Human Mismatch Repair

Bidirectional mismatch repair directed by a strand break located 3′ or 5′ to the mispair has been reconstituted using seven purified human activities: MutSα, MutLα, EXOI, replication protein A (RPA), proliferating cell nuclear antigen (PCNA), replication factor C (RFC) and DNA polymerase δ. In addition to DNA polymerase δ, PCNA, RFC, and RPA, 5′-directed repair depends on MutSα and EXOI, whereas 3′-directed mismatch correction also requires MutLα. The repair reaction displays specificity for DNA polymerase δ, an effect that presumably reflects interactions with other repair activities. Because previous studies have suggested potential involvement of the editing function of a replicative polymerase in mismatch-provoked excision, we have evaluated possible participation of DNA polymerase δ in the excision step of repair. RFC and PCNA dramatically activate polymerase δ-mediated hydrolysis of a primer-template. Nevertheless, the contribution of the polymerase to mismatch-provoked excision is very limited, both in the purified system and in HeLa extracts, as judged by in vitro assay using nicked circular heteroplex DNAs. Thus, excision and repair in the purified system containing polymerase δ are reduced 10-fold upon omission of EXOI or by substitution of a catalytically dead form of the exonuclease. Furthermore, aphidicolin inhibits both 3′- and 5′-directed excision in HeLa nuclear extracts by only 20–30%. Although this modest inhibition could be because of nonspecific effects, it may indicate limited dependence of bidirectional excision on an aphidicolin-sensitive DNA polymerase.

Mismatch repair stabilizes the genome by correction of DNA biosynthetic errors, by ensuring the fidelity of genetic recombination, and in mammalian cells by participation in the cellular response to some classes of DNA damage. Basic features of the reaction responsible for replication error correction are conserved from bacteria to mammals. Study of partial reactions has demonstrated that repair can be divided into three major steps: mismatch recognition, excision, and repair DNA synthesis (1)(2)(3)(4). Escherichia coli mismatch repair has been reconstituted using purified components. In the bacterial reaction MutS is responsible for mismatch recognition and recruits MutL to the heteroduplex in an ATP-dependent manner (5)(6)(7)(8)(9)(10)(11)(12)(13). Assembly of the MutL⅐MutS⅐heteroduplex complex activates the d(GATC) endonuclease activity of MutH, which incises the unmethylated strand of the heteroduplex (14). This strand break serves as an entry point for the exci-sion system, which is comprised of DNA helicase II and an appropriate single strand exonuclease (15)(16)(17)(18). A 3Ј to 5Ј exonuclease is required when the nick that directs excision is located 3Ј to the mismatch, whereas a 5Ј to 3Ј hydrolytic activity is necessary when the strand break is 5Ј to the mispair. DNA polymerase III holoenzyme is sufficient to repair the ensuing gap, and covalent integrity is restored to the helix by DNA ligase (19).
Analysis of nuclear extracts of human cells has indicated a similar excision repair mechanism for nick-directed mismatch correction in higher cells (20,21), and several reconstituted systems that rely on purified human proteins have been described that support nick-directed mismatch-provoked excision (22,23). In a simple system comprised of MutS␣ (MSH2⅐MSH6 heterodimer), EXOI, 3 and RPA, hydrolysis is mismatch-provoked and terminates upon mismatch removal, but always proceeds 5Ј to 3Ј from the nick that directs excision (22). Although MutL␣ (MLH1⅐PMS2 heterodimer) is not required for EXOI activation in this system, it does enhance the mismatch dependence of the reaction. Supplementation of MutS␣, MutL␣, EXOI, and RPA with PCNA and RFC yields a system that supports bidirectional excision, i.e. hydrolysis that is directed in an appropriate manner by a nick located either 5Ј or 3Ј to the mismatch (23). In contrast to the simpler 5Ј to 3Ј reconstituted reaction, 3Ј-directed excision in this 6-component system is absolutely dependent on MutL␣. RFC is thought to play two roles in 3Ј-directed excision (23). It functions as a clamp loader, with the loaded form of PCNA necessary to activate 3Ј-directed excision, but it also acts directly to suppress EXOI mediated 5Ј to 3Ј hydrolysis from a strand break located 3Ј to the mismatch.
Whereas analysis of these purified systems has yielded useful information on mechanism, other activities may play significant roles in mismatch repair in the cell. For example, HMGB1 has been implicated in mismatch repair in nuclear extracts (24), although it is not required in the purified systems (23). Furthermore, analysis of EXOI-depleted extracts and EXOI-deficient cell lines has indicated that other excision activities may function in a manner that is redundant with respect to EXOI (22,(25)(26)(27).
Depletion of human cell extracts and genetic studies in yeast have also implicated DNA polymerase ␦ in eukaryotic mismatch repair, with the former studies indicating involvement in the repair synthesis step of the reaction (28,29). This enzyme displays both polymerase and 3Ј-exonuclease activities (30), with the human enzyme comprised of at least four subunits (p125, p66, p50, and p12) (31)(32)(33)(34). Inasmuch as polymerase ␦ depends on PCNA for processivity (35), it is not surprising that PCNA has been implicated in the repair synthesis step of eukaryotic mismatch correction (36).
Analysis of mutant yeast strains with defects in the polymerase ␦ editing exonuclease has suggested that this activity may also play a role in mismatch-provoked excision (37,38). Although other studies have suggested that the genetic instability associated with such defects may be because of aberrant checkpoint activation (39), the observation that 3Ј-directed mismatch-provoked excision in HeLa extracts is partially inhibited by aphidicolin has also been attributed to editing exonuclease involvement in the excision step of mismatch repair (21).
In this study we demonstrate that a 7-component system comprised of MutS␣, MutL␣, EXOI, RPA, RFC, PCNA, and DNA polymerase ␦ supports bidirectional mismatch repair directed by a strand break located either 3Ј or 5Ј to the mispair. Whereas repair directed by a 5Ј-strand break can occur in the absence of MutL␣, this activity is required for mismatch correction directed by a 3Ј-strand break. Repair DNA synthesis by polymerase ␦ depends on the presence of RPA, PCNA, and RFC for optimal activity. Importantly, EXOI is required for both 3Ј-and 5Ј-directed repair, and no significant involvement of the polymerase ␦ editing exonuclease is evident in mismatch-provoked excision in the purified system. Analysis of aphidicolin sensitivity of 3Јand 5Ј-directed excision reactions in HeLa nuclear extracts also indicates that the editing functions of polymerases ␦ or ⑀ play little if any role in mismatch-provoked excision as scored on nicked circular heteroduplex DNAs.

MATERIALS AND METHODS
Proteins and Extracts-All fractionation was performed at 0 to 4°C in the presence of a set of protease inhibitors at the following final concentrations: 0.5 g/ml aprotinin, 1 g/ml E64, 1 g/ml leupeptin, 5 g/ml pepstatin, 100 g/ml Pefabloc, and 0.1% phenylmethylsulfonyl fluoride (relative to a saturated solution in isopropyl alcohol). With the exception of phenylmethylsulfonyl fluoride (U. S. Biochemicals), all protease inhibitors were from Roche. Human MutS␣, MutL␣, EXOIb, hydrolytically defective EXOIb D173A, RPA, PCNA, and RFC were isolated according to Dzantiev et al. (Ref. 23, and references therein). HeLa nuclear extracts were prepared as described (40). Partially purified human DNA polymerase ␦ was isolated from HeLa cells as described previously (28). T7 DNA polymerase was from Amersham Biosciences.
The column was eluted with a gradient of 250 -800 mM NaCl in Buffer C. Polymerase ␦ eluted at 0.51 M KCl and the final preparation was estimated to be ϳ70 -80% pure by Coomassie staining and contained all four subunits of polymerase ␦ as judged by Western analysis.
Goat polyclonal antibody against the catalytic subunit of polymerase ␦ p125 (C-20) was from Santa Cruz Biotechnology. Rabbit polyclonal anti-p50 against full-length p50 protein, rabbit polyclonal anti-p68 against peptide GHGPPASKQVSQQPKG, and rabbit polyclonal anti-p12 against peptide GLEPPPEVWQVLKYHPGD (31,33) were generous gifts from Marietta Lee (New York University).
Construction of DNA Substrates-5Ј-GT and 3Ј-GT heteroduplexes (and corresponding G⅐C and A⅐T homoduplex control DNAs, respectively) were prepared from f1 bacteriophage derivatives as described (20,23,41). The 5Ј-heteroduplex contained single strand break in the complementary DNA strand 128 base pairs 5Ј to the mismatch (shorter path in the circular DNA). 3Ј-Heteroduplex DNA contained single strand break in the complementary strand 141 base pairs 3Ј to the mispair. Gapped homoduplex DNA was prepared by a similar method (22,41). The latter DNA contained a 220-nucleotide gap in the complementary strand between the Sau96I and SwaI cleavage sites.
Primed single-stranded DNAs for assay of the polymerase ␦ editing exonuclease were prepared by hybridization of a 5Ј-32 P-labeled 35-residue oligonucleotide to circular single-stranded phage f1 MR3 DNA (6). Oligonucleotide C5650 (d(CCGAATTTCTAGACTCGAGAGCTT-GCTAGCAATTC)) is a perfect complement to the viral DNA, whereas the oligonucleotide C5650-GA (d(CCGAATTTCTAGACTC-GAGAGCTTGCTAGCAATGA)) yields a hybrid in which the 3Ј-terminal dinucleotide is unpaired. Annealing was at 60°C for 30 min in 50 mM NaCl, 10 mM Tris-HCl, 1 mM EDTA, pH 8.0, and the molar ratio of circular viral strand to oligonucleotide was 1.1:1 (110 nM viral DNA and 100 nM oligonucleotide).
Assay of Mismatch-provoked Excision and Repair-Mismatch repair and mismatch-provoked excision in nuclear extracts was performed as described previously (20,25,40). For analysis of aphidicolin inhibition, reactions were supplemented with 90 M drug. A 27 mM aphidicolin (A. G. Scientific) stock, which was used within 2 months, was prepared by dissolving the compound in Me 2 SO and storing as aliquots at Ϫ80°C. The stock solution was diluted to 900 M with H 2 O immediately prior to use.
Excision and repair synthesis tract end points were localized by indirect end labeling (23,25,41). Briefly, reaction products were cleaved with an appropriate restriction endonuclease, resolved through alkaline-agarose, and transferred to Hybond N ϩ membranes, which were probed with 32 P-labeled oligonucleotides that hybridize to complementary strand sequences on the appropriate side of the restriction enzyme site. 32 P-Labeled probe V5891 (d(ATTTAACGCGAATTTTAACAA)) was used to localize the 3Ј-termini on 5Ј-heteroduplex products following SspI digestion. 32 P-Labeled probe V194 (d(AACATGTTGAGCTA-CAGCACC)) was used to map the 3Ј termini on 3Ј-heteroduplex products after BstYI digestion.
Fine Mapping of Excision and Repair DNA Synthesis Tracts-Excision and repair reactions in nuclear extracts and with purified proteins were performed as described above. Excision reactions using HeLa nuclear extract were supplemented with 90 M aphidicolin to inhibit limited DNA synthesis supported by trace dNTP levels present in extract preparations (20). Repair reactions contained 20 M ␣-thio-dCTP (TriLink), 180 M dCTP, and 200 M each of dATP, dGTP, and dTTP. After incubation for 8 min at 37°C, reactions were processed as described above and DNA digested with SspI. The 10-l restriction digestion reaction mixture was diluted to 180 l with TE (10 mM Tris-HCl, pH 7.5, 1 mM EDTA) and divided into two 90-l aliquots. One aliquot was supplemented with 10 l of 50 mM iodine in ethanol, which cleaves at sites of ␣-thio-dCMP incorporation (42), with the other aliquot serving as control. After 5 min at room temperature, DNA was recovered by ethanol precipitation, resuspended in 5 l of 10 mM Tris-HCl, pH 8.0, 1 mM EDTA and supplemented with 10 l of 90% formamide (v/v), 89 mM Tris borate, 1 mM EDTA. Reaction products were separated on a 6% denaturing polyacrylamide gel, electrotransferred to a Hybond N ϩ membrane, and probed with V5891.
Assay of Polymerase ␦ Editing Exonuclease-Editing exonuclease was assayed using primed single strands that were either perfectly paired or contained a 3Ј-terminal unpaired dinucleotide. Buffer conditions were identical to those used for assay of reconstituted mismatch-provoked excision (23)

In Vitro Reconstitution of Bidirectional Mismatch
Repair-A singlestrand break located 3Ј or 5Ј to the mismatch is sufficient to determine the stand specificity of mismatch repair or mismatch-provoked excision in extracts of mammalian cells (20,21). Several reconstituted systems have been recently described that support mismatch-provoked excision using purified human proteins. 5Ј-Directed excision can be mediated by EXOI in a reaction that is controlled by MutS␣ and RPA, with specificity enhanced by MutL␣ (22). A 6-component system consisting of MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA is sufficient to support mismatchprovoked excision directed by a strand break located either 3Ј or 5Ј to the mismatch (23). Because previous work has implicated DNA polymerase ␦ in the repair synthesis step of mismatch correction in HeLa extract fractions (28), we have asked whether supplementation of this 6-component system with polymerase ␦ will yield a system that supports mismatch repair.
To address this question, we have used the recombinant, four-subunit form of the polymerase (34). As judged by immunoblotting, the preparation used ("Materials and Methods") contained all four subunits (p125, p66, p50, and p12) present in polymerase ␦ isolated from HeLa cells, with three of these subunits corresponding to major protein bands in the elecrophoretic profile of the isolated recombinant activity (Fig.  1A). The biochemical properties of this preparation are consistent with those described for this enzyme (33,34). As expected (Fig. 1B), the recombinant enzyme was activated by PCNA on a poly(dA)⅐oligo(dT) primer-template (34,35), with the level of activity similar to that observed with comparable amounts of DNA polymerase ␦ partially purified from HeLa cells (28). The recombinant enzyme also supported efficient synthesis on a gapped circular DNA that is similar to the excision intermediate produced during the course of in vitro mismatch repair (Fig. 1C). Although gap repair displayed only a modest dependence on RPA, RFC and PCNA were essential for the reaction, which was abolished by aphidicolin, consistent with previous findings (33,35).
Supplementation of MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA with recombinant DNA polymerase ␦ is sufficient to reconstitute nickdirected mismatch rectification of 5Ј-and 3Ј-heteroduplexes in vitro. As shown in Fig. 2B, repair of both 5Ј-and 3Ј-GT heteroduplexes was dependent on PCNA (lanes 1 and 7), RFC (lanes 2 and 8), and DNA polymerase ␦ (lanes 4 and 10), with RPA acting in a stimulatory fashion (compare lane 3 with 5, and lane 9 with 11). Aphidicolin inhibited repair of both heteroduplexes (lanes 6 and 12), an effect that we attribute to inhibition of the DNA synthetic activity of polymerase ␦ (see below). Because excision on a 5Ј-heteroduplex can occur in the absence of PCNA and RFC (22,23), the requirements for these two proteins in 5Ј-heteroduplex repair may solely reflect their involvement in the repair synthesis step. However, because 3Ј-directed excision requires both PCNA and RFC (23), reconstituted 3Ј-heteroduplex repair presumably depends on function of the two proteins in both the excision and repair synthesis steps of the reaction. Such a multifunctional role for RFC accounts for the anomalous product produced on the 3Ј-substrate in the absence of this activity (Fig. 2B, lane 8). Absence of RFC not only blocked repair DNA synthesis (Fig. 1), but also prevented activation of 3Ј-directed excision (23). Under these conditions, mismatch-provoked hydrolysis occurs with 5Ј to 3Ј polarity (23), yielding the product observed in lane 8.
MutS␣, MutL␣, and EXOI requirements for reconstituted repair mirror those previously observed for reconstituted excision (22,23). Thus, repair of the 5Ј-GT heteroduplex occurred efficiently in the absence of MutL␣, although MutS␣ and EXOI were required for the FIGURE 2. Reconstitution of bidirectional mismatch repair with purified human proteins. A, the assay for mismatch repair utilizes a 6,440-base pair circular heteroduplexes containing a GT mismatch within overlapping restriction sites for Hin-dIII and XhoI (6, 23) and a site-specific nick in the complementary DNA strand 141 base pairs 3Ј to the mismatch (shorter path, 3Ј-GT) or 128 base pairs 5Ј to the mispair (5Ј-GT). Because repair occurring on the incised strand renders the DNAs sensitive to HindIII, cleavage of repaired molecules with HindIII and ClaI yields two rapidly migrating fragments (arrows in panels B and C). The heteroduplexes contain a NheI site 5 base pairs from the mismatch. Because the gaps produced by mismatch-provoked excision span this site, conversion of the heteroduplex to an NheI-resistant form provides a simple assay for mismatch-provoked excision (23,25). B, reactions ("Materials and Methods") contained a 3Ј-GT or 5Ј-GT heteroduplex, MutS␣, MutL␣, and EXOI. DNA polymerase ␦, PCNA, RFC, RPA, and aphidicolin were present as indicated. C, reactions as in panel B contained DNA polymerase ␦, PCNA, RFC, and RPA. MutS␣, MutL␣, and EXOI were included as indicated. Although not shown, no detectable mismatch rectification was observed on the continuous strand of 3Ј-or 5Ј-heteroduplexes when incubated with a complete, 7-component system. reaction (Fig. 2C, lanes 1-3). As can be seen, significant background repair on the 5Ј-substrate occurred in the absence of MutS␣ (about 15% of that observed in the presence of the mismatch recognition protein; lanes 1 and 4). This is consistent with the previous finding that reconstituted 5Ј-directed excision displays a mismatch dependence of only 3-5-fold, an effect attributed to deficiency of as yet unidentified component(s) that contribute to the specificity of 5Ј-directed excision (22,23). By contrast, mismatch rectification directed by a 3Ј-strand break required both MutS␣ and MutL␣ (lanes 5 and 6) and was reduced 10-fold upon omission of EXOI (lane 7). It is also important to note that the repair reaction shown in Fig. 2 was strongly directed to the nicked DNA strand; no detectable repair on the continuous strand of the heteroduplex was observed in the presence of the complete 7-component system (not shown).
Dependence of Excision and Repair on DNA Polymerase ␦-As noted above, omission of EXOI from the reconstituted repair system largely abolished both 3Ј-and 5Ј-directed mismatch rectification, suggesting that the 3Ј to 5Ј hydrolytic function of DNA polymerase ␦ does not contribute significantly to mismatch-provoked excision in this system. A more comprehensive analysis of the dependence of the reconstituted reaction on the polymerase is shown in Fig. 3. In the presence of MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA, mismatch repair increased continuously with polymerase concentration up to 4 nM for both 3Ј-and 5Ј-heteroduplexes (Fig. 3A). In the absence of dNTPs, neither the 3Јnor the 5Ј-directed excision was altered to a significant degree by the presence of the polymerase. Identical results were obtained when polymerase ␦ isolated from HeLa cells was substituted for the recombinant enzyme used in these experiments (not shown). These findings indicate that the contribution of the polymerase ␦ editing exonuclease to excision in the purified system is limited.
To explore the possibility that EXOI hydrolysis might mask a significant polymerase ␦ contribution to excision, we performed similar experiments using hydrolytically defective EXOI D173A. This mutant exonuclease is well behaved and purifies like the wild type protein (23). It also inhibits excision in the presence of wild type EXOI, suggesting that it supports assembly of repair complexes. As shown in Fig. 3A, substitution of the mutant protein for wild type EXOI abolished excision, irrespective of the presence of the DNA polymerase, indicating that the latter activity plays little if any role in mismatch-provoked hydrolysis in the purified system.
Indirect end labeling was also used to assess effects of DNA polymerase ␦ on the fate of heteroduplex 3Ј termini during the course of reconstituted excision and repair reactions (Fig. 3, panels B and C). Under excision conditions in the absence of dNTPs, no significant 3Ј to 5Ј degradation was observed on 5Ј-homoduplex or heteroduplex substrates in the absence or presence of polymerase ␦ (Fig. 3B, lanes 1-3). However, mismatch-dependent extension of the 3Ј terminus of the strand was readily evident when the 5Ј-heteroduplex was incubated with increasing amounts of polymerase ␦ in the presence of dNTPs (Fig.  3B, lanes 5-8). This repair DNA synthesis terminated 100 -150 nucleotides beyond the location of the mispair, consistent with previous localization of 5Ј-directed excision tracts in HeLa nuclear extracts (20). Extension of the 3Ј terminus of virtually all molecules of the homoduplex control was also observed, but repair synthesis tracts in this case were much shorter (extension of the 3Ј terminus by about 65 nucleotides, compare lanes 1 and 4). Extension of the homoduplex 3Ј terminus  Fig. 2A) were performed as described under "Materials and Methods " except that hydrolytically defective EXOI D173A was substituted for EXOI as indicated. DNA polymerase ␦ concentration was varied as shown, and dNTPs were omitted from those reactions used to score excision. Mismatch repair (solid lines) was scored by cleavage with ClaI and HindIII. Mismatch-provoked excision was determined by cleavage with ClaI and NheI (dashed and hyphenated lines indicate excision occurring in the presence of wild type EXOI or hydrolytically defective EXOI D173A, respectively). B, reactions using a 5Ј-heteroduplex or an otherwise identical G⅐C homoduplex control (lanes 1 and 4) were performed as in panel A in the absence (lanes 1-3) or presence of dNTPs (lanes 4 -8). DNA polymerase ␦ concentration was varied as indicated. Reaction products were digested with SspI, denatured, resolved by electrophoresis through 1.8% alkaline-agarose, and probed with 32 P-labeled oligonucleotide V5891 ("Materials and Methods"). As illustrated schematically in the diagram on the right, this oligonucleotide hybridizes to the 5Ј-end of the incised strand of the SspI fragment that contains the mismatch. This probe thus permits visualization of the fate of the 3Ј terminus at the strand break. C, reactions in the absence (lanes 1-5) or presence of dNTPs (lanes 6 -10) and indicated amounts of DNA polymerase ␦ were performed as in panel B, except that a 3Ј-GT heteroduplex or the corresponding A⅐T homoduplex control (lanes 1 and 6) were used. After cleavage with BstYI and denaturation, products were separated and probed with 32 P-labeled oligonucleotide V194 ("Materials and Methods") to monitor the fate of the 3Ј termini in the incised strand of the heteroduplex.
in this manner may be because of conversion of the nick to small gaps via limited nonprocessive 5Ј to 3Ј hydrolysis by EXOI (22).
We have previously shown that 3Ј-directed hydrolytic tracts produced on heteroduplex DNA in the reconstituted excision system display a broader distribution of sizes than those observed in HeLa nuclear extracts (23). As shown in Fig. 3C, similar results are obtained in the absence of dNTPs when MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA are supplemented with DNA polymerase ␦ (Fig. 3C, lanes 2-5). As can be seen, hydrolytic tracts in the purified system extend from 100 to 500 nucleotides beyond the mismatch, whereas in HeLa extracts these tracts extend to only about 150 nucleotides beyond the mispair (20). Nevertheless, excision in this manner was highly mismatch-dependent (Fig.  3C, compare lane 1 with lanes 2-5), and DNA polymerase ␦ was without significant effect on either the extent of excision or the distribution of excision end points. In the presence of dNTPs, DNA products increased in size with increasing amounts of polymerase (lanes 7-10), presumably reflecting extension of recessed 3Ј termini that result from excision. At the higher concentrations of polymerase ␦, these DNA synthesis tracts extend to and just beyond the mismatch, within the error of the gel method used (lanes 8 -10). As observed with the 5Ј-homoduplex control (Fig. 3B, lane 4), significant extension of the 3Ј terminus of the strand break also occurred with 3Ј-homoduplex control (Fig. 3C, compare lanes 1 and 6). As noted above, this may be a consequence of limited nonspecific hydrolysis by EXOI.
Fine Structure Analysis of Repair Tracts-Excision repair tracts produced in HeLa nuclear extracts and in the reconstituted repair system were compared using high resolution sequencing gels. In these experiments excision tracts were determined under conditions of DNA syn-thesis block (20), whereas repair synthesis tracts were analyzed by partial substitution of [␣]S dCTP for dCTP. The latter method exploits the fact that sites of dCMP[␣]S incorporation can be visualized by virtue of the sensitivity of phosphothiodiester bonds to cleavage with iodoethanol (42). After digestion with SspI, which yields a 744-bp fragment spanning the mismatch and strand break, DNA products were treated with ethanol or iodoethanol, resolved on a urea polyacrylamide gel, and products were visualized by indirect end labeling using a [ 32 P]oligonucleotide complementary to the 5Ј-end of the incised strand of the SspI fragment. As shown diagrammatically in Fig. 4, this method monitors the fate of the 3Ј termini on the incised DNA strand. Fig. 4A shows such an analysis of excision and repair in HeLa nuclear extracts. Results obtained with 5Ј-heteroduplex and homoduplex controls are shown in pairs in lanes 4 -5, 8 -9, 12-13, and 16 -17. With the exception of ligation of a subset of molecules to produce a 744-nucleotide product, 3Ј termini in the majority of the 5Ј DNAs were stable in HeLa extract under conditions of DNA synthesis block, although very limited hydrolysis occurred on the 3Ј side of the strand break with heteroduplex, but not homoduplex DNA (compare lanes 4 and 5). As expected, the distribution of these products was unaffected by iodoethanol treatment (lanes 12 and 13). Under conditions permissive for repair DNA synthesis, unligated intermediates resulting from extension of 3Ј termini were produced on heteroduplex, but to a much lesser extent on homoduplex DNA (lanes 8 and 9). Iodoethanol treatment resulted in extensive fragmentation of the heteroduplex product within a region extending from the strand break to about 150 nucleotides beyond the original location of the mismatch (compare lane 16 with lane 8), implying that repair DNA synthesis occurred throughout this region. Iodo-  5 and 10 -13) were performed in a similar manner except that dNTPs were omitted and reactions were supplemented with 90 M aphidicolin to suppress DNA synthesis supported by endogenous nucleotides present in extracts (20). After digestion with SspI, DNA samples were untreated (lanes 2-9) or treated with 5 mM iodine in ethanol (lanes 10 -17), products were resolved on a denaturing gel and probed with 5Ј-32 P-labeled oligonucleotide V5891 ("Materials and Methods"). As illustrated schematically in the diagrams to each side of the gel, this probe hybridizes to the incised complementary strand adjacent to the SspI site and can be used to locate 3Ј termini on both 3Ј-and 5Ј-DNA substrates. Numerical coordinates shown in the diagram on the right indicate fragment size in nucleotides (left axis) or distance from the mismatch (right axis). The 744-nucleotide species is the product of ligation.  17 with lanes 9 and 16). This observation is in agreement with the previous finding that some background repair DNA synthesis occurs on homoduplex DNA in HeLa nuclear extracts (40).
Similar results were obtained with 5Ј-substrates in the reconstituted reaction (Fig. 4B), although differences from the HeLa extract reaction are evident in electrophoretic profiles because ligation cannot occur in the purified system. As in HeLa extracts, limited hydrolysis to the 3Ј side of the strand break is evident on heteroduplex DNA when repair synthesis was blocked by omission of dNTPs (lanes 4 -5 and lanes 12-13).
In the presence of dNTPs, about 90% of the 3Ј termini in the 5Ј-GT heteroduplex were extended, with repair synthesis intermediates apparent in the region from the strand break to well beyond the mispair (lane 8). The fact that some repair synthesis intermediates fail to reach the mismatch suggests that DNA polymerase ␦ may be limiting in this system, consistent with the results shown in Fig. 3A. As can be seen in lane 9, a large fraction of 3Ј termini in the homoduplex control were also extended in this system, but extension in this case was only about 50 nucleotides as compared with the much more extensive elongation that occurred on the heteroduplex. As mentioned above, this probably reflects nonprocessive 5Ј to 3Ј hydrolysis by EXOI from the homoduplex strand break, an effect that has been observed earlier (22). Treatment with iodoethanol revealed the sites of dCMP[␣]S incorporation because of these DNA biosynthetic events (lanes 16 and 17).
Similar experiments were performed with 3Ј-heteroduplex and homoduplex DNAs, but results in this case are more difficult to interpret because excision products that fail to undergo elongation by repair synthesis cannot always be distinguished from intermediates that undergo partial elongation. As observed previously (20), incubation of 3Ј substrates with HeLa nuclear extract under conditions of DNA synthesis block resulted in mismatch-dependent production of excision intermediates that extend to at least 200 nucleotides beyond the mispair (Fig. 4A, lanes 2 and 3). Bands corresponding to these excision products are generally attenuated under conditions permissive for DNA synthesis (lane 6), suggesting that the excision products shown in lane 2 serve as precursors for repair DNA synthesis. This idea is consistent with the fact that treatment of the repair products (lane 6) with iodoethanol resulted in extensive fragmentation throughout the region where excision end points are observed (compare lanes 2, 6, and 14). The homoduplex control also displays some iodoethanol sensitivity, but to a much lower degree than that observed with the GT heteroduplex (compare lanes 14 and 15). As mentioned above, significant background repair synthesis is known to occur on homoduplex DNA in HeLa nuclear extracts (40).
Comparable experiments with the 3Ј-heteroduplex in the purified system are shown in Fig. 4B. Whereas the homoduplex control was not subject to significant hydrolysis in the reconstituted reaction in the absence of dNTPs, extensive excision occurred on the heteroduplex under these conditions (lanes 2 and 3). Supplementation of reconstituted reactions with dNTPs resulted in appearance of a distinct product spectrum (compare lanes 2 and 6), with the major product located about 100 nucleotides 3Ј to the original location of the mismatch. This suggests that excision intermediates like those in lane 2 are subject to polymerase ␦ elongation to yield the repair synthesis products of lane 6.  14 with lanes 2 and 6). As mentioned above, significant dCMP[␣]S incorporation occurred on the 3Ј-homoduplex in HeLa extract (Fig. 4A,   lane 15). By contrast, incorporation of the thionucleotide into the homoduplex control in the defined system was almost nonexistent in the region 3Ј to the strand break (Fig. 4B, compare lanes 14 and 15). However, as observed with the 5Ј-homoduplex, the 3Ј terminus at the homoduplex strand break was also subject to modest elongation in the purified system (compare lanes 3 and 7).
Specific Requirement for Polymerase ␦ in Repair DNA Synthesis-A specific requirement for polymerase ␦ in mismatch repair has been suggested based on extract fractionation studies (28), and genetic studies have implicated the POL32 subunit of yeast polymerase ␦ in mismatch repair in this organism (29). To assess the specificity of the requirement for DNA polymerase ␦ in the purified system, we tested the ability of the DNA polymerase encoded by bacteriophage T7 to substitute for polymerase ␦ in the reconstituted mismatch repair system. The amount of T7 polymerase used in this analysis was determined based on its ability to support repair of the 220-nucleotide gap in the circular DNA used in Fig. 1C. As judged by this method, 4 fmol of the T7 enzyme is comparable in activity to the 80 fmol of DNA polymerase ␦ (in the presence of RFC and PCNA) that is used in the reconstituted mismatch repair system (not shown).
As demonstrated above, repair does not occur when DNA polymerase ␦ is omitted from the reconstituted system (Fig. 2B). However, if DNA products produced in the absence of the polymerase were deproteinized and then incubated with the enzyme in the presence of RFC, PCNA, and RPA, repaired molecules were produced with an efficiency similar to that of the complete system for both 3Ј-and 5Ј-heteroduplexes (Fig. 5A). Substitution of T7 DNA polymerase for polymerase ␦ in the reconstituted repair system yielded different results (Fig. 5B). As mentioned above, MutS␣, MutL␣, EXOI, and RPA are sufficient to FIGURE 5. T7 DNA polymerase cannot substitute for DNA polymerase ␦ in mismatch repair. A, reactions (20 l) were carried out in two stages. Stage I reactions containing 24 fmol of 5Ј-GT or 3Ј-GT heteroduplex DNA were performed as described ("Materials and Methods") in the presence or absence of 80 fmol of DNA polymerase ␦ as indicated. Incubation was for 8 min at 37°C. For stage II incubations, reactions corresponding to lanes 2 and 4 were deproteinized by treatment with proteinase K, phenol, and chloroform, collected by ethanol precipitation, and then incubated with RFC, PCNA, DNA polymerase ␦, and RPA as in stage I except that MutS␣, MutL␣, and EXOI were omitted. To score repair, DNA products from all reactions were digested with HindIII and ClaI, subjected to electrophoresis through 1% agarose, and visualized with ethidium bromide. Arrows indicate repair products. B, stage I reactions were performed as in panel A except that T7 DNA polymerase (4 fmol) was substituted for polymerase ␦, and RFC and PCNA were present only as indicated. After deproteinization, products corresponding to lanes 2, 4, 6, and 8 were incubated in stage II reactions with T7 DNA polymerase and RPA in the absence of other proteins. Repair was scored as in panel A. support 5Ј-directed excision (RFC and PCNA are not required) (22,23). Supplementation of these four proteins with T7 polymerase resulted in a modest level of repair, although repair products were aberrant in the sense that the smaller restriction fragment scored by the repair assay (see Fig. 2A) was recovered in low yield (lane 1). The addition of RFC and PCNA suppressed 5Ј-directed repair in the presence of the T7 enzyme to about 20% of that observed in the presence of polymerase ␦ (compare lane 3 in panel B with lane 1 in panel A). However, when 5Ј-excision products produced in the absence of T7 polymerase (in the presence or absence of RFC and PCNA) were deproteinized and then incubated with T7 DNA polymerase, repair products were produced in high yield (panel B lanes 2 and 4). Similar results were obtained with the 3Ј-heteroduplex (lanes [5][6][7][8]. No detectable repair occurred when the 3Ј-substrate was incubated with MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA in the presence of T7 DNA polymerase (lane 7), although incubation of the 3Ј-heteroduplex with the excision system, followed by deproteinization and then incubation with the polymerase did yield repair products (lane 8). As shown in lane 6, repair in this manner was dependent on the presence of RFC and PCNA during the initial incubation; these two proteins are required for 3Ј-directed excision (23).
We have ruled out two trivial explanations for these findings. The inability of the T7 enzyme to support repair in the presence of the other activities is not because of an inability to copy a RPA-bound template strand because the stage II T7 polymerase reactions shown in Fig. 5B were performed in the presence of the single-stranded DNA-binding protein. The possibility that T7 DNA polymerase might inhibit mismatch-provoked excision was also excluded by scoring the extent of excision in the stage I incubations of Fig. 5B; in the reactions shown in lanes 3 and 7, excision occurred on 71 and 47% of the heteroduplex molecules, respectively (not shown). These findings indicate that primer-template access in repair intermediates is restricted, presumably via interaction with one or more repair activities.
Aphidicolin Effects on Mismatch Repair and the Hydrolytic Function of DNA Polymerase ␦-DNA polymerases ␦ and ⑀ possess a 3Ј to 5Ј hydrolytic activity that provides a proofreading function during DNA synthesis (35). The polymerase activity of both enzymes is inhibited by aphidicolin (43)(44)(45)(46), and on certain DNA substrates, the 3Ј-exonuclease function is also subject to inhibition by the drug. Studies with polymerase ␦ have shown that aphidicolin inhibits the hydrolysis of oligo(dT) in a poly(dA)⅐oligo(dT) substrate, but fails to block hydrolysis of singlestranded oligo(dT) or removal of 3Ј-terminal mispaired nucleotides in a substrate of the form poly(dA)⅐oligo(dT)-(dGMP) 4 (43)(44)(45). Similar studies have been done with polymerase ⑀. As observed with polymerase ␦, hydrolysis of single-stranded DNA is not significantly inhibited by aphidicolin (46). However, in contrast to the results obtained with polymerase ␦ using poly(dA)⅐oligo(dT), polymerase ⑀ hydrolysis of a defined oligonucleotide primer-template was inhibited by only 20% (46).
These aphidicolin effects on editing exonuclease function have been exploited to ask whether this activity contributes to the excision step of mismatch repair. Wang and Hays (21) have reported that aphidicolin inhibits 3Ј-directed excision in HeLa nuclear extracts by about 70% but is without effect on 5Ј-directed hydrolysis (21). Because the aphidicolin effects on editing exonuclease function summarized above were determined in the absence of RFC and PCNA and because the studies described here have failed to indicate a significant role for polymerase ␦ in mismatch-provoked excision, we have re-evaluated effects of the drug on editing exonuclease function and on the excision step of mismatch repair as it occurs in nuclear extracts.
The polymerase ␦ 3Ј to 5Ј hydrolytic activity was determined on 35-residue oligonucleotides hybridized to circular single-stranded f1 MR3 DNA, such that the hybrid was either perfectly paired or contained an unpaired 3Ј-terminal dinucleotide. In the absence of RFC and PCNA, mispaired 3Ј-terminal residues were subject to removal by an aphidicolin-resistant reaction, but the perfectly paired hybrid was much more resistant to hydrolysis (Fig. 6, left panels), results that are in agreement with previous studies (45,46). However, different results were obtained in the presence of RFC and PCNA. The clamp loader and replication clamp promoted extensive hydrolysis of the hybridized oligonucleotide without regard to the paired or unpaired nature of 3Ј-terminal nucleotides (right panels). We attribute this hydrolysis to polymerase ␦ because RFC and PCNA failed to support detectable hydrolysis in the absence of the polymerase (not shown). Furthermore, hydrolysis of the perfectly paired hybrid was abolished by aphidicolin, and in the presence of the drug, hydrolysis of the mispaired hybrid was restricted to the 3Ј-termi-

Aphidicolin effects on rates and extents of excision in HeLa nuclear extracts
Extents of excision were determined in the absence of exogenous dNTPs in reactions ("Materials and Methods") containing 24 fmol of 5Ј-GT or 3Ј-GT heteroduplex DNA and 100 g of HeLa nuclear extract. Extents of excision, which were determined after a 5-min incubation at 37°C, are limited primarily by ligation of unprocessed substrate. Values shown are the average of 4 determinations Ϯ 1 S.D. Corresponding values for extents of repair determined in the presence of dNTPs (100 M each) but in the absence of aphidicolin were: 11.5 Ϯ 1 and 5.2 Ϯ 0.4 fmol for 5Ј and 3Ј substrates, respectively. Repair in the presence of aphidicolin was undetectable (Ͻ1 fmol). Initial rate values were determined by sampling scaled up-reactions and are based on the linear portion of the progress curve. Values shown are the average of two independent determinations, with errors shown corresponding to the range of values observed.  nal unpaired nucleotides. These findings indicate that in the absence of dNTPs, RFC and PCNA modulate activity of the polymerase ␦ editing exonuclease, presumably via interaction of the clamp with the enzyme. Despite the inhibitory effects of aphidicolin on duplex DNA hydrolysis by polymerase ␦ in the presence of RFC and PCNA, effects of the drug on both 5Ј-and 3Ј-directed excision in HeLa nuclear extracts are modest. As shown in TABLE ONE, aphidicolin inhibited the initial rate of 3Ј-directed excision by 30% and the extent of the reaction by 22%. However, these inhibitory effects were not restricted to the 3Ј-reaction. 5Ј-Directed excision was inhibited to a similar degree (22% inhibition of initial rate and 20% inhibition of extent). The results shown in TABLE ONE were obtained with nuclear extracts for which we were able to observe significant inhibition of the excision reaction. With some extracts, we have been unable to detect aphidicolin inhibition of either 3Ј-or 5Ј-directed excision (not shown). Our failure to observe substantial inhibition of the 3Ј-excision reaction is not because of a problem with the drug preparation used, because it was effective in abolishing DNA synthesis by polymerase ␦ (Fig. 1C), inhibiting of 3Ј to 5Ј hydrolysis of duplex DNA by polymerase ␦ (Fig. 6), as well as overall mismatch repair (TABLE ONE, legend). These findings differ from those of Wang and Hays (21), but the basis of this discrepancy is unclear. Our results suggest that an aphidicolin-sensitive DNA polymerase (␣, ␦, or ⑀) might contribute in a modest but undefined manner to both 3Ј-and 5Ј-directed, mismatch-provoked excision in nuclear extracts. However, it is also possible that the inhibitory effects described here, as well as that described by Hays and Wang (21) for 3Ј-excision, could be because of some sort of nonspecific effect on the extract system.

DISCUSSION
The studies described here are based on two previous observations: the demonstration that MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA are sufficient to support bidirectional mismatch-provoked excision (23), and the finding that DNA polymerase ␦ is required for human mismatch repair in nuclear extracts (28). The latter study implicated polymerase ␦ in the reaction, but involvement of polymerases ␣ and ⑀ was not excluded because the depleted extracts used contained a normal complement of polymerase ␣ and trace levels of ⑀. Whereas the experiments described here do not exclude significant participation of DNA polymerases ␣ and ⑀ in mismatch repair, they do show that the ␦ enzyme is sufficient to meet the polymerase requirement in a reconstituted system. As mentioned above, genetic studies in Saccharomyces cerevisiae have also indicated involvement of POL32, a subunit of polymerase ␦, in mismatch repair (29).
Analysis of the repair synthesis on gapped circular DNA and the use of staged reactions (Figs. 1 and 5) implicate polymerase ␦ in the repair synthesis step of mismatch repair in the reconstituted system. This reaction also requires PCNA and RFC, and is stimulated by RPA, proteins that also play key roles in mismatch-provoked excision (22,23,47). These conclusions are consistent with previous findings based on use of crude nuclear fractions. Thus, the depleted extracts used to demonstrate polymerase ␦ involvement in mismatch repair were shown to be defective in a step subsequent to mismatch-provoked excision (28), and PCNA has been implicated in both excision and repair synthesis steps of the crude extract reaction (36).
As further evidence for the specificity of the polymerase ␦ requirement in mismatch repair, we have found that T7 DNA polymerase substitutes poorly for the human polymerase in reconstituted mismatch repair. However, the T7 enzyme is capable of efficient gap repair synthesis on deproteinized excision intermediates in the presence of RPA. Based on this finding, we presume that access to the excision interme-diate is restricted in some way by one or more repair activities and that this nucleoprotein complex serves to recruit polymerase ␦ to the primer terminus.
MutS␣, MutL␣, EXOI, RPA, RFC, and PCNA are sufficient to support bidirectional mismatch-provoked excision in a defined system (23). Inasmuch as EXOI hydrolyzes DNA with 5Ј to 3Ј polarity (48 -50), it can account for 5Ј-directed excision mediated by these proteins. The activity responsible for 3Ј-directed hydrolysis in this reconstituted system has not been identified. Because the other components used were free of significant exonuclease activity, it was suggested that a cryptic 3Ј to 5Ј hydrolytic activity of EXOI might be responsible for 3Ј-directed excision in this system (23). However, the 7-component system described in this study does include a 3Ј to 5Ј hydrolytic activity, the editing exonuclease of DNA polymerase ␦. In fact, the strong synergistic increase in mutation rates observed with S. cerevisiae exoI pol3-01 and exoI pol2-4 double mutants, as compared with those observed with exoI, pol3-01, or pol2-4 single mutants, has led to the suggestion that in addition to their editing function, the 3Ј to 5Ј exonucleases of polymerases ␦ and ⑀ participate in the excision step of mismatch repair (37). Given the multiple functions of EXOI (51), it is also possible that these genetic results indicate co-participation of these exonucleases in a mutation avoidance pathway distinct from mismatch repair (38). Indeed, the hypermutability associated with the pol3-01 editing exonuclease defect is dependent on S phase checkpoint activation (39). Nevertheless, it has also been reported that 3Ј-, but not 5Ј-directed, mismatch-provoked excision in HeLa nuclear extracts is inhibited to a substantial degree by aphidicolin, an effect attributed to participation of polymerase ␦ and/or ⑀ editing exonuclease(s) in the excision step of mismatch repair (21).
In an attempt to further clarify possible function of the polymerase ␦ editing exonuclease in mismatch repair, we have asked whether this activity contributes to excision in the purified system. However, we were unable to detect significant excision in reactions that contained the DNA polymerase but lacked EXOI, or in reactions that included both the polymerase and a catalytically dead form of EXOI. These findings strongly suggest that the polymerase 3Ј to 5Ј hydrolytic activity does not contribute substantially to excision in the defined system. We have also repeated previous experiments on aphidicolin effects on mismatch-provoked excision (20,21). In our hands, the effects of the drug on mismatch-provoked excision in HeLa nuclear extracts are limited and extract dependent. With some extracts, we have been unable to detect any inhibition, and in the case of those extracts where significant inhibition was observed, the effect was modest and evident with both 3Јand 5Ј-heteroduplexes. One explanation for these results is that the aphidicolin effects on excision are nonspecific in nature, although these findings could also indicate a limited dependence of both 3Ј-and 5Ј-directed excision on an aphidicolin-sensitive DNA polymerase. It is important to note in this regard that the substrates used to study in vitro mismatch repair are non-replicating DNAs. If editing exonuclease involvement in mismatch repair was dependent on a replication fork context, the nicked circular heteroduplexes used for the in vitro assay may preclude detection of such an effect.