Simultaneous Disruption of Two DNA Polymerases, Polη and Polζ, in Avian DT40 Cells Unmasks the Role of Polη in Cellular Response to Various DNA Lesions

Replicative DNA polymerases are frequently stalled by DNA lesions. The resulting replication blockage is released by homologous recombination (HR) and translesion DNA synthesis (TLS). TLS employs specialized TLS polymerases to bypass DNA lesions. We provide striking in vivo evidence of the cooperation between DNA polymerase η, which is mutated in the variant form of the cancer predisposition disorder xeroderma pigmentosum (XP-V), and DNA polymerase ζ by generating POLη−/−/POLζ−/− cells from the chicken DT40 cell line. POLζ−/− cells are hypersensitive to a very wide range of DNA damaging agents, whereas XP-V cells exhibit moderate sensitivity to ultraviolet light (UV) only in the presence of caffeine treatment and exhibit no significant sensitivity to any other damaging agents. It is therefore widely believed that Polη plays a very specific role in cellular tolerance to UV-induced DNA damage. The evidence we present challenges this assumption. The phenotypic analysis of POLη−/−/POLζ−/− cells shows that, unexpectedly, the loss of Polη significantly rescued all mutant phenotypes of POLζ−/− cells and results in the restoration of the DNA damage tolerance by a backup pathway including HR. Taken together, Polη contributes to a much wide range of TLS events than had been predicted by the phenotype of XP-V cells.

Introduction DNA replication involves a rapid but fragile enzymatic mechanism that is frequently stalled by damage in the DNA template. To complete DNA replication, DNA lesions are bypassed by specialized DNA polymerases, a process called translesion synthesis (TLS) (reviewed in [1,2]). A number of TLS polymerases, including Polg and Polf, that are conserved throughout eukaryotic evolution, have been identified in yeast and mammals. Polg deficiency is responsible for a variant form of xeroderma pigmentosum (XP-V) [3,4] that is characterized by UV photosensitivity and a predisposition to skin cancer (reviewed in [5]). Deficiency in Rev3, the catalytic subunit of Polf, results in a considerably more severe phenotype, compared with Polg. In fact, Rev3 disruption is lethal to mouse embryogenesis [6]. Chicken DT40 cells deficient in Rev3 exhibit significant chromosome instability and hypersensitivity to a wide variety of DNA-damaging agents [7][8][9][10]. In addition to their role in TLS, both Polg and Polf can contribute to homologous DNA recombination (HR) in DT40 cells [9,11,12].
Exposure to UV induces cyclobutane pyrimidine dimers (CPDs) and 6-4 UV photoproducts in DNA. While Polg can efficiently and accurately bypass CPDs [4,[13][14][15], no single DNA polymerase has been shown to be capable of effectively bypassing 6-4 UV photoproducts in vitro. This suggests that the coordinate use of more than one polymerase is required to bypass such damage in vivo. In support of this notion, several biochemical studies have suggested that lesion bypass can be effected by two sequential nucleotide incorporation events [16,17]. For example, to bypass the 6-4 UV photoproduct, Polg inserts a nucleotide opposite the damage as a first step, followed by extension from the inserted nucleotide as a second step. This extension process has been shown to be catalyzed by yeast Polf and by human Polk [1,2,18]. The contribution of mammalian Polf to the extension step remains elusive, because functional Polf has not been purified [19,20]. Recently, replication of episomal plasmid DNA carrying various lesions was analyzed in mammalian cell lines to define the role of each DNA polymerase in TLS past individual DNA lesions [15,21,22]. Shachar et al. suggest the sequential usage of Polg and Polf in TLS past the cisPt-GG lesion [21], while results of others do not support this two-step TLS model [15,23]. Indeed, evidence for the two-step model in the replication of chromosomal DNA has so far been lacking. By contrast, both human and yeast Polg can bypass CPDs effectively in vitro without extension polymerases [4].
Combining genetic tractability with a number of sensitive phenotypic assays, the chicken DT40 B lymphocyte cell line provides a unique opportunity to precisely analyze the role of individual DNA polymerases in TLS as well as in HR. The immunoglobulin loci of DT40 cells undergo constitutive diversification in culture by a combination of gene conversion (which depends on HR) and point mutation (which depends on TLS [24]). This diversification is driven by activation-induced deaminase (AID) [25,26], which catalyzes the deamination of cytosine to generate uracil in the immunoglobulin loci. The uracil is then eliminated by uracil glycosylase to form abasic sites, which are thought to be the lesions that trigger bypass, either by gene conversion or by mutagenic translesion synthesis (reviewed in [27]). To study TLS in a different context, an episomal plasmidbased system was recently developed to examine the replication of a plasmid carrying site-specific lesions, in this case 6-4 UV photoproducts, in DT40 cells [28].
To investigate the functional interaction between Polg and Polf in the DT40 cell line, we created POLg 2/2 /POLf 2/2 DT40 cells (hereafter called polg/polf cells). Unexpectedly, depletion of Polg in the polf cells suppressed virtually all mutant phenotypes associated with the loss of Polf, including genome instability and hypersensitivity to DNA-damaging agents. Furthermore, the reconstitution of POLg 2/2 /POLf 2/2 cells with intact human Polg, but not the polymerase-deficient mutant carrying D115A/ E116A substitutions, increased their hypersensitivity to DNAdamaging agents to the level of the POLf 2/2 cells, indicating that Polg-dependent DNA synthesis is toxic in the absence of Polf. Remarkably, this alleviation of the polf phenotype was associated with the restoration of effective translesion synthesis. These data provide in vivo support for the two-step model of lesion bypass, with Polf playing a critical role in the extension step following nucleotide incorporation by Polg.

Results
Deletion of Polg reversed the hypersensitivity of the polf mutant to UV, ionizing radiation, MMS, and cisplatin We generated polg/polf cells by inactivating both REV3 alleles of the polg DT40 cells using a previously published gene-targeting strategy ( Figure 1A) [9,12]. The growth properties of the mutant cells were examined by measuring their growth rate and cell-cycle profile. As reported previously, the polg cells had a normal growth rate, whereas the polf cells proliferated more slowly, exhibiting an increase in the sub-G 1 fraction, indicative of spontaneous cell death during the cell cycle ( Figure 1C). The loss of Polf caused a significant increase in the number of spontaneous arising cH2AX foci, which represent replication collapse ( Figure S1). Interestingly, deletion of POLg in the polf cells reversed their growth retardation and reduced the rate of cell death ( Figure 1B and 1C). Similarly, the number of spontaneous chromosomal aberrations was significantly reduced in polg/polf cells, compared with polf cells (Table 1). Ectopic expression of human Polg in polg/polf cells diminished their growth rate to the level of polf cells. This observation does not reflect general toxicity of the overexpressed human Polg, since its ectopic expression caused pronounced growth retardation only in the polg/polf double mutant but not in wild-type cells ( Figure S2). These observations indicate that the growth defect of polf cells is dependent on the presence of Polg.
The sensitivity of polg/polf cells to genotoxic stresses was evaluated using a colony formation assay. polg cells showed a mild sensitivity to UV but not to the other genotoxic stresses, in agreement with the phenotype of mammalian XP-V cells [29,30]. In contrast, polf cells showed a marked sensitivity to UV, ionizing radiation, cis-diaminedichloroplatinum-II (cisplatin), and methylmethane sulfonate (MMS) (Figure 2), as previously described [9]. The polg/polf mutant cells were less sensitive to UV than were the polf cells. Furthermore, this double mutant showed significantly increased tolerance to ionizing radiation, cisplatin, and MMS, compared with the polf cells. This increased tolerance of the polg/ polf cells was reversed by ectopic expression of human Polg. To investigate whether this reversion depended on the polymerase activity of human Polg, we expressed human POLg cDNA carrying D115A/E116A mutations in the polg/polf cells. These mutations in the catalytic site abolish polymerase activity (data not shown), as previously reported in yeast [31]. The expression of the mutant Polg had no effect on the sensitivity of the polg/polf cells to the DNA-damaging agents (Figure 2), indicating that Polgdependent DNA synthesis sensitizes polf cells to these DNAdamaging agents.
polf cells have a more prominent defect in TLS past abasic site than polg/polf cells We wished to investigate if Polg and Polf could collaborate in TLS past specific types of DNA damage. To this end, we performed two sets of experiments: analysis of immunoglobulin hypermutation, which in DT40 provides a readout of the bypass of abasic sites [24], and analysis of the replication of a T-T (6-4) photoproduct on an episomal plasmid [28].
To induce Ig hypermutation, we overproduced AID in DT40 cells using a retrovirus vector [32,33]. This vector drives the monocistronic expression of AID and green fluorescent protein (GFP), allowing a comparative assessment of the level of ectopic AID expression. At 24 hours after infection with the AID retrovirus, virtually all cells from each line displayed a strong GFP signal, indicating that the deficiency of Polg and Polf did not affect the expression of AID ( Figure 3A). However, a substantial fraction of the polf cells died at day 3, and with the surviving polf Author Summary DNA replication is a fragile biochemical reaction, as the replicative DNA polymerases are readily stalled by DNA lesions. The resulting replication blockage is released by translesion DNA synthesis (TLS), which employs specialized TLS polymerases to bypass DNA lesions. There are at least seven TLS polymerases known in vertebrates. However, how they cooperate in vivo remains one of central questions in the field. We analyzed this functional interaction by genetically disrupting two of major TLS polymerases, Polg and Polf, in the unique genetic model organism, chicken DT40 cells. Currently, it is widely believed that Polg plays a very specific role in cellular tolerance to ultraviolet light-induced DNA damage. Polf, on the other hand, plays a key role in cellular tolerance to a very wide range of DNA-damaging agents, as POLf 2/2 cells are hypersensitivity to a number of DNA damaging agents. Our phenotypic analysis of POLg 2/2 /POLf 2/2 cells shows that, unexpectedly, the loss of Polg significantly rescued all mutant phenotypes of POLf 2/2 cells. The genetic interaction shown here reveals a previously unappreciated role of human Polg in cellular response to a wide variety of DNA lesions and two-step collaborative action of Polymerase g and f.
cells at day 10 showed a decrease level of GFP signals ( Figure 3A and 3B). Furthermore, polf cells, but not polg or polg/polf cells, displayed prominent chromosomal breaks at day 3 post-infection ( Figure 3C). Since the break sites on the chromosome were randomly distributed, the overexpressed AID protein may be targeting a number of different loci in addition to the Ig locus, in DT40 cells. These observations suggest that TLS past abasic sites created by the combined action of AID and uracil glycosylase may    be performed less effectively in polf cells, compared with polg/polf or polg cells, resulting in chromosome breakage and cell death.
To verify that Polg-dependent DNA synthesis was toxic to the AID-overproducing polf cells, we reconstituted polg/polf cells with either wild-type POLg or the catalytically inactive mutant polgD115A/E116A). Wild-type POLg expression sensitized the polg/polf cells to the overexpression of AID, whereas the mutant POLg had no impact on cell survival ( Figure 3A and 3B). We thus conclude that, in polf cells, TLS past abasic sites may be less effective because neither Polg nor any other polymerase can extend DNA synthesis following the Polg-mediated insertion of nucleotides opposite the abasic site.
We have shown that AID overexpression results in increased TLS-mediated hypermutation at G/C base pairs in Ig V segments [32]. To define the role of Polg and Polf in this TLS process, we determined the Ig V l nucleotide sequences of AID-overexpressing wild-type, polg, and polg/polf cells. polf cells were not analyzed because of the difficulty of ectopically expressing AID to the same extent as in the other lines. The number of non-templated point mutations (PM, Figure 4A) was somewhat lower in polg cells than in wild-type cells (This slight reduction is not significant (p = 0.15, ttest) [32]). Surprisingly, the level of Ig V mutations in polg/polf cells was comparable to that of wild-type cells. This observation is in marked contrast with the fact that Rev1, an essential factor for the function of Polf [20], plays a critical role in non-templated point mutations at the abasic site [34]. Moreover, Polf played the critical role in cellular tolerance to AID-mediated abasic sites (Figure 3). These observations indicate that in the absence of both Polg and Polf, other unidentified DNA polymerase(s) can participate in TLS past the abasic site. As the polg/polf cells displayed a significant increase in the proportion of G/C to A/T transitions in the non-templated Ig V mutations (12/25; 48%), compared with wild-type cells (2/18; 11%)( Figure 4B), this unidentified DNA polymerase(s) may preferentially incorporate adenine opposite the abasic site in the absence of both Polg and Polf.
Lack of TLS past the T-T (6-4) UV photoproduct in polf cells is reversed by the inactivation of Polg T-T (6-4) UV photoproducts represent the most formidable challenge to DNA replication, as they potently arrest replicative polymerases [35]. To analyze TLS past a T-T (6-4) photoproduct, we transfected two plasmids, pQTs and pQTo, [28] (Figure 5) carrying T-T (6-4) UV photoproducts into DT40 cells. At two days after transfection we recovered only replicated copies, and were thus able to analyze the replication of site-specific T-T (6-4) UV photoproducts in vivo. This photoproduct can be arranged in one of two ways. In the staggered conformation (pQTs) ( Figure 5A), the lesions are separated by 28 intervening nucleotides and placed opposite a GpC mismatch. Replicated copies can thus result from TLS on the top or bottom strand. Error-free template switching should result in GpC at the site of the photoproduct, while TLS  Figure 3. Cells were clonally expanded for two weeks. Three clones were analyzed in each data set. Note that the Amb mutation at A/T pairs represents short-tract gene conversion [32]. (B) Nucleotide substitution preferences (as a percentage of the indicated number in the total mutations) deduced from the V l sequence analysis of the clones shown in (A). The preferences are shown for mutations categorized as non-templated base substitutions (PM), which are caused by TLS past abasic sites [32]. doi:10.1371/journal.pgen.1001151.g004 past this photoproduct may insert ApA (accurate TLS) or other nucleotides (inaccurate TLS) at this site. Note that our experiment was done in a nucleotide-excision repair-deficient (xpa-deficient) background, and thus excluded the recovery of replicated copies generated by error-free nucleotide-excision repair. In the second, unphysiological, replication template, the lesions are placed opposite to each other (pQTo) ( Figure 5E). Using this conformation, replicated DNA copies can be recovered as a consequence of TLS, but not by template switching.
To assess the mode of bypass used when generating replicated copies of pQTs, we analyzed the nucleotide sequences of replicated copies of plasmids recovered from DT40 cells ( Figure 5B) and determined the proportion of TLS relative to error-free template switching ( Figure 5C). Overall replication efficiency was comparable among cells carrying the various genotypes used in the previous study and this study. Previous study found that 45% and 55% of the recovered plasmid copies resulted from TLS in xpa and xpa/polg cells, respectively [28]. In comparison, the efficiency of TLS in xpa/polf cells was significantly reduced, with less than 10% of the recovered plasmid generated as a consequence of TLS. All TLS events observed in the xpa/polf cells were associated with deletion at damaged sites ( Figure 5D). Thus, as found previously [28] (Figure 5C), we conclude that Polf is required for the successful bypass of T-T (6-4) UV photoproducts by TLS. Remarkably, the xpa/polg/polf cells displayed a normal TLS efficiency, indicating that the failure of TLS in polf cells is dependent on the presence of Polg.
We also classified the replication products obtained from the pQTo plasmid, where bypass can be effected only by TLS or deletion. As found in the previous study [28], the loss of Polf was frequently associated with the deletion of two or more nucleotides covering the site of the T-T (6-4) UV photoproduct ( Figure 5E and 5F). The loss of Polf did not impair TLS in the absence of Polg, but severely compromised it in the presence of Polg. A possible explanation, discussed below, is that the Polg-dependent insertion of nucleotides opposite the T-T (6-4) UV photoproduct inhibits the completion of TLS in the absence of Polf (presumably due to defective extension from inserted nucleotides), while other unidentified DNA polymerases can perform the complete TLS reaction in polg/polf cells.

Polk plays only a minor role in the damage tolerance of polg/polf cells
The milder phenotype of the polg/polf cells, compared with polf single mutants, led us to investigate the contribution of other TLS polymerases to TLS in polg/polf cells. We previously showed that polk/polf cells show a higher sensitivity to mono-alkylating agents, compared with polf cells, though polk cells show normal sensitivity [36], and thereby suggested that Polk can partially substitute for the loss of Polf. We therefore sought to determine whether Polk contributes to damage tolerance in polg/polf cells. To this end, we deleted the Polk gene in polg/polf cells and analyzed the phenotype of the resulting triple knockout polg/polk/polf clones ( Figure 6A). Deletion of the Polk gene tended to reduce growth kinetics. However, the polg/polk/polf clones exhibited only limited increase in DNA-damage sensitivity, compared with polg/polf cells ( Figure 6B). Likewise, overexpression of chicken Polk did not increase cisplatin tolerance in polf or polg/polf cells (data not shown). These observations imply that Polk does not play an important role in TLS in polg/polf cells.

polg/polf cells, but not polf cells, displayed increased numbers of UV-induced sister chromatid exchange events
Replication arrest can be released by two major mechanisms: HR and TLS [37]. HR-dependent release is initiated by homologous pairing between the 39 end of the arrested strand and the sister chromatid, followed by strand invasion and DNA synthesis to extend the invading 39 strand ( Figure 7D). To analyze the efficiency of HR-mediated release from replication blockage, we analyzed sister chromatid exchange (SCE) (Figure 6C and 6D) [38,39]. The level of SCE is likely to be determined by two factors: the number of DNA lesions that cause a replication block and the efficiency of HR-dependent release from replication blockage. As previously reported [9,39], the level of spontaneous SCE was slightly increased (1.5 to 2-fold) in all TLS mutants compared with wild-type cells, presumably because lesions are more frequently channeled to HR. The polg cells displayed a markedly greater increase in the level of UV-induced SCE (the number of spontaneous SCE subtracted from the number of SCE following UV irradiation shown in Figure 6D), which phenotype is attributable to defective TLS over UV-induced damage. In contrast, in the polf cells, the UV-induced SCE level was similar to that of wild-type cells, suggesting that the defective TLS may not be adequately compensated by HR [9]. SCE was induced more efficiently in the polg/polf cells than in the polf cells, which is consistent with the increased UV tolerance of polg/polf cells, compared with polf cells. Reconstitution of the polg/polf cells with wild-type POLg, but not with the catalytically inactive POLg, significantly reduced the number of UV-induced SCE events. Thus, the degree of UV tolerance correlated with the number of UV-induced SCE events at least in polf and polg/polf cells. This observation suggests that, in addition to TLS, HRmediated release from replication blockages contributes to a significantly higher UV tolerance in polg/polf cells than in polf cells.

Discussion
Polg is required for TLS across a much wider range of DNA lesions than indicated by the sensitivity of Polg-deficient cells. Analysis of XP-V cells indicates that Polg plays a major role in TLS past cyclobutane dimers and certain bulky adducts, but few other lesions. We demonstrate here that, in DT40 cells, Polg is also involved in the interaction of the replication machinery with different types of damage, such as those induced by chemical cross-linking agents (Figure 2), abasic sites (Figure 3 and Figure 4), and T-T(6-4) UV photoproduct ( Figure 5). Thus, the deletion of polg in the polf cells reversed their hypersensitivity to UV, MMS, cisplatin, and the ectopic expression of AID. Our finding that polg/polf cells were significantly more tolerant to all tested DNAdamaging agents compared with polf cells, reveals that neither of these polymerases is absolutely required for the tolerance of these types of damage during DNA replication. However, when Polg is present, Polf is also required for efficient recovery from the effects of these damaging agents.  XP-V cells show a modest phenotype, including moderate sensitivity to UV and cisplatin [40], only in the presence of caffeine, and do not show significant sensitivity to MMS. Consequently, it was once believed that Polg does not play a role in TLS past DNA lesions induced by alkylating agents. However, in vitro biochemical studies have shown that purified Polg can bypass a variety of lesions including 7, 8-dihydro-8oxoguanine (8-oxoG), O 6 -methylguanine, abasic sites, benzo pyrene adducts, and cisplatin intrastrand crosslinks [2,[41][42][43]. The present study helps demonstrate that Polg can indeed be involved in TLS past a wide variety of DNA lesions, even without caffeine treatment in DT40 cells. This idea is relevant to mammalian cells, since Polg focus formation is observed in mammalian cells following treatment with cisplatin and UV in the absence of caffeine [40].

The collaborative action of Polg and Polf in TLS
The improved damage tolerance of polg/polf cells, compared with polf cells, suggests following several possibilities. One possibility is that Polf somehow inhibits Polg action and that Polg does not actually have a significant role when Polf is present. However this possibility may be unlikely, since physical interaction between Polg and Rev1, and Polg dependent recruitment of Rev1 to the DNA damage site support the sequential actions of Polg followed by Polf rather than inhibitory action of Polf on Polg [44][45][46][47]. Thus, more likely possibility is that, Polg generates a replicative intermediate in an attempt to bypass the lesion, but cannot complete an effective bypass reaction without Polf ( Figure 7A). We suggest that the abortive intermediates generated by Polg in the absence of Polf lead to a difficult-to-rescue replication collapse, thereby accounting for the hypersensitivity of the single polf mutant ( Figure 7B). The modest phenotype of XP-V cells indicates that Polf may efficiently mediate TLS past a variety of DNA lesions, either in collaboration with other polymerases or possibly on its own. This situation can be explored further in the light of current TLS models. In the canonical model for TLS replication, arrest by agents such as UV, MMS, and cisplatin, leads to PCNA becoming mono-ubiquitinated [40,[48][49][50]. This increases the affinity of PCNA for Polg and other Y-family TLS polymerases by virtue of their UBM and UBZ ubiquitin-binding motif [44,48,[50][51][52] and likely contributes to the accumulation of Y-family polymerases at the sites of blocked replication forks. It has been suggested that these polymerases can compete with each other by mass action to attempt to carry out TLS. In the case of CPDs, if Polg wins this competition; bypass can occur without the need for a second polymerase. However, with other lesions such as a T-T (6-4) photoproduct, there is currently no evidence to show that any single polymerase can complete bypass. Polg may be able to start the bypass by incorporating opposite at least the 59 base of the lesion, but it cannot extend from the resulting mismatch ( Figure 7A). This would explain the abortive intermediate referred to above. To complete TLS, Polf is required to extend from the inserted nucleotides ( Figure 7A), an explanation that is consistent with the sequential action of Polg and Polf demonstrated in in vitro studies [2]. A further implication of this model is that no other polymerase can effectively substitute for Polf in this extension step ( Figure 7B).
The successive action of Polg and Polf might be mediated by the association of the two polymerases with Rev1 [45,53]. The idea of a Rev1-mediated switch from Polg to Polf is supported by the fact that Polg tightly interacts with Rev1 [1,46,47,53,54]. Moreover, DNA-damage-induced Rev1 focus formation appears to be dependent on Polg [46,47]. Adding to these findings, the present work establishes a role for Polg in the bypass of a wider range of DNA damage than previously thought and demonstrates the in vivo importance of the two-step bypass of many lesions.
The effect on TLS of depleting both Polg and Polf is distinctly different in DT40 cells than in human cells. Ziv et al. showed that depletion of Rev3 sensitized cells to UV more severely at two days after irradiation in Polg-deficient XPV fibroblasts than in Polg-proficient control cells [22], although its impact on Polgproficient DT40 cells was considerably stronger than on Polgdeficient DT40 cells. There are several explanations for this apparent difference. First, we favor the idea that DT40 may be a more reliable cell line than others to evaluate TLS by measuring cellular survival due to following reasons. The cell cycle distribution is distinctly different between DT40 and other mammalian cell lines. In DT40 cells, ,70% of the cells are in the S phase, and the G 1 /S checkpoint does not function at all [55]. In most of the mammalian cell lines, on the other hand, more than 50% of the cells are in the G 1 phase, and G 1 /S checkpoint works at least partially. Therefore, environmental DNA damage interferes with DNA replication more significantly in DT40 cells than in mammalian cell lines. Accordingly, TLS contributes to the cellular survival of the colony formation assay to a considerably higher extent in DT40 cells than in mammalian cell lines. Ziv et al., on the other hand, evaluated TLS by measuring the number of living cells at 48 hours after UV irradiation [22]. Since a majority of UV irradiated cells may have stayed outside the S phase at 48 hours, it is unclear whether this survival reflects the efficiency of TLS. The same laboratory also analyzed TLS past a cisplatin G-G intrastrand crosslink located in a gapped episomal plasmid [21]. Depletion of Rev3 with or without codepletion of Polg in U2OS cells resulted in an 80% reduction in TLS past the lesion, irrespective of the presence or absence of Polg. The relevance of this finding to TLS during the replication of chromosomal DNA remains elusive. Yoon et al. investigated TLS in a double-stranded plasmid containing a single 6-4 photoproduct as well as replication origins derived from the SV40 virus [23]. Depletion of Rev3 or Rev7 in NERdeficient XP-A fibroblasts reduced the efficiency of TLS in the episomal plasmid by approximately 50%, with a similar reduction obtained in XP-V fibroblasts. This result using human fibroblasts is clearly different from the data we obtained using DT40 cells. Given the close sequence similarity between the two polymerases in human and chicken cells, we consider it unlikely that the mechanisms of lesion bypass are fundamentally different between the two organisms. The apparent difference between mammalian cells and DT40 may be caused by the incomplete si-RNA mediated inhibition in human cells versus the null mutation we have used in DT40. Another possible reason to explain this difference is the active HR system in DT40 cells, and the different usage order of TLS DNA polymerases because the DT40 B lymphocyte line undergoes Ig V hypermutation through TLS [32]. The usefulness of the three episomal plasmid systems to the analysis of TLS occurring during replication of chromosomal DNAs should be further investigated [15,21,22,28]. Irrespective of the reason for this apparent difference, our data clearly indicate that, as discussed above, under some conditions Polg can hinder the efficient progress of the replication fork past lesions mediated by Polf.

Rescue of failed translesion synthesis by homologous recombination
It is remarkable that deletion of the three major TLS polymerase genes, POLg, POLk, and POLf, results in only a mild reduction in the growth kinetics of DT40 cells ( Figure 6A). This is in marked contrast with the immediate cell death associated with the massive chromosomal breaks generated upon deletion of Rad51 [56]. These observations imply that during DNA replication, if replication blocks are encountered, HR can at least partially compensate for defective TLS. The significant functional redundancy between TLS and HR is supported by our previous report, which concludes that the deletion of both RAD18 and RAD54, a gene involved in HR, as well as the deletion of both REV3 and RAD54, are synthetically lethal to cells [9,39]. We show here that depletion of Polg irrespective of the status of Polf markedly increases the level of UV-induced SCE ( Figure 6C), suggesting that DNA damage that cannot be resolved by TLS because of the absence of Polg may be resolved by HR, leading to increased SCE levels ( Figure 7D). However, SCE is not induced to the same level in polf cells ( Figure 6C). Thus, nucleotide incorporation by Polg appears to represent a point of commitment in the TLS reaction beyond which rescue by HR is problematic. This is likely to be explained by the creation of an intermediate, possibly the mismatched primer terminus, which can be efficiently extended by Polf, but which cannot readily initiate HR.
In summary, the significant increase in the cellular tolerance of polg/polf cells to DNA-damaging agents, compared with polf cells, can be partially attributable to more efficient HR in polg/polf cells than in polf cells. However, the normal level of TLS-dependent Ig V mutation and restoration of TLS during 6-4 photoproduct bypass in polg/polf cells ( Figure 4B) suggests that one or more other unidentified TLS polymerases can act as a substitute and carry out TLS when both Polg and Polf are missing ( Figure 7C).

Cell lines and cell culture
Generation of polf (rev3)-and polg-deficient DT40 cells was described previously [9,12]. To generate polg/polf cells, we sequentially introduced two rev3 gene-disruption constructs (rev3hygro and rev3-His) [9] into polg (Puro R /Bls R ) cells. A puromycinresistant XPA disruption construct was used to disrupt the single XPA allele and recreate the xpa/rev3 cell line. After removal of the Bls R marker gene from the polg/polf cells by the transient expression of CRE recombinase, a blasticidin-resistant XPA disruption construct was used to generate the xpa/polf/polg cell line. A puromycin-resistant POLk disruption construct was used to generate polg/polf/polk cells. To make a POLg expression plasmid, we inserted human POLg cDNA into the multi-cloning sites (MCS) of an expression vector, pCR3-loxP-MCS-IRES-GFP-loxP [57]. A mutant Polg that lacks polymerase activity (Mutant POLg) was generated by inserting D115A/E116A mutations into human Polg cDNA. The conditions for cell culture, selection, and DNA transfection were described previously [58]. The growth properties of cells were analyzed as described previously [58].

AID overexpression by retrovirus infection
For the retrovirus infection, the pMSCV-IRES-GFP recombinant plasmid was constructed by ligating the 5.2 kb BamHI-NotI fragment from pMSCVhyg (Clontech) with the 1.2 kb BamHI-Not1 fragment from pIRES2-EGFP (Clontech). Mouse AID cDNA [33] was inserted between the BglII and EcoRI sites of pMSCV-IRES-GFP. The preparation and infection of the retrovirus were carried out as previously described [33]. Expression of the GFP was confirmed by flow cytometry. The efficiency of infection was more than 90% as assayed by GFP expression.

Chromosome aberration analysis
Karyotype analysis was performed as described previously [56]. Cells were treated with colcemid for 3 hours to enrich mitotic cells.

Measurement of SCE level
Measurement of SCE level was performed as described previously [38]. For UV-induced SCE, cells were suspended in PBS and irradiated with 0.25 J/m 2 UV followed by BrdU labeling.

Ig V l mutation analysis
Genomic DNA was extracted at 14 days after subcloning. The rearranged V l segments were PCR amplified using 59-CAG-GAGCTCGCGGGGCCGTCACTGATTGCCG-39 as the forward primer in the leader-V l intron and 59-GCGCAAG-CTTCCCCAGCCTGCCGCCAAGTCCAAG-39 as the reverse primer in the 39 of the JC l intron. To minimize PCR-introduced mutations, the high-fidelity polymerase, Phusion (Fynnzymes) was used for amplification (30 cycles at 94u C for 30 s; 60u C for 1 min; 72u C for 1 min). PCR products were cloned using a Zero Blunt TOPO PCR Cloning Kit (Invitrogen) and subjected to sequence analysis with the M13 forward (-20) or reverse primer. Sequence alignment using GENETYX-MAC (Software Development, Tokyo) allowed the identification of changes from the parental sequences in each clone.
As described previously [59], all sequence changes were assigned to one of three categories: point mutation, gene conversion, or ambiguous. This discrimination is based on the published sequences of V l pseudogenes that can act as donors for gene conversion. For each mutation, the database of V l pseudogenes was searched for a potential donor. If no pseudogene donor containing a string .9 bp could be found, the mutation was categorized as a non-templated point mutation. If such a string was identified and there were further mutations that could be explained by the same donor, then all these mutations were categorized as a single long-tract gene conversion event. If there were no further mutations, indicating that the isolated mutation could have arisen through a conversion mechanism or could have been non-templated, it was categorized as ambiguous.