Identification of a cis-acting sequence in the human plasminogen activator inhibitor type-1 gene that mediates transforming growth factor-beta1 responsiveness in endothelium in vivo.

The mechanism of regulation of the plasminogen activator inhibitor type-1 (PAI-1) gene by transforming growth factor-β1 (TGF-β1) was studied in vitro and in vivo in endothelial cells. We constructed adenovirus vectors containing PAI-1 5′-flanking sequences driving expression of a β-galactosidase (β-gal) reporter gene. Cultured bovine endothelial cells were transduced with the vectors and treated with TGF-β1. β-Gal expression was up-regulated 10-20-fold by TGF-β1 when vectors contained 799-base pair (bp) of 5′-flanking sequence, but only minimally (2-3-fold) from a vector containing only 82-bp of 5′ PAI-1 flanking sequence. TGF-β1 up-regulated β-gal expression at the mRNA level, congruently with TGF-β1 up-regulation of expression of the endogenous PAI-1 gene. The constructs were transduced into intact rat carotid endothelium, and TGF-β1 was injected systemically. In vivo, TGF-β1 up-regulated endothelium-specific expression of β-gal 3-fold (p < 0.03) from a vector containing the 799-bp sequence, but did not alter expression from a vector containing the 82-bp sequence. The sequence between −799 and −82 mediates up-regulation of reporter gene expression by TGF-β1 in endothelial cells in vitro and in vivo. This general method permits the elucidation of mechanisms of gene regulation by physiologic stimuli delivered to the endothelium of intact animals.

Plasminogen activator inhibitor type-1 (PAI-1), 1 the major physiological inhibitor of tissue-type plasminogen activator and urokinase plasminogen activator (1,2), is thought to participate in the regulation of several plasmin-dependent processes including fibrinolysis, trophoblast implantation, cell migration, ovulation, angiogenesis, and wound repair (3). Among these potential biological roles of PAI-1, much attention has been devoted to its role in the arterial wall. In human tissues, PAI-1 is expressed by both vascular smooth muscle and endothelial cells (EC) (4). PAI-1 is up-regulated in vivo in association with atherosclerosis (5), endotoxemia (6), thrombosis (7), and vascular injury (8). Definition of the molecular mechanisms by which PAI-1 expression is up-regulated in these pathological states may lead to a better understanding of vascular pathobiology and could also lead to the development of novel approaches to the prevention of thrombosis and atherosclerosis.
Despite the likely importance of PAI-1 regulation by TGF-␤1 in EC in vivo, virtually all of the data on the mechanisms of regulation of PAI-1 gene expression have been produced in transformed hepatocyte and fibroblast lines in vitro (17)(18)(19)(20)(21)25). While these studies are informative, the extent to which one may extrapolate the results of gene regulation experiments across cell types and from in vitro to in vivo is unclear. For example, the mechanisms of regulation of vascular cell adhesion molecule-1 in cultured skeletal muscle and epithelial cells differ from those operative in cultured EC (26,27). Insulin and proinsulin up-regulate PAI-1 production in cultured HepG2 cells but not in cultured EC (28), and the effects of TGF-␤1 on PAI-1 mRNA half-life also differ between HepG2 cells and EC (18). Moreover, even if gene regulation studies were performed solely with cultured EC, patterns of gene expression and regulation in EC (both baseline transcript levels (29) as well as regulation of transcription in response to added cytokines (30)) are quite variable in vitro, leading to the generation of conflicting experimental results in different laboratories. Finally, PAI-1 protein is nearly undetectable in normal unstimulated endothelium in vivo (31); yet when placed in vitro, EC synthesize and secrete large amounts of PAI-1 (32). For all of these reasons, to understand the physiological mechanisms of regulation of PAI-1 expression in EC, it is most appropriate to perform experiments in vivo in endothelium. Until recently, however, this has not been feasible.
We recently reported an animal model of in vivo endothelium-specific gene transfer (33) and speculated on the utility of this model to permit the definition of molecular mechanisms of in vivo EC gene regulation. Here we report the use of this animal model to define a functional cis-acting sequence in the human PAI-1 promoter that mediates up-regulation of EC gene expression in response to TGF-␤1 administration in an intact animal.

EXPERIMENTAL PROCEDURES
Construction of Plasmids-We obtained the following plasmids: pSP72 (Promega, Madison, WI); pAdRSV4 and pAdBglII (34) (Dr. Blake Roessler and Dr. Beverly Davidson, University of Michigan); pLZ11 (35) (Dr. Joshua Sanes, Washington University); p3P, a derivative of pUC 13 containing a fragment of the human PAI-1 promoter extending from 3.0 kb 5Ј of the transcription start site to the EcoRI site in the 5Јuntranslated sequence (12) (Dr. Thomas Quertermous, Vanderbilt University); and pPK9A (36) (Dr. Anita Roberts, National Cancer Institute). To generate recombinant replication-defective adenovirus (Ad) vectors ( Fig. 1), five plasmids (pAdPAI800␤-Gal, pAdIAP800␤-Gal, pAdPAI82␤-Gal, pAdRSVnLacZ, and pAdRSVTGF-␤1) were constructed. Each of these plasmids contained elements of the 5Ј end (0 -1 and 9.2-16.1 map units) of the Ad 5 genome (37), with individual expression cassettes inserted at the site of the E1 deletion. pAd-PAI800␤-Gal, pAdIAP800␤-Gal, and pAdPAI82␤-Gal were constructed as follows. A BglII fragment containing the Rous sarcoma virus (RSV) long terminal repeat (LTR) promoter and SV40 polyadenylation sequences (termed "SVpA") was excised from pAdRSV4 and ligated into BamHI-digested pSP72. The resulting plasmid, designated pSP72BglII, was digested with SmaI and BamHI then ligated to a blunt-ended, BglII-XbaI fragment of pLZ11. This BglII-XbaI fragment of pLZ11 contained a nuclear-targeted Escherichia coli ␤-gal gene (nLacZ). The plasmid product of the ligation of the BglII-XbaI fragment to pSP72BglII, termed pSP72␤-Gal, contained nLacZ fused to SVpA, with cloning sites upstream for insertion of promoter fragments. To obtain fragments of the human PAI-1 promoter, we first digested p3P with HindIII and EcoRI, releasing an 874-bp (all base pair enumeration includes unpaired overhangs) sequence of the human PAI gene, including 799 bp upstream of the transcription start site and 75 bp of the first (untranslated) exon. This 874-bp fragment was blunt-ended and ligated into PvuII-cut pSP72, yielding pSP72PP, and restoring the 3Ј EcoRI site. PAI-1 promoter fragments were removed from pSP72PP either as: 1) a 887-bp XhoI-EcoRI fragment containing the original 874-bp promoter ϩ exon 1 fragment along with 13 bp of 5Ј-flanking polylinker sequence ("PAI800," containing 799 bp of PAI-1 sequence 5Ј to the PAI-1 transcription start site) or, 2) as a 157-bp BsrI-EcoRI fragment, truncated by 717 bp of 5Ј PAI-1 promoter sequence, but identical to the longer XhoI-EcoRI promoter fragment at the 3Ј end ("PAI82," containing only 82 bp of PAI-1 sequence 5Ј to the PAI-1 transcription start). PAI800 and PAI82 were blunt-ended, and ligated into the StuI site of pSP72␤-Gal. The resulting plasmids, pPAI800␤-Gal and pPAI82␤-Gal, contained expression cassettes consisting of either PAI800 or PAI82 promoter sequences driving expression of nLacZ-SVpA. pPAI800␤-Gal and pPAI82␤-Gal were then digested with BglII and XhoI, sites that flanked the PAI-␤gal-SVpA expression cassettes. These BglII-XhoI fragments were blunt-ended and ligated into a BglII-digested bluntended pAdBglII. Both orientations of the PAI800 expression cassette as well as the forward (with respect to the Ad genomic sequences) orientation of the PAI82 expression cassette were propagated as pAd-PAI800␤-Gal, pAdIAP800␤-Gal, and pAdPAI82␤-Gal, respectively. Both pAdPAI800␤-Gal and pAdIAP800␤-Gal were used to construct vectors because use of an exogenous expression cassette that was transcribed toward the Ad E1A enhancer resulted in elimination of baseline expression of the cassette. 2 Our in vitro experiments, however, demonstrated similar baseline levels of expression for both orientations of the PAI800 expression cassette (data not shown), therefore the pAdIAP800␤-Gal-derived vector was not used in in vivo experiments.
The plasmid pAdRSVnLacZ was constructed as follows. pSP72␤-Gal was digested with BglII and BamHI, releasing the nLacZ sequences. This fragment was ligated into the BamHI site of pAdRSV4, creating an expression cassette consisting of the RSV LTR promoter driving nLacZ-SVpA, oriented 5Ј 3 3Ј with respect to the Ad5 sequences of pAdRSV4.
The plasmid pAdRSVTGF-␤1 was constructed by digestion of pPK9A with BglII, releasing a 1.3-kb fragment containing an active form of porcine TGF-␤1. This BglII fragment was ligated into the BamHI site of pAdRSV4, creating an expression cassette consisting of the RSV LTR promoter driving TGF-␤1-SVpA, also oriented 5Ј 3 3Ј with respect to the Ad5 sequences of pAdRSV4. The fidelity of all plasmid constructions was confirmed by restriction mapping.
Recombinant Ad Production-Adenoviral vectors were generated by cotransfection of linearized plasmids pAdPAI800␤-Gal, pAdIAP800␤-Gal, pAdPAI82␤-Gal, pAdRSVnLacZ, or pAdRSVTGF-␤1 with the 35-kb ClaI fragment of Ad-dl327 (an E3-deleted Ad (38)). The resulting replication-defective Ad vectors are shown in Fig. 1. To screen for recombinant Ad vectors AdRSV␤-Gal, AdPAI800␤-Gal, AdIAP800␤-Gal, and AdPAI82␤-Gal, freeze-thaw lysates made from plaques were tested for ability to transfer ␤-gal expression to 293 cells. For screening putative AdRSVTGF-␤1 recombinants, lysates were tested for the ability to transfer the capacity to synthesize and secrete TGF-␤1 to 293 cells. TGF-␤1 was detected by assaying conditioned media with the Predicta TGF-␤1 ELISA kit (Genzyme, Cambridge, MA), without prior acid activation. Virus harvested from recombinant plaques was propagated and titered on 293 cells. An additional recombinant Ad, AdHV1.1, was used as a control in certain experiments. AdHV1.1 has a structure similar to the other vectors, containing a hirudin cDNA fused to SVpA, driven by the RSV LTR promoter (39). Titers of purified stocks of all viruses ranged from 5 ϫ 10 10 to 2 ϫ 10 11 plaque-forming units (pfu)/ml.
For experiments analyzing the time course of recombinant gene expression, CPAEC were infected with Ad vectors, then were incubated with growth medium containing 10 ng/ml TGF-␤1 or vehicle only. Endothelial cells were harvested either immediately (t ϭ 0) or at various time points after infection. Cell extracts were assayed for ␤-gal activity, as above.
In certain experiments, CPAEC transduced with one of the four ␤-gal expressing vectors were subsequently infected with a second virus, either AdRSVTGF-␤1 or AdHV1.1. Immediately following infection with a ␤-gal vector, EC were rinsed once with OPTI-MEM, incubated with AdRSVTGF-␤1 or AdHV1.1 for 90 min, then fed with growth medium. At defined time points, the EC were harvested and assayed for ␤-gal activity and protein. The conditioned media were also collected and assayed for TGF-␤1 antigen by ELISA. As a specificity control, in certain experiments a rabbit polyclonal neutralizing antibody to TGF-␤1 (100 g/ml; Promega, Madison, WI) was added at the time of growth medium addition.
The primary purpose of this study was to identify cis-acting elements in the PAI-1 promoter that are TGF-␤1-responsive in vivo, not to perform promoter deletion or reconstitution analyses to identify and quantitate cis-acting elements that are functional under baseline conditions. For this reason, in both in vitro and in vivo gene transfer studies, reporter gene expression data are normalized to cell or tissue protein, and comparisons are generally made between expression levels in cells receiving the same vector in the presence or absence of TGF-␤1 stimulation. We do not normalize data based on expression of a second reporter gene, as is usually done with plasmid-mediated transfection. There are two reasons for this. First, the in vitro transduction efficiencies are essentially 100%, and therefore do not require controls specif-ically for variations in efficiency. Second, in preliminary in vivo experiments, we transduced rat endothelium with a mixture of two adenovirus vectors: AdRSV␤-Gal and AdSV40Luc (33), in which the luciferase gene is driven by an SV-40 promoter. When normalization of the expression of either vector was done to total protein, the data were reproducible between animals. When normalization of the expression of either vector was attempted by using expression from the second vector as a denominator, the data became more variable (data not shown). This observation could potentially result from competition between the viruses for binding or entry into cells, a phenomenon not found with plasmid co-transfection. Because of this observation, and due to the above stated primary purpose of the study, we normalized expression data to protein content only, both in vitro and in vivo.
Northern Analysis-CPAEC were grown to 50% confluence in 100-mm plates and infected at a m.o.i. ϭ 250 with AdPAI800, AdIAP800, AdPAI82, or AdRSV␤-Gal for 90 min followed by replacement with growth medium. After 24 h, recombinant TGF-␤1 protein was added to growth medium for 1 h. Cells were then incubated in fresh growth medium for an additional 5 h, and total RNA was harvested using RNAzol B (Tel-Test, Inc., Friendswood, TX) and quantitated by absorbance at 260 nm. RNA (20 -30 g/lane) was electrophoresed into 1.2% formaldehyde-agarose gels, and blotted onto Nitroplus 2000 (0.22 m, Micron Separation, Westboro, MA). Blots were hybridized sequentially with cDNA probes to rat PAI-1, E. coli ␤-gal, and human glyceraldehyde phosphate dehydrogenase (GAPDH). Blots were hybridized in QuikHyb Hybridization Solution (Stratagene, La Jolla, CA) at 65°C overnight, then washed in 2 ϫ standard saline citrate, 0.1% SDS at room temperature followed by 30 min wash at 60°C in 0.1 ϫ standard saline citrate and 0.1% SDS, and autoradiographed at Ϫ70°C. cDNA probes were: PAI-1, a 2.4-kb HindIII fragment of plasmid pSKPAI53, which encodes rat PAI-1 (41); ␤-gal, a 3.0-kb StuI-XbaI fragment of plasmid pLZ11 (35); GAPDH, a 1.4-kb BamHI-HindIII fragment of GAPDH (a gift from Dr. Stephan Karlsson, NINDS). All probes were 32 P-labeled with the random primer method to specific activity of approximately 1 ϫ 10 9 cpm/g. An initial blot was subjected to autoradiography to obtain a figure for presentation. The experiment was then repeated six times, followed each time by Northern analysis and visualization of transcripts on blots with a FUJIX Bas1000 Bioimaging Analyzer. The radioactivity present in each band was quantitated with the aid of MacBas software. The extent of up-regulation of PAI-1 and ␤-gal transcripts following addition of 10 ng/ml TGF-␤1 (i.e. when compared to the intensity of bands generated from cells exposed to 0 ng/ml TGF-␤1) was determined following normalization of the intensity of each band to the intensity of the GAPDH signal in each lane. Specifically, for each of the four groups of adenovirus-infected cells we calculated PAI-1/GAPDH (at 10 ng/ml TGF-␤1) Ϭ PAI-1/GAPDH (at 0 ng/ml TGF-␤1) to obtain the up-regulation of PAI-1 mRNA. Similarly, we calculated ␤-gal/GAPDH (10 ng/ml TGF-␤1) Ϭ ␤-gal/GAPDH (at 0 ng/ml TGF-␤1) to obtain the up-regulation of ␤-gal mRNA.
In Vivo Transduction of Rat Carotid Arteries with Adenoviral Vectors-All animal procedures were approved by the Animal Care and Use Committee of the National Heart, Lung, and Blood Institute. Adult male Sprague-Dawley rats (Taconic Farms, Germantown, NY) weighing 270 -350 g were used. Anesthesia, endothelium-specific gene transfer, and postoperative care were performed as described previously (33). Briefly, the left carotid system was surgically exposed, proximal and distal control obtained, and an arteriotomy made on the external carotid artery. A 1-cm segment of common carotid artery was isolated and flushed clear of blood with approximately 1 ml of M-199 medium (Biofluids), infused through a 24-gauge polytetrafluoroethylene catheter. Aliquots of Ad were thawed immediately prior to use and diluted as required with "virus diluent" consisting of M-199 medium with 1 mg/ml rat serum albumin (Sigma). For each rat, 50 l of either Ad solution or virus diluent (as a control) were infused into the isolated carotid segment and allowed to dwell for 20 min. The solution was then withdrawn into the catheter, the catheter removed, the external carotid artery ligated, and blood flow re-established through the common and internal carotid arteries. This in vivo gene transfer technique results in recombinant gene expression that is highly (95%) specific for the luminal arterial endothelium, with approximately 35% of total luminal EC transduced along a 1-cm length of arterial surface (33).
In vivo gene transfer was performed with the AdPAI800␤-Gal, Ad-PAI82␤-Gal, and AdRSV␤-Gal vectors. All vectors were infused at a concentration of 3 ϫ 10 10 pfu/ml. The AdIAP800␤-Gal vector was not used in vivo because: 1) the structure of AdIAP800-␤-Gal, with the direction of transcription of the ␤-gal expression cassette toward the E1A enhancer, was less symmetric than that of AdPAI800␤-Gal with respect to the control vectors AdPAI82␤-Gal and AdRSV␤-Gal (Fig. 1) and 2) the anticipated advantage of lower baseline expression levels of ␤-gal (see above) with AdIAP800␤-Gal was not supported by our in vitro data.
Regulation of PAI-1 Promoter in Vivo by Systemic TGF-␤1-The inducibility of Ad constructs by TGF-␤1 was tested by intravenous injection of TGF-␤1 72 h post-gene transfer. Rats were sedated (as above), and the right external jugular vein of each animal was surgically exposed. The vein was cannulated with a 27-gauge needle mounted on a tuberculin syringe, and 300 l of either TGF-␤1 (100 or 200 g/kg of porcine TGF-␤1; purified from platelets and acid-activated, provided by Dr. Anita Roberts, National Cancer Institute) or control solution (4 mM HCl diluted with 1% bovine serum albumin in phosphate-buffered saline) were administered over 1 min. Hemostasis was attained with gentle pressure at the puncture site, and the wound was closed.
Fixation, Removal, and Explantation of Vessel Segments-Animals were killed 24, 48, or 72 h post-injection of TGF-␤1 or control (4, 5, or 6 days post-carotid gene transfer) by overdosing with pentobarbital, 600 mg/kg intraperitoneally. At the time of death, none of the animals had clinical evidence of wound infection, and all common carotid arteries were patent. For experiments involving histologic examination of vessels, perfusion fixation was carried out in situ; alternatively arteries were removed and frozen at Ϫ80°C for subsequent extraction and measurement of recombinant ␤-gal activity and antigen (42). Both assays were used because of a concern that reliance only on ␤-gal activity assays might be inappropriate due to endogenous activity in rat tissues. The ELISA (5Ј 3 3Ј Inc., Boulder, CO), while less sensitive than the activity assay, is highly specific for the E. coli versus mammalian enzyme (data not shown).
Protein Determinations-Protein concentrations of cell and tissue extracts were determined with the BCA assay (Pierce), using bovine serum albumin as a standard.
Statistical Analysis-One way ANOVA was used to test for the presence of differences in the behavior of individual vectors within the group of four vectors tested. Not all possible pairwise comparisons were made. Comparisons between any two vectors (or comparisons between ␤-Gal expression in the presence or absence of TGF-␤1) were based only on a priori hypotheses, and were made with Student's unpaired t-test, using a two-tailed ␣ of 0.05. Statistical analyses were performed on a microcomputer with the aid of the SigmaStat program (Jandel Scientific, San Rafael, CA).
Histochemical staining of EC transduced with each of the four vectors and treated with either vehicle or 10 ng/ml TGF-␤1 confirmed the up-regulation of ␤-gal expression only in Ad-PAI800␤-Galand AdIAP800␤-Gal-transduced EC (Fig. 3). This staining provided a visual correlate for the more quantitative data in Fig. 2. As with the primary data obtained with protein extracts (see above), higher baseline levels of expression from the AdPAI82␤-Gal and AdRSV␤-Gal vectors were evident (compare Fig. 3 E and F versus A and C). While these elevated baseline values could potentially result from variability in titering of the different viruses, they were reproducible in experiments performed with different lots of virus that were titered independently.
Two central premises underlie experiments in which fragments of genomic DNA are fused to reporter genes for the purpose of drawing inferences regarding the regulation of expression from intact genomic DNA: 1) increased reporter pro-tein levels are reflective of the abundance of corresponding mRNA and 2) the behavior of the largest promoter fragment (into which deletions and mutations are introduced) fused to the reporter gene accurately models the behavior of the endogenous promoter fused to the endogenous gene, in this case PAI-1. To verify these two premises for the PAI-1 promoter-␤gal constructs used in our study, we performed Northern analysis of CPAEC that had been infected with AdPAI800␤-Gal, AdIAP800␤-Gal, AdPAI82␤-Gal, or AdRSV␤-Gal and treated with 0, 1, or 10 ng/ml TGF-␤1 (Fig. 5). Endogenous PAI-1 mRNA was up-regulated by TGF-␤1 (at 10 ng/ml) in all cells to a similar extent (2-5-fold; p ϭ 0.35). Specific mRNA encoding ␤-gal was up-regulated by TGF-␤1 (at 10 ng/ml) in parallel with endogenous PAI-1 mRNA when expression of ␤-gal was driven by the 799-bp PAI-1 promoter fragment in either orientation, but not when expression of ␤-gal was driven by the 82-bp PAI-1 promoter (5.2-fold for the IAP800 promoter; 2.0-fold for PAI800 promoter versus 1.3-fold for PAI82 promoter; p ϭ 0.011 and 0.024 for IAP800 and PAI800 versus the 82-bp fragment, respectively). These mRNA data are in qualitative agreement with the protein data obtained with the constructs (Fig. 2). Use of the PAI800␤-Gal Ad vector and deletion mutants thereof, with measurement of ␤-gal activity, appears to represent a valid approach with which to study mechanisms of regulation of the endogenous PAI-1 gene.
Co-infection of CPAEC with a TGF-␤1-expressing Ad Vector-As a prelude to in vivo studies examining the ability of TGF-␤1 to up-regulate expression from PAI-1 promoter fragments transduced into intact endothelium, we tested whether co-infection of CPAEC with a TGF-␤1-expressing Ad vector could serve as a source of TGF-␤1 protein to up-regulate PAI-1 promoter activity in vitro. The CPAEC infected with Ad-PAI800␤-Gal, AdIAP800␤-Gal, AdPAI82␤-Gal, or AdRSV␤-Gal were exposed to AdRSVTGF-␤1 at m.o.i. of 25 or 250. As a control for nonspecific effects of infection with AdRSVTGF-␤1, parallel wells of transduced CPAEC were infected with the hirudin-expressing vector AdHV1.1, also at m.o.i. of 25 or 250. Delivery of TGF-␤1 by infection with AdRSVTGF-␤1 up-regulated ␤-gal expression in a manner both qualitatively and quantitatively similar to that found with addition of TGF-␤1 protein (Fig. 6A; compare to Fig. 2). A time course study of up-regulation of ␤-gal expression by AdRSVTGF-␤1 also produced results similar to those obtained by addition of recombinant TGF-␤1 protein ( Fig. 6B; compare to Fig. 4). The slower kinetics of induction of ␤-gal expression in these co-infection experiments is likely due to the gradual accumulation of TGF-␤1 in the medium (Fig. 6B, inset), compared to the immediately high levels of TGF-␤1 that are obtained following addition of recombinant protein to the culture dish.
Regulation of the Human PAI-1 Promoter in Endothelium in Vivo-To test the ability of fragments of the human PAI-1 promoter to respond to TGF-␤1 stimulation in vivo, we introduced AdPAI800␤-Gal, AdPAI82␤-Gal, or AdRSV␤-Gal into the endothelium of rat common carotid arteries. In initial studies we attempted to deliver TGF-␤1 to these rats by means of in vivo expression from the AdRSVTGF-␤1 virus. First we attempted co-local delivery: AdPAI800␤-Gal was infused for 15 min followed by infusion of either AdRSVTGF-␤1 or a control adenovirus, AdHV1.1 (3 ϫ 10 10 pfu/ml of either virus, also for 15 min). No up-regulation of ␤-gal expression was observed, although the interpretation of this negative result is clouded by our previous observations on potential interference between co-infecting viruses (see "Experimental Procedures"). We next attempted to achieve systemically elevated levels of TGF-␤1 protein by intravenous injection of a bolus of up to 8 ϫ 10 9 pfu of AdRSVTGF-␤1 (injection of higher amounts of AdRSV-TGF-␤1 was fatal). Although injection of 5 ϫ 10 9 pfu of AdRSV␤-Gal yielded the expected evidence of significant recombinant gene expression in hepatocytes (numerous blue hepatocyte nuclei; not shown), in no rat injected with up to 8 ϫ 10 9 pfu/ml AdRSVTGF-␤1 did we detect an elevated plasma TGF-␤1 level. We therefore attempted up-regulation of PAI-1 promoter activity by systemic injection of recombinant TGF-␤1 protein, a technique shown to up-regulate endogenous PAI-1 expression in mice (43).
In preliminary time course experiments, carotid arteries were transduced with AdPAI800␤-Gal. Three days later, rats were injected intravenously with 100 or 200 g/kg TGF-␤1, or vehicle only. Arteries were harvested 24, 48, or 72 h later. ␤-Galactosidase expression was up-regulated, as determined by X-gal staining followed both by en face viewing and by counting transduced EC in histologic sections (data not shown). Up-regulation appeared maximal 48 h after TGF-␤1 injection and was not further increased either at a dose of 200 g/kg TGF-␤1 or in arteries harvested at 24 and 72 h. Thus, we carried out a series of definitive in vivo experiments with the optimized protocol of injection of 100 g/kg TGF-␤1 and vessel harvest 48 h later.
Additional carotid arteries transduced with AdPAI800␤-Gal, AdPAI82␤-Gal, or AdRSV␤-Gal and taken from rats exposed to systemic injection of either TGF-␤1 or vehicle were harvested 48 h later and stained with X-gal. Arteries transduced with AdPAI800␤-Gal were viewed both en face and by histologic sectioning (Fig. 8). There was a visible increase in ␤-gal expression in the endothelium of TGF-␤1 exposed arteries (Fig. 8, B and D) when compared to that present in arteries taken from animals injected with vehicle only (Fig. 8, A and C). No increase in ␤-gal expression in response to TGF-␤1 was seen following X-gal staining of arteries transduced with either AdPAI82␤-Gal or AdRSV␤-Gal (not shown). Thus, an in situ assay of ␤-gal expression confirmed the results of tissue extract assays and provided additional data that localized the up-regulation of ␤-gal expression specifically to the endothelium. Taken together, the in vivo data identify a TGF-␤1-responsive element of the human PAI-1 promoter within a defined sequence of 717 bp (between HindIII and BsrI sites), and indicate that in the endothelium of an intact animal there are no DNA elements strongly responsive to TGF-␤1 present in the 3Ј 82 bp of the promoter sequence. DISCUSSION We used Ad-mediated gene transfer to identify a sequence in the human PAI-1 gene that is responsive to TGF-␤1 in endothelium in vivo. Our major findings were: 1) 799 bp of the human PAI-1 promoter, along with 75 bp of 5Ј-untranslated sequence, confer TGF-␤1-responsiveness on a reporter gene in arterial EC in vitro and carotid endothelium in vivo; the presence of E1A enhancer sequences adjacent to the promoter does not prevent TGF-␤1-responsiveness. 2) Transcripts from the 799-bp PAI-1 promoter fragment are up-regulated in parallel with transcripts of the PAI-1 gene in cultured EC, validating the use of these Ad-encoded constructs to study the regulation of PAI-1. 3) Recombinant TGF-␤1, delivered by co-infection with a TGF-␤1-expressing Ad vector in vitro, duplicates the up-regulation of PAI-1 promoter activity achieved with purified TGF-␤1 protein. 4) The proximal 82 bp of the human PAI-1 promoter are minimally responsive to TGF-␤1 in EC in vitro, and are unresponsive to TGF-␤1 in rat endothelium in vivo.
Two previous studies examined the regulation of PAI-1 expression in endothelium. Sawdey et al. reported that lipopolysaccharide, tumor necrosis factor-␣, or TGF-␤1 up-regulated PAI-1 transcription in cultured bovine EC; however, no functional analysis of the PAI-1 promoter was performed (44). Van Zonneveld et al. (12), using plasmid transfection into both cultured rat FTO2B hepatoma cells and bovine EC, analyzed the human PAI-1 promoter. In FTO2B cells, dexamethasoneresponsive sequences between Ϫ800 and ϩ75 mediated a 40fold induction of reporter gene expression. In bovine EC, PAI-1 promoter fragments of 187 bp and 1.5 kb were both functional, but no data were presented on their responsiveness to dexamethasone. Of note, we treated CPAEC transduced with Ad-PAI800␤-Gal with dexamethasone and (in contrast both to our results with TGF-␤1 and to those of van Zonneveld in FTO2B cells) found no up-regulation of ␤-gal expression. Thus, glucocorticoid regulation of the PAI-1 promoter may be cell typespecific, as reported elsewhere for PAI-1 regulation by insulin (28). These results underscore the importance of performing The CPAEC transduced with the indicated vectors were treated with either vehicle (TGF-␤1 ϭ 0 ng/ml) or TGF-␤1 protein (1 or 10 ng/ml) for 1 h. Five h later, total RNA was isolated, electrophoresed, and blotted. The blot was probed sequentially with ␤-gal, PAI-1, and GAPDH cDNA probes. The quantitative and statistical analyses presented under "Results" were performed during a separate series of experiments in which the PAI-1 and ␤-gal transcripts at 0 ng/ml TGF-␤ were not as low as in the experiment portrayed here (see "Results"). Thus, the calculated up-regulation of PAI-1 and ␤-gal transcripts was less than is apparent here. Based on ethidium bromide staining of the gel, detection of similar sized mRNA on this blot with other probes, and multiple independent repetitions of this experiment, the relative absence of GAPDH signal in lanes 11 and 12 is a technical artifact.
promoter analysis in the cell type of interest, in this case EC. To our knowledge, the present study is the first to provide a mechanistic analysis of the regulation of the PAI-1 promoter in EC either in vitro or in vivo.
The 82-bp PAI-1 promoter fragment had a consistently higher constitutive activity (mean 6-fold) than the 799-bp fragment in the in vitro transduction assays. In the in vivo experiments, this increase was still present, but was smaller (2-fold). We are cautious in our interpretation of these data because the experiments were not designed prospectively to test the relative strengths of promoter sequences. Nevertheless, this was a consistent finding and is similar to the 2-fold increase in activity of a 187-bp versus a 1.5-kb PAI-1 promoter fragment found in cultured bovine EC (12). A negative regulatory element that is functional in EC may be present upstream of Ϫ82 bp. Further prospective quantitative studies are required to address this conclusively.
In addition to defining TGF-␤1-responsive sequences in the PAI-1 gene that function in endothelium, one of our primary goals was to establish a quantitative system for studying gene regulation in vivo in endothelium. For these initial studies, we used constructs containing sequences that mediate TGF-␤1 responsiveness in vitro in other cell types (17)(18)(19)(20)(21). Accordingly, the finding that TGF-␤1 up-regulated expression from Ad-PAI800␤-Gal was to a certain extent anticipated. Nevertheless, our results are novel in that they elucidate an in vivo mechanism of PAI-1 regulation in endothelium, the existence of which was previously only hypothetical. The correlation of our in vitro and in vivo results is useful: the in vivo results prove biological relevance while the in vitro system permits a more robust experimental approach. In vitro strategies will be required, for example, for the eventual identification and cloning of transcription factors involved in gene regulation by TGF-␤1; in vivo approaches will be required for a confirmation that such factors not merely in vitro phenomena but are present and functional in intact animals.
The method described herein permits the definition of pathways of EC gene regulation by physiologic stimuli in intact animals. While exposure of EC to TGF-␤1 may be accomplished both in vitro and in vivo, other important physiologic stimuli that are delivered to the endothelium, such as alterations in blood pressure or flow (45), cannot be adequately modeled (46) (and therefore are not conclusively studied) in vitro. We anticipate that the Ad vector-reporter gene system described herein will permit further in vivo definition of cis-acting sequences that modulate EC gene expression in response to all types of physiologic stimuli, including flow and pressure. Our ability to measure significant up-regulation of gene expression with only 5-6 animals per group (Fig. 7) also demonstrates the utility of this somatic cell gene transfer approach in comparison to analogous experiments that might be performed by germ-line targeting of genetic material to the endothelium (47).
Our experimental system also has shortcomings. Preparation of the large amounts of purified agonists such as TGF-␤1 that must be injected systemically to produce biological effects at a distance can be time consuming and costly. Moreover, side effects of agonist delivery into the systemic circulation could potentially confound experimental results. For example, we cannot be certain that TGF-␤1 is the proximal effector upregulating PAI-1 expression in the transduced carotid endothelium. It is possible that other downstream agonists, induced in vivo by TGF-␤1, participate in the induction of PAI-1. Nevertheless, the consistency of our in vivo results with those obtained in the far more simple in vitro system supports a direct role for TGF-␤1 in vivo. A more conclusive demonstration of a proximal role for TGF-␤1 in vivo might be obtained by incorporating a dominant negative TGF-␤1 receptor into the Ad vector used to transduce endothelium; the consequent elimination of PAI-1 promoter up-regulation would support a role for TGF-␤1 at the surface of transduced EC. In addition, incorporation of promoter elements that are highly active in hepatocytes (48) may permit systemic delivery of agonists by intravenous administration of Ad vectors, eliminating the need for large amounts of purified recombinant proteins.
Side effects specific to the Ad delivery system are another potential difficulty inherent in our in vivo experimental system. While we have excluded certain significant alterations in the vascular phenotype consequent to Ad infusion in this rat system (33), we cannot eliminate the possibility that the artery wall is somehow altered by exposure to Ad, such that it is no longer representative of a "normal" artery. In a rabbit model, we recently reported marked alterations in vascular phenotype following exposure to Ad (49); similar findings have not been reported by us or by others in the rat. Nevertheless, no somatic gene transfer system is completely free of such considerations. The potential occurrence of confounding local biological phenomena must be considered in somatic gene transfer experiments, just as the potential for developmental activation of compensatory gene expression must be considered in germline transgenic and knockout experiments.
In summary, we defined a functional cis-acting sequence in the PAI-1 promoter in endothelium in vivo. Future experiments will include transduction of additional deleted and mutated promoter sequences as well as the co-transduction of dominant negative and constitutively active components of signal transduction pathways that participate in PAI-1 regulation (50). This general method permits definition of pathways that mediate gene regulation in endothelium in vivo. FIG. 8. Histochemical confirmation of up-regulation of expression from 800-bp human PAI-1 promoter in vivo in endothelium. Rat carotid arteries were transduced with AdPAI800␤-Gal, and rats were injected with vehicle (A and C) or TGF-␤1 protein (B and D). Arteries were then harvested, stained with X-gal, and sectioned. To ensure an informative comparison, in both control and experimental arteries only the areas with the most prominent blue staining are shown. In en face views of luminal endothelium only very faint blue dots are visible in a control artery (A); numerous blue dots of varying intensities are visible on the surface of the artery from a rat injected with TGF-␤1 (B). In a histologic section from a control artery (C), only occasional pale blue nuclei are seen. In a section of an artery from a rat receiving TGF-␤1 (D), blue nuclei are more numerous and stain more intensely. Original magnifications: ϫ 12 (A and B); ϫ 158 (C and D). Nuclear fast red counterstain: C and D.