Distinct Binding Sites in the Structure of α2-Macroglobulin Mediate the Interaction with β-Amyloid Peptide and Growth Factors*

α2-Macroglobulin (α2M) and its receptor, low density lipoprotein receptor-related protein (LRP), function together to facilitate the cellular uptake and degradation of β-amyloid peptide (Aβ). In this study, we demonstrate that Aβ binds selectively to α2M that has been induced to undergo conformational change by reaction with methylamine. Denatured α2M subunits, which were immobilized on polyvinylidene difluoride membranes, bound Aβ, suggesting that α2M tertiary and quaternary structure are not necessary. To determine whether a specific sequence in α2M is responsible for Aβ binding, we prepared and analyzed defined α2M fragments and glutathioneS-transferase-α2M peptide fusion proteins. A single sequence, centered at amino acids (aa) 1314–1365, was identified as the only major Aβ-binding site. Importantly, Aβ did not bind to the previously characterized growth factor-binding site (aa 718–734). Although the Aβ binding sequence is adjacent to the binding site for LRP, the results of experiments with mutated fusion proteins indicate that the two sites are distinct. Furthermore, a saturating concentration of Aβ did not inhibit LRP-mediated clearance of α2M-MA in mice. Using various methods, we determined that the K D for the interaction of Aβ with its binding site in the individual α2M subunit is 0.7–2.4 μm. The capacity of α2M to bind Aβ and deliver it to LRP may be greater than that predicted by theK D , because each α2M subunit may bind Aβ and the bound Aβ may multimerize. These studies suggest a model in which α2M has three protein interaction sites with distinct specificities, mediating the interaction with Aβ, growth factors, and LRP.

␣ 2 -Macroglobulin (␣ 2 M) and its receptor, low density lipoprotein receptor-related protein (LRP), function together to facilitate the cellular uptake and degradation of ␤-amyloid peptide (A␤). In this study, we demonstrate that A␤ binds selectively to ␣ 2 M that has been induced to undergo conformational change by reaction with methylamine. Denatured ␣ 2 M subunits, which were immobilized on polyvinylidene difluoride membranes, bound A␤, suggesting that ␣ 2 M tertiary and quaternary structure are not necessary. To determine whether a specific sequence in ␣ 2 M is responsible for A␤ binding, we prepared and analyzed defined ␣ 2 M fragments and glutathione S-transferase-␣ 2 M peptide fusion proteins. A single sequence, centered at amino acids (aa) 1314 -1365, was identified as the only major A␤-binding site. Importantly, A␤ did not bind to the previously characterized growth factor-binding site (aa 718 -734). Although the A␤ binding sequence is adjacent to the binding site for LRP, the results of experiments with mutated fusion proteins indicate that the two sites are distinct. Furthermore, a saturating concentration of A␤ did not inhibit LRP-mediated clearance of ␣ 2 M-MA in mice. Using various methods, we determined that the K D for the interaction of A␤ with its binding site in the individual ␣ 2 M subunit is 0.7-2.4 M. The capacity of ␣ 2 M to bind A␤ and deliver it to LRP may be greater than that predicted by the K D , because each ␣ 2 M subunit may bind A␤ and the bound A␤ may multimerize. These studies suggest a model in which ␣ 2 M has three protein interaction sites with distinct specificities, mediating the interaction with A␤, growth factors, and LRP.
Accumulation of ␤-amyloid peptide (A␤  and A␤  ) 1 in the brain plays a central role in the development and progression of Alzheimer's disease (AD) (1). Mutations in ␤-amyloid precursor protein (APP), which result in increased production of A␤, are associated with autosomal dominant forms of familial AD in humans (2)(3)(4). Mutated forms of human APP may also induce changes consistent with AD when expressed as transgenes in mice (5)(6)(7)(8). Furthermore, immunization with A␤  prevents progression of AD in animal model systems and may reverse symptoms by promoting resorption of A␤containing plaques (9,10). These results suggest that A␤ accumulation in the brain is a dynamic and reversible process. Proteins other than antibodies with the capacity to bind A␤ and promote its catabolism may influence disease progression.
␣ 2 -Macroglobulin (␣ 2 M) is a 718-kDa homotetrameric glycoprotein, which is well characterized as an extracellular proteinase inhibitor (11) and as a carrier of specific growth factors, including transforming growth factor-␤ (TGF-␤) and nerve growth factor-␤ (NGF-␤) (12,13). At least two separate polymorphisms in the A2M gene may be associated with increased risk of late-onset AD. The first involves a region within intron 17, at the 5Ј splice acceptor site for exon 18 (14). This exon is important because it encodes part of the bait region, where proteinases initiate reaction with ␣ 2 M by cleaving susceptible peptide bonds (15,16), and a segment of the growth factor binding sequence (17)(18)(19). In the second A2M gene polymorphism, Val-1000 is replaced by Ile (20). The linkage of A2M gene polymorphisms to late-onset AD remains incompletely understood, because the original observations have been confirmed in only a limited number of populations (21)(22)(23)(24)(25) and because there is no molecular explanation regarding how A2M gene mutations may affect ␣ 2 M structure, function, and expression. ␣ 2 M is expressed by microglia, which accumulate near amyloid plaques (26). Thus, locally synthesized ␣ 2 M may affect AD progression by regulating the activity of various proteinases or by binding important growth factors. The previously demonstrated ability of ␣ 2 M to bind and neutralize the activity of TGF-␤ (12,13,(27)(28)(29) may be detrimental in AD, because TGF-␤ stimulates A␤ clearance by microglial cells and reduces A␤ accumulation in the brain parenchyma of mice that overexpress human APP (30). Furthermore, TGF-␤ has been reported to antagonize the cytotoxic activity of A␤ (29,31,32).
Another mechanism whereby ␣ 2 M may regulate AD progression involves its ability to bind A␤, forming a complex that is internalized by the ␣ 2 M receptor, low density lipoprotein receptor-related protein (LRP) and then degraded (33)(34)(35). Du et al. (36) originally reported that A␤  and A␤  bind to native ␣ 2 M and to ␣ 2 M that has been transformed into the LRP-recognized or "activated" conformation by reaction with methylamine (␣ 2 M-MA). Narita et al. (33) subsequently reported selective binding of A␤  and A␤  to the activated conformation of ␣ 2 M. ␣ 2 M-MA apparently binds A␤  and A␤  with equivalent affinity (33). Hughes et al. (37) executed a yeast two-hybrid screen using A␤  as bait and identified a 250-amino acid peptide from the C terminus of ␣ 2 M as a strong and specific interactor. The same group also reported experiments confirming the interaction of A␤ with intact ␣ 2 M; however, they did not demonstrate that the sequence identified by yeast-two hybrid screen is responsible for the binding of A␤ to intact ␣ 2 M.
The growth factor binding site in ␣ 2 M is contained within a 16-amino acid peptide located ϳ500 amino acids N-terminal to the A␤-binding site identified by yeast-two hybrid screen (19). The growth factor binding sequence is composed mainly of hydrophobic amino acids with two potentially important acidic residues. TGF-␤, platelet-derived growth factor-BB (PDGF-BB), and NGF-␤ all interact with the growth factor-binding site in ␣ 2 M (18,19), despite the fact that these proteins demonstrate limited sequence identity. Based on this promiscuous behavior, we hypothesized that the growth factor-binding site in ␣ 2 M may also function as an A␤-binding site.
To test our hypothesis, we undertook a comprehensive molecular analysis to identify sequences in ␣ 2 M with A␤ binding activity. Our results demonstrate that a single sequence, located near the C terminus of the ␣ 2 M subunit, constitutes the only significant A␤-binding site. Importantly, this sequence is entirely distinct from the growth factor-binding site. The LRP recognition sequence is also located near the C terminus of the ␣ 2 M subunit (38 -42); however, our evidence indicates that the LRP recognition site and the A␤ binding sequence are distinct. Thus, in addition to the bait region, the ␣ 2 M subunit has at least three distinct "protein interaction sites" with distinct binding specificities. These sites mediate interactions with growth factors, A␤ and LRP.

MATERIALS AND METHODS
Proteins and Reagents-␣ 2 M was purified from human plasma by the method of Imber and Pizzo (43). ␣ 2 M-MA was prepared by dialyzing ␣ 2 M against 200 mM methylamine-HCl in 50 mM Tris-HCl, pH 8.2, for 12 h at 22°C and then exhaustively against 20 mM sodium phosphate, 150 mM NaCl, pH 7.4. Modification of ␣ 2 M by methylamine was confirmed by demonstrating the characteristic increase in ␣ 2 M electrophoretic mobility by non-denaturing PAGE (15). ␣ 2 M-MA was radioiodinated using IODO-BEADs (Pierce) and stored at 4°C for no more than 2 weeks. The specific activity was 0.5-1.0 Ci/g. Receptor-associated protein (RAP), which blocks binding of ␣ 2 M-MA to LRP (65), was expressed as a glutathione S-transferase (GST) fusion protein in bacteria and purified by chromatography on glutathione-Sepharose. A␤  was purchased from Bachem and radioiodinated using 125 I-labeled Bolton-Hunter reagent (di-iodinated, PerkinElmer Life Sciences). Biotinylated A␤  was prepared by reacting A␤  with 4 M sulfo-N-hydroxysuccinimide biotin (Pierce) for 2 h at 4°C in siliconized tubes. The reaction mixture was dialyzed extensively against water. Biotinylated A␤ was stored for up to 1 month at 4°C or frozen at Ϫ80°C and thawed once without affecting its ability to bind to ␣ 2 M. GST-specific IgG, bovine serum albumin (BSA, greater than 99% pure), dithiothreitol (DTT), and iodoacetamide were from Sigma Chemical Co. Bis(sulfosuccinimidyl) suberate (BS 3 ) and horseradish peroxidase-conjugated avidin were from Pierce. Polyclonal A␤-specific rabbit antibody was from Zymed Laboratories Inc.
Methods for Defined Fragmentation of ␣ 2 M-When ␣ 2 M is treated with papain under mildly acidic conditions, an 18-kDa fragment is released from the C terminus of each ␣ 2 M subunit (aa 1314 -1451) (38). The 18-kDa fragment includes the intact receptor-binding site and is thus referred to as the receptor binding fragment (RBF). The residual 600-kDa ␣ 2 M remnant retains the major structural features of the parent molecule (45). To obtain the 18-and 600-kDa ␣ 2 M fragments, 4.0 M ␣ 2 M-MA was treated with 2.4 M papain in 50 mM sodium acetate, 1 mM cysteine, pH 5.0, for 20 h at 22°C. The pH of the reaction mixture was increased to 7.4, and the products were purified by molecular exclusion chromatography on Ultrogel AcA-22.
Each ␣ 2 M subunit has a single thiol ester bond formed by the side chains of Cys-949 and Gln-952 (11,46,47). When ␣ 2 M is heated in the presence of SDS, the thiol esters react internally, and, as a result, the ␣ 2 M peptide backbone is cleaved (47,48). The products include a 120-kDa N-terminal heat fragment and a 60-kDa C-terminal heat fragment. To produce ␣ 2 M heat fragments, native ␣ 2 M was incubated at 100°C in 2% SDS (w/v) and 14 mM DTT for the indicated periods of time. The products were treated with iodoacetamide (70 mM) and subjected to SDS-PAGE.
All of the fusion proteins were partially purified from induced bacterial suspensions by selective detergent extraction, as previously described (17). The resulting preparations yielded clearly defined bands, with the correct molecular masses when assessed by Coomassie Blue staining of SDS gels or immunoblot analysis with GST-specific antibody. FP3, FP6, and FP6d-AA were purified to homogeneity by chromatography on glutathione-Sepharose. Fig. 1 shows the sequences of the GST fusion proteins used in this investigation.
Non-denaturing PAGE Analysis of A␤ Binding to ␣ 2 M-125 I-A␤ (2.5 nM) was incubated with native ␣ 2 M, ␣ 2 M-MA, or the purified 600-kDa fragment (0.3-1.0 M) in 20 mM sodium phosphate, 150 mM NaCl, pH 7.4, for 2 h at 37°C. In some experiments, increasing concentrations of the 18-kDa RBF (0.2-2.8 M) were co-incubated with 125 I-A␤ and ␣ 2 M-MA. Reaction mixtures were subjected to non-denaturing PAGE, using the buffer system described by Van Leuven et al. (49). 125 I-A␤ binding to ␣ 2 M was detected as radioactivity co-migrating with the ␣ 2 M band. In control experiments, free 125 I-A␤ did not migrate near ␣ 2 M. To quantitate 125 I-A␤ binding to ␣ 2 M, gels were subjected to PhosphorImager analysis using ImageQuant software. Non-denaturing PAGE preserves non-covalent interactions; however, the amount of binding detected may be influenced by dissociation of protein complexes during electrophoresis (13).
Determination of Apparent Equilibrium Dissociation Constants-Because A␤ binding to ␣ 2 M is reversible and probably subject to rapid dissociation when methods such as non-denaturing PAGE or chromatography are used, we utilized the BS 3 rapid cross-linking method to determine the apparent K D for the binding of A␤ to ␣ 2 M-MA. This method has been used previously to determine K D values for the interaction of ␣ 2 M with multiple growth factors and cytokines (12,13,50).
Increasing was then added for 5 min. Cross-linking reactions were quickly terminated by rapid acidification, followed by transfer to buffered SDS. Under pseudo-first order conditions, a constant fraction of the noncovalent 125 I-A␤⅐␣ 2 M-MA complex is covalently stabilized by the BS 3 (13). To quantitate the amount of covalently stabilized complex, BS 3treated and vehicle-treated samples were subjected to SDS-PAGE. 125 I-A␤ that was covalently cross-linked to ␣ 2 M-MA (bound) and free 125 I-A␤ (free), which includes free A␤ and A␤ that was bound to ␣ 2 M-MA but not cross-linked, were quantitated by PhosphorImager analysis. Results were analyzed according to the following equation (12), The cross-linking efficiency, z, is a constant, derived from the y intercept, for each set of proteins and conditions (12). z is referred to as the BS 3 -cross-linking efficiency but may also be affected if a fraction of the radioiodinated protein is incapable of binding to the ␣ 2 M. The apparent K D was determined from the slope when free/bound was plotted against 1/[␣ 2 M-MA]. This value is based on the assumption that there is a single binding site for A␤ in ␣ 2 M. Assuming one A␤-binding site/␣ 2 M subunit, as suggested by our data, then the K D must be corrected by multiplying the apparent K D by a factor of four.
Ligand Blotting-This method has been previously used to demonstrate specific and saturable binding of growth factors to denatured ␣ 2 M subunits, ␣ 2 M fragments, and GST-␣ 2 M-peptide fusion proteins (17)(18)(19). Protein preparations were denatured in 2% SDS or treated with 1 mM DTT in 2% SDS and then with 5 mM iodoacetamide for 2 h, as previously described (17). Samples were then subjected to SDS-PAGE and electrotransferred to polyvinylidene fluoride (PVDF) membranes. The membranes were blocked with 5% milk in 20 mM sodium phosphate, 150 mM NaCl, 0.1% Tween 20, pH 7.4, and probed for 2 h with 125 I-A␤ or biotinylated-A␤. 125 I-A␤ ligand blots were washed and subjected to PhosphorImager analysis. Biotinylated A␤ ligand blots were probed with horseradish peroxidase-conjugated avidin (1:5000 dilution). The membranes were then subjected to enhanced chemiluminescence (ECL) and densitometry. Equivalent loading and transfer of proteins were demonstrated by Coomassie Blue staining or, when applicable, by immunoblot analysis with GST-specific antibody (17).
A␤-peptide Immunoblotting-Denatured ␣ 2 M subunits and BSA were treated with 1 mM DTT and then with 5 mM iodoacetamide to block free sulfhydryl groups, subjected to SDS-PAGE, and electrotransferred to PVDF. The membranes were blocked with 5% milk. Unlabeled A␤  was incubated with the immobilized ␣ 2 M and BSA in PBS-T at 37°C for 2 h. The membranes were then washed extensively and probed with rabbit A␤-specific IgG (1:4000) in PBS-T and 0.1% milk (v/v), followed by anti-rabbit IgG-horseradish peroxidase conjugate (1: 10,000). Membranes were analyzed by ECL and densitometry.
Plasma Clearance Experiments in Mice-125 I-␣ 2 M-MA (20 nM) was incubated with 20 M A␤ or with vehicle for 2 h at 37°C. The 125 I-␣ 2 M-MA (0.3 Ci) was then injected, in the presence and absence of GST-RAP (40 or 80 g), into the lateral tail veins of CD-1 female mice (30 g). Blood samples (40 l) were withdrawn from the retro-orbital venous plexus, using heparinized capillary tubes, at the designated times (0.5-30 min). The radioactivity in each sample was determined using a gamma counter and expressed as a fraction of that present in the 0.5-min time point.  and A␤  function differently in the initiation and progression of AD; however, both forms of A␤ bind to ␣ 2 M equivalently (33). Thus, we conducted our analysis of A␤ binding to ␣ 2 M and its derivatives using one form of A␤ (A␤  ). To determine whether A␤ binding to ␣ 2 M is ␣ 2 M conformation-specific, as has been demonstrated with growth factors (12,13), 125 I-A␤ (2.5 nM) was incubated with native ␣ 2 M or ␣ 2 M-MA (each at 0.3 M) in solution. The products were analyzed by non-denaturing PAGE, which preserves non-covalent interactions. As shown in Fig. 2A, 125 I-A␤ bound to ␣ 2 M-MA, whereas binding was not detected with native ␣ 2 M. Free 125 I-A␤ migrated near the dye front. These results suggest that A␤ binds selectively to ␣ 2 M that has undergone conformational change.

A␤ Binding to ␣ 2 M Is ␣ 2 M Conformation-dependent-A␤
To estimate the K D for A␤ binding to ␣ 2 M-MA, we used the BS 3 -rapid cross-linking method, which has been used exten-sively to determine binding affinities for ␣ 2 M and growth factors (12,13,50). A major advantage of this method is that it is not necessary to resolve free and bound A␤, which typically involves the use of steps, such as chromatography or PAGE, that promote dissociation of non-covalent protein complexes. A representative study in which 125 I-A␤ was incubated with increasing concentrations of ␣ 2 M-MA is shown in Fig. 2B. The exact fraction of the non-covalent ␣ 2 M⅐A␤ complex, which was cross-linked by BS 3 (z), was determined from the y intercept, as previously described (12). In three separate experiments, z ϭ 0.06 -0.14, compared with z values that are typically in the range of 0.15-0.40 for the binding of growth factors to ␣ 2 M (12). z may be decreased if a fraction of the 125 I-A␤ was incapable of binding to ␣ 2 M-MA or if the A␤, which bound to ␣ 2 M-MA, multimerized so that individual A␤ monomers could not be cross-linked to the ␣ 2 M-MA. Neither of these effects would be expected to influence the calculated apparent K D .
The apparent K D for the binding of A␤ to ␣ 2 M-MA was 0.29 Ϯ 0.02 M (n ϭ 3). This value is based on the assumption that each molecule of ␣ 2 M has one binding site for A␤. If each ␣ 2 M subunit has a distinct A␤-binding site, as the evidence to be presented will suggest, then the K D for the binding of A␤ to the individual binding site is 1.2 M. Although this is a low affinity interaction, due to the homotetrameric structure of ␣ 2 M, the plasma concentration of ␣ 2 M subunits is 12-20 M (11).
Binding of 125 I-A␤ to Denatured ␣ 2 M Subunits-Native ␣ 2 M was denatured in SDS and DTT and treated with iodoacetamide. A similar protocol was executed with ␣ 2 M-MA and with BSA. The preparations were then subjected to SDS-PAGE. Coomassie Blue staining revealed the 180-kDa ␣ 2 M subunit as the major band in both the native ␣ 2 M and ␣ 2 M-MA preparations, as anticipated (Fig. 3). Faint bands with apparent masses of 120 and 60 kDa were observed in the native ␣ 2 M lane. These bands correspond to the ␣ 2 M heat fragmentation products that result from an internal reaction involving the thiol ester bonds at 100°C, as previously described (48).  125 I-A␤ bound to denatured ␣ 2 M subunits that were electrotransferred to PVDF membranes. No difference in 125 I-A␤ binding was observed with native ␣ 2 M and ␣ 2 M-MA, as was anticipated, because the difference in structure between these two forms of ␣ 2 M is mainly conformational. 125 I-A␤ did not bind to BSA, suggesting that the interaction with ␣ 2 M is specific. The interaction of 125 I-A␤ with ␣ 2 M, in the ligand blotting system, suggests that the individual ␣ 2 M subunit binds A␤ and that ␣ 2 M tertiary and quaternary structure are not necessary. In this respect, A␤ binding to ␣ 2 M resembles the interaction observed with growth factors (17) but not with proteinases (11,15).
To confirm that the interaction of A␤ with ␣ 2 M was not dependent on an unanticipated modification occurring during A␤ radioiodination, we developed alternative methods for detecting A␤ binding to PVDF-immobilized ␣ 2 M subunits. In the first protocol, unlabeled A␤ was used to probe the PVDF membranes. ␣ 2 M-associated A␤ was then detected by immunoblot analysis. In the second protocol, biotinylated A␤ was substituted for 125 I-A␤. In both cases, binding of A␤ to PVDF-immobilized native ␣ 2 M and ␣ 2 M-MA was detected whereas A␤ binding to BSA was not.
A␤ Binding to ␣ 2 M Heat Fragments-When ␣ 2 M is heated in the presence of denaturant, the thiol ester bonds, which are formed from the side chains of Cys-949 and Gln-952, react internally with the ␣ 2 M polypeptide backbone, causing scission of the ␣ 2 M subunit at residue 952 (46 -48). The N-terminal 120-kDa fragment includes the bait region and the growth factor binding sequence. The C-terminal 60-kDa fragment includes the LRP recognition sequence (38 -42)  To determine whether A␤ binding activity is localized to either or both of these denatured ␣ 2 M fragments, ␣ 2 M was subjected to heat fragmentation and analyzed by 125 I-A␤-ligand blotting. Only the 60-kDa ␣ 2 M heat fragment bound 125 I-A␤ (Fig. 4). The 120-kDa ␣ 2 M heat fragment was without activity. This result provides evidence that a specific sequence is responsible for the interaction of ␣ 2 M with A␤. Furthermore, this result suggests that the ␣ 2 M growth factor-binding site and the A␤-binding site are distinct.
The 18-kDa RBF Competes with ␣ 2 M-MA for A␤ Binding-A second method for defined fragmentation of ␣ 2 M involves papain treatment of the activated conformation under mildly acidic conditions. An 18-kDa fragment, which retains LRP binding activity, is dissociated from the C terminus of each ␣ 2 M subunit (aa 1314 -1451) (38 -40). The residual 600-kDa fragment retains the major structural characteristics of ␣ 2 M-MA, as determined by electron microscopy (45). The 18-and 600-kDa ␣ 2 M fragments were purified and assessed for their ability to bind A␤ without prior denaturation. When 125 I-A␤ was incubated with the 600-kDa fragment, in solution, binding was not detected by non-denaturing PAGE (Fig. 5A). Under equivalent conditions, binding was readily detected with intact ␣ 2 M-MA.
In separate experiments, 2.5 nM 125 I-A␤ was incubated with 0.3 M ␣ 2 M-MA and increasing concentrations of purified 18-kDa RBF in solution. 125 I-A␤ binding to ␣ 2 M-MA was decreased in the presence of the 18-kDa RBF, and the magnitude of the effect was dependent on the RBF concentration (Fig. 5B). The IC 50 was 0.7 M. Because of the relatively high concentration of ␣ 2 M-MA, the K D for A␤ binding to the 18-kDa RBF was ϳ2-fold lower (0.3-0.4 M) than the IC 50 . These results provide further evidence that an A␤-binding site is localized near the C terminus of the ␣ 2 M subunit and that this site is distinct from the growth factor-binding sequence.
A␤ Binding to GST-␣ 2 M-peptide Fusion Proteins-To comprehensively analyze the ␣ 2 M sequence with regard to A␤ binding, we utilized ligand blotting to screen a series of six previously described ␣ 2 M-peptide-GST fusion proteins (FP1-FP6) (18). In the intact ␣ 2 M subunit, the bait region and growth factor-binding site are located in FP3. The residues that comprise the thiol ester bond are located in FP4, and the LRP recognition sequence is in FP6. To assess A␤ binding, the fusion proteins were treated with iodoacetamide, without prior reduction, and subjected to SDS-PAGE. After electrotransfer to PVDF, only FP6 bound 125 I-A␤ (Fig. 6A).
Because intact disulfide bonds may allow partial restoration of non-denatured structure following protein electrotransfer to PVDF membranes, ligand-blotting experiments were also performed using FP3 and FP6 that were reduced with DTT and then alkylated with iodoacetamide. In these experiments, 125 I-A␤ binding was still detected only with FP6 and not with FP3 (Fig. 6B). Equivalent results were obtained when biotinylated A␤ was substituted for 125 I-A␤ (Fig. 6C). Based on these results, a model emerges in which the structure of ␣ 2 M includes at least two distinct protein interaction sites with differing specificity. A site located near the center of the ␣ 2 M subunit is responsible for the binding of growth factors whereas a separate site near the C terminus is exclusively responsible for the binding of A␤.

Resolution of the LRP-and A␤-binding Sites in ␣ 2 M-In
intact human ␣ 2 M, aa 1370 -1377 constitute an ␣ helix that is the center of the LRP recognition site (41,42). The ␣ helix is anchored in position by a ␤ sandwich so that the side chains of two critical Lys residues (aa 1370 and 1374) protrude at 45°a ngles and are surrounded by hydrophobic surface residues (41). This complex secondary and tertiary structure may explain why the 18-kDa RBF is recognized by LRP, whereas tryptic peptides corresponding to the same region and partially denatured forms of the 18-kDa RBF are not (38,41).
Because our ligand blotting results demonstrated that tertiary structure is not necessary for A␤ binding to FP6, we generated a new set of fusion proteins to explore the relationship between the LRP-and A␤-binding sites in ␣ 2 M. FP6c included all of the amino acids that form the LRP recognition ␣ helix, five amino acids N-terminal to the ␣ helix and the entire sequence C-terminal to the ␣ helix; however, FP6c did not bind A␤ (Fig. 7). FP6a, which included the N-terminal segment of FP6 but terminated five amino acids before the start of the ␣ helix, bound A␤, albeit at lower levels than FP6. These results suggest that the A␤-binding site is located N-terminal to the LRP binding ␣ helix. FP6b, which was equivalent to FP6a but extended through the ␣ helix to aa 1400, bound slightly increased levels of A␤, suggesting that amino acids 1365-1400 may impact positively on this interaction; however, ligand blotting results provide only an approximation of differences in binding affinity.
Fusion proteins, which correspond exactly to the sequence of the 18-kDa RBF, were generated and mutated to essentially eliminate the LRP-binding site (FP6d-AA) (39) or substantially reduce binding to LRP while eliminating binding to the previously described ␣ 2 M signaling receptor (FP6d-AR) (40). By A, PVDF membranes with FP1-FP6, which had been denatured in the absence of reductant and treated with iodoacetamide, were probed with 125 I-A␤. B, a representative study in which FP3 and FP6 were denatured in the presence of DTT, treated with iodoacetamide, and subjected to ligand blot analysis with 125 I-A␤. C, FP3, FP4, and FP6 were denatured in 2% SDS and DTT, treated with iodoacetamide, subjected to SDS-PAGE, and electrotransferred to PVDF membranes. The membranes were probed with biotinylated A␤ or subjected to immunoblot analysis with GST-specific antibody.

FIG. 7. Resolution of the A␤-and LRP-binding sites in the ␣ 2 M subunit.
A, fusion proteins containing amino acids that constitute the LRP recognition site and other fusion proteins that do not were subjected to ligand blot analysis with 125 I-A␤. As a control for load, blots were stained with Coomassie Blue or subjected to immunoblot analysis with GST-specific antibody. B, the results of at least eight separate experiments with each fusion protein were pooled. In each case, 125 I-A␤ binding to a fusion protein was standardized against that observed with FP6 (mean Ϯ S.D.). ligand blotting, both forms of FP6d retained A␤ binding activity, supporting our hypothesis that the A␤ and ␣ 2 M receptor recognition sequences in ␣ 2 M are non-identical. A slight decrease in the binding of A␤ to FP6-AA and FP6-AR, compared with FP6, may indicate that the mutated Lys residues, although non-essential, impact positively on the interaction.
Determination of the K D for A␤ Binding to FP6 -In the ligand blotting experiments, the concentration of 125 I-A␤, used as probe, was substantially lower than the likely K D value for 125 I-A␤ binding to any of the fusion proteins. Thus, assuming equivalent load and insignificant contributions from "low affinity" or "nonspecific" binding sites, the amount of binding observed is inversely proportional to the K D for each interaction. To more accurately assess the binding affinity of A␤ for FP6 and FP6d-AA, specific binding experiments were performed.
FP6 and FP6-AA were purified to homogeneity; however, unlike intact ␣ 2 M-MA and the 18-kDa RBF, the fusion proteins did not bind A␤ in solution, even when refolding protocols were executed. Possible explanations for this observation are provided under "Discussion." As an alternative approach, we immobilized purified FP6, FP6d-AA, and FP3 on PVDF after exposure to SDS and probed the membranes with 125 I-A␤ (0.1 M) and increasing concentrations of unlabeled A␤. Nonspecific binding was defined by the level of 125 I-A␤ binding observed in the presence of 30 M unlabeled A␤. As shown in Fig. 8, specific binding of 125 I-A␤ to both FP6 and FP6d-AA was detected. The K D values were 2.4 Ϯ 0.8 and 5.2 Ϯ 0.6 M, respectively (mean Ϯ S.E., n ϭ 3 with internal triplicate replicates). The B max , which is not an informative value when this method is used, was 20 -25% higher with FP6. Importantly, the binding of 125 I-A␤ to FP3 was entirely nonspecific.
Effects of A␤ on ␣ 2 M Recognition by LRP-Others have demonstrated that LRP mediates cellular uptake and degradation of A␤ by binding A␤⅐␣ 2 M complexes (33)(34)(35). Because of the close proximity of the A␤ and LRP recognition sequences in ␣ 2 M, we considered the possibility that A␤ might inhibit binding of ␣ 2 M to LRP even though the binding sites are nonidentical. Each ␣ 2 M tetramer has four independent LRP recognition sequences (38). Thus, even if some of the LRP-binding sites were blocked by A␤, receptor recognition may still occur. To address this question, we examined the plasma clearance of ␣ 2 M-MA in mice. In this well characterized system, conformationally modified forms of ␣ 2 M, such as ␣ 2 M-MA, clear from the plasma as a first-order process with a t1 ⁄2 of 3-5 min, and clearance competition is observed when 125 I-␣ 2 M-MA is injected in the presence of excess unlabeled ␣ 2 M-MA (51). The rapid plasma clearance of ␣ 2 M-MA represented a clear advantage for our experiments with A␤, compared with binding or endocytosis experiments performed in vitro, because of the opportunity to minimize the time period during which dissociation of A␤⅐␣ 2 M-MA complex may occur. Fig. 9 shows the plasma clearance of 125 I-␣ 2 M-MA (n ϭ 4) in the absence of competing ligand and in the presence of 40 or 80 g of GST-RAP. The GST-RAP inhibited the clearance of 125 I-␣ 2 M-MA from the plasma, as anticipated due to competition for plasma-accessible LRP, which is mainly located in the liver (51). To determine whether A␤ inhibits ␣ 2 M-MA binding to LRP, 125 I-␣ 2 M-MA was preincubated with a nearly saturating concentration of A␤ (20 M, 16-fold the K D ) for 2 h at 37°C and then injected intravenously in mice. The rate of 125 I-␣ 2 M-MA clearance was completely unchanged. We cannot rule out the possibility that A␤ dissociated from the 125 I-␣ 2 M-MA after the preparation was injected intravascularly, due to dilution in the bloodstream; however, given the rapid timeframe of the plasma clearance experiments, our results demonstrate that ␣ 2 M-associated A␤ either does not interfere with LRP recognition or rapidly dissociates from some sites to free up LRP recognition sequences and allow uptake of the remaining A␤. DISCUSSION LRP and its ligands, ␣ 2 M, apolipoprotein E4, and Kunitzproteinase inhibitor domain-containing isoforms of APP, form an intriguing functional family, the members of which have been implicated in familial or late-onset AD (52). Deciphering the mechanisms whereby these proteins affect AD may be difficult because of their multifunctional nature. For example, LRP may mediate the clearance of A␤ in association with ␣ 2 M and may be essential for A␤ transport across the blood-brain barrier (53). However, LRP may also mediate the transfer of APP into intracellular compartments where there is increased access to amyloidogenic proteinases (54).
Like LRP, ␣ 2 M expresses multiple activities that may be involved in AD progression. Because ␣ 2 M is a broad-spectrum proteinase inhibitor, which reacts with proteinases from all four major mechanistic classes (15,55), various proteinases that are involved in A␤ catabolism, such as neprilysin, insulysin, and plasmin (56 -59), may be ␣ 2 M targets. The possibility that ␣ 2 M regulates the activity of proteinases involved in APP processing has also been considered; however, De Strooper et al. (60) demonstrated that this does not occur. The lack of an effect of ␣ 2 M on APP processing is consistent with other studies demonstrating that ␣ 2 M is poorly reactive with proteinases functioning at or near the cell surface (61). The ability of ␣ 2 M to function as a carrier and deliver proteins to LRP for catabolism was first demonstrated with TGF-␤ and PDGF-BB (62,63). Only activated ␣ 2 M is functional in this capacity, because native ␣ 2 M is not recognized by LRP (51). Growth factors, such as TGF-␤, bind to native ␣ 2 M, as well as activated ␣ 2 M, and the resulting effects on growth factor activity are complicated. When bound to native ␣ 2 M, growth factors are typically inactive; however, because this interaction is reversible, ␣ 2 M-associated growth factors may provide a reservoir, buffering against changes in the free growth factor concentration (13,64).
In our experiments, A␤ bound selectively to the LRP-recognized or -activated form of ␣ 2 M, confirming the work of Narita et al. (33). To activate ␣ 2 M, we reacted the protein with methylamine. This reaction induces a conformational change in ␣ 2 M that is equivalent to the structural rearrangement induced by proteinases (11,15,44). In the ligand blotting system, denatured ␣ 2 M subunits retained A␤ binding activity. This result suggests that ␣ 2 M tertiary and quaternary structures are not necessary for A␤ binding. Instead, we hypothesize that ␣ 2 M activation reveals an otherwise cryptic linear sequence of amino acids that constitute a binding site for A␤. ␣ 2 M conformational change is also necessary for recognition by LRP; however, this interaction apparently requires retention of secondary and tertiary structure in the ␣ 2 M RBF (38,41).
Although the primary sequence in ␣ 2 M that is responsible for the binding of growth factors is fairly promiscuous, this sequence did not interact with A␤ and apparently did not contribute to the A␤ binding activity of intact ␣ 2 M. Instead, a distinct protein-interaction site was identified, and our analysis of GST fusion proteins suggests that the center of this site is located between amino acids 1314 and 1365. The A␤ binding sequence identified in our experiments may be equivalent to the candidate A␤-binding site identified by yeast-two hybrid screen (37). Based on these results, we now propose that ␣ 2 M contains at least two distinct protein-interaction sequences that are functional even when higher order ␣ 2 M structure is eliminated. These two binding sites demonstrate distinct ligand binding specificities, because the growth factor-binding site in FP3 does not bind A␤, and the A␤-binding site in FP6 does not bind TGF-␤, PDGF-BB, or NGF-␤ (18).
The LRP recognition sequence includes two Lys residues within an ␣ helix that includes amino acids 1370 -1377 (41,42). In the intact three-dimensional structure of the ␣ 2 M RBF, the two Lys residues are oriented so that the side chains are readily available for interaction with LRP. Furthermore, the Lys residues are surrounded by a high density of hydrophobic surface residues. All of our evidence indicates that the A␤binding site and the LRP recognition sequence are adjacent but distinct. FP6c, which includes the entire ␣ helix, did not bind A␤, whereas FP6a, which lacks the ␣ helix, did. Mutants of the RBF, which have been previously shown to not bind LRP (39,40), still bound A␤. Furthermore, a saturating concentration of A␤ did not inhibit the plasma clearance of ␣ 2 M-MA, which is mediated by LRP (51). We propose that the center of the A␤ binding sequence is located on the N-terminal side of the LRP recognition ␣ helix. Comparison of the structure of RBFs from various ␣-macroglobulins has demonstrated that one surface of the RBF is highly conserved (41). Jenner et al. (41) proposed that the conserved surface is divided into two patches, one of which constitutes the LRP recognition site. Only speculation was offered regarding the function of the second patch, which includes amino acids from our fusion proteins that bind A␤. The possibility that this second conserved surface patch represents an A␤-binding site merits consideration.
Determining the binding affinity of A␤ for ␣ 2 M and its de-rivatives is not straightforward. With regard to intact ␣ 2 M, protein conformation is clearly very important, and commercial sources of ␣ 2 M typically contain mixtures of the different conformational forms. Our evidence suggests that each ␣ 2 M subunit contains an independent A␤-binding site so that there are four sites per molecule. Finally, A␤ may multimerize when associated with ␣ 2 M, and this may have an effect on the total A␤ binding capacity of ␣ 2 M. Previously reported K D values for the A␤⅐␣ 2 M interaction were 380 pM (36) and 350 nM (37). The K D values determined here for A␤ binding to intact ␣ 2 M-MA, purified 18-kDa RBF, and FP6 are in good agreement with the value reported by Hughes et al. (37). In each case, we demonstrated that A␤ binding is saturable. Specificity in binding was demonstrated in control experiments with BSA and other fusion proteins, such as FP3. Although the affinity of the ␣ 2 M subunit for A␤ was low, we hypothesize that the ability of ␣ 2 M to function as an A␤ binding protein and shuttle A␤ to LRP is actually much higher than would be predicted, based on the K D , due to the fact that each ␣ 2 M has four A␤-binding sites and because multimerized A␤ may bind to ␣ 2 M. Although the purified 18-kDa RBF bound A␤ in solution, under equivalent conditions, FP6 and the FP6-derived fusion proteins demonstrated only limited A␤ binding activity. Apparently, exposure to denaturant (SDS) was necessary to fully reveal the A␤-binding site in the fusion proteins. In the absence of denaturant, the A␤-binding site, in FP6, may be partially buried or form a compatible surface for FP6 self-association.
From this study, a model emerges in which the ␣ 2 M subunit includes two linear sequences that form protein interaction sites with distinct specificities. The resolution of the growth factor-binding site and the A␤-binding sequence in ␣ 2 M is intriguing, given the accumulating evidence that ␣ 2 M may inhibit AD progression by promoting A␤ clearance while promoting AD progression by neutralizing TGF-␤. We hypothesize that it may be possible to generate ␣ 2 M derivatives, by sitedirected mutagenesis, with defective growth factor binding activity and intact A␤ binding activity. Such derivatives may provide an important tool for understanding the function of ␣ 2 M in AD.