Mechanisms in Eukaryotic Mismatch Repair*

Inactivation of the human mismatch repair system confers a large increase in spontaneous mutability and a strong predisposition to tumor development. Mismatch repair provides several genetic stabilization functions: it corrects DNA biosynthetic errors, ensures the fidelity of genetic recombination, and participates in the earliest steps of checkpoint and apoptotic responses to several classes of DNA damage (see refs. 1-3 for recent reviews). Defects in this pathway are the cause of typical and atypical hereditary nonpolyposis colon cancer (4), but may also play a role in the development of 15 to 25% of sporadic tumors that occur in a number of tissues (5). The system is also of biomedical interest because mismatch repair-deficient tumor cells are resistant to certain cytotoxic chemotherapeutic drugs (2,3), a manifestation of its involvement in the DNA damage response. Of the several mutation avoidance functions of mismatch repair, the reaction responsible for replication error correction has been the most thoroughly studied, and the discussion that follows is restricted to this pathway.

Inactivation of the human mismatch repair system confers a large increase in spontaneous mutability and a strong predisposition to tumor development. Mismatch repair provides several genetic stabilization functions; it corrects DNA biosynthetic errors, ensures the fidelity of genetic recombination, and participates in the earliest steps of checkpoint and apoptotic responses to several classes of DNA damage (see Refs. 1-3 for recent reviews). Defects in this pathway are the cause of typical and atypical hereditary nonpolyposis colon cancer (4) but may also play a role in the development of 15-25% of sporadic tumors that occur in a number of tissues (5). The system is also of biomedical interest because mismatch repair-deficient tumor cells are resistant to certain cytotoxic chemotherapeutic drugs (2,3), a manifestation of its involvement in the DNA damage response. Of the several mutation avoidance functions of mismatch repair, the reaction responsible for replication error correction has been the most thoroughly studied, and the discussion that follows is restricted to this pathway.

Mismatch Repair in Eukaryotic Cell Extracts
Correction of DNA biosynthetic errors requires targeting of mismatch repair to the newly synthesized strand at the replication fork. In contrast to Escherichia coli, where mismatch repair is directed by the transient absence of adenine methylation at d(GATC) sites within newly synthesized DNA, the strand signals that direct replication error correction in eukaryotes have not been identified. However, the function of the hemimethylated d(GATC) strand signal in E. coli mismatch repair is provision of a nick on the unmethylated strand, which serves as the actual signal that directs the reaction (2). Similarly, a strandspecific nick or gap is sufficient to direct mismatch repair in extracts of mammalian and Drosophila cells as well as Xenopus egg extracts (1)(2)(3). These findings, coupled with the observation that mismatch repair is more efficient on the lagging strand at the replication fork (6), suggest that DNA termini that occur as natural intermediates during replication (3Ј terminus on the leading strand; 3Ј and 5Ј termini on the lagging strand) may suffice as strand signals to direct the correction of DNA biosynthetic errors in eukaryotic cells.
Available information on the mechanism of eukaryotic mismatch repair is largely derived from analysis of the nick-di-rected repair of circular heteroduplexes in mammalian cell extracts. As shown in Fig. 1, the strand break that directs repair may reside either 3Ј or 5Ј to the mispair as viewed along the shorter path linking the two sites in the circular substrate, and processing of such molecules in extracts is largely restricted to this region. Examination of intermediates produced in HeLa nuclear extracts when repair DNA synthesis is blocked has demonstrated that mismatch-provoked excision removes that portion of the incised strand spanning the shorter path between the nick and the mismatch ( Fig. 1) with excision tracts extending from the strand break to a number of sites within a region Ϸ90 -170 nucleotides beyond the mispair (7,8). Radiolabeling of repair DNA synthesis tracts is also consistent with this view (9). The mammalian repair system thus displays a bidirectional capability in the sense that it responds to both 3Ј-and 5Ј-heteroduplex orientations, and functionality is retained at nick mismatch separation distances as large as 1000 bp (7).
The nicks that direct the E. coli mismatch repair also serve as sites for initiation of excision (2), and function of the strand break in the eukaryotic reaction has generally been interpreted in a similar manner (1-3). However, a distinct mechanism for mismatch-provoked excision has been proposed based on radiolabeling of repair DNA synthesis tracts in Xenopus egg extracts. In contrast to results obtained with the human system (9), radiolabeling of repair products in Xenopus extracts was significantly higher near the mismatch than the strand break (10). Based on this analysis, Varlet et al. (10) suggested that the nick that directs repair does not correspond to the site of excision initiation. Rather, a mismatch-activated strand-specific endonuclease is postulated to introduce a second nick near the mispair, with this nick serving as an entry site for the excision system. As described below, recent experiments suggest that the mammalian repair system supports both of these modes of excision.

Mammalian MutS and MutL Activities
The activities responsible for initiation of E. coli mismatch repair are MutS and MutL, which function as homo-oligomers (2). MutS is responsible for mismatch recognition and MutL serves to interface mismatch recognition by MutS to activation of downstream activities. Mammalian cells possess two MutS activities that function as heterodimers and share MSH2 as a common subunit (11)(12)(13)(14): MutS␣ (MSH2⅐MSH6 heterodimer) and MutS␤ (MSH2⅐MSH3 heterodimer). MutS␣, which represents 80 -90% of the cellular MSH2, preferentially recognizes base-base mismatches and insertion/deletion (ID) 2 mispairs in which one strand contains 1 or 2 unpaired nucleotides but is also capable of recognition of larger ID heterologies with reduced affinity (11)(12)(13)15). MutS␤ recognizes ID mismatches of 2 to about 10 nucleotides, weakly recognizes single-nucleotide ID mispairs, and is essentially inert on base-base mismatches (12,15). MSH2 and MSH6 defects have been impli-cated in tumor development, but the cancer predisposition conferred by MSH6 inactivation is less severe (4,16). The association of MSH3 defects with tumor development appears to be limited (4,5,16).
Three eukaryotic MutL activities have been identified and, like eukaryotic MutS activities, function as heterodimeric complexes with MLH1 serving as a common subunit. MutL␣, a heterodimer of MLH1 and PMS2, is the primary MutL activity in human mitotic cells and supports repair initiated by either MutS␣ or MutS␤ (17). MutL␣ accounts for Ϸ90% of the MLH1 in human cells (18,19), but two low abundance complexes involving MLH1 have also been identified. A human MLH1⅐PMS1 heterodimer (MutL␤) has been isolated, but involvement in mismatch repair has not been demonstrated (18). However, the MutL␥ MLH1⅐MLH3 complex has been reported to support modest levels of base-base and single-nucleotide ID mismatch repair in vitro, events that are presumably initiated by MutS␣ (19). Genetic inactivation of MLH1 or PMS2 confers cancer predisposition, but mutations in PMS1 do not (4,16). Involvement of MLH3 defects in tumor development is uncertain (4,5).

Other Activities in Mammalian Mismatch Repair
Yeast genetic studies and analysis of the mammalian extract reaction have implicated six additional activities in eukaryotic mismatch repair. The key finding that culminated in the reconstitution studies described below was the demonstration that exonuclease 1 (Exo1) participates in the reaction. Genetic evidence for Exo1 involvement in yeast mismatch repair (20,21) led to the subsequent demonstration that Exo1 is required for repair of base-base and single-nucleotide ID mismatches in mammalian cell extracts (22,23). Because Exo1 hydrolyzes duplex DNA with 5Ј to 3Ј polarity (24,25), the surprising feature of this requirement is that the enzyme is necessary for excision and repair directed by strand break located either 5Ј or 3Ј to the mismatch. This paradoxical requirement led to the suggestion that Exo1 might play a positive regulatory role in 3Ј excision or that the reaction may be mediated by a cryptic Exo1 3Ј to 5Ј hydrolytic activity (22). As discussed below, this issue has been recently resolved, and it is not necessary to invoke either of these possibilities.
Exo1 Ϫ/Ϫ mice display modest cancer predisposition, and Exo1 deficiency is associated with a 30-fold elevation of hypoxanthine-guanine phosphoribosyltransferase mutability, substantially less than the 150-fold increase observed with Msh2 Ϫ/Ϫ cells (23). Although extracts of Exo1 Ϫ/Ϫ mouse cells are virtually devoid of repair activity on basebase mismatches, they retain significant activity on one-or two-nucleotide ID mispairs (23). These findings imply the existence of one or more alternate excision activities, and several possibilities have been suggested. Involvement of the 3Ј to 5Ј editing exonuclease functions of DNA polymerases ␦ and ⑀ in mismatch repair has been proposed on genetic and biochemical grounds (26 -28), but this idea has been questioned (29,30). Using a small interfering RNA knockdown approach, Vo et al. (31) have suggested that the Mre11 3Ј to 5Ј-exonuclease participates in 3Ј-directed mismatch repair. Mre11 depletion was shown to reduce the efficiency of 3Ј-directed repair by Ϸ40%, and repair was restored to normal levels by the addition of partially purified Mre11. However, the involvement of other activities in repair restoration was not excluded because the Mre11 fraction tested was relatively crude. This is of concern because down-regulation of Mre11 also leads to relatively rapid chromosome breakage and can interfere with cell proliferation (32).
Experiments in human cell extracts and partially purified fractions have also indicated involvement of several DNA binding proteins in eukaryotic mismatch repair. The extract reaction is abolished by antibody against the single-stranded DNA binding protein RPA (33), which stimulates excision, stabilizes the ensuing gap against endonuclease attack, and promotes repair DNA synthesis (34). The non-histone chromatin protein HMGB1, which interacts with MutS␣, may also play an important role in the initiation of mismatch-provoked excision in nuclear extracts (35).
The PCNA replication clamp and DNA polymerase ␦ have also been implicated in mismatch repair in human cell extracts (36 -39). PCNA, which confers processivity on polymerase ␦ (40), plays multiple roles in mismatch repair. As might be expected, PCNA is necessary for repair DNA synthesis (38,39), but it is also required for mismatch-provoked excision (36). The most compelling evidence for PCNA involvement in the excision step of mismatch repair has been provided by p21 inhibition studies. By forming a stable complex with DNA-bound PCNA, p21 interferes with downstream PCNA-dependent events (41). Although p21 abolishes 3Ј-directed mismatch-provoked excision in HeLa cell extracts, only 40 -50% of 5Ј-directed excision events are subject to p21 inhibition, implying occurrence of at least two hydrolytic modes on 5Ј-heteroduplexes (36, 39, 42). In vitro analysis of eukaryotic mismatch repair has relied on the use of circular heteroduplexes containing a mismatch and a strand-specific nick or short gap. Because repair tracts are restricted to the shorter path linking the two DNA sites, circular heteroduplexes are assigned a polarity depending on whether the nick resides 3Ј (left) or 5Ј (right) to the mismatch along this path. Excision tracts produced in nuclear extracts of human cells span the distance between the nick and the mismatch (7-9) as indicated. However, those produced in Xenopus egg extracts are more localized to the vicinity of the mispair (10).

Mismatch-provoked Excision in Purified Systems
These findings have led to several reconstituted systems that rely on near homogeneous proteins and support mismatch-provoked excision and repair. The simplest excision system depends on MutS␣, MutL␣, Exo1, RPA, and ATP (39), and similar results have been obtained in a system that also contains HMGB1 (43). As illustrated in Fig. 2, 5Ј-directed excision in this system is mismatch-provoked and terminates upon mismatch removal. Analysis of this reaction has revealed several features of the hydrolytic mechanism. MutS␣ activates Exo1 hydrolysis on a 5Ј-heteroduplex in a mismatch-and ATP-dependent manner. In the absence of other proteins, 5Ј to 3Ј hydrolysis by Exo1 occurs by a distributive mechanism, but MutS␣ renders the enzyme highly processive, resulting in removal of Ϸ2,000 nucleotides prior to dissociation (39), an effect attributed to formation of a MutS␣ ⅐Exo1 complex. Hydrolysis by the MutS␣⅐Exo1 complex is controlled in part by RPA, which reduces processivity of the MutS␣⅐Exo1 complex to Ϸ250 nucleotides, and by binding to gaps, RPA controls access of Exo1 to 5Ј termini in excision intermediates/products (39). Although an RPAfilled gap is a very poor substrate for Exo1, MutS␣ promotes Exo1 loading at such sites provided that the gapped molecule contains a mismatched base pair. The ramifications of these RPA effects are 2-fold. Excision on 5Ј-heteroduplexes proceeds via a set of pseudo-discrete hydrolytic intermediates, which differ in size by about 250 nucleotides, an effect attributed to multiple reloading of MutS␣ and Exo1 (39,43). Second, hydrolysis is dramatically attenuated upon mismatch removal because MutS␣ can no longer promote Exo1 loading at the RPA-filled gap in the excision product. RPA thus has both negative and positive regulatory effects on this reaction; by suppressing processive behavior of the MutS␣⅐Exo1 complex and by restricting hydrolytic activity on excision products, it promotes turnover of the system after mismatch removal, allowing other heteroduplex molecules to participate in the reaction.
MutL␣ is not required for mismatch-and MutS␣-dependent activation of Exo1, but it does play a significant role in excision. By acting in concert with MutS␣ to suppress Exo1 hydrolysis on DNA that lacks a mispair, MutL␣ enhances the mismatch dependence of the reaction (39,44). MutL␣ also participates in excision termination in this system, but two different mechanisms have been proposed to account for its function in this regard. Genschel et al. (39) have attributed MutL␣ involvement in termination to its role in suppressing Exo1 activity on mismatchfree DNA. In this mechanism MutL␣ simply stabilizes excision products against nonspecific hydrolysis by Exo1. By contrast, Zhang et al. (43) have concluded that MutL␣, acting in concert with RPA, plays an active role in excision termination upon mismatch removal. This issue has not been resolved.
MutS␣, MutL␣, Exo1, and RPA also support mismatchprovoked excision on a 3Ј-heteroduplex. As in the case of a 5Ј-substrate, hydrolysis on a 3Ј-heteroduplex proceeds 5Ј to 3Ј from the strand break (Fig. 2), which is the wrong polarity for mismatch removal (22,45). The 5Ј to 3Ј directionality of this system has been referred to as a default polarity (2). Although PCNA has no significant effect on the restricted directionality of this system, supplementation with both PCNA and RFC (RFC loads PCNA onto the helix (40)) yields a system that supports mismatch removal from both 5Ј and 3Ј-heteroduplexes (45). Excision products obtained from a 5Ј-heteroduplex in this six-component system are similar to those produced by MutS␣, MutL␣, Exo1, and RPA. However, when the nick is located 3Ј to the mismatch, Exo1 5Ј to 3Ј hydrolysis initiating at the nick is largely repressed by RFC, and excision occurs with apparent 3Ј to 5Ј polarity resulting in mismatch removal. Although excision in this six-component system displays similarities to the bidirectional reaction that has been studied in nuclear extracts, the distribution of excision products in the purified system is more disperse than that observed in extracts. This purified system therefore lacks one or more activities that play significant roles in mismatch repair (45).
Because an Exo1 active site mutant failed to support both 5Јand 3Ј-directed excision in this system, mismatch removal in both cases was attributed to this exonuclease (45). It was suggested that a cryptic Exo1 3Ј to 5Ј hydrolytic function is responsible for 3Ј-directed excision. However, the necessity for a 3Ј to 5Ј-exonuclease in this purified system was rendered moot by the demonstration that MutS␣, RFC, and PCNA activate a latent MutL␣ endonuclease in an ATP-and mismatch-dependent manner (46). As shown in Fig. 3 incision by activated MutL␣ endonuclease occurs on both 3Ј-and 5Ј-heteroduplexes and is strongly biased to the nicked heteroduplex strand. For heteroduplexes with a nick mismatch separation distance of Ϸ100 bp, incision tends to occur on the distal side of the mismatch relative to the strand break but at larger separation distances readily FIGURE 2. 5 to 3 default hydrolytic system. Exo1, which is activated in a mismatch-and MutS␣-dependent manner, initiates 5Ј to 3Ј hydrolysis at the strand break (22,39,43). Excision on a 5Ј-heteroduplex terminates upon mismatch removal in a manner that depends on RPA and MutL␣. MutS␣ also activates Exo1 on a 3Ј-heteroduplex, but in this case hydrolysis proceeds with the wrong directionality for mismatch removal. Green arrows indicate a requirement for signaling between the mismatch, which activates excision, and the strand break where hydrolysis initiates.
occurs between the two DNA sites. 3 In the case of a 3Ј-heteroduplex, incision distal to the mismatch provides an initiation site for mismatch removal by the 5Ј to 3Ј action of MutS␣activated Exo1 (Fig. 3). Inasmuch as PCNA-dependent and independent modes of 5Ј-directed excision have been invoked in nuclear extracts (39,42), it is noteworthy that this PCNA-dependent endonucleolytic system also incises 5Ј-heteroduplexes (46).
The endonucleolytic mode of action of this system is reminiscent of the mechanism for mismatch repair in Xenopus egg extracts proposed by Varlet et al. (10). As discussed above, this model posits that the nick that directs repair serves as a strand signal but not as a site for excision initiation, which actually occurs at a strand break produced by a mismatch-activated endonuclease. This mode of excision is fundamentally different from that used by the E. coli methyl-directed pathway, where hydrolysis initiates at a 3Ј or 5Ј strand break that directs repair (2).
The probable active site of the MutL␣ endonuclease has been localized to a divalent metal binding site within the PMS2 subunit that is defined by a DQHA(X) 2 E(X) 4 E motif (46). Amino acid substitution mutations within this motif abolish MutL␣ endonuclease activity as well as the ability of the protein to support mismatch repair in nuclear extracts. This motif is conserved in homologs of eukaryotic PMS2 and in archaeal and eubacterial MutL proteins but is conspicuously absent in MutL proteins from bacteria like E. coli that rely on d(GATC) methylation to direct mismatch repair. The presence or absence of this MutL motif may therefore define two distinct classes of mismatch repair systems.

Mismatch Repair in Purified Systems
Several defined systems that support mismatch correction have also been described, but these differ in several respects. Zhang et al. (43) have shown that MutS␣, MutL␣, Exo1, RPA, HMGB1, and DNA polymerase ␦ are sufficient to support repair of 5Ј-heteroduplexes containing a G-T mismatch or a 3-nucleotide ID mispair and that covalently closed repair products are obtained upon supplementation of these proteins with DNA ligase I. As observed for 5Ј-directed excision (39), MutL␣ is not required for repair in this system (43). The functions of RPA and HMGB1 in this reconstituted reaction appear to be largely redundant because either protein is sufficient to support reconstituted repair, and excision in the presence of RPA is only modestly enhanced by the addition of HMGB1. Interestingly, substitution of MutS␤ for MutS␣ yields a system that supports 5Ј-directed excision and repair of a 3-nucleotide ID mismatch, implying that like MutS␣, MutS␤ can activate Exo1. One surprising feature of this reconstituted system is that repair is independent of RFC and PCNA. This is unexpected because the DNA synthesis step of 5Ј-heteroduplex repair in nuclear extracts is PCNA-dependent (38,39). Furthermore, in contrast to its activity on a 5Ј-heteroduplex, this purified system does not support 3Ј-directed excision or repair when supplemented with RFC and PCNA (43). A reconstituted repair system with somewhat different properties has been described by Constantin et al. (30). In contrast to the 5Ј-directed repair system described above (43), this system supports bidirectional mismatch repair dependent on MutS␣, MutL␣, Exo1, RPA, DNA polymerase ␦, RFC, and PCNA (30). MutL␣ is dispensable for 5Ј repair in this system but required for 3Ј-heteroduplex repair. The RFC and PCNA requirement for 5Ј-directed correction is because of their involvement in the repair DNA synthesis step, whereas both proteins are also required for excision on a 3Ј-heteroduplex.
The different results obtained by Zhang et al. as compared with those of Constantin et al. (30) and Dzantiev et al. (45) have been attributed to activity differences between the RFC preparations used (43,46  MutL␣ endonuclease, which is activated in a mismatch-MutS␣-, PCNA-, RFC-, and ATP-dependent manner, incises (red arrows) the discontinuous strand of 5Ј-or 3Ј-heteroduplex DNAs in an ATP-dependent manner (46). Incision of a3Ј-heteroduplex on the distal side of the mispair relative to the strand break yields a 5Ј terminus that serves as an entry site for MutS␣-activated Exo1, which removes the mismatch by the 5Ј to 3Ј hydrolytic reaction shown in Fig. 2. Excision tracts shown in this model span the mismatch and the original heteroduplex strand break, as observed in human cell extracts (7)(8)(9). However, excision in Xenopus egg extracts (10) would be restricted to the region between the breaks introduced by MutL␣. Because MutL␣ activation depends on a heteroduplex strand break and because endonuclease action is restricted to the nicked heteroduplex strand, signaling must occur between mismatch and heteroduplex strand break (green arrows). Adapted from Kadyrov et al. (46).

Some Unanswered Questions
The recent establishment of defined systems that support mismatch-provoked excision and repair reactions should facilitate study of this elaborate reaction. However, because the reconstituted reactions described to date are minimal systems, it is premature to assume that they can account for mismatch repair as it occurs in the eukaryotic cell. There is excellent evidence for participation of hydrolytic activities in addition to Exo1 (23), but these have not been convincingly identified. Activities that regulate the action of MutL␣ endonuclease and control 3Ј-directed excision have also been invoked (45,46), but these have not been characterized.
The defining feature of the replication error correction reaction is its strand-specific character, an effect that depends on the interaction of a mismatch and strand break that can be separated by hundreds of base pairs. Several models, which attempt to explain the molecular nature of this interaction, have been thoroughly debated in the literature (1-3), but the mechanism responsible for communication between the two DNA sites has not been established. However, the recent finding that mismatch-dependent incision by activated MutL␣ endonuclease is strongly biased to the nicked heteroduplex strand (46) suggests that interaction of the mismatch and strand break may involve keeping track of a DNA strand.
The sequence of events during the course of nick-directed mismatch repair is presumably dictated by the temporal course of protein-protein interactions that occur on the heteroduplex. A number of protein-protein interactions have been documented in this system, including MutS␣-MutL␣, MutS␣-PCNA, MutS␤-PCNA, MutL␣-PCNA, MutS␣-Exo1, MutL␣-Exo1, Exo1-PCNA, and PCNA-polymerase ␦ (1-3, 40). With the exception of interactions between MutS␣ and Exo1 and between PCNA and polymerase ␦, the significance of these protein-protein interactions in nick-directed mismatch repair has not been established.