Exonucleolytic proofreading enhances the fidelity of DNA synthesis by chick embryo DNA polymerase-gamma.

The high fidelity of chick embryo DNA polymerase-gamma (pol-gamma) observed during in vitro DNA synthesis (Kunkel, T. A. (1985) J. Biol. Chem. 260, 12866-12874) has led us to examine this DNA polymerase for the presence of an exonuclease activity capable of proofreading errors. Highly purified chick embryo pol-gamma preparations do contain exonuclease activity capable of digesting radiolabeled DNA in a 3'----5' direction, releasing deoxynucleoside 5'-monophosphates. The polymerase and exonuclease activities cosediment during centrifugation in a glycerol gradient containing 0.5 M KCl. In the absence of dNTP substrates, this exonuclease excises both matched and mismatched primer termini, with a preference for mismatched bases. Excision is inhibited by the addition of nucleoside 5'-monophosphates to the digestion reaction. In the presence of dNTP substrates to permit competition between excision and polymerization from the mismatched primer, the exonuclease excises mismatched bases from preformed terminal mispairs with greater than 98% efficiency. The preference for excision over polymerization can be diminished by addition of either high concentrations of dNTP substrates or nucleoside 5'-monophosphates to the exonuclease/polymerase reaction. To determine if this exonuclease is capable of proofreading misinsertions produced during a normal polymerization reaction, a sensitive base substitution fidelity assay was developed based on reversion of an M13mp2 lacZ alpha nonsense codon. In this assay using reaction conditions that permit highly active exonucleolytic proofreading, pol-gamma exhibits a fidelity of less than one error for every 260,000 bases polymerized. As for terminal mismatch excision, fidelity is reduced by the addition to the synthesis reaction of high concentrations of dNTP substrates or nucleoside 5'-monophosphates, both hallmarks of exonucleolytic proofreading by prokaryotic enzymes. Taken together, these observations suggest that the 3'----5' exonuclease present in highly purified chick embryo pol-gamma preparations proofreads base substitution errors during DNA synthesis. It remains to be determined if the polymerase and exonuclease activities reside in the same or different polypeptides.

The high fidelity of chick embryo DNA polymerasey (pol-y) observed during in vitro DNA synthesis (Kunkel, T. A. (1985) J. Biol. Chern. 260,[12866][12867][12868][12869][12870][12871][12872][12873][12874] has led us to examine this DNA polymerase for the presence of an exonuclease activity capable of proofreading errors. Highly purified chick embryo pol-y preparations do contain exonuclease activity capable of digesting radiolabeled DNA in a 3' + 5' direction, releasing deoxynucleoside 5'-monophosphates. The polymerase and exonuclease activities cosediment during centrifugation in a glycerol gradient containing 0.5 M KCl. In the absence of dNTP substrates, this exonuclease excises both matched and mismatched primer termini, with a preference for mismatched bases. Excision is inhibited by the addition of nucleoside 5'-monophosphates to the digestion reaction. In the presence of dNTP substrates to permit competition between excision and polymerization from the mismatched primer, the exonuclease excises mismatched bases from preformed terminal mispairs with greater than 98% efficiency. The preference for excision over polymerization can be diminished by addition of either high concentrations of dNTP substrates or nucleoside 5'-monophosphates to the exonuclease/polymerase reaction. To determine if this exonuclease is capable of proofreading misinsertions produced during a normal polymerization reaction, a sensitive base substitution fidelity assay was developed based on reversion of an M13mp2 lacZa nonsense codon. In this assay using reaction conditions that permit highly active exonucleolytic proofreading, pol-y exhibits a fidelity of less than one error for every 260,000 bases polymerized. As for terminal mismatch excision, fidelity is reduced by the addition to the synthesis reaction of high concentrations of dNTP substrates or nucleoside 5'-monophosphates, both hallmarks of exonucleolytic proofreading by prokaryotic enzymes. Taken together, these observations suggest that the 3' + 5' exonuclease present in highly purified chick embryo pol-y preparations proofreads base substitution errors during DNA synthesis. It remains to be determined if the polymerase and exonuclease activities reside in the same or different polypeptides.
A wealth of genetic and biochemical observations have demonstrated that low spontaneous mutation rates in prokaryotes result in part from proofreading of DNA synthesis errors by the 3' + 5' exonuclease activity associated with prokar-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "aduertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. yotic DNA polymerases (for review, see Ref. 1). Of the four classes of DNA polymerases found in higher eukaryotes (for review, see Ref. 21, DNA polymerase-d has an associated 3' -5' exonuclease activity (3,4) that has recently been shown to proofread errors (5). Highly purified preparations of two other classes of higher eukaryotic DNA polymerases, LY and (3, have generally been found to lack associated nuclease activity (2). Although several reports do describe exonuclease activities associated with these DNA polymerases (6-lo), a role for these exonucleases in proofreading has not been established.
The fourth class of DNA polymerases in animal cells is DNA polymerase-y (pol-y).' This enzyme is responsible for replication of mitochondrial DNA and may be involved in DNA repair processes as well (2). It is found in both the mitochondrion and the nucleus and is present as only 1-10% of the total cell DNA polymerase activity. Because of this low abundance as well as its heterogeneity and instability during isolation, pol-y has been difficult to purify to homogeneity for detailed characterization. Despite these difficulties the enzyme has been substantially purified from a variety of sources (11)(12)(13)(14)(15)(16)(17)(18)(19), and reviewed in Ref. 2), including to near homogeneity from chick embryos (20). The chick embryo poly preparation, purified 1,500,000-fold to a very high specific activity, is both highly processive (21) and highly accurate for several different types of errors during in vitro DNA synthesis with natural DNA (22)(23)(24). The fidelity results have led us to examine chick embryo pol-y preparations for an 3' + 5' exonucleolytic activity that could function to proofread errors. Such an activity, which fulfills the established criteria for a proofreading exonuclease, has been detected.
For simplicity mutant M13mp2 derivatives containing a single base substitution will be described by the new base present at a particular position within the viral (plus) strand. For example, the nonsense codon used in reversion studies presented here resulted from a change of a G to an A at position 89 (where position 1 is the first transcribed base of the lacZa sequence), creating a TGA opal codon. This mutant is therefore designated M13mp2A89 or simply A89. To determine the exact nature of the base change for any mutant, refer to the wild type M13mp2 lacZa sequence (26). (22,25,26). Mutant A102, containing a phenotypically silent base change, was constructed by a variation of the published procedure for oligonucleotide-directed mutagenesis (27), the details of which will be published elsewhere? The eight mutants used for the heteroduplex expression experiments were blue revertants of the opal codon that was created by the A89 mutation and were obtained from several sources' (5).

DNA Polymerases
The chick embryo DNA polymerase-y used in this study was purified by Yamaguchi and co-workers as described (20). Briefly, this entails purification from a crude extract by phosphocellulose chromatography, ammonium sulfate fractionation, and successive column chromatography steps using phosphocellulose, Sephadex G-200; hydroxylapatite, and finally double-stranded DNA-cellulose. Assayed with the preferred template-primer poly(rA),. oligo dT12.18, the final fraction, which was the fraction used in the present studies, was purified 1,500,000-fold to a specific activity of 570,000 units/mg protein. Over the course of these studies, three independently purified preparations of pol-y were used. Similar results (i.e. mismatch excision, high fidelity) were obtained with all three preparations. The source and purity of both rat and chick embryo pol-8 (26, 28) have been described. T4 DNA polymerase, pol I, and AMV pol were from PL-Pharmacia. Pol I (KO was from either PL Pharmacia or Boehringer Mannheim. Restriction endonucleases AuaII, KpnI, and PuuII as well as other reagents were from commercial sources as described (25,26,29).

Construction of Radiolabeled Substrates for Exonuclease Assays
To prepare the 3'-end-labeled substrate for 3' + 5' exonucleolytic activity, M13mp2 DNA containing a 390-base gap (constructed as described below) was used. Polymerization to fill the gap was performed in a 200-pl reaction containing 20 mM Tris-HC1 (pH 7.5), 2 mM dithiothreitol, 10 mM MgC12, 10 p M each of dATP, dTTP, dGTP, and [a-3ZP]dCTP (18,000 cpm/pmol), 5 pg of gapped M13mp2 DNA, and 5 units of Kf. Incubation was at 37 "C for 20 min and the reaction was stopped by addition of EDTA to 15 mM. Unincorporated radioactively labeled dCTP was removed by two successive cycles of chromatography on Sephadex G-50 columns (Boehringer Mannheim) following the manufacturer's instructions. The DNA was precipitated with ethanol, resuspended in TE buffer (10 mM Tris-HC1 (pH 8.0), 0.1 mM EDTA) and used as a substrate for the 3' + 5' exonuclease assays shown in Fig. L4.
To prepare the 5"end-labeled DNA substrate, an unphosphorylated 15-base oligonucleotide complementary to positions 177-191 of the lacZa-coding sequence in M13mp2 was phosphorylated at the 5' end in a 50-pl reaction containing 50 mM Tris-HC1 (pH 7.5),5 mM dithiothreitol, 10 mM MgC12, 50 pCi [y3'P]ATP (5000 Ci/mmol), 36 ng of oligonucleotide and 10 units of T4 polynucleotide kinase. Incubation was at 37 "C for 30 min, and the reaction was terminated by incubation at 70 "C for 5 min. To this labeled oligonucleotide was added 50 pg of single-stranded viral, circular M13mp2 DNA and standard sodium citrate (SSC) to a final 1 X concentration (i.e. 150 mM NaCI, 15 mM sodium citrate). Hybridization was performed by placing the mixture (80 p1 in a 1.5-ml microcentrifuge tube) into a 500-ml beaker of water at 70 "C and allowing the beaker to cool to 37 "C. To this was added 120 pl of polymerization mix to produce a labeled DNA described above, except that the dNTP substrates (at 2 0 0 4 reaction containing the same components as for the w3'P-100 p~) were all unlabeled. Incubation to permit polymerization was at 37 'C for 15 min, and the reaction was terminated by addition of EDTA to 15 mM. Unincorporated [Y-~~PIATP was removed and the DNA processed as described above. Analysis of the product DNA on an 8% polyacrylamide-sequencing gel demonstrated that all the 32Plabel was in high molecular weight (>500 bases) material. This 5'end-labeled DNA was used as a substrate for 5' + 3' exonuclease activity (Fig. l e ) . ' T. A. Kunkel and K. Bebenek, unpublished observations. ' Two peaks of DNA polymerase-y activity were resolved at this step, having molecular mass of 280,000 and 180,000 daltons. Further purification and characterization demonstrated no other obvious differences between these two preparations. Thus, during subsequent purifications of pol-y, including the preparations used in this study, these two peaks were combined and further purified as described (20).
Exonuclease Assays with End-labeled DNA Exonuclease assays were performed in 50-pl reactions containing 20 mM Hepes (pH 7.8), 2 mM dithiothreitol, 10 mM MgCIz, and either 200 ng of 3"end-labeled DNA (-550 cpm/ng), denatured just before use by incubating at 95 "C for 5 min, or 1.8 pg 5"end-labeled DNA (34,000 cpm/pg) and the amount of DNA polymerase indicated in the legend to Fig. 1. At the indicated times aliquots were processed for determination of acid-insoluble radioactivity as described (30). At the final time point, 2 p1 of the reaction were analyzed by polyethyleneimine cellulose chromatography as described (31).

Electrophoretic Analysis of Terminal Mismatch Excision
A 15-base oligonucleotide complementary to positions 106-120 of the lacZa sequence of M13mp2 was phosphorylated at the 5'-end as described above for the preparation of the 5"end-labeled exonuclease substrate. This oligonucleotide was hybridized (as above) to a %fold molar excess of either wild type M13mp2 viral DNA, to create a correct C(temp1ate) .G(primer)' base pair at the 3"OH end, or to an M13mp2 mutant viral DNA containing a minus C frameshift mutation at position 106, to create an incorrect A(temp1ate position 105). G(primer) mispair at the 3'-OH end. Terminal excision reactions (25 pl) contained 20 mM Hepes (pH 7.8), 2 mM dithiothreitol, 10 mM MgCl,, 300 ng of M13mp2 DNA (matched or mismatched) and 0.9 unit of pol-y. Aliquots (5 pl) were removed after 0, 5, 15, 30, and 60 min of incubation at 37 "C, into 5 pl of dye mix (90% formamide, 5 mM EDTA, 0.1% bromphenol blue, 0.1% xylene cyanol). Electrophoretic analyses of 3-pl aliquots were performed in 20% polyacrylamidesequencing gels (Fig. 2), which were dried and used to expose Kodak XAR film. The radioactivity in each band was quantitated by cutting the bands from the gel and counting radioactivity in a Beckman LS7800 liquid scintillation counter.

Glycerol Gradient Centrifugation
DNA polymerase-y, in a volume of 200 pl, was layered onto a 4.8ml linear 10-30% (v/v) glycerol gradient containing 50 mM Tris-HC1 (pH 7.6), 1 mM dithiothreitol, 0.5 mM EDTA, and 0.5 M KC1. Centrifugation in polyallomer tubes was at 50,000 rpm for 13 h at 2 "C in a Beckman SW 55 Ti rotor. For comparison, rabbit aldolase (8.3 S, 158,000 daltons) was used as a standard in a parallel gradient.
Seven drop (-140 pl) fractions were collected from the top using a Buchler Auto Densi-flow IIC fraction collector. Polymerase activity for a 60-min incubation at 37 "C was determined using 15 pl of each fraction in a 60-pl reaction containing 50 mM Tris-HC1 (pH 8.0), 5 mM 2-mercaptoethanol, 7.5 mM MgCl,, 0.5 mM EDTA, 150 mM KC], 100 pg/ml activated DNA, and 15 p~ each of dATP, dGTP, dCTP, and 3H-dTTP (470 cpm/pmol). Exonuclease activity in a 6 -4 aliquot of each fraction was determined during a 60-min incubation at 37 "C using the gel electrophoresis assay for terminal mismatch excision described above.
Preparation of M13mp2 DNA Substrates Three types of M13mp2 DNA substrates were used: gapped heteroduplex molecules containing 3"OH terminal mispairs (experiments shown in Fig. 3 and Table I), completely double-stranded heteroduplex molecules containing an internal single-base mispair (experiments shown in Tables I and II), and gapped molecules containing an opal codon within the single-stranded (template) DNA (experiments shown in Table HI). For each of these constructions, singlestranded viral (plus) DNA and double-stranded replicative form (RF) DNA were obtained as described in Refs. 26 and 29, respectively.

Construction of Gapped Heteroduplexes Containing 3 ' Terminal
Mispairs"Ml3mp2 G103 RF DNA, containing unique sites for both restriction endonuclease AuaII and KpnI (this site was actually created by the T + G base change at position 103) was digested in Base pairs will be described by listing the base in the template strand first, then the base in the primer strand.
individual reactions with each of these enzymes to determine the amount required to digest the DNA to completion (as determined by analysis in agarose gels). The reaction was then scaled up to digest 500 pg of DNA with both enzymes at once. This reaction produces two fragments, 6826 and 363 base pairs in length, each with one 4base (3'-OH) single-stranded end and one 3-base (5'-P) singlestranded end. These fragments are sufficiently different in size to permit separation by precipitation with polyethylene glycol by a variation (32) of the original procedure (33, 34) as follows. The digested DNA was phenol extracted, ethanol precipitated, and resuspended in TE buffer as described (29). The solution was adjusted to 6.0% PEG-8000, 0.55 M NaCl, and a final DNA concentration of 0.4 mg/ml. This mixture, in sterile 1.5-ml microcentrifuge tubes, was incubated at 37 "C overnight. The precipitated 6826-base pair fragment was pelleted by centrifugation for 10 min and the supernatant carefully removed. After resuspending the pellet in TE buffer, an aliquot of this sample and the supernatant were analyzed by agarose gel electrophoresis. The resuspended pellet fraction was highly enriched for the 6826-base pair fragment, containing only a trace of the 363-base pair fragment. Likewise, the supernatant was highly enriched for the small fragment and contained only a trace of the large fragment. The 6826-base pair fragment was precipitated with PEG a second time just as before, yielding a preparation lacking detectable small fragment as determined by agarose gel analysis. To remove residual PEG, the large fragment was precipitated with ethanol and then resuspended in TE buffer to a final concentration of 1 mg/ml. The final yield was -90-95% of the starting material.
To form the gapped terminal mismatched heteroduplexes, the large fragment was hybridized to single-stranded circular, viral (plus) DNA, either wild type, A103 or A102 (as shown in Table I) as follows. The fragment was diluted 10-fold with water to reduce the concentration of both DNA and salt and incubated at 70 "C for 5 min to denature the strands. Sufficient viral DNA was added just before removal from the 70 "C water bath to produce a 1:l molar ratio of minus strand fragment to viral, circular DNA. The mixture was placed on ice and SSC was added to a final 2 X concentration (300 mM NaCl, 30 mM sodium citrate). The mixture was incubated at 60 "C for 5 min, then placed on ice. The DNA was ethanol precipitated, resuspended in TE buffer, and a small aliquot analyzed by agarose gel electrophoresis. Using this 1:l fragment to viral DNA ratio, typically about one-half of the single-stranded DNA was converted to a heteroduplex molecule containing a 363-base gap from position -261 to 102, with a terminal mispair at position 103. The concentration of this heteroduplex was estimated by visual inspection of the gel in comparison to a known quantity of an RF DNA standard.

Construction of Complete Heteroduplexes Containing Internal
Mispairs-The appropriate mutant RF DNA was digested to completion with restriction endonuclease AuaII, which incises only once in M13mp2 at position -264. This genome-length linear DNA was used to form a completely double-stranded but nicked heteroduplex molecule by denaturing, then annealing the minus strand, to mutant single-stranded circular, viral (plus) DNA as described above.

Construction of Gapped Template Containing an
Opal Codon-M13mp2A89 RF DNA was digested to completion with restriction endonuclease PuuII to produce three blunt-ended fragments of 6835, 268, and 93 base pairs. As before, the largest fragment can be selectively precipitated with 6% PEG-8000, 0.55 M NaCl. However, unlike the previous separation of fragments with sticky ends which required two successive precipitations at 37 "C to produce uncontaminated large fragment, these blunt-ended fragments yielded essentially pure large fragment after a single PEG precipitation step at 0 "C overnight. Hybridization with viral M13mp2 A89 DNA was performed as above to produce a double-stranded homoduplex molecule with a 361-base single-stranded gap from position -216 to 145 and containing the TGA nonsense codon at positions 87,88, and 89.
Both the M13mp2 mismatch excision assay and the opal codon reversion assay for base substitution fidelity are described in detail under the "Results" section.

DNA Polymerase Reactions with Ml3mp2 Primer Templates
Reactions contained 20 mM Hepes (pH 7.8), 2 mM dithiothreitol, and 10 mM MgC12. The variables, including reaction volume, time, amount of DNA polymerase, amount of gapped DNA (containing either the terminal mismatch or the opal codon), and the concentrations of the deoxynucleoside triphosphates and deoxyguanosine monophosphate are given in the legends to the tables and figures. After incubation at 37 "C, reactions were terminated by addition of EDTA to a final concentration of 15 mM. Twenty pl of each reaction was analyzed by agarose gel electrophoresis as described (26). All polymerase reactions reported here generated products that migrated coincident with a replicative form 11, fully double-stranded DNA standard.

Preparation of Competent Celk, Transfection, and Plating
The products from DNA polymerase reactions with M13mp2 primer templates were used to transfect competent cells prepared from E. coli strain MC1061 by treatment with CaCI2 as described (35). This strain was chosen because of its high transfection efficiency relative to previously used strains. With this strain, one ng of untreated double-stranded M13mp2 DNA yields 10,000-50,000 (infective center) plaques. Single-stranded DNA is much less infective, typically yielding 100-1,000 plaques/nanogram. Plating was performed by adding the transfection mixture to 2.5 ml of 0.8% soft agar (in 0.9% NaCl and at 4&52 "C) containing 2.5 mg of 5-bromo-4chloro-3-indoyl-~-~-thiogalactoside, 0.24 mg of isopropylthio-D-Dgalactoside, and 0.4 ml of a mid-log phase culture of E. coli CSH50 (25). This mixture was poured onto plates containing 30 ml (solidified) 1.5% agar in minimal medium and allowed to solidify. The plates were inverted and incubated for 18-24 h at 37 "C, followed by an additional 24-48 h incubation at room temperature. For either the heteroduplex expression controls or the mismatch excision experiments, where subtle color distinctions are important, sufficient transfection mixture was used to obtain 100-500 plaques/plate. For the reversion assay, where more (colorless) events must be screened to observe rare (blue) revertants, plating was performed with a small amount of the transfection mixture on four to six plates to obtain countable plates (100-500 plaques) for calculation of the total plaques monitored. The remainder of the mixture was plated to achieve higher plaque densities (1,000-10,000 plaques/plate). Control experiments (not shown) demonstrated that even faint blue plaques can be quantitatively scored on confluent plates containing as many as 10,000 colorless plaques.

Comments on Improvements in M13mp2 Assay
Our efforts to probe fidelity mechanisms with natural DNA depend on the ability to construct precisely defined primer templates of high quality and in high yield. DNA must be prepared by methods least likely to generate cryptic damage to the template-primer, a concern for in vitro fidelity assays (23). High yield is important due to the expense of certain restriction endonucleases and since more DNA is required for reversion assays which focus on specific mutational pathways. For these reasons the following modifications of previously published procedures (26) are worth stressing. During the purification of RF DNA by the Birnboim and Doly procedure (36), the 60 "C incubation with LiCl was omitted (to reduce deamination) and, after RNase treatment, the DNA was simply phenol extracted, ethanol precipitated, and resuspended. Fragments required to produce gapped duplexes were purified not by gel electrophoresis but by selective PEG precipitation. This gives pure DNA in high yield without subjecting the DNA to lengthy treatments which could possibly introduce damage. Denaturation of the fragment at 70 "C (rather than 95 'C) was possible due to the low salt and DNA concentrations used. The template was not added until after the denaturation step so that the level of template cytosine deamination was reduced. Finally, the gapped duplex was not purified by gel electrophoresis to remove primer or residual single-stranded DNA, resulting in good yield while minimizing the possibility of damage. Gel purification is not required, since at the primer to template ratios used here (l:l), the contributions of either residual primer or residual single-stranded viral DNA to total biological activity are negligible relative to that of doublestranded circular DNA. This is true for cells made competent by either the CaC12 (35) or Hanahan (37) procedures.

M13mp2 Mutant Color Discrimination
We have established a scale of blue color intensities for M13mp2 mutants relative to that of wild type M13mp2. Using the plating conditions described above, wild type M13mp2 generates dark blue plaques which, on a scale of 0-4, are assigned a value of 4. Mutants can then be readily described as O+ (colorless), 1+ (faint blue), 2+ (medium blue), or 3+ (almost wild type). These subjective designations are assigned by directly comparing relative color intensities of two derivatives on the same plate, thus permitting subtle distinctions to be made reproducibly and with confidence.

3' + 5' Exonuclease Activity with End-labeled
DNA-Chick embryo pol-y was previously found to be accurate for certain mutational pathways during in uitro DNA synthesis (23,24). This led us to examine pol-y for the presence of an exonuclease that could proofread errors. We began by first determining if exonucleolytic activity capable of digesting radiolabeled DNA could be detected. As shown in Fig. lA, nuclease activity capable of digesting the denatured 3'-endlabeled M13mp2 DNA is detectable in the pol-y preparation. The products of this reaction were deoxynucleoside 5'-monophosphates, since upon analysis by thin layer chromatography using polyethyleneimine cellulose, all the released radioactivity migrated coincident with the appropriate deoxynucleoside 5'-monophosphate standard (dCMP). These observations indicate the presence of a 3' + 5' exonuclease. Further evidence that the exonuclease activity digested in the 3' 4 5' direction but not in the 5' + 3' direction is provided by the observation that the pol-y preparation digested little of the 5'-end-labeled substrate (Fig. 1B). The 3' + 5' exonuclease activity was not observed in the absence of MgCl, (data not shown). Both E. coli pol I and T4 DNA polymerase were used as positive controls for 3' + 5' exonuclease activity. All three DNA polymerases were also concomitantly assayed under identical conditions for DNA polymerase activity using activated DNA. From these data the polymerase to exonuclease ratios were calculated (see legend to Fig. 1) to be 35:l for T4 DNA polymerase, 88:l for the pol-y preparation, and 17001 for E. coli pol I. The 3' + 5' proofreading exonuclease activities associated with E. coli pol I (38,39) and pol I11 (40) and mammalian polb (3)(4)(5) are all inhibited in the presence of nucleoside monophosphates. We therefore examined the effect of monophosphates on the nuclease activity in the pol-y preparation. Addition of 10 mM AMP inhibited this nuclease activity by 54% (Fig. lA, open circle), while the 3' + 5' exonuclease activity of pol I was inhibited 80% (Fig. lA, open triangle).
Terminal Mismatch Excision by Gel Electrophoresis-We next examined the ability of the 3' + 5' exonuclease to excise a 3"terminal base from either a matched or mismatched end. T w o substrates were prepared, by hybridizing a 5"end-labeled 15-base oligonucleotide to two different single-stranded DNA templates, producing either a correct C. G base pair or an incorrect A.G mispair. Excision was performed over a 1-h time course and was monitored by gel electrophoresis in a 20% polyacrylamide-sequencing gel. The results (Fig. 2) confirm the presence of the exonucleolytic activity which excises in the 3' + 5' direction, since 5' + 3' excision would remove the label and the resulting products would not be detected. These results further demonstrate the absence of a 5' + 3' exonuclease, since there was no loss of total radioactivity upon counting all the bands for any lane shown in Fig. 2. The presence of a series of oligonucleotides from 15 to 10 bases in length and differing by one base supports the previous product analysis demonstrating that single mononucleotides are excised. Both matched (Fig. 2, lanes 2-5) and mismatched (lanes  7-10) bases are removed, with a 2-to 6-fold preference for excision of the G from the A. G mispair over excision from the matched C G base pair. Exonuclease activity is inhibited by addition of monophosphates (Fig. 2, lanes 12-15); the extent of inhibition using 10 mM dGMP ranges from 69% (lane 12 versus 7) to 42% (lane 15 versus 10).
Cosedimentation in Glycerol Gradients-To determine if the polymerase and exonuclease activities could be separated, their sedimentation profiles during glycerol gradient centrifugation in 0.5 M KC1 were determined. The two activities cosediment (Fig. 3), each as a single peak at 9.4 S (-180,000 daltons), and the ratio of polymerase to exonuclease remains relatively constant across the peak fractions.
Terminal Mismatch Excision in DNA Polymerase Reactions-We next examined the terminal mismatch excision capability under conditions where it is possible to establish a competition between excision and polymerization from the terminal mispair. To do this we constructed gapped, biologically active M13mp2 DNA substrates containing either a T. C or an A. C mispair, where the two bases of the mispair can be distinguished from each other by the intensity of blue color in an M13mp2 plaque produced upon transfection of the DNA. For example, for the T. C mismatch, polymerization to fill the gap without excision of the mismatched 3"terminal cytosine will produce a double-stranded heteroduplex, which upon transfection will yield both 2+ and 4+ blue plaques: 2+ when the cytosine-containing minus strand is expressed and 4+ when the thymine-containing plus strand is expressed. These two phenotypes are easily distinguished from each other. However, if the mispaired cytosine is removed prior to extension by the polymerase, subsequent correct incorporation of adenine opposite the template thymine will yield a homoduplex molecule having exclusively a 4+ blue plaque phenotype. The proportion of 2+ and 4+ blue plaques obtained upon  1.8, 1.1, 1.0, 1.1, 0.85, and 0.85, respectively, where the peak fraction is assigned an arbitrary value of 1.0 to describe the ratio of cpm digested to cpm incorporated. transfection of the reaction products thus describes the extent of terminal mismatch excision prior to polymerization. The logic is identical for the A-C mismatch, but the blue color intensities are different (see legend to Table I).
Both the T -C and A-C gapped, mismatched molecules were used as template-primers in polymerization reactions catalyzed by several DNA polymerases, including pol-y. The products of an aliquot of each of these reactions were analyzed by electrophoresis in an agarose gel to monitor gap-filling DNA synthesis. The 363-base gap was filled by pol-y to the extent that the product migrated coincident with the completely double-stranded DNA standard (data not shown). Within the limits of detection of this analysis, no uncopied DNA was observed. Similar results were obtained for each DNA polymerase and reaction condition used in these studies, both for terminal mismatch excision and for fidelity studies.
The remaining gap-filled DNA was used for transfection of competent cells to score the colors of the resulting plaques (Table I). Gap-filling polymerization by either AMV DNA polymerase or DNA polymerase-& both of which lack associated 3' + 5' exonuclease activity7 (41, 42), occurs without substantial excision of the T. C or A. C mismatches since the minus strand phenotype was observed at a frequency of 46-53%. In contrast, Kf removes 92% (T-C) or 88% (A-C) of the terminal mismatch prior to extension to fill the gap. Pol-y removes the mismatches even more efficiently, excising 98% (T. C) and 99% (A. C) of the mispaired cytosine prior to filling the gap. In all cases, the extent of excision prior to polymerization is calculated (see legend to Table I) by comparison to two uncopied DNA controls (Table I). First, polymerization without excision creates an internal mispair within a complete heteroduplex. Transfection of such a heteroduplex yields a minus strand phenotype of 60%; a value similar to the results obtained by extension with non-exonuclease-containing pol-/ 3 and AMV DNA polymerase. Second, the opposite extreme is the background frequency of minus strand phenotype plaques. This value is 0.37%, the lowest value attainable upon complete excision of the terminal cytosine prior to polymerization.
Two variations in reaction conditions were examined for their effects on the efficiency of excision of cytosine from the terminal A.C mispair by the exonuclease. First, the ratio of exonuclease to polymerase activity may be reduced by increasing the concentration of dNTP substrates in the reaction. The expected result is an increase in the 2+ blue (minus strand) phenotype. This is exactly what is observed for both Kf and pol-y (Fig. 4A). Comparing Kf reactions using 1 p M versus 1000 p~ dNTPs, the increase in minus strand phenotype frequency (the next nucleotide effect) is 70-fold. Pol-7 is even more efficient at terminal mismatch excision since at all comparable dNTP concentrations the minus strand expression values are lower than for Kf. This remains so even at the highest dNTP concentrations used (2000 pM). The maximum obtainable next nucleotide effect, comparing values at 10 p M versus 2000 p~ dNTPs, is 14-fold. Over the range of dNTP concentrations examined, there is no next nucleotide effect using AMV DNA polymerase, which lacks a 3' + 5' exonuclease activity.
In the absence of DNA synthesis, the 3' + 5' exonuclease activity in the pol-y preparation can be inhibited by addition ' T. A. Kunkel, unpublished observations. 8This value, as well as those shown in Table 11, represents an approximate 2-fold increase in minus strand expression over our previous measurements (23). We attribute this to the improved method of construction of gapped DNA described in the text. The same increase over previous results has also been observed for other mispairs as well as for one-base frameshift heteroduplexes. Under some conditions, a substantial number of colorless mutants were observed. These mutants (included here) were analyzed and found to result from aberrant synthesis from the cytosine without excision and thus were scored as minus strand phenotype. The complete details of this analysis are beyond the scope of this study and will be described elsewhere!
The value shown is for the complete heteroduplex containing the internal T. C mispair. A similar value was obtained for the A.C mispair.
The gapped heteroduplex used here is a control containing a T. C mispair followed by a template A, designed to determine the background 2+ blue minus strand phenotype resulting from undigested M13mp2G103 RF DNA used in the construction. The logic is as follows: this gapped heteroduplex should give a 3+ minus strand phenotype and a 4+ plus strand phenotype. However, the value listed in the table is actually the number of 2+ blue mutants observed upon transfection of this heteroduplex. These could result either from polymerase errors during gap-filling synthesis in the k Z a target or from undigested 2+ blue M13mp2G103 RF DNA. Sequence analysis of these 2+ blue mutants demonstrated the presence of the GGT (2+) codon of the G103 mutant rather than the scattered mutants that would result from polymerase errors. Since this outcome is not predicted by the sequence of either strand for the intended construction, these mutants presumably reflect incomplete digestion of G103 of nucleoside 5'-monophosphates to the reaction (Figs. 1 and  2). This inhibition of exonucleolytic excision is also observed during a polymerization reaction. At a constant 100 PM dNTP concentration, addition of increasing concentrations of dGMP to pol-y reactions resulted in decreased A.C terminal mismatch excision (Fig. 4B). As expected, the 3' + 5' exonuclease of Kf is also inhibited by dGMP. As with the next nucleotide effect, at a comparable dGMP concentration, Kf was more effectively inhibited than the exonuclease in the pol-y preparation. No monophosphate effect is observed with AMV DNA polymerase.
Since even the highest concentrations of dNTPs or dGMP used did not abolish terminal mismatch excision by Kf and the 3' + 5' exonuclease in the pol-y preparation, these two I ' ' ' ' ' " " ' ' ' ' ' " " ' 60 I" conditions were employed in concert. The results (Fig. 4A, open symbols) demonstrate that at 1 PM dNTP and 20 mM dGMP, Kf excision of cytidine from the A. C mispair is inhibited about 90%, while for pol-y excision was only inhibited 50% even at 2000 W M dNTPs and 20 mM dGMP.
Exonuclease Proofreading during Synthesis-To determine if proofreading occurs during DNA synthesis, we looked for next nucleotide and monophosphate effects on base substitution fidelity, using the TGA opal codon reversion assay. The experimental design is to construct a gapped molecule in which the gap contains a single base change in the lacZacoding sequence (G + A in the viral, template strand at position 89). This change creates an opal termination codon, resulting in a colorless plaque phenotype. Base substitution errors at the opal codon can be detected as blue plaque revertants upon transfection of the DNA polymerase reaction products, and the proportion of blue to total plaques (the reversion frequency) reflects the error frequency of DNA synthesis. Table I1 lists the consequences of single base substitution errors at this codon, including the mispairs, codons, amino acids, and relative blue color intensities. Heteroduplex expression experiments similar to those described previously (23) demonstrate the efficiency of minus strand expression for each mutational pathway and indicate no strong bias against any particular error. Eight errors can be observed; the only exception is the transition mispair (T. G) at the middle position, which generates an ochre codon which is colorless. The assay is quite sensitive since the background reversion frequency of uncopied DNA is low (Table 111).
Two experiments were performed to measure the fidelity of pol-y in this base substitution assay. In the first experiment each DNA polymerase reaction was performed using an equal concentration of all four dNTP substrates. As expected from previously studies (26, 46), chick embryo pol-@ is inaccurate, demonstrating the capability of the assay to detect in vitro polymerization errors. In contrast, Kf, used here as a positive control to detect proofreading, is substantially more accurate using a low concentration of dNTP substrates. The accuracy of Kf diminishes about 20-fold in reactions containing 1000 pM dNTPs and 20 mM dGMP, conditions that reduce the contribution of proofreading to fidelity. The magnitude of this

E. coli cells (made competent by the CaClz procedure) with complete
heteroduplexes. The values shown are the average of either four or five determinations, in each case scoring 500 or more total plaques. The new procedure for preparing double-stranded molecules (either gapped or complete) does not remove residual single-stranded viral DNA remaining from hybridizations performed at a 1:l primer to template ratio. In order to examine the influence of this residual single-stranded DNA on mutation frequencies and minus strand expression values, we systematically varied the primer to template ratio during the hybridization procedure from 1O:l (excess primer) to 1:lO (excess template) and measured minus strand expression values. Measurements were performed for 5 different ratios, both for the T. C and the A. C mispairs, and by transfection into cells made competent by both the CaClZ and Hanahan procedures. In only the extreme case of 10-fold template excess did the minus strand expression value diminish significantly. This is consistent with our reproducible observation of a much higher transfection efficiency for double-stranded M13 DNA compared to single-stranded DNA. In competent (CaCl,) cells, single-stranded viral M13mp2 DNA is reproducibly 30-fold less infective than is double-stranded DNA, and in Hanahan cells the difference is even greater (-100-fold). Because this double-stranded DNA: single-stranded DNA bias is greater than Hanahan cells, the results for minus-strand expression for the heteroduplexes shown in the table (shown for transfection of CaC1, cells) are all slightly higher when transfecting Hanahan cells (ranging from 67 to 86%), but show no bias for any mispair.  Base substitutwn fidelity of pol-y Reactions were performed as described under "Experimental Procedures," using 300 ng of gapped M13mp2 A89 (opal codon-containing) DNA per 50 pl of reaction. For experiment 1, the dNTP concentrations were equimolar and as indicated. The amounts of enzyme used per 50 pl of reaction and the incubation times were: chick pol-j3,0.8 unit, 37 "C, 60 min; Kf, 0.5 unit, 37 'c, 20 min (1 p M ) or 5 min (1000 p~ plus 20 mM dGMP); and chick pol-y, 3 units, 37 "C for 30 min (both conditions). In experiment 2, a constant 50-fold excess of dCTP over dATP, dTTP, and dGTP was employed and the concentration listed is that of dCTP. Only 4+ (wild type) blue mutants were scored, and all 4+ mutants were plaque purified and plated with wild type M13mp2 to confirm the phenotype. Sequence analysis of 10 of these revertants confirmed that all 10 had the predicted wild type TGG codon resulting from the misincorporation of C opposite A at position 89. Kf reactions used 0.5 unit of enzyme at 37 'C for 20 min. Pol-y reactions were as for experiment 1. The average fidelity of poly per base synthesized at the opal codon can be calculated (using as an example the reversion frequency for synthesis at 20 p M dNTPs from experiment 1, i.e. 8.3 X by subtracting the background reversion frequency (1.3 X lo-'), dividing by 0.6 (the approximate probability of expressing an error, see Table 11) and then dividing by 3, since errors can be monitored at all three bases of the opal codon. At this codon with equimolar substrate concentrations, the pol-y error frequency is one error for each 260,000 bases polymerized. proofreading effect is in fact similar to previous results with pol I in the 4X174arn3 assay (39). Chick pol-y is also highly accurate; at a dNTP concentration of 20 WM the reversion frequency of DNA copied by pol-y is only slightly above the background. Using reaction conditions (2000 p~ dNTPs, 20 mM dGMP) that were demonstrated to partially reduce terminal mismatch excision by the 3' + 5' exonuclease in the pol-y preparation (Fig. 3A), the reversion frequency increases more than 4-fold, a result that is consistent with diminished proofreading. However, even under this extreme reaction condition, pol-y remains more accurate than Kf. The second experiment focuses exclusively on misincorporation of C opposite A in the TGA codon by utilizing a pool imbalance in which dCTP is in 50-fold excess over the other three dNTP substrates. This is intended to specifically increase the error frequency for A.C mispairs by two mechanisms. First, incorrect dCTP is present in 50-fold excess over correct dTTP, increasing the misinsertion frequency. Second, dCTP is the next correct nucleotide to be inserted after the error, opposite the template G of the TGA codon. Its presence in high concentration is expected to increase the rate of polymerization from the mismatch, thus decreasing the likelihood of exonucleolytic removal. Stable misincorporation of C opposite A also produces a TGG, 4+ blue codon that is easily scored and differentiated from the lighter colors generated by the other revertants (Table 11). The reversion frequencies shown for experiment 2 therefore reflect only the 4+ phenotype. Kf is again used as a positive control for detection of proofreading activity. At a dCTP concentration of 100 pM (and 2 p~ for the other three dNTPs), Kf reactions produce a 4+ reversion frequency of 16 x Either a &fold increase in next nucleotide (while keeping the 50 X pool imbalance constant) or addition of dGMP to 20 mM substantially increases the reversion frequency, consistent with a reduced contribution of proofreading to fidelity. Again pol-y is highly accurate, since at 500 p~ dCTP the reversion frequency is only 2.1 X 10P. Even a 10-fold increase in dNTPs produced little effect on the reversion frequency. The absence of mutants under these conditions is not due to an inability to detect errors, since Kf generated mutants at high frequency with the same reagents under comparable conditions. Despite the high accuracy of pol-y, both a next nucleotide and a monophosphate effect are readily apparent. At 20 mM dGMP, a 10-fold increase in substrate concentration (from 500 to 5000 p~) produced a 5.7-fold increase in reversion frequency. At a constant 5000 p~ dCTP concentration, addition of 20 mM dGMP increased the reversion frequency 14-fold.

DISCUSSION
We initiated this study to determine if the high fidelity of chick embryo DNA polymerase-y (22-24) could result in part from 3' + 5' exonuclease proofreading. The results of this study show that three independently purified chick pol-y preparations indeed contain a 3' + 5' exonuclease activity which releases nucleoside 5'-monophosphates. Each of the observations presented here is consistent with a proofreading function for this exonuclease since 1) hydrolysis proceeds in the 3' + 5' direction, 2) the products are nucleoside 5'monophosphates, 3) the exonuclease exhibits a preference for mismatched rather than matched bases, 4) pol-y synthesis is highly accurate and, most importantly, 5) both exonuclease activity and the fidelity of pol-y-catalyzed DNA synthesis decrease in reactions containing high dNTP concentrations or dGMP. These conditions are known to inhibit proofreading by prokaryotic proofreading exonucleases, and their similar effect on pol-y fidelity suggests a coordinated action of the 3' + 5' exonuclease with chick embryo DNA polymerase-y to permit proofreading during DNA synthesis. Assuming that this interpretation is correct, chick embryo pol-y represents the second animal cell DNA polymerase preparation, along with calf thymus pol4 I1 (5), to be shown to proofread errors by direct measurements of fidelity during DNA synthesis on biologically active templates.
The data in Fig. 1 suggest that the exonuclease to polymerase ratio is quite high, being intermediate between that of E. coli pol I and T4 DNA polymerase. The terminal mismatch excision studies with M13mp2 DNA also suggest this, since more excision prior to polymerization is observed with the exonuclease in the pol-y preparation than for Kf at all dNTP or dGMP concentrations examined. Similarly, under all comparable conditions, pol-y is more accurate for base substitution errors during DNA synthesis at the opal codon than is Kf. The difference can be as much as 100-fold, depending on the conditions that are compared (e.g. Table 111, experiment 2, at 500 p~ dCTP: (after subtracting background) compare 229 X lop6 for Kf uersus 2.1 X for pol-y). The data in Table I11 can also be used to calculate the average base substitution fidelity of pol-y at the three template positions monitored by this reversion assay. At the lowest dNTP concentration examined (experiment 1, 20 p~ dNTPs), a condition which permits the exonuclease to be highly active, the pol-y error frequency at the opal codon is one error for each 260,000 bases polymerized (see legend to Table I11 for calculation). This represents the average frequency for all eight possible mispairs that can be detected at this codon (Table 11). Considering only the A . C mispair at position 89 (experiment 2, Table 111), pol-y is even more accurate, with an error frequency that is at least 10-fold lower than the overall average. Thus, the chick embryo pol-y used here exhibits accuracy which is similar to or even slightly higher than that of calf thymus pol4 I1 (5). The next nucleotide and monophosphate effects in the opal codon reversion assay further suggest that proofreading contributes at least 5to 10-fold to the overall accuracy of pol-y. This is likely an underestimate since, in the terminal mismatch excision experiments (Fig. 4A), the exonuclease activity remains highly active for excision of A. C and T. C mispairs even under extreme reaction conditions.
It is interesting to compare these estimates for the basesubstitution fidelity of pol-y, which are based on a reversion assay that detects eight mispairs at three template positions, to our previous results with pol-y using a forward mutational assay, which detects over 200 mispairs at over 100 template positions. In the forward assay, using 500 p~ dNTPs, the overall average base substitution fidelity of pol-y was estimated to be only %OO, a value considerably less accurate than the values obtained here. The primary reason for these seemingly discordant observations is that pol-y exhibits substantial differences in accuracy depending on the error being considered. In the previous studies, high fidelity was indeed observed for 9 of 12 possible mispairs (23) and for frameshift errors (24); it was this observation that led to the present study. However, pol-y produced three specific mispairs, (C . A, G. A, and G . G), at a high frequency. These errors obviously escaped proofreading even under conditions (500 p~ dNTPs) that do not fully eliminate exonuclease activity with other mispairs (our present results). Furthermore, the error frequency varied over more than 200-fold depending on the mispair (23) and over more than 16-fold for the same mispair, depending on its position (22). These observations, which are consistent with prokaryotic proofreading results (39, 40,44,45,47,48), demonstrate that the error frequency as well as the extent of proofreading depends on the composition and position of the mispair. Variations in reaction conditions as well as the mutational target DNA sequence should permit a further analysis of DNA context effects and of the parameters that determine pol-y insertion and proofreading discrimination, not only for base substitution errors but for frameshift and deletion errors as well.
The presence of an exonuclease activity in preparations of highly purified chick embryo pol-y leads to three questions. First, is the exonucleolytic activity physically associated with the DNA polymerization activity? Three observations suggest that this is likely. Pol-y has been purified to very high specific activity through six purification steps based on different separation principles (solubility, size, charge, and affinity), yet the exonuclease is still present with a ratio of exonuclease to polymerase comparable to that observed for prokaryotic exonuclease-containing DNA polymerases. Also the exonuclease and polymerase activities cosediment in glycerol gradients containing high salt which should minimize nonspecific interactions, and the polymerase/exonuclease ratio remains relatively constant across the peak. Finally, the proofreading results imply a functional association of these activities, which act in a coordinated manner during DNA synthesis at the template-primer site to achieve high fidelity.
Are the two activities in the same polypeptide or do they reside on separate subunits? In prokaryotic systems, precedent exists for both possibilities, since E. coli pol I and Tphage DNA polymerases contain both activities within a single polypeptide (see Ref. 43 for review) while the two activities reside on different subunits of E. coli pol 111 (49). Denaturing polyacrylamide gel electrophoresis of the pol-y preparations used here demonstrates the presence of a major polypeptide species of 47,000 daltons (86% of the total protein) and a second polypeptide of 135,000 daltons (12% ofthe total protein), both of which correlate with DNA polymerase activity (20). Based on native molecular weight, stoichiometry, and specific activity, Yamaguchi et al. (20) suggest that the native chick DNA polymerase-y may be a tetramer of four identical 47,000-dalton subunits. However, they do not exclude the possibility that the 135,000-dalton polypeptide has DNA polymerase activity. The exonuclease activity could reside in either of these two polypeptides or even in a minor species of polypeptide (although specific activity considerations make this last possibility seem less likely). Clearly, further attempts to purify and/or dissociate the exonuclease and polymerase activities are required.
Thirdly, are the observations presented here peculiar to preparations of chick embryo pol-y or do comparably purified preparations of pol-y isolated from other systems contain 3' -5' exonucleolytic activity? First purified from rat liver tissue as the mitochondrial DNA polymerase in 1968 (11,12), DNA polymerase-y has since been purified and characterized from a variety of sources (for review see Ref. 2). In most cases these preparations were not examined for nuclease activity. Three early reports, two for rat liver (11,12), and one for mouse myeloma cells (14), describe partially purified pol-y preparations devoid of detectable nucleolytic activity. These negative results should be interpreted with caution due to the sensitivity limitations imposed by the substrates and reaction conditions employed and the methods of detection available. Partially purified DNA polymerase-y from human placenta has been reported to contain detectable 3' 5' exonuclease activity that could be removed by glycerol gradient sedimentation (17). This exonuclease neither removed terminalmatched bases from synthetic polynucleotides under polymerization conditions, nor affected the fidelity of DNA synthesis with such templates. This latter observation on fidelity may not be particularly instructive, however, since with E. coli pol I, proofreading contributes little to the fidelity of synthesis on synthetic substrates (50).
Only two pol-y fidelity studies with natural DNA have been reported. HeLa cell pol-y was shown to be inaccurate for base substitution errors (A. C, A. A, and/or A. G) at position 587 in 0X174am3 DNA (46), implying that either exonuclease proofreading was not present, or, if present, was not particularly active at this site with these mispairs. In our initial fidelity studies of pol-y (22), a crude preparation of fetal calf liver pol-y was even more accurate in the M13mp2 forward mutagenesis assay than was the same preparation of chick embryo pol-y used here (22)(23)(24). This accuracy could reflect either high insertion fidelity or proofreading activity. As expected for an enzyme that has been difficult to purify due to low abundance, microheterogeneity and instability, no clear understanding emerges from the limited data available to date. Further work will be required to relate these chick embryo pol-y results to results obtained with pol-y isolated from additional sources.