Fidelity and error specificity of the alpha catalytic subunit of Escherichia coli DNA polymerase III.

Escherichia coli DNA polymerase III holoenzyme is the replicative enzyme primarily responsible for the duplication of the E. coli chromosome. This process occurs with high accuracy, less than 10−9 to 10−10 errors being committed per base pair per round of replication. As a first step in understanding the mechanisms responsible for the high fidelity of this process, we have purified the polymerase III α catalytic subunit, free of exonuclease activity, and analyzed its fidelity in vitro. We employed a newly developed gap-filling assay using the N-terminal 250 bases of the lacI gene as a forward mutational target. When synthesizing across this target, α subunit produced mutations at a frequency of 0.6%. DNA sequencing revealed that the mutants created in vitro consisted mostly of frameshift mutations, although some base substitutions were also observed. The frameshifts, occurring at more than 120-fold above the background, consisted largely of −1 deletions. Among them, about 80% were the deletion of a purine template base with a pyrimidine 5′-neighbor. These results suggest that the α subunit (i) has a relatively low ability to extend from misincorporated bases, accounting for the low level of observed base substitutions, and (ii) has a relatively high capability of extension after misalignment of a misincorporated base on the next (complementary) template base, accounting for the high level of frameshift mutations. This model is supported by an experiment in which α subunit was required to initiate DNA synthesis from a terminal mispair in a sequence context that allowed slippage on the next template base. Among the products of this reaction, frameshifts outnumbered base pair substitutions by greater than 70-fold. A comparison to in vivo mutational spectra suggests that the pol III accessory factors may play a major role in modulating the fidelity of DNA synthesis.

DNA replication in prokaryotic and eukaryotic organisms is a highly accurate process, only one error being committed per 10 9 to 10 10 nucleotides incorporated (1). This high fidelity of DNA replication is achieved by at least three critical steps (reviewed in Refs. 2 and 3): (i) the insertion fidelity of the DNA polymerase, which selects correct over incorrect nucleotides via template-directed discrimination, (ii) the 3Ј 3 5Ј (proofreading) exonuclease of the DNA polymerase, which removes incorrectly inserted bases before DNA synthesis continues; and (iii) postreplicative DNA mismatch repair, which corrects errors re-sulting from DNA replication errors by incising the newly synthesized strand.
DNA polymerase III holoenzyme is the main enzyme responsible for replication of the Escherichia coli chromosome, including the high fidelity of this process (4 -6). DNA polymerase III holoenzyme is a dimeric polymerase that simultaneously replicates the leading and lagging strands (4 -7). It consists of 10 different subunits, ␣, ⑀, , , ␥, ␦, ␦Ј, , , and ␤ with a presumed overall composition (␣⑀) 2 2 (␥ 2 ␦␦Ј)␤ 4 (6). The tightly bound ␣, ⑀, and subunits form the pol III core. The ␣ subunit, encoded by the dnaE gene, is the catalytic DNA polymerization subunit (8). The ⑀ subunit, encoded by the dnaQ gene, provides the 3Ј 3 5Ј proofreading activity (9). The function of the subunit is as yet unknown. The fine structure and dynamics of the pol III holoenzyme are under active investigation (6). Postreplication DNA mismatch repair is performed by the mutH, mutL, mutS, and mutU gene products (see Ref. 10, for review).
The relative contributions of base selection, proofreading, and DNA mismatch repair to in vivo fidelity have been estimated from mutant frequencies and sequenced mutation spectra in strains defective in the respective pathways (3,11). For example, using mismatch repair-defective strains, it was estimated that pol III holoenzyme replicates DNA at an average accuracy of 10 Ϫ7 errors per base pair per round of replication, of which proofreading may contribute an approximate factor of 10 Ϫ2 (3,12). Postreplicative mismatch repair improves overall fidelity by 200-to 300-fold (3,11), yielding the final mutation rate of 10 Ϫ9 to 10 Ϫ10 .
To gain insights into the mechanisms by which pol III holoenzyme achieves its high fidelity, we have initiated an in vitro investigation of its fidelity. Despite the fact that this enzyme is one of the best characterized replication complexes, its fidelity has not been studied in detail (13)(14)(15)). In the current study, we focus on the simplest unit, the ␣ catalytic subunit. One previous study on the fidelity of ␣ subunit has been reported, in which steady-state kinetic assays at a limited number of nucleotide sites were performed to determine nucleotide misincorporation rates by this enzyme (14). Here, we analyze the fidelity of ␣ subunit using a newly developed in vitro forward mutation assay that uses the N-terminal region of the lacI gene as a sequence target. This target allows a variety of mutations to be recovered, including base substitutions, frameshifts, deletions, and duplications. We show that in this assay ␣ subunit produces relatively few base pair substitutions, but a high level of single-base frameshift mutations. The implications of these findings for the overall fidelity mechanism used by pol III holoenzyme are discussed.
Overexpression and Purification of ␣ Subunit-A fresh overnight culture of E. coli JM109 harboring pDNAE OPI (dnaE ϩ ) was diluted 1:100 in LB broth and shaken at 37°C until A 600 reached 0.60. Isopropyl-1-thio-␤-D-galactoside was added to a final concentration of 1 mM and shaking was continued for another 4 h. A total of 8 liters of the induced cells were spun down and resuspended in 50 ml of buffer containing 50 mM Tris-HCl (pH 7.5), 10% sucrose, and 8 mM dithiothreitol (DTT). Lysozyme was added to a final concentration of 1.8 mg/ml. Lysis was achieved by incubating the suspension at 37°C for 7 min and then on ice for 15 min. Cell debris was removed by centrifugation. Ammonium sulfate was added to the supernatant to a final concentration of 0.25 g/ml followed by stirring for 30 min. The precipitated protein was recovered by centrifugation, and the pellet was resuspended in buffer A containing 50 mM Tris-HCl (pH 7.5), 20 mM NaCl, 1 mM EDTA, 5 mM DTT, and 20% glycerol. The resuspension was dialyzed against 1 liter of buffer A for 16 h with two changes. It was then loaded onto a HiTrap-Heparin column (bed volume, 5 ml) equilibrated with buffer A. The column was washed with 25 ml of buffer A. Protein was eluted with 50 ml of buffer A containing a linear gradient (0 -0.4 M) of NaCl at a flow rate of 0.1 ml/min. Peak fractions from two separate batches of HiTrap-Heparin chromatography were pooled and applied to a FPLC Mono Q 5/5 HR column (bed volume, 1 ml). The column was washed with 5 bed volumes of buffer A. Protein was eluted with 20 ml of buffer A containing a linear gradient (20 -400 mM) NaCl at a flow rate of 0.1 ml/min. The peak fractions were pooled and concentrated to 150 l by a Centricon 30 (30,000 M r cut-off) centrifugal microconcentrator and then gel-filtered through an FPLC Superose 12 column (HR 10/30) equilibrated with buffer A containing 500 mM NaCl at a flow rate of 0.15 ml/min. All operations were done at 0 -4°C. Polymerase activity assays (see Other Methods) and SDS-polyacrylamide gel electrophoresis were used to monitor ␣ subunit during the purification.
3Ј-Exonuclease Activity Assay-A 5Ј-32 P-labeled 15-mer oligonucleotide was annealed to M13mp2 single-stranded DNA to generate a mismatched substrate with an A(template)⅐G 3Ј-terminal mispair. The oligonucleotide was complementary to nucleotides 106 -119 of the lacZ␣ gene with the A⅐G mispair located at position 105. Annealing was done in 150 mM sodium chloride, 15 mM sodium citrate, by heating the mixture at 70°C for 10 min and then allowing it to cool down to room temperature. The DNA:oligonucleotide ratio was 1.5 to 1. Following hybridization, the product was purified through a Sephadex G-50 column. Exonuclease reactions (25 l final volume) were started by adding 10 pmol of ␣ subunit or 12 pmol of pol I Klenow fragment into a mixture containing 50 mM Tris-HCl (pH 7.8), 2 mM DTT, 10 mM MgCl 2 , and 300 ng of terminally-mismatched DNA. After incubation at 30°C for 10 and 60 min, 5-l aliquots were removed and mixed with stop solution (20 mM EDTA, 95% formamide, 0.05% bromphenol blue, 0.05% xylene cyanol FF). The sample was then run on a 20% polyacrylamide-6 M urea gel and analyzed by PhosphorImager (Molecular Dynamics, Sunnyvale, CA). Exonuclease activity was calculated as the rate of disappearance of the starting 15-mer.
Construction of mRS65 Gapped DNA-Gapped molecules were prepared essentially according to the method described by Kunkel and Soni (20). RF I DNA of phage mRS65 was isolated by Qiagen column. The double-stranded DNA was digested with restriction endonucleases MluI and SphI, which cut at positions Ϫ60 and ϩ379 of lacI, respectively, producing two fragments of size 439 and 8040 base pairs (bp). The 8040-bp fragment was separated from the small fragment by PEG precipitation using 6.5% PEG-8000 in the presence of 0.5 M NaCl and resuspended in TE buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA). Using these conditions, the 8040-bp preparation does not contain detectable 439-bp fragment. 0.38 pmol of 8040-bp fragment in TE buffer were heated at 70°C for 5 min; 0.25 pmol of single-stranded circular mRS65 DNA were added, and heating was continued for another 2 min. NaCl and sodium citrate (20 ϫ SSC) were added to a final concentration of 30 mM, and heating continued at 65°C for 5 min. The product was desalted using a Microcon-30 centrifugation device and analyzed via agarose gel electrophoresis. The gapped DNA obtained through this procedure did not contain detectable single-stranded DNA.
Gap-filling DNA Synthesis and Mutant Frequency/Error Rate Determination-␣ subunit from Superose fractions 35 and 36 (see Fig. 2) was used for gap-filling assays. Reactions were initiated by addition of ␣ subunit to a mixture (total volume, 25 l) containing 20 mM Tris-HCl (pH 7.5), 8 mM DTT, 10% glycerol, 10 mM MgCl 2 , 1 mM each of the four dNTPs and 60 fmol of gapped DNA. For the gap-filling reaction with mp2 DNAs, 28 pmol of ␣ subunit were added. For gap-filling reaction with mRS65 DNA, 5.4 pmol of ␣ subunit was used. The reaction was run at 30°C for 1 h and terminated by adding EDTA to a final concentration of 15 mM. A portion of the reaction mixture was run on an 0.8% agarose gel to confirm completeness of gap-filling. Another portion of the reaction mixture was diluted in sterile water and transfected into competent cells of strain MC1061 by electroporation. Competent cells were prepared as described (21). An appropriate amount of transfected cells was mixed with 3 ml of soft agar containing 2.5 mg of X-gal and 0.5 ml of indicator strain NR9099 and plated on minimal medium plates (20). The plates were incubated at 37°C for 24 h. Mutants were scored as colorless (or light-blue) plaques in case of mp2 DNA and as (light and dark) blue plaques in case of mp2A89, mp2(ϩT70), and mRS65. The mutant frequency was calculated by dividing the number of mutant plaques by the total number of plaques. The error rate per nucleotide polymerized in the lacI target was calculated by dividing the mutant frequency (after subtraction of the control frequency) by 0.55 (the average Ϫ-strand expression factor, see legend to Fig. 3) and then by the approximate number of nucleotides in the target (220 for the combined mutations or the frameshift mutations and 100 for base pair substitutions).
Terminal-mispair Utilization Experiments-A gapped mp2 DNA molecule containing a mismatched 3Ј primer terminus was prepared as described by Kunkel and Soni (20), except that the primer fragment was obtained by agarose gel purification, instead of PEG precipitation. The gapped primer template (gap size 366 nucleotides) contained a 3Јterminal T(template)⅐C mismatch at position 103 of the lacZ␣ gene (20,22). DNA synthesis by ␣ subunit (10.8 pmol) or by E. coli DNA polymerase I Klenow fragment (exo Ϫ ) (0.5 unit) was performed as described for gap-filling assays, above. The product DNA was electroporated into strain MC1061, followed by plating on strain CSH50 in the presence of X-gal and isopropyl-1-thio-␤-D-galactopyranoside and scoring for plaque-color phenotype (dark-blue, light-blue, or colorless) (22).
Other Methods-DNA polymerase assays used to monitor ␣ subunit during purification were done using poly(dA)/(oligo(dT) as templateprimer (150 M as nucleotide, dA:dT ϭ 2:1) as described (23). One unit of polymerase activity is defined as the amount that catalyzes the incorporation of 1 pmol of dTMP per min. SDS-polyacrylamide gel electrophoresis was done as described (24). Protein concentrations were determined using the Bio-Rad protein assay dye reagent with bovine serum albumin as a standard. Isolation of single-and double-stranded DNA was done by a published method (25). DNA sequencing of mRS65 mutants containing a mutation in the lacI gene was performed as described (26), using an 18-mer primer complementary to positions 259 to 276 of lacI (27). Mutations were recorded when residing in the region between position ϩ30 and ϩ250 (3,27).

RESULTS
Purification and Properties of ␣ Subunit-DNA polymerase III ␣ subunit was purified from strain JM109 harboring the dnaE-overexpressing plasmid pDNAE OPI. From 8 liters of culture we obtained 1.5 mg of ␣ subunit. The protein was judged to be pure (Ͼ95%) by SDS-polyacrylamide gel electrophoresis ( Fig. 1) and silver staining (not shown). The specific activity was 4.5 ϫ 10 6 units/mg of protein, a value comparable to those published by others (8).
Since our primary goal was to perform in vitro fidelity assays with the purified ␣ subunit, it was important to establish the level of 3Ј-exonuclease activity in the preparation. Contaminat-ing exonuclease might function as a proofreader and reduce the number of observable polymerase errors. Exonuclease activity was examined using a 3Ј-terminally mismatched DNA substrate. The reaction products were analyzed by electrophoresis in a 20% denaturing polyacrylamide gel (Fig. 2). After a 10-min incubation, no detectable exonuclease activity was observed in any of the peak fractions of the final gel filtration column. In contrast, the Klenow fragment of E. coli DNA polymerase I used as a positive control degraded almost half of the substrate in this time period. After a 1-h incubation, there was a trace of exonuclease activity in fractions 30 and 31. The activity in these two fractions was about 1/5000 of the activity in Klenow polymerase. Exonuclease activity in fractions 32-36 was still not detectable after a 1-h incubation. ␣ Subunit from fractions 35 and 36 was used for gap-filling assays (see below).
Mutant Frequencies in the M13mp2 Assay System-To determine the fidelity of DNA synthesis by the ␣ subunits, we first used the M13mp2 mutation assay system described by Kunkel (16) and Kunkel and Soni (20). This system uses a doublestranded M13mp2 DNA molecule containing a 361-nucleotide gap across the lacZ␣ gene required for ␣-complementation. The accuracy of gap-filling DNA synthesis is monitored by transfecting the completed DNA and plating on an ␣-complementation host. Two versions of the M13mp2 system exist. In the forward assay, the gap contains the lacZ␣ ϩ gene, and polymerase errors throughout the target are scored as lacZ␣ Ϫ mutants (colorless or light-blue plaques among the parental darkblue plaques). In the reversion assay, the gap contains a defined lacZ mutation, and DNA polymerase errors at (or near) the site of mutation are scored as lacZ␣ ϩ plaques (blue plaques among parental white plaques). The ␣ subunit was capable of filling the 361-nucleotide gap across the lacZ␣ gene, although a large molar excess of enzyme over template (Ͼ400:1) was required. Using the forward assay, the mutant frequency for ␣ subunit was 30 ϫ 10 Ϫ4 , 7-fold above the background level (Table I). We also examined the fidelity using two different reversion systems. Phage mp2A89 contains a G to A substitution in position 89 of lacZ␣, resulting in a TGA termination codon, and reversion can take place by any base-substitution error changing the termination codon into a sense codon (20). Using this assay, the reversion frequency was essentially at background level (Table I). We also measured the frameshift fidelity using mp2(ϩT70), which has an extra T added to a TTTT run at positions 70 -73 of lacZ␣, abolishing ␣-complementation (28). Reversion takes place by (Ϫ1) frameshift mu-tation at the 5 Ts or in a limited region around it (28). Table I shows that reversion frequency was about 30-fold above the background. The results suggest that the ␣ subunit is quite accurate for base substitutions, but relatively inaccurate for frameshift mutations. This is an unusual specificity among DNA polymerases tested (16, 22, 28 -35).
A drawback of using the mp2 system to study the fidelity of ␣ subunit is that a very large excess of enzyme over DNA was required (Ͼ400:1) to completely fill the gap. This amount is some 10-fold higher than generally required among enzymes tested (e.g. Refs. 32, 36, and 37). Further analysis (not shown) revealed that the high enzyme requirement was likely due to secondary structures in the single-stranded region of the gapped mp2 molecules, such as the palindrome of the lac operator, which is located about two-thirds down the gap (38,39). For this reason, additional, more detailed studies were performed using a new gap-filling assay based on the E. coli lacI gene, as described below.
lacI in Vitro Forward Mutation Assay-Phage mRS65 is a derivative of phage M13 containing, in addition to the lacZ␣ gene, the entire adjacent lacI gene (M13lacI ϩ Z␣ ϩ ) (17,39). Since the lacI gene encodes the lac repressor, ␣-complementation by this vector depends on the state of the lacI gene. When plated on ␣-complementation host strain NR9099 (which is lacI Ϫ ) in the presence of X-gal, mRS65 will produce colorless plaques, but any lacI Ϫ derivative will produce a blue plaque. Thus, this phage provides a forward mutational target where blue mutants can be readily observed against a background of colorless plaques. The outline for the lacI in vitro assay system is shown in Fig. 3. We created an mRS65 gapped molecule containing a 439-nt gap that includes the N-terminal portion (nucleotides Ϫ60 to ϩ379) of the lacI gene. This region of lacI was chosen because it contains the so-called lacI d region (ϩ30  a The typical background frequency for the mp2A89 reversion assay is ϳ2 ϫ 10 Ϫ6 (28,32). In the present case, we obtained 8 blue plaques among 614,000 total. For the copying reaction by ␣ subunit, we obtained 1 blue plaque among 798,000 total. through ϩ240), which has a high density of detectably mutable sites and has been the target of many in vivo mutagenesis studies (3, 11, 40 -43). There are a number of advantages to this lacI in vitro assay for measurement of DNA synthesis errors. First, smaller amounts of enzyme are sufficient for complete gap-filling (see "Experimental Procedures"), presumably due to less interference by secondary structures. Second, the lacI N-terminal region has been used extensively as a target for studies of mutagenesis in vivo, more than 4,000 lacI mutants having been sequenced to date (3, 11, 40 -43). Results obtained in vitro with E. coli DNA replication proteins can therefore be evaluated against this in vivo data base. Third, since mutants are selected as blue plaques against a background of colorless plaques, the scoring of forward mutants is facilitated.

Mutant Frequencies and Mutant Specificity in the lacI in Vitro Assay
System-The products of lacI gap-filling reactions by ␣ subunit were analyzed on agarose gels (Fig. 4). The band representing the gapped DNA clearly shifted to the position of double-stranded RFII DNA. The reaction products were transfected into competent cells. The mutant frequencies are shown in Table II. The frequency for uncopied DNA was 14 ϫ 10 Ϫ4 , while for copied DNA the frequency was 113 ϫ 10 Ϫ4 , indicating that ␣ subunit made errors at a level readily detectable above the background.
To learn the precise nature of the replication errors, we sequenced 100 independent lacI mutants produced by ␣ subunit. Our sequencing effort was restricted to the lacI d region of the gap (nucleotides ϩ30 to ϩ250). A mutation in this region was found for close to 60% of the mutants. The resulting corrected mutant frequency showed a 12-fold increase above the background (Table II). Both base substitutions and frameshifts were recovered. The mutational spectrum is presented in Fig.  5, and the results are tabulated in Table III. Base substitutions occurred at a frequency only about 2-fold above background (Table III). Therefore, ␣ subunit did not make many base substitution errors, as was also surmised from the mp2A89 reversion assay (Table I). Frameshifts outnumbered base substitutions 7.8:1 (after subtraction of the background, see Table  III). This is in contrast to the control, where base pair substitutions exceeded frameshifts by more than 10-fold. Thus, frameshifts were increased some 130-fold above the background. Based on these data, the overall error rate (per nucleotide) can be calculated to be 1/20,000. The specific rate for base substitutions is 1/91,000; that for frameshift errors 1/25,000. Thus, on a per nucleotide basis, frameshifts are about 3.5-fold more frequent than base pair substitutions. However, it is likely that the calculated base substitution rate is an overestimate (see "Discussion") and the difference between the two classes of mutations may be larger.
Almost all frameshifts were 1-base deletions, although a few (Ϫ2) frameshifts were also observed. Among the Ϫ1 deletions, deletion of G was the most frequent event, followed by deletion of A. Overall, deletion of a purine base was 4-fold more frequent than of a pyrimidine. This is not due to a bias within the target sequence, since it contains an almost equal number of purines and pyrimidines (116 versus 103). In Tables IV and V, we have further analyzed the DNA sequence dependence of the (Ϫ1) frameshifts. In general, there is no good correlation between the deletion frequency and the length of runs of identical bases (Table IV). For example, the deletion of G occurs at similar frequency whether the G is solitary or part of a GG or GGG sequence. The nearest neighbor analysis of Table V shows that for deletion of G, all 5Ј neighbors (20/20) were pyrimidines. Among the 3Ј neighbors, C was preferred.
Terminal Mismatch Utilization by ␣ Subunit-We also investigated the ability of ␣ subunit to continue synthesis from a mismatched 3Ј terminus (the immediate product of a misincorporation error). We used a gapped mp2 substrate containing a terminal T(template)⅐C mismatch at position 103 of the lacZ␣ gene (20,22), which has been used to investigate terminal mispair utilization by a number of different enzymes (20,22,28,32). One advantage of the T⅐C mismatch at position 103 is that its DNA sequence context permits distinction between two types of mismatch utilization (see Table VI): direct extension of the terminal C, generating a phage with the G 103 mutation (a light-blue plaque phenotype), or extension after misalignment of the terminal C on the next template G (providing a correctly paired terminus), thus creating a (Ϫ1) frameshift mutant (colorless plaque) (20,22). The data in Table VI show that uncopied DNA yielded no colorless plaques and only a few light-blue phages, demonstrating the poor in vivo expression of nonextended mispairs (the majority of plaques being dark-blue, representing the T-containing wild-type template strand). However, extension with either ␣ subunit or the exonucleasedeficient form of E. coli DNA polymerase I Klenow fragment yielded a high percentage of mutant phage. Interestingly, in the case of ␣ subunit, the plaques were in large majority colorless, whereas in the case of the Klenow fragment they were in the majority light-blue (as observed before (32)).  3. The mRS65 gap-filling assay system. The gap located in the 5Ј-region of the lacI gene is 439 nucleotides long between positions Ϫ60 and ϩ379 of lacI, numbered according to Farabaugh (27). The gap contains the lacI promoter region and the first 379 nucleotides of the lacI gene. Gap-filling synthesis proceeds from right to left. The dashed square represents E. coli competent cells. The expression of (Ϫ)-strand errors in this system as determined by heteroduplex experiments (20,38) is about 55% (data not shown), similar to what has been reported previously for the mp2 system (20). DNA sequencing was performed of mutants in the lacI d region (ϩ30 to ϩ250) (3).

DISCUSSION
In order to better understand the precise mechanisms that enable E. coli to replicate its chromosome with high fidelity, we have examined the properties of the ␣ (catalytic) subunit of E. coli DNA polymerase III holoenzyme. From an overproducing strain, ␣ subunit was purified essentially free of 3Ј exonuclease activity. The current experiments may therefore allow an assessment of the accuracy of this enzyme without interference by the proofreading activity provided by the normally tightly associated ⑀ subunit. Fidelity was assessed using in vitro gapfilling assays, which used either the lacZ␣ or the lacI gene as a mutational target. The lacI gap-filling assay was specifically developed to facilitate complete gap-filling synthesis by ␣, as the enzyme proved to have great difficulty synthesizing past presumed secondary structures in the lacZ template DNA. Such a synthesis problem may not be surprising for polymerases like ␣ that are accustomed to performing DNA synthesis in the presence of multiple accessory factors.
The salient result of our study is the seemingly low ability of ␣ subunit to produce base pair substitutions, while showing a high propensity for producing frameshift mutations. This discrepancy between base pair substitutions and frameshifts is seen in both the lacZ␣ reversion assay and the lacI forward assay. In the lacZ reversion assays, we observed no increase above the background (Յ10 Ϫ6 ) for the base substitution marker, but a significant increase (ϳ60-fold) for the frameshift marker. Among the sequenced lacI mutations, frameshifts outnumbered base substitutions by almost 8-fold (after subtraction of the background) (Table III).
It is furthermore possible that the small increase in lacI base substitutions (about 2-fold, see Table III) does not reflect increased base substitution replication errors by the ␣ subunit. When we attempted to enhance the frequency of the C 3 T transitions, the most frequent base pair substitution in the spectra (Table III) by a dNTP pool imbalance (i.e. a 20-fold lowering of the concentration of dGTP, the correct nucleotide opposite template C), we did not observe an increase in base substitutions, suggesting that the C 3 T do not result from normal C⅐A mispairings (data not shown). Instead, an increase in Ϫ1 frameshifts (loss of C) was observed. One possibility a In parentheses is given the absolute frequency of each class of mutations. This frequency was obtained by multiplying the fraction of each mutational class by the corrected mutant frequency of Table II.

TABLE V Nearest-neighbor analysis of minus-G frameshifts created
by ␣ subunit For the purpose of this analysis, when deletion of G occurred from a GG or GGG sequence, the 5Ј or 3Ј base adjacent to the run was used.

Sequence
Sites in target Number of mutations 5Ј-A-G 11 0 5Ј-T-G 20 10 5Ј-C-G 18 10 5Ј-G-A 12 3 5Ј-G-T 17 5 5Ј-G-C 20 12 might be that the base substitutions reflect a low level of errors forced by damaged bases introduced by the gap construction procedures. For example, deamination of cytosine would readily yield C3 T transitions independently of dNTP pool imbalances. Such a possibility is supported by our observation that transfection of a gapped DNA molecule generally yields a mutant frequency a few fold higher than the single-stranded DNA from which it derives, as was also observed by others. 2 Thus, the apparent frequency of base pair substitution mutations by ␣ subunit, as measured by gap-filling assay, may be essentially Յ10 Ϫ6 , as deduced from the experiments with the A89 marker in Table I. It is, however, unlikely that the base substitution error rate of ␣ is as low as the mutant frequency suggests. Sloane et al. (14) used a kinetic gel assay to determine single-site misincorporations by ␣ subunit on an oligonucleotide template. For the four mispairs tested (A⅐A, A⅐C, A⅐G, and T⅐G), they observed misinsertion frequencies ranging from 1.5 ϫ 10 Ϫ5 to 7.7 ϫ 10 Ϫ4 , depending on the mispair. This range is quite typical for most DNA polymerases examined by this method (32, 44 -47) and is consistent with the general substitution fidelity of nonproofreading enzymes measured by gap-filling assays (16, 22, 28 -35). Therefore, the low base pair substitution frequency of ␣ in the gap-filling reaction unlikely reflects its inability to produce such errors.
Instead, we suggest that the low base pair substitution frequency reflects the enzyme's limited ability to continue synthesis from the terminal mispairs that are created by the misincorporation event. In general, extension from mispairs is slow for most enzymes compared to correct pairs (46, 48 -50). The reduced rate of extension is considered one determinant of fidelity, as it allows efficient competition by proofreading (46, 48 -52) or, in case of non-proofreading enzymes, polymerase dissociation from the primer terminus. Nonextended terminal mispairs have a low probability of survival upon transfection into competent cells (20) causing the base substitution error to go undetected (see also Table VI). Kinetic experiments with bacteriophage T4 and T7 DNA polymerases (51,52) have shown that dissociation is capable of competing effectively with mispair extension, and fidelity measurements with T7 DNA polymerase showed that this enzyme, when deprived of its processivity factor thioredoxin, displayed an apparent antimutator effect for base pair substitutions, an effect attributed to its strongly reduced extension capability (34). Thus, it seems reasonable to assume that the isolated pol III ␣ subunit, devoid of its supporting subunits, may have great difficulty extending mispairs (53).
Ϫ1 frameshift mutations occurred at 60 to 120 times the background level in the lacI and lacZ␣ assays, indicating that the ␣ subunit readily makes such errors. Two general models have been proposed for the production of frameshifts during DNA replication: (Streisinger) slippage in runs (54,55) and misincorporation plus misalignment of the misincorporated base on the next template base (22,56). The latter can occur favorably if the misincorporated base is complementary to the next template base; the misalignment allows further synthesis to proceed from a correctly paired, although misaligned, 3Ј terminus. From which pathway frameshifts arise can often be discerned by an inspection of the DNA sequence context at which they occur. Streisinger slippage is associated with runs of identical bases, the frequency increasing with the length of the run (55,57). In contrast, frameshifts occurring via misincorporation occur readily at non-reiterated sequences in specific sequence contexts. Among the ␣ subunit-induced frameshifts, we observed no bias in favor of frameshifts in runs (Table IV), suggesting that most of the frameshifts do not result from direct slippage. Further, a striking feature of the ␣ subunit-induced frameshifts is that Ͼ90% (23/25) of the non-run deletions are the loss of a purine base that has a pyrimidine as its 5Ј-nearest neighbor (Fig. 5). This specificity has been observed for several other enzymes and is considered a hallmark of the misincorporation/realignment model (28,56,57). Purine⅐purine mispairs such as G⅐G or A⅐A are among the pairs extended the most poorly (46,50), thus providing the greatest opportunity for the misalignment to occur on the next (complementary) pyrimidine. These combined observations are consistent with the idea that the frameshifts generated by ␣ subunit proceed largely through the misincorporation/slippage model.
Direct evidence for the above model is provided by our experiments in which ␣ subunit is provided with a preformed terminal mispair (Table VI). While most other enzymes tested by this procedure (including polymerase I Klenow fragment used here) prefer (to varying degrees) direct extension to yield a base pair substitution (20,22,28,32), ␣ subunit almost exclusively favors misalignment on the next template base to yield a (Ϫ1) frameshift mutation.
One useful aspect of our current E. coli system is the possibility of comparing the in vitro error spectra with spectra of mutations observed in vivo in the same sequence target. For example, the in vivo mutation spectrum in a mismatch-repair defective mutL strain may be considered to reflect in vivo DNA replication errors. This spectrum is strongly dominated by (transition) base pair substitutions (3,40,58), although in one study a significant contribution (ϳ25%) of frameshift errors was also observed (58). The latter occurred at a run of five identical bases (135-139, see Fig. 5), an event not very frequent in the ␣ spectrum and likely representing a direct slippage mechanism. Thus, the error spectrum of the ␣ subunit in vitro is quite different from the in vivo mutation spectrum. One possible explanation for this difference is that exonucleolytic proofreading removes the misaligned intermediates or the mispairs that promote a misalignment. However, the spectrum of mutations in mutDmutL double-mutator strains, defective in both mismatch repair and proofreading, is still dominated by base pair substitutions (3).
Thus, interestingly, significant differences exist in the (fidelity) behavior of the ␣ subunit when acting by itself or when 2 T. A. Kunkel, personal communication.

TABLE VI
Terminal mismatch utilization by ␣ subunit Gap-filling synthesis was performed on a gapped construct containing a terminal T(template) ⅐ C mismatch at position 103 of lacZ␣ (see "Experimental Procedures"). Both ␣ subunit and pol I Klenow fragment were capable of completely filling the gap under standard gap-filling conditions (data not shown). After transfection, plaque phenotypes were scored. Light blue and colorless plaques result from extension of the indicated primer termini (aligned or misaligned). Total plaques include dark blue plaques, the phenotype of the T-containing (wild-type) template strand, which are produced from full-length heteroduplexes with 40 -60% efficiency (20). DNA sequencing of a subset of the light blue and colorless plaques revealed the correctness of the mutational assignment: 17 out of 17 light blue plaques produced by ␣ subunit contained the T 3 G base substitution at position 103, whereas 20 out of 20 colorless plaques (10 from each enzyme) contained the deletion of T from position 103-104 (22). acting as part of a greater assembly, such as the in vivo replication complex. The mechanisms underlying these differences are likely to be important for the precise mechanisms by which the E. coli replication machinery achieves its high fidelity. It is likely that analysis of the fidelity properties of higher order pol III assemblies, such as pol III core, pol IIIЈ, pol III*, or pol III holoenzyme in our current in vitro assay system will provide further insights into this question.