hMSH2-independent DNA mismatch recognition by human proteins.

Two distinct mismatch binding activities are detected using bandshift assays with human cell extracts and DNA with mispairs at defined positions. One requires hMSH2 protein and is absent from extracts of LoVo cells, which contain a partial deletion of the hMSH2 gene. The second activity is independent of hMSH2 and is present at normal levels in LoVo and three other cell lines, which are defective in in vitro hMSH2-dependent binding. The two mismatch recognition activities are distinguished by their sensitivity to polycations and can be resolved by chromatography on MonoQ. hMSH2-independent activity has been purified extensively from wild-type cells and from a cell line deficient in hMSH2-dependent binding. The purified material preferentially recognizes A•C, some pyrimidine•pyrimidine mismatches, and certain slipped mispaired structures. Binding exhibits some sequence preferences. The similar properties of the two mismatch binding activities suggest that they both contribute to mismatch repair.

The recognition and correction of non-Watson-Crick base pairs increases the fidelity of DNA replication (1) and prevents recombination between partially diverged (homeologous) DNAs (2). In Escherichia coli, a single mismatch recognition protein, MutS, initiates both the correction of DNA replication errors and the abortion of homeologous recombination intermediates (3). Mismatch repair is more complex in eukaryotes, and Saccharomyces cerevisiae expresses at least four functional MutS homologs (4). There is some specialization among these, and yeast MSH1 (for MutS Homolog) is mitochondrial, MSH2 participates in repair of nuclear genomic DNA, while MSH3 and MSH4 are involved in recombination (4). Yeast cells express multiple homologs of a second E. coli mismatch correction protein, MutL, which participates with MutS in the initial steps of repair. In this case, genetic and biochemical evidence indicates that the yeast MutL function is carried out by a heterodimer of MutL homologs, the products of the MLH1 and PMS1 genes (5).
Human mismatch correction proteins are particularly important because of their association with hereditary non-polyposis colorectal carcinoma (HNPCC), a syndrome in which there is a familial clustering of colorectal and other carcinomas (6). Germ-line mutations in three homologs of E. coli MutL, hMLH1, hPMS1, and hPMS2, have been implicated in HNPCC (7)(8)(9). A heterodimer between two of these gene products, hMLH1 and hPMS2, constitutes the human MutL function (10). Other human MutL-related genes have been identified, but their functions remain undefined, and there is no known association with disease (8). To date, four human MutS homologs have been described. Duc-1 is the product of a divergent transcript from the dihydrofolate reductase promoter (11). hMSH4 was identified on the basis of its sequence similarity to E. coli MutS. The functions of these proteins are unknown, and neither has been implicated in disease. Two other MutS-like proteins, hMSH2 and GTBP, a 160-kDa protein, participate together in mismatch binding (12,13). The former is of particular importance since germ-line mutations in hMSH2 may underlie about 40% of hereditary non-polyposis colorectal carcinoma cases (14).
It is likely that the extended repertoire of mismatch repair proteins in human cells reflects functional specialization. Biochemical assays provide an alternative to comparisons of sequence homology in the search for putative mismatch repair proteins. Using a bandshift assay with mismatched DNA molecules, we identified a human mismatch binding function that could be distinguished from a known G⅐T mismatch recognition activity (15) by its apparent preference for A⅐C and pyrimidine⅐pyrimidine mismatches (16). The G⅐T binding reaction is now known to be mediated by hMutS␣, a GTBP⅐hMSH2 heterodimer (12,13). Here, we describe the extensive purification of the other mismatch binding function. The activity is independent of hMSH2. It is present in several cell lines defective in hMSH2-dependent binding, including one that is homozygous for a partially deleted hMSH2 gene. Extracts of cells deficient in hMSH2-dependent binding but containing apparently normal hMSH2-independent (A⅐C) binding activity, are unable to repair A⅐C mismatches in vitro, suggesting that the A⅐C binding function does not simply replace hMutS␣ in the repair of A⅐C mismatches. A⅐C binding is carried out by a protein complex of similar size to hMutS␣. The two activities are resolved following extensive purification, and the partially purified binding functions are affected differently by polycations. Preferential binding by both mismatch recognition activities is influenced to some extent by the sequence context of the mismatch. The substrate preferences of the purified A⅐C binding activity provide clues to its possible role in mismatch repair, and the similarities between hMSH2-dependent and -independent binding suggest that the two might play complementary roles.

EXPERIMENTAL PROCEDURES
Materials-Biochemicals were obtained from Sigma except where stated otherwise. Oligonucleotides were synthesized on an Applied Biosystems 380B DNA synthesizer.
The generic 34-mer substrates for bandshift assays are shown below. The precise sequences of mismatched duplexes (oligonucleotides 1-7) are given in Table I. Oligonucleotides were end-labeled and annealed as described previously to generate either paired or mismatched duplexes. For most substrates, oligonucleotides synthesized in two or more independent * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ Recipient of a Postdoctoral Fellowship award under the Human Capital and Mobility Program of the European Union. batches were used. No differences in the ability to serve as substrates for binding were observed between batches.
Cell Culture and Extract Preparation-The thymidine kinase-deficient subline of the Raji Burkitt's lymphoma, its methylation tolerant derivative, RajiF12 (17), and HeLa cells were maintained in spinner culture in RPMI medium containing 5% fetal calf serum (Life Technologies, Inc.). Exponentially growing cells were harvested by centrifugation. The human colorectal adenocarcinoma lines LoVo and SW620 (kindly provided by C. Dixon) were grown, respectively, in Ham's F12 and Eagle's minimal essential medium supplemented by 10% fetal calf serum. LoVo and SW620 cells were detatched from the flasks by trypsin-free cell dissociation solution (Sigma) and harvested by centrifugation. Nuclear extracts for binding were prepared from fresh or frozen (Ϫ80°C) cells as described previously (16). Replication-competent cytoplasmic extracts for use in the mismatch correction assay were prepared from freshly harvested cells by a modification (18) of the method of Li and Kelly (19).
Bandshift Assay-Bandshift assays were performed as previously reported (16). Briefly, cell extract (10 -15 g of protein) was preincubated at 20°C with 40 fmol of matched non-radioactive 34-mer (oligonucleotide 1) in a 20-l reaction buffer comprised of 25 mM Hepes⅐KOH, pH 8.0, 0.5 mM EDTA, 0.1 mM ZnCl 2 , 10% glycerol, 50 g of poly(dI⅐dC)⅐poly(dI⅐dC). After 5 min, the reactions were supplemented with radioactive substrate (20 fmol), and incubation continued for a further 20 min. 10-l aliquots, supplemented with bromphenol blue, were analyzed by electrophoresis on 6% polyacrylamide gels as described (16). Reaction products were detected by autoradiography. When non-radioactive competitor oligonucleotides or spermine were included, they were present during the preincubation and subsequent incubation.
In Vitro Mismatch Correction-The substrates for in vitro mismatch correction are circular M13 heteroduplexes containing a single nick. The HK7 M13 derivatives used in their construction (kindly provided by Dr. Peter Brooks), the details of the heteroduplex cassette, which places mismatches in positions that create diagnostic restriction sites after correction, and the annealing conditions are described by Varlet et al. (20). Briefly, replicative form I M13 molecules (10 g) were linearized by digestion with AvaII. The purified linear duplexes were denatured and annealed in the presence of a 10-fold molar excess of appropriate single-stranded HK7 DNA. Remaining single strands were removed by chromatography on benzoylated napthoylated DEAE-cellulose. The resulting replicative form II heteroduplex preparations were free of single-stranded DNA and generally contained about 10% linear homoduplex molecules. The latter did not affect the analysis as the products of digestion of these molecules could be distinguished from the products of repair.
The assay mixtures (20 l) are based on that of Holmes et al. (21) and contained 20 mM Tris⅐HCl, pH 7.6, 5 mM MgCl 2 , 1 mM glutathione, 50 g/ml bovine serum albumin, 0.1 mM each dNTP, 1.5 mM ATP, 70 mM KCl, and 10 ng of heteroduplex DNA. The reaction was started by addition of extract (70 g) and continued for 60 min at 37°C. Reactions were terminated by the addition of 30 l of 25 mM EDTA, 0.5% SDS, and proteinase K (50 g/ml). After a further 15 min at 37°C, the mixture was extracted with phenol:chloroform followed by chloroform, and DNA was recovered by ethanol precipitation. Precipitated DNA was dissolved in buffer for digestion with AlwNI followed by the appropriate diagnostic restriction enzyme (at high concentrations, some diagnostic enzymes were able to cut mismatched DNA, and each batch of mismatched substrate was therefore titrated with the appropriate enzyme prior to assay to determine the minimum effective concentration and thereby avoid over-digestion). For analysis with XmnI, for which there are two sites in HK7 outside the heteroduplex cassette region, digestion with AlwNI was omitted. Digestion products were separated on agarose gels, denatured, and transferred to Hybond N ϩ filters (Amersham International), which were subsequently probed with radiolabeled randomprimed (Boehringer) M13 DNA. Radioactive DNA was localized by autoradiography.
Purification of Mismatch Binding Activities-All steps were performed at 0 -4°C. Extracts were prepared by homogenizing approximately 2.10 10 cells in buffer A (25 mM Hepes⅐KOH, pH 8.0, 1 mM EDTA, 2 mM ␤-mercaptoethanol, 0.5 mM spermidine, 0.1 mM spermine, 1 mM benzamidine, 0.2 mM 4,(2-aminoethyl)-benzenesulfonyl fluoride⅐HCl, 10 M leupeptin, and 1 M pepstatin). Protein precipitating from the extract between 5 and 55% ammonium sulfate saturation was dialyzed overnight against Buffer A containing 100 mM NaCl. The dialyzed material was applied to a column (6 ϫ 1 cm, diameter) of doublestranded DNA-cellulose equilibrated in buffer A. Bound material was eluted with the same buffer containing 300 mM NaCl. Active fractions were pooled and dialyzed against buffer A containing 100 mM NaCl. The dialyzed material was clarified by centrifugation (13,000 ϫ g, 10 min) and applied to a 1-ml Pharmacia HR5/5 MonoQ column equilibrated in buffer A. After removal of unadsorbed material by washing with buffer A, bound protein was eluted with a 20-ml linear gradient up to 500 mM NaCl in the same buffer. The flow rate was maintained at 0.5 ml/min throughout. Fractions containing A⅐C binding activity were pooled, diluted to 100 mM NaCl, and applied to a 1-ml Pharmacia HR5/5 MonoS column equilibrated in buffer A. Unbound material was removed by washing with buffer A. Bound protein was eluted at 0.5 ml/min with a 20-ml linear gradient to 500 mM NaCl in buffer A. Active fractions were pooled and concentrated to approximately 100 l by Centricon microconcentrator (Amicon). The concentrated material was applied to a precalibrated 25-ml Pharmacia HR10/30 Superose 12 gel filtration column equilibrated in buffer A. The column was developed at a flow rate of 0.25 ml/min. Active fractions were pooled, aliquoted, and stored at Ϫ80°C. Purified material retained full activity under these storage conditions for at least 3 months.
SDS-polyacrylamide gel analysis was performed using 6 or 8% gels. Staining was Coomassie or Silver Stain Plus (Bio-Rad). Protein concentrations were determined by the Bradford method or estimated from A 280 . (Table I, oligonucleotides 2 and 3) containing a G⅐T or A⅐C mismatch previously identified two different types of mismatch binding. G⅐T and A⅐C binding activities 1 can be distinguished 1 Since mismatch binding is by a complex of more than one protein, we refer to binding activities rather than to binding proteins. A mismatch binding activity that complements the mismatch repair defect of LoVo extracts is a heterodimer involving hMSH2. This complex has been designated hMutS␣ (12). The same heterodimer was purified as a G⅐T binding activity using the bandshift assay described here (22). For this reason, and since it is absent from extracts of LoVo cells, it is most likely that the G⅐T binding activity we observe is hMutS␣. We retain the descriptive term "A⅐C binding" as a convenient shorthand to describe OLIGONUCLEOTIDES 9 -12

hMSH2-independent Mismatch Binding
by the rate of migration of the respective DNA-protein complexes in the bandshift assay. Extracts of several mutator human cell lines, the methylation tolerant RajiF12, the colorectal carcinoma lines LoVo and DLD1/HCT15, and the methylation tolerant hamster line Chinese hamster ovary clone B, are defective in G⅐T binding (17,23,24). All contain A⅐C binding activity comparable to that seen in Raji (17) and the G⅐T binding proficient colorectal carcinoma cell lines SW620, HRA19, and LS174T ( Fig. 1, a and b and data not shown). Inclusion of various non-radioactive competitor oligonucleotides in the binding assay indicated that the complex formed by LoVo extracts with the A⅐C substrate is the same as complex A formed by the extracts of wild-type Raji cells. In particular, an unlabeled A⅐C heteroduplex was an efficient competitor and effectively abolished binding when present in 30-fold excess, whereas the same concentrations of homoduplexes and a G⅐T heteroduplex were not significantly inhibitory (Fig. 1b). These observations confirm the selective binding to mismatched DNA and the preference for A⅐C over G⅐T mispairs. Since LoVo cells are homozygous for a large deletion in the hMSH2 gene, the data also provide direct evidence that hMSH2 is not required for A⅐C binding.
In Vitro Mismatch Correction and the A⅐C Binding Activity-To investigate if the A⅐C binding activity might substitute for the G⅐T binding function in correction of A⅐C mismatches, we measured A⅐C mismatch repair in an in vitro assay. Correction of an A⅐C mismatch by extracts of the G⅐T binding-defec-tive DLD1, LoVo, and RajiF12 cells and control HeLa, SW620, and Raji cells was compared. Cell extracts were incubated with circular heteroduplexes that contained a single A⅐C mispair. The mismatch is positioned in a heteroduplex cassette sequence such that correction to a G:C base pair creates an MluI site, whereas correction to A:T generates a unique ClaI site (Fig. 2a). Mismatch repair in vitro by human cell extracts requires a nick in one of the DNA strands, and correction is directed to the nicked strand (21). The A⅐C heteroduplex contained a nick in the mismatched A-containing strand and was efficiently repaired by HeLa, SW620, and Raji cell extracts. After incubation with extract, substrate DNA was recovered, linearized with AlwNI and further digested by MluI. The DNA was cleaved into fragments of 4.2 and 3.2 kilobases, indicating repair of the A⅐C mismatch to G:C. No correction to A:T (ClaI sensitivity) was detected, confirming that repair was strandspecific (data not shown). In contrast to wild-type cells, extracts of G⅐T binding-defective LoVo, DLD1, or RajiF12 cells did not carry out detectable repair (Fig. 2b). In a second substrate, the mismatch was inverted, and the nick was positioned in the C-containing strand. In this case, nick-directed repair to a T:A base pair generates an additional XmnI site (Fig. 2a), whereas correction in the opposite orientation produces a unique EcoNI site. The presence of fragments of 3.7 and 1.4 kilobases in XmnI digests of substrate DNA recovered from extracts of HeLa, SW620, or Raji cells indicated the creation of a new XmnI site by nick-directed correction to a T:A base pair. Extracts of LoVo, DLD1, or RajiF12 cells did not carry out detectable repair (Fig.  2b). Thus, wild-type cell extracts can correct A⅐C mismatches in two different sequence contexts by a nick-directed process. The normal levels of A⅐C binding activity in G⅐T binding-deficient extracts do not compensate for the absence of the G⅐T recognition function in correction of A⅐C mismatches in the heteroduplex cassette substrate.
Effect of Polycations on Mismatch Binding-The A⅐C and G⅐T recognition activities could be distinguished by their differential sensitivity to polycations. When fractions enriched for both A⅐C and G⅐T binding by successive ammonium sulfate, heparin-Sepharose, and DNA-cellulose chromatography were assayed in the presence of increasing concentrations of spermine, the polycation stimulated G⅐T binding. In contrast, A⅐C binding was progressively inhibited over the same range of spermine concentrations (Fig. 3a) and was essentially abolished at 1 mM. Spermine was more effective than spermidine (data not shown), suggesting that charge neutralization might underlie the observed effects. When a similarly enriched fraction from RajiF12 cells was used, A⅐C binding was inhibited in an identical fashion to the wild-type fraction (Fig. 3b). As expected, no G⅐T binding was observed either in the presence or the absence of spermine. Essentially identical results were obtained using partially purified LoVo cell extracts (data not shown). The difference in susceptibilities of the two binding activities to spermine provides supporting evidence that G⅐T and A⅐C binding are carried out by different proteins or complexes. The effect of the polycation on A⅐C binding in F12 and LoVo further indicates that the A⅐C binding activities in the variant cells have similar properties to their wild-type counterpart.
Fractionation of Binding Activities-The standard assay indicates that A⅐C, but not G⅐T, binding activity is recovered by precipitation with ammonium sulfate between 5 and 35% saturation. The loss of G⅐T binding is not due to its inactivation since both functions are recovered in a 5-55% fraction. The A⅐C and G⅐T binding activities in an ammonium sulfate fraction of a wild-type Raji cell extract copurified through subsequent consecutive heparin-Sepharose, AcA34 gel filtration, and DNAcellulose chromatography. Both bound to heparin-Sepharose the hMSH2-independent binding function. This is not intended to imply, however, that this mismatch is necessarily its preferred substrate.
FIG. 1. A⅐C mismatch binding by colorectal carcinoma cell extracts. a, extracts (12 g) of the human colorectal carcinoma cell lines indicated were combined with radiolabeled A⅐C-mismatched duplex 34-mer oligonucleotide as described under "Experimental Procedures." Bound and free oligonucleotides were separated by electrophoresis on a 6% polyacrylamide gel and located by autoradiography. Complex A is the A⅐C-specific complex. b, extracts of Raji or LoVo cells were preincubated with a 30-fold excess of the non-radioactive duplex oligonucleotide indicated. Radioactively labeled A⅐C substrate was then added, and complex A formation was analyzed by gel electrophoresis.
hMSH2-independent Mismatch Binding from which they coeluted close to the end of a 0.1-1 M NaCl gradient. They were retained by double-stranded DNA-cellulose and were eluted at approximately 0.3 M NaCl (data not shown). A⅐C and G⅐T binding cochromatographed on AcA34 at the position corresponding to a monomeric M r of approximately 250,000 (data not shown). The elution position was unchanged over a range of NaCl concentrations from 0.1 to 1 M, suggesting that this apparent large size was not due to nonspecific protein aggregation. A similar estimated M r for both binding functions was obtained by gel filtration on Sephacryl 300. Thus, the two binding activities share several physical properties including a similar size.
A⅐C and G⅐T mismatch binding activities were resolved by chromatography on MonoQ. An extract of Raji cells enriched for both activities by sequential ammonium sulfate and DNAcellulose fractionation was applied to a MonoQ column in buffer containing 0.1 M NaCl. The bound protein was eluted by a gradient of NaCl to 0.5 M. Fractions were assayed for both A⅐C and G⅐T mismatch binding. A⅐C binding activity eluted early in the gradient at approximately 0.2 M NaCl. It was detected by the formation of complex A (Fig. 4a) with an A⅐C substrate. No G⅐T binding (complex B) was observed in these early fractions. The peak of A⅐C binding was partially coincident with the formation of a complex that migrated close to the well of the gel. This complex was not mismatch-specific and was formed to the same extent with both A⅐C and G⅐T substrates. In fractions containing very high levels of A⅐C binding activity, a small amount of complex A was also formed with the G⅐T substrate, suggesting that while formation of this complex with A⅐C mismatches is highly preferred, this preference is not absolute. Similar behavior has been observed with unfractionated ex-tracts (16). G⅐T binding (complex B) eluted after the A⅐C and this nonspecific binding activity. There was a slight overlap with the tail of the A⅐C binding peak (fraction 13, Fig. 4a), but the later G⅐T binding fractions did not bind detectably to A⅐C. Two other rapidly migrating complexes (C and D) were also seen. These complexes, which have been noted previously in crude extracts, are not mismatch specific and were formed to similar extents with both substrates. Thus, in addition to their differential susceptibility to spermine, the two binding activities also exhibit different behavior on ammonium sulfate fractionation and can be physically resolved on MonoQ.
A⅐C binding activity was purified from extracts of the G⅐T binding-defective RajiF12 cells by the same procedure. Fig. 4b shows the mismatch binding activity in MonoQ fractions. Complex A was detected with an A⅐C substrate in fractions 9 -14 and again partially overlapped fractions forming the slowly migrating nonspecific complex. No complex B was detected in any fraction. The nonspecific binding (complex D) that partially overlapped the G⅐T binding activity in wild-type preparations was also observed. This serves as an internal control for these fractions. Similar data were obtained from MonoQ chromatography of extracts of LoVo cells (data not shown). Thus, the A⅐C binding activity, which is present at wild-type levels in G⅐T binding-defective cells, has the same purification properties as the wild-type A⅐C binding activity. This suggests that the two mismatch binding activities are independent of one another.
Purification of A⅐C Binding Activity-A⅐C binding activity was purified from the MonoQ fraction of wild-type Raji cells by further chromatography on MonoS and Superose 12. A⅐C binding eluted from Superose 12 slightly later than the nonspecific binding activity with which it had copurified, at a position hMSH2-independent Mismatch Binding corresponding to a monomeric M r of approximately 250,000 -300,000 (Fig. 5). This value is in good agreement with previous estimates using less purified material on AcA34 and Sephacryl 300. The bandshift assay is semiquantitative, and the overall purification is difficult to assess, but we estimate the A⅐C binding activity to be enriched by more than 200-fold.
SDS-polyacrylamide gel analysis of several preparations indicated that active A⅐C binding fractions (complex A) were associated with prominent silver-staining protein bands of estimated M r ϭ 110,000 and 140,000. Examples of two preparations, purified by slightly different procedures, are shown in Fig. 6. Although several stained bands are apparent in the purified fraction, two relatively intense bands at 110 and 140 kDa were observed in these and several other independent preparations. A prominent protein larger than 250 kDa was observed in the fractions catalyzing the formation of the slowly migrating nonspecific complex. The intensity of this band from different column fractions was proportional to the extent of formation of the nonspecific complex but not of the A⅐C mismatch-specific complex A. We conclude that this protein is not responsible for A⅐C mismatch recognition. Since gel filtration indicates that the native M r of the A⅐C binding activity is around 250,000, a likely possibility is that the A⅐C binding complex is a heterodimer of approximately 250 kDa comprising the prominent 140-and 110-kDa proteins that were consistently observed in the purified fractions.
Substrate Preferences of the A⅐C Binding Activity-The wildtype A⅐C binding activity that had been purified through the MonoQ step was assayed for its ability to recognize other mismatches. When the A⅐C mismatch in the standard 34-mer duplex was replaced by T⅐T or T⅐C (Table I, oligonucleotides 4 and 5), complex A was formed indicating that these mismatches are alternative substrates for the A⅐C binding activity (Fig. 7a). This possibility was investigated further by including nonradioactive T⅐T or T⅐C competitor oligonucleotides in the assay. Only A⅐C and these pyrimidine⅐pyrimidine mismatches were effective competitors of complex A (Fig. 7b), confirming that they are efficiently recognized. G⅐T or G⅐G 34-mers were poor inhibitors of binding to the A⅐C substrate. It is notable that little binding was seen with C⅐C or the purine⅐purine mismatch in the same 34-mer sequence (Fig. 7a and data not shown) nor were these duplexes efficient competitors for A⅐C binding (Fig. 7b). These observations agree well with the specificity previously reported for the A⅐C binding activity in cell extracts.
Unpurified cell extracts also form complex A on certain looped mismatched structures (23). The same 40-mer substrate containing the sequence (AT) 2 in which one dinucleotide repeat is displaced from the duplex (oligonucleotide 9) was also recognized by the purified fraction, and complex A was formed (Fig.  7a). An analogous substrate (oligonucleotide 10) in which a CA loop displaced from (CA) 2 replaced the AT loop was not detectably recognized by the purified A⅐C binding activity (Fig. 7a). This substrate was bound, however, by a late-eluting fraction from MonoQ that contained the G⅐T binding activity (data not shown). These data provide further evidence for the different preferences for mismatch recognition by the two binding functions and suggest that some slipped mispaired intermediates are also substrates for the A⅐C binding activity.
A third set of mismatched substrates was based on the heteroduplex cassette sequence in the in vitro mismatch correction assay (oligonucleotides 11 and 12). Surprisingly, the purified A⅐C binding activity did not efficiently recognize a 48-mer duplex containing a single A⅐C mismatch at the same position as the repair substrate (Fig. 2a). Complex A was formed but only to a limited extent, and binding was similar to that seen with a G⅐T (Table I, oligonucleotide 3) substrate (Fig. 8a). When a MonoQ fraction enriched for G⅐T binding, but containing a small amount of A⅐C binding activity (far right lane, Fig. 8a), was used with the A⅐C mismatched 48-mer, complex B was formed. This preferential binding by the G⅐T activity to the A⅐C mismatch in this substrate was confirmed with unfractionated extracts (data not shown). Neither binding activity detectably recognized a substrate containing an AT loop in an (AT) 2 sequence placed in the heteroduplex cassette region. Thus, recognition of both single base mismatches and slipped/mispaired structures by one or either of the two mismatch binding activities exhibits some dependence on the sequence context of the mispairs.
Another striking example of sequence effects on mismatch recognition was provided by the standard 34-mer substrate. When the A⅐C mismatch in this substrate was inverted (Table  I, oligonucleotide 7), thus altering its surrounding sequence context, the A⅐C binding activity no longer recognized the A⅐C mismatch. This mismatch was also a substrate for the partially purified G⅐T binding activity, and complex B was formed (Fig. 8b). DISCUSSION The complexity of mismatch repair in eukaryotic cells is reflected in the abundance of different human mismatch repair proteins. Since there are at least four MutS homologs in yeast, some diversification of mismatch recognition is also likely in human cells. It is interesting to note that the constituents of hMutS␣, hMSH2 and GTBP, are both MutS homologs (12,13) and so each may contribute to mismatch recognition. The A⅐C binding activity may be an example of the diversity in mismatch recognition.

hMSH2-independent Mismatch Binding
Not all human mismatch recognition proteins are necessarily involved in mismatch repair. Mg 2ϩ and ATP-independent mismatch binding detected by bandshift assays is associated with mammalian DNA topoisomerase I (25). Topoisomerase I can be excluded as a candidate for the A⅐C binding activity on the basis of its size and its substrate preferences. It is a single polypeptide of M r ϭ 95,000, which is sometimes purified as a proteolysis product of approximate M r ϭ 80,000, whereas the A⅐C binding activity is purified as an apparently stable complex of estimated M r Ն 250,000. Topoisomerase I acts on all single base mismatches, whereas the purified A⅐C binding complex is more fastidious in its recognition. Two other human proteins that bind to mismatches and can be detected by the bandshift assay have been described. One of these, hMutY, the human homolog of the E. coli MutY protein, is a DNA glycosylase that removes A residues from A⅐G, A⅐7,8-dihydro-8-oxoguanine, and, with 30-fold less efficiency, A⅐C mismatches (26). hMutY and the recently described deoxyinosine 3Ј-endonuclease (27) can be excluded as candidates for the A⅐C binding activity both on the basis of their sizes (65 and 25 kDa, respectively) and of their substrate preferences. In initial fractionation studies of human cell extracts, A⅐C binding activity was associated with a protein, or complex, of approximately 100 kDa (16). It is possible that this reflects an autonomous binding activity of the smaller component of an A⅐C complex dissociated at relatively low protein concentrations.
Our data suggest that the A⅐C binding complex is a dimer of proteins of approximately 110 and 140 kDa. This composition is similar to that of the G⅐T binding complex of hMSH2 and GTBP (12,13) (104 and 160 kDa). Despite this similarity in size, the complexes formed with A⅐C and G⅐T substrates are resolved in the bandshift assay. The faster migration of the A⅐C complex suggests that the A⅐C binding proteins may be more acidic. A difference in surface charge is consistent with separation of the A⅐C and G⅐T binding activities on MonoQ and may also underlie the different effect of spermine on A⅐C and G⅐T binding. Polyamines can affect protein-DNA interactions in several ways, including inducing DNA bending (28) and altering the site FIG. 6. Proteins involved in A⅐C binding. Two independent preparations of A⅐C binding activity from Raji cells were analyzed by SDSpolyacrylamide gel electrophoresis followed by silver staining. The positions of the molecular weight standards are indicated on the right side of the figure. The arrows on the left side indicate the positions of the two most prominent silver-stained bands that were reproducibly observed in several preparations. Lane 1, Raji activity purified by ammonium sulfate (5-55%) fractionation followed by successive double-stranded DNA-cellulose, MonoQ, MonoS, and Superose 12 chromatography as described under "Experimental Procedures." Lane 2, Raji activity purified by an identical procedure except that a 5-35% ammonium sulfate fractionation was used, and the MonoS chromatography step was omitted.
hMSH2-independent Mismatch Binding occupancy by binding proteins (29). A contribution of one or more of these other properties to the different effect of spermine on A⅐C and G⅐T binding cannot be excluded.
The presence of A⅐C binding activity in RajiF12, LoVo, and DLD1 extracts, all of which have no detectable G⅐T binding, indicates that A⅐C binding is a separate function. This is supported by our partial purification of the A⅐C binding activity from RajiF12 and LoVo cells and the demonstration that its properties are indistinguishable from the wild-type activity. Since LoVo and DLD1 contain mutations in hMSH2 (14,30) and GTBP (31), respectively, both of which inactivate G⅐T binding, it appears that neither hMSH2 nor GTBP is required for the A⅐C binding reaction.
Although we have no direct evidence that the A⅐C binding activity participates in mismatch correction, their similarities in size and general properties suggest that the G⅐T and A⅐C binding complexes may play complementary roles. We previously reported that the A⅐C binding activity in unfractionated cell extracts efficiently recognizes T⅐C and T⅐T mismatches but binds to C⅐C only poorly (16). These substrate preferences are apparent in the purified A⅐C binding complex. Since of all single base mismatches, C⅐C mispairs are corrected least efficiently in vitro and in vivo (21,32), these observations are consistent with an involvement of the A⅐C binding activity in mismatch repair. The purified binding complex also recognizes AT dinucleotides displaced from a (AT) 2 sequence. This ability to bind to displaced dinucleotides is analogous to the recognition of displaced CA dinucleotides by the G⅐T binding function (23) and suggests that the A⅐C binding activity might also be involved in stabilizing microsatellite sequences.
Recognition by the A⅐C mismatch binding activity does not exhibit an absolute dependence on DNA sequence context because the A⅐C complex is formed on A⅐C mispairs and AT loops in completely unrelated sequences. The context of the mismatch can, however, influence its recognition. We observed that neither A⅐C mismatches nor AT loops in a third, unrelated, sequence were recognized by the purified A⅐C binding complex. Instead, these mismatches were substrates for the G⅐T binding activity. Several other observations are also consistent with a degree of overlap in mismatch recognition by the A⅐C and G⅐T binding activities. At high concentrations, the purified A⅐C complex binds to G⅐T mismatches. This can be seen in the MonoQ fractions that are highly enriched for A⅐C binding activity. In addition, purified hMutS␣ binds (albeit rather poorly) to A⅐C mismatches (22). The rules that govern recognition cannot be inferred from these few data, but local sequence determinants may contribute to the probability that mismatches are recognized by one or other of the binding heterodimers.
Data from the in vitro mismatch correction assays indicate, however, that mismatches cannot simply be divided into two groups, one comprising substrates for hMutS␣ and the other the mismatches recognized by the A⅐C complex. LoVo cell extracts cannot repair several mismatched structures in vitro, implicating hMSH2 in their correction. Our data indicate that the repair deficiency of LoVo extends to A⅐C (and C⅐A) mismatches, at least in the particular context of the heteroduplex cassette. The inability of LoVo, DLD1, and RajiF12 cell extracts to correct A⅐C mismatches in the in vitro assay is paralleled by the absence of recognition by the A⅐C binding activity of A⅐C mismatches in the heteroduplex cassette sequence. A⅐C mis- FIG. 7. Substrate preferences of A⅐C binding activity. a, The standard bandshift assay was carried out using 34-mer duplexes containing the single base mispair indicated (Table I) and the purified A⅐C binding activity. The A⅐C complex (A) is shown with an arrow. The substrates containing the AT and CA loops are shown under "Experimental Procedures" (oligonucleotides 9 and 10). b, the standard bandshift assay was carried out using the radiolabeled 34-mer duplex containing an A⅐C mispair (Table I, oligonucleotide 2). Non-radioactive competitor 34-mer duplexes were included in 30-fold excess as indicated.
FIG. 8. Effect of sequence context on A⅐C and G⅐T binding activities. a, the bandshift assay was carried out using the standard A⅐C mismatched 34-mer (oligonucleotide 2) (A⅐C, first and last lanes), the standard G⅐T mismatched 34-mer (oligonucleotide 3) (G⅐T), an A⅐C mismatched substrate based on the sequence in the heteroduplex cassette of the in vitro mismatch repair assay (A⅐C, lanes 3 and 6) (oligonucleotide 10), or an AT loop in the heteroduplex cassette sequence (oligonucleotide 12). The A⅐C and G⅐T complexes (A and B) are shown with arrows. Binding in the four left lanes is by the purified A⅐C binding activity. Binding in the four right lanes is by a MonoQ fraction enriched for G⅐T binding activity but which contains a low level of A⅐C binding activity. b, the MonoQ fraction containing both A⅐C and G⅐T binding activities assayed with the standard A⅐C mismatched 34-mer (A⅐C) (oligonucleotide 2), the standard G⅐T mismatched 34-mer (G⅐T) (oligonucleotide 3), or the standard 34-mer in which the A⅐C mismatch was inverted (oligonucleotide 7 in Table I). The A⅐C and G⅐T complexes (A and B) are shown with arrows. match recognition in this particular context is mediated by the G⅐T complex, which is absent from these cells. hMutS␣ and the A⅐C mismatch binding activity may turn out to have complementary but partially overlapping roles in mismatch recognition.
The in vitro assay provides a good indication as to whether particular mismatches are repaired. These assays are building up a picture of the requirements for hMutS␣ in the correction reaction. A caveat should be added, however. The assays provide a strand discrimination signal for correction in the form of a nick in the substrate DNA. The possibility that the mismatch binding activities participate in strand selection during correction of their preferred mispairs has not been ruled out. Provision of a pre-incised substrate may obviate the need for this selective recognition and may miss an important property of the mismatch recognition complexes.
Many DNA repair functions are partially duplicated by back-up activities with similar specificities. The A⅐C binding activity might be one of these and may serve a complementary function to the hMutS␣ complex in mismatch recognition and initiation of repair.