A Human Protein Containing a “Cold Shock” Domain Binds Specifically to H-DNA Upstream from the Human y-Globin Genes*

We previously determined that a region between positions -228 and -189 upstream from the human y-globin genes can form an intramolecular triplex (H-DNA) in supercoiled plasmids. To identify proteins that might interact with this DNA structure, we performed expression cloning using an adult bone marrow cDNA library and the single-stranded region of the H-DNA structure as a probe. We cloned molecules very similar to two previously identified cDNAs, dbpA and dbpB. The dbpB-like protein (called BP-8 in this study) interacts specifically (K, - 4 m) with two homopyrimidine “half-sites” in the single-stranded y-228 to -189 probe, but binds to double- stranded DNA containing the same sequence with 100-fold less affinity. We have also shown that supercoiled plasmids containing the 7-228 to -189 region contain a high affinity binding site for BP-8 that is stabilized by factors that stabilize H-DNA; two HPFH point mutations (-202 C - G or C - T) that destabilize the secondary DNA structure abolish the high affinity binding site. Col- lectively, these data show that dbpB/BP-8 binds specifically to homopyrimidine half-sites in single-stranded DNA, and that it also binds to H-DNA structures that contain homopyrimidine

organization of a transcription complex on the y promoter in erythroid cells) (Catala et al., 1989;Ulrich and Ley, 1990). However, several lines of evidence suggest that this region is critical for y-globin gene repression. First, the -200 region is identical upstream from the and G-y genes, and it is highly conserved in all primates (Gumucio et al., 1991b). Second, five distinct point mutations in the -200 region, at positions -202 (C --j G and C + T), -198, -196, and -195 are all strongly associated with the phenotype of Hereditary Persistence of Fetal Hemoglobin (HPFH), a syndrome in which the y-globin genes fail to be silenced in adult red blood cells (see Ulrich et aE., 1992). These mutations appear to be responsible for the phenotype, since only the gene directly downstream from the mutation (Ay or Gy) is expressed in adult red cells, and since the mutations are concordant with the HPFH phenotype (Collins et al., 1984;Waber et al., 1986). Third, several DNA-binding proteins interact with the -200 region; some of these proteins appear to be erythroid-specific and can recognize differences between wild-type and HPFH alleles (Gumucio et al., 1988; al., 1989al., , Gumucio et al., 1991aO'Neill et Sykes and Kaufman, 1990;Fischer and Nowack, 1990;McDonagh et al., 1991; al. 1991a;Jane et al., 1993). Finally, the region upstream from the y-globin genes seems to be involved with providing appropriate signals for the yolk sacspecific, erythroid-specific expression of the y-globin genes in transgenic mice (Chada et Kollias et Trudel et al., 1987;Perez-Stable and Costantini, 1990;Dillon and Grosveld, 1991, Lloyd et al., 1992; al., 1993).
The -200 y region consists of alternating homopurine and homopyrimidine tracts, and we suggested several years ago that this region might adopt an unusual DNA structure when it is placed under torsional stress (Gray and Ley, 1985). We recently showed that this region adopts an intramolecular triplex structure (H-DNA) when it is torsionally stressed in supercoiled plasmid DNA, four independent point mutations associated with HPFH (-202 C + G, -202 C + T, -196 C --j T, and -195 G + A) all seem to disrupt critical Hoogsteen base pairs that stabilize the intramolecular triplex structure (Ulrich et al., 1992). Since the structure is found upstream from wild-type y-globin gene promoters, it could be the target for trans-repressors in adult erythroid cells; mutations that destabilize the structure could cause repression t o fail, leading to persistent expression of the y genes in adult red blood cells.
In this study, we used a n expression cloning strategy to define several DNA-binding proteins that are capable of interacting with the single-stranded region of the H-DNA structure upstream from the y-globin genes. We hypothesized that the single-stranded loop-out on the "coding s t r a n d of DNA between -216 and -208 could be the binding site for the transrepressor itself. To our surprise, we consistently cloned molecules from the cold shock gene family, and now show that these molecules are nearly identical to the previously described proteins dbpB and dbpA ( S a k u r a et al., 1988). Finally, we show that the dbpB protein binds not only to homopyrimidine "halfsites" in single-stranded DNA, but that it also recognizes H-DNA structures that contain homopyrimidine tracts in the single-stranded and triple-stranded regions of these structures. To our knowledge, this represents the first description of a specific interaction between a DNA-binding protein and H-DNA.

MATERIALS AND METHODS
Library Screening-We screened a hgtll library containing human bone marrow cDNAs derived from normal adult human bone marrow RNA (Clontech, HL1058B, Palo Alto, CA). 400,000 plaques were screened using isopropyl-P-D-diothiolgalactopyranoside induction of fusion proteins, guanidinium denaturatiodrenaturation of proteins, and hybridization with radiolabeled DNA probes (Vinson et al., 1988). cDNAs were subcloned into pUC9 and sequenced using standard dideoxy chain termination methods with Sequenase 2.0 (U. S. Biochemical Corp., Cleveland, OH). To produce /3-gal fusion proteins, we plated 50,000 plaques/l50-ml plate on top agarose containing 0.4 mM isopropyl-P-D-diothiolgalactopyranoside. Plates were incubated at 42 "C until pinpoint plaques appeared and then were incubated at 37 "C overnight. Fusion proteins were directly eluted off the plates by placing 10 ml of elution buffer (20 m~ Tris pH 7.5,1% SDS, 10% glycerine, 1 pg/ml leupeptin, 1 pg/ml pepstatin, and 20 n~ phenylmethylsulfonyl fluoride) directly on the surface of the plate for 1 h at room temperature with gentle agitation. The eluate was cleared by centrifugation and analyzed directly on SDS-PAGE gels with Coomassie staining andor Southwestern blotting, as described below.
Probe Production-The sequences of DNA probes used for screening and analysis are listed in Tables I and 11. Probes were end-labeled using standard techniques (using polynucleotide kinase (U. S. Biochemicals) and [32PlATP) and purified on denaturing acrylamide gels before use.
Production of Recombinant Proteins in Escherichia coli-After sequencing the BP-5 and BP-8 inserts, a 1.1-kilobase EcoRI fragment containing the BP-8 insert was subcloned into PET 5A (containing the appropriate reading frame for the cDNA insert). PET 5A vector alone or PET 5A containing the BP-8 insert were transformed into E. coli strain BL21 (DE3) pLysS, and bacteria were plated overnight at 37 "C on L-broth plates (Studier et al., 1990). Fresh individual colonies were inoculated into 100 ml of L-broth containing 20 pg/ml ampicillin. When the OD,, reached 0.6, isopropyl-1-thio-P-D-galactopyranoside was added to a final concentration of 0.4 mM. Cultures were allowed to grow for an additional 2-3 h at 37 "C. The E. coli pellets were resuspended in 1 ml of 50 mM glucose, 10 mM EDTA, 25 mM Tris, pH 8.0, and bacteria were lysed by freeze-thawing for three cycles. Three ml of a n 8 M urea solution was then added, and the samples were rocked overnight at 4 "C. The extracts were then dialyzed in spectropore tubing against 1000 ml of HEMG (25 mM HEPES, pH 7.6, 0.1 mM EDTA, 12.5 mM MgCI,, 10% glycerine, and 1 mM dithiothreitol) for 24 h a t 4 "C. The crude proteln extracts were then cleared by centrifugation and stored at -20 "C. Purification of Recombinant BP-8-The crude BP-8 protein was applied to a heparin-agarose column in HEM buffer (25 mM HEPES, pH 7.6,O.l mM EDTA, and 12.5 mM magnesium chloride). The flow-through fraction and a batch eluted 0.2 M KC1 fraction were discarded. The BP-8 protein was recovered with a batch elution at 2.0 M KC1, and the eluted protein was dialyzed overnight at 4 "C against HEG buffer (25 mM HEPES, pH 7.6, 0.1 mM EDTA, 10% glycerine). The samples were cleared by centrifugation for 1 min at 10,000 x g and stored in aliquots a t -70 "C. Prior to use, samples were boiled -5 min and then cleared by centrifugation to remove all potential contaminating nucleases. This procedure was shown to have no effect on the binding activity of rBP-8.
Southwestern Blotting-Proteins were separated by electrophoresis on 10% SDS-PAGE gels and transferred to nitrocellulose membranes using standard protocols. Nitrocellulose filters containing transferred proteins were then subjected to guanidinium hydrochloride denaturatiodrenaturation as previously described (Vinson et al., 1988) (although this step was later shown to be unnecessary for DNA binding with recombinant BP-8 protein). The filter was blocked with 5% dry milk in binding buffer (Blotto), and then washed with binding buffer (50 mM KCl, 10 mM HEPES, pH 7.0, 1.0 mM EDTA, 6.4 mM MgCl,, and 1.0 mM dithiothreitol). Blots were hybridized in a final volume of 20 ml containing 2 x lo7 cpm of radiolabeled probe at specific activities of >1 x lo7 c p d p g of DNA. Hybridization was performed for 2 h at room temperature, and then blots were washed in binding buffer for 10-15 min each for three changes. Autoradiography of the nitrocellulose filters was performed a t -70 "C for 12-48 h. Coomassie stains of SDS-PAGE gels were performed with Rapid Coomassie stain (Diversified Biotech) using a protocol suggested by the manufacturer.
DNase I Footprinting-Approximately 20 PM of coding strand oligomers (either y-248 to -189 or -228 to -189) were 5' end-labeled with polynucleotide kinase (U. S. Biochemicals), and 50,000 cpm (0.2 pM/ reaction) was used in each footprinting analysis. The labeled oligomer was allowed to incubate in HEMG buffer containing 50 mM KC1 and 0.2 pg of poly(1C) (Pharmacia LKB Biotechnology Inc.) with or without recombinant BP-8 protein for 10 min at 23 "C. Then 2.5 p1 of a 1 mg/ml stock of DNase I (Sigma, DNase I-EP) was added to the reaction for 3 min at 23 "C. The reaction was stopped with the addition of 150 pl of stop buffer (1% SDS, 20 mm EDTA, 20 mM Tris, pH 7.5) and 130 p1 of water. Samples were extracted with phenolkhloroform, ethanol-precipitated with carrier tRNA, denatured, and analyzed by electrophoresis on a 12% polyacrylamide sequencing gel. G ladders were obtained from the same end-labeled, coding strand oligomers using standard Maxam and Gilbert reactions. Plasmid Preparation-Plasmids used for this study are described in Ulrich et al. (1992). Topoisomerase Uethidium bromide treatment of plasmids to increase negative superhelical density was performed as described in Ulrich et al. (1992) according to methods originally suggested by Singleton and Wells (1982). The superhelical density of plasmids was analyzed by chloroquine gel analysis, as previously described (Ulrich et al., 1992).
For a binding analysis, 300 pl of HEGK was mixed with 100 p1 of the appropriately diluted, unlabeled, competing oligonucleotide and 50 pl of the radiolabeled probe. The reaction was initiated with the addition of 50 pl of rBP-8 diluted to 100 ng/ml. Each 500-1.11 reaction was incubated at 37 "C for 5 min (the protein-DNA complex reaches equilibrium in <5 min and remains stable for >90 min -37 "C).
The binding reaction for competing plasmids was performed as described above except that the HEGK, plasmid, and rBP-8 were mixed and allowed to incubate at 37 "C for 1 h. Then, 50 p1 of the 32P-radiolabeled oligomer was added, and the reaction was allowed to incubate a t 37 "C for a n additional 5 min.
Each binding reaction was terminated by passing duplicate 2 0 0 4 aliquots of the reaction over Metricel" filters under low vacuum in a multiport manifold. The filters were washed with 500 p1 of HEGK and placed into a scintillation vial. Cerenkov counts were obtained for each filter by counting in a Beckman LS 1701 Liquid Scintillation Counter without scintillation fluid. All experiments were repeated at least three times.
For each assay, one sample containing no competing unlabeled probe was used to assess maximal binding, and another sample contained no rBP-8 or competing DNA to assess nonspecific adherence of the radiolabeled probe to the filter. All counts were transformed to the percentage of maximal binding for analysis. Nonspecific binding was 55% of the total binding in all experiments.

Expression Screening of an Adult Human
Bone Marrow cDNA Library with a Coding Strand y-228 to -189 Oligomer-400,000 protein plaques derived from a hgtll adult bone marrow cDNA library were screened with the radiolabeled y-228 to -189 coding strand oligomer (this region is identical for the eyand *y-globin genes). F r o m the initial screens, eight hybridizing clones were purified to homogeneity, and crude preparations of P-gal fusion proteins were analyzed by Southwestern analysis (Fig. 11, as described under "Materials and Methods." Two clones, BP-1 and BP-7, hybridized strongly with the y-228 terminus of these proteins. The loss of a GT dinucleotide at positions 1059 and 1060 results in a frameshift that deletes 7 amino acids from the COOH terminus of YB-1. In addition, there are 25 substitutions noted in the 3'-untranslated region of this cDNA, and it extends beyond the polyadenylation signal defined in BP-5. Existing data do not allow us to determine whether these differences are due to bona fide differences between two genes, sequencing errors, reverse transcription errors, or cloning artifacts. The cDNA called NSEP-1 is highly related to dbpB and YB-1, but this protein contains such a large number of differences that it is certainly a different member of this gene family. The black bar in the cDNAs in Fig. 2 represents the region of homology with E. coli cold shock 7.4 protein (Goldstein et al., 19901, and hence the designation cold shock domain.
Production of Recombinant BP-8 ProteinSouthwestern blotting studies revealed that the P-gal fusion proteins containing BP-5 and BP-8 behaved identically (Fig. l), and for convenience, we decided to use the EcoRI fragment defined by BP-8 for expression studies in E. coli. Based on the studies of Kolluri et al. (19921, the 10 amino acids encoded by sequences downstream from the EcoRI site are unimportant for the DNA binding function of NSEP-1. The amino acids downstream from the EcoRI site are the same in BP-8 and NSEP-1, and we assume that these amino acids probably do not contribute to DNA binding activity of the recombinant BP-8 protein. The BP-8 cDNA was subcloned into PET 5 4 the vector alone and vector containing the BP-8 insert were used to transform E. coli strain BL21 (DE3) pLysS, and recombinant BP-8 protein was made and purified as described under "Materials and Methods." In Fig. 3, the left panel reveals that the recombinant BP-8 protein migrates with a molecular mass of -50,000. The predicted molecular mass of dbpB is 35,924; the additional mass is contributed by PET 5A derived amino acids at the 5' and 3' ends of the cDNA, and by the 5"untranslated region of BP-8. We have shown that these additional sequences do not alter the binding behavior of the recombinant protein? rBP-8 was purified by batch chromatography over heparinagarose and then boiled to remove any residual nucleases; rBP-8 was purified in the fraction eluted between 0.2 and 2.0 M KC1, as shown in lane 3 of the Coomassie-stained gel. A Southwestern assay of the same samples is shown on the right. The -228 to -189 y-coding strand probe detects no specific binding proteins in the PET 5A extract alone (lane 1 ). The crude rBP-8 extract contains a 50-kDa protein that binds avidly to the probe; several smaller fragments, presumably representing proteolytic products of rBP-8, are also present in lane 2. Only the largest molecular weight form is purified with the heparinagarose (H. A.) step (lane 31, suggesting that heparin-binding domain is present only on the intact protein. The rBP-8 protein, purified in this manner, was used in all subsequent footprinting and filter-binding assays. rBP-8 Binds to Tho Homopyrimidine-rich Sequences in the y-228 to -189 Region-To define the DNA sequences in the y-228 to -189 probe bound by the BP-8 protein, we first performed DNase I footprinting analysis of the single-stranded oligomer, as shown in Fig. 4. In panel A, we 5' end-labeled the 7-228 to -189 probe and a nonspecific 40-mer derived from sequences just outside the EcoRI site in the polylinker of pUC9 GTG). The gel-purified probes were incubated with increasing amounts of rBP-8 (the concentration of protein in the purified preparation was -5 pglml, of which a t least 90% was rBP-8). In lanes 3-5, increasing amounts of rBP-8 clearly reduced the ability of DNase I to cleave a specific band between -193 and Crude extracts of adult human bone marrow-derived A gtll clones that hybridized with the y-228 to -189 probe were prepared as described under "Materials and Methods"; these protein extracts were subjected to SDS-PAGE and transferred to a nitrocellulose membrane. The membrane was subjected to guanidinium HCI denaturation-renaturation as described (Vinson et al., 1988), blocked with Blotto, and hybridized with the single-stranded y-228 to -189 coding strand oligomer. Lane 1 (a negative control) contains proteins from a h g t l l clone that did not hybridize with the probe during the primary screening. Lanes 2-9 contain extracts from A clones containing BP 1-8, representing eight independent clones that hybridized with the probe and were purified to homogeneity. BP-1 and BP-7 did not react with the probe in the Southwestern assay, even though they did react when filter-lifted. The cDNAs encoding BP-4 and BP-6 are highly related to dbpA; BP-5 and BP-8 are nearly identical to dbpB.

11-228 to -189
did not hybridize with the probe in Southwestern blotting assays (lanes 2 and 8). BP-2 (lane 3 ) hybridized with all oligomers tested and was therefore not further examined because of its lack of specificity. BP-3 (lane 4 ) was partially sequenced and found to be identical to a portion of the human nucleolin cDNA (Srivastava et al., 1990); the "RNP" consensus motif of nucleolin was present within this clone (Landsman, 1992). BP-4 and BP-6 (lanes 5 and 7) made fusion proteins of identical size and binding characteristics. BP-6 was sequenced in its entirety and was nearly identical to the long form of a protein previously called dbpA (Sakura et al., 19881.' Recombinant fusion proteins BP-5 and BP-8 bound most strongly to the y-228 to -189 probe on Southwestern blots; the inserts from both phage clones were purified and subcloned into pUC9 where they were sequenced on both strands in their entirety (see Fig. 2). BP-5 represents a full-length clone and is essentially identical to the previously described gene dbpB (Sakura et al., 1988) except for polymorphisms at positions 195 and 372 (all coordinates are defined by the dbpB cDNA numbering system). Both of these polymorphisms are silent at the amino acid level. BP-5 begins at nucleotide +40 with respect to dbpB, but contains additional 3'-untranslated sequence and a bona fide poly(A) tail. BP-8 is truncated at an intragenic EcoRI site approximately 10 amino acids upstream from the stop codon.
In Fig. 2, dbpB, BP-5, and BP-8 are compared with two highly related sequences called YB-1 (Didier et al., 1988) and NSEP-1 (Kolluri and Kinniburgh, 1991;Kolluri et al., 1992). YI3-1 and dbpB are essentially identical except for the carboxyl K. A. Maloney and T. J. Ley, unpublished observations. P. A. Honvitz and T. J. Ley, unpublished observation.   2. In lane 3, heparin-agarose purified rBP-8 is present. Note that the BP-8 is approximately 90% pure in the heparin-agarose-purified fraction. In the right panel of this figure, a photograph of a Southwestern blot of the same protein extracts is shown. The same samples shown in the left panel were blotted to nitrocellulose and hybridized with the 3ZP-labeled 7-228 to -189 coding strand oligomer. Note that the pET5A extract contains no proteins that react specifically with this probe, but that rBP-8 in the crude extract and heparin-agarose-purified preparations reacts specifically with this probe.

NSEP-
-200. In addition, sequences in the -200 to -204 region exhibit increased sensitivity to DNase I with increasing protein concentrations. In contrast, the pUC Eco oligomer's DNase I cleavage pattern is not significantly altered by rBP-8 protein.
Since DNase I produced no cleavage events between -208 and -228, we footprinted a second oligomer extending from -248 to -189, as shown in Fig. 4, panel B. In this panel, increasing amounts of rBP-8 protein result in decreased cleavage of bases between -219 and -212, and again produce increased cleavage of bases between -202 and -206. Collectively, these data suggest that rBP-8 protein is binding with specificity to the two homopyrimidine tracts located between -215 and -209, and between -203 and -194 (see Table I for sequence). In addition, the region between these two homopyrimidine tracts (five consecutive Gs) seems to become more accessible to DNase I cleavage in the presence of rBP-8, suggesting that the protein may induce a conformation shift in the oligomer upon binding.

Specific Binding Activity of Heparin-Agarose Purified rBP-8 Protein Preparations-The
Coomassie gel (Fig. 3) of the purified BP-8 preparation demonstrated minor contamination with other proteins. We therefore decided to determine whether the contaminating proteins contribute to binding of the y-228 to -189 oligomer in a filter binding assay. The binding of the y-200 oligomer to a control heparin-agarose purified extract (derived from E. coli containing PET 5Avector only) was compared with purified rBP-8. Fig. 5 shows that there is 20-fold greater binding to the BP-%containing extract than to the "vector only" extract (the binding was normalized to 1.0 pg of protein for comparison). However, the contaminating proteins contribute a maximum of 10% of the total proteins in the BP-8 extract but 100% of the protein in the control extract. Therefore, the ratio of specific BP-8 binding to contaminating protein binding is 2200:l.
Dissociation curves generated with the filter binding assay are shown in Fig. 6. The single-stranded 7-228 to -189 oligomer (which was also used for the expression cloning) binds rBP-8 according to a typical sigmoidal dose-response curve with a dissociation constant (K,) of 4.0 nM, similar to that of many other DNA binding proteins. The slope of the sigmoidal binding curve at the mid-point is mathematically equivalent to the Hill coefficient (Hill, 1910). The slope of the binding curves for all oligomers studied in the filter binding assays ranged from 1.55 to 1. 80. The slope of the binding curve of all plasmids studied (see below) ranged between 1.90 and 2.05. Collectively, this suggests that there is protein-protein interaction resulting in a binding mechanism involving positive cooperativity.
The single-stranded 7-189 to -228 non-coding strand oligomer binds to rBP-8 with a K, of 80 nM, which is approximately 5% that of the affinity of the coding strand oligomer ( Fig. 6 and Table I). The double-stranded 7-228 to -189 region binds to rBP-8 with a KD of 323 nM (-1% of the affinity of the coding strand probe). This suggests that BP-8 binds singlestranded DNA preferentially and that the recognition motif is contained primarily within the coding strand.
To identify the specific BP-8 recognition sequences within the -228 to -189 oligomer, four mutant single-stranded oligomers with pyrimidine to purine mutations in 10 nucleotide blocks were created, as described in Table I (mutants A-D). Mutant A contains mutations in nucleotides 1-10 and showed no significant change in affinity from the wild-type 7-200 oligomer. Mutant C, which contains mutations in nucleotides 21-30, and mutation D, which contains mutations in nucleotides 31-40, showed modestly decreased affinities. Mutant B, containing mutations of nucleotides 11-20, had only -25% of the affinity of the wild-type 7-200 oligomers, suggesting that this region is important for BP-8 recognition.
To evaluate whether other sequences play a role in binding, an oligomer (mutant E) containing pyrimidine to purine mutations at all positions except nucleotides 11-20 was synthesized; it demonstrated a significant reduction of binding (K, = 15.9 mM). Since footprinting data (Fig. 4) showed that the downstream homopyrimidine tract was protected by rBP-8, mutant F was synthesized (pyrimidine to purine substitutions from positions 26-35, which corresponds to the downstream homopyrimidine tract); this mutation revealed a significant reduction of binding, similar to that of mutant E. These data support the hypothesis that two homopyrimidine tracts (corresponding to the regions mutated in B plus F) comprise the binding site. Mutant BF, which contains the mutations of mutant B plus F in a single oligomer, demonstrated a 50-fold reduction of binding, clearly implicating these two regions as the "complete" binding site.
To determine whether precise sequence specificity is required, mutant G (containing pyrimidine to pyrimidine (C + T or T + C) mutations in the two homopyrimidine tracts) was created; it had full activity (K, = 4.7 nM). Therefore, nonspecific homopyrimidine tracts are sufficient for BP-8 recognition. To test this hypothesis further, the binding of a poly(CT) 40-mer and a poly(AG) 40-mer was assessed (Fig. 6). The homopyrimidine oligomer bound with a similar affinity (K, = 4.0 nM) to the wild-type 7-200 oligo, but the homopurine oligomer was unable to displace the radiolabeled probe at any concentration tested Comparison of BP-8 Binding to YB-1 and NSEP-1-The YB-1 (Didier et al., 1988) and NSEP-1 (Kolluri et al., 1992) Increasing amounts of heparin-agarose purified rBP-8 resulted in increasing protection of DNase I-cleaved bands between positions -219 and -212, as determined by the Maxim and Gilbert G ladder (shown in lane 9). The region between -202 and -206 demonstrates increased DNase I sensitivity with increasing BP-8 protein concentrations.
( K , 2 1,000 nM). designated unlabeled competing oligomers and subjected to the filter binding assay described under "Materials and Methods" and "Results." The Heparin-agarose purified rBp-8 was allowed to interact with a 32P-labeled WT-7-200 coding strand oligomer and increasing amounts of the apparent KD and % relative affinity were calculated as shown in Fig content in the extracts. Note that addition of the BP-8 insert dramati-Specific binding above background was normalized for a total protein cally increases specific binding, suggesting that the major y-228 to -189 oligomer binding activity is that of rBP-8. cDNAs are shown in Fig. 2. The YI3-1 protein has been shown to bind to a double-stranded "Y-box" region that contains a CCAAT recognition motif. NSEP-1 has been shown to bind to a double-stranded pyrimidine-rich sequence upstream from c-myc, but it did not bind to the double-strandedY-box sequence (Kolluri et al., 1992). Additionally, NSEP-1 binds to a CT-rich, single-stranded oligomer from the c-myc S1 nuclease-sensitive element (Kinniburgh 1989;Firulli et al., 1992), but not the G-rich complementary s t r a n d it did not bind well to either single-stranded oligomer containing the Y-box recognition motif. To compare the binding specificity of rBP-8 with these two proteins, we assessed the ability of each of the aforementioned single-stranded oligomers to bind to rBP-8 in the filter binding assay (Table 11). rBP-8 bound the C-rich single-strand of the c-myc NSE similarly to that of the 7-200 oligomer, while the G-rich strand was bound minimally. rBP-8 bound the Y-box (pyrimidine rich) "top" strand with slightly less affinity than the y-200 oligomer, while the purine rich '%bottom" strand did not bind to BP-8 (despite containing a single tract of four consecutive pyrimidines). We conclude that the Y-box top strand contains a single homopyrimidine half-site (TI""TCT) for BP-8 or NSEP-1 binding and that >4 pyrimidines must be present in a half-site.
The Secondary Structure of the y-200 Region in Supercoiled Plasmids Contains a High Affinity Binding Site for BP-8-A 335-bp AluI fragment containing the y-globin promoter (from positions -299 through +36) and a 40 bp region (from positions -228 through -189) were each subcloned into pUC9 and are referred to as the y promoter plasmid (py-pro) or the y-200 plasmid (py-ZOO), respectively. Each plasmid was assessed for its ability to displace the y-200 oligomer from rBP-8 in the filter-binding assay (Fig. 7). Using conditions favoring intramolecular triplex formation (increased negative superhelical density and 10 mM MgSO,), we determined that the affinity of rBP-8 for py-pro is 68-fold greater than pUC9 vector alone; for py-200 the affinity is 50-fold greater than pUC9. This suggests that a high affinity BP-8-binding site exists within these two plasmids. Furthermore, the py-200 and py-pro plasmids bind indistinguishably within the limits of this assay; the high affinity binding site within the promoter must therefore exist within the -200 region.
The Effect of Intramolecular Diplex "Stabilizers" on BP-8 Binding-Magnesium ions, negative superhelical stress, and acidic pH are physicochemical effectors known to stabilize intramolecular DNA triplexes by increasing the stability of Hoogsteen base pairs (Lyamichev et al., 1985;Kohwi and Kohwi-Shigemztsu, 1988;Htun and Dahlberg, 1989;Collier and Wells, 1990;Kang and Wells, 1992). We next wished to assess the influence of these effectors on BP-8 binding to the ply-pro plasmid.
The addition of 10 m~ MgSO, enhanced BP-8 binding to a highly supercoiled y promoter plasmid (U = 5 -0.05) by -8-fold (Table 111) We next examined the contribution of negative superhelical density for BP-8 binding to the y promoter plasmid (Table 111). A p y-pro plasmid with increased negative superhelical density ( a 5 -0.05) is bound 47-fold more avidly by rBP-8 than a y promoter plasmid of "native" superhelical density (the nonconstrained or "free" superhelical density at which the plasmid exists in E. coli is more relaxed, with u --0.03). However, enhancing the negative superhelical density of pUC9 did not alter its binding by rBP-8. Linearizing the plasmid with EcoRI reduced rBP-8 binding by more than 100-fold.
Data in Table I11 also demonstrate the effect of acidic pH on rBP-8 binding. In this experiment, rBP-8 binds the py-pro plasmid ( a 5 -0.05 in 10 m~ MgSO,) 80-fold greater than pUC9 at pH 7.0. The affinity for py-pro is increased more than 10-fold by lowering the pH to 4.5. In this acidic milieu, rBP-8 binding to the py-pro plasmid is about 1,000-fold greater than to pUC9.
HPFH Mutations Abolish Binding of BP-8 to Supercoiled y-Promoter Plasmids-p y-Pro plasmids containing a -202 C 4 T mutation or a -202 C + G mutation (naturally occurring mutations that have been associated with HPFH) do not form a stable intramolecular triplex in the y-200 region (Ulrich et al., 1992). Data presented in Table IV demonstrates that highly supercoiled py-pro plasmids containing these HPFH mutations are bound by rBP-8 with approximately 3% of the afinity of the wild-type py-pro plasmid; these binding affinities do not significantly differ from pUC9 (the superhelical densities of these plasmids were shown to be equivalent by chloroquine gel analysis, data not shown). However, these plasmids contained an alteration of the primary sequence and of the secondary structure. To address the potential contribution of the primary sequence, we examined rBP-8 binding to single-stranded y-200 oligomers with the same mutations ( Table IV). The y-200 oligomer containing the -202 C + T mutation binds indistinguishably from the wild-type oligomer; this was expected since this is a pyrimidine to pyrimidine substitution (see Table I). However, the y-200 oligomer containing the -202 C --j G mutation is bound with 10-fold less affinity. Since both mutations alter the binding to the plasmids, but only the purine substitution alters binding to the oligomers, the decreased binding of rBP-8 to both py-pro plasmids containing the -202 mutations must be due to disruption of the secondary structure itself.
Specificity of rBP-8 Binding to y-200 Region-We next wished to determine whether nonspecific single-stranded DNAbinding proteins, such as the E. coli single-stranded DNA-binding protein SSB (Lohman and Bujalowski, 19901, would bind to the single-stranded y-200 oligomer or the supercoiled py-pro plasmids. The SSB protein bound the y promoter plasmid and pUC9 with similar affinities, suggesting the lack of a specific high affinity recognition site for this protein in py-200 (data not shown). However, SSB protein does bind the single-stranded y-228 to -189 oligomer with high affinity (K, = 0.3 nM). Similar results were obtained with the bacteriophage T4 gene 32 gene product (Kowalczykowski et al., 1981;Newport et al., 19811, another nonspecific single-stranded DNA-binding protein (data not shown).
We also assessed the ability of plasmids containing other defined triplex structures to bind to rBP-8. A previously defined supercoiled plasmid that contains the sequence (CCC), as the single-stranded region of a n intramolecular triplex structure (pRW2383, Kang and Wells, 1992) displaces the y-200 oligomer from rBP-8 20 times more efficiently than a vector plasmid without the secondary structure (data not shown). A second plasmid containing the sequence (GAA), as the single-stranded region of an intramolecular triplex (pRW1701, Shimizu et al., 1989) displaces y-200 from BP-8 only minimally better than the vector backbone alone. rBP-8 will therefore bind to another intramolecular triplex structure that contains a homopyrimidine tract as the single-stranded region of the structure; like the y-200 region, the (CCC), plasmid contains a second, downstream homopyrimidine tract that contributes to the intramolecular triplex. DISCUSSION We have described the cloning of cDNAs that are nearly identical to the previously described dbpA and dbpB (Sakura et al., 1988). BP-8 specifically binds to the single-stranded coding strand of the y-globin -228 to -189 region via two homopyrimidine "half-sites." Although BP-8 does bind double-stranded DNA, it binds to single-stranded y-200 oligomer with much greater (>lOO-fold) affinity. BP-8 also specifically recognizes the secondary DNA structure within the y-globin -200 region. Magnesium ions, negative superhelical density, and acidic pH, which are known to stabilize intramolecular triplex formation all enhance BP-8 binding to the secondary structure within the y-200 region. Two HPFH-associated point mutations that disrupt the formation of the intramolecular triplex in the y-200 region (Ulrich et al., 1992) reduce BP-8 binding to background levels. Finally, rBP-8 recognizes other intramolecular triplex structures that contain homopyrimidine tracts within the single-stranded and triplex regions of the structure.

TABLE I1
Comparison of rBp8 binding to oligomers used in previously published studies Filter binding assays were performed with the designated oligomers exactly as described in Table I  Oligomers are named as in Kolluri et al., 1992. The wild-type y-200 CS oligomer is defined as 100%. See Table I Each point represents the mean of duplicate determinations and are representative of not less than three assays. Panel B , the data shown i n panel A ares represented in bar graph form as the relative affinity of py-pro and py-200 compared with pUC9, which was defined to have an affinity of 1.0. The error bars represent the standard deviation of replicate determinations. Note that the presence of the region from -228 to -189 in the py-200 plasmid confers a 50-fold increase in binding affinity, and that this does not differ significantly from the entire y-promoter (68-fold).
BP-8 binds to a single-stranded oligomer that contains the single-stranded region of the secondary structure. However, the y-228 to -189 region contains two homopyrimidine-binding sites; the 3'-homopyrimidine tract is incorporated within the intramolecular triplex in supercoiled plasmids. Binding of rBP-8 to the intramolecular triplex could not be directly measured in this study; the BP-8-binding site(s) within the secondary structure is therefore not yet known. The binding of BP-8 to supercoiled plasmids that contain a triplex-forming insert in which the single-stranded region is a homopyrimidine tract could support the hypothesis that rBP-8 binds to the singlestranded pyrimidine tract. However, this plasmid also contains a second homopyrimidine tract that contributes to the intramolecular triplex in a manner similar to that o f the y-globin -200 region. So, although BP-8 was initially isolated based on its affinity for single-stranded DNA, the homopyrimidine tract within the intramolecular triplex might also play a role in BP-8 recognition. BP-8 binding to the secondary structure of the y-globin -200 region may occur via single-stranded DNA binding, triplex binding, or possibly, a combination of both. A highly supercoiled py-pro plasmid, or the y-228 to -189 oligomer, were incubated with rBP-8 with or without 10 mM MgSO, in HEGK buffer for 60 min at 37 "C. 32P-Labeled y-200 oligomer was added to the mixture and incubated for 5 additional min. Binding in the absence of MgSO, was defined as 1.0 in each case. Note that the presence of MgSO, increases rBP-8 binding to the supercoiled py-pro plasmid, but does not affect binding to the y-200 oligomer itself.
' Relative affinities were determined from competition curves at 50% displacement as shown in Figs. 6 and 7. Each value represents the mean of replicate determinations (fS.D.1.
The highly supercoiled plasmids have u values of 'r -0.05. The native py-pro or pUC9 plasmids have u values of --0.03. Linear py-pro plasmid was linearized with EcoRI prior to analysis. Note that a high negative superhelical density of the py-pro plasmid increases binding significantly, suggesting that nonspecific secondary structures within the pUC9 backbone formed at native superhelical density must contribute to the background binding activity of these plasmids.
with rBP-8 in HEGK containing 10 mM MgSO,, pH 7.0, or in a buffer The highly supercoiled py-pro or pUC9 plasmids were each mixed containing 25 mM sodium acetate pH 4.5 with 5% glycerine and 10 mM MgSO,, pH 4.5. The relative affinity of pUC9 incubated with rBP-8 at pH 7.0 was defined as 1.0. Note that binding at pH 4.5 does not significantly affect the relative affinity of rBP-8 for pUC9. However, the py-pro plasmid demonstrates a >10 fold increase in binding affinity a t pH 4.5 compared with pH 7.0.
NSEP-1 binds to a region within the 5'-flanking DNA of the c-myc gene that is capable of forming a tripledsingle-stranded structure (Kinniburgh 1989;Firulli et al., 1992). This protein binds t o CT-rich double-stranded oligonucleotides derived from the c-myc upstream region, and also to CT-rich single-stranded probes . Kolluri et al. (1992) noted that these CT-rich elements all have strong purine/pyrimidine asymmetry and proposed that NSEP-1 binds to a "type" of DNA rather than to a primary sequence. In contrast, Hasegawa et al. (1991) proposed that the cold shock proteins bind to a wide range of DNA sequences and that they play a general role in cellular metabolism rather than a specific role in transcriptional regulation. Our data are most consistent with the hypothesis of Kolluri et al. (1992). BP-8/ dbpB, like NSEP-1, clearly binds most avidly to homopyrimidine tracts within single-stranded oligonucleotides. However, it also binds avidly to the H-DNA structure within the -200 region of the y-globin gene. Depending on the frequency of these structures in torsionally strained genomic DNA, dbpB binding could actually be very specific if the triplex structures represent physiologic targets. Several dbpB-like proteins have been cloned using expression strategies that have used a myriad of DNA sequences as probes (Sakura et al., 1988;Didier et al., 1988;Ozer et al., 1990;Tafuri and Wolffe, 1990;Hasegawa et al., 1991;Gai et al., 1992;Spitkovsky et al., 1992;Shaughnessy and Wistow, 1992). Although YB-1 was identified by virtue of its ability to bind CCAAT motifs (Didier et al., 19881, this point has remained controversial; the binding behavior of dbpB-like proteins to  rBP-8 binding to highly supercoiled (a 5 -0.05) py-pro plasmids was determined as described in the text.
* Relative affinities are as described in Table 111. e The relative affinities of y-228 to -189 oligomers containing C + T or C + G HPFH-associated substitutions at the -202 position are shown. Notice that the relative affinity of the C -T substitution is minimally different from that of the wild-type oligomer, while the C + G substitution reduces the relative affinity by more than 10-fold. Adult RBC double-stranded DNA containing CCAAT motifs appears to be variable (Kolluri et al., 1992;Gai et al., 1992). Unfortunately, none of the studies performed to date have precisely mapped contact points between dbpB-like proteins and target DNAs. Because of this, the precise DNA binding motifs for dbpB-like proteins have not yet been defined. This study, and the previous one by Kolluri et al. (19921, suggests that dbpB-like proteins bind with greatest afinity to single-stranded homopyrimidine tracts. We have shown that a single short homopyrimidine tract creates a good binding site and that paired homopyrimidine tracts create the highest affinity sites (Tables I and 11). One potential explanation for the binding to seemingly unrelated DNA sequences in previous studies may be that the doublestranded DNA probes used to perform screening contained contaminating single-stranded fragments with short homopyrimidine tracts. DNA probes prepared by random primer labeling, nick translation, or oligomer annealing can all be contaminated with single-stranded fragments of DNA. Indeed, a 20-mer containing the sequence TTTTTCT provides a good binding target for BP-8 and NSEP-1 (see Table 11, Y-box top oligomer).
In addition to the DNA binding properties of dbpB-like proteins, these proteins might also potentially bind to homopyrimidine tracts in RNA. All proteins containing cold shock motifs contain an 8-amino-acid conserved sequence called the RNP-1 domain (Landsman, 1992). This region is highly conserved in many RNA-binding proteins and is known to play a role in the DNA binding activity of dbpB-like proteins (Kolluri et al., 1992;Tafuri and Wolffe, 1992). Although RNA binding has not yet been directly demonstrated for human dbpB-like proteins, Tafuri et al. (1993) and Ranjan et al. (1993) have shown that murine andxenopus dbpB-like proteins bind to and selectively repress translation of specific germ-cell-derived mRNAs. The RNA binding properties of other dbpB-like proteins will also need to be carefully examined.
The binding of BP-8 to its target sequence reveals a Hill co-efficient greater than unity. This suggests that the binding of BP-8 occurs by a mechanism involving positive cooperativity. In our in vitro system, this phenomena could only occur via protein-protein interaction or by binding site-binding site interaction (e.g. two or more binding sites exist in a single protein molecule). In this regard, the crystal structure of the Bacillus subtilis cold shock protein, CspB, has been elucidated by x-ray diffraction (Schindelin et al., 1992(Schindelin et al., , 1993 and NMR spectros-  al., 1993). BP-8 and other eukaryotic cold shock proteins (e.g. FRG Y-1 and FRG Y-2) are much larger than bacterial Csp-B; this bacterial protein is comprised almost entirely of the highly conserved cold shock domain . Even so, the Csp-B structure serves as a model that is consistent with protein-protein interaction, i.e. homodimerization. Similarly, Xenopus proteins FRG Y-1 and FRG Y-2 clearly can multimerize in vitro . These observations suggest that BP-8 may also homodimerize (or perhaps heterodimerize with other cold shock proteins); this may result in conformational changes that cause the observed positive cooperativity of DNA binding. We suggest that the triplex formation and BP-8 binding could be involved with y-globin repression, in a model depicted in Fig. 8. We hypothesize that in fetal erythrocytes, the 7-200 region does not form a n intramolecular triplex and that this allows the LCR to productively interact with the control elements of the y-globin gene. In adult erythrocytes, formation of the y-200 triplex would be facilitated or stabilized by a cold shock protein. The protein-H-DNA complex could prevent assembly of transcription complexes on the y-globin promoter or might prevent a productive interaction of the y-globin promoter and the LCR; the LCR would therefore scan down the locus and interact with the P-globin gene. HPFH mutations that abolish triplex formation would eliminate the block to LCR interactions, allowing the y-globin promoters to remain active in adult red cells. By stabilizing the triplex, cold shock proteins could play a key role in this process. Experiments directed at determining the role of cold shock proteins for the transcriptional regulation of the y-globin genes are in progress.