DNA Binding Properties of Human pol γB*

We have recently reported the crystal structure of the accessory subunit of mitochondrial DNA polymerase, pol γB, and identified a region of the protein involved in DNA binding. The DNA employed in previous studies was presumed to be single-stranded, because it was generated by single-sided PCR. Further characterization of this DNA indicated that, due to a strand transfer event during synthesis by single-sided PCR, the DNA adopts a double-stranded hairpin conformation under native conditions. We used a series of double- and single-stranded oligonucleotides of different lengths to confirm that human pol γB prefers to bind double-stranded DNA longer than 40 bp with little apparent sequence specificity. Site-specific deletion mutagenesis identified clusters of basic residues in two surface loops required for DNA binding located on opposite sides of the symmetrical pol γB dimer. A heterodimer of pol γB that contains one mutant and one wild-type DNA binding region was shown to be unable to bind double-stranded DNA, suggesting that a single DNA molecule must contact both DNA binding sites in the pol γB dimer. The ability to bind double-stranded DNA is not essential for pol γB stimulation of pol γA activity in vitro, but may play a role in DNA replication or repair.

We have recently reported the crystal structure of the accessory subunit of mitochondrial DNA polymerase, pol ␥B, and identified a region of the protein involved in DNA binding. The DNA employed in previous studies was presumed to be single-stranded, because it was generated by single-sided PCR. Further characterization of this DNA indicated that, due to a strand transfer event during synthesis by single-sided PCR, the DNA adopts a double-stranded hairpin conformation under native conditions. We used a series of double-and singlestranded oligonucleotides of different lengths to confirm that human pol ␥B prefers to bind double-stranded DNA longer than 40 bp with little apparent sequence specificity. Site-specific deletion mutagenesis identified clusters of basic residues in two surface loops required for DNA binding located on opposite sides of the symmetrical pol ␥B dimer. A heterodimer of pol ␥B that contains one mutant and one wild-type DNA binding region was shown to be unable to bind double-stranded DNA, suggesting that a single DNA molecule must contact both DNA binding sites in the pol ␥B dimer. The ability to bind double-stranded DNA is not essential for pol ␥B stimulation of pol ␥A activity in vitro, but may play a role in DNA replication or repair.
Mitochondrial DNA is replicated by DNA polymerase ␥, an enzyme with a catalytic subunit, pol 1 ␥A, containing both 5Ј 3 3Ј polymerase and 3Ј 3 5Ј exonuclease activities, and an accessory subunit, pol ␥B, that affects a number of key properties of the catalytic subunit (1,2). pol ␥B is related both in primary sequence and structure to class IIa prokaryotic aminoacyl-tRNA synthetases (aaRSs) (1,3). Apart from its role in the stimulation of pol ␥, pol ␥B has DNA binding activity that may reflect properties of aaRSs. All aaRSs bind specific RNAs, although one, phenylalanyl-tRNA synthetase, has been shown to bind specifically to double-stranded DNA using an atypical helix-turn-helix domain (4). The so-called b5 domain that mediates this DNA binding is not found in most aaRSs. In preliminary experiments, we found that pol ␥B was able to bind to a DNA substrate generated by single-sided PCR that was presumed to have a mostly single-stranded conformation (3) similar to the H-strand region that serves as origin for lagging strand mtDNA replication (O L ) (5). These observations provided support for models suggesting that the DNA binding ability of pol ␥B might play a role in initiation of mtDNA replication (6). This sort of model has been suggested for Drosophila pol ␥ as well (7), although this enzyme appears to have a simple heterodimer structure with extensive contacts between the A and B subunits (8).
In this report we present the results of further experiments to characterize the nucleic acid binding properties of mammalian pol ␥B, by studying binding to a variety of single-stranded and double-stranded DNAs and by exploring the effects of amino acid changes on nucleic acid binding. The results show that wild-type pol ␥B binds only to one of two major DNA species generated by single-sided PCR extending through O L . pol ␥B prefers to bind to an aberrant PCR product that is substantially double-stranded due to a strand transfer event at the hairpin structure at O L . Binding titrations with a variety of single-stranded and double-stranded DNAs of different lengths confirmed that pol ␥B prefers to bind double-stranded DNA. We further show that clustered point mutations that convert basic residues to alanine residues in two nucleic acid binding loops alter the DNA binding properties of the protein. The pol ␥B dimer contains two DNA binding sites on opposite sides of the protein. We constructed a pol ␥B heterodimer containing one mutant and one wild-type DNA binding site and found that this heterodimer was unable to bind double-stranded DNA. Thus, we conclude that both sites are required for high affinity DNA binding, suggesting that an individual DNA molecule must wrap around pol ␥B to interact simultaneously with both sites.

EXPERIMENTAL PROCEDURES
DNA Clones and Oligonucleotides-The DNA template used for the synthesis of 129-mer and other single-stranded DNAs was first cloned by PCR from HeLa mitochondrial DNA using primers containing XbaI (HOL1) and XhoI (HOL2) restriction sites. This clone, containing human mitochondrial DNA sequences from position 5495 to 5920, spanning O L , was named pJAC64. Sequencing revealed a point mutation, G to A, at position 5773.
Labeling, Purification, and Annealing of Oligonucleotides-Oligonucleotides used for PCR and EMSA were gel-purified before labeling and again after labeling as described above. Concentrations of unlabeled oligonucleotides were calculated based on UV absorption. Labeling was carried out with polynucleotide kinase (New England BioLabs) and [␥-32 P]ATP, under standard conditions. Fractions of known amounts of labeled oligonucleotides were spotted onto DE-81 paper (Whatman), washed with 250 mM potassium phosphate, and counted in a scintillation counter to determine specific activities. To generate doublestranded oligonucleotides, equal amounts of complementary singlestranded oligonucleotides were mixed in a buffer containing 100 mM NaCl, 10 mM Tris, pH 8.0, 1 mM EDTA, heated at 90°C, and then cooled slowly to room temperature.
Synthesis of Single-stranded DNA-The general approach used for synthesis of single-stranded DNA by PCR has been described previously (3). To generate the 129-mer, the 228-bp double-stranded DNA used as a template was excised by restriction digestion with enzymes XhoI and HincII from clone pJAC64. Primer HOL3 was used in a standard PCR reaction using either Taq DNA polymerase (Fisher) or Pfu Turbo DNA polymerase (Stratagene) with buffers supplied by the manufacturer. To prepare other single-stranded DNA species, the template was obtained by PCR using pJAC64 DNA as template and the primers described above. 25 or 30 cycles were carried out at 94°C for 45 s, 54°C for 45 s, and 72°C for 20 s. To incorporate internal label, reactions were carried out using 100 ng of template DNA, 20 pmol of unlabeled primer, and 5 Ci of [␣-32 P]dATP. For 5Ј-end-labeled products, 5 pmol of kinased HOL3 were used in a PCR reaction under the same conditions as above. The PCR products were precipitated with ethanol and collected by centrifugation, and the pellet was resuspended in formamide loading buffer, boiled, and run in 8 -13% polyacrylamide-8 M urea sequencing gels. Bands were identified by autoradiography, excised from the gel, and crush-eluted by rotating end over end overnight in a buffer containing 0.3 M sodium acetate, 10 mM Tris, pH 8, 1 mM EDTA. After brief centrifugation in a microcentrifuge, the supernatant was filtered through 0.22-m Ultrafree-MC filters (Millipore). DNA was precipitated with ethanol and resuspended in 10 mM Tris, pH 8, 1 mM EDTA.
The concentration of internally labeled PCR products was calculated by measuring the incorporated radioactivity and the specific radioactivity of the precursors in the PCR mix, taking into account the sequence of the DNA. For 5Ј-end-labeled PCR products, the final concentration was estimated using the known specific radioactivity of the labeled primer.
Generation of Mutants and Purification of Recombinant Proteins-Human pol ␥B mutants HGB P1 (pJAC81) and HGB P2 (pJAC82) were generated using the QuikChange method (Stratagene), using pJAC44 DNA (6) as template and primers 32F and 32R (P1) and 40F and 40R (P2). Recombinant proteins were expressed and purified essentially as described (6), with the following changes: expression was carried out at room temperature, with addition of extra ampicillin (50 g/ml) every hour after induction, for a total of 3 h. Frequent addition of fresh ampicillin helped prevent loss of the plasmid by the BL21(DE3) cells and substantially increased the yield of recombinant protein. C-terminal His-tagged recombinant proteins were purified by Ni-NTA (Qiagen) affinity chromatography (6) except that the wash buffer was adjusted to contain 1 M NaCl instead of 300 mM NaCl to reduce contamination by bacterial proteins.
To generate pol ␥B heterodimers with differential affinity tags on the two monomers, a derivative of pET22b(ϩ) was generated with a Cterminal calmodulin binding protein tag in place of the His tag. Oligonucleotide CBPF containing a NotI recognition site and oligonucleotide CBPR containing an XhoI recognition site were used for PCR on plasmid pSH6 (9) to produce a 105-bp product encoding the calmodulin binding peptide tag sequence AAAKRRWKKNFIAVSAANRFK-KISSSG. Following cleavage of the PCR product with NotI and XhoI, the resulting fragment was cloned into pET22b(ϩ) vector cut with the same restriction enzymes. The resulting ampicillin-resistant vector was named pET22b(ϩ)/CBP. This vector was digested with NdeI and NotI to permit it to accept NdeI/NotI DNA fragments encoding wild-type human pol ␥B or the P2 mutant. To permit selection for heterodimeric pol ␥B, NdeI/NotI cassettes encoding wild-type and P1 mutant human pol ␥B were inserted into the kanamycin-resistant vector pET29a(ϩ), which supports synthesis of his-tagged proteins. Co-transfection of Escherichia coli BL21(DE3) with HGB P1 in pET29a(ϩ) and HGB P2 in pET22b(ϩ)/CBP and selection for both ampicillin and kanamycin resistance generated a strain capable of expressing both proteins. As a control, the two wild-type HGB clones in both pET22b(ϩ)/CBP and pET29a(ϩ) were also co-expressed. In each case, co-expression was expected to produce three forms of dimeric protein, the His-tagged and CBP-tagged homodimers and the heterodimer bearing both His and CBP tags.
Purification of the His-/CBP-tagged heterodimers was carried out first on a calmodulin column (Stratagene) and then on Ni-NTA (Qiagen). Bacterial cells were sonicated in lysis buffer containing 50 mM Tris, pH 7.4, 1 mM dithiothreitol, 150 mM NaCl, and 0.1% Triton X-100, 0.2 mM phenylmethylsulfonyl fluoride, 5 g/ml leupeptin, 1 M pepstatin, 1 mM imidazole, 1 mM MgCl 2 , and 4 mM CaCl 2 . The homogenate was clarified by centrifugation, and the supernatant was incubated with calmodulin affinity resin on a rotator for 2 h at 4°C. The beads were washed extensively with lysis buffer, and bound protein was step-eluted with the same buffer lacking CaCl 2 and MgCl 2 but containing 2 mM EGTA. The eluate was concentrated by ultrafiltration using a Centricon 30 and adjusted to 25 mM sodium phosphate, pH 8.0, 300 mM NaCl, 20% glycerol, 2 mM ␤-mercaptoethanol, and 0.2 mM phenylmethylsulfonyl fluoride. Additional purification by Ni-NTA affinity chromatography was performed as described (6). Quantitation of recombinant proteins was carried out by UV absorbance or by densitometry of Coomassie Blue-stained SDS-PAGE gels using commercial glutamate dehydrogenase as a standard.
Electrophoretic Mobility Shift Assays-Reactions were carried out in 10-l volumes containing 10 mM Tris, pH 8, 2.5 mM dithiothreitol, 1 mM EDTA, 150 g/ml bovine serum albumin, 10% glycerol, and 70 mM NaCl. Each reaction contained a total of 2 l of either protein or dialysis buffer (6), which supplied the glycerol and 60 mM salt to the reaction. 1 l of DNA was used, containing 100 mM NaCl, which was responsible for 10 mM salt in the reaction. Reactions were incubated at 30°C for 10 min and run in native polyacrylamide gels. Gels contained 6% acrylamide, 0.1% bis-acrylamide, 20 mM HEPES, pH 8, 0.1 mM EDTA. Running buffer was 20 mM HEPES, pH 8, 0.1 mM EDTA. Gels were pre-run at room temperature for 1 h at 80 V, using the miniprotean II system (Bio-Rad), and run under the same conditions. Gels were dried on DEAE paper (Whatman) and imaged using either Kodak XAR5 film or a phosphorimaging screen (Amersham Biosciences).
Purine Ladders-30 fmol (10,000 cpm) of end-labeled DNA was mixed with 10 g of tRNA carrier in a total volume of 20 l. 4 l of 4% pyridinium formate, pH 2, was added, and the mix was incubated at 48°C for 15 min. Following ethanol precipitation, the DNA was resuspended in 20 l of water, 80 l of 10% piperidine was added, and the mixture was incubated at 90°C for 10 min. After ethanol precipitation, the DNA was resuspended in formamide loading buffer, boiled, and run in 8% polyacrylamide-8 M urea sequencing gels. Gels were fixed in a solution containing 10% acetic acid, 12% methanol and dried on Whatman 3MM paper, and an autoradiogram was obtained.

Single-sided PCR through O L Produces Two Major DNA
Species, Only One of Which Binds Pol ␥B-We have previously suggested that the relationship of pol ␥B with some aminoacyl-tRNA synthetases could indicate a role in binding tRNA-like structures present at the mitochondrial origins of replication. To test this model we attempted to synthesize a singlestranded DNA spanning the light strand origin of replication (O L ). This heavy-strand DNA fragment was expected to contain the stem-loop structure known to be required for the initiation of light strand synthesis. We chose single-sided PCR as a quick method to generate single-stranded DNA. This method involves the use of PCR to extend an oligonucleotide primer using double-stranded DNA as template, generating a run-off product. The product, labeled by incorporation of radioactive dAMP, was then separated from the template DNA using denaturing electrophoresis in polyacrylamide-urea gels. Two closely migrating species were observed near the position expected for the single-stranded DNA (Fig. 1) in addition to other shorter extension products that appeared to terminate at secondary structures in the template. We assumed that the largest species was the expected run-off product of 129 nucleotides and the shorter one was produced by the polymerase stalling before reaching the end of the template. We refer to the slower migrating species as 129(ϩ) and to the faster migrating one as 129(Ϫ). 129(ϩ) was used as the presumed single-stranded DNA in previous binding studies (3).
In our attempts to delimit the region in the 129-mer necessary for pol ␥B binding, we generated shorter products using the same single-sided PCR technique, changing the primer and/or DNA template used for synthesis. In this way, we synthesized a 117-mer, a 98-mer, and an 83-mer ( Fig. 2A). In each case we obtained two major DNA species around the expected size (data not shown), resembling products we obtained for the 129-mer. Experiments employing the 83-mer are not included in this report.
We synthesized 5Ј-end-labeled 129(ϩ) and 129(Ϫ) species using 5Ј-end-labeled primers to permit chemical sequencing to determine whether they represented different conformations of the same DNA fragment or differed somehow in sequence. Following purification of the DNAs in denaturing gels, we carried out partial DNA sequencing by generating purine ladders for each of the DNA species. The ladders obtained with the 129(Ϫ) species matched the sequence expected for the 129-mer.
The ladder obtained with the 129(ϩ) species showed the identical sequence up to a point corresponding to the "descending" side of the O L stem, in the direction of synthesis. From that point on, the sequence was divergent (Fig. 3). Equivalent results were obtained with end-labeled 98(ϩ) and 98(Ϫ) PCR products, with the 98(Ϫ) ladder matching the expected sequence for the 98-mer and the 98(ϩ) one diverging at the same point seen with 129(ϩ) (data not shown).
As shown in Fig. 1, when the 129-mer was synthesized, a significant fraction of PCR products appeared to be aborted at a size around 89 nucleotides (upper asterisk). This is the point at which 129(ϩ) and 129(Ϫ) sequences diverge, coinciding with the descending half of the stem at O L , in the direction of synthesis. We reasoned that the 3Ј-region of those aborted molecules should be able to fold in a stem-loop structure, resembling the O L . The 3Ј-end of this stem could then prime synthesis by the polymerase, extending the size of the doublestranded stem back to the 5Ј-end of the primer (Fig. 4B). This would generate a large stem structure with a loop corresponding to the O L loop, i.e. a 70-bp stem with a 12-nt loop. In denaturing gels, this DNA species would behave as a singlestranded DNA of 152 nt, which agrees well with the mobility of the 129(ϩ) species seen in Fig. 3. The relative mobility of the hairpin PCR products varied somewhat with the gel temperature and the quantity loaded on gels containing 8 M urea. It is well known that urea is a rather weak denaturing agent that is not able to completely disrupt secondary structures (10). Analysis of the purine ladder generated from the 129(ϩ) species beyond the point of sequence divergence in Fig. 3 showed that it matched the expected sequences for the 152-mer DNA predicted by the mechanism shown in Fig. 4. The 129(ϩ) and 98(ϩ) DNAs would be expected to behave as double-stranded DNAs of approximately half the size under native conditions. This agrees with the faster mobility of the (ϩ) species in the native gels used for EMSA (see Fig. 2B).
We also probed the structure of these PCR products by di-  gestion with the restriction enzyme AluI. This enzyme should produce a labeled 42-nucleotide fragment from 5Ј-end-labeled 129(ϩ), when analyzed in denaturing gels (Fig. 4B). AluI should not cut the single-stranded 129(Ϫ) species. This is exactly what we observed (data not shown). We also confirmed that when internally labeled 129(Ϫ) is annealed to a complementary single strand, it can be cut by AluI, generating the expected products shown in Fig. 4A (data not shown).
The reason why the polymerase stalls in the region corresponding to the descending half of the O L stem loop is not obvious. The polymerase might be expected to stall preferentially when entering the stem structure (assuming such a structure exists during synthesis), but the products seen during synthesis due to stalling at this site are only minor species. Instead, the polymerase appears to stall and to engage in hairpin replication at the base of the loop at O L . Interestingly, the hairpin product was only observed when the light strand was the template, not when the polymerase was moving in the opposite direction using the heavy strand as template. There may be some unusual structure resulting from the G-rich tract at the base of the O L loop that facilitates hairpin replication by the polymerase. We have obtained different ratios of 129(ϩ) and 129(Ϫ) species using different polymerases (Taq, Pfu) and different concentrations of nucleotides, suggesting that different conditions could produce different amounts of the hairpin species. We tested whether pol ␥ is similarly prone to hairpin formation at this site but did not find evidence of hairpin products. 2 Collectively, these results demonstrated that the reported binding by pol ␥B to a singlestranded 129-mer (3) represented binding to a mostly doublestranded DNA of 70 bp, with a 12-nt loop at one end.
pol ␥B Binding to Synthetic Double-stranded DNA Oligonucleotides-The data presented in Fig. 2B indicate that pol ␥B binds more tightly to double-stranded (H117(ϩ), H98(ϩ)) than to single-stranded DNA (H117(Ϫ), H98(Ϫ), L117). To confirm the double-strand DNA binding preference of pol ␥B, we used a series of complementary oligonucleotides of different lengths. Oligonucleotides of 32, 40, 47, and 65 nt of unrelated sequences were annealed to their complementary oligonucleotides to generate double-stranded DNA or were used alone as single strands. Binding assays proved that pol ␥B prefers to bind double-stranded instead of single-stranded DNA (Fig. 5). More avid binding is observed with DNAs of 47 bp or larger, which indicates an approximate minimum DNA size requirement for pol ␥B binding. Also, the fact that pol ␥B was able to bind a variety of DNA sequences indicates that there is little or no sequence specificity for this reaction, although this aspect has not been studied in detail. To calculate an approximate K d for double-stranded DNA binding by pol ␥B we measured the disappearance of free DNA as the protein concentration was in-2 K. G. Pinz, unpublished observation. creased, because more than one complex can be seen. A binding titration with 1 nM ds47 DNA as shown in Fig. 6 was analyzed as a simple binding 1:1 interaction between the pol ␥B dimer and DNA, provided an apparent K d of 8.6 Ϯ 1.5 nM.
Identification of Residues in Pol ␥B Necessary for DNA Binding-We previously used deletion mutagenesis to identify two protein loops in pol ␥B required for DNA binding (3). We refer to these surface loops as loop I6, between ␤ strands 10 and 11, and I7, between ␤ strands 13 and 14. These two loops are closely apposed in the dimeric protein structure (3) and contain clusters of basic residues. The corresponding regions in threonyl-tRNA synthetase contribute to the RNA binding site for anticodon recognition (11). To identify residues in these loops necessary for DNA binding by pol ␥B, we generated alanine replacement mutants. In mutant P1, two basic residues in the I6 loop 302 RK 303 , were replaced with alanines; in mutant P2, three residues in the I7 loop, 337 RKK 339 , were replaced with alanines (see Fig. 7A).
We studied the abilities of the new mutants, HGB P1 and P2, as well as other mutants of pol ␥B, to bind the double-stranded 47-mer, ds47. Fig. 7B shows the EMSA results obtained with wild-type pol ␥B, deletion mutants I4, I5, I6, and I7 (described previously (3)) and point mutants, P1 and P2. These results indicate that the basic residues in loops I6 and I7 are required for pol ␥B binding to double-stranded DNA. The P1 and P2 mutants are able to stimulate pol ␥A activity in vitro (not shown), as has been shown for the I6 and I7 deletions (3), indicating that pol ␥B binding to double-stranded DNA is not necessary for stimulation of pol ␥A activity.
Double-stranded DNA Interacts with pol ␥B on Two Opposite Sides of the Protein-The results presented above suggest a working model to describe the DNA binding of pol ␥B whereby an I6 loop from one monomer and an I7 loop from the other  1 through 5, respectively). ss stands for single-stranded and ds for double-stranded DNA, followed by the size of the DNAs in nucleotides or base pairs. Note that ds47 in panel C contains a minor portion of ss47 that failed to anneal to its complement and did not bind pol ␥B.
FIG. 6. Titration of the binding of pol ␥B to duplex oligonucleotide ds47. Binding reactions were conducted under standard conditions with 1 nM ds47 DNA and 0, 2.5, 5, 10, 20, 40, 80, and 160 nM dimeric pol ␥B. The fraction of DNA bound to protein was estimated by the removal of free DNA from the unbound position. The smooth curve drawn through the data points was fit to the data using the Langmuir isotherm, DNA bound /DNA total ϭ [protein]/(K D ϩ [protein]), and Sig-maPlot software, resulting in the K D of 8.6 Ϯ 1.5 nM. The inset shows the phosphorimaging results of the gel analysis. Note that ds47 contains a minor portion of ss47 that failed to anneal to its complement and did not bind pol ␥B. create a binding site for double-stranded DNA. Because pol ␥B is a dimer, the complex would be expected to contain two potential I6/I7 binding sites on either side of the protein. When either the I6 or I7 loop is mutated, binding sites on both sides in the pol ␥B dimer are affected. Thus, the foregoing results do not permit us to determine whether one I6/I7 binding site is sufficient for binding to double-stranded DNA.
To test this model directly, we generated a pol ␥B heterodimer containing a mutated I6 loop (P1) in one monomer and a mutated I7 loop (P2) in the other as described under "Experimental Procedures." The strategy to accomplish this was to co-express two forms of pol ␥B with different C-terminal affinity tags in the same E. coli cells, as shown in Fig. 8A. We reasoned that successive chromatography on two different affinity matrices would permit purification of heterodimers containing one monomer with each type of affinity tag. We employed a C-terminal calmodulin binding protein (CBP) tag for this experiment. This is a convenient affinity tag, because the protein can be adsorbed to a calmodulin affinity column in the presence of calcium and desorbed by the replacement of calcium with the chelator EGTA (9). To provide a positive control, wild-type pol ␥B was cloned in the same two vectors and expressed under the same conditions. In each case, three types of dimers are expected to form: His-tagged homodimers, CBPtagged homodimers, and heterodimers containing one Histagged subunit and one CBP-tagged subunit. Only the heterodimers are retained on both types of affinity matrices; this was confirmed by matrix-assisted laser desorption time of flight mass spectrometry (data not shown). For the case of the P1/P2 heterodimer, the pol ␥B should have mutated DNA binding loops I6 and I7 on one side of the protein dimer and wild-type ones on the opposite side. Dimeric wild-type and P1/P2 mutant pol ␥B proteins were purified by chromatogra-phy on calmodulin affinity resin followed by Ni-NTA as described under "Experimental Procedures." As shown in Fig. 8B, the P1-His/P2-CBP heterodimer was unable to bind doublestranded DNA, although the wild-type heterodimer control was fully active in DNA binding. These results suggest that two wild-type I6 and I7 loops are necessary for DNA binding by pol ␥B and that DNA must loop around the protein in some fashion to permit this interaction. DISCUSSION The accessory subunit of DNA pol ␥, pol ␥B, stimulates the activity of the catalytic subunit under physiological buffer conditions (1). The finding that vertebrate pol ␥B is related to prokaryotic aminoacyl-tRNA synthetases (aaRSs) suggested that the nucleic acid binding properties of a tRNA synthetase might contribute to the function of the accessory factor. Because the sequences surrounding both origins of mtDNA replication have a high potential for forming complex secondary structures, we speculated that pol ␥B could be involved in the recognition of such structures, directing the polymerase to the origins of replication (6). As a first approach to test this hypothesis we studied pol ␥B binding to a DNA fragment derived from a region of the mitochondrial genome that contains the light strand origin of replication (O L ). We initially documented binding of pol ␥B to a DNA substrate synthesized by singlesided PCR (3). In this report we show that the PCR product that bound tightly to pol ␥B in these experiments was, in fact, a largely duplex hairpin generated by fold-back priming during PCR. To confirm the double-stranded DNA binding preference of pol ␥B, we used a series of oligonucleotides, either in singlestranded or double-stranded form. The data in Figs. 5 and 6 show that pol ␥B has a poor ability to bind short duplex oligonucleotides but is able to bind with high affinity (K d of ϳ8.6 nM) to a 47-mer oligonucleotide. Our measurement of the absolute affinity of this interaction is subject to technical limitations of the EMSA assay, and we are working to develop independent measurements of this affinity using other methods. Lim et al. (12) have previously observed binding of pol ␥B to a 34:38-mer primer-template, but the lowest protein concentration used in their experiments, 2 pmol in a 20-l binding reaction, did not permit determination of the K D for this interaction. Our results suggest that the presence of a 5Ј overhang in the primertemplate used by Lim et al. (1999) probably did not influence the binding. pol ␥B has shown a similar ability to bind other duplex fragments longer than ds47 (data not shown), suggesting that DNAs must exceed a minimal size between 38 and 47 bp to bind. pol ␥B binds very poorly to single-stranded DNA, such that only a few percent of input DNA is bound by 150 nM protein (Fig. 5). To date, we have identified no specific sequences that preferentially bind to pol ␥B. However, the finding that a single DNA molecule appears to interact with binding sites on both sides of the pol ␥B dimer suggests that DNA sequences with an intrinsic bend may be bound more avidly.
The binding of pol ␥B to double-stranded DNA provides a contrast to the binding of folded single-stranded RNA by tRNA synthetases. The affinity of pol ␥B for double-stranded DNA is much higher than that previously observed for the phenylalanyl-tRNA synthetase from Thermus thermophilus, which has been estimated to have a binding constant of 400 nM. This interaction also requires a longer minimal DNA size of ϳ80 bp and does not employ the same regions of the protein required for tRNA binding (4,13). Thus, it appears that there are significant differences between the DNA binding reported for phenylalanyl-tRNA synthetase and that reported here for pol ␥B.
DNA binding by mammalian pol ␥B depends on the dimeric structure of the protein and on two superficial loops initially identified by deletion analysis, I6 and I7 (3). The corresponding regions of threonyl-tRNA synthetase are involved in binding to the anticodon of tRNA, as depicted in Fig. 7. Both loops in pol ␥B contain basic lysine and arginine residues that appeared to be good candidates to play a role in DNA binding. Site-directed mutagenesis to convert these residues to alanines confirmed this model (Fig. 7). We conclude that the basic residues in the I6/I7 region are essential for the double-stranded DNA binding activity of pol ␥B. Because we observed that DNA binding requires a rather long segment of DNA, ϳ38 -47 bp (Fig. 5), we sought to test the model that a single DNA duplex must interact simultaneously with the I6/I7 loops on both faces of the pol ␥B dimer protein. We produced a heterodimer containing one pol ␥B polypeptide with point mutations in I6 and a second with point mutations in I7. The results shown in Fig. 8 revealed that this heterodimer was not able to bind DNA. The use of a dual-tagged control wild-type protein ruled out the trivial possibility that this deficiency was due to the nature of the tags employed in purification. Thus, we conclude that a single DNA molecule must contact basic residues on both sides of pol ␥B for stable binding.
The structural basis for the action of pol ␥B on the catalytic subunit is poorly understood. This reflects the fact that the structure of the catalytic subunit has not been determined, and the interactions between the large and small subunits have not been defined precisely. Both the mammalian pol ␥B, which has a dimeric structure, and its Drosophila homolog, which binds as a monomer to its cognate pol ␥A, resemble tRNA synthetases. Recently, Drosophila pol ␥B has been shown to make extensive contacts with the catalytic subunit (8). These extensive contacts may be critically important for the activity of the small subunit as a processivity factor (14). Among processivity factors, the ability of pol ␥B to bind duplex DNA is unusual, but not unprecedented. The toroidal "sliding clamp" processivity factors like proliferating cell nuclear antigen (PCNA) and E. coli DNA pol III ␤ subunit do not possess intrinsic DNA binding activity and must be loaded onto DNA by additional factors. However, the herpes virus UL42 protein does bind DNA non-specifically with high affinity (15). In this case, the nonspecific DNA binding activity of UL42 appears to con-tribute to the ability of the herpes virus DNA polymerase holoenzyme to conduct a one-dimensional scan along DNA to identify primer-template binding sites (16). Indeed, mutations in UL42 that abrogate nonspecific DNA binding also impair the ability of the protein to function as a processivity factor. This provides an interesting contrast to pol ␥B, where mutants deficient in DNA binding, point mutants PI and P2 and the related deletion mutants I6 and I7, are not impaired in their ability to stimulate in vitro DNA synthesis by the catalytic subunit on a poly(dA):oligo(dT) template:primer ((3) and data not shown). Thus, the role, if any, that is played in mtDNA maintenance by this double-stranded DNA binding of pol ␥B remains to be established.