Repression of platelet-derived growth factor A-chain gene transcription by an upstream silencer element. Participation by sequence-specific single-stranded DNA-binding proteins.

Platelet-derived growth factor A-chain is a potent mitogen expressed in a restricted number of normal and transformed cells. Transient transfection and deletion analysis in BSC-1 (African green monkey, renal epithelial) cells revealed that the -1680 to -1374 region of the A-chain gene repressed homologous and heterologous promoter activities by 60-80%. An S1 nuclease-hypersensitive region (5'SHS) was identified within this region (-1418 to -1388) that exhibited transcriptional silencer activity in BSC-1 and a variety of human tumor cell lines (U87, HepG2, and HeLa). Electrophoretic mobility shift assays conducted with 5'SHS oligodeoxynucleotide probes revealed several binding protein complexes that displayed unique preferences for binding to sense, antisense, and double-stranded forms of the element. Southwestern blot analysis revealed that the antisense strand of 5'SHS binds to nuclear proteins of molecular mass 97, 87, 44, and 17 kDa, whereas the double-stranded form of 5'SHS is recognized by a 70-kDa factor. Mutations within 5'SHS element indicated the necessity of a central 5'-GGGGAGGGGG-3' motif for protein binding and silencer function, while nucleotides flanking both sides of the motif were also critical for repression. These results support a model in which silencer function of 5'SHS is mediated by antisense strand binding proteins, possibly by stabilizing single-stranded DNA conformations required for interaction with enhancer sequences in the proximal promoter region of the A-chain gene.

three isoforms (AA, BB, and AB). PDGF B-chain is the product of the c-sis proto-oncogene and is located on human chromosome 22 (2,3), whereas the homologous A-chain gene is found on human chromosome 7 (4,5). Expression of the A-chain has been detected in normal endothelial, epithelial, smooth muscle, neuronal, and glial cells, suggesting various functions in cell differentiation, embryogenesis, and wound healing (4,6,7). PDGF is also strongly implicated as an etiologic factor in such disparate fibroproliferative diseases as atherosclerosis, glomerulonephritis, pulmonary fibrosis, and neoplasia (8 -10).
Transcription of the PDGF A-chain gene can be stimulated or inhibited by a variety of growth factors and cytokines (reviewed in 11). The transcriptional mechanisms mediating these effects and cell-specific patterns of A-chain expression are not fully understood, although accumulating evidence indicates that the transcriptional regulation of the A-chain gene is governed by a complex interaction of positive and negative regulatory elements. A highly GC-rich region in the proximal 5Ј-flanking sequence of the human A-chain promoter (Ϫ80 to Ϫ50) has been shown to contribute over 80% promoter activity (12)(13)(14), probably mediated through the binding of Sp1-like proteins. Deletion mutagenesis and transient transfection analyses (12,13) have indicated that the gene is also subject to transcriptional repression by two discrete negative regulatory elements located in more distal areas of the 5Ј-flanking DNA (Ϫ1853 to Ϫ1032 and Ϫ1031 to Ϫ883). In addition, an S1 nuclease-hypersensitive (SHS) region has been localized within the first intron (ϩ1605 to ϩ1630; intron SHS or int-SHS) and demonstrated to exhibit transcriptional silencer activity in HeLa, but not human glioblastoma (A172) cells, suggesting a cell-specific function (15).
In this study, we have employed deletion mutagenesis, S1 nuclease hypersensitivity assay, and transient transfection to identify a 31-bp silencer element in the distal 5Ј flanking region (Ϫ1418 to Ϫ1388; 5ЈSHS) that represses transcription of the A-chain promoter in a variety of cell lines. Mutational analysis of the 5ЈSHS element suggests that a centrally located GGG-GAGGGGG motif, found in both 5ЈSHS and int-SHS elements, is required both for binding of antisense strand-specific proteins and for full silencer function. Furthermore, we have determined that 5ЈSHS and int-SHS bind to the same antisense strand-specific binding proteins (p97, p87 and p17), strongly suggesting a common mechanism of transcriptional silencing by these topologically distant DNA elements. These results highlight the role of transcriptional repression in the regulation of A-chain gene expression and indicate the importance of single strand DNA-specific binding proteins in eukaryotic gene transcription.

MATERIALS AND METHODS
DNA Sequencing and Plasmid Constructs-Construction of the plasmids pAC261 and pUTKAT3 has been described (12,16). The plasmid pGL261 was constructed by inserting an XhoI-HindIII fragment of the PDGF A-chain promoter fragment (Ϫ261 to ϩ8) into pGL2-basic (Promega), a promoterless plasmid containing the luciferase reporter gene. To localize DNA elements in the upstream negative regulatory region involved in repression of PDGF A-chain transcription, the Ϫ1853 to Ϫ883 region was digested with restriction endonucleases AccI and BssHII into four fragments, designated A (Ϫ1853 to Ϫ1716), B (Ϫ1680 to Ϫ1374), C (Ϫ1373 to Ϫ1032), and D (Ϫ1031 to Ϫ883). The four fragments were inserted in both orientations into the XbaI site of pUTKAT3 and the NheI site of pGL261. DNA sequence in the Ϫ1853 to Ϫ883 region of the PDGF A-chain promoter was obtained by doublestranded DNA sequencing of plasmids containing fragments A-D by the Sequenase procedure (Amersham Corp.). Wild type and mutated 5ЈSHS oligodeoxynucleotides and their complementary strands were synthesized (Integrated DNA Technologies), annealed, and inserted in both orientations 5Ј to the cat gene into the XbaI site of pAC261. The wild type 5ЈSHS was also inserted in both orientations 3Ј to the cat gene into the BamHI site of pAC261, and 5Ј to the cat sequence at the XbaI site of pUTKAT3.
All plasmids used in transient transfection were purified by an ion-exchange chromatographic procedure (Qiagen). BSC-1 and HepG2 cells were transfected by calcium phosphate/DNA co-precipitation as described previously (12,17). Cells were harvested 48 h after transfection and were lysed for cat assays by three freeze-thaw cycles in 100 l of 0.25 mM Tris, pH 7.8, per dish. For luciferase assays, cells were lysed in 200 l of 1 ϫ reporter lysis buffer (25 mM Bicine, pH 7.6, 0.05% Tween 20, 0.05% Tween 80).
U87 and HeLa cells were transfected with Lipofectin reagent (Life Technologies, Inc.) as described (18). U87 cells were plated 24 h prior to transfection on fibronectin-coated (1 g/cm 2 for 1 h, Sigma) 60-mm diameter tissue culture dishes at a density of 4 ϫ 10 5 cells per dish. HeLa cells were plated at a density of 1.4 ϫ 10 6 cells per 60-mm dish 24 h prior to transfection. Lipofectin was added to DMEM at a concentration of 16 l/ml, mixed with an equal volume of DMEM containing 1.5 g of pRSV-␤gal, and 3.5 g of test plasmid, and the DNA-Lipofectin mixture was incubated at room temperature for 10 min. Cells were rinsed twice with 1 ϫ PBS and incubated with the Lipofectin-DNA mixture (1.6 ml per dish) at 37°C for 5 h. An equal volume of DMEM containing 10% FBS was then added to the dishes, and the cells were harvested 60 h after transfection.
CAT, ␤-Galactosidase and Luciferase Assays-Assays of cat and ␤-gal activity were performed on cell lysates as described (12,19,20). Relative cat activity represents the percent conversion of [ 14 C]chloramphenicol to acetylated products after correction for transfection efficiency by ␤-gal activity. Luciferase assays were conducted as described by the manufacturer (Promega). Twenty-l aliquots were assayed by injection of 100 l of reconstituted luciferase assay reagent ( S1 Nuclease Hypersensitivity Assay-S1 nuclease hypersensitivity (SHS) assays were performed as described (21). To localize regions of the upstream PDGF A-chain promoter that were hypersensitive to S1 nuclease, mapping of SHS sites was performed with 1 g of supercoiled pGL261ϩB plasmid DNA in 50 l of S1 buffer (30 mM NaOAc, pH 4.5, 100 mM NaCl, 0.2 mM EDTA, and 3 mM Zn(OAc) 2 ) with increasing amounts of S1 nuclease (0.1, 0.5, 1 or 5 units) at 37°C for 10 min. S1 nuclease-nicked DNA was end-labeled on the antisense strand by linearization with MluI and incubation with Klenow fragment of DNA polymerase (Life Technologies, Inc.) and [␣ 32 P]dCTP (DuPont NEN). Radiolabeled DNA was digested with XhoI and the 300-bp product isolated by electrophoresis. The sense strand was radiolabeled in similar fashion, except that S1 nuclease-nicked DNA was treated with XhoI prior to end labeling, followed by digestion with MluI. SHS sites were visualized by electrophoresis on a 6% denaturing gel followed by autoradiography. Sites of cleavage were determined relative to a sequencing ladder obtained by DNA sequencing of pGL261ϩB by the Sequenase procedure.
Nuclear Extract and Electrophoretic Mobility Shift Assay-Nuclear extracts were prepared from HepG2, U87, HeLa, BSC-1, and A172 cells by the method of Dignam et al. (22). Electrophoretic mobility shift assays (EMSAs) were conducted as described (23). A 36-bp irrelevant DNA sequence (5Ј-GGGCCGGCGGACTCGCACGGTGGGTTTTAACC-TTTT-3Ј) was used as a nonspecific competitor. Competitive EMSAs with wild type and mutated 5ЈSHS were scanned with an M2 densitometer (Imaging Research Inc.) to quantify radioactivity in protein-DNA and free probe bands.
Southwestern Blotting-Southwestern blotting was performed as described (24). Crude nuclear extract (50 g) was mixed with Dignam buffer D and 6 l of 4 ϫ sample loading buffer (8% SDS, 28% glycerol, 0.32 M Tris, pH 6.8, 0.2% bromphenol blue, 10 mM DTT) to a final volume of 24 l, heated at 95°C for 5 min, and then loaded onto a denaturing SDS-polyacrylamide gel (3% stacking, 10 or 15% separating) in a buffer of 25 mM Tris, 192 mM glycine, 0.1% SDS. After electrophoresis, gels were incubated for 30 min in gel transfer buffer (25 mM Tris, 192 mM glycine, 20% methanol) and blotted to nitrocellulose membranes with a Bio-Rad transfer cell at 20 V for 1 h. Membranes were blocked for 1 h at room temperature in TNED (10 mM Tris, pH 7.5, 50 mM NaCl, 0.1 mM EDTA, 1 mM DTT) containing 5% Carnation nonfat dry milk, washed in TNED for 5 min, and placed in a sealed plastic bag containing 5 ml TNED, 5 ϫ 10 5 cpm/ml oligodeoxynucleotide probe, with or without unlabeled competitor DNA. Membranes were incubated with the probe mixture for 1 h at room temperature, washed twice with TNED for 10 min, and air-dried. Bands were visualized by autoradiography.

Localization of Multiple Negative Regulatory Regions in the PDGF A-chain
Promoter-We previously employed 5Ј-end point deletion analysis to identify two regions in the 5Ј-flanking sequence of the A-chain promoter that exhibited negative regulatory activity, designated NRE1 (Ϫ1853 to Ϫ1032) and NRE2 (Ϫ1031 to Ϫ883) (12). The DNA sequence of the Ϫ1853 to Ϫ883 region was obtained (Fig. 1), and the region was digested into four fragments, A, B, C, and D, with restriction endonucleases AccI and BssHII ( Fig. 2A). These fragments were subcloned into pGL261, which contains the Ϫ261 to ϩ8 segment of the PDGF A-chain promoter and the luciferase reporter gene. BSC-1 cells, which exhibit considerable A-chain expression when maintained in basal serum-containing culture medium conditions (12), were transiently transfected with these constructs. Fragment A had no effect on luciferase activity when oriented correctly with respect to the A-chain promoter (sense) but modestly inhibited transcription in the opposite orientation (Fig. 2B). Fragment B displayed the most inhibitory activity of the four fragments, reducing luciferase activity to 35% of pGL261 when oriented correctly, although it failed to inhibit transcription in the opposite orientation. Fragment C did not affect transcriptional activity, whereas fragment D displayed a 50% inhibition in either orientation.
To test the ability of fragments A-D to repress transcription of a heterologous promoter, they were inserted into the plasmid pUTKAT3 that contains a promoter/enhancer segment of the thymidine kinase gene from herpes simplex virus (HSV-tk) and the cat reporter gene (16). Fragment A modestly inhibited tk promoter activity in both orientations (Fig. 2C). Fragments B and C were much more active than fragment A in their sense orientations, decreasing cat activity to 20 and 35% of the pUT-KAT3 parent vector, respectively. In their negative orientations, fragments B and C also inhibited the tk promoter, albeit to a lesser extent, decreasing cat activity to 55 and 75% of the parent construct. Fragment D also significantly inhibited tk promoter activity but only in the opposite orientation. These results indicate the presence of multiple subregions in the Ϫ1853 to Ϫ883 fragment of the PDGF A-chain gene that repress transcription. Fragment B exerted the strongest inhibitory activity upon both promoters, and its ability to inhibit transcription in a distance-independent manner suggested that it could be liberally characterized as a classical silencer element.

Localization of a Predominant SHS Site(Ϫ1416 to Ϫ1399) in the Upstream Negative Regulatory
Region-Hypersensitivity to S1 nuclease has proven useful in mapping cis-acting DNA transcriptional elements, since single-stranded, non-B form DNA structures often serve as recognition sites for binding of trans-acting factors. When a supercoiled plasmid containing the upstream negative regulatory region of the A-chain promoter (Ϫ1836 to Ϫ814) was treated with S1 nuclease and then digested with BamHI, a single, predominant S1 nuclease-hypersensitive (SHS) site was revealed within the fragment B region, at approximately Ϫ1400 (data not shown). To localize this SHS site, high resolution mapping was performed as described under "Materials and Methods" with the plasmid pGL261ϩB, which contains the fragment B region (see Fig.  2B). S1 nuclease hypersensitivity was observed on both DNA strands over an 18-bp region (Ϫ1416 to Ϫ1399), consistent with the location identified by crude mapping. The antisense strand was particularly accessible to S1 nuclease, with two adjacent nucleotides A and C displaying the most sensitivity and two nearby nucleotides T and C showing a lesser degree (Fig. 3, left). Cleavage induced at the site was directly correlated with increasing concentrations of S1 nuclease, while none was observed in the absence of enzyme (Fig. 3, lane 1). Two G residues on the sense strand were S1 nuclease-hypersensitive, with two nucleotides A and C exhibiting lesser sensitivity (Fig. 3, right). The DNA sequence in this region bore considerable homology to an SHS silencer element previously identified in the first intron of the A-chain gene, including an identical GGGGA-GGGGG motif ( Fig. 4A; Ref. 15). Hereafter, the upstream SHS region will be designated 5ЈSHS and intron silencer element as int-SHS.
The 5ЈSHS Element Functions as a Silencer in Multiple Cell Lines-To determine whether the 5ЈSHS region harbored transcriptional silencer activity, a double-stranded 31-bp oligodeoxynucleotide encompassing the 5ЈSHS (Ϫ1418 to Ϫ1388) was synthesized and inserted either 5Ј-or 3Ј-to the A-chain promoter in pAC261 and 5Ј-to the tk promoter in pUTKAT3. Transient transfection of BSC-1 cells with these plasmids revealed a significant ability of the 5ЈSHS sequence to repress the A-chain promoter, reducing cat activity to approximately 40% of control in both orientations when placed upstream of the promoter. The 5ЈSHS also repressed promoter activity in the correct orientation downstream of promoter and cat sequences but significantly enhanced transcription (1.8-fold) in the oppo-site orientation. The 5ЈSHS inhibited transcription from the heterologous tk promoter in both orientations, decreasing cat activity to less than 40% in the opposite orientation and to a lesser extent when oriented correctly (85%, Fig. 4B). The 5ЈSHS can thus be defined as a classical silencer element based on its ability to repress homologous and heterologous promoters, to function in a distance-independent manner, and to inhibit transcription in both orientations with respect to the A-chain promoter.
To assess 5ЈSHS silencer activity across a panel of cell lines that express varying amounts of A-chain, HeLa (low expression), HepG2 (intermediate), and U87 (high) cells were transiently transfected with plasmid constructs 5ЈSHS ϩ /pAC261 and 5ЈSHS Ϫ /pAC261. As previously observed in BSC-1 cells, the 5ЈSHS element reduced cat activity to 40 -50% of control in all three cell lines and both orientations (Fig. 4C). The ability to repress transcription in multiple, unrelated cell lines indicates a ubiquitous role of 5ЈSHS as a silencer.
5ЈSHS and int-SHS Bind the Same Single-stranded and Double-stranded DNA-binding Proteins from BSC-1 Cells-To identify nuclear factors that bind to 5ЈSHS, EMSAs were performed with BSC-1 cell nuclear extract and oligodeoxynucleotide probes derived from the 5ЈSHS. To address whether 5ЈSHS and int-SHS bind to the same proteins, an oligodeoxynucleotide containing the int-SHS sequence was also included. Double-stranded 5ЈSHS probe (ds-5ЈSHS) formed two distinct DNA-protein complexes in the absence of unlabeled competitor (Fig. 5A, lane 1). The intensities of these two bands were both reduced by more than 80% in the presence of unlabeled ds-5ЈSHS competitor at a 10-fold molar excess (lane 2), with complete displacement at a 100-fold molar excess (lane 3). Unlabeled, double-stranded int-SHS oligodeoxynucleotide (dsint-SHS) also competed at a 100-fold molar excess with ds-5ЈSHS probe for binding to the two proteins but was not as efficient as unlabeled ds-5ЈSHS at a 10-fold molar excess (lanes 4 and 5). Single-stranded (sense or antisense) forms of 5ЈSHS and int-SHS did not compete with ds-5ЈSHS probe for protein binding (lanes 6 -9), suggesting high affinity binding requires contacts furnished by both DNA strands. An irrelevant 36-mer oligodeoxynucleotide in single-stranded form did not compete with ds-5ЈSHS probe for binding to these two complexes (lane 10). The double-stranded form of the irrelevant oligodeoxynucleotide competed inefficiently at a 100-fold molar excess for binding to complex 1 (lane 11), at a level less than that seen with a 10-fold excess of homologous ds-5ЈSHS competitor. The irrelevant oligodeoxynucleotide competed somewhat more efficiently for binding to complex 2, suggesting this binding interaction was less sequence-specific.
The S1 nuclease hypersensitivity of 5ЈSHS suggested that proteins with specificity for single-stranded structures might bind to this region. Thus, EMSAs were conducted with singlestranded oligodeoxynucleotide probes derived from both strands of the 5ЈSHS. Antisense 5ЈSHS probe (as-5ЈSHS) formed two distinct DNA-protein complexes (Fig. 5B, lane 1) with significantly greater mobility than those formed with ds-5ЈSHS probe. Efficient dose-dependent competition of binding was observed with unlabeled homologous competitor (lanes 2 and 3), indicating a high affinity interaction between these two protein complexes and as-5ЈSHS. Unlabeled ds-5ЈSHS competed to a lesser extent (lanes 4 and 5), indicating the affinity of these complexes was higher for the antisense strand form. The as-int-SHS sequence competed effectively with as-5ЈSHS probe for protein binding (lanes 6 and 7), strongly suggesting that the antisense strands of both 5ЈSHS and int-SHS are recognized as high affinity binding sites for the same nuclear proteins. Double-stranded int-SHS competed less efficiently for binding of these proteins to the as-5ЈSHS probe at a 100-fold molar excess (lanes 8 and 9). Neither the sense strand of int-SHS (lane 10) nor nonspecific competitor (lanes 11 and 12) significantly displaced protein binding to the antisense 5ЈSHS probe, indicating the sequence specificity of these complexes. EMSAs conducted with 5ЈSHS sense probe (s-5ЈSHS) revealed the formation of numerous DNA-protein complexes, only one of which was specifically displaced by a 100-fold molar excess of unlabeled s-5ЈSHS competitor (Fig. 5C, lanes 1-3).
The ability of unlabeled int-SHS oligodeoxynucleotides to compete with radiolabeled 5ЈSHS probes for protein binding suggested that the 5ЈSHS and int-SHS sequences bind the same set of proteins. This conclusion was reinforced by EMSAs conducted with radiolabeled as-int-SHS probes, in which a doublet of bands identical in mobility to that formed with as-5ЈSHS probes was observed (data not shown). Taken together, the EMSA analyses suggest that the antisense strands of 5ЈSHS and int-SHS are recognized as high affinity binding sites for the same set of nuclear proteins, whereas sense and double-stranded forms of 5ЈSHS are recognized by their own cognate DNA-binding proteins.
Mutations within the 5ЈSHS Reveal Different Nucleotide Sequence Requirements for Binding of Double-strand, Antisense Strand, and Sense Strand DNA-binding Proteins-To identify individual nucleotides within the 5ЈSHS sequence required for protein binding, the ability of a panel of mutant 5ЈSHS oligodeoxynucleotides to compete for binding to radioactive 5ЈSHS probes was evaluated (Fig. 6). Disruption of the GGGGAGGGG core (m2) resulted in a complete loss in ability to compete for binding to both ds-5ЈSHS and as-5ЈSHS probes (Fig. 7, A and  B), whereas mutation of the 5Ј-and 3Ј-portions of the 5ЈSHS (m1 and m3, respectively) had little effect. This indicates that the GGGGAGGGG core comprises the recognition site for the 5ЈSHS binding proteins, whereas the nucleotides flanking the core are not critical for binding.
Smaller mutations revealed subtle differences in nucleotide sequence requirements for binding to double-stranded, sense, and antisense forms of the 5ЈSHS (Fig. 7, C-E). While a 2-bp mutation in the region 5Ј to the core sequence (m4) did not affect binding to any 5ЈSHS binding proteins, a 2-bp mutation (m5) just 3Ј to m4 significantly diminished binding to anti-  3. Identification of an S1 nuclease-hypersensitive site in the upstream negative regulatory region. Plasmid pGL261ϩB was treated with S1 nuclease and then 3Ј-end-labeled on either the antisense or the sense strand as described under "Materials and Methods" to reveal SHS sites. Bands corresponding to hypersensitive sites are indicated by black arrows, with arrow size indicating relative extent of cleavage. The DNA sequence of the SHS region is displayed at the bottom of the figure, with two inverted repeats (5Ј-GACGT-3Ј and 5Ј-TGCAG-3Ј) indicated by horizontal arrows.
sense-strand binding proteins (D) and abolished binding to the sense-strand binding protein (E). The ability of the m5 mutation to affect binding is somewhat unexpected, since these two nucleotides were altered in the m1 mutant without effect (A and B). The apparent disparity could be related to the different nucleotide substitutions employed in these two mutants. Mutation of the 5Ј portion of the core sequence (m6: GGGGA to CTCTT) inhibited binding to all three forms of 5ЈSHS probe, whereas mutation of the 3Ј-portion of the core sequence (m7: GGGGG to CTCTC) or the centrally located A residue of the core (m8) decreased binding to ds-5ЈSHS and as-5ЈSHS but not s-5ЈSHS. A 5-bp mutation in the region flanking the 3Ј-end of the core sequence (m9) did not affect binding of any 5ЈSHS binding proteins. Although oligodeoxynucleotides m5-m8 all exhibited partial reductions in affinity for the antisense binding protein complexes, none exhibited a complete loss, suggesting that the antisense strand DNA recognition site is multivalent over the GTGGGGAGGGGG core sequence.
The different binding specificities of these proteins can be summarized as follows. Binding of complexes 1 and 2 to ds-5ЈSHS (Fig. 5A) appears to require the entire core sequence as well as the central A residue. Similarly, complexes 1 and 2, which bind to the antisense strand of the 5ЈSHS (Fig. 5B), require both halves and central A of the core sequence but also appear to require the presence of a GT pair (nucleotides Ϫ1411 and Ϫ1412) just 5Ј to the core. Finally, binding of the single complex to the sense strand (Fig. 5C) requires the 5Ј-portion of the core and the adjacent GT pair but neither the central A residue nor the 3Ј-half of the core.
Both the Core GGGGAGGGGG Sequence and Nucleotides Flanking the Core Are Required for 5ЈSHS Silencer Function-To explore DNA sequence requirements for silencer function of 5ЈSHS, mutant oligodeoxynucleotides m1 through m9 were inserted into pAC261 in both orientations and analyzed by transient transfection of BSC-1 cells. Mutant oligodeoxynucleotides m1, m2, and m3, in which nucleotide sequence of the 5ЈSHS was disrupted in three large discrete blocks (Fig.  6), all displayed significant decreases in silencer activity in either orientation (Fig. 8A). m1 and m3, which contained mutations of the sequences flanking the 5Ј and 3Ј sides of the GGGGAGGGGG core, respectively, exhibited almost total loss of silencer activity, whereas mutation of the core sequence (m2) resulted in nearly a 50% decrease (Fig. 8A). A nearly total loss in silencer function was observed when two adjacent nucleotides 5Ј to the core (m4) were mutated (Fig. 8A). Taken together, these observations indicate that nucleotides required for 5ЈSHS function are dispersed over the entire DNA element. A second panel of oligodeoxynucleotides containing smaller mutations (m5-m9) failed to display any alterations in silencer activity, with the exception of a partial loss observed when m5 was inserted in the opposite orientation relative to the promoter (Fig. 8B). The loss in 5ЈSHS silencer activity seen when the core sequence was mutated in toto (m2) correlated with loss in binding of nuclear proteins to the antisense strand, suggesting that binding of these factors is required for full function of the element. In contrast, mutations that ablated binding of double-stranded (m6 and m7) and sense strand (m5) binding proteins had no appreciable effect on silencer activity, suggesting these proteins were not required. Two mutations that completely disrupted silencer function, m1 and m3, had no effect on binding of double-stranded or antisense forms (Fig. 7).
Southwestern Blot Analysis of Nuclear Proteins Binding to 5ЈSHS-To further characterize the proteins that bind the 5ЈSHS element, Southwestern blot analysis was conducted with the single-stranded and double-stranded probes employed in EMSA analyses. Four cell lines exhibiting a range of A-chain expression, HepG2 (intermediate A-chain expression), U87 (high), HeLa (low), and BSC-1 (high), were examined. When equal concentrations of nuclear extract from these cells were subjected to denaturing SDS-polyacrylamide gel electrophoresis, blotted to nitrocellulose membranes, and probed with radiolabeled as-5ЈSHS, four protein species of molecular mass 97,000, 87,000, 44,000, and 17,000 (p97, p87, p44, and p17) were identified (Fig. 9A). These bands represented specific DNA-protein complexes, as demonstrated by displacement with a 100-fold molar excess of unlabeled homologous competitor DNA (data not shown). Whereas nearly equivalent concentrations of p97 were observed across these cell lines, BSC-1 cells contained significantly less p44 (Fig. 9A). p87 was ob- served in HeLa nuclear extracts and to a lesser extent in BSC-1, but was not observed in HepG2, U87 (Fig. 9A), or A172 cells (data not shown). Hybridization of blots with as-int-SHS oligodeoxynucleotide probes revealed three proteins (p97, p87, and p17), identical in molecular weight to those seen with the as-5ЈSHS probe (data not shown), suggesting that these elements are recognized by the same set of three antisense strandspecific binding proteins. An additional as-int-SHS binding protein species (p37) was observed that was not found in 5ЈSHS-probed blots.
When ds-5ЈSHS was employed as probe, binding to a 70-kDa protein was seen at approximately equal concentrations in all cell lines examined (Fig. 9B). Two additional species were observed in HeLa cells (p48 and p44), with a small amount of p48 also seen in HepG2 cells. p97, p87, and p17 were not detected, however, indicating these proteins bind preferentially to the antisense strand of 5ЈSHS. Conversely, absence of p70 from the as-5ЈSHS-probed blot demonstrates specificity of this protein for the double-stranded form of 5ЈSHS (Fig. 9A). No specific DNA-protein complexes were detected with s-5ЈSHS probe (data not shown), suggesting that the sense strand binding protein detected by EMSA is sensitive to denaturing SDSpolyacrylamide gel electrophoresis.

DISCUSSION
In this report, deletion analysis and S1 nuclease hypersensitivity assay have been employed to localize a silencer element, designated 5ЈSHS, to nucleotides Ϫ1418 to Ϫ1388 in the upstream promoter region of the PDGF A-chain gene. The 5ЈSHS bears homology to a recently described silencer in the first intron of the A-chain gene (int-SHS, 15). Our study extends previous observations by demonstrating the contribution of single strand-specific DNA-binding proteins in the function of the A-chain silencers and by identifying the molecular weight species involved in binding to both elements. We have also employed site-directed mutagenesis to define DNA sequences required for binding of nuclear factors and silencer function. Multiple negative regulatory elements have been identified in the promoters of genes encoding human PDGF B-chain (25)(26)(27), chicken (28), hamster (29), and human (30) vimentin, human collagen II (31), interleukin-4 (32), and ⑀-globin (33), and rat glutathione P-transferase (34). The 5ЈSHS and int-SHS appear to be unique in their relatively large distance from the transcription start site (Ͼ1 kilobase) and locations both 5Ј (5ЈSHS) and 3Ј (int-SHS) to the start site. The appear- ance of the GGGGAGGGGG motif in these two topologically distant elements and the sharing of cognate binding proteins suggest possible interactions in repression of A-chain gene transcription.
Probing of several promoter regions by single strand-specific nucleases such as S1 has revealed the existence of non-B-form DNA, which often correlates with important transcriptional functions. However, the molecular mechanisms through which single-stranded DNA structures mediate the regulation of gene transcription remain obscure. Many transcription factors exhibit high affinity binding to single-stranded DNA, including myoD (35) and estrogen receptor (36,37). In addition, specific DNA sequences found in a number of promoters, such as ␤-casein (38), c-Myc (39), adipsin (40), and the serum response element of the PDGF A-chain (41) have been shown to exhibit considerable S1 nuclease sensitivity, suggesting the presence of single-stranded structure. More specifically, repressor/silencer function has been attributed to S1 nuclease-sensitive sequences and/or cognate binding proteins that exhibit preferential binding to single-stranded DNA, such as the intron 1 silencer (int-SHS) of the PDGF A-chain (15,42), and silencers of genes encoding lipoprotein lipase (43), mouse androgen receptor (44), vascular smooth muscle ␣-actin (45), and mouse myelin basic protein (46). A number of molecular structures have been proposed for unwound DNA, including simple single strands, junctions between B-and Z-DNA, and triplex forms such as H-DNA (reviewed in Ref. 47). In vivo formation and maintenance of these structures are presumed to require assistance in the form of superhelical DNA winding and/or the binding of effector molecules (i.e. protein, cRNA), since DNA melting is unlikely under normal conditions of intracellular pH, temperature, and ionic strength. Thus, it is noteworthy that the antisense strand of 5ЈSHS is S1 nuclease-hypersensitive when presented to the enzyme in supercoiled form but is not recognized by single-stranded DNA binding proteins in a double-stranded oligodeoxynucleotide. This suggests that the element may indeed require chromosomal superhelicity and/or binding of accessory factors, such as the 5ЈSHS antisense strand binding proteins for generation of single-stranded structure in vivo.
Mutation of the GGGGAGGGGG core sequence resulted in loss of both silencer activity and binding to antisense strandspecific binding proteins, suggesting an important role of these proteins in silencer function. Also implicating these proteins is the observation that the same antisense-specific 5ЈSHS binding proteins bound with high affinity to the int-SHS silencer element. Southwestern blot analysis has provided the first insight into the identity of proteins that bind to the 5ЈSHS and int-SHS silencers. The 5ЈSHS antisense binding protein p97 (Fig. 9A) is close in size to a 95-kDa species shown to bind a homologous silencer element (5Ј-CCCCCTCC-3Ј) in the vimentin promoter (28,30), raising the interesting possibility that p97 may represent a transcription factor or family of related factors involved in silencing expression of growth-related genes. Homologues of the 5ЈSHS core sequence (5Ј-CCCCTCCC-3Ј) in the rat prolactin and human urokinase-type plasminogen activator promoters have been proposed as binding sites for AP-2, although no AP-2 binding has been demonstrated (48 -50). Members of the heterogeneous nuclear ribonucleoprotein (hnRNP) family of proteins also exhibit affinity for DNA sequences similar to 5ЈSHS and int-SHS. hnRNP K is relevant since it binds DNA with higher affinity than RNA, exhibits preferred binding to C-rich, single-stranded DNA sequences such as the four direct repeats of the sequence CCCTCCCCA (CT element) of the c-myc gene (51), and has been recently implicated in c-myc transcriptional activation (51,52). hnRNP K is close in size (p68) to that of the 5ЈSHS binding protein p70 (Fig. 9), but its preference for single-stranded DNA is inconsistent with the double-stranded DNA specificity of p70. No hnRNPs have been identified in the molecular weight range of p97, p87, or p17. NSEP-1 also interacts with the C-rich strand of the c-myc CT element and is close in size (322 amino acids; Ref. 53) to the int-SHS binding protein p37. Detailed analyses will be required to conclusively identify the complex of transcription factors that interact with the 5ЈSHS and int-SHS silencers and to better understand the molecular mechanism mediating silencer function.
Mutation of regions flanking the core sequence (m1, m3, m4) inhibited silencer activity but did not affect binding to proteins specific for the 5ЈSHS, suggesting a role for these nucleotides in conferring a silencer-competent DNA structure and/or binding to proteins not detected by our EMSA analyses. Two inverted repeats (5Ј-GACGT-3Ј and 5Ј-TGCAG-3Ј) are found in the coreflanking sequence and contribute to a mirror symmetry over the whole region (Fig. 3), suggesting a potential for stabilizing a DNA triplex structure such as H-DNA. The prominent S1 hypersensitivity on the 5ЈSHS antisense strand in the region of the downstream inverted repeat is consistent with the H-DNA structure shown in Fig. 10, with a sharp turn of the DNA strand and relatively short length of single-stranded DNA on the antisense strand. Presumed contact points between DNA and binding proteins that were mapped by mutagenesis and EMSA studies are also represented in the model. Since many transcription factors bind to bent DNA, or may generate DNA bending upon binding (54 -58), it is tempting to speculate that formation of H-DNA or other single-stranded DNA structures in the 5ЈSHS region may serve to bring silencer element binding proteins into correct orientation with downstream positive regulatory elements.
The 5ЈSHS was equally efficient in repressing A-chain promoter activity in either orientation when placed 5Ј-to the promoter, conforming to a classical definition of a silencer element. The ability to function in upstream and downstream locations relative to the A-chain promoter in pAC261 further suggests position independence, although 5ЈSHS failed to repress transcription in the opposite orientation at a downstream location. The lack of strict adherence to the criteria of orienta- FIG. 10. A proposed model of 5SHS structure and sites of nuclear protein binding. Shown in this model are a non-B-DNA structure, consistent with the pattern of S1 nuclease hypersensitivity seen in Fig. 3, and contact points between nuclear factors that interact with the 5ЈSHS region, as mapped by mutagenesis and EMSA studies. The two inverted repeats are proposed to form a triplex that permits access of the antisense strand to S1 nuclease, hypersensitive sites of which are indicated by arrows. The proteins that interact with the 5ЈSHS sequence are abbreviated as follows: dsBP, double-stranded binding protein; sBP, sense strand binding protein; asBP, antisense strand binding protein.
The 5ЈSHS repressed transcription from promoter/enhancer fragments of the A-chain and TK genes, both of which contain GC-rich enhancers and TATAA boxes (64). This suggests an inhibitory interaction between 5ЈSHS binding proteins and enhancer transcription factors bound to the promoter-proximal GC-rich region (Ϫ82 to Ϫ42) and/or the general transcription factors of the RNA polymerase initiation complex. We showed previously that Sp1-like proteins bind with high affinity to the GC-rich region (12), suggesting this family of transcription factors may be the target of inhibitory action by 5ЈSHS binding proteins.
The 5ЈSHS silencer was active both in cell lines that express considerable amounts of A-chain (BSC-1, HepG2, U87) and in HeLa cells that express very little (15). This is consistent with our previous observation that the upstream negative regulatory region of the A-chain gene (Ϫ1853 to Ϫ883), which contains the 5ЈSHS, appears to function in cells regardless of their level of basal transcription (12). The ubiquitous silencer activity of the 5ЈSHS and its cognate binding proteins suggests a fundamental role in repressing A-chain expression, thus preventing inappropriate expression of this potent mitogen. In cells where A-chain expression is elevated, transcriptional silencing continues to oppose enhancement mediated by the proximal GC-rich region. Wang et al. (15) reported that the int-SHS functioned as a negative regulatory element in HeLa but not in human glioblastoma (A172) cells and ascribed a function to the sequence in the maintenance of cell-specific expression of the A-chain. Although the study compared silencer activity in only two transformed cell lines, it raises the possibility that tissue-and developmental-specific patterns of A-chain expression may be partially determined by transcriptional silencing. The Southwestern blot analysis has provided preliminary identification of proteins that may participate in these aspects of A-chain expression. Although p97 was expressed at similar levels in all cell lines examined, concentrations of p87, p44, p17 and the as-int-SHS-binding protein p37 varied considerably, suggesting cell-specific possible roles in transcriptional repression. Taken together, our observations of structural and functional similarity between the 5ЈSHS and int-SHS suggest that the 5ЈSHS and int-SHS and their cognate trans-acting factors may cooperate to affect cell-specific expression of the A-chain gene.
PDGF is established as a mediator of malignant growth in astrocytoma and sarcoma (65,66), with elevated expression of PDGF A-and B-chains often seen in transformed cells. Loss of silencer function and ensuing A-chain overexpression could be envisioned, therefore, to constitute an important step in tumor progression. Sharing of protein components (i.e. p97, p87, and p17) between two separate silencer elements (5ЈSHS and int-SHS) would appear to make the A-chain gene vulnerable to such a loss of transcriptional control. The recent observation that the 95-kDa silencer element binding protein of the vimentin gene is absent from the metastatic breast cancer cell, MDA-MB-231, indeed suggests that such a mechanism may mediate progression to the metastatic phenotype (67). Also relevant is a recent observation that overexpression of PDGF B-chain confers a metastatic phenotype to human T98G glioblastoma cells (68). Characterization of nuclear factors binding to the A-chain silencers and their roles in expression of A-chain in malignancy and tumor metastasis will be of interest in future studies.