No abstract

DNA mismatch recognition and binding in human cells has been thought to be mediated by the hMSH2 protein. Here it is shown that the mismatch-binding factor consists of two distinct proteins, the 100-kilodalton hMSH2 and a 160-kilodalton polypeptide, GTBP (for G/T binding protein). Sequence analysis identified GTBP as a new member of the MutS homolog family. Both proteins are required for mismatch-specific binding, a result consistent with the finding that tumor-derived cell lines devoid of either protein are also devoid of mismatch-binding activity.

In human cells, mismatch recognition is mediated by a heterodimeric complex, hMutSalpha, comprised of two members of the MutS homolog (MSH) family of proteins, hMSH2 and GTBP [1,2]. Correspondingly, tumour-derived cell lines defective in hMSH2 and GTBP have a mutator phenotype [3,4], and extracts prepared from these cells lack mismatch-binding activity [1]. However, although hMSH2 mutant cell lines showed considerable microsatellite instability in tracts of mononucleotide and dinucleotide repeats [4,5], only mononucleotide repeats were somewhat unstable in GTBP mutants [4,6]. These findings, together with data showing that extracts of cells lacking GTBP are partially proficient in the repair of two-nucleotide loops [2], suggested that loop repair can be GTBP-independent. We show here that hMSH2 can also heterodimerize with a third human MSH family member, hMSH3, and that this complex, hMutSbeta, binds loops of one to four extrahelical bases. Our data further suggest that hMSH3 and GTBP are redundant in loop repair, and help explain why only mutations in hMSH2, and not in GTBP or hMSH3, segregate with hereditary non-polyposis colorectal cancer (HNPCC) [7].

In extracts of human cells, base-base mismatches and small insertion/deletion loops are bound primarily by hMutSalpha, a heterodimer of hMSH2 and hMSH6 (also known as GTBP or p160). Recombinant hMutSalpha bound a G/T mismatch-containing oligonucleotide with an apparent dissociation constant Kd = 2.6 nM, while its affinity for a homoduplex substrate was >20-fold lower. In the presence of ATP, hMutSalpha dissociated from mismatched oligonucleotide substrates, and this reaction was attenuated by mutating the conserved lysine in the ATP-binding domains of hMSH6, hMSH2 or both to arginine. Surprisingly, this reaction required only ATP binding, not hydrolysis. The ATPase activity of hMutSalpha variants carrying the Lys-->Arg mutation in hMSH2 or in hMSH6 was severely affected, but these mutants were still proficient in mismatch binding and were able to complement, albeit to different extents, mismatch repair-deficient cell extracts. The mismatch binding-proficient, ATPase-deficient double mutant was inactive in the complementation assay and its presence in repair-proficient extracts was inhibitory. We conclude that although the ATPase activity of hMutSalpha is dispensible for mismatch binding, it is required for mismatch correction.

We tested the ability of recombinant hMutS␣ (hMSH2͞hMSH6) and hMutS␤ (hMSH2͞hMSH3) heterodimers to complement the mismatch repair defect of HEC59, a human cancer cell line whose extracts lack all three MutS homologues. Although repair of both base͞base mispairs and insertion-deletion loops was restored by hMutS␣, only the latter substrates were addressed in extracts supplemented with hMutS␤. hMutS␣ was also able to complement a defect in the repair of base͞base mispairs in CHO R and HL60R cell extracts. In these cells, methotrexate-induced amplification of the dihydrofolate reductase (DHFR) locus, which also contains the MSH3 gene, led to an overexpression of MSH3 and thus to a dramatic change in the relative levels of MutS␣ and MutS␤. As a rule, MSH2 is primarily complexed with MSH6. MutS␣ is thus relatively abundant in mammalian cell extracts, whereas MutS␤ levels are generally low. In contrast, in cells that overexpress MSH3, the available MSH2 protein is sequestered predominantly into MutS␤. This leads to degradation of the partnerless MSH6 and depletion of MutS␣. CHO R and HL60R cells therefore lack correction of base͞base mispairs, whereas loop repair is maintained by MutS␤. Consequently, frameshift mutations in CHO R are rare, whereas transitions and transversions are acquired at a rate two orders of magnitude above background. Our data thus support and extend the findings of Drummond et al.