Biochemical Characterization of the Interaction between theSaccharomyces cerevisiae MSH2-MSH6 Complex and Mispaired Bases in DNA*

The interaction of the Saccharomyces cerevisiae MSH2-MSH6 complex with mispaired bases was analyzed using gel mobility shift assays and surface plasmon resonance methods. Under equilibrium binding conditions, MSH2-MSH6 bound to homoduplex DNA with a K d of 3.9 nm and bound oligonucleotide duplexes containing T:G, +1, +2, +4, and +10 insertion/deletion loop (IDL) mispairs with K d values of 0.20, 0.25, 11, 3.2, and 0.55 nm, respectively. Competition binding experiments using 65 different substrates revealed a 10-fold range in mispair discrimination. In general, base-base mispairs and a +1 insertion/deletion mispair were recognized better than intermediate sized insertion/deletion mispairs of 2–8 bases. Larger IDL mispairs (>8 bases) were recognized almost as well as the +1 IDL mispair. Recognition of mispairs by MSH2-MSH6 was influenced by sequence context, with the 6-nucleotide region surrounding the mispair being primarily responsible for influencing mispair recognition. Effects of sequences as far away as 15 nucleotides were also observed. Differential effects of ATP on the stability of MSH2-MSH6-mispair complexes suggested that base-base mispairs and the smaller IDL mispairs were recognized by a different binding mode than larger IDL mispairs, consistent with genetic experiments indicating that MSH2-MSH6 functions primarily in the repair of base-base and small IDL mispairs.

There are four ways in which mispaired bases arise in DNA: chemical damage to DNA such as deamination of 5-methylcytosine; chemical damage to DNA precursors such as oxidation of dGTP; misincorporation during DNA synthesis; or as a result of heteroduplex recombination intermediates formed during recombination between DNAs containing sequence differences. There are two basic types of mismatch repair pathways known. One type of mismatch repair recognizes and repairs mispaired bases in DNA that result from chemical damage to DNA and DNA precursors such as A:8-oxoguanine mispairs (reviewed in Ref. 1). The second type of mismatch repair recognizes and repairs a broad spectrum of mispaired bases in DNA that arise due to errors during DNA replication and as a result of genetic recombination (reviewed in Refs. [2][3][4]. The recent observation that genetic defects in the broad spectrum mismatch repair pathways underlie the development of both inherited and sporadic cancers has underscored the importance of understanding the mechanism of eukaryotic mismatch repair (5)(6)(7)(8)(9)(10).
At this time, the best understood broad spectrum mismatch repair pathway is the Escherichia coli MutHLS mismatch repair pathway (reviewed in Refs. 1 and 3). In this case, the methylation state of a GATC site, specifically a hemimethylated GATC site, governs the repair of a mispair such that the mispaired base on the unmethylated strand is repaired (11,12). The MutHLS system has been reconstituted using hemimethylated DNA substrates containing mismatches, MutH, MutL, MutS, and UvrD (helicase II) proteins along with DNA polymerase III holoenzyme, DNA ligase, single-strand DNA binding protein, and any one of the single-stranded DNA exonucleases, exo I, exo VII, or RecJ protein (3,4,13). The reaction involves mismatch-dependent nicking of the unmethylated strand at a hemimethylated GATC site by the MutH protein after the MutS protein binds to the site of a mispaired base (14). No activity has been assigned to MutL, although it interacts with both MutS and UvrD, is required for activation of MutH and UvrD, and may enhance the specificity of MutS for mispaired bases (15)(16)(17). Excision initiates at the nick produced by MutH and requires UvrD and one of the single-stranded DNA exonucleases, exo I (3Ј-exo), exo VII (3Ј-and 5Ј-exo), RecJ protein (5Ј-exo), or possibly other exonucleases, depending on whether the nicked, unmethylated site is 5Ј or 3Ј to the mispair (18). Once excision has occurred, resynthesis is mediated by DNA polymerase III holoenzyme and single-strand DNA binding protein (1,3,4).
Many studies of mismatch repair in eukaryotes have focused on the identification of genes encoding homologues of the E. coli MutS and MutL proteins. Six genes encoding MutS homologues, MSH1-6, have been identified in Saccharomyces cerevisiae (19 -24), and genes encoding homologues of MSH2- 6 have thus far been identified in humans and mice (7)(8)(9)(25)(26)(27)(28)(29)(30)(31)(32)(33) and to some extent in other organisms. Genetic experiments in S. cerevisiae have provided evidence as to the roles of each of these gene products. MSH1 is thought to function in the suppression of mutations in mitochondrial DNA (20,34). MSH2, -3, and -6 are thought to function in mismatch repair in the nucleus (20,21,23,35). Analysis of human tumor cell lines containing mutations in different combinations of MSH2, -3, and -6 and analysis of cells and tissues derived from mice containing mutations in these genes have also supported this view (9, 36 -44). MSH4 and -5 do not appear to be required for mismatch repair but rather are required for efficient crossing over during meiotic recombination (22,24,28,45,46). Studies characterizing the mutator phenotypes caused by different combinations of mutations in MSH2, -3, and -6 and experiments in which MSH2 was demonstrated to interact with both MSH3 and MSH6 have led to a model for how these three proteins might function in mismatch repair (2,23). Mismatch repair was proposed to involve two different heterodimeric complexes, MSH2-MSH3 and MSH2-MSH6, that have unique but overlapping mispair recognition specificity. These two complexes have been demonstrated in both S. cerevisiae and human cells, and their properties have been partially characterized. Both complexes have been demonstrated to bind to mispaired bases in DNA (30,36,43,(47)(48)(49)(50), and information is available on the regions of MSH2 and MSH6 that interact and on the effect of nucleotide cofactors on MSH2-MSH6 mispair binding (51)(52)(53)(54)(55)(56)(57). Little is known about the specific roles of the different subunits, although it seems possible that MSH3 and MSH6 modify the intrinsic mispair recognition properties of MSH2 (58,59).
Much of what is known about the mispair recognition specificity of the MSH2-MSH3 and MSH2-MSH6 complexes has been predicted from the analysis of the mutator phenotypes caused by mutations in the S. cerevisiae genes encoding these proteins. These studies predict that the MSH2-MSH6 complex recognizes both base-base and insertion/deletion mispairs, whereas the MSH2-MSH3 complex only recognizes insertion/ deletion mispairs (23, 35, 60 -64). Studies of the effect of mutations in the MSH2, MSH3, and MSH6 genes on the accumulation of mutations in microsatellite sequences as a function of repeat unit length have suggested that the relative role of the MSH2-MSH3 complex in mismatch repair relative to the MSH2-MSH6 complex increases with increasing size of the insertion/deletion mispair (61). However, the ability of mismatch repair to suppress the accumulation of mutations in microsatellite sequences appears to decrease with increasing repeat unit length, suggesting that both complexes may have limited ability to recognize larger insertion/deletion mispairs (61).
The use of mutator phenotypes to predict the mispair recognition specificity of the MSH2-MSH6 and MSH2-MSH3 complexes is subject to some uncertainty. For example, a high rate of instability was observed in microsatellite instability experiments in wild type yeast cells when larger repeat units were analyzed, and mutations in MSH2, MSH3, and MSH6 caused little or no destabilization as the repeat size increased (61). This could either mean that large insertion/deletion mispairs are not recognized by mismatch repair or that the mutagenic pathway acting on large repeats involves a mechanism that does not produce mispairs. A second point is that analysis of in vivo mutator phenotypes reflects the ability of a specific protein complex to interact with a mixture of substrates produced by replication errors, and under these conditions substrates produced at high levels may compete for the repair proteins with the substrate being analyzed leading to a reduced level of repair, particularly if specific repair proteins are in limiting concentration. Indeed, results from recent in vitro complementation experiments suggest that the MSH2-MSH6 complex might recognize insertion/deletion mispairs having as many as 12 unpaired bases which is larger than predicted from genetic experiments (48). However, in these experiments there was a high level of MSH-independent repair that may have precluded observing MSH2-MSH6-or MSH2-MSH3-dependent repair of the largest insertion/deletion mispair tested that contained 27 unpaired bases. Indeed, there is evidence that insertion/deletion mispairs containing as many as 26 or 38 unpaired bases can be acted on by MSH2-dependent mismatch repair, which is larger than predicted from mutator phenotype analysis (65,66).
The MSH2-MSH6 and MSH2-MSH3 complexes have been demonstrated to bind to mispaired bases in DNA; however, there is only a limited amount of information available describing the basic properties of mispair recognition by these complexes. Surprisingly, there is limited information available about the spectrum of mispaired bases recognized by the purified MSH2-MSH6 complex as greater effort has been focused on the mispair recognition properties of the MSH2-MSH3 complex. In the following study, we describe a detailed analysis of the mispair recognition properties of the S. cerevisiae MSH2-MSH6 complex.

EXPERIMENTAL PROCEDURES
Materials-ATP, ADP, and poly(dI⅐dC) DNA were obtained from Amersham Pharmacia Biotech. [␥-32 P]ATP was obtained from NEN Life Science Products. Bovine serum albumin was obtained from New England Biolabs (Beverly, MA). Acrylamide was purchased from Schwarz/Mann (ICN) (Cleveland, OH), and bisacrylamide was purchased from Eastman Kodak Co. For chromatography, PBE94 and Q-Sepharose were purchased from Amersham Pharmacia Biotech, and ssDNA 1 cellulose was made according to the method of Ref. 67. For BIAcore analysis, P20 surfactant and streptavidin-coated SA sensor chips were obtained from BIAcore AB (Uppsala, Sweden). Growth media, including synthetic minimal dropout media, were prepared as described previously (20,58). Raffinose and galactose were purchased from Sigma.
Duplex DNA Binding Substrates-DNA duplexes used in this study were annealed by first heating 200 pmol of each oligonucleotide to 94°C for 5 min in 100 l of annealing buffer (0.5 M NaCl, 10 mM Tris-Cl, pH 7.5, 1 mM EDTA) and then slow cooling over 2 h to 25°C. DNA duplexes were then purified by high pressure liquid chromatography using a Waters GEN-PAK FAX column (Milford, MA) as described (68). Oligonucleotides used in this study, including biotinylated oligonucleotides, were synthesized by the Dana-Farber Cancer Institute Molecular Biology Core Facility using a PE/ABI 3948 Synthesizer (Perkin-Elmer) and are listed in Table I. Pairwise combinations of oligonucleotides used to create the duplex DNA binding substrates are listed in Table II.
Production and Purification of S. cerevisiae MSH2-MSH6 -S. cerevisiae MSH2-MSH6 was purified from cell extracts of galactose-induced cells by sequential chromatography on PBE94, ssDNA cellulose, and Q-Sepharose using a method similar to that described by Alani (47). All chromatography steps were performed at 4°C. Briefly, cells were grown in a fermenter (New Brunswick Scientific) in 10 liters of synthetic dropout medium lacking uracil and leucine and containing 2% raffinose, and production of MSH2 and MSH6 was induced by the addition of galactose to a final concentration of 2%. Galactose-induced cells (50 g) were harvested and then ground under liquid nitrogen using a motorized mortar and pestle (Retsch Grinder) (69). The ground cell paste was resuspended with 50 ml of buffer A (25 mM Tris-Cl, pH 7.6, 1 mM EDTA, 10 mM ␤-mercaptoethanol) containing 0.25 M NaCl and freshly added 1 mM phenylmethylsulfonyl fluoride, and a clarified supernatant was made by ultracentrifugation for 30 min at 50,000 ϫ g at 4°C. This yielded 95 ml of clarified cell extract, which was immediately applied to a PBE94 column (100 ml), which was then washed with 750 ml of buffer A containing 0.25 M NaCl, and eluted with a linear gradient of 0.25-1 M NaCl in buffer A. MSH2-MSH6 was found to elute at 0.45 M NaCl. The MSH2-MSH6 containing fractions (90 ml) were diluted with buffer A containing no NaCl until a conductivity equal to 0.2 M NaCl was obtained. The diluted PBE94 pool was applied to a 15-ml ssDNA Gel Mobility Shift Assay of DNA Binding-High pressure liquid chromatography-purified DNA substrates were 5Ј-end-labeled using [␥-32 P]ATP and T4 polynucleotide kinase and purified by centrifugation through a Centri-Sep column (Princeton Separations, Adelphia, NJ). DNA binding assays were performed by incubating MSH2-MSH6 (50 nM MSH2-MSH6 heterodimer, unless otherwise described) with the 32 P-labeled DNA substrate (10 nM, unless otherwise described) in a final volume of 30 l Binding Buffer (25 mM NaCl, 20 mM Tris-Cl, pH 7.5, 1 mM dithiothreitol, 5 mM MgCl 2 , 100 g/ml bovine serum albumin) for 30 min at 20°C. Glycerol was added to a final concentration of 5% just prior to gel electrophoresis on 4.5% polyacrylamide (60:1 bisacrylamide), 0.5ϫ TBE (45 mM Tris borate, 1 mM EDTA, pH 8.0), 5% glycerol for 3.25 h at 150 V at 4°C. Gels were dried and subsequently analyzed using a PhosphorImager and ImageQuant software (Molecular Dynamics, Sunnyvale, CA).
BIAcore Analysis-By using the BIAcore system (BIAcore 1000), high pressure liquid chromatography-purified biotinylated DNA duplexes were bound to the streptavidin-coated SA sensor chip by injecting 10 l of a 10 nM stock of biotinylated G:T or G:C duplex to yield resonance units (RU) of 73 and 154, respectively. MSH2-MSH6 was dialyzed in running buffer (0.15 M NaCl, 25 mM Tris-Cl, pH 7.5, 1 mM EDTA, 10 mM ␤-mercaptoethanol, 0.005% P20 surfactant) and injected at a flow rate of 50 l/min. This flow rate was maintained throughout each experiment, which consisted of binding and dissociation phases of 120 and 240 s, respectively, unless otherwise specified. Two washes of 50 l of 0.05% SDS, 1 M NaCl, 25 mM Tris-Cl, pH 7.5, 1 mM EDTA were used to regenerate the binding surface after each injection of MSH2-MSH6. All experiments were performed at 25°C. Data were collected at 1 Hz and analyzed using the BIAevaluation software (version 3.0) provided by BIAcore AB.

RESULTS
Purification of MSH2-MSH6 -To purify the S. cerevisiae MSH2-MSH6 complex, we first transformed the protease-deficient S. cerevisiae strain RKY2418 with the 2 m Gal1,10 expression vectors pRDK354 (MSH2) and pRDK568 (MSH6). Cell extracts were made from galactose-induced cells by grinding under liquid nitrogen as described under "Experimental Procedures," and MSH2-MSH6 was purified by sequential chromatography on PBE94, ssDNA cellulose, and Q-Sepharose using a procedure that was similar to that used by Alani (47) (Fig. 1). The final purity of the MSH2-MSH6 complex was determined by densitometry to be greater than 95%. The MSH2 and MSH6 subunits were found to be in a 1:1 molar ratio and to chromatograph as a complex of approximately 250 kDa during gel filtration, consistent with the MSH2-MSH6 complex being a heterodimer. 2 Binding of MSH2-MSH6 to DNA Mispairs-In previous experiments, MSH2 was found to function in a complex with MSH6 that binds to oligonucleotide duplexes containing a T:G mispair and small insertion/deletion loops (30,36,43,47). Here we have tested the ability of the MSH2-MSH6 heterodimer to bind to model sets of oligonucleotide duplexes containing different mispaired bases. Each substrate series was based on a common oligonucleotide duplex 37 base pairs long such that the mispaired bases were located at the center of the duplex, and DNA binding was detected using a gel mobility shift assay.
In experiments where increasing amounts of MSH2-MSH6 were added to a fixed amount of T:G mispair, a specific complex (labeled S, Fig. 2A) was observed at low protein concentrations. At higher protein concentrations, a second, more slowly migrating species was observed (labeled NS, Fig. 2A). Subsequent experiments (see below) indicated that the formation of the NS species did not require the presence of a mispair and likely represented some type of nonspecific complex. Over 90% of the mispair containing substrate was bound at high concentrations of MSH2-MSH6 (Fig. 2B). In experiments where a fixed amount of MSH2-MSH6 was incubated with increasing amounts of T:G-containing mispair, the binding of MSH2-MSH6 to the T:G-containing mispair was found to be saturating at approximately a 1:1 molar ratio of MSH2-MSH6 heterodimer to T:G substrate, consistent with the active species of MSH2-MSH6 being a heterodimer (Fig. 2C).
Parallel experiments were performed with increasing amounts of MSH2-MSH6 and oligonucleotide duplexes containing either a C:G base pair or ϩ1, ϩ2, ϩ4, or ϩ10 insertion/ deletion loops (Fig. 3). In the case of the C:G substrate only a small amount of the "S" species (compare with the T:G substrate, Fig. 2A) was seen at low protein concentrations. In contrast, at higher protein concentrations almost all of the C:G substrate was found in the "NS" species indicating that the formation of this protein-DNA complex did not require a mispaired base. In the case of the ϩ1 and ϩ10 insertion/deletion loop substrates, large amounts of S complex were formed at the same low MSH2-MSH6 concentrations where this complex was formed with T:G substrate, whereas less S complex was formed with the ϩ2 and ϩ4 insertion/deletion loop substrates and was also observed at higher protein concentrations. Each DNA duplex formed the nonspecific NS complex at higher protein concentrations. For the C:G, ϩ2 and ϩ4 insertion/deletion loop substrates, the formation of the NS complex was more efficient than the formation of the specific S complex. Based on the amount of S complex formed at a fixed MSH2-MSH6 concen-2 G. T. Marsischky and R. D. Kolodner, unpublished observations. tration, the relative affinity of MSH2-MSH6 for the different DNA substrates was T:G Ͼ ϩ1, ϩ10 Ͼ ϩ4, C:G and ϩ2 substrates (Fig. 3).
In previous mispair binding experiments with MSH2, MSH2-MSH3, and MSH2-MSH6, it was either not possible to achieve equilibrium binding conditions or equilibrium binding conditions were not investigated (30,43,47,49,52,54,55,58,59,70). In order to determine if equilibrium binding conditions had been achieved under the experimental conditions used here, two types of DNA binding experiments were performed. In one case, 32 P-labeled T:G substrate was mixed with different amounts of unlabeled T:G substrate. MSH2-MSH6 was added, and the amount of specific S complex formed was measured after 60 min of incubation. In the second case, 32 P-labeled T:G substrate was mixed with MSH2-MSH6 and incubated for 30 min to form a MSH2-MSH6-mispair complex. Then different amounts of unlabeled T:G substrate were added, and the amount of specific S complex remaining was measured after 15 min of additional incubation. The results (Fig. 2D) showed that essentially equal amounts of specific complex were present at each competitor concentration, independent of when the competitor was added. A parallel experiment using the ϩ1 insertion/deletion loop mispair as competitor gave identical results (Fig. 2E). In an experiment where the time course of dissociation of a preformed complex was measured after the addition of a large excess of competitor, dissociation was found to be complete in under 5 min (data not shown). These results indicate that under the binding conditions used here, multiple cycles of association and disassociation occur and that equilibrium binding conditions have been achieved.  The achievement of equilibrium binding conditions allowed the determination of equilibrium binding constants for the S complex by Scatchard analysis. A representative analysis for the T:G mispair is presented in Fig. 2F, and the K d values obtained in the analysis of the T:G, C:G, and ϩ1, ϩ2, ϩ4, and ϩ10 insertion/deletion loop substrates are presented in Fig. 3F. Virtually identical K d values were obtained when the data were analyzed using nonlinear least squares curve fitting. The K d values obtained cover an approximate 20-fold range between the T:G and C:G substrates. These values parallel the relative affinity of MSH2-MSH6 for the different substrates observed in direct binding experiments, where the amount of specific S complex was measured at a fixed MSH2-MSH6 concentration.
Specificity of Mispair Binding-To analyze further the specificity of MSH2-MSH6 binding to mispairs, a series of competition experiments were performed in which the 32 P-labeled T:G-containing mispair was mixed with different amounts of unlabeled T:G, C:G, and the ϩ1, ϩ2, or ϩ10 insertion/deletion duplex competitors, and the amount of complex formed between MSH2-MSH6 and labeled T:G containing substrates was measured using gel shift analysis. The relative specificity of mispair recognition was determined by the method of Chi and Kolodner (34) (Fig. 4A). MSH2-MSH6 was found to have a 6 -8-fold higher preference for T:G compared with C:G (theoretical competition was observed for the T:G competitor). The ϩ10 insertion/deletion loop again was found to be recognized by MSH2-MSH6 with the same specificity as the ϩ1 mispair with each being recognized with a 3-fold greater preference than C:G. The small insertion/deletion loop mispairs, ϩ2 and ϩ4, were recognized by MSH2-MSH6 with no greater specificity than C:G. Similar conclusions were reached when the level of competition at a fixed competitor concentration of 100 nM was analyzed (Fig. 4B).
The specificity of binding of MSH2-MSH6 to all possible base-base mispairs was also tested using a competition binding assay in which the ability of a fixed concentration (100 nM) of each oligonucleotide duplex to act as a competitor was measured. As shown in Fig. 5A, the ability of a mispair to compete with a T:G mispair for binding of MSH2-MSH6 was highly variable. For instance, some mispairs were as effective or more effective as competitors than T:G indicating they were recognized by MSH2-MSH6 with high affinity. In contrast, other mispairs were no more effective as competitors than homoduplex DNA, indicating that they were not recognized by MSH2-MSH6 as mispairs. It was also observed that the effectiveness of two of the mispairs, the C:A and G:A mispairs, depended on the orientation of the mispair in the heteroduplex. Hence, the strand and sequence context of a mispair is likely to contribute to their recognition by MSH2-MSH6 (see below).
Since the MSH2-MSH6 complex bound with high affinity to the ϩ10 insertion/deletion loop (Fig. 3E), it was of interest to determine if MSH2-MSH6 would bind to other insertion/dele- tion loop substrates. To investigate this, two series of insertion/ deletion loop substrates were compared with the C:G homoduplex as competitors for T:G mispair binding (Fig. 5, B and C). In one series, an oligonucleotide duplex containing a mononucleotide run of 5 G:C nucleotides was constructed, in which the length of the G strand was varied to contain up to 10 additional Gs that would be expected to form unpaired loops. In the second series, the loop was constructed within a GT dinucleotide repeat, and the length of the GT-containing strand was varied such that loops of up to 16 nucleotides were created. In each series, short loops on the opposite strand (the C or CA strands, respectively) were made by making the repeat unit of the G or A-E, MSH2-MSH6 (0 -250 nM) was incubated with 32 P-labeled 10 nM C:G homoduplex or the ϩ1-, ϩ2-, ϩ4-, or ϩ10 insertion/deletion loop heteroduplexes for 30 min as described under "Experimental Procedures" and analyzed by gel shift assay. F, summary of quantitative binding data. Fraction of heteroduplex DNA bound at 50 nM MSH2-MSH6 (data from A-E and Fig. 2, A and B). Equilibrium dissociation constants for heteroduplexes (data from Fig. 2F) and parallel experiments with other heteroduplexes (data not shown) were calculated as described for the T:G mispair in Fig. 2. GT strand shorter. In each case, insertion/deletion mispairs with a loop size greater than ϩ6 were more effective competitors than the small loop mispairs (ϩ2 to ϩ5) indicating they were well recognized by MSH2-MSH6. Interestingly, the ϩ1 C-loop was not as effective a competitor as the ϩ1 G-loop, suggesting that sequence context may be also be relevant to the recognition of ϩ1 insertion/deletion loop mispairs by MSH2-MSH6. It is possible that the effects of local sequence context explains one observation (55) that a ϩ1 loop was recognized by MSH2-MSH6 with higher affinity than a G:T mispair. An additional feature of these data is that the ϩ2 and ϩ4 insertion/ deletion mispairs in the G and GT series (Fig. 5) were better competitors that the ϩ2 and ϩ4 mispairs in the original substrate series (Fig. 4B) again supporting the view that similar mispairs can be recognized with different affinities, possibly depending on sequence context.
Effect of Sequence Context on Mispair Recognition-Two approaches were taken to explore the effect of sequence context on the recognition of A:C mispairs by MSH2-MSH6. First, the mispair and an increasing number of base pairs on both sides of the mispair were swapped from one strand to the other while maintaining strand polarity (rotational swap), and the resulting oligonucleotide duplexes were tested for their ability to compete with T:G for MSH2-MSH6 binding in order to define a minimum recognition element (Fig. 6). As described above, the A:C mispair was recognized by MSH2-MSH6 ϳ5-fold better than the C:A mispair, as reflected by relative ability to compete for binding with 32 P-labeled T:G mispair. When the A:C and one flanking base pair on each side were swapped, creating a C:A mispair with 2 flanking nucleotides of the well recognized A:C context, the resulting oligonucleotide duplex (AC1) was almost as effective a competitor as the original well recognized A:C mispair. Thus, a well recognized C:A mispair could be created by providing 2 base pairs of flanking A:C context. However, this C:A mispair was recognized about 2-fold less well than the original well recognized A:C. When additional flanking base pairs were swapped (Fig. 6, AC series), recognition was not improved until the entire A:C substrate had been recreated (i.e. the original A:C mispair).
In a complementary experiment, increasing numbers of flanking nucleotides were swapped along with the poorly recognized C:A mispair, resulting in the well recognized A:C within the context of increasing amounts of the original C:A flanking sequence. As an increasing number of pairs of flanking nucleotides were swapped, a poorly recognized, and hence poorly competing, A:C-containing oligonucleotide duplex was created (Fig. 6, CA series). Swapping of 3 base pairs on either side of the C:A was required to produce an A:C in a C:A context that was as poorly recognized and hence was a poor a competitor for binding as the original C:A substrate. Swapping additional nucleotides created even more poorly recognized substrates than the original C:A, and the original level of competition was not restored until the entire original C:A context was created. These results suggest that the mispair and up to 3 base pairs on either side of the mispair define the primary mispair recognition context and that nucleotides that are at least as far as 18 base pairs on either side of the mispair affect the ability of MSH2-MSH6 to interact with the DNA backbone in which the mispair is present.
In the second set of experiments, an increasing number of base pairs on both sides of the mispair were simply flipped to the opposite strand to create oligonucleotide duplexes containing new sequence contexts in which the original base composition was maintained but in which the polarity of the sequence (and, hence, sequence context) was increasingly changed (Fig.  7). Such oligonucleotide duplexes were derived from both the original well recognized A:C and poorly recognized C:A substrates. The ability of MSH2-MSH6 to recognize these was then determined by testing their ability to compete with T:G for MSH2-MSH6 binding. As expected, since none of the mispaircontaining oligonucleotide duplexes derived from either the original A:C or C:A substrates creates an approximation of either the original A:C or C:A sequence context, they do not approximate the effectiveness of the original A:C or C:A substrates as competitors. Some observations can be made, however. First, effects were again observed when changes were made close to the mispair. The effectiveness of the A:C mispair as a competitor markedly decreased by the simple exchange of only the two nucleotides on both sides of the A:C mispair to the opposite strand (mispair ACf2). Similarly, the limited ability of the C:A mispair to act as a competitor was somewhat improved by making the identical swap involving the 2 base pairs on both sides of the C:A mispair (mispair CAf2). Like the results obtained in the first series of experiments (Fig. 6), these results suggest that sequences within 1-3 base pairs of the mispair play important roles in mispair recognition. Second, as additional flanking nucleotides of the A:C and C:A substrates were flipped (ACf and CAf series), the resulting oligonucleotide duplexes showed an increased ability to act as competitors for binding to MSH2-MSH6 indicating that they gained an increased ability to interact with MSH2-MSH6. The ability of both types of substrates to act as competitors plateaued when between 10 and 15 base pairs on both sides of the mispair were flipped and the resulting A:C-and C:A-containing substrates had an approximately equal ability to interact with MSH2-MSH6. These results indicate that sequences as far as 10 -15 base pairs on either side of the mispair play important roles in mispair recognition and are similar to the effects seen in the first series of experiments (Fig. 6).
Effect of ATP and ADP on Mispair Binding-A number of studies have documented the interaction of MutS-related proteins, nucleotides like ATP, and mispair recognition. Several studies have shown that ATP added to mispair binding reactions at the time of MSH2-MSH6 addition reduces the amount FIG. 6. Effect of sequence context on mispair recognition examined by rotational swapping. Competition of binding of MSH2-MSH6 (50 nM) to 32 Plabeled T:G heteroduplex (10 nM) by 100 nM unlabeled duplex DNAs that contained varying amounts of AC or CA core sequence (rotational swap of core sequence). Samples were incubated and processed as described in Fig. 4. 100% is the amount of protein-DNA complex formed in the presence of C:G competitor. Representative DNA substrates are listed below each panel.
of MSH2-MSH6 mispair complex formed (43,47,51,52,54,55), whereas addition of ADP has little or no effect (52,55). One study has shown that ATP addition does not reduce the formation of MSH2-MSH6 mispair complexes in cases where the mispair is known to be poorly repaired in vivo suggesting that the ATP effect is indicative of functional mispair recognition (47). More recently, it has been suggested that ADP stabilized MSH2-MSH6 mispair binding and that exchange of ATP for ADP converts MSH2-MSH6 into an activated diffusable form that signals other components of the mismatch repair system (52,71). These results were obtained using a limited number of mispairs and are to some extent contradictory. It was of interest, therefore, to determine the effect of ADP and ATP on binding of MSH2-MSH6 to the mispaired base containing substrates studied here.
To investigate the effects of added nucleotides on mispair binding, MSH2-MSH6 was mixed with oligonucleotide duplexes to form MSH2-MSH6-DNA complexes, and the stability of these preformed complexes was challenged by the addition of competitor DNA and either no nucleotide, ATP, or ADP for 15 min, and then the amount of MSH2-MSH6-DNA complex remaining was determined using gel mobility shift assays (Fig. 8,  A and B). Note that binding in this experiment was done in buffer containing EDTA so the results should reflect the effect of nucleotide binding rather than hydrolysis. The addition of ATP decreased the amount of specific S complex formed between MSH2-MSH6 and the substrates containing T:G, C:G, and ϩ1, ϩ2, ϩ4 (not shown), and ϩ10 insertion/deletion loop mispairs. The reduction observed ranged from 2-to 3-fold, although in the case of the ϩ10 substrate there was less destabilization. ATP caused a smaller decrease in the amount of specific S complex (0.6-to 0.1-fold) observed for the larger insertion/deletion loop mispairs tested including a ϩ16, a ϩ31, and a "bubble" substrate containing an unpaired region of 31 nucleotides in the center of the duplex. These results indicate that ATP significantly increases the dissociation rate of the MSH2-MSH6-mispaired base-specific S complex formed with many but not all mispaired bases. No significant effect of ATP was observed on the nonspecific NS complex formed with each of the substrates. In no case was it found that the addition of ADP stabilized the specific or the nonspecific MSH2-MSH6 mispair complex, and in most cases, ADP destabilized these complexes slightly. We have also performed experiments where either ATP or ADP were added to binding reactions with each of the substrates studied here at the same time as the MSH2-MSH6 was added. In the case of ATP addition, the results (data not shown) paralleled the results obtained when ATP was added to preformed complexes, with addition of ATP reducing the recovery of MSH2-MSH6 mispair S complexes for the smaller mispairs and having little effect for the larger mispairs. However, in no case did ATP addition reduce the formation of the nonspecific NS complex. Addition of ADP at the start of the binding reaction had no effect with the results obtained (data not shown) being identical to those obtained in the ab- sence of nucleotide.
The effect of nucleotides on mispair binding was explored further with surface plasmon resonance analysis using the BIAcore system (BIAcore AB), a technique which allows realtime monitoring of binding between macromolecules within a flow cell (72,73). In these experiments, biotinylated T:G heteroduplex or C:G homoduplex DNA was first bound to the streptavidin-dextran matrix of the flow cell, and MSH2-MSH6 was subsequently passed over the bound DNA duplex. In the first experiment, the relative affinities of MSH2-MSH6 for the T:G heteroduplex and the C:G homoduplex were determined by passing increasing concentrations of MSH2-MSH6 through the cell (Fig. 9, A and B). MSH2-MSH6 was found to have a 7.5-fold higher affinity for T:G than C:G, which was primarily due to differences in association rate (see Fig. 9G). Also, at saturating MSH2-MSH6 concentrations, MSH2-MSH6 was found to bind to the duplex DNA in a 1:1 molar ratio (750 RU MSH2-MSH6 (after subtraction of bulk and drift components)/73 RU DNA for T:G which have molecular masses of 250 and 25 kDa, respectively). In a second experiment, the effect of ATP and ADP on the binding of MSH2-MSH6 to both the T:G and C:G duplex DNAs was tested by passing a single concentration of MSH2-MSH6 through the cell in the presence or absence of 1 mM ATP or 1 mM ADP (Fig. 9, C and D). For both T:G and C:G in the presence of ATP, there was a large decrease in the rate of association of the MSH2-MSH6-DNA duplex complex (14-and 3.9-fold for T:G and C:G, respectively) and little change in rate of dissociation. Thus, in the presence of ATP, values for k a and k d for both T:G and C:G are essentially the same, suggesting that ATP causes MSH2-MSH6 to recognize T:G as homoduplex DNA (Fig. 9G). The presence of ADP in the buffer had only small effects of the interaction of MSH2-MSH6 with either T:G or C:G (Fig. 9, C, D, and G). Finally, the effect of added nucleotide on the stability of MSH2-MSH6-DNA complexes formed in the absence of nucleotide was tested (Fig. 9, E-G). In this case, MSH2-MSH6 was bound to T:G or C:G duplex DNA on the chip, and the stability of the complex was challenged by continued flow of either buffer or buffer containing 1 mM ATP or ADP. For C:G, ATP and ADP caused similar but small increases in the rate of dissociation of the MSH2-MSH6-DNA duplex complex. In contrast, ADP increased the dissociation of MSH2-MSH6-T:G complex about as much as the MSH2-MSH6-C:G complex, whereas ATP caused a significantly greater increase (3-4-fold) in the dissociation of the MSH2-MSH6-T:G complex. The latter result corroborates the effect of ATP and ADP on the stability of MSH2-MSH6 binding to mispaired and homoduplex DNA observed in the gel shift experiments described above. Similar effects of the addition of nucleotides on the stability of preformed MSH2-MSH6-mispair complexes were recently reported by Blackwell et al. (57).

DISCUSSION
The MSH2-MSH6 complex plays a central role in the repair of mispaired bases in eukaryotes. The results presented here have demonstrated several features of this interaction. First, the MSH2-MSH6 complex binds to a wide variety of mispaired structures albeit with variable specificity. Whereas these interactions can appear highly specific in gel mobility shift assays, the K d values of binding to the best recognized mispaired bases differ by at most 20-fold from the K d of binding to homoduplex DNA. Second, there is a kinetic discrimination by MSH2-MSH6 between T:G and homoduplex DNA which is modulated by nucleotide binding. It was found that ATP affects the rate of association with T:G such that it is equivalent to that of homoduplex DNA, whereas ADP had little effect on the rate of association with T:G compared with no nucleotide. Also, ATP and, to a lesser extent, ADP affect the stability of MSH2-MSH6-mispair complexes, although the addition of ATP increases the off-rate of MSH2-MSH6 by no more than 3-4-fold. These effects may reflect an ATP binding-induced conformational change that is important for mismatch repair. Finally, the affinity of MSH2-MSH6 for a mispair is highly influenced by sequence context. These sequence context effects appear to involve 1-3 base pairs on either side of the mispair as well as base pairs that are as far as 15-18 base pairs on either side of the mispair. The DNA binding studies performed here have demonstrated that MSH2-MSH6 can form two types of complexes with DNA, a specific S complex that requires the presence of a mispair and a nonspecific NS complex that does not require a mispair for formation. Under the experimental conditions used, MSH2-MSH6 appears to be at equilibrium with the DNA substrates in solution and undergoes multiple association and disassociation cycles. These two types of complexes have been observed in other studies (52,54); however, equilibrium binding conditions have not been previously demonstrated. The observation of equilibrium binding conditions has allowed us to determine K d values for the interaction of MSH2-MSH6 with different DNAs. Our data generated using gel shift assays indicate that the K d values for the best recognized mispairs are on the order of 0.1 to 1 nM and that the differences between the K d values for the best recognized mispairs and base pairs are at most 20-fold. The K d values determined using BIAcore analysis were slightly higher than those determined using gel shift assays suggesting that there may be significant experimental differences between the two methods such as the use of higher salt concentrations in the BIAcore analysis to eliminate nonspecific binding of the protein to the chips. However, the affinity differences between T:G and C:G observed with the two methods were similar and were consistent with affinity differences seen by others (54). The affinity differences observed were also consistent with those observed in competition experiments. The ability of MSH2-MSH6 to discriminate between different mispairs was similar to that observed for MSH2 and MSH2-MSH6 for a more limited set of mispairs under conditions where equilibrium binding conditions were either not achieved or not investigated (30,43,47,52,54,55,58,59). These data clearly demonstrate that the ability of MSH2-MSH6 to distinguish between mispaired and correctly paired bases in solution is an intrinsic property of the protein.
We observed a large variability in the ability of MSH2-MSH6 to recognize base-base mispairs. This was expected in some cases, such as the case of the low affinity for C:C, which is known to be poorly repaired in vivo in many cases (74 -77), but seemed surprising for other base-base mispairs. Our analysis of the difference between the A:C and C:A mispairs has indicated that sequence context can affect mispair recognition in at least two different ways. First, bases within 1-3 bases of the mispair can dramatically affect mispair recognition. Such nearest neighbor effects have been observed on mismatch repair in vivo in E. coli (78) and have been attributed to base composition effects. Effects of neighboring sequence have also been observed in vitro with MutS (79,80) and MSH1 (34) and have been implied by a recent limited study of human MSH2-MSH6 (50). Such effects appear to be consistent with previous measurements on the effect of a mispaired base on the energetics of base pairing and base stacking of neighboring nucleotides (81)(82)(83)(84). Second, base pairs as far as 15-18 bases on either side of the mispair can affect mispair recognition. Footprinting studies of MSH2-MSH6 and MutS have indicated that these proteins can interact with DNA over a region of 25 base pairs surrounding a central mispair (52,85,86). Our data suggest that the exact bases at these flanking contact points can affect the overall stability of the MSH2-MSH6 mispair complex such that preferable flanking contacts might stabilize the interaction with a less well recognized mispair and less preferable flanking contacts might weaken the interaction with a well recognized mispair. Historically, studies of the in vivo repair efficiency of different mispairs have reported varying degrees of repair for different mispairs. In light of the data presented here, it seems possible that the variable repair observed reflects sequence context differences, rather than differences in the inherent ability of mismatch repair proteins to recognize mispaired bases.
Studies of mutation rates in S. cerevisiae, and to some extent in human and mouse systems, have suggested that mismatch repair can utilize two different mispair recognition complexes, MSH2-MSH6 and MSH2-MSH3 (23,33,35,41,42,44,60,61,87,88). These studies have suggested that MSH2-MSH6 is responsible for repair of base-base mispairs, that MSH2-MSH3 does not participate in the repair of base-base mispairs, that MSH2-MSH6 and MSH2-MSH3 can each be responsible for repair of 1 base insertion/deletion mispair, and the MSH2-MSH3 is relatively more important for repair of larger insertion/deletion mispairs. This type of analysis is in part a measure of the ability of the two MSH complexes to compete for different substrates within a mixture of different substrates produced by replication errors and does not necessarily indicate that a specific MSH complex cannot actually recognize and participate in the repair of a specific mispair to some extent. The studies presented here indicate that MSH2-MSH6 can bind to a broad diversity of base-base and insertion/deletion mispairs. Most of the insertion/deletion mispairs ranging from ϩ1 to ϩ16 base loops derived from GT, CA, G, or C repeats were recognized by MSH2-MSH6. The exceptions to this were the ϩ2 and ϩ4 loops containing GT and the Ϫ4 and Ϫ6 loops containing CA which were recognized better than C:G but less well than T:G. This somewhat lower affinity recognition of some of the insertion/deletion mispairs may reflect a sequence context effect as discussed above, although we have not specifically investigated this. This pattern of mispair recognition fits well with recent studies of human MSH2-MSH6 and MSH2-MSH3. These studies demonstrated that MSH2-MSH6 can promote the in vitro repair of base-base mispairs and insertion/deletion mispairs containing loops of up to 12 bases (48). The significance of our observations that MSH2-MSH6 can bind to insertion/deletion mispairs containing 16 or greater unpaired bases is unclear given that some studies can be interpreted as suggesting that such substrates may not be repaired by MSH-dependent mismatch repair (61). However, other studies have clearly demonstrated MSH-dependent repair of insertion/deletion mispairs having 26 and 38 unpaired bases suggesting that the recognition of the larger insertion/deletion mispairs seen here could be of functional significance (65,66).
A number of studies have documented that the addition of nucleotides like ATP to DNA binding reactions reduces the recovery of MutS and MSH2-MSH6 complexes with mispaired bases (43,47,51,52,54,55). One study has shown that ATP addition does not reduce the formation of MSH2-MSH6 mispair complexes in cases where the mispair is known to be poorly repaired in vivo, suggesting that the ATP effect is indicative of functional mispair recognition (47). Another study (52) has suggested that the ability of a mispaired base containing DNA to serve as a cofactor for the ATPase activity of MSH2-MSH6 and mediate the dissociation of MSH2-MSH6 mispair complexes reflects an important mechanistic feature of mismatch repair. Our results have extended these studies by demonstrating that MSH2-MSH6 cannot recognize mispairs in the presence of ATP and that ATP increases the off-rate of preformed MSH2-MSH6 mispair-specific complexes. No effect of ATP was observed on the formation or stability of the MSH2-MSH6 nonspecific NS complex. We have, however, been unable to demonstrate that ADP has a significant effect on mispair recognition. Given our results, the ability of MSH2-MSH6 to hydrolyze ATP in the absence of DNA (52) is likely to be what allows MSH2-MSH6 to interact with mispaired bases even if the ADP-bound form does not have an enhanced ability to recognize a mispaired base.
We have observed high affinity binding of MSH2-MSH6 to large insertion/deletion mispairs and reduced ATP inhibition of mispair recognition and reduced ATP destabilization of preformed complexes with increasing insertion/deletion mispair size; relatively small ATP effects were seen with insertion/ deletion loops of 16 bases and greater. We have little information about the in vivo or in vitro repair of the exact mispairs studied here. Base-base mispairs are known to be repaired by MSH2-MSH6-dependent mismatch repair, and insertion/deletion mispairs of up to 12 bases are also repaired by MSH2-MSH6-dependent mismatch repair although with decreasing efficiency (23,48,61). In the case of larger insertion/deletion mispairs, there are examples of inefficient repair of a 38-base insertion/deletion mispair by MSH2-MSH6-and MSH2-MSH3dependent mismatch repair (66) and a 26-base insertion/deletion mispair that is repaired in an MSH2-dependent fashion, although possibly by a different mechanism since RAD10 is also required (65). These results and the above discussed results on the relationship between ATP effects and functional mismatch repair suggest that the high affinity binding to large insertion/deletion mispairs and reduced ATP effects with increasing insertion/deletion mispair size observed here could reflect several possibilities as follows: a reduced ability of large insertion/deletion mispairs to serve as cofactors for mismatch repair; non-functional recognition of the mispairs; or possibly recognition of mispairs that is related to a different type of repair.
What role in mismatch repair might ATP-induced destabilization of MSH2-MSH6 mispair complexes reflect? Recently, it has been suggested that ATP binding converts MSH2-MSH6 into an activated form that leaves the mispair and signals other components of the mismatch repair system (52,71). It has subsequently been shown that MSH2-MSH6 can be converted to a form that slides along DNA, a conformational change that requires a mispair, ATP, and Mg 2ϩϩ . Surprisingly, a nonhydrolyzable ATP analog was found to support the conformational change as well (89). Another possibility that has been suggested is that ATP hydrolysis is required for the formation of DNA loops that might be involved in the coordination of mispair recognition and other steps in mismatch repair like initiation (90). These authors (57) also found evidence for a sliding form of MSH2-MSH6 but found that ATP hydrolysis was essential to its formation. An alternative possibility is that binding of ATP by MSH2-MSH6 induces a mismatch-dependent conformational change that facilitates interactions with other mismatch repair proteins and that in the absence of these other factors an abortive reaction occurs and the MSH2-MSH6 complex dissociates from the DNA. This latter possibility is consistent with recent observations that MutL increases the affinity of MutS for mispaired bases (15). It is also consistent with the observations that the assembly of MSH complexes onto mispairs in the presence of other mismatch repair proteins like MLH1-PMS1 (PMS2 in humans) and proliferating cell nuclear antigen results in higher order complexes that appear stable in the presence of ATP and Mg 2ϩϩ , conditions that destabilize the MSH2-MSH6 mispair complex (55,56). In this regard, an interesting feature of our data is the observation that MSH2-MSH6 formed a small amount of specific S complex with homoduplex DNA and that this complex was sensitive to ATP. In the gel mobility shift experiments, we cannot exclude the possibility that the homoduplex DNA contains a low percentage of mispairs due to problems during oligonucleotide synthesis, even though the method of oligonucleotide purification used should preclude this. However, in the BIAcore experiments, a higher level of binding to homoduplex DNA was observed than could be accounted for by contamination of the oligonucleotide duplexes with duplexes containing mispaired bases, and ATP addition destabilized these complexes. This observation suggests that MSH2-MSH6 can inefficiently form functional complexes on DNA lacking mispairs and that the ATP-induced disassociation reflects abortive turnover of MSH2-MSH6 DNA complexes that have been assembled in the absence of a mispaired base; whether other mismatch repair proteins could assemble onto such complexes is not known.
One striking feature of our data is the apparent contrast between the relatively low affinity differences for MSH2-MSH6 binding to mispairs and base pairs in vitro and the more striking requirement for distinguishing between base pairs and mispairs in vivo. For genes such as CAN1 the mutation rates in mismatch repair-defective mutants containing mutations in genes like MSH2, MLH1, and PMS1 are on the order of 1-5 ϫ 10 Ϫ6 per generation (20,76,91). Given the number of genes in S. cerevisiae, this suggests that there might only be one misincorporated base per 30 -150 cells. Assuming this analysis is correct, mismatch repair must be capable of productively recognizing a single mispair among 2.4 ϫ 10 7 correctly paired bases (the size of the haploid S. cerevisiae genome after DNA replication). It seems unlikely that the relative affinity differences of MSH2-MSH6 for normal base pairs and mispairs can alone account for this in vivo specificity. This raises several possibilities with regard to how mismatch repair specificity can be determined. 1) The MSH complexes might interact directly with other proteins implicated in mismatch repair, and these complexes may have increased affinity for mispairs. 2) By analogy to the recognition of transcriptional promoters by proteins and signal transduction, interaction of MSH2-MSH6 with a mispair may cause a conformational change that facilitates protein-protein contacts required to build a higher order protein-protein structure at the site of the mispair that is required for mismatch repair. Such complexes need not be assembled at the mispair once initial binding has occurred nor need they remain at the mispair once assembly has occurred. Indeed, the interaction of the MLH1-PMS1 complex with MSH-mispair complexes could possibly represent such an example. 3) Finally it has been proposed that binding of MSH2-MSH6 to a mispair may be required to convert MSH2-MSH6 to an activated form that diffuses to and activates other components of mismatch repair (71).