Comparison of promoters for the murine and human P-selectin genes suggests species-specific and conserved mechanisms for transcriptional regulation in endothelial cells.

P-selectin, an adhesion receptor for leukocytes, is constitutively expressed in megakaryocytes and endothelial cells. Tumor necrosis factor-alpha (TNF-alpha) or lipopolysaccharide (LPS) increases synthesis of P-selectin in murine but not in human endothelial cells. To identify potential species-specific and conserved mechanisms for regulation of expression of P-selectin, we cloned the 5'-flanking region of the murine P-selectin gene and compared its features with those previously reported for the human gene. The murine and human genes shared conserved Stat-like, Hox, Ets, GATA, and GT-IIC elements. In the murine gene, a conserved GATA element bound to GATA-2 and functioned as a positive regulatory element, whereas a conserved Ets element bound to GA-binding protein and functioned as a negative regulatory element. Significantly, the murine P-selectin gene had several features not found in the human gene. These included an insertion from -987 to -649 that contained tandem GATA and tandem AP1-like sequences, which resembled enhancers in beta-globin locus control regions. Both tandem elements bound specifically to nuclear proteins. The murine gene lacked the unique kappaB site specific for p50 or p52 homodimers found in the human gene. Instead, it contained two tandem kappaB elements and a variant activating transcription factor/cAMP response element site, which closely resembled sites in the E-selectin gene that are required for TNF-alpha- or LPS-inducible expression. TNF-alpha or LPS augmented expression of a reporter gene driven by the murine, but not the human, P-selectin promoter in transfected endothelial cells. Deletional analysis of the murine 5'-flanking region revealed several sequences that were required for either constitutive or inducible expression. These data suggest that both species-specific and conserved mechanisms regulate transcription of the human and murine P-selectin genes.

Leukocyte extravasation is mediated by the sequential engagement of adhesion molecules, whose expression and/or function are regulated by inflammatory mediators (1)(2)(3). For exam-ple, tumor necrosis factor-␣ (TNF-␣) 1 or LPS stimulates endothelial cells to transcribe the genes encoding the adhesion molecules E-selectin, intercellular cell adhesion molecule-1, vascular cell adhesion molecule-1, and mucosal addressin cell adhesion molecule-1 (2, 4 -9). These newly synthesized proteins are directly transported to the cell surface, where they mediate adhesion of leukocytes. In contrast, the expression of P-selectin is regulated by two distinct mechanisms. First, P-selectin is constitutively synthesized by megakaryocytes (the precursors of platelets) and endothelial cells, where it is sorted into the membranes of secretory granules (10 -14). When these cells are stimulated by agonists such as thrombin, P-selectin is rapidly translocated from granules to the plasma membrane (15), where it initiates tethering and rolling of leukocytes (16 -19). Second, inflammatory mediators such as TNF-␣, LPS, interleukin-4, or oncostatin M increase levels of P-selectin mRNA and protein in endothelial cells (20 -22). The increased synthesis of P-selectin may saturate the sorting pathway into secretory granules, leading to direct delivery of the protein to the cell surface (22,23). Notably, TNF-␣ or LPS increases expression of P-selectin in murine endothelial cells but not in human endothelial cells (22,24). The mechanism for this unusual speciesspecific response is not known.
We previously isolated and conducted a preliminary analysis of the 5Ј-flanking region of the human P-selectin gene (25,26). The sequence from Ϫ309 to Ϫ13 relative to the translational start site conferred tissue-specific expression of a reporter gene in cultured bovine aortic endothelial cells. Deletions or mutations of this sequence revealed at least three positive regulatory regions that included a B site, a GATA element, and at least two potential Ets elements. Most B sites are recognized by inducible homodimeric or heterodimeric NF-B/Rel proteins that participate in cytokine-stimulated gene expression. However, the B site in the human P-selectin promoter bound specifically to constitutively expressed homodimers of p50 or p52, whose functions were differentially regulated by the protooncogene product, Bcl-3 (26). GATA and Ets elements are often observed in genes whose expression is restricted to hematopoietic and/or endothelial cells (25,(27)(28)(29)(30)(31)(32)(33)(34). The human P-selectin GATA element bound to the protein GATA-2, and mutation of this element markedly inhibited gene expression (25). The function of the potential Ets elements in the P-selectin gene was not determined.
To better understand potential conserved and nonconserved mechanisms for expression of the P-selectin gene, we isolated the 5Ј-flanking region of the murine P-selectin gene and com-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM  pared its features with those previously reported for the human gene (25,26). Both genes share functional GATA and Ets sites. However, the murine gene has several unique putative regulatory elements that may explain why TNF-␣ or LPS augments transcription of the P-selectin gene in murine but not in human endothelial cells.
Genomic Cloning, Southern Blot Analysis, and Sequence Comparison-To clone the murine P-selectin 5Ј-flanking region, we designed the following two primers: 5Ј-ACATTTCTGGAAAGCGAATAGG-3Ј which spanned nucleotides 1-22 of the murine P-selectin cDNA sequence, and 5Ј-CATGTCCTCTCTGATCCGTCTG-3Ј which is complementary to nucleotides 105-126 of the cDNA sequence (21). Based on the known human P-selectin gene structure (35), we predicted that the murine P-selectin 5Ј-untranslated region and the translation initiator ATG would be encoded by one exon. We confirmed this prediction by sequencing the PCR product amplified from mouse genomic DNA using the above two primers. The primers were then used in a PCR screen to identify bacteriophage P1 clones encoding the murine 5Ј-flanking region (custom service provided by Genome Systems, Inc., St. Louis, MO). The PCR profile involved denaturation at 94°C for 1.5 min, primer annealing at 50°C for 2.5 min, and extension at 72°C for 1 min, with a final concentration of Mg 2ϩ at 1 mM. Two P1 clones named GS5512 and GS5572 were obtained. DNA restriction fragments derived from the GS5512 clone were subcloned into the plasmids pBluescript II SK (Stratagene) or pIBI20 (IBI) for restriction enzyme mapping and DNA sequencing. Southern blot analysis of mouse genomic DNA with a random-labeled probe spanning the entire 5Ј-untranslated region of the murine P-selectin gene was performed as described (36). Sequence comparison was carried out by using the Bestfit program from Genetics Computer Group (Madison, WI).
RNase Protection Assay-An RNase protection assay was performed with a RPAII TM kit (Ambion Inc., Austin, TX) according to the instructions of the manufacturer. The labeled 165-nucleotide cRNA probe extended from Ϫ164 to Ϫ13 of the murine P-selectin 5Ј-flanking region, plus 14 nucleotides transcribed from the plasmid. Poly(A) ϩ RNA from bEnd.3 cells stimulated with LPS (1 g/ml) and cycloheximide (10 g/ml) for 6 h was isolated as described (25).
Construction of Chimeric Luciferase Expression Vectors, Transfection, and Luciferase Assay-The parental vector for creation of deletion mutants was pMG2, in which a 2032-bp XbaI fragment that contained the 5Ј-flanking region, first exon, and part of intron 1 of the murine P-selectin gene was inserted between the XbaI site of pBluescript II SK with the 5Ј-flanking region toward the T3 promoter. Plasmid mp1379LUC was constructed in the following three steps. 1) A PCR product spanning the murine P-selectin 5Ј-flanking region from Ϫ164 to Ϫ13 was amplified with a sense primer carrying a native PstI site and an antisense primer with addition of an KpnI site at the 5Ј end. The PCR product was digested with PstI and KpnI and used to replace the shorter KpnI-PstI fragment in pMG2. 2) The fragment from Ϫ1379 to Ϫ13 was excised by digestion with KpnI and XbaI and inserted between the KpnI and XbaI sites of pIBI20 to create an additional HindIII site adjacent to the XbaI site. To make other constructs, the PstI site between the XbaI and HindIII sites was eliminated by ligation of the compatible XhoI and SalI sites. The resultant plasmid was designated mp1379IBI. 3) mp1379IBI was digested with KpnI and HindIII, and the HindIII-KpnI fragment carrying the 5Ј-flanking region from Ϫ1379 to Ϫ13 was then inserted between the HindIII and KpnI sites of the plasmid p0LUC (25,37). Plasmids mp593LUC, mp474LUC, mp349LUC, mp288LUC, and mp229LUC were constructed by replacement of the shorter PstI-HindIII fragment in mp1379IBI with the respective PCR fragment spanning from Ϫ593 to Ϫ164, Ϫ474 to Ϫ164, Ϫ349 to Ϫ164, Ϫ288 to Ϫ164, or Ϫ229 to Ϫ164. These fragments were generated with a sense primer carrying an additional HindIII site and an antisense primer carrying a native PstI site. The plasmids were digested with KpnI and HindIII, and the shorter KpnI-HindIII fragments were inserted between the HindIII and KpnI sites of p0LUC. Plasmids mp164LUC, mp148LUC, mp110LUC, and mp91LUC were constructed by inserting between the HindIII and KpnI sites of p0LUC the respective PCR product from Ϫ164 to Ϫ13, Ϫ148 to Ϫ13, Ϫ110 to Ϫ13, or Ϫ91 to Ϫ13 that was generated with a sense primer carrying an additional HindIII site and an antisense primer with an additional KpnI site. Plasmids mpMutGATA and mpMutETS1 were constructed in two steps as follows. 1) The shorter KpnI-PstI fragment in mp1379IBI was replaced with the respective PCR fragment carrying a mutation that was generated according to an overlap extension protocol (25). 2) The shorter HindIII-KpnI fragment from the resultant plasmid was excised and inserted between the HindIII and KpnI sites of p0LUC. The fidelity of all constructs was confirmed by restriction mapping, and the PCR-generated products were verified by sequencing. Preparation of plasmids, transfections, and luciferase assays were performed as described (26).
Gel Mobility Shift Assay-Nuclear extracts from bEnd.3 cells or BAEC were prepared as described (38). Nuclear extracts from HEL and CHRF-288 cells were prepared by a slightly different protocol (39). Gel mobility shift assays were performed as described (25).

Molecular
Cloning of the 5Ј-Flanking Region of the Murine P-selectin Gene-Two mouse P1 clones encoding the murine P-selectin gene were obtained using PCR screening. Southern blot analysis of these two clones revealed the same restriction patterns as those obtained with mouse genomic DNA, suggesting that there was no DNA rearrangement during cloning (data not shown). Restriction fragments derived from the P1 clone GS5512 were subcloned into plasmids (Fig. 1). The insert in subclone pMG2 was sequenced on both strands; it contained the first 1256 bp of the 5Ј-flanking region, all of exon 1, and part of intron 1. 2 Determination of the Transcription Start Site-We employed an RNase protection assay to determine the transcription start site of the murine P-selectin gene. Several protected products were seen from Ϫ134 to Ϫ75 relative to the ATG translation start codon (Fig. 2). These protected products were observed in mRNA from LPS/cycloheximide-stimulated bEnd.3 cells but not in control yeast tRNA. The longest product was the most abundant. These data indicate that transcription of the murine P-selectin gene is initiated primarily at Ϫ134 relative to the translation start site.
Comparison of Structural Features of the Murine and Human 5Ј-Flanking Regions- Fig. 3 aligns the 5Ј-flanking sequence of the murine P-selectin gene with the corresponding sequence of the human gene (25). There was 63% sequence identity between the two genes in the first 600 bp of the 5Ј-flanking region, exon 1, and the first 500 bp of intron 1. This degree of similarity is typical for the mouse and human homologues of a gene (40). Both genes ended the first exon with the translation start codon ATG and shared several putative reg-FIG. 1. Structure of the 5-flanking region of the murine Pselectin gene. Schematic representation of the restriction map for the P1 clone GS5512 encoding the murine P-selectin gene and the subclones of plasmids derived from the P1 clone. The insert in plasmid pMG2 encodes the 5Ј-flanking region, exon 1, and part of intron 1 and was sequenced in entirety on both strands. The inserts in pMG1, pMG3, pMG4, and pMG5 were partially sequenced. ulatory regions. The first conserved region from Ϫ419 to Ϫ409 in the mouse gene and from Ϫ413 to Ϫ403 in the human gene contained a HOX recognition element (41) and a Stat-like sequence (42). The second region from Ϫ337 to Ϫ319 in mouse and from Ϫ325 to Ϫ307 in human contained a HOX element and an Ets motif (43). The third conserved region from Ϫ163 to Ϫ151 in mouse and Ϫ162 to Ϫ150 in human contained a GATA element. The fourth conserved region from Ϫ130 to Ϫ92 in mouse and from Ϫ129 to Ϫ91 in human contained a sequence similar to the GT-IIC element of the SV40 enhancer (44) and to Stat consensus motifs, plus an adjacent Ets element flanked by 4-bp A/T sequences that typically bind to HMG I(Y) proteins (45).
Several unique aspects of the murine P-selectin gene were noted. First, unlike the human P-selectin gene (25), the murine gene initiated transcription predominantly at a single site, consistent with the presence of a weak TATA box ( Ϫ160 AGTAT-CAA Ϫ153 ) located 21 bp upstream of this start site. The weak TATA box overlapped the conserved GATA element. We referred to this murine sequence as a GATA/TATA element because of its possible dual functions. Second, the murine gene contained an insertion from Ϫ987 to Ϫ649 not found in the human gene. Data base analysis revealed that this insertion was a unique sequence. Within the insertion, there were tandem GATA elements from Ϫ875 to Ϫ862 and tandem AP1-like sequences from Ϫ913 to Ϫ895, which resembled the enhancers in the DNase I hypersensitive sites of the ␤-globin locus control regions (46). Third, the murine gene lacked the unique B site and surrounding putative regulatory elements previously identified in the human gene (25,26). Instead, it contained a B site in reverse orientation from Ϫ144 to Ϫ136 and an AP1 element from Ϫ262 to Ϫ256 (47, 48). It also had two tandem B elements from Ϫ542 to Ϫ521 and a variant activating transcription factor/cAMP sequence from Ϫ80 to Ϫ73 that resembled those required for cytokine-inducible expression of the human and murine E-selectin genes (49 -52).
The GATA/TATA Element Is Required for Optimal Murine P-selectin Promoter Activity-To test whether the murine Pselectin GATA/TATA element bound to nuclear proteins, we synthesized double-stranded oligonucleotide probes encompassing the wild-type sequence or a mutant sequence in which the core TATC was converted to TTAG (Fig. 4A). As a positive control, we used a human endothelin-1 GATA probe that is known to bind specific GATA proteins (25,53,54). The labeled P-selectin probes did not form detectable sequence-specific DNA-protein complexes. However, a 100-fold molar excess of the unlabeled wild-type probe, but not the mutant probe, weakly competed with the human endothelin-1 GATA element for binding to GATA proteins (Fig. 4B). These data suggest that the GATA/TATA element interacts weakly with GATA proteins.
To determine whether the GATA/TATA element was required for promoter activity, we made a construct (mp1379LUC) in which the murine 5Ј-flanking region from Ϫ1379 to Ϫ13 was linked to luciferase. We also made a mutant construct in which the TATCA of the wild-type sequence was converted to TTAGA, the same changes made in the mutant probe used for the gel shift assay. When transfected into BAEC, the mutant construct was expressed at levels Ϸ75% lower than those of the wild-type sequence (Fig. 4C). These results indicate that the GATA/TATA element is essential for optimal transcription of the murine P-selectin gene.
Nuclear Protein Binding Activities of Tandem GATA Motifs and Tandem AP-1-like Elements in the Insertion-As illustrated in Fig. 5A, the murine P-selectin sequence from Ϫ923 to Ϫ849 contained tandem GATA and tandem AP-1-like elements, which closely resembled the enhancers in hypersensitive sites 2 and 3 of the murine ␤-globin locus control region (46).
We used gel shift assays to determine whether the tandem GATA elements bound to nuclear proteins. A labeled probe encoding the tandem GATA elements formed a DNA-protein complex with nuclear extracts from bEnd.3 cells (Fig. 5B). Formation of the complex was sequence-specific, as addition of a 100-fold molar excess of the unlabeled probe, but not of an unrelated oligonucleotide, suppressed formation. Complex formation was dependent on the GATA core sequence, because a labeled probe with mutations introduced into tandem GATA elements did not bind nuclear proteins. The unlabeled human P-selectin and endothelin-1 GATA probes also competed with the murine tandem GATA probe for complex formation. Preincubation of HEL cell nuclear extracts with antibodies to GATA-2 diminished binding of the labeled tandem GATA probe to nuclear proteins, whereas antibodies to Est-1/Est-2 or to the Ets protein Fli-1 had no effect (Fig. 5C). These data indicate that the tandem GATA elements bind to members of the GATA protein family.
A labeled probe encompassing the tandem AP-1-like elements formed a DNA-protein complex with nuclear extracts from bEnd.3 cells (Fig. 5D). Complex formation was sequencespecific, as formation was suppressed by addition of a 100-fold molar excess of the unlabeled probe but not of an unlabeled murine H-2K b B recognition element (55). These data indicate that the region encompassing tandem AP-1-like elements also forms a sequence-specific DNA-protein complex.
Nuclear Protein Binding Activity and Transcriptional Function of the Proximal Ets Motif-The human and murine genes for P-selectin share conserved proximal and distal Ets elements. To investigate the role of the proximal Ets element, we determined nuclear protein binding activities with oligonucleo-FIG. 2. Determination of the transcription start site of the murine P-selectin gene. Ten micrograms of poly(A) ϩ RNA from LPS/ cycloheximide-stimulated bEnd.3 cells or 10 g of yeast tRNA was hybridized with a 165-nucleotide riboprobe. After treatment with RNase A/T1, the protected products were separated on a 6% sequencing gel. Arrows indicate the positions of the protected products. The length assigned to each protected band was estimated to be 10% larger than that predicted from DNA size markers, based on the relative mobilities of the riboprobe and a DNA marker of the same size. tides encoding the murine and human proximal Ets sequences (Fig. 6A). A labeled probe encoding the murine proximal Ets element formed two DNA-protein complexes with nuclear extracts from BAEC (Fig. 6B) and from bEnd.3, HEL, and CHRF-288 cells (data not shown). Formation of these two complexes was sequence-specific, as it was significantly diminished by addition of a 100-fold excess of the unlabeled probe, or of the proximal Ets element from the human gene, but not of an unrelated sequence. Complex formation further depended on the Ets core sequence, because a mutant probe in which the core sequence AGGAAG was converted to AGCTAG did not inhibit binding of the labeled wild-type probe and, when labeled, did not form specific DNA-protein complexes (Fig. 6C). Of note, formation of complex I varied with different batches of nuclear extracts, as previously noted for binding of the Etsrelated GABP to its cognate DNA sequences (56). We therefore carried out cross-competition gel shift assays with the proximal Ets probes and the HSV1 IE Ets probe that interacts with GABP (56). Addition of a 50-fold excess of the HSV1 IE Ets probe, but not a 200-fold excess of an unrelated sequence, prevented complex formation of the labeled proximal murine P-selectin Ets probe with BAEC nuclear extracts (Fig. 6D). The labeled HSV1 IE Ets probe formed three DNA-protein complexes with nuclear extracts from BAEC (Fig. 6E) or from HEL, CHRF-288, or bEnd.3 cells (data not shown). However, only complex A was consistently observed. Formation of complex A was prevented or significantly diminished by addition of a 50 -200-fold excess of the unlabeled HSV1 IE Ets element or the murine or human proximal Ets probe but not of the mutant proximal Ets probes. These data suggest that the proximal Ets elements bind to one or more members of the Ets family, including GABP.
To determine the potential function of the murine proximal Ets element, we changed the proximal Ets sequence AGGAA to AGCTAA in the mp1379LUC construct. These changes were the same as those made in the mutant probe that eliminated nuclear protein binding. When transfected into BAEC, the mutant construct was expressed at 2-10-fold higher levels than those of the wild-type sequence ( Fig. 6F and data not shown). These data suggest that the proximal Ets element plays a negative regulatory role in transcription of the murine P-selectin gene.

TNF-␣ or LPS Increases Luciferase Reporter Gene Expression Driven by the Murine but Not the Human P-selectin 5Ј-Flanking
Region-To test whether differences in the promoters of human and murine P-selectin gene accounted for the differential re- sponse of P-selectin to TNF-␣ or LPS, we compared the abilities of TNF-␣ or LPS to induce expression of a reporter gene driven by the murine or human P-selectin 5Ј-flanking region. Following transfection of the constructs into BAEC, the cells were treated with fresh medium in the presence or absence of LPS or TNF-␣ for various intervals, and the cells were then harvested for luciferase activity assays. As shown in Fig. 7, the reporter gene driven by the 5Ј-flanking region of the murine P-selectin gene from Ϫ1379 to Ϫ13 promoted constitutive expression of luciferase that was further increased 5-15-fold after the transfected cells were stimulated with TNF-␣ or LPS for 4.5 h. In sharp contrast, the reporter gene driven by the 5Ј-flanking region of the human P-selectin gene from Ϫ1339 to Ϫ13 promoted constitutive expression that was not further increased after the same treatment. Similar results were obtained when the transfected cells were stimulated with TNF-␣ or LPS for 6 or 8 h. The failure to induce expression of the human P-selectin reporter gene was not due to excessive basal expression, because the levels of constitutive expression were similar for both the murine and human P-selectin reporter genes (data not shown). These data demonstrate that the 5Ј-flanking regions of both the murine and human P-selectin genes confer constitutive expression in endothelial cells. However, only the murine sequence has elements that allow inducible expression in response to TNF-␣ or LPS.
Deletional Analysis of the Murine P-selectin 5Ј-Flanking Region-To localize regulatory elements required for the constitutive or inducible expression of the murine P-selectin gene, we prepared reporter constructs in which serially truncated fragments of the 5Ј-flanking region were linked to luciferase. Following transfection of the constructs into BAEC, the cells were incubated with or without TNF-␣ or LPS and then assayed for luciferase activity. As shown in Fig. 8, deletion to position Ϫ593 decreased constitutive expression of the luciferase reporter gene by Ϸ50% but did not affect the degree of induction by LPS or TNF-␣. Deletion to Ϫ474 did not further decrease constitutive expression but did significantly reduce inducible expression. Deletion to Ϫ349 slightly increased constitutive expression but did not further affect inducible expression. Deletions to Ϫ288, Ϫ229, Ϫ164 or Ϫ148 only modestly affected constitutive or inducible expression. Deletions to Ϫ110 or Ϫ91 drastically reduced both constitutive and inducible expression. These data suggest that the sequence from Ϫ1379 to Ϫ593 contains a positive regulatory element(s) for constitutive expression, and the sequence from Ϫ593 to Ϫ474 contains a regulatory element(s) for TNF-␣-or LPS-mediated inducible expression. Sequences proximal to Ϫ229 appear to be required for both constitutive and inducible expression. Fig. 4. Nuclear protein binding activity and function of the conserved GATA element. A, upper strand sequences of the oligonucleotides used as probes and competitors in the gel mobility shift assay. The murine P-selectin GATA/TATA element is from the indicated region of the 5Ј-flanking region of the murine P-selectin gene. The mutant oligonucleotide contains three nucleotide changes in the core GATA sequence. The human P-selectin and endothelin-1 GATA probes were characterized previously (25,53,54). B, competitive gel mobility shift assay. Nuclear extracts from HEL cells were incubated with a 32 Plabeled endothelin-1 GATA element in the presence of a 100-fold molar excess of the indicated unlabeled competitor. C, the same mutations of the GATA/TATA element were introduced into a luciferase reporter gene driven by the murine P-selectin 5Ј-flanking sequence from Ϫ1379 to Ϫ13. Luciferase activity was measured following transfection of the wild-type and mutant reporter genes into BAEC. The luciferase activity of the wild-type reporter gene was normalized to 100%. The data represent the mean Ϯ S.D. of three experiments. Three independent transfections were performed in each experiment.

DISCUSSION
We cloned the 5Ј-flanking region of the murine gene and compared its features with those previously reported for the human gene. Both genes shared several conserved regulatory elements, including functional GATA and Ets elements. However, several unique aspects of the murine gene were revealed, including several elements that may explain why TNF-␣ or LPS augments transcription of the P-selectin gene in murine but not in human endothelial cells.
The human P-selectin promoter initiates transcription at multiple sites (25). In contrast, the murine promoter initiates transcription predominantly at a single site. Both genes have functional proximal GATA sites. However, unlike the human GATA site, the murine site deviates one nucleotide from the consensus GATA sequence, leading to weaker binding to GATA proteins. Furthermore, the murine GATA site overlaps a weak TATA box located 21 bp upstream of the transcription start site. We refer to the murine GATA site as a GATA/TATA element, because it resembles the GATA/TATA elements in the rat platelet factor-4 and chicken ␤-globin genes that can bind to either TFIID or GATA proteins (28,57,58). This dual-binding property may allow the GATA/TATA element to communicate with a distal tissue-specific enhancer. In the chicken ␤-globin gene, the protein GATA-1, when bound to a proximal GATA/ TATA element, may interact with GATA-1 or other proteins bound to a distal 3Ј enhancer that contains both GATA ele-FIG. 5. Nuclear protein binding activities of the murine P-selectin tandem GATA and AP-1-like elements. A, sequence comparison of the enhancers in the hypersensitive sites 2 or 3 (HS2 or HS3) of the murine ␤-globin locus control region (46) and in the murine P-selectin insertion. Also indicated are the upper strand sequences of the wild-type and mutant oligonucleotides encoding the murine P-selectin tandem GATA elements or AP-1-like elements used in gel mobility shift assays. B, nuclear extracts from bEnd.3 cells were incubated with labeled wild-type or mutant (Mut.) P-selectin tandem GATA probes in the absence or the presence of the indicated unlabeled competitor. C, nuclear extracts from HEL cells were incubated with the labeled tandem GATA probe in the absence or the presence of the indicated antibody. D, nuclear extracts from bEnd.3 cells were incubated with the tandem AP-1-like probe in the absence or the presence of the indicated unlabeled competitor.

Promoters of Murine and Human P-selectin Genes
ments and NF-E2 (AP1-like) elements (58). This interaction allows the two control regions to form a DNA loop. Once the DNA loop is formed, TFIID and adapter proteins displace GATA-1 and produce an active initiation complex under control of the erythroid-specific enhancer. Intriguingly, the murine P-selectin promoter contains an upstream region composed of tandem GATA elements and tandem AP-1-like elements. Furthermore, the murine P-selectin tandem GATA elements bound to at least one GATA protein (GATA-2), and the tandem AP-1like elements bound to a still uncharacterized nuclear protein.
Deletion of the sequence from Ϫ1379 to Ϫ593, which contains these elements, decreased expression of a luciferase reporter gene in BAEC by 50%. These data suggest that the murine GATA/TATA element may cooperate with this putative enhancer, in a manner much like that proposed for the ␤-globin gene. The murine putative enhancer is located in an insertion that is not present in the corresponding portion of the human P-selectin gene. It is not known whether the human gene has a similar enhancer in a different location.
Ets elements, which are often associated with GATA elements, are in the promoters of many genes whose expression is restricted to hematopoietic and/or endothelial cells (25,(27)(28)(29)(30)(31)(32)(33)(34). The functions of most Ets elements have not been well characterized, although many Ets proteins are known transcriptional activators. Many Ets elements are adjacent to other regulatory elements, which refine their binding specificities for particular Ets proteins (43). Two conserved Ets elements were identified in both the mouse and human P-selectin genes. Interestingly, the distances between the two murine Ets elements (224 bp) and the two human elements (213 bp) differ by exactly one turn of the DNA helix. The distal Ets element is adjacent to a conserved AT-rich HOX recognition sequence (41), and the proximal Ets element is flanked by conserved 4-bp A/T sequences. Mutation of the proximal Ets element eliminated binding to an Ets protein, most likely the ubiquitously expressed GABP, and increased reporter gene expression directed by the murine P-selectin promoter by 2-10-fold in transfected BAEC. The ␣-subunit of GABP (GABP␣) shares sequence similarity with the Ets family of transcription factors, and the ␤-subunit contains four ankyrin repeats that mediate the interaction with GABP␣ (56). GABP␤ dimerizes to form an ␣2␤2 tetramer that binds with high affinity to two Ets elements separated by variable distances (56,59). Although the two conserved Ets elements in the P-selectin genes are separated by ϳ200 bp in linear sequence, a nucleosome (60,61) could bring the two elements sufficiently close to interact simultaneously with a GABP tetramer to form a DNA loop. By a similar mechanism, the three DNA-binding domains of the architectural HMG I(Y) protein may interact with the three A/T-rich sequences adjacent to the Ets elements to stabilize the loop (62). Formation of this putative loop may be regulated. For example, oxidation of GABP␣ destroys its DNA binding activity (63), a Hox protein may bind to the conserved 4-bp A/T sequence adjacent to the distal Ets element (41), and interleukin-4-dependent signaling may modulate the DNA binding of HMG I(Y) through phosphorylation (64). Mutation of the murine proximal Ets element increased reporter gene expression. This suggests that the putative Ets-mediated DNA looping may repress transcription. Transcriptional repression mediated by DNA looping has been observed in other genes (65). Although deletion of the distal Ets element did not alter reporter gene expression, large deletions may alter other regulatory elements. Furthermore, data from transient transfection of reporter genes may not fully reflect the architecture and function of genes in the intact chromosome (66,67).
Our findings also suggest a molecular explanation for the ability of TNF-␣ or LPS to augment expression of P-selectin in murine but not in human endothelial cells. TNF-␣ or LPS increased expression of a reporter gene driven by the murine, but not the human, P-selectin promoter in transfected BAEC. Serial deletions of the murine 5Ј-flanking region identified two segments that were required for the TNF-␣or LPS-mediated induction of transcription. The first segment from Ϫ593 to Ϫ474 contained two tandem putative B sites, whereas the second segment proximal to Ϫ229 contained a reverse-oriented B site and a variant activating transcription factor/cAMP sequence. In the accompanying paper (68), we demonstrate that all these elements are required for TNF-␣ or LPS to FIG. 6. Nuclear protein binding activity and function of the proximal Ets element. A, upper strand sequences of the oligonucleotides encompassing the proximal Ets elements from the 5Ј-flanking sequences of the human and murine P-selectin genes. The mutant oligonucleotides have two nucleotide changes (AGGAA to AGCTA) in the core region of the human or murine Ets element. The oligonucleotide encoding a GABP recognition site was derived from the HSV1 IE gene (56). B, the labeled murine proximal Ets oligonucleotide was incubated with BAEC nuclear extracts in the absence or presence of the indicated competitor. The positions of specific DNA-protein complexes are indicated with arrows. The amount of complex I varied with different batches of nuclear extracts. C, the labeled wild-type or mutant murine proximal Ets element was incubated with BAEC nuclear extracts in the absence or presence of the indicated competitor. D, the labeled murine proximal Ets element was incubated with BAEC nuclear extracts in the absence or presence of the indicated unlabeled competitor. E, the labeled HSV IE Ets element was incubated with BAEC nuclear extracts in the absence or presence of the indicated unlabeled competitor. The positions of specific DNA-protein complexes are indicated. Complexes B and C were variably detected using nuclear extracts from different cell types. F, the same mutations in the proximal Ets element were introduced into a luciferase reporter gene driven by the murine P-selectin 5Ј-flanking region from Ϫ1379 to Ϫ13. The mutant and wild-type constructs were transfected into BAEC, and the luciferase activities were determined. The data represent the mean Ϯ S.D. of one experiment with three independent transfections. In two other experiments, the mutant construct expressed 2-or 10-fold more luciferase activity than the wild-type construct. maximally induce transcription of the murine P-selectin gene. These sites are not present in the 5Ј-flanking region of the human P-selectin gene. Instead, the human gene has a novel B site that binds homodimers of p50 or p52 but not homodimers or heterodimers containing p65 (26). Thus, the differential ability of TNF-␣ or LPS to induce expression of Pselectin in mice, but not in humans, may result in part from differences in the promoters of the respective genes.
The species-specific transcriptional regulation of P-selectin in response to TNF-␣ or LPS is both unusual and surprising, given the apparent similarities of many selectin functions across species (69 -71). However, there are other known genes in which species-specific changes in regulatory elements affect responsiveness to particular mediators (72,73). Although the biological implications of these species-specific regulatory mechanisms remain unexplained, the presence of such mechanisms suggests caution in extrapolating the results of some inducible gene responses in mouse models to humans.
FIG. 8. 5 deletion analysis of the murine P-selectin promoter. BAEC were transfected with plasmids containing serially truncated segments of the murine P-selectin 5Ј-flanking region fused to luciferase as diagrammed on the left. The locations of putative regulatory elements are depicted on the upper part of the figure. The relative luciferase activity of each construct in unstimulated, TNF-␣-, or LPS-stimulated BAEC is plotted on the right. The constitutively expressed luciferase activity is represented as a percentage, with the activity of the longest construct listed as 100%. The fold induction of expression after stimulation with TNF-␣ or LPS is given relative to the constitutive expression for the same construct. Each experiment was carried out with three independent transfections for each construct. The data represent the mean of three experiments for constitutive expression and the average of two experiments for expression after stimulation with TNF-␣ or LPS. The S.D. or variation was less than 20%. ND, not determined.