The integrin alpha1 A-domain is a ligand binding site for collagens and laminin.

The integrin alpha1beta1 is a cell surface receptor for collagens and laminin. The alpha1 subunit contains an A-domain, and the A-domains of other integrins are known to mediate ligand binding. To determine the role of the alpha1 A-domain in ligand binding and the extent to which it reproduced the ligand binding activity and specificity of the parent molecule, we produced recombinant alpha1 A-domain and tested its ability to bind collagens and laminin. In solid phase assays, the A-domain from alpha1 was found to bind to collagen I, collagen IV, and laminin in a largely cation-dependent manner. The alpha2 A-domain, from the alpha2beta1 integrin, also bound to these ligands, but the binding hierarchy differed from that seen for alpha1. This is the first demonstration of laminin binding by A-domains. Specificity of A-domain-ligand binding was further investigated using the triple-helical proteolytic fragment of collagen IV, CB3, and its subfragments, F1 and F4. alpha1 A-domain bound to all three fragments, while the alpha2 A-domain bound CB3 less well and exhibited little binding to F1 and no binding to F4. These differences mirror previous reports of distinct integrin binding sites in collagen IV and for the first time identify a limited proteolytic fragment of a ligand that contains integrin A-domain binding activity. To gain insight into the contribution that the A-domain makes to ligand binding within the whole integrin heterodimer, we measured binding constants for A-domain-collagen interactions using surface plasmon resonance biosensor technology. The values obtained were similar to those reported for intact integrin binding, suggesting that the A-domain is the major collagen binding site in the alpha1beta1 and alpha2beta1 integrins.

Integrins are a family of ␣␤ heterodimeric cell surface receptors, responsible for cell-cell and cell-extracellular matrix interactions. The specificity and regulation of these interactions is critical to many biological processes, including embryonic cell migration, wound healing, and the immune response (1). The integrin family contains at least 16 ␣ subunits, seven of which contain an ϳ200 amino acid inserted domain in their N-terminal region (I or A-domain) (2,3). This domain is homologous to the von Willebrand factor A-domain, a module also found in a number of other membrane, plasma, and matrix proteins (2,4).
The integrin ␣1␤1 is a receptor for collagen I, collagen IV, and laminin (16). These interactions, like all integrin-ligand binding events, are cation-dependent and require Mg 2ϩ or Mn 2ϩ ; Ca 2ϩ , however, does not support binding (17). A similar pattern of ligand binding is found for the closely related integrin ␣2␤1, which also interacts with collagen I, collagen IV, and laminin (16). ␣1␤1 and ␣2␤1 have, however, been shown to differ in their relative affinities for ligands, with ␣1␤1 binding collagen IV and laminin with higher affinity (18). Integrins ␣1␤1 and ␣2␤1 also bind to a proteolytic fragment of collagen IV, CB3 (19), but are believed to have distinct binding sites within this fragment (18).
To investigate the role of the ␣1 A-domain in ligand binding, we have generated recombinant A-domain in a bacterial expression system and tested its binding to a range of integrin ligands. The specificity of A-domain ligand binding has also been examined by comparing recombinant ␣1, ␣2, and ␣M integrin A-domains. We find that integrin ␣1 A-domain binds to collagen I, collagen IV, and laminin in a saturable, concentration-dependent manner and that the binding can be inhibited by an anti-functional anti-␣1 monoclonal antibody (mAb). 1 ␣2 A-domain also binds these ligands, while ␣M A-domain does not. The ␣M A-domain does, however, bind to fibrinogen, while the ␣1 and ␣2 A-domains do not. This binding is largely cationdependent, and real time binding studies using surface plasmon resonance (SPR) analysis suggests that cation, A-domain, and ligand may form a ternary complex. By using proteolytic fragments of collagen IV, the ligand specificity of the closely related ␣1 and ␣2 A-domains has been compared. The results indicate that recombinant A-domains retain the specificity reported for intact receptors.
Production of Recombinant Integrin A-domains-The production of recombinant ␣2 A-domain has already been described (12). The ␣1 and ␣M A-domains were produced in a similar manner. DNA coding for the A-domains from integrins ␣1 and ␣M was produced by reverse transcriptase-polymerase chain reaction (RT-PCR). Total RNA prepared from the A375 human melanoma cell line (for ␣1) or human buffy coat lymphocytes (for ␣M) was a gift of L. J. Green (University of Manchester, UK). First strand cDNA was generated using a 3Ј primer spanning the predicted end of the integrin A-domains and incorporating a SalI site (␣1, 5Ј-TTTGTCGACTCAGGCTTCCAGG GCAAATATTCTTTCTC-C-3Ј; ␣M, 5Ј-TTTGTCGACTCAACCCTCGATCGCAAAGATCTTCTCC-CG-3Ј). Polymerase chain reaction (PCR) amplification of this cDNA was then carried out using the thermostable proof-reading DNA polymerase Pfu (Stratagene Ltd., Cambridge, UK) with the above 3Ј primer and a 5Ј primer designed to produce DNA coding for 17 amino acids preceding the predicted start of the integrin A-domain (21) and including a BamHI site (␣1, 5Ј-TTTGGATCCGTCAGCCCCACATTTCAAGT-CGTGAATTCC-3Ј; ␣M, 5Ј-TTTGGATCCAACCTACGGCAGCAGCCCC-AG-3Ј). 50 cycles, each consisting of 1 min at 94°C, 1 min at 55°C, and 2.5 min at 72°C, were carried out. PCR products, of the correct M r , were then excised from a 1% agarose gel. After digestion with BamHI and SalI, the products were ligated into pUC119 and used to transform Escherichia coli strain DH5␣FЈ. A-domain sequences from transformants were sequenced by the dideoxy chain termination method of Sanger (22) and compared with the published sequences (␣1 and ␣M, see Refs. 23 and 24, respectively). The sequenced DNA was then subcloned into the expression vector pGEX-4T3 (25) (Pharmacia, Milton Keynes, UK). Transformants were screened, and protein was produced as described for ␣2 (12) except that dithiothreitol was not required for GST removal from ␣1 and ␣M A-domains. For experiments using fusion protein, a single glutathione affinity step was performed and thrombin digestion was not carried out. The recombinant A-domains produced were 214 (␣1), 224 (␣2), and 211 (␣M) amino acids long. N-terminal amino acid sequencing of thrombin-cleaved ␣1 and ␣M A-domains confirmed the predicted starting sequence.
Biotinylation of Proteins-A-domain-GST fusion proteins, collagen I, and fibrinogen were biotinylated as follows. Protein was diluted to 1 mg/ml in 150 mM NaCl, 10 mM Tris-HCl, pH 7.5 (TBS), and dry sulfo-N-hydroxysuccinimido-biotin (Pierce, Chester, UK) was added to give a ratio of 1:1 (w/w) protein:biotin. The mixture was incubated for 1 h at room temperature and then dialyzed against TBS (or 0.1 M acetic acid for collagen I) to remove unincorporated biotin.
A-domain Binding Assays-Assays measuring binding of soluble Adomains to immobilized ligand were carried out essentially as above, but plates were coated with ligand (usually at 30 g/ml) instead of fusion protein and binding of biotinylated A-domain fusion protein was measured. For antibody detection of unbiotinylated A-domain fusion protein binding to ligand, the assay was performed as above up to the streptavidin step where 5 g/ml anti-GST antibody in TBS, 1 mg/ml BSA, 1 mM MnCl 2 was added for 1 h instead. Wells were washed three times with TBS, 1 mM MnCl 2 , and 1:500 peroxidase-linked goat antirabbit antibody (Sigma) in TBS, 1 mg/ml BSA, 1 mM MnCl 2 was added for 45 min. Wells were washed three times with TBS, 1 mM MnCl 2 , and binding was detected with ABTS as above.
Surface Plasmon Resonance-The kinetic parameters (apparent association and dissociation rate constants k a and k d , respectively) and the apparent equilibrium constant (K D ) for A-domain binding to collagen I and IV were measured using SPR on a BIAcore TM (Biacore, St. Albans, UK). This biosensor device was used in accordance with the manufacturer instructions. Briefly, collagen I or IV was covalently coupled via primary amine groups to the dextran matrix of a CM5 sensor chip. A-domain solution was flowed over the chip at 5 l/min, and binding was measured as a function of time. TBS, 1 mM MnCl 2 was used as running buffer throughout, and injections of TBS, 10 mM EDTA were used to remove bound A-domain, regenerating the surface for further binding experiments. The curve fitting software, BIAevaluation, was used to fit these results to the simple first order interaction model A ϩ B N AB. This produced values for k a and k d , allowing the apparent equilibrium constant K D to be derived from k d /k a . To measure binding in the presence of divalent cations other than Mn 2ϩ , a 10-min injection of TBS, 10 mM EDTA was passed over the chip (to remove any residual Mn 2ϩ ) immediately prior to injection of A-domain in the specified cation.

Generation of Integrin ␣1 and ␣M A-domain cDNA-
The reverse transcriptase-polymerase chain reaction was used to generate integrin ␣1 A-domain cDNA from A375 human melanoma cell line RNA and integrin ␣M A-domain cDNA from human buffy coat lymphocyte RNA. After cloning into pUC119, sequencing revealed that the ␣M A-domain cDNA matched the published sequence (24). Sequencing of ␣1 A-domain cDNA revealed three differences from the published sequence (23): 1) an inserted Thr at position 502; 2) a deleted Ala at position 511, which put the sequence back in frame after the earlier insertion, and 3) a Cys to Thr mutation at position 674. These differences result in two changes in the predicted amino acid sequence of the protein, a lysine to glutamate at position 170 and a threonine to isoleucine at 228, numbered from the start of the mature polypeptide. Repeating the RT-PCR using two different A375 RNA samples produced the same sequence, and RT-PCR using human smooth muscle RNA also gave the same sequence. As all PCR reactions used a proof-reading DNA polymerase and produced the same sequence from different RNA samples, we believe that it represents an accurate human ␣1 integrin sequence.
Expression of Integrin ␣1 and ␣M A-domains in E. coli-Integrin A-domain cDNAs were cloned into the pGEX-2T3 expression vector and used to transform E. coli. After induction, transformants expressed GST-A-domain fusion proteins of ϳ50 kDa as reported for the ␣2 A-domain (12). Fusion proteins were purified on glutathione-agarose columns and were either used directly or cleaved with thrombin and passed through a second glutathione-agarose column to produce purified 25 kDa A-domain. N-terminal sequencing of purified A-domains showed that cleavage had occurred at the expected site. SDS-PAGE showed the recombinant protein was at least 90% pure (data not shown).
All A-domains reacted specifically with previously characterized mAbs: the ␣2 A-domain bound a number of anti-␣2 integrin mAbs, including Gi9, an anti-functional mAb. The ␣1 A-domain was recognized by the anti-functional anti-␣1 integrin mAb 5E8D9 in ELISA, and ␣M A-domain bound the antifunctional anti-␣M mAb 44 (data not shown).
Collagen I Binds to ␣1 and ␣2 A-domains but Not to ␣M A-domain-The integrins ␣1␤1 and ␣2␤1 are reported to be collagen receptors, while integrin ␣M␤2 binds non-collagenous ligands. To investigate the role of A-domains in binding collagen, the binding of biotinylated collagen I to A-domain fusion proteins was measured. Fig. 1 shows that the A-domain fusion proteins from ␣1 and ␣2 integrins support dose-dependent, saturable binding of collagen I, while the ␣M A-domain exhibits very little binding. Collagen I shows a higher maximal binding to the ␣1 A-domain than to the ␣2 A-domain. The observed differences in binding were not due to different coating efficien-cies of the GST-fusion proteins as the coating concentration chosen (10 g/ml) gave very similar, almost maximal, coating of all three fusion proteins to the plate as measured by anti-GST antibody in ELISA (data not shown). The differences between ␣1 and ␣2 A-domains may instead reflect differences in the amount of correctly folded A-domain in the different samples rather than a difference in the number of binding sites per A-domain. Binding of collagen I was dependent on the conformation of the triple helix, as heat denaturation of the collagen I at 50°C for 30 min inhibited its binding to ␣1 and ␣2 A-domains (data not shown).
Fibrinogen Binds the ␣M A-domain Specifically in a Cationdependent Manner-To confirm the specificity of the A-domains for their ligands, we investigated binding of the known ␣M␤2 ligand fibrinogen to all three A-domains. Biotinylated fibrinogen bound to the ␣M A-domain in the presence of 1 mM Mn 2ϩ but did not bind to ␣1 or ␣2 A-domains, above the GST control (Fig. 2). Interestingly the binding to GST was higher than to BSA. Fibrinogen binding to ␣M A-domain was reduced to GST levels by 5 mM EDTA indicating that the binding was cation-dependent. The binding of biotinylated fibrinogen to ␣M A-domain was also reduced to the levels of the GST control by the anti-functional anti-␣M mAb 44 (data not shown).
Cation-dependence of Collagen I Binding to A-domains-Having shown that the ␣M A-domain interaction with fibrinogen required divalent cations, we investigated the effect of different divalent cations on biotinylated collagen I binding to ␣1 and ␣2 A-domains. Fig. 3A shows that collagen binding to ␣1 A-domain was completely inhibited by EDTA and, therefore, requires divalent cations. However, the nature of the divalent cation had little effect as Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ all supported similar levels of binding. Of the three cations, Mn 2ϩ supported the highest levels of binding. This cation profile does not match that reported for the whole integrin, where Mg 2ϩ was required for ligand binding and Ca 2ϩ did not support binding (17). Collagen binding to ␣2 A-domain was inhibited by EDTA (Fig.  3B); however, in this case Ca 2ϩ supported only very little binding, and Mg 2ϩ and Mn 2ϩ produced identical levels of collagen binding. This is in agreement with previously published data (12). Thus, while the A-domains from integrins ␣1 and ␣2 both bind collagen I, they differ in their cation specificities.
The binding of soluble A-domain to immobilized collagen was also investigated. This was carried out using biotinylated Adomain fusion proteins or by antibody detection of the GST moiety on the A-domain fusion protein. Both methods demonstrated concentration-dependent saturable binding of ␣1 A-domain to collagen I; however, EDTA only partially inhibited this interaction (Fig. 4A). Inhibition was similar in both biotinylation and antibody detection assays, with slightly more inhibition seen for the antibody detection assays (data not shown). In both assays, inhibition was maximal at lower A-domain protein concentrations. Similar results were obtained for ␣2 A-domain binding to collagen I (data not shown).
The discrepancy between cation-dependence of collagen binding to immobilized A-domain and A-domain binding to immobilized collagen is difficult to explain. The interaction appeared to be specific as, under these conditions, binding was still dependent on the triple-helical conformation of collagen and the native A-domain structure, as heat denaturation of either component strongly inhibited binding (Fig. 4B, and data not shown).
To address the apparent discrepancies in solid phase assay results, the cation dependence of the A-domain-collagen interaction was further investigated using SPR measurements on a BIAcore. Collagen I was covalently immobilized onto the sensor chip, and binding of A-domain was measured. These measure-ments were performed for both ␣1 and ␣2 A-domain fusion proteins at a range of concentrations (3-100 g/ml). Fig. 5 shows a typical sensorgram of ␣1 A-domain binding to collagen I in the presence of 1 mM Mn 2ϩ . Association and dissociation phases are marked and removal of bound A-domain with 10 mM EDTA is shown. EDTA injection removed 86 Ϯ 2% of the bound ␣1 A-domain (mean Ϯ S.E.; n ϭ 33) and 82 Ϯ 8% of the bound ␣2 A-domain (mean Ϯ S.E.; n ϭ 10), indicating that the presence of divalent cation is required for maintenance of the bound complex, not simply for A-domain-collagen binding. GST failed to bind collagen in these assays. Binding of fusion proteins in the presence of different cations was also investigated, and, as already seen for biotinylated collagen binding to A-domains, 1 mM Mn 2ϩ , Mg 2ϩ , and Ca 2ϩ all supported ␣1 A-domain binding while only Mn 2ϩ and Mg 2ϩ supported ␣2 A-domain binding. No binding occurred without addition of divalent cations (data not shown). This supports the ligand-binding assay results that demonstrated a cation-dependent interaction.
␣1 A-domain Binds to Collagen IV and Laminin-In addition to collagen I, ␣1␤1 is also reported to bind collagen IV and laminin (16). The binding of collagen IV could not be studied using immobilized A-domains because collagen IV is a poor biotinylation substrate. 2 Antibody detection of ␣1 A-domain fusion protein binding to immobilized collagen I, collagen IV, and laminin showed that the ␣1 A-domain bound all these ligands in a concentration-dependent, saturable manner (Fig.  6A). The ␣1 A-domain exhibited very similar levels of binding to collagen I and collagen IV, with laminin bound to a lesser extent. In addition, CB3, an integrin-binding proteolytic fragment of collagen IV, supported a similar level of binding to intact collagen IV although binding was less at lower A-domain concentrations. To test the specificity of the A-domain binding, 2 D. A. Calderwood, unpublished observation.

FIG. 4. ␣1 A-domain binding to immobilized collagen I.
A, microtitre plates were coated with substrate at 30 g/ml and blocked with BSA, and GST-A-domain fusion protein was added. After incubating for 3 h at 37°C, unbound fusion protein was washed off. Binding was detected using anti-GST antibody, a peroxidase-conjugated goat antirabbit IgG secondary antibody, and ABTS. Binding of ␣1 A-domain fusion protein to collagen I in 1 mM MnCl 2 (q) and 5 mM EDTA (f) and to BSA in 1 mM MnCl 2 (E). Shown are the binding of GST in 1 mM MnCl 2 to collagen I (å) and BSA (ç). Data are mean Ϯ S.E. and n Ն 9 from four experiments normalized for 30 g/ml ␣1 A-domain binding to collagen I. B, alternatively biotinlyated A-domain fusion protein was used, and binding was detected using ExtrAvidin-peroxidase and ABTS. The effect of heat denaturation of collagen I (hdCI), or ␣1 A-domain (hd A-dom.) by heating to 50°C for 30 min, on ligand binding was investigated. Results are mean Ϯ S.E. and n ϭ 9 from three experiments. Background binding to BSA has been subtracted and results normalized for native ␣1 A-domain binding to native collagen I in the presence of 1 mM MnCl 2 . Following the initial regeneration, most of the binding was removed, and further injections of EDTA did not remove the residual binding. The large change in response during regeneration is due to differences in the refractive index of the buffers.
we investigated cation dependence and antibody inhibition at low A-domain concentrations. Fig. 6B shows that 0.2 g/ml of biotinylated ␣1 A-domain bound collagen I, collagen IV, laminin, and CB3 and that binding was inhibited by EDTA and the anti-␣1 integrin mAb 5E8D9.
Collagen IV and Laminin Are Ligands for the ␣2 A-domain-The integrin ␣2␤1 is also a receptor for laminin and collagen IV, so we investigated the binding of ␣2 A-domain to these ligands and compared it with ␣1 A-domain. Fig. 7 shows that ␣2 A-domain bound to collagen IV and laminin; however, in this case, collagen I was a better ligand than collagen IV and laminin. Furthermore, CB3 supported much less binding than collagen IV, unlike the case with ␣1. Thus, while ␣1 and ␣2 A-domains bound the same range of ligands, comparison of Figs. 6A and 7 indicate that the relative binding to these ligands differed between the two A-domains. GST showed only background levels of binding to laminin, collagen I, and collagen IV (data not shown).
␣1 and ␣2 A-domain Binding to Collagen IV Fragments-Proteolytic fragments of collagen IV were used to further investigate the differences between ␣1 and ␣2 A-domains and the integrin binding sites in collagen IV. As described above, both ␣1 and ␣2 A-domains bound to CB3, which is consistent with results obtained for whole integrins in solid phase binding assays and cell attachment assays (12,18,19). Further proteolysis of this fragment produces four smaller fragments, F1-F4, and investigation of integrin binding sites in these fragments demonstrated that both ␣1␤1 and ␣2␤1 were able to bind to F1 but only ␣1␤1 bound to F4 (18). F4 lacks the N-and C-terminal regions present in CB3 and F1, which are thought to contain the ␣2␤1 recognition site. Antibody detection of fusion protein binding demonstrated that ␣1 A-domain bound both F1 and F4 in a concentration-dependent saturable manner, while ␣2 A-domain bound only poorly, if at all, to F1 and not at all to F4. This binding was inhibited by 5 mM EDTA (data not shown).

Measurement of Apparent Binding Constants for A-domain Collagen Interactions-Binding of ␣1 and ␣2
A-domain fusion proteins to collagen I and IV was investigated on a BIAcore. BIAevaluation, the kinetic analysis curve-fitting software supplied with the BIAcore, was used to determine the k a and k d values. Prior to fitting the association and dissociation phases of the curve, the binding of A-domain to a blank sensor chip was measured and subtracted from the binding curve. Binding to the uncoated surface was negligible. Binding of a range of fusion protein concentrations (3-100 g/ml) to collagen I and IV-coated chips was measured. Covalent coupling of collagen I to the chip was much more efficient than for collagen IV, and consequently, signals obtained for binding to collagen I were higher than those to collagen IV. Binding of ␣2 A-domain to collagen IV could be detected using the BIAcore; however, signals were too low to allow accurate curve fitting. Values obtained are shown in Table I. The BIAevaluation software provides a number of statistical parameters to judge the fitting of the binding model to the experimental data. These indicated that a simple association model provided a good approximation FIG. 6. ␣1 A-domain binding to immobilized collagen IV and laminin. A, binding of ␣1 A-domain fusion protein to collagen I (q), collagen IV (å), laminin (f), CB3 (ç), and BSA (E) was measured in antibody detection assays. Ligands were coated at 10 g/ml. Data are mean Ϯ S.E. and n Ն 6 from four experiments normalized for 30 g/ml ␣1 A-domain binding to collagen I. B, A-domain binding assays were used to measure binding of biotinylated fusion protein (0.2 g/ml) to immobilized ligands coated at 10 g/ml in the presence of 1 mM MnCl 2 , 1 mM MnCl 2 with 10 g/ml 5E8D9, 1 mM MnCl 2 with 10 g/ml 16B4, or 5 mM EDTA. Data are mean Ϯ S.E. and n Ն 9 from three experiments. Background binding to BSA has been subtracted from each column, and results are normalized for binding to collagen I in the presence of 1 mM MnCl 2 . CI, collagen I; CIV, collagen IV; Lam, laminin.

FIG. 7. A-domain binding to different ligands. Binding of ␣2
A-domain fusion protein in 1 mM MnCl 2 to collagen I (q), collagen IV (å), laminin (f), CB3 (ç), and BSA (E) was measured in antibody detection assays. Ligands were coated at 10 g/ml. Data are mean Ϯ S.E. and n Ն 6 from four experiments normalized for 30 g/ml ␣2 A-domain binding to collagen I. to the experimental data; however, residual plots, comparing the fitted data with experimental results, indicated that a more complex interaction may occur. That the kinetic constants produced provide a good measure of the molecular interactions involved was shown by comparison of a simulated binding curve, produced using the calculated constants, with the experimental results (Fig. 8). This demonstrated that both curves were very similar. The binding curves obtained for A-domain binding to immobilized collagens in solid phase assays can also be fitted to produce apparent K D values for the interaction (26,27). These values are shown in Table I and exhibit generally good agreement with those obtained using the BIAcore. DISCUSSION We have produced a recombinant A-domain from the ␣1 integrin and compared its ligand binding characteristics with recombinant A-domains from the ␣2 and ␣M integrins. Our key findings are that (i) the ␣1 integrin A-domain is a largely cation-dependent ligand binding domain, (ii) the ␣1 and ␣2 A-domains show similar but distinct ligand binding specificities, and (iii) measurement of binding affinities for A-domaincollagen interactions produces values comparable with those reported for integrin-collagen interactions, suggesting that Adomains are the major collagen binding sites in integrins.
Recombinant integrin A-domains and solid phase ligandbinding assays have been used to investigate A-domain-ligand interactions. The results showed that recombinant A-domains from ␣1 and ␣2 integrins bind collagen I, collagen IV, laminin, and the collagen IV fragment CB3, but not the ␣M␤2 ligand fibrinogen. This is the first direct demonstration that the ␣1 A-domain is a ligand-binding module, and the first observation of laminin binding by A-domains. Interestingly, while the ␣1 and ␣2 A-domains were found qualitatively to bind the same ligands, their binding profiles were quantitatively different, and their binding sites within collagen IV appeared to be separate. The specificity of these interactions was confirmed by demonstrating the requirement for native structure of both ligand and A-domain and through antibody inhibition studies.
Published reports of the ligand specificity of ␣1␤1 and ␣2␤1 match these findings with isolated A-domains. Thus, cell binding, antibody blocking, affinity chromatography, and solid phase assays have variously been employed to demonstrate ␣1␤1 binding to collagen types I, III, IV, and VI and to laminin (5, 17, 18, 28 -31). In addition, the CB3 fragment of collagen IV and fragments F1 and F4 of CB3 have been shown to contain the ␣1␤1 binding site (18,19). The ␣2␤1 integrin binds a similar range of ligands (16); however, the affinity of interaction with collagens is different from that of integrin ␣1␤1, and ␣1␤1 appears to bind laminin better than ␣2␤1 (18). The binding site for ␣2␤1 in collagen IV has also been localized to the CB3 fragment (19); however, the ␣1␤1 and ␣2␤1 binding sites are apparently separate (18). The A-domains thus mimic almost all of the ligand binding activity of the intact ␣1␤1 and ␣2␤1 integrins and appear to be key ligand-recognition mod-ules within the intact heterodimer.
To characterize the A-domain-collagen interactions further and address the relative contribution of the A-domain to integrin-ligand binding affinity, the kinetics of binding were measured using SPR. These results showed that the ␣1 A-domain has a higher affinity for collagens than does the ␣2 A-domain. Comparison of apparent binding affinities produced from analysis of solid phase binding data indicated a similar pattern of binding; however, the differences between ␣1 and ␣2 binding were less notable. The apparent binding constants for ␣1␤1 binding to collagen IV have been measured using both BIAcore and solid phase binding and inhibition assays. 3 These techniques produced K D values in the range of 1-5 nM, which is in good agreement with the figures obtained for ␣1 A-domain binding to collagen IV (6 -34 nM; Table I). Kern et al. (18) have obtained similar values for ␣1␤1 binding to CB3; however, this varied from 1 to 30 nM depending on the divalent cations present. ␣2␤1 binding to CB3 showed similar affinity but was more sensitive to cations (1-110 nM) (18). We measured the affinity of ␣2 A-domain for collagen IV as 115 nM. Taken together, these findings indicate that the binding affinity of ␣1 A-domain and ␣1␤1 binding to collagen IV are similar, suggesting that the A-domain is the major ligand binding site in the integrin. The variation in affinity reported for the intact receptor is dependent on divalent cation and suggests that the bind-  8. Comparison of experimental results for ␣1 A-domain fusion protein binding to collagen I with a simulated results curve generated using calculated k a and k d values. The experimental sensorgram (solid line) was recorded using a flow rate of 5 l/ml and a 10-min injection of 0.6 M ␣1 A-domain. The simulated sensorgram (dotted line) was modelled using the experimental parameters and a k a value of 4 ϫ 10 3 M Ϫ1 s Ϫ1 and k d value of 8 ϫ 10 Ϫ5 s Ϫ1 . To allow comparison of the curves, the signal prior to injection was adjusted to zero, and injection was considered to start at time 0.
ing site can be regulated by different cations. As discussed below, this may suggest allosteric regulation of the A-domain by cation-binding regions lying outside the A-domain. As the recombinant A-domain is free of this regulation, it may account for the variation between K D values measured in the presence of Mn 2ϩ for the A-domains and those reported for whole integrins. With this in mind, it should be noted that while the simple association model closely approximates the BIAcore data, some non-random variation from the model suggests that a more complex interaction may occur; this may be due to conformational changes in the A-domain.
Integrins are known to require divalent cations for ligand binding, and a number of regions have been proposed to act as cation-binding sites. The crystal structures of the ␣M and ␣L A-domains have now been solved and show a single divalent cation-binding site at one end of the domain (4,32). While it is accepted that intact integrins require cations for ligand binding, cation dependence, independence, and partial dependence have all been reported for isolated A-domain binding to ligands (11,12,33,34). Results reported here show largely cation-dependent binding; however, some variation in cation-dependence was observed. Binding of soluble A-domain to immobilized collagen is only partially cation-dependent, while binding of biotinylated collagen to immobilized A-domain is completely cation-dependent. BIAcore analysis of A-domain-collagen binding showed an absolute requirement for divalent cations, and binding that had already taken place could be reversed with EDTA, suggesting that there is a requirement for cation to remain bound during the integrin-ligand interaction. This suggests that divalent cations are normally required for A-domain ligand binding; however, some cation-independent binding may be seen due to non-native conformations of the recombinant A-domains.
The exact role of divalent cations in ligand binding remains unclear as it is difficult to determine whether the cation binding produces a ligand binding conformation in the A-domain or is itself required as a bridge between ligand and A-domains. Data from mutagenesis and peptide binding studies (14,11) suggest that cation is not absolutely required for A-domainligand binding but that it normally regulates integrin-ligand binding. Qu and Leahy (35) have recently shown that the crystal structures of the recombinant ␣L A-domain in the presence and absence of divalent cation are very similar, suggesting that the cation-dependence of ligand binding is not due to stabilization of a ligand binding conformation. The presence or absence of divalent cation does, however, have a profound effect on the charge distribution on the cation binding face of the A-domain. As epitope mapping and mutagenesis studies suggest that the ligand binding and cation binding sites are located on the same face of the molecule, surface charge might account for the cation requirement of ligand binding. Our finding that cation was required to maintain ligand binding is consistent with both this and a cation bridge model. In some recombinant A-domains, non-native structures may mean that even in the absence of divalent cation the conformation and charge distribution are such that ligand binding can occur.
In the intact integrin, the situation is further complicated by the presence of other cation-binding sites in the ␣ and ␤ subunits. Cation regulation of the whole integrin does not always match that seen in the isolated A-domain. For example, it is reported that Ca 2ϩ will not support collagen binding by ␣2␤1, or ␣1␤1; indeed, Ca 2ϩ may actually inhibit Mg 2ϩ -induced col-lagen binding (18,17). We have shown that, while Ca 2ϩ will support only low levels of collagen binding to ␣2 A-domain, it will allow binding to ␣1 A-domain. It is possible that, while the A-domain in the intact ␣1␤1 will bind Ca 2ϩ , potentially permitting ligand binding, Ca 2ϩ binding at other sites on the integrin normally precludes binding.
In conclusion, we have demonstrated that the A-domain from ␣1 integrin can bind ligands and that laminin is a ligand for A-domains. We demonstrate that distinctions between ␣1␤1 and ␣2␤1 binding to collagen IV are also observed with isolated A-domains and that the affinity of A-domain binding to collagen is similar to that reported for whole integrin binding. Finally, we show that A-domain-ligand binding is largely divalent cation-dependent and suggest that cation, ligand, and A-domain form a ternary complex.