Discrimination between Translesion Synthesis and Template Switching during Bypass Replication of Thymine Dimers in Duplex DNA*

The goal of this study was to determine whether bypass replication occurs by translesion synthesis or template switching (copy choice) when a duplex molecule carrying a singlecis,syn-cyclobutane thymine dimer is replicated in vitro by human cell extracts. Circular heteroduplex DNA molecules were constructed to contain the SV40 origin of replication and a mismatch opposite to or nearby the dimer. Control molecules with only the mismatch were also prepared. Heteroduplexes were methylated at CpG islands and replicated in vitro (30 min). Following bisulfite treatment, the nascent DNA complementary to the dimer-containing template was distinguished from the other three strands by methylation-specific polymerase chain reaction. Cloning and sequencing of polymerase chain reaction products revealed that 80–98% carried the sequence predicted for translesion synthesis, with two adenines incorporated opposite the dimer. The fraction of clones with sequence predictive of template switching was reduced when extracts deficient in mismatch repair or nucleotide excision repair activities were used to replicate the heteroduplex molecules. These results support the conclusion that lesion bypass during in vitroreplication of duplex DNA containing thymine dimers occurs by translesion synthesis.

The goal of this study was to determine whether bypass replication occurs by translesion synthesis or template switching (copy choice) when a duplex molecule carrying a single cis,syn-cyclobutane thymine dimer is replicated in vitro by human cell extracts. Circular heteroduplex DNA molecules were constructed to contain the SV40 origin of replication and a mismatch opposite to or nearby the dimer. Control molecules with only the mismatch were also prepared. Heteroduplexes were methylated at CpG islands and replicated in vitro (30 min). Following bisulfite treatment, the nascent DNA complementary to the dimer-containing template was distinguished from the other three strands by methylation-specific polymerase chain reaction. Cloning and sequencing of polymerase chain reaction products revealed that 80 -98% carried the sequence predicted for translesion synthesis, with two adenines incorporated opposite the dimer. The fraction of clones with sequence predictive of template switching was reduced when extracts deficient in mismatch repair or nucleotide excision repair activities were used to replicate the heteroduplex molecules. These results support the conclusion that lesion bypass during in vitro replication of duplex DNA containing thymine dimers occurs by translesion synthesis.
Genomes are constantly damaged through thermal, chemical, and radiation-induced reactions that cause modifications in DNA structure. These can lead to mutations during replication of the damaged DNA. A network of DNA repair pathways and checkpoint responses minimize this risk. Both prokaryotes and eukaryotes are also endowed with DNA damage tolerance pathways that increase survival by facilitating the duplication of the genome, even in the presence of unrepaired lesions. These pathways include lesion bypass during semi-conservative DNA synthesis by mechanisms that are not yet completely understood (1,2).
In the past 3 years, novel DNA polymerases have been identified and characterized in bacteria, yeast, and humans. These new enzymes have in common the property of extending primers beyond a template DNA lesion during replication in vitro.
In Escherichia coli, they include the product of the dinB gene, namely DNA polymerase (pol) 1 IV (3), and UmuDЈ 2 C (pol V) (4). Saccharomyces cerevisiae contains two or three specialized DNA polymerases encoded by the REV3, REV7, REV1, and RAD30 genes. Rev3p and Rev7p are subunits of pol (5). This DNA polymerase together with Rev1p, a primer-template-dependent deoxycytidyltransferase (6), appear to be responsible for most of the mutagenic events in S. cerevisiae (7). By contrast, the product of the RAD30 gene (pol ) was shown to bypass cyclobutane thymine dimers efficiently by inserting adenines opposite this photoproduct (8).
The recent cloning of the human homologs of these new DNA polymerases has drawn the suggestion that the process of bypass replication of DNA lesions share common mechanisms in prokaryotes and eukaryotes (9,10). Accordingly, Gibbs and collaborators (11,12) have shown that the expression of hREV3 or hREV1 antisense RNAs in human fibroblasts reduces UVinduced mutagenesis. Loss of pol (hRad30A) leads to the xeroderma pigmentosum variant (XP-V) syndrome (13)(14)(15). XP-V patients are prone to skin cancer in areas exposed to the sun (16), and their fibroblasts in culture are hypermutable by UV light (17)(18)(19)(20)(21). This indicates that pol in vivo protects human cells from UV-induced mutagenesis by supporting the bypass of UV-induced lesions in much the same way as Rad30p does in yeast (8,22). Also cloned recently were the human homolog of E. coli dinB (23), encoding pol (24), and another homolog of S. cerevisiae RAD30, termed hRAD30B (25). The latter is distinct from pol (hRAD30A) but appears to encode also a bypass DNA polymerase, pol (26). Whether these novel DNA polymerases contribute to the bypass of different DNA lesions is still under investigation.
The preferred model of lesion bypass during DNA replication calls for a step of translesion synthesis catalyzed by one of the novel DNA polymerases described above (9,10). One characteristic of these new enzymes is that they synthesize DNA with low processivity (27) and low fidelity (27,29). This suggests that the bypass polymerases might replace the replicative polymerase blocked at the damage site, catalyze the addition of a few nucleotides across the lesion (their active sites are thought to be more tolerant of DNA distortions), and then dissociate from the primer-template (27,28). At this point, the main replicase would return to the DNA growing point for efficient (processive and faithful) duplication of the undamaged DNA (9,10,27,28). An alternative model of lesion bypass calls for the melting from the damaged template of the 3Ј end of the blocked strand and re-annealing to the complementary nascent DNA strand (copy choice). This template-switching model was first proposed by Higgins et al. (30) and Fujiwara and Tatsumi (31) to explain bypass replication of UV-induced lesions in irradiated human cells. Such a damage-avoidance mechanism has been proposed to operate in bacteria (32) and yeast (33). There is strong evidence that bypass polymerases, such as pol (5) and pol (8,13,14,28,34), carry out translesion synthesis in primer-extension assays in vitro. However, the possibility that template switching might occur during the replication of a duplex DNA, when the synthesis of both leading and lagging strands takes place, has not been formally excluded.
The template-switching model did not receive much attention until recent years because it was assumed that interruption of leading strand synthesis by a template lesion would result in complete blockage of the replication fork (35,36). Studies using in vitro replication of duplex DNA, however, has provided evidence that uncoupling of leading and lagging strand synthesis occurs when leading strand synthesis is blocked by a template lesion (37)(38)(39)(40). Under these conditions, the replication fork continues to move beyond the damaged site, extending the lagging strand by 1 to 2 thousand nucleotides beyond the lesion (40,41). These findings, as well as evidence that extended single-stranded DNA regions are also formed during DNA replication in UV-irradiated mammalian cells (42,43), suggested that a template-switching mechanism was indeed quite plausible. The appeal of such a lesion bypass pathway was its potential for high fidelity and for being independent of the nature of the blocking lesion.
In this study we designed and carried out experiments to determine whether translesion synthesis or template-strand switching is the primary mechanism by which a site-specific thymine dimer is bypassed during in vitro replication of a duplex DNA. Heteroduplex circular molecules with or without a single cis,syn-thymine dimer [T ∧ T] on the template to the leading strand and a mismatch on the complementary template were fully methylated at CpG islands and replicated in vitro by several bypass-proficient human cell extracts. Leading strand was successfully distinguished from the lagging strand and methylated template strands by methylation-specific PCR, after the DNA was treated with bisulfite. Sequencing of cloned PCR products revealed that translesion synthesis, and not template switching, was the major mechanism by which cis, syn-thymine dimers were bypassed during in vitro replication.

EXPERIMENTAL PROCEDURES
Cell Lines, Culture Conditions, and Preparation of Cell-free Extracts-HeLa S3 cells were obtained from the Lineberger Comprehensive Cancer Center Tissue Culture Facility (University of North Carolina, Chapel Hill) or from the National Cell Culture Center (Minneapolis, MN). The IDH4 cell line was a gift from Dr. Jerry Shay (University of Texas Southwestern Medical Center). This cell line was generated by transformation with SV40 large T antigen of fibroblasts from an apparently normal human fetus (44). SV40-transformed XPA cells (XP12BE) were obtained from the NIGMS Human Genetic Mutant Cell Repository (GM4429). The human colorectal cancer cell line HCT116 (hMLH1 mutant, MMR defective; see Refs. 45 and 46) was obtained from Dr. Thomas A. Kunkel from the NIEHS, National Institutes of Health. Cells were grown in monolayer cultures at 37°C in an atmosphere of 5% CO 2 in air. HCT116 cells were maintained in a 1:1 mixture of Dulbecco's modified Eagle's medium and Ham's F-12 medium (Life Technologies, Inc.) with 10% fetal bovine serum (HyClone Laboratories). The other cell lines were grown in Eagle's minimal essential medium, supplemented with 2 mM L-glutamine and 10% fetal bovine serum. Growth medium for IDH4 cells also contained 1 M dexamethasone (Sigma). Replication-competent cell extracts for DNA replication in vitro were prepared as published (47,48).
Construction of Circular Heteroduplexes Molecules Containing a Mismatch and a Single T ∧ T-The base changes required for the formation of the mismatch in the final heteroduplex ( Fig. 1) were introduced in the sequence of the ϩ strand of M13LeaSV (40). Mutated ϩ strands for the construction of Lea-C and T ∧ T-C were prepared essentially as described (40), except that the 20-mer inserted into M13mp2SV (49) by site-directed mutagenesis (50) was changed by a single base (5Ј-GAGCTCACT-TAGTCAGCTGC-3Ј). The mutated ϩ strand in Lea-munI and T ∧ T-munI were produced with the QuickChange TM site-directed mutagenesis kit (Stratagene). The 35-bp oligonucleotides 5Ј-CCGGCTCGAG-CTCAATTGGTCAGCTGCGTATGTTG-3Ј and 5Ј-CAACATACGCAGC-TGACCAATTGAGCTCGAGCCGG-3Ј containing a point mutation at position 389 (bold and underlined) were used to introduce a unique MunI site into M13LeaSV. Mutated products were verified by sequencing of the ϩ strands. Mutated closed circular single-stranded DNA was purified and annealed with the oligonucleotide 3Ј-CTCGAGT ∧ TAATC-AGTCGACG-5Ј previously phosphorylated at the 5Ј end using T4 polynucleotide kinase (40). Then, second-strand synthesis, ligation, and purification in CsCl gradients were carried out according to published procedure (51). Control heteroduplex molecules (Lea-munI and Lea-C) were prepared in the same manner, using an oligonucleotide of identical sequence, but without the thymine dimer.
In Vitro DNA Replication-Heteroduplex DNA molecules were incubated with the bacterial SssI methylase (New England Biolabs) for 2-3 h at 37°C, under conditions suggested by the enzyme supplier for methylation of CpG sites. Complete methylation was confirmed by resistance of the closed circular heteroduplex molecules to digestion by AciI (New England Biolabs), a restriction enzyme that cuts only unmethylated DNA.
In vitro replication of the methylated heteroduplexes was carried out in 25-l reactions, as described previously (40), except that the concentration of dCTP was reduced to 50 M. Reaction mixtures lacking SV40 large T antigen were used as negative controls of in vitro DNA replication. After incubation at 37°C for 30 min, reactions were terminated by adding an equal volume of stop solution containing 2% SDS, 2 mg/ml proteinase K, and 50 mM EDTA. DNA was purified by using the QIAEX II Gel Extraction System (Qiagen). In experiments with T ∧ T-munI and Lea-munI, the purified replication products were digested with MunI (Roche Molecular Biochemicals) for 2-3 h at 37°C. Aliquots of purified DNA samples (ϳ10 ng) were fractionated in 1% agarose gels containing 0.2 g/ml ethidium bromide. Dried gels were exposed to a phosphor screen that was later scanned by a PhosphorImager TM (Molecular Dynamics, Sunnyvale, CA).
Bisulfite Treatment-Several protocols for bisulfite conversion of cytosine to uracil in single-stranded DNA have been described (52)(53)(54)(55)(56). In order to achieve efficient bisulfite conversion, the purified DNA was first digested with XmnI restriction endonuclease (New England Biolabs) for 2-3 h at 37°C. There are two XmnI recognition sequences in the heteroduplex DNA used in these experiments (Fig. 1). The restricted DNA (18 l, ϳ30 ng) was denatured by incubation in 0.3 M NaOH (addition of 2 l of freshly prepared 3 M NaOH) for 15 min at 37°C. Then, 200 l of freshly prepared 5 M bisulfite solution, pH 5.0 (a mixture of 2.5 M metabisulfite and 500 mM hydroquinone; Sigma), was added to each reaction tube, followed by an incubation of 4 h at 50°C. The bisulfite-treated DNA was desalted by using the Wizard® DNA clean-up system (Promega) and desulfonated by addition of 3 M NaOH to a final concentration of 0.3 M. After 15 min at 37°C, the solution was neutralized by addition of 7.5 M NH 4 Oac, pH 7.0, to a final concentration of 3 M. DNA was precipitated with ethanol and resuspended in 12 l of 10 mM Tris, containing 1 mM EDTA, pH 8.0, and used immediately or stored at Ϫ20°C.
PCR Amplification of the Bisulfite-treated DNA-The list of primers used in this study is shown in Table I. Primers were designed for both 5Ј33Ј and 3Ј35Ј strands. Strand-specific primers for the selective amplification of the replicated DNA (UNMET) or the methylated template (MET), nonselective primers, and the sequencing primer were designed using the GCG sequence analysis software package of the Wisconsin Genetics Computer Group (version 7). Primers were synthesized at the Pathology Oligonucleotide Synthesis Facility (University of North Carolina, Chapel Hill).
PCR was carried out in a Touchdown TM Thermal Cycler with a hot-start step that improved specificity. Reaction mixtures (50 l) containing 3 l of bisulfite-treated DNA, 10 pmol of each PCR primer, 10 nM of each dNTP, and buffer supplied by Qiagen were placed in the thermal cycler and hot-started at 94°C for 4 min. After addition of 2.5 units of Taq polymerase (Qiagen) the denaturation step was continued for 2 min, followed by the touchdown protocol (denaturation at 94°C for 30 s, annealing at X°C for 30 s, extension at 72°C for 30 s). The initial annealing temperature of 61°C was decreased at the rate of 1°C for every PCR cycle until the targeted temperature was reached (optimal annealing temperature for the primers used; see Table I). At the target temperature, 19 -20 regular PCR cycles were performed, followed by a final extension step of 10 min at 72°C. Negative controls for the strandspecific primers were performed with each PCR set. Aliquots (2 l) of the PCRs were mixed with Ficoll loading buffer (3% Ficoll 400, 25 mM EDTA, 0.025% Orange G) and subjected to electrophoresis in 1.2% agarose gel containing 0.2 g/ml ethidium bromide.
For direct sequencing, PCR products were purified using QIAquick TM PCR product purification system (Qiagen). PCR products obtained from 5Ј33Ј and 3Ј35Ј strands of template, and replicated DNA was directly sequenced using one of the PCR primers. PCR products obtained from the leading strand of replicated DNA were subcloned using TA Cloning Kit (Invitrogen). Plasmid DNA from individual clones were isolated using WIZARD PLUS Minipreps Kit (Promega), checked for the presence of an insertion, and sequenced using the sequencing primer (Table  I). DNA sequencing was done at the University of North Carolina, Chapel Hill, Automated DNA Sequencing Facility on a model 377 DNA Sequencer (Perkin-Elmer) using the ABI PRISM TM Dye Terminator Cycle Sequencing Ready Reaction Kit with AmpliTaq DNA polymerase. The sequence data were analyzed using the GCG sequence analysis software package of the Wisconsin Genetics Computer Group (version 7).

RESULTS
Substrate Design-Two sets of circular heteroduplex DNA molecules were prepared for in vitro DNA replication (T ∧ T-munI and T ∧ T-C), each containing a control molecule lacking the T ∧ T dimer (Lea-munI and Lea-C, respectively). In the T ∧ T-munI and T ∧ T-C substrates, the dimer was placed in the template to the leading strand (for the first replication fork encountering the lesion), 385 bp from the center of the SV40 origin of replication (Fig. 1). The two constructs differed only in the location of the mismatched bases. In T ∧ T-munI, a T:G mismatch was at position 389, 3 bp away from the T ∧ T dimer (3Ј-T ∧ TAAT-5Ј:5Ј-AATTG-3Ј). This mismatch created a unique MunI site in the replication product of the undamaged strand, but the replication product of the damaged strand remained resistant to MunI. The second heteroduplex (T ∧ T-C) contained a T:C mismatch opposite the T ∧ T dimer (3Ј-T ∧ TAA-5Ј:5Ј-ACTT-3Ј) at position 386. The single mismatch opposite (T ∧ T-C) or nearby (T ∧ T-munI) the T ∧ T dimer was used as the marker of the undamaged strand to discriminate whether DNA replication past the dimer occurred by translesion synthesis or template switching. The diagram in Fig. 2 illustrates how these two mechanisms could be distinguished by sequencing the 5Ј33Ј leading strand of nascent DNA synthesized from T ∧ T-munI. If translesion synthesis were the major mechanism by which the replication machinery bypassed the lesion, the expected sequence of the leading strand (complementary to the T ∧ T-containing template) would be AATTA (AATT with the T ∧ T-C construct). In case of template strand switching (replication across the lesion was avoided by using the undamaged lagging strand as template), the leading strand molecules would carry the information from the marked undamaged strand and display the sequence AATTG (ACTT with the T ∧ T-C construct).
In Vitro Replication and Modification of DNA Strands-We chose to carry out the in vitro replication reactions for 30 min to maximize lesion bypass but to minimize the probability of a second round of replication (38,40,41,57). Circular heteroduplexes were first treated in vitro with SssI, an enzyme that methylates cytosines exclusively at CpG dinucleotides in double-stranded DNA (58). Resistance to digestion by AciI confirmed that the heteroduplex DNA was efficiently methylated on both strands (there are 47 AciI recognition sites, 5Ј-CCGC-3Ј, in the 7.4-kilobase pair circular molecule). Next, we determined that the methylated heteroduplexes could be replicated in vitro by human cell-free extracts. SV40 large T antigen-dependent replication of the fully methylated heteroduplexes represented 25-75% (varying with the extract and substrate used) of that obtained in parallel reactions containing unmethylated DNA (results not shown).  (40), was modified to contain a single T ∧ T in the Ϫ strand and a mismatched base on the ϩ strand. The location of the T ∧ T relative to the SV40 origin of replication placed the dimer on the template to the leading strand for the first replication fork to encounter the lesion. In T ∧ T-C, a mismatched C was placed opposite the 5Ј-T of the dimer. In T ∧ T-munI, a G:T mismatch was placed 3 bp downstream from the dimer to create a MunI recognition site upon replication of the undamaged strand.

FIG. 2. Predicted intermediates of bypass replication of a T ∧ T blocking leading strand synthesis.
Uncoupling of lagging strand synthesis, displacement of the replication fork beyond the dimer, and formation of an extended single-stranded DNA region have been previously documented (41). This study sought to determine whether such structures could represent intermediates of bypass replication by template switching via the hypothetical configurations shown inside the boxed inset. The mismatched base on the DNA strand complementary to that carrying the T ∧ T provided the sequence marker needed to determine whether extension of the leading strand beyond the dimer (bypass replication) occurred primarily by translesion synthesis or template switching. DNA sequences surrounding the dimer in the T ∧ T-munI construct are shown in bold (template DNA) and in italics (daughter DNA) to illustrate the predicted sequence of the 5Ј33Ј leading strand following bypass replication according to one or the other of the two potential models. The completion of a MunI recognition site upon replication of the 5Ј33Ј template strand containing the mismatched G should allow for the digestion of the resulting duplex prior to amplification and sequencing of the nascent leading strand.
Genomic sequencing after bisulfite modification was developed by Frommer and collaborators (52,53). The method is based on sodium bisulfite-mediated conversion of unmethylated cytosine to uracil under conditions whereby 5-methylcytosines are resistant to this conversion. We used this approach to distinguish the products of DNA replication in vitro (3Ј35Ј and 5Ј33Ј newly synthesized strands) from each other and from their template strands. Single round of replication of double-stranded templates fully methylated at CpG islands yields two product molecules each containing one methylated parental strand and one unmethylated newly synthesized strand (Fig. 3A). Bisulfite reaction converts all unmethylated cytosines to uracils, whether they are in the newly synthesized or the template strands. There are two advantages of methylation and bisulfite modification of in vitro replicated DNA. After treatment, the duplex DNA strands are no longer complementary (Fig. 3B), and methylated template and unmethylated nascent DNA strands of identical sequence and polarity become divergent at CpG islands (Fig. 3C). Thus, bisulfitetreated products of replication of duplex DNA methylated at CpG islands become a mixture of four different strands that can be selectively amplified by PCR with strand-specific primers.
Primer Design-PCR primers were designed as described by Herman et al. (59). They are listed in Table I, and their relative hybridization positions on the target DNA strands are shown in Fig. 3D. These primers were able to amplify efficiently both the bisulfite-treated methylated and unmethylated strands of M13leaSV (the parent plasmid used to generate heteroduplexes). A nonselective primer pair was also designed to check the methylation status of the heteroduplex template strands after the incubation with human cell extracts. This was necessary to rule out possible contamination of leading strand DNA with template strands demethylated during the in vitro replication reaction. In control experiments, methylated T ∧ T-munI DNA was incubated for 30 min with either HeLa or IDH4 extracts in the absence of SV40 large T antigen (no semiconservative DNA replication). After bisulfite treatment, the 5Ј33Ј strands were PCR-amplified using nonselective primers, and the PCR products were purified and sequenced directly. Results (not shown) revealed that methylated molecules retained the CpG islands after incubation with human cell extracts. Sequencing histograms showed that each C peak at CpG islands contained only one signal. No evidence was found for a secondary (shadow) peak representing demethylated cytosines (i.e. uracils) at positions corresponding to CpG islands. These results clearly indicated that under conditions used for in vitro replication the template DNA molecules remained methylated and should not interfere with results obtained during the sequencing of the unmethylated newly synthesized DNA.
Reconstruction experiments were used to estimate the specificity of primers to amplify only the targeted strand among the four expected to be present in the in vitro replication mixture (Fig. 4). Use of touchdown PCR with a hot start improved the specificity of primers, and no PCR products were detected in the reactions containing negative controls. Unmethylated M13leaSV was premixed with 10ϫ the mass of methylated M13leaSV. This mixture was treated with bisulfite, and aliquots were subjected to PCR using the 5Ј-3Ј-UNMET primer set (Fig. 4A). Separate PCRs containing either bisulfite-treated unmethylated DNA or 10ϫ the mass of bisulfite-treated methylated DNA were used as controls. Note that only the bisulfitetreated, unmethylated DNA was amplified with the 5Ј-3Ј-UN-MET primer set. Fig. 4B illustrates the selective amplification of the bisulfite-treated methylated DNA with the 5Ј-3Ј-MET primer set. In this case, the methylated DNA that was treated with bisulfite was mixed with 10ϫ the mass of the untreated methylated DNA. Only the bisulfite-treated DNA was amplified with the 5Ј-3Ј-MET primer set.
Next, we examined the specificity of the 5Ј-3Ј-UNMET primers to amplify bisulfite-treated DNA from an in vitro replication mixture containing the IDH4 extract and methylated Lea-C as the substrate (Fig. 4C). DNA from in vitro replication reactions in which methylated Lea-C was incubated with the same extract in the absence of SV40 large T antigen (no replication) was used as an additional negative control. Efficient PCR amplification with the 5Ј-3Ј-UNMET primers was only observed with DNA from in vitro replication reactions when the addition of SV40 large T antigen promoted the replication of the methylated Lea-C substrate. Direct sequencing of PCR products obtained with 5Ј-3Ј-UNMET and 3Ј-5Ј-UNMET primer pairs revealed that all CpG dinucleotides were converted into UpGs during bisulfite treatment and then PCRamplified as TpGs. Direct sequencing also revealed that the 5Ј33Ј leading strand synthesized from the region of interest (carrying the mismatch 3Ј-TTAA-5Ј:5Ј-ACTT-3Ј) was 5Ј-AATT-3Ј, whereas the 3Ј35Ј lagging strand was 3Ј-TGAA-5Ј (underlining added for emphasis). Both template strands were amplified from the same in vitro replication reaction using 5Ј-3Ј-  Table I. Dotted lines represent DNA synthesized during PCR.
MET and 3Ј-5Ј-MET primer sets and sequenced directly. Results showed that all CpG dinucleotides were protected by methylation in both template strands during bisulfite reaction. The 5Ј-ACTT-3Ј template strand (mismatch-containing region) was amplified during PCR as 5Ј-ATTT-3Ј (unmethylated cytosine was converted to U and then to T) and 3Ј-TTAA-5Ј template strand was amplified as 3Ј-TTAA-5Ј.

Sequence Analysis of in Vitro Replication Products of Undamaged Heteroduplexes-Methylated
Lea-munI (3Ј-TTAAT-5Ј:5Ј-AATTG-3Ј) was replicated in vitro by HeLa and IDH4 extracts and treated with bisulfite after XmnI and MunI digestion. As emphasized above, the mismatch in this construction leads to the introduction of a recognition site for MunI in the replication product of the 5Ј33Ј template strand. After complete digestion with MunI, neither the 3Ј35Ј lagging nascent strand nor the 5Ј33Ј template strand in replicated molecules should be available for PCR amplification. After bisulfite treatment, the 5Ј33Ј leading nascent strand from three independent in vitro replication reactions with methylated Lea-munI was amplified with the 5Ј-3Ј-UNMET primers and directly sequenced. This analysis confirmed that the targeted replicated strands were amplified because ϳ90 -100% of the CpG dinucleotides in the amplified DNA was converted to TpGs. For sequencing analysis of individual molecules, PCR products were obtained from four independent in vitro replication reactions conducted with one of two human cell extracts (IDH4 and HeLa) and cloned. Approximately 20 -25 cloned molecules from each reaction were sequenced, and the results are shown in Table II. The data revealed that 91 (HeLa, n ϭ 45) and 98% (IDH4, n ϭ 46) of sequenced molecules displayed the expected sequence AATTA. Similar results were obtained when the second heteroduplex, Lea-C (3Ј-TTAA-5Ј:5Ј-ACTT-3Ј), was replicated by the IDH4 extract. Among the 18 sequenced clones, 17 (94%) displayed the expected sequence AATT and only one (6%) showed the ACTT sequence (Table IV). We were surprised to find leading strand clones carrying the sequence ACTT (Lea-C; 6% in IDH4, Table IV) or the sequence AATTG (Lea-munI; 9% in HeLa and 2% in IDH4, Table II) among products of replication of the undamaged constructs. Therefore, we consider whether these clones could reflect the activity of mismatch repair (MMR) on the heteroduplex prior to DNA replication (60). The experiments were repeated with an extract of HCT116, an MMR-deficient cell line, and the results showed that all 49 sequenced clones (two independent in vitro replication reactions) carried the sequence AATTA (Table II).
Sequence Analysis of in Vitro Replication Products of T ∧ T Dimer-containing Heteroduplexes-T ∧ TmunI heteroduplex molecules were incubated in vitro with bypass-proficient extracts from HeLa and IDH4. Newly synthesized leading strand molecules were analyzed as described for the corresponding undamaged heteroduplex molecules. PCR products obtained from three independent in vitro replications were sequenced directly. Sequence histograms showed that the peak corresponding to the 3Ј-A at the 5Ј-AATTA-3Ј region contained one strong A signal and a smaller G peak, indicating that the population of amplified molecules was heterogeneous in this region. After cloning of PCR products and sequencing of individual molecules, the data summarized in Table III were obtained. Among the molecules synthesized by HeLa (n ϭ 40) and IDH4 (n ϭ 46), 80 and 87% displayed the sequence AATTA, which was consistent with bypass replication taking place by translesion synthesis (Fig. 2). Clones carrying the sequence AATTG constituted 20% in HeLa and 13% in IDH4. According to our experimental design, the detection of AATTG sequences among the bypass products would indicate that direct damage bypass could be avoided by a template-switching mechanism. Before reaching the conclusion that such pathway was functioning during the replication of a small fraction of heteroduplex molecules, it was important to determine whether DNA repair activities could be distorting the experimental results. Both IDH4 and HeLa extracts are proficient in nucleotide excision repair (NER) and MMR. The heteroduplexes used in this study carry a compound lesion (a mismatch opposite the  Table I Figs. 1 and 3D) Underlined nucleotides represent positions protected from deamination by SssI methylation of template DNA but modified by bisulfite treatment in replicated DNA; sequence differences between primers and DNA not treated with bisulfite are in boldface. See Fig. 3D for schematic representation of the relative positions and directions of these primers.

CCAACTAACAAAAAAAAAATATACTACAAAACA
T ∧ T dimer), and such lesions are better substrates for NER than the simple T ∧ T (61). If some molecules of T ∧ T-munI (3Ј-T ∧ TAAT-5Ј:5Ј-AATTG-3Ј) were repaired by NER to 3Ј-TTAAC-5Ј:5Ј-AATTG-3Ј, the replication of the repaired 3Ј-TTAAC-5Ј strand would generate leading strand molecules with the sequence AATTG. If the repaired and replicated molecules escaped MunI digestion, then AATTG clones could be recovered in our experiments without template switching having occurred during bypass replication. We also considered that if AATTG clones resulted from bypass replication by template switching, but post-replication repair corrected the mismatch by copying the template strand (3Ј-T ∧ TAAT-5Ј:5Ј-AATTG-3Ј repaired to 3Ј-T ∧ TAAT-5Ј:5Ј-AATTA-3Ј), then our results could be distorted in favor of translesion synthesis. If that was the case, the percentage of AATTG clones should increase when MMR-deficient extracts were used to replicate the heteroduplex molecules. Therefore, these experiments were repeated with extracts from NER-deficient (XPA) and MMR-deficient (HCT116) cells. The data in Table III show that 93% of the leading strand molecules displayed the sequence AATTA and only 7% AATTG when in vitro replication was supported by the HCT116 extract. Therefore, a reduction and not an increase in the number of AATTG clones was observed in the absence of MMR activity. This observation strengthened the suspicion that prereplicative NER (and not template switching) could be the major source of the observed AATTG clones. This was confirmed by the results with an extract from XPA cells (NERdeficient). Among the 45 clones sequenced, only one (2%) displayed the sequence AATTG (Table III).
The results presented in Tables II and III strongly suggest that bypass replication of a single cis,syn-thymine dimer in duplex DNA occurs primarily by translesion synthesis. This conclusion was confirmed by the results obtained with the T ∧ T-C construct (3Ј-T ∧ TAA-5Ј:5Ј-ACTT-3Ј) and shown in Table  IV. After in vitro replication by IDH4 and HCT116 extracts, the sequence of the leading strand was found to be AATT in 84% (IDH4, n ϭ 56) and 82% (HCT116, n ϭ 39) of the clones. DISCUSSION The significance of the data reported here rests on the demonstration that human replication complexes carry out bypass replication of a pyrimidine dimer in duplex DNA by translesion synthesis. Although this conclusion may seem obvious to many, this is the first analytical study designed to test this assertion experimentally. Recent discoveries of several novel DNA polymerases with bypass replication capability in primer extension reactions reinforced the expectation that polymerization across non-instructive lesions in duplex DNA is the norm. In addition, the in vitro replication of single-stranded vectors containing a single acetylaminofluorene adduct demonstrated that translesion synthesis through and beyond the damaged site can be catalyzed by extracts from bypass-proficient human cells (34). However, other pieces of evidence kept pointing to an alternative mechanism, mainly template switching, as a plausible pathway of damage tolerance when lesions are encountered during the replication of duplex DNA.
The potential co-existence in the same replication fork of a newly synthesized daughter DNA, with the same sequence and polarity as the lesion-containing template, was thought to hold the key to the mechanism by which DNA polymerization could be guided beyond the blocking lesion (Fig. 2). When this model was initially proposed (30,31), however, it was met with skepticism. One of the experimental evidence presented in support of the model was the detection by electron microscopy of fourarmed replication forks (see boxed inset in Fig. 2). These structures, however, could have been formed in solution by DNA branch migration instead of representing true intermediates of bypass replication. In the absence of conclusive proof in favor of or against template switching, this model remained appealing because it envisioned an error-free alternative for the completion of DNA replication, despite the presence of DNA damage in the genome. In vitro replication studies aimed at detecting replication past site-specific DNA lesions demonstrated that blockage of the leading strand at a damaged site did not interrupt the progression of the replication fork (37)(38)(39)(40). In a large fraction of the replicating molecules, the attendant DNA synthesis resulted in polymerization of the lagging strand beyond the lesion, even before bypass could take place (41). These observations made clear that the necessary prerequisite for template switching, i.e. an undamaged 3Ј35Ј nascent strand to serve as the alternative template to the growing leading  strand, was indeed present at the damaged site. Therefore, it became important to design experiments to detect whether bypass replication by template switching was occurring in vitro.
The first experimental challenge was to mark the template strands, so that we could determine which one directed the synthesis of the DNA later found opposite the T ∧ T. This was the purpose of adding a mismatched base opposite to or nearby the dimer in the two constructs used in this study (Figs. 1 and 2). Once bypass had taken place, we needed to determine without ambiguity the sequence of the newly synthesized leading strand. By using technology developed to study DNA methylation patterns, it was possible to distinguish the four strands of the duplex DNA and to amplify them selectively. The principles of this approach are illustrated in Fig. 3. Methylation protection of bisulfite deamination of cytosines, in stretches of the template strands containing at least three CpG islands, provided sufficient sequence divergence from the newly synthesized DNA, which prior to chemical treatment had an identical sequence and polarity. The positive and negative controls illustrated in Fig. 4, and subsequent sequencing analyses, demonstrated that the strand-specific primers listed in Table I indeed amplified only the targeted strand. These results set the stage for the determination of the sequence of interest, that is the one carried by the 5Ј33Ј leading strand complementary to the dimer-containing template. We cloned PCR products and sequenced individual molecules so that both major and minor events could be detected.
The final results shown in Tables II-IV strongly support the conclusion that translesion synthesis, most likely catalyzed by pol , was responsible for the insertion of adenines opposite the T ∧ T in Ͼ80% of the bypass products, without transfer of sequence information from the alternative template. Furthermore, the fraction of bypass products with the sequence predictive of template switching was dependent on the capability of the extract for DNA repair. The use of human cell extracts that were deficient in NER or MMR resulted in reductions in the number of clones carrying the sequence information of the nascent lagging strand. This finding suggested that the AATTG (ACTT) clones were not the product of template switching during bypass but instead were generated by DNA repair occurring prior to DNA replication. Remaining to be explained is why the repaired and replicated molecules carrying the sequence 3Ј-TTAAC-5Ј:5Ј-AATTG-3Ј were amplified by methylation-specific PCR when they should have been digested by MunI. Although we optimized the reaction conditions for complete digestion of molecules carrying the MunI recognition sequence, it is possible that some products of in vitro replication could have escaped digestion due to proteins remaining bound to the replicated DNA, even after its purification with the QIAEX II Gel Extraction System. The interpretation that template switching can occur in a small fraction of the replicating molecules, however, cannot be completely dismissed at this point. It can be argued that replication-competent extracts prepared from different human cell lines might differ in their capability to support template switching relative to translesion synthesis.
The maintenance of genetic stability requires that DNA replication be carried out with the highest degree of fidelity. This is accomplished by the combined action of DNA polymerases with low frequency of errors, proofreading activities, and postreplication repair. These functions depend on the recognition of correct base pairing between the template and the nascent strands. However, when the replication machinery encounters damaged sites in DNA, less stringent base pairing conditions must be accommodated (28,29), if DNA synthesis across and beyond the lesion is to take place. This balance between high fidelity during the replication of normal DNA and the need to complete replication even in the presence of DNA lesions requires the concerted effort of different DNA polymerases. Even though the regulation of DNA polymerase switching at sites of blocked DNA replication is not yet understood, the results reported here demonstrate that translesion synthesis is the major mechanism by which a cyclobutane thymine dimer is bypassed during in vitro replication.