Sequence-specific DNA Binding and Transcriptional Regulation by the Promyelocytic Leukemia Zinc Finger Protein*

Chromosomal translocation t(11;17)(q23;21) is associated with a retinoic acid-resistant form of acute promyelocytic leukemia. The translocation fuses the RARα gene to thePLZF gene, resulting in the formation of reciprocal fusion proteins, hypothesized to play prominent roles in leukemogenesis. Promyelocytic leukemia zinc finger (PLZF) encodes a transcription factor with nine Krüppel-like zinc fingers, seven of which are retained in the t(11;17) fusion protein RARα-PLZF. We identified a specific DNA-binding site for the PLZF protein and showed that PLZF binds to this site through its most carboxyl seven zinc fingers. In co-transfection experiments, PLZF repressed transcription through its cognate binding site. This repression function of PLZF was mapped to two regions on the protein, including the evolutionarily conserved POZ domain. In contrast, the RARα-PLZF protein activated transcription of a promoter containing a PLZF response element. These results suggest that RARα-PLZF, generated in acute promyelocytic leukemia, is an aberrant transcription factor that can deregulate the expression of PLZF target genes and contribute to leukemogenesis.

Acute promyelocytic leukemia (APL) 1 is characterized by the clonal expansion of malignant myeloid cells blocked at the promyelocyte stage of development. Generally, APL is characterized by the reciprocal translocation t(15;17) (q22;21) which fuses the retinoic acid receptor ␣ (RAR␣) gene to the PML gene (reviewed in Refs. 1 and 2) generating the fusion protein PML-RAR␣. PML is a member of the RING finger family of proteins (3). PML-RAR␣ is an aberrant retinoid receptor with altered DNA binding and transcriptional activities that can act to block the action of wildtype RAR␣ in a dominant negative manner (4 -9). The PML-RAR␣ protein is central to the pathogenesis of t(15;17)-associated APL and may explain the unique responsiveness of this disease to therapy. APL patients with t (15;17) can be induced to undergo complete remission with all-trans-retinoic acid (ATRA) differentiation therapy and are generally responsive to conventional chemotherapy (10 -14). Experimentally, expression of the PML-RAR␣ fusion protein in myeloid and erythroid cells inhibits differentiation in the presence of low levels of ATRA (15)(16)(17)(18) and actually accelerates differentiation in the presence of pharmacologic doses of retinoic acid (18).
In contrast, a rare form of APL associated with t(11;17) (19 -21) is resistant to ATRA and conventional chemotherapy. In these cases the RAR␣ gene is broken prior to the third coding exon and joined to the PLZF (promyelocytic leukemia zinc finger) gene yielding a fusion protein linking the N-terminal 455 amino acids of PLZF to the B-F domain of RAR␣, including the DNA binding and ligand binding portions of the nuclear receptor. Reciprocal transcripts were also detected in these patients (8), encoding proteins containing either the A1 or A2 activation domain of RAR␣, generated from alternative promoter usage (22) fused to the last 7 zinc finger motifs of PLZF (Fig. 1). PLZF-RAR␣ is similar to PML-RAR␣ in its the ability to bind to retinoic acid response elements as a homodimer (5,23,24) and its capacity to form multimeric complexes with RXRs (5,24,25). Like PML-RAR␣, PLZF-RAR␣ is a ligand-dependent factor with altered transcriptional activity, and both proteins act in a dominant negative manner to inhibit the full transcriptional function of wild-type RAR␣ (6,7,9,23,24,26,27). These similarities do not explain why t (11;17) patients are resistant to therapy, suggesting that the disruption of the PLZF gene plays a role in the clinical phenotype.
The PLZF gene encodes a 673 amino acid transcription factor with nine Krü ppel-like C2-H2 zinc fingers located within a single region at the C terminus of the protein (19,24). During murine development, PLZF is highly expressed in the developing central nervous system, limb buds, and the perinatal kidney, liver, and heart (28,29). In the hematopoietic system, PLZF is expressed in early CD34 ϩ progenitor cells and in primitive multipotent hematopoietic cell lines (30). In addition, PLZF transcripts in myeloid HL60 and NB4 cells decline during retinoic acid-induced differentiation (19) suggesting that PLZF plays an important role in myeloid development.
To understand the function of the PLZF protein, we performed two sets of DNA-binding site selection. PLZF bound to a specific selected DNA sequence with high specificity through its most carboxyl seven zinc fingers that are also present in the t(11;17) fusion protein RAR␣-PLZF. In transient transfection experiments, a Gal4p-PLZF fusion protein was a potent transcriptional repressor. Mapping experiments revealed two noncontiguous repression domains within the N-terminal region of PLZF, one the evolutionarily conserved POZ domain. PLZF also repressed transcription through its cognate binding site. In contrast, the RAR␣-PLZF protein modestly activated tran-scription of a reporter containing five PLZF response sites. These data suggest that the deregulation of PLZF target genes by RAR␣-PLZF may be partially responsible for the aggressive clinical phenotype of t(11;17) APL.

MATERIALS AND METHODS
Plasmid Constructions-To construct GST-9ZF, the PLZF cDNA (19) was digested with NcoI to release the sequences encoding the N-terminal effector domain and recircularized. Sequences encoding the nine zinc fingers of PLZF were then excised by digestion with EcoRI and inserted into EcoRI-digested pGEX-3X (Pharmacia Biotech Inc.). GST-7ZF was synthesized by digesting the PLZF cDNA with SacI and EcoRI and inserting this fragment into pGEX-2T (Pharmacia) digested with BamHI and EcoRI utilizing an adapter of Sequence 1.
GST-2ZF was constructed by amplification of a portion of the PLZF cDNA with the N-terminal primer 5Ј GGATCCGGATCCCCGTGGGCAT-GAAGTCA 3Ј and the C-terminal primer 5Ј GAATTCGAATTCAGAACT-GCTGCTCTGGGT 3Ј. The amplified fragment was digested with BamHI and EcoRI and subcloned into pGEX-3X digested with BamHI and EcoRI. SG5-PLZF, RAR␣-PLZF (A1 domain), and PLZF-RAR␣ expression plasmids, described previously (24,27), contain the SV40 promoter/enhancer included in the SG5 plasmid (31). Gal4p-PLZF fusion constructs were made by amplification of segments of the PLZF coding sequence by PCR, digestion of these fragments with BamHI and XbaI, and insertion of these sequences into pGBX1 a plasmid encoding Gal4p amino acids 1-147, digested with BamHI and XbaI. Sequences encoding PLZF fragments beginning with amino acid 1 were amplified with the N-terminal primer 5Ј CGCGGATCCGTATG-GATCTGACAAAAATG 3Ј.
Sequences encoding fragments of PLZF beginning with amino acid 99 were amplified with the N-terminal primer 5Ј CGCGGATCCG-TCTGGATGACCTGCTGTAT.
Sequences encoding fragments of PLZF beginning with amino acid 200 were amplified with the N-terminal primer 5Ј CGCGGATCCGTA-AGGCTGCAGTGGACAGTTTG 3Ј.
Sequences encoding fragments of PLZF beginning with amino acids 300 were amplified with the N-terminal primer 5Ј CGCGGATCCGT-GAGGAGAGTCGCCTCGAGCAG 3Ј.
Sequences encoding fragments of PLZF ending with amino acid 100 were amplified with the C-terminal primer 5Ј GCTCTAGAGATCCA-GGGCCTCCGCCTTG 3Ј.
Sequences encoding fragments of PLZF ending with amino acid 200 were amplified with the C-terminal primer 5Ј GCTCTAGAGCTTGGT-GGGACTCATGGCTGA 3Ј.
Sequences encoding fragments of PLZF ending with amino acid 300 were amplified with the C-terminal primer 5Ј GCTCTAGAGCTCTCGC-CCATAGTGTAG 3Ј.
Sequences encoding fragments of PLZF ending with amino acid 400 were amplified with the C-terminal primer 5Ј GCTCTAGAGCCGGC-TCTCTGACTT 3Ј.
Reporter construct G5 tk-CAT was described previously (32). Reporter constructs PLZF 3 tk-CAT and PLZF 5 tk-CAT were constructed by ligating three or five copies of the duplex oligomer A (see Sequence 2) 5Ј TCGATATTGAAGCTAAAGTTTGATCTG 3Ј 3Ј ATAACTTCGATTTCAAACTAGACAGCT 5Ј SEQUENCE 2 in the SalI site of the polylinker region of pBLCAT5 (33). Binding Site Selection-Bacterial expression plasmids were transformed into the DnaJ-deficient Escherichia coli K12 strain CAG748 (New England Biolabs, Beverly, MA). To produce glutathione S-transferase (GST) fusion proteins, exponentially growing cultures of transformed bacteria were induced with isopropyl-1-thio-␤-D-galactopyranoside for 4 h at 37°C and lysed by sonication with the fusion protein purified on glutathione-agarose beads as described (34,35). GST protein-coated agarose beads were stored at Ϫ20°C in a buffer containing 50 mM HEPES, pH 7.5, 50 mM KCl, 5 mM MgCl 2 , 10 M ZnSO 4 , 1 mM dithiothreitol, and 50% glycerol (36). To purify GST fusion proteins for mobility shift assays, 250 l of agarose beads were incubated with 10 ml of crude bacterial extract for 1 h in 4°C and then were washed three times in 15 ml of ice-cold phosphate-buffered saline. The bound proteins were then eluted from the beads by incubation with a buffer containing 20 mM reduced glutathione, 100 l of 1 M Tris, pH 8.45, and 900 l of water for 1 h at room temperature. At the end of the incubation, the beads were centrifuged, and the supernatant was collected. The resulting protein concentration was determined by absorbance at 280 nM (35).
To prepare a pool of random oligomers for the first set of binding site selections, an oligonucleotide (500 ng) of the sequence 5Ј GGGACAAT-TCAACTGCCATCTAGGC (N) 20 ACACCGAGTCAGAAGGATCCTACG 3Ј was hybridized to primer RP2 (250 ng) of the sequence 5Ј CGTAG-GATCCTTCTGGACTCGGTGT 3Ј in a total volume of 10 l. Five l of the annealed DNA was extended to make a duplex random oligonucleotide pool with the Klenow fragment of DNA polymerase I (New England Biolabs) with a final concentration of 0.25 mM each dNTP at 37°C in a volume of 50 l. After 1 h of incubation, 5 l of 2.5 mM each dNTP was added to the extension reaction, and the incubation in 37°C was continued for another 30 min. The duplex oligomers were ethanolprecipitated, resuspended in TE (10 mM Tris, pH 8.0, 1 mM EDTA), phosphorylated with T4 kinase and [␥-32 P]ATP (6000 Ci/mM), and purified using a spin column (CLONTECH, Palo Alto, CA).
PLZF binding sites were selected by incubating the random oligomers with 50 l of a 50% slurry of GST-9ZF-coated agarose beads (about 3 g of protein) in siliconized Eppendorf tubes in a buffer containing 10 mM Tris-HCl, pH 7.5, 1 mM EDTA, 4 mM dithiothreitol, 1 mM MgCl 2 , 50 mM NaCl, and 5% glycerol (buffer A). After 1 h of incubation on ice, the beads were collected by centrifugation, the supernatant containing unbound DNA was removed, and the beads were washed twice with 1 ml of ice-cold buffer A. The oligomers retained by the beads were released by addition of 20 l of deionized water and boiling for 5 min. The resulting DNA was subjected to PCR amplification with primer RP2, described above, and primer RP3 of the sequence 5Ј GG-GAGAATTCAACTGCCATCTAGGC 3Ј. The reaction (50 l) contained 0.25 mM dATP, dGTP, and dTTP, 0.05 mM dCTP, and 3 l of [␣-32 P]dCTP (3000 Ci/mmol) to radiolabel the PCR products. Twenty cycles of PCR were performed, consisting of 1 min in 95°C, 1 min in 59°C, and 1 min at 72°C. Subsequently, 1 l of mixture of 2.5 mM of each dNTP was added to the reaction followed by one cycle of 5 min in 95°C, 2 min in 59°C, and 15 min in 72°C. The amplified DNA was purified by spin column and used in a subsequent round of binding site selection. The success of each selection round was monitored by the increasing percentage of radiolabeled probe oligomer retained of the GST-PLZF beads. After six rounds of selection, the selected DNA was reamplified using radiolabeled dCTP as above, purified, and incubated with approximately 3 g of GST-9ZF protein in a 20-l reaction containing buffer A. The resulting DNA-protein complexes were resolved by electrophoresis through a 6% (30:1, acrylamide:bisacrylamide) gel. The retarded DNA-protein complex was excised, crushed, and soaked overnight in 300 l of 0.5 M ammonium acetate, 1 mM EDTA. The eluted DNA was ethanol-precipitated, digested with BamHI and EcoRI, subcloned into the polylinker region of pBluescriptSK ϩ (Stratagene, La Jolla, CA), transformed into E. coli, and sequenced. A total of 29 clones was sequenced in this set of selections.
In the second set of binding site selection, the random oligomers used had the sequence 5Ј GGGACAATTCAACTGCCATCTAGGC (N) 13 Duplex oligomers were created by primer extension with the RP2 primer, and fours rounds of binding site selection was performed as described above. The selected DNA was digested with BamHI and EcoRI, subcloned into pBluescript SK ϩ , and 42 individual clones were sequenced.
Electrophoretic Mobility Shift Assay (EMSA)-A synthetic duplex oligomer probe A of the sequence indicated above was labeled by filling in with the Klenow fragment and [␣-32 P]dTTP (3000 Ci/mmol) and purified by spin column. Each binding reaction (20 l) contained approximately 2 g of batch-purified GST or GST-9ZF as measured by A 280 and 26 fmol of labeled DNA (approximately 0.6 ng) in a buffer of 20 mM ZnCl, 10 mM Tris-HCl, pH 7.5, 10 mM MgCl 2 , 50 mM NaCl, and 1 mM dithiothreitol. The binding reactions were incubated on ice for 1 h. In competition electrophoretic mobility shift assays, unlabeled DNA competitors were preincubated with GST-9ZF for 15 min on ice, followed by the addition of labeled duplex oligonucleotide and incubation on ice for 45 min. The resulting DNA-protein complexes were separated by electrophoresis at 200 V through either 4 or 6% polyacrylamide gels (30:1, acrylamide:bisacrylamide) for 3.5 h at room temperature. The competitors used in the EMSAs are as shown in Sequences 3-5.
Mutants of binding site A are as listed in Table III. Transient Transfections-CV1 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% calf serum and penicillin/ streptomycin in a 5% CO 2 environment. One day before transfection, 1 ϫ 10 6 cells were plated in 10-cm tissue culture dishes. Reporter and effector plasmids in the amounts indicated in the figure legends along with 1 g of a growth hormone internal control reporter gene (37) were co-transfected by calcium phosphate precipitation as described (35,38). At 48 h post-transfection, the cell media were collected and assayed for growth hormone (Nichols Institute, San Juan Capistrano, CA), and the transfected cells were harvested, and cell lysates were assayed for chloramphenicol acetyltransferase (CAT) (35). The percent conversion of unacetylated to acetylated chloramphenicol was quantified by analysis of chromatography plates on a PhosphorImager using Imagequant software (Molecular Dynamics, Sunnyvale, CA) or by densitometric analysis (NIH Image 1.56) of scanned images of the chromatograms. Graphics were scanned on an Arcus II scanner (Agfa, Germany) and assembled on a Power Macintosh 7100/66 computer (Apple, Cupertino, CA).

PLZF Binding
Site Selection-To identify the cognate binding site of PLZF, the sequence encoding the nine zinc fingers was fused in-frame to the glutathione S-transferase gene. The resulting fusion protein (GST-9ZF) ( Fig. 1) was immobilized on glutathione-coated agarose beads and incubated with a pool of random oligomers (Fig. 2B). After six rounds of binding site selection, the resulting pool of DNA was radiolabeled and incubated with GST-9ZF in an electrophoretic mobility shift assay. The DNA in the shifted band was eluted, amplified by PCR, and subcloned. Twenty-nine independent clones were sequenced (Table I), and although a consensus binding site sequence could not be formulated from the results of the first selection, it was observed that most of the selected sequences contained either TAA or TAAA. One of the sequences, 15A, was found in 5 out of 29 clones. EMSA assays with the 15A probe, however, indicated that the binding was relatively nonspecific as binding could be competed with an excess of unlabeled GAL4 operator, a non-TA-rich oligomer (data not shown). Because of the relative nonspecificity of binding, another set of selection was performed.
In the second set of selections, the sequence TAAA, frequently found in the first set of selected DNA fragments was embedded in middle of the random sequence region (Fig. 2B). Four rounds were performed; by the end of the fourth round 46% of the labeled probe bound to GST-9ZF beads. The selected DNA was subcloned, and the sequences for 42 clones were determined; only eight different sequences were found and one sequence (A) was found in 19 of the 42 clones (Table II). The affinity of GST-9ZF for this site was tested in a series of competition electrophoretic mobility shift assays (Fig. 3). GST-9ZF formed a highly specific complex with site A. A 10-fold molar excess of unlabeled A oligomer completely abolished labeled retarded complex, whereas addition of up to a 1000-fold molar excess of GAL4, p53, or WT1 binding sites did not significantly compete with site A for GST-9ZF binding.
To more precisely map residues important for PLZF-DNA interaction, six mutants of oligonucleotide A with short stretches of base transversion were synthesized (Table III) and used as unlabeled competitors in an EMSA (Fig. 4). The amount of radioactivity in the retarded PLZF-labeled A complex was quantified and plotted as a function of the competitor concentration present in the binding reaction (Fig. 5). The value of 100% was given to the amount of DNA bound in the absence of competitor, and the effectiveness of competition was measured by the amount of competitor required to displace 50% bound probe. As little as a 10-fold molar excess of unlabeled M1 or homologous probe A oligonucleotide virtually abrogated the retarded complex (Figs. 4 and 5). This suggests that the residues changed in M1 do not contribute significantly to PLZF-DNA interaction. M2 was approximately 4-fold less effective as competitor compared with M1 suggesting that residues changed in this mutant play some minor role in DNA binding by PLZF. In Fig. 5, the competition curves for both M4 and M5 overlapped each other suggesting that the residues change in both mutants contributed to PLZF DNA binding to the same extent. Both are at least 10-fold less effective as competitors compared with M2 and 40-fold less effective than M1 or homologous site A. The sequence TAAA was hypothesized to be important for PLZF DNA binding. When this sequence was mutated to GCCC in M3, the ability of this oligomer to compete was diminished 60-fold relative to wild-type oligonucleotide A. The least effective competitor was M6 which mutates the central sequence within binding site A, TA-AAGTTT. This oligomer was approximately 200-fold less effective in competing for PLZF binding than site A. Even a 3300fold molar excess of this competitor or M3 could not abolish binding of PLZF to site A. Combined, the results of these experiments suggest that the most important portion of the PLZF binding site A is TAAAGTTTGATCTGTTC and that the TAAA sequence, included in the random oligonucleotide, is essential for DNA binding by PLZF.

FIG. 3. Competition EMSA analysis.
EMSAs were performed as described under "Materials and Methods" using radiolabeled binding site A and competitors as indicated. Each competitor was added in molar excess of 10-, 100-, and 1000-fold. A, EMSA reactions were performed using GST-9ZF. B, EMSA reactions were performed using GST-7ZF.

TABLE I PLZF binding sites selected after the first set of binding site selection
Most of the sequences contained either TAA or TAAA (underlined). One of the sequences that contained TAAA was found in 5 out of 29 clones.

Sequence
Number of clones sesses the first two zinc fingers of PLZF and RAR␣-PLZF possesses the last seven. Therefore, either chimeric protein has the potential to bind to cognate PLZF sites. To test this possibility, sequences encoding the first two and the last seven zinc fingers of PLZF were fused in-frame to the glutathione Stransferase gene. The resulting fusion proteins (GST-2ZF and GST-7ZF, respectively) ( Fig. 1) were assayed for their ability to bind labeled site A by EMSA (Fig. 6). GST-7ZF but not GST-2ZF bound to site A. The affinity of GST-7ZF for binding site A was further tested by competition EMSA (Fig. 3B). As previously observed with GST-9ZF, as much as 1000-fold molar excess of unlabeled GAL4, p53, or WT1 binding sites did not abolish the GST-7ZF⅐A complex. This indicates that the se-quence-specific DNA-binding activity of PLZF resides primarily in the C-terminal seven zinc fingers. Transcriptional Regulation by PLZF and RAR␣-PLZF-To study the transcriptional effect of PLZF through its interaction with site A, the expression plasmids for PLZF, RAR␣-PLZF, or SG5 were co-transfected with a reporter construct containing three copies of the PLZF binding site A placed upstream of the herpes simplex virus thymidine kinase promoter (PLZF 3 tk-CAT). PLZF repressed expression of the reporter gene about 20-fold, whereas the RAR␣-PLZF fusion protein repressed the reporter gene by only approximately 50% (Fig. 7A). Neither PLZF nor RAR␣-PLZF had a significant effect on the parental expression vector lacking PLZF binding sites (Fig. 7A). PLZF-RAR␣ had no effect through site A (data not shown) as predicted by mobility shift assay (Fig. 6). Under certain experimental conditions RAR␣-PLZF could weakly activate transcription. When a 2:1 ratio of PLZF expression plasmid to PLZF 5 tkCAT reporter plasmid was co-transfected, PLZF again repressed transcription, and RAR␣-PLZF activated the reporter about 2-fold (Fig. 7B). Together these data indicate that the RAR␣-PLZF fusion protein generated in t(11;17)-associated APL possesses transcriptional activity which differs signifi- FIG. 4. EMSA analysis using transversion mutants of binding site A as unlabeled competitors. GST-9ZF was preincubated with the indicated competitors and then allowed to bind to radiolabeled site A. A, a 10-fold molar excess of competitors was added to each reaction. B, a 100-fold molar excess of competitors was added. C, each unlabeled competitor was added in a molar excess of 100-, 330-, 1000-, and 3300-fold. cantly from the wild-type PLZF protein.
Mapping the Functional Domains of PLZF-To functionally characterize the effector domains of PLZF, various segments of the effector region were fused to the DNA-binding domain of the GAL4 protein (Gal4p). Constructs encoding these chimeric proteins were co-transfected with a reporter (G 5 tk-CAT) containing 5 GAL4 operators (Fig. 8). The N-terminal effector domain of PLZF (amino acids 1-400) repressed the expression of the CAT reporter approximately 16-fold but did not significantly affect transcription of a tk reporter lacking GAL4 operators (data not shown). The repression function of PLZF mapped to two regions in the effector domain. One of these regions was the POZ (pox virus zinc finger)-broad complex, tramtrack, bric a brac) domain (39) (amino acids 1-100), which repressed transcription approximately 6-fold. When the POZ domain was absent from the PLZF effector domain, repression by the Gal4p-PLZF fusion protein dropped approximately 4-fold (Fig. 8, compare Gal4p 1-400 to Gal4p 100 -400). Another, more potent repression domain, PLZF localized within amino acids 200 -300, repressed transcription approximately 13-fold. Unlike the POZ domain, this region does not have homology to known protein motifs. Neither of these repression domains can repress transcription to the same extent as the entire effector domain. One region within the PLZF effector domain, amino acids 100 -200, rich in acidic residues, activated transcription approximately 3-fold. Transactivation by this domain, however, is weak because it can be masked by the presence of either the POZ domain (Fig. 8, compare Gal4p 100 -200 to  or the second repression domain (Fig. 8, compare Gal4p 100 -200 to Gal4p 100 -300). DISCUSSION The identification of a cognate binding site for the PLZF protein is an important step in the characterization of this transcription factor and its potential target genes. To accomplish this goal, we used a PCR-based method in two sets of selection experiments. The products of the first set of binding site selection clearly demonstrate the importance of a TAA or TAAA in the PLZF cognate element. Changing the TAAA on binding site A to GCCC rendered the duplex oligonucleotide virtually incapable of completely competing with unmutated site A for binding with PLZF. Although a strong consensus sequence was not established for PLZF binding, there are striking similarities between the three most highly selected sites of the second binding site selection (Table II). In site A, a TA-AAGT is present partially because of the engineered TAAA (Table II). In binding site B, present in 7 out of 42 clones, a TAA and TAAAGT were selected independent of the engineered TAAA. PLZF binding site C, found in 6 out of 42 clones, possesses a GTTC which is present in site A at exactly the same position, 10 bases 3Ј from the engineered TAAA. When this GTTC was mutated in binding site A (M5), the affinity of PLZF for this site was severely diminished (Figs. 4 and 5) which suggests that GTTC may be an actual site of protein-DNA contact. It is curious that the GTTC is one helical turn away from the TAAA sequence. This finding is reminiscent of the situation of the 9-zinc finger protein TFIIIA. This protein binds to its cognate site utilizing two groups of zinc fingers that specifically bind in the major groove of specific DNA sequences, whereas intervening zinc fingers may recognize a DNA structure rather than a specific sequence (40). Similarly, groups of PLZF zinc fingers may recognize patches of specific DNA sequence within a longer stretch of DNA. This may account for the difference in the sequences between the TAAA and GTTC in oligonucleotides A and C.
Very recently, a binding site for PLZF was fortuitously discovered in a yeast two-hybrid screening experiment. PLZF, fused to an acidic activation domain, was isolated by its ability to activate a bacterial lex operator-containing reporter gene in yeast (41). The lex operator sequence actually has some similarity to PLZF binding site A and the TAAAGT sequence selected in the second round of binding site selection (Table II). Furthermore, site A and the lex operator can be aligned with a PLZF binding site we selected from a human CpG island library and a potential PLZF binding site in the cyclin A2 promoter (42,43) (Fig. 9). A possible core consensus of A(T/G)(G/ C)T(A/C)(A/C)AGT can be derived from this comparison. The guanine residues at positions 4 and 9 of the alignment, shown to be important for lex operator-PLZF interaction by methylation interference, are well conserved among the sites. Furthermore, the thymidine residue at position 5 and adenine residue at position 8 are completely conserved (Fig. 9). The exact role of these residues in the context of binding site A will require further detailed studies of DNA-protein interaction. We found that PLZF binding site A can be recognized by the C-terminal seven zinc fingers retained in the RAR␣-PLZF fusion protein.
Similarly, the lex operator site can be bound by proteins containing the last 7 or 5 zinc fingers of PLZF (41). The exact role of the first two zinc fingers, retained in PLZF-RAR␣ in DNA binding, is not yet clear.
The binding site selection strategy yielded a relatively weak consensus that was modestly strengthened by comparison with other selected sites. The lack of a dominant consensus for a zinc finger protein is not without precedent. The Drosophila tramtrack protein, which possess two zinc fingers, binds to many sequences that have little in common other than a GGA (44,45). The WT1 zinc finger protein can also bind to DNA sequences with little in common (46,47). This can be partially attributed to the fact that WT1 protein isoforms recognize different DNA sequences depending on which subset of zinc fingers is employed (48). Similarly, PLZF might recognize more than one DNA sequence by utilizing different subsets of its nine zinc finger motifs.
The PLZF binding site lacks the guanine-rich sequence of the zif268 binding site suggesting that these proteins do not bind DNA through similar molecular arrangements. Although the base triplet recognition theory is useful in predicting the DNAbinding activity of some zinc finger proteins (49), zinc finger-DNA interactions may be more complex than once envisioned. X-ray crystallography data from tramtrack-DNA complex (45) demonstrated that DNA binding by zinc finger proteins may occur through different structural arrangements compared with those of zif268. In addition, DNA deformation especially at an AT-rich site may also play a role in zinc finger protein-DNA interactions. In the tramtrack binding site, this occurs at an ATA (45). In a similar vein, the TAAA present in PLZF binding sites may also contribute to DNA bending which may be necessary for PLZF zinc fingers to make base contact. X-ray crystallography of tramtrack zinc fingers also revealed base contacts on both strands of DNA. Whether this is true for PLZF remains to be investigated.
Transfection experiments revealed that PLZF is a transcriptional repressor that can act through the cognate binding site selected in these experiments or when tethered to a heterologous DNA-binding protein, Gal4p. RAR␣-PLZF, retaining the last 7 of 9 zinc fingers of PLZF, maintained the ability to bind to PLZF target sequences. However, RAR␣-PLZF was greatly impaired in its capacity to repress transcription and under certain circumstances activated transcription. It must be noted that addition of binding site A to pBLCAT5 (as in PLZF 3 tk-CAT) significantly increased reporter gene expression which suggested that a transcriptional activator present in CV1 might bind to site A or a site created by multimerization of site A (Fig. 7A). This protein may be widespread since in hematopoietic K562 cells, the A site also activated the tk reporter gene (data not shown). The presumed endogenous activator may bind to some sequences in common with PLZF in site A. This is suggested by the fact that the level of CAT activity generated by the PLZF 3 tk-CAT reporter in the presence of PLZF is less than the basal activity of the pBLCAT5 parental construct, whereas the activity of the PLZF 3 tk-CAT reporter in the presence of RAR␣-PLZF itself is above the base-line level of pBLCAT5 (Fig. 7A), but still repressed relative to the level of transcription generated from the reporter in the presence of the empty SG5 vector. This indicates that PLZF may repress transcription by both passive displacement of an activator and by active repression. In contrast, RAR␣-PLZF might repress a PLZF binding site-containing promoter only by displacement of an endogenous factor bound to the same or overlapping sequences.
Therefore, in t(11;17) APL, RAR␣-PLZF may compete with PLZF for binding to critical target genes and prevent repression by PLZF in a dominant negative manner. The weak activation domain within the RAR␣-PLZF fusion protein could, in addition, inappropriately activate some PLZF target genes. ATRA treatment theoretically could make the situation worse since the RAR␣2 promoter is stimulated by ATRA (50). In patients with t(11;17) this might stimulate the production of the RAR␣-PLZF fusion protein, further dysregulating genes normally repressed by PLZF. Together, these data suggest a possible mechanism for the resistant phenotype of the t(11;17) APL patients.
The POZ domain was identified as a domain that could inhibit DNA binding by the zinc finger protein ZID (39). This may not hold true for the PLZF protein, particularly since transfected PLZF could regulate a reporter gene in vivo. Moreover, Gal4p-PLZF fusion proteins that included the POZ domain clearly repressed transcription in a DNA-binding site dependent manner. Although our binding site selections were performed using only the DNA-binding domain of PLZF, we have also observed significant DNA-binding activity with fulllength PLZF (data not shown).
The observation that the POZ domain of PLZF acts as a transcriptional repression domain coincides with conclusions reached for the POZ domains of ZF5 (51) and BCL6 (52)(53)(54). BCL6 is encoded by a gene rearranged in the majority of cases of diffuse large cell lymphoma (55), further emphasizing the importance of the POZ class of zinc finger proteins in human disease. We showed that PLZF-RAR␣ can homodimerize and interact with wild-type PLZF through the POZ domain (23). Multimerization of PLZF with itself or other POZ domaincontaining proteins may play an important role in transcription regulation by PLZF. Hence, the PLZF-RAR␣ fusion protein could, in addition to inhibiting RAR function (24,27), further contribute to leukemogenesis by preventing PLZF homodimerization or by competing with PLZF for protein factors that bind to the POZ and/or other transcriptional effector domains. Alternatively, it is possible that the first two zinc fingers present in PLZF-RAR␣ may bind DNA specifically. If this is the case, PLZF-RAR␣ may also deregulate the expression of a subset of PLZF target genes. We have yet to define which sequences, if any, can be bound by these two finger motifs.
Like other zinc finger transcriptional repressors such as Drosophila Krü ppel (38,56) and the WT1 protein (57), PLZF contains contains both activation and repression domains. The PLZF activation domain is weak and is masked when linked to the repression domains of PLZF in Gal4p fusion constructs. Nevertheless, the presence of an activation domain within PLZF, corresponding with negatively charged acidic sequences, suggests that PLZF could potentially activate transcription depending on cell type, protein conformation, state of modification, the presence of other transcription factors or co-factors, and the nature of the target promoter. The presence of two repression domains within PLZF might lend added flexibility to the protein, allowing it to repress transcription stimulated by multiple types of transcriptional activators, perhaps through different molecular interactions and mechanisms.
Binding site A is an artificially optimized PLZF binding site that may not represent a biologically significant PLZF response element. To assess the biological relevance of binding site A, we searched promoters of possible target genes for this site and identified a PLZF binding site in the promoter of cyclin A2 (43,58). This putative PLZF response element is similar to PLZF binding site A, matching 12 out of 15 base pairs, spanning the TAAA region of site A. Preliminary data suggest that PLZF binds to this site with a higher affinity than to binding site A suggesting that PLZF may indeed act through this site to regulate cyclin A2 transcription. 2 Therefore, the definition of a PLZF binding site is beginning to assist in the identification of target genes influenced by PLZF during normal hematopoiesis and in acute promyelocytic leukemia.  9. Comparison of PLZF binding sites. The PLZF A binding site was compared with sites found within the cyclin A promoter, a human genomic DNA fragment, and the lex operator, recently shown to bind PLZF. *, guanine/cytidine residues important for PLZF binding as defined by methylation interference experiments with the lex operator (40). #, a thymidine residue important for PLZF binding to human genomic DNA (41).