ATF-2 and C/EBP (cid:97) Can Form a Heterodimeric DNA Binding Complex in Vitro

We screened an expression cDNA library with a radio- labeled C/EBP (cid:97) fusion protein and isolated three independent cDNAs encoding ATF-2, a bZIP protein that binds cAMP response elements (CRE). This interaction requires the respective bZIP domains, which form a typ- ical bZIP heterodimer with altered DNA binding selec-tivity. C/EBP (cid:97) and ATF-2 homodimers bind CRE sites, but ATF-2:C/EBP (cid:97) heterodimers do not. Heterodimers bind an asymmetric sequence composed of one consen- sus half-site for each monomer, and may thus have a unique regulatory function. As predicted, co-transfec- tion of ATF-2 with C/EBP (cid:97) results in decreased activation of transcription driven from consensus C/EBP-bind- ing sites. In contrast, C/EBP (cid:97) and ATF-2 function cooperatively to activate transcription driven by the asymmetric sequence. Both factors are expressed in liver, where immunoprecipitation experiments show that ATF-2 co-precipitates with C/EBP (cid:97) . These results are consistent with the interpretation that C/EBP (cid:97) and ATF-2 can associate in vivo . Moreover, the formation of ATF-2:C/EBP (cid:98) heterodimers suggests that cross-family dimerization with ATF-2 may be a general property for C/EBP family proteins. PAGE, electroblotted, reacted with C/EBP (cid:97) antibody, and developed with horseradish peroxidase-conjugated donkey anti-rabbit serum. The 42- and 29-kDa C/EBP (cid:97) translation products are retained only on the MBP-ATF-2 resin. The three bands observed between C/EBP 42 and 29 kDa are internal translation initiation products that have been observed previously (47, 48).

We screened an expression cDNA library with a radiolabeled C/EBP␣ fusion protein and isolated three independent cDNAs encoding ATF-2, a bZIP protein that binds cAMP response elements (CRE). This interaction requires the respective bZIP domains, which form a typical bZIP heterodimer with altered DNA binding selectivity. C/EBP␣ and ATF-2 homodimers bind CRE sites, but ATF-2:C/EBP␣ heterodimers do not. Heterodimers bind an asymmetric sequence composed of one consensus half-site for each monomer, and may thus have a unique regulatory function. As predicted, co-transfection of ATF-2 with C/EBP␣ results in decreased activation of transcription driven from consensus C/EBP-binding sites. In contrast, C/EBP␣ and ATF-2 function cooperatively to activate transcription driven by the asymmetric sequence. Both factors are expressed in liver, where immunoprecipitation experiments show that ATF-2 co-precipitates with C/EBP␣. These results are consistent with the interpretation that C/EBP␣ and ATF-2 can associate in vivo. Moreover, the formation of ATF-2:C/EBP␤ heterodimers suggests that cross-family dimerization with ATF-2 may be a general property for C/EBP family proteins.
CCAAT/enhancer-binding protein, C/EBP, 1 was purified as an activity that bound to consensus enhancer elements and to the CCAAT box motif (1). C/EBP protein is a member of the basic region-leucine zipper (bZIP) class of transcription factors (2,3). The bZIP domain consists of the leucine zipper, a heptad repeat of leucines, preceded by the basic region, a sequence with net positive charge (4,5). A combination of molecular and structural studies showed that the heptad repeat region is an amphipathic ␣ helix that mediates dimerization by forming a parallel coiled-coil (4 -15). The co-crystal structure of the bZIP domain of GCN4 bound to DNA showed that the coiled helices of the leucine zipper separate, positioning one helix from each chain for sequence specific interaction with DNA (10).
Leucine zippers accommodate both homotypic and heterotypic dimerization. For example, the first C/EBP protein char-acterized, C/EBP␣, is one of at least five gene products comprising the C/EBP gene family (16,17). These proteins show extensive sequence similarity that is restricted to the bZIP domain, such that each homodimer binds the same DNA element. Additionally, all proteins in the C/EBP family can pair with each other to form DNA binding heterodimers (16,17). When not bound to DNA, the subunits of bZIP dimers are in a rapid monomer:dimer equilibrium such that the lifetime of dimers is estimated in seconds (18,19). The ready dissociation of bZIP dimers in vitro suggests that heterodimers with potentially unique regulatory properties may form in vivo. In the case of Fos, a bZIP protein in the AP1 family, no homodimers form. Instead, Fos forms heterodimers with Jun, another AP1 family protein (20).
Sequence-specific interaction with DNA increases the lifetime of bZIP dimers more than 10-fold (18,19,21). However, the apparent DNA affinity constants are lower than might be anticipated (18). Surprisingly, GCN4, a protein that binds the core sequence TGAGTCA, exhibited similar affinity for a core motif containing an extra nucleotide, TGACGTCA (18). The structures of GCN4 and Fos:Jun bound to DNA revealed that most of the amino acids involved in base specific contacts are those that are highly conserved among all bZIP proteins (11), including those with distinct DNA binding specificity. Thus, the basic mechanism by which bZIP proteins interact with DNA is known, while the determinants of sequence discrimination are less clearly understood.
C/EBP␣ activates transcription of several liver and fat cellspecific genes (22)(23)(24)(25). Interestingly, over-expression of C/EBP␣ in cultured cells results in growth arrest (26 -29), a finding consistent with the observation that C/EBP␣ expression commences during the conversion of 3T3L1 preadipocytes into quiescent fat cells in vitro (17,30,31). The role of C/EBP␣ in the adipogenic program was further verified by antisense experiments (32), and by ectopic expression of C/EBP␣ in a variety of fibroblastic cell lines, where efficient promotion of fat cell differentiation was observed (33). Support for the notion that C/EBP␣ plays a central role in regulating energy homeostasis (34) was provided by targeted gene disruption. Homozygous C/EBP␣ knockout mice are born with apparently normal blood glucose levels, but become severely hypoglycemic within minutes (35). These animals exhibit glycogen storage defects and morphological anomalies in fat and liver tissues. Although these defects are consistent with a role for C/EBP␣ in energy homeostasis, expression of several putative C/EBP target genes was normal (35). Thus, it has been difficult to identify genes that are C/EBP␣ targets in vivo.
Activator proteins like C/EBP␣ bind specific DNA sequences located either upstream or downstream of the core promoter. In response to physiological cues, activators stimulate transcription initiation by interacting with general transcription factors, with TATA-associated factors, or with adaptor proteins (36,37). The observation that C/EBP␣ binds a range of DNA sequences, coupled with the observation that its subunits exchange rapidly, suggests that the monomer may be a target for regulation. We screened a gt11 expression cDNA library for proteins that physically interact with C/EBP␣. Three independent cDNA clones encoding ATF-2, a bZIP protein, were isolated. We mapped the interacting protein domains to the respective leucine zippers, and demonstrated formation of a bZIP heterodimer with restricted DNA binding selectivity, in vitro. C/EBP␣ and ATF-2 are expressed in liver, and transient transfection analysis shows that co-transfected ATF-2 impacts C/EBP␣ function. Together with immunoprecipitation results showing that ATF-2 coprecipitates with C/EBP␣, these results are consistent with the interpretation that functional ATF-2: C/EBP␣ heterodimers form in vivo.

EXPERIMENTAL PROCEDURES
Recombinant Plasmids-pGEX-2T (Pharmacia, Uppsala Sweden) was modified by the insertion of a protein kinase A phosphorylation site as described (38). pMSV-C/EBP⌬1-2 (24) was used for construction of N-terminal GST-C/EBP fusions. Truncated constructs lacking the leucine zipper and the bZIP domain were prepared by placing in-frame stop codons at amino acids 310 and 272, respectively (39). MluI digestion and fill-in created the fusion to amino acid 192. Constructs 218 -358 and 281-342 have been described (19). A diagram of the fusion proteins is shown (see Fig. 2A).
The reporter vector 2X C/EBP-Luc contains two direct repeats of the consensus C/EBP␣-binding site, TGCAGATTGCGCAATCTGCA, and was a gift from P. Rorth (Carnegie Institution of Washington). 2X C/EBP-Luc has the minimal thymidine kinase promoter (40) and a BamHI site for proximal insertion of transcription factor-binding sites. Three iterations of the chimeric binding site (5Ј-GCCGTGACG-CAATCTC-3Ј) were inserted into the thymidine kinase promoter/luciferase vector, producing the reporter 3X Chimera-Luc.
Protein Interaction Screening-A GST (41) fusion protein encoding C/EBP␣ amino acids 1-10 fused to amino acids 60 -358 (see Fig. 2A) was radiolabeled with protein kinase A (Sigma) as described (33). Radiolabeled protein was used for interaction screening immediately.
Expression and Purification of Recombinant Fusion Proteins-Fusion proteins were expressed in Escherichia coli host strain BL21. Log phase cultures were induced at an OD 600 of 0.8 for 2 h. Cells were harvested, and GST (41) or MBP (New England Biolabs, Beverly, MA) fusion proteins were purified according to established protocols. Detailed purification of "C/EBP short" was described previously (19).
Preparation of Antisera-The ATF-2 cDNA (EcoRI) was subcloned into the vector pMal (New England Biolabs), for protein expression in bacteria. MBP-ATF-2 was affinity purified, and the antigen was excised from a 10% SDS-PAGE gel. Purification of an NH 2 -terminally deleted C/EBP␣ protein (C/EBP␣ short) was described previously (19). This protein was used to prepare antiserum. Antisera were raised according to standard protocols, and appropriate reactivity was verified against recombinant proteins. The specificities of all antisera were verified against purified recombinant proteins. ATF-2 and C/EBP␣ antisera do not cross-react against purified recombinant proteins.
Purification and Biotinylation of Antibodies-2.5 mg of MBP-ATF-2 was coupled to CNBr-activated Sepharose 4B (Pharmacia, Uppsala, Sweden). Following the removal of lipids with Seroclear (CalBiochem, La Jolla, CA), clarified serum was passed 3 ϫ over a 0.2-ml bed of MBP-ATF-2 equilibrated in phosphate-buffered saline. Columns were washed with 10 ml of phosphate-buffered saline, and eluted with 750 l of 0.1 M glycine (pH 3.0), dripping directly into 750 l of 0.1 M Na 2 HPO 4 (pH 9.2) for neutralization. Affinity purified antibody was dialyzed into phosphate-buffered saline to remove glycine, and biotinylated as described (43). This protocol was adapted for purification and biotinylation of the antisera against intact C/EBP␣, kindly provided by Pernille Rorth, and C/EBP␣ short proteins.
Immunoblotting and Modified Western Blotting-Proteins from SDS-PAGE were electroblotted to Immobilon P membranes (0.45 M, Millipore) and developed with polyclonal antiserum. Secondary detection utilized horseradish peroxidase-conjugated donkey anti-rabbit serum (Amersham), which was visualized by chemiluminescence (ECL reagent, Amersham). For biotinylated antibodies, secondary detection utilized streptavidin-conjugated horseradish peroxidase (Amersham) and ECL reagent.
To map protein interaction domains, equal amounts of each truncated protein were separated on 10% SDS-PAGE gels, and transferred to nitrocellulose (0.45 m, Schleicher & Schuell). Membranes were blocked with 5% nonfat milk, and incubated in Hyb150 containing soluble GST-C/EBP (ATF-2 proteins on the membrane) or soluble GST-ATF-2 (C/EBP proteins on the membrane). After 4 washes, the interaction of soluble protein with membrane-bound protein was probed with specific antiserum.
Immunoprecipitation and Co-immunoprecipitation Analysis-Freshly prepared RLNE (45) was pre-cleared with an irrelevant antiserum and protein A-agarose beads (Life Technologies). After a brief spin, supernatants were separated and adjusted to 50 mM Tris (pH 8.0), 15% glycerol, 0.25 M NaCl, 1 mM MgCl 2 , 0.1 mM EDTA, 1% Nonidet P-40. 3 l of affinity purified antibody was added and incubated overnight at 4°C with rocking. Precipitates were collected onto protein A beads and washed 5 ϫ with 500 l of 50 mM Tris (pH 8.0), 150 mM NaCl, 1% Nonidet P-40. After SDS-PAGE and electroblotting, proteins were detected with biotinylated, affinity purified antibodies and visualized as described above.
For co-immunoprecipitation, primary precipitates were formed with the first affinity purified antibody (as above), separated by SDS-PAGE, electroblotted, and detected with a second antibody (affinity purified and biotinylated) as above. 20 pmol of double-stranded oligonucleotide-binding sites were included where indicated, and blots were developed as described above.
Transient Transfection Analysis-The hepatoma cell line Fao (46) was transfected by the standard calcium phosphate method (42). pCMX1-␤Gal, a gift from C. Thompson, was included in all transfections to normalize the data. Cells were harvested 48 h later, and relative luciferase (Analytical Luminescence Laboratory, Ann Arbor, MI) and ␤-Gal activities (Promega) were determined in duplicate, and the average was determined. Cells transfected with the reporter vector alone determined relative luminescence. All results are the average of three independent experiments.

RESULTS
Identification of a C/EBP␣ Interacting Protein-To search for proteins that physically interact with C/EBP␣, a rat liver expression cDNA library was screened with a radiolabeled GST-C/EBP␣ fusion protein. From 2 ϫ 10 6 phage screened, eight positive cDNAs were isolated. The phage inserts were subcloned and sequenced, revealing that three of them encoded the bZIP protein ATF-2, a member of the ATF/CREB transcription factor family. The interaction between ATF-2 and C/EBP␣ was specific as both GST-GCN-4, a yeast bZIP protein, and GST failed to interact with ATF-2 in this assay (not shown).
An ATF-2 Affinity Column Binds C/EBP␣ from RLNE-To test whether ATF-2 and C/EBP␣ would interact under more stringent conditions, rat liver nuclear extract (RLNE) was fractionated on a column displaying ATF-2 (fused to MBP). As a specificity control, a column displaying MBP alone was run in parallel. Equivalent percentages of the unbound fraction and the salt eluted fractions were separated by SDS-PAGE, electroblotted, and probed with C/EBP␣ antiserum. Multiple forms of C/EBP␣ are expressed in the liver, including the full-length protein (42 kDa) and several internal translation initiation products (47,48). The MBP-ATF-2 affinity column, but not the MBP column alone, selectively binds all forms of C/EBP␣ present in the nuclear extract (Fig. 1, compare the eluates from the two columns). Note that elution of C/EBP␣ from the ATF-2 column requires 0.5-1.0 M KCl, an indication of a relatively high affinity interaction.
Association of ATF-2 with C/EBP␣ Requires the bZIP Domains-To map the domains required for interaction between C/EBP␣ and ATF-2, a sequential deletion strategy was used. C/EBP␣ deletion constructs were expressed in E. coli as GST fusion proteins, and purified on glutathione-agarose. Equivalent amounts of each protein were separated on SDS-PAGE gels and transferred to nitrocellulose. The blot was incubated with purified GST-ATF-2, washed extensively, and subsequently developed with anti-ATF-2 serum. As shown in Fig. 2A, GST-C/EBP constructs containing the leucine zipper bound soluble GST-ATF-2 efficiently (lanes 1, 4, 5, and 6). Analogously, truncated versions of ATF-2, similarly prepared as GST fusions, were tested for the ability to bind soluble C/EBP␣ (Fig.  2B). Again, only those constructs containing the leucine zipper efficiently retained soluble C/EBP␣ (lanes 1 and 3). The weak binding observed in the absence of the leucine zippers (Fig. 2, A  and B, lanes 2) is likely due to the putative zinc finger motif in ATF-2 and will be addressed under "Discussion." As controls, each blot (Fig. 2, A and B, respectively) was incubated with antiserum without prior exposure to soluble proteins, showing that the antisera for C/EBP␣ and ATF-2 are not cross-reactive (not shown). These results show that the leucine zippers are sufficient to mediate interaction between these bZIP proteins.
ATF-2:C/EBP␣ Heterodimers Bind DNA-Formation of an ATF-2:C/EBP␣ heterodimer would bring together different DNA-binding domains, which may lead to a change in DNA binding selectivity. To assay for DNA binding heterodimers, three DNA sites were tested in electrophoretic mobility shift assays (EMSA): a consensus C/EBP-binding site; a consensus CRE-binding site; and a chimeric site consisting of one C/EBP half-site directly abutted to one CRE half-site. To distinguish heterodimeric DNA binding complexes from homodimeric DNA binding complexes, a truncated C/EBP␣ protein encompassing the bZIP domain (19) was used. DNA binding complexes formed with one short C/EBP␣ subunit and one full-length C/EBP␣, ATF-2, or CREB subunit migrate intermediate to DNA binding complexes composed of two full-length or two short subunits (homodimers), and demonstrate subunit exchange.
As shown in Fig When a radiolabeled CRE site is tested (Fig. 3B), a shift is observed with C/EBP␣ (lanes 2-4). When C/EBP␣ short and ATF-2 are mixed, it is predominantly ATF-2 homodimers that shift the probe (lanes 5-7). Although both homodimers bind, no heterodimeric DNA binding complex is observed (see lane 6). Analogous results are obtained when C/EBP␣ and CREB are mixed (lanes 8 -10), where the predominant shift is by CREB homodimers, and no heterodimeric complex is detected (lane 9).
Upon EMSA analysis using the chimeric binding site (Fig.  3C), a gel shift complex of intermediate migration is observed (lane 6), which is consistent with formation of an ATF-2:C/ EBP␣ heterodimer. In fact, C/EBP␣, ATF-2, and CREB homodimers (lanes 4, 7, and 10, respectively) all bind the chimeric sequence. The failure of ATF-2:C/EBP␣ heterodimers to form a gel shift complex on the consensus C/EBP-and CREB-binding sites suggests that these heterodimers have a restricted DNA binding selectivity.

FIG. 1. An MBP-ATF-2 affinity column binds C/EBP␣ from RLNE.
Equivalent amounts of MBP-ATF-2 or MBP alone were immobilized on maltose beads. 70 g of rat liver nuclear extract was applied to each column, followed by extensive washing. Equivalent percentages of unbound and eluted protein fractions were separated on 10% SDS-PAGE, electroblotted, reacted with C/EBP␣ antibody, and developed with horseradish peroxidase-conjugated donkey anti-rabbit serum. The 42-and 29-kDa C/EBP␣ translation products are retained only on the MBP-ATF-2 resin. The three bands observed between C/EBP 42 and 29 kDa are internal translation initiation products that have been observed previously (47,48).

ATF-2 Forms Heterodimers with C/EBP␤-Since C/EBP
family proteins dimerize interchangeably, we tested ATF-2 for heterodimer formation with C/EBP␤ (Fig. 4). As a control, we show that C/EBP␣ short and C/EBP␤ form a heterodimeric DNA binding complex in the EMSA assay. When C/EBP␤ was mixed with ATF-2, a heterodimeric complex formed on the chimeric binding site. Thus, the capability to form heterodimers with ATF-2 is a property that appears to be shared among C/EBP family proteins.

Co-transfection of ATF-2 with C/EBP␣ Results in Decreased Activation of a Minimal Promoter Driven by C/EBP-binding
Sites-To analyze the impact of ATF-2 on C/EBP␣ function, we transfected the hepatoma cell line Fao with mammalian expression vectors encoding these proteins. Transcription was analyzed with a simplified luciferase reporter vector (2X C/EBP Luc) driven by the minimal thymidine kinase promoter and two copies of the C/EBP-binding site. Hepatoma cells transfected with ATF-2 alone showed basal levels of reporter gene activity (Fig. 5, left panel), consistent with the observation that ATF-2 does not bind the consensus C/EBP site. In contrast, C/EBP␣ transfectants showed a 7-fold increase in reporter gene activity. When ATF-2 and C/EBP␣ were co-transfected, activation levels decreased 43 to 55% (Fig. 5, left panel). These results are consistent with the interpretation that formation of ATF-2:C/ EBP␣ heterodimers decreases the pool of C/EBP␣ homodimers available to bind the reporter construct, resulting in decreased transcription activity.
ATF-2 and C/EBP␣ Cooperate in the Activation of a Minimal Promoter Driven by Chimeric Binding Sites-The initial characterization of ATF-2 suggested that the protein lacked trans-activation activity altogether (49 -53). Subsequently, transactivation activity was demonstrated, but was subject to tight control (50,51). To characterize transcription from the heterodimer binding site, we cloned 3 copies of the chimeric sequence upstream of the minimal thymidine kinase promoter (3X Chimera Luc). Surprisingly, Fao cells transfected with ATF-2 alone showed reporter activity that was 12-18-fold higher than the control (Fig. 5, right panel). Similarly, cells transfected with C/EBP␣ alone showed 5-7-fold activation of the reporter. When ATF-2 and C/EBP␣ were co-transfected, the reporter gene was activated approximately 26-fold (Fig. 5, right  panel). First, these results show that ATF-2 activates transcription of the minimal promoter in Fao cells, which is surprising in that ATF-2 activity is repressed in many cell types. Second, these results are consistent with the interpretation that ATF-2 and C/EBP␣ form heterodimers that activate transcription from the chimeric sequence.
ATF-2 Is Expressed in the Liver and Co-Precipitates with C/EBP␣-Although ATF-2 mRNA is found in essentially all tissues tested, the protein has been characterized mainly in brain and thymus. Using affinity purified ATF-2 antibodies, we found that ATF-2 is expressed in rat liver, apparently as a doublet of about 68 kDa (Fig. 6A, lanes 2 and 4). ATF-2 is also expressed in thymus, while expression is minimal in L cell and spleen cell nuclear extracts (not shown). As C/EBP␣ is also expressed in liver, we tested for co-precipitation of ATF-2 with C/EBP␣. Freshly prepared rat liver nuclear extracts were subjected to immunoprecipitation with affinity purified C/EBP␣ antibodies. The 42-and 29-kDa forms of C/EBP␣ were precipitated by this reagent (Fig. 6A, lane 1). When this immunopre-

FIG. 2. Interaction of ATF-2 with C/EBP␣ requires the bZIP domains.
The interaction surfaces were mapped by a modified Western blotting technique. A, equivalent amounts of the indicated GST-C/EBP fusion proteins were separated by SDS-PAGE and electroblotted to nitrocellulose. The blot was incubated with purified GST-ATF-2, and washed extensively. Bound ATF-2 was detected with an ATF-2 antibody. Schematic representations of the proteins are shown below the blot. B, equivalent amounts of the indicated GST-ATF-2 fusion proteins were separated by SDS-PAGE and electroblotted. The blot was incubated with purified GST-C/EBP␣ and washed extensively. Bound C/EBP␣ was detected with a C/EBP␣ antibody. cipitate was blotted for co-precipitating ATF-2 reactivity, the characteristic doublet at 68 kDa was observed (lane 3). The same result was obtained when C/EBP␣ was immunoprecipitated with a different antibody, one directed against the COOH-terminal portion of the protein (lane 5). These results are consistent with the interpretation that ATF-2 and C/EBP␣ can form heterodimers in vivo.
Although the half-life of a bZIP dimer is measured in sec-onds, binding to a consensus DNA sequence results in a 10 -100-fold increase in the lifetime of the dimer (19). We reasoned that inclusion of a specific oligonucleotide-binding site during immunoprecipitation would stabilize heterodimers, facilitating their detection in the co-precipitation assay. The two binding sites we compared were the consensus C/EBP site and the chimeric site (see Fig. 3, A and C, for sequences). As shown previously, ATF-2 is readily detectable in RLNE (Fig. 6B, lane   FIG. 3. ATF-2:C/EBP␣ heterodimers bind DNA. A short C/EBP␣ protein containing only the bZIP domain (19) was mixed with full-length C/EBP␣, ATF-2, or CREB in the presence of a radiolabeled DNA probe. A, a radiolabeled C/EBP-binding site; B, a radiolabeled CRE-binding site (Jun2); or C, a radiolabeled chimeric binding site was incubated with proteins as indicated below the figures. The formation of a gel shift complex with migration intermediate to the homodimers is interpreted to indicate heterodimer formation. A reticulocyte endogenous protein binds the CRE probe (Retic., B, arrow). The position of migration of all homodimers is indicated by arrows, as is the ATF2:C/EBP␣ short heterodimer (C, lane 6). Free probe was run off the gels to better resolve heterodimeric complexes from homodimeric complexes. 1). A control immunoprecipitation with irrelevant antiserum precipitates neither C/EBP␣ nor ATF-2 (lane 2). Significantly, inclusion of the chimeric binding site during precipitation with C/EBP␣ antibodies facilitates detection of ATF-2 as a co-precipitant (compare lane 3 to 4). When a C/EBP-binding site is included during immunoprecipitation, the efficiency of ATF-2 co-precipitation (lane 4) is diminished. These precipitates (lanes 3 and 4) show equivalent reactivity with C/EBP␣ antibodies (not shown), indicating that enhanced co-precipitation of ATF-2 in the presence of the chimeric binding site cannot be explained by a simple difference in the amount of C/EBP␣ precipitated. These results suggest that the subunits of ATF-2 and C/EBP␣ are in dynamic equilibrium, and that the composition of dimers is a reflection of the stoichiometry of the individual subunits. DISCUSSION To identify proteins that interact with C/EBP␣, we screened an expression cDNA library with a radiolabeled GST-C/EBP␣ fusion protein. Of the four cDNAs we isolated, three encoded nuclear proteins and one was a novel gene product. Three independently isolated cDNAs encoded the bZIP transcription factor ATF-2. Neither GST-GCN4, a yeast bZIP protein, nor GST itself, interact with ATF-2 under these conditions, indicating that interaction between ATF-2 and C/EBP␣ is specific. The fact that we did not isolate any abundant cytoskeletal protein that possesses a coiled-coil motif is a further indication of the stringency of the screen.
Although the protein domains mediating association mapped to the respective leucine zippers (see Fig. 2, A and B), weak association of ATF-2 with an immobilized C/EBP␣ fusion protein containing the DNA-binding domain but lacking the leucine zipper was consistently observed (Fig. 2A, lane 2). Reciprocally, we observed weak association of soluble C/EBP␣ with immobilized ATF-2 constructs containing only the aminoterminal domain of the protein (Fig. 2B, lane 2). This is likely due to a putative zinc finger motif located near the amino terminus of ATF-2. This motif interacts with bZIP domains (54), and may be involved in negative autoregulation of ATF-2 function (55). Although it is possible that this motif augmented detection during our initial screen, it is clear that ATF-2 interacts with C/EBP␣ primarily by leucine zipper-mediated dimer formation.
ATF-2:C/EBP␣ heterodimers form a DNA binding complex with a target specificity that differs from the parental homodimers. The heterodimers do not bind symmetric DNA elements like consensus C/EBP and CRE sites, (see Fig. 3, A and  B), but bind to an asymmetric sequence consisting of one consensus half-site for each monomer (Fig. 3C). It is noteworthy that ATF-2:C/EBP␣ heterodimers do not bind the CRE sequence, considering that both parent homodimers bind this site (Fig. 3B). This suggests that heterodimer formation results in increased DNA binding selectivity, as C/EBP␣ homodimers bind all three sites, while ATF-2:C/EBP␣ heterodimers bind only one.
Oppositely charged side chains at specific positions in each helix of a leucine zipper form interhelical electrostatic contacts that stabilize the dimers (56). Comparison of the charges at appropriate positions within the C/EBP␤ and ATF-2 leucine zippers revealed the occurrence of potential stabilizing contacts. EMSA analysis revealed that such heterodimers can form (Fig. 4), and further suggest that ATF-2 subunits may be FIG. 4. ATF-2 forms heterodimers with C/EBP␤. A radiolabeled chimeric binding site was incubated with proteins as indicated below the figure. The ATF-2:C/EBP␤ heteromeric shift is indicated by the arrow (lane 5). The C/EBP␤ protein sample has a partial degradation product, which is indicated.
FIG. 5. ATF-2 affects C/EBP␣dependent transcription activation in a transient assay. The hepatoma cell line Fao was transfected with reporter plasmid alone, or with reporter plasmid plus the indicated mammalian expression vector. All transfection results were normalized to ␤-galacosidase activity, and represent the average of three independent experiments. The left panel shows luciferase activity driven from the minimal thymidine kinase promoter containing C/EBP-binding sites. The right panel shows luciferase activity driven from the minimal thymidine kinase promoter containing chimeric sites. A scrambled chimeric site reporter vector was used to verify activation dependence upon the chimeric binding site. The numbers indicate micrograms of plasmid transfected. capable of forming heterodimers with all C/EBP family proteins. Heterodimeric DNA binding complexes involving subunits from different bZIP protein families have been described previously (7, 58 -61). For example, C/EBP␤ was reported to form heterodimers with C/ATF, another CREB/ATF family protein. The heterodimer bound a different asymmetric CRE element which is found in the promoters of several liver genes (60).
The principle effect of ATF-2 on C/EBP␣ dependent transcription is transcriptional interference (Fig. 5, left panel). ATF-2:C/EBP␣ heterodimers cannot bind consensus C/EBP sites, and thus decrease the pool of C/EBP␣ homodimers available to regulate the reporter gene. Similar results were obtained upon heterodimer formation between C/EBP␤ and Jun (61), where "repression" of transcription from C/EBP sites was reported. This is in contrast to results obtained when the C/EBP-binding sites are replaced by the chimeric sequence.
When both ATF-2 and C/EBP␣ are co-transfected, reporter activity exceeds that observed with either transcription factor alone (Fig. 5, right panel). Taken with the results using the C/EBP site reporter, these findings are consistent with the interpretation that ATF-2:C/EBP␣ heterodimers form, and affect transcription activity.
Both ATF-2 and C/EBP␣ are expressed in liver. Using freshly prepared rat liver nuclear extracts, we showed that immunoprecipitates for C/EBP␣ react with affinity purified ATF-2 antibodies, and that the amount of ATF-2 co-precipitated could be enhanced or diminished in a predictable fashion by inclusion of specific DNA-binding sites. Together, the results of transient transfection and immunoprecipitation studies are consistent with the formation of ATF-2:C/EBP␣ heterodimers in vivo. The fact that ATF-2 can associate with C/EBP␤ indicates that this association can probably be extended to other C/EBP family proteins. One functional consequence of this interaction is inhibition of transcriptional activity from consensus C/EBP target sites. It is also likely that heterodimers have a positive impact on transcription from promoters containing chimeric DNA elements. A search of primate promoter sequences in the data base using the 8-nucleotide chimeric DNA element indicates that the promoters of the protein disulfide isomerase, glutathione S-transferase, and ornithine decarboxylase genes are candidates for this kind of regulation.
Heterodimer formation is appealing from a regulatory viewpoint because of the asymmetry that is generated. By selecting for asymmetric DNA elements, heterodimers bind their target in an orientation dependent fashion, presenting distinct surfaces for interaction with proteins bound to adjacent DNA sites. Both the position and the orientation of protein-binding sites within some enhancer elements are important for the formation of what has been termed the "stereospecific" complex (57). Although the function of ATF-2:C/EBP heterodimers is not known, it has been proposed that cross-family heterodimers serve as a common target for the integration of signals arising from different extracellular stimuli. In this model, each subunit of the heterodimer would be modified by a unique protein kinase in response to only one of the signals being transduced. Since each subunit of the heterodimer is independently modified in this scenario, signals from two pathways converge, leading to the appropriate changes in the pattern of gene expression (51,60,61). Lane 5, anti-ATF-2 blot of a precipitate formed with a C/EBP␣ antibody directed against the carboxyl terminus of the protein. The band in lanes 3-5 labeled Nsp is nonspecific, as it is also observed after precipitation with an irrelevant antiserum. B, C/EBP␣ immunoprecipitates were formed in the presence of 20 pmol of specific double-stranded DNAbinding sites as indicated below the figure. Lane 1, Western analysis of 30 g of RLNE to show the position of ATF-2. Lane 2, irrelevant antibody control. Lane 3, precipitation in the presence of the chimeric binding site. Lane 4, precipitation in the presence a consensus C/EBPbinding site. After electroblotting, the filter was reacted with biotinylated ATF-2 antibody, and developed with horseradish peroxidase-conjugated streptavidin. Molecular weights were determined by prestained standards included on the gels.