Regions flanking exon 1 regulate constitutive expression of the human antithrombin gene.

We have identified cis-acting elements and trans-acting factors that regulate constitutive expression of the human antithrombin gene. The activity of the sequences flanking the first exon of the gene was investigated using a luciferase-based reporter assay in transiently transfected HepG2, COS1, BSC40, and HeLa cells. Deletion analysis allowed the mapping of two elements able to promote antithrombin gene transcription in HepG2 and COS1 cells. The first element is located upstream of the first exon (−150/+68 nucleotides). The second element is in the first intervening sequence (+300/+700 nucleotides) and functions in an orientation opposite to that of the first. Footprint analysis showed three protected areas in the 5′ upstream element at −92/−68 (element A), −14/+37 (element B), and −126/−100 nucleotides (element C). These elements acted as enhancers in luciferase reporter assays. Gel retardation analysis demonstrated that two liver-enriched transcription factors, hepatocyte nuclear factor 4 (HNF4) and CCAAT enhancer-binding protein (C/EBPa), bound to the 5′ upstream element. HNF4 bound to elements A and C, whereas C/EBPa bound to element B. Element A also interacted with the ubiquitous nuclear hormone receptors chicken ovalbumin upstream promoter transcription factor 1 (COUP-TF1), thyroid hormone receptor α (TRα), peroxisome proliferator-activated receptor α(PPARα), and retinoid X receptor α (RXRα). In HepG2 and BSC40 cells, HNF4, C/EBPα, and RXRα activated luciferase expression from a reporter construct containing the 5′-upstream minimal antithrombin gene promoter, while COUP-TF1, TRα, and HNF3 (α or β) repressed such expression. Our results show that constitutive expression of the human antithrombin gene depends in part upon the interplay of these transcription factors and suggest that signaling pathways regulated by these factors can modulate antithrombin gene transcription.

We have identified cis-acting elements and trans-acting factors that regulate constitutive expression of the human antithrombin gene. The activity of the sequences flanking the first exon of the gene was investigated using a luciferase-based reporter assay in transiently transfected HepG2, COS1, BSC40, and HeLa cells. Deletion analysis allowed the mapping of two elements able to promote antithrombin gene transcription in HepG2 and COS1 cells. The first element is located upstream of the first exon (؊150/؉68 nucleotides). The second element is in the first intervening sequence (؉300/؉700 nucleotides) and functions in an orientation opposite to that of the first. Footprint analysis showed three protected areas in the 5 upstream element at ؊92/؊68 (element A), ؊14/؉37 (element B), and ؊126/؊100 nucleotides (element C). These elements acted as enhancers in luciferase reporter assays. Gel retardation analysis demonstrated that two liver-enriched transcription factors, hepatocyte nuclear factor 4 (HNF4) and CCAAT enhancer-binding protein (C/EBPa), bound to the 5 upstream element. HNF4 bound to elements A and C, whereas C/EBPa bound to element B. Element A also interacted with the ubiquitous nuclear hormone receptors chicken ovalbumin upstream promoter transcription factor 1 (COUP-TF1), thyroid hormone receptor ␣ (TR␣), peroxisome proliferator-activated receptor ␣(PPAR␣), and retinoid X receptor ␣ (RXR␣). In HepG2 and BSC40 cells, HNF4, C/EBP␣, and RXR␣ activated luciferase expression from a reporter construct containing the 5-upstream minimal antithrombin gene promoter, while COUP-TF1, TR␣, and HNF3 (␣ or ␤) repressed such expression. Our results show that constitutive expression of the human antithrombin gene depends in part upon the interplay of these transcription factors and suggest that signaling pathways regulated by these factors can modulate antithrombin gene transcription.
Human antithrombin (AT) 1 is a major inhibitor of a number of serine esterases implicated in blood coagulation (1,2). The AT gene contains 7 exons and 6 intervening sequences (IVS) in a 14-kilobase pair region of the long arm of chromosome 1. The mechanisms underlying AT gene expression are not well known. The AT gene is expressed primarily in the liver, with lower levels in the kidney and brain (2). AT gene expression is believed to be constitutive, but developmental and hormonal factors are known to influence AT levels in plasma (2). At the transcriptional level, an early report by Prochownik described enhancer activity of a Ϫ340/ϩ1200 nucleotide (nt) fragment of the AT gene (3), while Ochoa et al. (4), investigating the regulation of expression of the human transferrin gene, identified a sequence at Ϫ480/Ϫ458 nt that contained the core motif TCT-TTGACCT. This element, named TF-DRI, was shown to be homologous with an element at Ϫ89/Ϫ75 nt in the human AT gene, and generated shifts in liver nuclear extracts in a tissuespecific manner (4). These observations prompted us to characterize in greater detail elements in this area of the human AT gene which are involved in its transcriptional regulation.

MATERIALS AND METHODS
Plasmids and AT Gene Deletion Constructs-To detect AT gene promoter activity, a 6900-base pair HindIII fragment of a normal genomic human AT clone, starting 4800 nt upstream of the first exon and ending 2100 nt downstream of the first exon within the first IVS ( Fig. 1), was subcloned into pUC19 (Stratagene, La Jolla, CA). This subclone and deletions derived therefrom were inserted into pSVOA-L⌬5Ј, a luciferase reporter plasmid lacking eukaryotic regulatory sequences (5). Constructs of AT gene deletions downstream of Ϫ1100 nt were made following subcloning of the Ϫ1100/ϩ68 nt BamHI/EaeI and the Ϫ150/ϩ68 nt DraI/EaeI fragments respectively into pGEM-3Zf(ϩ) (Promega, Madison, WI). The restriction sites employed in the reporter assays are detailed in Figs. 1 and 2. Additional constructs were made by amplification from the initial genomic clone using the polymerase chain reaction (see Fig. 2). The orientations of the various constructs were confirmed by restriction analysis and/or DNA sequencing. Control luciferase reporter plasmids contained the SV40 early promoter-enhancer complex (pGL Prom-Enh, Promega) or the liver-specific proximal promoter from the carbamoyl-phosphate synthetase (CPS) gene (pCPS-luc; Ref. 6). A SV40 ␤-galactosidase-derived reporter plasmid (pSV-␤GAL, Promega) was used as an internal control for transfection efficiency. The parent luciferase plasmid pSVOA-L⌬5Ј was included as a control for background luciferase expression. Promoter, enhancer, or repressor activity of AT sequences was tested in the following vectors: pGL (Promega), a SV40 derivative; and pCPS Ϫ67/ϩ3 nt-luc and pTK Ϫ81/ϩ52 nt-luc, containing the minimal promoters of the CPS and thymidine kinase genes, respectively. Single and multiple copies of elements A and B were inserted upstream of the minimal promoters of pCPS Ϫ67/ϩ3 nt-luc and pTK Ϫ81/ϩ52 nt-luc. Single copies cloned in pUC19 were digested with KpnI/SalI and inserted between the KpnI/ XhoI sites of both reporter plasmids. Multiple copies of the elements obtained from self-ligation were subcloned into pBluescript (KS) ϩ (Stratagene), digested with KpnI/SacI or HindIII/SacI, and inserted into the KpnI/SacI sites of pCPS-luc or the HindIII/SacI sites of pTKluc. Vectors for the in vitro and in vivo expression of various transcrip-tion factors have been described: human COUP-TF1 (7), a truncated version of human COUP-TF1 (t-COUP-TF1) lacking its first 51 amino acids (7), rat HNF4␣ (8), rat PPAR␣ (9), and human RXR␣ (9). Vectors for the expression of mouse C/EBP␣ and rat HNF3␣ and ␤ were a generous gift of Drs. P. Hoodless and J. Darnell, The Rockefeller University. Large scale plasmid preparations were made either by two rounds of centrifugation on CsCl or by anion exchange chromatography (Quiagen, Chatsworth, CA).
Cell Culture and Transfection-HepG2, COS1, BSC40, and HeLa cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 100 g of streptomycin/ml, and 100 g of penicillin G/ml. Cells at 50 -60% confluence were transfected with 25 g of DNA (20 g of plasmid plus 5 g of carrier DNA) by calcium phosphate precipitation (10), followed by a 10% glycerol shock 16 h post-transfection (with the exception of COS1 cells). Lipofection transfections were performed as described (11). In cotransfection experiments, all cells were incubated for 24 h before and during transfection in medium without phenol red and containing 5% charcoal-stripped fetal bovine serum (9). Cells were cotransfected with 5 g of the AT minimal promoter construct pATϪ150/ϩ68 nt-luc and 1-8 g of the expression vectors for various transcription factors. The amount of plasmid transfected was normalized to 20 g with the parental vectors pSG5 or pRC/CMV (2.5 g each) and carrier DNA. Luciferase activity was measured 48 h post-transfection and was normalized to the protein content of the lysates and to ␤-galactosidase activity.
Footprint Analysis-The Ϫ150/ϩ68 nt DraI/EaeI fragment of the AT gene was made blunt and subcloned in both orientations into the plasmid pGEM-3Zf(ϩ). The plasmid was digested with EcoRI, and the 3Ј-ends were labeled with the Klenow fragment of DNA polymerase I and [␣-32 P]dATP. The plasmid was then digested with HindIII, and the probe was purified from agarose gel. DNase I cleavage and protection by total cellular extracts of HepG2, COS1, and HeLa cells were performed in the presence of bovine serum albumin, poly(dI-dC)⅐poly(dI-dC), and carrier DNA, as described (12). The amounts of extract and DNase I used are indicated in the corresponding figure legends.
Electrophoretic Mobility Shift Assays-Electrophoretic mobility shift assays (EMSA) were performed with nuclear extracts of HepG2, COS1, BSC40, and HeLa cells (13). The oligonucleotides tested (wild-type; mutant elements A, B, and C; and oligonucleotides for competition) are listed in Table I. Additional oligonucleotides for binding of CTF/NF1, NFB, TFIID, SP1, AP1, AP2, and AP3 were obtained from Promega. Each oligonucleotide and its complementary strand (with added restriction sites) were annealed and subcloned into pUC18 (Stratagene). For element B, a Ϫ28/ϩ47 nt NlaIV/BstXI fragment of the AT promoter subcloned into the plasmid pGEM-7Zf(ϩ) (Promega) was also tested. Restriction fragments of each element were labeled at their 3Ј-ends with [␣-32 P]dATP and the Klenow fragment of DNA polymerase I. Extracts containing 10 g of protein were incubated for 15 min with each probe (2.5 ng) at 25°C in the presence of 1 g of poly(dI-dC)⅐poly(dI-dC), 2 g of bovine serum albumin, and 0.5 g of carrier DNA (unless otherwise indicated). Buffer conditions were as described (14). Electrophoresis was carried at 4°C on prerun 4% polyacrylamide gels (30:1 acrylamide/N,NЈ-methylenebisacrylamide weight ratio) with 6.7 mM Tris, 6.7 mM boric acid, 1 mM EDTA as running buffer. In competition assays, unlabeled oligonucleotides were used at 100 molar excess of the test oligonucleotide (unless otherwise indicated). In supershift assays, 1 l of antiserum or 0.5 g of an IgG fraction was used. Antibodies were added immediately after addition of labeled probe, and the reaction was incubated for another 30 min unless otherwise indicated. In vitro transcription-translation assays were carried out with a rabbit reticulocyte lysate-coupled system according to the manufacturer's instructions (Promega). Reactions were done concomitantly with and without L-[ 35 S]methionine. The efficiency of the translation was determined by SDS-polyacrylamide gel electrophoresis of the labeled products. Unlabeled translation products were used for EMSA, and the total amount of lysate was maintained at 4 l with unprogrammed lysate. The specificity of the shifts was confirmed by supershift analysis. Fig. 1 shows the results of deletional analysis of cis-elements flanking exon 1 of the AT gene in the reporter plasmid pSVOA-L⌬5Ј. The data shown were generated following calcium phosphate-mediated transfection of HepG2 and COS1 cells. The largest construct assayed was a 6900-base pair HindIII fragment of the AT gene (Fig. 1A). This fragment was unable to promote luciferase activity. Removal of exon 1 and downstream sequences by cleavage with HindIII at Ϫ4800 nt and EaeI at ϩ68 nt, 3 nt upstream of the translational start site of the AT mRNA, provided a construct with detectable luciferase activity (Fig. 1D). Further 5Ј deletions to Ϫ2500 (Fig. 1E) or Ϫ1100 nt (Fig. 1F ) did not noticeably change luciferase values. Because slightly lower luciferase expression was obtained for a Ϫ4800/ϩ68 nt construct (Fig. 1D) than for a Ϫ2500/ϩ68 nt construct (Fig. 1E), the effect of sequences between Ϫ4800 and Ϫ2500 nt was tested directly in the pGL plasmid system. A Ϫ4800/Ϫ2500 nt construct did not reduce luciferase activity driven by the SV40 minimal promoter, confirming the absence of a negative element in this region of the gene (data not shown). To facilitate further deletional analysis, the activities of all constructs were reported relative to the activity of the Ϫ1100/ϩ68 construct, which was given an arbitrary luciferase value of 100% (Fig. 1F).

Mapping of Regulatory Elements Upstream of Exon 1-
5Ј deletions of the Ϫ1100/ϩ68 construct up to Ϫ150 nt did not noticeably alter luciferase activity in either HepG2 or COS1 cells (Fig. 2). Further deletions of the Ϫ150/ϩ68 nt element were selected to encompass the area protected in DNA footprints. Deletions to Ϫ101, Ϫ70, and Ϫ28 nt progressively abolished luciferase activity in HepG2 and COS1 cells. A Ϫ28/ϩ68 nt construct was unable to promote transcription by itself (Fig.  2). Deletions from the 3Ј-end and upstream of ϩ68 nt did not show markedly altered luciferase activity up to ϩ11 nt. In contrast, decreased luciferase activity was seen at, and upstream of, Ϫ28 nt (Fig. 2).
The three elements A, B, and C mapped in footprints were unable to promote luciferase expression when subcloned into pSVOA-L⌬5Ј in their natural orientation relative to the luciferase reporter gene (Fig. 3). In contrast, two tandemly repeated copies of element A or three tandemly repeated copies of element B enhanced luciferase expression driven by heterologous minimal promoters when inserted upstream of TK-luc or CPSluc (Fig. 3). Fig. 1 support the possibility of a second regulatory element within IVS1 of the AT gene. This element was identified following subcloning of the ϩ300/ϩ2100 nt region of IVS1 in an inverse orientation upstream of the luciferase gene (Fig. 1I ). Effectively, insertion of this fragment led to an increase in luciferase activity in pSVOA-L⌬5Ј. The presence of this second regulatory element was confirmed by further deletional analysis, allowing the narrowing of the active sequences to ϩ300/ ϩ700 nt (Fig. 1J). When the initial ϩ300/ϩ2100 nt construct, or its deletions, were inserted in their natural orientation relative to the luciferase gene, no increase in luciferase expression was observed (Fig. 1, H and K-M). The strength of the element in IVS1 to promote luciferase expression was about 50% and 25% relative to the level of expression of the 5Ј upstream element in COS1 cells and HepG2 cells, respectively. The lower efficiency in HepG2 cells as compared to COS1 cells was seen with both the original IVS1 construct and with deletions derived from it.

Mapping of Regulatory Elements in IVS1-Data presented in
Footprint Analysis of the Upstream Promoter Element-Figs. 4 and 5 define the areas protected by nuclear extracts in the upper and lower strands of the Ϫ150/ϩ68 nt probe from the 5Ј upstream promoter. Three elements, designated A, B, and C (in decreasing order of protected strength) were identified. Element A was strongly protected in both DNA strands and gave a very similar pattern for the three sources of nuclear extracts (HepG2, COS1, and HeLa). Element A extended from Ϫ89 to Ϫ65 nt in the upper strand (Fig. 4) and from Ϫ92 to Ϫ68 nt in the lower strand (Fig. 5). Element B was protected by nuclear extracts prepared from the three cell lines, but stronger protection was observed with extracts from COS1 cells and HepG2 cells than from HeLa cells. This was particularly evident in the upper strand when increasing amounts of DNase I, or of nuclear extract, were added (Fig. 4). In the upper strand, differences in the boundaries of element B were also observed (Fig. 4, lanes 2 (top dashed line), 8 (top solid line), and 10 (top double line, top solid line)); nuclear extracts from HeLa, HepG2, and COS1 cells were protected from ϩ3 nt, Ϫ10 nt (area Ϫ10/ϩ3 nt slightly protected), and Ϫ14 nt to ϩ37 nt, respectively. The protection of element B in the lower strand (Fig. 5, lanes 13-15) was similar (Ϫ8/ϩ30 nt) for extracts from the three cell lines. Element C (Ϫ124/Ϫ101 nt) was seen clearly only in the lower DNA strand with extracts from all three cell lines (Fig. 5). Variations in the DNase I concentration, or in the amount of Identification of Factors Acting in Trans with Elements A, B, and C-Incubation of double-stranded oligonucleotides corresponding to elements A, B, or C with nuclear extracts from COS1, HeLa and HepG2 cells resulted in electrophoretic mobility shifts (Fig. 6). Element A exhibited similar shifts with the three sources of extract (Fig. 6, A and B). Binding could be competed with an excess of unlabeled element A (Fig. 6A, lane 2) and with oligonucleotides corresponding to the binding sites of members of the nuclear hormone receptor superfamily, such as COUP-TF1 and HNF4 (Fig. 6A, lanes 3 and 4, respectively). Element A was also effectively competed with the peroxisome proliferator response element of the rat hydratase-dehydrogenase gene (HD-PPRE) (Fig. 6A, lane 6), which is very similar in sequence to element A. The HD-PPRE has been shown to interact with a number of nuclear hormone receptors, including COUP-TF1 and 2, HNF4, RXR, PPAR, TR, and LXR (9,10,15). No competition for binding to element A was seen with oligonucleotides corresponding to the binding sites for elements B, C, and the general transcription factors listed in Table I (e.g. CTF/NF1; Fig. 6A, lane 7). Supershift analysis with anti-COUP-TF1 antibodies confirmed the interaction of COUP-TF1 with element A in all cell extracts (Fig. 6B). Antibodies against HNF4 showed a supershift with HepG2 extract, but not with HeLa or COS1 extracts (Fig. 6B). Antibodies against RXR␣ and PPAR␣ also generated supershifts with extracts from HeLa and HepG2 cells (Fig. 6B). Similarly, antibody to TR␣, another ubiquitous nuclear hormone receptor shown to bind the HD-PPRE (12,15), reacted with element A and extract from HeLa cells (other sources of extracts not tested).
Mobility shifts were generated with element B and nuclear extracts from the three cell lines. Binding could be competed with excess unlabeled element B, but not elements A or C. Data for HepG2 nuclear extracts are presented in Fig. 6C. Competition assays suggested interactions of element B with HNF3 (lane 1) and members of the CCAAT-binding proteins, such as C/EBP␣ and CTF/NF1 (lane 2 or 10 and lane 4, respectively). The Epstein-Barr Virus nuclear antigen 1-nuclear factor 1 (EBNA1-NF1), another element known to bind members of this protein family, also competed strongly with element B for binding (Fig. 6C, lane 6). Antibodies to isoforms ␣, ␤, and ␥ of HNF3 did not generate supershifts with element B (Fig. 6D for  HNF3␣). In contrast, supershift analysis confirmed the interaction of C/EBP␣ with element B in HepG2 extracts after overnight incubation at 4°C of the antibody with the nuclear extract mixture before addition of the probe (Fig. 6D). Mutant B3 (ϩ3/ϩ27 nt; Fig. 6C, lane 12), or half-site mutant B1 (ϩ3/ ϩ13 nt; Fig. 6C, lane 9) could not compete for binding to element B, while mutant B2 (ϩ10/ϩ27 nt; Fig. 6C, lane 11) competed only slightly.
Mobility shifts were generated with element C and nuclear extracts from all three cell lines (Fig. 6E, data for HepG2 and HeLa nuclear extracts). Binding to element C could be competed with excess unlabeled element C and also partially with element A (Fig. 6E, lanes 3 and 1, respectively). Additional competition assays suggested interactions of element C with HNF4, HNF3, and C/EBP␣ (Fig. 6E, lanes 6 -10). Supershift analysis confirmed a partial interaction of element C with HNF4 (Fig. 6F).
Interaction of Individual Transcription Factors with Elements A, B, and C-Interaction of the AT 5Ј upstream promoter was observed with two liver-enriched factors, HNF4 and C/EBP␣. HNF4 bound to elements A and C (Fig. 7, A and D,  respectively), whereas C/EBP␣ bound to element B (Fig. 7B). 5Ј truncation of element A (removal of a nuclear hormone receptor consensus half-site at Ϫ92/Ϫ88 nt) did not suppress HNF4 binding (Fig. 7A). The additional element A mutations tested were targeted to the nuclear hormone receptor half-site TGACC at Ϫ75/Ϫ79 nt and to the 5 nt immediately upstream. None of the mutants tested could bind HNF4.
Interactions between ubiquitous nuclear hormone receptors and element A were also detected by EMSA. Addition of both PPAR␣ and RXR␣ with element A resulted in a shift, suggest- ing heterodimerization of these receptors (Fig. 7A). No binding was seen with either receptor alone. No mutant of element A bound the combination of both receptors. In contrast, binding of TR␣ was seen with the wild-type and all mutant A elements, although TR␣ binding to mutant A4 was greatly reduced (Fig.   7C). Fig. 7 also presents the binding profile of COUP-TF1 with the wild-type and mutant A elements. COUP-TF1 interacted strongly with element A, with additional species above the primary shifted species (Fig. 7A). COUP-TF1 dimerization on element A was assessed by incubation with full-length COUP-TF1 and with a truncated form of the same receptor (Tr-COUP). This truncated form has been shown previously to retain intact DNA binding properties (7). After EMSA, a new shifted species migrating between the COUP-TF1 and Tr-COUP species was seen, indicating dimerization of the COUP-TF1 receptor (Fig. 7C, arrow). 5Ј truncation of element A (Ϫ87/ Ϫ65 nt element) did not suppress COUP-TF1 binding (Fig. 7A). Mutant A1 (TGAgg at Ϫ76/Ϫ75 nt) and mutant A2 (ggaaT-GACC at Ϫ83/Ϫ80 nt) retained the ability to bind COUP-TF1 (Fig. 7A). Mutant A3, which combines the mutations in mutants A1 and A2, and mutant A4 (TcAgg modification at Ϫ78/ Ϫ76/Ϫ75 nt) abolished COUP-TF1 binding (Fig. 7A). These results suggest that the Ϫ79/Ϫ75 nt half-site is critical for COUP-TF1 binding and that the nucleotides immediately upstream of this motif also influence COUP-TF1 binding, although to a lesser degree.
We attempted to detect receptor heterodimerization on element A. There was no evidence of COUP-TF1 heterodimerization with any receptor (Fig. 7C). Likewise, HNF4 did not heterodimerize with TR␣ or PPAR␣, although preliminary evidence from supershift analysis suggested a RXR␣-HNF4 interaction. As seen in Fig. 8, addition of antibodies resulted in   7 and 11; double arrowhead). Anti-PPAR␣ antibodies did not form a readily discernible supershifted species; however, these antibodies did result in the disappearance in the shifted species formed with element A (Fig. 8, lane 8). Addition of antibodies against RXR␣ to a reaction containing element A and both HNF4 and RXR␣ formed a supershifted species (Fig. 8, lane 11). As RXR␣ alone was unable to bind element A, the results suggest that a complex of HNF4-RXR␣ can form on element A.

Effects of in Vivo Expression of Nuclear Hormone Receptors and C/EBP␣ on the Transcriptional Efficiency of the AT 5Ј
Upstream Promoter-Expression of HNF4 and C/EBPa in transient transfections of HepG2 and BSC40 cells activated transcription from the AT 5Ј-upstream promoter (Fig. 9). RXR␣ activated transcription in BSC40 cells, but almost not at all in HepG2 cells. COUP-TF1, TR␣, HNF3, and, to a lesser extent PPAR␣, repressed transcription in both cell types. Activation by HNF4 was reduced by coexpression of COUP-TF1, TR␣, and PPAR␣/RXR␣. In HepG2 cells, activation by HNF4 was reduced by coexpression of PPAR␣ or RXRa alone. In contrast in BSC40 cells, when HNF4 and RXRa were cotransfected, the increase in luciferase expression was up to 7-fold higher than the increase expected by the sum of activities of both factors, suggesting RXR␣-HNF4 synergy. The addition of all the receptors tested resulted in a marked decrease of AT gene transcriptional efficiency.

DISCUSSION
The 5Ј-upstream element, located at Ϫ150/ϩ68 nt, was shown to promote basal transcription. Deletion of the three areas protected in DNA footprint analysis provided evidence for the modular nature of this element. Deletion of element A at Ϫ92/Ϫ65 nt, element B at ϩ1/ϩ37 nt, and, to a lesser extent, element C at Ϫ124/Ϫ101 nt dramatically decreased luciferase reporter activity. Elements A, B, and C were not able to promote luciferase expression individually when inserted upstream of the luciferase gene in pSV0A-L⌬5Ј. A role for these sequences in transcription initiation is therefore unlikely. In contrast, tandem copies of the two elements A and B enhanced the transcriptional efficiency of two heterologous minimal promoters. This is in agreement with observations of a number of liver promoters, which contain a cluster of modular elements able to initiate and to modulate (principally enhance) transcription in close proximity to the start site(s) of transcription. Examples of promoters homologous to that of the human AT gene include the promoters of genes encoding several coagula- tion factors, their inhibitors (serpins), apolipoproteins, and transferrin (4, 16 -19).
When the AT gene region upstream of exon 1 was searched for classical eukaryotic control sequences, TATA box consensus sequences were found, but only upstream of the Ϫ150/ϩ68 nt element (first consensus sequence at Ϫ155 nt). In contrast, perfect matching for an initiator element (CCACCC) was found at Ϫ43 nt (20). As seen in Fig. 2, Ϫ28/ϩ67 nt-luc was unable to promote transcription, while Ϫ70/ϩ67 nt-luc retained 20 -30% of the activity of the entire Ϫ150/ϩ67 nt region. Residual activity (15-35%) was also seen for a 3Ј-deletion ending at Ϫ28 nt. These data suggest the presence of an initiator at Ϫ43 nt. Ongoing cotransfection experiments with the transcriptional activators identified herein and with deletion mutants of Ϫ150/ ϩ68 nt-luc should confirm the presence of this initiator. Furthermore, initiator elements mostly encompass mRNA start site(s) (20). Prochownik previously located a single AT mRNA start site 43 nt downstream of the putative initiator CCACCC (21). We wanted to confirm these findings and to reassess the 5Ј-end of the AT mRNA in tissues and cells of hepatic origin, using different experimental approaches. Mapping of the 5Јend of AT mRNA by S1 nuclease protection, primer extension, or 5Ј-rapid amplification of cDNA ends (RACE) generated a single major product, which placed the start site of transcription in agreement within 1-2 bases with the previously published start site (data not shown). The sequence encompassing this site, ACCAGTTT (Ϫ1/ϩ7 nt), is also homologous to the mammalian initiator of transcription consensus sequence Py-PyCAN(T/A)PyPy (20). We also observed protection in this region, participation of element B (ϩ1/ϩ37 nt) in enhancing promoter strength, and binding by transcription factors. Element B alone was nevertheless unable to initiate luciferase expression. This region, especially in the area immediately upstream of the start site at Ϫ14/ϩ1 nt, also shows homology with recently described GAGA boxes, control sequences for growth hormone-induced transcription (22). It has been shown that GAGA boxes also determine, in part, basal promoter activity in non-hormonal responses and bind specific zinc finger proteins such as PUR-1 and MAZ-1 (22,23).
The three elements A, B, and C are able to bind nuclear proteins from several cell lines. These results are summarized in Fig. 10. None of the elements was protected solely by HepG2 nuclear extracts, contrary to previous results (4). Only differences in protection strength, upstream boundary, or the sequence around the transcription start site in the upper strand could be detected. Our results are somewhat similar to the regulatory features of a transgene AT Ϫ680/ϩ24 nt upstream sequences linked to the apolipoprotein A-II gene (24). In the AT region, four protected areas (I-IV, corresponding to Ϫ138/Ϫ123, Ϫ112/Ϫ104, Ϫ89/Ϫ68, and Ϫ48/Ϫ22 nt, respectively, in the upper strand) were identified by footprint analysis using mouse liver extracts. Two of these areas (II and III) are similar to elements A and C, respectively, of the AT gene, except that element C extends further upstream to Ϫ124 nt. Protection of element B in our case and of element I of Tremp et al. (24) were not observed, because the 5Ј and 3Ј boundaries of the probes used for footprint analysis were located directly in these two areas. We did not observe protection of the putative initiator location (element IV at Ϫ48/Ϫ22 nt in Ref. 24) with extract from any cell line.
EMSA with nuclear extracts or in vitro translated factors allowed identification of interactions of the promoter with two liver-enriched factors: HNF4, a member of the nuclear hormone receptor superfamily, and C/EBP␣, a leucine zipper CCAAT-binding protein. These two factors have been implicated with the liver-enriched expression of many other genes (8,16,19). HNF4 bound to elements A and C in extracts of HepG2 cells, but not COS1 or HeLa extracts. Interactions of HNF4 with area II of the Ϫ680/ϩ24 nt AT sequences (equivalent to element A) have also been reported in the transgenic  (24). However, the absence of HNF4 in CV1-derived cells (COS1 and COS7) has been well documented (25). In element A, the HNF4 consensus binding site proposed by Sladek (8) matches in 10 of 13 nt of the Ϫ86/Ϫ74 nt region. In addition, binding data from mutants of element A with in vitro translated HNF4 strongly suggested that the Ϫ75/Ϫ83 nt region is involved in binding HNF4. In element C, 9 of 13 nt and 11 of 13 nt are identical with the HNF4 consensus binding sequence at Ϫ104/Ϫ116 nt and Ϫ111/Ϫ123 nt, respectively. All regions of binding identified in these two elements contain direct repeats of half-site nuclear hormone consensus sequences (ACTGG) separated by 1 nt (DR1 element). C/EBP␣, which gave only a slight supershift in HepG2 cells, bound to the ϩ1/ϩ37 nt region of element B. Mutation analysis suggested that the GTTTTCAGGC region at ϩ4/ϩ11 nt is involved in binding.
Footprint protection was also seen with extracts of cells of non-hepatic origin. This result suggests that ubiquitous factors also bind to the Ϫ150/ϩ68 nt element. Element A was shown to interact with several ubiquitous nuclear hormone receptors, e.g. COUP-TF1, RXR␣, PPAR␣, and TR␣. COUP-TF1 and TR␣ bound element A (in the HNF4-binding region), and data from the analysis of binding to mutant forms of element A supported their reported higher promiscuity of binding as compared to HNF4, RXR␣, and PPAR␣ (15,26). RXR␣ and PPAR␣ bound as heterodimers, as documented previously (9,15). Data presented in Fig. 8 also suggest RXR␣ and HNF4 interaction. In addition, control supershift experiments and direct immunoprecipitation with radiolabeled translated products confirmed that antibodies to RXR␣ and HNF4 were specific for their antigens and did not cross-react (data not shown). Moreover, although our observations are preliminary, synergistic effects were observed after cotransfection of these factors into BSC40 (and HeLa cells as well). Whether RXR␣ and HNF4 heterodimerize, interact through bridging proteins provided by the reticulocyte lysate, or associate via allosteric interaction (26 -28) remains to be determined. Differential effects of the combination of HNF4-RXR␣ in HepG2 and BSC40 cells could be explained by differences in cell milieu. For example, HNF4␣ and C/EBP␣ were present in HepG2 cells and absent in the other cell lines tested. In this regard, cotransfection effects in BSC40 cells were less likely to be modulated by endogenous factors. This cell line, in which only COUP-TF1 has been iden-tified (7), generated only 5-10% of the luciferase expression of HepG2 cells. Whether HNF4 and RXR␣ could interact directly in vivo to modulate constitutive, physiologic expression of the AT gene remains to be demonstrated.
Additional cotransfection experiments showed that HNF4, C/EBP␣, and RXR␣ were able to activate the AT 5Ј-promoter, while COUP-TF1, TR␣, and HNF3 had a repressive effect. Interaction with the transcription initiation machinery, either directly or through bridging cofactors, could explain the activating and repressive effects of these unliganded nuclear hormone receptors (29,30). HNF4 transactivation potential was also strongly reduced by COUP-TF1, TR␣, and PPAR␣ ϩ RXR␣. Direct binding competition between HNF4 and these four receptors has been demonstrated, at least in the induced transcription setting (15,26). Our study suggests that interactions of elements A and C with nuclear hormone receptors could directly modulate initiation of AT gene expression. It is likely that additional members of the nuclear hormone receptor superfamily will interact with elements A and C in a constitutive or an induced setting. For example, heterodimerization of RXR␣ with several other receptors has been shown to participate in the transcriptional responses induced by retinoic acid, retinoids, peroxisome proliferators, vitamin D, and triiodothyronine (7,9,10,15,26,27). Furthermore, interactions with subtypes or isoforms of the factors identified in this study or with newly described receptors, some of which might interact with half-site motifs (26,27), multiply the number of potential modulatory responses through elements A and C. Moreover, elements A and C contain several putative arrangements of TGACC half-sites in addition to direct repeats.
DNA-binding proteins other than nuclear hormone receptors, e.g. the single-stranded binding proteins PYBP or pTB (31,32) originally identified in studies of the promoters of the transferrin gene and other liver-enriched genes, also interacted with element A (data not shown). Mutation 2, which destroyed HNF4 or RXR␣-PPAR␣ binding to element A, also destroyed the pyrimidine strand immediately upstream of the Ϫ75/Ϫ79 nt TGACC half-site. A similar mutation in the rat aminotransferase gene has been reported to also destroy pTB binding (32). Whether or not PYBP or pTB could modulate nuclear hormone receptor interactions and/or binding to DNA is unknown. Moreover, ubiquitous factors interacting with elements B and C have not been characterized in this study. Preliminary data provided by competition assays (Fig. 6) suggest possible interactions of element B with ubiquitous CCAAT-binding proteins of the NF family. This result will have to be confirmed with supershifts and direct binding assays. For element B (Ϫ14/ϩ37 nt), putative crosstalk with C/EBP isoforms and subtypes involved in the acute phase response, and with signal transduction and hormonal pathways (e.g. growth hormone-responsive elements) should be investigated in the induced setting (22,23).
The detection of a promoter-like activity in IVS1 was surprising, but a number of regulatory sequences have been found in IVSs, including IVSs of genes encoding serpins, liver-expressed coagulation factors, and a number of apolipoproteins (33,34). The orientation of this element in the coding DNA strand of the AT gene suggested an open reading frame was present in the lower strand of IVS1. Computer analysis did not reveal any apparent coding regions. The search for long terminal repeat sequences in this region of the gene was also unsuccessful. The absence of reported activity of a construct including both the 5Ј-upstream and IVS1 elements could be due to the fact that both elements function in opposite directions and that promoter activity is quenched, at least in reporter assays. It could also suggest that the IVS1 element could silence expres- sion of the 5Ј-upstream element. Our findings to date are preliminary. Effectively, exon 1 and the first 300 nt downstream of it were shown to also decrease (Fig. 1G), but not abolish, the activity of the 5Ј-upstream promoter. Therefore, the actual role of the IVS1 element on AT gene expression remains to be determined.