Crystal structure of the WFIKKN2 follistatin domain reveals insight into how it inhibits growth differentiation factor 8 (GDF8) and GDF11

Growth differentiation factor 8 (GDF8; also known as myostatin) and GDF11 are closely related members of the transforming growth factor β (TGF-β) family. GDF8 strongly and negatively regulates skeletal muscle growth, and GDF11 has been implicated in various age-related pathologies such as cardiac hypertrophy. GDF8 and GDF11 signaling activities are controlled by the extracellular protein antagonists follistatin; follistatin-like 3 (FSTL3); and WAP, follistatin/kazal, immunoglobulin, Kunitz, and netrin domain–containing (WFIKKN). All of these proteins contain a follistatin domain (FSD) important for ligand binding and antagonism. Here, we investigated the structure and function of the FSD from murine WFIKKN2 and compared it with the FSDs of follistatin and FSTL3. Using native gel shift and surface plasmon resonance analyses, we determined that the WFIKKN2 FSD can interact with both GDF8 and GDF11 and block their interactions with the type II receptor activin A receptor type 2B (ActRIIB). Further, we solved the crystal structure of the WFIKKN2 FSD to 1.39 Å resolution and identified surface-exposed residues that, when substituted with alanine, reduce antagonism of GDF8 in full-length WFIKKN2. Comparison of the WFIKKN2 FSD with those of follistatin and FSTL3 revealed differences in both the FSD structure and position of residues within the domain that are important for ligand antagonism. Taken together, our results indicate that both WFIKKN and follistatin utilize their FSDs to block the type II receptor but do so via different binding interactions.

Growth differentiation factor 8 (GDF8; also known as myostatin) and GDF11 are closely related members of the transforming growth factor ␤ (TGF-␤) family. GDF8 strongly and negatively regulates skeletal muscle growth, and GDF11 has been implicated in various age-related pathologies such as cardiac hypertrophy. GDF8 and GDF11 signaling activities are controlled by the extracellular protein antagonists follistatin; follistatin-like 3 (FSTL3); and WAP, follistatin/kazal, immunoglobulin, Kunitz, and netrin domain-containing (WFIKKN). All of these proteins contain a follistatin domain (FSD) important for ligand binding and antagonism. Here, we investigated the structure and function of the FSD from murine WFIKKN2 and compared it with the FSDs of follistatin and FSTL3. Using native gel shift and surface plasmon resonance analyses, we determined that the WFIKKN2 FSD can interact with both GDF8 and GDF11 and block their interactions with the type II receptor activin A receptor type 2B (ActRIIB). Further, we solved the crystal structure of the WFIKKN2 FSD to 1.39 Å resolution and identified surfaceexposed residues that, when substituted with alanine, reduce antagonism of GDF8 in full-length WFIKKN2. Comparison of the WFIKKN2 FSD with those of follistatin and FSTL3 revealed differences in both the FSD structure and position of residues within the domain that are important for ligand antagonism. Taken together, our results indicate that both WFIKKN and follistatin utilize their FSDs to block the type II receptor but do so via different binding interactions.
Growth differentiation factor 8 (GDF8), 4 also known as myostatin, is a member of the TGF-␤ superfamily of ligands (1). GDF8 is a potent negative regulator of skeletal muscle growth where genetic deletion results in a hypermuscular phenotype (2,3). In the adult animal, several studies have established that down-regulation, or inhibition, of GDF8 induces muscle hypertrophy (2)(3)(4)(5)(6)(7). In contrast, transgenic overexpression of GDF8 causes muscle wasting, consistent with the idea that certain tumors increase GDF8 levels and contribute to cancer cachexia (8 -11). GDF11, which shares 90% identity at the amino acid level, has distinct functions and is important for proper anterior-posterior patterning (12,13). Recent evidence supports a beneficial role of GDF11 in neurogenesis; however, excessive treatment of mice with GDF11 results in muscle wasting, similar to GDF8 (14 -16). Given that GDF8 and GDF11 have prominent roles in tissue homeostasis, there has been a strong interest in developing therapeutics that modulate their signaling, especially ones that aim to boost muscle by targeting GDF8 in muscle-related pathologies (17)(18)(19)(20)(21)(22).
To signal, TGF-␤ ligands bind to, and coordinate the assembly of, two type II and two type I serine-threonine kinase receptors (1). GDF8 and GDF11 signal by using a select combination of receptors, ActRIIA or ActRIIB (type II) coupled with Alk4 or Alk5 (type I) (23)(24)(25). Ligand activity is selectively controlled by extracellular protein antagonists that bind to ligands and interfere with receptor binding. Although TGF-␤ ligands are structurally similar, antagonists have variable sizes and domain architectures (23, 26 -30). This structural diversity allows different antagonists to selectively inhibit subsets of TGF-␤ family members. Over the years, structural studies have provided insight into how different antagonists can adopt different mechanisms to bind and neutralize ligands. However, it appears that in most cases, antagonists block both the type I and type II receptorbinding interfaces (23, 29 -32).
WFIKKN is named from the conserved multidomain architecture: whey acidic protein (WAP), follistatin domain (FSD), Immunoglobulin domain (Ig), two tandem Kunitz (K1, K2) domains, and a netrin (N) domain (Fig. 1a). Most animals have two related versions of WFIKKN, WFIKKN1 and WFIKKN2, which share 56% identity (37,38). Differences in potency and binding stoichiometry have previously been reported, where WFIKKN2 has an IC 50 of 0.26 nM for GDF8 antagonism and is nearly 100-fold more potent than WFIKKN1 (38). In addition, WFIKKN2 forms a 1:1 complex with GDF8/11 (1 WFIKKN2 and 1 ligand dimer), whereas WFIKKN1 forms a 2:1 complex. Interestingly, removing the Kunitz 2 and netrin domains both reduces the potency of WFIKKN2 (IC 50 ϭ 7.2 nM) and shifts the binding stoichiometry to that of full-length WFIKKN1 (38). Whereas multiple domains are thought to interact with GDF8 and GDF11, previous studies have determined that the FSD plays a significant role in antagonism (34,38).
Follistatin domains are also functionally important for the antagonists follistatin and FSTL3. FSTL3 and follistatin contain two and three FSDs, respectively, which have been shown through multiple X-ray crystal structures to directly contact the ligand (23,32,39). Interestingly, each of the FSDs (FSD1-3) within follistatin/FSTL3 adopt different molecular conformations and functions differently at the ligand interface, indicating that the FSDs are not functionally redundant. In fact, biochemical experiments showed that changing the order of the FSDs can severely alter ligand binding and specificity (27,40).
Whereas WFIKKN selectively inhibits GDF8 and GDF11, the molecular basis for ligand selectivity has not been established. Given that WFIKKN and follistatin/FSTL3 each have FSDs important for binding, whether they exhibit unique or common ligand mechanisms has not been determined. To investigate this, we characterized the WFIKKN FSD and contrasted the binding features of the follistatin FSDs with the WFIKKN FSD. We found that the FSD of WFIKKN blocks the type II receptor-binding interface similar to the FSD2 of follistatin and FSTL3. However, WFIKKN2 FSD displays different structural features and residues that interact with the ligand that map to a different location of the FSD, indicating a different binding mode when compared with the FSDs of follistatin and FSTL3.

Production of WFIKKN2 follistatin domain
Recombinant murine WFIKKN2 FSD, containing residues 104 -172, was produced in bacteria. WFIKKN2 FSD protein formed inclusion bodies, which were solubilized and subjected to oxidative refolding to induce disulfide bond formation. Properly folded material was purified using reverse-phase chromatography, as depicted in Fig. 1b. First, native PAGE was used to analyze the refolded WFIKKN2 FSD and to evaluate binding interactions with GDF8. FSD alone migrated as a single band, indicating the isolation of a single refolded species. Similar to previously published results, GDF8 alone does not enter the native PAGE and is not visible (25). Upon mixing WFIKKN2 FSD with GDF8, an additional band appeared, indicating complex formation (Fig. 1c). Titration of GDF8 with increasing concentrations of WFIKKN2 FSD showed an increase in the complex band intensity. Analysis of the newly formed band by SDS-PAGE shows that both WFIKKN2 FSD and GDF8 are present, supporting complex formation (Fig. 1c). Similar results were observed with GDF11 where the complex band was significantly sharper (Fig. 1d). These results indicate that WFIKKN2 FSD is properly folded and can bind both GDF8 and GDF11, as previously implicated (34).

Comparison of WFIKKN2 FSD and WFIKKN2 full-length protein
We next wanted to compare the WFIKKN2 FSD with fulllength (FL) WFIKKN2 for their ability to bind and antagonize GDF8 and GDF11. Surface plasmon resonance (SPR) was used to measure binding affinity, where GDF8 and GDF11 were coupled to a CM5 SPR chip. Binding analysis was performed by injecting increasing concentrations of WFIKKN2 FL or WFIKKN2 FSD (Fig. 2, a-d). Similar to previous reports, WFIKKN2 FL bound GDF8 and GDF11 with a high affinity, as demonstrated by the slow dissociation rate (34,35). Consistent with the native PAGE analysis, WFIKKN2 FSD exhibited a significant interaction with GDF8 and GDF11, albeit with a much faster dissociation rate as compared with WFIKKN2 FL. The association and dissociation rate constants and equilibrium constant, K D , are reported in Table 1. Overall, WFIKKN2 FL exhibited a high-affinity interaction with GDF8 (K D ϭ 0.74 nM) and GDF11 (0.24 nM), whereas the WFIKKN2 FSD was ϳ1000-fold weaker (K D ϭ 0.66 M for GDF8 and 0.12 M for GDF11). WFIKKN2 FL and WFIKKN2 FSD were able to bind TGF␤1, albeit weakly, but have no affinity for activin A, consistent with previous studies (35). We next sought to determine whether the affinity of WFIKKN2 FSD for GDF8 and GDF11 was sufficient to inhibit signaling. Using a luciferase reporter assay responsive to GDF8 and GDF11, we titrated increasing concentrations of WFIKKN2 FL and WFIKKN2 FSD. Whereas WFIKKN2 FL inhibited GDF8 and GDF11 signaling with an IC 50 value of 0.34 and 0.13 nM, respectively, WFIKKN2 FSD exhibited a significantly reduced inhibition of GDF8 and GDF11 with an IC 50 value of 0.85 and 0.28 M, respectively (Fig. 2d), similar to the 1000-fold difference in binding affinity determined by SPR. This demonstrates that whereas the WFIKKN2 FSD binding affinity is severely reduced compared with full-length WFIKKN2, it is still able to antagonize GDF8 and GDF11, albeit with weaker potency. Binding and antagonism of WFIKKN2 FSD to GDF8. a-d, SPR of WFIKKN2 FL and WFIKKN2 FSD over immobilized GDF8 and GDF11. Red, experimental binding trace; black, data fit using a 1:1 binding model. a, WFIKKN2 FL applied to immobilized GDF8; b, WFIKKN2 FSD applied to immobilized GDF8; c, WFIKKN2 FL applied to immobilized GDF11; d, WFIKKN2 FSD applied to immobilized GDF11; e, WFIKKN2 FL (red) or WFIKKN2 FSD (black) applied to immobilized TGF-␤1. Shown is WFIKKN2 FL (gray) or WFIKKN2 FSD (orange) applied to immobilized activin A. f, luciferase assay using (CAGA) 12 HEK293 cells treated with GDF8/GDF11 alone or titrated with increasing concentrations of WFIKKN2 FL or WFIKKN2 FSD. Data were fit using nonlinear regression to determine an IC 50 and plotted as mean Ϯ S.D. (error bars) conducted at least twice with each point measured in triplicate.

WFIKKN2 follistatin domain structure Does the WFIKKN2 FSD bind GDF8 at the type II receptor epitope?
To signal, GDF8 binds the extracellular domain of the type II receptor, ActRIIB, with high affinity (K D ϳ1 nM) (41,42). Two ActRIIB receptors are expected to bind GDF8, or GDF11, at each knuckle region of the dimeric ligand, similar to the observed binary crystal structures of ActRIIB in complex with activin A and BMP7 (43,44). Previous studies have demonstrated that full-length WFIKKN2 interferes with type II receptor binding (26). To determine whether WFIKKN2 FSD could interfere with type II receptor binding, we performed a competition analysis using native PAGE. GDF11 mixed with WFIKKN2 FSD was titrated with increasing concentrations of the extracellular domain of ActRIIB (ActRIIB-ECD) and analyzed by native PAGE (Fig. 3a). GDF11 was used for analysis because the complex bands bound to WFIKKN2 FSD and ActRIIB-ECD are more distinct than those with GDF8. Our results show that ActRIIB-ECD easily displaces WFIKKN2 FSD, leading to the formation of the ActRIIB-ECD:GDF11 complex (Fig. 3a). A higher-order complex that contained all three components was not visible, indicating that the WFIKKN2 FSD and ActRIIB-ECD were mutually exclusive and likely bind to the same position on the ligand. Bands corresponding to WFIKKN2 FSD:GDF11 and ActRIIB-ECD:GDF11 were excised and verified by SDS-polyacrylamide gel under reducing conditions (Fig. 3b). These results indicate that the WFIKKN2 FSD and ActRIIB compete for the same binding interface of GDF11.
To further test the idea that WFIKKN2 FSD and ActRIIB bind GDF8 and GDF11 at similar locations, competition experiments were performed using SPR. Experiments were conducted using the extracellular domain of ActRIIB fused to the Fc portion of an antibody (ActRIIB-Fc). ActRIIB-Fc was captured onto a protein A chip, followed by an injection of GDF11 to form the binary ActRIIB-Fc:GDF11 complex. We next injected WFIKKN2 FSD at 500 nM and observed no increase in binding, indicating that the WFIKKN2 FSD could not bind GDF11 in the presence of ActRIIB (Fig. 3c). Interestingly, when similar experiments were performed with WFIKKN2 FL, a noticeable mass increase was observed, indicating that WFIKKN2 FL was able to bind GDF11 that was already in complex with ActRIIB (Fig. 3c).

Crystallization and structural determination of WFIKKN2 FSD
Because the WFIKKN2 FSD can bind and antagonize GDF8 and GDF11, we next wanted to determine the molecular structure and draw comparisons with the FSD of follistatin and FSTL3. The X-ray crystal structure of WFIKKN2 FSD was solved by single isomorphous replacement with anomalous scattering to 1.39 Å resolution. Data collection and refinement statistics are presented in Table 2. Fig. 4a depicts the overall structure of the WFIKKN2 FSD. Similar to other FSDs, the structure contains two subdomains: an N-terminal EGF-like portion followed by a Kazal-like protease inhibitor subdomain. The EGF-like portion contains anti-parallel ␤-strands (␤1 and ␤2), whereas the Kazal-like protease inhibitor subdomain contains a central helix (␣1) and another set of anti-parallel ␤-strands (␤3 and ␤4) that caps the Kazal subdomain. Similar to the FSDs from follistatin and FSTL3, the WFIKKN2 FSD contains five conserved disulfide bonds. Two are positioned in the EGF subdomain that connect the ␤1 and ␤2 strands (Fig. 4a). The other three are located in the Kazal subdomain; two link the ␣1 to the segment connecting the EGF and Kazal subdomains, and one links ␤3 to the C terminus. Values shown represent the association constant (k a ), dissociation constant (k d ), and equilibrium dissociation constant (K D ) of WFIKKN2 FL or the FSD injected over primary amine-coupled GDF8 or GDF11.

WFIKKN2
GDF8  Figure 3. Competitive binding between WFIKKN2 FSD and ActRIIB to GDF11. a, native PAGE analysis using preformed WFIKKN2 FSD:GDF11 complex mixed with increasing amounts of ActRIIB (ARIIB). b, reduced SDS gel of WFIKKN2 FSD, ARIIB, and GDF11 and excised bands represented in a by a black box. c, co-injection SPR binding experiment. First, ActRIIB-Fc was captured onto a protein A sensor chip, and baseline was normalized. Next, GDF11 was injected to form the binary ActRIIB:GDF11 complex. Subsequently, WFIKKN2 FL or FSD was injected, and binding to the ActRIIB:GDF11 complex was monitored.

WFIKKN2 follistatin domain structure WFIKKN2 FSD mutagenesis and inhibition
Utilizing the structure, we wanted to identify residues that are important for the interaction with GDF8 and GDF11. TGF-␤ ligands, including BMP2, BMP4, BMP7, activin A, activin B, GDF8, and GDF11, predominantly interact with receptors and extracellular antagonists using hydrophobic interactions (23,45). Therefore, we hypothesized that surfaceexposed hydrophobic residues of WFIKKN2 FSD might be important for ligand binding and antagonism. Several surfaceexposed hydrophobic residues in both the EGF and Kazal subdomains were readily apparent (Fig. 4b) (46). Residues were selected for alanine mutagenesis, including two that are positioned in the EGF domain (Phe-109 and Trp-121) and three positioned within the Kazal subdomain (Phe-139, Phe-153, and Ile-163) (Fig. 5a). To determine whether these residues were important for WFIKKN2 antagonism, we generated single point mutations of the selected residues in the full-length WFIKKN2. WT and mutant versions of WFIKKN2 FL were expressed in human embryonic kidney cells (HEK293F), purified, and tested for GDF8 inhibition in a cell-based luciferase reporter assay. HEK293T-CAGA cells were treated with 0.62 nM GDF8 and titrated with increasing concentrations of either purified WT or mutant WFIKKN2 FL (Fig. 5b). Data were fit to a dose-response curve to determine an IC 50 for WFIKKN2 FL and mutations. Mutations within the EGF domain (F109A and W121A) resulted in little to no change in the antagonism of GDF8. However, WFIKKN2 mutants tested within the Kazal subdomain (F139A, F153A, and I163A) all displayed weaker GDF8 inhibition. The most striking effects were seen with the mutations F139A and F153A, which were 10-and 18-fold weaker than WT WFIKKN2 (Fig. 5). Thus, the three residues that had a negative impact on GDF8 antagonism are located on the highly hydrophobic face of the WFIKKN2 FSD Kazal subdomain, as shown in Fig. 4.

Comparison of the WFIKKN2 FSD with the FSDs of follistatin
Using the structure of WFIKKN2 FSD, we can draw structural comparisons to the different FSDs in FS288 and FSTL3. Because the FSD1 and FSD2 in both FS288 and FSTL3 share similar structures, the comparison focuses on the FSDs within FS288. The WFIKKN2 FSD and the FSDs from FS288 and FSTL3 contain a similar domain architecture with EGF and Kazal protease inhibitor subdomains, including the spacing and alignment of the five disulfide bonds. In addition, all of the FSDs

WFIKKN2 follistatin domain structure
contain two highly conserved residues, Thr-146 and Tyr-147, highlighted in Fig. 6a, which are conserved in the broader Kazal domain of protease inhibitors (e.g. ovomucoids and serine protease inhibitor Kazal-type 1) (47)(48)(49). Despite these conservations, there are differences in the relative position of the subdomains, making it challenging to perform an overall alignment of the FSDs. However, the Kazal subdomains exhibit similar structures, where alignment results in a root mean square of 3.1, 2.1, and 2.9 Å 2 for FSD1, FSD2 and FSD3 of follistatin, respectively (40 ␣ positions), when aligned to WFIKKN2 FSD. The moderate differences of the C␣ positions between the Kazal subdomains localize in the loops connecting the ␣1-helix and the position of ␤-strands 3 and 4. Despite their similar structure, the Kazal subdomain of WFIKKN2 FSD only contains two of the three ␤-strands found in other FSDs (Fig. 6b).
Whereas the individual subdomains are structurally similar, there are significant conformational differences in the relative position of the subdomains. In retrospect, this explains why using the available FSD structures (FSD1-3) as search models failed to provide a molecular replacement solution for the WFIKKN2 FSD X-ray diffraction data. Thus, we aligned the structures using only the Kazal subdomain (Fig. 6b). Comparison of the structures shows that different FSDs can adopt either an "open" or "closed" conformation. An open conformation is observed in FSD1 and WFIKKN2 FSD, where the EGF ␤-strands are extended away from the Kazal domain, which creates a more linear appearance of the FSD domain. In contrast, a closed, more compact conformation is observed in FSD2 and FSD3, where the EGF domain, specifically ␤-strand 1, interacts with the Kazal domain. In FSD2, Thr-178, Val-180, and Val-181 within ␤-strand 1 of the EGF domain form hydrophobic interactions with the Kazal subdomain (Fig. 6b). Similarly, in FSD3, Lys-256, Leu-258, and Phe-261 of ␤-strand 1 in the EGF domain interact with the Kazal domain. Whereas the FSD of WFIKKN2 is in the open conformation, it should be noted that the N terminus contains tandem glutamine residues (Gln-113 and Gln-114) that wrap back into the Kazal subdomain to interact with the ␣1-helix. Residues at the interfaces of all of the FSDs are highly conserved across species, including Gln-113 and Gln-114, suggesting that they might serve a role to stabilize the different conformations of the FSDs.

Discussion
WFIKKN proteins are evolutionarily conserved in all vertebrates and can be found in other organisms, such as sea urchins and worms (37). When compared with other extracellular protein antagonists of the TGF-␤ family, WFIKKN proteins are remarkably specific for the ligands GDF8 and GDF11 (26,35). Although the basis for this specificity remains unknown, it is clear that the FSD plays an important role in binding and antagonism (34,38). For example, a construct containing only the WF domains of WFIKKN2 is sufficient for antagonism of GDF8, and the FSD of WFIKKN1 has a higher affinity for GDF8 than the Kunitz 2 and netrin domain (34,38). Interestingly, comparing the different domains of WFIKKN2, the FSD is the most conserved domain across mammalian species, with the lowest rate of substitution (37).
In other ligand antagonists, such as follistatin and FSTL3, FSDs play an important role in ligand binding and antagonism (32,40,50). However, whether the FSD from WFIKKN serves a similar role remains to be determined. To address this, we produced recombinant WFIKKN2 FSD and characterized its interaction with GDF8 and GDF11.
A complex between WFIKKN2 FSD and GDF8 or GDF11 was observed using native PAGE analysis, which SPR experiments revealed to involve submicromolar-affinity interactions. Using native PAGE, we also demonstrated that the addition of the type II receptor, ActRIIB, was able to dissociate a WFIKKN2 FSD:GDF11 complex, suggesting a direct competition between WFIKKN2 FSD and ActRIIB. Consistent with these results, SPR experiments showed that GDF11 in complex with ActRIIB-Fc was unable to bind WFIKKN2 FSD. Therefore, similar to their role in follistatin and FSTL3, the FSD of WFIKKN2 appears to block or at least compete for type II receptor binding on the ligand. However, full-length WFIKKN2 retained the ability to bind the GDF11:ActRIIB-Fc complex. This interaction is likely due to other domains of WFIKKN2 associating with regions of GDF8 and GDF11 apart from the type II-binding site. Thus, whereas the other domains of WFIKKN2 are important for potent antagonism, the WFIKKN2 FSD is sufficient to antagonize GDF8 and GDF11 by blocking type II receptor binding. Whether WFIKKN2 FSD or full-length WFIKKN2 interferes with type I receptor binding has not been

WFIKKN2 follistatin domain structure
resolved and is complicated by the low-affinity interaction of the type I receptor with GDF8 and GDF11 (25).
To help draw comparisons with the FSD of follistatin, we solved the crystal structure of WFIKKN2 FSD. The structure reveals a similar domain architecture with an N-terminal EGF subdomain and a C-terminal Kazal subdomain, as observed previously for other FSDs (23,32,40,51). Using the structure, we were able to identify hydrophobic residues at the surface that contribute to GDF8 antagonism. Single point mutations within the FSD of WFIKKN2 FL reduced the capacity to inhibit GDF8, presumably by reducing the ability of WFIKKN2 FL to interact with the type II receptor binding interface. These results are consistent with antagonists across the family that use hydrophobic interactions to engage the ligand surfaces important for binding type II receptors. However, it is possible that the residues identified could be important for interacting with other domains within WFIKKN2 FL, functioning to stabilize the WFIKKN2:ligand complex.
Comparison of WFIKKN2 FSDs across 54 species reveals a high sequence identity (68%) (Fig. S1). Interestingly, the residues identified (Phe-139, Phe-153, and Ile-163) within the FSDs as important for antagonism of GDF8 are highly conserved. Further, these residues are not conserved in other FSDs (Fig. 6), indicating that they are not necessary to the structural integrity of the Kazal subdomain. Thus, the decrease in antagonism is not likely due to a misfolded FSD within WFIKKN2. Taken together, these results support a functional rather than struc-tural role of Phe-139, Phe-153, and Ile-163 in the binding and antagonism of GDF8 and GDF11.
Interestingly, when compared with follistatin and FSTL3, WFIKKN2 interacts with the ligands through a different surface of the FSD. The WFIKKN2 FSD residues, Phe-139, Phe-153, and Ile-163, are located within the Kazal subdomain, centered on the hydrophobic face of the ␣1 opposite of the N-terminal loop (Figs. 4 and Fig. 7). In FSD1 and FSD2, residues that interact with the ligands are located in a completely different location of the domain, especially for FSD2, where the interaction surface is on the opposite side as compared with WFIKKN2 FSD (Fig. 7).
Although it is possible that the conformational differences observed between the EGF and Kazal subdomains of FSDs are related to crystallization, the linker region is well-ordered in the crystal lattice and exhibits similar temperature factors throughout the structure, suggesting that the conformation of the domain is stable. Consistent with the FSD having a stable structure, the open conformation of the follistatin FSD1 was observed in both the bound and unbound state, indicating that the domain structure is not affected by ligand binding and/or alternative crystallization conditions.
In addition to WFIKKN, follistatin, and FSTL3, FSDs are found in several proteins with various roles in TGF-␤ signaling. For instance, FSDs are also found in agrin, tomoregulin, and FSTL1, all of which are implicated in regulating TGF-␤ ligand activity (52)(53)(54). This raises a question as to the general func-

WFIKKN2 follistatin domain structure
tion of the FSD and whether the domain serves a common role in ligand interactions. Whereas the FSD in both WFIKKN and follistatin/FSTL3 both interact at the type II receptor interface, differences are readily apparent in both their structures and the placement of residues utilized for this interaction. Thus, it does not appear that the FSD represents a common binding domain, but rather a stable scaffold that allows the presentation of surface-exposed hydrophobic residues that can be important for protein-protein interactions.
Given the differences in domain architecture and the differences in the position of surface residues important for ligand binding, we anticipate that the FSD of WFIKKN2 does not resemble the ligand binding scheme of any of the three FSDs of follistatin. Certainly, resolution of WFIKKN2 FSD or FL in complex with a ligand will ultimately resolve these differences and help us to understand how WFIKKN specifically engages GDF8 and GDF11 over other TGF-␤ ligands.

WFIKKN2 FSD production and purification
Mouse WFIKKN2 FSD (residues 104 -172, 97% identical to human) with a cleavable N-terminal His 6 tag was cloned into pET28a(ϩ) and expressed in BL21 Rosetta cells. Cells were spun down and resuspended in PBS before sonication for cell lysis. Pellets were washed two times with PBS, 0.1% Triton X-100, followed by a final wash with PBS. Inclusion bodies containing the WFIKKN2 FSD were solubilized using 10 mM sodium tetrathionate, 100 mM sodium sulfite, 100 mM Tris, pH 8.5, 8 M urea, and 100 mM DTT. Solubilized inclusion bodies were dialyzed into 4 M urea, 50 mM sodium acetate, pH 4.5, 100 mM NaCl, 50 mM Tris, 15 mM imidazole and rapidly diluted into refolded buffer containing 100 mM Tris, pH 8.5, 150 mM NaCl, 1 mM EDTA, 0.1 mM oxidized GSH, 0.5 mM reduced GSH, and 0.5 M arginine to a final concentration of 0.1 mg/ml. After 3 days, protein was diluted 1:5 in nickel-nitrilotriacetic acid running buffer (50 mM Tris, pH 8, 500 mM NaCl, and 15 mM imidazole) and applied to a HisTrap TM Excel column (GE Healthcare). WFIKKN2 FSD was eluted with 500 mM imidazole and dialyzed into 10 mM HCl prior to separation on a C18 reversephase column (Phenomenex).

ActRIIB-ECD production and purification
ActRIIB-ECD was produced within SFϩ insect cells using pFastBac expression plasmid. Purification was conducted as described previously (44). In short, ActRIIB-ECD containing a His 6 tag was purified from SFϩ conditioned medium using a nickel-nitrilotriacetic acid affinity column. Bound protein was eluted with 20 mM Tris, pH 7.4, 500 mM NaCl containing 500 mM imidazole and subsequently applied to an S75 size exclusion column. Fractions containing the ActRIIB-ECD were pooled and used for native PAGE analysis.

Native PAGE and Western blotting
3 g of GDF11/8 was mixed with WFIKKN2 FSD at different molar ratios, starting with 4:1 (WFIKKN2 FSD:GDF11/8), and WFIKKN2 FSD was serially diluted 1:2 five times for a final ratio of 0.13:1. Native polyacrylamide gels (12%) were run at 20°C for 140 min at 110 V and stained using colloidal Coomassie Blue. For staining, gels were fixed using 40% EtOH and 10% acetic acid for at least 1 h before washing three times using distilled H 2 O. A working dye solution was used containing 80% colloidal Coomassie Blue and 20% methanol and stained overnight.

Native PAGE to SDS-PAGE transfer
Native polyacrylamide gels were run as described above. However, rather than a titration, one condition was repeated in seven lanes. One lane was cut, stained, and realigned to the unstained gel. The desired stained band was used as a guide to excise the other six unstained bands. Excised gel was then placed in a dialysis bag with 1 ml of SDS running buffer and electro-eluted with 180 V for 1 h. The buffer containing the eluted band was then concentrated and subjected to SDS-PAGE under reducing conditions followed by colloidal staining as described above.

Luciferase assay
Luciferase assays were conducted in HEK293 (CAGA) 12 cells as described previously by our laboratory (25,38,55). For the assay, 20,000 cells were seeded in growth medium in a 96-well format on poly-D-lysine-coated plates (catalog no. 655940,

WFIKKN2 follistatin domain structure
Greiner Bio-One GmbH). Cells were grown at 37°C with 5% CO 2 until reaching 75-85% confluence. Medium was then removed and treated with 100 l of serum-free medium containing GDF8 in the presence or absence of antagonists. GDF8 was kept at a constant concentration (0.62 nM) while antagonists (WFIKKN2 FSD and FL) were titrated in using 1:2 dilutions. After 1 day, cells were lysed using 1ϫ passive lysis buffer (E1941, Promega) on a plate shaker (800 rpm, 20 min, 20°C). Lysates were transferred to black and white 96-well plates, and 40 l of LAR (E1501 and E1960, Promega) was added. Firefly luminescence was measured using the Synergy H1 Hybrid Plate Reader (BioTek). All experiments were conducted independently at least twice with all data points being done in triplicate. The concentration of antagonists at which 50% of GDF8 activity is lost, or IC 50 , was calculated using nonlinear regression with variable slope using GraphPad Prism version 5 software.

Surface plasmon resonance
SPR experiments were conducted on the Biacore T200 microfluidic system. Ligands were primary amine-coupled to the GE series S CM5 sensor chip (catalog no. BR-1005-30) using the manufacturer's protocols. GDF8, GDF11, activin A, and TGF␤1 at 1 g/ml was sufficient to achieve ϳ500 response units bound to the chip. SPR experiments were conducted in HBS-EP buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% P-20 surfactant (Biacore AB)). WFIKKN2 fulllength or WFIKKN2 FSD was diluted in HBS-EP buffer to a concentration of 500 nM and then serially diluted 2-fold 10 times and applied to the chip at 15 l/min. Association was measured for 180 s followed by 120 s of buffer to measure dissociation. The dissociation equilibrium constant (K D ) was calculated using Biacore T200 software using a 1:1 binding model. For the receptor-binding experiments, ActRIIB-Fc at a concentration of 312.5 ng/ml (ϳ4.2 nM) was immobilized on a protein A chip. GDF11 was then applied at a concentration of 250 nM for 180 s. Subsequently, binding of 125 and 500 nM WFIKKN2 full-length or WFIKKN2 FSD to the receptor:ligand complex was measured.

Structural determination and experimental phasing
Crystals of WFIKKN2 FSD were grown in 2.5 M AmNO 3 and 0.1 M sodium citrate, pH 4.6, via hanging drop vapor diffusion at 4 mg/ml. Rectangular crystals grew to a size of 100 ϫ 200 m. Diffraction data were collected at Advanced Photon Source (APS) beamline GM/CA 23-ID-D at Argonne National Laboratory. Molecular replacement techniques for phasing were unsuccessful, requiring experimental phasing via single isomorphous replacement with anomalous scattering (SIRAS). In short, crystals were soaked in 5 mM KCl 4 Pt for 96 h before being transferred to a cryogenic solution containing mother liquor and 35% PEG 550 and flash-frozen. Heavy atom positions were identified using phenix AutoSol (56,57). Phasing and model refinement were carried out using AutoSol and phenix.refine (58). Model validation was performed using the program Mol-Probity (59,60). Coordinates are available at the Protein Data Bank, accession code 6MAA. Hydrophobicity depicted in Fig. 4 was determined using the Color_H PyMOL script (46).

Mutagenesis and purification of full-length WFIKKN2
Full-length WFIKKN2 was purified as described previously (38). In short, conditioned medium from CHO cells stably expressing WFIKKN2 was collected and subjected to butyl-Sepharose and heparin columns. Elutions containing WFIKKN2 were dialyzed into 50 mM Tris, pH 7.4, 20 mM NaCl, 1 mM EDTA and applied to a MonoQ 10/100 GL column and eluted with a linear gradient of NaCl. Following Mono Q, elutions were subjected to S2000 size-exclusion chromatography to obtain pure WFIKKN2. Full-length WFIKKN2 was cloned into pcDNA4 mammalian expression vector and subjected to site-directed mutagenesis. WFIKKN2 mutants were transfected and produced via HEK293F cells. Following expression, WFIKKN2 mutants were purified similarly to full-length WFIKKN2 produced in CHO cells.