Mismatch Extension by Escherichia coli DNA Polymerase III Holoenzyme*

The in vitro fidelity ofEscherichia coli DNA polymerase III holoenzyme (HE) is characterized by an unusual propensity for generating (−1)-frameshift mutations. Here we have examined the capability of HE isolated from both a wild-type and a proofreading-impaired mutD5 strain to polymerize from M13mp2 DNA primer-templates containing a terminal T(template)·C mismatch. These substrates contained either an A or a G as the next (5′) template base. The assay allows distinction between: (i) direct extension of the terminal C (producing a base substitution), (ii) exonucleolytic removal of the C, or (iii), for the G-containing template, extension after misalignment of the C on the next template G (producing a (−1)-frameshift). On the A-containing substrate, both HEs did not extend the terminal C (<1%); instead, they exonucleolytically removed it (>99%). In contrast, on the G-containing substrate, the MutD5 HE yielded 61% (−1)-frameshifts and 6% base substitutions. The wild-type HE mostly excised the mispaired C from this substrate before extension (98%), but among the 2% mutants, (−1)-frameshifts exceeded base substitutions by 20 to 1. The preference of polymerase III HE for misalignment extension over direct mismatch extension provides a basis for explaining the in vitro (−1)-frameshift specificity of polymerase III HE.

The error rate of a DNA polymerase depends not only on the efficiency with which it discriminates against incorrect nucleotides during the insertion step but also on its efficiency in continuing synthesis from the mismatched primer-terminus that it created by misinsertion. This is most obvious for enzymes containing an associated proofreading activity. Proofreading and mismatch extension compete, and a slow extension rate is likely to lead to very few mismatches being able to escape removal by the exonuclease. Thus, the ability to extend mismatches is an important factor in determining polymerase fidelity.
The accuracy of the duplication of the genetic material in the bacterium Escherichia coli depends largely on the fidelity of the DNA polymerase III holoenzyme (HE), 1 which is responsible for its chromosome replication (1). The HE is a large dimeric complex composed of 10 distinct subunits that is capable of simultaneously synthesizing the leading and lagging strands of the replication forks (2)(3)(4). In vitro fidelity studies of purified HE using either single-stranded phage templates (5,6) or oligonucleotide substrates (7) have revealed that HE is quite accurate for base substitution errors. Interestingly, gap-filling synthesis of M13 DNA (6) using a forward mutational assay capable of detecting various types of synthesis errors revealed HE to be relatively inaccurate for (Ϫ1)-frameshift mutations. In fact, (Ϫ1)-frameshifts were the major class of errors generated by HE in this assay (6).
Studies of (Ϫ1)-frameshift mutations have lead to the proposal of several general mechanisms for their formation. One model proposed by Streisinger et al. (8) postulates the occurrence of slippage of the nascent DNA strand in homopolymeric sequences (the direct slippage model). Within this model, the frameshift mutation frequency increases with the length of the run, because larger runs produce more (and more stable) misaligned intermediates. This model has been suggested to contribute to spontaneous frameshift mutations in vivo (9 -12) as well as frameshifts produced in vitro by a variety of DNA polymerases (13).
A second mechanism, the "misincorporation model" proposed by Kunkel and Soni (14), states that frameshift mutations can be initiated by misinsertion of a nucleotide. If the misinserted nucleotide is complementary to the next template base, then its 1-base forward misalignment can form a frameshift intermediate containing a correctly base-paired terminus (and an unpaired extra base in the template strand). As in the direct slippage model, this frameshift intermediate would be fixed into a (Ϫ1)-frameshift mutation by further extension. The misincorporation mechanism has been shown to operate during in vitro DNA polymerization by the Klenow polymerase (15) and human immunodeficiency virus type I reverse transcriptase (16). A third possible model for (Ϫ1)-frameshifts generated by polymerase III HE in vitro results from recent work by Bloom et al. (7), who observed elevated misincorporation by polymerase III HE in cases where the misincorporated base was complementary to the next template base. The authors proposed a transient misalignment before incorporation, in which the incoming (incorrect) dNTP is aligned on the next template base (dNTP-mediated misalignment). Continued DNA synthesis from the misaligned intermediate (after incorporation of the misaligned dNTP) would also generate a (Ϫ1)-frameshift.
Based on a detailed analysis of in vitro mutation spectra produced by E. coli DNA polymerase III holoenzyme (both wild-type and MutD5), we had suggested that most of the observed (Ϫ1)-frameshift mutations produced by this enzyme occur via the misincorporation plus slippage mechanism (6). This conclusion was based on the following observations: (i) most of the (Ϫ1)-frameshifts occurred at nonreiterated tem-plate positions, (ii) the (Ϫ1)-frameshift mutations were proofread with the same efficiency as the base substitution mutations, and (iii) (Ϫ1)-frameshift errors significantly increased under conditions of biased dNTP pools, specifically at template positions where the 5Ј neighbor is complementary to the dNTP provided in excess (6). We further suggested that the high level of (Ϫ1)-frameshift mutations created by this enzyme reflects a general inability to directly extend terminal mismatches and a greatly enhanced extension efficiency if the terminal mismatch can be misaligned on the next template base.
In this study, we have further investigated the mismatch extension capability of the wild-type and MutD5 holoenzyme by using a fidelity assay specifically designed to measure the relative probabilities by which a polymerase chooses each of the three possible pathways upon encountering a mismatched primer-terminus: (i) direct extension (yielding a base substitution), (ii) misalignment extension (yielding a Ϫ1-frameshift), or (iii) exonucleolytic proofreading (yielding no mutation). We show that for a mispaired primer-template in which misalignment is not possible, HE preferentially proofreads the mispaired base (Ͼ99%). However, for a primer-template in which the terminal primer base is complementary to the next template base, the polymerase extends from the slipped intermediate to yield a (Ϫ1)-frameshift mutation in a significant fraction of the cases. These data support models in which the in vitro frameshift specificity of HE results from its relatively low efficiency in direct mispair extension compared with its efficiency in misalignment and extension of the misaligned intermediate.

EXPERIMENTAL PROCEDURES
Materials and Chemicals-Bacteriophage M13mp2 and its mutant derivatives containing a T3 G substitution at position 103 (mp2G103) or a G3 A substitution at position 102 (mp2A102) in the lacZ␣ gene were obtained from Dr. Katarzyna Bebenek (National Institute of Environmental Health Sciences, Research Triangle Park, NC). Strains MC1061, which was used to prepare competent cells, and CSH50, which was used as an ␣-complementation host to score plaque color, were stocks of our laboratory. The holoenzymes from a wild-type and a mutD5 strain were purified as described previously (6). Klenow polymerase (exo Ϫ ) and restriction endonucleases BamHI and KpnI were from New England Biolabs (Beverly, MA). Ultrapure dNTP, ATP, and E. coli single-stranded binding protein were purchased from Pharmacia Biotech, Inc.
Construction of M13mp2 Substrates with a 3Ј Terminal Mismatch-M13mp2G103 RF DNA containing unique sites for restriction endonucleases BamHI and KpnI (15) was doubly digested with the two enzymes to produce 4061-and 3135-bp-long fragments. These fragments were separated by electrophoresis on a 0.8% agarose gel, and the 3135-bp fragment was purified from the gel by electroelution. The fragment was desalted and concentrated using a Microcon-50 microconcentrator (Amicon Inc., Beverly, MA). To form the terminally mismatched heteroduplexes, the purified fragment was hybridized to single-stranded circular viral mp2 DNA that was either wild-type (G102, T103) or altered at position 102 (A102, T103) (see Fig. 1), as follows. The fragment was diluted to 5 g/ml with water and incubated at 70°C for 5 min to denature the strands. Single-stranded circular viral DNA was added (5 g/ml, final concentration) just before removal from the 70°C water bath, and the mixture was placed on ice. After 5 min, SSC was added to a final 2ϫ concentration (300 mM NaCl and 30 mM sodium citrate), and the mixture was incubated for 5 min at 60°C, after which it was placed on ice. The resulting partial duplex DNA was desalted and concentrated by a Microcon-50 microconcentrator. Using a 1:1 ratio of fragment to viral DNA, about one-half of the single-stranded DNA was converted to partially heteroduplex molecules containing the terminal T⅐C mismatch at position 103 (Fig. 2, lane 1).
In Vitro DNA Synthesis Reactions-Mismatch extension experiments were performed in a 25 l volume at 30°C in 1.5-ml Eppendorf tubes containing 45 fmol of partially heteroduplex substrate, 30 mM HEPES, pH 7.5, 16 mM Tris, pH 7.5, 6 mM dithiothreitol, 10 mM MgCl 2 , 100 g/ml bovine serum albumin, 500 M ATP, 1 g of single-stranded binding protein, and either 50 or 1000 M each of the four dNTPs. Reactions were initiated by adding 135 units of MutD5 HE or 120 units of wild-type HE (1 unit ϭ 1 pmol of nucleotide incorporated per minute). At defined time points (from 15 s to 5 min), the reaction mixtures were quenched by adding 2 l of 200 mM EDTA. Extension reactions by the Klenow fragment (0.5 unit/reaction; 1 unit ϭ 10 nmol of nucleotide incorporated per 30 min) were carried out for 30 min in the absence of single-stranded binding protein and ATP. The reactions were extracted twice with an equal volume of phenol:chloroform:isoamyl alcohol (25: 24:1) and separated by electrophoresis on a 0.8% agarose gel (70 V, 4 h). Full-length reaction products (RF II band) were cut out from the gel, and the DNA was extracted using a QIAquick Gel Extraction Kit (Qiagen Inc., Chatsworth, CA). The extracted DNA was desalted and concentrated into a 10-l final volume using a Microcon-50 microconcentrator.
Transfection and Plating-2 l of gel-purified RF II DNA were used to transfect 50 l of competent cells prepared from E. coli strain MC1061 by electroporation with a Bio-Rad Gene Pulser (Bio-Rad Laboratories, Inc.) set at 2.01 kV, 400 W, 25 F. Immediately after electroporation, 1 ml of SOC medium was added. Plating was performed by adding the transfected cells to 3 ml of melted soft agar (42°C) containing 2.5 g of 5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside, 0.24 mg of isopropyl-1-thio-␤-D-thiogalactoside, and 0.25 ml of a mid-log culture of indicator strain CSH50. This mixture was poured onto minimal agar plates. The plates were inverted and incubated for 18 -24 h at 37°C, followed by an additional 24-h incubation at room temperature. Plaques were counted and classified according to their dark blue, light blue, or colorless phenotype. The frequencies of mispair extension and misalignment extension were calculated by dividing the frequency of light blue plaques or colorless plaques, respectively, by 0.55. The latter represents the average 55% expression probability of a mutation contained in the (Ϫ)-strand of a full-length RF II molecule (17,18). The use of a mismatch repair-proficient strain, such as MC1061, is preferred in this assay because it minimizes the occurrence of mixed (blue/colorless) plaques that make it difficult to score the light blue phenotype (17,18).

RESULTS
To examine how E. coli DNA polymerase III HE processes terminal mismatches in vitro, we used a fidelity assay that permits distinction between the different pathways available to the polymerase upon encountering a primer-template containing a 3Ј terminal mismatch. The mismatched substrates used in this study are described in Fig. 1. They are based on the M13mp2 lacZ ␣-complementation system and use differential plaque colors to distinguish the various reaction products. They represent a modification from those originally described by Kunkel and co-workers (17, 19 -21), which have been used to assess terminal mispair utilization by a number of different polymerases. Instead of using duplex DNA containing a small gap, we have constructed half-duplex M13mp2 molecules consisting of a 3135-bp duplex region and a 4061-bp singlestranded region. The 3135-bp primer contains the terminal mismatch. Upon synthesis by polymerase III HE, the fulllength RF II products can be readily separated from the starting molecules using a 0.8% agarose gel (see Fig. 2), thus allowing us to recover and separately analyze the full-length extension products. This feature is important for assaying rapid and highly processive polymerases such as HE, because it permits the analysis of extension products at very short reaction times, even when not all substrates are bound or extended.
Two substrates were prepared, both of which contained a T⅐C mismatch at the 3Ј end of the primer strand (position 103 of the lacZ␣ sequence) (Fig. 1). In one substrate, the next template nucleotide to be copied is a G, in the other substrate, it is an A. In both cases, the template strand is phenotypically wild-type (dark blue plaque), because the G3 A change at position 102 is silent (17). However, the mismatched C in the primer strand is derived from a T3 G base substitution mutant at position 103, which produces a light blue plaque. There are three possible outcomes for these two substrates when processed by HE. First, HE may simply extend the mispaired terminus C, creating a RF II heteroduplex molecule containing a base substitution mutation in the (Ϫ)-strand. Upon transfection, this molecule will yield a light blue plaque. (Actually, due to the action of mismatch repair on the internal mismatch, upon transfection into the competent cells, the probability of producing a light blue plaque is about 55%; the remaining 45% carry the (ϩ)-strand genotype and will be dark blue. The 55% (Ϫ)-strand expression efficiency has been determined experimentally for a large series of base⅐base and frameshift mismatches (17,18)). Second, in the case of the G-containing substrate, the terminal mismatched C may first misalign on the next template base G, followed by the extension of this misaligned structure. Upon transfection, this will yield (again at 55% efficiency) a mp2 frameshift mutant with a colorless plaque phenotype. Third, instead of extending the terminal mismatch, HE may remove the mispaired C by exonucleolytic proofreading. Upon extension, this would produce a wild-type dark blue plaque. Thus, a simple enumeration of light blue, colorless, and dark blue plaques reveals the fate of the terminal mismatch upon polymerization.
Extension reactions were performed for each DNA substrate with wild-type or MutD5 HE at two different dNTP concentrations: 50 and 1000 M. The reaction products were separated by agarose gel electrophoresis (see Fig. 2), and full-length RF II molecules were purified and analyzed by transfection. Timecourse experiments showed that the RF II product had already appeared after 15 s of reaction time, as expected for a rapid and highly processive enzyme such as HE (data not shown). As a control, we also performed extension reactions with the Klenow fragment (exo Ϫ ; see "Experimental Procedures").
The data in Fig. 3 show that with the substrate containing the T⅐C mispair followed by a template A (right panel), most of plaques scored (Ͼ99%) in reactions with either the wild-type or MutD5 HE were dark blue, indicating that proofreading of the terminal base was the predominant outcome. The failure to detect light blue or colorless plaques even for the MutD5 HE, which possesses only limited (ϳ5%) exonuclease activity (5), indicates that HE is very inefficient in extending directly from the mispaired terminus.
Extension by MutD5 HE of the substrate containing the T⅐C mispair with the 5Ј neighboring template G (Fig. 3, left panel) yielded high numbers of colorless and dark blue plaques and significantly fewer light blue plaques. Sequencing of randomly  selected colorless and light blue plaques (10 each) showed that all colorless mutants had lost template T at the site of mispair, and that all light blue mutants carried the expected T to G base change at position 103. The proportion of colorless and light blue plaques was lower when the dNTP concentration was decreased from 1000 to 50 M, reflecting the stimulatory effect of high dNTP concentrations on the two forward reactions as they compete with the proofreading step (6). The data with the G-containing substrate demonstrate that MutD5 HE is highly proficient in forming and extending a (Ϫ1)-frameshift intermediate (when permitted by the sequence context) as compared with direct extension of the mismatch, which is inefficient. The wild-type enzyme at either dNTP concentration produces a majority of dark blue plaques, indicating that proofreading is the predominating pathway. Nevertheless, among the mutants, colorless plaques outnumber the light blue plaques, indicating that misalignment extension is strongly preferred over direct extension in this case also. Note that the Klenow fragment greatly prefers direct extension over misalignment extension, as seen from the excess of light blue plaques versus colorless plaques (Fig. 3).
In Table I, we present the calculated probabilities for the G-containing substrate for each of the three pathways. At 1000 M dNTP, the MutD5 HE chooses the misalignment extension pathway in 61% of the cases. The direct extension pathway is chosen in only 6% of the cases, while 33% of the mismatches are proofread. At 50 M dNTP, the proportion of proofread molecules increases to 80.8%, whereas 17.8% are processed by misalignment extension, and only 1.5% are directly extended. Thus, the ratio between direct extension and extension from a misaligned intermediate at either dNTP concentration is about 1:10. The wild-type HE, which has a strong exonucleolytic proofreading activity (6), prefers to excise the mispaired C before polymerization. However, at 1000 M dNTP, about 2% of the primer-templates are processed by HE to yield the (Ϫ1)frameshift mutation, whereas only 0.1% is directly extended. Reactions sampled at different time points between 15 s and 5 min yielded the same ratios between direct extension, misalignment extension, or proofreading (data not shown). Because smaller HE subassemblies (core, polymerase IIIЈ, polymerase III*) cannot synthesize the required long stretch (ϳ4000 bp) of DNA within a short time (and in the presence of singlestranded binding protein), there can be little doubt that the observed products are made by HE. The results obtained with HE are in contrast to those obtained with the Klenow polym-erase: this enzyme efficiently extended the mispaired C (95%) and only rarely chose the misalignment pathway (5%).

DISCUSSION
The data obtained in this study reveal properties of E. coli DNA polymerase III HE that appear to be unique among the polymerases tested to date. First, when presented with a preformed terminal mismatch, direct extension of the mismatch by the enzyme to yield the expected base substitution is extremely infrequent. For example, for the A-containing substrate (Fig.  1), only about 0.2% is extended to yield the expected base substitution (Fig. 3), with the vast majority (Ͼ99.8%) being removed by proofreading. This is true even for the proofreading-impaired HE purified from the mutD5 mutator strain, in which only 5% or less of the exonuclease activity remains (6). Secondly, when presented with a terminal mismatch in a sequence context in which the terminal base in the primer strand is complementary to the next template base (the G-containing substrate; see Fig. 1), extension is greatly facilitated but now takes place from the misaligned intermediate, yielding a (Ϫ1)frameshift instead of the base substitution mutation. In the case of the proofreading-impaired enzyme, this second pathway significantly exceeds the exonuclease pathway (61 versus 33%; see Table I). For the wild-type enzyme, proofreading still predominates; however, most significantly, misalignment extension exceeds direct extension by 20:1 (Table I). This strong bias favoring misalignment extension over direct extension is unique to polymerase III, because other enzymes for which misalignment extension has been documented, such as Klenow fragment (15,21), avian myeloblastosis virus and Moloney a Probabilities of misalignment extension and direct extension were calculated from the data in Fig. 3, making a correction for the average (Ϫ)-strand expression value of 55% (see Refs. 17 and 18 and the text). The proofreading probability is calculated as 100% minus the sum of the percentage of misalignment extension and direct extension.

FIG. 3. Terminal mismatch utilization by polymerase III holoenzyme.
The two terminally mismatched substrates used in this study are shown. The next template base to be copied after the terminal mismatch is underlined (G and A in the left and right panel, respectively). The numbers represent the observed numbers of dark blue, light blue, or colorless plaques seen after transfection of the purified RF II product from the indicated extension reactions (see the text for a more detailed explanation). murine leukemia virus reverse transcriptases (19), mammalian polymerase ␤ (19), and yeast polymerase I (20), generally produce the frameshift mutation less efficiently than the base substitution. To illustrate the extreme tendency of polymerase III HE in this respect, we compared the enzyme side by side with the Klenow fragment. This enzyme also performed misalignment extension, but at only 5% of the rate of the direct extension (Table I).
The present data on mismatch extension by HE are relevant for explaining the unusual spectrum of in vitro errors produced by HE in a forward mutagenesis assay using the lacI gene as a mutational target (6). In these spectra, wild-type and MutD5 HE produced a majority of (Ϫ1)-frameshift mutations at nonreiterated sequences. We proposed (6) that these frameshifts originated as base misinsertion errors that failed to be extended by the enzyme but were processed efficiently by extension from the slipped intermediate in sequence contexts where the next template base is complementary to the misinserted base. The present data demonstrating the clear preference for misalignment extension over direct extension when HE is faced with a (preformed) 3Ј terminal mismatch are fully consistent with this model. Although a preformed terminal mismatch may not resemble a mismatch created during ongoing DNA synthesis in all respects, the behavior displayed by HE is striking and must reflect some property of HE that is likely relevant during ongoing DNA synthesis as well.
An important alternative context in which to consider the tendency of HE to create (Ϫ1)-frameshifts is provided by the "dNTP-stabilized misalignment" model proposed by Bloom et al. (7). These investigators noted increased misincorporation by HE on an oligonucleotide template under conditions in which the misincorporating nucleotide was complementary to the next template base. The findings were interpreted to indicate that the polymerase is able to read ahead and use the information provided by the next template position to direct (mis)incorporation. First of all, this model provides an alternative mode of misincorporation by HE compared with direct base⅐base misinsertion. If correct, both base⅐base mismatching and misincorporation by dNTP-stabilized misalignment may provide the terminally mismatched substrates that HE may extend from either the aligned or misaligned state. Secondly, provided that the primer-terminus remains in the misaligned state after phosphodiester bond formation, the reaction product of a misincorporation via dNTP-stabilized misalignment is the same as the intermediate that we suggested to arise from a standard misincorporation and subsequent forward misalignment. This feature of the dNTP-stabilized misalignment model for misincorporation makes it a potentially attractive pathway for the generation of the (Ϫ1)-frameshifts observed in vitro (6). However, if the forward extension rate is slow relative to the interconversion of the aligned and misaligned states, then any distinction with regard to the initial mode of misincorporation would disappear. Finally, it is likely that the preferred usage of misaligned intermediates by HE and misincorporation by dNTP-stabilized misalignment are manifestations of the same, unusual enzymological property of HE that allows the enzyme to generate and/or use misaligned intermediates with relatively high efficiency. This property, which is obviously relevant for the fidelity of the enzyme, is most intriguing and deserves further investigation.
In addition to providing insight into the mechanism of frameshift mutagenesis by HE, our present study also provides information about the contribution of proofreading to in vitro fidelity. The data from Table I indicate that for the MutD5 enzyme, 33 and 81% of the terminal mismatches are proofread at 1000 and 50 M dNTP, respectively; for the wild-type en-zyme, these numbers are 97.8 and 99.6%, respectively. Converted into fidelity factors (the fold reduction in mutant fraction due to proofreading), the proofreading activity of MutD5 HE contributes 1.5-fold [100/(100 -33)] and 5.3-fold [100/(100 -81)] to the accuracy of extension synthesis at the two dNTP concentrations, respectively, whereas for the wild-type enzyme, these factors are 45-and 250-fold. Thus, the efficiency of the MutD5 proofreading activity is reduced between 30-and 47fold compared with the wild-type enzyme. These data are consistent with direct measurements of the exonuclease deficiency associated with the MutD5 enzyme taken in the absence of dNTPs that indicated that the exonuclease activity was decreased 26-to 47-fold (6). Thus, it appears that in the present experiments, which use a preformed terminal mismatch, proofreading correlates directly with the strength of the exonuclease. Interestingly, this correlation does not appear to hold for the proofreading contribution during ongoing in vitro DNA synthesis, in which we measured only a 4-to 6-fold difference in the error rates (for both base substitutions and frameshifts) between the wild-type and the MutD5 enzyme. This may be due to a different binding mode of the enzyme in the two assays. Possibly, upon initial binding to a primer, HE preferentially binds with the primer in the exonuclease subunit, as has been shown to be the case for T7 DNA polymerase (22). In contrast, during ongoing DNA synthesis, the terminal nucleotide is in the polymerase active site, and a kinetic barrier may exist for transfer to the exonuclease site.
Whereas this study provides new insights into the fidelity behavior of the replicative complex of E. coli, numerous questions still remain. Poor extension from terminal mismatches can be readily seen as a valuable attribute for a high fidelity enzyme because it will greatly increase the potential for exonucleolytic removal, but it is not clear why a high tendency to transform a terminal mismatch into a frameshift intermediate is particularly useful. This tendency seems to be a property intrinsic to the ␣ subunit, because a similar frameshift predominance was observed in fidelity experiments performed with the isolated ␣ subunit (23). Apparently, the numerous other subunits present in the HE do not play a significant role in preventing these frameshifts. However, because high levels of frameshift mutagenesis are not observed in vivo (even before the action of mismatch repair) (24), the question of how they are prevented inside the cell is interesting. We suggest that this additional mode of error prevention is somehow related to the functioning of HE at the in vivo replication fork. Further experiments will be needed to address this important question.