Ligand- and cation-induced structural alterations of the leukocyte integrin LFA-1

In αI integrins, including leukocyte function–associated antigen 1 (LFA-1), ligand-binding function is delegated to the αI domain, requiring extra steps in the relay of signals that activate ligand binding and coordinate it with cytoplasmic signals. Crystal structures reveal great variation in orientation between the αI domain and the remainder of the integrin head. Here, we investigated the mechanisms involved in signal relay to the αI domain, including whether binding of the ligand intercellular adhesion molecule-1 (ICAM-1) to the αI domain is linked to headpiece opening and engenders a preferred αI domain orientation. Using small-angle X-ray scattering and negative-stain EM, we define structures of ICAM-1, LFA-1, and their complex, and the effect of activation by Mn2+. Headpiece opening was substantially stabilized by substitution of Mg2+ with Mn2+ and became complete upon ICAM-1 addition. These agents stabilized αI-headpiece orientation, resulting in a well-defined orientation of ICAM-1 such that its tandem Ig-like domains pointed in the opposite direction from the β-subunit leg of LFA-1. Mutations in the integrin βI domain α1/α1′ helix stabilizing either the open or the closed βI-domain conformation indicated that α1/α1′ helix movements are linked to ICAM-1 binding by the αI domain and to the extended-open conformation of the ectodomain. The LFA-1–ICAM-1 orientation described here with ICAM-1 pointing anti-parallel to the LFA-1 β-subunit leg is the same orientation that would be stabilized by tensile force transmitted between the ligand and the actin cytoskeleton and is consistent with the cytoskeletal force model of integrin activation.

Lymphocyte function-associated antigen 1 (LFA-1, 3 integrin ␣L␤2) is important in leukocyte diapedesis, migration within tissues, and recognition processes requiring cell-cell adhesion. Two other integrins with the same ␤2-subunit, macrophage antigen 1 (Mac-1, ␣M␤2) and ␣X␤2, function as complement and danger receptors and are primarily expressed on myeloid cells (1,2). Mutations in their common ␤2 integrin subunit in leukocyte adhesion deficiency result in life-threatening bacterial infections. LFA-1 binds intercellular adhesion molecules (ICAMs), a subfamily of cell surface molecules that contain tandem immunoglobulin-like domains. ␤2 integrins bind ligands through the ␣I domain, which is inserted in the ␤-propeller domain in the ␣-subunit (Fig. 1, A-C). The highaffinity, open conformation of the ␣I domain is stabilized by binding of an internal ligand to a binding site at the interface between the ␤-propeller and ␤I domains in the integrin head (3) (Fig. 1C). The ␣I and ␤I domains are structurally homologous and undergo similar conformational change between low-affinity closed and high-affinity open conformations (4,5). Integrins that lack ␣I domains (␣I-less integrins) bind external ligands at the same site to which ␣I-integrins bind the internal ligand. How signals are transmitted through integrin ␣L␤2 from the actin cytoskeleton to ICAM-1 on the surface of another cell and the structural characteristics of relay of activation between the ␣I and ␤I domains are major topics of this paper.
Integrins have three overall conformational states ( Fig. 1, A-C), as shown with both ␣I-less integrins and the ␣I integrin ␣X␤2 (2). Visualization of ␣X␤2 in negative-stain EM bound to Fab fragments of allosteric, conformation-specific, activating or inhibitory antibodies, together with the effects of these Fabs on cell adhesion to the complement fragment iC3b, showed that the extended-open conformation of ␣X␤2 is adhesive, whereas the bent-closed and extended-closed conformations are not (6,7). In negative-stain EM, the conformation of the headpiece can be assigned based on whether the ␤-subunit hybrid domain is swung away from or toward the ␣-subunit (Fig. 1, B and C); however, ␣I-domain conformation can only be inferred based on ligand-binding activity. One of the most curious features of ␣I integrins is how allostery can be relayed between the ␤I and ␣I domains despite evidence for marked flexibility in the ␣I domain (3, 8 -11). One of the questions we address here is whether the ␣I domain adopts a defined orientation when it couples to the ␤I domain through its internal ligand during allostery relay. Tensile force is exerted on the cytoplasmic domain of LFA-1 when it mediates lymphocyte migration on ICAM-1 substrates (12). Thus, the specific orien- . The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article contains Figs. S1-S3. 1 To whom correspondence may be addressed. E-mail: msen2@central.
cro ARTICLE tation of the ␣I domain should be compatible with the orientation for force transmission through the integrin-ligand complex between the actin cytoskeleton and the substrate.
Whereas the ␤-propeller and ␤I domains have an extensive interface and highly stable orientation with respect to one another and may be termed a "platform," crystal structures show that the ␣I domain can differ in orientation by up to 150°w ith respect to this platform (3,8,10) (Fig. 1, D-F). In each case, orientation is stabilized by specific contacts of the ␣I domain with other molecules in the crystal lattice, whereas in other cases, ␣I domain density is lacking, suggesting that it is flexible. Flexibility is shown schematically in Fig. 1 (A and B) as three different positions of the ␣I domain. In the first ␣X␤2 ectodomain crystal structure, all domains in the ectodomain were visualized, except most independent molecules in crystal lattices lacked density for the ␣I domain (8). The visualized ␣I domain was in the closed conformation, and crystal contacts stabilized a tilt toward the integrin ␤-subunit (Fig. 1F). In another ␣X␤2 ectodomain structure, the crystal lattice fortuitously stabilized the ␣I domain in the open conformation and in an orientation midway between the integrin ␣and ␤-subunits (3) (Fig. 1E). The internal ligand at the C terminus of the ␣I domain was reshaped relative to the closed conformation and bound to its pocket between the ␤-propeller and ␤I domains (Fig. 1C). Furthermore, aside from internal ligand binding, there was little contact between the open ␣I domain and the remainder of the integrin head, suggesting that the ␣I domain would be capable of substantial tilting and rotation relative to the ␤-propeller/␤I-domain platform. In a crystal structure of the ␣L␤2 headpiece, the ␣I domain was in the closed conformation, and crystal lattice contacts stabilized an extreme tilt toward the ␣-subunit (10) (Fig. 1D). These findings raise the question of whether the ␣I domain has a well-defined orientation, either in the closed conformation when not bound to the platform through the internal ligand or in the open conformation when bound through its internal ligand to the platform, and illustrate the importance of the use of techniques orthogonal to crystallography, such as EM and small angle X-ray scattering in solution (SAXS), as in the current study.
Our knowledge about how ligand binding and divalent metal ion binding influence the conformation of the headpiece in ␣I integrins is incomplete. Complexes thus far visualized are with isolated ␣I domains, with the ligand iC3b bound to the isolated ␣X I domain, with the ligand iC3b bound to the ␣M␤2 headpiece, or with the ligand iC3b bound to the ␣X␤2 ectodomain, which was prestabilized in the extended-open conformation with a combination of extension-stabilizing and opening-stabilizing Fabs (2,13,30). The headpiece conformation of ␣L␤2 has not previously been visualized in the presence of a biological ligand.
␣I and ␤I domains contain a Mg 2ϩ ion held in a metal iondependent adhesion site (MIDAS) that coordinates to external or internal ligands (2). Additionally, the ␤I domain MIDAS is flanked by an adjacent to MIDAS (ADMIDAS) that binds Ca 2ϩ . Mn 2ϩ activates integrins by competing with Ca 2ϩ at the ADMIDAS (14,15). Mn 2ϩ along with soluble ICAM-1 also enhances binding of antibodies to LFA-1 (15, 16) that stabilize extension and headpiece opening (6, 7). However, the effect of Mn 2ϩ on ␤2 integrin conformation has not been addressed with structural studies.
Here, EM, SAXS, and mutational studies provide views of the headpiece of LFA-1 and its complex with ICAM-1 that are complementary to and extend previous structural studies on ␤2 integrins and address the open questions described above.

Purified proteins and structural methods
We studied a headpiece fragment of LFA-1 containing the ␣I, ␤-propeller, and thigh domains in the ␣L-subunit and the ␤I, hybrid, PSI, and I-EGF1 domains in the ␤-subunit (schematized in color in Fig. 1 (A-C) and shown in ribbon representations in the same colors in Fig. 1, D-F). The LFA-1 headpiece was expressed in HEK293S GnTI Ϫ/Ϫ cells with C-terminal ACID-BASE coiled-coil peptides on the ␣and ␤-subunits, respectively, followed by purification tags. After affinity purification using nickel-nitrilotriacetic acid-Sepharose, coiled-coil and tags were removed by 3C protease digestion, and the headpiece was purified to homogeneity by gel filtration. SDS-PAGE of gel filtration fractions showed bands corresponding only to the LFA-1 ␣Land ␤2-subunits ( Fig. 2A).
The ICAM-1 ectodomain contains five Ig-like domains (D1-D5). To facilitate complex formation with LFA-1, we utilized D1-D5 of Hi3-ICAM-1, which contains five substitutions in the binding site for LFA-1 in D1 that increase affinity 20-fold (17,18). Hi3-ICAM-1 with a C-terminal His tag was made in HEK293S GnTI Ϫ/Ϫ cells and purified using nickel-nitrilotriacetic acid-Sepharose, anion exchange, and gel filtration. ICAM-1 ran as a single band in SDS-PAGE (Fig. 2B). Multiangle light scattering showed a glycoprotein molecular mass of M r 66,400 (Fig. 2C). Thus, compared with the protein mass of M r 50,600, glycans contributed M r 15,800. Assuming all nine potential N-linked glycosylation sites were utilized, this corresponds to M r 1,800 per N-glycan, or 2 GlcNAc and 8.6 mannose residues per N-glycan, close to the expectation for high-mannose glycans.
Both the LFA-1 headpiece and ICAM-1 gave single, monomeric peaks in gel filtration (Fig. 2D). When the headpiece was incubated with an excess of Hi3-ICAM-1 in 1 mM Mn 2ϩ , 0.2 mM Ca 2ϩ , an early-eluting complex peak was formed, and the peak at the position of the free headpiece was depleted (Fig. 2D).
We characterized the LFA-1 headpiece, the ICAM-1 ectodomain (Ig-like domains 1-5), and their complex by both EM (Fig.  3) and SAXS (Fig. 4). We describe the results by biological unit, not by technique. Thus, we first describe the structure using both techniques of the headpiece and then ICAM-1 and then the LFA-1 headpiece-ICAM-1 complex. The LFA-1 headpiece was characterized in 5 mM Mg 2ϩ , 1 mM Ca 2ϩ and also, to promote activation, in 1 mM Mn 2ϩ , 0.2 mM Ca 2ϩ . To maximize complex formation, we used 1 mM Mn 2ϩ , 0.2 mM Ca 2ϩ for the LFA-1 headpiece complex with ICAM-1 and for comparison with ICAM-1 also used 1 mM Mn 2ϩ , 0.2 mM Ca 2ϩ . In EM, 6,800 -10,600 particles were subjected to multireference alignment and averaging into 20 or 50 classes (Figs. S1 and S2). Representative class averages are shown in Fig. 3. SAXS data sets extended to q values ranging from 0.61 to 0.15 Å Ϫ1 (Table S1).

Orientation of LFA-1 bound to ICAM-1
Guinier and Kratky plots are shown in Fig. S3. Multiple ab initio molecular models constructed from the SAXS data were superimposed and averaged to calculate molecular envelopes (Fig. 4).

The LFA-1 headpiece
In Mg 2ϩ /Ca 2ϩ , EM class averages of the LFA-1 headpiece predominantly exhibited the closed-headpiece conformation (Fig. 3A, panels 1-3, representing 70% of all particles). Clear densities were present for the ␣I, ␤-propeller, and thigh domains in the ␣L-subunit and the ␤I, hybrid, and PSI/I-EGF1 unit in the ␤-subunit (Fig. 3A, panel 1). Only 4% of particles showed the open conformation, and 18% of particles oriented on their sides on grids, preventing classification as open or closed (Fig. 3A, panel 4). Side views contained clear density for the ␣I and ␤-propeller domains in the ␣L-subunit and two leg domains that appear to correspond to the hybrid and PSI/I-EGF1 unit. The presence of at least four distinct densities in side views demonstrated that side views contained both subunits, because each subunit contained only three distinct densities. LFA-1 appears to have a greater tendency than other studied integrins, such as ␣X␤2, to electrostatically adsorb to grids on its side, as seen previously with the complete LFA-1 ectodomain (6).
In activating conditions in Mn 2ϩ /Ca 2ϩ , 38% of particles exhibited the open-headpiece conformation (Fig. 3B, panels 2 and 3). Some particles remained closed (4%; Fig. 3B, panel 1). Furthermore, a higher proportion of particles than in Mg 2ϩ oriented on their sides (53%; Fig. 3B, panel 4). Additionally, a large proportion of particles adopted an oblique orientation, intermediate between the view on the side and the view with the    3)). We speculate that in Mn 2ϩ , the ␣I domain orients more out of the headpiece plane than in Mg 2ϩ , giving rise to oblique orientations and a higher proportion of side orientations on EM grids. SAXS also showed that in Mg 2ϩ /Ca 2ϩ , the LFA-1 headpiece was closed (Fig. 4, A-D). Fits of closed LFA-1 headpiece models to the SAXS envelope in Mg 2ϩ showed that the middle position of the ␣I domain ( Fig. 1E) fit better ( 2 ϭ 1.8) than ␣and ␤-tilt positions ( 2 ϭ 2.6 -2.8), and manual fitting of the ␣I domain was only slightly better than the middle tilt ( 2 ϭ 1.4).
In Mn 2ϩ /Ca 2ϩ , the headpiece was substantially more open, with an increase in the proportion of interatomic distances in a range from 75 to 140 Å and an increase in radius of gyration (R g ) to 5.5 nm (Fig. 4, E-H) compared with values in Mg 2ϩ /Ca 2ϩ (Fig. 4, A-D). Correspondingly, the ab initio molecular envelope of the headpiece was broader in Mn 2ϩ than in Mg 2ϩ (Fig. 4, compare H and D). A model of the open headpiece of LFA-1, made using the swung-out orientation of the hybrid domain from the open crystal structure of integrin ␣IIb␤3 (4), was used to fit the SAXS data in Mn 2ϩ . The fit was reasonable, although the model showed greater separation between the upper integrin legs than implied by the distance distribution data or the ab initio molecular envelope (Fig. 4, F and H). Fits to the open headpiece with different ␣I domain orientations showed reasonable fits to a manually docked ␣I domain (Fig. 4F, 2 ϭ 2.8), reasonable fits to middle and ␤-tilts and to an orientation optimized for complexes with ICAM-1 described below (3.1-3.2), and a poorer fit to an ␣I ␣-tilt ( 2 ϭ 3.6). As the EM data suggested both closed-and open-headpiece conformations were present in Mn 2ϩ , we used nonlinear regression analysis sampling of the back-calculated P(r) graphs of the open and closed LFA-1 headpiece models to approximate the P(r) graph generated from the experimental Mn 2ϩ SAXS data. The best approximation, with a correlation coefficient of 0.9994, was achieved using an ensemble containing 24% closed-headpiece and 76% open-headpiece conformations (Fig. 4F). The results suggest that Mn 2ϩ by itself stabilizes the open-headpiece relative to the closed-headpiece conformation but is not sufficient to induce a complete shift to the open conformation.
We also used EM to examine ␣I domain orientations. Filtered molecular envelopes at 20 Å resolution from molecular models with three alternative ␣I domain orientations ( Fig. 1, D-F) and either the closed or open headpiece were used to calculate regularly spaced projections at 2°intervals and were cross-correlated with EM class averages using SPIDER (19) (Fig. 5). In Mg 2ϩ , differences in ␣I domain position were apparent in class averages, and class averages were found that crosscorrelated best with ␣I domains in ␣-tilt, middle, and ␤-tilt positions (Fig. 5A, panels 1, 2, and 3, respectively). By contrast, in Mn 2ϩ , the best cross-correlation with all representative class averages was with the ␤-tilted ␣I domain (Fig. 5, B (panels 1 and 2) and C (panels 1 and 2)). Thus, it appears that in Mn 2ϩ , the ␣I domain is more constrained than in Mg 2ϩ /Ca 2ϩ and is more proximal to the ␤I domain, to which it couples by binding of the internal ligand in the ␣I domain to the ␤I domain (Fig. 1C).

Orientation of LFA-1 bound to ICAM-1
domains were often present. Many class averages, including the most populous one (Fig. 3C, panel 1), displayed a bend about two-fifths along the D1-D5 ectodomain axis.
SAXS of Hi3-ICAM-1 showed an R g of 5.8 nm and an extended molecular envelope with two kinks (Fig. 4, M-P). ICAM-1 has been crystallized as either D1-D2 or D3-D5 fragments (20 -22); however, D1-D5 of ICAM-5 are 50% identical in amino acid sequence to ICAM-1, and superposition of ICAM-1 D1-D2 and D3-D5 fragments on a D1-D4 crystal structure of ICAM-5 (23) allowed us to build an ICAM D1-D5 model. The model fit into the molecular envelope well, at a low 2 value ϭ 1.2 for the experimental and back-calculated solution scattering data (Fig. 4, M-P). The largest kink observed in both ICAM-1 and ICAM-5 crystal structures and in an early EM study that mapped antibody binding sites is at the D3-D4 junction (21,23,24). The kink helps define the orientation of D1-D5 in the SAXS envelope (Fig. 4, O and P); flipping the end-to-end orientation resulted in protrusion of D1 from the SAXS envelope and a poor fit. The overall kinked, S-shaped SAXS structure is in agreement with the bent structure observed here in EM (Fig. 3C) and in early EM structures (24,25).

The LFA-1-ICAM-1 complex
EM of the LFA-1 headpiece-Hi3-ICAM-1 complex peak from gel filtration in Mn 2ϩ /Ca 2ϩ (Fig. 2D) showed class averages corresponding to ICAM-1-LFA-1 complexes, headpiece alone, and ICAM-1 alone (Fig. 3D). Dissociation of receptorligand complexes is a common problem in negative-stain EM and may relate to the low protein concentrations required by the technique. Remarkably, all three class averages of the complex showed nearly identical ICAM-1-LFA-1 orientations (Fig.

Orientation of LFA-1 bound to ICAM-1
3D, panels 1-3). The rod-shaped ICAM-1 molecule bound to the face of the ␣I domain adjacent to the ␤-subunit rather than the face of the ␣I domain most distal from the head. Crystal structures show that the LFA-1 ␣I domain binds to an edge ␤-strand in D1 of ICAM-1 (5), as shown in the ribbon diagrams in Fig. 3 (F and G). Thus, the ␣I domain must be tilted toward the ␤-subunit. This inference is independently supported by cross-correlations of the headpiece moiety from complexes with Hi3-ICAM-1, which correlate better with the ␤-tilt position than the other two positions tested (Fig. 5C, panels 3 and  4). Density for individual domains of ICAM-1 was not evident in complexes, but the length of the rod of 16 nm suggests that D1-D4 were visualized, that D4 was fainter, and that D5 was averaged out. ICAM-1 D1-D4 extended along a line that was almost anti-parallel to the line made by the three densities in the integrin ␤-subunit that correspond to the ␤I domain, the hybrid domain, and the PSI plus I-EGF1 domain that attach to the same end of the hybrid domain (compare Fig. 3D (panels 1-3) with Fig. 1C). In headpiece complexes with ICAM-1, the hybrid and PSI/I-EGF1 domains were clearly swung away from the integrin ␣-subunit in the open conformation (Fig. 3D, panels [1][2][3]. For SAXS, the LFA-1-ICAM-1 complex in 1 mM MnCl 2 and 0.2 mM CaCl 2 was isolated by gel filtration at a ϳ20-fold higher concentration than used for EM and further concentrated to 3.2 mg/ml for data collection. The R g of the complex was 8.4 nm. Averaging of ab initio GASBOR models showed a molecular envelope with clear density for the ICAM-1 and LFA-1 headpiece moieties (Fig. 4, I-L). The open-headpiece moiety fit the envelope well with a solvent-filled gap between the knee-proximal ␣-subunit thigh and ␤-subunit PSI and I-EGF1 domains (Fig. 4L). Comparison with the molecular envelope of the headpiece alone in Mn 2ϩ (Fig. 4H), which appeared to correspond to an ensemble of closed and open conformations, suggested that binding to ICAM-1 completely shifted the headpiece into the open conformation. A thin, elongated protrusion of the molec-ular envelope clearly corresponded to ICAM-1. Using the model of ICAM-1 D1-D5 described above, the structure of the LFA-1 ␣I domain bound to ICAM-1 D1-D2 (5), and the model of the LFA-1 headpiece in the open conformation, we constructed a model of the LFA-1-ICAM-1 complex. The model fit the molecular envelope well, and its back-calculated scattering showed excellent fit to the experimental SAXS data with a 2 value of ϭ 1.6 (Fig. 4J). In SAXS, the long axis of ICAM-1 was anti-parallel to the integrin ␤-subunit (Fig. 4L), just as seen in EM. The ICAM-1-integrin binding orientations seen in SAXS and EM are remarkably similar, as shown in the comparison in Fig. 3 (F and G). These results establish the overall structure of LFA-1 bound to its physiological ligand, ICAM-1, and that binding results in a well-defined orientation of ICAM-1 with respect to the integrin. Moreover, the particular orientation found aligns ICAM-1 anti-parallel to the LFA-1 ␤-subunit leg, as would occur in the presence of tensile force exerted physiologically when actin retrograde flow exerts force on the integrin that is resisted by ICAM-1 on the surface of another cell.

Mutational studies
The above experiments suggested that binding of LFA-1 to ICAM-1 is associated with opening of the integrin headpiece. We used mutations to further support this conclusion and to relate it to movement of residues in the ␤I domain ␣1 and ␣1Ј helices that alter position upon opening of the ␤ 2 ␤I domain (3). During opening, axial displacement of the ␣1 and ␣1Ј helices is associated with displacement of Val 124 by Leu 127 (Fig. 6A), as shown in a structure of partially open integrin ␣X␤2 (3). Val 124 is more buried in the closed than partially open conformation, whereas Leu 127 is more buried in the partially open than closed conformations (Fig. 6A); therefore, the mutations V124A and L127A are predicted to stabilize the open and closed conformations, respectively. We also mutated residues Leu 132 and Leu 135 to Ala; however, their effects were difficult to predict in advance because both residues are similarly buried in the partially open

Orientation of LFA-1 bound to ICAM-1
and closed conformations (Fig. 6A). Studies on integrin ␣IIb␤3 showed that movements in the region corresponding to residues 124 and 127 occur earlier in the opening process than movements in the region corresponding to residues 132 and 135 (26); thus, the positions of residues 132 and 135 in partially open ␤ 2 integrins are not predictive of their positions in the open state. We hypothesized that Leu 132 and Leu 135 would move during complete headpiece opening and that their mutation might differentially stabilize the closed and open conformations. Mutant ␤2-subunits were co-transfected with ␣Lor ␣M-subunits to study ligand-binding function of LFA-1 and Mac-1, respectively, in HEK293T cells. All mutants were expressed at levels comparable with those of WT (Fig. 6, B  and C).

Orientation of LFA-1 bound to ICAM-1
adhesion to fibrinogen and ICAM-1 substrates (Fig. 6, F-I). L132A strongly activated LFA-1 transfectant adhesion to ICAM-1 and had much less effect on binding to soluble ICAM-1-Fc, which reflected a trend shared with the V124A mutation (Fig. 6, D and E). L132A similarly activated Mac-1 function in all three assays (Fig. 6, F-I). In contrast, the L135A mutation inhibited manganese-stimulated binding and adhesiveness of LFA-1 and Mac-1 in all assays (Fig. 6, D-I).
We correlated these mutational effects on ligand binding function of the LFA-1 and Mac-1 ␣I domains with exposure of epitopes that measure ␤ 2 integrin extension (Kim127 epitope) and ␤I domain opening (m24 and MEM148 epitopes). Activating mutations V124A and L132A greatly increased LFA-1 and Mac-1 extension measured with the Kim127 epitope in both Mg 2ϩ and Mn 2ϩ (Fig. 6, J and K). V124A and L132A also greatly enhanced headpiece opening measured with m24 and MEM148 in both Mg 2ϩ and Mn 2ϩ , with the MEM148 epitope showing less exposure than m124 by the L132A mutation in Mg 2ϩ but comparable exposure in Mn 2ϩ (Fig. 6, J and K). L127A and L135A mutations decreased exposure of all three epitopes in both Mg 2ϩ and Mn 2ϩ and in both LFA-1 and Mac-1 (Fig. 6, J and K). Thus, mutations demonstrate that ligand binding to the ␣I domain, including ICAM-1 binding to the ␣I domain of LFA-1, is regulated by movement of residues in the ␤I domain ␣1 and ␣1Ј helices and is correlated with ␤I domain opening and integrin extension.

Discussion
Here, we have used techniques orthogonal to crystallography to investigate how the ␣I domain of LFA-1 is linked to the other two domains in the integrin head, the influence of the activating metal ion Mn 2ϩ and the ligand ICAM-1 on the orientation of the ␣I domain and the conformation of the headpiece. Moreover, we investigated the relevance of movements of the ␣1 and ␣1Ј helices in the ␤I domain to these linkages between the ␣I domain and the conformation of the headpiece and the ectodomain. Perhaps most strikingly, we have visualized how the headpiece of integrin LFA-1 binds to its physiological ligand, ICAM-1, and demonstrated a remarkably well-defined orientation in EM as well as in solution that we discuss in relation to the pathway for force transmission through the integrin between ICAM-1 and the actin cytoskeleton.

The LFA-1 headpiece and ICAM-1
Previous crystal structures of ␣X␤2 and ␣L␤2 have revealed three markedly different ␣I domain orientations (Fig. 1, D-F)  (3, 8, 10). Among three distinct crystal lattices in one study with a total of 10 independent integrins, only two integrins showed density for the ␣I domain (8). Variations in lattice dimensions altered ␣I domain lattice contacts and orientation whether the ␣X␤2 ectodomain was associated with an open, internal ligandbound ␣I domain (3) or a closed ␣I domain (8). Reconstruction of the negatively stained ␣M␤2 ectodomain suggested a distinct orientation for the ␣I domain from that seen in one ␣X␤2 lattice (9), which, given the marked variation in ␣I domain orientation in distinct lattice environments, is hardly surprising. In basal conditions in Mg 2ϩ , we found here that the ␣L␤2 headpiece was predominantly in the closed conformation, as shown by both EM and SAXS. In the closed conformation, we found class averages that differed in ␣I domain position relative to the head and that were consistent with crystal structures in which the ␣I domain is tilted toward the ␣-subunit or ␤-subunit side of the headpiece or in a middle position. In SAXS, we saw no evidence for broadening of the envelope for the ␣I domain, and it was closest to the middle position. An interpretation consistent with all of the above studies is that in the closed conformation, 1) the ␣I domain has a preferred orientation near the middle of the variation seen among crystal lattices, 2) there is enough variation in ␣I domain position among integrin ensembles to prevent resolution in crystal structures where lattice contacts are absent, and 3) dynamic motion of the ␣I domain relative to the remainder of the headpiece is sufficiently limited so that an average position for the center of mass of the ␣I domain in solution is evident in SAXS molecular envelopes. In the closed conformation of the ␣I domain, its flexible connection to the ␣-subunit ␤-propeller domain in which it is inserted may enable it to sample multiple orientations to bind external ligand, as well as multiple orientations in which the ␣I domain may reshape into the open conformation, and its internal ligand can bind to its pocket in the integrin head between the ␤-propeller and ␤I domains.
In activating conditions in Mn 2ϩ , EM showed that the LFA-1 headpiece had predominantly the open conformation, with the hybrid, PSI, and I-EGF1 domains swung away from the ␣-subunit. SAXS also showed opening, and the best fit to the scattering data was obtained with an ensemble containing 76% open and 24% closed conformations. This was consistent with ab initio molecular envelopes that showed greater width between the ␣and ␤-subunit legs than in Mg 2ϩ and less width than expected for an open-headpiece model or seen with the LFA-1 headpiece in Mn 2ϩ when complexed with ICAM-1. In Mn 2ϩ , EM class averages of the headpiece alone or the headpiece bound to ICAM-1 showed better correlation to a crystal structure with an ␣I domain tilted toward the ␤-subunit. In Mn 2ϩ , SAXS results on the headpiece alone were consistent with either a middle ␣I domain position or tilt toward the ␤-subunit. These results support tilting toward the ␤-subunit of the ␣I domain in the open-headpiece conformation, which is consistent with binding the internal ligand at the C terminus of the ␣I domain to a pocket at the interface between the ␤I and ␤-propeller domains (3).
SAXS showed a relatively well-defined molecular envelope for the five Ig-like domains in the ectodomain of ICAM-1, with a clear kink. A kink was also evident in EM about two-fifths of the way along the rodlike length of the five tandem Ig-like domains (i.e. either between domains 2 and 3 or between domains 3 and 4). The bend had previously been localized to between domains 3 and 4 by antibody mapping in EM (24) and between domains 3 and 4 in crystal structures of D3-D5 of ICAM-1 and D1-D4 of ICAM-5 (21,23,27). A model of D1-D5 of ICAM-1 fit the SAXS molecular envelope well and confirmed the kink position. Crystal structures of ICAM-1 have shown variation in angle between D1 and D2 in lattices of up to 17° (20) and between D3 and D4 of up to 18° (21). This interdomain flexibility is far less than the ϳ150°variation in orientation seen for the ␣I domain among integrin structures (10). The

Orientation of LFA-1 bound to ICAM-1
well-defined molecular envelope of ICAM-1 in SAXS suggests that in solution, there are preferred orientations between the tandem Ig-like domains in ICAM-1.

The LFA-1-ICAM-1 complex
Given the potential range of motion found in crystal lattices of the integrin ␣I domain relative to the remainder of the integrin head, and between tandem Ig-like domains in ICAM-1, we found remarkably well-defined orientations between the ICAM-1 moiety and the headpiece moiety in complexes. Crystal structures of the isolated LFA-1 ␣I domain bound to D1-D2 fragments of ICAM-1, ICAM-3, and ICAM-5 have all shown a well-defined interface of the ␣I domain bound to D1 of the ICAM fragment, with the MIDAS Mg 2ϩ ion of the ␣I domain at the center of the interface (5,28,29). The Mg 2ϩ ion coordinates a Glu side chain in an edge ␤-strand in D1 of ICAM-1. In agreement, our EM class averages show that the ␣I domain binds to one side of the rod-shaped ICAM-1 molecule and at a position on the rod consistent with binding to D1 (Fig. 3, H and I). The ␣I domain appears as a round density in EM, and thus its orientation is impossible to assign based on the shape of its density. However, because it is known that the MIDAS-bearing face of the ␣I domain binds D1 of ICAM-1, and the orientation of the ␣I domain complex with D1 of ICAM-1 is clearly visible in each LFA-1-ICAM-1 complex class average, the ␣I domain with its MIDAS-bearing face must be markedly tilted toward the ␤-subunit side of the integrin head. In other words, while the ␣I domain is inserted in the ␤-propeller domain, it is tilted toward the ␤I domain interface with the ␤-propeller domain, to which its internal ligand is bound in the open conformation of the ␣I domain (3). The tilt toward ␤ of the ␣I domain in the complex with ICAM-1 and in the open LFA-1 headpiece in Mn 2ϩ in EM class averages was independently confirmed by cross-correlation with a crystal structure with a ␤-tilted ␣I domain.
The LFA-1 headpiece and ICAM-1 moieties were also wellresolved in complexes with ICAM-1. The headpiece was fully open, with the leg of the ␤-subunit swung away from the ␣-subunit, in both EM class averages and SAXS. In contrast, the headpiece in Mn 2ϩ alone was best represented as an ensemble with about 76% open and 24% closed conformations. The openheadpiece conformation of a ␤2 integrin remains to be defined with a crystal structure, but swing-out of the hybrid domain with the PSI and I-EGF1 domain matched well a model based on integrin ␣IIb␤3 (4). The ICAM-1-stabilized open headpiece resembled the open headpiece previously seen with a minority of integrin ␣X␤2 and ␣L␤2 ectodomain class averages in Mg 2ϩ (6), the open ␣X␤2 ectodomain conformation stabilized with activating Fab (7), the open ␣X␤2 ectodomain simultaneously bound to activating Fabs and the ligand iC3b (13), and the open headpiece visualized in complexes of the ligand-bound ␣X␤2 headpiece (30). The orientation of the rodlike density for ICAM-1 relative to the integrin headpiece was remarkably similar among three independent EM class averages and was uniform enough in solution to give a rodlike shape, and the orientation in solution obtained in ab initio SAXS models was essentially identical to that seen in EM, as shown by superposition of the model that fit the SAXS envelope with an EM class average (Fig. 3, F and G). The uniform orientation seen in three EM class averages confirms that the five Ig-like domains of ICAM-1 have a well-defined average position, as independently shown with SAXS of ICAM-1 alone.
We can infer that the LFA-1 ␣I domain in the LFA-1 complex with ICAM-1 is activated. There is no direct evidence at the resolution of the EM and SAXS studies here that the LFA-1 ␣I domain is in the open conformation or that its internal ligand is bound to its pocket at the ␤-propeller-␤I domain interface. However, it is known that the open ␣I domain binds with ϳ1,000-fold higher affinity than the closed ␣I domain to ICAM-1 (18, 31); that stabilizing the ␣L␤2 headpiece in the open conformation with activating Fabs to the ␤-subunit increases ␣I domain affinity for ICAM-1 by ϳ1,000-fold (18); and that mutation of LFA-1 ␣I domain internal ligand residue Glu-310 or antagonists directed to the pocket that binds the internal ligand abolish activation of LFA-1 adhesiveness, including by activating Fabs to the ␤-subunit (32)(33)(34). This previous evidence, together with the evidence here that the LFA-1 headpiece is open with its ␤-leg swung out when it binds ICAM-1, suggests with great confidence that in the complex with ICAM-1 visualized here, the ICAM-1-bound ␣I domain is open and its internal ligand is bound to its pocket at the interface between the ␤-propeller and ␤I domains.

␣I domain orientation and linkage to the headpiece
We may further deduce that binding of the internal ligand of the open ␣I domain to its pocket enforces a highly preferred orientation of the ␣I domain in which it is tilted toward its internal ligand binding pocket at the ␤-subunit side of the head (Figs. 1C and 3 (F and G)). This contrasts with the middle orientation of the ␣I domain (Fig. 1B) in a structure of the ␣X␤2 ectodomain in which the crystal lattice enforced the open ␣I domain conformation and in which the ␣I domain internal ligand bound to its pocket and the ␤I domain was in an intermediate, partially open conformation (3). The ␣I domain was held away from the remainder of the head by lattice contacts, and its ␣7 helix was largely unwound to span the distance to the internal ligand-binding pocket. It was therefore predicted that the open ␣I domain should be able to substantially rotate while remaining allosterically engaged (3). This prediction is consistent with the middle ␣I domain orientation observed in the latter structure and the ␤-tilt of the ␣I domain in the ICAM-1engaged ␣L␤2 headpiece structure observed here. The unwinding of the ␣I domain ␣7 helix must have been energetically unfavorable, and imposed by the crystal lattice, which held the ␣I domain 15 Å farther away from the binding pocket for the internal ligand than would have been possible without ␣7 helix unwinding (3). It is reasonable to assume that in the LFA-1 complex with ICAM-1, the ␣I domain ␣7 helix remains helical, which would bring it ϳ15 Å closer to the internal ligand binding pocket and tilt the ␣I domain toward the ␤-subunit, exactly as observed here in the LFA-1-ICAM-1 complex.
We previously visualized in EM a complex between the ectodomain of integrin ␣X␤2, three activating Fabs, and fragments of complement component C3 that are ligands of ␣X␤2, iC3b, and C3c (13). The ␣X␤2 ␣I domain bound to a specific site on the "key ring" moiety of C3c, which lies predominantly in one plane. Both the ␣X␤2 ectodomain and iC3b/C3c had planar Orientation of LFA-1 bound to ICAM-1 orientations on the EM grid. In some class averages, the orientation of C3c was flipped, whereas interaction of the ␣I domain was maintained. We speculated that the orientation of the iC3b/C3c key ring and extended-open integrin planes might have been more perpendicular than co-planar in solution and that adsorption on the grid forced them to be co-planar and resulted in occasional rotation of the ␣I domain C3c moiety relative to the remainder of the integrin ectodomain (13). This result did not imply lack of a preferred orientation of the ligandbound ␣I domain in solution. In the case of binding of ICAM-1 to the LFA-1 headpiece, our SAXS data suggest that in solution, the ICAM-1 rod is in the same plane as the LFA-1 headpiece. Thus, there should be no tendency for ICAM-1 to flip during adsorption to the substrate, consistent with the excellent agreement among distinct EM class averages of ICAM-1-LFA-1 complexes and the SAXS molecular envelope.
Our mutational data demonstrate that movements in the ␣1 and ␣1Ј helices of the ␤I domain are coupled to adhesiveness, ligand binding, headpiece opening, and extension of integrins LFA-1 and Mac-1 on the cell surface. Shifts in these helices were noted in a crystal structure of an intermediate conformation of the ␣X␤2 ␤I domain (3). As predicted by the structure and confirmed by mutation of ␣X␤2, ␤2 V124A and L127A mutations activated and inhibited, respectively, ligand binding and opening and extension of ␣X␤2. Here, we extended these results to LFA-1 and Mac-1. Furthermore, we tested two residues that are more C-terminal in the ␣1Ј helix, Leu 132 and Leu 135 . This region underwent less movement in the intermediate conformation but is expected to be involved in more extensive, ratchet-like exchanges in position in an open ␤I domain conformation. We found that mutations L132A and L135A activated and inhibited, respectively, binding to soluble ICAM-1 and adhesiveness to ICAM-1 of LFA-1, adhesiveness to ligands including rosetting with iC3b-sensitized erythrocytes of Mac-1, and exposure of epitopes associated with headpiece opening and ectodomain extension on cell surfaces of both LFA-1 and Mac-1. These results suggest that movements may occur throughout the length of the ␣1 and ␣1Ј helices in the open conformation of the ␤2 I domain and predict that burial of Leu 132 is important to stabilize the open conformation, whereas burial of Leu 135 (perhaps in the same ratchet pocket) is important to stabilize the closed conformation of the ␤2 I domain. Overall, the mutational results extend our observations on purified fragments to LFA-1 on cell surfaces and show that high-affinity binding to ICAM-1, headpiece opening, and extension are linked and that this linkage requires specific conformational changes in the LFA-1 ␤I domain.
The anti-parallel orientation of ICAM-1 and the integrin ␤-leg in complexes visualized here is relevant to force-dependent activation by the actin cytoskeleton (35). Actin polymerization and myosin-dependent filament contractility apply force through adaptor proteins to integrin cytoplasmic domains, which is then transmitted through integrins to ligands, which resist the applied force as a consequence of anchorage on the cell surface or in the extracellular matrix (36). Tensile force measurements in the ␤-subunit cytoplasmic domain of LFA-1 show that tension generated by the actin cytoskeleton is required to induce and stabilize adhesion to ICAM-1 (12). Ten-sile force stabilizes more extended and open integrin conformations because of their increased length in the direction of force transmission (37,38). Our study shows that the ICAM-1-LFA-1 headpiece complex has a preferred orientation in which the pathways for force transmission through ICAM-1 and the upper ␤-subunit leg are parallel to one another. As force is transmitted through receptors and their ligands, force balance requires that the pathway for force transmission become aligned with the direction of force application. Flexible domains and flexible linkages between ectodomains and the plasma membrane will straighten out, like links in a tow chain, to align the receptor-ligand complex with the points where force is applied by the cytoskeleton to the integrin and resisted by the ligand. Fluorescence polarization microscopy shows that LFA-1 is aligned in the same direction as, and is tilted by, the force applied by actin retrograde flow (39). The structure of the ICAM-1-LFA-1 complex shows that ICAM-1 and the integrin ␤-leg are pre-aligned in an orientation that requires little readjustment after force application, which makes the complex well-suited for stabilization by cytoskeletal force of the highaffinity, extended-open integrin conformation.

Synchrotron SAXS measurements
Purified LFA-1 headpiece in 5 mM Mg 2ϩ , 1 mM Ca 2ϩ at 1 mg/ml and purified LFA-1 headpiece, ICAM-1 ectodomain, and their complex in 1 mM Mn 2ϩ , 0.2 mM Ca 2ϩ at concentrations of 1.4, 0.85, and 3.2 mg/ml, respectively, were shipped on ice to the synchrotron and incubated at ambient temperature for 15-20 min before X-ray solution scattering measurements were performed. Small and wide angle X-ray scattering measurements were collected at beam line X9 at the Brookhaven National Synchrotron Light Source (Upton, NY). The detector was a high sensitivity 300K Pilatus, at a sample-to-detector distance of 3.5 m. Samples were passed through a flow capillary, collecting data at 20-s exposures in triplicate at the specified concentrations. Guinier analysis showed no signs of radiation damage or aggregation. I(0) and the pair distance distribution function P(r) were calculated by circular averaging of the scattering intensities I(q) and scaling using the software GNOM (40). LFA-1 and ICAM-1 alone scattering data were processed to a q(A Ϫ1 ) of 0.15 and 0.2, respectively, and 12 ab initio models were generated using DAMMIF (41). The LFA-1-ICAM-1 complex scattering data were processed to a q(A Ϫ1 ) of 0.61.
Orientation of LFA-1 bound to ICAM-1 GASBOR (42), which can utilize wide-angle scattering, was used to generate 12 ab initio models. Models were superimposed and averaged using the program DAMAVER (43). The theoretical solution scattering of the crystal structures of LFA-1 headpiece in the closed and open states, ICAM-1, and the open LFA-1 headpiece-ICAM-1 complex were calculated using CRYSOL (44).

Molecular models
LFA-1 headpiece models utilized PDB entry 5E6U (10). Because the LFA-1 headpiece crystal structure lacks a thigh domain, that of ␣X␤2 was added from PDB entry 5ES4. Three different orientations of the ␣I domain were utilized: PDB entries 5E6U for the ␤I-tilt, 4NEH for the mid-␣I tilt, and 5ES4 for the ␣I-tilt. To model the ICAM-1 D1-D5 complex with the ␣ L I domain, crystal structures of ICAM-1 D3-D5 (27) and the ␣ L I domain-ICAM-1 D1-D2 complex (5) were superimposed onto the ICAM-5 D1-D4 structure (23). The ICAM-1 D1-D5/␣ L I domain model was used to model the LFA-1 headpiece-ICAM-1 D1-D5 structure. To obtain the model of the ICAM-1 D1-D5 ectodomain alone, the ␣ L I domain was removed from the ICAM-1 D1-D5/␣ L I domain structure.
Swing-out motion of the hybrid domain relative to the ␤3 I domain was modeled into the ␤2-subunit (4) as follows. Comparison of the closed ␤2and ␤3-subunits after superimposing only the ␤I domains shows that the closed ␤2 and ␤3 hybrid domain positions differ by an in-plane angle of ϳ20º, with the ␤2 hybrid domain more swung in. Therefore, the same amount of rotation as seen between the closed and open conformations of ␤3 was applied to the ␤2 hybrid domain to swing it out.
Two models, 1) the ICAM-1 D1-D5 model that contains the ␣ L I domain and 2) the open LFA-1 headpiece lacking the ␣ L I domain, were simultaneously docked to the SAXS profile using the FOXSDOC server. The position of the LFA-1 headpiece was then manually adjusted, which decreased the value from 1.7 to 1.6.

Rosetting assay
Sheep erythrocytes were sensitized with IgM anti-Forssman and human C5-deficient serum as described previously (46). E-IgM-iC3b and E-IgM as control were assayed for binding to ␣X␤2 HEK293T transfectants (10 5 cells/well in 12-well cluster plates) (47). Briefly, E-IgM-iC3b or E-IgM in HBS was incubated for 1.5 h at 37°C in the presence of 1 mM Mg 2ϩ /Ca 2ϩ or Mn 2ϩ . After three washes, the percentage of cells with rosette formation (Ͼ10 erythrocytes/HEK293T cell, over 100 HEK293T cells examined) was assessed by microscopy.

V-bottom cell adhesion assays
Cell adhesion to a V-bottom-well plate was assayed as described previously (48). Briefly, V-bottom 96-well plates (Corning) were coated with a chimera containing the five Ig domains of ICAM-1 fused to the Fc portion of IgA (ICAM-1-Fc␣) or human fibrinogen (1 g/ml) at 4°C overnight and then blocked with Hepes-buffered saline, pH 7.4, and 2% BSA for 1 h at ambient temperature. LFA-1 and Mac-1 HEK293T transfectants were labeled for 30 min at 37°C with 2Ј,7Ј-bis-(carboxyethyl)-5-(and-6)-carboxyfluorescein acetoxymethyl ester (Molecular Probes), washed, resuspended in HBS (5 ϫ 10 4 cells/50 l) with 2 mM Mg 2ϩ /Ca 2ϩ or Mn 2ϩ or 10 mM EDTA, and incubated for 30 min at ambient temperature. 50 l of transfectants were added to each well and incubated for 30 min and then centrifuged at 200 ϫ g for 5 min at ambient temperature.

Soluble ICAM-1 binding
HEK293T transient transfectants were washed with HBS (20 mM HEPES, 150 mM NaCl, pH 7.3) containing 5 mM EDTA and resuspended in HBS buffer containing Mg 2ϩ /Ca 2ϩ or Mn 2ϩ or EDTA. Binding of dimeric soluble ICAM-1 was assayed as follows. ICAM-1-Fc␣ was added to the cells at 10 g/ml and incubated at 37°C for 30 min in the presence of 1 mM Mg 2ϩ /Ca 2ϩ or Mn 2ϩ . The cells were washed and incubated with a 1:100 dilution of goat anti-human IgA-FITC (Zymed Laboratories Inc.) for 30 min at room temperature, washed, and analyzed by flow cytometry.

Epitope exposure
For immunofluorescent flow cytometry (49), HEK293T transfectants were stained with FITC-labeled CBR LFA-1/7 and biotinylated KIM127, MEM148, or m24, referenced elsewhere (7), in the presence of 1 mM MgCl 2 /CaCl 2 or 1 mM MnCl 2 and then with phycoerythrin (PE)-streptavidin to recognize the activation-dependent epitopes. An FITC gate was set to define transfected cells and used to collect PE fluorescence. Mutant PE mean fluorescence intensity (MFI) was normalized by multiplying it by (FITC MFI of WT)/(FITC MFI of mutant).
Author contributions-M. S. and T. A. S. conceived the work and wrote the manuscript; A. C. K. carried out and wrote sections on SAXS data collection and analysis; M. S. expressed and purified ␣L␤2 and high-affinity ICAM-1, prepared samples, analyzed SAXS and EM data, fit molecular models, tested ␤2 mutations for functional analysis, and wrote the paper; K. Y. made and tested ␤2 mutations in the binding assays; J. W. acquired EM data.