Frequency and fidelity of translesion synthesis of site-specific N-2-acetylaminofluorene adducts during DNA replication in a human cell extract.

We have previously analyzed the effects of site-specific N-2-acetylaminofluorene (AAF) adducts on the efficiency and frameshift fidelity of SV40-based DNA replication in a human cell extract (Thomas, D. C., Veaute, X., Kunkel, T. A., and Fuchs, R. P. P.(1994) Proc. Natl. Acad. Sci. U.S.A. 91, 7752-7756). Here we use two sets of substrates to examine the probability of replication termination and error-free and error-prone bypass of AAF adducts. The substrates contained site-specific adducts at one of three guanines in a NarI sequence (5′-GGCGCC-3′) placed within the lacZα reporter gene and located on the template for either leading or lagging strand replication. The presence of the adduct at any position strongly reduces the efficiency of a single round of replication in a HeLa cell extract. Product analysis reveals preferential replication of the undamaged strand and termination of replication of the damaged strand occurring one nucleotide before incorporation opposite either a leading or lagging strand adduct. Products resistant to restriction endonuclease cleavage at the adducted site were generated in amounts consistent with 16-48% lesion bypass during replication. Most of this bypass was error-free. However, two-nucleotide deletion errors were detected in the replication products of DNA containing an AAF adduct in either the leading or lagging strand, but only when present at the third guanine position. Collectively, the data suggest that the replication apparatus in a HeLa cell extract generates a template-primer slippage error at an AAF adduct once for every 30-100 bypass events.

DNA replication in eukaryotic cells requires multiple proteins that function in an asymmetric manner to coordinately synthesize the leading and lagging strands (Kornberg and Baker, 1992). Given the asymmetry of replication of duplex DNA, unrepaired DNA adducts may have different potentials for blocking replication or promoting replication infidelity depending on whether the adduct is located on the leading or lagging strand. To examine this, we have been using the SV40 origin-dependent replication system (for review, see Stillman (1994), and references therein) as a model for human chromosomal replication. With this system, replication of undamaged DNA in mammalian cell extracts is highly accurate (Roberts and Kunkel, 1988;Hauser et al., 1988;Thomas et al., 1991). However, replication of DNA containing randomly placed cyclobutane pyrimidine dimers (CPDs) 1 is highly mutagenic Carty et al., 1993). The assay system was adapted  to determine whether, as a result of the asymmetry of replication, CPDinduced errors arise at different rates during leading and lagging strand synthesis. Overall average error rates were equal for the leading and lagging strand replication machinery, with strand-specific differences in fidelity observed at some nucleotide positions .
For several years, we have studied the effects of an encounter between the replication fork and the major adduct of a known carcinogen, N-2-acetylaminofluorene (AAF) located at one template position. Studies in Escherichia coli have revealed that AAF, which binds primarily at the C-8 position of guanine, is a strong mutagen for Ϫ1 and Ϫ2 frameshift mutations (Fuchs et al., 1981;Koffel-Schwartz et al., 1984;Schaaper et al., 1990), particularly within repetitive sequences that permit slippage to form misaligned intermediates (Burnouf et al., 1989;Lambert et al., 1992). Mutagenesis within these sequences is strongly dependent on adduct location. Template modification at the 3Ј most base in a run of guanines or at the third guanine in the NarI recognition sequence 5Ј-GGCGCC-3Ј is much more mutagenic than at the other guanine residues because stable intermediates can form during replication (Lambert et al., 1992;Garcia et al., 1993;Milhé et al., 1994). More recently, we have extended these studies in human cell extracts using the SV40based replication system. We examined the efficiency and frameshift fidelity of replication of DNA containing a sitespecific AAF adduct present at one of the three guanine positions in the NarI sequence, each on the leading strand template relative to the first fork to encounter the adduct (Thomas et al., 1994). The primary effect of the adduct was to inhibit replication. When replication of the damaged template did occur, evidence was obtained for error-free lesion bypass with all three substrates, and for error-prone lesion bypass with the substrate that carries the AAF adduct at the third guanine.
When a ColE1-derived plasmid is replicated in E. coli, sitespecific AAF adducts induced both Ϫ1 and Ϫ2 frameshift mutations at a 20-fold higher frequency when the adduct is located on the lagging strand as compared to the leading strand (Veaute and Fuchs, 1993). We hypothesized that the asymmetry of leading and lagging strand replication enzymology in human cells might also affect the frequency of mutations induced by AAF adducts. For this study we constructed two sets of four DNA substrates, each containing the SV40 replication origin and either no damage or a single, site-specific AAF adduct at one of the three guanines in a NarI sequence. In both sets of substrates, the adducts are located asymmetrically relative to the origin of replication, such that they will be encountered first by either the leading or lagging strand replication apparatus. After replication in a human cell extract, we performed product analyses to quantitatively compare the probabilities of replication termination, error-free bypass and sequence context-dependent, error-prone bypass.

EXPERIMENTAL PROCEDURES
Materials-E. coli strains and construction of substrates containing site-specific AAF adducts have been described previously (Burnouf et al., 1989;Veaute and Fuchs, 1993). The substrates (Fig. 1A) are described below. Unmodified substrate (G0) was prepared by the same procedure as for AAF-containing DNAs, starting with mock treatment of the oligonucleotide used to insert the adduct. Radionuclides were from Amersham Corp.; T antigen was from Molecular Biology Resources. Enzymes were from New England Biolabs or U. S. Biochemical Corp. Extracts were prepared as described (Li and Kelly, 1985).
Replication Reactions and Product Analysis-Reactions (25 l) containing [␣-32 P]dCTP, 100 M dNTPs, 10 -20 ng of DNA, and other components as described  were incubated for 1 h (or for times indicated in Fig. 2) at 37°C. One-tenth of each reaction was collected on filters following precipitation with trichloroacetic acid to determine total incorporation. Following purification of the DNA , an aliquot of each replication reaction was analyzed by electrophoresis (55 V) on a 1.1% agarose gel containing 0.2 g/ml ethidium bromide. The dried gel was exposed to a phosphor screen for scanning and products quantitated using a Molecular Dynamics PhosphorImager.
Measurement of Repair in Extracts-Repair 2 of adducts was measured by incubating 32 P-radiolabeled substrate in a standard reaction mixture in the absence of T antigen for 1 h. Following purification of the DNA by the same procedure used for replication products, the samples were treated with 1 unit of restriction endonuclease BsaHI (an isoschizomer of NarI) for 1 h at 60°C. The samples were resolved by agarose gel electrophoresis, and the dried gel was quantitated as described above. Repair (expressed in percent) was calculated by dividing the sum of the two cleavage products (cut) by the sum of the intensities of the band resistant to cleavage (uncut) plus the two cleavage products, after subtracting background values for a blank region of the gel.
Analysis of Lesion Bypass-Purified DNA samples were digested with 1 unit of AvaII for 1 h at 37°C to linearize the DNA and then with 1 unit of BsaHI for 1 h at 60°C. The samples were resolved by agarose gel electrophoresis, and the dried gel was quantitated as described above. Lesion bypass was calculated as described in the legend to Table I.
Determination of Reversion Frequencies-Following purification, replicated DNA samples were digested with DpnI. The lacZ ␣-complementation reversion frequencies for control and replicated DNAs were determined by electroporation of E. coli strain JM103 to score blue and white colonies on LB indicator plates containing ampicillin, as described (Burnouf et al., 1989). The assay scores errors that restore the correct reading frame (lacZ ϩ ) from the ϩ2 reading frame (lacZ Ϫ ) of the starting DNA as blue colonies. White colonies represent the remaining total of replicated products.
Analysis of Replication of the Two Strands-Replicated samples were processed as usual and digested with PvuII and EcoRI in a 30-l reaction, using conditions optimal for PvuII. The samples were desalted, lyophilized, dissolved in 10 l of formamide loading dye, and resolved on an 8% polyacrylamide sequencing gel. Bands on the dried gel were quantitated as above.

RESULTS
Description of Substrates-Two sets of DNA substrates were prepared (Fig. 1, A and B). Each set consists of a control lacking an AAF adduct and substrates with an AAF adduct on one of the three different guanines in a NarI sequence G 1 G 2 CG 3 CC in the lacZ␣ reporter gene of pUC8. All substrates contain the SV40 origin of replication to support T antigen-dependent DNA replication in a human cell extract. The two sets differ only in the relative orientation of the restriction fragment containing the reporter gene. When modified with AAF, the pMKBNar substrate contains the adduct on the leading strand template with respect to the closest replication fork emanating from the SV40 origin (Fig. 1C). When pMZBNar is used, the adduct will be located on the lagging strand template (Fig. 1D). The dis- 2 We refer to repair of AAF adducts as any process leading to loss of the adduct such that the NarI sequence is rendered sensitive to BsaHI restriction endonuclease.
FIG. 1. Maps of plasmids pMKBNar and pMZBNar and the location of unique AAF adducts on the leading and lagging strands. A and B, maps of plasmids containing site-specific AAF adducts. Vectors contained either no modification or unique AAF adducts in the lacZЈ gene (lacZЈ refers to the modified lacZ sequence containing the NarI sequence). The precise locations of the adducts are shown in Fig. 5 (A and B). For reference, pMKBNar contains the G3 adduct at nucleotide 419, while pMZBNar contains the G3 adduct at nucleotide 484. Denoted are the locations of restriction sites used in bypass (BsaHI, AvaII) and uncoupling (PvuII, EcoRI) measurements. Ap is the resistance gene for ampicillin, ␤-lactamase. C and D, schematic showing location of site-specific AAF adducts on the leading or lagging strands with respect to the nearest fork emanating from the SV40 origin. This fork must replicate 520 or 585 nucleotides as the leading or lagging strand, respectively, before encountering the G3 adduct in pMKBNar or pMZBNar. Thus, 19% (pMKBNar) or 21% (pMZBNar) of the plasmid is replicated before the fork reaches the G3 adduct. Only a portion of the entire plasmid is presented in these linear drawings.
tances that the closest fork must travel from the SV40 origin to the first nucleotide of the NarI site are 520 and 585 nucleotides for the leading and lagging strand vectors, respectively, distances representing 19 and 21% of the total size of the plasmid. For simplicity, the substrates of each series are referred to as leading or lagging strand substrates and identified by the position of the adduct (G0, G1, G2, or G3). The lacZ␣ gene coding sequence of all substrates is in the ϩ2 reading frame, yielding white colonies on indicator plates. Mutations restoring the correct reading frame (e.g. two-nucleotide deletions) are detected as blue colonies.
Replication Time Course-Previous replication reactions with substrates containing site-specific AAF adducts were performed for 4 h (Thomas et al., 1994), potentially allowing enough time for more than one round of replication. This could significantly alter the product distribution obtained with substrates in which only one of the two strands contains a bulky adduct. Therefore, we began the present study by analyzing the replication products obtained at different incubation times to measure the fraction of products representing more than one round of replication. Both strands of the input DNA are methylated at adenines in 5Ј-GATC-3Ј sequences. A single round of replication generates hemimethylated DNA, whereas multiple rounds yield unmethylated DNA. The proportion of unmethylated, replicated DNA can be quantitated by digesting the radiolabeled reaction products with restriction endonuclease MboI. This enzyme cleaves only unmethylated DNA, leaving hemimethylated products intact.
The results of a time course for T antigen-dependent replication of undamaged and damaged DNA are shown in Fig. 2. Radiolabeled replication products are detected after a 30-min reaction, and product yield significantly increased in incubations of 1 h or longer. No MboI cleavage products were detected in 30 min. As the incubation time is increased, substantially more MboI cleavage products were observed. In 1-h reactions with G0 DNA, these products comprise approximately 7% of the total. This increases to 29% at 2 h and 38% by 3 h. For G3 DNA, these values were about 5% at 1 h, 29% at 2 h, and 51% at 3 h. Based on these data, we chose to perform subsequent replication reactions for 1 h, because the products at this time point predominantly reflect one round of replication and the yield of replicated DNA was adequate for further analysis.
Measurement of Adduct Removal in the Extract-To more accurately quantitate lesion bypass efficiency and replication fidelity, we next examined the extent of repair of AAF adducts by the same HeLa cell extract used for replication. We took advantage of an earlier observation that the AAF-modified NarI site is insensitive to cleavage by restriction endonuclease BsaHI (Thomas et al., 1994). Radiolabeled control and modified DNAs were incubated in the extract in the absence of T antigen. The products were purified, treated with BsaHI, and resolved on an agarose gel (Fig. 3). The undamaged G0 sample (lagging strand substrate) was completely cleaved. In contrast, while most of the G1 and G3 DNA products were insensitive to cleavage, some digestion was observed. Quantitative analysis of the results in Fig. 3 suggests that after a 1-h incubation in the extract, 14% of the G1 and G3 modified substrates were rendered adduct-free.
Replication of Modified Substrates in the Extract-T antigendependent replication reactions were performed for 1 h in the same HeLa cell extract. In three separate experiments, all modified substrates were replicated ϳ30% less efficiently than was the unmodified control substrate when total T antigen-dependent nucleotide incorporation was considered. Following precipitation and extraction of the products, a portion of each sample was subjected to electrophoresis on an agarose gel (Fig.  4). The alternate lanes marked (Ϫ) for BsaHI ϩ AvaII treatment represent the undigested products of reactions with G0, G1, and G3 DNAs. This product distribution is typical of the FIG. 2. Time course for replication of unmodified or AAF-containing plasmids in a HeLa cell extract. Standard replications reactions were performed containing 10 ng of unmodified (G0) or modified (G3) pNar DNA (Thomas et al., 1994) for varying incubation times as indicated. Samples were processed as described, and aliquots were left untreated or treated with MboI and were resolved on an agarose gel as described under "Experimental Procedures." The positions of RF I, RF II, and the two largest MboI cleavage products (cut) are indicated. To calculate the extent of synthesis beyond one round of replication, the sum of the radioactivity of the MboI cleavage products (extrapolated from the size of the largest fragment) was divided by the total radioactivity in the lane from RF I DNA to the top of the gel, after subtracting background values for a blank lane. The results are expressed as percentage.
FIG. 3. AAF adduct removal in a HeLa cell extract. Unmodified pMZBNar (G0) or modified pMZBNar (G1 and G3) DNA were incubated in a standard reaction mixture minus T antigen for 1 h. Samples were processed as usual, and equal aliquots were treated or not with restriction endonuclease BsaHI and resolved on an agarose gel as described under "Experimental Procedures." The positions of RF I, II, and III are indicated. Also denoted is the largest BsaHI restriction fragment (uncut, 1,978 bp) resulting from cleavage at the sites indicated in Fig. 1B except the site within lacZЈ (which is refractory to incision when the site is modified with AAF or is changed) and the largest cleavage product of this fragment (cut, 1,732 bp). Repair is calculated as described under "Experimental Procedures." pattern obtained previously with damaged or undamaged DNA Thomas et al., 1994) and includes covalently closed monomer-length products (RF I), as well as relaxed and linear DNAs and some high molecular weight species. To avoid overexposure, the volume of material loaded in each lane varied, and product yields were calculated by extrapolation after quantitation with the PhosphorImager. The yields of covalently closed monomer replication products were reduced on average by about 55% for each of the damaged DNAs relative to the undamaged DNA controls. Similar results were obtained with both series of substrates.
Estimation of Lesion Bypass Efficiency-To determine if the replicated, damaged DNA retained the AAF adduct, replication products were digested first by MboI, to cleave products having undergone more than one round of replication, and then by AvaII and BsaHI. AvaII was included to achieve better resolution of fragments of interest, while BsaHI-resistant products are inferred to have retained the AAF adduct (or to have a sequence change at the NarI site; see below).
The results are shown in Fig. 4, in the alternate lanes marked (ϩ) for BsaHI ϩ AvaII treatment. Digestion of the products of replication of the undamaged G0 substrates with AvaII and BsaHI yielded several bands as predicted from complete digestion and from the location of the known sites ( Fig. 1 and legend to Fig. 4). The largest of these bands is shown in the autoradiogram (indicated as cut), while higher mobility bands are not shown. Bands at these same positions were observed with the G1 and G3 substrates in both orientations. However, a new band (indicated as uncut) was observed with the G1 and G3 substrates. For the lagging strand G1 and G3 substrates, the mobility of this band is as predicted by lack of incision at the NarI site containing the AAF adduct. The same is true for the leading strand vector, but here the mobility of the "uncut" band is slightly lower, as predicted by the different location of the adducted NarI site relative to its neighbors. The data are consistent with the interpretation that this band represents replicated DNA containing the AAF adduct. To estimate the amount of translesion replication inferred from this result, the uncut band intensities were determined after subtracting background values for minus-T antigen controls (not shown) and the G0 sample. This value was then used to calculate the percentage of BsaHI-resistant product, after correcting for the amount of adduct removed in a 1-h incubation in the extract (Fig. 3, and see legend to Table I). Estimates ranged from 17 to 24% and were similar for all four substrates (Table I).
Differential Replication of the Damaged and Undamaged Strands-Following a replication reaction for 1 h in a HeLa cell extract, samples were purified, digested with PvuII and EcoRI, and then separated on a denaturing polyacrylamide gel. Staggered incision by EcoRI generates strands differing in length by four nucleotides. Fig. 5 (A and B) depicts the expected lengths of the two strands for the fragment containing the adduct and for a flanking fragment. A PhosphorImager print of the gel is shown in Fig. 5C. Note the relative intensities of the two bands representing each full-length restriction fragment. For the undamaged substrates the relative band intensities are about equal. In contrast, for the damaged substrates one band is substantially less intense than the other. For both fragments and each damaged substrate in either orientation, the band of lesser intensity represents the replication product synthesized from the damagecontaining template strand. Quantitative analysis of band intensities (not shown), including an adjustment for cytosine content (the radiolabeled nucleotide), shows that both strands of the undamaged substrates are replicated with equal efficiency. In contrast, with the AAF-modified substrates the undamaged strand is replicated preferentially over the damaged strand, by factors ranging from 2.2 to 3.3, during a 1-h incubation.
Also observed in Fig. 5C are unique bands in each of the lanes containing products of replication of the modified substrates. The lengths and relative positions of these bands strongly suggest that they are the products of termination of replication of the adducted template strand, as suggested by our previous study (Thomas et al., 1994). The exact lengths of the bands in Fig. 5C were unequivocally determined by mixing an aliquot of each sample with the products of a chain termination sequencing reaction. The results (not shown) suggest that with the G1, G2, and G3 substrates in both orientations, replication terminated after incorporation opposite the template nucleotide immediately preceding the adducted guanine. The intensities of these bands were quantitated and compared to the values for the full-length strand (120 nucleotides) and the complementary strand (124 nucleotides) after adjusting for FIG. 4. Estimation of bypass of AAF adducts by restriction analysis. Following replication of AAF-modified plasmids, samples were extracted and treated or not with BsaHI plus AvaII restriction enzymes, and resolved on an agarose gel as described under "Experimental Procedures." To avoid overexposure, approximately equal amounts of radioactivity, based on RF I products, were loaded in each lane. The positions of RF I, II, and III and the uncut and the cut fragments resulting from BsaHI and AvaII cleavage are indicated. The uncut fragment for the lagging strand vector (1,580 bp) is the largest of six fragments generated by BsaHI and AvaII when the NarI site is modified by AAF (or no longer exists). The smaller fragments are not seen in this figure. Cleavage of the unmodified NarI site yields the one large product indicated (cut, 1,334 bp) and a smaller fragment not shown (246 bp). The uncut fragment for the leading strand vector (1,888 bp) represents the largest of five fragments generated by BsaHI and AvaII when the NarI site is modified by AAF (or no longer exists). Again, the smaller fragments are not seen in this figure. Cleavage of the unmodified NarI site yields the one large product indicated (cut, 1,403 bp) and a smaller fragment not shown (485 bp). Lesion bypass was calculated as described in the legend to Table I. a The percentage of BsaHI-resistant replication products was calculated from the data in Fig. 4 by first dividing the intensity of the band resistant to cleavage (uncut) by the sum of the intensity of the two cleavage products (cut) plus the resistant band, after subtracting background values for a blank region of the lane and background values measured for a minus-T antigen lane. The net value for the cut products was reduced by 14% to adjust for DNA repair (see Fig. 3). Finally, the value obtained for G0 by this calculation (without adjusting for repair) was subtracted from the values obtained for damaged DNA to obtain net bypass, expressed as percentage.
cytosine content and DNA repair (see above). For the products of replication of all six modified substrates, the sums of the pixel values for the termination bands and the 120-nucleotide product are similar to the values for the complementary 124-nucleotide fragment (Table II). The relative values are not exactly 100%, perhaps reflecting low levels of second-round synthesis and/or some replication by the fork arriving from the other direction.
Values for the ratio of termination band intensity to the 124-nucleotide fragment intensity ranged from 69 to 95% (Table II). We suggest that these represent the probability that replication terminates upon encountering an adduct. Correspondingly, the ratio of the 120-nucleotide fragment intensity to the 124-nucleotide fragment intensity may approximate the probability of lesion bypass, which is needed for synthesis of full-length restriction fragment. These probabilities ranged from 16 to 27%.
AAF Adduct-induced Error Rates with Opposite-orientation Substrates-We next compared frameshift replication error rates when the AAF adduct was present on the leading or lagging strand relative to the closest fork emanating from the SV40 origin. Reactions were performed as above, and replicated samples were digested with DpnI to remove unreplicated or incompletely replicated molecules. Controls included untreated substrates and those incubated in the extract but with-FIG. 5. Measurement of strand asymmetry and termination bands among replication products. A and B, map of restriction sites and the strands they generate and the location of unique AAF adducts in the target sequence for leading and lagging strand adducts. Replication products were digested with PvuII and EcoRI to generate two fragments whose individual strands differ in length by four nucleotides. The exact length in nucleotides is noted above and below each strand, and the location of the AAF-modified guanines in the NarI site in the 124-nucleotide strand is indicated for each vector. C, PhosphorImager print of denaturing polyacrylamide gel containing resolved strands of PvuII-EcoRI fragments depicted in A and B. The length of each strand is indicated to the left. The sites of termination for each modified sample are indicated according to their position opposite the bases within the NarI sequence shown on the bottom left. The exact length of bands representing sites of termination for each substrate can be determined by comparison to a sequence ladder (not shown). out T antigen. These controls were not subsequently treated with DpnI, as this would destroy the unreplicated (fully methylated) molecules. DNA samples were introduced by electroporation into an E. coli ␣-complementation host, and cells were plated to score white colonies as well as blue colonies resulting from restoration of the correct reading frame in the lacZ reporter gene.
As noted previously, transfection of unreplicated G1 or G2 DNAs, or the same DNAs incubated with the extract in the absence of T antigen, yielded revertant frequencies only slightly above the background established with undamaged G0 DNA (Thomas et al., 1994). We have noted similar results with the G1 and G2 substrates in either orientation (Table III). Control values obtained for the G3 DNAs were higher than the other DNAs, ranging from 4.0 to 6.6 ϫ 10 Ϫ4 (Table III). These higher values are consistent with earlier studies demonstrating that the G3 adduct is mutagenic in E. coli (Burnouf et al., 1989;Veaute and Fuchs, 1993).
Only the products of replication of the G3 substrates yielded revertant frequencies above background levels (Table III). For the leading strand vector the frequencies were about 6-fold above the minus-T antigen control, indicating that most of the blue colonies reflected errors generated during replication in the HeLa cell extract. Revertant frequency values for the lagging strand vector were roughly 3-fold higher than with the leading strand vector and ϳ20-fold above background reversion frequencies. Sequence analysis of revertants from the G3 reaction products demonstrated the deletion of a GC dinucleotide from the NarI site.
The data showing preferential replication of the undamaged strand (Fig. 5C) led us to next examine whether multiple rounds of replication of the undamaged strand lead to an underestimate of AAF-dependent replication infidelity. To test this, we treated the G3 product DNA with MboI to digest second-round replication products. The effect on revertant frequencies was negligible with both G3-modified substrates (Table III, values in parentheses).
The appearance of BsaHI-resistant replication products consistent with lesion bypass (Fig. 4) suggests that AAF adduct blockage of replication (Fig. 5C) is not absolute. Therefore, we next examined whether longer incubation times influenced revertant frequencies. The effect on reversion frequencies for the substrate containing the lagging strand adduct were negligible (values of 94, 120, and 110 ϫ 10 Ϫ4 , respectively, for incubations of 60, 90, and 120 min). However, the reversion frequencies for the substrate containing the leading strand adduct were in-creased (values of 33, 51, and 77 ϫ 10 Ϫ4 , respectively, for incubations of 60, 90, and 120 min). Thus, the ϳ3-fold difference between leading and lagging strand values for a 1-h incubation was reduced to only a 1.4-fold difference after 2 h.
Influence of Proofreading on AAF-induced Replication Errors-We next attempted to determine whether the rate of AAF-induced deletion errors is affected by replication conditions that modulate proofreading. The contribution of proofreading to the fidelity of replication of undamaged DNA (reviewed by Roberts and Kunkel (1995)) has been assessed by increasing the dNTP concentration in the reaction. This selectively stimulates the rate of polymerization from a mispaired or misaligned intermediate, allowing less opportunity to edit the mistake. Alternatively, proofreading exonuclease activity can be inhibited by adding to the reaction a dNMP, the end product of exonuclease action. Both approaches were used in the present study for replication of the two G3-modified substrates (Table IV). With the leading strand substrate, the revertant frequency of replication products from reactions containing 1000 M dNTPs was 12-fold higher than from reactions containing 10 M dNTPs, while an intermediate revertant frequency was obtained for the 100 M reaction. A similar effect was seen for the lagging strand substrate, but the maximum difference was 4.4-fold. Note that in a 1-h reaction containing 1000 M dNTPs, the two substrates are copied with about equal fidelity (Table IV, compare 470 ϫ 10 Ϫ4 to 440 ϫ 10 Ϫ4 ). The addition of 2 mM dGMP to reactions containing 100 M dNTPs had little effect on reversion frequencies with either G3-modified substrate.

DISCUSSION
The present investigation is an extension of our earlier study (Thomas et al., 1994) of the efficiency and fidelity of replication a Pixel values (rounded to three significant figures) are corrected by subtracting a background value for a blank portion of each lane. 120 nt refers to the 120-base strand seen in Fig. 5C, Term represents the termination bands at the bottom of Fig. 5C, and Total is the sum of these values. 124 nt refers to the 124-base strand and is corrected for cytosine content in order to compare to the values for the 120 base and termination bands. The values for "120 nt" and "Term" were adjusted for DNA repair (14%) by decreasing (120 nt) or increasing (Term) by 14% of the 124-nt value. in human cell extracts of DNA containing unique AAF adducts located at defined template positions. The two main objectives here were to be as quantitative as possible and to compare the effects of adducts on replication by leading and lagging strand replication proteins. Thus, we first established conditions to obtain primarily one round of replication in vitro (1-h incubation, Fig. 2) and then determined the level of adduct repair activity in the extract (14%, Fig. 3) under these conditions. Subsequent analyses of the replication products (Figs. 4 and 5, Tables II-IV) then permit a quantitative estimate of the probability of replication termination and error-free and errorprone translesion synthesis.
The primary effect of an AAF adduct is termination of replication. The present study precisely maps the site of termination as occurring after incorporation opposite the template nucleotide immediately preceding the adduct, in all six sitespecifically modified substrates examined (Fig. 5C). We did not observe detectable termination at any upstream template positions or after incorporation opposite the adduct itself. Thus, any incorporation that does occur opposite the adduct likely has one of three fates: (i) correct extension, (ii) misalignment followed by extension (see below), or (iii) excision by a nuclease. The termination pattern observed during synthesis by the human replication apparatus is much simpler than that observed with the replicative DNA polymerase III holoenzyme of E. coli (Belguise-Valladier et al., 1994). That multisubunit enzyme complex terminates synthesis at several upstream locations preceding site-specifically AAF-adducted guanines. The pattern observed in Fig. 5C is also simpler than that observed with exonuclease-deficient DNA pol ␣ and pol ␤, which terminate synthesis both before and opposite AAF adducts (Rabkin and Strauss (1984), and references therein). Incorporation studies with AAF-adducted substrates using eukaryotic DNA polymerases having intrinsic 3Ј 3 5Ј exonuclease activities have not been reported.
The presence of BsaHI-resistant radiolabeled products (Fig.  4) suggests that AAF adducts are present in the replicated DNA and, thus, are not absolute blocks to replication. Quantitative analysis shows that approximately 17 to 24% of the products are BsaHI-resistant with all substrates examined. (Note that the ϳ1% mutagenesis for two-base deletions at the NarI site observed with the G3 vectors (Table III) should not contribute significantly to the population of BsaHI-resistant products). Since these products are derived from replication of both the damaged and undamaged strand, the efficiency of bypass per encounter with an AAF adduct in the damaged strand is 2-fold higher than this value, i.e. 34 -48%. Lesion bypass efficiencies estimated from quantitative analysis of fulllength restriction fragments versus termination bands (Fig.   5C) are also high, ranging from 16 to 27% (Table II). The data suggest that the human replication apparatus in a HeLa cell extract can bypass one-sixth to one-half of all AAF adducts encountered. This is remarkable given the evidence for high fidelity replication with undamaged substrates (Izuta et al. (1995), and references therein). It is also remarkable from a structural perspective. Although the structures of the major eukaryotic cellular replicative DNA polymerases are not yet known, some information does exist for Klenow DNA polymerase (Joyce and Steitz, 1994), human immunodeficiency virus type 1 reverse transcriptase (Kohlstaedt et al., 1992;Jacobo-Molina et al., 1993), T7 RNA polymerase (Sousa et al., 1993), and pol ␤ (Pelletier et al., 1994). These studies suggest numerous contacts between DNA polymerases and template-primers that might be severely perturbed by the presence of an adduct as bulky as the AAF adduct studied here. A bulky adduct might be expected to strongly reduce polymerase binding and/or translocation, as evidenced by the AAF-induced termination of synthesis observed with numerous DNA polymerases (e.g. see above). Thus, frequent bypass by the multiprotein replication complex implies a major role for replication accessory proteins in assisting lesion bypass. Logical candidates for modulating lesion bypass during eukaryotic chromosomal replication of damaged DNA are proteins that influence polymerase⅐ template-primer binding and/or translocation, such as proliferating cell nuclear antigen, which confers high processivity to pol ␦ (for review see Stillman (1994), and references therein).
The data in Table III demonstrate that replication of the G3-modified substrate is mutagenic, whereas replication of the other two modified substrates is not. This is consistent with a model wherein the replication apparatus first incorporates cytosine opposite the modified guanine (Lambert et al., 1992). The template-primer then slips such that the two terminal bases in the primer (3Ј-CpG-5Ј) hydrogen-bond with a repeated downstream complementary 5Ј-GpC-3Ј dinucleotide in the template, thus forming a misaligned intermediate (Schaaper et al., 1990). Continued synthesis from this intermediate leads to a two-base deletion error. This model predicts that mutagenesis should result from adducts in the G3 but not the G1 or G2 positions, because, given the sequence at the NarI site, only the former allows formation of a misaligned intermediate stabilized by two terminal base pairs. This prediction was realized in E. coli (Burnouf et al., 1989;Veaute and Fuchs, 1993), in our previous study of T antigen-dependent replication in a HeLa cell extract (Thomas et al., 1994) and in this study.
The revertant frequencies in Table III are for G3 adduct-dependent replication errors among total replication products generated in a single round of replication and can therefore be used to estimate quantitatively the probability of error-free versus error-prone bypass. Given the average ϳ3-fold bias in replication of the undamaged strand (Fig. 5C), a revertant frequency of 1% for the lagging strand G3-modified substrate (Table III) suggests that 3% of the products of replication of the damaged strand are mutagenic with this substrate in a 1-h incubation. A similar approach suggests that ϳ1% of bypass events is mutagenic in a 1-h reaction with the leading strand G3-modified substrate. These estimates are limited to errors that restore the correct reading frame, which in this and our previous study are two-base deletions.
Two asymmetries are observed during T antigen-dependent replication of AAF adduct-modified DNAs in the extract. These are a 3-fold difference in AAF adduct-induced replication infidelity with the two opposite-orientation G3-modified substrates (Table III) and preferential replication of the undamaged strand (Fig. 5C). The latter could reflect uncoupling of the first replication fork to encounter the adduct, such that repli- cation of the undamaged strand continues while replication of the damaged strand is completely or transiently blocked. Evidence exists that uncoupling can occur under some circumstances in E. coli (Koffel-Schwartz et al., 1987) 3 and during SV40 replication in vivo (Burhans et al., 1991). Alternatively, the first fork may not replicate either strand beyond the lesion, with preferential replication of the undamaged strand accomplished by the fork arriving from the other direction in the circular substrate. By this same logic, the 3-fold difference in revertant frequencies can be explained in more than one way. If bypass is catalyzed by the first fork to encounter the adduct, then the difference suggests that the leading strand replication apparatus is 3-fold more accurate than the lagging strand apparatus. Alternatively, error-prone lesion bypass replication may be accomplished only by the fork arriving from the other direction. For example, it is formally possible that much or all of the error-prone lesion bypass synthesis with both G3-modified substrates is catalyzed by the lagging strand replication apparatus. Consistent with this idea, mutagenesis with the leading strand G3 vector increases with increasing incubation time, while that with the lagging strand substrate remains about the same. Given two replication forks and use of a circular substrate, several explanations are consistent with the existing data. It is also possible that the lesion bypass synthesis in the extract is catalyzed by proteins other than those normally comprising a replication fork. It may be possible to resolve these issues in the future using reactions reconstituted from purified proteins or using much larger replicons or substrates containing a replication termination signal. Several observations suggest that the misaligned intermediate believed to be responsible for the AAF-dependent deletion errors might be edited by proofreading during replication. First, the evidence presented here for substantial bypass implies that incorporation opposite an AAF adduct is possible, but the bands in Fig. 5C indicate termination of incorporation one base before the adduct but not opposite the adduct itself. This may be interpreted in light of the observation that proofreading-proficient Klenow polymerase generated a termination pattern similar to that seen here, while the exonuclease-deficient polymerase terminated synthesis after incorporation opposite the adduct (Belguise-Valladier et al., 1994). This indicates that a proofreading exonuclease can remove incorporations opposite the adduct. Moreover, the presumed premutational intermediate in this study contains two terminal base pairs and an unpaired modified GpC dinucleotide in the template strand. Several studies suggest that undamaged misaligned intermediates formed during synthesis by proofreading-proficient DNA polymerases Thomas et al., 1991; and during T antigendependent replication in extracts  are subject to editing by exonucleases. In this study, the rate of AAF-induced deletion errors was substantially increased by increasing the dNTP concentration in the reaction (Table IV). This observation is consistent with selective stimulation of polymerization from a misaligned intermediate, allowing less opportunity to proofread the error. The fact that the dNTP-dependent increase was greater with the substrate containing the adduct in the leading strand relative to the first fork to encounter the adduct is consistent with more active proofreading by the leading strand replication apparatus. As discussed above, other explanations are also possible. The second approach to modulate deletion error rates by modulating proofreading, inhibition of exonuclease activity by adding dGMP to the reac-tion, yielded very small effects (Table IV). Lack of a monophosphate effect on frameshift fidelity during T antigen-dependent replication was also observed with undamaged substrates in an earlier study . The reason for this is unclear, especially since addition of dGMP to a replication reaction clearly reduces base substitution fidelity with normal substrates Izuta et al., 1995) and with a dNTP analog (Minnick et al., 1995).
The estimates of termination and error-free and error-prone bypass presented here are for replication in a HeLa cell extract. It will be interesting to obtain similar estimates using extracts of normal human fibroblasts (Boyer et al., 1993) or cell lines having defects in mismatch repair or nucleotide excision repair.