Nuclear protein interactions with the human KDR/flk-1 promoter in vivo. Regulation of Sp1 binding is associated with cell type-specific expression.

The endothelial cell type-specific tyrosine kinase KDR/flk-1 is a receptor for vascular endothelial growth factor and a critical regulator of endothelial cell growth and development. To study mechanisms of endothelial cell differentiation and gene regulation, we have analyzed the topology of the proximal promoter of human KDR/flk-1. A protected sequence between base pairs −110 and −25 was defined by in vitro DNase I footprinting analysis in human umbilical vein endothelial cells (HUVECs). Purified Sp1 alone produced similar protection, and electrophoretic mobility shift assays demonstrated that Sp1 was indeed the major nuclear protein binding to this region. Despite the cell type specificity of KDR/flk-1 expression, no cell type differences were observed in DNA-protein interactions in vitro. In contrast, in vivo footprinting assays demonstrated marked differences in core promoter interactions between cell types. Protection of Sp1 binding sites was observed in HUVECs by in vivo DNase I footprinting, whereas in human fibroblasts and HeLa cells a pattern consistent with nucleosomal positioning was observed. In vivo dimethylsulfate footprinting confirmed that DNA-protein interactions occurred within Sp1 elements in HUVECs but not in nonendothelial cells. It is possible that distant elements coordinate Sp1 binding and chromatin structure to regulate cell type-specific expression of KDR/flk-1.

KDR/flk-1 is a membrane-bound receptor of the tyrosine kinase family with expression restricted predominantly to endothelial cells (1). KDR/flk-1 and a similar tyrosine kinase, flt-1, are receptors for the specific endothelial cell mitogen and angiogenic peptide vascular endothelial growth factor (VEGF) 1 (1)(2)(3). Both receptors are expressed early in murine development, with KDR/flk-1 appearing a full day earlier (day 7.0 -7.5 of the developing mouse embryo) than flt-1 (4,5). Of the two, KDR/flk-1 has a wider pattern of expression among endothelial cell populations than does flt-1 (6). In addition, only KDR/flk-1 has been shown to autophosphorylate in the presence of VEGF (1), and signal transduction mechanisms downstream of the two receptors differ (7). These differences in expression and signaling suggest that the two receptors mediate different physiologic functions of VEGF in endothelial cell growth and angiogenesis.
Recent studies in which the genes for the two VEGF receptors were deleted in mice by homologous recombination have provided critical data about KDR/flk-1 function (8,9). In mice with homozygous deletions of flt-1, endothelial cells develop normally, but vessel formation is impaired, and lethality occurs at the midsomitic stages of development. In contrast, homozygous deletion of KDR/flk-1 also results in early embryo death (at day 8.5-9.5), but histologic examination reveals that embryonic endothelial cells are completely absent. Thus, KDR/ flk-1 appears to be upstream of flt-1 in the cascade of vascular development; moreover, the presence of KDR/flk-1 is absolutely required for the development of endothelial cells from hemangioblastic precursors.
In addition to its important developmental role, KDR/flk-1 is implicated in the pathogenesis of a number of diseases with significant angiogenic components. For example, expression of both KDR/flk-1 and its ligand VEGF are up-regulated in neoplastic processes (10,11), and administration of a dominantnegative form of KDR/flk-1 inhibits angiogenesis and experimental tumor growth (12). Likewise, a critical role for KDR/ flk-1 has been established in angiogenesis associated with proliferative retinopathies (13).
In view of the importance of KDR/flk-1 in endothelial cell differentiation and angiogenesis, and to address the mechanisms of endothelial cell type-specific gene regulation, we have cloned and begun to analyze the regulatory elements of the human KDR/flk-1 gene (14). We have previously demonstrated in transient transfection assays that maximal promoter activity of the 5Ј-flanking region resides within a fragment from Ϫ225 to ϩ127 relative to the transcription initiation site, and that deletions from Ϫ95 to Ϫ37 result in complete loss of promoter activity, defining this segment as the core promoter for human KDR/flk-1.
Because putative binding sites for AP-2, NFB, and Sp1 are found within the KDR/flk-1 core promoter, we have now examined which factors interact with the proximal promoter region in vitro and in vivo and how nonspecifically expressed trans-acting factors might function to regulate an endothelial cell type-specific gene. We demonstrate by electrophoretic mobility shift assays (EMSAs) and in vitro DNase I footprinting experiments that a large nucleoprotein complex forms over the human KDR/flk-1 core promoter, and that purified Sp1 alone is sufficient to recapitulate this binding pattern. Nuclear ex-tracts from endothelial and nonendothelial cells produce the same binding pattern over the core promoter, as would be expected, since Sp1 is expressed ubiquitously in mammalian cells (15). In marked contrast, we show by in vivo dimethylsulfate (DMS) and DNase I footprinting experiments using ligation-mediated polymerase chain reaction (LM-PCR) that Sp1 or Sp1-like proteins bind to the human KDR/flk-1 core promoter in endothelial cells but not in fibroblasts or HeLa cells in vivo, despite the presence of Sp1 in fibroblasts and HeLa cells. Moreover, we provide evidence that the exclusion of Sp1 from the KDR/flk-1 core promoter in nonendothelial cells is associated with changes in chromatin structure. Our data support a model whereby proximal promoter elements are regulated by chromatin structure to provide access to or to exclude ubiquitous trans-acting factors for the regulation of cell type-specific gene expression.

MATERIALS AND METHODS
Cell Culture-Primary culture human umbilical vein endothelial cells (HUVECs) were obtained from Clonetics Corp. (San Diego, CA) and were grown in M199 medium supplemented with 20% fetal calf serum (HyClone, Logan, UT), 60 g/ml endothelial cell growth supplement (Collaborative Biomedical, Bedford, MA), 50 g/ml heparin, 100 units/ml penicillin, and 100 g/ml streptomycin in gelatin-coated tissue culture plates. Human fetal fibroblasts and HeLa cells were obtained from the American Type Culture Collection and were cultured in Dulbecco's modified Eagle's medium (JRH Biosciences, Lenexa, KS) supplemented with 10% fetal calf serum. Primary culture cells were passaged every 4 -6 days, and experiments were performed on cells three to six passages from the primary culture. Nuclear extracts were prepared as described (16); protein concentrations in nuclear extracts were measured with the Bio-Rad protein assay system.
Electrophoretic Mobility Shift Assay-EMSA was performed as described previously (17). Probes consisted of annealed synthetic complementary oligonucleotides corresponding to the published human KDR/ flk-1 sequence (14). The sequences of the probes used were: GS1, 5Ј-CGGGTGAGGGGCGGGGCTGGCCGCACG-3Ј; GS2, 5Ј-GCACGGGA-GAGCCCCTCCTC-3Ј; GS3, 5Ј-TCCTCCGCCCCGGCCCCGCCCCGC-ATG-3Ј; and GS4, 5Ј-CATGGCCCCGCCTCCGCGCTCTAGAG-3Ј. Prior to annealing, the oligonucleotides were labeled with [␥-32 P]ATP using polynucleotide kinase (Boehringer Mannheim). A typical binding reaction contained 20,000 cpm of DNA probe, 1 g of poly(dI-dC)⅐poly(dI-dC), 25 mM HEPES (pH 7.9), 40 mM KCl, 0.1 mM EDTA, 1 mM dithiothreitol, 10% glycerol, and 10 g of nuclear extract in a final volume of 25 l. The reaction mixture was incubated at room temperature for 20 min and fractionated on a 5% native polyacrylamide gel in 0.25 ϫ Tris borate/EDTA buffer (22 mM Tris base, 22 mM boric acid, and 0.5 mM EDTA). To determine the specificity of the DNA-protein complexes, we performed competition assays using 10-or 100-fold molar excess of an unlabeled double-stranded oligonucleotide encoding a consensus Sp1 binding site (specific inhibitor) or a 100-fold molar excess of an unrelated double-stranded oligonucleotide of comparable length (nonspecific inhibitor). To characterize specific DNA-binding proteins, we incubated nuclear extracts with a polyclonal anti-human Sp1 antibody or a similar anti-AP-2 antibody (Santa Cruz Biotechnology, Santa Cruz, CA) for 3 h at 4°C before addition of probe. In some experiments, recombinant Sp1 (Promega, Madison, WI) was used in place of nuclear extract.
In Vitro DNase I Footprinting-In vitro DNase I footprinting was performed as described by Wildeman et al (18). DNA fragments containing bp Ϫ323 to ϩ5 (for coding strand analysis) and bp Ϫ225 to ϩ56 (for noncoding strand analysis) of the human KDR/flk-1 gene were labeled with [␣-32 P]dGTP using the Klenow fragment of DNA polymerase I. About 10,000 cpm (30 pg) of the labeled DNA fragment were incubated with 25 g of nuclear extract or BSA and 1 g of poly(dI-dC)⅐poly(dI-dC) for 25 min on ice and then 2 min at room temperature.
In some experiments, recombinant human Sp1 or AP-2 (2 footprinting units, Promega) was used in addition to BSA. Samples were treated with increasing doses of DNase I (0.005-0.0005 Kunitz units with BSA and 0.05-0.005 Kunitz units with nuclear extract) at room temperature for 2 min. Samples were analyzed as described (18) using a 6% denaturing polyacrylamide/urea gel. For comparison, a G ladder of the same end-labeled DNA fragment was generated by Maxam-Gilbert chemistry (19).
In Vivo Footprinting by LM-PCR-In vivo DMS treatment was performed according to the method of Mueller and Wold (20) with modifications. For methylation, cells were treated with 0.1% DMS at room temperature for 3-5 min, washed, and deproteinized in lysis buffer (300 mM NaCl, 50 mM Tris-Cl, 25 mM EDTA, 0.2% SDS, and 0.2 mg/ml proteinase K) at 37°C for 4 h. For in vitro methylation of genomic DNA, naked DNA was treated with 0.125% DMS for 2 min. Methylated DNA was subsequently cleaved with piperidine.
In vivo DNase I treatment of cultured cells was performed according to the method of Pfeifer and Riggs (21) with modifications. For in vivo DNase I treatment, cells were permeabilized with 0.05% lysolecithin (Sigma) followed by treatment with DNase I (75 or 150 Kunitz units/ml) for 4 min at room temperature. Genomic DNA was treated in vitro with DNase I (15 Kunitz units/ml) for 4 min.

DNA-Protein
Interactions in the Core Promoter of the Human KDR/flk-1 Gene-We have previously defined within the 5Јflanking region of the human KDR/flk-1 gene a fragment extending from Ϫ95 to Ϫ37 that, when deleted, reduces activity of a luciferase reporter construct to background in transient transfection assays (14). These data establish that elements within this region, which contains potential binding sites for Sp1, AP-2, and NFB, are critical for expression of KDR/flk-1 and that these elements likely serve as the core promoter of this TATA-less gene (Fig. 1). To define DNA-protein interactions in this region and to determine the proteins binding the KDR/flk-1 promoter, we first performed in vitro DNase I footprinting assays using a labeled fragment from Ϫ323 to ϩ5 of the KDR/flk-1 promoter as a probe to disclose interactions protecting the coding strand. A densely protected nucleoprotein complex formed in the presence of HUVEC nuclear extract (but not BSA) spanning nucleotides Ϫ106 to Ϫ30 ( Fig. 2A). This region includes four potential binding sites for Sp1, two for AP-2, and one for NFB but does not include the upstream binding sites for AP-2, Sp1, and NFB located between Ϫ143 and Ϫ118, which were not protected. On the coding strand, hypersensitive sites flanked the region of protection and were also noted at positions Ϫ85 and Ϫ84 ( Fig. 2A, arrows).
Footprint analysis of the noncoding strand revealed a similar pattern, with protection from Ϫ25 to Ϫ110 (Fig. 2B). On this strand, the region from Ϫ90 to Ϫ75 was only partially protected and contained hypersensitive bases; in addition, guanine residues between positions Ϫ58 and Ϫ54 were also strongly hypersensitive (Fig. 2B, arrows). On both strands, protection of Ϫ110 to Ϫ25 was nearly complete, suggesting that the affinity of protein binding to DNA is relatively strong. No protection was observed upstream or downstream of Ϫ110 to Ϫ25 on either strand of the probes used.
Sp1 Binds to the KDR/flk-1 Core Promoter-As a first step to determine the nuclear protein(s) present in HUVEC nuclear extract that contribute to this nucleoprotein complex, we compared the binding activity of HUVEC nuclear extract with that of recombinant Sp1 in in vitro footprinting assays using the human KDR/flk-1 fragment from Ϫ323 to ϩ5 as a probe. To our surprise, Sp1 had essentially the same binding activity as did HUVEC nuclear extract (Fig. 3), raising the possibility that it (or similar proteins) are the principal components of HUVEC nuclear extract binding to the human KDR/flk-1 core promoter. We also examined whether recombinant AP-2 contributed to the binding activity of HUVEC nuclear extract. Interestingly, AP-2 did not bind within the region bound by the nuclear extract but did bind immediately upstream of this region, between bp Ϫ126 to Ϫ106 (Fig. 3). On examination of the sequence in the region protected by AP-2, we found a motif (TCCCCGCCG) that differs from the consensus AP-2 binding site (YCSCCMNSSS) (24) by 1 bp; AP-2 may bind to these bases but does not seem to bind any of the exact AP-2 consensus sequences in the KDR/flk-1 promoter. These results demonstrate that Sp1 binding to the KDR/flk-1 promoter is specific; furthermore, they indicate that AP-2 does not contribute to the binding activity of HUVEC nuclear extract, although there is a possibility that AP-2 is involved in inducible regulation of KDR/flk-1 through this upstream binding site.
To confirm that Sp1 is indeed the nuclear protein in HUVEC nuclear extract binding to the KDR/flk-1 promoter, we performed EMSA. The protected region of the KDR/flk-1 promoter was divided into four overlapping probes (GS1-GS4), and their mobilities were analyzed after incubation with HUVEC nuclear extract. Three specific DNA-protein complexes (Fig. 4, a-c) were formed with probes GS1, GS3, and GS4, which all contain consensus Sp1 binding sites. A fourth high mobility complex (Fig. 4, d), was variably present in nuclear extract preparations and may represent a degradation product of nuclear proteins participating in the larger complexes. These complexes were abolished by competition with a 10-or 100-fold molar excess of a nonradioactive Sp1 consensus sequence but not by nonspecific competitor DNA. The upper complex (Fig. 4, a) was supershifted by incubating nuclear extracts with an anti-Sp1 antibody prior to the binding reaction, whereas an anti-AP-2 antibody (Fig. 4) or an anti-p65 antibody (not shown) had no effect on complex formation. Finally, incubation of probes GS1, GS3, and GS4 with recombinant Sp1 produced a complex that co-migrated with complex a. Taken together, these results indicate that the most abundant complex, a, is Sp1. Preliminary data indicate that the lesser abundant complexes, b and c, may be the closely related zinc finger protein Sp3 (data not shown), which recognizes the same sequence motif as Sp1 and is a competitive inhibitor of Sp1 binding and trans-activation (25). When GS2 (which does not contain an Sp1 motif) was used as a probe, a single complex was formed that could be abolished by both specific and nonspecific competitor DNA. No specific binding to this fragment was observed under any conditions tested. Of note, the sequence contained in GS2 is also poorly protected by DNase I footprinting in vitro (Fig. 2) and in vivo (below).
Analysis of Cell Type Specificity of DNA-Protein Interactions in Vitro-Because KDR/flk-1 is expressed in a cell type-specific manner, it was important to test whether nuclear proteins interacted with the core promoter in a cell type-specific manner. We therefore compared the binding activity of HUVEC and human fibroblast nuclear extracts in in vitro DNase I footprinting experiments. As shown in Fig. 5, nuclear extracts from human fibroblasts produced a footprinting pattern on the KDR/flk-1 promoter identical to that of HUVEC nuclear extracts. Taken together with our other studies, it appears that Sp1 is the predominant nuclear protein binding to the KDR/ flk-1 proximal promoter. Moreover, ubiquitously expressed Sp1 from different cell types appears to bind to this promoter with equal avidity.
In keeping with the convention of Gidoni et al. (26), we have labeled the Sp1 sites that are bound by nuclear protein Sp1 I-IV in order of their proximity to the transcription initiation site (Fig. 5). Our initial experiments define the importance of Sp1 interactions with these sites in the KDR/flk-1 core promoter; however, our in vitro experiments provide no information about the mechanisms of the cell type-specific expression of the gene.
In Vivo Analysis of the KDR/flk-1 Promoter by LM-PCR-We hypothesized that important differences between in vitro and in vivo DNA-protein interactions might explain how ubiquitous trans-acting factors could regulate a cell type-specific gene such as KDR/flk-1. To explore this possibility, we performed in vivo footprinting of the proximal promoter of KDR/flk-1 by LM-PCR in cells treated with DNase I after permeabilization with lysolecithin. This permeabilization technique does not inhibit transcription initiation and should therefore maintain a faithful promoter architecture for probing with DNase I (27). Compared with in vitro DNase I-treated purified genomic DNA, HUVEC genomic DNA treated in vivo was strongly protected over the four Sp1 sites defined as being important in our previous in vitro experiments (Fig. 6). Sequences surrounding the pro- tected regions were hypersensitive to DNase treatment (Fig. 6,  arrows), and no other upstream protection was noted. Because these results correspond closely to the pattern of protection defined by in vitro footprinting experiments (Figs. 2 and 3) and because of the relative abundance of Sp1 compared with other species that may bind these sequences in vitro (Fig. 4), we speculate that these sequences are indeed bound predominantly by Sp1 in HUVEC in vivo.
In vivo DNase I reactivity within the KDR/flk-1 proximal promoter was markedly different for human fibroblasts and HeLa cells in comparison with either the pattern of protection noted in HUVECs or DNase I sensitivity of naked genomic DNA. In contrast to the random pattern of naked DNA or the protected pattern of HUVECs, a third pattern of alternating hypersensitive (Fig. 6, black triangles) and hyposensitive cleavage was observed. The periodicity of in vivo DNase I-hypersensitive sites in nonendothelial cells was between 10 and 11 bp. On the basis of these experiments, two equally plausible explanations for our results exist. Nuclear proteins different from those interacting with the KDR/flk-1 promoter in endothelial cells may bind this promoter in nonendothelial cells, perhaps exerting a silencing effect on transcription. Alternatively, the periodic pattern of DNase I hypersensitivity may reflect wrapping of DNA around positioned nucleosomes, which either move or are dissolved when the gene is activated in endothelial cells. Similar DNase I sensitivity associated with nucleosomal positioning has been described for the phosphoglycerate kinase-1 locus (21).
To explain the differences between endothelial and nonendothelial cells in patterns of in vivo DNase I sensitivity within the KDR/flk-1 promoter, we performed LM-PCR on genomic DNA from cells treated with DMS in vivo. DMS footprinting provides information complementary to that generated by DNase I footprinting in that DMS is a smaller molecule and can therefore identify nucleotides interacting with protein with single base pair resolution. In addition, DMS interactions are not affected by DNA positioning on nucleosomes (28). On the coding strand, G residues at Ϫ103, Ϫ102, and Ϫ98 were protected from DMS modification, and the G residue at Ϫ97 was strongly hypersensitive (Figs. 7A and 8). These residues are contained within the Sp1 IV site, which is the only Sp1 site that is G-rich on the coding strand (Fig. 9). No other DMS-sensitive residues were noted within at least 100 bp upstream or downstream of these residues. It should be noted that some G residues (for example, Ϫ101) are poorly reactive with DMS both in vitro and in vivo; this may be a consequence of the high G-C content of the KDR/flk-1 5Ј-flanking sequence.
On the noncoding strand, 14 G residues were protected from DMS modification in HUVECs but not naked DNA, and 2 were hypersensitive (Fig. 7B). Two protected G residues immediately flank the three 3Ј-Sp1 sites (Sp1 I-III), and the remainder are contained within canonical sequences for these Sp1 sites, all of which are G-rich on the noncoding strand. In marked contrast, no differences were noted in DMS modification in human fibroblasts or HeLa cells compared with naked genomic DNA on either strand (Figs. 7 and 8). We conclude that nuclear proteins (most likely Sp1) interact with the KDR/flk-1 proximal promoter in endothelial cells, but no DNA-protein interactions are detected in nonendothelial cells in vivo. These results strongly support the contention that nucleosomal wrapping rather than differential transcription factor binding accounts for the differences between endothelial and nonendothelial cells in DNase I protection of the KDR/flk-1 proximal promoter in vivo. promoter of human KDR/flk-1. We have demonstrated that the important Sp1 sites are bound in vivo in endothelial cells but not in nonendothelial cells. Furthermore, our data suggest that the lack of Sp1 binding to this promoter in nonendothelial cells is associated with changes in chromatin structure.
The results of in vivo DMS and DNase I footprinting in HUVECs demonstrate that four Sp1 motifs (Sp1 I-IV) within the KDR/flk-1 promoter are occupied in vivo, and that contacts are made predominantly with the G-rich strand of each motif. Our in vivo results correspond closely to in vitro binding of purified Sp1 to the SV40 promoter (26), in which DMS protection is also restricted to the G-rich strand of Sp1 elements. In addition, although the crystal structure of Sp1 binding to DNA has not yet been solved, the coordinates for binding of Egr-1, a related zinc finger protein, to DNA have been resolved and are also restricted to G residues on one strand (29), suggesting that binding to the G-rich strand is the mechanism of DNA interaction for three-zinc finger nuclear proteins. Our results are therefore entirely consistent with Sp1 binding to the KDR/ flk-1 promoter in endothelial cells in vivo.
On the basis of our results, the human KDR/flk-1 promoter bears strong similarities to the SV40 promoter (26). The SV40 promoter contains repeated Sp1 elements, and Sp1 binds cooperatively to these elements to induce bending toward the minor groove of the DNA helix; this bending may bring distant regulatory elements in proximity to the core transcriptional apparatus (30). It is thought that cooperative binding of Sp1 on the SV40 promoter involves alignment of protected residues along a single face of the DNA helix (26). The periodicity between the axes of protected residues in the Sp1 I-III sites in the KDR/ flk-1 promoter is approximately 14 bp (Fig. 9), suggesting that if similar cooperative alignment occurs in this promoter, the helix may be slightly unwound, perhaps due to steric or sequence-related considerations. Alignment may also explain why the Sp1 IV site has a greater periodicity between the axis of its protected residues and those of Sp1 III; since Sp1 IV is G-rich and therefore protected on the opposite strand from the other three Sp1 sites, an additional half turn of the helix would be necessary to bring it into alignment. In the SV40 promoter, all Sp1 sites are located on the same strand; whether the orientation of Sp1 IV on the opposite strand from the other Sp1 sites in the KDR/flk-1 promoter has any functional significance remains to be determined.
Although Sp1 has primarily been considered a transcription factor for housekeeping genes, convincing evidence suggests that Sp1 is functionally regulated (31,32) and participates in cell type-specific gene expression (33)(34)(35). Moreover, Sp1 is regulated during development (36), is important for the expression of -globin in differentiating erythroid cells (37), and is highly expressed in areas of vasculogenesis such as the developing hearts of mouse embryos (36). It is plausible, then, that Sp1 may participate not only in constitutive expression of KDR/flk-1 in human endothelial cells, but also in the developmental expression of this gene as endothelial cells differentiate from hemangioblastic precursors. If this is the case, it is likely to interact functionally with other, more cell type-specific transcription factors, as is true for developmentally regulated erythroid genes (38).
With the exception of areas of Sp1-induced protection, the pattern of in vivo DNase I sensitivity of the human KDR/flk-1 5Ј-flanking sequence in endothelial cells is essentially identical to that of genomic HUVEC DNA exposed to DNase I in vitro, suggesting that the active promoter is not bound by nucleosomes (39). In contrast, the 10 -11-bp helical repeating pattern of in vivo DNase sensitivity in conjunction with the lack of protection from DMS modification in vivo indicates that the promoter is likely to be associated with positioned nucleosomes in nonexpressing cells. Indeed, helical periodicity determined by in vivo DNase I footprinting is a defining feature of nucleosomes present in the 5Ј-flanking regions of a number of genes (21,40). Future studies will be necessary to determine the exact rotational and/or translational positioning of nucleosomes throughout the KDR/flk-1 5Ј-flanking sequence in nonendothelial cells and how such nucleosomal positioning is regulated.
The general effect of nucleosomal positioning on Sp1 binding and trans-activation remains unclear. Although other nuclear factors may require Sp1 for efficient nucleosomal disruption and transcription initiation (41), it is unlikely that Sp1 itself is responsible for nucleosomal disruption, since Sp1 is ubiquitous in mammalian cells (15), and Sp1 alone does not disrupt preformed nucleosomes in vitro (42). It also has not been demonstrated that Sp1 can interact with binding sites that are bound by nucleosomes in vivo. Sp1 interacts with nucleosomal DNA in vitro, but its affinity for nucleosomal DNA is reduced by greater than 1 order of magnitude compared with its affinity for naked DNA (42), and transcription initiation by Sp1 from preformed nucleosomes is strongly impaired or prevented (43). Conversely, nucleoplasmin-induced disruption of nucleosomal FIG. 8. Concordance of in vivo footprinting by DNase I and DMS. In vivo DMS and DNase I footprinting were performed as described in Figs. 6 and 7 and analyzed simultaneously on the same polyacrylamide gel. An example of reactions using the P5Ј-primers is shown. Rightward arrows denote protected guanine residues, and leftward arrows mark hypersensitive residues. The most 5Ј-region protected from DNase I attack in HUVECs (corresponding to Sp1 IV) is denoted with brackets. cores has been shown to enhance Sp1 binding in vitro (44).
Our data indicate that Sp1 is specifically excluded from binding the KDR/flk-1 promoter in nonendothelial cells that do not express this gene, and we have evidence that this exclusion is associated with changes in chromatin structure. A direct association between the two is difficult to prove in vivo, however. In this regard, results of transgenic experiments in mice using a 4.0 kilobase KDR/flk-1 promoter:LacZ fusion gene are informative. All progeny containing this transgene fail to express LacZ in all cell types examined including endothelial cells, although KDR/flk-1 mRNA is easily detectable in endothelial cells of these mice. 2 This would suggest that the proximal 4 kilobases of 5Ј-flanking sequence alone, although potent transcriptionally in transient assays of gene expression (14), is transcriptionally silent in a chromosomal context, and that sequences outside this region are required for normal expression of KDR/flk-1. A plausible model, therefore, is that distant sequences regulate KDR/flk-1, either by disrupting chromatin structure to allow Sp1 to bind the core promoter, or by allowing Sp1 to bind before chromatin structure is established. In either event, an open structure for the gene is preserved in endothelial but not in nonendothelial cells.
ALthough a number of genes are expressed with varying degrees of specificity in endothelial cells, we know of no transcription factors that function exclusively in endothelial cells. We postulate that novel factors, or a precise dosage of more general factors, initiate KDR/flk-1 expression in angioblastic precursors early in development in the course of their commitment to the endothelial cell lineage, thus rendering this population of cells sensitive to the mitogenic and vasculogenic effects of VEGF. In our model, changes in chromatin structure within the core promoter, presumably regulated by elements at a distance from the transcription initiation site, are associated with cell type-specific regulation of this gene. A similar model, based on regulation via a locus control region, is used to explain gene expression within the ␤-globin locus in the developmentally related erythroid lineage (45,46). If, as our results suggest, KDR/flk-1 is indeed regulated by a mechanism similar to that of the globin locus, then methods used to characterize regulatory elements within the globin locus, such as DNase I hypersensitivity assays and heterologous and homologous transgenic promoter analysis, may be particularly useful in elucidating the lineage-restricting mechanisms governing KDR/flk-1 as well. In addition, the results of our experiments indicate that one or more functional elements should regulate chromatin structure such that the KDR/flk-1 gene is accessible and expressed specifically in endothelial cells. Delineation of these, presumably promoter-distal, elements is of paramount importance to understand the mechanisms of lineage-specific expression of the human KDR/flk-1 gene.