Alternative Binding Modes Identified for Growth and Differentiation Factor-associated Serum Protein (GASP) Family Antagonism of Myostatin*

Background: GASP-1 and GASP-2 are highly specific antagonists for the TGF-β ligand myostatin, a negative regulator of muscle growth. Results: GASP-1 and GASP-2 form asymmetric and symmetric complexes with myostatin, respectively. Conclusion: Despite the different binding modes, the GASP proteins retain a high specificity for myostatin. Significance: Inhibition of myostatin can be achieved using different binding modes and may facilitate future development of novel anti-myostatin therapeutics. Myostatin, a member of the TGF-β family of ligands, is a strong negative regulator of muscle growth. As such, it is a prime therapeutic target for muscle wasting disorders. Similar to other TGF-β family ligands, myostatin is neutralized by binding one of a number of structurally diverse antagonists. Included are the antagonists GASP-1 and GASP-2, which are unique in that they specifically antagonize myostatin. However, little is known from a structural standpoint describing the interactions of GASP antagonists with myostatin. Here, we present the First low resolution solution structure of myostatin-free and myostatin-bound states of GASP-1 and GASP-2. Our studies have revealed GASP-1, which is 100 times more potent than GASP-2, preferentially binds myostatin in an asymmetrical 1:1 complex, whereas GASP-2 binds in a symmetrical 2:1 complex. Additionally, C-terminal truncations of GASP-1 result in less potent myostatin inhibitors that form a 2:1 complex, suggesting that the C-terminal domains of GASP-1 are the primary mediators for asymmetric complex formation. Overall, this study provides a new perspective on TGF-β antagonism, where closely related antagonists can utilize different ligand-binding strategies.

lished that myostatin is a strong negative regulator of muscle. Myostatin-null mice have three times more functional muscle mass than WT mice (1,2). Further, the same phenotype can be recapitulated using naturally occurring myostatin antagonists and myostatin dominant negatives (3). These findings have prompted the development of anti-myostatin therapeutics to treat disorders in which loss of muscle mass is either a complication or a comorbidity.
Most TGF-␤ ligands, including myostatin, are covalently linked dimers with a propeller-like morphology. Their signaling activity is regulated by a number of extracellular binding proteins or antagonists. Antagonists can range from small single domain proteins to large multidomain proteins but appear to utilize similar paradigms for ligand inhibition where they block ligand-receptor interactions (4 -9). However, significant differences occur in both the structures of the antagonist and the specific mechanisms or strategies used to block ligand-receptor interactions. For example, the antagonist noggin is a covalent dimer that binds ligands through a single disulfide-rich domain and an extended N terminus (8). In contrast, the antagonist follistatin (FS) utilizes multiple domains and two FS molecules to encircle both myostatin and activin ligands (4,6). Despite these structural differences, both noggin and FS bind dimeric ligands in a symmetrical fashion. This ensures full inhibition of the TGF-␤ ligand, because all receptor-binding sites are occupied.
Analysis of GASP-2 identified the FS domain as the predominant domain responsible for myostatin binding, whereas the C-terminal domains alone (K2 and netrin-like domain) showed limited binding (17). However, the full-length protein was necessary for the nM affinity, suggesting that multiple domains contribute to ligand binding (17). It should be noted that unlike FS, which inhibits multiple ligands (e.g. activin A, activin B, BMP-7, and myostatin), both GASP-1 and GASP-2 selectively inhibit myostatin and GDF-11 (17, 24 -27). Therefore, apart from the myostatin pro-domain, GASP-1 and GASP-2 are the only known molecules to be highly specific for myostatin. Despite this, little is known about how GASP molecules interact with myostatin at the molecular level and, further, how they compare with the other ligand antagonists such as FS and noggin.
To address this, we present the first low resolution solution structure of GASP-1 and GASP-2 in complex with myostatin. Our evidence shows that although GASP-1 and GASP-2 are structurally similar, they use two different binding modes to antagonize myostatin. Through biophysical characterization, we show that GASP-1 preferentially binds myostatin with a 1:1 stoichiometry, whereas GASP-2 preferentially binds myostatin with a 2:1 stoichiometry. Finally, we show that the progressive truncation of domains from the C terminus of GASP-1 result in less potent molecules with a shift to a 2:1 stoichiometric complex with myostatin, similar to GASP-2.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-CHO cells stably overexpressing myostatin, GASP-1 and GASP-2 were used as previously published (4,19). Myostatin conditioned medium (CM) was concentrated ϳ10-fold using tangential flow and concomitantly buffer exchanged into 50 mM Tris pH 7.4, 500 mM NaCl and applied to a Lentil Lectin-Sepharose 4B (Amersham Biosciences) column. Myostatin was eluted with the same buffer with the addition of 500 mM methyl mannose. Eluted protein was then dialyzed against 20 mM trisodium citrate pH 5.0, 20 mM NaCl. Myostatin was then applied to a HiPrep SP FF 16/10 column (GE Life Sciences) and eluted using the same buffer with the addition of 1 M NaCl. The eluted protein was then dialyzed against 20 mM trisodium citrate pH 5.0, 20 mM NaCl. Next, the protein was adjusted to 5% acetonitrile, 0.1% trifluoroacetic acid, 4 M guanidinium HCl and applied to a Sepax C4 reverse phase HPLC column. Myostatin was eluted using an acetonitrile gradient.
GASP-1 and GASP-2-GASP-1 and GASP-2 was expressed and purified as published earlier with some minor modifications (19). Following application to butyl-Sepharose and heparin columns, the heparin eluent containing either GASP-1 or GASP-2 was dialyzed extensively into 50 mM Tris pH 7.4, 20 mM NaCl, 1 mM EDTA, applied to a MonoQ 10/100 GL column and eluted with a linear NaCl gradient.
GASP-1 C-terminal Truncation Mutants-The full-length mouse GASP-1 cDNA fragment was inserted into pFastBac1 followed by subsequent insertion of a stop codon at the desired location for C-terminal truncation. The truncations consisted of the following amino acids (a.a.): WF (30 -198), WFI (30 -314), WFIK (30 -375). The maltose binding protein (MBP)-WFIK fusion construct consisted of MBP positioned on the N terminus linked to WFIK (30 -375) by three alanines. Baculovirus production and protein expression was performed according to the manufacture's protocol (Invitrogen). Following expression in SF9 insect cells, conditioned medium was adjusted to 750 mM ammonium sulfate and applied to a butyl-Sepharose column. The eluent was subsequently applied to a Nickel-Sepharose HiTrap column (GE) followed by extensive dialysis into 50 mM Tris pH 7.4, 20 mM NaCl, 1 mM EDTA, applied to a MonoQ 10/100 GL column and eluted with a linear NaCl gradient. MBP-WFIK was purified using the same strategy except that the buffers utilized contained 5 mM maltose as an additive.
Luciferase Reporter Assays-The luciferase reporter assays were performed as previously described (28,29). Briefly, HEK293 CAGA 12 cells were plated in a 96-well plate and grown for ϳ24 h. Subsequently, the growth medium was removed and myostatin at a concentration of 0.62 nM was mixed with the antagonist (GASP-1, GASP-2, or GASP-1 truncations) in serum free medium and applied to the cells for ϳ18 h. The cells were lysed and luminescence was recorded immediately using a Synergy H1 Hybrid plate reader (BioTek). The inhibition data were imported into GraphPad Prism and fit using nonlinear regression with a variable slope. Inhibition curves shown are one representative experiment of three independent conducted assays.
Surface Plasmon Resonance (SPR)-SPR analysis was performed similar to previous studies (5). Briefly, Surface plasmon resonance (SPR) measurements were carried out in HBS-EP buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% P-20 surfactant (BIAcore AB)) on a BIAcore 3000 optical biosensor system operated with BIAevaluation 4.1 software. Myostatin was immobilized on a CM4 research grade sensor chip (BIAcore AB) by amine coupling chemistry using the manufacturer's protocol at 25°C (2785 RUs). For kinetic measurements, GASP-1 and GASP-2 were diluted in HBS-EP buffer to a concentration of 2 M. The proteins were then diluted in a 2-fold dilution series and applied to the chip at a flow rate of 20 l/min. Protein was injected for an association time of 6 min, and then dissociation was monitored for 9 min. After each measurement, the chip surface was regenerated with four 15 l pulses of 2 M guanidine HCl at a flow rate of 100 l/min. SPR sensorgrams were globally analyzed using a distribution model for continuous affinity and rate constant analysis using the program EVILFIT (30).
Size Exclusion Analysis-GASP-1, GASP-2, GASP-1⅐myostatin, and GASP-2⅐myostatin were applied independently to a Phenomenex HPLC S2000 size exclusion column equilibrated with 20 mM HEPES pH 7.4, 500 mM NaCl, 1 mM EDTA to obtain their retention volume. Sizing standards were run under identical conditions for determination of apparent molecular weight. Fractions from each peak were analyzed using SDS-PAGE followed by Western analysis to ensure both proteins were present.
Analytical Ultracentrifugation (AUC) Sedimentation Velocity-Experiments were performed with a Beckman ProteomeLab XL-1 fitted with absorbance optics and a four-hole rotor. Samples and matched buffer (20 mM HEPES pH 7.4, 500 mM NaCl, 1 mM EDTA; 5 mM maltose was added to the buffer for the GASP-1 MBP-WFIK:myostatin experiment) for buffer subtraction were loaded in a two-channel, carbon-filled, epon centerpieces at 48,000 rpm at 20°C. Absorbance was monitored at 230 nm. Absorbance data were processed using the program Sedfit (31) to determine the sedimentation coefficient (c(s)) and sedimentation coefficient-frictional ratio c(s,f/f o ) distributions (32). Values for buffer density and viscosity and protein partial specific volume were calculated using Sednterp. Because of the presence of both N-linked and O-linked glycosylation of GASP-1 and GASP-2, the partial specific volume was calculated as a weight-averaged value.
Small Angle X-ray Scattering (SAXS)-SAXS data were collected at beam line 12-ID-B at the Advanced Photon Source at Argonne National Laboratory and using the SIBYLS beam line mail-in program (Berkeley, CA) (33). Purified GASP-1, GASP-2, GASP-1⅐myostatin, GASP-2⅐myostatin and GASP-1 MBP-WFIK:myostatin were purified as described above. Three different concentrations of each sample were analyzed to determine whether concentration dependent effects exist. Data were collected in 20 mM HEPES pH 7.4, 500 mM NaCl, 1 mM EDTA at 10°C. Additionally, for the GASP-1 MBP-WFIK construct, the buffer contained 1 mM maltose. Four exposure times of 0.5 s, 1 s, 2 s, and 5 s were collected. Exposures exhibiting radiation damage were discarded. Buffer matched controls were used for buffer subtraction. ScÅtter (SIBYLS) and the ATSAS program suite (EMBL) were used for data analysis. The online DAMMIF server (EMBL-Hamburg) was used for generation of 20 independent ab initio molecular envelope reconstructions (34). P1 symmetry operators were employed to improve the quality of the fit for GASP-1, GASP-2, GASP-1⅐myostatin whereas a P2 symmetry operator was used for GASP-2⅐myostatin. Final models were then averaged using DAMAVER pipeline (35). Graphics were generated using UCSF Chimera and Pymol.

Characterization of Purified GASP-1 and GASP-2-For
structural analysis, we produced and purified murine GASP-1 and GASP-2 as previously described (Fig. 1B) (19). Analysis by SDS-PAGE shows that GASP proteins migrate at higher than expected molecular weight. Subsequent analysis by mass spectrometry reveals that GASP-1 and GASP-2 exhibit mass increases of ϳ8.7 and 5.3 kDa, respectively, over their predicted molecular mass ( Table 1). The majority of this mass is attributed to both O-and N-linked glycosylation and is consistent with previous studies describing that GASP-1 is both O-linked and N-linked glycosylated ( Fig. 1B and Table 1) (36). Because of a significant decrease in solubility upon deglycosylation, all structural analysis was performed on the fully glycosylated form of the GASP proteins.
Purified GASP proteins were biologically active and maintained specificity for myostatin. Using a HEK293-CAGA 12 luciferase reporter assay, we determined the myostatin IC 50 values for GASP-1 and GASP-2 (Fig. 1C). Similar to previous studies, both GASP-1 and GASP-2 potently blocked myostatin signaling with IC 50 values of 4.4 Ϯ 2.6 ϫ 10 Ϫ10 and 1.2 Ϯ 0.5 ϫ 10 Ϫ8 M, respectively (25). The nearly 100-fold difference in inhibitory activity is consistent with previous studies (17,25). SPR analysis confirms that our purified GASP-1 and GASP-2 bind myostatin with an apparent low pM and nM K d values, respectively, and both display a nearly irreversible off rate (Fig.  1D). GASP-1 and GASP-2 did not inhibit activin A or activin B signaling (data not shown), consistent with previous reports that GASP proteins are highly selective for myostatin (17,24,25).
Size Exclusion Chromatography (SEC) Reveals Differences in GASP-1⅐Myostatin and GASP-2⅐Myostatin Complexes-We next analyzed the GASP proteins alone and in the presence of myostatin by SEC. GASP-1 and GASP-2 had similar retention times with symmetrical peaks but eluted at smaller than expected molecular masses, suggesting that there may be nonspecific interactions with the SEC resin ( Fig. 2A).
To form the GASP⅐myostatin complex, GASP proteins were mixed at a 2.25:1 (GASP⅐myostatin) molar ratio and analyzed by SEC. Excess GASP was used to ensure full saturation of the myostatin dimer. In both cases, a new peak formed that contained both myostatin and GASP, indicative of complex. In addition, the traces also contained a peak from unbound GASP, indicating that myostatin in the complex peak is saturated with GASP. Interestingly, the peak from the GASP-1⅐myostatin complex shifted to the left by 1.1 ml and overlapped with the free GASP-1 peak, whereas there was a distinct separation of the GASP-2⅐myostatin peak from the unbound GASP-2. As such, regression analysis revealed significant differences in the apparent molecular mass of the complexes between GASP-1 and GASP-2 (Fig. 2B). However, because of the inherent limitations of SEC, analysis of these differences required further, more rigorous investigation. Nonetheless, the qualitative difference in retention volume between GASP-1⅐myostatin and GASP-2⅐myostatin suggested differences in the two complexes.
Analytical Ultracentrifugation (AUC) of Purified Components-To more accurately define the molecular mass of the complexes, we turned to analytical ultracentrifugation ( Fig. 3 and Table 2). Sedimentation velocity showed that GASP-1 and GASP-2 contained single species with sedimentation coefficients of 3.6 and 3.36 S, respectively. Each exhibited a similar frictional ratio of ϳ1.47, suggesting that they are more elongated than globular in shape ( Fig. 3A and Table 2). Importantly, the apparent molecular mass for GASP-1 and GASP-2 were more consistent with the mass spectrometry results than the SEC-derived molecular masses (Tables 1 and 2).
Samples containing the GASP⅐myostatin complexes were pooled from SEC (high molecular weight peak) and analyzed by sedimentation velocity. For GASP-1⅐myostatin the c(s) profile exhibited two peaks with sedimentation coefficients of 3.73 and 5.24 S, indicative of excess GASP-1 and the GASP-1⅐myostatin complex, respectively ( Fig. 3B and Table 2). Derivation of molecular mass from sedimentation velocity data can be less accurate when multiple species are present. Therefore, we utilized a recently developed method known as c(s, f/f 0 ) analysis to more accurately determine the molecular mass for the GASP-1⅐myostatin complex (32). The c(s, f/f 0 ) analysis separately fits both the frictional ratio and sedimentation coefficient for each species to provide a more accurate estimation of molecular mass (32). Applying this analysis, the apparent molecular mass for the GASP-1⅐myostatin complex is 97.1 kDa (Table 2). This indicates that GASP-1 preferentially binds myostatin in a 1:1  MARCH 20, 2015 • VOLUME 290 • NUMBER 12 complex. We suspect the excess GASP-1 detected in our AUC experiment is linked to our inability to fully resolve the GASP-1⅐myostatin complex from free GASP-1 using SEC.

Binding Modes for GASP Antagonism of Myostatin
In contrast, analysis of the GASP-2⅐myostatin complex by sedimentation velocity resulted in a more homogenous sample with a major species at 5.81 S (Fig. 3B and Table 2). A minor peak at 3.97 S was also detected, suggestive of unbound GASP-2; however, this only accounted for ϳ8% of the total signal (Fig. 3B). Again, using c(s, f/f 0 ) analysis, we determined the apparent molecular mass for the GASP-2⅐myostatin complex to be 137 kDa (Table 2), consistent with a 2:1 complex stoichiometry. These data suggest that GASP-1 and GASP-2 preferentially form different stochiometric complexes with myostatin. In fact, closer inspection of the SDS-PAGE gels by densitometry analysis also suggests a difference in stoichiometry where the intensity of GASP-2 (3.1) is almost twice that of GASP-1 (1.8) when normalized to myostatin ( Fig. 2A).

SAXS Analysis of GASP-1 and GASP-2-
The SAXS scattering profiles for GASP-1 and GASP-2 are shown in Fig. 4A, along with the corresponding analysis in Table 3. Samples were well behaved in solution and did not show evidence of interparticle repulsion or aggregation over multiple protein concentrations (Fig. 4, A and B). The data are of high quality because (a) the I(0) increases proportionally to the concentrations tested, (b) the Gunier range has a linear appearance at low scattering angles, and (c) the radius of gyration (R g ) does not significantly change in response to changes in protein concentration ( Fig. 4 and Table 3).  From the Gunier analysis, we determined that GASP-1 has a slightly higher R g than GASP-2, 52 Ϯ 1.1 versus 45.6 Ϯ 0.75 Å, respectively ( Fig. 4B and Table 3). This suggests that GASP-1 is more elongated in solution than GASP-2. In support, this observation is readily apparent in the pairwise distribution plot (P(r); Fig. 4C). Further transformation of the SAXS data indicates that both GASP-1 and GASP-2 are inherently flexible as shown by a plateau in the q 3 ⅐I(q) versus q 3 (Å Ϫ1 ) 3 plot compared with a plateau observed in the q 4 ⅐I(q) versus q 4 (Å Ϫ1 ) 4 , which would indicated a globular, rigid structure (Fig. 4, D and E). The apparent molecular masses for GASP-1 and GASP-2 derived from SAXS are 76.4 Ϯ 9.2 and 70 Ϯ 8.4 kDa and are in agreement with the expected molecular mass (Tables 1 and 3).
The SAXS data were then used to perform ab initio modeling to generate molecular envelopes as shown in Fig. 5. Both GASP-1 and GASP-2 have acceptable normalized spatial discrepancy values (0.94 Ϯ 0.06 and 0.83 Ϯ 0.047, respectively; Table 3). In agreement with the P(r) analysis, the generated GASP-1 molecular envelope has an elongated and rather featureless shape (Fig. 5A). Along these same lines, the ab initio results for the GASP-2 molecular envelope revealed a slightly less elongated shape that adopted a more compact structure (Fig. 5B), suggesting that there may be interdomain contacts within GASP-2, resulting in a more globular-like appearance.
SAXS Analysis of the GASP-1⅐Myostatin and GASP-2⅐Myostatin Complexes Supports Differences in Binding Stoichiometry-Because previous methods to purify the GASP-1⅐ myostatin complex resulted in contamination of unbound antagonist, we attenuated the ratio of GASP-1 to reflect the 1:1 stoichiometry identified by sedimentation velocity. Therefore, GASP-1 was mixed with myostatin in a 1.1:1 molar ratio, which resulted in the loss of the excess GASP-1 AUC peak (data not shown). No change to the GASP-2⅐myostatin ratio was required, and the complex was purified using a mixing ratio of 2.25:1.
Scattering profiles for GASP-1⅐myostatin and GASP-2⅐myostatin are shown in Fig. 4F. The data are of high quality and free of aggregation (Fig. 4, F and G, and Table 3). Gunier analysis and inspection of the P(r) plot reveals that the GASP-1⅐myostatin complex is more compact than GASP-1 alone as indicated by a significant decrease in both the R g and the maximum particle distance (D max ; Fig. 4, F and G, and Table 3). This is in stark contrast to the GASP-2 complex, which has a significantly larger R g and D max than GASP-2 alone (Table 3).
Similar to the flexibility analysis applied to the myostatinfree GASP proteins, both complexes display a plateau in the q 3 ⅐I(q) versus q 3 (Å Ϫ1 ) 3 plot compared with the q 4 ⅐I(q) versus q 4 (Å Ϫ1 ) 4 plot, indicating that both complexes are flexible in solution (Fig. 4, I and J). However, the plateau appeared more prominent for the complexes than the GASP proteins alone, suggesting that some rigidity is gained when complex is formed (37).
Analysis of the scattering profiles provides several lines of evidence to support that GASP-1 and GASP-2 form different stoichiometric complexes with myostatin (Table 3). For example, the GASP-2⅐myostatin complex had ϳ25% more overall scattering intensity compared with GASP-1⅐myostatin, denoting a significant difference in overall particle mass. In addition, the R g of the GASP-1⅐myostatin complex is significantly smaller (48.5 Ϯ 0.85 Å) as compared with the GASP-2⅐myostatin (60.4 Ϯ 2.25 Å) complex (Table 3). Furthermore, the P(r) plot shows that the GASP-1⅐myostatin complex is globular with a smaller D max , whereas the GASP-2⅐myostatin complex is much more extended (Fig. 4H and Table 3). Finally, the SAXS derived apparent molecular mass for the GASP-1⅐myostatin complex was determined to be 103 Ϯ 8.0 kDa compared with the much larger GASP-2⅐myostatin complex (187 Ϯ 25 kDa; Table 3).
A summary of the theoretical molecular masses and experimentally determined molecular masses for each technique is displayed in Table 1. Taken together, it is evident that GASP-1 preferentially binds myostatin in a 1:1 configuration, whereas GASP-2 preferentially binds myostatin in a 2:1 configuration.
The SAXS data were subsequently used to perform ab initio modeling to derive low resolution molecular envelopes for both the GASP-1⅐myostatin and GASP-2⅐myostatin complexes (Fig.  5). The envelopes for each complex had acceptable normalized spatial discrepancy values of 1.034 Ϯ 0.057 and 1.488 Ϯ 0.478, respectively ( Table 3). The GASP-1⅐myostatin molecular envelope resembles the featureless characteristics of myostatin-free GASP-1 but is more compact with a slightly more "full" shape (Fig. 5A). The additional volume contribution is most likely from the bound myostatin. In striking contrast, the GASP-2⅐myostatin complex adopts a symmetrical conformation and has a more featured appearance (Fig. 5B), whereas the molecular envelope has a symmetrical resemblance to previously determined antagonist⅐ligand structures (e.g. two molecules of FS wrapping around the ligand myostatin). However, it appears that the GASP-2 domains are splayed away from a centrally positioned ligand as opposed to the compact FS⅐ligand structures (4 -7).
Truncation of GASP-1 C-terminal Domains Changes Binding Stoichiometry-Previous truncation analysis of GASP-2 revealed that the whey acidic protein and FS domains of GASP-2 are the primary mediators for myostatin binding, suggesting that the C-terminal domains are expendable (17). Given the differences in stoichiometry, we next wanted to determine whether GASP-1 exhibits a similar behavior. Therefore, we generated a number of C-terminal truncations of GASP-1 and tested their ability to inhibit myostatin in the luciferase reporter assay (Fig.  6A). For simplicity, we will use the following abbreviations to describe the domain composition of the GASP-1 C-terminal truncation constructs: W indicates whey acidic protein domain, F indicates follistatin domain, I indicates immuno-globulin-like domain, and K indicates kunitz domain 1, unless otherwise stated. Following progressive removal of the C-terminal domains, we observed a corresponding decrease in the IC 50 of GASP-1 for myostatin (Fig. 6A). Interestingly, the C-terminal truncation constructs WFIK (IC 50 ϭ 2.6 Ϯ 1.0 ϫ 10 Ϫ9 M) and WFI (IC 50 ϭ 5.1 Ϯ 1.7 ϫ 10 Ϫ9 M) displayed an inhibitory curve strikingly similar to full-length GASP-2, whereas the WF construct showed a significantly weaker IC 50 (6.5 Ϯ 2.3 ϫ 10 Ϫ8 M; Fig.  6A). Furthermore, the GASP-1 truncations did not inhibit activin A or activin B (data not shown).
We next wanted to determine whether the truncated versions of GASP-1 maintained the 1:1 stoichiometry. WFI or WFIK was mixed with myostatin at 2.5:1 (antagonist⅐ligand) ratio to ensure full saturation of myostatin and analyzed by sedimentation velocity. As expected, we observed two peaks in the c(s) profile. The larger peak is consistent with complex formation, whereas the smaller peak is consistent with sedimentation analysis of WFI or WFIK alone (Fig. 6B). The sedimentation coefficients for the WFI⅐myostatin and WFIK⅐myostatin complexes were determined to be 4.2 and 5.1 S, respectively (Fig. 6B). These results are similar to the large shift in the sedimentation coefficient of the GASP-2⅐myostatin complex versus unbound GASP-2. Analysis of the c(s,f/f 0 ) distribution revealed that both complexes share an apparent molecular mass of 104 kDa, consistent with a 2:1 binding stoichiometry (Fig. 6, B and E). We would expect that the WFI⅐myostatin complex to be of lower molecular mass than the WFIK⅐ myostatin complex; however, the exact molecular mass for the WFI⅐myostatin is difficult to accurately determine because of the breadth of the peak in the c(s) profile (Fig. 6B).
We further analyzed a fusion protein of MBP-WFIK in complex with myostatin by AUC and SAXS ( Fig. 6 and Table 3). The   A, ab intio model generated from SAXS data of GASP-1 (dark blue, left) and GASP-1⅐myostatin (light blue, right). A P1 symmetry operator was used for both GASP-1 and GASP-1⅐myostatin. B, ab intio model generated from SAXS data of GASP-2 (red, left) and GASP-2⅐myostatin (pink, right). A P1 symmetry operator was used for GASP-2 and P2 symmetry operator for GASP-2⅐myostatin. All envelopes were generated using DAMMIF and averaged using DAMAVER. The averaged envelope is shown. See also Table 3. MARCH 20, 2015 • VOLUME 290 • NUMBER 12 fusion partner was used to enhance production and solution behavior necessary for SAXS. Importantly, the MBP-WFIK fusion construct maintained a similar myostatin inhibitory activity as WFIK (IC 50 ϭ 7.2 Ϯ 2.8 ϫ 10 Ϫ9 M; Fig. 6A). Similar to GASP-2⅐myostatin complex preparation, we mixed MBP-WFIK at a 2.5:1 (antagonist⅐ligand) ratio with myostatin and purified the complex using SEC. Fractions from the peak were pooled, concentrated, and subjected to AUC and SAXS. Consistent with the other GASP-1 C-terminal truncations, both AUC (Fig. 6B) and SAXS (Fig. 6, C and D, and Table 3) analysis revealed a 2:1 MBP-WFIK⅐myostatin complex ( Fig. 6E and Table 3). Taken together, we have shown that progressive removal of the C-terminal domains of GASP-1 results in decreased inhibitory potential, similar to that of GASP-2, a significant increase in the sedimentation coefficient upon complex formation with myostatin, and an apparent molecular mass for each complex that is consistent with formation of a 2:1 antagonist⅐ligand complex.

DISCUSSION
At the onset of our study, we operated under the assumption that GASP would bind myostatin in a similar fashion to previous known antagonists (FS, FSTL3, noggin, etc.), where two antagonist molecules would interact with one ligand dimer (4 -8). However, upon complex isolation using SEC and AUC analysis, we were surprised by the dramatic difference in apparent molecular mass of the two GASP⅐myostatin complexes. This prompted us to structurally characterize these differences using SAXS and provided the first low resolution solution structures of GASP-1 and GASP-2 in their myostatin-free and myostatin-bound states. To extend our findings, we progressively truncated the C-terminal domains from GASP-1, which resulted in less potent GASP-1 molecules that preferentially bind myostatin in a 2:1 fashion similar to full-length GASP-2. Taken together, these data support that, although GASP-2 binds in a similar manner to other known antagonist by using 2 molecules to sequester the ligand, GASP-1 preferentially binds using only one molecule.
Over the years, efforts have started to unravel the different mechanisms of TGF-␤ antagonism. Interestingly, extracellular antagonists have evolved different approaches to regulate ligand signaling. Not surprisingly, because of the dimeric architecture of the ligand, antagonists prefer a 2:1 binding mode (e.g. FS-type⅐ligand, noggin⅐BMP-7, CV2⅐BMP-2) (4 -9). Although this paradigm is consistent with GASP-2, we have uncovered that GASP-1 diverges from the more standard 2:1 complex and forms an asymmetric complex with the ligand.
GASP-1 is clearly a more potent myostatin inhibitor than GASP-2. However, when the C-terminal domains of GASP-1 were removed, myostatin inhibition was reduced. Accordingly, the truncated versions of GASP-1 exhibited a 2:1 binding stoi- chiometry. This suggests that the very potent anti-myostatin activity of GASP-1 could be coupled with its ability to form a 1:1 complex (e.g. through a cooperative binding mechanism or interaction using multiple binding surfaces). Certainly future experiments will be needed to determine the mechanistic differences in GASP-1 versus GASP-2 myostatin complex formation.
It should be noted that similar to GASP-1, the BMP antagonist chordin also binds to its ligand in a asymmetrical 1:1 fashion (38,39). However, unlike GASP, Chordin is composed of four highly conserved von Willebrand C (vWC) domains that are shown to occupy receptor sites in a symmetrical fashion (i.e. vWC1 binds to one side, whereas vWC3 binds to the other side of the ligand). Although we have shown that GASP-1 interacts with myostatin in a 1:1 fashion, the lack of repetitive binding domains within GASP-1 emphasizes a completely asymmetric mechanism where unrelated domains within GASP-1 are likely to bind each side of the symmetrical dimer. Overall, the results of this study extend our understanding of the different mechanisms utilized for TGF-␤ inhibition. Furthermore, to our knowledge, this is the first example of where closely related extracellular TGF-␤ antagonists interact with ligands utilizing different binding modes. With myostatin inhibitors in high demand, knowledge that alternative binding strategies exist yet are still able to achieve high ligand specificity may offer additional therapeutic development options.