (Fig. 4C, lower  panel) showed that the ␣L subunits are present in equal amounts in the four transfectants, suggesting that the different levels of free cysteines in the four samples were not due to differences in immunoprecipitation.
Expression of the KIM127 epitope in LFA-1 is associated with an active extended conformation (10,25). To test the reactivity of KIM127 with the LFA-1 variants, transfectants were surface-labeled with biotin (22), and the labeled cells were incubated at 37°C with KIM127 and Mg 2ϩ /EGTA. Cells were subsequently lysed, and biotin-labeled LFA-1 was precipitated with protein A-Sepharose. KIM127 was able to immunoprecipitate the wild-type LFA-1 and the ␣L␤2G33C and ␣L␤2G33C/L487C LFA-1 variants but not the ␣L␤2G33C/G486C LFA-1 variant (Fig. 4D). This was not due to inefficient labeling of LFA-1 variant ␣L␤2G33C/ G486C, because the control mAb MHM23 precipitated LFA-1 and its three variants. Thus, the lack of reactivity of KIM127 with LFA-1 variant ␣L␤2G33C/G486C suggests that the engineered disulfide bridge, as a result of G33C/ G486C mutations, restrains LFA-1 in a conformation that is incapable of full extension in the presence of activating Mg 2ϩ / EGTA. It should be noted that wild-type LFA-1 under physiological divalent cation concentration in culture medium is not precipitated by KIM127 as previously reported (28). Furthermore, MEM48 can only precipitate the wild-type but not the ␣L␤2G33C/G486C variant in Mg 2ϩ /EGTA. In contrast, the disulfide bond does not affect the ␣L␤2G33C/G486C variant precipitation by KIM185.
An attempt was made to reduce the introduced disulfide bond in the ␣L␤2G33C/G486C variant with varying concentrations of DTT. Even at 30 mM DTT, the ␣L␤2G33C/G486C cannot be precipitated with KIM127 in the presence of Mg 2ϩ / EGTA, although under these conditions, the precipitation of wild-type ␣L␤2 is easily detected (Fig. 4E). At 100 mM DTT, the epitope is abolished from the wild type, presumably by the reduction of other disulfide bonds.

DISCUSSION
A number of reports have described conformational changes accounting for the allosteric mechanism of integrin activation (for reviews, see Refs. 29 -31). The ␣V␤3 integrin crystal structure revealed an overall V-shaped compact molecule, but only the bend in the ␣V subunit could be resolved (4). The junction between I-EGF1 and I-EGF2 of the ␤3 subunit was not visible in the electron density map, but the location of the bend could be deduced from the positions of other domains, namely the PSI and the hybrid domain on one side, and the I-EGF3 domain on the other (4,13). No atomic structure is available for either the ␣ or the ␤ subunit in their extended conformation. In this study, we determined the structures of the PHE2 and PHE3 fragments of the integrin ␤2 subunit. An acute bend is seen in the PHE2 structure between the I-EGF1 and I-EGF2 domain, whereas PHE3 assumes an extended structure. How do these structures relate to the bent and extended conformations of a complete integrin LFA-1 molecule?
The extended structure of PHE3 may faithfully reflect the extended conformer of LFA-1 because of its reactivity with a panel of conformation-sensitive mAbs, in particular, with KIM127, which only detects an extended LFA-1 conformer (10,25). By contrast, it is not immediately apparent from the lack of KIM127 reactivity with PHE2 that this structure depicts a bent conformer of LFA-1, because the expression of the KIM127 epitope requires, in addition to I-EGF2, I-EGF3, which is absent in PHE2 (26). To verify that the L-shaped conformation in the PHE2 structure also occurs in an intact integrin, we introduced two cysteine residues, one to replace Gly 33 in the PSI domain and the other to replace Gly 486 in the I-EGF2 domain of the ␤2 subunit. The distance between the C␣ atoms of these two residues was 16 Å in the extended (PHE3) form but only 4.2 Å in the bent (PHE2) structure. Thus, we expect that this disulfide bond would only be formed if these two cysteine residues are brought in close proximity, as seen in the compact PHE2 structure. Formation of this engineered disulfide bond in this LFA-1 variant is supported by data shown in Fig. 4. In addition, the epitope of KIM127 was not expressed in the presence of Mg 2ϩ /EGTA, which is also expected if the LFA-1 variant is locked by the engineered disulfide into a bent conformation. Thus, these data provide direct evidence for the presence of a bent conformation adopted by LFA-1 integrins at the cell surface that mirrors the PHE2 structure. On the other hand, in the absence of a disulfide bond such as in the LFA-1 variants having ␣L␤2G33C and ␣L␤2G33C/L487C substitutions, the extended conformations can be detected by the KIM127 monoclonal antibody.
The MEM48 epitope, which also requires both I-EGF2 and I-EGF3 for expression in wild-type LFA-1 and is not sensitive to Mg 2ϩ /EGTA, is not expressed in the ␣L␤2G33C/G486C variant. This is not unexpected, because the introduced disulfide bond would lock I-EGF2 into a conformation that does not allow the I-EGF2 and I-EGF3 to bind MEM48 by an induced-fit mechanism. On the other hand, the expression of the epitope of KIM185, which maps to the I-EGF4/␤-TD domains (25), is not affected.
Our attempt to reduce the introduced disulfide bond on the ␣L␤2G33C/G486C variant was not successful. The expression of the KIM127 epitope was not affected in 30 mM DTT in the wild-type LFA-1. Under these conditions, re-expression of the KIM127 epitope was not detected on the ␣L␤2G33C/G486C variant, suggesting that the introduced disulfide between the PSI and I-EGF2 domains is not accessible for reduction. At 100 mM DTT, expression of the epitope was abolished in the wildtype LFA-1, presumably by the reduction of other disulfide bonds in the ␤2 subunit.
The ␣L␤2G33C/G486C variant can mediate adhesion to ICAM-1, suggesting that the ligand binding site is intact. Furthermore, it also suggests that leg extension is not an absolute requirement for ligand binding, at least under this particular assay system.
In the PHE2 and PHE3 structures, the I-EGF2 domain assumes two different conformations involving alternative dihedral torsion angles of a disulfide bond between Cys 461 and Cys 492 and the presence or absence of a short ␣-helix (Fig. 3B). This difference may represent the conformational switch within the leg of the integrin ␤2 subunit associated with a large reorientation of the I-EGF1 and I-EGF2 domains. An activation energy barrier of ϳ7 kcal/mol has to be overcome to switch the disulfide bond from one favorable conformer to the other (32). However, this kinetic energy barrier would not contribute to affinity regulation when the system is at thermodynamic equilibrium. Whether an exogenous disulfide isomerase is involved in the transition remains to be clarified (33). It is also possible that the integrin may have intrinsic disulfide isomerase activity (34).
The two different structures of PHE2 and PHE3 reported here clearly bear relevance to the transition between the bent and the extended conformation of the LFA-1 integrin molecules. Conformational changes within the I-EGF2 domain may be a key feature for this transition. How this conformational change is coupled with others described, including leg separation (35), removal of a dead-bolt (36), swing-out of the hybrid domain (6), and conformational changes in the I-domains of the ␣ and ␤ subunits (37, 38, and for reviews see refs. 29 -31) will require further examination. Further work is also necessary to assess whether comparable transitions occur in the ␤1 and ␤3 subunits, whose family members undergo similar conformational changes during activation.