Long Patch Base Excision Repair with Purified Human Proteins

Among the different base excision repair pathways known, the long patch base excision repair of apurinic/apyrimidinic sites is an important mechanism that requires proliferating cell nuclear antigen. We have reconstituted this pathway using purified human proteins. Our data indicated that efficient repair is dependent on six components including AP endonuclease, replication factor C, proliferating cell nuclear antigen, DNA polymerases δ or ε, flap endonuclease 1, and DNA ligase I. Fine mapping of the nucleotide replacement events showed that repair patches extended up to a maximum of 10 nucleotides 3′ to the lesion. However, almost 70% of the repair synthesis was confined to 2–4-nucleotide patches and DNA ligase I appeared to be responsible for limiting the repair patch length. Moreover, both proliferating cell nuclear antigen and flap endonuclease 1 are required for the production and ligation of long patch repair intermediates suggesting an important role of this complex in both excision and resynthesis steps.

The constant danger of damaging DNA is counteracted in nature by a variety of different repair mechanisms such as nucleotide excision repair, base excision repair (BER), 1 doublestrand break repair, mismatch repair, and recombinational repair. Among these BER is an essential mechanism that guarantees a rapid and accurate removal of damaged bases in the genome. BER involves the removal of a single base and its replacement by a chain of enzymatic steps. Efficient repair of a uracil-guanine base pair present in a duplex oligonucleotide can be achieved in vitro, via replacement of a single nucleotide (short patch BER), by the sequential action of the human proteins uracil-DNA glycosylase, the apurinic/apyrimidinic (AP) endonuclease HAP 1, DNA polymerase (pol) ␤, and either DNA ligase III (1) or DNA ligase I (2). The second BER pathway which involves gap filling of several nucleotides (long patch BER) shares some common proteins with the short patch pathway (e.g. the base damage recognition and cleavage enzymes) but presents also some specific ones. Long patch BER requires specifically PCNA (Refs. 3 and 4, reviewed in Ref. 5), its matchmaker protein replication factor C (RF-C) (Ref. 6, reviewed in Ref. 7), and the flap structure-specific endonuclease 1 (FEN1) (Ref. 8, reviewed in Ref. 9). The PCNA dependence of this pathway as well as the efficient long patch BER detected in pol ␤-deleted mouse cells (10) strongly suggested the involvement of either pol ␦ and/or ⑀ in the resynthesis step. At a reduced AP site, however, the long patch BER synthesis was strongly inhibited by antibodies against pol ␤ in an in vitro repair assay with cell-free extracts (8). In these experiments a linear double-stranded DNA oligonucleotide was used as substrate. A linear substrate, however, abolishes the requirement for RF-C in loading PCNA onto the DNA and relies on a simple sliding of the ring-shaped trimeric PCNA protein from the free DNA ends (11). We have recently demonstrated by using fractionated cell extracts from pol ␤-deleted mouse cells that either pol ␦ or pol ⑀ participates in long patch BER (6). PCNA and RF-C, but not RP-A were required for the Pol ␦/⑀-dependent long patch repair supporting a direct involvement of these polymerases in the repair of AP sites in cells of higher eukaryotes. Both pol ␦ and pol ⑀ were also able to replace a single nucleotide at the lesion site, but the repair intermediates were inefficiently ligated.
The issue of the identity of polymerases involved in the DNA synthesis step that fills in gaps created during repair has also been addressed in other DNA repair systems. Pol ␦ and/or pol ⑀ have been implicated in the synthesis step of at least three other repair processes, including nucleotide excision repair (12), mismatch repair (13), and double-strand break repair (14), which in addition all require PCNA and RF-C in the case of nucleotide excision repair (12) and double-strand break repair (14).
FEN1 is a structure-specific endonuclease that cleaves single-stranded flaps at the junction to double-stranded DNA (9). This enzyme was also shown to act as a 5Ј-3Ј exonuclease and it interacts with PCNA (15). More recent studies demonstrated that PCNA must be located below the 5Ј flap on a forked template with the C-terminal side facing toward the flap to stimulate FEN1 (16). This nuclease has been demonstrated to be involved in the PCNA-dependent long patch BER pathway (17) and it was shown that PCNA can facilitate excision in long patch BER through its interaction with FEN1 (18).
In this paper we show that purified HAP 1, PCNA, RF-C, pol ␦/⑀, FEN1, and DNA ligase I are all required for efficient long patch BER at an AP site and that this set of proteins is the core  1 The abbreviations used are: BER, base excision repair; AP, apurinic/ apyrimidinic; RF-C, replication factor C; FEN1, flap structure-specific endonuclease 1; pol, polymerase; PAGE, polyacrylamide gel electrophoresis; PCNA, proliferating cell nuclear antigen; nt, nucleotide(s); IS, internal standard. machinery for long patch BER. Moreover, our data demonstrated that the repair intermediate pattern by pol ␦ and ⑀ is different in the absence of DNA ligase I. The apparent role of DNA ligase I in controlling the length of the repair patches is discussed. In the accompanying paper, Matsumoto et al. (42) describe the reconstitution of pol ␦-mediated repair of AP sites with purified components. Their findings are in good agreement with our results.

Nucleic Acid Substrates
Poly(dA) 200 and oligo(dT) 16 were purchased from Amersham Pharmacia Biotech and Microsynth, respectively. Poly(dA):oligo(dT) (base ratio 10:1) was prepared as described (19). Oligonucleotides for preparing the substrate for the BER assay were purchased from M-Medical. Closed circular DNA containing a single abasic site was produced as described previously (4) by priming single-stranded (ϩ) pGEM-3Zf DNA (Promega) with a 30-fold molar excess of the uracil-containing oligonucleotide 5Ј-GATCCTCTAGAGUCGACCTGCA-3Ј and incubating with T4 DNA polymerase holoenzyme, single-stranded DNA-binding protein and T4 DNA ligase (Roche Molecular Biochemicals). Closed circular DNA duplex molecules were purified by cesium chloride equilibrium centrifugation and then digested with Escherichia coli uracil-DNA glycosylase to produce a single abasic site.

Enzymes and Proteins
Restriction enzymes were from New England Biolabs or Roche Molecular Biochemicals.
Recombinant Proteins from E. coli-Purified recombinant human AP endonuclease (HAP 1) was a gift from I. D. Hickson. Purified uracil DNA glycosylase was a gift from S. Boiteux. Human PCNA was produced in E. coli using the plasmid pT7/hPCNA (gift of B. Stillman) and purified to homogeneity as described (20). Human FEN1 and human DNA ligase I were expressed in E. coli and purified as described in detail earlier (16).
Proteins Purified from HeLa Cells-RF-C was purified from nuclear extract of 60 g of harvested HeLa cells as described (21) with some modifications (22). Pol ␦ and pol ⑀ from HeLa cells were prepared from cytoplasmic extract of 60 g of HeLa cells according to Ref. 23 and subsequently purified as described in Ref. 19.

Enzymatic Assays
Assay for HAP 1-The assay mixture contained 40 mM Hepes/KOH (pH 7.9), 75 mM KCl, 5 mM MgCl 2 , 0.5 mM dithiothreitol, 2 mM ATP, 40 mM phosphocreatine, 2.5 g of creatine phosphokinase, 18 g of bovine serum albumin, a duplex oligonucleotide carrying an AP site ( 32 Plabeled at the 5Ј-end of the strand containing the AP site), and enzyme to be tested. Reactions were incubated at 30°C for 30 or 120 min and incised products were separated from unincised oligonucleotides by electrophoresis on 15% denaturing PAGE and visualized by autoradiography.
Pol and RF-C Assays-The poly(dA):oligo(dT) RF-C independent assay was carried out according to Ref. 19 and the insoluble radioactive material determined as described in Ref. 24. One unit of pol activity is defined as the incoporation of 1 nmol of dTMP into acid insoluble material in 60 min at 37°C. The RF-C dependent pol assay was used according to Ref. 22. One RF-C unit is defined as the incorporation of 10 pmol of dNMP into acid insoluble material in 60 min at 37°C. FEN1 Endonuclease Assay-FEN1 assays were performed as described (25).
DNA Ligase I Assay-DNA ligase activity was assayed using a 5Ј radiolabeled 31-mer and 5Ј-phosphorylated 30-mer oligonucleotides annealed to a 2-fold molar excess of a complementary 61-mer oligonucleotide. Each assay contained 2 pmol of annealed substrate in 10 l of the BER assay reaction buffer (see below) or T4 DNA ligase buffer, and enzyme fraction to be tested. Samples were incubated for 1 h at 30°C. The ligated products (61 nt) were separated from the unligated 31-mer by electrophoresis on 15% denaturing PAGE and visualized by autoradiography.
BER Assay-The repair reactions were carried out essentially as described in Refs. 26 and 6 by using highly purified components.
Briefly, reaction mixtures (50 l) contained 40 mM Hepes/KOH (pH 7.9), 75 mM KCl, 5 mM MgCl 2 , 0.5 mM dithiothreitol, 20 M of each dNTP, 2 mM ATP, 40 mM phosphocreatine, 2.5 g of creatine phosphokinase, 18 g of bovine serum albumin, 300 ng of plasmid DNA containing a single AP site, 2 Ci of either [␣-32 P]dTTP or [␣-32 P]dCTP as indicated, and 2.5 ng of HAP 1, 0.8 units of RF-C, 0.2 units of pol ␦ or 0.1 units of pol ⑀, 50 ng of PCNA, 25 ng of FEN1, and 200 ng of DNA ligase I. All components were titrated against the others at saturation. The individual activities of pol ⑀, pol ␦, RF-C, FEN1, and DNA ligase I were determined as described above. After 120 min at 30°C the plasmid DNA was recovered and digested with the appropriate restriction enzymes. The digestion products were resolved on a denaturing 20% PAGE. The BER products were visualized by autoradiography and analyzed by electronic autoradiography (Instant Imager, Packard).

Antibodies, Electrophoresis, and Immunoblot Analysis
Polyclonal antibodies against mouse pol ␦ were prepared from rabbit as described by Ref. 27. Antibodies against the 140-and 40-kDa subunits of RF-C were gifts of B. Stillmann and J. Hurwitz, respectively. Monoclonal antibody against human pol ⑀ catalytic subunit was a gift of J. Syvä oja. Polyclonal antibody against bovine DNA ligase I was a gift from T. Lindahl. Monoclonal antibody against PCNA was purchased from Roche Molecular Biochemicals and polyclonal antibody against FEN1 was a gift of M. Lieber. SDS-PAGE was as described by Laemmli (28), stained with either Coomassie Blue or silver (29) and immunobloting was done according to the manufacturer's (Pierce Chemical Co.) protocol for ECL blotting.

Purification of Six Different Protein Components Required for Long Patch BER at an AP
Site-Initial studies suggested that a minimal set of six proteins (HAP 1, PCNA, RF-C, pol ␦/pol ⑀, FEN1, and DNA ligase I) is required to perform long patch BER in vitro at an AP site (6,8). We have now highly purified all these proteins from either recombinant or natural source ( Fig. 1). All final protein fractions were tested for enzymatic cross-contamination by the six others in specific enzymatic assays for HAP 1, PCNA, RF-C, pol ␦, pol ⑀, FEN1, and DNA ligase I as outlined under "Materials and Methods." Furthermore, all proteins except the recombinant HAP 1 were tested in immunoblots for their bona fide identifications. Pol ␦ did not show a signal in an immunoblot for pol ⑀ and vice versa (data not shown).
DNA Polymerase ␦ Performs Significant Processive DNA Synthesis Beyond the Lesion in the Absence of DNA Ligase I-The purified proteins HAP 1, RF-C, PCNA, and FEN1 were tested in the presence of pol ␦ in an in vitro BER assay (4) using an AP site-containing circular duplex plasmid molecule as substrate (outlined in Fig. 2A). This DNA substrate has been constructed to allow fine mapping of the repair patches at the lesion site by restriction digestions and/or by using different labeled dNTPs in the repair reaction. In particular, long patch BER is identified by the incorporation of dCMP within the BamHI-HindIII restriction fragment containing the AP site. Additionally, the incorporation of dTMP in this construct is useful for detecting intermediates containing one incorporated nucleotide. These intermediates can be identified on the gel as a radioactive band which comigrates with a 13-mer. Long patch BER intermediates are expected to measure 14 or more nucleotides, whereas complete repair products comigrate with the full-length BamHI-HindIII fragment, a 30-mer. To analyze the yield and type of repair intermediates produced by different concentrations of the purified proteins, the experiment displayed in Fig. 2B was performed in the absence of DNA ligase I. dTTP was used as labeled precursor in the 2-h repair reaction at 30°C. A 5Ј-end labeled 60-mer (indicated as internal standard, IS) was added in all reaction tubes in order to correct the repair incorporation values, as measured by electronic autoradiography, for DNA recovery. Repair intermediates strictly confined to 1-nt replacement events were observed in the ab-sence of FEN1 (Fig. 2B, lane 1). At concentrations of pol ␦ which do not allow detectable synthesis at the incised AP site (lane 2), an increase in FEN1 concentration (lane 3) stimulated the synthesis of a ladder of repair intermediates. Replacements of 1 nt and Ͼ18 nt (indicated by the 30-nt products) were the dominant events. In the absence of DNA ligase, the radioactive 30-nt products represent elongation by pol ␦ beyond the Hin-dIII restriction site converted into 30-nt fragments by digestion with BamHI and HindIII. The yield of the repair intermediates and of the radiolabeled 30-nt product increased at increasing concentrations of either pol ␦ or FEN1 or both (lanes 2-7). The yield of 30-nt product was also enhanced by increasing RF-C concentration in the presence of 25 ng of FEN1 (data not shown). When DNA ligase was added to the reaction (lane 8) most of the repair intermediates were converted into radioactive 30-nt products that represent fully repaired molecules as determined by mapping of the repair patches (see below). These results suggest that nick translation by pol ␦/FEN1 occurs at the lesion site in the absence of DNA ligase I.
DNA Polymerase ␦-dependent Repair of AP Sites: Omission of Each of the Six Core Protein Factors Lead to Incomplete Repair-The purified proteins HAP 1, RF-C, PCNA, FEN1, and DNA ligase I were tested in the presence of pol ␦. Fig. 3 shows the repair products obtained when dTTP is used as labeled precursor in the 2-h repair reaction at 30°C. In the absence of HAP 1 (lane 1) some incorporation of radiolabeled dTMP into the 30-nt fragment was detected. This synthesis is likely to occur at the nicks originated by spontaneous hydrolysis of the abasic site. The AP site-containing vector is in fact fully converted into nicked forms following a 2-h incubation in the repair synthesis buffer (data not shown). In the presence of HAP 1 (lanes 2-7) spontaneous hydrolysis of the abasic site is negligible since the enzymatic cleavage reaches completion in 15 min (data not shown). When RF-C was omitted from the reaction (lane 2), only 1-nt repair intermediates were visible on the autoradiogram. In the absence of PCNA (lane 4), short repair intermediates, predominantly 1-2 nt long, were synthesized by pol ␦. When FEN1 was omitted from the repair reaction (lane 5), besides 1-nt repair intermediates, an increase of the labeled 30-mer product was observed. This increase is likely due to ligation of short repair intermediates which do not require FEN1 for repair completion (in the absence of DNA ligase the omission of FEN1 impedes synthesis of longer products, Fig. 2B, lane 1). These ligation events seem to require PCNA since no increase of the 30-nt product was observed when this accessory protein was omitted from the reaction (lane 4) or when its matchmaker RF-C was missing (lane 2). The synthesis of longer repair intermediates (beyond 4-nt long) required the presence of both PCNA and FEN1 (lane 6). When all the components are present with the exception of DNA ligase I (lane 6) a ladder of repair intermediates is visible on the gel. The most represented intermediates correspond to 1-2-nt replacement reactions. As already shown in Fig. 2B, synthesis overriding the HindIII site (30-nt product) is also abundant under these experimental conditions. When DNA ligase I is added to the repair reaction (lane 7) an increase in the yield of the 30-nt product is observed. Interestingly, in the presence of DNA ligase I, the repair intermediates that are still visible on the gel belong exclusively to very short synthesis products (1-4 nucleotides, see below). The weak background incorporation in the absence of Pol ␦ (lane 3) most likely mirrors a slight contamination of RF-C by a Pol. This contamination was too low to be detected by an RF-C-dependent DNA synthesis assay or by immunoblotting thus excluding the presence of pol ␦ and/or ⑀ (data not shown).
Pol ␦ was able to perform repair synthesis at AP sites supporting our earlier observation that partially purified pol ␦ and pol ⑀ from a pol ␤-deficient mouse fibroblast cell line (30), which were free of pol ␣, could support BER in vitro (6). For efficient repair synthesis pol ␦ required the five accessory components HAP 1, RF-C, PCNA, FEN1, and DNA ligase I. Omission of each component led to incomplete or no repair of the AP site.
Fine Mapping of the Repair Intermediates Is Consistent with the Function of DNA Ligase I in Ligating FEN1 Trimmed Repair Intermediates Synthesized by DNA Polymerase ␦-To determine the size of the ligated repair intermediates synthesized by pol ␦ we performed restriction mapping of the repair patches. As shown in Fig. 4, the incorporation of either dTMP (lanes 1-5) or dCMP (lanes 6 -9) clearly showed that, in the presence of DNA ligase I, the repair synthesis was targeted to the position where the AP site was originally located and extended up to a maximum of 10 nt 3Ј to the lesion (lack of radioactivity in the C fragments, lanes 3 and 8). No incorporation was detected 5Ј to the abasic site (B fragments, lanes 2 and  7). The only labeled products visible in these lanes comigrate with the 1-2-nt repair intermediates that are also present in lanes 1 and 6, respectively. From the quantitation of the radioactivity in the restriction fragments (Net cpm in Fig. 4) it emerged that the replacement reactions were mostly confined to 2-4-nt patches (almost 70% of the total repair synthesis). The remaining 30% of the total repair incorporation was due to nucleotide replacements extending up to the T residue located 7 nt 3Ј to the lesion (T 7 ). The fact that the 1-nt repair intermediates, which were quite abundant at the end of the repair reaction (see Figs. 2B and 3), were not or very inefficiently ligated is demonstrated by similar levels of radioactivity due to T 1 incorporation, 136 cpm (fragment E, lane 5) and C 2 incorporation, 162 cpm (as calculated by subtracting the radioactiv-ity present in fragment D, lane 9, from that of fragment A, lane 6). T 1 incorporation should exceed C 2 incorporation in the case of predominant 1-nt repair events. We thus conclude that the processive DNA synthesis by pol ␦ beyond the lesion, which overrides the HindIII site (Fig. 3), is abolished in the presence of DNA ligase I (Fig. 4). These results suggested that DNA ligase I acts as a "patch size mediator" for pol ␦ in BER, which is in accordance with our earlier findings that DNA ligase I selectively affects DNA synthesis by pol ␦ and pol ⑀ in the presence of PCNA (31). Alternatively, DNA ligase I could simply close the nicks at the moment when pol ␦ dissociates from the template (see also "Discussion"). . DNAs were digested with: BamHI-HindIII (fragment A) to release the 30-base pair fragment originally containing the lesion, BamHI-SalI (fragment B) to release the 12-bp fragment containing nucleotide residues 5Ј to the AP site, PstI-HindIII (fragment C) to release the 8-bp fragment containing nucleotide residues 3Ј to the AP site, HincII-PstI (fragment D) to release the 8-base pair fragment containing nucleotide residues 3Ј to the AP site and BamHI-HincII (fragment E) to release the 14-base pair fragment containing the AP site and nucleotide residues 5Ј to it. Proteins used in this experiment were: HAP 1 (2. Similarly to what was reported in the depletion experiment with pol ␦, in the absence of FEN1 (lane 5) an increase of the 30-nt product was observed. When DNA ligase I was omitted from the repair reaction (lane 6) longer repair intermediates approaching the 30-nt product were synthesized by pol ⑀. The repair synthesis due to spontaneous hydrolysis of the AP site in the absence of HAP1 (lane 1) as well as the background activity (lane 3) has already been discussed in the section of pol ␦. Finally, Fig. 5B demonstrates the differences between the repair intermediates produced by pol ␦ and ⑀ in the absence of DNA ligase I. Processive DNA synthesis overriding the HindIII restriction site was observed with both pol ␦ (lane 1) and pol ⑀ (lane 3). However, remarkable differences in the ladder of nucleotide incorporation events at the AP site were visible on the gel: pol ⑀ synthesizes a more homogeneous ladder whereas pol ␦ has a higher specificity for 1-2 nt incorporation.
Fine Mapping of the Repair Intermediates Is Consistent with the Function of DNA Ligase I in Ligating FEN1 Trimmed Repair Intermediates Synthesized by DNA Polymerase ⑀-The fine mapping of the repair patches made by pol ⑀ was performed using the same experimental strategy as applied to pol ␦-mediated repair. As shown in Fig. 6, the incorporation of either dTMP (lanes 1-5) or dCMP (lanes 6 -9) clearly showed that the repair synthesis was targeted, as in the case of pol ␦, to the position where the AP site was originally located and extended up to a maximum 10 nt 3Ј to the lesion (lack of radioactivity in the C fragments, lanes 3 and 8). No incorporation was detected 5Ј to the abasic site (B fragments, lanes 2 and 7). The labeled products visible in these lanes are the repair intermediates resulting from 1 to 4-nt gap filling. The same products are also present in lanes 1 and 6. From the quantitation of the radioactivity in the restriction fragments (Net cpm in Fig. 6) the size of the repair patches surprisingly turned out to be identical to that observed with pol ␦. The DNA synthesis events were indeed mostly confined to 2-4-nt patches with some incorporation (approximately 30%) extending up to T 7 . The intrinsic properties of pol ␦ and pol ⑀ which were displayed by the different ladders of repair intermediates produced in the absence of DNA ligase I were abolished in its presence thus . DNAs were digested with: BamHI-HindIII (fragment A) to release the 30-base pair fragment originally containing the lesion, BamHI-SalI (fragment B) to release the 12-base pair fragment containing nucleotide residues 5Ј to the AP site, PstI-HindIII (fragment C) to release the 8-base pair fragment containing nucleotide residues 3Ј to the AP site, HincII-PstI (fragment D) to release the 8-base pair fragment containing nucleotide residues 3Ј to the AP site and BamHI-HincII (fragment E) to release the 14-base pair fragment containing the AP site and nucleotide residues 5Ј to it. Proteins used in this experiment were: HAP 1 (2.5 ng), RF-C (0.8 units), pol ⑀ (0.1 units), PCNA (50 ng), FEN1 (25 ng), and DNA ligase I (200 ng). The levels of incorporation in the restriction fragments were measured by electronic autoradiography and corrected for DNA recovery (net cpm). Bottom, scheme of the restriction mapping.
confirming the role of this enzyme as the "patch size mediator" in BER.
PCNA-FEN1 Interaction Is Required for Efficient BER Mediated by DNA Polymerase ␦ and ⑀-Since it was previously shown that the damaged strand is excised by FEN1 in PCNAdependent BER (17), we wanted to investigate the functional role of this enzyme in our reconstituted system. As shown in Fig. 7, FEN1 is strictly required "to push" pol ⑀ synthesis

DISCUSSION
In this paper we describe the reconstitution of long patch BER with the purified human enzymes HAP 1, PCNA, RF-C, DNA pol ␦ or ⑀, FEN1, and DNA ligase I. Complete and efficient repair reaction was dependent on all six proteins. Omission of either PCNA or RF-C resulted in a complete elimination of longer repair intermediates. Synthesis was restricted mainly to 1-2 nt gap filling in the case of pol ␦ while a less stringent dependence on these factors was displayed by pol ⑀ which was able to synthesize 1-6 nt patches (Figs. 2B and 5A). Moreover, our data clearly show that pol ␦ is able to synthesize longer patches in the absence of DNA ligase I, starting at the position of the lesion and subsequently overriding the HindIII site. Pol ␦ forms a processive holoenzyme complex with its auxiliary factors PCNA and RF-C on primed single-stranded circular DNA templates in the presence of a single strand-binding protein such as replication protein A (RP-A). Pol ⑀ was shown to build up such a holoenzyme complex as well, but (similarly to what is observed for BER synthesis) in this case pol ⑀ is able to do some processive background synthesis even in the absence of the auxiliary factors (32). However, the situation is different and more complex in the case of BER. Here, the pol holoenzymes are not elongating primers on a single-stranded template, but participate in the replacement of the old damaged strand starting at the nick which was previously produced by the AP-specific endonuclease and proceeding in the 3Ј direction. The ability of pol ␦ holoenzyme to displace a downstream strand during DNA synthesis was previously shown in vitro in the presence of RP-A, using a gapped DNA template (33). However, in our in vitro assay, synthesis by pol ␦ beyond 1-2-nt patches was not only dependent on PCNA and RF-C, but also on the presence of FEN1. Pol ␦ holoenzyme is not able to displace the downstream strand alone. Therefore FEN1 seems to be the first actor at the incised AP site, excising nucleotides 3Ј to the lesion site, in order to allow the pol machinery to build up and to start DNA synthesis. Indeed, by means of an excision/ incision BER assay, Matsumoto and co-workers (17) showed excision by FEN1 at an incised AP site even in the absence of DNA synthesis. We show in our experimental system that, in the absence of DNA ligase I, processive DNA synthesis overriding the HindIII restriction site occurs with both pol ␦ and pol ⑀ but the ladder of repair intermediates produced by the two polymerases is significantly different. Since pol ⑀ was recently shown to interact in a dual mode with PCNA in primer binding and DNA synthesis (34) it might appear that this varying interaction pattern is responsible for the difference in the patch size distribution between pol ␦ and ⑀.
The tendency of pol ␦ and pol ⑀ to perform processive synthesis was completely abolished when DNA ligase I was present in the reaction. Fine mapping experiments with various combinations of restriction enzymes clearly showed that both pols ␦ and ⑀ perform only limited repair synthesis with patches that never exceed 10 nt. In previous work (31) we showed that DNA ligase I is able to inhibit the strand displacement activity of pol ␦ on a gapped template even in the absence of a ligatable intermediate. Therefore it seems that this effect is not due to simple competition for the DNA substrate between the pol and the ligase but more the result of a complex regulatory mechanism that is most probably mediated by PCNA. This might be of importance also in vivo. In a living cell, nick translation beyond a lesion site should be limited since it could heavily interfere with other important reactions during gene expression or DNA replication. Moreover the resynthesis events at damaged sites should involve a minimal number of nt to preserve template DNA from the introduction of pol errors. Finally, the coupling of repair synthesis with ligation avoids that unligated repair intermediates could destabilize the highly organized structure of the chromatin by sealing them as quickly as possible.
The finding that the repair patches made by pol ␦ and ⑀ are mostly confined to 2-4 nt has important implications as far as the interpretation of previous repair experiments performed with whole cell extracts is concerned. By using crude extracts from human and mouse cells a highest frequency of 2-4 nt replacement reactions was detected among the so called "long patches" detected after in vitro repair of either AP sites (4) or oxidized purines like 8-oxo-7,8-dihydroguanine (35). This observation is in agreement with an important role for pol ␦/⑀ in long patch BER at these lesions in the cells. However, recently Dianov et al. (36) have found that pol ␤ can also participate in long patch BER at uracil residues. In this case the excision products appear to be predominantly 5ЈdRP-3-nt. It might well be that the relative contribution of the different polymerases to the long patch repair varies depending on the cell cycle stage (37) or on the cell type-specific pol expression levels (38).
In our previous work (6) we already showed that, besides pol ␤, pol ␦ and pol ⑀ are also able to replace a single nucleotide at the lesion site in vitro, but the completion of the repair reaction was delayed by the slow processing of the 5Ј-deoxyribose phosphate moieties. The addition of the ␤-elimination catalyst spermine was in fact required to complete short patch BER with pol ␦ and pol ⑀ containing partially purified fractions from a pol ␤-deleted mouse cell line. The accumulation of repair intermediates that results from the 1-nt gap-filling reaction was also observed in this work with highly purified proteins. Again, these intermediates are not or only inefficiently ligated. This is in agreement with the previous finding that FEN1 is unable to remove a deoxyribophosphate residue alone but only together with its 3Ј-adjacent nucleotide or as a part of an oligonucleotide (39). However, some of them are ligated in the reconstituted system when DNA ligase I, RFC and PCNA are present (FEN1 is not required). Interestingly an increase of the ligated product was also observed in our previous study (6) when PCNA was added to the reaction. The importance of the interaction between DNA ligase I and PCNA (31) in this phenomenon is strongly suggested. What is the role of these unligated short patch intermediates? There are two possible explanations: first they represent in vitro artifacts and are not important for the living cell. Second, this kind of intermediate can only be completely processed in the presence of an additional enzyme which carries a dRPase function or molecules which can catalyze ␤-elimination reactions such as histones.
The role of the interaction between FEN1 and PCNA in BER has been studied using a reconstituted system with the proteins from Xenopus laevis (17). The results showed that FEN1 is the factor responsible for the excision of the 5Ј-incised AP site in the PCNA/pol ␦-dependent BER pathway. As already mentioned, these authors were able to demonstrate that DNA synthesis was not required for the excision activity of FEN1 in the presence of PCNA and RF-C. More recently, the same group provided evidence that during the long patch BER PCNA facilitates excision of the damage-containing strand by interacting with FEN1 (18). Our data show that FEN1 is required for repair synthesis extending beyond 1 nt and support a stimulatory effect of PCNA in the excision step when DNA ligase I is present. Additionally PCNA stimulates more extensive synthesis events by pol ⑀ or ␦ in the absence of DNA ligase I. These long patches that may result from extensive nick translation reaction are a sort of "abortive" repair products since DNA ligase I impedes their formation. However, also in this case the stimulatory effect of PCNA on DNA synthesis is mediated by its interaction with FEN1. Therefore the strict PCNA dependence of the long patch BER could also reside on PCNA-FEN1 interaction and not only on PCNA stimulation of primer recognition and/or DNA synthesis by pol ⑀ or ␦ as in the case of nucleotide excision repair (40).
In conclusion, the presented results suggest that long patch BER by pol ␦ and ⑀ is not a simple step-by-step mechanism like it was shown to be the case for short patch BER by pol ␤ at least on a linear substrate (41) but rather the action of a highly organized multiprotein machinery similarly to what has been identified for other DNA transactions such as DNA replication, nucleotide excision repair, and postreplicative mismatch repair. Since all the proteins of long patch BER (except HAP 1) were shown to directly interact with PCNA we hypothesize that PCNA might act as coordinator for their actual functions in BER.