Chicken Ovalbumin Upstream Promoter-Transcription Factor II, a New Partner of the Glucose Response Element of the L-type Pyruvate Kinase Gene, Acts as an Inhibitor of the Glucose Response*

Transcription of the L-type pyruvate kinase (L-PK) gene is induced by glucose in the presence of insulin and repressed by glucagon via cyclic AMP. The DNA regulatory sequence responsible for mediating glucose and cyclic AMP responses, called glucose response element (GlRE), consists of two degenerated E boxes spaced by 5 base pairs and is able to bind basic helix-loop-helix/leucine zipper proteins, in particular the upstream stimulatory factors (USFs). From ex vivo and in vivo experiments, it appears that USFs are required for correct response of the L-PK gene to glucose, but their expression and binding activity are not known to be regulated by glucose. A genetic screen in yeast has allowed us to identify a novel transcriptional factor binding to the GlRE, i.e. the chicken ovalbumin upstream promoter-transcription factor II (COUP-TFII). Binding of COUP-TFII to the GlRE was confirmed by electrophoretic mobility shift assays, and COUP-TFII-containing complexes were detectable in liver nuclear extracts. Neither abundance nor binding activity of COUP-TFII appeared to be significantly regulated by diets. In footprinting experiments, two COUP-TFII-binding sites overlapping the E boxes were detected. Overexpression of COUP-TFII abrogated the USF-dependent transactivation of an artificial GlRE-dependent promoter in COS cells and the glucose responsiveness of the L-PK promoter in hepatocytes in primary culture. In addition, a mutated GlRE with increased affinity for USF and very low affinity for COUP-TFII conferred a dramatically decreased glucose responsiveness on the L-PK promoter in hepatocytes in primary culture by increasing activity of the reporter gene in low glucose condition. We propose that COUP-TFII could be a negative regulatory component of the glucose sensor complex assembled on the GlRE of the L-PK gene and most likely of other glucose-responsive genes as well.

Carbohydrates are the major source of energy for mammalian cells. High carbohydrate intake results in the conversion of excess carbohydrates to triglycerides in the liver and adipocyte. This process is accompanied by the induction of many enzymes of the glycolytic and lipogenic pathways. This glucose response has been found to be, at least in part, at the transcriptional level, e.g. in the case of the L-pyruvate kinase (L-PK) 1 (1), aldolase B (2), spot 14 (S14) (3), acetyl-CoA carboxylase ACC (4), hormone-sensitive lipase HSL (5), and fatty acid synthase (6) genes. However, little information has been reported so far on the transacting factors that confer transcriptional glucose responsiveness on these genes.
Our laboratory has extensively studied the regulation of the L-PK gene by glucose. Doiron et al. (7) suggested that glucose is metabolized into active intermediates that in turn activate specific transcriptional factors that enhance the L-PK gene promoter. We defined a sequence responsible for mediating the positive response to glucose and the negative responses to cAMP in the proximal L-PK promoter from the position Ϫ172 to Ϫ142 bp, ex vivo by transient expression assays in hepatocytes in primary culture and in vivo in transgenic mice (8 -11). This DNA-binding site was termed the glucose response element (GlRE). This GlRE is closely related to the carbohydrate response element (ChoRE) described by Towle's group (12) in the regulatory region of the S14 gene and also to the glucose response element recently identified upstream of the glucagon receptor gene (13). In the context of the wild-type L-PK promoter, the complex that binds to this element cooperates with an adjacent DNA-binding site, the auxiliary site L3 (8,10), to confer a strong glucose responsiveness. This auxiliary site has been shown to bind mainly hepatocyte nuclear factor 4 (HNF4) in the liver but has also some affinity for chicken ovalbumin upstream promoter transcription factor (COUP-TF) (10,14) and nuclear factor 1 (15). However, our group and Towle's group showed that a GlRE multimer is sufficient to confer both carbohydrates and cAMP responses in the absence of the auxiliary site (8,10). The GlRE consists of two palindromic noncanonical E boxes (CANNTG) separated by 5 bp (12). The full activity of the L-PK GlRE seems to require the cooperation between these two E boxes (8,10). We have found that these E boxes are able to bind the upstream stimulatory factors (USFs), which are members of the basic helix-loop-helix/leucine zipper family (10,16). In vivo footprinting on the Ϫ183-PK promoter suggested that both E boxes are occupied in the liver (17). In vitro experiments showed that, in the liver as well as in most cells and tissues tested (18), USF binding activity is mainly accounted for by the USF1/USF2 heterodimer. By different approaches in cell culture and in vivo, we have shown that USF was implicated in the response of the L-PK gene to glucose (8,9,10,19). Accordingly, response of the L-PK, S14, and fatty acid synthase genes to dietary glucose is abnormal in USF2defective knock-out mice (20,21). In USF1Ϫ/Ϫ knock-out mice, only dietary response of the fatty acid synthase gene was impaired, whereas that of the L-PK and S14 genes was normal (21,22). However, USFs alone cannot account for the transcriptional response to glucose because most genes whose promoters include USF-binding E boxes are not regulated by glucose (23). In addition, the glucose responsiveness conferred by GlREs/ ChoREs is clearly not parallel to their affinity for USFs (24). Finally, gel shift assays with the L-PK GlRE as probe can detect, under some conditions, other binding proteins than USFs (Refs. 15 and 25 and this paper). We therefore undertook research of novel partners of the glucose response complex interacting with the L-PK GlRE.
Using a one-hybrid system in yeast with the GlRE as the target sequence, we cloned the COUP-TFII orphan nuclear receptor (also referred to as ARP-1) (26). COUP-TF is a member of the steroid/thyroid hormone receptor (TR) superfamily and was first identified as a homodimer that binds to a direct repeat regulatory element (DR1) in the chicken ovalbumin gene promoter (27). Indeed, COUP-TFII is able to bind to the GlRE in vitro, and COUP-TF-containing complexes interacting with the GlRE are detected in liver nuclear extracts. DNA binding activities of these complexes do not seem to be modulated by diets. COUP-TF-and USF-binding sites are overlapping, and consequently binding of these factors mutually interferes one on the other. Here, we demonstrate that overexpression of COUP-TFII not only inhibits USF-dependent transactivation of the L-PK promoter but also represses its stimulation by glucose in hepatocytes. Furthermore, a mutant GlRE with very low affinity for COUP-TFII conferred an impaired glucose response, because of increased activity under low glucose conditions. We propose that the glucose responsiveness mediated by the GlRE could involve a complex interaction between USF transactivators and COUP-TFII, probably in association with other, as yet not characterized partners. COUP-TFII-containing dimers could be involved in a glucose sensor system abrogating transactivation by USFs in the absence of glucose.

MATERIALS AND METHODS
Yeast Strain-The Saccharomyces cerevisiae strain used in this study was the CD156 strain (MATa,ade2,his3,leu2,lys2,trp1,ura3,gal4,gal80,cbf1⌬). The CYC1-HIS3 gene fusion used for the one-hybrid screen was as described in Blaiseau et al. (28). Wild-type GlRE motifs of the L-PK gene was cloned as two oriented copies, 200 bp upstream of the TATA box owing to the unique cloning site XhoI. The resulting plasmid was linearized with StuI (within the URA3 marker) and used to transform the CD156 strain. Stable uracil prototroph transformants were selected, and correct integration events were verified by Southern blot analysis.
Screening of a Mouse Embryo cDNA-VP16 Fusion Library-About 10 9 cells of strain YMV1 (cbf1::hisg, ura3::GlRE-HIS3::URA3) were transformed with 90 g of the cDNA fusion library (kindly provided by Dr. A. Vojtek) (29) in the presence of 300 g of salmon sperm carrier DNA according to Bartel and Fields (30). After transformation, the cells were recovered for 1 h in complete yeast medium and plated, and histidine prototroph, aminotriazole (AT, 30 mM)-resistant transformants were selected. From a screen of about 2 ϫ 10 6 transformants, 60 AT-resistant clones were selected after 3-10 days at 30°C. The corre-sponding plasmids were recovered and then tested for specificity by transforming yeast cells containing either the GlRE-HIS3 or the unrelated MEF3-HIS3 gene construct integrated in the genome. From this second round of selection, only one plasmid was shown to result in the specific activation of the GlRE-HIS3 reporter gene. The cDNA insert contained within this plasmid was sequenced using the Sanger dideoxy termination method adapted to the automated sequencer 373A from Applied Biosystems.
Plasmid Constructions-All plasmids were constructed by using standard DNA cloning procedures. The constructs were verified by nucleotide sequencing.
(LL) 2 -54PK/CAT and (MM) 2 -54PK/CAT constructs comprise two oriented copies of the L-PK GlRE (also called LL and L4 boxes) (8,10) or MM (10) ligated to the Ϫ54 to ϩ11 proximal base pairs of the L-PK minimal promoter (31). L-PK firefly luciferase (Luc) constructs are novel. Restriction or oligonucleotidic fragments of the L-PK gene corresponding to the regions Ϫ54, Ϫ96, or Ϫ183 to ϩ11 nucleotides with respect to the transcriptional start were subcloned into the basic plasmid pGL3 (Promega). (LL) 3 -96PK/Luc, (MM) 3 -96PK/Luc, (L3) 3 -54PK/ Luc, and (LL) 3 -54PK/Luc plasmids consist of three oriented tandem repeats of L-PK GlRE motif inserted in front of the Ϫ96 PK/Luc or Ϫ54 PK/Luc constructs, respectively. To correct for variations because of the transfection procedure and cell types, we cotransfected each test plasmid with a RSV/Rluc vector. This vector contains a cDNA encoding Renilla luciferase gene (pRL-null/Promega) driven by the RSV promoter.
For cell transfection experiments, we used the human COUP-TFII cDNA cloned into an eukaryotic expression vector, pMT2, driven by the adenovirus major late promoter and the simian virus 40 enhancer region (kindly provided by Dr. S. Karathanasis). Human USF2a was cloned into an eukaryotic expression vector, pCR3, driven by the cytomegalovirus immediate-early promoter region (18). The USF2 gene gives rise predominantly to full-length USF2a subunits, associated with two minor forms, USF2b (resulting from an alternative splicing out of an exon in the NH 2 domain) and mini-USF2, a NH 2 truncated form.
COUP-TFII used for in vitro transcription/translation was obtained by inserting the COUP-TFII EcoRI cDNA fragment into the pCR3 vector. The recombinant rat COUP-TFII expression vector pGEX-KG-COUP-TFII was a gift from J.-M. Boutin. The NcoI-XhoI fragment, containing the entire coding frame, was inserted in the same sites in the pGEX-KG vector (32). All plasmids were purified with a Qiagen kit.
Animals-Three-month-old Harlan Sprague-Dawley rats were subjected to different previously described nutritional and hormonal treatments to study L-PK expression in the rat liver (1). Animals were fasted for 48 h and then separated into three groups and refed for 24 h with different diets: high carbohydrate for the first group, high protein for the second group, and high fat for the third group. As previously demonstrated, we observed that the endogenous L-PK mRNA was abundant in the animals of the first group and scarcely detectable in the animals of the other two groups (data not shown). Other rats were fed with a regular chow ad libitum.
In Vitro Transcription and Translation-The plasmids pCR3/COUP-TFII and pCR3/USF2a were linearized with NotI and XbaI, respectively, and transcribed with T7 RNA polymerase in the presence of m 7 G(5Ј)ppp(5Ј)G (Roche Molecular Biochemicals). The resulting mRNAs were translated in vitro with rabbit reticulocyte lysate according to the manufacturer's instructions (Promega).
Recombinant Protein Production-GST-COUP-TFII protein was expressed in Escherichia coli BL-21 (DE3). Overnight cultures of bacteria that were newly transformed with the plasmid pGEX-KG-COUP-TFII were diluted with 10 volumes of medium, cultured for several hours to an optical density of 0.6 at 600 nm, and induced with 1 mM isopropyl-␤-D-thiogalactopyranoside at 37°C for 3 h. Bacteria from 500 ml of culture were harvested and resuspended in 10 ml of NTEN [(20 mM Tris-HCl, pH 8, 100 mM NaCl, 2 mM EDTA, 0.5% Nonidet P-40, 10 g of leupeptin/ml, 10 g of pepstatin/ml, 10 g of aprotinin/ml, 1 mM phenylmethylsulfonyl fluoride). The lysates were sonicated, and after centrifugation, the supernatants were mixed with glutathione-Sepharose 4B beads (500 l, Amersham Pharmacia Biotech) at 4°C for 30 min in NTEN buffer. After two washes with TG buffer (20 mM Tris-HCl, pH 8, 10% glycerol, 1 mM dithiothreitol), the GST-COUP-TFII was eluted with the TG buffer containing 30 mM glutathione. USF2a protein was expressed in E. coli tagged with histidine residues at its N terminus, allowing purification over Ni 2ϩ affinity resin as described (18).
Extract Preparation and EMSA-Nuclear extracts for EMSA were prepared at a concentration of about 10 mg/ml. Different methods of nuclear extract preparation were tested. Results present nuclear extract preparation according to the procedure described in Hasegawa et al. (25) in the presence of phosphatase inhibitors, as described in Ref. 33. EMSA were performed at room temperature with 1 ng of doublestranded oligonucleotides as probes, end-labeled and blunt-ended using Klenow enzyme with the appropriate radionucleotide or labeled in the presence of [␥-32 P]ATP and polynucleotide kinase. Incubation reactions were done on ice in the presence of 10 mM Tris-HCl, pH 7.5, 1 mM EDTA, 1 mM dithiothreitol, 5% (v/v) glycerol, 1 g of poly(dI-dC) (100 ng for recombinant proteins), and either 5 g of liver nuclear extracts or 1-4 l of reticulocyte lysate extracts and recombinant proteins. In Fig.  5B, the reaction buffer used was 20 mM Hepes/KOH, pH 7.9, 50 mM KCl, 0.2 mM EDTA, 5 mM dithiothreitol, 0.5 mM phenylmethylsulfonyl fluoride, and 10% (v/v) glycerol as reported by Hasegawa et al. (25). For competition assays, 10 -50 ng of different oligonucleotides were used as specific competitors, and 50 ng of MEF3 oligonucleotide was used as an unrelated competitor. For antibody supershift analysis, 0.5-1 l of the antiserum were included in the binding reactions.
Each reaction mixture was then electrophoresed in a native 6% (w/v) polyacrylamide gel in 0.25 ϫ Tris-borate-EDTA (or 5% in the case of the in vitro translated product and recombinant proteins experiments or 4.5% in Fig. 5B). Gels were prerun for 30 min and run for 3-4 h at 140 V.
Anti-COUP-TFII antibody, used to characterize the retarded complexes in supershift or Western blotting experiments, was provided by Dr. S. Karathanasis. RXR␣ and TR␣ antibodies were purchased from Santa Cruz. Anti-USFs (USF1 and LZ2) antibodies were produced in our laboratory (18).
DNase I Footprinting-A 100-bp XbaI-NheI rat L-PK 5Ј-flanking region (Ϫ196 to Ϫ96 containing the GlRE) was used for DNase I footprinting analysis. The labeled fragment was obtained by digestion of the Ϫ183-PK/CAT plasmid by NheI (coding strand) or by XbaI (noncoding strand), followed by filling in with Klenow fragment of DNA polymerase in the presence of [ 32 P[dCTP and [ 32 P]dTTP and then digested with XbaI or NheI. The DNase I footprinting was performed with a footprinting kit (Amersham Pharmacia Biotech). Before digestion with 0.1 unit of DNase I, the labeled fragments were incubated with various amounts of COUP-TFII recombinant protein (10, 25, and 50 ng). The DNA fragments were analyzed on a 7.5% (w/v) denaturing polyacrylamide sequencing gel. The chemical reaction for GϩA on the same fragments were included as markers.
Western Blot Analysis-The eluted GST-COUP-TFII recombinant protein (100 ng) or liver nuclear or whole COS-7 cell extracts (20 g) were separated by 10 -12% (w/v) SDS-polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes in a Bio-Rad apparatus according to the manufacturer's instructions. After blocking of nonspecific protein-binding sites for 1 h at room temperature in TBST (50 mM Tris-HCl, pH 8, 150 mM NaCl, 0.05% (v/v) Tween 20) containing 5% (w/v) nonfat dry milk, the blots were incubated with the primary hCOUP-TF antibody at a 1:500 dilution (a generous gift from Dr. S. Karathanasis) or with primary affinity-purified hUSF2 (LZ2) at a 1:500 dilution and (18) or with GST antibody at a 1:1000 dilution for 3 h at room temperature. Following three washes of 20 min each in TBST, the secondary antibody, peroxidase-conjugated swine anti-rabbit (Dako), was added at a 1:4000 dilution and left for 45 min at room temperature. The blots were washed again, and peroxidase activity was detected by autoradiography with ECL enhanced chemiluminescence system (Amersham Pharmacia Biotech). To ensure comparable loading of the samples, blots performed with nuclear and whole cell extracts were incubated with anti-annexin V antibody at a 1:1000 dilution.
Cell Culture Conditions, Transfections, and CAT and Luciferase Assays-COS-7 were plated at 70% of confluence on 6-cm-diameter dish and cultured at 37°C in Dulbecco's modified Eagle's medium supplemented with 5% (v/v) fetal calf serum and antibiotics. COS-7 cells were transfected 15 h after plating by the lipofection method, using the N-[1-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium methylsulfate transfection reagent (Roche Molecular Biochemicals). Cells were transfected with a total amount of 6 g of plasmid DNA divided into 2 g of reporter constructs, various amounts of expression vector, 0.2 g of the pRSV-luciferase plasmid as an internal transfection control (34), and pKSBluescript as a carrier to keep constant total amount of DNA. Cells were harvested 36 h after transfection, and cellular lysates were prepared for CAT assays as described previously (31). The luciferase activity was assayed as described by De Wet et al. (34) using a Lumat LB 9507 luminometer (EGG-Berthold). Results are calculated from the ratio of CAT/luciferase activity expressed in arbitrary units.
Adult male Harlan Sprague-Dawley rats (160 -200 g) were fed with regular ad libitum chow when used in the experiments. Hepatocytes were isolated as described (31). One and a half million freshly isolated cells were plated in 6-cm dishes in 3 ml of 199 medium containing 5 mM glucose supplemented with 1 M triiodothyronine, 1 M dexamethasone, 20 nM insulin, and 10% (v/v) fetal calf serum, replaced 2 h later by 199 medium plus hormones and 10% fetal calf serum. 6 h after plating, transfection was performed by the lipofection method (N-[1-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium methylsulfate) in 199 medium supplemented with 1 M triiodothyronine, 1 M dexamethasone, and 20 nM insulin. Cells were transfected with a total amount of 5 g of plasmid DNA, e.g. 4 g of reporter constructs, various amounts of expression vector, and pKSBluescript as a carrier to keep total amount of DNA constant. The concentration of expression vectors used was determined in preliminary experiments as the maximal concentration before occurrence of nonspecific squelching phenomena. The medium containing the liposome-DNA complex was removed 10 h later and replaced by glucose-free 199 medium supplemented with 1 M triiodothyronine, 1 M dexamethasone, 20 nM insulin, and either 5 mM glucose or 25 mM glucose. Luciferase extracts were prepared 30 h after transfection. Luciferase activities assays were performed using the dual luciferase reporter assay system (Promega). The total amount of protein, was determined in each cell extracts by a Bradford assay. Because no or very low variation was measured, results were calculated from the ratio of luciferase/Renilla activities expressed in arbitrary units. Results were corrected for by the Ϫ54 (see Fig. 7) or Ϫ96 L-PK activities (see Fig. 6B). Each experiment was made in triplicate and repeated three to five times.

Molecular Cloning of COUP-TFII by One-hybrid Selection in
Yeast-To identify transcriptional factors other than USF that could recognize the L-PK GlRE (also named L4 or LL boxes), we used the one-hybrid method, which allows, in yeast cells, the cloning of eukaryotic DNA-binding factors. First, we observed by mobility shift assays that wild-type yeast cell extracts do contain proteins capable of binding to the L-PK GlRE with a high affinity (data not shown). It was previously established that, in S. cerevisiae, the major CACGTG binding activity is accounted for by the multifunctional factor cbf1p (35). To increase the sensitivity of our one-hybrid screen, we decided to perform the screen in cells that do not express CBF1, thus avoiding competition between the endogenous GlRE binding activity and the factors encoded by the transfected cDNA libraries. The CBF1 gene was disrupted by using the hisg-URA3hisg marker construct (36). In a second step, the URA3 marker was removed from the cbf1 locus because of recombination between the two hisg direct repeats by selecting for uracil auxotroph cells on 5-fluoroorotate-containing medium (37), yielding the CD156 strain (cbf1::hisg). As expected, the cbf1⌬ cells were methionine auxotroph and exhibited only a residual GlRE DNA binding activity (data not shown). A GlRE-HIS3 reporter gene was then integrated into the genome of the CD156 strain at the URA3 locus. The GlRE-HIS3 fusion consisted of two tandem copies of the GlRE inserted 200 bp upstream of the TATA box of the CYC1 promoter, with its own UAS deleted, and placed in front of the HIS3 gene. The resulting strain (YMV1) exhibited a leaky Hisϩ phenotype that was suppressed by the addition of 10 mM AT to the medium. The YMV1 strain was transformed by an adult rat liver cDNA expression library that coded for oriented poly(A) tail protein segments fused to the Gal4p transcriptional activation domain. Transformants were selected directly for growth in the presence of 10 mM AT. From a screen of about 2 ϫ 10 6 transfor-mants, 30 colonies capable of growing in the presence of 10 mM AT appeared over a course of 8 days. Plasmid DNA was recovered from these colonies and used to retransform the YMV1 and the YMEF3 control strains (the latter comprising the unrelated integrated MEF3-HIS3 gene fusion). Only one plasmid gave rise to AT-resistant clones when transformed in the YMV1 strain but not in YMEF3 control strain. Sequencing of this plasmid revealed that it encodes an USF2 cDNA deleted from exons 1 and 2 and fused in frame to the Gal4p activation domain. This result thus confirmed that in yeast cells, USF is indeed capable of binding to the L-PK GlRE and provided us with a positive control of our screening.
To account for the fact that this first screen did not reveal a new GlRE-binding factor, we next surmised that if such proteins exist, they might be more abundant in tissues other than liver. Therefore, we performed a new one-hybrid screen in the YMV1 strain with another library that expresses short fragments of mouse embryo cDNA fused to the VP16 activation domain (29). In this case, the transformants were selected directly for growth in the presence of 30 mM AT. From a screen of about 2 ϫ 10 6 transformants, 60 colonies capable of growing in the presence of 30 mM AT appeared over a course of 10 days, and plasmids were recovered. After retransformation of both the YMV1 and YMEF3 strains, only one plasmid was shown to encode a fusion protein capable of specifically activating transcription of the GlRE-HIS3 reporter gene. Sequencing revealed that this plasmid expressed the VP16 activation domain fused to an 80-amino acid open reading frame comprising the COUP-TFII DNA-binding domain flanked by additional residues.
COUP-TFII Protein Binds to the L-PK GlRE-The DNA binding capacity of the entire COUP-TFII protein was assessed by EMSA with in vitro translated COUP-TFII. Doublestranded oligonucleotide containing wild-type L-PK GlRE was retarded by in vitro translated COUP-TFII (Fig. 1A, lane 1). A polyclonal antiserum raised against COUP-TFII specifically recognized the protein-DNA complex (Fig. 1A, lane 2). We also demonstrated that in vitro translated COUP-TFII bound to a DR1 probe was competed for by the L-PK GlRE oligonucleotide (data not shown). In addition, as previously demonstrated USF2a was retarded and displaced by anti-USF2 (Fig. 1A,  lanes 3 and 4). The faint band migrating faster was also displaced by anti-USF2 antibody (Fig. 1A, lanes 3 and 4). This USF protein was described as mini-USF (⌬USF2) (18) and was supposed to be generated by internal translation initiation from internal methionines.
Binding of COUP-TFII to the GlRE was also evaluated by DNase I protection assays, using recombinant COUP-TFII protein and the Ϫ196 to Ϫ96 region of the L-PK gene promoter as probe. A protection pattern was observed between bases Ϫ172 and Ϫ142 encompassing the GlRE site ( Fig. 2A for coding strand and Fig. 2B for noncoding strand). A second region was strongly protected between bases Ϫ142 and Ϫ126 corresponding to the L3 box. This L3 region contains a DR1 site that has been previously described to bind not only HNF4 but also COUP-TF (10). Examination of the GlRE area suggested that it could encompass at least two COUP-TFII-binding sites. COUP-TFII strongly protected a central half-site between Ϫ162 and Ϫ157 and more weakly 5Ј and 3Ј half-sites, generating direct repeats DR1 or DR7 in the GlRE (half-sites are delineated by arrows in Fig. 2C). These two DR1 and DR7 sites have been shown by EMSA to have similar affinity for COUP-TFII (data not shown). A G to T mutation of the second position of the central half-site (mutant MM) practically abolished any affinity for COUP-TFII (Fig. 1B, lane 1), whereas this mutation created a consensus E box (CACGTG) with high affinity for USF (Fig. 1B, lane 2). These results suggested that COUP-TFII and USF binding to the GlRE could interfere one with the other.
Interference between COUP-TFII and USF in Binding to the GlRE-To understand more precisely the interaction between COUP-TFII and USF transcription factors and the GlRE, we performed band shift experiments using COUP-TFII and USF2 recombinant proteins. In these experiments, a combination of both proteins in various ratios was used. Recombinant COUP-TFII and USF2a proteins formed two major distinct complexes (Fig. 3, lanes 1 and 6). In lane 6 of Fig. 3, the major faster migrating complex was a USF2a homodimer, and the slower complex corresponded to two homodimers, each binding to one of the two palindromic E boxes separated by 5 bp. Accordingly, both can be competed for by a USF-binding site and displaced by anti-USF antibodies (data not shown). In lane 1 of Fig. 3, COUP-TFII corresponded to a major fast migrating complex (truncated, tCOUP-TFII), whereas a minor slower migrating band (full-length, fCOUP-TFII) was, on this gel, at the limit of detection. Fig. 3B shows a Western blot analysis of recombinant COUP-TFII, demonstrating that the major form was a proteolytic product. The same result has been previously reported by Malik and Karathanasis (26), namely that recombinant full-length COUP-TFII (rARP-1) preparations were very sensitive to proteolytic cleavage, leading to a truncated form retaining DNA binding activity.
Upon addition of increasing amounts of USF2a, an increase in USF2a-DNA complexes and a corresponding decrease in the COUP-TFII retarded complex was observed (Fig. 3, lanes 2-4). Conversely, upon addition of increasing amounts of COUP-TFII, an increase in COUP-TFII-DNA complex and a corresponding decrease in the USF2a retarded complexes was observed (Fig. 3, lanes 7-9). These results suggest a competition between USF and COUP-TFII for binding to the GlRE. However, we could observe a minor complex detected only when both COUP-TFII and USF are present. This complex was clearly detectable at a ratio of 1 COUP-TFII/2 USF2 (Fig. 3,  lanes 4 and 7), displaced by DR1 (Fig. 3, lane 5) and MLP sites (data not shown). This result suggests that a ternary complex consisting of the GlRE probe, one USF dimer, and one COUP-TF dimer can be detected in vitro in the presence of concentrated recombinant factors. Fig. 2C shows that, indeed, binding of a USF dimer on the downstream E box and a COUP-TF dimer on the DR1 overlapping the upstream E box is in principle conceivable.
Liver Nuclear Extracts Contain Different Factors Able to Bind the L-PK GlRE-We performed band shift experiments using the GlRE oligonucleotide and liver nuclear extracts from rats fed a regular diet ad libitum. EMSA were performed with a novel binding buffer as compared with the one previously used in our laboratory (18) and the one described by Hasegawa et al. (25). This buffer is characterized, in particular, by increased EDTA concentration (to 1 mM) and absence of KCl. In this new binding conditions, up to seven DNA-protein complexes were detected with the GlRE probe (Fig. 4, lane 1). Identification and specificity of these complexes were analyzed by competition with unlabeled oligonucleotides and depletion or supershift by specific antibodies. All of these complexes were specific because they were displaced by 10 ng of unlabeled GlRE site (Fig. 4, lane 3). The complex 2 corresponded to binding of the USF1/USF2 dimers because depletion by anti-USF antibodies (Fig. 4, lane 2) and competition with an unlabeled MLP site (data not shown and Ref. 18) suppressed formation of this complex. Complexes 1 were likely to belong to the Sp1 family of transcriptional factors as demonstrated by competition with a GC box-containing oligonucleotide (lane 4). Finally, the complex 3 binding activities remaining after USF depletion and Sp1 DNA binding activity displacement were competed for by an excess of a DR1 motif (Fig. 4, lanes 12 and  13) and of the L3 element of the L-PK gene promoter, which also contains a DR1 site (Fig. 4, lanes 14 and 15) (10); in contrast the complex 3 binding activities were not displaced by an unrelated probe (Fig. 4, lane 11). These results suggest that the complexes 3 could contain members of the nuclear receptor superfamily. To determine whether the complex 3 binding activities actually corresponded to COUP-TFII, we used an antibody toward COUP-TFII. The addition of this antibody to liver nuclear extracts, depleted of USF DNA binding activity, supershifted the complex 3 binding activities (Fig. 4, lane 5), whereas the addition of anti-RXR␣ or anti-TR␣ antibodies did not affect them (Fig. 4, lanes 6 and 7). After depletion of USF and Sp1 DNA binding activities, we observed that the addition of COUP-TF supershifted both complexes 3 (Fig. 4, lane 8), whereas the addition of anti-RXR␣ or anti-TR␣ antibodies did not affect them (Fig. 4, lanes 9 and 10). All other antibodies specific to various members of the nuclear receptor superfamily tested so far were without any effect on the complex 3-DNA binding activities (data not shown). These results confirm that the two COUP-TFII containing complexes binding to the GlRE can be easily detected in liver nuclear extracts, provided that binding conditions are optimized. Fig. 5A, we compared the GlRE binding activities in liver nuclear extracts of rats fed either L-PK gene inducing diet (i.e. high carbohydrate, lanes 1 and 2) or L-PK gene inhibiting diet (i.e. high protein in lane 3 and high fat in lane 4). The complexes described in Fig. 4 were still observed, in particular complex 2, i.e. USF, supershifted by anti-USF antibodies (Fig. 5A, lane 5), and complexes 3 supershifted by anti-COUP-TFII antibodies (Fig. 5A, lane 6) but not by anti-RXR antibodies (Fig. 5A, lane 7) and competed for by a DR1 motif (Fig. 5A, lane 8). With respect to the NF-Y (rat albumin CAAT box) (Fig. 5C) and nuclear factor 1 (data not shown) complexes characterized in parallel to control quality and quantity of the nuclear extracts, none of the GlRE binding activities seemed to be detectably affected by diets. We also found that the relative binding activity of the different GlREbinding complexes depends on the binding conditions; Fig. 5B shows the patterns obtained with extracts from carbohydraterefed and protein-rich diet-refed rats using the binding buffer described by Hasegawa et al. (25). Here, as in our previous publication (18), USF is the major binding activity, whereas COUP-TF-containing complexes and GC-box binding activities are scarcely detectable (compare lanes 1 and 3 of Fig. 5A and  lanes 1 and 2 of Fig. 5B).

In Vitro Binding of COUP-TFII-containing Complexes Is Not Significantly Modulated by Diets-In
In Fig. 5D, we evaluated the presence of COUP-TFII and USF2 in carbohydrate and protein-refed liver nuclear extracts by Western blot analysis with specific antibodies. The amounts of COUP-TFII and USF2 proteins were similar under both dietary conditions used.
COUP-TFII Represses USF2-dependent Transcription of the L-PK Promoter in COS-7 Cells-COUP-TF has been demonstrated to be a repressor of transcription of many genes (27). Therefore, we explored the possibility that COUP-TFII homodimers might counteract the activity of USFs on a promoter depending on the GlRE element. To test this hypothesis, COS-7 cells were transiently transfected with L-PK reporter plasmids and both USF2a and COUP-TFII expression vectors. The (LL) 2 -54-PK/CAT reporter plasmid was directed by two copies of the wild-type GlRE, while the (MM) 2 -54-PK/CAT construct was directed by two copies of the MM mutant GlRE. This latter mutant was characterized by very low affinity for COUP-TFII and high affinity for USF (Fig. 2C). Fig. 6A showed that expression of USF2a increased dramatically the CAT activity of the LL reporter construct (lane 3), as previously reported (19). As expected USF2a also increased activity of the MM mutant reporter gene (Fig. 6A, lane 10) because canonical E boxes were created. COUP-TFII overexpression in transfected COS-7 cells inhibited this USF2a-mediated activation of the reporter genes in a dose-dependent fashion, but this inhibition was more important on the LL (Fig. 6A, lanes 4 -7) than on the MM construct (lanes 11-14). Western blot controls allowed us to assure that the COUP-TFII inhibitory effect was not due to inhibition of the USF2a expression vector; USF2a abundance was constant in cells transfected with a same amount of USF2a expression vector and increasing amounts of COUP-TFII expression vector (Fig. 6A, lower panels). These results demonstrate that COUP-TFII is capable of antagonizing transactivation by USF2a of a reporter gene directed by the wild-type GlRE but that this antagonism effect is strongly reduced by a GlRE mutation decreasing affinity for COUP-TFII.
COUP-TFII Represses GlRE-mediated Glucose Induction in Hepatocytes-To determine whether COUP-TFII could play a role in mediating the glucose responsiveness, we used the nat-ural Ϫ183 L-PK promoter and a reporter gene carrying three copies of the GlRE placed upstream of the Ϫ54 L-PK minimal promoter, (LL) 3 -54-PK/Luc in hepatocytes in primary culture. To increase sensitivity of the assay after transient transfection in hepatocytes, we constructed different pGL3 luciferase chimeric L-PK constructs. As can be seen in Fig. 7, both the natural Ϫ183 L-PK promoter and the artificial promoter with oligomerized GlRE (LL)-binding sites were specifically stimulated by glucose, 8-and 50-fold, respectively, whereas a construct directed by oligomerized box L3, (L3) 3 -54-PK/Luc was insensitive to glucose. Cotransfection of hepatocytes with both glucose-responsive L-PK/Luc constructs and COUP-TFII expression vector repressed the response to glucose, strongly for the (LL) 3 -54-PK construct and totally for the Ϫ183-PK construct. Because the inhibitory effect of COUP-TFII overexpression on the glucose-dependent promoter activation was observed with the (LL) 3 -54 plasmid devoid of the L3 element, which was found to bind HNF4 and COUP-TF factors (10), as well as with the Ϫ183-PK construct, this effect could not be ascribed to competition of COUP-TFII for binding of HNF4 to element L3 but rather to interaction with the GlRE. Accordingly, COUP-TFII did not repress activity of the (L3) 3 -54 construct in hepatocytes. To confirm these data, we also studied the effect of HNF4 overexpression on the inhibition by COUP-TFII of the glucose-dependent Ϫ183-PK/Luc activation; although HNF4 overproduction increased the glucose responsiveness, as already described (38), it did not block the inhibitory action of COUP-TFII (data not shown). Therefore, these experiments demonstrate that COUP-TFII is an inhibitor of glucose responsiveness that acts through the GlRE. This conclusion was supported by the decreased glucose responsiveness of a L-PK/Luc construct directed by three copies of the MM GlRE mutant with decreased affinity for COUP-TF and increased affinity for USF (Fig. 6B). Although a (LL) 3 -96-PK/Luc construct (directed by three copies of the wild-type GlRE upstream of the L-PK HNF1-binding site) (16) was activated 26 Ϯ 6-fold by glucose in hepatocytes, the MM construct was only activated 3.2 Ϯ 1.2-fold. This decreased glucose re- sponsiveness was due to a strong residual MM promoter activity under low glucose conditions.

DISCUSSION
The most intensively investigated glucose/carbohydrate response elements are the GlRE of the L-PK gene promoter (8 -10, 12, 39) and the ChoRE located 1448 bp upstream of the S14 gene (12,24). Both are able to bind USF transactivators, but this property alone does not easily explain the glucose responsiveness because USF-binding sites exist in regulatory regions of numerous genes insensitive to nutrient regulation. Moreover, Kaytor et al. (24) reported that a mutant ChoRE that had lost its ability to bind USFs still retained its glucose responsiveness. However, in our hands, this mutant conserved a clearly detectable affinity for USFs. 2 In any case, affinity for USFs is clearly not parallel to the efficacy as a glucose response element. Although our results in cell culture (19) and in vivo in knock-out mice indicate that endogenous USFs are important for a kinetically normal activation of various dietary-dependent genes by glucose (20 -22), they cannot explain by themselves the transcriptional regulation of glucose-responsive genes by glucose. Because both S14 and L-PK gene ChoRE/GlRE have the same structure, characterized by two more or less degen-erated E boxes separated by a 5-bp spacer, it could be hypothesized that this unique arrangement was needed for binding of both USF dimers and other components of a multimolecular glucose sensor component. Therefore, we undertook a search for additional GlRE-binding proteins that could act as components of the glucose sensor system.

COUP-TFII Binds to the GlRE, in Vivo and in Vitro-
Our one-hybrid screen in yeast allowed us to detect two GlREbinding proteins: USF2, starting from an adult rat liver cDNA library, and COUP-TFII from a mouse embryo cDNA library. The liver-derived library was constructed using oligo(dT) as the primer for first strand cDNA synthesis, which probably explains why COUPTF-TII was not found here. Indeed, the DNAbinding domain for this factor is very upstream from the 2 M. Vasseur-Cognet, unpublished results. poly(A) tail in the mRNA and was therefore highly underrepresented in the liver library. In contrast, the embryo library was derived from cDNAs primed with random hexamers, such that cDNAs for 5Ј parts of the messengers had more chance to be present.
COUP-TFII protein synthesized in a cell-free transcriptiontranslation system binds efficiently to the GlRE and also to the S14 ChoRE (data not shown). EMSA of the GlRE incubated with liver nuclear extracts revealed complexes displaced by a DR1 oligonucleotide in excess and supershifted by anti-COUP-TF antibody. Finally, footprinting experiments localize two COUP-TF DNA-binding sites within the GlRE that overlap the E boxes: a DR1 in 5Ј (that does not bind HNF4; data not shown) and a DR7 in 3Ј. Both contained a common central half-site (Fig. 2C).
In addition to USF and COUP-TF-containing complexes, the GlRE also gave Sp1-type complexes displaced by a Sp1-specific oligonucleotide competitor. Because binding of Sp1 has been proposed to be involved in the response of the ACC gene to glucose in cultured adipocyte cell line (40), we asked whether Sp1 binding might be a component of the GlRE/ChoRE. In fact, the S14 ChoRE has no affinity for Sp1, and we never cloned Sp1-cDNAs in our one-hybrid screen, which could indicate that in vivo, in yeast, the GlRE does not interact with members of the Sp1 family. In addition, we previously showed that a GlRE construct devoid of the Sp1 site conserved its property to mediate the glucose responsiveness (19). Finally, the putative GC-rich Sp1-binding site in the L-PK GlRE is not conserved between rat, human, and mouse (GenBank TM accession numbers are X05684 for rat and Z18922 for Homo sapiens; for mouse, the sequence is unpublished data).
COUP-TF II Counteracts Transactivation by USF-COUP-TF proteins are orphan nuclear receptors that are most generally considered to be transcription inhibitors (27). Accordingly, COUP-TFI has been shown to interact with corepressors such as N-CoR and SMRT (41). COUP-TF can also inhibit gene transcription by heterodimerization with RXR, the common partner of several nuclear receptors, and by competing with them for binding to common sites (27). However, several reports also show that COUP-TFs act as gene activators, either by directly binding to DNA elements and cooperating with contiguous factors (27,42,43) or by interacting with other transcription factors (e.g. HNF4 or Sp1) through protein-protein contacts (44,45). COUP-TFII is widely expressed, in particular in hepatocytes (46); its defect in knock-out mice results in embryonic lethality because of inappropriate angiogenesis and heart development (47). COUP-TFI is abundant in the central nervous system, and its defect is associated with abnormal development of glossopharyngeal ganglion and defective axonal projection and arborization, resulting in perinatal death (48).
When COS cells are co-transfected with COUP-TFII and USF2a expression vectors, USF-dependent transactivation through the GlRE was inhibited. Footprinting experiments demonstrate that COUP-TF-binding sites overlap E boxes and therefore are expected to compete with basic helix-loop-helix/ leucine zipper family proteins for binding to the GlRE. Accordingly, there is competition between USF2 and COUP-TFII for binding to this element in EMSA experiments. However, at high concentration of recombinant COUP-TFII and USF2 proteins, a ternary complex consisting of the GlRE binding both factors could be documented; it is likely that in this complex, USF binds the downstream E box while COUP-TFII binds the upstream one overlapping with the DR1 site. It is noteworthy that we were unable to detect any direct protein-protein interaction between USF2a and COUP-TFII in a two-hybrid test in yeast (data not shown).
COUP-TFII Inhibits the Glucose Responsiveness of the L-PK Promoter-The Ϫ183 L-PK promoter confers a good glucose responsiveness on a reporter gene in hepatocytes in primary culture. Overexpression of COUP-TFII in these cells inhibited the glucose response. This inhibition was found again on an artificial promoter in which three GlRE motifs were oligomerized upstream of the minimal L-PK promoter. This indicates that COUP-TFII acts through the GlRE and not through the contiguous L3 HNF4-binding site. In vitro, L3 is a relatively strong COUP-TF-binding site, but we hypothesize that the high HNF4 content of hepatocytes precludes any significant interaction of this type in hepatocytes. ARP-1/COUP-TFII was reported to bind to the C3P element of the apolipoprotein CIII gene promoter (46) with a K d value ϭ 13.4 nM. We have confirmed this result with recombinant COUP-TFII and found that COUP-TFII had about 3-fold less affinity for the L3 site and 4-fold less for the GlRE site than for the C3P site (data not shown). Accordingly, the L3-dependent reporter activity was not significantly affected by COUP-TFII overexpression, and HNF4 overexpression, although improving the glucose response in the absence of COUP-TF expression vector, did not relieve COUP-TFII-dependent inhibition of this response as would have been expected if COUP-TFII and HNF4 competed for binding to L3.
We have identified a mutant GlRE that lost its ability to bind COUP-TFII but conserved and even increased its affinity for USF. This mutant was called MM to refer to Diaz Guerra et al. (10). In the context of the natural L-PK promoter and thus in association with the element L3, this MM mutant GlRE was able to maintain glucose responsiveness (10). However, we show here that when oligomerized in front of the minimal L-PK promoter, the MM element conferred a dramatically decreased glucose responsiveness in hepatocytes. This was explained by increased activity of the MM-dependent reporter gene under low glucose condition. This result was precisely the expected one if we consider that COUP-TFII binding under low glucose conditions could contribute to counteract the transactivating effect of a transactivator responsible for glucose-induced transcriptional stimulation.
Is COUP-TF II a Component of the Glucose Sensor?-COUP-TFII binds to the L-PK GlRE in vivo in yeast as well as in vitro. In vitro, it also binds to the S14 ChoRE (data not shown). In EMSA, liver nuclear extracts gave with the GlRE two fast migrating complexes displaced by a DR1 oligonucleotide competitor and supershifted by anti-COUP-TFII antibodies. The faster one exhibited approximately the same migration as COUP-TFII homodimers synthesized in COS cells transfected with a COUP-TFII expression vector (data not shown). The slower complex might correspond either to a heterodimer with an undetermined partner or to a post-translational modification of COUP-TF. In fact, O'Malley and co-workers (49) discussed the fact that orphan steroid receptors may be activated independently of any ligand fixation through external cues acting on membrane receptors and intracellular signaling pathway, resulting in receptor phosphorylation. Accordingly, COUP-TFII activity could be modulated by post-translational modifications induced by nutrient or hormonal regulations, e.g. through PKA or 5Ј AMP-activated protein kinase (11).
In transient transfection experiments, overexpression of COUP-TFII counteracts the transactivating effect of USF and inhibits the glucose response of the L-PK promoter in hepatocytes in primary culture; in addition, intact COUP-TFII DNAbinding site is required for correct glucose response. Therefore, COUP-TFII is obviously a very good candidate to participate in the transcriptional regulation by glucose.
Recently, different groups reported new data concerning GlRE DNA-binding proteins. Hasegawa et al. (25) reported that carbohydrate refeeding resulted in increase of a GlRE binding activity in the liver and cultured hepatocytes and excluded USF, Sp1, and c-Myc factors as being involved in this activity. These assumptions were based exclusively on the use of commercial antibodies. We reproduced the results of these authors and found that under their described nuclear extract preparations (precipitation by 25% polyethylene glycol, eventually after discarding of the 10% polyethylene glycol precipitate and use of the nuclear extract precipitates without dialysis) and EMSA conditions, Sp1 family members, USF, and COUP-TFcontaining complexes were the only GlRE-binding proteins detected (Fig. 5B). In the proteins precipitating between 10 and 25% polyethylene glycol, we found that the major GlRE binding activity was USF, displaced by anti-USF1, anti-USF2 antibodies, and an excess of the MLP USF-binding site. We are currently determining with a large number of nuclear extract samples, conditions which may modulate these binding activities.
Yamada et al. (50) described two novel purified GlRE-binding proteins that are different from USF. The DNA binding activity of these 24-and 26-kDa proteins disappeared after preincubation for 5 min at 60°C. We found that COUP-TFII either translated in vitro or from liver nuclear extract is labile after a preincubation at 60°C for 5 min, whereas USFs were stable. Yamada et al. (50) observed that the binding profile of their purified proteins was apparently migrating faster than that from the starting materials and indicated that purification of this GlRE binding activity was very difficult because it corresponded to an extremely unstable protein. All these results suggest that this novel activity is proteolytically unstable as described for COUP-TF (26). Therefore, we propose that this novel GlRE binding activity could be COUP-TFII.
Finally, Ferré and co-workers (51) recently proposed that the basic helix-loop-helix/leucine zipper family SREBP-1c/ADD1 factor could be responsible for the transcriptional activation of lipogenic genes, including the L-PK gene, by glucose. However, the SREBP gene is activated by insulin, regardless of the presence of glucose (51) and binds very poorly to the L-PK GlRE. 2 In fact, we have shown that SREBP-1c was a very inefficient transactivator of a reporter gene directed by oligomerized GlREs in hepatoma mhAT3F cells (38).
The regulation of the transcriptional activity by glucose through the GlRE could be due to competition between glucoseinducible activators and the inhibitor COUP-TFII for binding to the response element. This hypothesis is in line with the observation that under low glucose conditions the GlRE behaves rather as a negative cis-acting element, probably because it binds mainly COUP-TFII, which could be displaced by activators upon glucose refeeding. Such an alternative binding of COUP-TFII or basic helix-loop-helix activators on the GlRE E boxes could still explain why this element seems to be always occupied in in vivo footprinting experiments in fasted as well as refed animals. 3 In addition, the increased activity at low glucose and the decreased activation by glucose conferred by the oligomerized GlRE mutant consisting of two canonical E boxes with high affinity for USF and very low affinity for COUP-TFII are also consistent with our hypothesis.
The putative role of COUP-TFII in the glucose response complex as a binding competitive inhibitor of glucose-dependent activators is reminiscent of its reported capability to antagonize different members of the nuclear receptor family, either ligand-dependent (estrogen receptor) (52,53), peroxisomeproliferator-activated receptor (54,55) or still orphan but sensitive to cAMP (steroidogenic factor) (56). In conclusion, although we are aware that confirmation or invalidation of these hypotheses still require extensive studies, we have demonstrated in this paper that COUP-TFII does bind to the L-PK GlRE and inhibits the glucose response and is therefore likely to be an important player of the glucose sensor system of glucose sensitive genes.