Transcriptional repression by v-Ski and c-Ski mediated by a specific DNA binding site.

The Ski oncoprotein has been shown to bind DNA and activate transcription in conjunction with other cellular factors. Because tumor cells or myogenic cells were used for those studies, it is not clear that those activities of Ski are related to its transforming ability. In this study, we use a nuclear extract of c-ski-transformed cells to identify a specific DNA binding site for Ski with the consensus sequence GTCTAGAC. We demonstrate that both c-Ski and v-Ski in nuclear extracts are components of complexes that bind specifically to this site. By evaluating the features of the sequence that are critical for binding, we show that binding is cooperative. Although Ski cannot bind to this sequence on its own, we use cross-linking with ultraviolet light to show that Ski binds to this site along with several unidentified cellular proteins. Furthermore, we find that Ski represses transcription either through upstream copies of this element or when brought to the promoter by a heterologous DNA binding domain. This is the first demonstration that Ski acts as a repressor rather than an activator and could provide new insights into regulation of gene expression by Ski.

The Ski oncoprotein has been shown to bind DNA and activate transcription in conjunction with other cellular factors. Because tumor cells or myogenic cells were used for those studies, it is not clear that those activities of Ski are related to its transforming ability. In this study, we use a nuclear extract of c-ski-transformed cells to identify a specific DNA binding site for Ski with the consensus sequence GTCTAGAC. We demonstrate that both c-Ski and v-Ski in nuclear extracts are components of complexes that bind specifically to this site. By evaluating the features of the sequence that are critical for binding, we show that binding is cooperative. Although Ski cannot bind to this sequence on its own, we use cross-linking with ultraviolet light to show that Ski binds to this site along with several unidentified cellular proteins. Furthermore, we find that Ski represses transcription either through upstream copies of this element or when brought to the promoter by a heterologous DNA binding domain. This is the first demonstration that Ski acts as a repressor rather than an activator and could provide new insights into regulation of gene expression by Ski.
c-Ski is an 84-kDa nuclear protein that has been shown to bind DNA when a complementing activity is supplied by a nuclear extract (1). v-Ski is a truncated form of the c-Ski protein that is missing 20 amino acids from the amino terminus and 292 amino acids from the carboxyl terminus (2)(3)(4). This truncation, which removes a carboxyl-terminal dimerization domain, plays no role in the activation of ski as an oncogene (5,6). Overexpression of either c-ski or v-ski induces transformation in chicken embryo fibroblasts (CEFs) 1 and either muscle differentiation or transformation in cultured quail embryo fibroblasts (QEFs), depending on the growth conditions (7,8). The ability of ski to induce both transformation and muscle differentiation in the same cells (QEFs) is an intriguing para-dox and suggests that ski plays a pivotal role in regulating cell growth and differentiation. That such a role for ski is conserved in other organisms is demonstrated by the phenotype of v-ski transgenic mice, which have increased muscle mass caused by hypertrophy of type II fast muscle fibers (9). Ski has been shown to affect the proliferation and differentiation of cells outside of the myogenic lineage as well. For example, v-ski transforms a myeloid-erythroid hematopoietic multipotential progenitor cell from avian bone marrow (10,11). Recently, mice that are homozygous for a null mutation in the c-ski gene have been generated. These mice die at birth and show a variety of developmental defects, including defective closure of the rostral neural tube and decrease in skeletal muscle mass (12).
Because of its effect on muscle differentiation, upstream regulatory regions of muscle-specific genes were used in some initial studies of Ski's transcriptional regulatory properties. One such study showed that Ski stimulates transcription from the enhancers of both myosin light chain 1/3 and muscle creatine kinase by 2-3-fold in myoblasts in an E-box-dependent fashion (13). However, given the diverse biological consequences of Ski overexpression and loss of function, it is likely that Ski also regulates genes outside the myogenic lineage. Kelder and co-workers (14) have shown that v-Ski, overexpressed in mouse L-cells, activates the SV40, human cytomegalovirus immediate early, and RSV long terminal repeat enhancers (14). In addition, we have shown that Ski can interact with the DNA binding site for the nuclear factor I (NFI) transcription factor family by protein-protein interaction with the NFI protein and potentiate NFI-stimulated transcription of a reporter that has multimerized NFI binding sites upstream of a TATA box element (15). 2 Recently, it has become clear that some transcription factors can regulate gene expression by interacting with multicomponent protein complexes (16 -18). This allows a single transcription factor to interact with different binding sites, depending on the cellular context. The ability of Ski to affect transcription through E-box elements and to interact with the NFI binding site through interaction with the NFI protein suggests such a mechanism for Ski transcriptional regulation. The interaction of Ski with the NFI binding site was identified by cyclic amplification and selection of targets (CASTing), using a nuclear extract from v-ski-transduced mouse L-cells as the source of Ski protein (15,19). Those cells are highly transformed, and v-ski overexpression appears to suppress their transformed phenotype. 3 In the present work, we describe the identification of a second DNA binding site for the Ski protein that is more likely to be relevant to the process of transformation. For this purpose, we used a nuclear extract from c-ski-transformed CEFs (c-ski-CEFs), because CEFs, unlike L-cells, are transformed by overexpression of c-ski or v-ski. The binding site we have identified using this strategy has the sequence GTCTA-GAC. There are no known cellular factors that have been previously characterized as binding to this sequence. We show that both v-Ski and c-Ski bind this element, and we examine the affinity of the interaction. Here we also show that Ski is able to repress transcription through multimerized copies of the GTCTAGAC binding site cloned upstream of a minimal promoter. This is the first demonstration of Ski acting as a repressor rather than an activator and could provide insight into the ability of Ski to induce both transformation and differentiation in cells depending on the cellular environment.

MATERIALS AND METHODS
Cell Culture, Preparation of Nuclear Extracts, and Western Blotting-Culture and infection of chicken embryo fibroblasts (CEFs) with the replication-competent avian retrovirus, RCASBP, carrying c-ski or v-ski was performed as described previously (7,20). Nuclear extracts were prepared from normal CEFs or CEFs infected with RCASBPc-ski or RCASBPv-ski retroviruses using the method of Dignam (21). Nuclear extracts were analyzed for expression level and integrity of Ski proteins by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blotting with the anti-Ski monoclonal antibody (mAb), G8 (7), and the chemiluminescent detection system of Tropix. The relative amounts of exogenous or endogenous Ski in the extracts was determined by densitometry of exposed film.
Oligonucleotides-The oligodeoxynucleotide used for identification of the c-Ski binding site (CASTing oligonucleotide) contains a central 22 bases of random sequence flanked by nonrandom sequence on both sides. The 16-base nonrandom sequence at the 5Ј-end contains a recognition site for the restriction enzyme BamHI, and the 14-base nonrandom sequence at the 3Ј-end contains a recognition site for PstI. The double-stranded CASTing oligonucleotide that was used in the first round of binding and selection was prepared by annealing a 16-base pair primer (RO16 -2) to the 3Ј-end of the CASTing oligonucleotide and extending with the Klenow fragment of DNA polymerase I. Oligomers that were selected after each round of CASTing were amplified using primers RO16 and RO16 -2. The sequences of these oligonucleotides are as follows: CASTing oligonucleotide, ggcggatccacctaca . . . N 22 . . . tgtgcactgcagtg; RO16, ggcggatccacctaca; RO16 -2: gccactgcagtgcaca. The sequences of oligonucleotides used as probes and competitors in electrophoretic mobility shift assays (EMSAs) are given in the figures.
Selection of Ski-bound Oligonucleotides-Anti-Ski mAb beads were produced by overnight incubation of Dynal sheep anti-mouse IgGcoated magnetic beads with either G8 or M6 mAb in antibody buffer (1 ϫ phosphate-buffered saline, 0.1% Nonidet P-40, 0.02% sodium azide, and 100 g/ml bovine serum albumin). Following incubation, the beads were washed with antibody buffer and resuspended (6.7 ϫ 10 8 beads/ml) in antibody buffer without Nonidet P-40. c-ski-CEF nuclear extract (20 g) was incubated for 20 min at room temperature with 5 g of double-stranded CASTing oligonucleotide and 5 g of poly(dI-dC) in CASTing buffer (25 mM Hepes, pH 7.5, 100 mM NaCl, 10% glycerol, 0.2 mM EDTA, 0.1% Nonidet P-40, 2 mM dithiothreitol, 0.5 mM phenylmethylsulfonyl fluoride, and 0.5 mM Pefabloc from Boehringer Mannheim). Anti-Ski mAb-coated beads (200 l), or sheep ␣-mouse IgG-coated beads (200 l) for the mock reaction, were equilibrated in CASTing buffer and then resuspended in a 10-l volume of this same buffer. The DNA binding reaction was then added to the beads, and the mixture was incubated for 1 h at 4°C on a rotating platform. The beads were then washed six times with 250 l of CASTing buffer and resuspended in 100 l of PCR buffer. A portion of this suspension (20 l) was added to a PCR reaction with a total volume of 30 l and amplified using Boehringer Mannheim Taq polymerase and the primers RO16 and RO16-2. To avoid overamplification, triplicate reactions were prepared and amplified for 10, 15, or 20 cycles (19). One-third of the amplified material (10 l) from each reaction was run on a 12% polyacrylamide gel. The reaction in which amplified material was first visible on the ethidium bromide-stained gel was determined, and the remaining 20 l from this reaction was used as the starting material for the next round of CASTing. Five additional rounds of binding, washing, and amplification were performed.
Cloning and Sequencing of Selected Oligonucleotides-After the sixth round of CASTing, the amplified oligomers were digested with BamHI and PstI and cloned into a modified pUC18 vector (15). This modification caused a disruption of the translational reading frame, which rendered the pUC-encoded ␣-peptide of ␤-galactosidase nonfunctional. Insertion of a single oligomer fragment restores the reading frame, so that recombinant plasmids produce blue bacterial colonies on plates containing 5-bromo-4-chloro-3-indolyl ␤-D-galactopyranoside and isopropyl ␤-thiogalactoside (22). Positive (blue) colonies were picked at random, and the inserted DNA sequenced by the dideoxy chain termination method (23).
EMSA and Competition Analysis-32 P-Labeled probe was prepared according to the method of Mertz and Rashtchian (24). Briefly, oligomer inserts were amplified from pUC18 clones using the RO16 and RO16 -2 primers in the presence of 6 mM dCTP, dGTP, and dTTP, and 50 Ci of 3000 Ci/mmol [␣-32 P]dATP in a 20-l PCR reaction. For the GTCT/2 probe, clone number 6-6 was amplified. A single copy insert was produced by recombining clones 6-6 and 6-9 at the XbaI site, which comprises the central six base pairs of the binding site. The GTCT/1 probe was made by PCR amplification of the resulting single copy insert. GTCT/2 and GTCT/1 probe sequences are shown in Fig. 2A. GTCT/1.5 probes were made by PCR amplification of the pUC clones, numbers 6-7 and 6-15, and the sequences are shown in Fig. 2C. PCR labeling reactions were electrophoresed on a 7% polyacrylamide gel. Probe bands were cut out, and probe was eluted by soaking in CASTing buffer without glycerol overnight. Final probe concentration was approximately 2 fmol/l with a specific activity of 5 ϫ 10 7 cpm/pmol.
The probe (4 fmol) was added to a 20-l binding reaction containing 8 g of nuclear extract, 500 ng of poly(dI-dC), and 500 ng of RsaIdigested bovine DNA. Final salt and glycerol concentrations were adjusted to CASTing buffer concentrations. After a 20-min incubation at room temperature, reactions were loaded directly onto a pre-electrophoresed (90 min at 10 V/cm) 4% 60:1 polyacrylamide gel. The gels were electrophoresed at 10 V/cm for 30 min and then at 4 V/cm until the xylene cyanol loading dye was three-fourths of the way down the gel (approximately 14 h). The gel and chamber buffer was 0.5 ϫ TBE (45 mM Tris, 45 mM boric acid, and 1.25 mM EDTA). For antibody supershifts, mAb G8 or M6 (0.5 l of protein A-purified mAb at 2 g/l) were added to the binding reactions after a 15-min incubation with probe. Samples were vortexed and incubated for an additional 10 min at room temperature prior to loading onto the gel.
GST-⌬Ski1-136 protein, for supershift competition, was prepared as described previously for other GST-Ski fusion proteins (5). This protein was added to the binding reactions prior to the addition of probe.
For competition assays, the binding conditions were as described above, except that unlabeled competitors (amounts described in the figure legends) were added and incubated for 5 min on ice prior to the addition of the 32 P-labeled probe. Oligonucleotide competitors were prepared by annealing the oligonucleotides shown (Fig. 4D) to a complementary strand and purifying the duplexes by gel electrophoresis and elution as described for probe purification. PCR competitors were prepared by amplification of inserts from the indicated clones with the RO16 and RO16 -2 primers (Figs. 2A and 4B). Reactions were precipitated and resuspended at appropriate concentrations for competition. Relative binding compared with the reaction with no competitor was determined by quantitation of complexes 1 and 2 by PhosphorImager (Molecular Dynamics) analysis of dried gels as described in the legend for Fig. 4.
Plasmid Construction-The tkCAT reporter construct was a gift from H. L. Grimes and P. N. Tsichlis (25) and contains the fragment of the herpes simplex virus thymidine kinase promoter from Ϫ105 to ϩ51. Oligonucleotides complementary to the GTCT/2, GTCT/1, and mutGC/2 oligonucleotides shown in Fig. 4D were synthesized such that the annealed oligonucleotides would have a BamHI-compatible overhang at one end and a BglII-compatible overhang at the other end. The doublestranded binding site oligonucleotides were then cloned into the BamHI to BglII-digested tkCAT reporter plasmid, placing the binding site immediately upstream of the TK promoter.
The G5tk-luciferase reporter (pGL3.2G5tkLuc) was generously provided by S. B. Cohen in the Stavnezer laboratory. pGL3.2G5tkLuc is based on the pGL3 Basic plasmid (Promega). A HindIII to BglII fragment of the pG5tkCAT plasmid (26), in which the HindIII site had been filled in using dATP and dGTP only, containing the TK promoter and the five Gal4 binding sites, was cloned into NheI to BglII-digested pGL3 Basic in which the NheI site had been filled in using dCTP and dTTP only. To eliminate a putative Gal4 binding site in the luciferase cassette (27), a HindIII to BsrG1 segment of pGL3 Basic was replaced with a similar fragment from pGL2 Basic.
The RSVPL, RSVc-ski, RSVv-ski, and SG424c-ski expression plasmids were generously provided by P. Tarapore in the Stavnezer laboratory. The RSVPL expression plasmid was derived from RSVCAT (28). The chloramphenicol acetyltransferase (CAT) gene was removed as a HindIII to BanI fragment, the HindIII and BanI sites were filled in, and a PCR-amplified fragment containing the Bluescript (Statagene) polylinker was cloned into this location. This fragment was amplified using the SK and KS primers (Stratagene). An XbaI fragment containing c-ski from pMexNeoc-ski29 was cloned into the XbaI site of the RSVPL polylinker, and a BglII fragment containing v-ski from pCRpolski was cloned into the BamHI site to generate the RSVc-ski and RSVv-ski expression plasmids, respectively. The SG424c-ski expression plasmid was generated by filling in the ends of an NcoI to XbaI fragment containing c-ski29 from RSVc-ski and cloning this into the SmaI site of SG424.
The introduction of a unique NheI restriction site into the coding regions of the v-ski mutants, ⌬AH2 and ⌬AH4, has been described previously (29). These two mutants were recombined at the NheI site such that the region between was deleted. This deleted v-ski is designated ⌬AH2H4. To generate the plasmid for the expression of the glutathione S-transferase (GST) fusion with Ski amino acids 1-136, an NcoI to KpnI fragment from this deleted version of v-ski, in which the KpnI site was blunted by treatment with T4 DNA polymerase, was then cloned into NcoI to SmaI-digested pGEX-2T.
Reporter Gene Assays-The UMN-SAH/DF#1 chicken fibroblast cell line was a gift from D. Foster of the University of Minnesota. These cells were seeded at a density of 2.5 ϫ 10 5 cells/35-mm plate in Dulbecco's modified Eagle's medium with 10% fetal bovine serum. Cells were cultured overnight prior to transfection. Triplicate or duplicate plates were co-transfected with each combination of CAT reporter (600 ng) and the indicated amounts of ski expression plasmid. RSVPL or pUC18 plasmid DNA was added, as indicated in the figure legends, so that the total amount of ski-containing plus empty vector DNA was always 600 ng. The DOTAP liposomal transfection reagent was used according to the manufacturer's directions (Boehringer Mannheim). The DOTAP-DNA mixture was left on the cells for 12 h, at which time they were washed and refed. Cells were harvested 60 -72 h after transfection. For CAT assays, cells were harvested and lysates were prepared as described previously (15). CAT assays were performed by the liquid scintillation method as described previously (15,30). For luciferase assays, lysates were prepared and assayed for luciferase activity using the Promega luciferase assay system according to the manufacturer's directions. The protein content of each lysate was determined by the Bio-Rad protein assay, and these values were used to normalize the CAT and luciferase activity data. In addition, at least two different DNA preparations were tested for all expression plasmids and reporters. We did not employ an internal control for transfection efficiency, because our earlier work had shown that Ski activates expression of all the commonly used control plasmids.
Cross-linking and Immunoprecipitation of Ski DNA Binding Complex-The probe for DNA-protein cross-linking was prepared by annealing an excess of GTCTX primer (TGCTAGTCTAGAC) to 5 pmol of GTCT/2 binding site oligonucleotide ( Ci of 3000 Ci/mmol [ 32 P]dATP. Because the GTCTX primer anneals to one of the GTCTAGAC elements, bromodeoxyuridine (BrdUrd) is incorporated into the complementary strand of only a single GTCT element. The filled in probe was purified by electrophoresis on a 7% polyacrylamide gel as described above for EMSA probes. The specific activity of the resulting probe is approximately 1 ϫ 10 7 cpm/pmol. Probe binding reactions were carried out the same as for EMSA except that final probe concentration was 2.5 nM. Samples were transferred to a 96-U-bottom-well microassay plate for incubation. After incubation, this plate was placed on ice and irradiated with ultraviolet light at a distance of 10 cm for 5 min. in a UV Stratalinker 1800 (Stratagene). SDS-loading buffer was added to some samples, which were boiled and loaded directly onto a 7% SDS-polyacrylamide gel. For immunoprecipitations, five 20-l cross-linking reactions were pooled, diluted 1:3, and adjusted to either high stringency buffer (20 mM Tris, pH 7.5, 150 mM NaCl, 0.5% SDS, 0.5% Nonidet P-40, 0.5% sodium deoxycholate, 1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride) or low stringency buffer (20 mM Hepes, pH 7.9, 100 mM NaCl, 0.5% Nonidet P-40, 10% glycerol). 1 l of G8 or M6 mAb ascites or 1 l of the polyclonal antibody 32360 (a gift from S. Hughes at NCI-Frederick Cancer Research and Development Center), which reacts with fulllength c-Ski but not with v-Ski 4 was added to the diluted cross-linking reaction. These reactions were incubated on ice for 1 h, 50 l of a 1:1 protein A-agarose slurry was added, and incubation was continued for an additional 1 h at 4°on a rotating platform. The beads were then washed 5 times with 1 ml of high stringency buffer or low stringency buffer and resuspended in 60 l of SDS-loading buffer. Samples were boiled and centrifuged, and the bead supernatant was loaded onto the same gel along with the samples that were not immunoprecipitated.

In Vitro Selection of a DNA Binding Site Using a Nuclear
Extract of c-ski-CEFs-Nuclear extracts from CEFs infected by a c-ski retrovirus (RCASc-ski29) were used as the source of Ski protein for binding site selection. DNA oligomers bound by Ski-containing complexes were purified using magnetic beads coated with the anti-Ski mAbs, G8 and M6. As a negative control, magnetic beads coated with sheep anti-mouse IgG were used in a parallel selection.
Starting with a 54-base pair double-stranded oligomer containing a central 22 base pairs of degenerate sequence (see "Materials and Methods"), six cycles of DNA binding, antibody purification, and amplification were carried out. The population of oligomers purified with G8 or M6 mAbs was cloned, and randomly selected clones were sequenced. The resulting sequences (Fig. 1, A and B, indicated by 6-N) revealed a very strong consensus for the dyad symmetrical sequence GTCTA-GAC; 24 out of 29 selected oligomers contain this sequence, and among these, 18 carry two tandem repeats of the sequence (Fig.  1A). This 16-base consensus can be extended further to include an A at the 3Ј-end. In most cases, the other end of the consensus directly abuts the nonrandom flanking sequence, so it is difficult to determine if there is any base preference at this end. We refer to the tandem repeat version of the consensus as a GTCT/2 binding site. Oligomers from the mock selection with no mAb were also cloned and sequenced, and no matches with the GTCTAGAC element were found in this group.
Each of the remaining six GTCT-containing oligomers, identified after round six, contain a single perfect copy of the GTCTAGAC element plus at least one additional partial copy (Fig. 1B). This version of the binding site, referred to as GTCT/ 1.5, is underrepresented relative to GTCT/2 (Fig. 1, A and B), probably because six rounds of CASTing selected for only the highest affinity sites. Since no oligonucleotides contain only a single GTCT dyad (GTCT/1), we felt that the partial site must have contributed to selection and that analysis of these sequences in GTCT/1.5 elements would provide information as to which bases are most critical for binding by the Ski-containing complex. We reasoned that this information should be obtainable by cloning and sequencing oligomers that had been saved after the fifth round of CASTing. As shown in Fig. 1B (indicated  by 5-N), this was indeed the case; 6 out of 12 oligomers cloned after five rounds are of the GTCT/1.5 type, only two contain the perfect GTCT/2 binding site (Fig. 1A), and none contains only GTCT/1. These results suggest that there is a progressive enrichment for the GTCT/2 version of the binding site that does not predominate until the final round of selection (round six). An alignment of GTCT/1.5 sites identified after five and six rounds of selection (Fig. 1B) yields the consensus TGTCTAGACAN 1-3 TGTCTGG. In the partial sites, the half of the inverted repeat closest to the whole site is completely conserved, whereas the last two bases in the inverted repeat are less conserved. Surprisingly, A 5 is replaced by a G in all of the partial sites. This substitution never occurs in the neighboring whole site nor in any of the GTCT/2 elements.
Protein Complexes in the c-ski-CEF Extract Show Specific Interaction with the GTCTAGAC Binding Site-The complexes that form with the GTCT/2 and GTCT/1 probes and c-ski-CEF nuclear extract are compared by EMSA in Fig. 2A. The GTCT/2 probe yields four shifted bands ( Fig. 2A, lanes 1-8). Two of these (labeled 1 and 2) can be competed efficiently with an excess of unlabeled GTCT/2 ( Fig. 2A, lanes 3-5). A third band (complex X), which is also competed efficiently, overlaps with a nonspecific complex on this gel ( Fig. 2A, lanes 3-5). The fourth and fastest migrating band (complex Y) is competed with equal efficiency by both specific and nonspecific competitor and is not reproducible ( Fig. 2A and data not shown). Unlabeled GTCT/1 also competes for the formation of complexes 1, 2, and X although less efficiently than GTCT/2 ( Fig. 2A, lanes 6 -8). The GTCT/1 probe yields only one specific shifted band, which corresponds in mobility to complex 1 produced by the GTCT/2 probe ( Fig. 2A, lanes 9 -16). This suggests that complex 1 formed with the GTCT/2 probe results from binding of a protein complex to only one of the two tandem 8-base pair elements, whereas complex 2 results from the binding of a protein complex to both copies of the GTCT element. The fact that the GTCT/1 band and complex 1 with GTCT/2 are much fainter than the complex 2 band with the GTCT/2 probe suggests that cooperativity plays a role in the binding of this complex to the tandem GTCTAGAC element.
The lower affinity interaction with the GTCT/1 probe helps explain why none of the oligomers we sequenced contained just a single copy of the 8-base pair inverted repeat and suggests that the partial sites in the GTCT/1.5 elements should participate in binding. To determine whether this is the case, we compared gel shift complexes that form with two GTCT/1.5 probes to the complexes that form with the GTCT/1 and GTCT/2 probes (Fig. 2C). Both GTCT/1.5 probes form two bands (Fig. 2C, lanes 5 and 7), which are identical in mobility to GTCT/2 complexes 1 and 2 (Fig. 2C, lanes 1 and 3). This suggests that the complexes that bind to the nontandem partial repeat probes and the GTCT/2 probe have the same composition. There is some variation in the ratio of complexes 1 and 2 that is dependent on the divergence of the partial sequence from the consensus. With the GTCT/2 probe, complex 2 predominates, and this is also true, but to a lesser extent, for the 6-7 probe, which has a 6 of 9 match with the consensus. With the 6-15 probe (5 of 9 match to the consensus), the amounts of the two complexes are approximately equal. Therefore, the closer the second copy of the binding site is to a perfect consensus GTCT/1 site, the greater the amount of complex 2 formed. The amount of binding to either of the GTCT/1.5 probes is much greater than the amount of binding to the GTCT/1 probe, demonstrating that the separation of the two copies of the inverted repeat element by a three-nucleotide spacer does not prevent binding to these sites in a cooperative manner.
Ski is Part of a Complex That Binds to the GTCTAGAC Element-To determine if Ski protein participates in the specific GTCTAGAC binding, antibody supershifts were performed (Fig. 2, B, C, and D). Both complexes 1 and 2 produced by nuclear extract from c-ski-CEFs are supershifted by anti-Ski mAbs, G8 or M6 (Fig. 2B, lanes 1-3). On this gel, and most others, complex X is resolved into at least two bands (Fig. 2B,  lanes 1-9) that are specifically competed by cold GTCT competitor (data not shown). Complex X (seen more clearly in lanes 4 -9) is not supershifted by the anti-Ski mAbs and must contain proteins other than Ski that bind specifically to the GTCTA-GAC sequence.
To determine whether endogenous c-Ski binds the GTCT element, EMSAs were performed with a nuclear extract from normal CEFs. In this case, the GTCT/2 probe produces a single complex that is supershifted by both anti-Ski monoclonal antibodies (Fig. 2B, lanes 4 -6 and lanes 7-9). This complex is much fainter than either of the complexes formed with extract from RCASc-ski-transformed CEFs and can be seen more clearly after a longer exposure (Fig. 2B, lanes 7-9). The endogenous complex has a mobility that is intermediate between the two c-ski-CEF complexes (compare complex 1 and the location of asterisks). Endogenous Ski from mouse and human cell lines also binds the GTCTAGAC element (data not shown).
The specificity of the G8 mAb supershift is demonstrated in Fig. 2D. These results show that the supershift formed with G8 mAb is eliminated by adding GST-⌬Ski1-136 fusion protein to the reaction (Fig. 2D, lanes 1-5), and the extent of this effect is dependent on the amount of fusion protein added. This fusion protein consists of the first 136 amino acids of Ski fused to GST and contains the epitope for the G8 mAb. The supershift is not affected by the addition of the GST protein by itself (Fig. 2D,  lanes 6 -8). Furthermore, the effect of the GST-Ski fusion protein is not due to indirect action on the complex itself, because it has no effect when added to the shift reaction in the absence of antibody (Fig. 2D, lanes 11 and 12).
A nuclear extract of v-ski-transformed CEFs (v-ski-CEFs) produces a single complex with the GTCT/2 probe that is supershifted by M6 mAb but is disrupted by G8 mAb (Fig. 3A,  lanes 2-4). The v-Ski complex has a greater mobility than either of the c-Ski complexes, which run near the top of the gel shown in Fig. 3A, lane 1. The fact that the v-Ski gel shift is disrupted by the G8 mAb, although the c-Ski gel shift is not, indicates that c-Ski forms a more stable complex with the GTCT binding site than v-Ski. Neither v-Ski nor endogenous Ski form a detectable complex with the GTCT/1 probe (data not shown), suggesting that in both of these cases, the single complex that forms with the GTCT/2 probe results from binding to both copies of the GTCT binding site. The faster migration of the v-Ski complex relative to the exogenous c-Ski complex 2 is expected (Fig. 3A, lanes 1 and 2); whereas, the faster migration of the endogenous c-Ski complex relative to its exogenous counterpart is more surprising (Fig. 2B, lanes 1, 4, and 7). It is possible that this complex contains different proteins than the exogenous c-Ski complex 2 and that high level expression of c-Ski in the CEFs actually changes the composition of the GTCT/2 binding complex in some way.
The relative expression levels of RCASc-ski, RCASv-ski, and endogenous ski as determined by Western blotting are shown in Fig. 3B. c-Ski and v-Ski are expressed at similar levels, so the differences in quantity of binding to the GTCT/2 probe cannot be accounted for by differences in protein expression. However, endogenous Ski is expressed at approximately 1 ⁄10th to 1 ⁄20th the level of the retrovirally expressed proteins, which accounts for the very faint gel shift band produced by nuclear extract from normal CEFs. Due to alternative splicing (exon 2 minus transcript), endogenous Ski is present as a doublet in chicken cells, with the lower band of the doublet being 37 residues shorter and 2-5 times more abundant than the upper band (4). When expressed in CEFs via a retroviral vector, the smaller form (Ski2/29) produces the same complexes with the GTCT/2 probe as the full-length c-Ski protein (Ski29) used in these studies (data not shown).
Competition Analysis with GTCT/1.5 Sites and GTCT/2 Site Mutants-The results presented above indicate that bind-

FIG. 2. c-Ski binds specifically to different versions of the GTCT binding site.
EMSAs were performed with 32 P-labeled probes prepared by PCR amplification of cloned inserts as described under "Materials and Methods." The sequence of the variable central region plus 3 bases of nondegenerate sequence on either end is shown for probes and competitors. The GTCT/2 probe was used unless indicated otherwise. A, EMSAs with GTCT/2 and GTCT/1 probes and competitors were performed as described under "Materials and Methods." Specific competitors were present at 2, 5, and 20 nM (10, 25, and 100 times the amount of probe). The nonspecific competitor (NS) was a CASTing oligonucleotide with central sequence unrelated to GTCT and was used only at 100 times the amount of probe. In addition to complexes 1 and 2, the specific Ski-minus complex X and the nonspecific complex Y are indicated. B, antibody supershifts of GTCT/2 complexes were performed with 0.5 l of protein A-purified anti-Ski mAbs, G8 and M6 (2 g/l), as indicated. Lanes 7-9 are a longer exposure of lanes 4 -6. C, analysis of complexes that form with GTCT/1.5 probes prepared from clones 6 -7 and 6 -15 (Fig. 1). EMSAs were performed with (ϩ) or without (Ϫ) the addition of mAb M6. D, blocking of G8 mAb supershift of complex 2 by the addition of GST-⌬Ski1-136 fusion protein (1, 3, and 5 g).
ing to the GTCT element is dependent on the number of sites and their divergence from the consensus. To obtain a more quantitative measurement of the effects of these variables on binding, we performed competition experiments (Fig. 4, A-D). In the first experiment, EMSAs were performed using unlabeled GTCT/2, GTCT/1, or GTCT/1.5 oligomers to compete for binding to GTCT/2 (Fig. 4, A and B). The amount of complexes 1 and 2 was quantitated and is presented graphically in Fig. 4A and summarized as relative amount required for 50% competition (I 0.5 ) in Fig. 4B. The results indicate that GTCT/1 competes approximately 100-fold less efficiently for binding than GTCT/2. Among the GTCT/1.5 elements, the two with the partial site sequence TGTCTGGNC (6-7 and 6-12) are similar to GTCT/2 in their competition efficiencies. The loss of a single match relative to this sequence (the C 8 residue) produces oligonucleotides that match the partial site consensus TGTCTGG (6-9 and 6-21) but results in a 5-fold decrease in competition efficiency. As indicated by comparisons of 6-9 with 6-21 and of 6-12 with 6-7, oligonucleotides with one-nucleotide or threenucleotide spacers between the two copies of the binding site have the same relative binding affinity.
To systematically evaluate the contribution of each of the consensus base pairs to the binding efficiency of the Ski complex, we performed competitive binding studies with a series of oligonucleotides in which each of the base pairs was mutated individually (Fig. 4, C and D). The two-copy binding site was used for this analysis, so each base was mutated in both copies of the binding site. The results are summarized in Fig. 4D. In this case GTCT/1 is approximately 70-fold less efficient as a competitor than the GTCT/2 site, but surprisingly, none of the mutants with 7 of 8 matches to the consensus in both copies of the tandem elements competes better than GTCT/1. The best competitor among these is mutT4, which shows an 80-fold reduction in competition relative to GTCT/2. It is interesting that on the complementary strand the substitution made in mutT4 is equivalent to the A 5 to G change that was seen in all of the selected partial sites. The remaining base pairs appear to be critical for complex formation with the GTCT element, since mutations in each of them reduce competition by more than 200-fold.
Two additional competitors that were used, GTCT/1.5A and GTCT/1.5B, have one complete copy of the binding site in association with either the first four bases or the last four bases of an otherwise tandem repeat element. Despite the fact that both of these competitors actually have a greater number of mutated base pairs than the double substitution mutants, they compete more efficiently than the double substitution mutants (Fig. 4, C and D). This is a surprising result, because oligonucleotides of this type were not obtained by CASTing, although they appear to bind as well as those with a second nontandem partial site (Fig. 4B). The results of both competition experiments show that substitutions can be tolerated to a much greater extent in a second copy of the binding site as long as there is a single copy present with no substitutions. This conclusion is supported by the results of the CASTing experiment in which no oligomers were identified that carry two partial FIG. 4. Relative binding affinity to variants of the GTCT element. Competition EMSA was carried out as described under "Materials and Methods." A, binding to the GTCT/2 probe was competed with unlabeled GTCT/2, GTCT/1, and several of the GTCT/1.5 oligomers isolated in the CASTing experiment. Probes and competitors were prepared by PCR amplification of cloned inserts. Competitors used were at 2, 5, 20, and 200 nM (10,25,100, and 1000 times the amount of the probe) except for GTCT/1, which was used at 4, 10, 40, and 400 nM. Relative binding compared with the reaction with no competitor (lane 1) was determined by quantitation of complexes 1 and 2 by Phosphor-Imager analysis of dried gels, and the results of two experiments are shown on the same graph. The gray bars on the graph represent the gel that is shown. B, sequences and relative binding affinities of the competitors used in A are shown. A plot of the values for relative binding given in A versus the concentration of competitor was used to determine the concentration that reduced binding to 50% (I 0.5 ). This is expressed as a ratio to the 50% inhibitory concentration of GTCT/2 itself; thus, higher values correspond to lower affinity binding. C, binding to the GTCT/2 probe was competed with unlabeled GTCT/2-containing doublestranded oligonucleotides in which a single base was changed in both copies of the binding site. Concentration of competitors used and quantitation of binding are the same as described for A. D, sequences and relative affinities of competitors used in C are shown. I 0.5 was determined relative to GTCT/2 as described in B.
copies of the GTCT binding site after round five. Together the results provide strong support for cooperative binding of the Ski complex to the GTCT element.
Ski Can Be Cross-linked to the GTCT Binding Site Along with Two Other Proteins-Irradiation of protein-DNA complexes with UV light causes covalent bonds to form between the DNA and proteins that are in close contact with the DNA (31). This reaction occurs more efficiently in the presence of halogenated analogs of thymidine, such as BrdUrd. If the DNA probe is labeled with 32 P, the label will be covalently attached to the cross-linked protein, so that components of the binding complex can then be separated by SDS-PAGE, and cross-linked proteins can be visualized by autoradiography (32,33). We used this method to analyze the components of the Ski-GTCT binding complex in c-ski-CEFs and to determine if Ski contributes directly to DNA binding.
The probe used for this experiment was produced by priming DNA synthesis from a GTCT/2 template using a complementary 32 P-labeled GTCT/1 primer and a dNTP mix containing equal concentrations of dTTP and BrdUTP (Fig. 5A). Because there are only two positions for T or BrdUrd in the single synthesized GTCT, this method provides for the incorporation of one BrdUrd per GTCT/2 probe molecule. Consequently, only one polypeptide should be cross-linked per probe molecule, thereby allowing us to enumerate the polypeptides bound to the GTCT element. After binding and UV irradiation, the crosslinked products were analyzed by SDS-PAGE. At least five proteins appear to cross-link specifically to the probe, as shown by their disappearance in the presence of unlabeled specific competitor, but not in the presence of mutant competitor (Fig.  5B, lanes 5-7). The largest cross-linked species, which is marked by a closed arrow in Fig. 5B, migrates more slowly than in vitro translated c-Ski (Fig. 5B, lane 3) but could represent c-Ski protein, which has reduced mobility due to the added mass of the cross-linked probe. This is apparently the case, because this is the only protein detected following immunoprecipitation of the cross-linked complexes in high stringency buffer (see "Materials and Methods") with three different Skispecific antibodies (Fig. 5B, lanes 8, 10, and 12) but not with preimmune serum or protein A-agarose alone (Fig. 5B, lanes 14  and 16).
To determine whether the other cross-linked proteins are in the same complex as Ski, we performed immunoprecipitation of cross-linked complexes in a low stringency buffer that should not dissociate protein-protein interactions (see "Materials and Methods"). Under these conditions, all four of the cross-linked species are co-immunoprecipitated with each of the three Skispecific antibodies (Fig. 5B, lanes 9, 11, and 13). These bands were not detected in samples that were immunoprecipitated with preimmune serum or protein A-agarose alone under the same conditions (Fig. 5B, lanes 15 and 17). It is difficult to make accurate estimates of the sizes of these proteins due to the unknown effect of the cross-linked probe on their mobilities. If we use the change in Ski mobility to correct for this effect, we estimate that these proteins range in size from 40 to 60 kDa. These results indicate that the Ski protein is in close enough contact with the binding site to allow cross-linking; however, there are additional proteins present in the Ski-GTCT binding complex that also make contact with the DNA.
The GTCTAGAC Element Mediates Transcriptional Repression by Ski-To determine if Ski can regulate transcription through the GTCTAGAC binding site, reporters were made in which copies of the binding site were cloned upstream of the herpes simplex virus thymidine kinase promoter (tkCAT). Cotransfection with the RSVc-ski expression plasmid into the chicken fibroblast cell line, UMN-SAH/DF#1, represses these reporters in proportion to the number of GTCT binding sites cloned upstream of the promoter (Fig. 6). Reporters with one, two, and four GTCT/1 sites are repressed 8-, 10-, and 12-fold respectively, and a reporter with two GTCT/2 sites is repressed 19-fold. There was some variation in the basal activity of the different reporters from which the values of -fold repression FIG. 6. Ski represses tkCAT reporters in proportion to the number of upstream GTCT/1 or GTCT/2 binding sites. UMN-SAH/ DF#1 cells were transfected with CAT reporter plasmids (600 ng) containing the indicated number of GTCT/1, GTCT/2, or mutGC/2 upstream binding sites. Co-transfection with these reporters was performed with either 600 ng of RSVPL (empty expression vector) or 200 ng of RSVc-ski plus 400 ng of RSVPL. Transfections were performed in triplicate and assayed for CAT activity as described under "Materials and Methods." CAT activity was normalized to protein concentration, and values for -fold repression in the presence of RSVc-Ski were calculated relative to the CAT activity of each indicated reporter co-transfected with 600 ng of RSVPL.
were calculated, but these differences do not correlate with the number of binding sites or the degree of repression observed in the presence of Ski. In addition, we consistently observe that c-Ski represses reporters with no binding sites or with mutant binding sites no more than 3-4-fold. This same experiment was done in primary CEFs and in the liver hepatoma cell line, HepG2, with similar results.
Like primary CEFs, the CEF cell line used for these experiments is transformed by overexpression of c-ski and v-ski (data not shown). Therefore, if repression of GTCT reporters is related to the transforming activity of Ski proteins, both c-Ski and v-Ski should be able to repress transcription from these reporters. To test this, we compared the relative -fold repression produced by different amounts of RSVc-ski and RSVv-ski on the tkCAT and GTCT/2X2tkCAT reporters (Fig. 7, A, B, and  D). Over a wide range of expression plasmid amounts (20 -600 ng), repression increases linearly with the amount of co-transfected c-ski expression plasmid, reaching a maximum of 35-40fold when equal amounts of reporter and c-ski expression plasmid are cotransfected (Fig. 7A). As noted above, c-Ski also represses expression from the reporter lacking upstream GTCT sites, but this activity actually decreases with higher levels of co-transfected RSVc-ski, dropping to less than 1 ⁄10th the value obtained with the GTCT reporter. v-Ski also represses transcription from the GTCT reporter in a dose-dependent manner, giving a 12-15-fold repression at maximum dose of 600 ng (Fig.  7B). Unlike c-Ski, v-Ski does not repress the TK reporter, which lacks GTCT binding sites. The 2-3-fold lower level of repression activity observed with v-Ski compared with c-Ski is consistent with the observed differences in their binding to this sequence. It also correlates well with their relative potency of transformation in CEFs, suggesting a role for transcriptional repression through the GTCT binding site in Ski's oncogenic activity.
There are a variety of mechanisms by which Ski could re-press transcription through the GTCT binding site (34,35). One possibility is that Ski might not have its own repression activity but instead might be repressing transcription through the GTCT binding site by competing with another protein for binding to this sequence. To test this possibility, we fused c-Ski to the DNA binding domain of the Gal4 protein and tested its effect on a reporter with five Gal4 binding sites cloned upstream of the TK promoter (G5tk-luciferase). We found that Gal4-Ski represses transcription from this reporter in a dosedependent fashion, reaching a maximum of 25-fold (Fig. 7, C and D). The result shown is a representative experiment. The same experiment was done with similar results using a G5tkCAT reporter in HepG2 cells. This demonstrates that Ski does contain a repression activity that can function independently of the GTCT binding site interaction. Once again, Ski repressed transcription to a lesser extent from a TK reporter with no Gal4 binding sites, showing that Ski may also be able to repress transcription by a mechanism that does not require DNA binding.

DISCUSSION
It is now clear that protein-protein interactions among DNA binding transcription factors play important roles in modulating both the DNA binding properties and transcriptional regulatory properties of such factors (16 -18). Carrying out binding site selection using a nuclear extract as the source of protein has the advantage that the transcription factor can interact with a given DNA binding site while maintaining many of the same protein-protein interactions that would occur inside the cell. Using this approach, we have identified a novel DNA binding site for the Ski protein with the consensus sequence GTCTAGAC. Exogenous c-Ski in CEF nuclear extracts binds to this sequence with high affinity and specificity. v-Ski also binds to this binding site, but with lower affinity than c-Ski, probably due to the loss of the carboxyl-terminal dimerization domain. Furthermore, endogenous Ski proteins from chicken, mouse, and human all bind to this GTCT element, suggesting that binding to this site may be important for the normal function of Ski.
Structure of Binding Site and Analysis of Binding Site Mutants-The GTCT consensus is extraordinarily conserved in the oligomers that we identified by CASTing. There is only a single base change in one of the 20 GTCT/2 oligomers, and none of the GTCT/1.5 oligomers had any substitutions in the first copy of the consensus. We were surprised that none of the binding sites contained two imperfect copies of the GTCT element. The reason for this observation became apparent when we analyzed binding affinities by competition. All of the symmetrical single base substitutions that we made in the GTCT/2 binding site dramatically reduced the binding affinity, such that it was equivalent to or less than that of a single copy binding site. On the other hand, GTCT/1.5 sites with 0 -3 base pairs between GTCT elements and two or more base substitutions in one element have only slightly lower binding affinity than GTCT/2. This demonstrates that at least one perfect copy of the binding site is essential for stable Ski complex formation and provides strong evidence for cooperative binding to multiple GTCT elements.
Composition of the Two Ski DNA Binding Complexes-Neither purified bacterially produced Ski nor in vitro translated Ski binds the GTCT element. Possible explanations for this failure include the following: 1) bacterially produced and in vitro translated Ski protein may not fold properly; 2) Ski may contain an autoinhibitory domain that prevents DNA binding; 3) post-translational modification such as phosphorylation may be required; 4) a protein cofactor may be required to stabilize Ski's interaction with the DNA. The first explanation is unlikely because bacterially produced and in vitro translated Ski proteins have been shown to dimerize efficiently, and analysis of in vitro translated Ski by partial proteolytic digestion indicates that it is correctly folded (29). Autoinhibitory domains have been shown to regulate DNA binding of such transcription factors as ETS-1 and p53; however, in these cases the inhibition of DNA binding is relieved by deletion of the autoinhibitory domains (36 -38). We have performed EMSAs with various deleted forms of in vitro translated Ski, but no deletions have been identified that enable the Ski protein to bind the GTCT element (data not shown). Previous work by Sutrave et al. (20) has shown that chicken c-Ski is a phosphoprotein, whereas v-Ski is not. Because both c-Ski and v-Ski can bind to DNA, it is unlikely that phosphorylation is required for DNA binding of chicken Ski proteins. Ishii and co-workers (1) have presented evidence indicating that Ski requires a cofactor present in a nuclear extract from Molt 4 cells to bind DNA. The Ski homologue, SnoN, which has been shown to heterodimerize efficiently with Ski (5, 6) is an obvious candidate for a Ski-GTCT co-binding partner. For this reason, we performed EMSAs with Ski and Sno heterodimers formed by co-in vitro translation, but no binding to the GTCT site was detected (data not shown).
In this study, we provide evidence for additional factors that interact specifically with the GTCT element. The non-Ski-containing complex X, which binds GTCT/2 specifically, migrates faster than Ski-containing complexes in EMSAs. It is possible that Ski associates with the components of this complex to form a ternary complex with further reduced mobility. This would account for the reduced level of complex X in c-ski-CEFs relative to normal CEFs. Interestingly, there is more complex X visible in gel shifts with v-ski-CEF extract than with c-ski-CEF extract, suggesting that v-Ski might form an unstable association with complex X. It is possible that we have identified the components of complex X as the four species (two sets of dou-blets) that cross-link specifically to the BrdUrd-substituted probe in addition to Ski.
The fact that Ski can be UV-cross-linked to a GTCT probe indicates that it probably contains a domain that makes contacts within the Van der Waals radius of the DNA (31). There are a number of mechanisms by which Ski could co-bind to the GTCT site along with other proteins and still play a role in specific sequence recognition. For example, another protein could bind in an overlapping complex with the Ski protein and stabilize Ski binding by cooperative interaction. The yeast proteins Mcm1 and MAT␣2 (␣2) provide an example of proteins that bind cooperatively to overlapping DNA binding sites (39,40). Although ␣2 binds to DNA on its own (41), interaction with Mcm1 increases the affinity of ␣2 for operator sites and extends the DNA binding specificity of ␣2 (42). Alternatively, there are transcriptional co-regulatory proteins, such as the B-cell-specific factor Bob-1/OBF-1/OCA-B or the viral transactivator VP16, that associate with site-specific transcription factors and can stabilize interaction of these factors with a subset of binding sites (18,43). It is possible that in the case of Ski, such an intimate association with the DNA binding domain of another protein could bring Ski into close enough proximity with DNA bases to allow UV cross-linking. It is true that cross-linking of Ski to the GTCT binding site seems to be less efficient than that of the unidentified components of the GTCT complexes.
Role of Ski Dimerization in Cooperative Binding to the GTCT Element-Exogenous c-Ski forms two complexes with GTCT/2 and GTCT1.5 probes (complex 1 and complex 2). The faster migrating form has the same mobility as the single complex that forms with the GTCT/1 probe. Since Ski is present in both complexes 1 and 2, the simplest model for the composition of these complexes is that complex 1 represents a Ski monomer plus associated proteins bound to a single GTCT site and complex 2 is a dimer of these elements. This suggestion fits nicely with the observations that binding of Ski to the GTCT/2 element is highly cooperative and that c-Ski dimerizes efficiently through a carboxyl-terminal dimerization domain (5,6). This model also agrees with the observation that v-Ski, which lacks this high affinity dimerization domain, binds to the GTCT/2 site much less efficiently than c-Ski. On the other hand, it predicts that the two Ski forms should be very similar in binding to GTCT/1, because in this case dimers would not be required. However, we investigated binding of v-Ski to the GTCT/1 element and were unable to detect the formation of a shifted band by EMSA.
The weak binding of v-Ski to GTCT/2 and its failure to bind GTCT/1 suggest that dimerization of Ski is important for binding to both single copy and double copy binding sites. This conclusion is contrary to the model described above and provides support for an alternative model in which a Ski dimer and associated proteins bind to each copy of the GTCT element. Cooperative binding to the GTCT/2 site then occurs by interaction between adjacent Ski dimers bound to DNA. Because the rate-limiting step in cooperative binding is the first step, the carboxyl-terminal dimerization domain of c-Ski greatly enhances cooperative binding by stabilizing dimers so that they can bind effectively to a single element. Accordingly, v-Ski is impaired in binding GTCT/1 because its dimers are unstable, but it is able to bind GTCT/2 because the cooperative interaction on DNA stabilizes two weakly associated dimers. This model is consistent with our previous data that demonstrated low affinity multimer formation by v-Ski (5).
Ski Represses Transcription through the GTCT Binding Site-Here we demonstrate for the first time that both c-Ski and v-Ski can repress transcription. c-Ski is shown to be a stronger repressor than v-Ski, and this difference is consistent with their relative abilities to bind the GTCT element and is mirrored by their relative transforming activity. Repression by Ski cannot be explained by competition with an activator for binding to the GTCT site, because c-Ski linked to an unrelated DNA binding domain is able to repress transcription of a cognate reporter. Therefore, Ski can be classified as an active repressor (35). Mechanisms of active repression include interfering with the activity of a DNA-bound activator (44) and interacting with the basal transcription machinery (45) or with histone deacetylases (46). Often these interactions are mediated by corepressors, as in the case of thyroid hormone receptor and the Krü ppel-associated box domain-containing proteins (17,47). Since the minimum transforming region of Ski has been localized to the N-terminal 304 residues (29), it will be interesting to determine whether this region contains the repression domain.
In light of the fact that Ski can activate transcription from some muscle-specific and viral enhancers and promoters, it was surprising to find that Ski represses transcription through the GTCT binding site. However, there are now numerous examples of transcription factors that can act as either activators or repressors depending on protein-protein interactions dictated by the promoter and physiological context (48). Although Ski appears to activate transcription by interacting with other DNA-bound transcription factors, including the myogenic regulatory factors (13) and members of the NFI family (15), 2 we have been unable to identify an activation domain by fusing various regions of Ski with an independent DNA binding domain (data not shown). In the case of repression through binding to the GTCT binding site, Ski's action appears to be different. Although Ski requires cofactors to bind to this sequence, it appears to possess a repression domain that can function independently of these proteins. However, further characterization of the factors that co-bind to the GTCT site along with Ski will be required to determine the role that Ski plays in repressing transcription through this binding site.
The ability of Ski to cause oncogenic transformation and induce terminal muscle differentiation in the same cells has been a paradox. Previous results showing transcriptional activation by Ski are now contrasted by our findings that it can act as a transcriptional repressor. It is interesting that the duality inherent in the biological activities of the Ski protein is thereby reflected at the transcriptional level. This makes it tempting to speculate that the opposite biological activities are coupled to the opposite transcriptional activities. However, this will remain only an attractive speculation until the key genes activated and repressed by Ski are identified.