The CCAAT-binding proteins CP1 and NF-I cooperate with ATF-2 in the transcription of the fibronectin gene.

We have previously proposed a molecular interaction between the liver factors that bind to the cyclic AMP response element (CRE) and CCAAT sites of the fibronectin (FN) gene based on the following evidence: (i) the close spacing of 20 base pairs between CRE and CCAAT elements is conserved in the FN genes from rats, mice, and humans; (ii) footprinting competitions showed that CRE oligonucleotides are able to detach both liver factors; (iii) CCAAT binding and transcriptional activity of liver extracts are reduced when the distance between the CRE and CCAAT elements is increased; and (iv) CCAAT-binding is stimulated by the addition of a liver extract fraction containing the CRE-binding factor ATF-2. This report provides binding and immunochemical evidence that nuclear factor I (CTF/NF-I) and CP1 (NF-Y or CBF) are the only liver factors that bind to the -150 CCAAT element of the FN gene, forming distinct complexes. We show that these factors bind less efficiently to the CCAAT site of a FN promoter in which the -170 CRE has been disrupted by site-directed mutagenesis and that each element contributes positively to the liver transcriptional activity assessed in vitro with a G-less cassette construct and in vivo by transfection of hepatoma cells with CAT constructs. Furthermore, using a method that combines UV cross-linking and immunoprecipitation, we show that antibodies specific to ATF-2 are able to specifically precipitate protein-protein-DNA complexes containing NF-I and CP1. This simple method preserves weak macromolecular interactions, avoiding the disruptive electrophoresis conditions of gel mobility shifts assays.

Advances in our knowledge of the physiology of gene regulation will ultimately require the precise definition of the multimolecular processes that are built up on a given gene's DNA by both the well known and still unknown factors. Ideally, the model gene should be widely expressed and regulated at the levels of transcription and splicing. Besides fitting these re-quirements (1), the fibronectin (FN) 1 gene offers a suitable model because it has an essential role in development (2), and changes in its expression pattern are involved in pathology, particularly of the liver (3). The importance of studying FN regulation in liver is highlighted by the fact that although many tissues express FN, liver is the major source of the plasma form of FN (4), made of splicing variants different from those of the extracellular matrix forms. Moreover, liver fibrosis, a common pathological condition preceding cirrhosis, is accompanied by dramatic changes in FN expression (3), whose mechanisms are still poorly understood.
The Ϫ170 cyclic AMP response element (CRE; 5Ј-TGACGTCA-3Ј) of the FN gene appears to be involved in different regulations of the promoter. Besides mediating cAMP stimulation (5-7), this element was unexpectedly shown to be essential for responsiveness to serum (8) in an additive way and therefore independent from cAMP. Viral induction by adenovirus E1a oncoprotein has also been shown to be mediated through the FN-CRE site (9).
DNase I footprints of the proximal region (Ϫ220 to ϩ65) of the FN gene differ considerably between liver and other cell type extracts. Although the Ϫ170 CRE is occupied by all tested extracts, strong binding to the Ϫ150 CCAAT is only observed with liver extracts. Three facts suggested an interaction between the liver factors that bind to these CRE and CCAAT sites: (i) A close spacing of 20 bp between CRE and CCAAT elements is conserved in the FN genes from rats, mice, and humans. (ii) Footprinting competitions showed that CRE oligonucleotides were able to detach both liver factors, suggesting that binding of a CRE factor to its cognate site is needed for the occupation of the neighbor CCAAT site placed two helical turns away. (iii) CCAAT binding and in vitro transcriptional activity of liver extracts were reduced when the distance between the CRE and CCAAT elements was increased in a series of spacing mutants (10).
The FN-CRE factor that cooperates with CCAAT binding in liver was shown to be an heterodimer between a 43-kDa polypeptide and the 70-kDa ATF-2 (11). ATF-2 is a ubiquitous transcription factor containing a basic region-leucine zipper motif that interacts with transcriptional activators that lack sequence-specific DNA binding activities like the retinoblastoma protein (Rb) (12) and viral activators such as adenovirus E1a (13) and HTLV-1 Tax (14), which are then tethered to CRE-containing promoters. ATF-2 heterodimerizes with c-JUN and this heterodimer has been implicated in the viral regulation of the ␤-interferon gene (15) through synergistic interac-tions between the ATF-2 subunit and the transcription factors HMG-I (Y) and NF-B. Phosphorylation of ATF-2 by the cJUN NH 2 -terminal protein kinase has been shown to modify three ATF-2-mediated transcriptional activations: induction by serum and trans-activations by Rb and E1a (16). The central role of ATF-2 is evidenced by the severe abnormalities observed in ATF-2 knock-out mice (17).
The FN-CCAAT-binding proteins have not been identified before. At least four families of transcription factors recognize the CCAAT motif. The best characterized is that of CCAAT/ enhancer binding protein (C/EBP), a leucine zipper dimeric factor related to myc, fos, and CREB, with defined roles in terminal cell differentiation (18) and long term memory (19). A second family is represented by nuclear factor-I (called CTF/ NF-I or simply NF-I), with the dual role of transcription factor for RNA polymerase II and initiation factor for adenovirus DNA replication (20). A third group includes a ubiquitous factor known as CP1 (21), NF-Y (22), or CBF (23). It is now clear that CP1 is a heterotrimer (22). Two of its subunits, CP1 A (also named NF-YB or CBF-B) and CP1 B (also named NF-YA or CBF-A) are homologous to the yeast CCAAT factors Hap3 and Hap2, respectively, which control nuclear genes required for mitochondria oxidative function (24). The fourth group is represented by CP2 (21), a heterodimer that controls the ␣-globin gene transcription in erythroid cells (25).
This report provides binding and immunochemical evidence that NF-I and CP1 are the liver factors that bind to the Ϫ150 CCAAT element of the FN gene, forming distinct complexes. We show that these factors bind less efficiently to the CCAAT site of a FN promoter in which the Ϫ170 CRE has been disrupted by site-directed mutagenesis. Furthermore, using a new method that combines UV cross-linking and immunoprecipitation, we show that antibodies specific to ATF-2 are able to recognize specific protein-DNA complexes containing NF-I and CP1.

EXPERIMENTAL PROCEDURES
Antibodies-Anti-NF-YA (COOH-terminal peptide) and NF-YB (whole recombinant protein) sera were generously provided by R. Mantovani and D. Mathis (26). Anti-CTF/NF-I serum was a gift from N. Tanese and R. Tjian. Anti-recombinant CP-2 IgG was obtained from M. Sheffery (25). Anti-CBF-A (CP1-B) IgG was a gift of B. De Crombrugghe (27). Anti-ATF-2 and anti-CREB IgGs were purchased from Santa Cruz Biotechnologies; anti-ATF-2 (catalog number sc-187 X, supershift reagent) is directed to an epitope corresponding to amino acids 487-505 of the carboxyl terminus of human ATF-2, and anti-CREB (catalog number sc-186 X, supershift reagent) is directed to an epitope corresponding to amino acids 295-321 mapping near the carboxyl terminus of human CREB-1. Anti-human ATF-2 and anti-human CREB are also reactive to the rat and mouse counterparts.
Protein Extracts-Nuclei were isolated from perfused rat liver according to Gorski et al. (28). Nuclear proteins were extracted according to Dignam et al. (29) in 0.37 M NaCl, in the presence of a mixture of protease inhibitors: 2 mM benzamidine, 0.5 mM phenylmethylsulfonyl fluoride, 5 g/ml aprotinin, and 5 g/ml leupeptin, concentrated by ammonium sulfate precipitation, dialyzed, frozen, and stored at Ϫ70°C.
Western Blots-Crude nuclear extracts (100 g of protein) were separated by sodium dodecyl sulfate gel electrophoresis in 10% polyacrylamide gels, and the proteins were electrotransferred to nitrocellulose membranes. The filters were probed with specific antibodies raised against ATF-2, CTF/NF-I, and NF-YA and developed using the Vectastain kit (peroxidase reaction, Vector Laboratories).
Oligonucleotides-Oligodeoxynucleotides used in this work are shown in Table I. Double stranded oligodeoxynucleotides were labeled FIG. 1. In vitro transcription of a human FN/Xenopus albumin promoter hybrid construct. Evidence that competitions with either FN-CCAAT or FN-CRE inhibit transcriptional activity elicited in vitro by liver extracts is shown. Top, schematic drawing of the promoter and regulatory regions of the construct pwt-G-free used as template. Bottom, autoradiograph of the in vitro transcription products of pwt-G-free and of a Rous sarcoma virus G-free control construct, in which a G-free cassette is under the control of the Rous sarcoma virus LTR. Reactions were carried out using 10 g of protein of liver nuclear extract and 0.5 g of plasmid DNA as template. The percentage of activity was estimated from densitometric scanning quantitation of the autoradiographs.
Band Shift Assays-Nuclear extracts (up to 5 g of protein) were preincubated for 10 min on ice in the presence of 0.5-1.0 g poly(dI-dC), 8 mM MgCl 2 /spermidine, 10 mM Hepes, pH 7.9, 30 mM KCl, 0.125 mM EDTA, 0.065 mM EGTA, 10% (v/v) glycerol, 1 mM dithiothreitol, and the "mixture" of protease inhibitors mentioned above, in a total volume of 20 l. When indicated, antibodies or competitor oligonucleotides were added to the preincubation. Binding reactions were started by the addition of 1 ng of the labeled probe and proceeded for 10 min. Complexes were separated on a 6% polyacrylamide, 0.25 ϫ TBE (1 ϫ TBE is 90 mM Tris borate, 2.5 mM EDTA, pH 8.3), native gel as described by Garner and Revzin (30).
Mutagenesis-In the construct named mut CRE, we changed the GTGACGTCA sequence to aTGgCtTCA by site-directed mutagenesis, using a U. S. Biochemical Corp. kit. In a second construct, named i⌬CRE for internally deleted CRE, the CRE sequence TGACGTCA was shortened to TGCA, as described (36). The construct named mut CCAAT was obtained by site-directed mutagenesis by changing the CCAAT sequence to CCccT. Double stranded oligonucleotides carrying these mutations (FN-mutCRE, FN-i⌬CRE, and FN-CCccT; Table I) were at least 100-fold less efficient in binding liver nuclear factors in gel shifts (not shown).
Cross-linking and Immunoprecipitation-This new method was inspired in the procedure reported by McKay (31). 20-l binding reactions of 32 P-labeled FN-CCAAT oligonucleotide to proteins in nuclear extracts were carried out in 1.5-ml Eppendorf tubes and irradiated with a 254 nm hand UV lamp (Spectroline, model ENF 240 C/F), for 30 min. The lamp was placed on top the uncapped tubes, which were kept in an ice/water bath. After this, 1 l of anti-ATF-2 purified IgG (see above) or other antibodies were added, and incubation proceeded at 0°C for further 30 min. Immunocomplexes were separated by incubation for 30 min with a 10% (w/v) suspension of fixed Staphylococcus aureus (Immunoprecipitin, BRL) and centrifugation at 4,000 rpm for 5 min in an Eppendorf microfuge. The amount of immunoprecipitated probe was determined by measuring Cerenkov radiation in the pellets, with or without previously washing the pellets with phosphate-buffered saline by resuspension and centrifugation.
DNase-I Protection Assays-For the footprinting analysis (32), 388-bp XbaI/HindIII fragments, which include FN promoter sequences between Ϫ220 and ϩ65, were isolated from wild type or mutant constructs (a detailed description of the constructs used for obtaining the promoter fragments will be published elsewhere). 2 Fragments were end-labeled with [␣-32 P]dATP and Klenow at their HindIII sites. Footprinting reactions were as previously reported (10), with marker GϩA ladders generated as described (33).
In Vitro Transcription-In vitro transcription was performed using as template the construct pwt-G-free described in Ref. 10. This construct has the proximal 220 bp of the FN promoter without its TATA box but contains the Xenopus albumin gene TATA box. In vitro transcription was performed as described by Sawadogo and Roeder (34)  Hepes, pH 7.6, and 10 g of protein extract. After incubating for 45 min at 30°C, transcription was stopped by the addition of 5 mM EDTA, 1% (w/v) SDS, 250 mM NaCl, and 20 mM Tris-HCl, pH 7.4 (final concentrations), and samples were treated with proteinase K, ethanol-precipitated, and analyzed in a 6% polyacrylamide urea sequencing gel.
Transfections and CAT Assays-Segments of the FN promoter (-224 to ϩ64) carrying either wild type or mutated CRE and CCAAT sites were linked to the CAT gene. These constructs were used to transfect the human hepatoma cell line Hep3B in triplicates. Cells were cotransfected with a Rous sarcoma virus-␤ galactosidase reporter plasmid as measure of transfection efficiency. Transfection conditions and CAT assays were as described previously (46).
Quantitation of Autoradiograms-Bands in autoradiograms were quantitated densitometrically in a Ultroscan XL Enhanced Laser Densitometer (Pharmacia Biotech Inc.). Quantitation of protein binding equilibria in DNase I footprints and calculations of the fractional protection factor f were as described in Ref. 10.

RESULTS
The Liver Ϫ170 CRE-and Ϫ150 CCAAT-binding Proteins Contribute Positively to the Transcriptional Activity of the FN Gene-We have previously demonstrated that increasing the distance between the CRE and CCAAT sites of the FN gene reduced liver transcriptional activity in vitro (10). Fig. 1 shows that 50-fold molar excesses of either ds FN-CRE or FN-CCAAT oligonucleotides inhibited in vitro transcription of liver extracts upon a G-less cassette template under the control of the FN promoter. At the same molar excess (50ϫ), FN-CRE is more efficient in bringing down transcription (37% remaining activity) than FN-CCAAT (63% remaining activity), confirming the preponderant role of the CRE in the composite site. Competition with FN-CCccT, a FN-CCAAT mutant oligonucleotide unable to bind liver factors, does not affect transcription (not shown). The Rous sarcoma virus-LTR was used as reference promoter. It is worth noting that although LTR transcription is not inhibited by FN-CRE (in fact it is activated by 50%), it is negatively affected by competition with FN-CCAAT. We believe that this is due to the fact that viral LTRs, as well the adenovirus major late promoter also tested as control (not shown), contain DNA elements recognized by CCAAT-binding proteins. Anyway, the inhibition caused by FN-CCAAT on the FN-promoter is more important than that of the LTR. These results indicate that the CRE and CCAAT sites are not dispensable for transcription in liver and that the factors interacting with them are positive regulators.
The Fibronectin "NF-I Site" Not Only Binds NF-I-The liver factors that bind to the Ϫ170 CRE sequence have been characterized by Southwestern analysis and include polypeptides of 120, 70, and 43 kDa (10). The 120-kDa protein is also found in HeLa cells, and little is known about its functions (35). The 43-kDa factor is thermoresistant and resembles a member of the CREB family. The 70-kDa polypeptide was identified as ATF-2 and seems to be more interesting because it promotes the occupation of the neighboring Ϫ150 CCAAT element (11). This observation prompted us to identify the liver factors that recognize the CCAAT site. Band shifts of liver factors to the FN-CCAAT probe systematically show a pattern characterized by a sharp band on top of a smear of higher mobility ( Fig. 2A,  lanes 1 and 4 -9). Both binding complexes are competed efficiently by homologous FN-CCAAT oligonucleotides ( Fig. 2A,  lanes 2 and 3) but not by as much as 100-fold excess of FN-CRE ( Fig. 2A, lanes 4 and 5) or the core of the SV40 enhancer ( Fig.  2A, lanes 8 and 9) used as controls.
The Ϫ150 CCAAT element of the FN gene had been described as a NF-I site (8) based on sequence comparison. However, whereas competition with cold FN-CCAAT abolished binding to both observed complexes (Fig. 2, A, lanes 1-3, and B,  lanes 1 and 2), competition with an oligonucleotide with the consensus sequence for adenovirus NF-I only eliminated the bottom smear, leaving the top band unaffected (Fig. 2B, lane 3). Accordingly, the liver complexes observed with a 30-bp NF-Ilabeled probe co-migrate with the bottom smear of the 32-bp FN-CCAAT probe (Fig. 2B, lane 4). These results showed that besides NF-I, the Ϫ150 CCAAT is interacting with other CCAAT factors that form the top band complex.
The Ϫ150 FN-CCAAT Also Interacts with CP1 (NF-Y/CBF) but Not with C/EBP nor CP2-Binding to FN-CCAAT is not competed by the consensus sequence for the C/EBP (Fig. 2A,  lanes 6 and 7). We conclude that C/EBP does not interact with FN-CCAAT, which is consistent with the fact that C/EBPbinding activity is resistant to heat inactivation, whereas the liver FN-CCAAT-binding activity is inactivated by heating at FIG. 3. Effect of the i⌬CRE mutation on the binding of liver nuclear factors. DNase I protection assays of FN promoter fragments with liver nuclear extracts are shown. DNA fragments were 32 P-labeled, incubated in the absence or the presence of nuclear extracts from rat liver, treated with DNase I, and electrophoresed on 8% denaturing gels as described under "Experimental Procedures." Nucleotide positions were determined by running a G ϩ A Maxam and Gilbert degradation (not shown) obtained by a rapid method (33). Numbering of DNasesensitive bands is referred to nucleotide positions in the wild type. The Ϫ154 and Ϫ174 positions indicate the bands that disappear after protection of the CCAAT and CRE sites, respectively. All binding reactions were performed with the same amount of DNA probe (1 ng) and protein extract (about 5 g). Note that all hypersensitive bands in lane 3 are stronger, due to a higher recovery in the precipitation of the digested probe.
65°C for 10 min (10). Comparison of the consensus sequences for the three CCAAT factors C/EBP, NF-I, and CP-1 with that of FN-CCAAT (Fig. 2F) confirms that the latter exhibits more sequence identity with the sites for NF-I and CP1 than for that of C/EBP. This observation led us to investigate the participation of CP1 (also named NF-Y or CBF) in the binding to FN-CCAAT, with the aid of specific antibodies. Results in Fig. 2C, clearly show that an antiserum raised specifically against NF-I was able to supershift the smear complex without affecting the top band (lane 2), whereas antibodies to the A chain of CP1 (anti-NF-YB) supershifted the top band complex without altering the bottom smear (lane 3). The top band is also supershifted by antibodies to the B chain of CP1 (anti-NF-YA, Fig. 2D, lanes  8 and 9) and by an antibody raised independently against the A chain of factor CBF (CBF-A, equivalent to CP1-B) (Fig. 2D,  lanes 2 and 3). An antiserum against the CCAAT-binding protein CP2 found to interact with the ␣-globin promoter (25) does not alter FN-CCAAT binding (Fig. 2D, lanes 4 and 5).
To confirm the previous results we depleted liver extract aliquots from specific factors by incubating them with specific antibodies as described under "Experimental Procedures." The remaining FN-CCAAT binding activities were visualized in gel shift assays. The extracts immunodepleted by anti-NF-YB and anti-NF-I presented bottom smear and top band patterns, respectively (Fig. 2E, lanes 1 and 2). A standard gel shift pattern was observed with extracts treated either with no antibody, rabbit normal serum, or antibodies specific to the TATA-binding protein (Fig. 2E, lanes 3-6), used as controls.
Additional evidence favoring involvement of CP1 is provided by co-migration of the top FN-CCAAT band with the complex formed by a CP1 probe, of equivalent size, with liver nuclear factors (see Fig. 7, top, lanes 1 and 2). Altogether these results support the conclusion that NF-I and CP1/NF-Y are the liver factors that bind to the Ϫ150 CCAAT element of the FN gene and that C/EBP and CP2 are not involved.
Disruptive Mutations of the Ϫ170 CRE Affect the Binding to the Ϫ150 CCAAT-We have previously shown that increasing the spacing between the CRE and CCAAT elements inhibited the occupation of the CCAAT box. A similar effect was obtained with a construct in which all promoter sequences upstream of the CCAAT box, including the Ϫ170 CRE, were deleted (named ⌬CRE in Ref. 10). These experiments did not exclude that the sequences upstream of the CRE cooperate with the CCAAT site. To rule out this possibility and get a more direct evidence of the role of the CRE site, we prepared two different disruptive mutations of the Ϫ170 CRE, within the context of a Ϫ220 ϩ65 fragment of the FN promoter (see "Experimental Procedures"). DNase I footprints of the wt and i⌬ CRE promoters are shown in Fig. 3. In the wild type, both the Ϫ170 CRE and Ϫ150 CCAAT are protected. In the CRE mutant, the Ϫ170 CRE remains unprotected, and this disruption affects negatively the binding to the neighbor Ϫ150 CCAAT box. Quantitation of four separate experiments is expressed as fractional occupation at the bottom of Fig. 3. Similar results were obtained with the mut CRE constructs (not shown).
Immunoprecipitation of CCAAT-binding Proteins Complexed with ATF-2-The cooperative effect of the CRE site upon the binding to the CCAAT box might involve specific protein-protein interactions between ATF-2 and CP1 or NF-I. However, we failed to demonstrate this interaction in gel shifts, probably due to disassembly of complexes during electrophoresis.
In our search for a nonelectrophoretic method that could FIG. 4. A, schematic representation of the immunoassay for evaluating ATF-2/ CCAAT-binding protein complexes through precipitation of the 32 P-labeled FN-CCAAT oligonucleotide. The scheme depicts ATF-2 as part of a heterodimer (11) and shows that anti-ATF-2 is directed to an epitope of the dimerization domain (manufacturer's technical information). Representation of the CCAATbinding proteins is speculative. B, gel electrophoresis (20% polyacrylamide) of the radiolabel immunoprecipitated by anti-ATF-2 in a typical immunoassay of a liver nuclear extract. Binding reactions were competed with a 20ϫ molar excess of the indicated ds oligonucleotides. The arrow marks the position of the 32-bp FN-CCAAT probe run as standard. C, gel retardation assay of a FN-CCAAT probe in the absence or the presence of anti-ATF-2 in the preincubation. D, Western blot to proteins present in liver nuclear extracts, reduced with ␤-mercaptoethanol, separated by size on SDS-10% polyacrylamide gel electrophoresis, and developed with specific antibodies to CP1 (anti-NF-YA), NF-I, and ATF-2. E, SDS-polyacrylamide gel electrophoresis of liver nuclear proteins cross-linked to 32 P-labeled FN-CCAAT and FN-CRE probes and immunoprecipitated by anti-ATF-2. Conditions were described under "Experimental Procedures."

ATF-2/CCAAT-binding Protein Cooperation
detect weak but specific protein-protein interactions, we explored the hypothesis that antibodies to ATF-2 would be able to precipitate a labeled FN-CCAAT probe bound to liver nuclear proteins after the formation of a "four-member sandwich" (anti-ATF-2/ATF-2/CCAAT proteins/ 32 P-FN-CCAAT), such as the one depicted in Fig. 4A. This new approach was inspired in the "three-member sandwich" immunoprecipitation (antibody/ transcription factor/labeled DNA) described by McKay (31). We found that indeed anti-ATF-2 was able to quantitatively precipitate 32 P label when added to a FN-CCAAT binding reaction. Nevertheless, a series of control experiments were necessary to assure us that this was not an artifact. Fig. 4B shows that the immunoprecipitated radioactivity corresponds to a bona fide FN-CCAAT oligonucleotide and not to a degradation product. It is also shown that the amount of precipitated oligo decreases when binding is competed by unlabeled FN-CCAAT but not by the mutant FN-CCccT. The inability of anti-ATF-2 to supershift FN-CCAAT complexes and to induce additional bands in the gel shift shown in Fig. 4C clearly indicates that anti-ATF-2 does not interact directly with the CCAAT-binding proteins nor with the FN-CCAAT probe alone and that ATF-2 does not bind the FN-CCAAT probe. Most importantly, in Western blots of whole liver extracts, anti-ATF-2 only recognizes the 70-kDa ATF-2 polypeptide, not being able to react directly neither with NF-I (about 50 kDa) nor with CP1 (NF-YA, 38 kDa), also present in the extract (Fig. 4D).
To improve the efficiency of these experiments, we stabilized the DNA protein interaction by covalent cross-linking with 254 nm UV light, as described under "Experimental Procedures." In these conditions, the FN-CCAAT probe was precipitated by anti-ATF-2 and, as expected, by anti-CP1 (a mixture of anti-NF-YA and anti-NF-YB) and anti-NF-I, with significantly higher efficiencies than that of a control serum (Fig. 5A). However, an anti-CREB was not effective in precipitating the CCAAT probe (Fig. 5C). This result is important because it confirms the distinct role of ATF-2 and not CREB as the CREbinding protein cooperating with the CCAAT site (11). Anti-CREB was also a good negative control, because both anti-CREB and anti-ATF-2 are "supershift" IgGs, prepared at the same protein concentration by the same manufacturer (see "Experimental Procedures"). The possibility that the immunoprecipitated label is the result of a direct interaction between ATF-2 and the FN-CCAAT probe was ruled out by two control experiments: (i) immunoprecipitation of the CCAAT probe is not affected by competition with a 50ϫ molar excess of the CRE oligonucleotide (Fig. 5A); (ii) SDS-polyacrylamide gel electrophoresis of UV-cross-linked complexes immunoprecipitated by anti-ATF-2, obtained with either FN-CRE or FN-CCAAT probes showed distinct band patterns (Fig. 4E). Another critical point is to demonstrate that anti-ATF-2 does not precipitate any labeled ds oligonucleotide of the same specific activity as FN-CCAAT. Fig. 5A (right) shows that the mutant FN-CCccT probe is not precipitated by anti-ATF-2, confirming the specificity of our assay. The FN-CRE is precipitated by anti-ATF-2 but not by anti-CP1 (NF-Y) and anti-NF-I (Fig. 5B), the same antisera that proved to be active in precipitating labeled FN-CCAAT (Fig. 5A, left). This might indicate that unlike anti-ATF-2, anti-CP1 (NF-Y) and anti-NF-I interfere with complex formation. Experiments shown in Fig. 6 provide further support for specificity; competitions with a 25-fold excess of unlabeled FN-CCAAT significantly inhibit the amount of labeled FN-CCAAT precipitated by anti-ATF-2 and anti-CP1 antibodies. The FN-CCccT mutant had no effect. This competition is particularly important because it certifies that ATF-2 interacts with factors that discriminate between wild type and mutant FN-CCAAT oligos. These factors should be NF-I, CP1, or both. To address their nature we assayed NF-I and CP1 probes in immunoprecipitation experiments. Fig. 7 (bottom) shows that anti-ATF-2 is also able to precipitate CP-1 and NF-I labeled probes, specifically bound and cross-linked to liver nuclear proteins, as evidenced in the gel shift of Fig. 7 (top).
All these experiments and their controls suggest that immunoprecipitation of the CCAAT probe by anti-ATF-2 is reflecting a specific interaction between ATF-2 and the liver FN-CCAATbinding factors CP-1 and NF-I. The fact that addition of the FN-CRE ds oligonucleotide to binding reactions did not stimulate immunoprecipitation of the FN-CCAAT probe by anti-ATF-2 suggests that protein complexes are formed in solution in the absence of the target DNA for ATF-2. However, this procedure does not allow us to tell if the assembly of ATF-2/ CCAAT-factors complexes occurs in the absence of FN-CCAAT DNA.
Transient Expression in Human Hepatoma Cells-The cooperative effect observed in vitro with liver nuclear extracts seems to correlate with the in vivo expression of FN-CAT constructs transfected in Hep3B cells. Fig. 8 shows that mutating both the CRE and CCAAT sites decreases promoter activity by 89%. The CCAAT mutation alone inhibits transcription by 40%. If the CRE and CCAAT sites were to act in an additive way, one would expect the CRE mutation to bring down activity by about 49%. However, the CRE mutant alone causes a 70% inhibition of transcription, suggesting a synergistic function of the CRE and CCAAT sites, where altering the CRE site might affect basal promoter activity by preventing not only factor binding to CRE but also to CCAAT.

DISCUSSION
The liver-specific occupation of the CRE and CCAAT sites of the FN promoter provides a useful model for studying tissuespecific cooperation between transcription factors. The liver factors that bind to each of the two sites are positive modulators as shown by competition of in vitro transcriptions (Fig. 1). This agrees with the fact that disrupting either the CRE or the CCAAT sites inhibits transcriptional activity in transfected hepatoma cells.
Before studying the functional role of the CRE-CCAAT interaction in response to regulatory signals, it is important to characterize the liver proteins that participate in this site interaction. In a previous report we found that liver ATF-2 was responsible for the cooperative effect (11). We now show that NF-I and CP1 but not C/EBP or CP-2 are the liver factors that recognize the Ϫ150 FN CCAAT. A series of complementary experiments involving competitions with sequence-specific oligonucleotides, coincidental migration of specific complexes in band shifts, supershifts, and immunodepletions with highly specific antibodies support our conclusions. It becomes clear that at least in liver, the Ϫ150 FN-CCAAT site is not simply a NF-I site (8), because after competition with a huge excess of NF-I oligonucleotide or supershifting with an anti-NF-I antibody, there still remains an important amount of bound FN-CCAAT probe. Experiments shown in Figs. 2 and 7 demonstrate that this factor is CP1. Ruling out of C/EBP is clear from experiments in Fig. 2 and from our previous observation that liver FN-CCAAT-binding factors are thermolabile (10), whereas C/EBP is typically thermostable. In fact, liver C/EBP is more likely to cross-talk with the Ϫ170 CRE, rather than with the Ϫ150 CCAAT site, as shown by Bakker and Parker (37). In any case, when antibodies specific to NF-I and CP1 were added together to a gel shift reaction, they abolished completely FN-CCAAT binding (not shown), indicating that the presence of other FN-CCAAT-binding proteins in liver extracts is highly unlikely.
Interaction between the CRE and CCAAT elements was first proposed by Muro et al. (10). Independently, Miao et al. (36) suggested a similar interaction between these two sites on the rat FN promoter. We provide here further and stronger evidence for cooperativity at the FN CRE and CCAAT sites in liver: two different mutations that disrupt the CRE site also inhibit occupation of the neighboring CCAAT box. On the other hand we found that anti-ATF-2 is able to precipitate the 32 Plabeled FN-CCAAT oligonucleotide cross-linked to NF-I and CP1 in liver extracts. This precipitation is inhibited by competitions with FN-CCAAT but not with a mutant oligonucleotide (FN-CCccT). The supershift and Western experiments in Fig. 4 respectively demonstrate that anti-ATF-2 does not react neither directly with the FN-CCAAT probe nor with NF-I or CP1. These and other experiments shown in the text not only put in evidence macromolecular assemblies between ATF-2, NF1, and CP1 but also illustrate the use of a simple, reliable, and fast method for studying those specific protein-protein interactions that are affected by disruptive electrophoretic conditions of gel shift assays.
These in vitro studies seem to correlate well with a mild but reproducible (see triplicates of transfections in Fig. 8) synergistic function observed in vivo in Hep3B human hepatoma cells transfected with wt and mutant FN-CAT constructs.
To our knowledge, this is the first evidence of a cooperation between NF-I, CP1, and ATF-2. We have not yet used purified proteins to test it. We did try, and failed, to detect proteinprotein interactions between rabbit reticulocyte lysates programmed with synthetic mRNAs for ATF-2 and CP1 (NF-YA ϩ NF-YB, not shown). This could be due to the absence in reticulocyte lysates of co-factors present in liver nuclear extracts such as the C subunit of CP1 (23) or the lack of an appropriate protein modification. Besides, recombinant ATF-2 made in reticulocyte lysates is a unmodified homodimer, whereas natural ATF-2 requires phosphorylation for binding DNA and activating transcription (16) and acts as a heterodimer. Preliminary results indicate that the liver partner of ATF-2 is not c-Jun, contrary to what was found in other systems (15). CP1 (NF-Y or CBF) is a multimeric factor that is required for tissue-specific activation of several genes, including albumin, and that plays a negative role in the transcription of the liverspecific aldolase B promoter (38). CP1 is not itself tissue-specific but is ubiquitous, and it is thought to act through interaction with appropriate tissue-specific or regulated factors to make possible the correct pattern of expression. This is particularly well illustrated by the strict need for CP1 in the steroldependent transcription of farnesyl diphosphate synthase and HMG-CoA synthase genes. In both cases, CP1 is not the regulatory protein but is essential for optimal activation, playing a role similar to SP1 in sterol-dependent regulation of the LDL receptor gene (39). Both CP1 and SP1 contain glutamine-rich activation domains and would cooperate with SREBP, the sterol response element-binding protein. None of the CP1 subunits shows homology with leucine zippers, helix-loop-helix domains, or the known protein dimerization domains. CP1 subunits are bound together by unique interaction motifs (23). This implies that if ATF-2 were to contact CP1, it would not dimerize through its leucine zipper. NF-I roles have been studied more extensively. This factor exists in several forms, one of them is particularly enriched in liver, and it is usually referred to as NF-I liver . Jackson et al. (40) demonstrated that NF-I liver and the hepatocyte-specific transcription factor HNF3␣ bind simultaneously to adjacent sites within the liver-specific enhancer of the serum albumin gene, creating a composite regulatory element with properties that differ from those of either factor bound alone. NF-I modulates positively or negatively HNF3␣ activity depending on the context of the albumin enhancer. NF-I liver also cooperates with HNF1, another liver-specific transcription factor, in liver-specific transcription from the tyrosine aminotransferase proximal promoter (41).
Two domains of NF-I have attracted interest for their surprising functions. One of them bears high sequence identity with the heptapeptide repeats of carboxyl-terminal domain of RNA polymerase II, and most importantly, the carboxyl-terminal domain-like domain is essential for the functioning of the proline-rich transcriptional activation domain of NF-I (42). The second domain is a subsegment of the proline-rich activation domain, called TRD (for TGF-␤-responsive domain), that mediates TGF-␤ induction in NIH-3T3 cells. TGF-␤ does not affect NF-I-TRD directly but up-regulates a specific and stimulatory interaction between TRD and histone H3, establishing an unprecedented link between growth factor signaling, chromatin dynamics, and NF-I (43).
In conclusion, a wide variety of cooperative effects that are relevant in basal, regulated, and tissue-specific transcription have been described for CP1 and NF-I. Accumulated evidence suggests that the CRE and CCAAT elements of the FN gene take part of a composite site. To demonstrate that it is a composite response element, such as the one described by Yamamoto and colleagues (for a review see Ref. 44), it becomes imperative to assess the role of the Ϫ150 CCAAT box, NF-I, and CP1 in modulating signals that regulate FN transcription through the Ϫ170 CRE, like cAMP, serum, and E1a, as well as the in vivo role of ATF-2. Another important question to investigate is the contribution of liver individual cell populations (hepatocytes, Kupffer cells, endothelial sinusoidal cells, and lipocytes) to the CRE and CCAAT-binding activities observed in whole extracts, focusing in the dramatic changes in FN expression that take place in healing processes triggered by TGF-␤ (1, 3), where NF-I is an obvious candidate for signal transduction.