The CKGF Domain and Its Families

The cystine knot (CK) growth factor (CKGF) domain has a specific three-dimensional fold and sequence. Other proteins contain CKs and thus the CK alone does not define the CKGF domain; the definition of this domain also includes the topological relationship between β-strand order in β-sheets and amino acid sequence. The CKGF domain can thus be recognized both by sequence and by structure (Roch and Sherwood 2014), similarly to epidermal growth factor (EGF) or immunoglobulin (Ig) domains. And just as the latter families have diverse functions, many proteins containing CKGF domains lack growth factor activity. Most TGF-β family members act as differentiation factors rather than growth factors and some such as inhibins and Leftys antagonize activities of other members. However, just as TGF-β is eponymous for the family as a whole, its signaling moiety is also eponymous for what we term the GF domain. All family members contain a GF domain that can form a GF dimer; thus, we term this the GF domain whether or not it has signaling activity.

The CKGF domain contains two long β-ribbons (β-sheets with two antiparallel β-strands) and a CK (GF monomer, Box 1). Two closely spaced pairs of cysteines in adjacent, parallel β-strands disulfide-link to form a ring composed of two peptide backbone segments and two disulfides. Another disulfide passes through the ring, linking two additional polypeptide segments (cystine knot, Box 1). Thus, the knot ties together four polypeptide segments that are distal in sequence, and forms a highly stable CK core from which three long loops emanate (GF monomer, Box 1).

The CKGF domain superfamily is present in the earliest metazoans. Like the Wnt/Frizzled and Delta/Notch families, the CKGF family is not found in protists, and first appears in primitive metazoans that lack a bilateral axis. Representatives are present in four distinct prebilaterian phyla, that is, Porifera (sponges), Cnidaria (coral and hydra), Ctenophora (comb jellies), and Placozoa (Trichoplax) (Roch and Sherwood 2014). The CKGF superfamily is composed of six groups (families) with the CKGF domain as the primary structural feature, that is, with or without a prodomain and without any other folded domain (Adamska et al. 2007; Roch and Sherwood 2014). These are (1) the TGF-β family, (2) the Dan family of BMP antagonists, (3) the glycoprotein hormone family (GPH), for example, the pituitary hormones follicle-stimulating hormone, luteinizing hormone, and thyroid-stimulating hormone, (4) bursicon hormones, which are limited to invertebrates, (5) the platelet-derived growth factor (PDGF) family, including vascular endothelial growth factors (VEGFs), and (6) the nerve growth factors (NGFs) (Roch and Sherwood 2014). CKGF domains are also found in noggin, in the CCN family, and as carboxy-terminal CK domains.

From Prebilaterians to the Emergence of TGF-β Itself in Deuterostomes

Of these six families, the first five are represented in prebilaterian phyla. Moreover, sequences characteristic of specific Dan family members represented in Cnidaria, Porifera, Ctenophora, and Placazoa are detectable in humans. In other words, statistically significant relationships between sequences can be used to infer orthology so that prebilaterian Dan family members can be linked to specific proteins or subfamilies of related proteins in humans, including gremlin, sclerostin, and uterine sensitization-associated gene 1 (USAG1) (Roch and Sherwood 2014). TGF-β family members and their type I and type II receptors are also present in nonbilaterians, but orthology with particular family members in chordates is unclear. In contrast, Smad orthologues corresponding to Smad 1 or 5, Smad 2 or 3, the co-Smad Smad4, and inhibitory Smads (I-Smads) can be identified in nonbilaterians (Herpin et al. 2004; Huminiecki et al. 2009; Pang et al. 2011).

The next step in evolution is marked by the definition of a new body axis, the bilateral axis, which creates dorsoventral asymmetry and also enables development of left–right asymmetry. Bilaterian phyla are grouped by nucleotide and protein sequence into three branches that extend previous embryological classification (Blair and Hedges 2005; Dunn et al. 2008; Hejnol et al. 2009). The protostome branch is now subdivided into the Ecdysozoa (e.g., Caenorhabditis elegans and Drosophila melanogaster) and Lophotrochozoa (e.g., molluscs and annelids) branches. Deuterostomes represent a later evolving group of phyla. Molecular phylogeny confirms deuterostomes as a branch with three phyla, Echinodermata (e.g., sea urchins), Chordata (vertebrates, urochordates, and cephalochordates), and Hemichordata (acorn worms).

Key innovations in the evolution of deuterostomes from lower bilaterians include the emergence of pharyngeal gill slits and an inversion in the dorsoventral axis of the body plan (Holland et al. 2015; Lowe et al. 2015). Notably, the BMP and chordin gradients that establish dorsoventral polarity also invert in deuterostomes (Lowe et al. 2015).

Despite the presence of multiple TGF-β family members, TGF-β family receptors, and Smads, TGF-β is clearly absent from early bilaterians. TGF-β itself first appears in the genome-based evolutionary record with the emergence of deuterostomes. Lower deuterostomes, including sea urchins, tunicates, and hemichordates, have only a single TGF-β, and in all cases its prodomain has an RGD motif (Robertson and Rifkin 2013). Mammals have three TGF-βs. TGF-β1 and -β3 contain RGD motifs and bind and are activated by integrins. TGF-β2 diverged from TGF-β3 after the divergence of TGF-β1. Thus, the lack of the Arg of the RGD motif in TGF-β2 is an acquired, not ancestral, characteristic. Furthermore, the acidic Asp residue retained in TGF-β2 is the only key residue required for integrin binding, and thus it is possible that TGF-β2 is activated by a yet unidentified integrin.

These evolutionary events shed important light on the TGF-β family and answer one of the commonly asked questions in the field, “Why are there so many antagonists of BMPs and activins but not TGF-βs?” Currently sequenced nonbilaterian metazoans have as many or more members of the CKGF Dan family than of the CKGF TGF-β family itself. Therefore, in prebilaterians, BMP antagonist and TGF-β family members coevolved in the presence of one another (Roch and Sherwood 2014), and in the presence of other developmental pathways such as Wnt (Adamska et al. 2007). These signaling families matured and diversified much further in bilaterian invertebrates, to the point where members in Drosophila can be grouped with specific TGF-β subfamilies in humans (TGF-β GFs, Box 1). In contrast, TGF-β itself evolved much later, long after the diversification of bilaterian phyla, at the emergence of deuterostomes. Integrins, like the CKGF superfamily, are present in both prebilaterian and early bilaterian metazoans, in which they are already well diversified. Thus, TGF-β emerged in a context in which the key tasks of bilaterian axis and organ specification by complex signaling networks were well established. Within this context, TGF-β evolved to take advantage of a new regulatory mechanism, in which integrins activate signaling. Such a new mechanism may have been required because the regulatory hubs involving BMP antagonists were already well established in the specification of body axes and organs, and their use by a radical new CKGF member might have thrown a wrench into this machinery.

Family Trees

The family trees in Box 1 show the diversification of family members in humans and D. melanogaster for TGF-β family GFs and prodomains (separately), BMP antagonists, and TGF-β family type I and type II receptors. These trees were constructed using rigorous methods appropriate for defining phylogenetic relationships, as implemented in MAFFT (Katoh and Standley 2013). Scale bars show the extent of sequence divergence, which for any two members is given by adding the lengths of the branches that separate them. However, the trees show ontogenetic relationships, not phylogenetic relationships. Thus, the trees in Box 1 show sequence divergence as a consequence of functional and structural diversification, rather than as a consequence of separation of animal species in evolutionary time. Furthermore, the shapes of the trees in Box 1 provide insights into how function and structure have driven diversification of the subfamilies. Separate trees for the TGF-β family prodomains and GFs emphasize that the prodomains are more divergent; when the scale bars in Box 1 are taken into account, the prodomains are about fourfold more divergent than the GFs. Thus, the prodomains have been under greater functional or structural pressure for divergence than the GFs. Also, these pressures must differ, because branch lengths and clustering within subfamilies differ; however, because of higher sequence identity for the GFs, subfamilies can be defined with more statistical significance.

Clustering using neighbor joining with bootstrap calculations (here) or Bayesian analysis (Pang et al. 2011) support significant relationships among TGF-β family GF domains (color-coded in Box 1). A BMP clade includes six subfamilies. Human BMP-5, 6, 7, and 8 and Drosophila Glass bottom boat (Gbb) and Screw cluster in one group that is significantly related to the cluster with human BMP-2 and 4 and Drosophila Decapentaplegic. Four other clusters contain BMP-9 and 10; GDF-5, 6, and 7; GDF-9 and BMP-15; and GDF-1 and 3. Nodal falls in the same BMP clade. An activin clade includes three subgroups: Drosophila activin and human inhibin β-subunits, which dimerize to form activins, GDF-11, myostatin (GDF-8) and possibly Drosophila Myoglianin, and BMP-3 and GDF-10. Further family members in the left side of the GF family tree in Box 1 are difficult to place, but lie closer to the activin than BMP clade. All have unique features. TGF-βs and anti-Müllerian hormone (AMH) each have unique type II receptors. AMH and GDF-15 have atypical prodomains that are the longest and shortest in the family, with 434 and 170 residues, respectively. Lefty 1 and 2 are cleaved by furin after the α2-helix in the prodomain, rather than between the prodomain and GF domain (Fig. 1), and are thus predicted to have their type II receptor binding sites permanently blocked (Shi et al. 2011; Mi et al. 2015). Inhibin-α is inhibitory; when it heterodimerizes with inhibin β-subunits it forms inhibin.

Open in a separate window

Open in a separate window

Figure 1.

Sequence alignment of the transforming growth factor β (TGF-β) family. Sequences were aligned with MAFFT (Katoh and Standley 2013), as described (Shi et al. 2011), with slight structural realignment of TGF-β1 and bone morphogenetic protein (BMP)-9 procomplexes (Mi et al. 2015). Structure elements of pro-TGF-β1 (Shi et al. 2011) and pro-BMP-9 (Mi et al. 2015) are labeled and shown as colored lines (upper and lower lines, respectively). Potential proprotein convertase (PC) and tolloid cleavage sites are marked with green and red circles, respectively. Potentially N-glycosylated asparagines in all family members and known phosphorylated serines in mature BMP-15 and growth and differentiation factor 9 (GDF-9) (Tibaldi et al. 2010) are shown in bold red. All 33 human TGF-β family polypeptides are shown, except BMP-8B; the almost identical BMP-8A is shown as BMP-8. Three well-characterized Drosophila melanogaster BMPs are included: Decapentaplegic (Dpp), Screw (Scw), and Glass bottom boat (Gbb). Closely related family members appear in adjacent rows.

Type I and type II receptors for TGF-β family GFs have structurally homologous ligand-binding ectodomains and homologous dual specificity kinase cytoplasmic domains, and thus may be compared in the same tree (TGF-β receptors, Box 1). The longer branches of type II receptors show that they are more diverse than the type I receptors; the receptor trees reflect primarily the cytoplasmic domain because of its much greater number of aligned positions. The human type I receptors show three clades, each with a Drosophila relative. Activin receptor-like kinase (ALK) 1 and 2 are closely related to Saxophone (Sax), ALK3 and 6 are related to Thickveins, and ALK4, 5 and 7 are relatives to Baboon. The ALK1/2, ALK3/6, and ALK4–7 subfamilies couple to two distinct groups of receptor-regulated Smads (R-Smads), with the ALK1/2 and ALK3/6 subfamilies coupling to R-Smad1, 5, 8 and the ALK4/5/7 subfamily coupling to R-Smad2 and 3. Type II receptors also show three groups. BMPRII and AMHRII relate to Wishful thinking, and ActRII and ActRIIB ally with Punt. In contrast, among all type I and type II receptors, only TβRII lacks a Drosophila relative, consistent with the recent deuterostome origin of TGF-β. Moreover, TβRII has no close human relatives.

The Dan family of antagonists against BMPs and Wnts has seven members in humans and shares some similarity with noggin in its CKGF domain (BMP antagonists, Box 1). Although present in many prebilaterians, the Dan family was lost in insects.

Structures of Prodomain–GF Complexes

Structures of TGF-β1 (Shi et al. 2011) and BMP-9 (Mi et al. 2015) procomplexes, which share only 11% prodomain sequence identity, illuminate TGF-β family diversity (Fig. 2A,B). The prodomain contains an arm domain flanked by shorter, partially α-helical, amino- and carboxy-terminal straitjacket elements (Fig. 2C,D). TGF-β1 and BMP-9 procomplexes reveal overall crossed-arm and open-arm conformations, respectively, with markedly different orientations between arm domain monomers (Fig. 2A,B). In the open-arm conformation of pro-BMP-9, the two arm domains point away from one another and are not in contact (Fig. 2B). In the crossed-arm conformation of pro-TGF-β1, the arm domains point toward one another and dimerize at the bowtie (Fig. 2A). In contrast to the arm domain, the segments amino-terminal (α1-helix, latency lasso, and fastener) and carboxy-terminal (α5-helix) to the arm domain interact differently with the GF domain in the two conformations (Fig. 2A–F).

Open in a separate window

Figure 2.

Procomplexes of transforming growth factor (TGF)-β1 and bone morphogenetic protein 9 (BMP-9) have crossed-arm and open-arm conformations, respectively. Crystal structures are shown in cartoon representation. Some labels are shown in the color of the corresponding proteins or protein segment. Cysteine side chains are shown as yellow sticks and in some cases with sulfur atom spheres. (A,B) Overall structures of pro-TGF-β1 (Shi et al. 2011) (A), and pro-BMP-9 (Mi et al. 2015) (B) in identical orientations with superimposition on GF dimers. Yellow spheres in A show Cys residues that disulfide-bond to latent TGF-β binding protein (LTBP) or GARP. (C,D) The prodomains in identical orientations after superimposition on the arm domains. Arm domain secondary structural elements common or unique to TGF-β1 and BMP-9 are labeled in black and red, respectively. Amino-terminal and carboxy-terminal appendages to the arm domain are labeled and are, in order, the α1-helix, latency lasso, α2-helix, and fastener (all amino-terminal) and the α5-helix (carboxy-terminal). (E,F) Interacting regions of the prodomain and GF shown in identical orientations after superimposition on GF dimers. (G–I) Competition of the prodomain with receptor and inhibitor binding. Complexes are shown in identical orientations after superimposition on GF dimers. For clarity, GFs in receptor and inhibitor complexes are omitted. (G) TGF-β1 procomplex (Shi et al. 2011) superimposed on TGF-β1 complex with ALK5 (R type I) and TβRII (R type II) (Radaev et al. 2010). (H,I) BMP-9 procomplex (Mi et al. 2015) superimposed on (H) BMP-9 complex with ALK1 (R type I) and ActRIIB (R type II) (Townson et al. 2012) or superimposed on (I) BMP-2 complex with a crossveinless 2(CV2)VWC domain fragment (Zhang et al. 2008).

Despite differences in orientation, the arm domains of TGF-β1 and BMP-9 show overall similar folds. TGF-β1 and BMP-9 arm domains have two β-sheets that partially overlap in the hydrophobic core. Long meandering loops and a substantial α4-helix cover the remainder of the core (Fig. 2C,D). The β1-strand of the arm domain hydrogen bonds to the β7-strand in the GF finger to form a super-β-sheet (Fig. 2E,F). This super-β-sheet knits the prodomain and GF together. Substantial twisting of the arm β1-strand between the crossed-arm and open-arm conformations enables the arm domain to reorient by ~90°, while nonetheless maintaining equivalent super-β-sheet hydrogen bonds in TGF-β and BMP-9 procomplexes (Fig. 2A,B,E,F).

A long loop at the opposite end of the arm domain special to pro-TGF-β1, β2, and β3 contains cysteines that disulfide link the two arm domains together to stabilize the crossed-arm procomplex conformation (Fig. 2A). The disulfide-linked region is termed the bowtie knot; the region that follows the knot and includes the RGD motif is termed the bowtie tail. The bowtie tail greatly changes in conformation when bound to integrins (Dong et al. 2014; X Dong, B Zhao, and TA Springer, unpubl.).

Elements amino terminal to the arm domain surround the GF dimer in the crossed-arm conformation of pro-TGF-β and form a straitjacket (Fig. 2A,E) (Shi et al. 2011). The α1-helix is buried between the two GF monomers in an extensive interface that includes the amphipathic, carboxy-terminal end of the α1-helix (Fig. 2A,E). Hydrophobic residues in three turns of the α1-helix interface with aromatic residues on one GF monomer. α1-helix burial helps stabilize the crossed-arm conformation with its unique arm domain orientation, and displaces the GF α3-helix from its usual position in the interface between the two monomers. The latency lasso loosely surrounds the GF finger (Fig. 2A,E). The fastener, a short segment just before the arm domain, binds to the α1-helix and completes GF encirclement (TGF-β1 homodimer, Box 1, Fig. 2A,C,E). These elements form a straitjacket that prevents receptor binding (Fig. 2G) and maintains TGF-β latency, as verified by mutation (Shi et al. 2011).

In native pro-TGF-β complexes with LTBP and the cell-surface leucine-rich repeat protein 32 (LRRC32 or GARP), cysteines in an “association region” that comprises the most amino-terminal portion of each prodomain (Figs.1 and ​and2A)2A) link to two cysteines in a single LTBP or GARP molecule in a 2:1 complex (Shi et al. 2011; Wang et al. 2012). Thus, the prodomains are covalently linked together, directly through the bowtie knot at one end of the wreath-shaped procomplex, and indirectly through an intermediate protein on the opposite side of the procomplex (Fig. 2A). Unpublished pro-TGF-β crystal structures show that, in the absence of an LTBP or GARP partner, the association region of ~10 residues is unstructured or varies in structure depending on the crystal lattice (X Dong, B Zhao, and TA Springer, unpubl.). Proper disulfide bond formation is guided by noncovalent interactions surrounding the cysteines. Because the TGF-β binding-like (TB3) domain in LTBP and the LRRC domain in GARP that disulfide-link to pro-TGF-β have no structural similarity, the association region must adopt different conformations when linked to these different partners. Furthermore, because the complex is 2:1, each prodomain monomer must associate with a different cysteine and surrounding region in the binding partner, resulting in a different structure in the association region of each monomer (shown in Fig. 3A,B).

Open in a separate window

Figure 3.

Models for growth factor (GF) latency, activation, and binding to receptors. (A,B) Schematic representation of how anchoring molecules present pro-transforming growth factor β1 (TGF-β1) in the extracellular matrix (ECM) or on cell surfaces, and how the association region at the amino terminus of the prodomain must take on different conformations to bind distinct anchoring molecules and two different sites within each anchor molecule. (C) Schematic representation of pro-TGF-β1 activation (Shi et al. 2011). Integrins, such as αvβ6, attached to the actin cytoskeleton exert traction force against RGDLXXL/I motifs in pro-TGF-β1 and -β3, which are covalently linked to anchor molecules in the ECM or on the cell surface. The most force-susceptible structural region is the straitjacket and its latency lasso, which will be elongated. Consequently, TGF-β1 is released, and shifts conformation to its free form. (D) Models for conformational change between crossed-arm, latent, and open-arm nonlatent conformations of pro-BMPs (Mi et al. 2015). (1) Open-arm, nonlatent conformation; (2) putative crossed-arm conformation of pro-BMPs stabilized by association with ECM components; (3,4) step-wise binding to type I (3) and type II (4) receptors. Putative pathways for secretion (a,b) and conformational interconversion (c) are marked.

In the open-arm conformation of pro-BMP-9, the prodomain α1-helix and latency lasso are not visualized (Fig. 2B) (Mi et al. 2015). Provocatively, the α1-helix sequence and especially its amphipathic signature are well conserved between BMP-9 and TGF-β procomplexes and among the majority of TGF-β family members (Fig. 1). This suggests that pro-BMP-9 may assume an alternative, crossed-arm conformation with an ordered α1-helix that resembles pro-TGF-β. Models show that such a conformation is plausible for pro-BMP-9 (schematized in Fig. 3D), and movies show a pathway for interconversion between open-arm and crossed-arm conformations of pro-BMP-9 (see Movie 1) (Mi et al. 2015).

The sequence of the prodomain α2-helix is conserved in all 33 TGF-β family polypeptides (Fig. 1). The α2-helix has a unique role at the interface between the arm and GF domains. The α2-helix is amphipathic, and its hydrophobic face shields hydrophobic residues in the GF fingers from solvent. Between the crossed-arm and open-arm conformations, the α2-helix maintains its position relative to the GF fingers (Fig. 2A,B,E,F). In contrast, the α2-helix moves ~90° with respect to the arm domain between crossed- and open-arm conformations (Fig. 2C,D). Reorientation of the α2-helix in pro-BMP-9 allows it to interact with the arm domain in a way not seen in pro-TGF-β1 (Mi et al. 2015).

The α5-helix in pro-BMP-9 is much longer than in pro-TGF-β1 (Fig. 1), orients differently relative to the arm domain, and binds to a similar region of the GF domain as the α1-helix in pro-TGF-β1 (Fig. 2A,B,E,F) (Mi et al. 2015). The BMP-9 prodomain α5 helix inserts into the hydrophobic groove formed by the fingers of one GF monomer and α3-helix of the other GF monomer (Fig. 2F). Thus, instead of displacing the GF α3-helix as the prodomain α1-helix does in pro-TGF-β1, the α5-helix in pro-BMP-9 binds adjacent to the GF α3-helix (Fig. 2F), enabling a relaxed, mature-like GF conformation in pro-BMP-9. In pro-TGF-β1 the α5-helix also binds the GF; it fills a small gap between the arm domain and α1-helix (Fig. 2E) (Shi et al. 2011).

A remarkable feature of the pro-TGF-β dimer is a swap in the GF monomer that each prodomain monomer embraces. The swap was suggested based on the structure of PC-cleaved pro-TGF-β1, in which prodomain residues preceding the PC cleavage site are disordered (Shi et al. 2011). In a structure with the PC site mutated (B Zhao, X Dong, and TA Springer, unpubl.), the linkage is also flexible, as expected for a polypeptide that is susceptible to PC cleavage. However, the disordered polypeptide is only long enough to span one of the two possible prodomain–GF connections (TGF-β1 monomer, Box 1). This establishes that the most intimate contacts are between pro- and GF domains of different polypeptide chain monomers in the precursor in the ER. Thus, the straitjacket, α2-helix, arm domain, and α5-helix from prodomain monomer 1 surround the GF from monomer 2 (Fig. 2A).

Prodomain–GF swapping has important biological implications, because it can provide a mechanism for preferential formation of heterodimers over homodimers when a cell synthesizes monomers for two different TGF-β family members. In zebrafish embryos, BMP-2/7 heterodimers pattern the dorsoventral axis, whereas the respective BMP-2 and BMP-7 homodimers are insufficient (Little and Mullins 2009). Overexpression of recombinant human BMPs also suggests greater activity of BMP-2/5, 2/6, 2/7, and 4/7 heterodimers than the corresponding BMP homodimers (Israel et al. 1996). Complementary interactions between swapped prodomains and GF domains must also be important among inhibin α- and β-subunits, which form both activating activin β homodimers and inhibitory inhibin αβ heterodimers (Walton et al. 2009).

Competition between Prodomains, Receptors, and Antagonists for GF Binding

Prodomains encoded by the 33 TGF-β family genes have important yet poorly understood roles, including extending the range of BMP signaling in vivo (Cui et al. 2001; Harrison et al. 2011; Akiyama et al. 2012; Constam 2014). Many prodomains form strong complexes with their GFs (Brown et al. 2005; Sengle et al. 2008), and some of these bind tightly enough to compete for binding of the GFs to their cognate receptors (Sengle et al. 2008; Walton et al. 2009; Mi et al. 2015). Interestingly, the prodomains seem to compete mainly with type II receptors but rarely with type I receptors (Sengle et al. 2008; Hauburger et al. 2009; Mi et al. 2015).

The strong effect of BMP prodomains on type II receptor binding can be explained by the pro-BMP-9 structure (Fig. 2H). The prodomain arm domain and α2-helix occupy the type II receptor-binding site. In contrast, only the prodomain α5-helix blocks the type I receptor binding site (Fig. 2H). The α5-helix binds weakly as suggested by the relatively small surface on the GF that it buries, and by its weak electron density in certain crystallization conditions (Mi et al. 2015). These structural results together with receptor competition experiments suggest that, among BMPs, type I receptors may displace the prodomain α5-helix from the GF, whereas the prodomain α2-helix and arm domain remain bound to the GF and inhibit type II receptor binding (Sengle et al. 2008; Mi et al. 2015).

Chordin and chordin-related BMP inhibitors contain one or several von Willebrand factor C (VWC) domains as BMP-inhibiting moieties. Most interestingly, the binding of such a domain of the chordin-member crossveinless 2 to BMP-2 (Zhang et al. 2008) is reminiscent of prodomain binding to BMP-9 (Fig. 2I). The VWC domain binds to the GF fingers similarly to the α2-helix and arm domain, whereas an amino-terminal appendage called clip binds to the same site as the prodomain carboxy-terminal appendage, the α5-helix (Fig. 2I, further described in section on Chordin Family Members). It is, therefore, theoretically possible that prodomains could help deliver GFs to receptors by shielding GFs from inhibitors; a similar mechanism has been shown for chordins (Ashe and Levine 1999; De Robertis et al. 2000; Harland 2001).

Integrin- and Force-Dependent Release of TGF-β1 from the Prodomain

TGF-βs evolved in deuterostomes with a unique integrin-dependent mechanism of activation. In the crossed-arm conformation of pro-TGF-β1, binding to both type I and type II receptors is completely blocked (Fig. 2G) (Shi et al. 2011). An integrin-binding RGDLXX(L/I) motif in the prodomains of TGF-β1 and 3 locates to the shoulder region of their arm domains, between the β7- and β10-strands of the arm domain (Fig. 1). Integrins αvβ6 and αvβ8 bind this motif with ~10 nM affinity, which is extremely high for integrins (Dong et al. 2014). Integrin binding is not sufficient to release and thus to activate TGF-β (Shi et al. 2011). Anchoring pro-TGF-β covalently to LTBP in the extracellular matrix (ECM) or covalently to GARP on cell surfaces is required for activation (Munger et al. 1999; Annes et al. 2004; Yoshinaga et al. 2008; Wang et al. 2012). Furthermore, activation requires the cytoplasmic domain of the integrin β-subunit, which connects to the actin cytoskeleton and exerts traction (Munger et al. 1999). These and other experiments (Wipff et al. 2007; Wipff and Hinz 2008) suggest that traction force exerted by integrins on the RGDLXX(L/I) motif in the arm domain, and its resistance on the opposite side of the prodomain by association region cysteines that link to LTBP and GARP, is required for activation (Fig. 3C). The results suggest a model in which applied traction force distorts the straitjacket region and results in release of TGF-β (Shi et al. 2011).

Conformational Change in Procomplexes as a Mechanism for Regulating Activity in the TGF-β Family

Some members of the TGF-β family may be able to access both open-arm and crossed-arm procomplex conformations (Fig. 3D). BMP-9 may be one of these; its procomplex structure shows an open-arm conformation (Fig. 3D, structure 1), whereas strong conservation of its α1-helix sequence suggests the presence of a distinct, putative crossed-arm conformation (Fig. 3D, structure 2) (Mi et al. 2015). Modeling of a crossed-arm BMP-9 procomplex shows that the prodomain α1-helix and GF can associate in a crossed-arm conformation that recapitulates the hydrophobic interface in pro-TGF-β1 (see Movie 1) (Mi et al. 2015). BMP-9 regulates vascular development. Mutations in its prodomain can cause syndromes resembling hereditary hemorrhagic telangiectasia (HHT) caused by mutations in receptors for BMP-9 (David et al. 2009; Castonguay et al. 2011; Wooderchak-Donahue et al. 2013). The location of mutations associated with an HHT-like condition in the BMP-9 prodomain α1-helix and in the subsequent loop supports the hypothesis of a crossed-arm conformation. A similar embrace between the prodomain α1-helix and mature domains in inhibin α and βA is shown by mapping of disruptive mutations to the hydrophobic face of the amphipathic α1-helix (Walton et al. 2009). Thus, multiple family members besides pro-TGF-β1 may adopt an α1-helix–associated, putative crossed-arm conformation.

Conformational changes between latent, crossed-arm conformations and nonlatent, open-arm conformations may regulate whether procomplexes are stored in the ECM in a latent form or released in a nonlatent form for signaling (Fig. 3D) (Mi et al. 2015). Because the crossed-arm conformation is tensed with more surface area buried between the GF and prodomain, and with the GF surrounded by a straitjacket, it may correspond to a latent conformation. In contrast, the open-arm conformation may correspond to a nonlatent conformation. This idea is consistent with latency of crossed-arm pro-TGF-β1 (Shi et al. 2011) and nonlatency of open-arm pro-BMP-7 and -9 (Mi et al. 2015). Although most members of the TGF-β family are nonlatent when overexpressed as recombinant proteins (Fig. 3D, pathway a, structure 1) (Brown et al. 2005; Sengle et al. 2011; Mi et al. 2015), latency may differ in vivo, in which cells cosynthesize TGF-β family members with components of the ECM (Fig. 3D, pathway b). Prodomains can bind to ECM components, including heparin, proteoglycans, LTBP, and fibrillin (Gregory et al. 2005; Anderson et al. 2008; Sengle et al. 2008, 2011; Li et al. 2010); one or more of these may bind to procomplexes, and thereby stabilize the crossed-arm conformation (Fig. 3D, structure 2) and enable the GF domain, which is very short-lived in vivo, to reach storage concentrations as high as 100 ng/g in demineralized bone (Pietrzak et al. 2006). The short lifetimes of free BMPs, together with significant storage depots of BMPs in bone, and colocalization of GFs with prodomains (Sampath and Reddi 1981, 1984; Gregory et al. 2005; Pietrzak et al. 2006), suggest latency in vivo. Furthermore, purification of BMPs showed a 60-fold increase in total activity in the first two steps, consistent with possible purification away from an inhibitor such as the prodomain (Luyten et al. 1989). Release from storage depots in vivo may be associated with conformational change to the open-arm conformation, which is ready for receptor or inhibitor binding (Fig. 3D, pathway c).

Among different family members, differences in structural elements, including those corresponding to the straitjacket and fastener, may regulate whether crossed-arm or open-arm conformations predominate. The prodomain α1 and α5-helices are likely to compete for overlapping binding sites on the GF in pro-BMP-9 (Mi et al. 2015). Therefore, the relative strengths of α1- and α5-helix binding among different TGF-β family members may regulate which conformation is preferred (Fig. 3D, pathway c; see Movie 1). In a single family member, binding to ECM components, and proteolytic removal by PC or tolloid proteases of helical segments amino- or carboxy-terminal to the arm domain may regulate which state predominates. Many family members contain known or putative PC or tolloid/BMP protease cleavage sites in these segments (Fig. 1). The open-arm conformation of BMP-9 can readily bind to type I receptors, with displacement of the α5-helix (Fig. 3D, structure 3) (Mi et al. 2015). The final step in signaling could then be complete prodomain dissociation followed by binding to type II receptors (Fig. 3D, structure 4).

Structures of Type I and Type II Receptor Ectodomains

TGF-β family GF dimers initiate signaling by binding to type I and type II receptors. The receptor ectodomains of ~100 residues are cysteine-rich. Homology in sequence and structure is detectable between type I and type II receptors in both their ectodomains and kinase domains, and thus they arose from a common ancestor (Box 1, Fig. 5). Four of the seven mammalian type I receptors and four of the five mammalian type II receptors are structurally characterized (Table 1, Fig. 6). Because the receptor ectodomain fold has three fingers (RI and RII, Box 1) and the superfamily includes neuro- and cardiotoxins, it is sometimes termed the three-finger toxin fold (Greenwald et al. 1999). The fold comprises a central three-stranded β4-β3-β5 antiparallel β-sheet flanked on its convex surface by a smaller two-stranded β1-β2 antiparallel β-sheet, and along the edge of β-strand 5 by an extended loop of variable structure that connects β-strands 4 and 5 (Fig. 6).

Open in a separate window

Figure 5.

Alignment of type I and type II receptor ectodomains by structure and sequence. Type I and type II receptors with structures (labeled with PDB codes) were aligned to one another structurally using DeepAlign (Wang et al. 2013); that is, only residues that are in equivalent positions in three-dimensional structures are aligned in the same column in the sequence alignment. Cα atom root mean square deviation (RMSD, Å) of each type I and type II receptor structure to their average position in the structural superposition is shown at each position in the alignment. Gaps were closed to minimize the length of the alignment; corresponding positions lack an RMSD value. Asterisks in the RMSD row show positions in which type I receptor cysteines are poorly aligned with one another structurally and were manually aligned. Human type I and type II receptors were also aligned separately from one another by sequence using MAFFT (Katoh and Standley 2013). Sequences not represented by structures were then aligned to those in the structural alignment. Color code is modified from ClustalX. Known disulfide bonds are color coded and lettered, except for the “h” disulfide in AMHRII, which is predicted by the structural alignment (see section on Structures of Type I and Type II Receptor Ectodomains). β-strand arrows above the alignment shown with dark or light shading show positions where all or some structures, respectively, have β-structure.

Open in a separate window

Figure 6.

Type I and type II receptors of the transforming growth factor β (TGF-β) family. Structures of type I (A–D) and II (E–H) receptor ectodomains are shown as ribbon diagrams, with disulfide bonds depicted as sticks with yellow sulfur atoms. Disulfide bond letters and β-strand numbers correspond to those in Fig. 5. A loop without reliable electron density in ALK3 is shown as a dashed line. Structures are for ALK6 (Kotzsch et al. 2009), ALK3 (Kirsch et al. 2000), ALK1 (Townson et al. 2012), ALK5 (Groppe et al. 2008), ActRII (Greenwald et al. 1999), ActRIIB (Thompson et al. 2003), BMPRII (Mace et al. 2006), and TβRII (Boesen et al. 2002).

Type I and type II receptors all share a common set of four disulfide bonds termed a, b, c, and d (Figs. ​(Figs.55 and ​and6).6). The e–g disulfides, which are present in only some receptors (Figs. ​(Figs.55 and ​and6),6), alter structures to enable distinctive GF binding modalities. Thus, the f disulfide, which is present in ActRII, ActRIIB, and BMPRII (Figs. ​(Figs.55 and ​and6E–G),6E–G), braces the amino-terminal portion of the β4-β5 loop so that it points away from the convex surface of the β4-β3-β5 sheet. This is important for the function of ActRII (Greenwald et al. 2003; Allendorph et al. 2006) and ActRIIB (Thompson et al. 2003; Weber et al. 2007), and likely BMPRII as well (Yin et al. 2008), as these receptors use the exposed face of the β4-β3-β5 sheet, along with the β4-β5 loop, to bind their cognate GFs. The e disulfide is present in all type I receptors, as well as TβRII (Figs. ​(Figs.55 and ​and6A–D,H).6A–D,H). The e disulfide in type I receptors fulfills an analogous role as the f disulfide in type II receptors, as it precludes the β1-β2 loop from occluding the convex surface of the β4-β3-β5 sheet, which along with the critical β4-β5 loop, binds cognate GFs (Kirsch et al. 2000).

The TGF-β type II receptor, TβRII, is unique among the type I and type II receptors of the family in that it binds its cognate GFs through a distinct interface, namely, an edge β-strand on the smaller β1-β2-sheet, the β2-strand, and the flanking β1 and β1′ strands (Figs. ​(Figs.6H6H and ​and7F)7F) (Hart et al. 2002; Groppe et al. 2008; Radaev et al. 2010). Use of this alternative interface is promoted by two important structural differences in TβRII compared with ActRII, ActRIIB, and BMPRII. First, the β4-β5 loop is extended by seven to eight residues in TβRII relative to other type II receptors and lacks the f disulfide (Figs. ​(Figs.55 and ​and6H).6H). The absence of this disulfide allows the extended β4-β5 loop to fold onto the concave surface of the β4-β3-β5 sheet, thus blocking GF binding through this interface. Second, the β1-β2 loop in TβRII is extended by more than 10 residues compared with all other type I and type II receptors of the family, except AMHRII (Fig. 5). The β1-β2 loop is further stabilized by both the e disulfide at its base, as in type I receptors, and the TβRII-specific g disulfide near its tip (Figs. ​(Figs.55 and ​and6H).6H). The extended loop wraps onto the convex surface of the β4-β3-β5 sheet and packs tightly against it. This unusual structural feature is likely important for proper positioning and bracing of the short β2-strand, which forms the primary structural element used by TβRII to bind the three TGF-β isoforms.

Open in a separate window

Figure 7.

Representative growth factor (GF) complexes with type I and type II receptors. Structures are shown for representative binary (A–C) or ternary (D–F) complexes as ribbon diagrams with pink and light blue GF monomers, and type I and type II receptor ectodomains in orange-red and green, respectively. Orthogonal views are shown above and below one another in each panel. The ActRIIB:activin A structure model with a closed activin GF is made by superimposing open monomer complexes from the ActRIIB:activin A structure (Thompson et al. 2003) on the closed activin dimer from the activin A:Fst288 structure (Thompson et al. 2005). Panel F (inset) shows the interactions of Pro55, Asp57, Arg58, and Pro59 in the ALK5 pre-helix extension with Lys97 on TGF-β3 and Asp118 in TβRII, and of ALK5 with Phe24 in the amino-terminal tail of TβRII. Structures are for ALK6:GDF-5 (Kotzsch et al. 2009), ActRII:BMP-7 (Greenwald et al. 2003), ActRIIB:activin A (Thompson et al. 2003), ActRIIB:ALK3:BMP-2 (Weber et al. 2007), ActRIIB:ALK1:BMP-9 (Townson et al. 2012), and TβRII:ALK5:TGF-β3 (Groppe et al. 2008) complexes. N-term, amino terminal.

The AMHRII type II receptor is unusual in that it includes only one of the two cysteines that form the f disulfide found in other family type II receptors, including ActRII, ActRIIB, and BMPRII (Fig. 5). However, this cysteine is predicted to form a disulfide unique to AMHRII with another cysteine that immediately precedes the carboxy-terminal cysteine of the conserved a disulfide (Fig. 5). This putative h disulfide is predicted by the structure and sequence-based alignment in Figure 5, because aligned residues in homologous type II receptors have Cα atom separations and side chain orientations consistent with disulfide bond formation.

Structures of GF Complexes with Receptors

TGF-β family GF dimers greatly outnumber the type I and type II receptors, and thus, multiple GFs must share the same type I or II receptors, or in some cases, share both type I and type II receptors. Nonetheless, sequence and structure provide a basis for understanding this sharing; particular subfamilies of type I and type II receptors (TGF-β receptors, Box 1) preferentially bind and transduce signals for particular subfamilies of GFs (TGF-β GFs, Box 1). Thus, the closely related type I receptors ALK3 and ALK6 bind and transduce signals for the BMP clade comprising the BMP-5, 6, and 7 subfamily, the BMP-2 and 4 subfamily, or the GDF-5, 6, and 7 subfamily, yet not for the activin or the myostatin and GDF-11 subfamilies in the activin clade or for TGF-β (ten Dijke et al. 1994b). The type I receptors ALK4 and ALK5, which diverge from ALK3 and ALK6 (TGF-β receptors, Box 1), show the opposite GF binding properties; they bind and transduce signals for the activin subfamily and the myostatin and GDF-11 subfamily in the activin clade, and for TGF-β, but not for BMP-5, 6, and 7, BMP-2 and 4, or GDF-5, 6, and 7 in the BMP clade (ten Dijke et al. 1994a). The three type II receptors ActRII, ActRIIB, and BMPRII bind and transduce signals for BMP-5, 6, and 7, BMP-2 and 4, GDF-5, 6, and 7. Furthermore, the receptors ActRII and ActRIIB also interact with activin A and B (receptor binding data for activin C and E is unclear) as well as myostatin, and thus act across clades, yet do not bind or signal for the TGF-βs or AMH (Liu et al. 1995). In contrast, TβRII and AMHRII are highly specific or exclusive for binding and transducing signals for the TGF-βs and AMH, respectively (Baarends et al. 1994; Liu et al. 1995). TβRII does not ally with any other type II receptors in the family tree and lies closest to the type I receptors; however, AMHRII is the most divergent from the consensus type II receptor sequence (TGF-β receptors, Box 1).

Among receptor ectodomain:GF complexes (Table 1), binary type I or II receptor:GF complexes and ternary type I receptor:type II receptor:GF complexes show nearly identical binding positions for type I and type II receptors (Fig. 7). Type II receptors bind the distal ends of the GF dimer, at the knuckle (ActRII and ActRIIB) or fingertips (TβRII) of the GF. Thus, residues within a single monomer determine type II receptor binding specificity. In contrast, type I receptors bind across the dimer interface at the wrist, and thus can show selectivity for open versus closed homodimers or homodimer versus heterodimer (Fig. 7). The GF:receptor structures provide insight into many patterns of receptor:GF binding that have been reported.

Promiscuous binding of many GFs from the BMP and activin clades by ActRII and ActRIIB, for example, is evident from their identical orientations for binding BMP-7, activin A, BMP-2, and BMP-9 dimers (Fig. 7B–E) (Greenwald et al. 2003, 2004; Thompson et al. 2003; Allendorph et al. 2006; Weber et al. 2007; Townson et al. 2012). Consistent with their similar β4-β3-β5 sheet concave surfaces and β4-β5 loop occluding f disulfide (Fig. 6E,F), ActRII and ActRIIB complement the convex knuckle of the GF in a manner that is nearly indistinguishable (Fig. 7B–E). Superpositions of ActRIIB binding to three different GFs and of ActRII and ActRIIB binding to two different GFs further show essentially identical orientations in all cases (Fig. 8E–G). Promiscuity of ActRII and ActRIIB for binding a large number of GFs agrees with the similarity in the receptor structures and knuckle regions of the GFs, which are among their most structurally conserved features. Promiscuity is further promoted by the predominantly hydrophobic composition of the interface (Fig. 9E–G). Among 23 residues in BMP-2 that directly contact ActRIIB in the structure of the BMP-2:ActRIIB:ALK3 complex, substitution with alanine of only six residues perturbed binding (Kirsch et al. 2000; Weber et al. 2007). Among these, only a Leu is conserved among the diverse GFs bound by these receptors (Leu residues 92, 90, and 125 in activin A, BMP-2, and BMP-7, respectively, in Fig. 9E–G). Thus, the similar overall shape and hydrophobicity of the knuckle in GFs that bind ActRII and ActRIIB enables the promiscuity that is possible for predominantly hydrophobic interfaces (Weber et al. 2007).

Open in a separate window

Figure 8.

Relative positioning of type I and type II receptor ectodomains on growth factor (GF) dimers. Positions of type I receptors on GF dimers in complexes are compared by superimposition using GF dimers on the ALK3:BMP-2 complex (A–D); similarly, positions of type II receptors on GF dimers in complexes are compared by superimposition using GF dimers on the ActRIIB:BMP-2 complex (E–H). Receptor names are color-coded as in their ribbon diagrams. GFs used for superposition are shown with black names below names of their bound receptors. GF dimers are shown as surfaces, with monomers shaded pink and light blue. Overlaid type I and type II receptors are in yellow, whereas ALK3 and ActRIIB are in orange–red and green, respectively. Complex structures are for ALK3:GDF-5 (Klammert et al. 2015), ALK6:GDF-5 (Kotzsch et al. 2009), ALK1:ActRIIB:BMP-9 (Townson et al. 2012), ALK5:TβRII:TGF-β3 (Groppe et al. 2008), ALK3: BMP-2 (Kirsch et al. 2000), ActRIIB:ActA (Thompson et al. 2003), ActRII:BMP-7 (Greenwald et al. 2003), and ALK3:ActRIIB:BMP-2 (Weber et al. 2007).

Open in a separate window

Figure 9.

Type I and type II receptor–GF interfaces. Interfaces show growth factors (GFs) bound to type I receptors (A–D) or type II receptors (E–H). Cartoon diagrams show GFs as pink and light blue monomers, and type I and type II receptors in red–orange and green, respectively, with key side chains in stick with carbons in the same colors, red oxygens and blue nitrogens. Structures correspond to those in the Figure 8 legend.

How TβRII specifically binds TGF-β1–3 is evident from the structure of the TGF-β3:TβRII:ALK5 ternary complex (Figs. ​(Figs.6H,6H, ​H,7F,7F, and ​and9H)9H) (Groppe et al. 2008). TβRII binds in the cleft between the fingers of the GF, using primarily the β2, but also the flanking β1 and β1′ strands. Remarkably, the TβRII binding site on the GF does not overlap with those of ActRII and ActRIIB (Fig. 8H). The ability of TβRII to bind a distinct site on the GF is driven by differences in receptor structure that change the overall shape of the binding surface (see section on structures of type I and type II receptor ectodomains); sequence and loop shape differences in the finger region of TGF-βs relative to other GFs enable specificity (Fig. 4B, lower) (De Crescenzo et al. 2006; Baardsnes et al. 2009). The interface that stabilizes TβRII in the cleft between the TGF-β fingertips is unique relative to that used by ActRII and ActRIIB, and is characterized by an inner hydrophobic portion, with Ile53 of TβRII packing into a shallow pocket in TGF-β formed by Trp32, Tyr90, and Val92 (Fig. 9H) (Groppe et al. 2002; Radaev et al. 2010). The inner hydrophobic portion is flanked on either edge by hydrogen-bonded ion pairs between the side chain carboxylate groups of Asp32 and Glu119 on TβRII and the side chain guanidinium groups of Arg25 and Arg94 on the fingertip loops of TGF-β (Fig. 9H). The two arginine residues contribute >30% of the total binding energy and are invariant in TGF-β1 and 3, which bind TβRII with high affinity (De Crescenzo et al. 2006; Baardsnes et al. 2009). The two arginine residues are conservatively replaced by lysine in TGF-β2, which binds TβRII with low affinity, and substitution of these with arginine increases the affinity of TGF-β2 to that of TGF-β1 and 3 (De Crescenzo et al. 2006; Baardsnes et al. 2009). Relative to all other GFs, TGF-βs have a one-residue insertion in the loop bearing Arg25, and differ in the length and sequence of the loop bearing Arg94 (Fig. 1), correlating with the different shapes of fingertip loops (Fig. 4B, lower) and creating high specificity for TβRII.

Type I receptor ectodomain:GF complexes also provide insights into specificity, but the differences are more subtle. Thus, complexes containing ALK1, 3, 5, and 6 with GFs as diverse as GDF-5, BMP-2, BMP-9 and TGF-β3, show that the type I receptors are positioned similarly, although not identically, at the wrists of the GFs (Fig. 7A,D–F) (Kirsch et al. 2000; Groppe et al. 2008; Kotzsch et al. 2009; Townson et al. 2012). The closely related type I receptors ALK3 and ALK6 (TGF-β receptors, Box 1) bind similarly at the wrist and contact each GF monomer roughly equally (Kirsch et al. 2000; Keller et al. 2004; Kotzsch et al. 2009; Klammert et al. 2015). ALK1 and ALK5 belong to subfamilies distinct from that containing ALK3 and ALK6 (Box 1). The structural features that cause ALK1 and ALK5 binding positions to shift relative to the ALK3/6 subfamily (Fig. 8A–D) are discussed in the next two paragraphs.

The type I receptor ALK1 is divergent in sequence from ALK3 and ALK6 (TGF-β receptors, Box 1) and binds and transduces signals for BMP-9 and BMP-10, but not other BMPs and GDFs (David et al. 2007). ALK1 binds both monomers at the wrist, but the directionality of the strands that comprise its central β4-β3-β5 sheet is rotated relative to that of ALK3 and ALK6 (Mahlawat et al. 2012; Townson et al. 2012), as shown by the dashed lines in Figure 7D,E. Repositioning ALK1 relative to ALK3 and ALK6 is primarily driven by a two-residue shortening in the segment between the Cys of the c disulfide at the carboxy-terminal end of β-strand 4 and the Cys of the d disulfide in the 3₁₀ helix of the β4-β5 loop (Figs. ​(Figs.55 and ​and6).6). Shortening causes this segment of ALK1 to be more rigid and shifted closer to β-strand 5 than in ALK3 and ALK6 (Fig. 6A–C) (Mahlawat et al. 2012; Townson et al. 2012). Loop shortening in ALK1, together with conformational changes in the prehelix extension and other loops in the wrist region of BMP-9 (Fig. 4B), leads to steric clashes when ALK1 is positioned onto BMP-9 in the same manner as ALK3 on BMP-2 (Mahlawat et al. 2012; Townson et al. 2012). The structure of the BMP-9:ActRIIB:ALK1 complex shows that these clashes are alleviated when ALK1 binds BMP-9 by a rotation of ~20° relative to ALK3 bound to BMP-2 (Fig. 7D,E) (Townson et al. 2012).

ALK5 (also known as TβRI) is in a type I receptor subfamily distinct from the ALK1 and ALK 3/6 subfamilies (TGF-β receptors, Box 1). Compared with these receptors, ALK5 binds more toward the TGF-β fingertips, which enables it to additionally contact TβRII (Fig. 7F, upper panel) (Groppe et al. 2008). TβRII has a longer amino-terminal segment than other type II receptors (Fig. 5), and a portion of this segment becomes ordered by this contact, with Val22 and Phe24 in TβRII binding to a hydrophobic pocket in ALK5 (Fig. 7F, inset). The repositioning of ALK5 relative to ALK3 and ALK6 is, like ALK1 repositioning, driven by a modification of the segment between the c disulfide cysteine in the carboxy-terminal end of β-strand 4 and the d disulfide cysteine in the 3₁₀ helix within the β4-β5 loop, but in this case with a five-residue extension rather than a two-residue shortening (Figs. ​(Figs.55 and ​and6).6). The PRDRP sequence motif in the prehelix extension, between the c and d disulfide in Alk5, is rigid and forms a tight turn at its amino-terminal cis Pro-55 (Figs. ​(Figs.5,5, ​,6D,6D, and ​and7F,7F, inset) (Zuniga et al. 2011). The extension precludes ALK5 from binding similarly to ALK3 due to clashes with hydrophobic residues in the wrist. These clashes cause ALK5 to reposition so that it shifts toward the fingertips in which it binds in the cleft between TGF-β and TβRII. Repositioning orchestrates a number of interactions with TGF-β and TβRII that are required for cooperative assembly of the TGF-β signaling complex (Fig. 7F, inset) (Groppe et al. 2008; Radaev et al. 2010; Zuniga et al. 2011). In summary, alternative positioning of ALK1 and ALK5 provide distinct interfaces and stabilizing interactions that enable segregation of GF specificities of the ALK1 and ALK5 subfamilies from the ALK3 and ALK6 subfamily.

Further details on amino acid sidechain and structural features that underlie the selectivities displayed by type I and type II receptors, despite promiscuity in binding multiple GFs, are summarized in Table 2. These details extend to differences among ActRII, ActRIIB, and BMPRII, and between ALK3 and ALK6 (Table 2).

Noggin—The BMP Antagonist Paradigm

Noggin, a component of the Spemann organizer (Smith and Harland 1992), antagonizes the ventralizing activity of BMPs in dorsoventral patterning and regulates skeletogenesis (Zimmerman et al. 1996; Brunet et al. 1998). Noggin binds to the same sites on BMP-7 as the BMP type I and type II receptors, and thereby impedes BMP signaling (Fig. 12A–D) (Groppe et al. 2002). Noggin contains a CKGF domain like TGF-β GF domains, but has a 10- rather than 8-membered CK ring (Fig. 12A). Furthermore, its monomers dimerize through a Cys residue in a position (cysteine 11 in Fig. 11G,H) that differs from that in TGF-β family proteins, and through an additional α-helical domain (Fig. 12A). The helices lie on the convex side of noggin, opposite the concave BMP binding site, and support noggin’s C-clamp-like curvature (Fig. 12A,B). Thus, whereas the TGF-β CKGF dimer is linked by monomers that overlap in the palm and heel (GF homodimer, Box 1), and dimerize through cysteine 6, the noggin dimer links the two CKGF domain monomers more distally toward their carboxy-terminal end at cysteine 11 (Fig. 11F,G), leading to a longer and more curved dimer that straddles and clamps the BMP dimer (Fig. 12A,B).

Open in a separate window

Figure 12.

The bone morphogenetic protein (BMP) modulator noggin, members of the Dan family, and sclerostin. (A) Noggin binds BMPs (PDB 1M4U [Groppe et al. 2002]) in a clamp-like manner with the fingers of its cystine knot (CK) growth factor (CKGF) domain (red) blocking type II receptor binding, and a peptide segment termed clip (green) shielding the type I receptor epitopes. Hydrophobic Pro35 of noggin, engaging in the knob-into-hole interaction motif is shown as a gold sphere. The noggin dimer is formed by interactions between the α-helical dimerization domains and stabilized by an intermolecular disulfide bond (asterisk). Disulfides are shown with gold sticks and the CK motif (box) is shown in detail. (B) As in A but viewed down the dyad axis of the noggin and GF dimers. (C and D) A prototypical BMP ligand-receptor complex, with type I and type II receptors as green and red ribbons, respectively, and the GF dimer as gray surface (PDB 2H64 [Weber et al. 2007]) shown in similar orientations as in A and B. (E) Phylogenetic tree of the Dan family (upper panel) and schematic of domain organization (lower panel) with the CKGF domain boxed, its eight disulfides indicated by black lines, and the segments forming three loops colored as in panels F-H. The highly diverse segments amino- and carboxyl termini to the CKGF domain are shown as horizontal lines. (F) Sclerostin NMR structure (PDB 2KD3 [Weidauer et al. 2009]). The cartoon representation (upper) shows the NXI motif (red spheres) and the Cα atom trace (lower) shows the ensemble of 15 structures highlighting the flexibility of loop 2. Loops are colored as in E (lower). (G) PRDC (PDB 4JPH [Nolan et al. 2013]). The cartoon representation (upper, with one monomer colored as in E, lower) shows cysteines as gold sticks, with the unpaired Cys boxed. The lower panel shows PRDC as a surface, with the α-helical amino-terminal segments as blue Cα atom traces. Hydrophobic residues, which on exchange for alanine attenuate binding to BMP binding, are colored red. (H) Two Nbl1/Dan crystal structures (PDB 4X1J [Nolan et al. 2015], upper) and (PDB 4YU8, Table 1, lower). One monomer is colored as in E (lower). The other monomer(s) associates as in PRDC (gray, upper) or in a different manner, with two nearby monomers in the crystal lattice shown in two shades of gray (lower).

Distinct structural elements in noggin block the BMP type I (wrist) and type II (knuckle) receptor epitopes (Fig. 12A–D). A 25-residue segment within the amino-terminal region of noggin is spatially separated from the CKGF and α-helical domains in a complex with BMP, fastens to more than one BMP monomer in a U-shape, and is therefore termed the “clip” (Fig. 12A) (Groppe et al. 2002). In this segment, residues Pro35 to Ser38 snugly fold into the center of the wrist epitope, and block binding to type I receptors. The clip then turns over the fingers of the BMP and runs over the periphery of the knuckle epitope. In contrast, the noggin fingers at the end of its C-clamp extend over the fingertips of the BMP and shield a large part of the knuckle epitope, thereby blocking access and binding of type II receptors. Hydrophobic interactions dominate both the fingertip and clip interfaces, similarly to type I and type II receptor interfaces with BMPs. Moreover, Pro35 in the noggin clip, which is highly conserved among different species, extends into a hydrophobic pocket in the wrist epitope of BMP-7, mimicking the knob-into-hole interaction found in various BMP-type I receptor interactions (Fig. 12A–D) (Hatta et al. 2000; Kirsch et al. 2000; Groppe et al. 2002; Kotzsch et al. 2009). The importance of Pro35 for the BMP-neutralizing activity of noggin is highlighted by its mutation in skeletal malformation diseases (Gong et al. 1999; Mangino et al. 2002).

The mechanism of noggin binding to BMPs preserves much of the chemistry of receptor binding to BMPs, including shared hydrophobic interfaces and the knob-into-hole packing seen with most type I receptors; however, details of the distribution and arrangement of the structural elements differ. Remarkably, the noggin–BMP complex partially shares its building principle with the TGF-β1 proprotein complex; the folded noggin CKGF domain, like the prodomain arm domain, blocks the type II receptor site, and the single element clip in noggin, like the prodomain straitjacket, blocks type I receptor access to the GF (Fig. 12A; Box 1). In both noggin and the TGF-β1 prodomain, a central domain together with intermolecular disulfide bonds forms a covalent homodimer, which explains the extremely tight binding to the GF. Moreover, both interact with the ECM; the TGF-β1 prodomain is covalently linked to LTBP, and noggin has heparin binding sites to interact with the ECM (Groppe et al. 2002; Paine-Saunders et al. 2002) that can locally retain a bound GF to limit its radius of action.

Noggin binds in vitro to most BMPs with no apparent dissociation, which suggests that the concentration of active ligand would be effectively decreased for all BMPs that it targets (Zimmerman et al. 1996; Seemann et al. 2009; Song et al. 2010; Schwaerzer et al. 2012). However, in cell-based assays some BMPs, for example, BMP-6, have been described as noggin-resistant, although their affinity to noggin is in the same range as for nonresistant ligands (Song et al. 2010). Furthermore, mutations that render BMPs noggin-resistant only marginally decrease their affinity to noggin (Song et al. 2010; Schwaerzer et al. 2012), possibly indicating that the inhibition mechanism is more complex in vivo.

The Dan Family—A Heterogeneous Group of Antagonists for BMPs and/or Wnt Factors

With seven known members in mammals, Dan modulators are the largest family of TGF-β family antagonists (Nolan and Thompson 2014). They include neuroblastoma suppressor of tumorigenicity 1 (Nbl1, Dan, DAND1), gremlin/Drm (DAND2), protein related to Dan and cerberus (PRDC, DAND3, gremlin2), cerberus (DAND4), and coco/dante (DAND5) (Fig. 12E). These five members are the smallest proteins known to modulate BMP signaling. Sclerostin and USAG1 (also known as wise, ectodin, and SOSD) also belong to the Dan modulator family (BMP antagonists, Box 1, Figs. ​Figs.1111 and ​and12E),12E), but they antagonize Wnt and not BMP signaling. Dan family members share a central CKGF domain of 90 residues that may be substantially elaborated with amino- and carboxy-terminal extensions that add another ~100 residues (Fig. 11A–C, 12E, lower). Extensions include α-helices in PRDC (Nolan et al. 2013) and unstructured regions in sclerostin (Veverka et al. 2009; Weidauer et al. 2009). These might become structured when bound to GFs or ECM components and might resemble in this respect the long amino-terminal extensions in many TGF-β family prodomains (Fig. 1).

All Dan and sclerostin family proteins contain an 8-membered CK ring with the same sequence motif as in TGF-β family CKGF domains (C-X-G-X-C and C-X-C) (Figs. ​(Figs.1111 and ​and12E).12E). Besides the six CK cysteines, Dan proteins contain two additional cysteines (designated C2 and C8 in Fig. 11) that disulfide-link the two β-ribbon fingers at the fingertips (Figs. ​(Figs.1111 and ​and12F,G).12F,G). The same eight Cys residues are shared with carboxy-terminal CK (CTCK) domains, which are closely related to CKGF domains. CTCK domains have been structurally characterized in von Willebrand factor (Zhou and Springer 2014) and norrin (Ashe and Levine 1999; Chang et al. 2015). Three long loops in CKGF and CTCK domains link the CK residues (GF monomer, Box 1, Fig. 11). The Dan family (Nolan et al. 2013; Nolan et al. 2015), sclerostin (Veverka et al. 2009; Weidauer et al. 2009), and CTCK domains have similarly curved, long β-ribbon fingers (loops 1 and 3) stabilized by the fingertip C2-C8 disulfide (Figs. ​(Figs.11A–C,E11A–C,E and ​and12F–H).12F–H). Human chorionic gonadotrophin (HCG) in the hormone CKGF family and noggin have similarly long, disulfide-stabilized fingers, but the disulfides are between nonequivalent cysteines (Fig. 11D,G,H). Loop 2 of TGF-β family GF domains forms the heel α-helix; in contrast, loop 2 of PRDC, Nbl1, and CTCK domains extends far from the CK, forms a β-ribbon in some members, and continues the curvature of loops 1 and 3 (Figs. 11 and ​and1212G,H).

Different loop 2 structures correlate with the amount of overlap between monomers when they dimerize. TGF-β family CKGF domains dimerize through the cysteine in position C6 (C6-C6′ disulfide), whereas CTCK domains dimerize through reciprocal C5-C6′ and C6-C5′ disulfides and an additional C11-C11′ disulfide (Fig. 11). Because C5 and C6 are two residues apart in a C5-X-C6-C7 sequence that includes C7 of the CK, the dimer interface slides two residues in CTCK dimers so that their knots are two β-ribbon positions further apart than in TGF-β GF dimers. The dimer interface in the Dan family proteins PRDC and Nbl1 and in HCG slides similarly to that in CTCK domains, correlating with their similarly shaped loop 2 (Fig. 11). However, these Dan proteins and HCG form dimers that are not disulfide-linked. Furthermore, although PRDC and Nbl1 contain one or two Cys residues equivalent in sequence and structure to those in positions C5 and C11 in CTCK domains, they do not form equivalent disulfides. PRDC has a C5 cysteine that curiously has no disulfide bond partner, whereas in Nbl1 C5 and C11 form an intramolecular disulfide that contrasts with intermolecular disulfides in CTCK domains (Fig. 11A,B,E).

In contrast to Dan proteins, which are shown or considered to be dimers (Stanley et al. 1998; Pearce et al. 1999; Kattamuri et al. 2012), sclerostin and USAG1 are both monomeric (Kusu et al. 2003; Winkler et al. 2003; Lintern et al. 2009). In the absence of a dimer interface, loop 2 of sclerostin is not stabilized by van der Waals interactions and is highly flexible and disordered (Fig. 12F). Although sclerostin was initially described as a BMP antagonist (Kusu et al. 2003; Winkler et al. 2003), it functions as an antagonist of the canonical Wnt pathway and binds and blocks the Wnt coreceptor LRP5/6 (van Bezooijen et al. 2004, 2007; Krause et al. 2010; Boschert et al. 2013). Similarly, the close sclerostin relative USAG1 inhibits Wnt signaling (Itasaki et al. 2003). Structure–function studies show that the NXI motif, present in loop 2 of sclerostin and USAG1 but not the Dan clade (Figs. ​(Figs.11H11H and ​and12F),12F), binds to the first β-propeller domain in the extracellular domain of LRP6 and thereby competes with binding of Wnt GFs (Bourhis et al. 2011; Holdsworth et al. 2012; Boschert et al. 2013).

In contrast to the monomeric Wnt antagonists sclerostin and USAG1, the CTCK family member norrin is a dimeric Wnt agonist. Norrin stimulates canonical Wnt signaling by binding frizzled4, LRP6, and glycosaminoglycans (GAGs). A synthetic heparin glycan mimetic binds to both Frizzled and the norrin fingertips, providing a model for how ECM components may contribute to the formation of signaling complexes (Fig. 11E) (Ke et al. 2013; Chang et al. 2015). As described in the section on the brief history of TGF-β family evolution, the TGF-β and Wnt signaling pathways coevolved in primitive metazoans. Remarkably, CKGF/CTCK domains evolved to serve as both agonists and antagonists in each of these pathways. The ECM has an important organizing role in both families. Binding sites for GAGs are at the amino termini of some GFs (Ruppert et al. 1996), in the prodomains of activin A and myostatin (Li et al. 2010; Sengle et al. 2011), in the dimerization domain in noggin (Groppe et al. 2002), and along one side of the extended sclerostin molecule harboring residues from all three loops (Veverka et al. 2009). Therefore, such sites must have evolved independently multiple times.

Although all dimeric Dan proteins are effective BMP antagonists, coco and cerberus also block nodal and Wnt signaling (Glinka et al. 1997; Piccolo et al. 1999; Bell et al. 2003). Cerberus binds and inhibits Wnt, nodal and BMPs at mutually noncompetitive sites. Furthermore, the amino-terminal portion of cerberus is required for BMP binding and Wnt antagonism whereas the carboxy-terminal portion including the CKGF domain is sufficient for nodal antagonism (Piccolo et al. 1999).

The prototypical Dan protein PRDC forms a homodimer that is not disulfide-linked, despite its odd number of Cys residues (Fig. 12G) (Nolan et al. 2013). Dimerization of PRDC is between extremely long β-strands in each monomer that begin in loop 1 and extend through the CK into loop 2 (Figs. ​(Figs.1111 and ​and12G).12G). These join with antiparallel β-strands in the other halves of loops 1 and 2 to form an extensive “intermolecular” four-stranded β-sheet, and build a large dimer interface in common with CTCK domains and HCG (Fig. 12G) (Ke et al. 2013; Zhou and Springer 2014; Chang et al. 2015). Additionally, fingers 1 and 2 form two β-ribbons that extend from the central four-stranded β-sheet, reminiscent of the wrist in TGF-β GFs (Nolan et al. 2013). However, instead of an overall butterfly-shaped architecture like TGF-β dimers, the PRDC monomers (Fig. 11A) assemble to form a longer arc-like structure (Fig. 12G).

The structure of the CKGF domain of Nbl1 is very similar to that of PRDC, despite only 35% sequence identity (Figs. ​(Figs.11A,B11A,B and ​and12G,H)12G,H) (Nolan et al. 2015). Like PRDC, Nbl1 fails to form disulfide-linked dimers, with the C5 and C11 cysteines forming an intramolecular disulfide (Fig. 11B). This is remarkable because the C5 Cys in one Nbl1 monomer is positioned very close to the C11 Cys of the other monomer in the dimer, and thus only slight structural adjustment would be required to form intermolecular disulfides similar to those in the CTCK domains of VWF and norrin (Fig. 11E) (Ke et al. 2013; Zhou and Springer 2014; Chang et al. 2015). Dimerization in all of these structures is over essentially identical interfaces, and gives the dimers an arc-like shape, although the Nbl1 dimer is more curved and shows an S-like shape compared with PRDC (Fig. 12G,H, upper) (Nolan et al. 2015). Both PRDC and Nbl1 form noncovalent dimers in solution (Kattamuri et al. 2012). However, another structure of Nbl1 (PDB entry 4YU8, Table 1) fails to show the same dimerization interface in crystal lattices (Fig. 12H, lower).

Without structural data on a Dan protein bound to BMP or nodal, it is unclear whether Dan proteins inhibit GF signaling by impeding binding of both receptors like noggin, or of only one receptor type, as shown for Tsg (Zhang et al. 2007). Furthermore, the architecture of PRDC and Nbl1 differs from other structurally characterized BMP modulators and lacks C-clamp- or clip-like structures. However, a contiguous patch of hydrophobic residues on the convex surface of PRDC is required for binding and inhibition of BMP signaling in vitro (Fig. 12G, lower) (Nolan et al. 2013). Several of these residues are conserved in Nbl1, which binds BMP-2 and 7 and GDF-5 with different affinities compared with PRDC. The importance of these residues in encoding BMP selectivity was shown by exchange of patch residues from PRDC into Nbl1, which yielded an Nbl1 variant mimicking PRDC (Nolan et al. 2015). Interestingly, in PRDC the hydrophobic patch responsible for BMP binding seems to be covered by a dynamic α-helix in the amino-terminal segment (Fig. 12G, lower). A similar helix is not observed in the Nbl1 structure (Fig. 12H) (Nolan et al. 2015).

Follistatin Antagonizes Activins as Well as BMPs

Follistatin (Fst) was initially described as a protein that antagonizes activin signaling. However, follistatin also binds BMPs and GDFs, including BMP-2, 4, 6, and 7 as well as myostatin/GDF-11, albeit with lower affinities (Iemura et al. 1998; Amthor et al. 2002, 2004; Glister et al. 2004; Schneyer et al. 2008). Fst is a modular protein comprising a unique 65-residue amino-terminal domain (Fs0) followed by three follistatin domains (Fs1, Fs2, Fs3) of 70 to 85 residues each (Fig. 13A,B). Follistatin domains are in turn each composed of EGF-like and Kazal subdomains (Hohenester et al. 1997; Schneyer et al. 2004).

Open in a separate window

Figure 13.

Modulators of the follistatin and chordin families. (A,B) Two follistatin-288 molecules shown in ribbon cartoon wrap around myostatin shown as a surface (PDB 3HH2 [Cash et al. 2009]). Fs0 (green) blocks access to the type I receptor epitope (light green) with a knob-into-hole interaction indicated by a gold sphere; Fs1 (red) and Fs2 (magenta) block the type II receptor-binding site (pink). (C–E) Open and closed conformations of activin A. (C,D) Shows activin monomers as light and dark gray surfaces in different orientations (dashed lines) in an open complex with an Fs1-2 fragment of follistatin (C, PDB 2ARP [Harrington et al. 2006]) and in a closed conformation with FSTL3 (D, PDB 3B4V [Stamler et al. 2008]). (E) Compares monomer–monomer orientations in activin A and myostatin dimers by superimposing on one monomer (gray) and showing the orientation of the other monomer (colored and numbered). The colored and numbered monomers are from dimeric structures of myostatin bound to follistatin-288 (red, 1), activin A bound to FSTL3 (cyan, 2) and activin-A bound to the follistatin Fs1-2 fragment (blue, 3). (F) Chordin family members are mosaics of tandem domains. Although von Willebrand factor C (VWC) domain(s) mediate bone morphogenetic protein (BMP) binding, other domains present include chordin (CHRD), von Willebrand type D (VWD), trypsin inhibitor-like (TIL), insulin-like growth factor binding protein (IGFBP), thrombospondin (TSP) and cystine knot growth factor (CKGF). (G) VWC domain of collagen IIA (left) and VWC domain 1 of CV2 (right). Amino-terminal peptide segments and VWC subdomains 1 and 2 are green, red and blue, respectively. A structurally variable segment of VWC subdomain 1 is shown in yellow. (H) The VWC1 domain of CV2 shown in ribbon with knob-into-hole residue shown as gold sphere bound to BMP-2 shown as surface with type I and type II receptor sites colored green and pink, respectively (PDB 3BK3 [Zhang et al. 2008]). (I) Similarity in architecture of VWC1 domain of CV2 bound to BMP-2 in crystals (left) and free in solution (right panel, NMR ensemble of 10 structures, PDB 2MBK [Fiebig et al. 2013]).

Fs domains together with other domains are present in follistatin-like (FSTL) proteins. Although FSTL-1 and 3 are known to regulate activin and BMP signaling, FSTL-4, FSTL-5, SMOC1, and SMOC2 are not yet characterized for effects on activin and BMP signaling. Furthermore, SMOC1 and SMOC2 contain additional domains characteristic of the SPARC family (Schneyer et al. 2004; Bradshaw 2012). The follistatin family thus includes modulators with similar mechanisms of action (binding to GFs and blocking receptor binding) and similar specificities for activins and BMPs, yet its members are built from several types of domains that are assembled into distinctive architectures. Three isoforms of follistatin, Fst288, Fst303, and Fst315 have identical amino-terminal regions, but end at carboxy-terminal residues 288, 303, or 315, respectively, thus creating further complexity. They bind activin with different affinities, and longer isoforms partially lack the ability to bind to heparin and proteoglycans (Kumar 2005).

Structures of follistatin and FSTL-3 bound to activin A or myostatin provide insights into inhibitory mechanisms and specificity (Thompson et al. 2005; Harrington et al. 2006; Cash et al. 2009, 2012). Although follistatin and FSTL-3 block type I and type II receptor binding sites, they bind the GF domain vastly differently from noggin (Figs. ​(Figs.12A12A and ​and13A).13A). Follistatin wraps around activin on the perimeter of the dyad (two-fold rotation axis) of the GF dimer (Fig. 13B). In contrast, noggin binds on one side of the dyad axis. Viewed normal to this axis (as in Fig. 12A), noggin resembles a C-shaped clamp with the ends of the clamp gripping the GF fingers. In the case of follistatin, two modulator molecules embrace the GF such that almost its entire surface seems covered (Fig. 13A,B). In contrast to noggin, follistatin is a monomer when not bound to the GF. However, on binding, two follistatin monomers (structures of follistatin 288 are shown) tightly interact in a head-to-tail fashion and form a stable, noncovalent dimer (Thompson et al. 2005). The type I receptor epitope of activin is blocked by the unique amino-terminal domain (Fs0) of follistatin (Fig. 13A). Here, the key element is Phe52 located in the second α-helix of the Fs0 domain, which makes a knob-into-hole interaction with the wrist epitope of activin A similarly to type I receptors that bind BMPs (Hatta et al. 2000; Keller et al. 2004). Fs1, and primarily Fs2, shield the type II receptor epitope by mainly using hydrophobic interactions as found in complexes of activins and BMPs bound to their type II receptors (Fig. 13A,B). The final domain, Fs3, wraps around the fingertips of activin A and engages in predominantly polar contacts with the amino-terminal Fs0 domain of the other follistatin monomer, thus stabilizing encirclement of the ligand (Fig. 13A,B).

The selectivity of follistatin for activins and BMPs is explained by a central core of GF amino acids in the binding sites for the Fs0 and Fs1/2 domains that are conserved in activins and BMPs but not in TGF-βs (Thompson et al. 2005). The contribution of individual follistatin domains to binding differs depending on the GF (Cash et al. 2009). For example, when bound to myostatin compared with activin A, the Fs0 domain shifts more into the GF wrist epitope to better interact. To avoid a clash with Fs0, Fs3 in the other monomer shifts away from myostatin. Additionally, the ability of activin A to show both closed and open conformations (see section on GF structures) is reflected in differences in monomer–monomer orientation in complexes with different follistatin fragments (Fig. 13C–E) (Thompson et al. 2003, 2005; Greenwald et al. 2004; Harrington et al. 2006).

The Fs0 domain in follistatin and FSTLs may play a role in discriminating between high-affinity ligands (i.e., activins, myostatin, and GDF-11), and low-affinity binders i.e., BMP-2, 4, 7, and others). The type I receptor ALK3 still interacts with BMP-2 when the latter is bound to follistatin, suggesting that the Fs0 domain does not bind to or cannot block the wrist epitope in BMPs (Iemura et al. 1998). Lacking this interaction might also explain the lower affinities of follistatin binding to BMPs and indicate that the inhibition of BMPs by follistatin is functionally different from that of activins (see also Stamler et al. 2008). A role for the Fs3 domain in BMP binding is suggested by the finding that FSTL-3, which lacks Fs3 and thus cannot fully encircle activin A and myostatin (Fig. 13D) (Stamler et al. 2008; Cash et al. 2012), binds activin A and myostatin with high affinity, but cannot bind BMPs (Sidis et al. 2006).

T.A.S. is deeply grateful to Dr. Li-Zhi Mi for help with Figures 1 to ​to33 and is supported by National Institutes of Health (NIH) Grants AR067288 and CA210920. A.P.H is supported by NIH grants GM58670 and CA054174. T.D.M. is supported by Deutsche Forschungsgemeinschaft (DFG) grants MU1095/3-2 and MU1095/5-1.