Functional Human Mitochondrial DNA Polymerase γ Forms a Heterotrimer*

Mitochondrial DNA polymerase γ (pol γ) is responsible for replication and repair of mtDNA and is mutated in individuals with genetic disorders such as chronic external ophthalmoplegia and Alpers syndrome. pol γ is also an adventitious target for toxic side effects of several antiviral compounds, and mutation of its proofreading exonuclease leads to accelerated aging in mouse models. We have used a variety of physical and functional approaches to study the interaction of the human pol γ catalytic subunit with both the wild-type accessory factor, pol γB, and a deletion derivative that is unable to dimerize and consequently is impaired in its ability to stimulate processive DNA synthesis. Our studies clearly showed that the functional human holoenzyme contains two subunits of the processivity factor and one catalytic subunit, thereby forming a heterotrimer. The structure of pol γ seems to be variable, ranging from a single catalytic subunit in yeast to a heterodimer in Drosophila and a heterotrimer in mammals.

Mitochondrial DNA polymerase ␥ (pol ␥) is responsible for replication and repair of mtDNA and is mutated in individuals with genetic disorders such as chronic external ophthalmoplegia and Alpers syndrome. pol ␥ is also an adventitious target for toxic side effects of several antiviral compounds, and mutation of its proofreading exonuclease leads to accelerated aging in mouse models. We have used a variety of physical and functional approaches to study the interaction of the human pol ␥ catalytic subunit with both the wild-type accessory factor, pol ␥B, and a deletion derivative that is unable to dimerize and consequently is impaired in its ability to stimulate processive DNA synthesis. Our studies clearly showed that the functional human holoenzyme contains two subunits of the processivity factor and one catalytic subunit, thereby forming a heterotrimer. The structure of pol ␥ seems to be variable, ranging from a single catalytic subunit in yeast to a heterodimer in Drosophila and a heterotrimer in mammals.
Mitochondria contain a single DNA polymerase, pol 4 ␥, responsible for replication and repair of mtDNA (reviewed in Ref. 1). Human pol ␥ is isolated from mitochondria as a complex containing two subunits, a catalytic subunit, pol ␥A, of 139 kDa and an accessory subunit, pol ␥B, of 53 kDa (2)(3)(4)(5). The catalytic subunit is a family A DNA polymerase with separate polymerase and 3Ј-5Ј exonuclease domains. Two recent developments have stressed the importance of pol ␥. Mutations in the catalytic subunit of human pol ␥ cause mitochondrial disorders (6,7), and a transgenic mouse engineered to express an error-prone form of DNA pol ␥ lacking the 3Ј-5Ј proofreading ability accumulates errors in mtDNA and undergoes accelerated aging (8,9).
The processivity and substrate binding properties of pol ␥A are enhanced by complex formation with the accessory subunit (2,4,5). Most interestingly, the presence of the accessory subunit has been shown to decrease the fidelity of DNA synthesis by the catalytic subunit because it increases the ability of the enzyme to extend a mismatched primer (10). Initial characterizations of pol ␥ suggested that the enzyme forms a heterodimer containing one copy of each subunit. However, when we solved the crystal structure of mouse pol ␥B, it became apparent that this accessory factor is itself a homodimer with remarkable structural similarity to prokaryotic tRNA synthetases (Protein Data Bank code 1G5H (11)). We considered it to be very unlikely that this dimerization is a crystal packing artifact, especially because we were able to show that wild-type pol ␥B sedimented more rapidly than a deletion derivative lacking a major portion of the dimerization interface (11).
The incorporation of pol ␥B in the pol ␥ holoenzyme appears to be a relatively recent event in evolutionary terms. The protein has not been reported in yeast, where pol ␥ was first cloned as a product of the mip1 gene (12), and efforts to find evidence for it have been unsuccessful (13). Drosophila pol ␥ differs from the enzyme in vertebrates as it is reported to contain only one subunit each of pol ␥A and pol ␥B (14,15), because the amino acids required for dimerization of human pol ␥B are absent in the Drosophila ortholog. Thus, pol ␥ provides an interesting example of a eukaryotic DNA polymerase with variable quaternary structure. A number of publications have speculated on the possibility that pol ␥A or pol ␥B alone may have roles in mtDNA replication or repair independent of the other subunit (5,10,16,17). Therefore, we have undertaken the studies described here to examine the stoichiometry and association kinetics for human pol ␥B by using a range of different methods. Our studies represent the first extensive investigation of the interaction between subunits of human pol ␥ and show that the native enzyme is a heterotrimer that contains one molecule of the catalytic subunit and two molecules of the processivity factor. The tight binding we observe between the two subunits suggests that the free subunits do not exist in vivo and that individual subunits do not play important roles in mitochondrial DNA replication and repair.

EXPERIMENTAL PROCEDURES
Purification of Proteins-His-tagged recombinant exo Ϫ pol ␥A was purified from Sf9 cells infected with the pVL1393 baculovirus engineered to express exo Ϫ pol ␥〈 as described by Longley et al. (18). Cells were grown in suspension culture in JRH Excell 420 culture medium supplemented with 2.5% fetal calf serum and 100 IU penicillin plus 100 g/ml streptomycin. 5-Liter batches of cells were harvested on the 4th day of infection yielding ϳ40 g of cells. Cells were lysed by sonication in 150 ml of Buffer I (20 mM Hepes, pH 8.0, 5% glycerol, 1 mM EDTA, 5 mM 2-mercaptoethanol, 0.2 mM PMSF, 1 g/ml pepstatin A, 5 g/ml leupeptin) with 0.3 M KCl and 0.1% Triton X-100. The lysate was centrifuged at 96,000 ϫ g for 1 h at 4°C. The supernatant was treated with 0.04% polyethyleneimine and centrifuged for 30 min at 96,000 ϫ g to remove the DNA. The lysate was adjusted to contain 60 mM imidazole and 5 mM MgCl 2 and loaded onto Ni-IDA HisTrap resin (Amersham Biosciences) equilibrated in Buffer II (20 mM Hepes, pH 8.0, 5% glycerol, 0.2 mM PMSF, 1 g/ml pepstatin A, 5 g/ml leupeptin) with 0.2 M KCl and 50 mM imidazole. After washing with 10 column volumes of Buffer II, the protein was eluted with a linear gradient of imidazole (0.05-1 M) in Buffer II with 0.1 M KCl. Fractions of the eluate containing pol ␥A were diluted with Buffer I to adjust the conductivity to below 0.1 M KCl.
The enzyme was applied to a 1.0-ml HR 5/5 Mono S column equilibrated in Buffer I with 0.08 M KCl. After washing with 5 column volumes, the protein was eluted with a 15-column volume linear gradients of 0.08 to 0.7 M KCl in Buffer I. Fractions containing peak absorbance at 280 nm were loaded onto a Hiload 16/60 Superdex 200 column (Amersham Biosciences) equilibrated in Buffer I containing 0.3 M KCl. The exo Ϫ pol ␥A eluted from this column at 62 ml. All chromatographic steps were performed at 4°C. The enzyme was concentrated to 1 mg/ml using a 100,000 M r Vivaspin ultrafiltration device. The yield of the protein was about 0.7 mg/liter culture.
The His-tagged recombinant pol ␥B and pol ␥B⌬I4 mutant proteins were expressed in Escherichia coli BL21 RIL codonϩ cells (Novagen) transfected with plasmid pET22bϩ constructs as described by Carrodeguas et al. (11). When proteins were prepared without a His tag, the construct was modified by introducing a stop codon at the NotI restriction site of the pET22bϩ vector. Pellets from 2 liters of bacterial culture expressing pol ␥B were suspended in 40 ml of Buffer I with 0.3 M KCl and 0.1% Triton X-100, and cells were broken in a French press. The lysate was centrifuged at 16,000 ϫ g for 30 min at 4°C. The supernatant was treated with 0.14% polyethyleneimine and centrifuged under the same conditions to remove the DNA. The lysate was adjusted to contain 50 mM imidazole and 5 mM MgCl 2 , and the protein was purified by nickel affinity, cation exchange, and gel filtration chromatography as described for pol ␥A above. The enzyme was concentrated to 8 mg/ml using a 30,000 M r Centricon ultrafiltration device. The yields of pol ␥B and pol ␥B⌬I4 proteins were about 2 and 8 mg/liter culture, respectively.
Reconstituted pol ␥ holoenzyme was prepared by mixing 1 volume of Sf9 lysate containing His-tagged exo Ϫ pol ␥A and 0.7 volume E. coli lysate containing recombinant pol ␥B without a His tag. The relative volumes used for this reconstitution varied somewhat from one preparation to another based on the specific protein concentrations of the lysates as assessed by SDS-PAGE. The lysates were mixed for 30 min at 4°C on a rotator and adjusted to contain 60 mM imidazole and 5 mM MgCl 2 . The holoenzyme was purified and concentrated using the same protocol as for pol ␥A. The extinction coefficients used to quantify protein were determined following Gill and von Hippel (19). The ⑀ 280 for pol ␥A was 243,790 A 280 /M and that used for both WT pol ␥B and pol ␥B⌬I4 was 71,940 A 280 /M, because the deletion did not remove any residues that contribute to A 280 . The absorbance measurements were routinely made on native protein preparations and were blanked against the dialysate used in preparing the protein. Controls indicated that absorbances at 280 nm were identical within 3% when the proteins were denatured with guanidine hydrochloride.
Analytical Ultracentrifugation-Sedimentation equilibrium experiments were conducted using a Beckman XL-I Optima analytical ultracentrifuge and a Ti-60a rotor. Six-channel, charcoal-filled Epon centerpieces with quartz windows were filled with 100 l of sample (in 20 mM Hepes, pH 8.0, 300 mM KCl, 1 mM EDTA, 2 mM ␤-mercaptoethanol, 5% glycerol) at protein concentrations between 1.88 and 5.64 M. Absorbance profiles were acquired at 280 nm. Sedimentation equilibrium studies were performed at 4°C, using three different speeds (15,000, 18,000, and 22,000 rpm for wild-type pol ␥B and 20,000, 25,000, and 30,000 rpm for pol ␥B⌬I4). The data were analyzed by two programs, Heteroanalysis (National Analytical Ultracentrifugation Facility at the University of Connecticut) and the XL-A/XL-I data analysis software (Beckman), and consistent results were obtained. Solvent density was calculated as 1.026 (Sednterp software (20)), and the partial specific volumes of protein were approximated from their amino acid compositions (0.7299 for wild-type pol ␥B and 0.7283 for pol ␥B⌬I4) (Sednterp).
Surface Plasmon Resonance-Surface plasmon resonance (SPR) to analyze binding kinetics was performed on a Biacore 2000 instrument (Biacore AB, Uppsala, Sweden) at 25°C. pol ␥A was immobilized on a Sensor Chip S CM5 using the amine coupling kit (Biacore AB) in 10 mM phosphate buffer, pH 6.0, with immobilization levels of 1400 resonance units (RU). Analytes (WT pol ␥B and pol ␥B⌬I4) were injected at concentrations ranging from 0 to 320 nM over the sensor chip with immobilized pol ␥A for 3 min followed by 8 min of dissociation and 10 s of regeneration using 3 mM NaOH, 1 mM KCl. Mass transfer and linked reaction experiments were performed. The kinetics of pol ␥B and pol ␥B⌬I4 interaction with immobilized pol ␥A was evaluated using alternative predefined models in the BIAevaluation 4.1 software (Biacore AB). A model was judged appropriate if it fit to every experiment with 2 Ͻ1.5. The conformational change model assumes a two-state reaction, where analyte B binds to ligand A followed by a conformational change of the complex AB to AB*. The overall dissociation constant K D is calculated from the rate constants as follows: K D ϭ 1/(k a1 / k a2 ϫ (1 ϩ k a2 /k d2 )). For stoichiometry determinations, a sensor Chip S CM5 with immobilization levels of pol ␥A at 600 RU was used. Analytes (pol ␥B and pol ␥B⌬I4) were injected at increasing concentrations (from 50 to 500 nM) over immobilized pol ␥A for 5 min followed by 8 min of dissociation and 10 s of regeneration using 3 mM NaOH, 1 mM KCl. When the saturated concentration was achieved, R max values for pol ␥B (420 RU) pol ␥B⌬I4 (217 RU) were measured, and the stoichiometry was calculated based on the equation: Sn ϭ (M r(pol ␥A) /M r(pol ␥B) ) ϫ R max / RL, where RL is the resonance signal at a given ligand concentration.
The affinity of the enzymes for DNA was investigated using the streptavidin-coated sensor chip (SA chip; Biacore). A solution (2 nM) of a 5Ј-biotinylated oligonucleotide (5Ј-BioTEG) 26:45-mer (Qiagen) in HBS-EP BIAcore buffer was used to yield an increase of 10 RU (1 min) on flow cell 1 and 150 RU (10 min) on flow cell 2. Two other flow cells without an oligonucleotide were used as reference cells. After immobilization, running buffer with low salt was used (30 mM KCl, 0.005% surfactant P20, 10 mM Hepes, pH 7.4). To compare the affinity of enzymes for DNA at different salt concentrations, 50 nM analytes (pol ␥〈, pol ␥〈B 2 , and pol ␥〈B⌬I4) in buffers with different ionic strengths (30 -200 mM for KCl and 1-40 mM for MgCl 2 ) were injected over immobilized oligonucleotide for 8 min followed by 30 s of regeneration using 1 M KCl. In experiments with varied MgCl 2 concentration, 30 mM KCl was used.
Isothermal Titration Calorimetry-Isothermal titration calorimetry (ITC) experiments were performed with a VP-ITC calorimeter (Microcal) at 25°C with a mixing speed of 302 rpm. The WT pol ␥B (160 M) or pol ␥B⌬I4 (130 M) proteins were titrated as 5-l injections (first injection 2 l) into the sample cell containing 7.2 M exo Ϫ pol ␥A. Samples were prepared by dialyzing all interacting components extensively against a buffer containing 20 mM Hepes, pH 8.0, 300 mM KCl, 5% glycerol, 1 mM EDTA, 5 mM 2-mercaptoethanol, 0.2 mM PMSF, 1 g/ml pepstatin A, 5 g/ml leupeptin. The heats of dilution of both proteins with buffer were determined and subtracted prior to analysis. Data were analyzed using the ORIGIN software supplied with the instrument according to the one-and two-binding sites models.
Electrophoretic Mobility Shift Assays-Native gel electrophoretic mobility shift assays (EMSA) were conducted as described (11) using the 26:45-mer primer-template labeled on the 5Ј end of the 45-mer (16), except that the assays were conducted as challenge assays to assess the stability of the DNA-protein complexes. Binding reactions were assem-bled in 30 l containing 4 nM labeled primer-template, 4 nM pol ␥A, and where indicated, 8 nM pol ␥B or pol ␥B⌬I4 (both forms of pol ␥B calculated as a monomer). Binding reactions continued for 10 min. At this time 4 l were removed as a binding control, and 3 l of 200 nM cold 26:45-mer were added as a competitor to bind any protein that dissociated from the labeled complex. A second 4-l sample was removed within 10 s after the competitor was mixed into the reaction and was considered a 0.2-min time point during the time series. The binding control and 0.2-min sample were loaded directly into the lanes of a running native gel. Additional 4-l samples were removed at 1, 2, 4, 8, 16, and in some cases 32 min after addition of the competitor. Following electrophoresis, gels were dried onto Whatman DE81 paper and exposed to a PhosphorImager to quantify the radioactivity of bound and free DNA.
DNA Polymerase Assays-DNA polymerase assays were performed as described (2) with 0.2 pmol of pol ␥A and varied amounts of pol ␥B or pol ␥B⌬I4 in 30-l reactions using either 120 g/ml activated calf thymus DNA or 10 g/ml poly(dA)-oligo(dT) as template. The standard polymerase reaction buffer was 50 mM KCl, 8 mM MgCl 2 , 10 mM Tris, pH 8, 2 mM dithiothreitol, 100 g/ml bovine serum albumin. Where indicated, reactions contained 140 mM KCl. Primer-extension processivity experiments were performed as described by Carrodeguas et al. (2) and in the legend of Fig. 9.

High Level Expression of Both Subunits of Human pol ␥ and
Reconstitution of the Holoenzyme-We and others have been unable to express pol ␥A in bacteria in an intact form. To obtain quantities of recombinant enzyme sufficient for biophysical studies, we expressed the protein in insect cells from a baculovirus construct. Although it is possible to express the holoenzyme entirely in insect cells by co-expressing the A and B subunits, we chose to express the B subunit in bacteria using the T7 system. This permits us to test the activity of a number of pol ␥B variants without making an independent baculovirus construct for each protein of interest. In this study, we compared the binding of wild-type pol ␥B to a deletion derivative that we refer to as pol ␥B⌬I4. This derivative, which contains a glycine residue in place of residues 147-178 of pol ␥B, was first reported by Carrodeguas et al. (11) to test the role of the four-helix bundle observed in the mouse pol ␥B crystal structure in dimerization of the accessory factor. This derivative has some ability to bind to primer-templates along with pol ␥A but has little ability to stimulate activity of the catalytic subunit. Thus, we set out to compare the association of wild-type pol ␥B and pol ␥B⌬I4 to pol ␥A in this study.
Mixing a lysate of insect cells containing His-tagged pol ␥A with a lysate of bacterial cells expressing nontagged pol ␥B permits association of the subunits and purification of pol ␥ holoenzyme by using a threestep procedure. As described under "Experimental Procedures," this procedure consists of affinity chromatography on a Ni-IDA (His-trap) resin followed by ion exchange chromatography and gel filtration. In a typical preparation beginning with 21 g of SF9 cells expressing pol ␥A and 15 g of BL21 cells expressing pol ␥B, we obtained a final yield of ϳ4.5 mg of pol ␥ holoenzyme. pol ␥A is prepared similarly without the addition of pol ␥B. These procedures have provided a sufficient quantity of enzyme for the biophysical studies of the association of the polymerase subunits reported here. The purity of typical proteins is shown in Fig. 1. It is apparent that the amount of the small subunit bound to pol ␥A is diminished by approximately half when the reconstitution is performed with pol ␥B⌬I4 rather than with wild-type pol ␥B.
We previously compared the sedimentation properties of pol ␥B and pol ␥B⌬I4 by using velocity sedimentation in glycerol gradients using a preparative ultracentrifuge (11). In this paper, we characterized these proteins in more detail by using equilibrium ultracentrifugation as  shown in Fig. 2. For wild-type pol ␥B, sedimentation equilibrium distributions were fit to a single species. When the molar mass was treated as the unknown parameter in the global analysis, an average molecular mass of 105,747 Da was obtained (compared with the expected molecular mass of 106,916 Da for dimeric pol ␥B). Attempts to fit the data to a self-association model failed to yield better results. This is consistent with other data suggesting that the wild-type protein is a tightly bound homodimer with effectively no monomer present under these conditions. We estimated that the pol ␥B dimer association is characterized by a K D somewhat tighter than 0.1 nM. In contrast, analysis of data for the I4 mutant was complex. When the data were analyzed using a singlecomponent model the apparent molecular mass calculated by the software was 70 kDa, a value intermediate between a monomer and a dimer. Moreover, the residuals indicated that the single-component model was   Table 1.
a poor fit to the data (data not shown). A better fit was obtained using a self-association model for a monomer-dimer equilibrium. The fitted monomer molecular mass was 48,370 Da, and the dissociation constant (K D ) was 5.8 M. The fact that pol ␥B⌬I4 retains a weak ability to dimerize is consistent with the inference drawn from the crystal structure of mouse pol ␥B. In this model, we estimated that domain 2, which is deleted in pol ␥B⌬I4, contributes 2675 Å 2 to the total dimerization interface of 6850 Å 2 . Thus, it is reasonable that pol ␥B⌬I4 retains a weak ability to dimerize at high protein concentrations but behaves as a monomer at protein concentrations significantly below the K D value, as in most in vivo or in vitro reactions involving pol ␥.
Because the final step in our purification is a high resolution gel filtration step, we applied this as an analytical method to study the stoichiometry of the holoenzyme. Fig. 3 shows the results of a titration in which known molar quantities of pol ␥A and pol ␥B, as determined using the calculated extinction coefficients, were applied to a 60-cm Amersham Biosciences Superdex 200 gel filtration column. The elution profiles of free pol ␥A and pol ␥B are shown in the background of each panel. All added pol ␥B was bound to pol ␥A until saturation was reached at a 1:2 molar ratio of A to B. As the quantity of pol ␥B was further increased, a peak of excess free pol ␥B was apparent. This is consistent with the crystallographic finding that pol ␥B is a dimer. A similar binding titration performed with pol ␥A and pol ␥B⌬I4 showed saturation at a ratio of 1:1 (data not shown).
Isothermal Titration Calorimetry Confirms the 1:2 Stoichiometry of A and B Subunits and Estimates the Affinity of the Interaction-Isothermal titration calorimetry (ITC) was used to measure the enthalpy of holoenzyme formation, stoichiometry of the complex, and the dissociation constant. The ionic strength of the reaction buffer was chosen based on dynamic light scattering data, which demonstrated that aggregation was minimal under these conditions (data not shown). ITC was performed in parallel with both wild-type pol ␥B and pol ␥B⌬I4 (Fig. 4). K D values for complexes of native pol ␥B and pol ␥B⌬I4 subunits with pol ␥A (Table 1) were similar and within experimental error. A 2-fold difference in stoichiometry of the complexes formed by WT pol ␥B or the pol ␥B⌬I4 mutant in binding to pol ␥A was observed, which is a logical consequence of the proposed monomeric nature of the I4 mutant. These results, along with our size exclusion chromatography data (Fig. 3), confirm that the stoichiometry of the complex between pol ␥A and WT pol ␥B is 1:2. Interaction of pol ␥A with the pol ␥B dimer is characterized as an exothermic process with ⌬H of Ϫ6.4 kcal/mol and a K D of ϳ0.15 M. Association of pol ␥A with the monomeric pol ␥B⌬I4 mutant protein is nearly as tight as it is with the wild-type protein.
Because this experiment was performed under stoichiometric binding conditions, i.e. with a pol ␥A concentration (7 M) in excess of the K D value, we considered that this experiment overestimates the K D for the interaction between the subunits.
Surface Plasmon Resonance Analysis of the pol ␥A-pol ␥B Interaction-SPR analysis was employed to study the interaction of pol ␥A with either pol ␥B or pol ␥B⌬I4 in real time. pol ␥A was injected onto the activated sensor chip surface to immobilize the protein to the carboxymethyl-activated dextran to the extent of 1400 resonance units (RU), corresponding to ϳ1.4 ng/mm 2 of covalently linked protein.
According to mass transfer tests, the influence of diffusion on binding of pol ␥B to pol ␥A was negligible in our experimental conditions. Baseline drifts because of covalent linkage ruptures between the immobilized pol ␥A and the dextran matrix were also minimal for at least 10 consecutive ligand injections.
Typical association and dissociation curves characterizing the interaction of pol ␥B with pol ␥A are shown in Fig. 5A. Association and dissociation rate constants were obtained using curve fitting as implemented in the BIAevaluation software version 3.0. Because standard Langmuir kinetics modeling resulted in a poor fit ( 2 Ͼ20), a two-state reaction model was used, which led to considerably better results. Kinetic and equilibrium constants obtained as well as fitting details are summarized in Table 2. Similar experiments with the I4 mutant, which is unable to form a dimer at these low protein concentrations, demonstrated similar association and dissociation kinetics (Fig. 5B). The overall dissociation constant for the wild-type protein is 27 nM. The association between pol ␥B⌬I4 and pol ␥A is weaker by approximately a factor of 2. These measurements are comparable with the apparent K D of 35 nM for the pol ␥B-pol ␥A interaction estimated from polymerase stimulation kinetics by Johnson et al. (5).  Table 2. The overall dissociation constant K D is calculated from the rate constants: K D ϭ 1/(k a1 /k d2 ) ϫ (1ϩ k a2 /k d2 )). The I4 Mutant Stabilizes the Interaction of pol ␥A with Primer-Template DNA Nearly as Well as Wild-type pol ␥B-We were surprised at the relatively small difference in affinity of pol ␥A for binding pol ␥B⌬I4 as compared with the wild-type subunit. Therefore, we conducted experiments to examine the relative ability of these proteins to stabilize the interaction of pol ␥A with DNA primer-templates. One of the standard methods used to study DNA polymerases is to characterize their interactions with oligonucleotide primer-templates, often using an electrophoretic mobility shift assay. We have used this assay in the past to study the interaction of pol ␥B mutants with pol ␥A in primer-template complexes (2,11,16). We have shown that the I4 mutant participates with pol ␥A in binding primer-template, but we have not studied previously the stability of this interaction. We therefore designed a binding challenge assay in which complexes containing different forms of pol ␥ bound to radiolabeled primer-template were assembled and then challenged with a large excess of cold primer-template to follow the kinetics of dissociation. Following addition of a 50-fold excess of cold primertemplate, any polymerase released from the labeled complex would be much more likely to bind the unlabeled competitor than the original radiolabeled oligonucleotide. The kinetics of the dissociation reaction can be followed by withdrawing aliquots from the reaction at various times after addition of competitor and loading them directly on a running native gel. Similar use of EMSA to study dissociation kinetics has been employed since the initial development of this method (21). Fig. 6A shows that pol ␥A alone dissociates quickly from a 26:45 primer-template. The pol ␥AB 2 holoenzyme complex migrates more slowly on the native gel, as expected, and shows considerable stabilization, with a half-life of almost 2 min (Fig. 6B). A parallel EMSA challenge assay performed with pol ␥ reconstituted with the I4 mutant protein revealed that this complex was almost as stable as that formed with wild-type pol ␥B (Fig. 6C).
One limitation of the EMSA to study nucleic acid-protein interactions is that this method provides little ability to vary solution conditions because ions in the reaction buffer are quickly separated from macromolecules as complexes enter the gel. We used SPR to study the primer-template binding of pol ␥A alone and complexes of pol ␥A with either wild-type pol ␥B or pol ␥B⌬I4 at varied concentrations of monovalent and divalent ions (Fig. 7). Each curve in Fig. 7    a For k a1 , k d1 , k a2 , and k d2, the rate constants were obtained from nonlinear least squares global fitting of the respective sensorgrams using SP Revolution software version 3.0 and a model permitting conformational change. b K D (M) is the equilibrium dissociation constant derived from the rate constants by the equation K D ϭ 1/( k a1 /k d1 ) (1 ϩ k a2 /k d2 ). c The errors shown in parentheses are the means Ϯ S.D.
The foregoing results showing that pol ␥B⌬I4 binds pol ␥A with high affinity and stabilizes the polymerase on primer-template oligonucleotides were somewhat surprising in light of our previous work showing that this monomeric derivative had very limited ability to stimulate DNA synthesis by pol ␥A (11). Previous polymerase reactions with this mutant used only low quantities of pol ␥B⌬I4, Ͻ400 fmol, and used only poly(dA)-oligo(dT) as the primer-template. Therefore, we reinvestigated the ability of this derivative relative to wild-type pol ␥B to stimulate pol ␥A activity at higher protein concentrations and on heteropolymeric DNA as well as the synthetic homopolymer. As can be seen in Fig.  8A, in reactions using 0.2 pmol of pol ␥A (6.7 nM) and a poly(dA)oligo(dT) primer-template, DNA synthesis is maximally stimulated by 1 pmol of pol ␥B (0.5 pmol of B 2 dimer; 16 nM). Significantly higher concentrations of pol ␥B⌬I4 were required to stimulate pol ␥A, and the maximal activity obtained was ϳ50% compared with wild-type pol ␥B. This reduced ability to stimulate pol ␥A was observed when activated DNA was used as substrate as well, but a dependence on pol ␥B was only seen at elevated KCl concentrations, as reported previously (2, 4) (Fig.  8B). Although the monomeric pol ␥B⌬I4 did retain some ability to stimulate pol ␥A, the overall reaction rate was always diminished.
We performed single-round primer-extension reactions (Fig. 9) to study polymerase processivity in greater detail. pol ␥ was preincubated with a primer-template consisting of 5Ј-32 P-oligo(dT 16 )-poly(dA). Extension of primers bound by pol ␥ was initiated by addition of TTP mixed with a nonradioactive primer-template to serve as a binding competitor to trap polymerase molecules that dissociated from the radioactive primers. This 26:45-mer oligonucleotide would require dATP as the first incoming nucleotide and is incapable of supporting polymerization with only TTP. Lanes 5 and 6 of Fig. 9 show that wildtype pol ␥ extended primers to products as long as 200 -300 nucleotides (median extension about 60 -80 nucleotides) before dissociation within the first 1.5 min. The size range of products did not continue to increase when the reaction time was doubled. These results and the controls in lanes 7 and 8 of Fig. 9 indicate that the oligonucleotide competitor provided an effective trap to prevent re-utilization of labeled primers. Thus, we conclude that the distribution of product lengths shown in lanes 5 and 6 of Fig. 9 represent the products synthesized following a single polymerase binding event. When pol ␥AB⌬I4 was used in parallel reactions, the primers were typically extended by fewer than 40 nucleotides, indicating that the monomeric pol ␥B⌬I4 was greatly impaired as a processivity factor. Virtually no extension of the primer was seen using pol ␥A alone.

DISCUSSION
pol ␥ Holoenzyme Subunit Stoichiometry-We report the first extensive investigation of the assembly of the pol ␥ holoenzyme. Our sedimentation equilibrium data show that pol ␥B forms a dimer in solution, while analytical gel filtration and isothermal titration calorimetry establish that the pol ␥B dimer binds tightly to the pol ␥A monomer to form a heterotrimer with the structure AB 2 . This is the first detailed study of the association of these subunits in the absence of DNA. Interestingly, the K D of 27 nM for the subunit interaction, which we determined by using SPR, agrees well with the binding affinity measured in the presence of DNA in the enzyme kinetic studies of Johnson et al. (5). The recombinant pol ␥AB 2 complex would have a calculated mass of 245  kDa. This appears to be somewhat larger than estimates of the molecular mass of the polymerase isolated from mitochondria (3,4). We cannot rigorously rule out the possibility that differences in post-translational modification of the recombinant polypeptides or their inclusion of His tags may influence the higher order structure of the polymerase. However, considering the high biological activity of the recombinant enzyme, we think it is more likely that the smaller apparent size of the enzyme purified from mitochondria may be due to the fact that the hydrodynamic measurements in these studies used much lower protein concentrations. It is possible that the holoenzyme may be in equilibrium with A and B 2 subunits during overnight glycerol gradient sedimentation of dilute protein samples, leading to an underestimate of the sedimentation coefficient.
Unfortunately, there is currently no way to estimate the concentrations of polymerase subunits in mitochondria. The interaction between the pol ␥A and B 2 subunits is probably sufficiently tight to prevent frequent dissociation of the holoenzyme to the free subunits in vivo. Our previous experiments revealed no excess of free pol ␥A or pol ␥B subunits when the holoenzyme was purified from crude mitochondrial lysates (2,16). Therefore, we think it unlikely that models for the role of pol ␥ in mtDNA replication and repair should consider any independent roles for the pol ␥ subunits under most circumstances. An exception to this generalization derives from the fact that these two subunits are synthesized and imported into mitochondria separately. Very little is known about the kinetic process by which single subunits are assembled into a holoenzyme in the organelle. Although we cannot rule out the possibility that single subunits may be free to engage in nucleic acid binding, no pool of free subunits has been documented.
We also compared the behavior of wild-type pol ␥B with that of a mutant form designated I4 that lacks a two-helix bundle domain important for dimerization. Analytical ultracentrifugation experiments showed that dimerization of this mutant protein is markedly impaired, with an apparent K D of 5-14 M for the monomer-dimer equilibrium. Nevertheless, pol ␥B⌬I4 retains the ability to bind to pol ␥A in a 1:1 complex with an affinity comparable with that of the wild-type pol ␥B dimer. Despite the difference in stoichiometry, both complexes are formed with nearly the same ⌬G 0 (see Table 1). This implies that most of the interactions that pol ␥A forms with pol ␥B are preserved when it binds pol ␥B⌬I4. As has been suggested for the Drosophila pol ␥ subunits, it is likely that human pol ␥A and pol ␥B share an extensive interaction interface (22). One simple model for the structure of the holoenzyme might propose that a dimeric pol ␥B may provide two identical interfaces for the interaction with pol ␥A. Such a model would be consistent with the assembly of an A 2 B 2 tetramer. This structure, however, is not compatible with our results, which clearly indicate the formation of a heterotrimeric complex. We suggest an asymmetrical model for the interaction between pol ␥A and pol ␥B. According to this model, pol ␥A mainly interacts with one pol ␥B subunit in the pol ␥B dimer, but this interaction sterically inhibits interaction of a second catalytic subunit with the other pol ␥B subunit. When pol ␥A binds to the pol ␥B⌬I4 mutant it is possible that pol ␥A forms additional contacts with pol ␥B surfaces that are exposed in the monomeric pol ␥B⌬I4 permitting binding of this complex to primer-template but inhibiting polymerization. Both models are consistent with our data, and it will be extremely interesting to obtain a crystal structure of the pol ␥ holoenzyme to understand how this unusual processivity factor functions to stimulate the catalytic subunit.
Influence of pol ␥B on pol ␥〈 DNA Binding and Polymerase Activity-We have taken advantage of the contrast between wild-type pol ␥B and pol ␥B⌬I4 to explore the contribution of the accessory subunit to the interaction of the polymerase with DNA primer-templates. By using both electrophoretic mobility shift assays and SPR experiments, we showed that the heterodimeric enzyme formed by association of pol ␥B⌬I4 to pol ␥A binds tightly to primer-template and dissociates slowly in the presence of competitor primer-template. The apparent K D value for the binding of wild-type holoenzyme to a 26:45-mer oligonucleotide at 30 mM KCl is 0.06 nM, although pol ␥A and the pol ␥A-pol ␥B⌬I4 complex bind the same substrate with K D values of 0.6 and 0.4 nM, respectively. These binding affinities are considerably tighter than those reported by Johnson et al. (5) based on polymerase kinetic measurements made at a higher salt concentration, 100 mM NaCl. Several labs have shown that activity of pol ␥A is reduced at high salt but is stimulated by pol ␥〉 (2, 4). Our SPR experiments show both pol ␥B and pol ␥B⌬I4 dramatically stabilize binding of pol ␥A to primer-template at higher salt concentrations (Fig. 7). Like other DNA-binding proteins, pol ␥A presumably displaces cations upon binding to DNA (23). pol ␥B and, to a lesser extent, pol ␥B⌬I4 appear to modify the interaction of pol ␥A with DNA to enable it to resist competition by free cations.
Although pol ␥B clearly increases the processivity of pol ␥A, it affects other activities of the catalytic subunit as well. Johnson et al. (5) have shown that the accessory subunit decreases the K m value for binding nucleotides and increases the polymerization rate, although Longley et al. (10) have demonstrated that these effects result in decreased polym- The reaction shown in lanes 3 and 4 was conducted in the same manner, except that the nonradioactive 26:45 competitor was added to the preincubation mixture before polymerase was added. Lanes 5-8 and 9 -12 show the products of reactions similar to those in lanes 1-4, except that pol ␥AB 2 and pol ␥AB⌬I4 were used, respectively. A PhosphorImager analysis of a polyacrylamide-urea gel to analyze the products of primer extension is shown. Lane 0 shows the starting labeled oligonucleotide in the primer-template. Lane M shows mobility markers of end-labeled MspI fragments of pUC 18 DNA labeled according to size in nucleotides. erase fidelity. Our physical studies comparing the interaction of pol ␥B and pol ␥B⌬I4 with pol ␥A have not directly addressed the influence of these proteins on kinetic parameters of polymerase function. However, we note that the tight binding of pol ␥B⌬I4 to pol ␥A and the consequent stabilization of the enzyme on primer-templates does not lead to stimulation of polymerase activity (Figs. 8 and 9). Indeed, the monomeric pol ␥B⌬I4 may have a weak ability to act as a dominant negative form of pol ␥B, because it competes well for binding to pol ␥A but is very ineffective in stimulating polymerization. Experiments to test this proposal are under way.
Comparison of pol ␥B to Other Processivity Factors-The available data suggest that the interaction between pol ␥B and pol ␥A is structurally distinct compared with that of most other pairs of processivity factors with their cognate polymerases. pol ␥B is structurally unlike any of the sliding clamp processivity factors such as proliferating cell nuclear antigen, E. coli pol III ␤, and T4 phage glycoprotein 45 that stabilize polymerase binding to DNA by encircling the DNA duplex (24). In these examples, as well as in the binding of thioredoxin to T7 phage DNA polymerase, the interaction of the accessory subunit with the polymerase is mainly mediated by a small interface that can be delimited to a peptide domain (25,26). This is also true for cytomegalovirus UL44 (27,28) and herpes simplex virus UL42 (29) proteins that do not form complete toroidal rings. Breyer and Mathews (30) have suggested that processivity can be conferred in either of two ways, by topological linkage, as illustrated by sliding clamps, or through the involvement of a large interaction surface. We suggest that pol ␥ may exemplify this latter category.
As noted in the Introduction, pol ␥ provides a unique example of an essential DNA polymerase that has a variable structure in different organisms. The obvious relationship between mammalian pol ␥B and homodimeric prokaryotic tRNA synthetases (11) has been cited as an illustration of horizontal gene transfer during evolution (31). An exceptional contrast is provided by Drosophila pol ␥, which is a heterodimer containing one copy each of the A and B subunits (1,15). Although Drosophila pol ␥B shares only 15% sequence identity with human pol ␥B, the only significant internal deletion in the alignment of the two proteins suggests that the Drosophila protein contains a discrete deletion of domain 2 of the mammalian proteins containing the two-helix structure, a deletion similar to that used to generate human pol ␥B⌬I4. We suggest that the Drosophila protein may have experienced a deletion of this domain during evolution from a common ancestor with the mammalian lineage. Our observation that pol ␥B⌬I4 is able to stabilize pol ␥ on primer-templates suggests that this deleted protein could retain substantial function. Additional mutations may have occurred to enable monomeric Drosophila pol ␥B to stimulate its catalytic partner. These mutations may have occurred either before or after deletion of this domain. It will be of great interest to compare the structures of Drosophila and mammalian pol ␥ holoenzymes.