Smad1 Recognition and Activation by the ALK1 Group of Transforming Growth Factor-β Family Receptors*

Two structural elements, the L45 loop on the kinase domain of the transforming growth factor-β (TGF-β) family type I receptors and the L3 loop on the MH2 domain of Smad proteins, determine the specificity of the interactions between these receptors and Smad proteins. The L45 sequence of the TGF-β type I receptor (TβR-I) specifies Smad2 interaction, whereas the related L45 sequence of the bone morphogenetic protein (BMP) type I receptor (BMPR-I) specifies Smad1 interactions. Here we report that members of a third receptor group, which includes ALK1 and ALK2 from vertebrates and Saxophone from Drosophila, specifically phosphorylate and activate Smad1 even though the L45 sequence of this group is very divergent from that of BMPR-I. We investigated the structural elements that determine the specific recognition of Smad1 by ALK1 and ALK2. In addition to the receptor L45 loop and the Smad1 L3 loop, the specificity of this recognition requires the α-helix 1 of Smad1. The α-helix 1 is a conserved structural element located in the vicinity of the L3 loop on the surface of the Smad MH2 domain. Thus, Smad1 recognizes two distinct groups of receptors, the BMPR-I group and the ALK1 group, through different L45 sequences on the receptor kinase domain and a differential use of two surface structures on the Smad1 MH2 domain.

The Smad family of proteins play a central role in signal transduction by the transforming growth factor-␤ (TGF-␤) 1 family (1)(2)(3). Smad proteins are directly phosphorylated by type I TGF-␤ family receptors, an event that induces their accumulation in the nucleus where they activate transcription of specific genes. Smad proteins act as signal transducers for different members of the TGF-␤ family, including TGF-␤ itself, the activins, and the bone morphogenetic proteins (BMPs). The type I receptors for TGF-␤ and activin, which are known as T␤R-I and ActR-IB, respectively, signal via Smad2 and its close homolog Smad3. The BMP type I receptors BMPR-IA and BMPR-IB signal via Smad1 and possibly its close homologs, Smad5 and Smad8. Upon receptor-mediated phosphorylation, and on their way to the nucleus, all these Smad proteins associate with Smad4, a member of a separate subclass that is required for the formation of transcriptional complexes. The Drosophila orthologs of BMP (Dpp), BMPR-I (Thickveins), Smad1 (Mad), and Smad4 (Medea) are functionally linked in a similar fashion (4 -6).
The maintenance of specificity in this system requires that each member of the type I receptor family be able to discriminate among different groups of Smad proteins. The specific interaction between Smad proteins and type I receptors is determined by two structural elements, namely, the L3 loop in the carboxyl-terminal domain (or MH2 domain) of Smad proteins and the L45 loop in the kinase domain of the receptors (7,8). Both elements consist of a short amino acid sequence that is highly conserved among Smad proteins or receptors of similar specificity but differs on a few critical residues between functionally distinct Smad proteins or receptors. Thus, the L45 loop sequence of the T␤R-I/ActR-IB group (which also includes the orphan receptors ALK7, XTrR-I and Atr-I) is compatible with the L3 loop sequence common to Smad2 and -3, allowing functional interactions between these receptors and Smad proteins. A similar relationship exists between the L45 loop of the BMPR-I group (which also includes Thickveins) and the L3 loop of Smad1, -5, -8, and Mad (see Figs. 1A and 4A for L45 and L3 sequences).
An important question left open by previous studies is about the Smad specificity of a third group of type I receptors. This group includes ALK1, ALK2, and Drosophila Saxophone. Saxophone is essential for dorsal closure of the Drosophila embryo and is believed to mediate Dpp signals along with Thickveins (9 -12). ALK1 (also known as TSR-I) is highly expressed in vascular endothelial cells (13). Inactivating mutations of ALK1 in humans cause hereditary hemorrhagic telangiectasia (14). Hereditary hemorrhagic telangiectasia is an autosomal dominant disorder characterized by epithelial vascular dysplasia with a high propensity to hemorrhage in the nasal and gastrointestinal mucosa. ALK2 (also known as ActR-I or Tsk7L) is a broadly expressed receptor that can bind BMPs, activin, and, under certain conditions, TGF-␤ in vitro (1-3, 13, 15-19). However, ALK2 does not mediate activin responses like those mediated by ActR-IB (20). In Xenopus, the activity of ALK2 (21) is akin to that of BMPR-I (22,23), because both receptors signal ventral mesoderm induction, and defects in ventral mesoderm formation caused by a dominant-negative ALK2 construct can be rescued by overexpression of Smad8 (24).
In this study, we have investigated the Smad specificity of this third group of type I receptors. We were intrigued by the fact that the L45 sequence of these receptors is very different from that of the T␤R-I group or the BMPR-I group (see Fig. 1A). Nonetheless, we observed that ALK1 and ALK2, like BMPR-I, recognize and activate Smad1. This was paradoxical because T␤R-I, whose L45 sequence is much closer to that of BMPR-I, does not recognize Smad1. Insights from the crystal structure of the Smad MH2 domain allowed us to address this paradox. EXPERIMENTAL PROCEDURES R1B/L17, COS-1, and HepG2 cells were maintained as described previously (8,25). Mouse embryonic P19 cells were maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum and 2 mM glucose. Mutagenesis of Smad proteins and receptors was performed by polymerase chain reaction using appropriate oligonucleotides, and verified by DNA sequencing. T␤R-I(LA) contains a replacement of the L45 sequence ADNKDNGTW with the sequence SD-MTSRHSS. Smad2(H1-1) contains the mutations A323S, V325I, and M327N, and Smad2(HL1) contains the mutations A323S, V325I, M327N, R427H, and T430D. Other constructs have been reported previously (7,8).
Activation of the 3TP-luciferase reporter (20) and Mix2 A3-luciferase reporter (26) by receptors were analyzed in R1B/L17 cells as described previously (8,25). To measure the activity of a XVent2-luciferase reporter (27), P19 cells were transfected with this construct, and the constitutively active forms of type I receptors, using Lipofectin (Life Technologies, Inc.) according to manufacturer's instructions. Luciferase activity was measured 40 h after transfection.
Metabolic labeling of transfected cells with [ 32 P]orthophosphate or [ 35 S]Met/Cys was performed as described previously (8,25). Immunoprecipitates with monoclonal anti-Flag M2 antibody (IBI, East Kodak) were visualized by SDS-polyacrylamide gel electrophoresis followed by autoradiography. Subcellular localization of Smad proteins was examined in HepG2 cells transfected with N-terminally Flag-tagged Smad proteins and the constitutively active receptors. Immunofluorescence was carried out as described previously (8). At least 200 cells were scored per assay. The percentage of Flag-positive cells with predominantly or exclusively nuclear immunofluorescence is plotted in the figures.

RESULTS AND DISCUSSION
ALK1 and ALK2 Transduce BMP-like Signals-To investigate the signaling specificity of human ALK1 and ALK2, we used transcriptional assays that discriminate between BMPlike signaling and TGF-␤/activin-like signaling. Since the natural ligands for ALK1 and ALK2 have not been conclusively defined, we generated ALK1 and ALK2 mutant constructs [ALK1(QD) and ALK2(QD)] containing a Gln to Asp mutation in the penultimate residue of the regulatory domain (GS domain), which activates type I receptors in a ligand-independent manner (28 -30). The corresponding mutant forms of the BMP type I receptors, BMPR-IA(QD) and BMPR-IB(QD), activate luciferase expression when cotransfected with the XVent2-Luc reporter construct into mouse P19 cells (Fig. 1B). XVent2-Luc contains a BMP-responsive region from the Xenopus Vent.2 gene driving expression of luciferase (27) and can be activated via Smad1 but not Smad2 (8). ALK1(QD) and ALK2(QD) were also able to activate XVent2-Luc, whereas the constitutively active TGF-␤ receptor construct, T␤R-I(TD), had no effect (Fig.  1B).
Smad Specificity of ALK1 and ALK2-These results raised the possibility that ALK1 and ALK2 may be able to phospho-rylate and activate Smad1 but not Smad2. To investigate this possibility, receptor constructs were cotransfected with N-terminally Flag-tagged Smad1 or Smad2 into R1B/L17 cells, and Smad phosphorylation was determined by anti-Flag immunoprecipitation from the [ 32 P]orthophosphate-labeled transfectants. The basal phosphorylation of Smad1 and Smad2 observed in the absence of receptor activity in these cells ( Fig. 2A, left lanes) is caused by MAP kinases acting on phosphorylation sites unrelated to the receptor-mediated phosphorylation sites (37). 3 ALK1(QD) and ALK2(QD) increased the phosphorylation of Smad1 but not that of Smad2, which is similar to the effect of BMPR-IB(QD) and opposite to the effect of T␤R-I(TD) (Fig.  2A). These results were consistent with ALK1 and ALK2 acting as specific activators of Smad1 but not Smad2.
To confirm that ALK1(QD) or ALK2(QD) specifically activated Smad1, we determined the ability of these constructs to induce nuclear accumulation of Smad proteins. Consistent with the above results, expression of ALK1(QD) or ALK2(QD) induced nuclear accumulation of Flag-Smad1 but not Flag- activates the BMP-responsive XVent-Luc reporter but not the TGF-␤responsive 3TP-lux reporter. XVent-luciferase and 3TP-luciferase reporters were transfected into P19 or R1B/L17 cells, respectively, with wild type or mutant T␤R-I as well as T␤R-II. Cells were then incubated with 1 nM BMP2 (B) or 0.5 nM TGF-␤ (T) for 1 day, and luciferase activity was determined. Data are the average of three or more assays Ϯ SD. Smad2, which again is similar to the effect of BMPR-IB(QD) and opposite the effect of T␤R-I(TD) (Fig. 2B). Under these conditions, nuclear localization of Flag-Smad proteins was observed in only a fraction of the cells. The incompleteness of this response may be because of limitations in the ability of the receptors to quantitatively activate the overexpressed Smad proteins. In untransfected cells, TGF-␤ induces a quantitative translocation of endogenous Smad2 and -3 into the nucleus. 3 We repeated these assays using a Smad1 construct [Smad1(AAVA)] that contains three Ser to Ala mutations in the C-terminal sequence SSVS (38). These serines are the sites phosphorylated by BMPR-I, and their mutation to alanine prevents receptor-mediated phosphorylation and activation of Smad1 by BMP (38). The Smad1(AAVA) mutant failed to accumulate in the nucleus when cotransfected with ALK1(QD) or ALK2(QD) vectors (data not shown). Therefore, ALK1 and ALK2 can mimic the ability of BMP type I receptors to specifically phosphorylate and activate Smad1, leading to BMP-like transcriptional responses.
Determinants of Specificity in ALK1 and ALK2-The sequence of the L45 loop in the kinase domain of T␤R-I determines TGF-␤ signaling activity (39). Furthermore, the L45 loop  (40). Subtype-specific residues are boxed. The corresponding sequences of Smad4 is also shown. Numbering of the last residue in each sequence corresponds to the Smad species not in the parentheses. B, a close-up view of the Smad4 MH2 structure showing the L3 loop (yellow) with subtype-specific residues (red) and the ␣-helix 1 (purple) with subtype-specific residues (green). The insert shows a frontal view of the location of the L3 loop and helix 1 of each MH2 monomer in the crystallographic trimer. of T␤R-I and BMPR-I specifies the choice of Smad proteins by these receptors in the cell (8) . Five of nine residues in this sequence are conserved between T␤R-I and BMPR-I (see Fig.  1A). Thus, the three nonconserved residues are responsible for the different Smad specificity of these two receptors. Mutant T␤R-I and BMPR-I constructs with these residues swapped show a switch in their ability to recognize and activate Smad1 and Smad2 (8). The L45 loop is conserved between ALK1 and ALK2 but is very divergent between these receptors and T␤R-I or BMPR-I (see Fig. 1A). Because the Smad specificity of ALK1 and ALK2 is similar to that of BMPR-I receptors, we investigated whether this specificity is determined by the L45 sequence of ALK1 and ALK2 despite their lack of similarity to the BMPR-I L45 sequence. To this end, we generated a mutant T␤R-I [T␤R-I(LA)] containing the L45 sequence SDMTSRHSS, which corresponds to the ALK2 L45 sequence. The signaling specificity of this construct was compared with those of the wild type T␤R-I and the previously described mutant T␤R-I(LB) containing the L45 sequence of BMPR-I (8). T␤R-I(LA) was similar to T␤R-I(LB) and BMPR-I in its ability to induce nuclear accumulation preferentially of Smad1 (Fig. 3A) and its pattern of activation of XVent2-Luc (Fig. 3B) and 3TP-Lux (Fig.  3C). These results suggest that the ability of the BMPR-I L45 sequence to specify an interaction with Smad1 is shared by the very divergent L45 sequence of ALK1 and -2 but not by the closely related L45 sequence of T␤R-I.

Role of the ␣-Helix 1 in Smad Recognition of ALK1 and
ALK2-In light of this, we wondered whether Smad1 might recognize ALK1 and -2 and BMPR-I by different mechanisms. The ability of Smad1 and Smad2 to recognize BMPR-I and T␤R-I, respectively, is determined by the sequence of the L3 loop (7). As inferred from the crystal structure of the Smad4 MH2 domain, the L3 loop protrudes from the surface of this domain (40). The sequence of the L3 loop is identical within the Smad2/3 and Smad1/5/8 subgroups, but differs at two critical positions between these two subgroups (Fig. 4, A and B). Swapping these two residues between Smad1 and Smad2 switches their ability to interact with specific receptors (7). Thus, a Smad2 mutant containing the two Smad1-specific residues in the L3 loop [Smad2(L1) construct] is preferentially activated by BMPR-I (7) (Fig. 5A). Using this approach, we determined that ALK1(QD) and ALK2(QD) are relatively weak inducers of nuclear accumulation of Smad2(L1) even though they induce nuclear accumulation of Smad1 (Fig. 5A). These results indicate that the L3 sequence alone is not sufficient to specify Smad1 recognition by ALK1 or ALK2.
We searched Smad MH2 domains for other regions containing subtype-specific residues, that is, residues that would be different between Smad1 and -2 but conserved between Smad1, -5, -8, and Mad or between Smad2 and -3. The region immediately preceding the C-terminal phosphorylation sites contains four such residues (7). However, introduction of these mutations into Smad2, along with the L3 loop mutations, did not improve the ability of ALK1(QD) or ALK2(QD) to induce nuclear accumulation of the resulting construct (Fig. 5A, Smad2(LC1) construct). Another region with subtype-specific residues is a sequence corresponding to ␣-helix 1, which is a surface structure in Smad4 (40). The corresponding sequence is conserved between Smad1 and Smad2 except for two residues that are subtype-specific (Fig. 4A). These residues in Smad4 (Glu-374 and Arg-378) are separated by one turn of the ␣-helix and are exposed to solvent in the vicinity of the L3 loop (40) (Fig. 4B). Because of their properties and location, we investigated whether these residues play a role as determinants of the interaction of Smad1 and ALK1 and -2. A Smad2 mutant containing these two residues from Smad1 [Smad2(H1-1)] did not respond to BMPR-IB(QD), ALK1(QD) or ALK2(QD) in the nuclear accumulation assay (Fig. 5A). However, a Smad2 construct containing the two ␣-helix 1 residues and the two L3 loop residues from Smad1 [Smad2(HL1)] underwent a similar extent of nuclear accumulation when cotransfected with ALK1(QD), ALK2(QD) or BMPR-IB(QD) (Fig. 5A). These results suggest a critical role of ␣-helix 1 in Smad1 recognition by Smad1 recognition by ALK1 and ALK2 requires a structure, the ␣-helix 1, which is not essential for Smad1 interaction with BMPR-I (8). Smad proteins are represented as homotrimers based on the Smad4 MH2 structure (40). See text for additional details. ALK1 and ALK2.
To verify that both ␣-helix 1 and the L3 loop are required for Smad1 recognition by ALK1 and 2, wild type and mutant Smad2 constructs were transfected into HepG2 cells with the constitutively active forms of receptors, and phosphorylation of Smad2 was determined. Consistent with the pattern of nuclear accumulation, coexpression of BMPR-IB(QD) increased the phosphorylation level of Smad2(L1) and Smad2(HL1), but not wild type Smad2, whereas coexpression of ALK1(QD) and ALK2(QD) only enhanced phosphorylation of Smad2(HL1) (Fig. 5B).
From these results, we conclude that two groups of Smad1activating receptors coexist in vertebrates and possibly Drosophila, the BMPR-I group and the ALK1 group. A similar conclusion was recently reached by others using a different BMP reporter assay (41). Furthermore, the present work shows that the recognition and activation of Smad1 by ALK1 and ALK2 is specified by the L45 loop on the receptors and the L3 loop together with the ␣-helix 1 on Smad1 (Fig. 6). It is likely that Saxophone recognizes Mad in a similar fashion because the L45 sequences of Saxophone and the L3 sequence and ␣-helix 1 sequence of Mad are very similar or identical to those of their counterparts in vertebrates. It is important to note that the structural elements identified here and previously (7,8), although playing a major role in dictating the specificity of the recognition between type I receptors and Smad proteins, may be only part of larger regions mediating the association between these proteins. Additionally, the correct assembly of receptor complexes and Smad protein complexes and/or the possible intervention of "adaptor" molecules might be important for the association of receptors and Smad proteins.
The BMPR-I group and the ALK1 group recognize Smad1 through related but different mechanisms which involve different L45 sequences on the receptor kinase domain, and a differential use of two surface structures on the Smad1 MH2 domain. Smad1 recognition and activation by these two mechanisms might be qualitatively different in ways that cannot be discerned by the present assays. In a given cellular context, such differences might make one receptor more suitable than the other as an activator of the Smad1 pathway, or both receptors could activate Smad1 in complementary ways or with different response outcomes.