Ternary Complex of Transforming Growth Factor-β1 Reveals Isoform-specific Ligand Recognition and Receptor Recruitment in the Superfamily*

Transforming growth factor (TGF)-β1, -β2, and -β3 are 25-kDa homodimeric polypeptides that play crucial nonoverlapping roles in embryogenesis, tissue development, carcinogenesis, and immune regulation. Here we report the 3.0-Å resolution crystal structure of the ternary complex between human TGF-β1 and the extracellular domains of its type I and type II receptors, TβRI and TβRII. The TGF-β1 ternary complex structure is similar to previously reported TGF-β3 complex except with a 10° rotation in TβRI docking orientation. Quantitative binding studies showed distinct kinetics between the receptors and the isoforms of TGF-β. TβRI showed significant binding to TGF-β2 and TGF-β3 but not TGF-β1, and the binding to all three isoforms of TGF-β was enhanced considerably in the presence of TβRII. The preference of TGF-β2 to TβRI suggests a variation in its receptor recruitment in vivo. Although TGF-β1 and TGF-β3 bind and assemble their ternary complexes in a similar manner, their structural differences together with differences in the affinities and kinetics of their receptor binding may underlie their unique biological activities. Structural comparisons revealed that the receptor-ligand pairing in the TGF-β superfamily is dictated by unique insertions, deletions, and disulfide bonds rather than amino acid conservation at the interface. The binding mode of TβRII on TGF-β is unique to TGF-βs, whereas that of type II receptor for bone morphogenetic protein on bone morphogenetic protein appears common to all other cytokines in the superfamily. Further, extensive hydrogen bonds and salt bridges are present at the high affinity cytokine-receptor interfaces, whereas hydrophobic interactions dominate the low affinity receptor-ligand interfaces.

mal development, immune function, and carcinogenesis (1)(2)(3). TGF-␤s are the founding members of a highly diversified family of signaling ligands and receptors, known as the TGF-␤ superfamily. To date the superfamily consists of more than 30 growth factors and cytokines, including TGF-␤s, bone morphogenetic proteins (BMPs), activins, inhibins, nodal, Müllerian inhibiting substance, growth differentiation factors, and others (4). TGF-␤s and related factors signal through two single-pass transmembrane receptors, known as the type I and type II receptors. These two receptor types have the same overall domain structure, including an extracellular ligand-binding domain displaying a three-finger toxin fold, a single transmembrane helix, and a cytosolic serine-threonine kinase domain. Signaling is initiated by the ligand, which binds the receptor extracellular domains, bringing them together and triggering a phosphorylation cascade, whereby the type II phosphorylates the type I, and the type I phosphorylates Smads, the cytoplasmic effectors of the pathway (3).
Specificities have been determined based on cell-based affinity labeling studies with radiolabeled ligands and have enabled the identification of major ligands for most receptors of the superfamily, including those specific for BMPs, TGF-␤s, activins, and Müllerian inhibiting substance (4). Structural studies of the BMP and TGF-␤ receptor extracellular domains complexed to their cognate ligands have revealed that although ligands and receptors of the different subfamilies share the same overall fold, they nevertheless bind and assemble their receptors in ways that are entirely distinct (5)(6)(7)(8)(9). The distinct mode of ternary complex assembly for BMP-2 and TGF-␤ underscores the complexity governing the ternary complex assembly. That also raises the question about which mode of type II receptor binding is realized for other cytokines in the superfamily and what are the critical factors determining receptor specificity and promiscuity. In addition, the cytokines outnumber their receptors in the family with at least 29 ligands in mammals signaling through seven type I and six receptors (3, 10 -12), raising the question of combinatorial ligand recognition in the superfamily. Further, little is known regarding the underlying mechanisms by which ligands of particular subfamilies induce their specific activities. Although the three TGF-␤ isoforms, TGF-␤1, -␤2, and -␤3, signal through the same receptors and share significant sequence (71-79% identity) and structural similarity (backbone root mean square deviations (r.m.s.d.) Ͻ 1.5 Å) (13)(14)(15)(16), they nevertheless carry out unique functions in vivo as shown by the severe yet distinct phenotypes of the isoform-specific TGF-␤ null mice (17)(18)(19)(20). These differences have been attributed to distinct patterns of expression (17)(18)(19)21), yet some evidence suggests that it might also be due to differences in the ligands themselves. For example, the addition of purified exogenous TGF-␤s has been shown to lead to different outcomes in a bilateral palatal shelf closing assay, with TGF-␤3 promoting complete fusion and TGF-␤1 and -␤2 promoting only partial fusion (22). The application of purified TGF-␤1 and -␤3 has been further shown to lead to dramatic differences in cutaneous wound healing, with TGF-␤3 preventing and TGF-␤1 promoting scarring (23). The objective of this study was to investigate the mechanism by which TGF-␤s bind and assemble T␤RI and T␤RII into a signaling complex, to define the structural principles that underlie TGF-␤ isoformspecific function through comparison with the TGF-␤3 ternary complex, and to define the structural principles governing the combinatorial ligand recognition among TGF-␤ superfamily receptors.
Crystallization and Structure Determination-TGF-␤1 was first mixed with T␤RII and then with T␤RI at a molar ratio of ϳ1:4:4. The ternary complex was separated from the excess of T␤RI and T␤RII by size exclusion chromatography in 50 mM NaCl, 20 mM Tris-HCl, pH 8.0. Crystals of the TGF-␤1⅐T␤RI⅐T␤RII ternary complex were obtained by vapor diffusion in hanging drops at room temperature in 8 -15% polyethylene glycol 4000 -8000 at pH 6.0 -8.0. The complex crystals diffracted to 3.0 Å resolution, belonged to a triclinic space group P1 with cell dimensions a ϭ 37.70 Å, b ϭ 99.35 Å, c ϭ 102.7 Å, ␣ ϭ 64.01°, ␤ ϭ 84.47°, and ␥ ϭ 84.34°. The x-ray data sets were collected at the Southeast Regional Collaborative Access Team 22-ID Beamline at the Advanced Photon Source, Argonne National Laboratory. Supporting institutions are listed at the Southeast Regional Collaborative Access Team website. The data were processed and scaled with HKL2000 (43).
The structure was solved by the molecular replacement method using the structures of TGF-␤2 (Protein Data Bank code 2TGI), the extracellular domains of T␤RII (Protein Data Bank code 1M9Z), and BMPR-Ia (Protein Data Bank code 1REW) as search models. The solutions for TGF-␤1 and T␤RII were obtained with the program Phaser, and the solution for T␤RI was obtained using Evolutionary programming for molecular replacement (44,45). There are two ternary complexes in each asymmetric unit that were modeled using the programs O and COOT (46,47). The structure was refined using a maximum likelihood target function of CNS v1.1 (48) with a 2-fold noncrystallographic (NCS) constraint between the two ternary complexes. Final rounds of the refinement were carried out with phenix.refine (49) without NCS constraints. Several programs from CCP4 program suite were used throughout model building and refinement (50).
Surface Plasmon Resonance-Binding studies were performed with a BIAcore 3000 instrument (GE Healthcare) and were analyzed using the software package Scrubber2 (Biologic Software). TGF-␤s were biotinylated and captured on carboxymethyl dextran (CM5) chips to which 5000 response units (RU) streptavidin had been covalently attached to all four flow cells using an amine coupling kit (GE Healthcare). TGF-␤2 was biotinylated in 25 mM MES, pH 4.8, by first activating it with a 10-fold molar excess of 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (GE Healthcare) in the presence of a 25-fold molar excess of N-hydroxysuccinimide (GE Healthcare) and then by adding a 100-fold molar excess of amine-PEO 3 -biotin (Pierce). TGF-␤1 and -␤3 were biotinylated by first complexing the protein with an excess of T␤RII (4 equivalents) or T␤RI and T␤RII (4 equivalents each) followed by the addition of a 10-fold molar excess of sulfo-N-hydroxysuccinimide-LC-LC-Biotin (Pierce). Singly biotinylated TGF-␤1 and -␤3 were separated from doubly and multiply biotinylated forms by applying them to a Source S cation exchange column (GE Healthcare) in the presence of 30% isopropanol at pH 4.0. To ensure the reliability of the results, surface densities of captured TGF-␤s were kept at 50 -300 RU. T␤RII binding data for TGF-␤1 and -␤3 were collected using ligands that had been modified in the presence of T␤RII, whereas T␤RI binding data (both alone and in the presence of 4 M T␤RII) were collected using ligands that had been modified in the presence of both T␤RI and T␤RII.
Binding assays were performed by injecting 2-fold serial dilutions of the receptor in duplicate or triplicate in HBS-EP buffer (GE Healthcare) at a flow rate of either 5 l/min (equilibrium experiments for interactions with slow association times) or 50 -100 l/min (equilibrium experiments for interactions with fast association times and kinetic experiments). The surfaces were regenerated by a brief injection of 10 mM glycine, pH 2.5 (30-s contact time at a flow rate of 50 -100 l/min). Instrument noise was removed by referencing the data against at least three blank buffer injections. A very small background signal, caused by the nonspecific absorption of the receptors to the surfaces, was removed by referencing the data against a flow cell containing only immobilized streptavidin. Equilibrium analyses were performed on steady state measurements using the equilibrium binding response near the end of the injection. Kinetic analyses were performed by global fitting with a simple 1:1 model. Standard errors were obtained from the variation among the derived parameters from independent measurements using Origin Software.

RESULTS AND DISCUSSION
Structure of TGF-␤1⅐T␤RI⅐T␤RII Ternary Complex-The structure of a human TGF-␤1⅐T␤RI⅐T␤RII ternary complex was determined to 3.0-Å resolution (Table 1). One TGF-␤1 homodimer binds to two T␤RI and two T␤RII receptors forming a 2:2:2 complex (Fig. 1). Each asymmetric unit contains two complexes related by a NCS 2-fold symmetry. Electron densi-ties for all 112 residues in both chains of TGF-␤1 homodimer, for residues 20 -126 of T␤RII, and for residues 9 -85 of T␤RI were well defined except for two loop regions of T␤RI between residues 36 and 38 and between residues 64 and 71 that correspond to the tips of finger 2 and finger 3 of the three-finger toxin fold. All interface residues between the three components of the complex were well resolved. The overall assembly of the ternary complex is very similar to the reported structure of TGF-␤3⅐T␤RI⅐T␤RII complex (Protein Data Bank code 2PJY) with an r.m.s.d. of 1.14 Å for 357 C␣ atoms (supplemental Fig. S1) (7). In the center of the complex, butterfly-like shaped TGF-␤1 homodimer (TGF-␤1 A and TGF-␤1 B subunits) interacts with two T␤RII and two T␤RI receptors (Fig. 1). The structure of the TGF-␤ monomer was originally described as a slightly curved left hand with ␣1 and ␣3 helixes forming the thumb and the heel of the hand and two antiparallel ␤-sheets forming its four fingers (13). Both receptors T␤RI and T␤RII demonstrate three-finger toxin fold with some differences in the shape, length, and secondary structure of their first and third fingers. The ternary complex defines three pairwise interaction interfaces between TGF-␤1 and T␤RI, TGF-␤1 and   T␤RII, and between T␤RI and T␤RII, burying 1518-, 940-, and 740-Å 2 solvent-accessible areas, respectively. There are no significant conformational changes in TGF-␤ and T␤RII upon complex formation. Superposition of the current receptor with the unbound T␤RII (Protein Data Bank code 1M9Z) resulted in an r.m.s.d. of 0.5 Å for 100 C␣ atoms. Similarly, the receptorbound TGF-␤1 can be superimposed with unbound TGF-␤1 (minimized average NMR structure, Protein Data Bank code 1KLC), TGF-␤2 (Protein Data Bank code 2TGI), and TGF-␤3 (Protein Data Bank codes 1TGK and 1TGJ), resulting in an r.m.s.d. of 0.9 -1.5 Å for 100 -109 C␣ atoms.
Recognition of TGF-␤1 by the Type I Receptor-The type I receptor contacts both monomers of TGF-␤1, generating two primarily hydrophobic patches of the interface burying 370 and 1150 Å 2 of accessible-solvent area with TGF-␤1 A and TGF-␤1 B , respectively ( Fig. 2 and supplemental Table S1). The interface between TGF-␤1 A and T␤RI consists of Trp 30 , Trp 32 , Tyr 90 , and Leu 101 of the "palm" side of TGF-␤1 A fingers and Ile 54 , Pro 55 , and Phe 60 from T␤RI. This interface is well conserved in the structure of TGF-␤3 ternary complex as well as among the sequences of three TGF-␤ isoforms (Figs. 3A and 4A). The second patch consists of Ala 1 , Leu 2 , Asn 5 , and Tyr 6 from ␣1-helix and Ile 51 , Gln 57 , and Lys 60 from ␣3-helix of TGF-␤1 B contacting His 15 , Leu 16 , Lys 19 , Phe 31 , Ile 54 , and Val 61 from T␤RI with one hydrogen bond between the side chain of Tyr 6 of TGF-␤1 and His 15 of T␤RI. Interestingly, in the TGF-␤3 ternary complex structure, T␤RI is rotated ϳ10°away from TGF-␤ as calculated by program HINGE (Fig. 1C and supplemental Fig. S1B) (24), resulting in a partial solvent expo-sure and a 400-Å 2 reduction of buried solvent-accessible area at this interface between T␤RI and TGF-␤3 B . There is, however, one hydrophobic interaction between Thr 67 of TGF-␤3 B and Val 71 of T␤RI, which is absent in current TGF-␤1 complex because of a partial disorder of ␤4-␤5 loop around Val 71 of T␤RI.
T␤RI and BMPR-Ia dock to their ligands at a closely related site (Site I) with a 10 -30°difference in their receptor orientations (Fig. 5A). However, both the identity and position of interface residues vary considerably between the type I receptors (Fig. 4), suggesting a promiscuous recognition at Site I. Nevertheless, no cross-reactivity was observed between BMPR-Ia and T␤RI (1,25). Previously (7), a prehelix loop region (between the ␤4-␤5 strands) was suggested to be partially responsible for T␤RI specificity. The current structure points to the new area of interactions between the N-terminal loop of T␤RI (␤1-␤2 loop) and the ␣1-helix of TGF-␤1. The corresponding loop in BMPR-Ia contains insertions that potentially would hinder BMPR-Ia binding to the ␣1-helix of TGF-␤ (Fig. 5A). This ␣1-helix is stabilized by a Cys 7 -Cys 16 disulfide bond in TGF-␤s but is either absent or disordered in the structures of all BMPs, activin, and growth differentiation factor 5 (5-6, 26 -29). Interestingly, ALK1 and ALK2 also have a short ␤1-␤2 loop that correlates with their permissive recognition of both TGF-␤s and BMPs.
Recognition of TGF-␤1 by the Type II Receptor-The interactions between TGF-␤1 and T␤RII involve five TGF-␤1 residues (Arg 25 , His 34 , Tyr 91 , Gly 93 , and Arg 94 ) at the tips of its fingers and seven T␤RII residues (Phe 30 , Asp 32 , Ser 49 , Ile 50 , Ser 52 , Ile 53 , and Glu 119 ) on the base of the toxin-fold fingers of the receptor ( Fig. 2 and supplemental Table S1). Although Arg 94 and Tyr 91 of TGF-␤1 form hydrophobic contacts with Ile 50 , Ile 53 , and Phe 30 of T␤RII, the majority of the interactions at this interface are either salt bridges or hydrogen bonds ( Fig. 2 and supplemental Table S1). In particular, Arg 25 and Arg 94 of TGF-␤1 form salt bridges with Glu 119 and Asp 32 of T␤RII, respectively, that are also conserved in T␤RII⅐TGF-␤3 binary and T␤RI⅐TBRII⅐TGF-␤3 ternary complexes (Fig. 3C).
A major structural difference between the TGF-␤ and BMP receptor complexes is the association with their respective type II receptors. T␤RII and ActRII not only bind to different sites on their ligands, Site IIa and IIb, respectively, they also use distinct residues for their ligand recognition (Figs. 4 and 5B). T␤RII contacts TGF-␤ via residues from ␤1and ␤2-strands. The secondary structure conformation of this receptor region is stabilized by a disulfide bond between Cys 38 and Cys 44 that is FIGURE 3. Structural superposition between TGF-␤1 and TGF-␤3 ternary complexes. A and B, interface contacts between T␤RI and TGF-␤1 A (A) and between T␤RI and the helix ␣1 of TGF-␤1 B (B). TGF-␤1 complex is colored in light and dark blue for TGF-␤1 A and TGF-␤1 B monomers and red for T␤RI, whereas TGF-␤3 complex is in gray. C and D, the T␤RII/TGF-␤ interface (C) and the T␤RI/T␤RII interface region (D) with TGF-␤1 complex colored in light blue for TGF-␤1, red for T␤RI, and yellow for T␤RII, whereas the TGF-␤3 complex is in gray.
unique to T␤RII. The corresponding region in all other type II receptors, including BMPRII and ActRII, lacks both ␤1Ј, ␤1Љstrands and the Cys 38 -Cys 44 disulfide bond. It would result in a conformation that would sterically clash with their cytokines at Site IIa (Fig. 5, C and E). Therefore, we can conclude that Site IIa at the "tip of cytokine fingers" is unique and restricted only to TGF-␤ complex assembly. On the other hand, BMPs use their "knuckle site" of their fingers, Site IIb, to bind the ␤4 and ␤5 strands of their receptors. This region in T␤RII contains a unique 5-8-amino acid insertion forming a hook blocking its ␤4 and ␤5 strands from binding to Site IIb (Figs. 4B and 5D). Except T␤RII, other known type II receptors have significantly shorter ␤4-␤5 loops that adopt similar conformation (Fig. 4B), suggesting knuckle site on cytokine is a common type II receptor binding site in TGF-␤ superfamily. Moreover, the ␤4-␤5 region on type II receptors is responsible for their specificity. Here, as in the above mentioned type I receptor recognition, the shorter ␤4-␤5 loop is characteristic for promiscuous type II receptors, such as ActRII and ActRIIB, whereas the ␤4-␤5 loop in BMPRII, which is 3 residues longer, restricts its recognition to BMPs only.
The type I and II receptors in TGF-␤ superfamily share a common three-finger toxin fold, yet have distinct binding sites, and do not appear to cross-react. Site I binding involves the residues on the first ␤1-␤2 loop of both T␤RI and BMPR-Ia. This loop is not only 7-13 residues shorter in all type I receptors ( Fig. 4) but is restrained by a conserved Cys 14 -Cys 17 disulfide bond unique to type I receptors. Likewise, Site IIb binding is supported by a conserved disulfide bond (Cys 72 -Cys 84 , ActRIIA numbering) between the ␤5 and ␤6 strands present in all type II receptors but absent in type I receptors (Figs. 4 and 5F). The corresponding region in type I receptors assumes a different conformation stabilized by a conserved disulfide bond (Cys 62 -Cys 76 ; Fig. 4C) present only in the type I receptors. Thus, the ligand recognition by the type I and II receptors of TGF-␤ superfamily appears to rely primarily on structural compatibilities, such as insertions and deletions at their receptor-ligand interfaces and unique disulfide bonds stabilizing specific secondary structure conformations. In general, insertions at the interface restrict the recognition, whereas deletions generate promiscuity.
TGF-␤1, -␤2, and -␤3 Exhibit Distinct Receptor Preferences but Comparable Ternary Assembly Affinities-Despite high sequence and structure similarity among TGF-␤ isoforms (Fig.  4), they often cause distinct outcomes in embryonic and adult tissues (17,19,21,30). These differences in TGF-␤s actions might be caused by their distinct interactions with their receptors. TGF-␤1 and -␤3 have previously been shown to bind and assemble T␤RI and T␤RII into complexes in an ordered manner, first by forming a stable binary complex with T␤RII and then by recruiting T␤RI (31,32). Current structures of the ternary complexes support this interdependent manner of assembly (7). To examine the receptor binding properties of the TGF-␤s, we used surface plasmon resonance with TGF-␤s immobilized on the sensor surface between 50 and 300 RU. The initial measurements were aimed at defining the relative affinities of T␤RI and T␤RII for the three isoforms. The sensorgrams obtained upon injection of T␤RII over the TGF-␤1, -␤2, and -␤3 surfaces were characterized by relatively fast on and off rates and could be readily fit using a simple 1:1 binding model (Fig. 6A). The derived K D values were consistent with expectations, with both TGF-␤1 and -␤3 (K D values of 190 and 140 nM, respectively) having an affinity more than a 100-fold greater than TGF-␤2 (22.4 M) ( Table 2). Structurally, the T␤RII interface residues Arg 25 and Arg 94 of TGF-␤1 and -␤3 are replaced with lysines in TGF-␤2 (Fig. 4). This likely weakens TGF-␤2 binding to T␤RII because of a shorter lysine side chain and A, sequence comparison of several mammalian TGF-␤s, BMPs, activin, nodal and growth differentiation factor 5 (GDF5). The numbering is consistent with the sequence of TGF-␤1. B, sequence comparison of several type II receptors. The numbering is according to the T␤RII sequence. C, sequence alignment of several type I receptors. The numbering is consistent with the T␤RI sequence. The secondary structure elements are illustrated as arrows and cylinders for ␤-strands and ␣-helices, respectively. The residues involved in interactions in the TGF-␤1 and BMP-2 ternary complexes are highlighted cyan and magenta, respectively. Disulfide bonds critical for receptor/ligand specificity and compatibility and highlighted yellow, numbered, and connected by thick yellow lines. Secondary structure elements crucial for receptor/ ligand pairing are boxed. Arg 25 and Arg 95 are marked by stars.
T␤RI, to our surprise, yielded detectable signals when injected over the TGF-␤2 and -␤3 but not TGF-␤1 surfaces (Fig. 6B). The TGF-␤2 sensorgrams could be readily fit to a simple 1:1 binding model with a K D of 11.2 M, whereas the TGF-␤3 sensorgrams could not. The steady state fitting of TGF-␤3 sensorgrams yielded a K D of 2.4 M ( Table 2 and supplemental Fig. S2). Interestingly, TGF-␤2 binds T␤RI with 2-fold higher affinity than T␤RII. To attempt to quantify the weak interaction between T␤RI and TGF-␤1, an equilibrium experiment was performed with a higher density TGF-␤1 surface (686 RU). However, the response was barely detectable over the range of T␤RI concentrations sampled (0 -16 M). An estimate based on equilibrium response indicates a K D greater than 70 M (35). In addition to the type I receptor affinity differences between TGF-␤1 and -␤3, the kinetic association and disassociation rates for TGF-␤3 are also slower (Fig. 6). This could contribute to the functional difference between the TGF-␤ isoforms. Thus, T␤RI also displays ligand preferences with TGF-␤3 Ͼ TGF-␤2 Ͼ Ͼ TGF-␤1. Although T␤RI binds TGF-␤3 tighter than TGF-␤1, the interface between T␤RI and TGF-␤3 is 400 Å 2 smaller than that between T␤RI and TGF-␤1 because of a 10°d ifference in T␤RI docking orientation. The structure shows that 8 of 13 TGF-␤ residues are conserved at the T␤RI interface among the three isoforms. Asn 5 , Ile 51 , Gln 57 , Lys 60 , and Gln 67 vary with Ile 51 and Gln 67 unique to TGF-␤1 and thus may contribute to its weaker T␤RI binding (Figs. 2D and 4A). Despite the fact that T␤RI forms a larger interface area than T␤RII with TGF-␤1 and -␤3, it binds both TGF-␤s weaker compared with T␤RII. Structurally, TGF-␤1 interacts with its type II receptor mostly through hydrogen bonds and salt bridges but with its type I receptor via primarily hydrophobic contacts ( Fig. 4 and supplemental Table S1). A smaller interface yet higher affinity interaction between T␤RII and TGF-␤1 suggests the importance of hydrogen bonds in achieving high receptor-ligand affinity. Similarly, BMP-2 forms 16 hydrogen bonds and salt bridges with its high affinity receptor BMPR-IA but mostly hydrophobic FIGURE 5. Structural determinants critical for receptor/ligand recognition in TGF-␤ superfamily. A, alignment of TGF-␤1 and BMP-2 and their type I receptors. TGF-␤1 is colored cyan. For clarity only one monomer of BMP-2 is shown in magenta. T␤RI and BMPR-Ia are in red and blue, respectively. ␤1-␤2 and ␤4-␤5 loops as well as disulfide bonds on receptor and ligand that are critical for receptor/ligand compatibility are marked. Receptor II binding sites in TGF-␤1 and BMP-2 ternary complex (Site IIa and Site IIb) are outlined as ovals. B, schematic representation of TGF-␤-type and BMP-type ternary complex with receptor I and II binding sites marked as Site I, Site IIa, and Site IIb. C, T␤RII in surface representation (yellow) with TGF-␤1 binding site painted cyan. BMPRII ␤1-␤2 loop is represented as a blue ribbon that blocks the binding site, thus prohibiting BMPRII from binding to Site IIa. D, ActRII in surface representation (gray) with BMP-2 contact area colored magenta. T␤RII unique extension of ␤4-␤5 loop (yellow ribbon) and unique conformation of ␤4-␤5 loop in type I receptors (red ribbon) prevent binding of type I receptors and T␤RII to Site IIb. E, surface representation of T␤RI (red) with TGF-␤1 contact area marked cyan. ActRII ␤1-␤2 loop is shown as a gray ribbon that blocks one of the binding sites illustrating impossibility of any type II receptor to bind to Site I. F, critical differences in the conformation between type I (red) and type II (gray) receptors.
contacts with the low affinity ActRII. Because hydrogen bonds, in general, provide more specific receptor/ligand recognition than hydrophobic interactions, the dominance of hydrogen bonding interactions at the TGF-␤1/T␤RII and BMP-2/BMPR-IA interfaces is consistent with their higher affinity and thus the preferential recognition of type II receptor for TGF-␤s and type I receptor for BMPs. Hydrogen bonds and salt bridges also dominate the high affinity receptor ␣ chain binding to hematopoietic cytokines, such as interleukin-2 and -4, and van der Waal's contacts primarily occupy low affinity receptor-cytokine interfaces (36 -38).
The subsequent binding studies were aimed at quantifying the extent to which one receptor type potentiated the binding of the other. To accomplish this, T␤RII was included in the buffer at a concentration of 4 M, whereas T␤RI was injected over a range of concentrations. The sensorgrams obtained were characterized by slow association and dissociation rates and could each be easily fit to a simple 1:1 binding model (Fig. 6C). The derived K D values were 70, 16, and 14 nM for TGF-␤1, -␤2, and -␤3, respectively (Table 2). Thus, higher T␤RI binding affinities were observed for all three TGF-␤ in the presence of T␤RII, reflecting their similar efficiencies in assembly of the ternary complexes. Except for the case of TGF-␤2 where 4 M T␤RII is 5-fold lower in concentration than their K D , the K D values measured for TGF-␤1 and TGF-␤3 reflect the binding of T␤RI to the T␤RII⅐TGF-␤ binary complex. Structurally, the cooperative receptor binding reflects the favorable contacts between T␤RI and T␤RII FIGURE 6. Surface plasmon resonance sensorgrams and kinetic fits for binding of the T␤RI and T␤RII extracellular domains to TGF-␤1, -␤2, and -␤3. A, sensorgrams obtained as T␤RII was injected. The traces correspond to triplicate measurements of 2-fold serial dilutions of the receptor over the concentration ranges shown. The surface densities were 185, 339, and 165 RU for TGF-␤1, -␤2, and -␤3, respectively. The red curves correspond to global fits of each data set to a 1:1 binding model using Scrubber 2 software. B, sensorgrams and kinetic fits obtained as T␤RI was injected. Surface densities were 242, 339, and 595 RU for TGF-␤1, -␤2, and -␤3, respectively. Sensorgrams obtained for TGF-␤3 indicated heterogeneity that could not be fit to a simple 1:1 model, and hence no fit is shown. C, sensorgrams and kinetic fits obtained as T␤RI was injected in the presence of 4 M T␤RII. The surface densities were 498, 339, and 595 RU for TGF-␤1, -␤2, and -␤3, respectively.
The close resemblance between the TGF-␤1 and TGF-␤3 ternary complex structures is consistent with their similar binding affinities. However, there are some important differences with regard to assembly. Although TGF-␤1 and TGF-␤3 both bind T␤RII with high affinity, TGF-␤1 binds T␤RI much more weakly than TGF-␤3. This difference in T␤RI binding also persists in the context of the binary complexes, with the T␤RII⅐TGF-␤3 complex having a 5-fold greater affinity for binding and recruiting T␤RI compared with the T␤RII⅐TGF-␤1. To better describe the receptor preference and the cooperative contribution of individual receptors in TGF-␤ signaling complex assembly, we define the receptor preference as the ratio of the dissociation constants between the T␤RI and T␤RII for each TGF-␤. For example, TGF-␤3 binds to T␤RI and T␤RII with 2.4 and 0.17 M affinities, respectively, resulting in a 14-fold receptor preference for T␤RII. This T␤RII preference is estimated to be greater than 200 for TGF-␤1 and vanished completely (i.e. 0.5) for TGF-␤2. In other words, there is no preferential binding to the type II versus type I receptor in TGF-␤2 receptor recruitment. These results demonstrate that although all three TGF-␤s can effectively assemble their ternary complexes, their receptor preferences and the contribution of each receptor to the cooperative assembly appear distinct. TGF-␤1, because of its strong preference for binding T␤RII over T␤RI, assembles ternary complex in a prototypical manner, first binding T␤RII with high affinity (K D ϭ 190 nM) and then and only then binding and recruiting T␤RI (K D ϭ 70 nM). TGF-␤3 also preferentially binds T␤RII over T␤RI, although with less preference compared with TGF-␤1 (Table 2). Therefore, TGF-␤3 also likely binds and assembles its receptors in a largely prototypical manner. In contrast, TGF-␤2 displays no receptor preference, and both receptors contribute nearly equally to the assembly of its ternary complex. This suggests that TGF-␤2, instead of following the sequential receptor recruitment paradigm, engages either T␤RI or T␤RII and then recruits the complementary receptor or requires additional co-receptors to stabilize the binary cytokine-receptor complex. It is also possible that T␤RI and T␤RII associate into a preformed dimer, although no direct binding can be detected in solution between the two receptors. These results could explain the 100 -1000-fold lower potency of TGF-␤2 in inducing functional responses in cells lacking the TGF-␤ co-receptor betaglycan, because the fraction of ligand initially captured on the surface would expected to be lower for TGF-␤2 compared with TGF-␤1 and TGF-␤3 (primarily because of its lower affinity for T␤RII). Based on this, betaglycan likely functions as enhancer of cellular sensitivity to TGF-␤2 by its demonstrated ability to promote binding of TGF-␤2 to T␤RII (39); this in turn should endow TGF-␤2 with the capacity to bind and recruit T␤RI in a manner comparable with that of TGF-␤1 and -␤3.
In summary, the structure of TGF-␤1⅐T␤RI⅐T␤RII ternary complex and its comparison with the TGF-␤3 ternary complex showed a common ligand recognition mode in the TGF-␤ family of cytokines by their receptors. In particular, the low affinity type I receptor interacts with both TGF-␤ and T␤RII but with slightly different orientations at site I in the two structures. Among the type I receptors, their ligand specificity appears to correlate with the lengths of their ␤1-␤2 ligand binding loops, with receptors with shorter ␤1-␤2 loop being more promiscuous. The high affinity T␤RII binds to a conserved site (IIa) at the tip of TGF-␤1 and TGF-␤3. The T␤RII binding site at the tips of the cytokine fingers is restricted to TGF-␤ only. All other type II receptors in TGF-␤ superfamily bind to the common site IIb at the knuckles of cytokine. As with type I receptors, the length of the ␤4-␤5 region is a determining factor for type II receptor specificity and promiscuity, whereby the receptors with shorter ␤4-␤5 loop display promiscuous ligand recognition. Hydrogen bonds and salt bridges rather than hydrophobic interactions appear to be critical for the high affinity receptor recognition, in this case T␤RII and TGF-␤1. Unlike TGF-␤1, both TGF-␤2 and TGF-␤3 exhibited significant affinities to T␤RI. Although all three TGF-␤s form their ternary receptor complexes equally well, the variations in the type I and II receptor preferences among the TGF-␤s likely modulate the kinetics of ternary complex assembly. As a result, TGF-␤2 likely recruits the type I and II receptors simultaneously rather than sequentially. The differences in the kinetic assembly of the type I and II TGF-␤ receptors suggest a potential functional variation among TGF-␤s with respect to cellular and tissue distributions of their receptors.