An isoleucine-based allosteric switch controls affinity and shape shifting in integrin CD11b A-domain.

In response to cell activation signals, integrins switch from a low to a high affinity state. Physiologic ligands bind to integrins through a von Willebrand Factor A-type domain. Crystallographic studies revealed two conformations of this domain, "closed" and "open." The latter crystallizes in complex with a pseudoligand or ligand, suggesting that it represents the high affinity state; data linking structure and activity are lacking however. In this communication, we expressed stable low and high affinity forms of integrin CD11b A-domain and determined their binding isotherms and crystal structures. The low affinity form, generated by deleting an N-terminal extension extrinsic to the domain, did not bind to physiologic ligands, and crystallized in the closed conformation. The high affinity form was generated by either deleting or substituting an invariable C-terminal Ile(316), wedged into a hydrophobic socket in the closed form, but displaced from it in the open structure. Both mutants crystallized in the open conformation, and the Ile(316) --> Gly-modified integrin displayed high affinity. Structural differences between the low and high affinity forms were detected in solution. These data establish the structure-function correlates for the CD11b A-domain, and define a ligand-independent isoleucine-based allosteric switch intrinsic to this domain that controls its conformation and affinity.

Integrins are heterodimeric receptors that mediate vital cellcell and cell-matrix adhesive interactions (1). Integrins bind to physiologic ligands in a divalent cation-dependent manner and require a solvent-exposed acidic residue in their respective ligands for binding. Integrin interactions with physiologic ligands are dependent on inside-out signals, which switch integrins from a "low" to a high affinity state (2). This functional up-regulation is associated with conformational changes in the extracellular regions of integrins (2) that involve two major ligand binding sites, a von Willebrand Factor (vWF) 1 A-type domain (named A-or I-domain) present in the ␣ subunits of nine integrins, and an A-like domain embedded in all eight integrin ␤ subunits (3,4).
The integrin A-domain assumes a dinucleotide-binding fold (3,(5)(6)(7)(8), with a metal ion-dependent adhesion site (MIDAS) on one end (defined as the top of the domain) and is connected through the adjacent N and C termini on the opposite end (bottom) to the body of the integrin. MIDAS and its surrounding surface-exposed side chains form the binding site for several physiologic ligands (4, 6, 9 -12) and for the antagonist, neutrophil inhibitory factor, NIF (13,14). The A-domain has been crystallized in two conformations, "open" and "closed". In the open conformation, three noncharged residues in the protein directly coordinate the metal ion in MIDAS; a pseudoligand or ligand glutamate residue (3,6,15) completes metal coordination. In the closed form, the amphipathic C-terminal ␣7 helix is shifted upwards by 10 Å. This large shift is associated with a change in metal coordination, where one of the three coordinating residues, a threonine, is replaced with an aspartate, and a water molecule replaces the glutamate in completing the metal ion coordination sphere (16).
Crystal or NMR structures of four integrin A-domains (CD11b, CD11a, CD49a, and CD49b) have been reported to date (3,(5)(6)(7)(8). All, with the exception of integrin CD11b A-domain (11bA), were found only in the closed noncomplexed form, leading to the suggestion that the open form is a noninformative crystal artifact (17). However, recent studies provided evidence that the open and closed forms correspond to the high and low affinity states (6,15,18). The finding that one A-domain, CD49b, crystallized in the closed form in the absence of ligand, but in the open form in its presence, further suggested that the ligand may trigger or stabilize the high affinity state (15). Other evidence, however, suggest that integrins exist in the high affinity state even in the absence of ligand (2) and that ligand binding affinity in heterodimeric integrins can be effected in an allosteric manner (6,12,18,19). In this communication we describe the binding isotherms and structure of stable and homogeneous low and high affinity forms of the CD11b A domain. The low affinity form generated by removing an extrinsic N-terminal extension of the domain, crystallized in the closed conformation. The high affinity form was generated by unlocking an invariable C-terminal Ile 316 from a hydrophobic socket, which normally coordinates this residue in the closed conformation. The resulting domain crystallized in the open conformation and induced a high affinity state when introduced in the CD11b/CD18 receptor. Thus, an intrinsic isoleucine-based switch regulates affinity and shape shifting in the CD11b A-domain. We suggest that a similar mechanism may regulate integrin affinity in response to inside-out signals.

EXPERIMENTAL PROCEDURES
Reagents, Mutagenesis, and Protein Purification--Restriction and modification enzymes were purchased from New England Biolabs Inc. (Beverly, MA), Roche Molecular Biochemicals, or Life Technologies, Inc. The anti-CD11b mAbs 904 and 44a, the anti-CD18 mAb TS1/18, purified fibrinogen, recombinant CD54 (D1-5), NIF, and iC3b have been * This work was supported by National Institutes of Health Grants DK48549 and HL54227. 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.
Crystallization, Data Collection, and Structure Determination-Crystals were grown using 10 mg/ml protein and the hanging drop vapor diffusion method as described previously (6). 11bA 123-315 and 11bA Ile-3163Gly formed crystals in the presence of a reservoir solution containing 15% polyethylene glycol 8000, 0.1 M Tris-HCl, pH 8.2, 5 mM CaCl 2 , at room temperature. Crystals started to form within a week, grew to a typical size of 0.3 ϫ 0.05 ϫ 0.04 mm in 2 weeks, and belonged to the tetragonal space group P4 3 (Table I). 11bA 123-321 did not crystallize under the above conditions, but formed crystals at room temperature using 10% polyethylene glycol 4000, 0.1 M sodium acetate, pH 4.5, 5 mM MnCl 2 as precipitant; 11bA 123-315 and 11bA Ile-3163Gly did not form crystals in this buffer. The 11bA 123-321 crystals belong to space group P2 1 2 1 2 ( Table I). The two crystal forms are not related.
Single 11bA 123-315 and 11bA Ile-3163Gly crystals were used to collect, respectively, 2.3 and 3.0 Å resolution data sets, at 100 K, on beamline X12B of the National Synchrotron Light Source at Brookhaven National Laboratory using a charge-coupled device detector. A 2.6-Å resolution data set was collected from a single 11bA 123-321 crystal, using an in-house rotating anode generator/imaging plate system. Data were integrated and reduced with the HKL package (23). Structures were determined by molecular replacement. We used the refined 1.8-Å Mg 2ϩ structure (residues Asp 132 to Lys 315 , Protein Data Bank accession code 1ido) (3) as a starting model for 11bA  , and the resulting refined 11bA 123-315 model was then used as the starting model for 11bA Ile-3163Gly . The refined structure of F302W 11bA (residues Glu 131 to Gly 321 ), 2 which is similar to the Mn 2ϩ structure (Protein Data Bank accession code 1jlm) (16), was used as the starting model for 11bA  . The models were modified by deleting ions and water molecules. Rigid body refinement was initially used to improve each solution. Subsequent refinement rounds consisted of alternating cycles of torsion-angle dynamics and restrained individual isotropic B factor refinement protocols in XPLOR (24). In each case, 5% of the data were excluded from the refinement to monitor the free R-factor (24). After each refinement round, the models were inspected with (2F o Ϫ F c ) and (F o Ϫ Fc) electron density maps and modified using O (25). For 11bA 123-321 , NCS restraints were used in early refinement and gradually released as the resolution increased. Solvent and metal ions were added during later stages of model refinement based on peaks Ͼ 3 in (F o Ϫ F c ) difference maps, reasonable hydrogen bond distances, and refined temperature factors of less than 50 Å 2 . The final models comprise all non-hydrogen atoms of residues Asp 132 to Lys 315 for 11bA 123-315 and 11bA Ile-3163Gly , and residues Asp 132 to Gly 321 , for 11bA 123-321 . Data collection and refinement statistics are shown in Table I.
Protein Binding to 11bA-Ligand binding was measured using surface plasmon resonance (BIAcore AB, Uppsala, Sweden). The physiologic ligands iC3b, fibrinogen, CD54, the antagonist NIF, and the mAbs 904 or 44a were covalently coupled individually via primary amines to the dextran matrix of separate CM5 sensor chips. BSA immobilized in the same way was used as a control surface. 11bA 123-321 , 11bA 123-315 , or 11bA Ile-3163Gly domains were flowed over the chip at 5 l/min at different times. TBS (20 mM Tris-HCl, pH 8.0, 150 mM NaCl) with 2 mM MgCl 2 and 0.005% P20 (BIAcore AB) was used as the running buffer unless otherwise indicated. 1 M NaCl in 20 mM Tris-HCl, pH 8.0, was used to remove the bound proteins and to regenerate the surface. Binding was measured as a function of time, and binding isotherms were determined as described previously (26), after subtracting background binding to the BSA-coated chip.
Transfections and Analyses-COS M7 simian fibroblastoid cells at ϳ70% confluence were transfected with supercoiled cDNAs encoding full-length wild-type (WT) or CD11b Ile-3163Gly together with full-length CD18. Heterodimer formation and binding of iC3b-coated erythrocytes to WT and Ile-3163Gly CD11b/CD18 holoreceptors were carried out as described previously (13). Specific binding of iC3b to the holoreceptors was obtained by subtracting background binding to mock-transfected COS cells. Binding to Ile-3163Gly CD11b/CD18 was expressed as a percentage of binding to WT, after correcting for the degree of surface expression using binding of mAb 904 (13).

Generation of Stable High and Low Affinity Forms of 11bA in
Solution-In attempting to express stable and homogeneous forms of the low and high affinity forms of CD11bA, we took note of the following unexplained observations. First, integrin A-domains that bind to their respective physiologic ligands contain an N-terminal amino acid extension (16 amino acids N-terminal to Gly 127 in CD11b); domains lacking this extension bind physiologic ligands poorly (4,(27)(28)(29)(30). Second, one major difference in the crystal structure of the open and closed forms is the position of Ile 316 close to the C terminus. In the closed form, Ile 316 is part of the last turn of helix ␣7 and fits tightly into a conserved hydrophobic pocket (Fig. 1, a-d). In the open structure, Leu 312 replaces Ile 316 in this pocket (Fig. 1, c-f). Ile 316 is strictly conserved in all integrin A-domains (Fig. 1g). Third, some crystal and NMR structures of other integrin Adomains (3,(5)(6)(7) show structural flexibility in the C-terminal helical turn of ␣7, suggesting a potential physiologic regulatory role. We evaluated the structural and functional consequences of modifying Ile 316 in integrin 11bA that lacks the bulk of the extrinsic N-terminal extension.
The ligand binding properties of 11bA 123-321 and 11bA 123-315 were determined using surface plasmon resonance (6). 11bA  showed no binding to the activation-dependent physiologic ligands, complement iC3b, fibrinogen, and CD54 (ICAM-1) (Fig. 2a, c, and e) in the presence of MgCl 2 . In contrast, 11bA 123-315 displayed high affinity binding to all three ligands (Fig. 2, b, d, and f). The divalent cations Mn 2ϩ and Mg 2ϩ , but not Ca 2ϩ (each at 2 mM), supported the interaction of 11bA 123-315 with iC3b (Fig. 2b); no binding of iC3b to 11bA 123-315 was observed in the presence of 2 mM EDTA (Fig.  2b). Physiologic ligand binding to 11bA 123-321 in the presence of Mn 2ϩ , Mg 2ϩ , Ca 2ϩ , or EDTA was minimal (Fig. 2a). The structural differences observed between the low and high affinity forms did not require the presence of a physiologic ligand, as mAb 44a, which binds in the C-terminal region of 11bA (12), displayed a ϳ500-fold difference in affinity between the low and high affinity forms (Fig. 2, k and l). Both A-domain forms bound equally well to the activation-independent antagonist NIF (Fig. 2, g and h), and to mAb 904 (Fig. 2, i and j), indicating that the differences observed are not caused by variations in A-domain concentrations. These data establish the feasibility of expressing stable and homogeneous low and high affinity forms of 11bA and show that structural differences between the two can be detected in solution in the absence of physiologic ligands.
Crystal Structures of 11bA 123-321 and 11bA 123-315 -To provide a structural basis for the functional differences observed in solution, we determined the crystal structures of 11bA 123-321 and 11bA 123-315 . The crystal structure 11bA 123-321 was that of the closed conformation (Fig. 3, a, c, and e), with a Mn 2ϩ ion occupying MIDAS, as shown previously (16). On the other hand, 11bA 123-315 crystallized in the open form (Fig. 3, b, d,  and f). Although the crystallization conditions for both proteins were different, our observation that 11bA 123-321 did not form crystals under the 11bA 123-315 crystallization condition strongly suggests that the introduced C-terminal deletion played an essential role in generation of the open state. These data, combined with the ligand binding results, indicate that the high affinity state assumes the open conformation.
In the high affinity 11bA 123-315 open structure, MIDAS is occupied by a metal ion. We assigned a calcium ion at this position based on the following observations. First, crystals grew in the calcium-containing crystallization buffer; direct metal ion quantitation using the inductively coupled argon plasma method revealed predominantly calcium, with no other divalent cation present among 17 sought (including Mg 2ϩ , Mn 2ϩ , Ni 2ϩ , Zn 2ϩ , and Cd 2ϩ , not shown). Second, the typical octahedral coordination distinguishes it from solvent (31). Third, the average unrestrained distance from this metal ion to coordinating ligands (2.28 Å) is more close to Ca 2ϩ in the octahedral coordination (2.35 Å) than Mg 2ϩ (2.07 Å) or Mn 2ϩ (2.17 Å) (32). The refined B-factor for calcium is slightly lower than that of the surrounding side chains but is still within the margin of error for a 2.3-Å resolution. These data suggest that Ca 2ϩ may be coordinated in MIDAS in the open conformation, at least under the nonphysiologic crystallization conditions used here. An alternative interpretation of the above structural and functional data is that Ca 2ϩ does bind under physiologic conditions to the open form but exerts a suppressive effect on physiologic ligand binding. Measurements of Ca 2ϩ affinity to the high versus low affinity A-domain states may help in differentiating between these two possibilities. The above data also support previous findings that the nature of the metal ion per se is not sufficient to induce the open state (6,7).
An Ile 316 to Gly Substitution in 11bA Crystallizes in the Open Form and Generates High Affinity 11bA and Holoreceptor-Ile 316 is invariable in all integrin ␣ A-domains cloned to date (Fig. 1g). We determined the effect of removing the hydrophobic side chain "finger" of this isoleucine on affinity and structure of the A-domain. The 11bA Ile-3163Gly form exhibited "high affinity" (Fig. 4a). The same substitution created in the holoreceptor dramatically increased its ligand binding activity (Fig. 4b). The 11bA Ile-3163Gly crystal structure (Table I) was essentially identical to that of the open 11bA 123-315 form, including the predicted presence of calcium in MIDAS; the two structures can be superimposed onto each other with an root mean square deviation of 0.17 Å for main chain atoms. Taken together, these data indicate that an Ile 316 -based switch, intrinsic to this domain, acts allosterically to regulate its ligand binding affinity and topology. A conserved hydrophobic intramolecular socket (SILEN, socket for isoleucine), fastens the Ile 316 finger in the closed conformation; Ile 316 is replaced by Leu 312 at this site in the open structure (Fig. 1, a-f). SILEN is formed by the hydrophobic side chains of Ile 135 , Leu 164 , Ile 236 , and Tyr 267 both in the closed and open conformations (Fig. 1, c-f) and are either identical or conserved in all the other A-domains. In the   FIG. 3. Crystal structure of 11bA 123-321 and 11bA 123-315  CD11b/CD18 integrin, certain mutations that lie outside MI-DAS produce gain-of-function effects in the holoreceptor and are believed to act allosterically (12,18,19,33). Similar data were recently reported in the CD11a/CD18 heterodimer, while this manuscript was in preparation (33). All these studies were carried out in the holoreceptors and therefore did not determine whether allosteric regulation is intrinsic to the A-domain, since potential interdomain interactions and/or other quaternary effects in the holoreceptor may also be operative. While the transition from the closed to the open conformation undoubtedly involves multiple intradomain realignments, the data presented here suggest that interference with SILEN may be a key component in allosteric regulation. We note that mutations that up-regulate integrin function occur in or around SILEN. For example, substitution of the ␣1Ϫ␤B loop of CD11b with that of CD11a generates a constitutively active integrin (19). This region includes Leu 164 , one of the SILEN residues. Integrin activation also occurs in an Leu 164 to Phe substitution (18), which predictably makes SILEN smaller and therefore less accommodating to Ile 316 . Other activating mutations involving Glu 131 , Asp 132 , Lys 231 , and Phe 234 lie at the bottom of the structure, in close proximity to SILEN (18), and may thus exert their effects through interference with the proper coordination of the Ile 316 finger in SILEN. The inhibitory effect of certain mAbs with epitopes on the opposite side of MIDAS (e.g. mAb 44a, the epitope of which spans residues on the top of SILEN (12)) may similarly be explained through stabilization of the SILEN pocket. The binding site for the CD11a/CD18 inhibitor lovastatin (34) includes two of the four SILEN residues (Leu 132 and Tyr 257 in CD11a, which correspond to Ile 135 and Tyr 267 , respectively, in CD11b), and may therefore act by stabilizing the low affinity state. Interestingly, an Ile 306 to Ala substitution in CD11a (equivalent to Ile 316 in CD11b) was found to increase ligand binding affinity of the CD11a/CD18 heterodimer (33), suggesting that the present observations may extend to other A-domains.
Implications for Mechanisms of Activation by Inside-out Signaling-Integrins exist in low and high affinity states in the absence of ligand (2). However, the high affinity state of the ligand binding A-domain has so far been seen crystallographically only in complex with a pseudoligand (3,6) or ligand (15), leading to the suggestion that the ligand triggers or stabilizes the active state. The low and high affinity forms of 11bA reported here assume different structures even in solution and in the absence of ligand, as revealed by the differential binding of mAb 44a. Although not strictly proven, it is likely that the differences detected in solution correspond to the open and closed conformers. Generation of the high affinity 11bA in a stable form should now permit an examination of this question.
Previous studies in all integrin A-domains expressed to date have shown that the presence of a short N-terminal amino acid extension extrinsic to the A-domain itself allows the respective domain to bind to its physiologic ligands. Removal of this Nterminal extension or reducing it to four residues, as shown here, generates a low affinity state that assumes the closed form. These data suggest that the N-terminal extension plays a role in affinity regulation within the domain. The underlying structural basis for this effect is unknown, since none of the residues in the N-terminal extension are included in the threedimensional structures of this domain. It has been observed, however, that residues within this extension regulate ligand binding in A-domains. First, naturally occurring point mutations in this segment of the vWF A1 domain cause gain-offunction phenotypes in patients with type IIB vWf disease (35). This region also contains an activating mutation in CD11b (18). Second, structural data from the CD49b A-domain also show that three residues that extend beyond the ␣7 helix can pack into a crevice formed in part by N-terminal residues, bringing the N and C termini into close spatial proximity (5). Flexibility of the C-terminal residues in ␣7 has also been observed in the crystal (36) and NMR (8) structures of the CD11a A-domain. Based on these data, we speculate that the N-terminal extension in the isolated A-domain may loosen Ile 316 coordination into SILEN, thus allowing some molecules to exist in the high affinity open form. A similar mechanism may be operative in the holoreceptor; inside-out signals may through intramolecular repacking, free the N-terminal extension to access and therefore modify SILEN. Validation of such a mechanism will require the three-dimensional structure determination of a whole integrin.
The ability to express a stable form of the high affinity A-domain and to identify the structural basis of A-domain affinity switching should have broad biologic and pharmaceutical applications. Elucidation of the structures of other A-domains in their open conformation in the absence or presence of physiologic ligands may now be feasible. Also, integrin affinity is increased in many common diseases. High affinity forms of integrin A-domains could be of value therapeutically or utilized in high throughput screens to develop effective small molecule antagonists.
Acknowledgments-We thank Dr. Malcolm Capel and his staff at the Brookhaven National Laboratory for assistance in data collection. We acknowledge the Massachusetts General Hospital crystallography facility for providing in-house data collection equipment.
where T is a test set containing a randomly selected 5% of the reflections omitted from the refinement.