Evidence that monoclonal antibodies directed against the integrin beta subunit plexin/semaphorin/integrin domain stimulate function by inducing receptor extension.

The overall structure of integrins is that of a ligand-binding head connected to two long legs. The legs can exhibit a pronounced bend at the "knees," and it has been proposed that the legs undergo a dramatic straightening when integrins transit from a low affinity to a high affinity state. The knee region contains domains from both alpha and beta subunits, including the N-terminal plexin/semaphorin/integrin (PSI) domain of the beta subunit. The role played by the knee domains in the regulation of integrin-ligand binding is uncertain. Here we show that: (i) monoclonal antibodies (mAbs) N29 and 8E3 have epitopes in the beta(1) subunit PSI domain and stimulate ligand binding to alpha(5)beta(1); (ii) N29 and 8E3 cause long range conformational changes that alter the ligand binding activity of the head region; (iii) the stimulatory action of these mAbs is dependent on the calf-1 domain, which forms part of the alpha subunit knee; and (iv) the epitopes of 8E3 and N29 map close to the extreme N terminus of the PSI and are likely to lie on the side of this domain that faces the alpha subunit. Taken together, our data suggest that the binding of these mAbs results in a levering apart of the PSI and calf-1 domains, and thereby causes the alpha and beta subunit knees to separate. Several major inferences can be drawn from our findings. First, the PSI domain appears to form part of an interface with the alpha subunit that normally restrains the integrin in a bent state. Second, the PSI domain is important for the transduction of conformational changes from the knee to head. Third, unbending is likely to provide a general mechanism for control of integrin-ligand recognition.

Integrins provide a crucial bridge between the inside and outside environments of the cell by linking the surrounding matrix of a cell to its cytoskeletal framework (1). These receptors are ␣,␤ heterodimers, and both subunits have large extracellular domains and short intracellular regions. Integrins carry a two-way flow of information (inside the cell to out, and outside to in). To achieve this bidirectional signaling integrins must convey shape changes over a long distance, from the intracellular domains to the extracellular regions and vice versa (2,3). Furthermore, in most cases binding of integrins to their extracellular ligands has to be tightly controlled. For example, the interaction of ␣ IIb ␤ 3 with fibrinogen during platelet aggregation needs to be restricted to sites of vessel injury. Regulation of ligand binding is achieved by switching of an integrin between a constitutive low affinity (inactive) state and a high affinity (primed) state. In addition, the interaction of ligands with integrin stabilizes the high affinity state and may cause further shape shifting (ligand-activated state) (4,5). However, the molecular basis of the conformational changes involved is currently uncertain.
The recent crystal structures of the extracellular domains of ␣ V ␤ 3 (6, 7) have provided new insights into integrin function. Overall, the integrin structure resembles that of a "head" on two "legs." The head region contains a seven-bladed ␤ propeller in the ␣ subunit, the upper surface of which is in close association with a von Willebrand factor type A domain in the ␤ subunit (␤A). 1 ␤A (also referred to as the I-like domain or ␤I domain) contains a central ␤ sheet encircled by seven ␣ helices. ␤A is connected at its N and C termini to an immunoglobulinlike "hybrid" domain and forms an extensive interface with it.
The key regions involved in ligand recognition are loops on the upper surface of the ␤ propeller and the top face of ␤A, which contains a metal ion-dependent adhesion site (MIDAS). The ␤A domain can exist in low affinity and high affinity states, and the conformation of this domain is the critical determinant of ligand binding affinity (8 -11).
An unexpected feature of the ␣ V ␤ 3 structure was a cramping bend in both the ␣ and ␤ subunits at a region termed the "genu" (or "knee"), such that the head region was folded down between the legs. The knee region involves the thigh and calf-1 domains in the ␣ subunit, and the plexin/semaphorin/integrin (PSI) domain and EGF repeats 1 and 2 in the ␤ subunit. The ␤ subunit knee domains were not clearly resolved in the structure, suggesting that the knee may be flexible rather than rigid. Initially, the bent ␣ V ␤ 3 structure presented a puzzle of how transmission of conformational change from the cytoplasmic tails to the head domains could take place in the native integrin, particularly in view of the rather flexible knees. Furthermore, in the bent state the head region would be pointing toward the cell surface and would not be in the appropriate orientation to interact with extracellular ligands. Small structural movements were observed in an ␣ v ␤ 3 crystal structure soaked with a Arg-Gly-Asp ligand-mimetic peptide (7), but probably because of crystal contact constraints, these changes were limited to the head region and did not provide a mechanism for long range propagation of conformational change.
Recently, it has been proposed that the bent state of the integrin represents a low affinity conformation and that acquisition of the high affinity conformation involves an unbending of the knees to form an extended state (12). Major support for this model comes from studies of soluble recombinant integrins by electron microscopy (13), which show that ␣ V ␤ 3 is bent under conditions in which it is poorly active (e.g. in the presence of Ca 2ϩ ) and extended in the presence of Mn 2ϩ or ligand peptides (which promote the high affinity conformer). It has also been shown that constraining ␣ V ␤ 3 or ␣ IIb ␤ 3 in a bent form with a disulfide bond results in a low affinity state (13), whereas promoting unbending with a glycan wedge favors the high affinity state (14). The presence of activation state-dependent epitopes throughout the head and leg regions also provides evidence for large scale conformational rearrangements (15). It is hypothesized that unbending of the knee regions allows an outward movement of the hybrid domain which shifts the ␤A domain into a high affinity state (9,13).
Although considerable evidence has been presented that ␤ 3 and ␤ 2 integrins undergo unbending at the knees (12,13,16,17), data showing that this shape change is relevant to the function of ␤ 1 integrins (which make up half of all known integrin heterodimers) have been sparse. Furthermore, the precise nature of the conformational changes that take place during unbending are unclear. Among the domains found in the knee region, the PSI is particularly mysterious. Although this domain is conserved in integrins throughout all metazoa, its function is unknown. The PSI domain is found at the N terminus of the ␤ subunit and precedes the hybrid domain. The epitopes of some stimulatory anti-␤ 3 mAbs map to the PSI (18), and it is also an important site of drug-induced epitopes. Induction of these epitopes leads to platelet destruction by the immune system in some patients treated with ␣ IIb ␤ 3 antagonists (19), and therefore an understanding of the role of the PSI domain in integrin priming is also of clinical importance. A previously characterized anti-␤ 1 mAb whose epitope lies in this region of the subunit is N29 (20). N29 stimulates cell adhesion, and agents such as Mn 2ϩ and dithiothreitol, which are known to stimulate integrin function, induce expression of its epitope (21). A novel mAb 8E3, developed in our own laboratory, also stimulates cell adhesion, and its epitope appears to lie in the N-terminal region of ␤ 1 (22).
Here we demonstrate that N29 and 8E3 stimulate ligand binding through a mechanism that requires the calf-1 domain of the ␣ subunit leg. Both mAbs have epitopes at the extreme N terminus of the PSI domain, and we show that binding of these antibodies to the integrin is likely to wedge apart the ␣ and ␤ knee regions, resulting in unbending.
Expression Vector Construction and Mutagenesis-C-terminally truncated human ␣ 5 constructs encoding residues 1-613 (TR␣ 5 ), 1-694, 1-795, or 1-951 (FL␣ 5 ), and C-terminally truncated human ␤ 1 constructs encoding residues 1-455 (TR␤ 1 ) or 1-708 (FL␤ 1 ) were generated as described previously (22). An ␣ 5 construct containing residues 1-603 (i.e. lacking all of calf-1 and calf-2) was created using the same procedures. To create a construct containing only the thigh and calf-1 domains (C1␣ 5 ) the truncation position between calf-1 and calf-2 domains (amino acid Ala 749 ) was chosen based on alignment of the ␣ 5 subunit sequence with the ␣ V subunit structure (6). Both ␣ and ␤ constructs were fused in-frame to the hinge regions and C H 2 and C H 3 domains of human IgG␥1 (i.e. the Fc portion of the heavy chain). The sequence of the constructs was verified by DNA sequencing. To aid the formation of heterodimers, the C H 3 domain of the ␣ 5 construct contained a "hole" mutation, whereas the C H 3 domain of the ␤ 1 constructs carried a "knob" mutation as described previously (22). Oligonucleotides were purchased from MWG Biotech (Southampton, UK).
Reactivity of ␤ 1 mAbs with Recombinant PSI Domain-96-well plates (Costar 1 ⁄2-area EIA/RIA, Corning Science Products, High Wycombe, UK) were coated with PSI domain fusion protein at a concentration of 10 g/ml in 50 l/well Dulbecco's phosphate-buffered saline for 16 h. Wells were then blocked for 1 h with 200 l of 5% (w/v) bovine serum albumin, 150 mM NaCl, 0.05% (w/v) NaN 3 , 25 mM Tris-Cl, pH 7.4 (blocking buffer). The blocking solution was removed, and the wells were then washed three times with 200 l of 150 mM NaCl, 25 mM Tris-Cl, pH 7.4, containing 1 mg/ml bovine serum albumin (buffer A). Anti-␤ 1 mAbs (5 g/ml in buffer A) were then added (50 l/well). The plate was incubated for 2 h at room temperature and then washed three times in buffer A. Peroxidase-conjugated anti-mouse Fc secondary antibody (1:1,000 dilution in buffer A; Jackson Immunochemicals) was added (50 l/well) for 30 min, the plate washed four times in buffer A, and color was developed using ABTS substrate. Background binding of mAbs to wells treated with blocking buffer alone was subtracted from all measurements. Measurements obtained were the mean Ϯ S.D. of four replicate wells.
Purified ␣ 5 ␤ 1 Ligand Binding Assays-96-well plates (Costar 1 ⁄2area EIA/RIA) were coated with K20 at a concentration of 2 g/ml in 50 l/well Dulbecco's phosphate-buffered saline for 16 h. Wells were then blocked for 1-3 h with blocking buffer and then washed three times with buffer A. Biotinylated 3Fn6 -10 (0.1 g/ml) in buffer A containing 1 mM MnCl 2 , 5 mM MgCl 2 , 1 mM MgCl 2 plus 1 mM CaCl 2 , or 1 mM CaCl 2 was added to the plate (50 l/well), either alone, or in the presence of mAb JB1A, N29, 8E3, or 12G10 (5 g/ml). The plate was then incubated at 30°C for 2 h. Unbound ligand was removed and the wells washed three times with buffer A. Bound ligand was quantitated by the addition of a 1:500 dilution of ExtrAvidin peroxidase conjugate (Sigma) in buffer A with 1 mM MnCl 2 (buffer B) for 20 -30 min at room temperature (50 l/well). Wells were then washed four times with buffer B, and color was developed using 50 l/well ABTS substrate. Each experiment shown is representative of at least three separate experiments.
Effect of N29 and 8E3 on Binding of 12G10 and SNAKA51-Plates were coated with K20 and blocked as described above. The blocking solution was removed, and placental ␣ 5 ␤ 1 (approximately 1 g/ml in Dulbecco's phosphate-buffered saline) was added (50 l/well) for 1-2 h. The plate was washed three times with 200 l/well buffer A, and 0.1 g/ml biotinylated 12G10 or 1 g/ml biotinylated SNAKA51 in buffer B was added (50 l/well) in the absence or presence of N29, 8E3, or JB1A (5 g/ml). The plate was incubated for 2 h at 30°C and then washed three times in buffer A. Bound 12G10 or SNAKA51 was quantitated by the addition of a 1:500 dilution of ExtrAvidin peroxidase conjugate (Sigma) in buffer B for 20 -30 min at room temperature (50 l/well). Wells were then washed four times with buffer B, and color was developed using 50 l/well ABTS substrate. Background binding of biotinylated mAbs to wells treated with blocking buffer alone was subtracted from all measurements. Measurements obtained were the mean Ϯ S.D. of four replicate wells.
Recombinant ␣ 5 ␤ 1 -Fc Ligand Binding Assays-96-well plates (Costar 1 ⁄2-area EIA/RIA) were coated with goat anti-human ␥1 Fc (Jackson Immunochemicals, Stratech Scientific, Luton, UK) at a concentration of 2.6 g/ml in Dulbecco's phosphate-buffered saline (50 l/well) for 16 h. Wells were then blocked for 1-3 h in blocking buffer and then incubated with supernatants from CHO cell transfections (20 -180 l/well) for 1-3 h at room temperature. Wells were then washed three times with buffer A. Biotinylated 3Fn6 -10 (0.1 g/ml) in buffer B was added to the plate (50 l/well) alone or in the presence of mAb N29, 8E3 or 12G10 (5 g/ml) or in the presence of a control mAb (K20 or JB1A at 5 g/ml). The plate was then incubated at 30°C for 2 h. Unbound ligand was aspirated and the wells washed three times with buffer B. Bound ligand was quantitated by the addition of a 1:500 dilution of ExtrAvidin peroxidase conjugate in buffer B for 20 -30 min at room temperature (50 l/well). Wells were then washed four times with buffer B, and color was developed using ABTS substrate (50 l/well). Background binding to wells treated with supernatant from mock-transfected cells was subtracted from all measurements. Measurements obtained were the mean Ϯ S.D. of four replicate wells. To make a comparison between the different ␣ 5 ␤ 1 -Fc proteins the binding of 5 g/ml mAb TS2/16 was used to normalize for any differences between the amounts of the different heterodimers bound to the wells (22). Each experiment shown is representative of at least three separate experiments.
Epitope Mapping of 8E3 and N29 -Substitution of human residues with the corresponding residues in murine ␤ 1 within the PSI domain was performed using a PCR-based mutagenesis kit (Gene Taylor, Invitrogen) according to the manufacturer's instructions, or by overlap extension PCR (22). The substitutions made were E4K, S31T, and P56Q/D58S (as a dual substitution) in the TR␤ 1 -Fc construct. Oligonucleotides for mutagenesis were purchased from MWG Biotech (Southampton, UK). CHO-L761h cells were transfected with wild-type or mutant TR␤ 1 -Fc, together with wild-type TR␣ 5 -Fc, and supernatants were harvested as described above. Plates were coated with anti-human Fc and blocked as described above. The blocking solution was removed, and cell culture supernatants were added (25 l/well) for 1-2 h. All supernatants were assayed in triplicate, and supernatant from mocktransfected cells was used as a negative control. The plate was washed three times with 200 l/well buffer B, and anti-␤ 1 mAbs N29, 8E3, and TS2/16 (5 g/ml in buffer B) were added (50 l/well). The plate was incubated for 2 h and then washed three times in buffer B. Peroxidaseconjugated anti-mouse Fc secondary antibody (1:1,000 dilution in buffer B; Jackson Immunochemicals) was added (50 l/well) for 30 min, the plate washed four times in buffer B, and color was developed as above. All steps were performed at room temperature. Background binding of mAbs to wells incubated with supernatant from mock-transfected cells was subtracted from all measurements. Measurements obtained were the mean Ϯ S.D. of three replicate wells.
Competitive ELISA Experiments-Plates were coated with anti-human Fc and blocked as described above. Wells were then incubated with supernatant from cells transfected with 20 l/well FL␣ 5 FL␤ 1 -Fc for 1-2 h at room temperature. Wells were then washed three times with buffer B. Biotinylated HUTS-4 (0.5 g/ml) or 8E3 (0.1 g/ml) in buffer B was added to the plates (50 l/well) either alone or in the presence of unlabeled N29, 8E3, HUTS-4, 15/7, K20, or TS2/16 (10 g/ml). The plates were then incubated at room temperature for 2 h. The wells were washed three times with buffer B, and the amount of biotinylated HUTS-4 or 8E3 bound was quantitated by the addition of a 1:500 dilution of ExtrAvidin peroxidase conjugate in buffer B for 20 -30 min at room temperature (50 l/well). Wells were then washed four times with buffer B, and color was developed using 50 l/well ABTS substrate. Background binding of biotinylated mAbs to wells incubated with supernatant from mock-transfected cells was subtracted from all measurements. Measurements obtained were the mean Ϯ S.D. of four replicate wells.
To examine the ability of a Fab fragment of 8E3 to inhibit competitively the binding of other ␤ 1 mAbs, plates were coated with anti-human Fc, blocked, and then incubated with supernatant from cells transfected with 20 l/well FL␣ 5 FL␤ 1 -Fc for 1-2 h at room temperature. Wells were then washed three times with buffer B. mAbs (N29, 8E3, HUTS-4, 15/7, K20, or TS2/16) at 1 g/ml in buffer B were added to the wells either alone or in the presence of 10 g/ml 8E3 Fab. The plates were then incubated at room temperature for 2 h. The wells were washed three times with buffer B, and the amount of IgG bound was quantitated using a 1:1,000 dilution of goat anti-mouse Fc peroxidase conjugate (Sigma) in buffer B for 30 min at room temperature (50 l/well). Wells were then washed four times with buffer B, and color was developed as above. Background binding of mAbs to wells incubated with supernatant from mock-transfected cells was also measured. Measurements obtained were the mean Ϯ S.D. of four replicate wells. Data are expressed as shown in Equation 1.
Each experiment shown is representative of at least three separate experiments.

Monoclonal Abs N29 and 8E3 Have Epitopes in the PSI Domain and Increase the Ligand Binding Activity of
is a previously characterized mAb that stimulates ligand binding to ␤ 1 integrins (20). The epitope of N29 lies in the Nterminal region of ␤ 1 (amino acids 1-57) but has not been mapped precisely (21). We recently reported a mAb 8E3 that binds to ␤ 1 constructs containing the N-terminal region of ␤ 1 but does not bind to a construct lacking the first 119 amino acids of ␤ 1 , suggesting that its epitope could lie in the same region of the subunit as N29 (22). We found that both mAbs bound strongly to a recombinant protein containing the PSI domain (residues 1-60 of ␤ 1 ) in an ELISA, whereas other ␤ 1 antibodies did not react with this protein (Fig. 1A). Hence, the epitopes of both N29 and 8E3 appear to lie entirely within the PSI domain.
N29 and 8E3 increased binding of a fibronectin fragment to purified ␣ 5 ␤ 1 from human placenta (Fig. 1B), both under conditions where the integrin displayed high constitutive ligand binding activity (in Mn 2ϩ ) and where the integrin had low constitutive activity (Mg 2ϩ or Mg 2ϩ /Ca 2ϩ ). Both mAbs had little stimulatory effect in Ca 2ϩ alone. In these experiments, the hybrid domain mAb JB1A (27) was used as a negative control, 2 whereas the potent stimulatory mAb directed against the ␤A domain, 12G10 (8,28), was used as a positive control.
N29 and 8E3 Cause Long Range Conformational Changes in ␣ 5 ␤ 1 -To elucidate the mechanism by which N29 and 8E3 promote ligand binding by ␣ 5 ␤ 1 , we tested whether these mAbs affected the expression of activation epitopes on the ␣ 5 and ␤ 1 subunits (Fig. 2). The 12G10 epitope in the ␤A domain is expressed strongly only on the high affinity conformation of ␤A, and an increase in the binding of this mAb reports a movement of the ␣ 1 helix of ␤A (8). We found that N29 and 8E3 markedly mAb binding in the presence of 8E3 Fab Ϫ background binding mAb binding in the absence of 8E3 Fab Ϫ background binding ϫ 100% (Eq. 1) increased 12G10 binding, suggesting that both mAbs can induce conformational changes in ␤A. The SNAKA51 epitope lies in the calf domains of the ␣ 5 subunit, and expression of this epitope is enhanced in the primed and ligand-activated forms of the integrin (29). Both N29 and 8E3 caused a partial induction of the SNAKA51 epitope. The control mAb JB1A did not increase the binding of 12G10 or SNAKA51. Taken together, these data show that the PSI domain mAbs induce shape changes that are conveyed over a large distance, affecting both the head and leg regions. Furthermore, these conformational changes favor the high affinity state of the integrin and correlate with increased ligand binding activity of the head region. Stimulation of Ligand Binding by N29 and 8E3 Requires the ␣ Subunit Leg Region-Despite their ability to enhance ligand binding by purified ␣ 5 ␤ 1 , we observed previously that N29 and 8E3 do not stimulate fibronectin binding to a recombinant truncated ␣ 5 ␤ 1 that lacks the leg regions of the ␣ and ␤ subunits (30). To shed further light on the mechanism by which N29 and 8E3 modulate ligand recognition by ␣ 5 ␤ 1 , we created a number of recombinant ␣ 5 ␤ 1 constructs with truncations in the ␣ and/or ␤ leg regions, using a soluble Fc expression system (22) (Fig. 3A). All of these constructs were expressed well by CHO cells, and each protein reacted strongly with 8E3 and N29, suggesting that the epitopes of these mAbs were expressed at levels essentially identical to that of the constitutively expressed epitope of TS2/16 (data not shown). Recombinant receptors were tested for their ability to bind a fibronectin fragment and for the effect of N29 and 8E3 on this interaction. In these experiments a nonfunction altering mAb (K20 or JB1A) was used as a negative control, whereas 12G10 was used as a positive control. In the presence of Mn 2ϩ , a heterodimer containing the full-length extracellular domains of the ␣ and ␤ subunits (FL␣ 5 FL␤ 1 -Fc) was highly active for ligand binding in the absence of function-modulating mAbs but nevertheless showed a statistically significant increase in ligand recognition after treatment with N29 or 8E3 (Fig. 3B). Both mAbs strongly stimulated fibronectin binding to FL␣ 5 FL␤ 1 -Fc in the presence of Mg 2ϩ or Mg 2ϩ /Ca 2ϩ (data not shown). A construct containing the full-length extracellular domain of the ␣ subunit but lacking the leg region of the ␤ subunit (EGF repeats 1-4, and TD) (FL␣ 5 TR␤ 1 -Fc) again had high constitutive ligand binding activity, but this activity was increased further by N29 or 8E3 (Fig. 3C). In contrast, a construct containing the full-length extracellular domain of the ␤ subunit but lacking the leg region of the ␣ subunit (calf-1 and calf-2 domains) (TR␣ 5 FL␤ 1 -Fc) had low constitutive ligand binding activity, and this activity was not increased by N29 or 8E3, even though 12G10 was able to rescue the ligand binding activity fully (Fig. 3D). These results suggest that the ability of N29 and 8E3 to stimulate ligand binding by ␣ 5 ␤ 1 is dependent on the presence of the ␣ subunit leg domains but not the ␤ subunit leg domains.
Stimulation of Ligand Binding by N29 and 8E3 Requires the ␣ Subunit Calf-1 Domain-To narrow the region of the ␣ 5 subunit leg required for the stimulatory effect of N29 and 8E3, FIG. 1. mAbs N29 and 8E3 react with recombinant PSI domain and stimulate ligand binding to purified ␣ 5 ␤ 1 . A, recombinant PSI domain-thioredoxin fusion protein was coated onto ELISA plate wells. Binding of N29, 8E3, or other anti-␤ 1 mAbs (Lia1/2, TS2/16, or 12G10) to the fusion protein was measured. B, ␣ 5 ␤ 1 purified from human placenta was captured onto ELISA plate wells using the nonfunction perturbing ␤ 1 mAb K20. Binding of 3Fn6 -10 fragment to ␣ 5 ␤ 1 was measured in the presence of the control mAb JB1A (open bars), or mAb N29 (light gray bars), 8E3 (dark gray bars), or 12G10 (filled bars) in an assay buffer containing 1 mM Mn 2ϩ , 5 mM Mg 2ϩ , 1 mM Mg 2ϩ plus 1 mM Ca 2ϩ (Mg 2ϩ /Ca 2ϩ ), or 1 mM Ca 2ϩ . No ligand binding was observed in an assay buffer containing 2 mM EDTA (data not shown).
FIG. 2. N29 and 8E3 increase the expression of activation epitopes recognized by 12G10 and SNAKA51. ␣ 5 ␤ 1 from human placenta was captured onto ELISA plate wells using mAb K20. Binding of biotinylated 12G10 or biotinylated SNAKA51 to ␣ 5 ␤ 1 was measured in the absence of other mAbs (Con; open bars) or in the presence of the control mAb JB1A (diagonally shaded bars), mAb N29 (cross-hatched bars), or 8E3 (horizontally shaded bars) in an assay buffer containing 1 mM Mn 2ϩ . *, p Յ 10 Ϫ4 by Student's t test, relative to control. A similar stimulation of 12G10 and SNAKA51 binding by N29 and 8E3 was observed in an assay buffer containing 5 mM Mg 2ϩ in place of Mn 2ϩ (data not shown).
we created a construct of ␣ 5 which was truncated at the end of the calf-1 domain (C1␣ 5 ; Fig. 3A). A recombinant heterodimer containing C1␣ 5 together with the full-length extracellular domain of ␤ 1 (C1␣ 5 FL␤ 1 -Fc) had high constitutive ligand binding activity, which was increased further by N29 and 8E3 (Fig. 4A). N29 and 8E3 retained the ability to stimulate ligand binding to a protein containing C1␣ 5 together with the TR␤ 1 subunit (C1␣ 5 TR␤ 1 -Fc), even though this protein had low constitutive ligand binding activity (Fig. 4B). Hence, the results suggest that the calf-1 domain is sufficient for the stimulatory action of these mAbs. We also tested the ability of N29 and 8E3 to stimulate ligand binding to constructs containing other partial truncations of the ␣ subunit (Table I). The results suggested that the whole of the calf-1 domain was necessary for the stimulatory action of N29 and 8E3, although it is also likely that any partial truncation of calf-1 may lead to incorrect folding of the domain.
The Epitopes of N29 and 8E3 Map to the Extreme N Terminus of the ␤ 1 Subunit and Are Spatially Overlapping with the Epitope of HUTS-4 in the Hybrid Domain-To gain further insight concerning the mechanism of action of N29 and 8E3 we mapped precisely the epitopes of these mAbs using human/mouse substitution mutations (Table II). Human and murine ␤ 1 differ in sequence at only four positions in the PSI domain (residues Glu 4 , Ser 31 , Pro 56 , and Asp 58 ). Mutation of Ser 31 or Pro 56 and Asp 58 had no effect on the binding of N29 or 8E3. In contrast, substitution of Glu 4 resulted in a complete loss of binding by both mAbs. Hence the epitopes of N29 and 8E3 map to the extreme N-terminal region of the subunit.
The PSI domain is closely associated with the hybrid domain, and outward swing of the hybrid appears to be a central pivot point for affinity regulation because this movement alters the affinity state of the ␤A domain (9,14). The epitopes of HUTS-4 and 15/7 lie on the side of the hybrid domain, and increased exposure of these epitopes correlates with the outward pivoting of the domain (9). We therefore tested whether N29 and 8E3 could influence binding of the HUTS-4 antibody. Surprisingly, we found that both PSI domain mAbs strongly blocked binding of HUTS-4 in a competitive ELISA (Fig. 5A). As expected, the binding of HUTS-4 was blocked completely by both unlabeled HUTS-4 and 15/7 but not by the control mAb K20. In the   FIG. 3. N29 and 8E3 stimulate ligand binding in the absence of the ␤ subunit leg but not the ␣ subunit leg. A, schematic representation of a straightened form of ␣ 5 ␤ 1 (␣ subunit on the left, ␤ subunit on the right) showing the domain structure of the integrin and the major recombinant constructs used in these studies. FL, full-length extracellular domain; TR, truncated extracellular domain containing the head region but lacking the leg region; C1␣ 5 , ␣ 5 subunit truncation containing the calf-1 domain but not the calf-2 domain. B-D, recombinant integrin-Fc fusion proteins were captured onto ELISA plate wells using goat anti-human Fc. Binding of 3Fn6 -10 fragment was measured in the absence of mAbs (Con) or in the presence of the control mAb K20, or mAbs N29, 8E3, or 12G10. B, FL␣ 5 FL␤ 1 ; C, FL␣ 5 TR␤ 1 ; D, TR␣ 5 FL␤ 1 . All experiments were performed in an assay buffer containing 1 mM Mn 2ϩ . *, p Ͻ 0.001; **, p Ͻ 10 Ϫ5 by Student's t test, relative to control; statistical analysis shown only for N29 and 8E3. No ligand binding was observed in an assay buffer containing 2 mM EDTA (data not shown).
converse assay (Fig. 5B), the binding of labeled 8E3 was blocked completely by unlabeled 8E3 or N29, again showing that the epitopes of these mAbs localize to the same region of the PSI domain. More importantly, HUTS-4 (and also 15/7) partially attenuated binding of 8E3. The relatively weak inhibition of 8E3 binding by HUTS-4, compared with that of HUTS-4 binding by 8E3, is probably caused by the higher affinity of 8E3 binding to ␤ 1 (data not shown). In an additional experiment (Fig. 5C), the binding of HUTS-4 (and also 15/7) was strongly perturbed by a Fab fragment of 8E3, whereas binding of JB1A (whose epitope lies on the opposite side of the hybrid domain to that of HUTS-4 and 15/7) was only weakly inhibited. 8E3 Fab did not affect binding of mAbs TS2/16 (against the ␤A domain) or K20 (against the EGF repeats). Taken together, these data indicate that the epitopes of N29 and 8E3 are spatially overlapping with that of HUTS-4. The epitope of HUTS-4 is on the side of the hybrid domain that faces the ␣ subunit ␤ propeller (9). The spatial overlap between the N29/8E3 and HUTS-4 epitopes implies that the epitopes of N29 and 8E3 must lie on the same side of the ␤ subunit as the HUTS-4 epitope, i.e. the side that faces the ␣ subunit. DISCUSSION The role played by the PSI domain in the regulation of integrin function has been particularly uncertain, partly because this domain was not clearly resolved in the ␣ V ␤ 3 structure (6). Here we elucidate the molecular mechanisms involved in the control of ligand binding activity by this region of the ␤ 1 subunit using mAbs N29 and 8E3. Our major findings are as follows: (i) The epitopes of N29 and 8E3 lie the PSI domain, and these mAbs stimulate ligand binding to ␣ 5 ␤ 1 in the presence of Mn 2ϩ or Mg 2ϩ ; (ii) N29 and 8E3 cause long range conformational changes that alter the ligand binding activity of the ␤A domain; (iii) their stimulatory action is dependent on the presence of a portion of the ␣ subunit leg, the calf-1 domain; (iv) the epitopes of N29 and 8E3 map to near to the extreme N terminus of the PSI domain and are closely overlapping with the epitope of HUTS-4 in the hybrid domain. Critically, this evidence suggests that, like HUTS-4, the 8E3 and N29 epitopes lie on the side of the ␤ subunit that faces the ␣.
How Do N29 and 8E3 Stimulate Ligand Recognition?-An examination of the structure of ␣ V ␤ 3 (6) shows that the calf-1 domain is very close to the knee region of the ␤ subunit in the bent state; however, in an extended state these regions would be spatially distant. Hence, the epitopes of N29 and 8E3 are close to calf-1 only in the bent conformation (Fig. 6). Because 8E3 and N29 require the calf-1 domain to stimulate ligand binding, our data suggest that binding of 8E3 and N29 to the FIG. 4. N29 and 8E3 stimulate ligand binding to constructs containing the ␣ 5 subunit calf-1 domain but lacking the calf-2 domain. Recombinant integrin-Fc fusion proteins were captured onto ELISA plate wells using goat anti-human Fc. Binding of 3Fn6 -10 fragment was measured in the absence of mAbs (Con) or in the presence of the control mAb JB1A, or mAbs N29, 8E3, or 12G10. A, C1␣ 5 FL␤ 1 ; B, C1␣ 5 TR␤ 1 . Experiments were performed in an assay buffer containing 1 mM Mn 2ϩ . *, p Ͻ 0.001 by Student's t test, relative to control; statistical analysis shown only for N29 and 8E3. No ligand binding was observed in an assay buffer containing 2 mM EDTA (data not shown).

TABLE I
Analysis of ␣ 5 subunit truncations for constitutive ligand binding activity and for stimulation of ligand binding activity by N29 and 8E3 CHO-L761h cells were transfected with FL␤ 1 -Fc and different truncations of ␣ 5 -Fc. Numbers following ␣ 5 refer to the C-terminal position of the truncation (for example, ␣ 5 603 is truncated at Asp 603 ). Binding of Fc-captured receptors to the 3Fn6 -10 fragment was assessed in the absence or presence N29 or 8E3. Although N29 and 8E3 failed to stimulate ligand binding to ␣ 5 603FL␤ 1 -Fc or ␣ 5 694FL␤ 1 -Fc, 12G10 strongly promoted fibronectin binding to both receptors (data not shown). All constructs reacted equally well with a panel of anti-␣ 5 and anti-␤ 1 mAbs directed against the ␤ propeller and ␤A domains (not shown).

TABLE II
Analysis of N29 and 8E3 reactivity with ␤ 1 PSI domain substitution mutants CHO-L761h cells were transfected with TR␣ 5 -Fc and wild-type or mutant TR␤ 1 -Fc. Cell culture supernatants were analyzed for reactivity with anti-␤ 1 mAbs by sandwich ELISA. Results are expressed as a percentage of binding to wild-type TR␣ 5 TR␤ 1 -Fc and are the mean Ϯ S.D. from a single experiment (four replicate wells), representative of three separate experiments. A value of 0% indicates that mAb reactivity was identical to, or slightly lower than, reactivity with supernatant from mock-transfected cells. All mutants bound well to the anti-␣ 5 mAbs JBS5 and 16, and none of the mutations affected recognition of the 3Fn6 -10 fragment (data not shown). PSI domain may result in a prying apart of the PSI and calf-1 and thereby cause the ␣ and ␤ subunit knees to separate. This splaying apart of the knee regions would destabilize the bent state (in which the legs are together) and favor the extended state (in which the legs are apart) (12,13). Alternatively, because the bent and unbent states are in conformational equilibrium, when bound to a transient conformation in which the legs are separated, 8E3/N29 would obstruct the legs from coming together, thereby maintaining the primed state (Fig. 6). We mapped the epitopes of N29 and 8E3 to the N-terminal region of the PSI domain, including the residue Glu 4 . Our results for N29 appear to be in conflict with previous mapping data, which suggested that the N29 epitope did not lie within residues 1-14 of the ␤ 1 subunit (21). However, this conclusion was based on the inability of a synthetic peptide containing residues 1-14 to inhibit the binding of N29 to the ␤ subunit. Because of the tightly disulfide-bonded structure of the PSI domain it is unlikely that the peptide would be able to reproduce the conformation of the N-terminal region.
It has been reported previously that exposure of the N29 epitope on cell surface ␣ 4 ␤ 1 and ␣ 5 ␤ 1 correlates with the functional status of these integrins (21). N29 expression is low on nonadherent cells but high on spontaneously adherent cells. Furthermore, agents that stimulate cell adhesion, namely Mn 2ϩ and the reducing agent dithiothreitol, increase exposure of the N29 epitope. These findings are consistent with a role for unbending in ␤ 1 integrin function because the proximity of the N29 epitope to the ␣ subunit in the bent state may cause its epitope to be partially masked. Nevertheless, both N29 and 8E3 epitopes appeared to be constitutively expressed on all the recombinant ␣ 5 ␤ 1 -Fc constructs, suggesting that Glu 4 is solvent-exposed under most conditions.
A mechanism of action involving unbending is also consistent with the ability of N29 and 8E3 to stimulate ligand binding both in Mn 2ϩ and in Mg 2ϩ . Although Mn 2ϩ promotes unbending, a considerable proportion of the integrin may remain bent under these conditions (13). The majority of the receptors are likely to be bent in Mg 2ϩ , Mg 2ϩ /Ca 2ϩ , or Ca 2ϩ only (13). Both mAbs were effective at stimulating ligand binding in Mg 2ϩ and Mg 2ϩ /Ca 2ϩ , but not in Ca 2ϩ alone. The failure of N29 and 8E3 to stimulate ligand binding in the presence of Ca 2ϩ is likely to be because occupancy of the MIDAS site of ␤A by Ca 2ϩ is not permissive for ligand binding (8,31).
N29 and 8E3 had only a small stimulatory effect on ligand binding to FL␣ 5 FL␤ 1 -Fc in the presence of Mn 2ϩ . A possible explanation is that, for this construct, nearly all of the integrin may be unbent in Mn 2ϩ because this construct is lacking transmembrane and cytoplasmic tail interactions that help to constrain the native receptor in a low affinity state (3,(32)(33)(34). The ratio of unbent to bent integrin is also likely to vary between the different recombinant ␣ 5 ␤ 1 -Fc constructs. We do not currently have an explanation as to what causes some of these constructs to have low constitutive ligand binding activity, whereas others have high activity. The Fc domain contains a flexible hinge region and probably does not impose any large constraints upon the positions of the ␣ and ␤ subunit legs because cleavage of the Fc regions does not cause low activity constructs to have high activity. 3 The failure of N29 and 8E3 to stimulate ligand binding to constructs lacking the calf-1 domain was probably not caused by the low constitutive activity of these constructs because these mAbs did promote ligand binding to other low activity constructs such as C1␣ 5 TR␤ 1 -Fc (this report) and FL␣ 5 FL␤ 1 -Fc with the ADMIDAS site mutation D138A (30). 4 Furthermore, both antibodies stimulated ligand binding to purified ␣ 5 ␤ 1 under conditions where the integrin had low constitutive activity.
The binding of N29 and 8E3 to ␣ 5 ␤ 1 caused long range conformational changes that affected epitopes in the head and leg regions, i.e. at opposite ends of the molecule. Such global effects on integrin conformation strongly support a mechanism of action that involves the large scale shape changes that accompany unbending. Supporting evidence that unbending is involved in the regulation of ␤ 1 integrins comes from the increased exposure of other epitopes in the leg regions when the integrin is in a high affinity state (e.g. 9EG7 (35) and AG89 (36)). Exposure of these epitopes is predicted to be low in the bent state of the integrin where the legs are together but high in the extended state where the legs are apart (12). Changes in the fluorescence resonance energy transfer of cell surface ␣ 4 ␤ 1 have also been measured upon receptor activation (37), and there is a strong correlation between the affinity of ligand binding and the degree of receptor extension away from the cell membrane (38).
What Role Does the PSI Domain Play?-Our data suggest that the PSI domain is likely to form part of an interface with the ␣ subunit which is important for restraining the integrin in a low affinity state. This hypothesis is supported by several previous studies of ␤ 2 and ␤ 3 integrins. AP5 is a stimulatory anti-␤ 3 mAb whose epitope maps to residues 1-6 of the PSI domain (18) (i.e. in the region equivalent to N29 and 8E3). Significantly, the binding of AP5 is suppressed by Ca 2ϩ (which favors the bent state) and enhanced by ligand binding (which favors the extended state). These results can be readily explained if the epitope of AP5 is masked in the bent state because of its proximity to the ␣ subunit. Mutation of Cys residues in the ␤ 3 PSI domain has also been shown to result in priming of ␣ IIb ␤ 3 (39). Disruption of disulfide bonds may per-turb the structure of the domain and thereby weaken interactions with the ␣ subunit knee region, resulting in a shift toward the extended state. Conformational changes have also been detected in the region near the interface between the PSI and hybrid domains in an activated form of ␣ IIb ␤ 3 , using protease digestion (40). The PSI domain of the ␤ 3 subunit is also a key site of drug-induced neoepitopes on ␣ IIb ␤ 3 (19), suggesting that there are changes in the exposure of this domain upon ligand recognition. In further support of the role of the extreme Nterminal region of the PSI domain in stabilizing the bent form, it has been shown that Thr 4 in ␤ 2 (equivalent to Glu 4 in ␤ 1 ) is one of the residues involved in restraining activation of ␣ X ␤ 2 (41).
In addition, our data also suggest that the PSI domain is important for the transfer of conformational changes from the ␤ knee to the head region (which includes ␤A). Unbending is proposed to regulate ligand binding mainly through an effect on the hybrid domain (13): in the bent form the position of the hybrid domain is fixed through interactions with other domains; in the extended form the hybrid domain is released from these constraints, and an outward swing of this domain causes shape shifting in the ␤A domain, mainly in the region of the ␣ 1 and ␣ 7 helices (8,9,42), leading to the high affinity state of ␤A. A study of the head region of ␣ 5 ␤ 1 by electron microscopy showed that, upon fibronectin binding, the PSI domain and hybrid domain swing outward en bloc (43). Hence, a movement of the PSI domain is likely to be linked directly to hybrid domain movement and vice versa. Therefore, an outward displacement of the PSI domain caused by an opening of the knees is likely also to cause hybrid domain swing and thereby increase the ligand binding activity of ␤A.
We anticipated that binding of 8E3 and N29 to the ␤ subunit would cause hybrid domain movement, which could be detected by increased binding of HUTS-4 (9). Unexpectedly, however, both mAbs strongly inhibited HUTS-4 binding. This inhibition appears to be the result of the spatial overlap between the epitopes of these three mAbs; hence we could not test directly whether the binding of N29 and 8E3 to ␣ 5 ␤ 1 results in hybrid domain movement. However, we were able to demonstrate that the PSI domain mAbs increased binding of mAb 12G10, showing that N29 and 8E3 alter the conformation of ␤A. Hence, it is very likely that 8E3 and N29 do transfer conformational changes to ␤A via hybrid domain movement. Although we cannot rule out the possibility that the transduction of shape change takes place through the ␣ subunit, the rigid nature of the thigh domain and ␤-propeller (6) disfavors such a mechanism.
PSI domains are also found in a wide range of other proteins and, as in integrins, are often associated with an immunoglobulin fold (44,45). PSI domains are likely to be important modules for transfer of conformational changes in other receptors, for example, the long range shape changes in plexins (46).
During preparation of this manuscript, crystal structures of the PSI domain in ␣ V ␤ 3 (47) and ␣ IIb ␤ 3 (48) were published. In the bent state of the integrin (␣ V ␤ 3 structure) the PSI domain is positioned at the bend of the ␤ subunit knee and is very close to the calf-1 domain. Significantly, as our data suggested, residue 4 of the ␤ 3 PSI domain (equivalent to Glu 4 in ␤ 1 ) lies on the side of the domain that faces calf-1. In the ligand-activated state of the integrin (␣ IIb ␤ 3 structure) the hybrid domain and PSI domain are swung away from the ␣ subunit, and the knee regions are separated by a large distance of about 70 Å. Hence, both structures give strong support to the mode of action of N29 and 8E3 proposed here. These structures also suggest that the PSI domain is linked rigidly to the hybrid and acts like a stiff connecting rod that transmits and amplifies conformational changes to and from the head regions.
In summary, we have shown that the PSI domain mAbs N29 and 8E3 stimulate ligand binding through a mechanism that requires the ␣ subunit calf-1 domain and is likely to involve a separation of the ␣ and ␤ subunit knees. Our results indicate that regulation of ␤ 1 integrins, like ␤ 2 and ␤ 3 integrins, involves unbending, suggesting that this movement is likely to provide a general mechanism for control of ligand recognition. Our findings are not consistent with an alternative model of integrin affinity regulation that does not involve unbending (49); however, it is likely that many intermediate states of receptor extension exist (5). In the future, it will be important to define the precise molecular basis of the movements that occur in knee regions. This knowledge should aid in the development of novel integrin antagonists that do not cause unbending and thereby circumvent autoimmune side effects (50).