NM23-H1 and NM23-H2 Repress Transcriptional Activities of Nuclease-hypersensitive Elements in the Platelet-derived Growth Factor-A Promoter*

The platelet-derived growth factor (PDGF)-A promoter is regulated by a number of GC-rich regulatory elements that possess non-B-form DNA structures. Screening of a HeLa cDNA expression library with the C-rich strand of a PDGF-A silencer sequence (5 (cid:1) -S1 nu-clease-hypersensitive site (SHS)) yielded three cDNA clones encoding NM23-H1, a protein implicated as a suppressor of metastasis in melanoma and breast carci-noma. Recombinant human NM23-H1 cleaved within the 3 (cid:1) -portions of both 5 (cid:1) -SHS strands in either single-stranded or duplex forms. In contrast, NM23-H2, known as a transcriptional activator with a DNA cleavage function, cleaved within the 5 (cid:1) -portions of both strands, re-vealing that NM23-H1 and NM23-H2 cleave at distinct sites of the 5 (cid:1) -SHS and by different mechanisms. NM23-H1 and NM23-H2 and CAT, (cid:3) -Galactosidase Assays— HepG2 cells were obtained from the American Type Culture Collection. Construction of plasmids pAC1800, pAC880, pAC261, pAC-F11, pAC-F47, 5 (cid:1) -SHS (cid:7) /pAC261 (8, 11), and pCMV (cid:3) gal (28) has been described. To construct the NM23-H1 expression plasmid pL2-H1, a 700-bp Eco RI- Kpn I cDNA fragment containing the full coding sequence of NM23-H1 was excised from pBluescript SK (cid:2) (HeLa (cid:4) -ZAP clone 2-1) and inserted into the Eco RI site of pL2. The NM23-H2 expression plasmid pL3-H2 was constructed by insertion of a 650-bp Eco RI/ Hin dIII NM23-H2 cDNA fragment from pSPORT (Invitrogen) into the corre- sponding site in pL3. Expression of both proteins was driven by the cytomegalovirus (CMV) promoter/enhancer. The cells were transfected in 60-mm dishes by the calcium phosphate/DNA co-precipitation method (29). The precipitates were prepared using 2 (cid:2) g of promoter/ CAT reporter plasmid, 4 (cid:2) g of NM23 expression plasmid, and 2 (cid:2) g of the (cid:3) -galactosidase-expressing plasmid pCMV (cid:3) . CAT and (cid:3) -galactosid- ase assays were performed as described (11).

The platelet-derived growth factor (PDGF) 1 family consists of three structurally similar glycoproteins (M r 30,000) that induce proliferation and other growth-related effects in cells of mesenchymal origin. These proteins arise from covalent dimerization of two PDGF subunits, designated the A-chain and B-chain, yielding the heterodimer PDGF-AB and two homodimers, PDGF-AA and PDGF-BB (1,2). PDGF was implicated in tumorigenesis following the discovery of high sequence homology between the PDGF B-chain (PDGF-B) and the viral oncogene, v-sis (for a review, see Ref. 3). Other studies suggest that both PDGF-A and PDGF-B may also mediate tumor progression to the metastatic phenotype (4,5).
Transcription of the PDGF-A gene is regulated by several enhancer and silencer elements that are poly-purine/pyrimidine-rich and possess a high degree of single-stranded, non-B DNA structure. Other laboratories (6,7) as well as our own (8) have demonstrated that a highly GC-rich and nuclease-hypersensitive element (PDGF-A NHE) in the proximal 5Ј-flanking sequence of the PDGF-A promoter (Ϫ82 to Ϫ40) contributes most of the basal transcriptional activity of the gene. This activity is mediated by the binding of members of the Sp1 family of transcription factors and can be induced in vascular endothelial cells by phorbol ester treatment through displacement of Sp1 and Sp3 by the early growth response factor Egr-1 (7,9) or repressed by binding of the Wilms' tumor gene product WT1 (10). More recently, we localized a GC-rich, nucleasehypersensitive silencer element in a more 5Ј-distal region of the PDGF-A promoter (Ϫ1488 to Ϫ1388), which we designated the 5Ј-S1 nuclease-hypersensitive silencer or 5Ј-SHS, that represses transcription in a variety of normal and transformed cell lines (11). The noncoding (pyrimidine-rich) strand of the 5Ј-SHS has been shown by Southwestern blot analysis to bind specifically to proteins of M r 97,000, 87,000, 44,000, and 17,000, and binding of these proteins is well correlated with silencer function.
The current study was undertaken to identify these 5Ј-SHS binding proteins and better characterize their potential transcriptional functions. To that end, a HeLa cell cDNA expression library was screened with a radiolabeled probe derived from the pyrimidine-rich strand of the 5Ј-SHS. This approach yielded multiple cDNA clones, each encoding the 17-kDa protein, NM23-H1. The nm23-H1 gene belongs to a family of which eight human members have been identified to date (reviewed in Ref. 12), including nm23-H1, nm23-H2, DR-nm23, nm23-H4, and nm23-H5. These genes appear to play critical roles in cellular proliferation (13,14), embryonic development (15,16), and differentiation (reviewed in Ref. 17) and have also received attention for their potential roles in oncogenesis and tumor metastasis (13, 18 -21). All NM23 proteins possess nucleoside diphosphate kinase activities, which direct the transfer of ␥-phosphates between nucleoside triphosphates and diphosphates (22). In addition, NM23-H2 was shown to be identical to PuF, a transcriptional activator of the c-myc protooncogene that binds to the nuclease-hypersensitive element of the c-myc promoter (c-myc NHE) (23). NM23-H2/PuF has recently been identified as an activator of several genes required for early myeloid differentiation through recognition of c-myc NHE-like elements (24). However, no DNA binding or transcriptional function has yet been ascribed to NM23-H1.
In the current study, we find that NM23-H1 recognizes both the 5Ј-SHS silencer and the NHE basal promoter element of the PDGF-A gene as substrates for DNA cleavage, directing its activity to the 3Ј-portions of the individual strands of these nuclease-hypersensitive elements. Interestingly, NM23-H2 also cleaves both elements, but its activity is directed to their respective 5Ј-terminal regions. NM23-H1 and NM23-H2 also repress transcriptional activity of the PDGF-A promoter via functional interactions with both elements, suggesting a potentially important role for these proteins in dampening expression of an oncogenic and metastasis-promoting growth factor.

EXPERIMENTAL PROCEDURES
Screening of a HeLa cDNA Expression Library for 5Ј-SHS Silencer Element-binding Proteins-A HeLa S3 cDNA expression library constructed in the expression vector Uni-Zap XR (Stratagene) was employed for DNA recognition site cloning. An oligodeoxyribonucleotide corresponding to the noncoding strand of the 5Ј-SHS silencer element was end-labeled with T4 polynucleotide kinase and [␥-32 P]ATP (3000 Ci/mmol), purified by polyacrylamide gel electrophoresis, and used for screening of protein replica filters as described (25). Clones (5 ϫ 10 5 ) were screened on 150-mm plates containing 0.7% agarose at a density of 5 ϫ 10 4 plaque-forming units/plate. Three to four cycles of limiting dilution and screening were conducted to obtain purified plaques containing cDNAs of interest. cDNA inserts were sequenced by the Sequenase procedure (U.S. Biochemicals, Inc.).
DNA Binding and Cleavage Assays-DNA cleavage assays and mapping of DNA cleavage sites were performed as described (26), using 10 -50 fmol of 32 P-labeled oligodeoxyribonucleotide/15 l of incubation mixture. To promote DNA cleavage activities of the NM23 proteins, 2.7 mM Mg 2ϩ and 27 mM KCl were included in the incubation buffer. Double-stranded probes were prepared as indicated in the text by either (a) radiolabeling at both 5Ј-termini after annealing or (b) radiolabeling the strand of interest followed by annealing to the unlabeled opposite strand. Annealing of single-stranded oligodeoxyribonucleotides into double-stranded forms was performed by denaturation of individual strands for 2 min at 94°C, followed by a gradual decrease to 23°C over a 4-h time span. Protein-DNA complexes, radiolabeled probes, and cleavage products were resolved by electrophoresis through 6% polyacrylamide gels in 0.5ϫ TBE buffer (45 mM Tris borate, pH 8.3, 1 mM EDTA), whereas mapping of DNA cleavage products was performed on denaturing 16% sequencing gels. Sequencing ladders were generated by established procedures (27). Radiolabeled bands were visualized with a PhosphorImager (Molecular Dynamics). The nucleotide sequences of oligodeoxyribonucleotides employed (only sense strand sequences are shown) were: 5Ј-SHS, 5Ј-CTAGAGACGTGGGGAGGGGGCCTGCAGG-TGTGT-3Ј; and PDGF-A NHE, 5Ј-CTAGAGGGGGCGGGGGCGGGGG-CGGGGGAGGGG T-3Ј.

Screening of a HeLa Cell cDNA Expression Library with the
Pyrimidine-rich Strand of the 5Ј-SHS Yields Three cDNA Clones, Each Encoding nm23-H1-Our previous studies indicated that the noncoding, pyrimidine-rich strand of the 5Ј-SHS silencer element (5Ј-SHS Py) bound to a number of proteins with specificity for single-stranded DNA and that binding by those proteins appeared to mediate silencer function (11). To identify those proteins, the method of DNA recognition site screening described by Singh et al. (25) was undertaken using a HeLa cDNA expression library (-Zap II, Stratagene) and a single-stranded, 32 P-labeled oligodeoxyribonucleotide 5Ј-SHS Py probe. Screening of ϳ5 ϫ 10 5 independent plaques yielded three clones that exhibited strong retention of the radiolabeled 5Ј-SHS Py probe. DNA sequencing and a search of the human genome revealed that all three contained a 456-bp open reading frame encoding NM23-H1 (for a review see Ref. 30). Two of the clones (HL1 and HL7) expressed identical 784-bp cDNA inserts containing a short segment of 5Ј-untranslated sequence (7 bp) and 321 bp of 3Ј-untranslated sequence (Fig. 1). The 3Ј-untranslated segment was 122 bp longer than that previously reported (31), suggesting a novel pattern of processing of the primary NM23-H1 transcript. The third clone (HL2) contained a 660-bp insert that contained shorter 5Ј-untranslated (2 bp) and 3Ј-untranslated (189 bp) segments, of which the latter was identical to that described previously (31).
NM23-H1 Cleaves the 5Ј-SHS Silencer Sequence, but with a Pattern of Nucleotide Specificity Distinct from NM23-H2-To study the interactions of NM23-H1 with the 5Ј-SHS element in detail, the protein was expressed in E. coli and was purified to near homogeneity, using sequential steps of anion exchange and HTP column chromatography (see "Experimental Procedures"). NM23-H1 eluted as a doublet of ϳ19 kDa from the HTP column with a gradient of increasing potassium phosphate ( Fig. 2A). Resolution of NM23-H1 into a doublet is related to the presence of phosphate, because a single 19-kDa band is observed after phosphate removal (data not shown). Using a 32 P-labeled, single-stranded 5Ј-SHS Py strand as the DNA substrate and conditions shown previously for NM23-H2 to promote its DNA cleaving activity and to reduce formation of stable complexes with DNA (see "Experimental Procedures"; Ref. 26), strong cleaving activity was demonstrated in NM23-H1-containing fractions (Fig. 2B). The peak in DNA cleaving activity co-eluted precisely with NM23-H1 protein at ϳ150 mM of the phosphate gradient, strongly suggesting that NM23-H1 was indeed the active species.
The HTP-purified preparation of NM23-H1 was next ana-lyzed in more detail for its ability to cleave either of the individual strands of the 5Ј-SHS (5Ј-SHS Pu and 5Ј-SHS Py) in their single-stranded forms. In addition, considering the probability that these single-stranded oligodeoxyribonucleotides would form secondary structures not found in the natural context of duplex DNA, the double-stranded form of 5Ј-SHS was also evaluated. NM23-H1 cleaved both the single-stranded 5Ј-SHS Pu and 5Ј-SHS Py DNAs in a dose-dependent manner (Fig. 3A, lanes 1-12). Increasing amounts of NM23-H1 protein (0.1-2.5 g) yielded bands of progressively faster electrophoretic mobility from both individual 5Ј-SHS strands, suggesting a processive nuclease activity. The ability of NM23-H1 to cleave such distinct single-stranded DNA sequences (Pu-rich versus Py-rich) indicated recognition of structural features distinct from the primary nucleotide sequence per se. Co-incubation of NM23-H1 with double-stranded 5Ј-SHS that was radiolabeled at both 5Ј-termini yielded a pattern of bands similar to those obtained from the individual strands alone (lanes [13][14][15][16][17][18], suggesting that the signals for cleavage were the same in either single-stranded or duplex DNA forms. Based on the previously characterized ability of NM23-H2 to cleave the c-myc NHE sequence (26,32), we next examined the extent to which NM23-H2 could cleave the structurally similar 5Ј-SHS sequence. Indeed, NM23-H2 cleaved the 5Ј-SHS sequence in both its single-stranded and double-stranded forms (Fig. 3B). Most of the 5Ј-SHS Py strand was converted into a single fragment with high mobility, although small amounts of three fragments of lesser mobility were also observed (lanes 1-6). Co-incubation of increasing amounts of NM23-H2 with the 5Ј-SHS Pu strand yielded increasing levels of one rapidly migrating species and lesser amounts of two others (lanes 7-12). The pattern of cleavage obtained with a 5Ј-SHS doublestranded DNA substrate was consistent with that seen with the single-stranded 5Ј-SHS substrates (lanes [13][14][15][16][17][18]. NM23-H2 appeared to be somewhat more potent than NM23-H1 under the experimental conditions employed, with 5Ј-SHS cleaving activity appearing at concentrations as low as 100 ng/reaction as compared with 250 -500 ng with NM23-H1. Interestingly, NM23-H2 yielded a small number of DNA fragments that in general were not converted to progressively smaller sizes with increasing NM23-H2 concentration in the manner seen with NM23-H1. Cleavage of single-and double-stranded forms of the 5Ј-SHS by NM23-H1 and NM23-H2 were of relatively high affinity, because cleavage of each was inhibited with low concentrations of unlabeled, homologous DNAs (IC 50 ϭ 0.5-5 nM; data not shown).
Mapping of NM23-catalyzed Cleavage Sites in the 5Ј-SHS and NHE Sequences-Although nondenaturing polyacrylamide gels proved useful in the analysis of NM23-generated cleavage products, they did not provide the resolution necessary to localize specific sites of cleavage. Thus, we next analyzed the 5Ј-SHS cleavage products on denaturing sequencing gels, restricting our focus to the duplex form. To obtain an independent analysis of the individual strands within the 5Ј-SHS duplex, substrates were prepared for which either the Py or the Pu strand was selectively radiolabeled (see "Experimental Procedures"). In addition, we extended our analyses to the PDGF-A NHE located between Ϫ82 and Ϫ42, which has been shown to be critical for basal transcriptional activity. The PDGF-A NHE was considered a likely target of NM23 because of its considerable paranemic character (i.e. hypersensitivity to nuclease and chemical modification) and its high degree of nucleotide sequence identity with both the 5Ј-SHS silencer and the c-myc NHE, most notably the perfect conservation of a GGGGAGGGGG motif at its 3Ј-terminus.
NM23-catalyzed Cleavage of the 5Ј-SHS-As expected from the nondenaturing gel analyses (Fig. 3), NM23-H1 and NM23-H2 cleaved both the Pu-rich and Py-rich strands of the duplex 5Ј-SHS sequence, but the cleavage patterns obtained with the two proteins were quite different ( Fig. 4A; results summarized in Fig. 4B). NM23-H1 generated an array of Py strand fragments (lane 2), with cleavage occurring predominantly at two sites located 4 and 5 nucleotides from the 3Јterminus. Cleavage of the Pu strand by NM23-H1 was similar to that of the Py strand, characterized by multiple fragments with varying lengths of 3Ј truncation (lane 5). In contrast, NM23-H2 produced a more limited set of fragments resulting from cleavage within the 5Ј-portions of both 5Ј-SHS strands (Fig. 4A, lanes 3 and 6). The Py strand was cleaved predominantly between nucleotides located 7-11 residues from the 5Ј-terminus (lane 3), although minor cleavage sites were detected within the 3Ј-terminal portion. A single predominant cleavage site was mapped to 9 nucleotides from the 5Ј-terminus of the Pu strand (lane 6) but again with minor amounts of cleavage in the 3Ј-region. Interestingly, the cleavage sites mapped for NM23-H1 and NM23-H2 were not localized to the CCCCCTCCCC motif shared by the 5Ј-SHS, c-myc NHE, and the PDGF-A NHE. The observation that NM23-H2 cleaved the Py and Pu strands at similar distances from the 5Ј-terminus but within very different sequence contexts strongly suggests that cleavage of the 5Ј-SHS by NM23-H2 is position-dependent rather than strictly sequence-specific.
Cleavage of the PDGF-A NHE-The patterns of NHE cleavage by NM23-H1 were quite similar to those observed with the 5Ј-SHS sequence, with cleavage directed to the 3Ј-portions of both strands. The Py strand appeared to be a better cleavage substrate than the Pu strand, however, being characterized by a more extensive array of 3Ј-truncated products (lanes 8 and 11). NM23-H2 produced a single Py strand fragment resulting from cleavage at a site 9 nucleotides from the 5Ј-terminus (lane 9). The Pu strand was also a poorer substrate for NM23-H2 than the Py strand, although minor amounts of cleavage products were observed (lane 12).

Overexpression of NM23-H1 and NM23-H2 Represses Transcriptional Activity of the Basal Promoter and Upstream Negative Regulatory Regions of the PDGF-A Gene in HepG2
Cells-To measure the extent to which NM23-H1 and NM23-H2 interact with the upstream silencer and the proximal promoter regions of the PDGF-A gene, effects of NM23-H1 and NM23-H2 overexpression were assessed by transient transfection analysis. The HepG2 cell line (human hepatoma) was employed because it exhibits significant levels of basal promoter activity and PDGF-A expression, as well as silencer activity of the 5Ј-SHS element and the larger negative regulatory region (NRR; Ϫ1853 to Ϫ883). The plasmid pAC1853, which contains the NRR in the overall context of a 1853 to ϩ8 fragment of the PDGF-A gene, yielded only 40% of the CAT activity obtained with pAC883, from which the NRR has been deleted (Fig. 5A). These trends were consistent with previous observations of NRR activity in BSC-1 (African green monkey epithelial cells; Ref. 8) and HeLa (6) cell lines. Also, as seen previously (11), relocation of the 33-bp 5Ј-SHS silencer sequence to a site just upstream of the basal A-chain promoter (Ϫ261 to ϩ8) reduced CAT activity to 55% of that seen with the basal promoter alone.
Overexpression of either NM23-H1 or NM23-H2 repressed each of the PDGF-A promoter constructs studied (Fig. 5A). These repressive effects were selective for the PDGF-A promoter, with little or no effect seen on a co-transfected ␤-galactosidase reporter plasmid (pCMV-␤gal, data not shown). Although the inhibition obtained with NM23-H1 and NM23-H2 on the pAC261, pAC261/5Ј-SHS and pAC881 constructs ranged between 45 and 60%, NM23-H1 and NM23-H2 appeared to repress pAC1853 more strongly (33 and 26% of the activity of pAC1853, respectively), suggesting that additional repressive activity might be directed to the NRR. The presence of the 5Ј-SHS silencer (pAC261/5Ј-SHS) did not result in additional NM23-mediated repression, possibly because of relocation of this element from its natural upstream location to a position much closer to the basal promoter.
The previous localization of transcriptional activity of pAC261 to the Ϫ82 to Ϫ40 region (8,33), taken together with the current demonstration of efficient cleavage of this region in vitro by NM23-H1 and NM23-H2, strongly suggested that the repressive effects of NM23-H1 and NM23-H2 on pAC261 were directed to this element. To address this question more directly, a plasmid (pAC-F11) bearing the Ϫ82 to ϩ8 promoter fragment that contains only the NHE and TATA motifs of the FIG. 3. NM23-H1 and NM23-H2 cleave both strands of the 5-SHS silencer. A, EMSAs were conducted with the indicated amounts of NM23-H1 protein (g) and 32 P-labeled 5Ј-SHS oligodeoxynucleotides. Single-stranded 5Ј-SHS DNAs (5ЈSHS Py, pyrimidine-rich coding strand, lanes 1-6; 5Ј-SHS Pu, purine-rich, noncoding strand, lanes [7][8][9][10][11][12] were radiolabeled at their respective 5Ј-termini, whereas the double-stranded form (5Ј-SHS ds, lanes [13][14][15][16][17][18] was end-labeled at both 5Ј-termini. B, EMSAs were conducted with NM23-H2 and the same single-and double-stranded 5Ј-SHS probes used in A. basal promoter was analyzed. Both NM23-H1 and NM23-H2 reduced CAT activities of F11 by ϳ50% (Fig. 5B), localizing the inhibitory effects to within the minimal Ϫ82 to ϩ8 region of the PDGF-A promoter. Co-expression of NM23-H1 and NM23-H2 consistently resulted in less CAT reporter gene repression than was seen with either protein alone; however, this apparent repressive effect was cancelled by a concomitant inhibition of the transfection efficiency control plasmid, pCMV-␤gal (data not shown), presumably because of a cooperative transcriptional effect of NM23-H1 and NM23-H2 on the CMV promoter.

FIG. 4. Mapping of NM23-induced cleavage sites within the 5-SHS and NHE sequences of the PDGF-A gene in vitro.
A, analysis of NM23-generated DNA fragments by denaturing polyacrylamide gel electrophoresis. Duplex oligodeoxyribonucleotides corresponding to the 5Ј-SHS silencer element (lanes 1-6) or the NHE basal promoter element (lanes 7-12) of the PDGF-A gene were end-labeled selectively with 32 P on either the Pu-rich or Py-rich strand prior to incubation with 1 g of NM23-H1 or NM23-H2. Also included in the figure were the individual radiolabeled DNAs with no NM23 treatment (lanes 1, 4, 7, and 10). Shown adjacent to each lane are nucleotide sequences corresponding to sequencing ladders generated by the Maxam  CAT activity of a minimal Ϫ42 to ϩ8 promoter fragment was below the limit of detection (data not shown), consistent with previous studies (8). DISCUSSION Unwound or non-B-form regions of DNA, termed paranemic structures by Watson and Crick (34), are often associated with both enhancement and repression of transcription in eukaryotic genes (35). Our studies have demonstrated that NM23-H1 (H1) interacts with two such structures in the PDGF-A gene, the 5Ј-SHS silencer and the NHE basal promoter element, recognizing them as binding/cleavage substrates and repressing their transcriptional activity in cultured cells. These sequences can be added to a rapidly growing list of elements shown to be targets of NM23 proteins, including the NHE of the c-myc gene (23), as well as elements located within five different genes involved in myeloid-specific differentiation (myeloperoxidase, CD11b, CD11c, CD54, and CCR5) (24).
Although considerable nucleotide sequence identity is seen among the NM23 target sequences identified to date (e.g. the GGGGAGGGG motif found in the c-myc NHE, and the NHE and 5Ј-SHS of PDGF-A), a clear consensus sequence for DNA cleavage is not apparent. However, evidence of sequence selectivity has been observed; for example, NM23-H2 binds a 104-bp fragment of the c-myc NHE better than 50-or 34-bp duplex oligonucleotides (24). Furthermore, methylation interference footprinting has revealed specific contacts between NM23-H2 and the repeated palindromic sequence GGGTGGG of the cmyc NHE (36,37). On the other hand, NM23-H2 binds similarly to both the Pu-rich and Py-rich strands of the c-myc NHE, as well as to single-and double-stranded DNA forms (24). Furthermore, cleavage of the PDGF-A NHE and 5Ј-SHS appears to be determined by distance from the 3Ј-terminus (H1) or 5Ј-terminus (H2) rather than by specific nucleotide sequence. What do NM23 proteins recognize as the signal for cleavage in DNA, if not a strict nucleotide sequence per se? NM23 target sequences were originally identified by virtue of their hypersensitivity to nucleases, suggesting that these proteins may recognize paranemic DNA structures. Consistent with this notion is the efficient binding and cleavage of singlestranded DNA substrates by NM23-H1 and NM23-H2 (Fig. 3 and Ref. 38).
The repressive effects of NM23-H1 and NM23-H2 upon PDGF-A transcriptional elements contrast with the enhancement of c-myc NHE and myeloid differentiation gene activities by NM23-H2 (39). The repressive activities observed in the current study are not exceptions to a general rule, however, because NM23-H1 and NM23-H2 also inhibit the SV40 early promoter/enhancer and the Rous sarcoma virus 3Ј-long terminal repeat in the HepG2 cell line (data not shown). The abilities of these NM23 proteins to both repress and enhance transcription, depending on cell type and promoter context, would seem to indicate an accessory function rather than strictly defined on/off switches.
The NM23-induced repression observed in the current study was unlikely to be nonspecific, because the cytomegalovirus promoter used for constitutive ␤-galactosidase expression was unaffected by either NM23-H1 or NM23-H2 overexpression. The specificity of this activity is further indicated by more recent observations that a DNA binding-deficient form of NM23-H2 (Arg 34 to Ala) (59) exhibits impaired repression of PDGF-A regulatory elements. 2 Moreover, a relationship between the DNA cleaving and transcription-regulating activities of NM23-H1 and NM23-H2 has been demonstrated through a dominant-negative effect of DNA cleavage-deficient NM23-H1 and NM23-H2 on the transcription of myeloid differentiation genes (24). In this regard, an important goal of future studies will be to better understand the relationship between DNA cleavage and transcriptional repression.  5. NM23-H1 and NM23-H2 repress PDGF-A promoter activity. A, HepG2 cells were transfected with the PDGF-A chain promoter-CAT reporter constructs illustrated at left. The cells were co-transfected with NM23 expression plasmids as indicated under the column labeled NM23; Ϫ represents the parent expression plasmid pL2, whereas H1 and H2 represent the constructs pL2-H1 and pL3-H2, respectively. Relative CAT Activity represents the ratio of CAT activity to ␤-galactosidase activity obtained with a control vector, with the results representing the means ϩ S.E. of five independent transfection experiments (two replicate dishes/experiment). All results were normalized relative to activity obtained with pAC261 (100%). the asterisks denote significant effects (p Յ 0.05) of NM23 overexpression on individual CAT reporter constructs, as assessed by analysis of variance and the least square means separation procedure. Silencer activities of the 5Ј-SHS (pAC261 versus pAC261-5Ј-SHS) and negative regulatory region (pAC883 versus pAC1853) were also significant (p Յ 0.05), with pertinent comparisons denoted using brackets. B, transient transfections were conducted, and the data are presented as in A, using the CAT expression plasmids pAC261 and pAC-F11, the latter containing a Ϫ82 to ϩ8 fragment of the PDGF-A promoter.
could have profound implications for the types of multiprotein complexes that would assemble at such locations. Our studies suggest that such transitions may be favorable for the recruitment of repressor complexes, possibly containing histone deacetylase activities. An elegant example of how structural transitions in DNA can affect transcriptional activity is illustrated by a 30-bp nuclease-hypersensitive element in the vascular smooth muscle ␣-actin gene (40). The transcriptional activity of this element is repressive in non-␣-actin-expressing cell types (e.g. fibroblasts) but enhancing in vascular smooth muscle cells, and this switch in activity is closely correlated with DNA conformation. Two single-stranded DNA-binding proteins termed VACssBF-1 and VACssBF-2 repress transcription by stabilizing a local single-stranded conformation, whereas activation in vascular smooth muscle cells is mediated by the double-stranded DNA binding protein TEF-1. The contributions of NM23 proteins to DNA conformation and transcriptional activity should be facilitated through the study of well characterized elements as the PDGF-A basal promoter region, whose DNA structure and individual protein components (e.g. the SP-1 family, WT-1, and EGR-1) have been well characterized.
Although NM23-H1 and NM23-H2 overexpression resulted in a significant loss in activities of the NHE and upstream negative regulatory region (Ϫ1853 to Ϫ881) of the PDGF-A gene, they did not affect 5Ј-SHS activity when this silencer was relocated much closer to the basal promoter from its natural upstream location (approximately Ϫ1400). If NM23s do indeed play an important role in determining DNA conformation, it seems plausible that such an artificial relocation of the 5Ј-SHS could affect its dependence on these proteins for proper orientation with downstream positive regulatory elements. In fact, the 5Ј-SHS element, like many other silencers (for a review, see Ref. 41), does exhibit some orientation dependence, functioning in both orientations when placed upstream of the promoter but only in one orientation when placed downstream of a CAT reporter gene (11).
The differences observed in the cleavage characteristics of NM23-H1 and NM23-H2 suggest they may have complementary functions in transcriptional regulation. Consistent with that notion, co-expression of NM23-H1 and NM23-H2 inhibited CAT reporter activity more than either protein alone, although the effect could not be verified because of concomitant effects on the CMV-driven ␤-galactosidase plasmid used for a transfection efficiency control. The repressive effects of NM23 overexpression were modest, possibly because of the high and potentially saturating NM23-H2 protein levels in the HepG2 cell line studied (data not shown). Nevertheless, even modest effects on individual elements could have a significant cumulative impact on PDGF-A transcription, considering the multiplicity of NHEs identified thus far in this gene. Moreover, small changes in PDGF expression resulting from alterations in NM23 status might potentially have profound implications for cellular growth and malignancy.
The DNA cleaving activity of NM23-H2 was recently suggested to be mediated by a glycosylase/lyase-like mechanism (32), a feature of base excision enzymes that function in DNA repair. The relationship of NM23 to both transcriptional regulation and DNA repair represents an interesting parallel with BRCA1, the breast-and ovarian-specific tumor suppressor protein to which both transcriptional and repair functions have also been recently ascribed (for a review, see Ref. 42). A repair function for BRCA1 is strongly suggested by its enrichment at sites of DNA repair and its direct association with such DNA repair factors as Rad51, Rad50, and proliferating cell nuclear antigen (43,44), whereas a role in transcriptional regulation is strongly suggested by its association with the RNA polymerase II holoenzyme (45) and histone deacetylase complex (46). These functional parallels between the structurally distinct BRCA1 and NM23 proteins are intriguing in light of their shared roles as suppressors of malignant progression, and further investigation may reveal the relative importance of the transcriptional and repair activities to their suppressor functions.
The inhibitory interactions of NM23-H1 and NM23-H2 with the basal promoter and upstream silencer regions of the PDGF-A gene shown in the current study suggest a molecular mechanism that underlies the metastasis-suppressing function of NM23. The PDGFs were the first secreted polypeptides to be characterized as oncogenic (47,48), with PDGF overexpression being a hallmark of many different cancers (for reviews, see Refs. 3 and 49). Moreover, PDGF has been directly linked to neoplastic growth properties in sarcoma and astrocytoma (50 -52), as well as progression to the metastatic phenotype (53). Of particular relevance are the autocrine and paracrine actions of PDGF-A in melanoma (54 -56), in which the metastasis-suppressing activity of NM23-H1 was first observed (57,58). Our observations suggest the possibility that NM23 proteins may contribute to repression of PDGF-A and perhaps other growthrelated genes. Conversely, loss of NM23 expression could lead to derepression of growth-related genes, an event that could favor progression to increasingly malignant phenotypes, including metastasis. In this regard, we have observed that PDGF-A silencer activity is compromised in two human metastatic cell lines (WRO82, thyroid carcinoma; NKM-1, neuroblastoma) that express very low levels of NM23-H1 and NM23-H2 (data not shown). Additional studies will be directed to experimental testing of this molecular model and whether it can be extended to other growth-promoting genes whose derepression could also contribute to malignant progression.