Transcriptional repression by binding of poly(ADP-ribose) polymerase to promoter sequences.

Poly(ADP-ribose) polymerase (PARP) is a DNA-binding enzyme that plays roles in response to DNA damage, apoptosis, and genetic stability. Recent evidence has implicated PARP in transcription of eukaryotic genes. However, the existing paradigm tying PARP function to the presence of DNA strand breaks does not provide a mechanism by which it may be recruited to gene-regulating domains in the absence of DNA damage. Here we report that PARP can bind to the DNA secondary structures (hairpins) in heteroduplex DNA in a DNA end-independent fashion and that automodification of PARP in the presence of NAD+ inhibited its hairpin binding activity. Atomic force microscopic images show that in vitro PARP protein has a preference for the promoter region of the PARP gene in superhelical DNA where the dyad symmetry elements likely form hairpins according to DNase probing. Using a chromatin cross-linking and immunoprecipitation assay we show that PARP protein binds to the chromosomal PARP promoter in vivo. Reporter gene assays have revealed that the transcriptional activity of the PARP promoter is 4-5-fold greater in PARP knockout cells than in wild type fibroblasts. Reintroduction of vectors expressing full-length PARP protein or its truncated mutant (DNA-binding domain retained but lacking catalytic activity) into PARP(-/-) cells has conferred transcriptional down-regulation of the PARP gene promoter. These data provide support for PARP protein as a potent regulator of transcription including down-regulation of its own promoter.

Poly(ADP-ribose) polymerase (PARP, 1 EC 2.4.2.30) is a chromatin-associated enzyme that catalyzes the transfer of successive units of the ADP-ribose moiety from NAD ϩ to itself and other nuclear acceptor proteins (1). PARP is a zinc fingercontaining protein, which allows enzyme binding to either double or single strand DNA breaks without any apparent se-quence preference (2,3). The catalytic activity of PARP is strictly dependent on the presence of strand breaks in DNA and is modulated by the level of automodification (4,5). Data from many studies show that PARP is involved in numerous biological functions, all of which are associated with breaking and rejoining DNA strands, and it plays a pivotal role in DNA damage repair (2, 6 -8).
Recent studies have implicated PARP in transcription of eukaryotic genes (9 -16). PARP-dependent gene regulation involves poly(ADP-ribosyl)ation of transcription factors, which, in turn, prevents their binding to specific promoter sequences (10). The basal transcription factors TFIIF and TEF-1 as well as transcription factors TATA box-binding protein, YY1, SP-1, cAMP-response element-binding protein, p53, and NFB are all highly specific substrates for poly(ADP-ribosyl)ation (10,11,14,16). PARP may also interact directly with gene promoters. For instance, recombinant full-length PARP bound the DNA sequences within the MCAT1 regulatory element (11) and to the DF4 protein binding site of the Pax-6 gene neuroretinaspecific enhancer (17). Furthermore, PARP involvement in the active transcriptional DNA-protein complex formation on Reg promoter has been recently reported (12). Together these observations suggest that PARP may exert its function in transcription through direct binding to the gene-regulating sequences and through modification of transcription factors by poly(ADP-ribosyl)ation. However, total dependence of PARP function on DNA strand breaks (5) does not provide a mechanism by which it may ADP-ribosylate transcription regulators and be recruited to gene-regulating sequences in the absence of DNA damage.
Based on the ability of PARP to interact with partially unwound DNA (18,19), we reasoned that DNA secondary structures with single-stranded character may provide potential binding sites for PARP in gene-regulating sequences in the absence of DNA strand breaks. In this work we investigated the interactions between PARP protein and DNA structures of different complexity such as DNA heteroduplexes carrying stable secondary structures and superhelical DNA containing PARP promoter sequences. We found that PARP can recognize noncanonical conformations (hairpins) in a DNA end-independent fashion, and it is capable of in vitro binding to the PARP promoter sequences where the dyad symmetry elements may form the cruciforms. Using a chromatin cross-linking and immunoprecipitation assay we show that the human PARP promoter is an in vivo target for PARP protein. Further, we show that PARP protein down-regulates its gene promoter and that DNA binding activity of PARP is essential for its function in transcription.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-The plasmid pPR-PARP was constructed by cloning the 5Ј-flanking region of the human PARP gene (from Ϫ899 to ϩ156) fused to a chloramphenicol acetyltransferase reporter (20) into pcDNA 3.1 (Invitrogen) modified to remove the cytomegalovirus promoter. The 5Ј-deletion mutant of the PARP promoter (p⌬PR-PARP) was generated as described previously (20). The expression plasmid pCD12 containing cDNA for human PARP has been described previously (21). pPARP-DBD was constructed by cloning the PCR-generated fragment of cDNA (22) for human PARP-DBD (amino acids 1-303) tagged at its carboxyl terminus with a sequence encoding four FLAG epitope tags into pcDNA 3.1. The integrity of all constructs was confirmed by sequence analysis.
DNA Heteroduplex Formation and Isolation-Heteroduplex formation between 301-bp PvuII-PvuII fragments of pUC8 and a similar fragment of pUC8F14C and isolation of the heteroduplex isomers were performed as described previously (23). Briefly, 10 l of hybridization mixture containing 1 pmol of each DNA fragment in 100 mM NaCl, 50 mM Tris-HCl, pH 7.9, 1 mM dithiothreitol, 10 mM MgCl 2 were incubated stepwise at 100°C (1 min), 85°C (10 min), and 70°C (60 min) and then cooled to room temperature. Hybridization products were run in a 5% native polyacrylamide gel in 90 mM Tris borate (pH 8.3), 2.5 mM EDTA, and bands of heteroduplex fragments, which migrate slower than correctly annealed parental fragments (23), were excised. After an additional purification step using an UltraClean 15 DNA purification kit (MoBio, Solana Beach, CA), isolated heteroduplexes were resuspended in 60 l of TE (10 mM Tris-HCl, 1 mM EDTA, pH 7.8), and aliquots were taken for strand identification by sequencing and atomic force microscopy (AFM) analysis.
Supercoiled Topoisomer Preparation-Each of eight fractions of differently supercoiled DNA (topoisomers) was prepared by incubating 5 g of plasmid DNA purified by CsCl density gradient with 20 l of topoisomerase I-containing nuclear extract from HeLa cells (24) in the presence of appropriate concentrations of ethidium bromide (0 -13 M) in 200 l of reaction buffer (100 mM NaCl, 10 mM Tris-HCl, 1 mM EDTA, pH 7.6) (25). Average superhelical densities of resultant topoisomer fractions were calculated as ϭ 10.5/N, where N is the number of base pairs in the plasmid, and is the number of superhelical turns determined by the band counting method after topoisomer separation in an agarose gel in the presence of chloroquine (26).
Assay for Base-unpaired Sites-The sequence of the 1.1-kb insert was analyzed for potential hairpin formation using MFOLD software. 2 The free energies of potential hairpins were calculated for single-stranded DNA at 37°C in a solution containing 150 mM monovalent cation and 1 mM Mg 2ϩ . To detect unwound regions in supercoiled DNA, 1 g of each topoisomer prepared in a reaction with topoisomerase I was incubated on ice with 0.5 units of nuclease P1 (Invitrogen) in 10 mM Tris-HCl (pH 7.6), 10 mM MgCl 2 , 50 mM NaCl at 37°C for 10 min. The reaction was terminated by phenol/chloroform extraction, and DNA was recovered by ethanol precipitation. Following the EcoRI digestion to release a promoter-containing 1.1-kb insert, DNA was 3Ј-end-labeled using [␣-32 P]dATP and the Klenow fragment of DNA polymerase from Escherichia coli (New England Biolabs). The resultant products were separated in their single-stranded forms on a 1.5% alkaline agarose gel in 50 mM NaOH (pH 12.5), 1 mM EDTA.
PARP Binding Reactions-A recombinant full-length human PARP (R&D Systems) was used in DNA binding reactions at a 4:1 molar ratio (protein to DNA) under the ionic conditions required for optimal PARP activity (4,21). The heteroduplex DNA (23) containing stable 50-bp hairpin arms was used in PARP binding reactions. Parental duplexes (fragments of pUC8 and pUC8F14C plasmids) were used as controls in these experiments. For PARP binding reactions with the supercoiled or topologically relaxed DNA, plasmids were predigested with exonuclease III to exclude the presence of nicks in the DNA template (19). To analyze the interactions of PARP protein with the promoter region in supercoiled plasmids, bound PARP was cross-linked to DNA with 0.5% glutaraldehyde for 2 min at 37°C, and the 1.1-kb EcoRI-EcoRI fragment containing the PARP promoter region was isolated and purified on Sephadex G25 spin columns equilibrated with the deposition buffer (10 mM HEPES, pH 7.3, 1 mM MgCl 2 ).
Chromatin Cross-linking and Immunoprecipitation-Ewing's sarcoma cells A4573 (kindly provided by Dr. T. Kinsella, University of Wisconsin, Madison) were grown and maintained in Eagle's minimal essential medium (Invitrogen). Formaldehyde (Fisher) was added di-rectly to the cell culture medium to a final concentration of 1%, and fixation proceeded at 37°C for 10 min as described in the ChIP assay protocol (Upstate Biotechnology). Immunoprecipitation was performed with rabbit polyclonal anti-PARP antibody (Cell Signaling Technology). Cross-links were reversed by heating to 65°C for 4 h in the presence of 200 mM NaCl followed by PCR analysis of DNA for the detection of the PARP promoter sequences using upstream (5Ј-TGTCA ACCCA GAGAT GGCAT-3Ј) and downstream (5Ј-AACTA CTCGG GAGGC TGAA-3Ј) PCR primers designed according to the reported sequence data for the PARP 5Ј-region of the human PARP gene (27). Immunocapture of PARP from cross-linked chromatin was analyzed by immunoblotting with goat polyclonal anti-PARP antibody (1:1000, R&D Systems) as described previously (20).
Sample Preparation and Imaging with AFM-DNA samples or PARP-DNA binding reaction product in Mg 2ϩ -containing buffer (28) were deposited on an anatomically flat mica surface, allowed to adsorb for 1 min, rinsed with deionized water, and dried in a gentle nitrogen flow. The AFM images were obtained using a NanoScope IIIa instrument equipped with E-scanner (Digital Instruments, Santa Barbara, CA) operating in a tapping mode in air as described previously (28). The tapping frequency of the 125-m silicon cantilever was 300 -400 KHz, and the nominal scanning rate was set at 1-2 Hz. No less than 150 uncomplexed DNA molecules and 100 PARP-DNA complexes were analyzed in each experiment.
Transfections and Reporter Assays-Mouse embryonic fibroblasts derived from both wild type and PARP knockout mice (29) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, penicillin (100 units/ml), and streptomycin (100 g/ml). DNA transfections were carried out using a SuperFect reagent (Qiagen) according to the protocol of the manufacturer. The total amount of DNA transfected was held constant with the pcDNA 3.1 (Invitrogen) empty vector. Chloramphenicol acetyltransferase reporter assays were performed as described previously (20) and normalized for transfection efficiency using a co-transfected pSV-␤-gal vector (Promega) as an internal control. Each experiment was repeated at least three times, in duplicate, with independent plasmid preparations to assess reproducibility.

RESULTS AND DISSCUSSION
PARP Binds to Hairpins in DNA Heteroduplexes-To investigate the interactions of PARP with DNA, we used AFM, which allows direct visualization of protein and DNA molecules at nanometer resolution (30 -32). This approach was preferred to biochemical assays to address the hypothesis that PARP binding to DNA sites other than strand breaks was directed to single strand regions as observed in unwound structures in double-stranded DNA. Alternative DNA secondary structures are not thermodynamically stable in linear DNA fragments and, therefore, are not amenable to investigations of their functional transactions such as protein binding. Accordingly, our experimental approach used model heteroduplex constructs carrying stable DNA secondary structures. We used three-way junction heteroduplexes that contain 106-bp inverted repeats in one DNA strand (23) to form hairpin-like DNA structures (Fig. 1A). A representative AFM image shows that heteroduplex molecules have extrusions of the size expected for the 50-bp hairpin in the B conformation and bends at the junction (Fig. 1B).
After allowing full-length PARP protein to bind to the model hairpin-containing DNA, AFM images revealed a high incidence of DNA-protein complexes (ϳ60% of all DNA molecules) that were divided into two types based on their locations in the heteroduplexes. In complexes of the first type, PARP associated primarily with DNA ends and less frequently dimerized heteroduplexes end-to-end (Fig. 1D) consistent with our previous observations that PARP can link DNA fragments into chainlike structures (28). The most striking observation was the occurrence of the second type, internal DNA-protein complexes (Fig. 1, C, D, and E). Proteins in these complexes resided at the junction site and were not observed in other internal regions of the long arms of the model DNA. Moreover, no internal PARP-DNA complexes were formed with control DNA duplexes (301-bp fragment of pUC8 and 401-bp fragment of pUC8F14C), thus indicating the specificity of PARP binding to hairpincontaining regions in double-stranded DNA. This finding presents a challenge to the generally accepted view that PARP binds only to strand breaks in DNA.
In the presence of NAD ϩ , PARP bound to DNA strand breaks undergoes auto(ADP-ribosyl)ation, acquiring a high negative charge. Due to the charge repulsion the protein rapidly dissociates from DNA (4,33,34). Therefore, we next tested the ability of PARP to bind hairpin-containing DNA under conditions conducive to PARP automodification. Similar to our previous observations of PARP binding to DNA ends (28), NAD ϩ significantly decreased PARP affinity to the hairpins. Reversal of this effect was observed in the presence of 3-aminobenz-amide (Fig. 1F), a potent inhibitor of PARP catalytic activity. The relatively low yield of hairpin-protein complexes suggests that PARP has higher affinity to DNA ends than to hairpins in DNA fragments. These observations indicate that (i) PARP is capable of binding to certain secondary structures (e.g. hairpincontaining regions) in double-stranded DNA independently of the presence of DNA ends and (ii) NAD ϩ -dependent automodification of PARP results in inhibition of its hairpin binding activity.
PARP Protein Binds to the 5Ј-Flanking Region of the PARP Gene-Accumulating evidence supports the involvement of DNA secondary structures such as hairpins and cruciforms in transcription (34 -38). We reasoned that PARP affinity for stem-loops in DNA might influence regulation of transcription in undamaged cells by binding to such domains in promoter regions. To test this hypothesis, we investigated interaction of the PARP protein with the 5Ј-flanking region of the PARP gene (20). Structurally, the PARP gene promoter is TATA-deficient and G ϩ C-rich, typical of promoters that contain dyad symmetry elements with high propensity to form secondary structures such as cruciforms (39). Secondary structures are favored when DNA is negatively supercoiled and are not thermodynamically stable in linear DNA fragments (40). Therefore, we examined the PARP interactions with supercoiled ( ϭ Ϫ0.050) and topologically relaxed ( ϭ 0) pPR-PARP plasmids (Fig. 2, A  and B). PARP binding reactions were performed using the same DNA to protein molar ratio (4:1) as in experiments with hairpin-containing DNA heteroduplexes. AFM imaging of DNA-protein interactions revealed that PARP is capable of binding to supercoiled plasmid in a DNA end-independent fashion. Further, a quantitative evaluation of the AFM images revealed a 3-4-fold higher yield of DNA-protein complexes on a supercoiled plasmid compared with topologically relaxed DNA. These data suggest that the preferential binding of PARP to supercoiled plasmid is attributable to the formation of recognition sites for PARP in torsionally stressed DNA.
To examine PARP protein-promoter interactions in vitro, bound proteins were cross-linked to superhelical plasmid ( ϭ Ϫ0.050) with 0.5% glutaraldehyde, and the 1.1-kb fragment containing the promoter region was isolated and examined by AFM. An average of 1.2 protein molecules were bound to the promoter-containing DNA duplex, indicating that PARP recognizes certain relatively infrequent sites in the promoter region (Fig. 2C). Although the PARP binding site(s) in its own promoter is yet to be identified, our data might conceivably reflect polymerase interaction with the regions of single-stranded character that can be formed in superhelical DNA. One potential option is the formation of cruciform-like structures since several imperfect inverted repeats have been identified in the promoter sequence by the computer algorithm MFOLD (Fig.  3A). In support of this, we observed the appearance of yet unidentified sites in the promoter region that are recognized by the single strand-specific nuclease P1. These sites are generated by unwinding torsional stress in supercoiled DNA with a threshold value of superhelical density ϭ Ϫ0.050 (Fig. 3B) and were not detected in relaxed covalently closed plasmid DNA. Based on the size of P1 nuclease-generated fragments, the positions of the putative unwound sites correspond to imperfect inverted repeat (nt Ϫ325/Ϫ290) or an AT-rich region with dyad symmetry (nt Ϫ418/Ϫ403) in the PARP promoter sequences. Although these data suggest that the 5Ј-flanking region of the PARP gene has the ability to adopt unwound or alternatively base-paired structures, further studies are required to assess functional transactions between PARP protein and such structures and to map PARP binding sites on the promoter.
To analyze the PARP protein-DNA interactions at the human PARP promoter in vivo we performed formaldehyde crosslinking and immunoprecipitation experiments. This approach permits analysis of DNA-binding proteins in eukaryotic cells under physiological conditions (41,42). We observed that anti-PARP antibody effectively immunoprecipitated endogenous PARP protein and the 5Ј-flanking region of the PARP gene promoter (Fig. 4) from Ewing's sarcoma cells that constitutively express PARP protein (20). This observation indicates that PARP protein is recruited to the human PARP promoter sequences in vivo. It remains to be determined whether PARP protein binds to the promoter sequences as a monomer or forms a heterodimer with yet to be identified transcriptional regulator(s). In support of the latter possibility, the physical association of PARP with transcription factors TEF-1, B-MYB, and AP-2 and its involvement in the active transcriptional DNAprotein complex on Reg and Pax-6 promoters have been recently demonstrated (11,12,17,43,44).
Transcriptional Autoregulation of the Human PARP Gene-The functional significance of PARP interactions with its gene promoter was evaluated by transient transfection assays using immortal fibroblasts (PARP Ϫ/Ϫ ) derived from PARP knockout mice (29). We found that the transcriptional activity of the PARP promoter was 4 -5-fold greater in PARP Ϫ/Ϫ cells than in wild type (PARP ϩ/ϩ ) fibroblasts (Fig. 5A). Introduction of plasmid pCD12 carrying PARP cDNA into PARP Ϫ/Ϫ cells conferred transcriptional down-regulation of the PARP gene promoter (Fig. 5B). These data are in accord with the previously reported observations that inducible PARP expression in PARP-producing cells also inhibited PARP promoter activity (45), thus suggesting intrinsic autoregulation of PARP expression. Next we observed that deletion of the Ϫ899 to Ϫ95 region from the PARP promoter sequences alleviated PARP-mediated transcriptional inhibition (Fig. 5C) thus indicating that at least some of the functional sites that are required for PARP-mediated down-regulation of transcription may reside upstream of the minimal PARP promoter (nt from Ϫ95 to ϩ156). This suggestion agrees with our earlier observations that the PARP promoter region (nt Ϫ420/Ϫ290), harboring two putative unwound sites (at nt Ϫ418/Ϫ403 and Ϫ325/Ϫ290) (Fig. 3), is involved in negative control of the PARP promoter in cells naturally overexpressing PARP protein (20). To address the question whether catalytic activity of PARP is required for transcriptional down-regulation, the amino-terminal fragment of human PARP (amino acids 1-303) encompassing the region that encodes two zinc fingers of the enzyme and the proximal (amino acids 200 -220) helix-turn-helix motif (22) was transiently expressed in PARP Ϫ/Ϫ cells. Co-transfection of the reporter gene (pPR-PARP) and a vector (pPARP-DBD) expressing a truncated PARP mutant (that contains the DNA-binding domain but lacks catalytic activity) resulted in transcriptional down-regulation of the PARP promoter in cells with a PARPnegative background (Fig. 5B), thus indicating that PARPmediated inhibition of transcription was independent of PARP catalytic activity. Together these data demonstrate that PARP protein is a potent repressor of transcription when targeted to promoter and that its DNA binding activity is necessary and sufficient for transcriptional repression. However, we cannot rule out the possibility of cooperative interactions between PARP and other regulatory proteins for this repressive effect.
To conclude, the interactions of PARP protein with the promoter of its own gene result in suppression of transcription. PARP binding to secondary structures in DNA may reflect a potential mechanism by which it is recruited to the gene promoter. Furthermore, our data suggest that a hierarchy of PARP function may exist under which transcriptional repression may be abrogated in response to DNA damage due to a higher affinity of PARP for DNA breaks and its dissociation from DNA following protein automodification (Fig. 6). This concept integrates PARP functions in DNA repair (a nickprotection mechanism) (4, 33) and in transcriptional control of gene(s) involved in immediate cellular response to ionizing radiation and DNA-damaging drugs. Although the evidence supporting such a mechanism is not yet available, it is conceivable that the sharing of components such as PARP by DNA repair and transcription allows both events to control cellular survival in response to ionizing radiation and DNA-damaging treatments. In support of this mechanism, PARP-dependent inhibition of transcription elongation by RNA polymerase II in undamaged cells and up-regulation of mRNA synthesis in response to DNA damage have been recently demonstrated both in vitro and in vivo (13). Studies testing this hypothesis are underway.  striped box). Such macromolecular interactions between PARP protein and a promoter region constitute a repressor function for PARP in transcription. II, in response to DNA damage, PARP binding to the DNA ends triggers its catalytic activity. Subsequent poly(ADP-ribosyl) ation of free and bound PARP in the presence of intracellular NAD ϩ prevents its interaction with the promoter regions. This alleviates the PARP-mediated block on the promoter and up-regulates transcription of its own and other genes involved in the DNA damage response. III, the DNA binding activity of PARP is restored following DNA damage repair and the degradation of the ADP-ribose polymers by poly(ADPribose) glycohydrolase leading to reassembly of PARP-promoter complexes and inhibition of transcription.