Proofreading of DNA Polymerase h -dependent Replication Errors*

,

The UmuC/DinB superfamily of DNA polymerases includes human DNA polymerase (pol ), 1 the product of the XPV (Rad30A) skin cancer susceptibility gene (1,2). Human pol has the ability to efficiently copy cis-syn thymine-thymine dimers in template DNA (1). Mutations in XPV that inactivate pol (1,2) render cells hypermutable by UV radiation (3)(4)(5)(6)(7) and defective in replicating DNA containing UV photoproducts (Ref. 8 and references therein). These facts demonstrate an important role for pol in modulating UV-induced mutagenesis and in reducing the risk of human skin cancer. Kinetic analysis reveals that human pol inserts incorrect nucleotides opposite undamaged (9,10) and damaged (10) template bases more efficiently than most other DNA polymerases. Moreover, human pol lacks an intrinsic proofreading exonuclease activity (9), and its base substitution error rates when copying undamaged DNA are much higher than are those of most other eukaryotic polymerases, whether they have proofreading activity or not (9). We (9) and others (10) have suggested that this generally relaxed discrimination ability during DNA synthesis may be critical to the ability of human pol to bypass certain DNA lesions that impede synthesis by other DNA polymerases (1,8,11,12).
Current models suggest that pol competes with other replicative polymerases for 3Ј-OH termini at a replication fork (reviewed in Refs. [13][14][15]. Given the intrinsically low fidelity of pol , mechanisms may exist to prevent errors by pol from reducing the accuracy of chromosomal replication. We previously suggested (9) two obvious error correction mechanisms, exonucleolytic proofreading of pol mistakes by a separate exonuclease(s) and post-replication DNA repair of mismatches generated by pol . To test the proofreading hypothesis, here we examine the effects of pol on the accuracy of replication of double-stranded DNA catalyzed by the replication apparatus in extracts of human cells. For this we used the SV40 replication system, whose protein requirements are similar to those of human chromosomal replication (16). We first establish replication conditions under which pol reduces replication fidelity, suggesting that pol can indeed compete with other polymerases during semiconservative DNA replication. We then demonstrate that pol has two intrinsic biochemical properties, low processivity and slow mismatch extension, that could allow a separate exonuclease to compete for mismatched primer termini at the replication fork. Finally, we demonstrate that pol -induced replication infidelity depends on the dNTP concentration and is increased in the presence of a deoxynucleoside monophosphate, both classical hallmarks of exonucleolytic proofreading.
SV40 Replication Fidelity Measurements-SV40 origin-dependent replication reactions were performed using extracts of human TK6 and HeLa cells as described previously (18,20). When analyzed by agarose gel electrophoresis, the products generated in the reactions listed in Table I were similar to those seen in earlier studies (16 -18, 21). Analysis of the lacZ mutant frequency of the replication products by introduction into an E. coli lacZ ␣-complementation host strain and plating to score wild-type (blue) and mutant (colorless and light blue) plaques was performed as described previously (20).
Kinetic Analysis of Mismatch Extension-Reactions (25 l) were as above except that they contained 200 nM template primed at 1.2 to 1 molar ratio with a 5Ј-32 P-labeled primer, 2 nM pol , and either dATP or * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
dGTP. The template-primers are shown in the legend to Table III. Aliquots were removed at 2, 4, 6, and 8 min and products separated by electrophoresis in 16% polyacrylamide gels and quantified by phosphorimagery.

RESULTS AND DISCUSSION
Previous studies indicated that the fidelity of replication of undamaged M13mp2 DNA by extracts of human TK6 or HeLa cells is high (17)(18)(19)21). In this study, inclusion of human pol in a replication reaction reduced fidelity, as indicated by the concentration-dependent increase in the frequency of lacZ mutants among the M13mp2 products of semiconservative replication (Table I, Experiment 1). No increase in mutant frequency was observed when excess human pol ␤ was included or when pol was incubated with the extract in the absence of the SV40 large T-antigen that is required for replication from the SV40 origin. In fact the mutant frequencies under both reaction conditions were within the range of frequencies (3-7 ϫ 10 Ϫ4 ) for control DNA that had not been replicated in the extract.
DNA sequence analysis of 28 independent lacZ mutants from the reaction containing 100 nM pol revealed the presence of 17 single-base substitutions, 11 of which were consistent with incorporation of dGTP opposite T. One clone had a tandem double-base substitution and seven contained a single nucleotide deletion. This error specificity is remarkably similar to that of pol during gap-filling synthesis (9). A 14-nucleotide insertion was also recovered and three of the lacZ mutants contained two widely separated sequence changes, circumstances not encountered in previous studies of replication fidelity. Single base substitution and frameshift error rates calculated from these data are higher than for replication in the absence of added pol (Table II, Experiment 1). Overall, these data suggest that pol is capable of competing with other DNA polymerases at the replication fork and generates both base substitution and frameshift mutations.
We previously suggested that the mutagenic potential of pol in human cells might be reduced if pol -dependent errors were proofread (9). Several experiments were conducted to test this hypothesis. First, primer extension reactions were performed using the lacZ template and a 1000-fold molar excess of template-primer over enzyme, a condition that results in a single cycle of processive synthesis. Analysis of reaction products (Fig. 1) demonstrated that pol polymerizes one to ten nucleotides per cycle of enzyme binding dissociation and that

TABLE II Specificity of replication errors induced by human pol
Error rates were calculated as described (17,19,21), using the mutant frequencies shown in Table 1, the lacZ mutant sequence information described in the text, and the target sizes for scoring particular classes of errors that are described in Refs. 19  the probability of termination of processive synthesis after each incorporation varies between about 40 and 70%. This result quantitatively confirms the earlier observation (8) that human pol has low processivity, a property that could provide an exonuclease access to a template-primer containing a mismatch. All DNA polymerases studied to date extend mismatched template-primers less efficiently than matched termini. To determine whether this is also the case with human pol , we performed extension reactions to obtain steady-state kinetic constants from which extension efficiencies for matched and mismatched termini were calculated. The results (Table III) indicate that pol extended mismatched termini less efficiently than matched termini by factors of from 3-fold (T•G mismatch) to more than 100-fold (G•A mismatch). These data are consistent with the recent qualitative demonstration that human pol extends mismatched termini less efficiently than matched termini with undamaged DNA and at sites of DNA damage (8) and with a recent report that yeast pol also extends mismatched termini less efficiently than matched termini (22).
For polymerases having intrinsic proofreading exonucleases, slow polymerization increases the opportunity for movement of the primer terminus to an exonuclease active site for excision of a misinserted nucleotide (recently reviewed in Ref. 23). While pol lacks an intrinsic proofreading activity (9), it is possible that editing could be performed by a separate exonuclease that is physically associated with the replication machinery, as is the case in E. coli. To determine whether replication errors induced by exonuclease-deficient pol could be edited by an extrinsic exonuclease, replication fidelity was examined in a human extract under conditions known to modulate proofreading activity. One hallmark of proofreading is the "next nucleotide effect" (reviewed in Ref. 24). Since the probability of polymerization from a mismatch (or a misalignment) depends on the concentration of the next correct nucleotides to be incorporated, at high dNTP concentrations polymerization is favored over editing, and fidelity is reduced. This approach has already been used successfully to detect proofreading of base substitution (18,19) and frameshift errors (19,25) produced by the human replication complex. However, in reactions containing equimolar dNTPs, a next nucleotide effect is difficult to detect with the M13mp2 forward mutation assay, because replication fidelity is very high (17)(18)(19)(20)(21)25). Thus, replication in an extract to which excess pol was not added generated products with lacZ mutant frequencies that were similar to unreplicated DNA control values (Table I, Experiment 2). In contrast, when replication reactions were performed in the presence of pol , fidelity decreased as the dNTP concentration was increased from 10 to 1000 M (Table I, Experiment 2). The fact that the fidelity was higher in reactions lacking exogenous pol indicates that the majority of errors being proofread were dependent on pol .
A second hallmark of proofreading activity is reduced replication fidelity in the presence of a high concentration of dNMP (18 -20), the product of proofreading exonucleases. In the present study, inclusion of 2 mM dGMP in a replication reaction containing 100 M dNTPs and to which pol had been added FIG. 1. Processivity of human pol . Reactions and product analysis were performed as described under "Experimental Procedures." Lanes 1-3, products of primer extension reactions incubated for 5, 15, and 30 min, respectively. Termination probabilities are shown on the left and are expressed in percent, as the ratio of products at each site to the products at that site plus all greater length products. In 5 min, the enzyme extended 14% of the primer, indicating that the enzyme cycles, because the template-primer is in 1000-fold excess over enzyme. The probability of termination at each template position remained constant through the time of the reaction, demonstrating negligible initiation of synthesis on previously used template-primers. Lane 4, unextended primer. Lanes marked T, A, C, and G are markers depicting the sequencing ladder for this template-primer.

TABLE III
Kinetic analysis of mismatch extension by human pol-Reactions were performed as described under "Experimental Procedures." Duplicate determinations were performed using seven different concentrations of nucleotide, and kinetic constants were derived as described previously (32). The template primers used were: (a) dATP 2XTCTTTTGGGACCGCAATGG-5Ј (where X ϭ A, G, or C).  (Table I, Experiment 3) also reduced replication fidelity (lacZ mutant frequency of 45 ϫ 10 Ϫ4 compared with 20 ϫ 10 Ϫ4 for the equivalent reaction without dGMP). Again, the lower fidelity was dependent on pol . Both results imply that pol -induced replication errors can be proofread. We also examined whether pol competes for primer termini on the leading strand, the lagging strand, or both. Here we employed a strategy used previously (19), in which excess dGTP is included in a replication reaction to monitor misincorporation of dGTP during lagging strand replication of the M13mp2 viral (ϩ) strand or during leading strand replication of the complementary (Ϫ) strand. A reaction containing 50 nM pol and a 50-fold molar excess of dGTP generated replication products whose lacZ mutant frequency was strongly elevated relative to control values (Table I, Experiment 4). Sequence analysis of 29 independent lacZ mutants recovered from this reaction revealed 33 single base substitutions, 31 of which reflected misincorporation of dGMP. Thirty were due to misincorporation opposite template T, the most frequent error generated by pol during gap-filling DNA synthesis (9). Of these, nine were consistent with pol misincorporation during replication of the leading strand and 21 reflected pol misincorporation during replication of the lagging strand (Table II). The error rates calculated from these data are much higher than those observed for replication with excess dGTP in the absence of excess pol (Table II). Notably, seven independent lacZ mutants contained the same T to C substitution at template nucleotide 121. At this site, misinsertion of dGTP opposite this T is followed by correct incorporation of dGTP, which was present at 1000 M. Thus, this hot spot may reflect misinsertion by pol followed by suppression of proofreading due to the high dGTP concentration. Overall, these data imply that pol competes with the enzymatic machinery that replicates both the leading and lagging strands.
At first glance, the idea that pol is intrinsically inaccurate seems paradoxical in light of its role in suppressing UV-induced mutagenesis (3-7) and lowering the incidence of skin cancer (1,2). In fact, we believe that the properties of pol are ideally suited for this critical role. We have suggested that the intrinsically low fidelity of pol is a reflection of relaxed selectivity that facilitates highly efficient bypass of UV photoproducts that block other replicative polymerases. This implies that the low fidelity observed in vitro is not simply due to the lack of an accessory protein that might enhance pol selectivity in vivo. We further suggest that a high probability of termination of processive synthesis (Fig. 1) and slow extension of mismatches (Table III) facilitates proofreading (Table I). Proofreading could be performed by exonucleases intrinsic to replicative DNA polymerases ␦ and ⑀ or by a separate exonuclease (26,27). In the former case, the exonuclease active site of pol ␦ or pol ⑀ could directly bind the mismatch for immediate removal, as seen with the replicative T7 DNA polymerase (28). Even if pol extends a mismatch for a few nucleotides (note that discrimination for extension of a T•G mismatch in Table  III is only 3-fold), pol ␦ or ⑀ may still be able to edit a mismatch. This is suggested by studies (reviewed in Ref. 23) showing that some polymerases contact the DNA for five base pairs upstream of the polymerase active site, such that the presence of an embedded mismatch could still promote excision over polymerization. It is also possible that errors that escape proofreading could be corrected by mismatch repair (29,30), thereby further contributing to the fidelity of the overall bypass process. Finally, our data indicating that pol can compete for 3Ј-OH termini during replication by leading and lagging strand replication proteins suggest that genome instability could result from conditions that promote this competition, such as a change in the ratio of pol relative to other polymerases. This could result from overexpression of pol , from a reduction in polymerization efficiency due to mutations, or from reduced expression of other polymerases. As one example of the latter, decreased cellular levels of pol ␦ have recently been shown (31) to promote genomic instability by an unknown mechanism.