NMR Solution Structure of the Receptor Binding Domain of Human α2-Macroglobulin*

Human α2-macroglobulin-proteinase complexes bind to their receptor, the low density lipoprotein receptor-related protein (LRP), through a discrete 138-residue C-terminal receptor binding domain (RBD), which also binds to the β-amyloid peptide. We have used NMR spectroscopy on recombinantly expressed uniformly13C/15N-labeled human RBD to determine its three-dimensional structure in solution. Human RBD is a sandwich of two antiparallel β-sheets, one four-strand and one five-strand, and also contains one α-helix of 2.5 turns and an additional 1-turn helical region. The principal α-helix contains two lysine residues on the outer face that are known to be essential for receptor binding. A calcium binding site (K d ∼ 11 mm) is present in the loop region at one end of the β-sandwich. Calcium binding principally affects this loop region and does not significantly perturb the stable core structure of the domain. The structure and NMR assignments will enable us to examine in solution specific binding of RBD to domains of the receptor and to β-amyloid peptide.

Human ␣ 2 -macroglobulin (␣ 2 M) 1 is a highly abundant, very high molecular mass (ϳ720 kDa) plasma protein that is best known as a broad spectrum inhibitor of proteinases (1). ␣ 2 M is also reported to bind to certain growth factors (2) and to be involved in binding and clearance of the 42-residue ␤-amyloid peptide (3,4) that is thought to contribute to the etiology of Alzheimer's disease through formation of fibrils. The essential nature of ␣ 2 M is indicated by the failure to find any individuals with ␣ 2 M deficiency (5). The means of inhibition and clearance of proteinases by ␣ 2 M is through a unique proteinase-induced conformational change in the ␣ 2 M that physically traps the proteinase within a cage-like interior cavity (6,7). The massive extent of the conformational change and the nature of the sequestration have been likened to the action of a Venus fly trap closing around its prey. Because inhibition by ␣ 2 M involves sequestration of the proteinase from the surrounding milieu rather than direct and specific binding to the proteinase active site, the inhibitor is not limited to inhibition of one type or class of proteinase, but can instead, uniquely, inhibit proteinases of all four mechanistic classes.
The trapping conformational change not only results in sequestration of the proteinase, but also exposure of a receptor binding region on ␣ 2 M that is solely responsible for mediating binding to the clearance receptor LRP (low density lipoprotein receptor-related protein) (8). It has been shown that the receptor binding region is located within a discrete domain that comprises the last 138 residues of the ␣ 2 M monomer (9,10). This domain, termed the receptor binding domain (RBD), can be preparatively isolated from methylamine-transformed human ␣ 2 M by limited digestion with papain (9) or endo-Lys proteinase (10). Cleavage by either proteinase occurs between lysine and glutamate in the highly charged sequence Glu-Lys-Glu-Glu, which suggests that this tetrapeptide is an exposed linker that may act as a hinge to permit movement of RBD upon conformational transformation of ␣ 2 M and, hence, exposure of the LRP-binding epitope on the domain. Isolated RBD retains the ability to bind specifically to LRP (9) and to be taken up by the receptor (11) and, hence, represents an appropriate species for studying the interaction between ␣ 2 M and its receptor. Although the affinity for LRP is reduced ϳ200-fold compared with intact ␣ 2 M-proteinase complexes, from 0.5 to ϳ100 nM, this is likely to be because of the difference in higher ligand valency of tetrameric ␣ 2 M compared with the monomeric RBD (11).
The receptor to which these ␣ 2 M-proteinase complexes bind is LRP, a mosaic protein that is a member of the LDL receptor family. It contains clusters of epidermal growth factor (EGF)like repeats and of complement-like repeats (abbreviated to CR) (12). The latter are thought to be the sites of binding of different protein ligands both in LRP and LDLR (8,13). In LRP there are four such clusters, designated I-IV, containing 2, 8, 10, and 11 CR domains, respectively. ␣ 2 M-proteinase complexes have been shown to bind to the second cluster of repeats-domains CR3-CR8 (14). Given both the relatively small dimensions of these repeats compared with the size of the protein ligands that bind, together with the very broad range of proteins that LRP is able to bind, it may be that a protein ligand binding site on LRP is composed of interactions with two or more CR domains, where each interaction is relatively weak but where together they represent a high affinity ligand binding site.
Although there is good evidence for binding of the growth factor TGF-␤ to human ␣ 2 M (2), such binding is not thought to involve the receptor binding region. In contrast, the noncovalent interaction of ␤-amyloid peptide with human ␣ 2 M has been localized to the extreme C-terminal portion of the ␣ 2 M mono-mer that includes the receptor binding region and may well involve specific interactions with this domain (15). No further information on the location of the binding site is presently available.
Given the critical role of RBD in clearance of proteinases through binding of the exposed RBD to LRP, and its possible involvement in binding ␤-amyloid peptide, it is important to know not only the structure of this region of human ␣ 2 M but to have a means of examining the details of interactions between RBD and LRP or between RBD and ␤-amyloid peptide at the molecular level. Whereas crystallography in principle affords such a means, it requires the ability to crystallize not only RBD alone but also complexes formed with domains from LRP or with ␤-amyloid peptide. Probing the specificity of interaction of different CR repeats, where individual domains may bind only weakly, may be problematic because it may be hard to cocrystallize weak binding protein-protein complexes. For human ␣ 2 M, it has proved impossible to obtain satisfactory crystals even of RBD alone (16). NMR, however, provides a means of not only determining the structure of human RBD, but of subsequently examining in solution interactions with other protein or peptide ligands, even when the strength of the interaction may be weak. Thus, despite the recent availability of a crystal structure of bovine RBD (17), we have used uniformly 13 C/ 15 Nenriched recombinant human RBD to determine the threedimensional structure in solution of RBD from human ␣ 2 M, using multinuclear three-dimensional NMR spectroscopy. This has allowed us to make a comparison between crystal and solution structures, between the structures of the bovine and human proteins, and to characterize a calcium binding site in the protein that may be of structural significance in intact ␣ 2 M. We report here the results of these studies. In the future, the NMR assignments and structure will allow detailed examination of the solution interactions between human RBD and domains from LRP to which it might bind and between RBD and ␤-amyloid peptide.

EXPERIMENTAL PROCEDURES
Expression and Purification of Human RBD-Recombinant human RBD was expressed and purified as described previously (18). Briefly, RBD was expressed in E. coli as a 165 residue fusion protein that, in addition to RBD, contained an N-terminal 6-histidine tag, a thrombin cleavage site, and an extra 10 residues from ␣ 2 M at the N terminus that had been thought to be necessary for stability of RBD (19). These extra residues turned out to promote oligomerization of the refolded protein and so were removed with papain following refolding. The N terminus was confirmed by sequencing to be Glu-1314, using the numbering of the intact ␣ 2 M subunit (20). Plasma RBD, used for calcium binding experiments, was prepared by papain cleavage of methylamine-treated human plasma ␣ 2 M, as described previously (9). Residues of RBD are numbered consecutively from 1 to 138. This corresponds to residues 1314 to 1451 of the intact ␣ 2 M monomer.
NMR Sample Preparation-For the structure determination experiments, purified recombinant RBD was dialyzed into NMR buffer (100 mM phosphate, pH 5.1, 300 mM NaCl), to which 10% D 2 O was added. The final concentration for NMR studies was 0.9 mM for the 15 N-labeled sample and 1.2 mM for the 13 NMR Spectroscopy-NMR experiments were carried out at 25°C on a Bruker DRX600 at the University of Illinois at Chicago or a Bruker DMX600 at the University of Wisconsin, Madison. Both spectrometers were equipped with four channels and a pulsed-field gradient accessory. NMR data were processed and analyzed using Triad, Version 6.2 software (Tripos, Inc., St. Louis, MO).
Assignments-The backbone assignments were obtained according to standard procedures and have been published (18). Three-dimensional FIG. 1. Schematic representation of the secondary structure of human RBD using the numbering of the isolated domain. The structure is composed of two sheets with four and five strands in ␤-conformation (arrows) and two regions of ␣-helix (filled cylinders). The location of the two cysteines that form the single disulfide (residues 16 and 131) are marked with a dot, and the single site of glycosylation (residue 88) is marked with an asterisk. Structure Calculations-Distance constraints were estimated using the criteria of small, medium, and large NOEs. Upper limits for backbone NOE correlations were set at 5.0, 4.0, and 3.0 Å, whereas values of 5.5, 4.5, and 3.5 Å were used for backbone-sidechain and sidechainsidechain NOEs. These values were also the base values used to make pseudo-atom corrections for moieties where stereospecific assignments were not available. The input of this calculation used a total of 1142 upper distance constraints distributed as follows: 454 intraresidue, 331 sequential, 81 medium, and 276 long range correlations and 125 torsion angle constraints. Torsion angle restraints were derived from HNHA measurements of 3 J H N H ␣ coupling constants (21) and chemical shift index-based assignment of secondary structure (22). Structure calculations were performed with the torsion angle dynamics annealing simulation program DYANA (23). The final ensemble of structures that represented the 20 best DYANA conformers from an input of 100 initial structures was analyzed, and figures were generated by MOLMOL (24). The stereochemical quality of the protein was assessed with the program PROCHECK (25) for the 20 best structures. Atomic coordinates have been deposited with the Protein Data Bank, code 1BV8.
Materials-( 15 NH 4 ) 2 SO 4 was from Cambridge Isotopes, 13 C glucose was from Isotec, and D 2 O was from Sigma. Papain was from Roche Molecular Biochemicals. 99.997% CaCl 2 and GdCl 3 were from Johnson Matthey Inc.

RESULTS AND DISCUSSION
Secondary Structure of Human RBD-We have previously reported on the secondary structure of human RBD based on 1 H and 13 C chemical shift indices and some inter-strand NOEs (18). This showed the presence of eight major strands of ␤-sheet and one major ␣-helix of 2.25 turns. In completing the threedimensional structure determination, we found evidence for an additional small ninth strand of ␤-sheet from residues 53 to 55 (S4) and a 1.25-turn ␣-helical region from residues 19 to 23 (H1). Neither of these additional regions was previously found from secondary structure predictions of human RBD or other RBDs from seven other ␣-macroglobulins (18). The complete secondary structure present in the final three-dimensional structure is presented schematically in Fig. 1. In addition to the two short ␣-helices, the protein consists of two antiparallel ␤-sheets. One ␤-sheet is composed of strands S1-(5-12), S2-  Fig. 3. The front ␤-sheet of human RBD is composed, from left to right, of strands S5, S6, S3, S8, and S9. The rear sheet is composed, from left to right, of strands S4, S7, S2, and S1. Helix H2 is on the extreme left of the molecule.  The secondary structure composition and topology are identical to those of the crystal structure of bovine RBD, with only minor differences in the lengths of some of the secondary structural elements (17). Strand S5 is one residue longer in the human protein, and strands S8 and S9 are one residue shorter in the human protein.
Both helices are one residue longer in the human protein.
Three-dimensional Structure of RBD-Standard three-dimensional NMR experiments on doubly labeled human RBD were used to assign the backbone resonances and most of the side chain resonances and to obtain NOE constraints and some backbone torsion angle constraints (Table I). A total of 1142 NOE constraints and 125 torsion angle constraints were used. The family of twenty best structures obtained by simulated annealing is shown in Fig. 2, and a ribbon representation of the mean structure is shown in Fig. 3. The extreme N-and Cterminal regions, which lie at opposite ends of the molecule, are relatively poorly defined. Such poor definition for the N terminus is as expected from it being the region that links RBD to the central body of the ␣ 2 M tetramer (6) and that is susceptible to specific proteolytic cleavage (9,10). The poor definition of the C terminus parallels the behavior of the bovine RBD in the crystal, in which there is sufficient disorder that the last six residues cannot be traced. The quality of the structures in regions of secondary structure is good, with RMSD of 0.41-0.52 Å for regions of secondary structure backbone atoms. Overall the backbone RMSD is 1.78 Å for residues 5-131 (i.e. excluding the disordered N and C termini) ( Table I). For heavy atoms, the RMSDs are 1.36 -1.57 Å for regions of secondary structure and 2.73 Å for all heavy atoms for residues 5-131 (Table I). There are few distance constraint violations, with an average of 2.55 per structure for violations Ͼ0.2 Å. Ramachandran analysis shows that, for regions in defined secondary structure (S1-S9 and H1-H2), 89.2% of residues are in the most favored region, and 100% of residues are within either this region or the additional allowed region. For the whole structure, these percentages fall to 71.8 and 96.8%, respectively.
The structure consists of a sandwich of two ␤-sheets of 4 and 5 strands, edged by one ␣-helix of 2 turns and with a second 1-turn ␣-helix between strands S1 and S2. The major ␣-helix contains two lysine residues, 57 and 61, on the outward-facing flank with their terminal ⑀-amino groups no more than 11 Å apart. These lysines have been shown by others to be required for high affinity binding to the receptor and are in a good, well exposed position to interact with receptor. The loops that link the strands of the two ␤-sheets cluster above and below the sandwich. It has been pointed out by others that binding of protein ligands such as ␣ 2 M-proteinase complexes and apolipoproteins to complement-like repeats of LRP or other members of the LDL receptor family may involve extensive hydrophobic interactions in addition to ion pair interactions (17,26). There is, however, another positively charged residue, other than lysines 57 and 61, that may be involved in binding of human RBD to LRP. This is arginine 65, which is close to these two lysines and forms an almost continuous positive patch on the surface of helix 2 that might interact with carboxyl groups on complementlike domain(s) of the receptor. The precedent for the involvement of multiple positively charged residues interacting with such repeats is of apolipoprotein E binding to the LDL receptor, where eight basic residues in close proximity to one another on the face of an ␣-helix have been suggested to constitute a major part of the binding site (27), and more recently for the binding of PAI-1-uPA complexes to LRP (28,29) where four basic residues on PAI-1 have been implicated in receptor binding (28).
Although we are not yet in a position to fully define the binding surface on RBD for the receptor, we can use the present structure to evaluate a previous attempt, using epitope mapping to define this site. The study used phage display to identify possible discontinuous epitopes that might form part of the LRP binding site on RBD (30). The two discontinuous regions identified map to loop regions of RBD at opposite ends of the molecule. One segment, DEPK, is located between S1 and H1 (including the first residue of H1), whereas the second segment, SRS, is ϳ50 Å away at the far end of the molecule, between S2 and S3. Because the identification of these regions was from a heptapeptide that included both portions of the proposed binding site, the two epitopes should be contiguous in space, which clearly they are not. It is therefore questionable whether either portion of the epitope represents a part of the binding site for LRP.
A comparison was made between the mean NMR structure of human RBD and the crystal structure of bovine RBD (17), using coordinates generously provided by Dr. Nyborg (Protein Data Bank code 1AYO). Overall, there is high similarity between the two structures (Table II and Fig. 3), with one difference being in the orientation of helix H2 relative to the body of the protein. This may be because of the presence of an extra residue (histidine 68) in the loop that links the C-terminal end of the helix to the body of the protein in human RBD (Fig. 4). Whereas the C-terminal seven residues were not visible in the crystal structure, they do show some NOEs to define their position in the NMR structure. This high similarity between the two structures occurs despite the structure of bovine RBD being determined at higher pH, in the presence of 20 mM Ca 2ϩ and on a naturally glycosylated form. The lack of effect of Ca 2ϩ is consistent with our present findings that calcium binds to a loop region and has no structural effect on the core of the domain (see below). Similarly, the lack of effect of carbohydrate is consistent with our previous findings of the invariance of conformation-sensitive regions of the one-dimensional 1 H spectra in H 2 O of glycosylated human plasma ␣ 2 M-derived RBD and nonglycosylated recombinant human RBD (18).
Characterization of the Calcium Binding Site-Calcium is known to be essential for binding of ␣ 2 M-proteinase complexes and other ligands to their receptor LRP (12). While this is thought to result primarily from the presence of calcium binding sites in each of the complement-like repeats of LRP (31), the finding that there is a calcium ion shared between the two molecules of bovine RBD present in the asymmetric unit of the crystal structure pointed out the need to establish whether such a calcium binding site exists in solution in the receptor binding domain of human ␣ 2 M and, if it does, whether it is of sufficient affinity to be physiologically significant and what effect it has on the structure of RBD. The latter point is also important because the three-dimensional structure of RBD we had determined was in the absence of calcium. The HSQC spectrum of 15 N-labeled human RBD at pH 7.4 showed that, while calcium does cause some perturbation of backbone resonances (Fig. 5), the magnitudes of the perturbations are relatively small (Ͻ0.09 ppm for 1 H and Ͻ 0.52 ppm for 15 N) and involve relatively few of the cross peaks. Although the resonance assignments used above in the structure determination were made at pH 5.1 whereas the present spectra were recorded at physiological pH, there is sufficient similarity in the appearance of the spectra that we can assign with reasonable confidence the resonances in the pH 7.4 spectrum that are most strongly perturbed by calcium. We found that those resonances with significant perturbations arose from the loop regions at the "bottom" of RBD (orientation of Fig. 2), whereas the core of the molecule, which contains all of the secondary structure, as well as the "top" loop regions were very little affected. This was reinforced by the effect of Ca 2ϩ on the one-dimensional proton spectrum. Although there are few significant effects on the main part of the aliphatic region of the spectrum (not shown), indicating no major conformational change, the most sensitive regions of the spectrum, i.e. the upfield methyl region and the aromatic region, show some small perturbations (not shown). In particular, the presence of a cluster of aromatic side chains in the calcium-sensitive 115-122 loop results in one of the resolved aromatic resonances undergoing a readily measurable shift. This resonance was used to monitor calcium binding and gave a saturable change in chemical shift of ϳ0.04 ppm, indicating a specific binding interaction. This titration could be well fitted to a single site binding process and gave a K d for the calcium site of ϳ11 mM. Such an affinity is sufficiently high for there to be at least partial occupancy under physiological conditions, even if the presence of the remainder of the ␣ 2 M polypeptide does not enhance the affinity. In addition to showing that calcium binds to human RBD, these results show that the core secondary structure is not affected by calcium, so that the structure determined in the absence of calcium should have equivalent secondary and tertiary structure in the core region as when calcium is bound. Finally, the results suggest that any specific calcium binding site is probably located in the bottom loop region that connects the strands of the ␤-sandwich.
To better localize this low affinity calcium site, we used the distance-dependent paramagnetic broadening effect of Gd 3ϩ and the fast exchange conditions that apply to the calcium site. HSQC spectra of RBD were recorded at physiological pH in the absence or presence of low concentrations of added Gd 3ϩ . As expected from the presence of a well defined metal binding site, very specific broadening effects were seen (Fig. 6) that were confined to resonances solely from the bottom loop regions of the molecule. The residues that gave the largest effects on amide resonances all mapped to the 34 -42 loop of RBD (Fig. 7), very close to the location of the "bridging" calcium site found in the crystal structure of bovine RBD. It may be significant that the location of the site is close to the region that links RBD to the body of the ␣ 2 M, so that, in intact ␣ 2 M, it may be at the interface with another ␣ 2 M domain and may have higher affinity and serve a structural and/or functional role. In this regard, it might be noted that methylamine-transformed ␣ 2 M is only susceptible to the specific proteolytic cleavage that releases RBD at low pH. This requirement for low pH to render the linker accessible may result from dissociation of this calcium from the interface, thereby increasing mobility of the RBD and its linker.
Conclusion-We have used multinuclear NMR spectroscopy to determine the structure in solution of the recombinantly expressed 138-residue RBD from human ␣ 2 M and to characterize its calcium binding site. Comparison of the present struc-ture of human RBD with the recently solved crystal structure of RBD from bovine ␣ 2 M (17) shows that there is high similarity between them, despite the structure of bovine RBD being determined at higher pH, in the presence of 20 mM Ca 2ϩ and on a naturally glycosylated form. The structure of human RBD itself, together with the assignments of backbone and side chain NMR signals will, in the future, enable us to examine in solution the interactions of human RBD with domains from LRP and with ␤-amyloid peptide.