Isolation of MutSβ from Human Cells and Comparison of the Mismatch Repair Specificities of MutSβ and MutSα*

A human MSH2-human MSH3 (hMSH2·hMSH3) complex of approximately 1:1 stoichiometry (human MutSβ (hMutSβ)) has been demonstrated in several human tumor cell lines and purified to near homogeneity. In vitro, hMutSβ supports the efficient repair of insertion/deletion (I/D) heterologies of 2–8 nucleotides, is weakly active on a single-nucleotide I/D mispair, and is not detectably active on the eight base-base mismatches. Human MutSα (hMutSα), a heterodimer of hMSH2 and hMSH6, efficiently supports the repair of single-nucleotide I/D mismatches, base-base mispairs, and all substrates tested that were repaired by hMutSβ. Thus, the repair specificities of hMutSα and hMutSβ are redundant with respect to the repair of I/D heterologies of 2–8 nucleotides. The hMutSα level in repair-proficient HeLa cells (1.5 μg/mg nuclear extract) is approximately 10 times that of hMutSβ. In HCT-15 colorectal tumor cells, which do not contain hMSH6 and consequently lack hMutSα, the hMutSβ level is elevated severalfold relative to that in HeLa cells and is responsible for the repair of I/D mismatches that has been observed in this cell line. LoVo tumor cells, which are genetically deficient in hMSH2, lack both hMutSα and hMutSβ, and hMSH3 and hMSH6 levels are less than 4% of those found in repair-proficient cells. Coupled with previous findings (J. T. Drummond, J. Genschel, E. Wolf, and P. Modrich (1997) Proc. Natl. Acad. Sci. U. S. A. 94, 10144–10149), these results suggest that hMSH2 partitions between available pools of hMSH3 and hMSH6 and indicate that hMSH2 positively modulates hMSH6 and hMSH3 levels, perhaps by stabilization of the polypeptides upon heterodimer formation.

Correction of mismatched base pairs, resulting from replication errors, contributes significantly to the genetic stability of human cells (reviewed in Refs. [1][2][3]. Mutations in the gene that encodes the human MutS homolog MSH2 (hMSH2) 1 confer genetic instability and have been implicated in hereditary nonpolyposis colon cancer and sporadic tumors (4 -6). The muta-tion rate at the HPRT locus is greatly increased in the MSH2 Ϫ/Ϫ LoVo colorectal tumor cell line. The majority of the selectable HPRT mutations that occur in LoVo cells are transitions (80%), with the remainder largely single-nucleotide frame shifts in mononucleotide repeat sequences (7), although (CA) n microsatellite repeats are also highly unstable in this cell line (8). Extracts prepared from LoVo cells are deficient in repair of heteroduplexes containing either base-base or small insertion/deletion (I/D) mispairs (8,9), and a mismatch recognition activity that restores repair of a G-T base-base mismatch and TG I/D mispair to LoVo nuclear extracts has been isolated from HeLa cells (9). This activity, designated hMutS␣, is a heterodimer of hMSH2 and hMSH6 (also called p160 or GTBP (9,10)).
Nuclear extracts of the HCT-15 colorectal tumor cell line (11) and the alkylation-tolerant MT1 lymphoblastoid cell line (12) are also deficient in mismatch repair due to MutS␣ deficiency (9,13), but the MutS␣ defects in these cell lines are a consequence of MSH6 mutations (14). Extracts of HCT-15 and MT1 cells are deficient in repair of base-base mismatches and singlenucleotide I/D mispairs but retain partial proficiency in the correction of 2-, 3-, and 4-nucleotide I/D heterologies (9,13). The HPRT mutation rate is elevated 60-fold in MT1 cells (12), which harbor missense mutations in both MSH6 alleles (14), and 300-fold in HCT-15 cells (15) in which both MSH6 alleles have been inactivated by frame shift mutations. HCT-15 cells also contain a sequence change in a conserved region of one copy of the gene that encodes DNA polymerase ␦ (11), but this mutation does not appear to contribute significantly to the HCT-15 mutator phenotype (16). Although dinucleotide repeats are relatively stable in these MSH6 Ϫ/Ϫ cell lines, mononucleotide repeats are prone to mutation (13,14,17), but not to the extent observed with HCT-116 cells (13,15), which are defective in base-base and I/D mismatch repair (18,19) due to mutations in both alleles of MLH1 (20).
The proficiency of MSH6 Ϫ/Ϫ cells in the repair of I/D mismatches suggests the existence of a second mismatch recognition activity distinct from hMutS␣ in mammalian cells. Since hMSH2-deficient cell lines are defective in the repair of both base-base mispairs and I/D heterologies, this hMSH6-independent mismatch activity would appear to require hMSH2. There are two candidates for such an activity: a heterodimeric complex of hMSH2 and hMSH3, with the locus encoding the latter protein being the first MutS homolog gene identified in mammalian cells (21,22), or free hMSH2 in an unknown oligomeric state. Recombinant hMSH2 has been reported to bind to I/D mismatches (23), but recent work indicates that this is an extremely low affinity interaction (24,25), and free hMSH2 has not been detected in human cells (26). However, a complex between recombinant hMSH2 and hMSH3 polypeptides has been demonstrated (24,25), and a similar complex has been isolated from methotrexate-resistant human cells in which the DHFR-MSH3 region of chromosome 5 is highly amplified (26).
The hMSH2⅐hMSH6 complex has been shown to bind specifically to a G-T mispair and to 1-, 2-, and 3-nucleotide I/D mismatches and to efficiently restore repair of a base-base and a dinucleotide I/D mismatch to hMSH2-deficient nuclear extracts (9,24). By contrast, the recombinant hMSH2⅐hMSH3 hMutS␤ complex was shown to bind weakly to a single-nucleotide insertion/deletion mismatch and with high affinity to heteroduplexes containing 2, 3, 4, or 10 unpaired nucleotides but not to the several base-base mismatches tested (24,25). These in vitro binding specificities indicate that hMutS␣ and hMutS␤ have overlapping specificities for I/D mismatches and that the residual I/D heteroduplex repair activity observed in extracts of MSH6 Ϫ/Ϫ cells may be due to hMutS␤. We show here that this is in fact the case and also describe the specificities of hMutS␣ and hMutS␤ in strand-specific mismatch correction.
Mismatch Repair Assays-Mismatch repair assays were performed as described previously (9,27). Briefly, 50 g of nuclear extract was incubated with 100 ng (24 fmol) of a f1-derived heteroduplex DNA substrate in 10 -15 l at 37°C for 15 min. The final salt concentration in the assay was 100 -110 mM KCl. Assays of immunodepleted extracts were performed similarly except that reactions contained 100 g of extract protein, and incubation was extended to 60 min. Complementation of deficient extracts was achieved by the addition of purified hMutS␤ or hMutS␣ (9) as indicated. The hMutS␤ preparations used for complementation were isolated from HL60-R cells as described below. Circular phage f1 heteroduplexes were prepared as described (28,29) and contained a nick 181 nucleotides 3Ј to the mispair (short path) in the case of the 3Ј substrates or 125 nucleotides 5Ј to the mispair (short path) in the case of the 5Ј substrates. Heteroduplex substrates containing I/D heterologies of different sizes have the following nonpaired sequences: dA (1 nucleotide), d(CA) (2 nucleotides), d(CTG) (3 nucleotides), d(CTCGA) (5 nucleotides), d(ACACTCGA) (8 nucleotides), d(A-CACACACTCGA) (12 nucleotides), d(TTTCTAGACTCGACAGCTG-GCTAGCAA) (27 nucleotides). I/D heteroduplex substrates are sometimes described by the unpaired sequence that forms the loop, e.g. CA I/D. For 5Ј-I/D heteroduplexes the extra nucleotides were in the nicked complementary DNA strand, except for the A I/D heteroduplex, which contained an unpaired adenine in a run of six adenines within the continuous viral strand (9). For the 3Ј-CA I/D heteroduplex, the two extra nucleotides were present in the continuous DNA strand.
Immunological Methods-Antisera against hMSH3 and hMSH6 were generated by immunizing rabbits with MAP-conjugated peptides (30). Peptide sequences were TEIDRRKKRPLENDGPVKKK (21) for hMSH3 and MQRADEALNKDKIKRLELAV (10) for hMSH6. Blood obtained from rabbits was clotted overnight at room temperature, the clot and red blood cells were removed by centrifugation, and the resulting serum was heated for 20 min at 56°C to inactivate proteases and nucleases. Serum was stored in aliquots at Ϫ20°C. A monoclonal mouse antibody against hMSH2 (Ab-1) was purchased from Calbiochem. Secondary antibodies conjugated with horseradish peroxidase or alkaline phosphatase were obtained from Sigma.
For Western analysis, samples were separated on a 6% (w/v) denaturing polyacrylamide gel and transferred to polyvinylidene difluoride membrane (Immobilon P, Millipore Corp.), and proteins were detected after incubation with the appropriate primary and secondary antibodies using chemiluminescence reagents. The ECL system (Amersham Pharmacia Biotech) was used in the case of peroxidase-conjugated antibodies. For alkaline phosphatase-conjugated antibodies, the phosphatase substrate CSPD (Boehringer Mannheim) was employed according to the manufacturer's instructions. Chemiluminescence was detected by exposing BioMax film (Eastman Kodak Co.) at room temperature and quantitated using a cooled charge-coupled device imager (Photometrics). The integrated intensities of the bands were compared with a standard curve obtained with samples of purified hMutS␣ and hMutS␤ included in the immunoblots. The correlation of integrated intensity and the amount of standard protein was roughly linear.
For the immunodepletion of nuclear extracts, serum from the rabbits immunized with the hMSH3-specific peptide was bound to protein A-Sepharose (Amersham Pharmacia Biotech) at a ratio of 3 ml of serum/ml of wet gel. The charged gel was equilibrated in buffer A (25 mM Hepes/KOH, pH 7.5, 0.1 mM EDTA, 0.1% (v/v) saturated phenylmethylsulfonyl fluoride (Boehringer Mannheim) in isopropyl alcohol, 1 g/ml leupeptin (Boehringer Mannheim), 0.5 g/ml E-64 (Boehringer Mannheim), 0.1 g/ml aprotinin (Boehringer Mannheim)) containing 100 mM KCl, 1 mM dithiothreitol (DTT; Amersham Pharmacia Biotech) and incubated with nuclear extracts for 1 h at 4°C (50 mg of extract/ml of gel). The depleted extract was recovered by pouring the slurry into a 2-ml disposable column (Bio-Rad) and collecting the effluent. Depleted extracts were used immediately.
Purification of hMutS␣ and hMutS␤-hMutS␣ was purified to greater than 95% purity from nuclear extracts of HeLa cells as described (9). Preparations were free of detectable hMutS␤ as judged by Western analysis for hMSH3.
hMutS␤ was isolated from nuclear extracts of either HCT-15 or HL60-R cells by a procedure similar to that used for hMutS␣. All steps were performed at 4°C. Crude nuclear extract was treated with (NH 4 ) 2 SO 4 (194 g/liter), and the precipitate was removed by centrifugation. The supernatant was again treated with (NH 4 ) 2 SO 4 (226 g/liter). The precipitate, which contained more than 95% of the hMutS␤, was collected by centrifugation; resuspended in buffer A containing 100 mM KCl, 2 mM DTT; dialyzed against the same buffer; quick-frozen in liquid nitrogen; and stored at Ϫ80°C (fraction I). Fraction I (approximately 100 mg of nuclear protein) was thawed on ice by the addition of 10 volumes of buffer A containing 250 mM KCl, 1 mM DTT and loaded at 50 ml/h on to a Q-Sepharose column (10.5 cm ϫ 4.9 cm 2 , Amersham Pharmacia Biotech) equilibrated with buffer A containing 250 mM KCl, 1 mM DTT. The column was washed with 100 ml of buffer A containing 250 mM KCl, 1 mM DTT. This step is useful for separation of hMutS␤ from hMutS␣, since hMutS␣ binds to Q-Sepharose under these conditions while hMutS␤ remains in the flow-through (approximately 400 mM KCl is required to elute hMutS␣ from Q-Sepharose). The effluent of the Q-Sepharose column was applied to a single-stranded DNA-agarose column (2.7 cm ϫ 0.4 cm 2 , Life Technologies, Inc.) equilibrated with buffer A containing 250 mM KCl, 1 mM DTT via a direct connection between the two columns. The single-stranded DNA-agarose column was disconnected from the Q-Sepharose column and washed at 50 ml/h with 50 ml of buffer A containing 300 mM KCl, 1 mM DTT and subsequently with 50 ml of buffer A containing 300 mM KCl, 1 mM DTT, 2.6 mM MgCl 2 . hMutS␤ was eluted from the single-stranded DNA-agarose with 50 ml of buffer A containing 300 mM KCl, 1 mM DTT, 2.6 mM MgCl 2 , 1 mM ATP (fraction II). Fraction II was diluted with buffer A containing 1 mM DTT to a conductivity equivalent to 100 mM KCl and loaded at 75 ml/h on to a Q-Sepharose column (0.35 cm ϫ 0.28 cm 2 ) equilibrated in buffer A containing 100 mM KCl, 1 mM DTT. The column was washed with 4 ml of the starting buffer and then eluted with 0.7 ml of buffer A containing 250 mM KCl, 1 mM DTT. The eluate (0.7 ml) was diluted with 1.1 ml of buffer A containing 1 mM DTT to a conductivity equivalent to 100 mM KCl and loaded onto a 1-ml Mono-Q fast protein liquid chromatography column (HR 5/5, Amersham Pharmacia Biotech). The column was washed with 10 ml of buffer B (25 mM Hepes/ KOH, pH 7.5, 0.1 mM EDTA) containing 100 mM KCl and developed with a 20-ml linear gradient of 100 -370 mM KCl in buffer B at 0.5 ml/min. Fractions containing hMutS␤, which eluted at approximately 220 mM KCl, were supplemented with 0.1% (v/v) saturated phenylmethylsulfonyl fluoride in isopropyl alcohol, 1 g/ml leupeptin, 0.5 g/ml E-64, 0.1 g/ml aprotinin, and 0.5 mg/ml bovine serum albumin (samples for electrophoretic analysis were saved prior to the serum albumin addition). Pooled fractions were dialyzed against buffer A containing 100 mM KCl, 2 mM DTT, 20% (w/v) sucrose and quick-frozen in liquid nitrogen. Small aliquots were stored at Ϫ80°C. Eighteen g of HL60-R cells (wet weight) yielded 30 -50 g of hMutS␤ with a purity of 95%, and the yield from HCT-15 cells was about 5-fold lower. The hMutS␤ preparations obtained in this manner were free of detectable hMutS␣ as judged by Western blotting for MSH6.
hMutS␤ was immunopurified on 0.05-ml protein A-Sepharose columns charged with hMSH3-specific immunoglobulin. Columns equilibrated with buffer A containing 100 mM KCl were incubated with 500 g of nuclear extract in 1 ml of buffer A containing 100 mM KCl for 2 h, and the resin was then washed with 5 ml of the same buffer. Bound protein was specifically eluted by incubating the washed resin with 1 ml of a 0.2 mM solution of the hMSH3-specific peptide in buffer A containing 100 mM KCl for 1 h. The eluted protein was precipitated with trichloroacetic acid and used both for denaturing gel electrophoresis and immunoblots. For each experiment, half of the obtained protein was used for silver-stained denaturing gel electrophoresis. One quarter each was used for separate immunoblots with antibodies against hMSH2 and hMSH3.
Protein content in nuclear extracts and purified fractions was determined using a Bradford assay with bovine serum albumin as a standard (31). The stoichiometry of hMSH2 and hMSH3 in samples of purified hMutS␤ was estimated using a cooled charge-coupled device imager (Photometrics) following denaturing gel electrophoresis and staining with Coomassie Brilliant Blue.

Presence of hMutS␤ in Extracts of Human Cells-MSH6
Ϫ/Ϫ cell lines like HCT-15 and MT1 are selectively defective in the repair of base-base mispairs and single-nucleotide I/D mismatches but are proficient in correction of I/D heterologies of 2, 3, and 4 nucleotides (9). Since hMSH6 is a required subunit of hMutS␣ (9, 10), we reasoned that repair of I/D heterologies larger than 1 nucleotide in these cell lines must depend on a second mismatch recognition activity. To isolate this activity, HCT-15 nuclear extract was resolved by chromatography, and fractions were tested for their ability to restore repair of a dinucleotide I/D heteroduplex to nuclear extract of LoVo colorectal tumor cells, which are devoid of hMSH2 due to partial deletion of the structural gene (8). This cell line is also free of detectable hMSH3 and contains only trace levels of hMSH6 (see below and Ref. 26).
The activity isolated in this manner (see "Experimental Procedures") restored repair of a CA I/D heteroduplex (3.1 fmol/15 min with LoVo nuclear extract and 25 ng of purified protein) but was inactive on a substrate containing a G-T base-base mispair (Ͻ0.3 fmol/15 min in an identical assay). Activity is associated with two polypeptides of approximately 104 and 125 kDa, which were identified as hMSH2 and hMSH3 by Western blot (Fig. 1A). Apparent molecular weights of the two proteins agree well with those predicted from the cDNA sequences (10,21). This activity was also isolated from methotrexate-resistant HL60-R promyelocytic leukemia cells (21), in which hMSH3 levels are highly elevated due to amplification of the DHFR-MSH3 region of chromosome 5 (26). Polypeptide composition and mismatch repair activity of the protein isolated from this cell line are similar to those of the HCT-15 activity, but yields are about 5-fold higher (Fig. 1A). Integration of Coomassiestained species after denaturing electrophoresis indicated that hMSH3 and hMSH2 are present in roughly equal amounts in the HL60-R activity (0.7-0.9 mol of hMSH3/mol of hMSH2), suggesting a 1:1 complex as proposed previously for the hMutS␤ complex produced from baculovirus-expressed subunits (24). The hMSH3:hMSH2 ratio was lower for the protein isolated from HCT-15 cells (0.4 mol of hMSH3/mol of hMSH2), perhaps a consequence of hMSH3 loss or proteolysis during fractionation due to lower levels of protein present in this cell line. A minor impurity of 120 kDa (ϳ5%) was present in some preparations isolated from both cell lines (Fig. 1). The impurity does not cross-react with anti-hMSH2 or anti-hMSH3 antibodies used, but it is possible that it is a carboxyl-terminal fragment of hMSH3, since the anti-MSH3 antibody used is directed against the amino terminus (residues 81-100). A proteolytic origin from hMSH3 is consistent with electrophoretic analysis across the hMutS␤ elution profile from the last Mono-Q column. Early eluting fractions from this column were free of the contaminant and characterized by a hMSH3:hMSH2 molar ratio near unity, whereas the molar ratio was reduced in later eluting fractions containing the contaminant.
Further evidence for presence of the hMutS␤ complex in extracts of human cells was provided by immunopurification using immobilized antibody against hMSH3. As shown in Fig.  1B, hMSH2 co-purifies with hMSH3 from nuclear cell extracts of repair-proficient HeLa cells, from the hMSH6-deficient cell line HCT-15, and from the hMSH3-overproducing cell line HL60-R. However, hMSH3 could not be isolated from nuclear extracts of the hMSH2-deficient LoVo cell line, which harbors mutations in both alleles of MSH2 (8) but is also phenotypically deficient in hMSH3 and hMSH6 (see below and Ref. 26).
hMutS␤ Is More Abundant in HCT-15 than in HeLa Cells-hMSH3 and hMSH6 were quantitated via immunoblotting in nuclear extracts of HeLa, LoVo, and HCT-15 tumor cell lines (Fig. 2). Neither protein was detectable in MSH2 Ϫ/Ϫ LoVo cells (Ͻ0.03 g of MSH6 and Ͻ 0.005 g of MSH3 per mg of nuclear extract, Fig. 2), consistent with an earlier observation with this cell line (26). Since hMSH3 and hMSH6 were only detected in cell lines expressing hMSH2 and since both proteins associate FIG. 1. Subunit composition of hMutS␤. A, samples containing hMutS␤ isolated from HCT-15 (200 ng) and two different preparations from HL60-R (200 ng for preparation 1 and 1 g for preparation 2) were separated on 8% denaturing gels and stained with Coomassie Brilliant Blue. Molecular weight markers (Sigma) are also shown. The lower portion of the panel shows results of Western blotting with antibodies specific for hMSH2 and hMSH3. Panel B, immunopurification of hMutS␤ from nuclear extracts. Nuclear extracts prepared from LoVo, HeLa, HCT-15, and HL60-R cells were incubated with protein A-bound hMSH3-specific antibody; bound material was eluted with a hMSH3specific peptide; and samples were analyzed on 5% denaturing gels, which were silver-stained (see "Experimental Procedures"). Purified hMutS␤ (40 ng) was included as a standard. The lower portion of the panel shows the results of Western blots with antibodies specific for hMSH2 and hMSH3. tightly with hMSH2 during fractionation ( Fig. 1 and (9, 24, 26), immunological quantitation of hMSH3 and hMSH6 can be used to estimate levels of hMutS␣ and hMutS␤ in nuclear extracts. In HeLa cells, the hMSH3 level is equivalent to about 0.15 g of hMutS␤/mg of nuclear extract protein, while the hMSH6 level is equivalent to about 1.5 g of hMutS␣/mg. The hMutS␣: hMutS␤ ratio of 10:1 found here for HeLa cells is similar to the value of 6:1 estimated for HL-60 cells based on chromatographic resolution of the two heterodimers (26). In HCT-15 cells, the hMSH3 level was equivalent to about 0.5 g of hMutS␤/mg of nuclear extract, but hMSH6 was undetectable. This confirms the absence of hMutS␣ in this cell line.
hMutS␤ Is Responsible for I/D Mismatch Repair in HCT-15 Extracts-Nuclear extracts of HCT-15 cells were immunodepleted by adsorption to an anti-MSH3 support (see "Experimental Procedures"). As shown in Fig. 3, this resulted in loss of repair activity on a CA I/D heteroduplex, but activity on the dinucleotide I/D heteroduplex substrate was restored upon the addition of purified hMutS␤. By contrast, a G-T heteroduplex was not repaired by either the depleted or the mock-depleted extract, even upon the addition of hMutS␤. As expected, the addition of hMutS␣ to immunodepleted extracts restored repair of both the base-base and the I/D heteroduplexes. The addition of hMutS␣ also restored G-T heteroduplex repair activity to the mock-depleted extract, but repair of the CA I/D substrate was not significantly increased above endogenous levels. hMutS␣ therefore has a unique role in G-T heteroduplex repair, but hMutS␣ and hMutS␤ are redundant with respect to correction of the dinucleotide I/D substrate. The finding that immunodepletion of HeLa nuclear extracts with the hMSH3 antiserum did not affect repair of either heteroduplex (data not shown) is also consistent with this view.
Mismatch Specificities of hMutS␣ and hMutS␤ in Strandspecific Repair-Inasmuch as MSH2 Ϫ/Ϫ LoVo cells are also phenotypically deficient in hMSH3 and hMSH6 (see Fig. 2 and Ref. 26), extracts of this cell line can be used to establish the mismatch repair specificities of hMutS␣ and hMutS␤, since the potential for subunit exchange is precluded. As noted above (see "Experimental Procedures"), the hMutS␣ and hMutS␤ preparations used in this study are free of detectable crosscontamination. As shown in Fig. 4, hMutS␣ restored near normal levels of repair for each of the eight base-base mismatches to LoVo extract, but comparable amounts of hMutS␤ did not detectably increase repair above base line in any case. The amounts of hMutS␣ and hMutS␤ used in these experiments was 0.8 g/mg LoVo extract, corresponding to about 50 and 500% of the levels of hMutS␣ and hMutS␤ in HeLa nuclear extract, respectively (see above).
Purified hMutS␣ and hMutS␤ were also compared with re-spect to their ability to restore repair of I/D heteroduplexes to LoVo nuclear extract. As shown in Fig. 5, heterologies of 1, 2, or 3 nucleotides were not processed to a significant degree by LoVo nuclear extract, although repair of I/D mismatches of 5-27 nucleotides was observed, increasing with the size of the unpaired region. Since correction of I/D mismatches of 8 and 16 nucleotides has also been observed in extracts of MLH1 Ϫ/Ϫ HCT-116 cells (32), we attribute this activity to a pathway distinct from the mismatch repair system that is dependent on hMutS␣, hMutS␤, and hMutL␣. In the experiment shown in the upper panel of Fig. 5, LoVo extract was supplemented with equivalent amounts of hMutS␣ or hMutS␤ (0.4 g/mg of extract, corresponding to about 25 and 250% of hMutS␣ and hMutS␤ concentrations in HeLa extract, respectively). Since the rate of repair under these conditions is limited by the MutS homolog concentration, the values shown provide a direct comparison of the specific activities of the two heterodimers in I/D heteroduplex repair. As can be seen, hMutS␣ restored repair on the single-nucleotide I/D mispair, but both proteins complemented LoVo extract to increase repair on I/D mismatches containing 2-8 nucleotides. hMutS␣ supported repair of all I/D mismatches in this class, but hMutS␤ was somewhat more active in promoting correction of the 3-, 5-, and 8-nucleotide heterologies tested. Furthermore, the repair spectrum of hMutS␤-complemented LoVo extract was virtually identical to that observed with extract prepared from MSH6 Ϫ/Ϫ HCT-15 cells (compare Fig. 5 and Table I).
While experiments with equal amounts of hMutS␣ and hMutS␤ are useful for determination of relative activities of the proteins on particular substrates, these conditions do not accurately reflect the situation in the repair proficient cell lines, which, as noted above, contain considerably more hMutS␣ than hMutS␤. Supplementation of LoVo nuclear extract with disparate amounts of hMutS␣ and hMutS␤ (2 and 0.4 g/mg of extract, corresponding to about 125 and 250% of hMutS␣ and hMutS␤ levels in HeLa extract, respectively) resulted in significant increases in the level of hMutS␣-promoted I/D mismatch relative to that observed with hMutS␤. As shown in the lower panel of Fig. 5, such increases were evident in all cases except for the 5-nucleotide heterology.
Although hMutS␤ did not support significant repair of the single-nucleotide I/D mismatch under standard assay conditions, low level hMutS␤-directed repair of the single-nucleotide heterology was detected in HCT-15 nuclear extract and in hMutS␤-supplemented LoVo extract at reduced ionic strength (reduction of KCl concentration from 110 to 70 mM). 2 However, even under reduced salt conditions, the rate of hMutS␤-supported correction of the dinucleotide I/D heteroduplex was 4 -10 times that observed with the single-nucleotide I/D mismatch, as judged by competition experiments in which both substrates were present in the same reaction. hMutS␤ therefore displays low but detectable activity on single-nucleotide I/D mispairs.
Like hMutS␣, hMutS␤ Supports Bidirectional Mismatch Repair-Human strand-specific mismatch repair can be directed by a single-strand break located either 5Ј or 3Ј to the mispair on the incised strand (29). The experiments summarized in Figs. 4 and 5 used 5Ј-heteroduplexes. In similar experiments, supplementation of LoVo extract with either hMutS␣ or hMutS␤ restored repair of a 3Ј-CA I/D heteroduplex (3.9 fmol/15 min for hMutS␣ and 4.2 fmol/15 min for hMutS␤). However, repair of a 3Ј G-T heteroduplex was restored only upon the addition of hMutS␣ (4.9 fmol/15 min for hMutS␣ and Ͻ0.3 fmol/15 min for hMutS␤). Like hMutS␣ (9), hMutS␤ therefore supports bidirec-2 S. J. Littman and J. Genschel, unpublished results. FIG. 2. hMSH3 and hMSH6 levels in nuclear extracts. Samples (50 g) of nuclear extract from HeLa, LoVo, and HCT-15 cells were separated by denaturing gel electrophoresis. 100, 50, 25, and 10 ng of hMutS␣ were included in the gel used to quantitate hMSH6 (upper panel), and the same amounts of hMutS␤ were included in a separate gel to quantitate hMSH3 (lower panel). Amounts of hMSH3 and hMSH6 in the nuclear extracts were quantitated after Western blotting with antibodies specific for each protein by comparison of the immunological response to the included standards. tional repair, and its mismatch specificity is evidently orientation-independent. DISCUSSION Two phenotypes have been described for mismatch repairdeficient human cell lines with mutations in genes that encode MutS homologs. MSH2 mutations confer hypermutability at selectable loci and destabilize simple repeats such as (A) n and (CA) n (5,14). MSH6 Ϫ/Ϫ mutants, on the other hand, show an increased HPRT mutation rate and instability of single-nucleotide repeat sequences, but mutations in dinucleotide repeats are rare (13)(14)(15)17). hMSH3 deficiency has been reported in hematological malignancies (33) and sporadic cancers (34,44), but the effect of hMSH3 deficiency on genetic stability has not been carefully evaluated.
The differential nature of the mutation spectra of MSH2 Ϫ/Ϫ and MSH6 Ϫ/Ϫ human cells can be understood in terms of the repair specificities of hMutS␣ and hMutS␤ that we describe here. We have shown that hMutS␣ supports repair of all eight base-base mismatches, as well as each I/D mispair tested ranging from 1 to 8 unpaired nucleotides. By contrast, hMutS␤ is inactive in base-base mismatch repair and is weakly active on the single-nucleotide I/D mismatch tested but supports efficient repair of I/D mismatches of 2 to ϳ8 unpaired nucleotides. It is noteworthy that hMutS␣ supported the repair of each I/D mismatch that was also corrected in a hMutS␤-dependent reaction. This indicates extensive overlap in I/D mismatch specificity for the two activities, although we cannot exclude the possibility that hMutS␣ and hMutS␤ may differentially respond to other I/D mispairs depending on sequence context or the nature of the unpaired nucleotides (24).
hMutS␣ and hMutS␤ repair spectra are also in general accord with chromosome transfer experiments in which chromosome 5 bearing MSH3 ϩ or chromosome 2 bearing MSH2 ϩ and MSH6 ϩ genes was introduced into the MSH3 Ϫ/Ϫ MSH6 Ϫ/Ϫ HUAA double mutant cell line. Mono-, di-, tri-, and tetranucleotide repeats are unstable in HUAA cells (34). Introduction of MSH3 on chromosome 5 stabilized a tetranucleotide repeat and dinucleotide repeats, but failed to stabilize a d(A) n repeat (34). HUAA extracts are defective in repair of base-base and I/D mispairs, but extracts prepared from HUAA cells containing wild type MSH2 and MSH6 genes repair base-base mismatches as well as 1-, 2-, and 4-nucleotide I/D mismatches (35). Extracts derived from HUAA cells harboring a wild type copy of MSH3 were inactive on base-base mismatches but were found to repair mononucleotide and tetranucleotide I/D mispairs, provided that the unpaired nucleotides were in the incised DNA strand (34). These findings differ somewhat from our results with hMutS␤. We have found that a 3Ј-CA I/D heteroduplex containing the unpaired dinucleotide within the continuous strand is a good substrate for repair mediated by hMutS␤ (above) and have previously shown that di-, tri-, and tetranucleotide I/D heteroduplexes are good substrates for nickdirected correction in MSH6-deficient HCT-15 and MT1 nuclear extracts without regard to location of the unpaired nucleotides in the incised or continuous strand (9). The latter experiments also demonstrated that an unpaired d(T) in the nicked heteroduplex strand is a poor substrate for hMutS␤ in HCT-15 and MT1 cell extracts. It is possible that some of these discrepancies are due to differences in assay conditions or to the effect of sequence context, since we have observed weak but significant hMutS␤-directed repair of an A I/D mismatch under conditions of reduced ionic strength. Furthermore, while mononucleotide repeats are highly unstable in hMSH6-deficient human cells (13,14,17), the degree of destabilization of such sequences is not as great as that observed with MLH1 Ϫ/Ϫ cells (13,15), suggesting that some processing of mononucleotide FIG. 3. Immunodepletion of nuclear extracts from HCT-15 with a hMSH3specific antibody. Nuclear extracts from HCT-15 were treated using either serum containing hMSH3-specific antibodies (depleted) or preimmune serum (mock) and assayed for mismatch repair as described under "Experimental Procedures." Assays were supplemented with 100 ng of hMutS␤ or hMutS␣ as indicated. Heteroduplex substrates contained a G-T mispair and a nick 5Ј to the lesion (G-T) or a CA I/D mismatch and a nick 3Ј to the lesion ( / CA { ).
FIG. 4. hMutS␣ but not hMutS␤ restores repair of base-base mismatches to LoVo nuclear extract. MSH2 Ϫ/Ϫ LoVo nuclear extract was assayed for repair of base-base mismatches as described under "Experimental Procedures" in the absence (solid bars) or in the presence of 40 ng of either hMutS␣ (cross-hatched bars) or hMutS␤ (gray bars). The heteroduplex substrates used in these experiments contained a single-strand break 5Ј to the mispair. mismatches by hMutS␤ does occur. Interestingly, mononucleotide repeats appear to be relatively stable in MSH6 Ϫ/Ϫ murine cells (35).
While the mismatch repair specificities of hMutS␣ and hMutS␤ described here are consistent with mutation spectra and in vitro assay of hMSH6-deficient and hMSH3-deficient cells, they differ significantly from hMutS␣ specificity deduced from gel shift assay using a heterodimer produced by in vitro transcription/translation (25). These experiments failed to detect interaction of hMutS␣ with I/D heteroduplexes of 2 or 10 unpaired nucleotides, whereas hMutS␤ produced in a similar manner was found to bind to both. Since other studies have shown that hMutS␣ efficiently binds I/D heterologies with one, two, and three unpaired nucleotides (9,24), it is possible that hMutS␣ produced by in vitro transcription/translation is not fully active. In addition, since other activities are involved in mismatch rectification, mismatch binding may not provide a completely accurate indicator of the specificities of hMutS␣ and hMutS␤ in the overall repair reaction. Repair, on the other hand, is expected to reflect the influence of these other activities.
Our findings and the chromosome transfer experiments discussed above also suggest that significant differences exist between the specificities of hMutS␣ and hMutS␤ and their Saccharomyces cerevisiae counterparts. As in the case of hMutS␣, genetic evidence and gel shift data support a role for yeast MSH2⅐MSH6 in the repair of base-base mispairs and single-nucleotide and dinucleotide I/D mismatches (36 -38). Yeast MSH2⅐MSH3 is also able to support the repair of singlenucleotide and dinucleotide I/D mismatches, but only this complex appears to be involved in the repair of larger I/D heterologies (39 -41).
Heterodimer formation between hMSH2 and hMSH6 or between hMSH2 and hMSH3 results in two activities with distinct, partially overlapping specificities that are highly active in mismatch binding and repair (9,10,24,25). What determines the mispair binding of these two complexes? An intriguing possibility is that both subunits of the heterodimer contribute elements to the mismatch binding site. In this model, the mismatch binding site created in the hMSH2⅐hMSH6 complex would be able to accommodate both I/D and base-base heterologies, whereas only I/D heterologies would fit well into the binding site created by hMSH2⅐hMSH3 heterodimerization. Alternatively, the mismatch binding site may reside within one subunit, with the other subunit serving to activate this binding center.
This and other studies suggest that hMutS␣ and hMutS␤ participate in a common mismatch repair pathway. We previously demonstrated that hMutS␣ is present in a 6-fold molar excess over hMutS␤ in exponentially growing HL-60 cells (26) and have shown here that in mitotically active HeLa cells, 90% of the nuclear hMSH2 is present in the hMutS␣ heterodimer. The partitioning of hMSH2 between these two complexes appears to be important for genetic stabilization, since overproduction of hMSH3, which increases the hMutS␤ pool at the expense of hMutS␣, is associated with a large increase in mutation rate (26). As shown here, the hMutS␤ pool is also elevated in HCT-15 cells, which do not produce hMutS␣ due to genetic inactivation of MSH6. Several other lines of evidence indicate regulation of hMutS␣ and hMutS␤ pools in human cells. MSH2 Ϫ/Ϫ LoVo cells, which lack detectable levels of the hMSH2 polypeptide, are phenotypically deficient in hMSH3 and hMSH6 proteins ( Fig. 2 and Ref. 26). The distribution of the common hMSH2 subunit between hMutS␣ and hMutS␤ may be a consequence of mass action, with hMSH2 partitioning between hMSH3 and hMSH6 according to pool size and heterodimer formation stabilizing the hMSH3 and hMSH6 subunits. However, other forms of regulation have not been ruled out.
The relative abundance of hMutS␣ and its broad mismatch specificity suggest that the hMSH2⅐hMSH6 heterodimer is the primary mismatch recognition activity for correction of DNA biosynthetic errors. While hMutS␤ may complement the specificity of hMutS␣ for I/D heterologies, depending on sequence context of the mismatch and the nature of the unpaired nucleotides, it is also possible that the former activity may have distinct functions in DNA metabolism. For example, Haber and a Mismatch repair assays were performed as described under "Experimental Procedures." b Heteroduplex substrates used in these experiments contained a single-strand break 5Ј to the mismatch. The sequence of the unpaired region in each substrate is given under "Experimental Procedures." colleagues (42,43) have shown that S. cerevisiae MSH2 and MSH3, but not MSH6, are involved in the processing of nonhomologous ends during recombinational double-strand break repair.