A novel adaptation of the integrin PSI domain revealed from its crystal structure.

Integrin beta-subunits contain an N-terminal PSI (for plexin-semaphorin-integrin) domain that contributes to integrin activation and harbors the PI(A) alloantigen associated with immune thrombocytopenias and susceptibility to sudden cardiac death. Here we report the crystal structure of PSI in the context of the crystallized alphaVbeta3 ectodomain. The integrin PSI forms a two-stranded antiparallel beta-sheet flanked by two short helices; its long interstrand loop houses Pl(A) and may face the EGF2 domain. The integrin PSI contains four cysteine pairs connected in a 1-4, 2-8, 3-6, 5-7 pattern. An unexpected feature of the structure is that the final, eighth cysteine is located C-terminal to the Ig-like hybrid domain and is thus separated by the hybrid domain from the other seven cysteines of PSI. This architecture may be relevant to the evolution of integrins and should help refine the current models of integrin activation.

Integrins are heterodimeric (␣␤) cell-matrix and cell-cell adhesion receptors, with each subunit containing a large extracellular domain, a single-pass transmembrane domain, and a short cytoplasmic tail (1). Integrins are often expressed on the cell surface in an inactive state (unable to bind physiologic ligands) but can be rapidly activated by intracellular signals (inside-out activation) (2). Once liganded, integrins cluster and initiate outside-in signals similar to classical receptors that modify cellular functions. The precise mechanism of integrin activation is incompletely understood.
Insights into structure-activity relationships in integrins were greatly aided by our determination of the crystal structure of the ectodomain from integrin ␣ V ␤ 3 alone and in complex with the prototypical ligand RGD (3,4). The structure has four domains in the ␣ V subunit: an N-terminal seven-bladed propeller followed by an Ig-like "thigh" domain and two co-linear ␤-sandwich domains calf-1 and calf-2. The ␤-subunit ectodomain consists of eight domains. The N-terminal PSI domain (5) is followed by an Ig-like "hybrid" domain (with the ligandbinding vWFA domain (␤A) emerging from the loop connecting its two ␤-sheets). The hybrid domain is then connected to four EGF 1 -like domains and a novel ␤-tail domain (␤TD). An unexpected feature of the crystal structure is that ␣ V ␤ 3 is genuflexed at the ␣ V and ␤ 3 "knees" such that the head abuts the legs. Current models suggest that a straightening of the genu is required for physiologic ligand binding (in the switch-blade model of activation (6)) or for ligand-induced outside-in signaling (in the deadbolt model (2)). A better understanding of the basis of integrin activation requires structure determination of the activation-sensitive domains PSI, EGF1, and EGF2 (reviewed in Ref. 7), which are all located in the genuflexed ␤-subunit but were not resolved in the published structure. Although some features of the PSI domain, including two short helices, were visible in our electron density maps, our main chain tracing was inconsistent with published cysteine pairing (8) and predictions (5) especially between Cys 5 and Cys 435 . Since several cysteines cluster close together, we were initially unable to build the domain with certainty. The density for EGF1 and EGF2 was even less well defined.
The ϳ50-amino acid PSI domain, first recognized based on primary sequence alignments, is present in one or more copies in more than 500 proteins (see smart.embl-heidelberg.de/smart/ get_members.pl?WHATϭNRDB_COUNT&NAMEϭPSI). It is most commonly found in plexins, semaphorins, and integrins, glycoproteins that mediate cell growth, migration, and differentiation. Two recently determined structures of semaphorin 4D (SEMA4D) (9) and the plexin MET (10), each containing a PSI domain, have now allowed us to build the integrin PSI domain into our ␣ V ␤ 3 maps without ambiguity. The salient and unexpected features of this structure as it relates to integrin architecture, activation, and disease are presented and discussed here.

MATERIALS AND METHODS
The original ␣ V ␤ 3 structure was determined using a combined phasing approach, with a lutetium derivative data set used for multiple anomalous diffraction (MAD) phasing and a platinum data set used for single isomorphous replacement with anomalous scattering (SIRAS) (3). To solve the PSI structure, three maps were calculated: a MADphased map, a SIRAS-phased map based on the platinum derivative, and another SIRAS-phased map based on the lutetium data set collected at the peak wavelength. The three electron density maps were averaged using the RAVE package (11) and displayed using program O (12). The averaged map showed clear density for a large portion of the PSI domain, including two helices and the central tryptophan residue (Trp 25 ), allowing us to trace the PSI chain. The new model for ␣ V ␤ 3 , including the PSI domain, was refined using the existing 3.1-Å data set (3) and the program CNS (13). This resulted in R-factors of 29.0 (work set) and 35.6% (free set) and good model geometry.

RESULTS AND DISCUSSION
Structure Determination of the Integrin PSI-Our initial chain trace of the PSI did not agree with the published cysteine pairing, especially between Cys 5 and Cys 435 . However, the * This work was supported by National Institutes of Health Grants HL70219 and DK48549. 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 recent structure determinations of two PSI domains (9, 10), allowed us to establish the ␤ 3 PSI fold without ambiguity (Fig.  1A). We fitted the cysteine cores of SEMA4D and MET structures into our maps, and this allowed us to establish points of reference for the PSI domain fold. It became clear that Cys 13 , and not the published Cys 5 , pairs with Cys 435 (Fig. 1B). It also became clear that the region around Cys 435 substitutes for a portion of the PSI domain seen in the SEMA4D and MET structures, and thus Cys 435 is integral to PSI. We were able to build residues 1-50 of the PSI domain by using the corrected disulfide bond pattern. The linker between PSI and hybrid domains (residues 51-53) is not well defined in our density maps and was not included. In our efforts to connect the PSI domain with the hybrid domain, we also noticed that the Nterminal strand of the hybrid domain was out of register by one residue. We have corrected this error, which has no bearing on previously published interpretations of ␣ V ␤ 3 structure and function. The location of the PSI domain in relation to the rest of the ␣ V ␤ 3 ectodomain is shown in Fig. 1C.
Architecture of the Integrin PSI and Its Relationship to the Hybrid Domain-The integrin PSI domain forms a twostranded antiparallel ␤-sheet, with two flanking short helices, connected by disulfides to the central sheet, and an N-terminal segment that may also be helical (Fig. 1B). The small hydrophobic core of the domain is formed by the highly conserved side chain of Trp 25 . The core integrin PSI structure can be largely superimposed onto those of SEMA4D and MET, including all three conserved cysteine pairs (Fig. 2, A and B). The fourth disulfide bridge is lacking in semaphorins and has a somewhat different conformation in plexins and integrins, perhaps a result of an additional C-terminal ␣-helix in the integrin PSI (Fig. 2, A and B). A key difference between the integrin PSI domain and the semaphorin and plexin PSI domains is a distinctively longer interstrand AB loop (Figs. 1B and 2B). While some of the side chains in this loop are solvent exposed, others may face the EGF2 domain, suggesting potential interactions. It is also apparent that the overall structures of the three PSI domains are quite different in the C-terminal half of the domain (Fig. 2, A and B). This suggests that this portion of PSI has a function defined by its specific structural context. Alignment of the PSI domains of SEMA4D, MET, and ␣ V ␤ 3 reveals that Cys 435 of ␤ 3 , located at the C terminus of the hybrid domain, is an integral part of the integrin PSI ( Fig. 2A). Thus the hybrid domain is an insertion into the last loop of PSI. This unexpected architecture through which the PSI and hybrid domains are connected may explain the incorrect predictions of cysteine pairing of the PSI domain in integrins (5,8). The hybrid AЈB loop contains an arginine (Arg 93 ), conserved between integrin ␤-chains, that contributes to the interface with the PSI domain.
Functional and Disease Implications-The integrin PSI domain contributes to integrin activation as evidenced by binding of activation-sensitive monoclonal antibodies such as AP5 (which binds the N-terminal six amino acids of PSI of ␤ 3 (14)) and activating amino acid substitutions (15,16) such as an alanine substitution of the Cys 435 , which is immediately Nterminal to EGF1 (Fig. 1B) (reviewed in Ref. 7). The PSI domain also harbors clinically important alloantigens that cause immune thrombocytopenias. A leucine-proline dimorphism at amino acid 33 (Leu 33 3 Pro) known as autoantigen Pl(A) is the one most frequently implicated in syndromes of immune-mediated platelet destruction, particularly neonatal alloimmune thrombocytopenia and post-transfusion purpura. Substitution of the common Pl A1 alloantigen at Leu 33 with Pro (Pl A2 allele found in 15% of Caucasians) may also predispose to arterial thrombosis frequently manifested as acute coronary events that can lead to sudden death at a relatively young age (17). It is suspected that this predisposition is secondary to enhanced activation of platelet integrins (18), enhanced thrombin generation, and impaired antithrombotic action of aspirin at the site A Novel Adaptation of the Integrin PSI Domain 40253 of microvascular injury (19). It has also been reported that the Pl A2 allele is a risk factor for acute renal allograft rejection (20). The PSI domain also contributes to the binding of certain drugs that trigger drug-dependent antibodies to ␣ IIb ␤ 3 , precipitating thrombocytopenia (21). The present crystallographic studies clarify a number of observations about structure-activity relationships in integrins. First, they establish the correct pairing of cysteines and domain boundaries for the integrin PSI. This will allow a reinterpretation of many functional studies and will be invaluable in devising new ones to evaluate the salient features of the structure. Second, the structure reveals that the N terminus of the activation-sensitive AP5 epitope is solvent-exposed (Fig.  1B), which may explain the ability of the AP5 monoclonal antibody to bind the ectodomain in solution (not shown). The Cys 13 -Cys 435 disulfide bridge contributes to the PSI/hybrid interface and likely helps to restrict movement of the PSI with respect to the hybrid domain. The activating effect of the Cys 435 to alanine substitution is expected to make this interface more flexible, and it may also allosterically alter putative contacts between PSI and EGF1/2, thus accounting for the activating nature of this mutation. Interruption of the adjacent Cys 406 -Cys 433 disulfide bond in the hybrid domain may have a similar functional outcome. Third, Leu 33 is located in a hydrophobic segment of the distinctively long AB loop (between strands A and B of the PSI domain, Fig. 1B). Its replacement with Pro, a conformationally restricted residue, may alter the structure of this loop leading to an autoantibody response and immune thrombocytopenia. Elucidation of the three-dimensional structure of the Pl A1 alloantigen may be useful in averting the autoimmune consumption of platelets through drug design. Of note is that Leu 33 is located in the AB loop that likely faces EGF2; its replacement with Pro may also alter this interface and promote integrin activation. Fourth, three quinine-dependent antibodies known to cause the precipitous destruction of platelets if a patient is exposed to the drug require a conformational epitope in the PSI, which includes Ala 50 (Fig. 1B) and Arg 62 and Asp 66 in the first strand of the hybrid domain of ␤ 3 .
The present structure reveals that all three residues are located on the same side of the integrin. Finally, it is now apparent that two insertions have contributed to the current architecture of the integrin ␤-subunit, one in the last loop of PSI and the second in between the two sheets of the hybrid domain, setting the stage both for formation of the integrin heterodimer and its regulated ability to bind physiologic ligands.