The extracellular calcium‐binding domain (positions 138–286) of the matrix protein BM‐40 possesses a binding epitope of moderate affinity for several collagen types. This epitope was predicted to reside in helix αA and to be partially masked by helix αC. Here we show that deletion of helix αC produces a 10‐fold increase in collagen affinity similar to that seen after proteolytic cleavage of this helix. The predicted removal of the steric constraint was clearly demonstrated by the crystal structure of the mutant at 2.8 Å resolution. This constitutively activated mutant was used to map the collagen‐binding site following alanine mutagenesis at 13 positions. Five residues were crucial for binding, R149 and N156 in helix αA, and L242, M245 and E246 in a loop region connecting the two EF hands of BM‐40. These residues are spatially close and form a flat ring of 15 Å diameter which matches the diameter of a triple‐helical collagen domain. The mutations showed similar effects on binding to collagens I and IV, indicating nearly identical binding sites on both collagens. Selected mutations in the non‐activated mutant ΔI also reduced collagen binding, consistent with the same location of the epitope but in a more cryptic form in intact BM‐40.
Collagens are a large protein family with 19 members (types I–XIX) identified so far. They are localized in the extracellular space or exist as transmembrane forms (Brown and Timpl, 1995; Prockop and Kivirikko, 1995; Olsen and Ninomiya, 1998). They all share a triple‐helical domain of variable length which is formed from three chains and requires Gly at every third position and a high imino acid content in order to stabilize a polyproline II‐like helix (Brodsky and Shah, 1995). Further stabilization is achieved by a hydrogen‐bonded water network, which also requires 4‐hydroxyproline, as shown for a synthetic collagen‐like peptide (Bella et al., 1994, 1995). As a consequence of this structure, all of the side chains are exposed on the surface of the triple helix and are available for various intermolecular associations.
Interactions mediated through the triple helix include homo‐ and heterotypic self‐assembly processes, which result in the formation of large fibrils, filaments, networks and some other supramolecular associations, depending on the collagen type (Brodsky and Shah, 1995; Brown and Timpl, 1995; Prockop and Kivirikko, 1995; Olsen and Ninomiya, 1998). To a large extent, these assembly forms determine the mechanical strength of extracellular matrices, but they have many additional functions, frequently mediated through the binding of non‐collagenous ligands to the triple helix. These ligands include cell surface receptors such as α1β1 and α2β1 integrins (Vandenberg et al., 1991; Gullberg et al., 1992; Kern et al., 1993; Pfaff et al., 1993) and NG2 proteoglycan (Burg et al., 1996), several growth factors (Taipale and Keski‐Oja, 1997), small matrix proteoglycans (Oldberg et al., 1989; Bidanset et al., 1992) and other matrix proteins such as nidogen (Aumailley et al., 1989), BM‐40 (Lane and Sage, 1994) and von Willebrand factor (Ruggeri and Ware, 1993). They also include cell surface proteins, adhesins, which are present on some staphylococcal strains and are considered to allow invasion of host tissues (Switalski et al., 1989). The binding of these non‐collagenous ligands is either specific for a single collagen type or, more frequently, includes several different collagens (e.g. adhesins, integrins). Most of these interactions require an intact triple‐helical conformation and also, as shown for α1β1 integrin (Eble et al., 1993), a specific array of amino acid side chains. Such arrays are thought to interact with specific collagen‐binding epitopes, as has been shown in more‐detailed molecular analyses of adhesins (Patti et al., 1995; Symersky et al., 1997), α2β1 integrin (Kamata and Takada, 1994; Emsley et al., 1997), von Willebrand factor domain A3 (Bienkowska et al., 1997; Huizinga et al., 1997) and BM‐40 (Sasaki et al., 1997).
BM‐40, which is also referred to as SPARC or osteonectin, is an abundant 33 kDa extracellular calcium‐binding protein and is considered to modulate cellular phenotypes (Lane and Sage, 1994; Sage, 1997). It contains a flexible N‐terminal domain I, a follistatin‐like (FS) domain and a C‐terminal extracellular calcium‐binding (EC) domain (Maurer et al., 1995). The structure of the FS and EC domains has been elucidated recently by X‐ray crystallography (Hohenester et al., 1996, 1997) and thus set the stage for a molecular analysis of binding epitopes. Calcium‐dependent binding of BM‐40 to the triple helix of several fibrillar collagen types and basement membrane collagen type IV is of moderate affinity (Kd ∼1–2 μM) and has been mapped to the EC domain (Maurer et al., 1995, 1997a; Sasaki et al., 1997). The cleavage of a single peptide bond in helix αC of the EC domain was accompanied by a 10‐fold increase in collagen affinity. This suggested that the underlying helix αA should contribute to collagen binding (Sasaki et al., 1997). In the present study, a similar increase in affinity could be achieved by recombinant deletion of helix αC. This has allowed us to map the binding epitope in the activated and non‐activated forms of BM‐40 to two non‐contiguous sequence regions of the EC domain and to explain structurally how activation may occur.
Deletion of helix αC of BM‐40 increases collagen affinity
Previous data strongly suggested that the increase in collagen affinity of BM‐40 after a single proteolytic cleavage in helix αC of its EC domain was due to the removal of steric constraints (Sasaki et al., 1997). In order to verify this prediction, we carried out a recombinant deletion of helix αC (position 196–203) using BM‐40 mutant ΔI which lacks the flexible N‐terminal domain I (Maurer et al., 1995). This choice was instrumental in allowing crystallographic studies (Hohenester et al., 1997) and facilitating isolation of the new mutant ΔI,αC, as well as further mutants, in a single chromatographic step and in high yields from transfected mammalian cell medium. Purified ΔI,αC showed a single electrophoretic band with slightly higher mobility than ΔI. It also had a defined structure, as shown below. As a second, more versatile assay for structural integrity, we used a radioimmuno‐inhibition assay, which detected primarily conformation‐dependent epitopes (Figure 1). This demonstrated nearly identical activities for BM‐40 mutants ΔI, ΔI,αC and ΔI,II, the latter comprising the EC domain alone (Maurer et al., 1995).
The binding activities of BM‐40 ΔI and ΔI,αC were compared by surface plasmon resonance assay with collagens I–V as immobilized ligands (Table I). For BM‐40 ΔI, Kd values in the range 0.5–1.5 μM were observed, slightly better than those (∼2 μM) determined previously for intact BM‐40 (Sasaki et al., 1997). A clear 10‐fold increase in affinity was observed for BM‐40 ΔI,αC (Kd = 0.04–0.2 μM), mainly due to a 3‐ to 10‐fold increase in the association rate constants (Table I). These observations are consistent with a similar increase in affinity found after proteolytic activation (Sasaki et al., 1997) and also with the interpretation that steric restrictions that impair binding have been removed.
Crystal structure of BM‐40 ΔI,αC lacking domain I and helix αC
X‐ray crystallography was used to analyse the structural changes accompanying activation of collagen binding. BM‐40 ΔI,αC crystallized in a new crystal form and the structure was solved at 2.8 Å resolution by molecular replacement with a truncated BM‐40 ΔI model (Hohenester et al., 1997). The BM‐40 ΔI,αC structure and a superposition of the structures of BM‐40 ΔI,αC and BM‐40 ΔI is shown in Figure 2. The overall structures are very similar, but deletion of the major part of helix αC (residues 196–203) in BM‐40 ΔI,αC results in a significant structural rearrangement that extends far beyond the immediate site of deletion. If residues 162–205 are excluded from the superposition, the FS–EC domain pairs of BM‐40 ΔI,αC and BM‐40 ΔI can be superimposed with an r.m.s.d. of 0.6 Å, which drops to 0.4 Å if only the EC domains are compared. This is due to some segmental flexibility in the FS domain, especially in the N‐terminal hairpin (Figure 2B; Hohenester et al., 1997). Deletion of αC is accompanied by structural rearrangements in the αB–αD loop, as expected. Surprisingly, however, we observe that these changes are propagated as far as the αA–αB loop. Helix αB is rotated by ∼20° relative to the BM‐40 ΔI structure and is extended by two additional turns at its C‐terminus. Concomitantly, the αA–αB loop in BM‐40 ΔI,αC assumes a dramatically different conformation, with an additional helical turn (αA') not present in the BM‐40 ΔI structure. Five polar residues at the tip of the newly formed αB–αD loop in BM‐40 ΔI,αC are not seen in the electron density map and are presumed to be disordered.
In the BM‐40 ΔI,αC crystals, the most prominent packing interaction involves a bound calcium ion (Ca3; B‐factor 11 A2) shared by two symmetry‐related molecules. Three metal ion ligands are contributed by the first molecule of BM‐40 ΔI,αC (the carbonyl oxygens of P241 and I243 and the side chain carboxylate of E246) and two additional ligands are provided by the second molecule (the carbonyl oxygen of T160 and the side chain carboxylate of E163). The average metal–ligand distance is 2.3 Å. The co‐ordination geometry is that of an incomplete octahedron; the missing ligand may be a water molecule, and we indeed observe an appropriately positioned peak in the final difference Fourier map (not shown). The calcium‐mediated association of two molecules of BM‐40 ΔI,αC in the crystals results in 2‐fold symmetric dimers, in which the C‐termini of helices αA contact the metal ion bound to the αE–αF loop. We do not know whether such dimers exist in solution. It is noteworthy that in the BM‐40 EC domain structure (BM‐40 ΔI,II; Hohenester et al., 1996), a metal ion, putatively assigned as potassium, was observed bound to the tip of the αE–αF loop in a location identical to that of Ca3 in the present structure. This suggests the presence of a general cation site on BM‐40 that may be involved in ligand binding (see below).
Site‐directed mutagenesis of the collagen‐binding epitope in the activated form of BM‐40
The BM‐40 mutant ΔI,αC was chosen for analysis of the collagen‐binding epitope for two particular reasons. Because of its higher affinity, it would allow the detection of changes after mutagenesis over a broader range by the available assays. In addition, the mutant may resemble an analogue of biological activation as discussed later. All the mutants designed (Table II, Figure 3) could be obtained in good yields, indicating that there was no folding problem. The purified proteins showed a single band in electrophoresis with a mobility identical to that of BM‐40 ΔI,αC (data not shown). In addition, with two exceptions, they did not differ significantly from the parent molecule in the radioimmuno‐inhibition assay (Figure 1). Thirteen positions of the EC domain were chosen for alanine mutagenesis, based on several considerations (Figure 3). Three residues in particular are mostly buried in BM‐40 ΔI, but are essentially accessible in BM‐40 ΔI,αC: R149, D152 and N156 (and, to a lesser extent, L148 and K155). We argued that these residues were good candidates for participating in collagen binding and should be mutated to define this epitope. Additional residues on αA were chosen to map the extent of the epitope (L139, S141, E142, T144, E145, R151, D152 and T160). W153, whose side chain is partly accessible, was not mutated as it contributes significantly to the hydrophobic core of BM‐40. We also noted that several residues of the αE–αF loop, as well as Ca3, are directly adjacent to the putative epitope on helix αA. These residues, namely L242, M245 and E246 (a ligand of Ca3), were therefore included in our series of point mutants (Figure 3).
The analysis of a total of 13 alanine mutants revealed very similar changes of binding to collagens I and IV in surface plasmon resonance assays, as evaluated from Kd values (Table II). Mutants R149A, N156A and M245A showed a 50‐ to 100‐fold reduction in affinity, which was caused either by a strong decrease in the association rate constant (R149) or an increase in dissociation (N156, M245). Two more mutants (L242A, E246A) showed essentially no binding in the concentration range tested. A similar lack of binding was observed for double mutations at positions 149/156 and 156/245, demonstrating an additive effect. Three of these essential residues were examined further by introducing additional mutations. Substitution of R149 by Lys or Leu was even more deleterious than the Ala mutation, indicating a very strict structural requirement. Mutant N156Q still showed a 20‐ to 30‐fold loss of activity, indicating the need for a perfect fit to the length of the side chain. However, mutant M245Y was only of 3‐ to 4‐fold lower affinity compared with the parental ΔI,αC. A similar small reduction of binding was also observed for D152A. Ala mutations of L139, S141, E142, T144, R151 and T160 caused no significant reduction in affinity. Mutant E145A even showed a 3‐fold increase in binding to collagen IV, but not to collagen I (Table II).
The binding of these mutants was also analysed by solid‐phase assays with immobilized collagens I and IV. A few typical binding curves are shown in Figure 4 for collagen IV, and demonstrate examples of no or only slight changes, moderate loss or activation and a rather drastic inactivation. These assays were evaluated by comparing the concentration of soluble ligands required for half‐maximal binding, which showed a 10‐fold difference for BM‐40 ΔI,αC interactions with collagen I and IV. Irrespective of this difference, all mutants showed a similar change in relative binding activity in both assays (Table III). The magnitude and direction of changes were basically the same as observed for the Kd values (cf. Table II). The only exception was mutant M245Y, where the loss of activity was ∼10‐fold more accentuated in the solid‐phase assay. Together, these data demonstrated that five residues (R149, N156, L242, M245 and E246) are crucial for collagen binding. These residues occupy a restricted region constituted by helix αA and a link region on the surface of the activated BM‐40 variant ΔI,αC (Figures 3 and 5A).
Site‐directed mutagenesis of a non‐activated form of BM‐40
Fragment ΔI, which still possesses helix αC, has a low collagen‐binding affinity comparable with that of BM‐40, as previously shown (Maurer et al., 1995; Sasaki et al., 1997). This fragment was now used to introduce some of the same mutations as in fragment ΔI,αC as well as a P244A mutation. Analysis of the collagen IV binding of these mutants by surface plasmon resonance assay demonstrated similar but less marked affinity changes (Table IV). These included a distinct decrease for the BM‐40 ΔI mutants R149A, N156A and M245A, a more moderate one for S141A and D152A, and no significant change for T144A, E145A and P244A. Similar changes were also observed for collagen I binding and were also confirmed for collagen IV by the solid‐phase assay (data not shown). This strongly suggests that most, if not all, critical binding residues are the same for activated and non‐activated forms of BM‐40.
The calcium‐binding EC module of BM‐40, which is also shared in modified form by five related proteins, was shown previously to harbour the collagen‐binding epitope of this abundant extracellular matrix protein (Maurer et al., 1995). It was shown further that cleavage of a single peptide bond of helix αC of the EC domain either by several matrix metalloproteinases (Sasaki et al., 1997) or by unknown endogenous proteases (Maurer et al., 1997a) increased collagen affinity by a factor of 10. Based on the X‐ray structure of the EC domain (Hohenester et al., 1996), this was interpreted to indicate the removal of steric constraints imposed by helix αC on the underlying helix αA. In the present study, we have confirmed this hypothesis by deleting helix αC (fragment ΔI,αC) and showing that this causes a comparable increase in collagen affinity. X‐ray crystallography of this deletion mutant demonstrated a rearrangement in the shortened link region between helix αA and the pair of EF hands exposing the central part of helix αA (Figure 2). The effect of the deletion of eight residues of helix αC on the BM‐40 structure is not confined to the connection between helices αB and αD. It appears reasonable to assume that a very similar situation exists in BM‐40 when proteolytically nicked at the N‐terminus of αC (Sasaki et al., 1997). BM‐40 ΔI,αC, therefore, seems to be a good model of activated forms of BM‐40 as they may exist in tissues (see below).
The collagen‐binding epitope of BM‐40, as defined by site‐directed mutagenesis, is shown in Figure 5A. Two residues on helix αA (R149 and N156) and three residues on the loop connecting helices αE and αF (L242, M245 and E246) were identified as important for binding of both collagen I and IV. These five residues are arranged in a ring of ∼15 Å diameter, surrounded by residues that were shown not to contribute greatly to collagen binding. Figure 5A suggests that two additional residues may be part of the epitope. The hydrophobic side chains of W153 and P241 are at least partially accessible and are located within the perimeter of the five critical residues. W153 and P241 were not mutated, as we expected their mutation to Ala to be accompanied by major structural changes.
The surface of BM‐40 in the region containing the collagen‐binding site is relatively flat and does not feature a pronounced groove, as might have been expected for a binding site for triple‐helical collagens. A model showing a complex of BM‐40 ΔI,αC with a collagen triple helix is shown in Figure 5B. The extent of the collagen‐binding epitope, as defined by site‐directed mutagenesis, is seen to match the diameter of the collagen triple helix surprisingly well. The model also suggests that the complementary binding site on collagens should not extend over more than two Gly‐X‐Y triplet repeats (Brodsky and Shah, 1995). The flatness of the epitope places significant structural constraints on the interaction. The only real variable is the orientation of the axis of the triple helix in the plane defined by the five critical residues, although some arrangements can be ruled out because they would lead to clashes with helix αD and the αA–αD linker region.
The collagen‐binding site of BM‐40 consists of both polar and non‐polar residues, suggesting that collagen binding is due to a combination of van der Waals and electrostatic interactions. Two critical residues, R149 and N156, are not exposed in uncleaved BM‐40 (Hohenester et al., 1996, 1997), and this is proposed to account for the relatively low affinity of intact BM‐40 for collagens (Sasaki et al., 1997). Binding of collagen by intact BM‐40 may first involve the association of collagen with the solvent‐accessible part of the epitope (L242, M245 and E246), followed by a conformational change in the region of αC, thereby uncovering R149 and N156 and consolidating the interaction between BM‐40 and collagen. Proteolytic cleavage at the N‐terminus of αC would remove the second, energetically costly step and thereby increase association rates and, consequently, collagen affinity (Sasaki et al., 1997).
The observation of a calcium ion (Ca3) involved in a crystal contact and co‐ordinated by a residue shown to be important for collagen binding (E246) raises the interesting possibility that a metal ion may contribute to collagen binding. The side chain of E246 lies slightly below the surface containing the epitope. E246 could interact with either a lysine or an arginine residue of the collagen ligand. Alternatively, it may provide a crucial ligand for a calcium ion mediating the association of BM‐40 and a collagen triple helix. This situation would be rather similar to the metal ion‐dependent adhesion site (MIDAS) of integrin I‐domains (Lee et al., 1995). The recently determined structure of the collagen‐binding I‐domain of α2β1 integrin shows a groove around the MIDAS motif, and the authors speculate that collagen binding by the I‐domain may involve the co‐ordination of the resident magnesium ion by a glutamic acid of the collagen ligand (Emsley et al., 1997). Collagen binding to BM‐40 has been shown to be calcium dependent (Maurer et al., 1995), but this may reflect the requirement for calcium bound to the EF hand pair for the structural integrity of the EC domain. Further studies are required to define precisely the role of calcium in collagen binding to BM‐40.
Several extracellular matrix proteins share a related EC domain with BM‐40 (see Maurer et al., 1995; Hohenester et al., 1996). A sequence comparison of those regions critical for collagen binding (Figure 6) shows that proteins QR1 and SC1 (hevin) have a remarkable conservation including all five essential residues. Since these two proteins also have a long connecting region between helices αA and αD, we would predict that they share collagen binding with BM‐40 which may also be dependent on proteolytic activation. Testican and tsc36, on the other hand, differ by at least one substitution (N156A) in this region and would not be expected to be active. Recently, the EC domain of testican has been shown to lack significant collagen binding (Kohfeldt et al., 1997). Caenorhabditis elegans BM‐40 seems to represent a special case as it binds to mammalian collagens through its EC domain with an affinity (Kd = 0.5−2.2 μM) comparable with mammalian BM‐40 (Maurer et al., 1997b). However, our mutant data suggest that an N156Q substitution in the nematode BM‐40 (Figure 6) should result in a lower affinity (Table IV). Since the nematode BM‐40 lacks a long connecting region (αA–αD), however, it may exist in a permanently activated form and should then have an affinity comparable with BM‐40 ΔI,αC mutant N156Q (Kd = 1.6–2.8 μM), which is in fact the case.
To date, few structural studies of collagen‐binding sites have been reported for other proteins. Recently, the structure of the collagen‐binding domain of Staphylococcus aureus adhesin became available (Symersky et al., 1997). Docking and mutagenesis studies identified, as here, a crucial Arg and suggested a prominent and relatively narrow groove as the binding site for a collagen triple helix (Symersky et al., 1997). Such a pronounced groove may be advantageous for binding generic collagen triple helices, but may not be an element of more specific mammalian collagen‐binding proteins. This is suggested by two recent structure determinations of the A3 domain of von Willebrand factor (Bienkowska et al., 1997; Huizinga et al., 1997), where the collagen‐binding site seems to reside on a relatively flat face of the domain (Huizinga et al., 1997). By contrast, the related I‐domain of α2β1 integrin, which also binds collagen triple helices, has a groove in the corresponding region that also contains the MIDAS motif (Emsley et al., 1997).
The biological consequences of BM‐40 binding to collagens in tissues are only poorly understood, but could include a storage function as known for several cytokines (Taipale and Keski‐Oja, 1997) or the modulation of various anti‐adhesive or anti‐proliferative activities reported for BM‐40 (Lane and Sage, 1994; Sage, 1997). A proteolytic activation mechanism in order to increase collagen affinity could therefore represent an important controlling step. This may involve several matrix metalloproteinases, which all cleave at position 197–198 in helix αC (Sasaki et al., 1997), as well as unknown endogenous proteases cleaving at position 198–199 (Mann et al., 1987; Maurer et al., 1997b). BM‐40 was shown, furthermore, to increase the production of several matrix metalloproteinases in cell culture (Tremble et al., 1993), suggesting an autocrine or paracrine way of controlling activation. For further biological studies, we have recently generated antisera against the N‐terminal part of the two identified cleavage sites using synthetic peptide antigens (T.Sasaki and R.Timpl, unpublished). These antibodies distinguished not only between intact and activated BM‐40 but also between the two different cleavage sites. These new reagents and the precise mapping of the collagen‐binding site as described here will therefore offer novel molecular approaches to study BM‐40 function in a biological context.
Materials and methods
Sources of proteins
Recombinant human BM‐40 (Nischt et al., 1991) and its deletion mutants ΔI and ΔI,II (Maurer et al., 1995) have been described previously. Pepsin digests of human placenta were used to purify the fibrillar collagens I, III and V (Miller and Rhodes, 1982) and the basement membrane collagen IV (Vandenberg et al., 1991). Recombinant cartilage collagen II (Fertala et al., 1994) was kindly provided by Dr D.J.Prockop.
Construction of expression vectors
The insert of the construct for human BM‐40 deletion mutant ΔI (Maurer et al., 1995) was used to introduce further mutations based on oligonucleotide primers by polymerase chain reaction (PCR) using Vent polymerase (New England Biolabs). The two terminal primers were F1 (GTCAGCTAGCAAATCCCTGCCAGAAC) for the 5′ end and R1 (GTCACTCGAGTTAGATCACAAGATCCTTGTCG) for the 3′ end. Deletion of helix αC was introduced by primers DelC1 (GTTCTTCTCGGGGTGGTCTC) and DelC2 (GAGACCACCCCGAGAAGAAC) which were combined with F1 and R1, respectively, to generate the 5′ and 3′ fragments. The gel‐purified fragments were fused by annealing overlapping ends and amplified with primers F1 and R1 to give the final PCR product (Vallejo et al., 1994). Point mutations were introduced into the inserts encoding BM‐40 ΔI or ΔI,αC by using appropriate central overlapping primers (not shown) following the same strategy. A few mutations were close to an internal EcoRI site, which was used occasionally for insertion into the parental vector. The PCR products were cloned into blunt‐ended vector pUC18 by using the Sure Clone Kit (Pharmacia), and the correctness of the sequence was confirmed by the Dye Terminator Cycle Sequencing Ready Reaction Kit (ABI). The NheI–XhoI‐released inserts of correct clones were isolated and inserted into corresponding sites of the expression vector pCEP/Pu, which contains the BM‐40 signal peptide sequence (Kohfeldt et al., 1997).
Recombinant production and purification
The expression vectors were used for the episomal transfection of human EBNA‐293 cells (Invitrogen). Resistant cells were selected with puromycin (0.5 μg/ml) and used for the collection of serum‐free conditioned medium. The medium (200–300 ml) was passed over a DEAE–cellulose column (2.5×25 cm) which was equilibrated in 0.05 M Tris–HCl, pH 8.6, and eluted with a linear gradient of 0–0.4 M NaCl (500 ml). All recombinant products eluted as a sharp peak around 0.2 M NaCl, and were concentrated by ultrafiltration and dialysed against 0.05 M Tris–HCl, pH 7.4, 0.1 M NaCl. Electrophoresis demonstrated a purity of >95%. The yields were in the range 3–10 mg protein.
Protein concentrations were determined after hydrolysis with 6 M HCl (16 h, 110°C) on an LC amino acid analyser. SDS–gel electrophoresis in polyacrylamide gradient gels (10–20%) followed established protocols. A radioimmunoassay with a rabbit antiserum against human BM‐40 has been described (Nischt et al., 1991).
Crystallization and X‐ray data collection
BM‐40 ΔI,αC did not crystallize under the conditions used for BM‐40 ΔI (Hohenester et al., 1997). Crystals of BM‐40 ΔI,αC were obtained at room temperature by the hanging drop vapour diffusion method: 3 μl of a protein stock solution (8 mg/ml in 10 mM Tris–HCl pH 7.5, 2 mM CaCl2) were mixed with 2 μl of reservoir solution [12–14% (w/v) polyethylene glycol 4000, 10% (v/v) 2‐propanol, 100 mM Na‐HEPES pH 7.5] and equilibrated against 1 ml of the latter solution. The crystals belong to space group P212121 with unit cell constants a = 51.9 Å, b = 88.1 Å, c = 153.7 Å. There are two molecules of BM‐40 ΔI,αC in the asymmetric unit, resulting in a solvent content of ∼65%. The crystals diffracted weakly to 2.8 Å resolution. Diffraction data were collected at 100 K using a MAR image detector mounted on an Elliot GX‐21 rotating anode generator operated at 4 kW. Crystals were transferred into artificial mother liquor with 10, 20 and 30% glycerol added (three consecutive soaks of 10 s) and frozen in the cold nitrogen stream. Data were integrated with MOSFLM (Leslie, 1994) and reduced with programs of the CCP4 suite (Collaborative Computing Project No. 4, 1994). Data collection statistics are summarized in Table V.
Solution and refinement of the structure
The self‐rotation function did not show any peaks attributable to the non‐crystallographic symmetry (NCS) relating the two molecules of BM‐40 ΔI,αC in the asymmetric unit. The native Patterson map showed a strong non‐origin peak at ≈(0, 0.5, 0.5), suggesting the presence of two similarly oriented molecules. The structure of BM‐40 ΔI,αC was solved by molecular replacement with AMoRe (Navaza, 1994). The search model was the BM‐40 ΔI structure (Hohenester et al., 1997), with residues 168–208 and the NAG sugar moieties removed. The cross‐rotation function (15–3.5 Å) showed one prominent peak at 6.8σ, compared with 2.9σ for the highest noise peak. The highest peak in the translation function (6–2.8 Å) resulted in a solution with an R‐factor of 0.511. The first molecule was fixed at the respective position and a partial translation function was calculated, which revealed the position of the second molecule, shifted by 0.53b + 0.47c relative to the first one, as expected. The R‐factor of the two independent molecules after rigid‐body refinement (6–2.8 Å) was 0.433 and the correlation coefficient was 0.635.
The electron density map calculated from the molecular replacement solution was improved by 2‐fold NCS averaging with RAVE (Kleywegt and Jones, 1994). The resulting map allowed most of the residues missing from the search model to be built, with the exception of a few residues around the deletion site. The structure was refined with X‐PLOR (Brünger, 1992) using all observed (F>0) data between 6 and 2.8 Å resolution; strict NCS was imposed throughout. The final model consists of residues 54–286, three calcium ions and two NAG sugar moieties attached to N99. Residues 192–195 and 204–205 are not defined by the electron density and have been modelled stereochemically. About 10% of the unit cell content is carbohydrate that is not accounted for by the crystallographic model, but is partially ordered and makes a non‐negligible contribution to the diffraction at low resolution; therefore, data below 6 Å resolution were excluded from the refinement. A similar problem had been encountered in the structure determination of BM‐40 ΔI (Hohenester et al., 1997). According to the definition of PROCHECK (Laskowski et al., 1993), 91.8% of the residues are in core regions of the Ramachandran plot, with the remaining 8.2% in additionally allowed regions. The NCS real‐space correlation coefficient of the final averaged map was 0.94 and the R‐factor of the back‐transformed map with the observed amplitudes was 0.194. Refinement statistics are summarized in Table V. Co‐ordinates and structure factors have been deposited in the Brookhaven Protein Data Bank (accession code 1NUB) and will be made available 1 year after publication.
Surface plasmon resonance assays were performed with BIAcore instrumentation (BIAcore AB Uppsala) and various immobilized collagens. Collagens were immobilized by covalent coupling to CM‐5 sensor chips (research grade) (Maurer et al., 1995), resulting in binding of 6000–8000 RU. Binding assays were performed in 50 mM Tris–HCl, pH 7.4, 0.15 M NaCl, 2 mM CaCl2 and 0.05% P20 detergent (BIAcore) at 25°C and a flow rate of 20 μl/min. Preliminary assays showed that this flow rate was high enough to avoid mass transport problems. The different soluble ligands were applied as far as possible in a concentration range around the Kd value, resulting in an uptake of 20–3000 RU. The association curve was followed for 3 min, then the protein solution was replaced by buffer and the dissociation curve was monitored for another 15 min. Since most of the spontaneous dissociations were not quantitative, the chips were completely regenerated by treatment with 10 μl of 10 mM HCl. This treatment did not affect the immobilized layer, as controlled by repeated injections of the same amount of soluble ligands. Non‐specific binding to the surface was excluded for all ligands by carrying out binding assays on chips without collagens. The integrity of the immobilized collagens was tested from time to time with known ligands. All measurements were repeated several times with different batches of immobilized and soluble ligands.
Kinetic constants were calculated by non‐linear fitting of the association and dissociation curves according to the 1:1 model A + B = AB using BIAevaluation software version 2.1, following the manufacturer's instructions. For evaluation of the dissociation rate constants, the time ranges from ∼30–50 s (to avoid bulk refractive index effects at the very beginning of the dissociation) and 300–400 s (analyte rebinding can affect the later part of the dissociation) after the start of dissociation were used. Plots of In(R0/R) against t were linear over a wide time span, indicating that a 1:1 dissociation model should be reasonably applicable. For fitting of the association curve, the range where ln(dR/dt) was linear and the noise level not too high was selected. The fit led to a direct calculation of kass. As a second plot for the association phase, kobs [slope of ln(dR/dt) against t] versus C and its linear fit were used to calculate kass from the slope and kdiss from the intercept. This linear fitting allows control of the non‐linear fitting and of the applicability of the 1:1 model.
Solid‐phase assays with plastic‐immobilized collagens followed a previously described protocol (Aumailley et al., 1989). All binding tests were performed in the presence of 2 mM CaCl2.
We are grateful for the expert technical assistance of Mrs Mischa Reiter, Vera van Delden and Christa Wendt. The study was supported by EC contract No. BIO4‐CT96‐0537 and by a long‐term fellowship to E.H. from the Human Frontier Science Program.
- Copyright © 1998 European Molecular Biology Organization