Introduction
Transforming growth factor‐β (TGF‐β), bone morphogenetic proteins (BMPs), activins and related factors signal their many responses via pairs of transmembrane serine/threonine kinase receptors. In the ligand‐induced complex, one of the two kinases, called the type II receptor, phosphorylates and activates the other kinase, the type I receptor, which then phosphorylates substrates propagating the signal (
Massagué, 1996). The type I receptor substrates include a recently identified protein family, the SMAD proteins, that plays a central role in the relay of TGF‐β signals from the receptors to target genes in the nucleus (
Derynck and Zhang, 1996;
Wrana and Attisano, 1996;
Massagué et al., 1997).
A subclass of SMADs known as ‘receptor‐regulated’ SMADs are phosphorylated by specific receptors in a ligand‐dependent manner (
Hoodless et al., 1996;
Kretzschmar et al., 1997b). The receptor‐regulated SMADs physically associate with the ligand‐activated receptor complex (
Macias‐Silva et al., 1996;
Zhang et al., 1996) and undergo phosphorylation at the C‐terminus, release from the receptor, association with the related protein Smad4, which acts as a shared partner. This complex translocates into the nucleus and participates in transcriptional complexes (
Eppert et al., 1996;
Hoodless et al., 1996;
Lagna et al., 1996;
Liu et al., 1996,
1997;
Macias‐Silva et al., 1996;
Zhang et al., 1996,
1997;
Kretzschmar et al., 1997b). Although much progress has been made in understanding the TGF‐β/SMAD pathway, nothing is known about the protein structures that determine the specificity of this pathway at the level of receptor–SMAD interaction.
Specificity in this system is provided by the ability of the receptors to discriminate among SMADs. In vertebrates, Smad1 (
Graff et al., 1996;
Hoodless et al., 1996;
Liu et al., 1996), and presumably its close homologs Smad5 (
Yingling et al., 1996) and Smad8 (
Chen et al., 1997;
Watanabe et al., 1997), are phosphorylated by BMP receptors (
Hoodless et al., 1996;
Kretzschmar et al., 1997b) and mediate BMP responses (
Graff et al., 1996;
Liu et al., 1996;
Thomsen, 1996). Smad2 and its close homolog Smad3 are phosphorylated by TGF‐β receptors (
Eppert et al., 1996;
Macias‐Silva et al., 1996;
Zhang et al., 1996;
Kretzschmar et al., 1997b;
Nakao et al., 1997a) and mediate TGF‐β and activin responses (
Baker and Harland, 1996;
Eppert et al., 1996;
Graff et al., 1996;
Lagna et al., 1996;
Macias‐Silva et al., 1996;
Zhang et al., 1996). In
Drosophila, Mad (a close homolog of Smad1 and the founding member of this family) mediates the effects of the BMP‐like factor, Dpp (
Sekelsky et al., 1995;
Wiersdorff et al., 1996;
Newfeld et al., 1997). In
Caenorhabditis elegans, Sma‐2 and Sma‐3 function downstream of Daf‐4, which is a receptor for a BMP‐related factor (
Savage et al., 1996).
SMAD proteins consist of three regions: a conserved N‐terminal domain (referred to as the N or MH1 domain), a conserved C‐terminal domain (the C or MH2 domain) and a more divergent linker region (
Massagué et al., 1997). The MH2 domain has effector function as determined in transcriptional assays (
Liu et al., 1996) and in
Xenopus mesoderm formation assays (
Baker and Harland, 1996). The MH1 domain inhibits this effector function (
Baker and Harland, 1996;
Liu et al., 1996;
Hata et al., 1997). This inhibitory effect is relieved by receptor‐mediated phosphorylation on the C‐terminal sequence SS(V/M)S present in receptor‐regulated SMADs (
Macias‐Silva et al., 1996;
Kretzschmar et al., 1997b). Recently, the MH1 domain of Mad has been shown to bind to DNA (
Kim et al., 1997). Therefore, the MH1 domain may also subserve the gene activation role of the SMADs, and the MH1 and MH2 domains may have a mutually inhibitory effect in the basal state. The linker region contains serine residues whose phosphorylation by MAP kinases in response to tyrosine kinase receptor activation inhibits Smad1 translocation to the nucleus (
Kretzschmar et al., 1997a). Smad6, Smad7 and
Drosophila Dad, which act as antagonists of TGF‐β signaling, have a more divergent MH1 domain (
Hayashi et al., 1997;
Imamura et al., 1997;
Nakao et al., 1997b;
Topper et al., 1997;
Tsuneizumi et al., 1997;
Hata et al., 1998).
Smad2 and
Smad4 are tumor suppressor genes in chromosome 18q21, and mutations in these genes have been found in various types of cancers, principally carcinomas of the pancreas and colon (
Eppert et al., 1996;
Hahn et al., 1996). Many of the tumor‐derived missense mutations, as well as missense mutations in defective alleles of
Mad,
Sma‐2 and
Sma‐3, map to the MH2 domain (summarized in
Shi et al., 1997). The recent resolution of the crystal structure of the Smad4 MH2 domain has provided insights into how the naturally occurring mutations interfere with SMAD function (
Shi et al., 1997). Smad4 MH2 domain monomers assemble into a trimer, with each monomer resembling an open‐sided cradle containing a core β‐sandwich structure. Combining structural and functional insights, naturally occurring MH2 domain mutations can be grouped into those that disrupt the core structure of the MH2 domain and destabilize the protein, those that disrupt the trimer interface, and those that fall in a protruding structure referred to as the L3 loop. The latter mutations inhibit Smad4 from associating with receptor‐activated Smad2, thus preventing the formation of a functional Smad2–Smad4 complex (
Shi et al., 1997). Mutations in the MH1 domain augment its affinity for the MH2 domain, thus increasing the autoinhibitory function (
Hata et al., 1997).
Protein–protein interactions are critical determinants of specificity and fidelity in signal transduction pathways (
Hill and Treisman, 1995). Various protein modules have been identified that mediate specific protein–protein interactions in signal transduction pathways in a wide range of cellular processes (
Pawson and Scott, 1997). In order to identify protein structures that may mediate and specify protein–protein interactions in the receptor serine/threonine kinase signaling system, we have investigated TGF‐β and BMP signaling pathways whose surface receptors are prototypic of this system. We report here that the L3 loop in the MH2 domain of receptor‐regulated SMADs is crucial for their specific interaction with the receptors.
Discussion
Specificity is an essential property of signal transduction pathways. In the TGF‐β signaling system, specificity is determined by ligand activation of a particular receptor combination which, in turn, recruits and phosphorylates a particular subset of SMAD proteins. In the present study, we have investigated the SMAD–receptor interaction and the molecular basis for its specificity. Our results identify the L3 loop as a discrete surface structure in SMAD proteins that is necessary for the SMAD–receptor interaction and sufficient to dictate its specificity.
The differential ability of Smads 1 and 2 to associate with the TGF‐β receptor complex is consistent with their known responsiveness to these receptors: Smad2, which mediates TGF‐β signaling, associates with the TGF‐β receptor complex ∼10‐fold better than Smad1, which is primarily a mediator of BMP signaling (
Derynck and Zhang, 1996;
Wrana and Attisano, 1996;
Massagué et al., 1997). This receptor interaction is required for Smad2 phosphorylation, since docking‐defective mutants of Smad2 are not phosphorylated in response to TGF‐β. However, the Smad2 phosphorylation sites themselves, along with the adjacent sequence in the 11 amino acid C‐tail region, are dispensable for the receptor interaction. This conclusion is based on our observation that the TGF‐β receptor can associate with a Smad2 deletion mutant lacking the C‐tail.
We have identified a structural motif, the L3 loop, as an important determinant of SMAD recognition by the receptor. The L3 loop is a highly conserved region within the MH2 domain that, by analogy to the crystal structure of the Smad4 MH2 domain, is predicted to form a highly solvent‐exposed structure that is poised for protein–protein interactions (
Shi et al., 1997). Introduction of various mutations into the L3 loop, including developmental mutations previously observed in
Drosophila Mad (
Sekelsky et al., 1995) and
C.elegans Sma‐2 and ‐3 (
Savage et al., 1996), diminishes the ability of Smad2 to associate with the TGF‐β receptor complex. None of these mutations has appreciable effects on the Smad2 expression level or its ability to homo‐oligomerize, as predicted from the fact that the L3 loop is not part of the SMAD MH2 domain core structure (
Shi et al., 1997).
The sequence of the L3 loop, which is invariant among TGF‐β‐activated SMADs (Smads 2 and 3) and among SMADs thought to be activated by BMPs (Smads 1, 5 and 8) or Dpp (Mad), differs at two positions between these two groups. These two amino acids also differ in Smad4 as well as in Smads 6 and 7 (
Figure 1A). In Smad4, these two positions are highly exposed (
Figure 1B), and the same is likely to occur in other SMADs given their overall structural similarity to Smad4 (
Shi et al., 1997). As further testament to the importance of the L3 loop, switching these two amino acids in Smads 1 and 2 induces a gain or a loss, respectively, in their ability to bind to the TGF‐β receptor complex. This gain or loss of TGF‐β receptor binding is reiterated in TGF‐β receptor‐mediated phosphorylation of these SMADs, indicating that the L3 loop‐dependent receptor interaction is necessary and sufficient for TGF‐β receptor phosphorylation. Thus, it appears that the L3 loop‐mediated SMAD–TGF‐β receptor interaction is crucial or rate‐limiting for TGF‐β receptor‐mediated phosphorylation of its SMAD substrates. It should be noted, moreover, that the homologous C‐tail containing the phosphorylation sites and the adjacent sequence may ensure an optimal TGF‐β receptor‐mediated phosphorylation.
Testing whether the BMP receptor shares the same structural requirements for SMAD phosphorylation as the TGF‐β receptor revealed an intriguing difference. Whereas Smad2 containing the Smad1 L3 loop lost TGF‐β‐induced phosphorylation completely, Smad1 containing the Smad2 L3 loop was still phosphorylated in response to BMP, albeit to a lesser extent than wild‐type Smad1. This difference suggests that, in selecting among different SMAD substrates, the BMP receptor may have a more permissive requirement for the optimal L3 loop sequence, with the SMAD C‐tail region playing a complementary role in substrate recognition. Consistent with this notion, physical interaction between the BMP receptor complex and the SMADs appears to be much more transient than that between the TGF‐β receptor complex and the SMADs, as both affinity‐labeled wild‐type and kinase‐defective BMPR‐IA or ‐IB, in conjunction with wild‐type BMPR‐II, could not be co‐immunoprecipitated with Smad1 (our unpublished observations). In any case, swapping both the L3 loop and the C‐tail allows Smad1 and Smad2 to be phosphorylated respectively by the TGF‐β and BMP receptors, and a switch in agonist‐induced association with Smad4 and nuclear translocation accompanies this switch in phosphorylation.
Unlike the receptor‐regulated SMADs, Smad4 lacks a C‐terminal SS(V/M)S phosphorylation motif and does not appear to associate with the receptors on its own (
Macias‐Silva et al., 1996;
Zhang et al., 1996). What then is the function of the L3 loop in Smad4? Based on structural considerations and the observation that a mutation (G508S) in the Smad4 L3 loop abolishes the ability of Smad4 to associate with Smad2, we have proposed that the Smad4 L3 loop mediates the association with receptor‐activated SMADs (
Shi et al., 1997). We have corroborated the importance of the Smad4 L3 loop for Smad2–Smad4 interaction by showing that mutations of other residues in the Smad4 L3 loop (Y513A; and RQ515,516AA) also lead to the loss of TGF‐β‐inducible Smad2–Smad4 association in transfected COS‐1 cells (our unpublished observations). Smad4 is required for various responses to TGF‐β, activin and BMP by acting as a partner for the corresponding receptor‐activated SMADs (
Lagna et al., 1996;
X.Chen et al., 1997;
Liu et al., 1997;
Zhang et al., 1997). In addition, Smad4 can associate with these SMADs in yeast, suggesting that the interaction may be direct (
Hata et al., 1997;
Wu et al., 1997). SMAD L3 loops, therefore, are implicated in two distinct types of interactions. Among the receptor‐regulated SMADs, the L3 loop may mediate SMAD–receptor interactions, whereas the more divergent Smad4 L3 loop (see
Figure 1A) may mediate Smad4 interaction with receptor‐activated SMADs. It will be interesting to determine whether the L3 loop of receptor‐regulated SMADs has a dual function as a receptor‐interacting region and, upon phosphorylation of the C‐tail, as a Smad4‐interacting region.
Since the C‐tail of receptor‐regulated SMADs serves as a substrate for the type I receptor kinase, it must physically contact the receptor. However, this interaction apparently does not contribute significantly to the stability of the interaction that precedes phosphorylation, at least as determined with Smad2 and the TGF‐β receptor. In fact, the TGF‐β receptor–Smad2 interaction is weakened upon phosphorylation by the receptor, as either phosphorylation‐defective Smad2 mutants or a kinase‐defective TGF‐β type I receptor mutant enhance SMAD–receptor association (
Macias‐Silva et al., 1996; this work). It is not clear how SMAD phosphorylation may promote its dissociation from the receptor. A gain of affinity for Smad4 might contribute to Smad2 dissociation from the receptor upon phosphorylation. However, the Smad2(3A) mutant still showed an elevated receptor‐binding activity as compared with the wild‐type Smad2 in the Smad4‐deficient colorectal carcinoma cell line SW480.7 (our unpublished observation). Thus, an increased affinity for Smad4 may not be the only event driving dissociation of the phosphorylated Smad2 from the receptor complex.
Although two residues in the L3 loop are sufficient to dictate the specificity of the SMAD–receptor interaction, the entire L3 loop may not be sufficient to support this interaction fully. Attempts to demonstrate direct binding between receptors and SMADs or their L3 loops have not yet provided concrete evidence. It could be that a direct SMAD–receptor interaction is weak and requires oligomeric forms of both the receptors and the SMADs for cooperative binding. Alternatively, the SMAD–receptor interaction might be indirect, requiring a hitherto unidentified adaptor protein. Regardless of the mechanism, the evidence at hand identifies the L3 loop as a critical determinant of specific SMAD–receptor interactions.