Direct Interaction of Ski with Either Smad3 or Smad4 Is Necessary and Sufficient for Ski-mediated Repression of Transforming Growth Factor-β Signaling*

The oncoprotein Ski represses transforming growth factor (TGF)-β signaling in an N-CoR-independent manner. However, the molecular mechanism(s) underlying this event has not been elucidated. Here, we identify an additional domain in Ski that mediates interaction with Smad3 which is important for this repression. This domain is distinct from the previously reported N-terminal Smad3 binding domain in Ski. Individual alanine substitution of several residues in the domain significantly affected Ski-Smad3 interaction. Furthermore, combined mutations within this domain, together with those in the previously identified Smad3 binding domain, can completely abolish the interaction of Ski with Smad3, while mutation in each domain alone retained partial interaction. By introducing those mutations that abolish direct interaction with Smad3 or Smad4 individually, or in combination, we show that interaction of Ski with either Smad3 or Smad4 is sufficient for Ski-mediated repression of TGF-β signaling. Furthermore our results clearly demonstrate that Ski does not disrupt Smad3-Smad4 heteromer formation, and recruitment of Ski to the Smad3/4 complex through binding to either Smad3 or Smad4 is both necessary and sufficient for repression.

Recently, several studies have proposed structural models of complex formation between Ski and the MH2 domains of Smad3/4 (12,13). The minimal domain in Ski that defines interaction with Smad3 MH2 domain was mapped to the amino terminus (amino acid residues 17-45 in human c-Ski) and was demonstrated to preferentially bind to Smad3 when it was in the active trimeric complex (12). They also reported that this Ski fragment directly binds to Smad3 allowing Smad3-Smad4 heteromer formation in the complex. In another study the crystal structure of the Smad4 binding domain in Ski in complex with Smad4 MH2 domain was resolved and lead to a model in which Ski competes with Smad3 for binding to Smad4 and thus disrupts the functional integrity of the complex (13). Therefore at the present time studies on the formation of complexes involving Ski and Smads remain controversial, and clearly more work is needed to clarify the interaction of Ski with the Smad3-Smad4 complex.
Here we identify an additional domain in Ski that affects interaction with Smad2/3, which is distinct from the previously identified Smad3 binding domain (12). Our results also provide evidence that binding of Ski to Smad3/4 does not disrupt Smad3-Smad4 heteromer formation, rather the Smad heteromer provides Ski with several discrete sites to mediate interaction with the Smad3/4 complex. Thus interaction of Ski with the Smad3/4 complex through binding to either Smad3 or Smad4 is sufficient for the repression activity of Ski on Smaddependent transcriptional activation.

EXPERIMENTAL PROCEDURES
Chemicals and Oligonucleotides-All chemicals were purchased from Sigma or Fisher Scientific unless indicated. Oligonucleotides were purchased from Integrated DNA Technologies.
Cell Culture and Transfection-COS-1 and mink lung epithelial CCL-64 (Mv1Lu) cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal bovine serum, penicillin G (100 units/ml), and streptomycin (100 mg/ml). Expression plasmids were introduced into the cells using FuGENE 6 (Roche Applied Science) as described by the manufacturer.
Reporter Gene Assay-CCL-64 cells in 24-well plate were transfected with luciferase reporter construct 4ϫSBE-Luc (50 ng/well), ␤-galactosidase (TK-␤Gal, 50 ng/well), plus effector plasmids (100 ng/well). Total amount of transfected DNA was equalized with vector (pEGFP-C1). Luciferase activities were measured 24 h after transfection and normalized to the ␤-galactosidase activities according to the manufacturer's instructions (Promega). The results are mean values and S.D. from three independent experiments. For ligand stimulation, cells were treated with TGF-␤1 (1 ng/ml, Sigma) or appropriate solvent 12 h after transfection.

Identification of an Additional Domain in Ski That Mediates
Smad3 Binding-In the previous study, analysis of a Ski single point mutant Ski L110P, which was unable to bind to N-CoR, revealed that the Ski/N-CoR interaction was not required for repression of TGF-␤ signaling, whereas it was for nuclear hormone signaling (4). This finding led us to analyze further how Ski L110P could repress Smad-dependent transcriptional activation without N-CoR/SMRT. As previously reported (4), Ski L110P can associate with Smad4 and Sin3 to a similar extent to the wild-type Ski. Since Smad2/3 are also important factors for Ski-mediated repression, we investigated the interaction of Ski L110P with Smad2/3. We performed co-immunoprecipitation assay using lysates containing Ski (wt-Ski or mt-Ski), Smad2/3, and a constitutively active type I receptor T␤R-I(TD) (Fig. 1A). Surprisingly, the mutation L110P significantly (but importantly not completely) reduced the binding capacity to Smad3 (Fig. 1A, upper panels, lane 6) in comparison with the wild-type ( Fig. 1A, upper panels, lane 3). Similar results were obtained by using Smad2 (Fig. 1A, lower panels). These results indicate that the mutation L110P in Ski substantially affects interaction with Smad2/3 in addition to N-CoR/SMRT.
Analysis of the proximal amino acid sequence of the Leu 110 site revealed that it resides in a hydrophobic residue-rich region that includes a potential amphipathic ␣-helix highly conserved in Ski/Sno/Dachshund family proteins ( Fig. 1B) (14). To test the notion that these residues may facilitate the interaction with Smad3, we performed individual alanine substitution mutagenesis and examined the capacity of these mutants to associate with Smad3. Co-immunopreciptation assays using lysates containing each Ski mutant, Smad3, and T␤R-I(TD) showed that the mutations P107A, Q108A, I109A, V113A, and L114A significantly reduced Ski's binding ability to Smad3 (Fig. 1C, lanes 4, 5, 6, 8, and 9), indicating that these residues are required for effective Smad3 interaction. Interestingly, the L110A mutation had little effect and even appeared to slightly increase the interaction with Smad3 (Fig. 1C, lane 7), while L110P significantly decreased the interaction (Fig. 1C, lane 10). We obtained similar results using Smad2 (data not shown), and recently using glutathione S-transferase-pull-down assays this region was also shown to bind to Smad2. 2 Each mutation tested had no effect on Smad4 binding (data not shown). Our data strongly suggest that this domain from amino acids 107-114, which is distinct from the previously identified Smad3 binding domain (amino acid residues 17-45 in human c-Ski) (12), also mediates interaction of Ski with Smad2/3. Analysis of this region in Ski proteins from other species, and also in more distant family members such as SnoN and Dachshund (Dach2), revealed that this region is very highly conserved. Fig. 1B shows that there is 100% protein sequence identity between human, mouse, and frog c-Ski, and this region is also conserved in human SnoN and human Dach2 (Fig. 1B). Thus, this degree of conservation indicates it is likely that this region plays an important role in Ski-Smad2/3 interactions.
Two Distinct Domains in Ski Mediate Smad3 Interaction-The partial binding ability shown in the Ski mutants described above (Fig. 1) suggested that two distinct domains in Ski may independently mediate interaction with Smad3. In addition they raised the possibility that the remaining Ski-Smad3 binding found in the mutants could be via Ski-Smad4 interaction, which potentially could provide an indirect link between Ski and Smad3 through binding to the Smad3-Smad4 heteromer. To examine these possibilities, we generated a series of Ski mutants having independent mutations for binding to either Smad3, and/or Smad4, or in combination. Mutation of L19A (equivalent mutation L21A in human c-Ski) was shown previously to abolish interaction with Smad3 (12). Mutation W255E (equivalent mutation W274E in human c-Ski) was shown to disrupt Smad4 interaction (13). Therefore, we introduced mutations L19A, L110P, and W255E individually or in combination to generate targeted disruption of Smad interactions ( Fig. 2A).
Co-immunoprecipitation assay was performed using lysates containing Ski, Smad3, and T␤R-I(TD) to examine the effect of each mutation on Smad3 binding (Fig. 2B). L19A (LA) alone significantly reduced but did not completely abolish Smad3 binding (Fig. 2B, lane 2). This was similar to the result with L110P (LP) (Fig. 2B, lane 3), albeit to a greater extent. W255E (WE) alone had no effect on Smad3 binding (Fig. 2B, lane 4). Interestingly, combination of mutations LA/LP completely abolished Smad3 interaction (Fig. 2B, lane 5). Mutations LA/ WE, LP/WE, and LA/LP/WE showed no noteworthy difference from LA, LP, and LA/LP, (Fig. 2B, lanes 6 -8), indicating that mutation WE has no effect on Smad3 interaction. This result was consistent with the previous observations that Ski interacts with Smad3 and Smad4 through discrete binding sites (7, 8, 13). To confirm the effect of the mutations on Smad4 binding, similar co-immunoprecipitation assay was performed using Smad4 in-2 L. Li and E. Stavnezer, personal communication.

FIG. 1. Identification of an additional domain in Ski that mediates Smad2/3 interaction. A, L110P mutation in c-Ski affects
Smad2/3 interaction. Wild-type and mutant (L110P) c-Ski (wt-Ski and mt-Ski) were co-immunoprecipitated with Myc-Smad3 (upper panels) or Myc-Smad2 (lower panels) using anti-Myc antibody or mouse normal IgG. The immune complexes (IP) and 10% input were detected by Western blotting using combination of anti-T7 and anti-Myc antibody. B, alignment of the proximal protein sequence at position L110 in chicken c-Ski and the equivalent region of human, mouse, frog c-Ski, human SnoN, and Dachshund2 (GenBank TM accession numbers P49140, CAA33288, AAL30825, CAA48642, CAA33289, and AAL26561, respectively). Black shading indicates identity, and gray shading indicates similarity. Numbering above corresponds to amino acid positions in chicken c-Ski. An ␣-helix structure beneath was determined by Kim et al. (14). C, alanine substitution analysis to identify residues in the region that affect Smad3 interaction. A series of Ski point mutants as indicated were co-immunoprecipitated with Myc-Smad3 using anti-Myc antibody. The immune complexes (IP) (upper panels) and 10% input (lower panels) were detected as described for A. stead of Smad3 (Fig. 2C). LA, LP, and LA/LP mutations had no substantial effect on Smad4 binding (Fig. 2C, lanes 2, 3, and 5), while WE alone was sufficient for complete abrogation of Smad4 interaction (Fig. 2C, lane 4), as reported previously (13). These results indicate that while each Smad3 binding mutation alone (LA or LP) is not sufficient, combined mutations (LA/LP) disrupt Smad3 interaction independent of the Smad4 binding mutation (WE). Taken together, our data strongly suggest that under these experimental conditions, two distinct domains in Ski can independently mediate interaction of Ski with Smad3. Interestingly, Mizuide et al. (15) recently identified two separate sites in the MH2 domain of Smad2/3, which mediate interaction with Ski/ Sno. Thus, it is possible that two distinct interfaces of Ski/Sno and Smad2/3 facilitate interaction with each other.
Effects of Independent or Combined Mutations That Affect Interaction of Ski with Smad3 or Smad4 on Smad3/4-dependent Transcription-To investigate the effect of each Ski mutation on Smad-mediated transcriptional activation, we performed an established reporter gene assay using Smad3/4dependent 4ϫSBE-Luc reporter construct in CCL-64 cells (9) (Fig. 3). Surprisingly, regardless of the significantly reduced binding ability to Smad3 as evidenced by the co-immunoprecipitation assay (Fig. 2B), LA alone, LP alone, and LA/LP repressed transcription as effectively as the wild-type (WT) (Fig. 3, LA, LP, LA/LP, and WT). Furthermore, WE alone also repressed transcription to a similar extent to WT (Fig. 3, WE and WT), regardless of the complete lack of interaction with Smad4 (Fig. 2C). However, combined mutations LA/WE and LP/WE partially reversed repression (up to 50 and 30% of vector control, respectively) (Fig. 3, LA/WE and LP/WE). Clearly indicating that these mutants retain partial repression activity. In contrast, LA/LP/WE, which completely abolished individual interaction with Smad3 and Smad4 (Fig. 2, B and  C), nullified the repression (Fig. 3, LA/LP/WE). These results indicate that interaction of Ski with either Smad3 or Smad4 is necessary and sufficient for repression of Smad3/4-dependent transcriptional activation.
Ski Allows Direct Smad3-Smad4 Heteromer Formation, Enabling Recruitment of Ski to the Smad3/4 Heteromeric Complex through Either Smad3 or Smad4 Binding-Smad3 and Smad4 form heteromers in response to TGF-␤ signaling (16). Thus, it is possible that Ski mutants, which lack direct interaction with either Smad3 or Smad4, can be recruited to Smad complex through the Smad3-Smad4 heteromer. To test this notion, we performed co-immunoprecipitation assay (Fig. 2, B and C) in the presence of both Smad3 and Smad4 (Fig. 4A). Interestingly LA/LP in the presence of Smad4 allowed this mutant to interact with Smad3 (Fig. 4A, lane 5), in clear contrast to its inability of direct interaction with Smad3 (Fig. 2B, lane 5). Similarly, in the presence of Smad3 the WE Ski mutant could now interact with Smad4 (Fig. 4A, lane 4), in contrast to its inability to directly interact with Smad4 (Fig. 2C, lane 4). However, LA/WE and LP/WE significantly reduced interaction with Smad3 and Smad4 (Fig. 4A, lanes 6 and 7), and LA/LP/WE completely abolished the interaction (Fig. 4A, lane 8). The partial interaction shown by LA/WE and LP/WE (Fig. 4A, lanes  6 and 7) is consistent with the partial transcriptional repression activity seen in the luciferase assay (Fig. 3, LA/WE and LP/WE). It is possible that exogenously expressed mutant Ski can dimerize with endogenous Ski through its C-terminal coiled-coil domain (17); thus, the presence of endogenous wildtype Ski may contribute to recruit Smad3/4 to the complex. However, this possibility is ruled out, as the interactions are not observed in our co-immunoprecipitation assays using the triple mutant form of Ski (LA/LP/WE), which retains this coiled-coil domain (lanes 8, Fig. 4A and Fig. 2, B and C). These results provide evidence that binding of Ski still permits Myc-Smad3 was co-immunoprecipitated with T7-tagged Ski mutants as indicated using anti-T7 antibody. The immune complexes (IP) (upper panels) and 10% input (lower panels) were detected by Western blotting using anti-Myc or anti-T7 antibody. C, mutations LA, LP, and LA/LP minimally affect interaction of Ski with Smad4. Myc-Smad4 was coimmunoprecipitated with T7-tagged Ski mutants as indicated using anti-T7 antibody. The immune complexes (IP) (upper panels) and 10% input (lower panels) were detected by Western blotting using anti-Myc or anti-T7 antibody. Smad3-Smad4 heteromer formation, and interaction with either Smad3 or Smad4 is necessary and sufficient for Ski-mediated transcriptional repression in TGF-␤ signaling.
Our finding that there are two distinct domains in Ski involved in Smad2/3 interaction allowed us to generate a series of informative Ski mutants completely lacking direct binding to either Smad3 (LA/LP), Smad4 (WE), or both (LA/LP/WE) (Fig.  4B). These Ski mutants have provided strong evidence that either Smad3 or Smad4 interaction with Ski is necessary and sufficient for repression of TGF-␤ signaling. This is mediated by the Smad3-Smad4 heteromer enabling indirect links between Ski mutants (LA/LP and WE) and Smads (Smad3 and Smad4, respectively). This notion was further supported by the fact that only the Ski mutant LA/LP/WE, which binds to neither Smad3 nor Smad4, completely abolished repression activity of Smadmediated transcriptional activation (Fig. 3, LA/LP/WE). Contrary to our model, Wu et al. (13) recently proposed that Ski disrupts heteromeric complex formation of Smad3-Smad4, based on a structural observation that Ski binds to the L3 loop of Smad4. It is not clear why Smad3-Smad4 complex formation was not detected in their biochemical analyses, using equivalent mutants to our LA, WE, LA/WE mutants in the presence of both Smad3 and Smad4. In contrast, their report of the effect of the mutations on Smad-mediated transcriptional activation appeared consistent with our results, in that LA and WE repressed as efficiently as the wild-type Ski, and LA/WE still retained partial repression activity (up to 50% of vector control) (13). While their proposed model did not completely explain the transcriptional properties of those mutants, our model, in which Ski allows direct Smad3-Smad4 heteromer formation in the complex, is consistent with the transcriptional properties of these mutants. Consistent with our results, another group reported the heteromer formation of Smad3-Smad4 in the presence of Ski (12).
We have shown that Ski requires interaction with either Smad3 or Smad4 for repression of TGF-␤ signaling. As previously shown by several groups using gel shift assays (7)(8)(9)(10)18), Ski can be recruited to the Smad-binding elements (SBE) on DNA through interaction with Smads. However, how Ski can exert its repression activity on Smads remains unknown. We recently reported that this Ski-mediated repression is N-CoRindependent (4). Ski was shown to link Smad3 and Sin3, forming a complex including HDAC1 (8). Thus it is possible that corepressors other than N-CoR play a role in Ski-mediated repression in TGF-␤ signaling. Furthermore, Ski was suggested to interfere with Smad-CBP/p300 coactivator complex formation (8) and thus may function by inhibiting the binding of coactivators rather than recruiting corepressors. Hence these lines of repression by Ski cannot be ruled out either. Regardless of the details our data clearly show that binding of Ski to the Smad3/4 complex is central to its ability to repress.