Localization of the Binding Site for Transforming Growth Factor-β in Human α2-Macroglobulin to a 20-kDa Peptide That Also Contains the Bait Region*

α2-Macroglobulin (α2M) functions as a major carrier of transforming growth factor-β (TGF-β) in vivo. The goal of this investigation was to characterize the TGF-β-binding site in α2M. Human α2M, which was reduced and denatured to generate 180-kDa subunits, bound TGF-β1, TGF-β2, and NGF-β in ligand blotting experiments. Cytokine binding was not detected with bovine serum albumin that had been reduced and alkylated, and only minimal binding was detected with purified murinoglobulin. To localize the TGF-β-binding site in α2M, five cDNA fragments, collectively encoding amino acids 122–1302, were expressed as glutathione S-transferase (GST) fusion proteins. In ligand blotting experiments, TGF-β2 bound only to the fusion protein (FP3) that includes amino acids 614–797. FP3 bound125I-TGF-β1 and 125I-TGF-β2 in solution, preventing the binding of these growth factors to immobilized α2M-methylamine (α2M-MA). The IC50 values were 33 ± 5 and 26 ± 6 nm for TGF-β1 and TGF-β2, respectively; these values were comparable with or lower than those determined with native α2M or α2M-MA. A GST fusion protein that includes amino acids 798–1082 of α2M (FP4) and purified GST did not inhibit the binding of TGF-β to immobilized α2M-MA. FP3 (0.2 μm) neutralized the activity of TGF-β1 and TGF-β2 in fetal bovine heart endothelial (FBHE) cell proliferation assays; FP4 was inactive in this assay. FP3 also increased NO synthesis by RAW 264.7 cells, mimicking an α2M activity that has been attributed to the neutralization of endogenously synthesized TGF-β. Thus, we have isolated a peptide corresponding to 13% of the α2M sequence that binds TGF-β and neutralizes the activity of TGF-β in two separate biological assays.

and TGF-␤2 in fetal bovine heart endothelial (FBHE) cell proliferation assays; FP4 was inactive in this assay. FP3 also increased NO synthesis by RAW 264.7 cells, mimicking an ␣ 2 M activity that has been attributed to the neutralization of endogenously synthesized TGF-␤. Thus, we have isolated a peptide corresponding to 13% of the ␣ 2 M sequence that binds TGF-␤ and neutralizes the activity of TGF-␤ in two separate biological assays.
Human ␣ 2 -macroglobulin (␣ 2 M) 1 is a 718-kDa glycoprotein that was originally characterized as a broad spectrum proteinase inhibitor (1). The structure of ␣ 2 M consists of four identical subunits, each with 1451 amino acids (2). The subunits are linked into dimers by disulfide bonds and into intact homotetramers by noncovalent interactions (3,4). Proteinases react with ␣ 2 M by cleaving any of a number of susceptible peptide bonds in the "bait region," which includes amino acids 666 -706 (1,3,5). Bait region cleavage causes ␣ 2 M to undergo a major conformational change, which effectively "traps" the attacking proteinase in a complex that is nondissociable, even when the proteinase and the inhibitor are not covalently linked (1, 6 -8). Conformational change also reveals binding sites for the ␣ 2 M receptor/low density lipoprotein receptor-related protein (LRP) (9). These binding sites have been localized to 18-kDa peptides at the C terminus of each ␣ 2 M subunit; Lys-1370 and Lys-1374 play particularly important roles (10 -13).
Like the complement components, C3 and C4, each ␣ 2 M subunit contains a novel thiol ester bond, which is formed from the side chains of Cys-949 and Glu-952 (14 -16). The thiol esters may be instrumental in determining the conformational state of ␣ 2 M (17,18). When ␣ 2 M reacts with a proteinase, the thiol esters emerge from within hydrophobic, solvent-restricted clefts and are cleaved by nucleophiles or H 2 O (14,18). Small primary amines, such as methylamine, penetrate the hydrophobic clefts and react with ␣ 2 M thiol esters independently of proteinases, inducing an equivalent or nearly equivalent conformational change (6,7).
In addition to its activity as a proteinase inhibitor, ␣ 2 M functions as a major carrier and regulator of certain cytokines, including isoforms of the transforming growth factor-␤ (TGF-␤) family. O'Connor-McCourt and Wakefield (19) first identified ␣ 2 M as a physiologically significant carrier of TGF-␤ in human serum (19). Their studies demonstrated that nearly all of the TGF-␤1 in serum is associated with ␣ 2 M and that the bound TGF-␤1 is inactive. Huang et al. (20) confirmed the role of ␣ 2 M as a TGF-␤-carrier and demonstrated that the TGF-␤ binding activity of ␣ 2 M depends on its conformational state.
More recent studies have demonstrated the function of ␣ 2 M as a TGF-␤-carrier in animal model systems. When radioiodinated TGF-␤1 is injected intravascularly in mice, the cytokine is cleared rapidly at first; however, this is followed by a slow clearance phase, during which time the TGF-␤ is almost entirely ␣ 2 M-associated (21)(22)(23). In cell culture systems, ␣ 2 M neutralizes both exogenously added and endogenously synthesized TGF-␤ (24 -28). Neutralization of endogenously synthesized TGF-␤ results in altered gene expression, including greatly increased expression of inducible nitric-oxide synthase by murine macrophages and increased expression of plateletderived growth factor ␣-receptor by vascular smooth muscle cells (27,28). ␣ 2 M gene knockout mice demonstrate increased tolerance to endotoxin challenge (29); this characteristic is most likely explained by the enhanced function of TGF-␤ as an immunosuppressant, in the absence of ␣ 2 M (30). The function of ␣ 2 M as a significant modulator of TGF-␤ activity in vivo and in vitro has prompted us to elucidate the ␣ 2 M-TGF-␤ interaction on a molecular level.
Binding of TGF-␤ to ␣ 2 M is initially noncovalent and reversible; however, the complex can become covalently stabilized as a result of thiol-disulfide exchange (23). The latter reaction is observed primarily with conformationally altered ␣ 2 M, since native ␣ 2 M lacks free thiol groups (23,31,32). We have used a number of complementary methods to determine equilibrium dissociation constants (K D ) for the interaction of TGF-␤ with ␣ 2 M (23, 31, 33). The K D values for the binding of TGF-␤1 and TGF-␤2 to native ␣ 2 M are 300 and 10 nM, respectively; the K D values for the binding of TGF-␤1 and TGF-␤2 to methylaminemodified ␣ 2 M (␣ 2 M-MA) are 80 and 10 nM, respectively. These binding constants accurately predict the ability of ␣ 2 M to neutralize TGF-␤ in cell culture systems (26,30,34,35).
The mechanism by which ␣ 2 M binds cytokines remains unclear. Early studies, suggesting a prominent role for the thiol ester-derived Cys-residues, were not confirmed for TGF-␤1 and TGF-␤2 (32). When ␣ 2 M-MA was treated with papain to release the 18-kDa receptor binding domains, the TGF-␤-binding activity remained with the residual 600-kDa ␣ 2 M fragment (36). Thus, the cytokine-and LRP-binding sites are not co-localized. The goal of this study was to determine whether TGF-␤-binding can be localized to a specific region in the structure of ␣ 2 M. An alternative model is that the central cavity in the structure of ␣ 2 M, which serves as the proteinase trap, also nonspecifically binds cytokines. Arguments in support of the alternative model include the complex quaternary structure of ␣ 2 M, the known trapping mechanism by which ␣ 2 M interacts with proteinases, and the large number of structurally unrelated cytokines that have been reported to associate with ␣ 2 M (37).
In this study, we present evidence demonstrating, for the first time, that TGF-␤ binding to ␣ 2 M is not dependent on ␣ 2 M quaternary structure and thus not dependent on the ␣ 2 M trap. Furthermore, we localize the TGF-␤-binding site to a single 20-kDa peptide that also contains the ␣ 2 M bait region. The 20-kDa peptide, when expressed as a GST fusion protein, binds both TGF-␤ isoforms with equivalent or increased affinity, compared with intact ␣ 2 M. The peptide also neutralizes the activity of TGF-␤ in endothelial cell proliferation assays and macrophage NO synthesis experiments. Thus, this 20-kDa peptide mimics the TGF-␤-regulatory activity of intact homotetrameric ␣ 2 M.
␣-Macroglobulins and Related Derivatives-Human ␣ 2 M was purified from plasma by the method of Imber and Pizzo (39). Murinoglobulin (MUG) was purified from the plasma of CD-1 female mice as described previously (30). SDS-PAGE analysis of purified MUG revealed a single band with an apparent mass of 180 kDa. ␣ 2 M-MA was prepared by dialyzing human ␣ 2 M against 200 mM methylamine-HCl in 50 mM Tris-HCl, pH 8.2, for 12 h at 22°C followed by extensive dialysis against 20 mM sodium phosphate, 150 mM NaCl, pH 7.4 (PBS), at 4°C. Complete modification of native ␣ 2 M by methylamine was confirmed by loss of trypsin binding activity (greater than 96%) (40) and by the characteristic increase in electrophoretic mobility, when analyzed by nondenaturing PAGE (41,42). Monomeric ␣ 2 M was prepared by exposing the native form of the protein to a high concentration of DTT (2 mM) under nondenaturing conditions, as described by Moncino et al. (43). Incompletely dissociated ␣ 2 M was separated from the monomers by FPLC on Superose-6. Monomeric ␣ 2 M, which is prepared as described, does not reassociate at 22°C (44).
Preparation of Constructs Encoding GST-␣ 2 M-Peptide Fusion Proteins-The human ␣ 2 M cDNA in pAT153/PvuII/8 (pAT-␣ 2 M) was obtained from the ATCC (16). Restriction digest analysis revealed an additional SacI cleavage site, which was not predicted by the published sequence (16), due to a single base substitution at nucleotide 2431 (C 3 T). To generate a construct encoding GST-␣ 2 M peptide fusion protein-1 (FP1), a fragment from pAT-␣ 2 M that encodes amino acids 122-415 was excised with BstXI, blunt-ended with T4 DNA polymerase, and ligated into pGEX-3X at the SmaI site. The construct encoding FP2 was prepared by digesting pAT-␣ 2 M with EcoRI and NsiI, to yield a partial cDNA encoding amino acids 364 -712, which was further digested with SacI, to generate a cDNA encoding amino acids 364 -613. This fragment was blunt-ended and ligated into pGEX-2T at the SmaI site. Constructs encoding FP3 and FP4 were prepared by isolating cDNAs, from a SacI digest of pAT-␣ 2 M, corresponding to amino acids 614 -797 and 798 -1082, respectively. These cDNAs were blunt-ended and ligated into the SmaI site of pGEX-2T. The construct encoding FP5 was prepared by digesting pAT-␣ 2 M with XhoI and PstI. A resulting cDNA, which encodes amino acids 1053-1302, was blunt-ended and ligated into pGEX-2T at the SmaI site. Restriction digest analysis of the five constructs confirmed that the ␣ 2 M cDNA inserts were in the correct orientation. Fig. 1 shows the relationship of the five peptides to the intact structure of ␣ 2 M.
Purification of GST-␣ 2 M Peptide Fusion Proteins-BL21 cells harboring pGEX-␣ 2 M-peptide expression constructs were induced with 0.1 mM isopropylthio-␤-D-galactoside for 3 h at 37°C, harvested by centrifugation, and resuspended in 50 mM Tris-HCl, 100 mM NaCl, 10 mM EDTA, 1 mM EGTA, pH 8.0. Nearly pure fusion protein preparations were generated by treating bacterial suspensions with 1 mg/ml lysozyme for 15 min on ice. The suspensions were then sonicated and subjected to centrifugation at 12,000 ϫ g for 10 min. All five fusion proteins remained in the insoluble fraction. These fractions were suspended in 10 mM deoxycholate for 2 h, sonicated, and subjected to a second centrifugation step. The fusion proteins, which again remained in the insoluble fractions, were solubilized by sonication in 2.0% SDS. To block free sulfhydryls, each fusion protein was reacted with 1 mM IAM in SDS for 2 h at 25°C. The IAM was then removed by dialysis. Final fusion protein preparations were stored in SDS. Protein concentrations were determined by the bicinchoninic acid (BCA) method.
Highly purified preparations of FP3 and FP4 were isolated in the absence of SDS by treating the original lysozyme extracts with 1.5% (w/v) Sarkosyl and 5 mM DTT. The FP3 and FP4, which solubilized in the Sarkosyl, were passed sequentially through 18-and 25-gauge needles and subjected to centrifugation at 12,000 ϫ g. The supernatants, which contained the fusion proteins, were treated with Triton X-100 (2% v/v) to sequester the Sarkosyl (45) and then subjected to affinity chromatography on glutathione-Sepharose 4B. FP3 and FP4, which eluted from the column, were dialyzed against 1.5% Sarkosyl, 1 mM DTT and treated with IAM (5 mM) for 2 h at 25°C to block free sulfhydryl groups. The final preparations were then dialyzed extensively against PBS.
Ligand Blotting-Native ␣ 2 M, ␣ 2 M-MA, MUG, and BSA were incubated in 2% SDS, in the presence or absence of 1 mM DTT, for 30 min at 37°C. To block free sulfhydryls, some samples were treated with 5 mM IAM for 2 h at 25°C. Equivalent amounts of each protein (5 g) were subjected to SDS-PAGE on 5% slabs. IAM-treated GST-␣ 2 M-peptide fusion proteins were subjected to SDS-PAGE as well. All samples were electrotransferred to PVDF membranes. The membranes were blocked with 5% milk and 0.1% Tween 20 in PBS for 12 h at 4°C and then rinsed with 0.1% Tween 20 in PBS (PBS-T). Membranes with native ␣ 2 M, ␣ 2 M-MA, MUG, and BSA were probed with 125 I-TGF-␤2 (20 pM), 125 I-TGF-␤1 (20 pM), or 125 I-NGF-␤ (50 pM) for 2 h at 25°C; membranes with the five GST-␣ 2 M peptide fusion proteins were probed with 125 I-TGF-␤2. The TGF-␤1 and TGF-␤2 were radioiodinated, to a specific activity of 100 -200 Ci/g, as described previously (46). NGF-␤ was radioiodinated with IODO-GEN, to a specific activity of 2-5 Ci/g, Western Blot Analysis-GST-␣ 2 M peptide fusion proteins were subjected to SDS-PAGE and electrotransferred to nitrocellulose membranes. The membranes were blocked with 5% milk in PBS-T for 12 h at 4°C, incubated with a polyclonal antibody that recognizes GST and then with peroxidase-conjugated goat anti-rabbit IgG. Binding of secondary antibody was detected by enhanced chemiluminescence (Amersham Pharmacia Biotech).
Binding of 125 I-TGF-␤2 to FP3 and FP4 as Determined by FPLC-125 I-TGF-␤2 (0.5 nM) was incubated with FP3 or FP4 (0.5 M) in PBS for 30 min at 37°C. The FP3 and FP4 were purified by glutathione affinity chromatography, treated with IAM, and free of detergents. 125 I-TGF-␤2-fusion protein complexes were separated from free 125 I-TGF-␤2 by FPLC on prepacked Superose-12 columns. The flow rate was 0.4 ml/ min. Elution of FP3 or FP4 was detected by monitoring the absorbance at 280 nm. 125 I-TGF-␤2 was detected in elution fractions using a ␥-counter. To calibrate the FPLC, the following proteins were subjected to chromatography on the same column: soybean trypsin inhibitor (M r ϳ21,500, V e of 14.1 ml), ovalbumin (M r ϳ45,000, V e of 12.9 ml), BSA (M r ϳ66,000, V e of 12.1 ml), and BSA dimer (M r ϳ132,000, V e of 10.9 ml). 125 I-TGF-␤ Binding to Immobilized ␣ 2 M-MA-␣ 2 M-MA (1 g in 100 l) was incubated in 96-well microtiter plates for 4 h at 22°C, as described previously (33). This procedure results in the immobilization of approximately 90 fmol of ␣ 2 M-MA. The wells were washed three times with PBS-T and blocked with PBS-T for 16 h at 4°C. As a control, some wells were blocked with PBT-T without first immobilizing ␣ 2 M-MA. 125 I-TGF-␤1 or 125 I-TGF-␤2 (0.1 nM) was incubated with the immobilized ␣ 2 M-MA in the presence of increasing concentrations of FP3 or FP4 (4 -250 nM) for 1 h at 22°C. The fusion proteins were purified and detergent-free. The wells were then washed three times with PBS-T. 125 I-TGF-␤, which was associated with the immobilized phase, was recovered in 0.1 M NaOH, 2% SDS and quantitated in a ␥-counter. Results were analyzed by plotting the specific binding of 125 I-TGF-␤ versus the log of the fusion protein concentration. In these experiments, the concentration of TGF-␤ (TGF-␤1 or TGF-␤2) was at least 100-fold lower than the K D for TGF-␤-binding to immobilized ␣ 2 M-MA. Thus, TGF-␤-binding was linearly related to the free TGF-␤ concentration ([␤ F ]), according to the following equation: In the presence of a fusion protein (FP) that binds TGF-␤, the total concentration of TGF-␤ ([␤ T ]) was related to [␤ F ], at equilibrium, as follows: If the fusion protein-TGF-␤ complex did not bind to immobilized ␣ 2 M-MA, then TGF-␤-binding was reduced by 50% (the IC 50 ) when [␤ T ]/[␤ F ] ϭ 2, and the fusion protein concentration that yielded the IC 50 was equal to the K I .
Endothelial Cell Proliferation Assays-FBHE cells were cultured in DMEM supplemented with 10% FBS, 20 ng/ml acidic fibroblast growth factor, and 80 ng/ml basic fibroblast growth factor and passaged at subconfluence with trypsin-EDTA. To perform proliferation assays, the cells were plated at a density of 2 ϫ 10 4 /well (24-well plates) in DMEM supplemented with 0.2% FBS. The cells were pulse-exposed to TGF-␤1 or TGF-␤2 (10 pM), in the presence and absence of FP3 or FP4 (200 nM), for 1 h. The fusion proteins were preincubated with the TGF-␤ for 15 min and then added to the cultures. At the completion of an incubation, the cultures were washed three times with serum-free DMEM and then allowed to incubate in DMEM with 0.2% FBS for 30 h. [ 3 H]Thymidine was added for an additional 18 h; the cells were then harvested, and [ 3 H]thymidine incorporation was quantitated.
Nitric Oxide Synthesis-NO synthesis by RAW 264.7 cells was quantitated by measuring the stable NO oxidation product, nitrite, in conditioned medium, as described previously (47). Cells were plated at a density of 10 4 /well in 96-well plates and cultured in RPMI 1640 with 10% FBS for 24 h and then in RPMI 1640 without serum (SFM) for an additional 24 h. ␣ 2 M-MA, FP3, FP4, or GST was added separately to the cultures in SFM. The fusion proteins were purified and detergent-free. After 24 h, conditioned medium (100 l) was recovered, and nitrite was measured. We previously demonstrated that ␣ 2 M increases RAW 264.7 cell NO synthesis by neutralizing endogenously produced TGF-␤ (27). The ␣ 2 M-induced increase in RAW 264.7 cell NO synthesis is inhibited by the nitric-oxide synthase inhibitor, N G -monomethyl-L-arginine.

RESULTS
Ligand Blot Analysis of 125 I-TGF-␤ Binding to ␣ 2 M-Native ␣ 2 M, ␣ 2 M-MA, and BSA were denatured in SDS (with or without reductant), subjected to SDS-PAGE, and electrotransferred to PVDF membranes. Some samples were treated with IAM prior to electrophoresis. The membranes were stained with Coomassie Blue, demonstrating nearly equivalent electrotransfer of the three proteins (results not shown). Unreduced ␣ 2 M migrated as a single band with an apparent mass of 360 kDa, as expected; reduced ␣ 2 M migrated as a single major band with an apparent mass of 180 kDa (3). Methylamine treatment did not alter the mobility of ␣ 2 M (14,15,48). 125 I-TGF-␤2 bound to native ␣ 2 M and ␣ 2 M-MA, which were immobilized on PVDF membranes (Fig. 2). 125 I-TGF-␤2 binding was unchanged when the ␣ 2 M was treated with DTT or with IAM prior to electrophoresis. 125 I-TGF-␤2 also bound to BSA; however, this interaction was observed only after DTT treatment and was eliminated by treating the BSA with IAM. Thus, binding of 125 I-TGF-␤2 to reduced BSA probably involves free sulfhydryl groups that are not available in the native BSA structure. The ability of isolated ␣ 2 M subunits to bind 125 I-TGF-␤2, by an IAM-insensitive mechanism, suggests that the ligand blotting system accurately models the interaction of TGF-␤ with nondenatured ␣ 2 M and that ␣ 2 M quaternary structure is not necessary for this interaction.
To further assess the growth factor binding activity of isolated ␣ 2 M subunits in the ligand blotting system, studies were performed with 125 I-TGF-␤1 and 125 I-NGF-␤. These two cytokines bind to nondenatured ␣ 2 M with similar affinity (31). As shown in Fig. 2, 125 I-TGF-␤1 and 125 I-NGF-␤ both bound to immobilized ␣ 2 M by an IAM-insensitive mechanism. Reductant-treated BSA also bound 125 I-TGF-␤1 and 125 I-NGF-␤; however, this interaction was eliminated when the BSA was treated with IAM.
Ligand Blot Analysis of the Binding of 125 I-TGF-␤2 to MUG-MUG is a monomeric murine homologue of human ␣ 2 M. Although tetrameric murine ␣ 2 M, in its native form, and human ␣ 2 M bind TGF-␤1 and TGF-␤2 similarly, MUG does not bind either TGF-␤ isoform with significant affinity (K D ϳ1.0 M) (30). Thus, we compared the binding of 125 I-TGF-␤2 to human ␣ 2 M and MUG, as another test of the validity of the ligand blotting method. As shown in Fig. 3, only trace levels of 125 I-TGF-␤2 bound to MUG, and the amount of binding was decreased when the MUG was treated with IAM. These results support the hypothesis that ligand blotting is a valid method for the analysis of cytokine binding to ␣-macroglobulins. Apparently, MUG does not contain a cryptic TGF-␤-binding site that is exposed by SDS treatment.
TGF-␤2-Binding to GST-␣ 2 M Peptide Fusion Proteins-The five fusion proteins were subjected to SDS-PAGE and electrotransferred to PVDF. The electrophoretic mobility of the major Coomassie-stained band, in each preparation, indicated a molecular mass that was identical to the mass of the monomeric fusion protein predicted by the cDNA sequence (Fig. 4). Western blot analysis with a GST-specific antibody confirmed that the major band in each lane was a GST fusion protein. The low mobility bands also bound GST-specific antibody and thus most likely represent SDS-insensitive fusion protein aggregates. In ligand blotting experiments, only FP3 bound 125 I-TGF-␤2. Since all five fusion proteins were IAM-treated, free sulfhydryl groups in FP3 did not account for the 125 I-TGF-␤2 binding. FP1, FP2, FP4, FP5, and purified GST (not shown) did not bind 125 I-TGF-␤2.
In separate ligand blotting experiments, affinity-purified FP3 and FP3 that was stored in SDS bound TGF-␤2 equivalently (results not shown). Thus, the two preparations were interchangeable when analyzed by this method. In order to demonstrate that 125 I-TGF-␤-binding to FP3 is noncovalent and specific, 125 I-TGF-␤ was incubated with PVDF-immobilized FP3 in the presence of excess solution phase FP3 or unlabeled TGF-␤. Binding of 125 I-TGF-␤2 to FP3 in Solution-125 I-TGF-␤2 (0.5 nM) was incubated with FP3 or FP4 (0.5 M) in solution, in the absence of detergents. Free and fusion protein-associated 125 I-TGF-␤2 were separated by FPLC on Superose-12. We previously demonstrated that free TGF-␤ interacts substantially with Superose and thus is recovered slowly at volumes that Samples that were DTT-or IAM-treated are designated by plus signs. All samples were subjected to SDS-PAGE and electrotransferred to PVDF membranes. The membranes were blocked and incubated with 125 I-TGF-␤2 for 2 h at 25°C. 125 I-TGF-␤2-binding was detected by PhosphorImager analysis. exceed the totally included volume (36). As shown by the absorbance tracings (280 nm) in Fig. 5, FP3 and FP4 eluted at volumes suggesting that these fusion proteins are dimers. The V e values were 11.4 and 11.2 ml for FP3 and FP4, respectively, corresponding to apparent masses of 95-and 107-kDa. Other GST fusion proteins are also expressed as noncovalent dimers (49). Substantial amounts of radioactivity co-eluted with FP3; 42% of the 125 I-TGF-␤2 was recovered with this fusion protein (n ϭ 2). By contrast, only 6% of the TGF-␤2 co-eluted with FP4. FPLC is a nonequilibrium method for assessing protein-protein interactions. The amount of complex detected may be significantly lower than that which was initially present (23).  Table I summarizes studies comparing the activities of FP3, native ␣ 2 M, monomeric ␣ 2 M, ␣ 2 M-MA, and thrombospondin as solution phase inhibitors of the binding of TGF-␤ to immobilized ␣ 2 M-MA. Some of the results were presented previously (31,33); however, all of the IC 50 values were determined by an identical method. The IC 50 values determined for FP3, native ␣ 2 M, and ␣ 2 M-MA, in studies with TGF-␤2, were all similar although the effective sequence in FP3 is contained in quadruplicate within the structure of intact, homotetrameric ␣ 2 M. The IC 50 determined for TGF-␤1 and FP3 was slightly lower than that determined with ␣ 2 M-MA and substantially lower than that determined with native ␣ 2 M. Interestingly, monomeric ␣ 2 M bound TGF-␤1 with slightly increased affinity compared with native ␣ 2 M, although each mol of native ␣ 2 M contains 4 mol of monomer. This result may be explained if the affinity of TGF-␤1 for its binding site in tetrameric ␣ 2 M is decreased compared with the affinity for the same site in monomeric ␣ 2 M and/or the number of available TGF-␤-binding sites within tetrameric ␣ 2 M is less than four.
FBHE Cell Proliferation Assays-To determine whether FP3-binding inhibits TGF-␤ activity, FBHE cell proliferation assays were performed. The cells were pulse-exposed to TGF-␤1 or TGF-␤2 (10 pM  was included in the medium. By contrast, FP3 nearly completely inhibited the activities of both TGF-␤1 and TGF-␤2, increasing [ 3 H]thymidine incorporation to within 3 and 6% of the control values. Regulation of NO Synthesis by FP3 and FP4 -␣ 2 M neutralizes TGF-␤, which is synthesized and activated endogenously by RAW 264.7 cells, and thereby induces expression of inducible nitric-oxide synthase (27,30). In order to determine whether FP3 neutralizes the activity of endogenously synthesized TGF-␤, we assessed the ability of the fusion protein to induce the production of nitrite in RAW 264.7 cell-conditioned medium. As shown in Fig. 7, FP3 increased NO synthesis in a concentration-dependent manner and, at low concentrations, was more active than ␣ 2 M-MA. We previously demonstrated that the increase in NO synthesis, which is induced by 280 nM ␣ 2 M-MA, is comparable with that observed with 10 ng/ml interferon-␥ (27). FP4 and purified GST (250 nM) did not increase nitrite production by the RAW 264.7 cells.

DISCUSSION
The TGF-␤ family of cytokines regulates diverse processes including cellular growth, differentiation, wound healing, and inflammation (for review, see Refs. 50 and 51). At the cellular level, TGF-␤ response is mediated by or regulated by a variety of receptors and binding proteins, including the type I and type II receptors, which are serine/threonine kinases, ␤-glycan, and endoglin. TGF-␤ activity is also regulated by processes that alter delivery of the active cytokine to the cell surface. For example, TGF-␤ is secreted as a large latent complex that includes latency-associated peptide and a second gene product, latent TGF-␤-binding protein (52)(53)(54). Conversion of latent TGF-␤ into active 25-kDa homodimer requires dissociation of latency-associated peptide and latent TGF-␤-binding protein in reactions that may be mediated by proteinases (55), thrombospondin (56), the mannose 6-phosphate/insulin-like growth factor-II receptor (57) and acidic microenvironments (58). Once activated, the 25-kDa form of TGF-␤ may bind to ␣ 2 M, once again forming a complex that is unavailable for receptor binding.
The fate of ␣ 2 M-associated TGF-␤ depends on the ␣ 2 M con-  formation. Native ␣ 2 M, which is the predominant form of ␣ 2 M present in the plasma and probably in most extravascular microenvironments, binds TGF-␤ reversibly and noncovalently (23,31,32). Thus, native ␣ 2 M may buffer tissues against rapid changes in TGF-␤ levels by binding or slowly releasing the cytokine in response to the free TGF-␤ concentration. Based on the K D value, we predict that approximately 95% of the TGF-␤1 in plasma is ␣ 2 M-associated under equilibrium conditions, even though TGF-␤1 binds to native ␣ 2 M with lower affinity than TGF-␤2 (31). Conversion of ␣ 2 M into the transformed conformation, which probably occurs most frequently at sites of inflammation due to the increase in cellular proteinase secretion, alters the mechanisms by which TGF-␤ is regulated. First, transformed ␣ 2 M has free Cys residues and thus undergoes thiol-disulfide exchange with TGF-␤ (23,31,32), eliminating the potential for release of active cytokine. Second, ␣ 2 M-proteinase complexes bind to the endocytic receptor, LRP; bound TGF-␤ is internalized with the ␣ 2 M-proteinase complex and probably delivered to lysosomes (22,36).
The goal of the present investigation was to identify the binding site for TGF-␤ in ␣ 2 M. Our original ligand blotting experiments, with human ␣ 2 M, demonstrated that intact quaternary structure is not necessary for TGF-␤ binding. Treatment of ␣ 2 M with IAM did not inhibit TGF-␤-binding, indicating that free Cys residues, which arose either as a result of thiol ester aminolysis or DTT treatment, are not involved. We also studied the binding of TGF-␤ to purified MUG by ligand blotting, since nondenatured MUG, unlike human ␣ 2 M and tetrameric murine ␣ 2 M, does not bind TGF-␤ with significant affinity (30). MUG bound only trace levels of TGF-␤ in ligand blotting studies, supporting our hypothesis that ligand blotting provides a valid model of TGF-␤-␣-macroglobulin interactions that occur under nondenaturing conditions. The inability of TGF-␤ and NGF-␤ to bind to reduced and alkylated BSA further supports the use of ligand blotting as a valid model system.
When the majority of the ␣ 2 M cDNA was expressed in a series of five GST fusion proteins, TGF-␤-binding was localized exclusively to FP3. The other four fusion proteins and purified GST did not bind TGF-␤. Selective binding of TGF-␤ to affinitypurified FP3, and not to FP4, was demonstrated by FPLC and by radioligand-binding competition assay. FP3 was more effective than native ␣ 2 M or ␣ 2 M-MA at inhibiting TGF-␤1 binding to immobilized ␣ 2 M-MA. This result is intriguing for at least three reasons. First, in comparing FP3 and intact ␣ 2 M, we used the concentrations of intact ␣ 2 M tetramer and FP3 monomer, although our FPLC results suggested that FP3 is a noncovalent dimer. If, instead, we had based the IC 50 values on the concentration of the ␣ 2 M "subunit," then the difference between FP3 and intact ␣ 2 M would have been 4-fold greater. Second, the experimentally determined IC 50 values accurately estimate the K I only if one molecule of competitor is sufficient to completely prevent TGF-␤-binding to immobilized ␣ 2 M-MA; otherwise, the K I is lower than the IC 50 . Although it is possible that two copies of FP3 or ␣ 2 M are required to neutralize TGF-␤, given the homodimeric structure of TGF-␤, this possibility is considered less likely with intact ␣ 2 M, due to its large size and complex structure, as discussed below. Also, as discussed below, our studies suggest that tetrameric ␣ 2 M may bind more than one molecule of TGF-␤. Finally, we cannot be certain that FP3 adopts a secondary and tertiary structure that is optimal for TGF-␤ binding. Taken together, these results suggest that a specific sequence in FP3 binds TGF-␤ with relatively high affinity. The equivalent sequence may be partially masked within intact ␣ 2 M, accounting for the observed decrease in TGF-␤ binding affinity. The masking of the TGF-␤-binding site in intact ␣ 2 M may also explain why ␣ 2 M conformational change markedly alters TGF-␤ binding affinity (31).
Human ␣ 2 M and bovine ␣ 2 M bind TGF-␤2 with increased affinity compared with TGF-␤1 (24,31), explaining why TGF-␤1 is preferentially active in certain cell culture assays that require serum-supplemented medium (24 -26). Danielpour and Sporn (24) provided evidence that the ␣-macroglobulins from rabbit also preferentially bind TGF-␤2. By contrast, murine ␣ 2 M binds TGF-␤1 and TGF-␤2 with equivalent affinity (30). In this study, we demonstrated that TGF-␤1 and TGF-␤2 bind to FP3 with equivalent affinity as well. This result suggests that the isoform specificity in TGF-␤ binding to certain ␣-macroglobulins may be due to the ability of TGF-␤2 to preferentially access the FP3-binding site in the intact ␣-macroglobulin. When the structural constraints of intact ␣ 2 M are eliminated, as in FP3, isoform specificity in TGF-␤ binding is no longer observed. We do not understand why the binding site for TGF-␤2 may be "less masked" in the structure of intact human ␣ 2 M compared with the binding site for TGF-␤1; however, NMR and x-ray crystallography studies have demonstrated the presence of small differences in the overall shape and structure of TGF-␤1 and TGF-␤2 (59 -61).
In addition to the TGF-␤-binding site, FP3 also contains the ␣ 2 M bait region. Models have been developed regarding the location of the bait region within the complex three-dimensional structure of ␣ 2 M based on electron microscopy (62,63); the x-ray crystal structure, which has been solved at 10-Å resolution (64); NMR and EPR spectroscopy studies (65,66); and fluorescence resonance energy transfer studies (67). The overall structure of ␣ 2 M resembles a hollow cylinder with a two-compartment central cavity. In ␣ 2 M-proteinase complexes, the proteinases occupy the central cavities. The bait regions are located within the central cavities, toward the center of the ␣ 2 M structure, and within 11-17 Å of the Cys residues (Cys-949) that form the thiol ester bonds (64). If, in fact, the bait region and the TGF-␤-binding site are equivalent or overlapping, then the TGF-␤-binding site may be accessible only from within the ␣ 2 M central cavity. TGF-␤-specific antibodies fail to recognize ␣ 2 M-associated TGF-␤ (19,24), supporting the hypothesis that TGF-␤ occupies the central cavity; however, it is not clear whether the ␣ 2 M, which was studied in the antibody experiments, was in the native or conformationally altered form. Thus, the location of the FP3-binding site for TGF-␤, within intact ␣ 2 M, remains unresolved. The bait region is known as an area of extreme sequence variability among ␣-macroglobulins from different species (5). Since TGF-␤-binding is conserved among many ␣-macroglobulins, with the exception of MUG (30,31,34), one can argue that the bait region and TGF-␤-binding site are unlikely to be equivalent. Further studies will be necessary to determine the relationship between these two important functional regions.
The stoichiometry of cytokine binding to ␣ 2 M has been estimated at 1:1 or 2:1 (19,68). Our results suggest that the binding site contained within a single ␣ 2 M subunit may be sufficient to bind TGF-␤. Thus, an estimate of four cytokinebinding sites per ␣ 2 M does not seem unreasonable. Limitations in the number of cytokine-binding sites in intact ␣ 2 M may result from steric hindrance. If ␣ 2 M-associated cytokines occupy the central cavity, then the number of cytokines that bind may be limited by the available cavity space. Of equal importance is the possibility that a high affinity complex between ␣ 2 M and TGF-␤ requires that the cytokine engage two equivalent copies of FP3 on different subunits. K D values, determined by the ␣ 2 M-MA immobilization method and by our previously described BS 3 -cross-linking method (31,33), assume a single cytokine-binding site per ␣ 2 M tetramer. If there are two independent binding sites, then the K D for each site would be increased by a factor of 2; however, our reported binding constants may still be most useful for predicting the cytokineneutralizing activity of ␣ 2 M in biological assays.
In FBHE cell proliferation assays, we demonstrated that FP3 not only binds TGF-␤1 and TGF-␤2 but also neutralizes the activities of these cytokines. When added to RAW 264.7 cell cultures, FP3 promoted the accumulation of nitrite more effectively than ␣ 2 M-MA. Since we previously demonstrated that the induction of NO synthesis by ␣ 2 M is due to the neutralization of TGF-␤ (27), we hypothesized that the increased potency of FP3 may be due to its increased binding affinity for TGF-␤1. To test this hypothesis, we measured the secretion of TGF-␤1 and TGF-␤2 by RAW 264.7 cells using isoform-specific enzymelinked immunosorbent assays. In medium that was conditioned for 24 h, the concentrations of active and total (active plus latent) TGF-␤1 were 2 and 10 pM, respectively. The concentrations of active and total TGF-␤2 were 1 and 4 pM, respectively. The active TGF-␤ levels reported here are only slightly lower than those determined previously using an endothelial cell growth assay (27). More importantly, the enzyme-linked immunosorbent assays confirm that RAW 264.7 cells express both TGF-␤ isoforms but higher levels of TGF-␤1, supporting the hypothesis that the increased potency of FP3 reflects its increased capacity to neutralize TGF-␤1.
In summary, we have identified a single peptide from the structure of ␣ 2 M that contains the binding site for TGF-␤1 and TGF-␤2. The high affinity of FP3 for both TGF-␤ isoforms and the substantial potency of FP3 in two TGF-␤ neutralization assays suggests that the TGF-␤-binding sequence may be partially masked in intact ␣ 2 M. Like TGF-␤1 and TGF-␤2, NGF-␤ bound to dissociated ␣ 2 M subunits, suggesting that intact quaternary structure and the resulting ␣ 2 M central cavity or trap is not necessary. However, at this time, we have not determined whether the NGF-␤-binding site or the binding site for any other cytokine is contained within FP3. The fusion proteins generated in this study will represent excellent templates for defining other cytokine-binding sites in ␣ 2 M and for further refinement of the TGF-␤ binding sequence.