Identification of the growth factor–binding sequence in the extracellular matrix protein MAGP-1

Microfibril-associated glycoprotein-1 (MAGP-1) is a component of vertebrate extracellular matrix (ECM) microfibrils that, together with the fibrillins, contributes to microfibril function. Many of the phenotypes associated with MAGP-1 gene inactivation are consistent with dysregulation of the transforming growth factor β (TGFβ)/bone morphogenetic protein (BMP) signaling system. We have previously shown that full-length MAGP-1 binds active TGFβ-1 and some BMPs. The work presented here further defines the growth factor–binding domain of MAGP-1. Using recombinant domains and synthetic peptides, along with surface plasmon resonance analysis to measure the kinetics of the MAGP-1–TGFβ-1 interaction, we localized the TGFβ- and BMP-binding site in MAGP-1 to a 19-amino acid–long, highly acidic sequence near the N terminus. This domain was specific for binding active, but not latent, TGFβ-1. Growth factor activity experiments revealed that TGFβ-1 retains signaling activity when complexed with MAGP-1. Furthermore, when bound to fibrillin, MAGP-1 retained the ability to interact with TGFβ-1, and active TGFβ-1 did not bind fibrillin in the absence of MAGP-1. The absence of MAGP was sufficient to raise the amount of total TGFβ stored in the ECM of cultured cells, suggesting that the MAGPs compete with the TGFβ large latent complex for binding to microfibrils. Together, these results indicate that MAGP-1 plays an active role in TGFβ signaling in the ECM.

The extracellular matrix (ECM) 2 is a composite biomaterial that contains informational signals that influence cell phenotypes. The changing composition of the ECM during development and tissue repair provides necessary feedback signals used by cells to build functional tissues. Information from the ECM is transmitted to cells through direct interaction with specific receptors (e.g. integrins), through mechanical signals that reflect the material properties of the ECM and cellular microenvironment, and through regulation of growth factor accessibility and signaling. Many ECM proteins bind growth factors and regulate their activity by providing a pericellular substrate for presenting growth factors to specific receptors on cells or by sequestering active and latent forms away from the cell for later utilization (1). Growth factors bind to discrete domains and motifs in ECM proteins, and there is remarkable specificity to these binding interactions. As the ECM changes during development, so does the repertoire of ECM-associated growth factors that influence cell phenotypes.
The microfibril is an extracellular matrix structure made up of fibrillin as the core unit. Fibrillin appeared early in evolution and, over time, developed both structural and signaling functions (2). Fibrillin monomers assemble into fibers of 12-15 nm diameter that provide strength to tissues by forming fiber bundles. An example includes the ciliary zonule in the eye, where microfibrils form the suspensory ligaments that support the lens (3,4). Microfibrils are also an important structural component of primitive vertebrate and invertebrate arteries, where they provide a degree of elastic recoil (5)(6)(7).
Another key function of microfibrils is modulation of growth factor signaling through fibrillin interactions with the latent TGF␤-binding proteins (LTBPs). LTBPs covalently bind the small latent complex of TGF␤ to form the large latent TGF␤ complex, which binds to fibrillin (8,9). By sequestering latent TGF␤ in the matrix, microfibrils indirectly regulate TGF␤ signaling by providing a store of inactive growth factor that can be utilized through activation by proteolysis or by mechanical strain (10).
With the arrival of vertebrates in evolution, new proteins appeared that interact with fibrillin and modify microfibrillar function. These include elastin and microfibril-associated glycoprotein-1 and -2 (MAGP-1 and MAGP-2). Elastic fiber formation requires microfibrils (11) and, because MAGP-1 can bind both tropoelastin (12,13) and fibrillin (14), the MAGPs were initially thought to promote the tropoelastin-microfibril interaction necessary for elastin assembly. However, inactivation of the genes for MAGP-1 and MAGP-2 in mice did not affect elastic fiber formation. Instead, mice lacking these proteins had abnormalities in bone, fat, hematopoiesis, and wound healing that were associated with abnormal TGF␤ signaling (15)(16)(17). We have previously documented a direct binding interaction between TGF␤ family growth factors and the MAGPs (18), but the sequence within MAGP that mediates this interaction was unknown. In this report, we localize the growth factor-binding domain in MAGP-1 to a sequence near the protein's N terminus and address the functional consequences of the TGF␤-MAGP interaction.

Characterization of recombinant purified full-length MAGP-1 and MAGP-1 fragments
Full-length mouse MAGP-1 (exons 3-9) was expressed in Escherichia coli along with a series of N-terminal truncation mutants containing an N-terminal His 6 tag. The truncations were based on the exon structure of the gene encoding the mature molecule shown in Fig. 1A (the protein sequence for mouse MAGP-1 can be found in Fig. S1). The gene for MAGP-1 contains nine exons, but only seven encode the secreted fulllength protein. Exon 1 is untranslated, and exon 2 encodes the majority of the signal peptide, which is absent from the mature molecule. After confirming the correct sequence for each construct by DNA sequence analysis, the constructs were expressed, and the proteins purified. The results of the purification are shown in Fig. 1B. The full-length protein has a predicted mass of 20.6 kDa, but by SDS-PAGE analysis migrates ϳ36 kDa. Similarly, the bacterially expressed domain 3-4 construct has a predicted mass of ϳ7.5 kDa but runs at ϳ16 kDa. Purified domain 4 -9 runs slightly higher than expected (19 kDa compared with the predicted 17.5 kDa), and the domain 5-9 protein runs true to the predicted mass of ϳ16 kDa. Because the MAGP-1 protein and fragments are expressed in bacteria, the irregular migration is not because of posttranslational modifications. Anomalous migration of full-length MAGP-1 on SDS-PAGE has been described (19). Our results suggest that the sequence within domains 3-4 is responsible for this irregular behavior.

TGF␤ binding to MAGP-1 fragments
Our goal in this study was to localize the region of MAGP-1 that interacts with TGF␤ growth factors. To this end, we performed a set of surface plasmon resonance (SPR) experiments where purified TGF␤-1 was injected over immobilized fulllength MAGP-1 or over the purified truncation fragments coupled to CM-5 sensor chips. Fig. 2A shows binding isotherms for soluble TGF␤-1 interacting with immobilized full-length MAGP-1 in the SPR-based assay. The affinity constant calculated using 1:1 Langmuir binding was 15 nM. When tested for binding to the MAGP-1 fragments, TGF␤-1 bound the domain 3-4 -coupled sensor chip with a slightly higher affinity than the full-length protein, as seen in Fig. 2B. The affinity constant calculated using 1:1 Langmuir binding was 3 nM. Binding of TGF␤-1 to domain 4 -9 showed substantially lower affinity (190 nM), and there was no TGF␤-1 binding to a chip coated with domain 5-9 (Fig. 2, C and D, respectively). These results suggest that the primary binding site for TGF␤-1 is located in domain 4 of MAGP-1 with an affinity contribution from domain 3.

TGF␤ binds to a 19-residue peptide sequence spanning domains 3 and 4
Fine mapping of the MAGP-1-binding sequence for TGF␤-1 was done using overlapping synthetic peptides that spanned the sequence encoded by exons 3 and 4. Each peptide was purified by HPLC, and the sequence confirmed by MS (data not shown). Surprisingly, TGF␤-1 failed to bind to a CM-5 sensor chip coated with a synthetic peptide made from the sequence of domain 3-4. The sequence was the same as the bacterially expressed domain 3-4 peptide except for an N-ter- A, full-length, secreted, MAGP-1 protein is encoded by seven exons (exons [3][4][5][6][7][8][9]. In this report, each exon defines a protein domain, and truncations used for functional analysis were constructed based on exon boundaries. The schematic shows the domains included in the recombinant fragments (d followed by exon numbers). B, SDS-PAGE analysis of recombinant full-length MAGP-1 (FL) and purified truncated peptides shown in panel A. Proteins were separated on 8 -25% Phastgels under reducing conditions and detected with Coomassie Brilliant Blue. The observed and expected molecular weights of each fragment are described in "Results."

TGF␤-binding sequence in MAGP-1
minal His 6 tag on the bacterially expressed protein. The His 6 tag itself was not an active binding sequence because the expressed peptide 5-9 did not bind TGF␤-1 even though it also had a His 6 tag at the N terminus.
Although unable to bind TGF␤-1 when bound to a solid substrate, domain 3-4 synthetic peptide (p1-35) did inhibit binding of TGF␤-1 to a full-length MAGP-1-coated sensor chip, thereby establishing its activity (Fig. S2A). Therefore, for subsequent mapping studies, we used equilibrium titration experiments to assess the inhibitory activity of a series of peptides summarized in Fig. 3A. The peptide based on the first 18 residues of domain 3 (p1-18) showed no inhibitory activity. A peptide derived from residues 22-36 (p22-36) and a domain 4 peptide (p26 -35) both failed to inhibit TGF␤-1 binding to fulllength MAGP-1 (the inhibitory curve for peptide p22-36 is shown in Fig. S2B as representative of no binding inhibition). A peptide with residues 12-37 (not shown) and a smaller peptide spanning residues 17-35 (p17-35) were both inhibitory.
A detailed investigation into the dose response of the smallest peptide with inhibitory activity (p17-35) is shown in Fig. 3B. There is a dose-dependent decrease in TGF␤-1 binding to fulllength MAGP-1, with half-maximal inhibition at ϳ30 M p17-35 (Fig. 3C). These results are consistent with data from studies with recombinant MAGP-1 fragments and show that the TGF␤-1 binding activity of MAGP-1 is defined by a 19amino acid sequence in domains 3 and 4. A control peptide with conformational restraints containing four proline residues (p17-35pro) failed to inhibit TGF␤-1 binding to full-length MAGP-1 (Fig. 3D).
Secondary structure predictions for the MAGP-1 p17-35 peptide indicated that the sequence does not have extensive secondary structure elements, with most residues predicted to be a random coil (data not shown). IUPred results predicted a high degree of disorder from the primary sequence. PEP-FOLD models tended to be elongated or compact coils with little secondary structure. This region of the MAGP-1 protein is likely intrinsically disordered and cannot adopt an ordered state without binding to a binding partner such as a member of the TGF␤ superfamily.

MAGP-1 binds active, but not latent, TGF␤-1
TGF␤ is produced as a disulfide-linked homodimer that is inactive because of the presence of the N-terminal latency-associated peptide. This small latent TGF␤ complex is usually associated with LTBP1 to produce the large latent TGF␤ complex, which is covalently bound to fibrillin-containing microfibrils in the ECM (9). To determine whether MAGP-1 interacts with latent TGF␤-1, we injected the small latent complex form of TGF␤-1 over a sensor chip coated with full-length MAGP-1. As shown in Fig. 4, latent TGF␤-1 at 15 nM failed to bind to the full-length protein. However, after acid activation for 15 min, the same latent TGF␤-1 preparation showed a robust interaction with MAGP-1.

TGF␤-binding sequence in MAGP-1 Neither MAGP-1 nor p17-35 inhibits TGF␤-1 signaling
To investigate whether TGF␤-1 can still interact with its cellular receptor when complexed with MAGP-1, we pre-mixed MAGP-1 or MAGP-1 fragments with active TGF␤-1 and then added the complex to the reporter cell line MFB-F11. MFB-F11 cells are embryonic fibroblasts from Tgfb1 Ϫ/Ϫ embryos stably transfected with a reporter plasmid consisting of TGF␤-responsive Smad-binding elements coupled to a secreted alkaline phosphatase reporter gene (20). The secreted alkaline phosphatase can be detected with a chemiluminescent substrate and provides a sensitive assay for TGF␤ activity. We found that full-length MAGP-1 does not influence TGF␤-1-stimulated alkaline phosphatase release from the MFB-F11 reporter cells when tested over a wide range of MAGP-1 concentrations (Fig.  5A). Similarly, MAGP-1 peptide p17-35 that blocks binding of TGF␤-1 to full-length MAGP-1, presumably by binding to the MAGP-1-binding site on the growth factor, does not affect the signaling potential of TGF␤-1 (Fig. 5B).

Predicted MAGP-1-binding site on TGF␤
Molecular docking experiments consistently placed the MAGP-1 peptide p17-35 in contact with the finger region of TGF␤-1 (Fig. 6). The peptide prefers to dock near a cluster of positive residues in this region (Arg-94, Lys-95, and Lys-97) that are solvent-exposed and, hence, in a position to attract and bind the negatively charged MAGP-1 peptide. The MAGP-1 peptide remains a highly flexible coil even when bound, suggesting that binding occurs through a promiscuous chargeinteraction between the highly positive region of TGF␤-1 and the negative residues of MAGP-1. The critical positively charged residues are inaccessible in the inactive TGF␤-LAP complex, because of the presence of the latency lasso in the straight-jacket region of the large latency-associated complex (21). Masking of the MAGP-1 binding site by the latency peptide explains why MAGP-1 binds active TGF␤ but not the small latent complex.

MAGP forms a ternary complex with fibrillin-2 and TGF␤-1
Fibrillin interacts with LTBP1 to covalently anchor the TGF␤ large latent complex into the ECM (9,22). Whether active TGF␤-1 can bind to fibrillin is unclear. To explore a possible binding interaction between the two proteins, we injected 50 nM of active TGF␤-1 over full-length fibrillin-2 immobilized on an SPR chip. Fibrillin-2 was purified from a baculoviral expression system and showed a single band by SDS-PAGE (Fig. S3). Fig. 7A shows that no significant interaction was detected between TGF␤-1 and fibrillin-2. Soluble MAGP-1, however, bound tightly to fibrillin-2 with an apparent affinity constant of 100 pM (Fig. 7B). The binding curve fits 1:1 Langmuir binding with a noticeable slow dissociation rate of Ϫ2.64 ϫ 10 4 s Ϫ1 and a half-life of 2600 s. We took advantage of this slow release rate to follow MAGP-1 injection onto the fibrillin-2 chip with 50 nM of TGF␤-1 and found robust binding of the growth factor to the fibrillin-2-MAGP-1 complex (Fig. 7C). That MAGP-1 mediated the interaction was established by inhibiting TGF␤ binding with the active MAGP-1 peptide p17-35 (Fig. 7C). The binding of TGF␤-1 to the MAGP-1-fibrillin-2 complex showed similar binding kinetics to TGF␤-1 binding directly to a MAGP-1-coated chip (compare the top curve in Fig. 3B to Fig.  7C).

BMP-2 binds to the same MAGP-1 sequence as TGF␤
Our previous studies found that MAGP-1 bound to BMP-7 with an affinity approximately equal to TGF␤ (18). To determine whether other BMP family members interact with MAGP-1, we performed SPR experiments with BMP-2. Fig. 8A shows BMP-2 binding to MAGP-1 with an affinity of 25 nM. The binding was inhibited by the active MAGP-1 peptide p17-35, suggesting that TGF␤s and BMPs recognize the same sequence on the MAGP-1 protein. The control peptide p17-35pro (Fig. 8B) and the N-terminal peptide p1-18 (not shown) both failed to inhibit BMP-2 binding to full-length MAGP-1. We were unable to determine whether BMP-2 could form a ternary complex with fibrillin-2 and MAGP-1 because BMP-2 bound directly to fibrillin-2 with an affinity of 8 nM (Fig. S4).

Matrix devoid of MAGPs has higher TGF␤ incorporation
The majority of TGF␤ in the ECM is in the form of the large latent complex bound to fibrillin. MAGP-1 and the large latent complex bind to the same region of fibrillin (23), suggesting a possible competition between the two proteins for binding to microfibrils. To determine whether the absence of MAGP allows for more latent complex to accrue in the ECM, we compared total and active TGF␤ levels in cultures of fibroblasts from WT and MAGP knockout cells. Because MAGP-2 can also bind TGF␤ (16) and might compensate in the absence of MAGP-1, we evaluated TGF␤ levels in MAGP-1 and MAGP-2 knockout MEFs as well as MEFs from MAGP-1:MAGP-2 dou-

TGF␤-binding sequence in MAGP-1
ble knockout mice. After 14 days of culture, there was two to three times more total (latent plus active) TGF␤ activity in acidactivated cell-ECM extracts of all MAGP knockout cells compared with WT ( Fig. 9, black bars). TGF␤ activity detected without acid activation (indicative of active TGF␤) was also increased in the single knockout cell lines but was significantly lower in the double knockout MEFs (Fig. 9, gray bars). Quantitative PCR confirmed that TGF␤ expression was similar for WT and MAGP knockout cells. Cell number was also equivalent for all cell types (Fig. S5).

Discussion
The bioavailability of active TGF␤ is tightly regulated at multiple levels, both inside and outside the cell. It is a crucial regulator of ECM production (24) yet, in turn, is influenced by the ECM itself in ways that provide spatiotemporal control of growth factor activity (1,25,26). In the ECM, TGF␤ family growth factors bind avidly to sulfated sugars on proteoglycans as well as to ECM proteins (1,26). Fibrillin-containing microfibrils are particularly important in modulating TGF␤ signaling and the mechanisms whereby microfibril-associated proteins regulate growth factor availability and activity are complex. We and others have shown that the fibrillin-binding partners MAGP-1 and MAGP-2 work together with fibrillin to modulate TGF␤ availability in the ECM (16,(27)(28)(29). Mutations in MAGP genes lead to phenotypes in mice and humans that are associated with abnormal growth factor signaling (15, 16, 18, 30 -33).
The MAGPs are functionally defined by the amino acid composition of their front and back halves. The back half of MAGP-1 contains 13 cysteine residues, the first 7 of which define the matrix-binding domain that mediates interactions with fibrillin (34,35). The matrix-binding region is the only conserved structural motif shared with MAGP-2 and is conserved in all species of MAGP-1 and MAGP-2. The cysteinefree N-terminal half of MAGP-1 is acidic, enriched in proline, and contains tyrosine residues that undergo sulfation and threonine residues that are sites for O-glycosylation. Sequences in this region also mediate noncovalent interactions with tropoelastin (12,13), the ␣3 chain of collagen VI (13), decorin (36), and biglycan (37). Although secondary structure prediction algorithms suggest a random configuration for the N-terminal

TGF␤-binding sequence in MAGP-1
sequences, the absence of cysteine residues, the high negative charge density, and the binding of numerous proteins to this area of the molecule suggests a structure that is accessible and adaptable.
In this report, we used an overlapping peptide mapping strategy to localize the growth factor-binding domain to 19 amino acids in the N-terminal region of MAGP-1. The TGF␤ binding site is contained in domain 4 but requires a part of domain 3 for full affinity (Fig. 2). The binding sequence is highly acidic because of four aspartic acid residues, one glutamic acid, and three tyrosine residues that are sites for sulfation (with the sulfate group adding more negative charge). The acidic residues form a structural repeat, with each separated by two nonpolar amino acids. The functional significance of this repeating struc-ture is not known. The high charge density at this site resembles characteristics of heparin, which binds TGF␤ with high affinity through negatively charged sulfated sugars that interact with lysine residues on TGF␤ (38). In fact, the predicted MAGP-1binding site on TGF␤ is the same as that postulated for heparin (38,39). Although the charged residues in MAGP-1 are important for growth factor binding, there are conformational and flexibility requirements as well, because the biologically active p17-35 peptide no longer interacts with TGF␤ when prolines are inserted to introduce structural constraints.

TGF␤-binding sequence in MAGP-1
MAGP-1 is one of the few matrix proteins that binds active TGF␤ without the involvement of glycosaminoglycans. Some ECM proteins, such as fibronectin, thrombospondin, fibulins, tenascins, and fibrillins, bind the large latent complex form of TGF␤ through interaction sites on the LTBPs (9, 40-43), although several bind the active isoform as well. In general, binding sites on matrix proteins for active TGF␤ are not wellcharacterized, but the properties of the MAGP-1 growth factor-binding domain may be instructive for identifying active sequences in these as well as other proteins. The microfibrillar protein emilin-1, for example, binds specifically to the proTGF␤ precursor and prevents its maturation by furin convertases (44). Growth factor-binding activity was mapped to the protein's EMI domain, but the precise sequence was not identified nor is it clear whether the interaction is with the active or latent regions of the growth factor. Similarly, TGF␤binding activity in the matricellular protein SPARC was localized in one study to the last 37 amino acids in the C-terminal EC domain (45), but other work found that SPARC interacted with TGF␤ only when bound to its cognate type II receptor (46). Interestingly, both emilin-1 and SPARC have clusters of acidic amino acids in the putative TGF␤-binding regions that create a negatively charged sequence similar to the growth factorbinding domain in MAGP-1.
Our mapping studies create a clearer picture of the functionality of the MAGP-1 N terminus. The first four amino acids in the TGF␤-binding peptide p17-35 overlap with the last four amino acids of a peptide used to define the binding site for tropoelastin and collagen VI (amino acids 12-21 in our numbering scheme) (13). It is not known, however, if TE or collagen VI can affect the binding of TGF␤. Although this N-terminal region seems to be available to interact with numerous proteins noncovalently, there is a specificity to the binding because MAGP-1 does not bind TGF␤-3; collagens I, II, or V; FGF9; FGF21; or the cytokines TNF␣; IL-4; IL-13; and RANKL (13,17).
Our finding that MAGP-1 binds active TGF␤ but not the small latent precursor form of the growth factor suggests that MAGP-containing microfibrils have the potential to store active TGF␤ in addition to the large latent form that covalently binds to fibrillin. The association of TGF␤ with MAGP-1 does not alter the growth factor's signaling potential, nor does an association between MAGP-1 and fibrillin alter MAGP-1's ability to bind TGF␤. Thus, activated TGF␤ bound to MAGP-1 on microfibrils is capable of signaling when interacting with its receptor on cells and does not require an additional activation step to do so.
It is noteworthy that the absence of MAGP-1 or MAGP-2 (or both) is sufficient to raise the amount of total TGF␤ stored in the ECM of cultured cells. This finding is consistent with studies by Massam-Wu et al. (23), who showed the MAGP-1 could directly compete with LTBP1 for binding to fibrillin. In the absence of MAGP, one would, therefore, expect more total TGF␤ to be associated with the ECM, which is what we found using MAGP knockout cells. Importantly, the presence of latent but not active TGF␤ in the ECM of MAGP-1:MAGP-2 double knockout cells suggests that in the absence of the MAGPs, active TGF␤ is not retained in the matrix.
The mechanistic and functional mapping studies described in this report, together with our previous characterization of MAGP knockout mouse phenotypes indicating increased TGF␤ activity, provide strong evidence for regulation of the TGF␤ pathway as one of MAGP's major functional roles in tissues.

Cell culture and TGF␤-1 activity assay
The TGF␤ reporter cell line MFB-F11 was a generous gift from Tony Wyss-Coray, Stanford University. The cells were routinely cultured in DMEM supplemented with nonessential amino acid, 10% fetal calf serum, and 100 g/ml hygromycin B (Invitrogen). On the afternoon before an assay, the cells were split into a 96-well assay plate (4 ϫ 10 4 cells per well) and cultured overnight in the absence of hygromycin. For the assay, cells were rinsed in DMEM then incubated for 1 h in DMEM-BSA (DMEM ϩ 1.0 mg/ml heat-denatured BSA). To determine whether MAGP-1 or the p17-35 peptide inhibits TGF␤ signaling, TGF␤ (80 pM) was mixed with varying concentrations of MAGP-1 protein or p17-35 peptide in DMEM-BSA for 15 min at room temperature then incubated with the reporter cells for 24 h. Soluble alkaline phosphatase released into the media in response to TGF␤ was measured using the Phospha-Light TM SEAP Reporter Gene Assay System (Life Technologies).

MEF and MFB-F11 co-culture
Levels of active and total TGF␤ in the ECM of cultured MEFs were assessed using established methods (20,49). Embryonic fibroblasts from WT and MAGP-1 knockout mice were generated from explants as described previously (18). Early passage cells (passage 4 or less) were maintained in DMEM with 10% fetal calf serum supplemented with nonessential amino acids and routinely passaged 1:3 at confluence. For growth factor measurements, cells were plated at confluence in replicate 12-well tissue culture dishes and maintained for 14 days prior to TGF␤ assays. The cells were rinsed with PBS and incubated for 1 h in DMEM containing 1 mg/ml of heat-denatured BSA. For measurement of active TGF␤ levels, ϳ1 ϫ 10 5 MFB-F11 reporter cells in DMEM heat-denatured BSA were added to the cultured MEF cells and incubated overnight. Supernatant from this co-culture was assayed for alkaline phosphatase using the Phospha-Light TM SEAP Reporter Gene Assay System (Life Technologies).
To determine total protein and total TGF␤, MEF cells were rinsed with PBS and the cell layer incubated in 250 l 0.2 M HCl per well for 1 h on ice. An aliquot was removed for total protein TGF␤-binding sequence in MAGP-1 determination and the remaining solution neutralized by the addition of an equimolar amount of NaOH. The cell-layer extract was diluted 1:100 in DMEM containing 1 mg/ml heatdenatured BSA and assayed for TGF␤ using MFB-F11 cells as described above. The samples taken for protein determination were dried and hydrolyzed in 6N HCl for 48 h, and amine content was measured with ninhydrin (50).

Smad2 phosphorylation
RFL-6 cells were plated at a confluent density and incubated overnight in Ham's F12 medium containing 0.5% horse serum followed by a 1-hour incubation with Ham's F12 containing 0.1 mg/ml heat denatured BSA to decrease background TGF␤ activity. TGF␤-1 (50 pM) was mixed with 0, 0.4, 20, or 1000 nM MAGP-1 in Ham's F-12 containing 0.1 mg/ml heat-denatured BSA for 1 h at 37°C then added to the RFL-6 cells for 15 min. Cells were lysed in buffer containing 20 mM Tris, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM ␤-glycerophosphate, 1 mM Na 3 VO 4 , 1 mM NaF 1ϫ protease inhibitor (Roche Applied Science). Lysates were cleared by centrifugation, and protein concentration determined. Eight g of lysate was run per lane on 10% SDS-PAGE gels then transferred to nitrocellulose membranes. Membranes were blocked in 1% casein then incubated overnight with the indicated primary antibodies. Antibodies were detected with secondary antibodies conjugated to horseradish peroxidase and visualized using chemiluminescent substrate (Millipore WBKLS0100). The standards used were Precision Plus Dual Color protein standards (Bio-Rad 161-0374).

MAGP-1 recombinant expression constructs
Mouse MAGP-1 cDNA was amplified from a full-length clone (IMAGE Consortium Clone:5249342) using a forward primer with a SphI restriction site added to the 5Јend (fulllength forward primer with Sph1) and a reverse primer with SacI restriction site added to 3Ј end (full-length reverse primer with SacI). The resultant amplification product along with pQE31 was digested with SphI and SacI, gel purified, and appropriate bands ligated. The ligation mix was used to transform the M15 strain of E. coli, and a bacterial colony containing a plasmid with the appropriate sequence was used for expression.
A fragment of MAGP-1 representing domains 3-4 was amplified from the full-length pQE31 construct using a forward primer (truncation forward) from pQE31 sequence and a reverse primer from the MAGP-1 sequence containing a SacI restriction site (exon 4 reverse with SacI). The amplification product, along with pQE31, was digested with EcoRI and SacI, gel purified, and appropriate bands extracted and ligated. Fragments for domain 4 -9 and 5-9 truncations were amplified from the full-length construct using a reverse primer from pQE31 (truncation reverse) sequence and a forward primer from MAGP-1 sequence containing a BamHI restriction site (exon 4 forward with BamH1 or exon 5 forward with BamHI). The amplification products were digested along with pQE31 with BamHI and SacI, gel purified, and the appropriate bands ligated. The ligation mix from each construct was used to transform M15 E. coli. Colonies were picked from the transformations and sequence verified. The primer sequences used to con-struct His 6 full-length and truncated MAGP-1 are as follows: full-length forward primer with Sph1, TAGCATGCTCAGGG-CCAATATGACCTGGA; full-length reverse primer with SacI, TAGAGCTCCTAGCAGCCCCCACAGCTCCTG; truncation forward, CACACAGAATTCATTAAAGAGGAG; exon 4 reverse with SacI, CAGAGCTCTCTTGGTAGTCATAGTA-GTCTG; exon 4 forward with BamH1, CAGGATCCTGA-CAACGCAGACTACTATG; exon 5 forward with BamH1, CAGGATCCAGAAGTGAGTCCTCGG; truncation reverse, TTAAGCTTGGCTGCAGGT.

Peptides, protein expression, and purification
His 6 -tagged fusion proteins were expressed and purified using the Qiagen Expression System (Valencia, CA). Bacteria expressing full-length or truncated protein were expanded, and while in log-phase, induced with isopropyl ␤-D-thiogalactopyranoside at 1 mM final concentration for 4 h at 30°C. The bacteria were collected by centrifugation, extracted overnight at 4°C with gentle agitation in TRIS-phosphate buffered urea (8 M, pH8.0) containing protease inhibitors. After clarifying the extract by centrifugation (12,000 ϫ g for 20 min), recombinant His 6 -containing proteins were isolated with nickel resin chromatography. Proteins were refolded on the column (51) and eluted with a 0 -300 mM imidazole gradient in 50 mM TRIS pH7.4. Fractions were analyzed by SDS-PAGE, and fractions containing protein were dialyzed against water and lyophilized. The fusion proteins were further purified using HPLC over a PRP-3 reverse-phase column using a 0 -30% acetonitrile gradient in 0.05% triethylamine. The protein content of purified proteins was determined by absorbance at 280 nm using an extinction coefficient based on the protein sequence (DNASTAR, Madison WI). The purity of the expressed proteins was assessed using SDS-PAGE with reducing (10 mM DTT) sample buffer on 8 -25% gradient Phastgels (GE Healthcare Life Sciences) and stained with SimplyBlue (Invitrogen). Peptides were synthesized by GenScript (Piscataway, NJ) and further purified using HPLC over a PRP-3 reverse-phase column using a 0 -30% acetonitrile gradient in 0.05% triethylamine. Fractions were dried and analyzed by amino acid analysis.

RNA isolation and quantitative PCR
RNA was isolated using a RNeasy kit (Qiagen 74104) after cell homogenization using QIAshredder (Qiagen 79656). cDNA was generated using the high-capacity RNA-to-cDNA kit (Applied Biosystems 4387406). Quantitative PCR was performed on a ViiA7 (Applied Biosystems) using reagents and TaqMan probes supplied by Applied Biosystems for GAPDH (Mm99999915G1) and TGF␤ (Mm01178820_m1). Quadruplicate measurements from 100 ng RNA were performed for each sample.

Purification of fibrillin-2
Fifty ml of serum-free conditioned medium from SF-9 cells expressing full-length fibrillin-2 3 was applied at a flow rate of 4 ml/min onto a 1.0 ϫ 9.5 cm Source 15Q strong anion exchange column (GE Healthcare) at room temperature using a Pharmacia FPLC system. The column was washed with 50 ml buffer A followed by a 10 ml gradient from 0 -25% buffer B. Fibrillin-2, which remained bound to the resin, was eluted with a 50-ml 25-50% linear buffer B gradient before the column was recycled with 15 ml 100% buffer B and 100 ml buffer A. The elution stream was monitored for protein at 228 nm and NaCl content using conductivity. Each fraction was assessed for fibrillin-2 using an anti-fibrillin-2 antibody via dot-blot and densitometry with ImageJ. The three peak fractions containing fibrillin-2 were pooled and further purified on a 1.6 cm ϫ 1.6 m Sephacryl S-500 column (fractionation range 40 -20,000 kDa) equilibrated in 25 mM HEPES, 125 mM NaCl, 0.1 mM CaCl 2 , pH 7.4. The column flow rate was 1 ml/min at 4°C, and 6 ml fractions were collected. The elution stream was monitored for protein at 280 nm and fibrillin-2 content for each fraction was determined by dot-blot. The final product was analyzed by SDS-PAGE on a 4% polyacrylamide gel with Laemmli buffer in the presence and absence of a reducing agent. Duplicate gels were either stained with SimplyBlue Coomassie G-250 or transferred to nitrocellulose and stained for fibrillin-2 by Western blotting. The purity of the fibrillin-2 is shown in Fig. S3. The standards used were Precision plus dual color protein standards (Bio-Rad). The Western blotting detection used a secondary donkey anti-rabbit antibody-HRP conjugate (Amersham Biosciences Inc.) that was visualized on film using a chemiluminescent substrate (EMD Millipore).

Surface plasmon resonance
Initial mapping of the TGF␤-1binding site in MAGP-1 was accomplished using a Biacore X to measure the change in surface plasmon resonance because of the interaction of TGF␤-1 in solution with expressed full-length MAGP-1 or truncated MAGP-1 fragments coupled to a CM-5 chip. For coupling, the protein was dissolved at 20 g/ml in 5 mM acetate buffer, pH4.5, and injected for 10 min onto a CM-5 sensor chip previously activated for 10 min with 0.05 M N-hydroxysuccinimide-0.2 M EDC. Reactive sites on the chip were quenched with 100 l of 100 mM ethanolamine, pH 8.0, then the chip was equilibrated in HBST running buffer (10 mM HEPES 150 mM NaCl, pH 7.4, containing 0.01% Triton-100). Varying concentrations of TGF␤-1 diluted in running buffer were then injected. The volume of TGF␤-1 injected was 60 l, and the flow rate was 20 l/min. The precise localization of the growth factor-binding site in MAGP-1 was determined using inhibition of TGF␤-1 binding to full-length MAGP-1 by synthetic peptides. These experiments and the latent TGF␤ binding studies were performed on a Reichert SR7000DC SPR system. Full-length MAGP-1 was coupled to a carboxymethyl dextran hydrogel surface sensor chip using the same coupling conditions used for the Biacore CM-5 chip. TGF␤-1 or BMP-2 (50 nM final concentration) was mixed with varying concentrations of MAGP-1 synthetic peptide at room temperature for at least 15 min prior to injection. The flow rate for the localization studies was 50 l/min, and the injection volume was 100 l.
Latent TGF␤-1 (5 M stock solution) was diluted to 15 nM in HBST and injected over the MAGP-1 chip. To activate the latent TGF␤-1, HCl was added to the latent TGF␤-1 stock solu-tion to a final concentration of 100 mM for 15 min at room temperature followed by neutralization with the addition of an equimolar concentration of NaOH. The activated TGF␤-1 was diluted to 15 nM in HBST and injected over the full-length MAGP-1 chip.
In ternary complex studies, fibrillin-2 was diluted 1:5 in 5 mM sodium acetate and coupled with repeat 250 l injections onto a CM-5 activated and blocked as described above. Approximately 4000uRU were coupled to the chip. Direct binding of TGF␤-1 and BMP-2 to fibrillin-2 was initially assessed prior to loading MAGP-1 onto the fibrillin-2 chip. After direct binding experiments, the chip was recycled, and MAGP-1 was loaded onto the chip with duplicate MAGP-1 injections (100 l injection at 500 nM) where ϳ130 RU bound. Binding of TGF␤-1 and BMP-2 to the complex of fibrillin-2 and MAGP-1 was then re-assessed. No recycling was performed between injections on the fibrillin-2 MAGP-1 complex. HBST was used as the sample dilution buffer as the mobile phase. Unless otherwise noted, chips were recycled with 50 l injections of 0.2 M glycine pH, 2.3.

MAGP-TGF␤ interaction predictions
TGF␤ family structures were downloaded from the Protein Data Bank: 1TGK (58), 4KV5 (59), 1KLA, 1KLC, and 1KLD (60). Structures were edited to delete any additional ligands or binding partners. When appropriate, structures were subjected to energy minimization in NAMD (61) to produce a relaxed structure. Based on experimental SPR results from the lab, the MAGP-1 p17-35 sequences were chosen as docking candidates. The sequences were treated as flexible peptides and were docked to the TGF␤ structures using GalaxyDock (62) and CABS-dock (63). Predicted structures were evaluated by score and feasibility. Structures were manually compared with structures of known TGF␤-containing complexes, specifically receptor-bound TGF␤ (PDB ID 3KFD) (64) and TGF␤ in the large latent complex (PDB IDs) 3RJR (21) and 5VQP (65). Structures were visualized using VMD (66) and VMD was used to generate all images.