Mechanism of the Tumor Necrosis Factor α-mediated Induction of Endothelial Tissue Factor

This study examines the regulation of the human tissue factor (TF) promotor in vitro and in vivo. Transient transfections were performed in bovine aortic endothelial cells to investigate the role of two fundamentally different AP-1 sites and a closely located NF-κB site in the human TF promotor. The NF-κB site is functionally active, since overexpression of NF-κB(p65) resulted in induction of TF mRNA and activity. Promotor analysis showed that NF-κB induction was dependent on the integrity of the region from base pair −188 to −181. Overexpression of Jun/Fos resulted in TF induction of transcription and protein/activity. Functional studies revealed that the proximal AP-1 site, but not the distal, was inducible by Jun/Fos heterodimers. The distal AP-1 site, which has a G → A switch at position 4, was inducible by Jun homodimers. Electrophoretic mobility shift assays, using extracts of tumor necrosis factor α (TNFα)-stimulated bovine aortic endothelial cells, demonstrated TNFα-inducible binding to the proximal AP-1 site, comprising JunD/Fos heterodimers. At the distal AP-1 site, only minor induction of binding activity, characterized as proteins of the Jun and ATF family, was observed. Consistently, this site only marginally participates in TNFα induction. Functional studies with TF promotor plasmids confirmed that deletion of the proximal AP-1 or the NF-κB site decreased TNFα-mediated TF induction to a higher extend than loss of the distal AP-1 site. However, integrity of both AP-1 sites and the NF-κB site was required for optimal TNFα stimulation. The relevance of these in vitro data was confirmed in vivo in a mouse tumor model. Expression plasmids for a dominant negative Jun mutant or I-κB were packaged in liposomes. When either mutated Jun or I-κB were injected intravenously 48 h before TNFα, a reduction in TNFα-mediated TF expression in the tumor endothelial cells was observed. Simultaneously, fibrin/fibrinogen deposition decreased and free blood flow could be restored. Thus, TNFα-induced up-regulation of endothelial cell TF depends on a concerted action of members of the bZIP and NF-κB family.

Unstimulated endothelial cells express no tissue factor (TF) 1 in vitro and in vivo (1)(2)(3)(4)(5)(6). Recent studies show that TF expression in vitro can be induced by TNF␣ (7)(8)(9)(10)(11). In vivo, however, expression of TF has only been shown in selected vascular beds: in the splenic endothelium in a septicemia model (12) and in tumors such as Meth-A and Karposi sarcomas (13,14). The human TF promotor has been characterized, and its function has been extensively studied in monocytes/macrophages (15)(16)(17)(18). The porcine TF promotor has been described in endothelial cells (11). The data available from the human TF promotor suggest that two AP-1 and the NF-B site are central in the endotoxin-dependent regulation of TF expression (16 -18). In contrast, the study of the porcine TF promotor shows that mainly the induction of NF-B is responsible for lipopolysaccharide-and TNF␣-mediated induction of endothelial TF expression (11).
The discrepancy in the role of AP-1 in the human and porcine TF promotor might be due to differences in the sequence composition of the AP-1 binding sites of the various species. Sequence alignments revealed important differences between the AP-1 sites of the different species (11). The porcine TF promotor has two non-canonical AP-1 sites that differ from the defined AP-1 consensus sequence in one central base (11). Noncanonical AP-1 sites have been reported of being weak binding sites (19). In contrast, the proximal AP-1 site in the human (and the mouse) TF promotor contains the core of the consensus sequence (11,15,16) and thereby represents a high affinity site for AP-1 binding. This indicates that different AP-1-like proteins may be involved in the regulation of TF expression in different species. If the porcine model (11) would be relevant for human disease, then blocking of NF-B activation would provide a powerful way to prevent excess TF expression in human disease. If the human model (15)(16)(17)(18) is relevant, then inhibition of NF-B activation might lead to increased c-Fos transcription (20,21) and thereby to AP-1-mediated TF induction. To resolve this issue, we studied the role of both AP-1 sites and their cooperative action with the NF-B site in the human TF promotor.
A number of homo-and heterodimers can recognize AP-1 sites (22)(23)(24)(25) but exhibit different affinities for different motifs (26). These proteins have been termed bZIP family (27) due to their ability to dimerize via an ␣-helical leucine "zipper." The members of the Jun subfamily c-Jun, JunB, and JunD are highly homologous in their dimerization and binding domains and can compete for the same AP-1 sites (28). The transactivating capacities of c-Jun and JunD are dramatically increased in combination with c-Fos (29,30). The functional homologues of c-Fos, Fra-1, Fra-2, and FosB can also dimerize with Jun proteins (30 -33). In addition, members of the ATF/CREB family like ATF-2, ATF-3, and ATF-4 (but not ATF-1) are capable of binding to proteins of the AP-1 transcription factor family (25,26,34). Thus, it was our hypothesis that the distal noncanonical AP-1 site in the human TF promotor and the proximal canonical AP-1 site bind different members of the AP-1/bZIP family.
Tissue Culture-Bovine aortic endothelial cells (BAEC) were cultured in DMEM supplemented with 10% FCS as described previously (41,42) and characterized as endothelial cells by the expression of von Willebrand factor, thrombomodulin, low basal levels of TF, and morphologic features. Cells were passaged every 8 -10 days without showing gross morphological changes until passage 15. Passages 4 -8 were used. All experiments were performed with cells that had been confluent for 3-5 days. For transfection experiments, cells growing in the logarithmic phase were used.
Determination of Tissue Factor (TF) Activity by One-stage Clotting Assay-One-stage clotting assays were performed as described (1). After washing, cells were removed non-enzymatically by scraping in barbital buffer, pH 7.6, collected by centrifugation for 5 min at 1000 rpm, and resuspended in 100 l of the same buffer. After addition of 100 l of citrated bovine plasma 100 l of citrated bovine plasma, and 100 l of 25 mM CaCl 2 solution (Behring, Marburg, Germany) the samples were incubated at 37°C. The time from addition of CaCl 2 to the first defined fibrin strand was determined. TF activity was calculated by comparing the measured clotting time with a standard curve made with known amounts of TF (1). The measured amount of TF was expressed as picograms of TF/10 6 cells Ϯ S.D. All experiments were performed at least three times, and each experiment was done in triplicates.
Determination of Tissue Factor (TF) Activity by Monitoring the Hydrolysis of S2222 Synthetic Peptide Substrate-TF activity was also assessed by monitoring the hydrolysis of the synthetic peptide substrate Bz-Ile-Glu-Gly-Arg-p-nitroanilide (S2222, KabiVitrum, Stockholm, Sweden) as described previously (43). Cells were harvested in PBS, pH 7.4, and incubated with 30 l of human factor VII (final concentration 35 g/ml) (44) and 10 l of factor X (final concentration 200 g/ml) (44) in the presence of 10 mM CaCl 2 . Aliquots (30 l) were added to 470 l of 50 mM Tris, pH 7.9, 175 mM NaCl, 5 mM EDTA, and 0.5 mg/ml BSA. Factor Xa formation was assessed using 100 l of S2222 added to the entire 0.5-ml sample. Hydrolysis was monitored at room temperature by measuring the change in absorbance at 405 nm, using a Beckmann DU 7400 spectrophotometer (Beckmann, Dreieich, Germany). Factor Xa (final concentration 20 g/ml) (44) served as control. Each experiment was repeated three times.
Statistical Analysis-Data were analyzed with the aid of SIGMA PLOT software (Jandel Scientific). Levels of significance were determined by Student's t test. Any p value of 0.05 and below was considered to be significant. For calculation of p values in Fig. 1c, the Mann-Whitney U test was used.
Nuclear Run-on Transcription Assay-Nuclear run-on transcription assays were performed essentially according to the procedure of Greenberg and Ziff (45) as described elsewhere (46 -48). In brief, nuclei were harvested from 2 ϫ 10 7 cells after 42 h of transfection. Run-on reactions were performed in 0.7 M KCl, 50 mM MgCl 2 , 50 mM Tris-HCl, pH 8.0, 25 mM DTT, 1 mM EDTA in the presence of 250 Ci of [␣-32 P]UTP (3000 Ci/mmol) and incubated for 30 min at 30°C. The synthesized mRNA was incubated with DNase I for 5 min at 30°C, treated with proteinase K (10 mg/ml), and extracted with 0.45-m Millipore filters (type HA). The RNA was collected by EtOH precipitation and redissolved in 300 l of DEPC-H 2 O, 1 l was counted, and equal numbers of Cerenkov counts were made up to 2 ml of hybridization solution and added to the previously prepared slot blot filters. Hybridization was performed without prehybridization in 50% formamide, 5 ϫ SSC, 5 ϫ Denhardt's solution, 1% SDS for 4 days at 42°C. Filters were washed three times at room temperature for 10 min in 2 ϫ SSC and once for 10 min at 60°C in 1 ϫ SSC. Blots were exposed to Amersham Hyperfilms for 1-4 weeks at Ϫ80°C with intensifying screens. The density of autoradiographic signals was quantitated using a Beckman DU 7400 densitometer (Beckman, Dreieich, Germany) (46 -48). For preparation of filters, 3 g of human TF plasmid DNA or the respective household and control plasmids (GAPDH, Meth-tRNA, pSPT18, p19-Luc, TNF␣ cDNA) were applied to Hybond-N membranes using a slot blot apparatus (Schleicher & Schü ll, Dassel, Germany) as described previously (46 -48).
Electrophoretic Mobility Shift Assay-For electrophoretic mobility shift assays (EMSA), nuclear proteins were harvested by the method of Andrews (52) as described previously (47). Approximately 2 ϫ 10 7 were harvested in cold PBS, pH 7.4, and pelleted at 1500 rpm for 5 min. The pellet was resuspended in 400 l of cold buffer A (10 mM HEPES-KOH, pH 7.9 at 4°C, 1.5 mM MgCl 2 , 10 mM KCl, 0.5 mM DTT, 0.2 mM PMSF), and incubated for 10 min on ice. The samples were centrifuged for 10 s at highest speed, the supernatant was discarded and the pellet was resuspended in 100 l of cold buffer C (20 mM HEPES-KOH pH 7.9, 25% glycerol, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 0.5 mM DTT, 0.2 mM PMSF) and incubated on ice for 20 min. After centrifugation (2 min, 4°C, highest speed) the supernatant was quick-frozen at Ϫ80°C. Protein concentration was determined using a colorimetric assay based on Bradford (53). When organ tissue was used, the above protocol was modified according to Deryckere and Gannon (54). Large pieces of tissue (0.1 ϫ 0.2 cm) were frozen in liquid nitrogen, broken mechanically with a hammer, and transferred to a 50-ml Falcon tube containing 5 ml of cold buffer A (10 mM Hepes-KOH, pH 7.9, at 4°C, 10 mM KCl, 1.5 mM MgCl 2 , 0.5 mM DTT, 1 mM EDTA, 0.2 mM PMSF, 0.6% Nonidet® P-40). The tissue was homogenized in an Ultrathurax (Wheaton, Millville, NJ) for 1 min, transferred to an 15-ml tube, and centrifuged for 30 s at 2000 rpm, 4°C to remove tissue debris. The supernatant was incubated on ice for 10 min and centrifuged for 5 min at 8000 rpm at 4°C. The supernatant was discarded, and the nuclear pellet was resuspended in 100 l of buffer B (25% glycerol, 20 mM Hepes-KOH, pH 7.9 at 4°C, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 0.5 mM DTT, 0.2 mM PMSF, 2 mM benzamidine, 5 mg/ml leupeptin) and incubated on ice for 20 min. Cellular debris was removed by 2 min of centrifugation at 4°C and the supernatant was quick-frozen at Ϫ80°C. Protein concentrations were determined as above.
Oligonucleotides, listed in Table II, were synthesized on a Gene Assembler Plus (Pharmacia, Freiburg, Germany) and purified on histidine gels (55). They were labeled by kinasing to a specific activity Ͼ5 ϫ 10 7 cpm/g DNA. Binding of AP-1 was performed in 25 l of 10 mM HEPES, pH 7.9, 0.1 mM EDTA, 75 mM NaCl, 4 mM MgCl 2 , 2 mM DTT, 17.5% glycerol, 1 mg/ml BSA (DNase-free) in the presence of 0.01 g/l poly(dI-dC) (47). For organ preparations poly(dI-dC) was scaled up to 0.15 g/l. When recombinant proteins produced in rabbit reticulocyte lysates (Promega) were included in the reaction, the poly(dI-dC) concentration was increased to 0.05 g/l. When recombinant human c-Jun (Promega) was used, poly(dI-dC) was replaced by 0.01 g/l AP-3 oligonucleotides (Promega) according to the manufacturer's instruction. NF-B binding was performed in 10 mM HEPES, pH 7.5, 0.1 mM EDTA, 100 mM NaCl, 1 mM ZnCl 2 , 4 mM MgCl 2 , 2 mM DTT, 17.5% glycerol, 1 mg/ml BSA (DNase-free), and 0.1 g/l poly(dI-dC) in a total of 25 l (47). For organ preparations, the poly(dI-dC) concentration was increased to 0.3 g/l. 8 -10 g of nuclear extract were incubated on ice for 20 min in the appropriate binding buffer before adding approximately 1 ng of labeled oligonucleotide. A typical binding reaction contained 50,000 cpm (Cerenkov). The samples were incubated at room temperature for and additional 15 min. Protein-DNA complexes were separated from the free DNA probe by electrophoresis through 4% (AP-1) or 5% (NF-B) native polyacrylamide gels containing 2.5% glycerol and 0.5 ϫ TBE buffer (47). The gels run at room temperature with 30 mA for approximately 2.5 h. Gels were dried under vacuum on Whatmann D-81 paper (Schleicher and Schü ll, Dassel, Germany) and exposed for 12-48 h to Amersham Hyperfilms at Ϫ80°C with intensifying screens. Specificity of binding was ascertained by competition with a 160-fold molar excess of cold consensus oligonucleotides. For supershifting experiments 2.5 g of the respective antibody were applied to the reaction mixture at the time the labeled oligonucleotide was added.
Synthesis of Recombinant Proteins-c-Jun and c-Fos RNA were synthesized under the control of the T7 promotor according to the method of Melton and Krieg (56). Yeast inorganic pyrophosphatase was included in the reaction to increase the yield of RNA obtained (57). The size of in vitro transcribed RNA was assessed by 5% polyacrylamide gels. About 3.6 g of c-Jun or c-Fos RNA were added to 70 l of rabbit reticulocyte lysate (Promega) to translate proteins in vitro by the method of Sassone-Corsi (36). When c-Jun and c-Fos were cotranslated to form heterodimers, 2 g of each RNA was used. Efficient translation was monitored in a parallel reaction using [ 35 -S]Methionine as substrate. 0.3-10 l (approximately 0.15-5 g) of in vitro translated proteins were used in EMSA. Unprogrammed lysate served as control.
Transient Transfection of Endothelial Cells-Logarithmically growing endothelial cells were transfected as described by Lee (58,59) with minor modifications (47). Cells were grown in DMEM containing 10% FCS to 70% confluence. 1.4 g of the appropriate plasmids/ml of medium were transfected by the calcium phosphate (CaPO 4 ) method. Cells were exposed to the precipitate for 6 h (HUVEC) to 8 h (BAEC). Medium was changed, and cells were incubated for 42 h. Cells were harvested in the appropriate buffers.
Plasmid DNA used in transfections was isolated by alkaline lysis, followed by CsCl equilibrium centrifugation (50). For promotor studies 0.5 g of luciferase promotor constructs/ml of medium were transfected. To correct for variability in transfection efficiency 0.15 g of pSV-␤-Gal plasmid/ml of medium were included. For transactivation experiments, 0.25 g/ml pSV-c-Jun, pBK28(c-Fos), NF-B(p65), I-B, or mutated Jun were cotransfected with 0.5 g of luciferase containing promotor constructs. Reactions were filled up with pCAT-control (serving as mock control) to give the final DNA concentration of 1.4 g/ml medium. Cell extracts were prepared by lysis in 25 mM Tris phosphate, pH 7.8, 2 mM DTT, 2 mM 1,2-diaminocyclohexane-N,N,N,NЈ-tetraacetic acid, 10% glycerol, 1% Triton X-100 and assayed directly for luciferase activity (60). ␤-Galactosidase activity was determined in the same lysis buffer (61). Luciferase and ␤-galactosidase activity were determined for each sample. The ratio of luciferase activity to ␤-galactosidase activity served as a measure for normalized luciferase activity. For each experiment the normalized luciferase activity of the promotorless luciferase plasmid was subtracted from this quotient. The result was multiplied by 1000. To compare different transfections for each series of experiments, the relative Luc units of the triplicate were divided by those of the control construct pGL2-control. The quotient was multiplied by 100 and expressed as percentage of pGL2-control. In addition relative luciferase units were calculated as percentage of pHTF(Ϫ278)Luc basal expression (47). Each experiment was performed in triplicate. The data presented are the mean of at least three independent transfections performed. Standard deviations are given as vertical error bars.
Intravenous Somatic Gene Transfer-25-30 g of plasmid DNA were dissolved in 150 l of DMEM, added to 100 g of DOTAP/150 l of DMEM, and incubated at room temperature for 20 min (14). 300 l of this solution were injected intravenously via the tail vein of adult female C 3 H mice (18 -20 g) at the day, Meth-A sarcoma (10 6 cells/ animal) were implanted. A second DNA:DOTAP injection was performed 2 days before application of TNF␣. As soon as the tumors reached an average diameter of 0.5 cm (12-14 days after planting), TNF␣ (5 g/animal) was injected intravenously. Mice were sacrificed at the times indicated in the figure legend. Before harvesting the tumor tissues, mice were perfused with 30 -40 ml of PBS by intracardiac injection of PBS into the left ventricle (14).
Determination of Blood Flow-5 ϫ 10 5 of colored microspheres (10 mM, E-Z-Trac Ultraspheres, E-Z-Trac Inc., Los Angeles, CA) were injected for 10 -20 s into the left ventricle of the anesthetized mouse before the mice was sacrificed. Tumors were harvested, weighted, and hydrolyzed in sodium hydroxide-SDS solution (E-Z-Trac Inc., Los Angeles, CA) at 80°C overnight. Microbeads were isolated according to the manufacturer's instructions, counted microscopically, and expressed as microspheres per gram of tumor (Ϯ S.D.) (14,48).
Immunohistochemistry-Tumors were fixed in 4% formaldehyde. After cutting, sections were incubated with anti-mouse TF antibodies (48) for 2 h at room temperature as described previously (13,14,48). After three washes with 150 mM NaCl, 100 mM Tris-HCl, pH 7.5, sections were incubated for 1 h at room temperature with a peroxidase-conjugated second antibody. After washing color development was performed with 3-amino-9-ethylcarbazole and H 2 O 2 (13,14,48). Negative controls included omission of first or second antibodies and the substitution of the first antibody by nonspecific antibodies (data not shown). For immunofluorescence staining anti-fibrin-fibrinogen antibodies (fluorescine conjugate, Cappel, West Chester, PA) were used in a 1:8 dilution (14). Sections were incubated with the antibody for 45 min at room temperature, followed by three washes with PBS. 10 cuts from at least three different experiments were analyzed independently. 2 In Situ Hybridization-In situ hybridization was performed as de-2 Y. Zhang and Y. Deng, unpublished results.
scribed before (13,14). In brief, tumor sections were fixed in 4% paraformaldehyde, 0.5% glutaraldehyde in PBS, pH 7.4, and incubated with proteinase K (100 g/ml) for 10 min. Hybridization was performed for 16 h at 50°C against a digoxigenin-labeled TF antisense riboprobe, synthesized from a mouse tissue factor cDNA fragment (bp 721 to bp 1043). Hybridization solution was 0.6 M NaCl, 1 mM EDTA, 10 mM Tris-HCl, pH 7.6, 10% dextran sulfate, 1 ϫ Denhardt's, 50% formamide, 0.25% SDS, 100 mM DTT, 50 g/ml salmon sperm DNA, 100 g/ml tRNA, and 5-10 ng of labeled RNA probe. At the end of hybridization, samples were incubated for 30 min at 37°C with RNase A (20 g/ml) and washed once at 37°C for 30 min with 2 ϫ SSC, 50% formamide and twice for 30 min at 50°C with 0.2 ϫ SSC. Detection was performed immunologically with the DIG nucleic acid detection system according to the manufacturer's instruction. A TF sense riboprobe served as negative control (data not shown). TF was studied at several levels.

Tissue Factor Is Induced by Overexpression of Proteins of
(i) Nuclear run-on assays ( Fig. 1a) revealed that induction of transcription of the TF gene occurred.
(ii) The biological activity was characterized by its factor VII-dependent activation of factor X (synthetic substrate assay; Fig. 1b) and by a coagulant assay (one-stage clotting time; Fig.  1c). Since almost no factor X activation was observed in the absence of factor VII (data not shown), most of the coagulant activity induced is TF. But it cannot be absolutely excluded that other proteins involved in activation of coagulation are also inducible by overexpression of AP-1 or NF-B(p65). When BAEC were cotransfected with AP-1 and NF-B(p65), an additive effect in TF induction was observed in all systems tested ( Fig. 1).
Structural analysis of the human TF promotor has demonstrated two linked AP-1 sites, a proximal canonical site at Ϫ210 (A2), a distal non-canonical site at Ϫ223 (A1), and an NF-B site (N) at position Ϫ188 (15)(16)(17)(18). The availability of promotor constructs (Table I) allows for definition of the areas involved in activation of TF by AP-1 or NF-B(p65) (Fig. 2). The plasmids containing both AP-1 sites responded to c-Jun/Fos overexpression in the presence or absence of the NF-B site (Fig.  2a). However, deletion of the NF-B binding site reduced the induction by c-Jun/Fos (Fig. 2a), suggesting that endogenously present proteins (presumably p65/c-Rel) capable to bind to this site, act in concert with AP-1. To test whether optimal induction by c-Jun/Fos was dependent on the presence of endogenous NF-B, we cotransfected BAEC with plasmids overexpressing I-B. Overexpression of I-B reduced the c-Jun/Fos induction as long as the NF-B site was present (Fig. 2c).
When BAEC were transfected with a plasmid overexpressing NF-B(p65) (Fig. 2b), induction was maximal, when both AP-1 sites and the NF-B site were present. NF-B(p65) inducibility was reduced, but not lost, when the AP-1 sites were deleted (Fig. 2b). Deletion of the NF-B site resulted in loss of inducibility by NF-B(p65). To test, whether optimal NF-B induction was equally dependent on the presence of endogenous AP-1, we cotransfected BAEC with plasmids overexpressing NF-B(p65) and the Jun-specific inhibitor mutated Jun. This negatively dominant c-Jun point mutant (39) is able to dimerize with other members of the AP-1/bZIP family; however, it is unable to bind to the DNA recognition site. Since c-Fos does not bind to DNA at all, due to its failure to homodimerize (31), overexpression of mutated Jun leads to a significant reduction of Jun/Fos heterodimer binding. When NF-B(p65) and mu-tated Jun were cotransfected, a reduction of the NF-B(p65)mediated stimulation was seen (Fig. 2c). Thus AP-1 and NF-B(p65) act in concert in inducing TF (Figs. 1 and 2), which seems to be dependent on the presence of DNA sequences previously described as binding sites for AP-1 and the NF-B subunits c-Rel and p65 (11,(15)(16)(17)(18)62).
Both AP-1 Sites Differ in Their Affinity for Jun/Fos Heterodimers and Jun Homodimers-The human TF promotor and NF-B(p65) induces TF transcription, activity, and antigen. a, nuclei were extracted from BAEC transiently transfected with CAT (ϭ mock control), c-Jun, c-Fos, and/or NF-B(p65) overexpressing plasmids. Nuclear run-on experiments were performed as described under "Materials and Methods" to allow in vitro synthesis of [␣-32 P]UTP-labeled mRNA. This was hybridized against filters onto which cDNAs for TF (top), Meth-tRNA (bottom), pSPT18, and TNF␣ (negative controls, data not shown) had been fixed. b, activation of factor X by BAEC transfected with CAT (ϭ mock control), c-Jun, c-Fos, and/or NF-B(p65). Cells were transiently transfected with the respective plasmids, harvested after 42 h, and assayed for factor X activation in the presence of factor VII (35 g/ml). The generation of factor Xa was measured spectrophotometrically over a period of 15 min. S2222 served as substrate as described under "Materials and Methods." Two experiments were performed in triplicates with identical results. One typical experiment is shown. c, BAEC were transiently transfected with CAT (ϭ mock control), c-Jun, c-Fos, and/or NF-B(p65), harvested 42 h after transfection, and assayed for procoagulant activity as described under "Materials and Methods." TF activity of each sample was determined based on comparison with a standard curve established with known amounts of recombinant TF. The data Ϯ S.D. represent the mean of three independent experiments performed in triplicate (p values: control versus Jun/Fos ϭ 0.008; control versus NF-B ϭ 0.009; control versus Jun/Fos/NF-B ϭ 0.009) contains a distal non-canonical and a proximal canonical AP-1 binding site. Therefore EMSA were performed to compare the binding capacity of both AP-1 sites for Jun/Fos heterodimers (AP-1; Fig. 3). Decreasing amounts of c-Jun/Fos programmed rabbit reticulocyte lysate were incubated with either the proximal (A2; Fig. 3a) or the distal (A1; Fig. 3b) AP-1 site. The proximal canonical AP-1 site (A2) demonstrated the expected high affinity binding of c-Jun/Fos heterodimers detectable down to 2.5 l of programmed lysate (Fig. 3a). However, the distal non-canonical AP-1 site (A1) showed only very weak, nearly undetectable binding of c-Jun/Fos heterodimers (Fig.  3b). More rapidly migrating bands on gel shift assays were observed in all samples and represented binding not inhibited by excess unlabeled AP-1 oligonucleotides (data not shown) and are therefore marked as "nonspecific lysate bands." These observations were confirmed by functional promotor studies looking at the inducibility of mutants containing either the proximal (A2) or the distal (A1) AP-1 site (Fig. 3c). EMSA were performed to compare the binding capacity for both AP-1 sites for Jun homodimers. Decreasing amounts of recombinant Jun homodimers were incubated with either the canonical proximal AP-1 site (A2) (Fig. 4a) or the non-canonical distal AP-1 site (A1) (Fig. 4b). Binding to the proximal AP-1 site was detectable only down to 0.55 g of recombinant c-Jun homodimers, while binding was still detectable down to 2.5 l (ϭ 1.25 g) of Jun/Fos programmed lysate (Fig. 4a, lane 1). At the distal AP-1 site (A1), prominent binding of c-Jun homodimers, even at only 0.1 g, was detected (Fig. 4b). Our observations using EMSA were extended by transient transfection studies of BAEC using TF promotor mutants. Mutants containing only the distal AP-1 site (A1) were more strongly induced by c-Jun overexpression than mutants containing only the proximal AP-1 site (A2) (Fig.  4c). Therefore, the distal AP-1 site (A1) has properties of a Jun site rather than an AP-1 site.
In contrast, in nuclear extracts of BAEC three complexes (marked I, II, and III) were observed at the non-canonical distal AP-1 (Fig. 5b). When exposition of the films was extended for up to 8 days, a weak band, marked as complex I, occurred (Fig.  5b). The weak binding and its migration in the gel confirmed the results shown in Fig. 3b, i.e. this band is due to weak binding of Jun/Fos heterodimers.
Complex II (Fig. 5b) does not represent AP-1 heterodimers based on the following criteria. (i) The bands in control extracts (Fig. 5b, lane 1) and extracts from cells overexpressing Jun/Fos (Fig. 5b, lane 6) were only weakly (Fig. 5b, lanes 2 and 7) inhibited in the presence of mutated Jun. (ii) Complex II seen after TNF␣ induction (Fig. 5b, lane 3) was not competed by a 160-fold molar excess of cold AP-1 consensus oligonucleotides (Fig. 5b, lane 5) or by mutated Jun (Fig. 5b, lane 4), suggesting differences between control and TNF␣-stimulated cells. These data further indicate that complex II does not contain c-Fos or Fos-related proteins; however, they do not exclude the involvement of other members of the bZIP family, i.e. members of the ATF family (see Fig. 6b and Table II), which are able to bind DNA and therefore are only minorly influenced by mutated Jun. (iii) Complex II migrated more rapidly (Fig. 5b) than the complex formed with c-Jun homodimers (Fig. 4b).
Since complex III (Fig. 5b) was not at all competed by unlabeled consensus AP-1 oligonucleotides (see below) and also present in unprogrammed rabbit reticulocyte lysate (Fig. 3b), it is regarded as nonspecific as depicted in Fig. 3b.
To analyze the complexes I and II that were observed at the distal non-canonical AP site (Fig. 5b), characterization with supershifting antibodies was performed (Fig. 6a). The upper

TABLE I TF promotor plasmids used in this study
The organization of the TF promotor, which was cloned in front of the reporter gene luciferase, is given in the upper part of the table (15,16). The positions of the two AP-1 sites and the NF-B site along with the TATA box and the luciferase transcription start are indicated. The 5Ј-boundaries of the various TF promotor plasmids, which sequentially remove these transcription factors (row 1), published names of the constructs (row 2), and a short description of the cloning strategy (row 3) are shown. gel shift band (complex I ; Fig. 6a, lane 1) was reduced in the presence of pan-Jun antibodies (Fig. 6a, lane 2) and anti-JunD antibodies (Fig. 6a, lane 4) and suppressed in the presence of anti-c-Jun (Fig. 6a, lane 3) and anti-ATF-2 antibodies (Fig. 6a,  lane 7). In addition, anti-c-Fos (Fig. 6a, lane 5) and anti-ATF-1 antibodies (Fig. 6a, lane 6) slightly decreased the shift. Complex II was suppressed when anti-ATF-2 antibodies were included in the reaction (Fig. 6a, lane 7) and reduced in the presence of anti-pan-Jun and anti-c-Jun antibodies (Fig. 6a,  lanes 2 and 3), while anti-JunD, anti-c-Fos, and anti-ATF-1 antibodies did not affect binding (Fig. 6a, lanes 4 -6). Thus complex I and II also consist of different members of the AP-1/bZIP family. Intensity of the lower band (complex III), previously characterized as nonspecific (Fig. 5b), was not affected by any of the antibodies.
To further confirm this hypothesis, recombinant Jun and recombinant ATF-2 (0.5 g each), produced as inclusion bodies in Escherichia coli and able to heterodimerize, were used in binding reactions and their migration was compared to the complexes seen in extracts of control and TNF␣ stimulated cells (Fig. 6b). Consistent with the above data Jun homodimers bound to the distal non-canonical AP site (Fig. 6, panel a, lane  8, and panel b, lane 3) forming complexes that migrated in the gel at the same position as complex I (Fig. 6, panel a, lane 1,  and panel b, lane 2). The faster migrating ATF-2 homodimers demonstrated stronger binding to the distal non-canonical AP-1 site (Fig. 6b, lane 4) than Jun homodimers (Fig. 6b, lane  3). When equimolar amounts of recombinant ATF-2 and recombinant c-Jun were coincubated in the binding reaction, a slightly faster migrating complex was observed (Fig. 6b, lane  5). No significant binding was observed, when programmed Jun/Fos lysate was included in the binding reaction (Fig. 6b,  lane 6). The observed binding, seen in lane 6, is also present in unprogrammed lysate (Figs. 3b and 4b) and therefore regarded as nonspecific. To further characterize the TNF␣-inducible  (Table I) and a control plasmid (CAT ϭ "mock") or c-Jun and c-Fos (0.5 g each) and cultivated for 42 h. After harvest, luciferase activity was determined and normalized for transfection efficiency to the amount of ␤-galactosidase expressed by the plasmid pSV-␤-gal (Promega). The normalized data are expressed as relative Luc units and represent the mean of three independent experiments Ϯ S.D. performed in triplicate. The inducibility of the various TF promotor plasmids by c-Jun/Fos heterodimers is shown. The level of basal expression (transfected with CAT as control) is indicated with Basal.
complexes, a 500-fold molar excess of unlabeled AP-1 oligonucleotides (instead of 160-fold; Fig. 5b, lane 5) was included in the binding reaction with TNF␣ stimulated nuclear extract (Fig. 6b, lane 7). This unusual high excess of AP-1 competitor abolished binding of complexes I and II, but not of complex III. Since unlabeled oligonucleotides did not compete binding of complex III that was also detected in unprogrammed lysate (Figs. 3b and 4b), we defined complex III as nonspecific. The data shown in Fig. 6 (a and b) indicate that the distal noncanonical AP-1 site forms complexes with members of the Jun and ATF family. However, more detailed studies have to be performed to elucidate the nature and the functional significance of the proteins involved in complex I and II.
Complex II was the major specific binding observed after TNF␣ stimulation. Therefore the time course of complex II induction by TNF␣ was studied (Fig. 6c). TNF␣-mediated binding of complex II to the distal AP-1 site was biphasic, with a fast initial response at approximately 5 min and a slower response, maximal between 2 and 6 h (Fig. 6c). No signal was observed in the absence of nuclear extracts (Fig. 6c, lane 9).
At the proximal canonical AP-1 site TNF␣ induced time-dependent (Fig. 7a) and dose-dependent (Fig. 7b) induction of proteins, which reached a maximum between 30 min and 2 h. These proteins were characterized as AP-1 by (i) competing the binding with an excess of cold AP-1 consensus oligonucleotides (Fig. 7, panel a, lane 8 and panel b, lane 7) and (ii) by reducing binding activity by overexpression of mutated Jun (Fig. 5a,  lane 4). For the characterization using polyclonal antibodies, extract of TNF␣-stimulated cells (Fig. 7c, lane 1) was incubated with the antibodies (Fig. 7c, lanes 2-7). Migration was compared to the shift observed with recombinant Jun homodimers (Fig. 7c, lane 8) and Jun/Fos programmed lysate (Fig. 7c, lane  9). Pretreatment with anti-pan-Jun (Fig. 7c, lane 2) or anti-JunD antibodies (Fig. 7c, lane 4) resulted in supershifted bands; pretreatment with anti-c-Fos antibodies abolished the observed binding (Fig. 7c, lane 5). No reaction was observed  (Table I) and a control plasmid (CAT ϭ "mock") or c-Jun (0.5 g) and cultivated for 42 h. After harvest, luciferase activity was determined and normalized for transfection efficiency to the amount of ␤-galactosidase expressed by the plasmid pSV-␤-gal (Promega). The normalized data are expressed as relative Luc units and represent the mean of three independent experiments Ϯ S.D. performed in triplicates. The inducibility of the various TF promotor plasmids by c-Jun homodimers is shown. The level of basal expression (transfected with CAT as control) is indicated with Basal.

FIG. 5. TNF␣ induces binding of different proteins to the proximal (A2) and the distal AP-1 site (A1) of the human TF promotor:
BAEC were transiently transfected with CAT (ϭ "mock"), mutated Jun, or c-Jun and c-Fos (AP-1) overexpressing plasmids and cultivated for 42 h. Where indicated, TNF␣ (1 nM) was added 1 h before harvest. Nuclear extracts (10 g/binding reaction) were prepared as described under "Materials and Methods" and assayed in EMSA for binding to the proximal (a) or the distal (b) AP-1 site. To confirm AP-1 binding, TNF␣-induced nuclear extract was competed with a 160-fold molar excess of cold consensus AP-1 (lane 5). In addition, a parallel binding reaction was performed with 10 l of Jun/Fos programmed lysate (lane 8). a, the AP-1 complex binding to the canonical proximal AP-1 site is indicated with an arrow. b, the complexes binding to the distal noncanonical AP-1 site are termed I, II, and III and indicated by arrows (see "Results"). with antibodies directed against members of the ATF family (Fig. 7c, lanes 6 and 7). Consistent with the above results (Fig.  4a), no binding was observed with 0.3 g of Jun homodimers (Fig. 7c, lane 8). Therefore, the TNF␣-induced binding to the proximal canonical AP-1 site comprises JunD/Fos heterodimers. Although c-Jun-specific antibodies did not result in FIG. 6.

TABLE II
Sequences of double-stranded synthetic oligonucleotides used in this study Oligonucleotides were synthesized as sense and antisense strands (sense sequence is shown). The sense strand was labeled by kinacing with [␥-32 P]ATP and annealed at 68°C to a 10-fold excess of the cold antisense strand as described under "Materials and Methods." The AP-1 and NF-B binding motifs are underlined. Bases that differ from the consensus sequence are highlighted by bold letters. Fig. 6. Characterization of complexes I, II, and III, formed at the distal AP-1 site (A1). a, initial characterization of the nuclear complexes I, II, and III formed at the distal AP-1 site (Table II) derived from the human TF promotor indicates that complex I and complex II contain members of the Jun and ATF family; complex III did not react with the antibodies used and was defined as nonspecific (see "Results"). Characterization was performed six times, using three different nuclear extract preparations, with identical results. One typical experiment is shown. 10 g of nuclear extract were included in each binding reaction:  1-8). To demonstrate that the observed bands were not due to the oligonucleotide preparation used, a reaction without nuclear extract was included (lane 9). 10 g of nuclear extract were used in each binding reaction. DNA-protein complexes were analyzed on native 4% polyacrylamide gels. The very weak binding of complex I, the TNF␣-inducible complex II, and the nonspecific complex III are indicated by arrows.
supershifted or reduced binding, c-Jun might contribute to the observed complexes, since it might be possible that the anti-c-Jun antibody fails to recognize c-Jun in Jun/Fos heterodimers. 3 TNF␣-induced Binding of p65/c-Rel to the TF-derived NF-B Site-Stimulation with TNF␣ resulted in time-and dose-dependent binding of two distinct protein DNA-complexes to the TF derived NF-B site, which were identified as NF-B (data not shown). Consistent with previous studies, binding of the slower migrating complex was characterized as NF-B(p65/ c-Rel) complex (11,18,62,63), while in the faster migrating band only NF-B(p65) seemed to participate in the binding complex (11) (data not shown). Binding of NF-B(p65/c-Rel) was already detected after 5 min, reached its maximum after 10 min, and decreased to basal levels after 1 h (data not 3 N. Mackman, personal communication.  1-7). 10 g of nuclear extract were included in each binding reaction. DNA-protein complexes were analyzed on 4% native polyacrylamide gels. EMSA detected AP-1 binding to the proximal AP-1 site (Table II) derived from the human TF promotor. The TNF␣-inducible AP-1 complex is indicated with an arrow. Specificity of binding was ascertained by competing with 160-fold molar excess of cold AP-1 consensus oligonucleotides (Table II) 1-6). Nuclear extracts were prepared, and 10 g of this extract were included in each binding reaction and analyzed as above. The TNF␣-inducible AP-1 complex (JunD/Fos) is indicated with an arrow. Specificity of binding was ascertained by competing with 160-fold molar excess of cold AP-1 consensus oligonucleotides (Table II)  shown). The p65-containing faster migrating complex was induced between 5 min and 3 h (data not shown). Taken together, these data indicate that activation of the human TF promotor follows a kinetic that includes first activation of NF-B(p65/c-Rel), followed by activation of JunD/Fos heterodimers that bind to the proximal AP-1 site. While NF-B(p65/c-Rel) activation declines, the distal AP-1 site exhibits maximal binding of AP-1/bZIP proteins, characterized to contain Jun and ATF proteins. This might explain, why TF transcription is still enhanced after 4 -6 h (Refs. 9 and 10 and data not shown), even TNF␣-induced NF-B(p65/c-Rel) activation has already dropped to base-line level.
TNF␣ Mediates Induction of Endothelial TF by a Concerted Action of the AP-1 and NF-B Sites-To confirm that different members of the AP-1/bZIP and NF-B family functionally act on the human TF promotor and contribute to TF induction by TNF␣, transient transfections of BAEC with TF promotor mutants (Fig. 8) were performed. When BAEC were stimulated with TNF␣, promotor activity compared to basal expression was highest, when the proximal AP-1 (A2) and the NF-B site were present (Fig. 8, a and b). Loss of the proximal AP-1 site or the NF-B site resulted in significantly decreased TF induction (Fig. 8, a and b). Loss of the non-canonical distal AP-1 site had a less prominent effect, as expected (Fig. 8, a and b). Promotor mutants with only the proximal AP-1 site (A2) cloned in front of the minimal promotor were still inducible by TNF␣, while the distal AP-1 site (A1) alone was unable to confer TF induction (Fig. 8, a and b). The proximal AP-1 significantly contributed to TF basal expression (Fig. 8a). Therefore mutants that contained the NF-B, but not the proximal AP-1 site, were more inducible by TNF␣ than mutants comprising the proximal AP-1 site (Fig. 8b); however, the overall TF expression of mutants without proximal AP-1 was lower (Fig. 8a). Highest TF expression was observed only when both AP-1 sites and the NF-B site were intact, consistent with participation of all three sites in regulation of TF transcription (Fig. 8a). The TNF␣-induced activity of the construct A1-A2-N, spanning the complete promotor, is higher (396 Luc units) than the added activity of the constructs containing each element alone (A1 ϭ 50, A2 ϭ 125, N ϭ 106 Luc units; total 281 Luc units). Thus, these elements act in concert on the human TF promotor in mediating TNF␣-induced TF transcription.
The concept that both AP-1 sites and the NF-B site act in concert was further supported by studies overexpressing specific inhibitors of NF-B(p65)(I-B) or AP-1/bZIP (mutated Jun). Overexpression of I-B reduced TNF␣-mediated TF induction as long as the NF-B site was present (Fig. 8c). Overexpression of mutated Jun reduced TF induction by TNF␣ as long as the proximal AP-1 site was present (Fig. 8d).
TNF␣-mediated TF Induction Is Dependent on AP-1/bZIP and NF-B Proteins in Vivo-No data are available to support the concept of AP-1/bZIP-and NF-B-mediated TF induction in vivo. Since in the human and the mouse TF promotor the same sequences are found for the distal and the proximal AP-1, as well as for the NF-B recognition motif, we used a mouse model to study the dependence of TNF␣-mediated TF expression of endothelial cells on Jun and NF-B in vivo. In this model, TNF␣ induces the expression of TF and fibrin formation on  (Table I; for detail see "Materials and Methods") for 36 h before TNF␣ (1 nM) was added for 6 h, where indicated. After harvest luciferase activity was determined in the cell lysates and normalized for transfection efficiency to the amount of ␤-galactosidase activity expressed by the control plasmid pSV-␤-Gal (Promega). Corrected values were expressed as relative luciferase units. The results represent the mean of at least three independent experiments Ϯ S.D. that were performed in triplicate. a, functional analysis of TF expression in unstimulated BAEC compared with TF expression in TNF␣ stimulated BAEC; the mean of six independent experiments Ϯ S.D. performed in triplicate is shown. b, the inducibility by TNF␣ relating to basal expression is shown. The level of basal expression is indicated with B. c, to directly demonstrate the role of NF-B(p65) in TNF␣-mediated TF induction, various TF promotor plasmids (Table I) were cotransfected with plasmids overexpressing CAT (ϭ mock) or the NF-B(p65) specific inhibitor I-B and cultivated for 36 h, before TNF␣ (1 nM) was added to the cells for 6 h. After harvest luciferase activity and transfection efficiency were determined as above. The data represent the mean of three different experiments performed in triplicate. d, to directly demonstrate the role of JunD/Fos (AP-1) in TNF␣-mediated TF induction, various TF promotor plasmids (Table I)  endothelial cells of the tumor vasculature (14,64). 3 h after intravenous injection of TNF␣ TF expression was induced, based on in situ hybridization (Fig. 9a) and immunohistochemistry (Fig. 9b). TF was expressed by subendothelial structures including tumor cells and also by endothelial cells. With respect to the topic of this study, we focused on endothelial cells. More than 150 vessels were evaluated in this part of the study.
When animals were treated by intravenous somatic gene transfer with mutated Jun 24 h prior to TNF␣ injection, a decrease in the endothelial response to TNF␣ was observed by in situ hybridization (Fig. 9a) and immunohistology (Fig. 9b) compared to vector-transfected animals. Thus, by blocking the interaction of Jun with other members of the bZIP family by somatic gene transfer with a plasmid overexpressing mutated Jun, endothelial TF induction could be partially reduced (Fig.  9, a and b). Similar data were obtained when a plasmid overexpressing I-B was used (Fig. 9, a and b). Mutated Jun and I-B both reduced the inducibility of TF in endothelial cells in this tumor model in vivo. The tumor model was further used to examine the functional effect of TF; when the fibrin/fibrinogen deposition in response to TNF␣ was studied in animals perfused with 30 -40 ml of PBS (see "Materials and Methods"), a reduction by mutated Jun and I-B was demonstrated in some, but not all vessels (Fig. 9c). Thus TF expression in vivo is under the control of AP-1/bZIP and NF-B-like proteins.
Successful transfection with mutated Jun or I-B was monitored in EMSA of tumor tissue (Fig. 10). Tumors derived from animals transfected with vector DNA prior to TNF␣ had a stronger AP-1 binding activity than tumors derived from animals transfected with mutated Jun (Fig. 10, top left). Consistently, tumors from I-B transfected animals demonstrated reduced NF-B binding activity at the TF derived NF-B site (Fig. 10, top right) compared to vector controls. In addition, Northern blot of mRNA, derived from whole tumors, showed decreased TF mRNA levels, when the animals had been transfected with mutated Jun or I-B prior to TNF␣ application (Fig.  10, bottom). However, the suppression obtained was only partial, since (i) members of the Jun and ATF family are less responsive to inhibition by overexpression of mutated Jun than c-Fos, (ii) the in vivo involvement of other transcription factors can not be excluded, and (iii) transfection did not reach all cells.
To give a picture of the overall efficiency of transfection, microbeads were used for measuring blood flow of the whole organ, avoiding potential artifacts due to selection of a single area in histological studies. When microbeads were injected into animals, a high number of beads was present in tumors of animals not treated with TNF␣ (Fig. 11). The number of beads reflecting tumor perfusion was clearly decreased after TNF␣ injection with previous somatic gene transfer with vector DNA (Fig. 11). This indicated that TNF␣ treatment resulted in loss of free blood flow, potentially due to TF-mediated microvascular thrombosis. Therefore this method adds to the histological study by providing data about the effect of I-B and mutated Jun on the whole organ. Gene transfer with I-B or mutated Jun partially reversed this effect of TNF␣ (Fig. 11). Hence blocking TF on the transcriptional level reduced not only TF induction by TNF␣, but also reduced the fibrin/fibrinogen deposition and restored free blood flow. DISCUSSION Tissue factor (TF) is a potent initiator of the coagulation cascade (1,2,4,65,66) and normally is not expressed by quiescent endothelial cells (1,3,6,11,12). In vitro data showed induction of TF synthesis in endothelial cells by inflammatory mediators such as endotoxin, phorbol esters, oxygen-free radicals, or cytokines (7)(8)(9)(10)(11)(67)(68)(69)(70). Recently members of the NF-B and the AP-1/bZIP family have been reported to be involved in the lipopolysaccharide-and cytokine-mediated TF induction in monocytes (15)(16)(17)(18)62) and porcine endothelial cells (11). It has been more difficult to show endothelial TF in vivo (3,67); however, recent studies demonstrate that in selected areas of the vascular bed activators of the host response or TNF␣ lead to the synthesis and expression of TF (12)(13)(14)71). This study addresses the molecular mechanisms that underlie the regulation of the human TF promotor in response to the proinflammatory cytokine TNF␣.
We used bovine aortic endothelial cells (BAEC), which exhibit lower basal AP-1 and NF-B activity than porcine (PAEC) or human (HUVEC) endothelial cells, and the human TF promotor. A striking difference between the human and the porcine TF promotor is seen at the proximal AP-1 site, which resembles a canonical high affinity site in the human TF promotor (15,16), while it is a low affinity non-canonical site in the porcine promotor, due to a G 3 A switch at position 4 of the AP-1 heptamer (11). Thus, the proximal AP-1 site of the human promotor is more prominently involved in the TNF␣-mediated up-regulation of TF than it is in the porcine promotor and, as a consequence, in addition to NF-B activation, AP-1 activation may be relevant for human disease.
In transient transfection studies, the highest TNF␣ inducibility was only observed when the NF-B(p65/c-Rel) site was present in the TF promotor plasmids. Enhanced NF-B(p65/c-Rel) binding to its TF-derived motif was detectable within 5 min after TNF␣ stimulation and rapidly down-regulated after 1 h (data not shown). This fast activation of NF-B(p65/c-Rel) reflects that TF mRNA can be rapidly induced in the absence of protein synthesis; therefore, TF has been classified as an immediate early gene (72). The dependence of TF induction by inflammatory mediators on NF-B(p65/c-Rel) activation may insure that this induction is transient, since it has been recently reported that increased NF-B levels lead to increased expression of the inhibitor I-B (73,74), followed by NF-B inactivation. This might prompt the activated endothelial cells to return to a quiescent state.
Therefore, the existence of increased TF mRNA levels in HUVEC and BAEC 4 -6 h after TNF␣ stimulation (9, 10) cannot be explained solely on the basis of NF-B activation and demands the involvement of other inducible transcription factors. Functional studies demonstrated that optimal TF induction by TNF␣ was also mediated by both AP-1 sites in the human TF promotor (Fig. 8). EMSA revealed that the TNF␣inducible complex bound to the canonical proximal AP-1 site of the human TF promotor consisted mainly of JunD/Fos (Fig. 7c); however, it cannot be excluded that other members of the Jun family are also involved. Highly vascularized organs (spleen, lung, intestine, ovary, and brain) express high levels of JunD (30). While the expression of c-Jun and Jun B is rapidly upregulated by various stimuli, JunD is only modestly induced by growth factors and phorbol esters (26,30,75,76). Transactivation by JunD homodimers is significant lower than by c-Jun homodimers (30). In cooperation with c-Fos, however, JunD has transactivation capacities similar to those of c-Jun (29,30). The results displayed here demonstrate that in cultured endothelial cells TNF␣ induces JunD/Fos heterodimers that recognize the proximal AP-1 site of the human TF promotor and thereby enhance TF transcription. In this respect the human and the porcine system differ significantly. Binding of JunD/ Fos-containing complexes to the proximal AP-1 site is already detected 30 min after TNF␣ stimulation. This rapid response excludes newly synthesized JunD or Fos and indicates the rapid activation of preexisting proteins. This availability of JunD/Fos heterodimers therefore is a limiting factor. Consistently, the canonical proximal AP-1 alone was not able to confer high TNF␣-mediated induction in transient transfection experiments and required the presence of the NF-B(p65/c-Rel) site (11,15,16,18) for optimal TF expression. These data imply that the disposal of JunD/Fos heterodimers is not sufficient for maximal induction by TNF␣ and need to recruit NF-B (p65/c-Rel) nuclear binding activity. Since NF-B(p65/c-Rel) translocation into the nucleus precedes JunD/Fos activation only by 20 min, one might speculate that binding of one transcription factor facilitates binding of the other.
The proximal high affinity AP-1 site of the human TF promotor, which is missing in the porcine TF promotor, is of particular importance with respect to therapeutic interventions. A great variety of antioxidative agents has been reported to suppress activation of NF-B in vitro and in vivo (77) and therefore might potentially be used for reducing TF activity under certain pathophysiological conditions. However, recent studies elucidated that changes in the cellular redox system by radical scavengers suppress very fast NF-B, but at the same time induce time-dependent AP-1 activation (Jun/Fos) (20,21). Antioxidative conditions strongly induce c-Fos, which can form reactive heterodimers with preexisting Jun homodimers (20,21). As pointed out before, tissues with high endothelial portions contain constitutively high amounts of JunD homodimers (30) and are therefore primed to generate large amounts of JunD/Fos heterodimers under antioxidative therapy.
To define the role of the non-canonical distal AP-1 site in human TF regulation is more difficult. This site is a low affinity site for AP-1 binding and resembles the two non-canonical AP-1 sites of the porcine TF promotor (11). In accordance with these data, several independent approaches demonstrated that Jun homodimers, but not Jun/Fos heterodimers, bind to this site (Figs. 3, 4, and 6). Differences in the structure of the DNA binding domains for Jun homodimers and Jun/Fos heterodimers have been described (78); therefore, it seems likely that a G 3 A switch at position 4 of this site facilitates Jun binding and excludes significant Jun/Fos binding. Furthermore, specific properties of the regions outside the defined AP-1 binding sites might be responsible for preference in binding of the various homo-and heterodimer complexes (79). EMSA demonstrated (Fig. 6, a-c) that TNF␣ also induced protein complexes that were different from Jun homodimers. These complexes have been characterized to contain Jun and ATF family proteins (Fig. 6, a-c). In contrast, Moll et al. (11) recently reported constitutive binding of c-Jun, JunD, and possibly Fra2 complexes to the porcine TF promotor-derived noncanonical AP-1 sites. Since basal AP-1 binding activity is low in BAEC compared to PAEC), this might explain why the study presented here detected TNF␣-inducible binding at the noncanonical distal AP-1 site of the human TF promotor. Consistent with our observations, Donovan-Peluso and co-workers mentioned that in THP-1 cells large differences between the distal and the proximal AP-1 site were detected in EMSA, which indicate the involvement of different heterodimers (17). This finding differs from previous observations in HUVEC, where Jun homodimer and Jun/Fos heterodimer binding occurs at the distal and the proximal AP-1 site (47,63). This might be due to (i) a greater availability of Jun homodimers, (ii) to a different composition of the complexes induced, or (iii) to species differences in HUVEC versus BAEC. The low affinity distal AP-1 site of the human TF promotor only marginally participates in TNF␣-induced TF expression, consistent with the data described for the two low affinity AP-1 sites in the porcine TF promotor (11). However, the non-canonical AP-1 site significantly supports NF-B-mediated TF induction, even when the proximal AP-1 is deleted (Fig. 8). Since recently a cooperative action of ATF proteins and NF-B family members has been demonstrated (80), one might speculate that proteins bound to the distal AP-1 site support and facilitate NF-B activity. Therefore a set of different transcription factors has to be activated at the same time before endothelial TF is successfully induced.
The in vivo data presented ( Fig. 9 -11) support this concept. Intravenous somatic gene transfer with plasmids overexpressing I-B or mutated Jun reduce TF induction in vascular endothelial cells of the tumor. They also decrease deposition of fibrin/fibrinogen. The antibody used does not discriminate between fibrin and fibrinogen. Therefore, the animals were perfused with 30 -40 ml of PBS (see "Materials and Methods") prior to harvest of the organs to remove non-clotted material. The reactive material represents at least in part fibrin, since we observed striking differences in fibrin/fibrinogen deposition between the different animal groups corresponding to the perfusion studies with microbeads (the later ones giving a better view of the overall efficiency of I-B and mutated Jun). How-ever, in these experiments cells other than endothelial cells may be affected. Nevertheless, the in situ hybridization and immunohistochemical studies (Fig. 9) showed that endothelial cell expression of TF is under control of NF-B and AP-1 in the animal model used. The incomplete suppression of TF and fibrin/fibrinogen deposition can be explained (i) by the expected low to moderate transfection efficiency, (ii) by local differences in endothelial cells (capillaries still growing versus already grown vessels, dividing vessels versus non-dividing endothelial cells), (iii) by the involvement of other transcription factors than AP-1 and NF-B, and (iv) other EC genes influenced by cytokines. Hence the TNF␣-mediated activation of endothelial TF transcription occurs in vitro and in vivo by members of the NF-B and AP-1/bZIP family.