Smad6 as a transcriptional corepressor.

Smad6 and Smad7, a subgroup of Smad proteins, antagonize the signals elicited by transforming growth factor-beta. These two Smads, induced by transforming growth factor-beta or bone morphogenetic protein (BMP) stimulation, form stable associations with their activated type I receptors, blocking phosphorylation of receptor-regulated Smads in the cytoplasm. Here we show that Smad6 interacts with homeobox (Hox) c-8 as a transcriptional corepressor, inhibiting BMP signaling in the nucleus. The interaction between Smad6 and Hoxc-8 was identified by a yeast two-hybrid approach and further demonstrated by co-immunoprecipitation assays in cells. Gel shift assays show that Smad6, but not Smad7, interacts with both Hoxc-8 and Hoxa-9 as a heterodimer when binding to DNA. More importantly, the Smad6-Hoxc-8 complex inhibits interaction of Smad1 with Hoxc-8- and Smad1-induced transcription activity. These data indicate that Smad6 interacts with Hox transcription factors as part of the negative feedback circuit in the BMP signaling pathway.

Members of TGF-␤ 1 superfamily transduce their signals into the cell through a family of mediator proteins called Smads. Receptor-regulated Smad1, Smad5, and Smad8 mediate BMP signaling, whereas Smad2 and Smad3 respond to TGF-␤ (1)(2)(3)(4). Upon phosphorylation by their type I receptors, The receptorregulated Smads interact with the common partner, Smad4, and translocate to the nucleus where the complex recruits DNA-binding protein(s) to activate specific gene transcription (5)(6)(7)(8)(9). Smad6 and Smad7 are struturally divergent Smads as antagonists of TGF-␤ family signaling (5,6). They can associate with activated TGF-␤ and BMP type I receptors, thereby preventing phosphorylation of receptor-regulated Smads (11)(12)(13). In addition, Smad6 has also been demonstrated to interact with phosphorylated Smad1 to prevent the formation of an active signaling complex of Smad1 and Smad4, preferentially inhibiting the signaling pathways activated by BMPs (14,15). Studies on the mechanism by which Smads mediate TGF-␤-/ activin-regulated gene transcription have led to the discovery of several Smad-interacting nuclear transcription factors and their cis-acting DNA elements. In particular, the Xenopus forkhead activin signal transducer-1 (FAST-1) binds to an activin response element upstream of the homeobox gene mix2. The transcription activation requires the presence of activin and assembly of a FAST-1-Smad2-Smad4 complex (9). The mammalian homolog FAST-2 activates the hox gene goosecoid where formation of a higher order complex of FAST-2-Smad2-Smad4 is also essential for transactivation (17). Transcription factor F3 (TFE3) binds to the E-box of the plasminogen activator inhibitor-1 (PAI-1) promoter, whereas Smad3 and Smad4 bind to a sequence adjacent to the TFE3 binding site to cooperatively activate PAI-1 gene transcription (10).
We have reported that Smad1 interacts with homeodomain transcription factor Hoxc-8 in response to BMP signaling (8). Hoxc-8 belongs to a highly conserved hox gene family and is expressed in limbs, backbone rudiments, the neural tube of mouse mid-gestation embryos, and in the cartilage and skeleton of newborns (19 -21). The interaction domains between the two proteins were characterized. Two regions within the amino-terminal 87 amino acid residues of Smad1 were mapped to interact with the homeodomain of Hoxc-8. Stable expression of recombinant cDNAs encoding the Hoxc-8 interaction domains of Smad1 in 2T3 osteoblast precursor cells stimulated osteoblast differentiation-related gene expression and lead to mineralized bone matrix formation. In this communication we show that Smad6 interacts Hoxc-8 as a complex when binding to DNA, thereby inhibiting Smad1-mediated transcriptional activity as negative feedback loop in the nucleus.
Immunoprecipitation and Western Blot-FLAG-tagged full-length, amino-terminal domain with linker region (Smad6NL), and carboxylterminal domain (Smad6C) of Smad6 were subcloned into a mammalian expression vector pcDNA3 (Invitrogen). HA-tagged Hoxc-8 expression vector was constructed previously (8). Constitutively active BMP type IA (ALK3) expression plasmid was provided by Dr. Jeffrey L. Wrana (The Hospital for Sick Children, Toronto, Ontario, Canada). COS-1 cells were transfected with expression constructs as indicated in Fig. 1 using Tfx-50 according to the manufacturer's description (Promega). Cells were lysed 48 h post-transfection, and lysates were immunoprecipitated with anti-FLAG M2 antibody (Eastman Kodak) and immunoblotted with anti-HA antiserum (Babco).
Gel Shift Assay-Gel shift assays were performed as described previously (18). Smad1/-4 cDNAs were obtained from Dr. R. Derynck (University of California, San Francisco, CA). GST fusion constructs of Smad1, -4, and Hoxc-8 were generated in our previous study (8). Smad6 cDNA, obtained from Dr. Ali Hemmati-Brivanlou (The Rockefeller University), and Smad7 cDNA, obtained from Dr. Peter ten Dijke (Ludwig Institute for Cancer Research, Sweden), were cloned into pGEX-KG vector. The GST constructs described above were transformed into BL21. The expression and purification of the fusion proteins were performed as described (16). OPN5 DNA fragments were used as the probe for the gel shift assays (8).
Transfection-The Hox-pGL3 reporter bearing the Hoxc-8 binding site (Ϫ290 to Ϫ166) was constructed into pGL3 control vector (Promega). The Hox recognition core TAAT was replaced with GCCG in Hox-pGL3 by polymerase chain reaction to create mutant Hox-pGL3 (mHox-pGL3) (8). Mv1Lu cells (5 ϫ 10 4 cells/12-well plate) were trans-* This work was supported by National Institutes of Health Grant DK53757 (to X. C.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
fected using Tfx-50 with 0.5 mg of luciferase reporter plasmid (Hox-pGL3 or mHox-pGL3) and different expression plasmids as indicated. Total DNA was kept constant by addition of pcDNA3 plasmid. Luciferase activities were assayed 48 h post-transfection using the Dual-Luciferase™ assay kit (Promega) according to the manufacturer's direction. Luciferase values shown in the figures are representative of transfection experiments performed in triplicate in at least three independent experiments. For immunolocalization, Mv1Lu cells were cotransfected with HA-tagged Hoxc-8 and GFP-Smad6 in the presence or absence of BMP-4. 48 h after transfection, cells were fixed with 4% paraformaldehyde and permeablized with 0.2% Triton X-100. Hoxc-8 was stained with monoclonal antibody against HA tag (Babco) and detected with secondary antibody conjugated to Cy3 TM (Jackson Immu-noResearch Laboratories, Inc.).

RESULTS AND DISCUSSION
We have previously demonstrated that Smad1 interacts with Hoxc-8 in response to BMP stimulation (8). Hoxc-8 functions as a transcription repressor in BMP signaling. The interaction of Smad1 with Hoxc-8 dislodges Hoxc-8 binding from its element resulting in initiation of gene transcription (8). To characterize the Hoxc-8-mediated transcription mechanism in BMP-induced gene activation, we used a yeast two-hybrid system to identify transcription factors that interact with Hoxc-8. An intact Hoxc-8 cDNA fused with the Gal4 DNA binding domain was used as a bait plasmid to screen a human U2 OS osteoblast-like cell cDNA library constructed in pACT2 plasmid. After two rounds of screening, we obtained 26 positive clones. DNA sequence analysis identified one clone as Smad6 (Table  1). Smad6 and Smad7 are immunolocalized in the nucleus of rat epiphyseal plate (22), Xenopus embryo (23), and Mink lung epithelial (Mv1Lu) cells (24). The interaction of Smad6 with Hoxc-8 suggests that Smad6 may have a novel antagonistic function in the nucleus.
The initial Smad6 cDNA clone (Smad6C in Table 1) encodes amino acids 281 to 496 of a 496-amino acid protein. The interaction between Hoxc-8 and Smad6 was further confirmed with a ␤-gal filter lift assay (data not shown) and quantified by a liquid ␤-gal assay (Table 1). When the full length of Smad6 fused with the GAL4 transcriptional activation domain was tested in the two-hybrid system, it showed a weaker interaction with Hoxc-8 in comparison with the carboxyl-terminal domain (Smad6C). Deletion of the Smad6 amino-terminal domain may change the protein conformation in a way that the carboxylterminal region becomes easier to interact with Hoxc-8. The assays of both empty bait vector (pGBT9) with Smad6C or Smad6 full-length cDNAs in prey plasmids and empty prey vector (pACT2) with full-length Hoxc-8 in bait vector showed very little activity. Compared with the interaction between Smad1 and Hoxc-8, the interaction of Smad6 with Hoxc-8 is about 5 times stronger (Table 1).
To investigate the interaction of Smad6 with Hoxc-8 in mam-malian cells and the effect of BMP stimulation on this interaction, COS-1 cells were transiently co-transfected with expression plasmids for FLAG-Smad6, HA-Hoxc-8, and/or constitutively active BMP type IA receptor ALK3 (Q233D). The cell lysates were immunoprecipitated with anti-FLAG M2 antibody and immunoblotted with anti-HA antibody. The results represented in Fig. 1 demonstrate that Smad6 (see Fig. 1, lanes 7 and 8) was co-immunoprecipitated with HA-Hoxc-8. BMP induces Smad6 mRNA expression (25,26), and overexpression of ALK3 (Q233D) did not significantly change the interaction of Smad6 with Hoxc-8 (lane 8), indicating that BMP stimulation is not required for the interaction between Smad6 and Hoxc-8. Our initial Smad6 clone only encodes the carboxyl-terminal domain, indicating that this region of the protein may be involved in the interaction with Hoxc-8. To further investigate this observation, two FLAG-tagged Smad6 truncation expression plasmids were constructed. As shown in Fig. 1, Smad6C exhibits a strong interaction with Hoxc-8 ( Fig. 1, lanes 4 and 5).
In contrast, the Smad6 amino-terminal with the linker region (Smad6NL) failed to bind to Hoxc-8 in immunoprecipitation assay (Fig. 1, lane 6). Smad proteins contain highly conserved amino-and carboxyl-terminal domains (referred to as MH1 and MH2 domains, respectively). The MH1 domain inhibits biological activities of the MH2 domain because of interactions between these two distal sites (27). Like other regulatory Smads, Smad6 also contains a conserved MH2 domain and short segments of MH1 domain homology (28). Therefore, our results suggest that the carboxyl-terminal domain of Smad6 interacts with Hoxc-8 and that the amino terminus negatively regulates interaction between the two proteins. We examined the effect of the interaction between Hoxc-8 and Smad6 on Hoxc-8 DNA binding activity. Gel shift assays were performed with purified GST-Smad6 and GST-Hoxc-8 fusion proteins using osteopontin Hoxc-8 DNA binding element as a probe. As expected, Hoxc-8 protein binds to the DNA probe, which is inhibited by Smad1 (Fig. 2a, lanes 5 and 6). Smad6 alone did not bind to the DNA element (lane 4). Interestingly, Incubation of both Hoxc-8 and Smad6 proteins yields a distinct shifted band with a molecular weight higher than Hoxc-8 binding alone, indicating that Hoxc-8 and Smad6 bind to the DNA element as a complex (lane 7). More importantly, the formation of the Smad6-Hoxc-8 complex eliminated the inhibitory effects of Smad1 on Hoxc-8 DNA binding (lane 8). Yeast two-hybrid assays already demonstrated that the interaction between Hoxc-8 and Smad6 is much stronger than that between Hoxc-8 and Smad1 (Table 1).
Hox proteins have been demonstrated to interact with Smad1, but not Smad2 and -3, in response to BMP stimulation (8). To examine whether the interaction between Smad6 and

A Novel Function of Smad6 in the Nucleus 8268
Hoxc-8 is also specifically involved in the BMP signaling pathway, Smad7 was tested for its interaction with Hox proteins (Fig. 2b). Like Smad2 and -3, Smad7 did not interact with either Hoxc-8 or Hoxa-9 (Fig. 2b, lanes 4, 7, and 10). Considering Smad6 as only interacting with phosphorylated Smad1 in the cytoplasm (14), our results also suggest that Smad6 be preferentially involved in BMP signaling. Smad4, the common partner for all receptor-regulated Smads and interaction with Hoxc-8, was examined for the same purpose (Fig. 2c). In comparison with Smad1, the complex of Smad6 and Hoxc-8 did not block the interaction of Smad4 with Hoxc-8 completely (Fig. 2c,  lanes 7 and 9). In fact, Smad4 can only be passively translocated into the nucleus by forming hetero-oligomers with any of the receptor-regulated Smads (29). Again, these results support that Smad6 is an important antagonist preferentially for the BMP signaling pathway.
To investigate whether the Smad6-Hoxc-8 complex inhibits the interaction of Smad1 with Hoxc-8 in activating gene transcription, we utilized the model described in our earlier studies (8). Overexpression of the Smad1-Hoxc-8 interaction domain linked to a nuclear localization signal (Smad1B) stimulates BMP downstream gene expression and induces osteoblast differentiation from osteogenic cells (8,30). When the BMP-inducible construct (Hox-pGL3) was co-transfected in Mv1Lu cells with the Smad1B expression plasmid, the luciferase activity was stimulated in a dose-dependent manner (Fig. 3a). This model provides an ideal assay to directly examine the Smad6 antagonistic function in the nucleus. Because Smad1B mimics BMP-induced gene transcription without BMP receptor phosphorylation involving and interaction with Smad6 of Smad1 (13,14,30), this assay avoids Smad6 inhibitory function in the cytoplasm. Hox-pGL3 construct was co-transfected in Mv1Lu cells with Hoxc-8 and/or Smad6 expression plasmid. As shown in Fig. 3b, overexpression of Hoxc-8 or Smad6 alone moderately inhibited Smad1B-induced transcription activity. Most importantly, co-transfection of both Hoxc-8 and Smad6 plasmids completely abolished the Smad1B-induced luciferase activity. To validate this observation, we transfected Mv1Lu cells with a mutated construct, mHox-pGL3, in which the core nucleotides of the Hoxc-8 binding site were mutated from TAAT to GCCG. Transfection of the mutant construct dramatically reduced Smad1B-induced reporter activity. As expected, the inhibition mediated by co-transfection of Smad6 and Hoxc-8 was also reduced (Fig. 3c).
A Novel Function of Smad6 in the Nucleus 8269