Functional Analysis of Human MLH1 Variants Using Yeast and In vitro Mismatch Repair Assays

The functional characterization of nonsynonymous single nucleotide polymorphisms in human mismatch repair (MMR) genes has been critical to evaluate their pathogenicity for hereditary nonpolyposis colorectal cancer. We previously established an assay for detecting loss-of-function mutations in the MLH1 gene using a dominant mutator effect of human MLH1 expressed in Saccharomyces cerevisiae. The purpose of this study is to extend the functional analyses of nonsynonymous single nucleotide polymorphisms in the MLH1 gene both in quality and in quantity, and integrate the results to evaluate the variants for pathogenic significance. The 101 MLH1 variants, which covered most of the reported MLH1 nonsynonymous single nucleotide polymorphisms and consisted of one 3-bp deletion, 1 nonsense and 99 missense variants, were examined for the dominant mutator effect by three yeast assays and for the ability of the variant to repair a heteroduplex DNA with mismatch bases by in vitro MMR assay. There was diversity in the dominant mutator effects and the in vitro MMR activities among the variants. The majority of functionally inactive variants were located around the putative ATP-binding pocket of the NH₂-terminal domain or the whole region of the COOH-terminal domain. Integrated functional evaluations contribute to a better prediction of the cancer risk in individuals or families carrying MLH1 variants and provide insights into the function-structure relationships in MLH1. [Cancer Res 2007;67(10):4595–604]

Hereditary nonpolyposis colorectal cancer (HNPCC) is one of the most common familial cancer syndromes caused by mainly germ-line mutations in DNA mismatch repair (MMR) genes, such as MLH1, MSH2, MSH6, and PMS2. MMR contributes to genome integrity by correcting replication errors, particularly mismatch base pairs or slippages in simple repeat sequences. The MMR system is well conserved from Escherichia coli to mammals, and the E. coli MMR system, where MutS, MutL, and MutH complexes function, has been well analyzed. In mammalian cells, heterodimers of MutS homologues (MSH2-MSH6 and MSH2-MSH3) recognize replication errors, and the heterodimer of the MutL homologue (MLH1-PMS2) interacts with MutS homologues and recruits further repair proteins.

In the two MMR genes, MLH1 and MSH2, which account for the majority of HNPCC kindred mutations, 31.6% of the MLH1 mutations and 19.4% of the MSH2 mutations are missense (nonsynonymous single nucleotide polymorphisms) according to the InSiGHT database.⁴

Pathogenic significances of nonsynonymous single nucleotide polymorphisms are not easily evaluated without a functional assay. Generally, this makes genetic diagnosis more difficult, especially when phenotype-genotype segregation analysis is limited because of an ethical issue, the small number of family members, or other reasons. Therefore, functional analysis has been needed to interpret the pathogenicity of MLH1 variants in genetic diagnoses of HNPCC.

Several groups, including ours, have attempted to resolve this issue by analyzing the functional significance of MLH1 variants by various methods (1–8). We found a dominant mutator effect (DME) of wild-type MLH1 on interference in the yeast MMR system and evaluated the pathogenicity of 20 MLH1 missense variants using this effect for a yeast-based functional assay (1). One of the other strategies was focusing on the well-defined functions of MLH1, ATPase activity in the NH₂ terminus, and the binding abilities with PMS2 in the COOH terminus (2, 4, 7). Another strategy was based on the assessment of total MMR activity such as in vitro MMR assay, monitoring the MMR rate of cell extracts for heteroduplex DNA-containing mismatch bases (4, 5), or yeast system measuring the replication error rate resulting from the expressed yeast-human hybrid proteins or equivalent yeast variants (3). An alternative approach was to analyze the effects of an MLH1 variant for the expression of MLH1 mRNA and protein and the proliferation rate of cells when an MLH1 variant was introduced into the cells (6). Recently, MLH1 variants were characterized for the multiple functional properties of wild-type MLH1, including protein expression, subcellular localization, MLH1-PMS2 interaction, and MMR efficiency (8). However, it still seems to be important to establish the database on the functional effects of as many variants as possible by various methods and integrate the accumulated data for better understanding of MLH1 variants.

Beyond functional analyses, crystal structure analyses of E. coli MutL were used to predict the structural alterations of MLH1 variants (9, 10). Besides the NH₂ terminus, the crystal structure of the COOH terminus of E. coli MutL was identified recently (11). Consideration of the protein structure permits more comprehensive analysis of MLH1 and is thought to provide useful information for interpreting the pathogenesis of MLH1 variants in genetic diagnoses.

In this study, we examined 101 MLH1 variants in yeast assays and an in vitro MMR assay, to compare these two assays and to develop the functional database for a large number of variations. We also analyzed the distribution of loss-of-function variants on the secondary MLH1 structure or the crystal structure, to predict the pathogenicity of the MLH1 variants, and to get the structural basis for defects of MLH1 function.

Yeast strains and plasmids. The Saccharomyces cerevisiae haploid strains used in this study were as follows: YPH499 from Stratagene, and yCA23 (leu2, his3, trp1, can1, pCYC1-28C::ADE2) and yCA25 [leu2, his3, trp1, can1, pCYC1-(CA)₁₅::ADE2]. yCA23 and yCA25 were made by integrating plasmids pCA2 and pCA3 into the URA3 locus of a haploid derivative of ASZ3 (12). pCA2 and pCA3 were derived from pLS210 (13) and contain out of frame C₂₈ and (CA)₁₅ repeats at the amino terminus of ADE2 expressed by the CYC1 promoter. Constructs of the low-copy MLH1 expression plasmid, pCI-ML10, the gap repair vectors, pCI-ML22 and pCI-ML24, and the reporter LacZ plasmid, pCIZ1, were described previously (1). The mutation-reporter green fluorescent protein (GFP) plasmid, pCGFP2μ(URA3), was constructed by inserting an enhanced GFP (EGFP) coding sequence with a poly C (C₂₂) tract instead of the β-galactosidase gene, including the poly-GT tract of pCIZ1. The wild-type MLH1 human expression plasmid was constructed by inserting wild-type MLH1 cDNA into the BamHI site of a pCMV-Neo-Bam (14), and each variant MLH1 expression plasmid is identical to the wild-type MLH1 expression plasmid but contains each mutagenized MLH1 cDNA.

Site-directed mutagenesis. One hundred one MLH1 variants consisted of 99 missense variants, one nonsense variant (W714X), and one 3-bp in-frame deletion (K618del), and were constructed by site-directed mutagenesis (Table 1; ref. 1). The information on the majority of MLH1 variants were based on the InSiGHT and SWISS-PROT⁵

databases, whereas others were based on various publications (15–30). Each of the MLH1 sequences mutated specifically was confirmed by DNA sequencing. These MLH1 variants included 90 germ-line variants found in HNPCC patients fulfilling the Amsterdam criteria (AC+) or individuals not fulfilling the Amsterdam criteria (AC−), three somatic variants found in individuals with sporadic cancer and eight major or putative polymorphisms.

Yeast-based assays. For the LacZ and GFP assays, competent yeast cells YPH499 were transformed with an MLH1 expression plasmid and a reporter plasmid [pCIZ1 for LacZ assay, pCGFP2μ(URA3) for GFP assay], as described previously (1). In the LacZ assay, the MMR ability of each yeast transformant could be monitored by the appearance of a blue color on the Xgal plate after 5 days incubation at 37°C (1). In the GFP assay, the MMR ability could be monitored by the appearance of green-fluorescent colonies observed under fluorescent microscopy after 3 days incubation at 30°C. In the ADE2 assay, yCA23 or yCA25 cells were transformed with an MLH1 expression plasmid, and the MMR ability could be monitored by the appearance of a red color on a low-adenine plate after 2 or 3 days incubation at 30°C (31).

In vitro MMR assay and Western blot analysis. Human colon cancer HCT116 cells were cultured in DMEM with 10% fetal bovine serum at 37°C. HCT116 cells (3.0 × 10⁶) were transfected with 5 μg of each MLH1 expression plasmid, 5 μg of a PMS2 expression plasmid, and 1.2 μg of pEGFP-C3 (BD Biosciences) using FuGENE6 (Roche). After 24-h incubation, cytoplasmic extracts were prepared and MMR activity in human cell extracts containing heteroduplex DNA was measured in vitro as described previously (32). M13mp2 heteroduplex DNA with the T-G mismatch was used in a MMR reaction with HCT116 cell extracts transfected with each MLH1 expression plasmid. Each experiment was done twice or more and in vitro MMR activity (%) is shown as the mean ± SD. The extract was also examined in Western blot analyses using an anti-MLH1 antibody (G168-728, BD Biosciences), an anti-PMS2 antibody (A16-4, BD Biosciences), and an anti-GFP antibody (MMS-118P, Covance). Bands of MLH1 and GFP were quantified by densitometry.

Prediction of protein activity and structure. The Sorting Intolerant from Tolerant (SIFT) algorithm was adopted to predict the effects of amino acid substitutions on protein functions (33). Based on the crystal structure of the E. coli NH₂ terminus of MutL (9, 10), the MLH1 structure was predicted by a homology modeling method (34) using software ICM-Molsoft (Molsoft L.L.C.). The secondary structure alignment of the MutL homologues was based on reports (11, 35) and predicted using the PSIPRED server (36).

DME was detected in the three yeast assays. We previously developed a yeast-based MMR assay using a LacZ reporter plasmid with dinucleotide (GT) repeats (Fig. 1A,, left; ref. 1). To establish simpler methods to detect yeast MMR deficiency, we also developed three distinct assays: a GFP assay using mononucleotide (C) repeats (Fig. 1A, center) and two ADE2 assays using mononucleotide (C) repeats and dinucleotide (CA) repeats (Fig. 1A, right). In each assay, replication errors of mononucleotide or dinucleotide repeats were observed in yeast colonies transformed with wild-type MLH1 due to the DME and expressed reporter genes, whereas those were not observed in yeast colonies transformed with an empty vector (Fig. 1B). The results of the two ADE2 assays coincided in most of the MLH1 variants and, therefore, the result of the ADE2 assay using mononucleotide (C) repeats was adopted as the representative ADE2 assay. The 101 MLH1 variants were classified into one of the following four categories depending on the results of the LacZ, GFP, and ADE2 assays: (a) DME positive in all three assays (DME3+) contained 43 variants, (b) DME positive in two assays (DME2+) contained 16 variants, (c) DME positive in one assay (DME1+) contained 10 variants, and (d) DME negative in all three assays (DME−) contained 32 variants. The results of the DME phenotypes of all variants are summarized in Table 1.

In vitro MMR activity of human cell extracts expressing MLH1 variants. Although the dominant mutator assays are simple and quick methods to detect functional alterations of MLH1 variants, DME deficiency may not be directly linked to pathogenicity on HNPCC onset. To measure the MMR activities of MLH1 variants, we adopted a well-established in vitro MMR assay to analyze the ability of cell extracts to repair DNA substrates with mismatched bases (4, 5). In the assay, cytoplasmic extracts of HCT116 cells had no MMR activity (0%) in the in vitro MMR assay because the cells were defective in MLH1 and PMS2 expressions. The transient expression of wild-type MLH1 and PMS2 in the cells complemented the defect and the MMR activity (79.7 ± 7.8%) was comparable with that of MMR-proficient HCT116+Ch3 cells (71.3 ± 11.7%). The results indicated that exogenous MLH1 was able to restore the MMR function of MLH1-deficient cells as also described in other reports (4, 5). The MMR activities of the 101 MLH1 variants expressed in HCT116 cells were then evaluated. The results indicated that there was diversity (0–90.2%) in the ability to repair mismatch DNA among the MLH1 variants (Fig. 2A). All of the eight putative polymorphisms had >60% of in vitro MMR activity. Therefore, 51 variants that showed 60% or more of MMR activity were classified into “MMR+,” and 50 showing <60% of MMR activity were classified into “MMR−,” in this study.

Protein expression levels among the majority of MLH1 variants. In many previous studies on MLH1 protein, nonsynonymous single nucleotide polymorphisms have been shown to affect protein stability (5, 8, 30, 37). To evaluate the level of expressed MLH1 protein, the 101 variants were transiently coexpressed with PMS2 in HCT116 colon cancer cells. The two MMR proteins were detected by Western blot analyses and the transfection efficiency was normalized by the cotransfected GFP level. The representative results of 11 variants for the MLH1 protein levels are shown in Fig. 2B. Among these, the protein levels of R100P, G67R, T117M, K618del, and R659P variants were lower than those of the wild-type. The level of coexpressed PMS2 was also reduced in the four variants except for R100P. In S193P and E23D, PMS2 expressions were reduced although MLH1 expressions were comparable with wild-type MLH1. Comparing in vitro MMR activities with MLH1 expression levels, some variants, such as R100P, G67R, T117M, K618del, and R659P, showed MMR− with low MLH1 and/or PMS2 expression levels, whereas other variants, such as E102D, showed MMR− despite the retained expression levels of the two proteins (Fig. 2B). In all, only a slight correlation (R = 0.225, P = 0.024) between in vitro MMR activities and MLH1 expression levels was observed (Fig. 2C), suggesting that the MMR activities of MLH1 variants may not solely depend on MLH1 expression levels.

Correlations between DME phenotype and in vitro MMR activities. To examine whether the functions of the DME of MLH1 variants correlate with the ability to repair mismatched DNA, we compared the results of yeast assays with the in vitro MMR assay. The in vitro MMR activities were higher in the DME-positive variants than in the DME-negative variants in any yeast assay (P < 0.01 by two-sided Mann-Whitney's U test; Fig. 3A). Moreover, in the three yeast assays, more DME-positive results correlated with higher in vitro MMR activity (Fig. 3A). The scatter diagram in which 101 variations are plotted by in vitro MMR activities and DME showed also correlation between these two assays (Fig. 3B). The majority of MMR+ variants showed DME3+ (35 of 51) and MMR− variants showed DME−, DME1+, or DME2+ (42 of 50; Table 1; Fig. 3B).

Relationship between the function and structure of MLH1. To elucidate the function-structure relationship of MLH1 variants, we first compared the in vitro MMR activities with the SIFT score of the MLH1 variants. SIFT is a program that predicts the effect of amino acid substitutions on protein function, based on sequence conservation during evolution and the nature of amino acids substituted in a gene of interest (33). As shown in Table 1, a majority of variants sorted as tolerant by SIFT showed MMR+ (38 of 48;79%) and those sorted as intolerant showed MMR− (38 of 51;75%), suggesting that SIFT had a good accuracy for predicting functions of variants and there was correlation between amino acids conservations and functions. Second, we focused on the relationships between MLH1 functions and the functional domain. As shown in Table 1, a large number of MLH1 variants showing DME negative in each yeast assay and/or MMR− were located within either of the two functional domains, the NH₂-terminal ATPase domain or the COOH-terminal PMS2-interactive domain, whereas those showing DME+ or MMR+ were distributed throughout the whole gene product. Finally, we constructed a predicted three-dimensional structural model of the MLH1 NH₂-terminal domain by a homology modeling method to map variants in the predicted structure (Fig. 4A–C). Most of the variants predicted to be around the ATP-binding pocket were DME negative in at least one of the yeast assays or MMR− (Fig. 4B). In particular, variants mutated in residues thought to be critical for ATP binding, such as M35R, N38D, S44F, G67R, I68N, C77R, F80V, T82I, K84E, R100P, and I107R, affected both DME and in vitro MMR activity (Fig. 4C). Unlike the NH₂-terminal ATPase domain, sequence identity among the COOH-terminal domains of MutL homologues is not high enough to construct three-dimensional structure by homology modeling; however, the detailed analysis of secondary structure predictions and sequence alignments indicates that the structures of the COOH-terminal domains of MutL homologues are similar (Fig. 4D; refs. 11, 35). Missense variants were classified by the assays and mapped on a COOH-terminal domain alignment of four MutL homologues, showing that variants in the COOH-terminal domain were found more frequently in the internal subdomain, especially within and around the αC helix, than in the external subdomain and functionally inactive variants were distributed through the whole region of the COOH terminus (Fig. 4D).

Relations between MLH1 functions and clinical features. To investigate the relationship between the functional phenotype in the assays and clinical phenotypes previously reported in the databases or papers, at first, in vitro MMR activities were compared between AC+ variants and AC− variants, considering the total number of families reported for each variant. The AC+ group showed significantly lower in vitro MMR activities than AC− group (P < 0.01; Fig. 5A). Second, the MLH1 protein expression levels in HCT116 cells were compared between AC+ and AC− groups. The significant difference of the protein level was not shown between the two groups (P = 0.68; Fig. 5B). These results suggested that there is correlation between MMR defects and a strong family history but no correlation between the MLH1 protein instability and a strong family history.

In this study, we evaluated the functional significance of 101 MLH1 variants by yeast-based assay and in vitro assay to provide useful information for understanding of pathogenicity. This functional characterization of a large number of variants allowed us to compare the property of two assays and to analyze the structural basis of functional deficiency.

We showed that the yeast assay distinguished the majority of variations that retained or lost the in vitro MMR activity. These data confirmed the accuracy and usefulness of the yeast assay as a simple method having an advantage to analyze a large number of variations without laborious steps. The three kinds of reporter systems identified some variants showing different DME phenotype among yeast assays (DME1+ or DME2+), although the majority (75 of 101, 74.3%) of the MLH1 variants showed consistent DME (DME− or DME3+). These discrepant variations showing DME1+ or DME2+ were supposed to be functionally subtle because these difference could be explained by the distinct thresholds of the reporter genes (LacZ, GFP, or ADE2), the target nucleotide repeats (mono- or di-), and/or the location of the repeats (a plasmid or a chromosome). This is supported by our finding that the average values of the in vitro MMR activities depended on the degree of the DME (from DME− to DME3+; Fig. 3A). Thus, a combination of the three yeast assays has the ability to evaluate functionally subtle variants as well as the in vitro MMR assay. Among variants showing discrepant results between the yeast assays and in vitro MMR assay, variants showing both DME3+ in the yeast assays and MMR− in the in vitro MMR assay (DME3+/MMR−) should be estimated to be pathogenic because MMR deficiency should link directly with carcinogenesis by causing genome instability regardless of the DME in yeast. DME−/MMR+ variants are difficult to be interpreted but possibly defect some unknown function in human cells, because the phenotype of these variants in yeast cells are quite different from that of a wild-type. We consider that DME1+ or DME2+ variants can also lose some function partially for the same reason.

Other functional analyses examining a part of 101 variants agree with our functional evaluation (2–5, 7). For example, 34 MLH1 variants were recently analyzed for four distinct functional properties containing in vitro MMR assay, protein expression, protein localization, and interaction with PMS2, and were evaluated for their pathogenicity by integrating all of the results (8). Among 20 variants assayed in common with this study, the in vitro MMR activities were low (<48.7% activity) in their 12 pathogenic variants, and high (>60.7%) in their eight nonpathogenic variants. DME phenotype in this study also corresponded with their interpretation in 16 of 20 variants. The consistencies in the functionally known MLH1 variants supported reliability of the functional evaluation for the newly analyzed variations in this study.

Our final purpose of analyzing the large number of MLH1 variants is to establish a database for understanding the pathogenicity of the gene alterations. One of the major problems in using our data for clinical purpose is that the appropriate cutoff value is difficult to set up, because we cannot estimate how intermediate or subtle functional defects contribute to the pathogenesis in HNPCC. However, the information on common polymorphisms found in the normal populations can be useful to consider the tolerable level predicted to be functionally proficient. The most common polymorphism is I219V, which was reported from various countries, although the allele frequency is varied from 3% to 36% (24, 38, 39). According to our data, I219V retained ∼60% of MMR activity and DME3+. The other seven putative polymorphisms retained also >60% MMR activity, the seven of the eight showed DME3+ except H718Y showing DME2+. Therefore, we currently propose that both DME3+ and 60% of MMR activity is the reasonable cutoff value to estimate the variations not associated with pathogenicity in HNPCC. Then, 50 MLH1 variations with more than this MMR activity are categorized into MMR+ ones, and 35 DME3+/MMR+ variants are thought to be functionally proficient. The D132H also retains DME3+/MMR+, which is a polymorphism among Israeli populations not associated with HNPCC but the sporadic colorectal cancer predisposition (23). Thus, we cannot exclude the possibility that even the variations showing DME3+/MMR+ might be involved in the sporadic carcinogenesis.

Another critical problem is that amino acid substitutions in MLH1 affected both protein expression levels and functions. Our data showed that there was just a slight correlation between MLH1 protein levels and in vitro MMR activities (Fig. 2C). Therefore, we predict that there are at least two mechanisms inactivating MLH1 function by amino acid substitutions, the shortage of MLH1 protein by protein instability, and functional inactivation by structural alteration. This suggested that the cutoff values are required in both functional level and protein level for overall evaluation. Recent studies have indicated that the immunohistochemical analysis of MMR proteins is one of the most efficient and sensitive screening method to detect abnormalities in MMR proteins (40, 41). In our data, in vitro MMR activities were impaired without reducing MLH1 protein levels in some MLH1 missense variants. This observation suggested that these pathogenic variants can be detected positively by immunohistochemistry, although we cannot directly link the protein amount of transient expression in cell lines with the endogenous protein level in tumors of the corresponding mutation carrier. Among the 50 variants with MMR−, 20 (40%) retained >75% of the MLH1 expression level of a wild-type. We speculate that these 20 variants (E23D, I25F, P28L, G54E, N64S, F80V, T82I, K84E, E102D, E102K, S193P, G244V, D304V, H329P, Q542L, I565F, L574P, P640T, and R755S) will be the candidate pathogenic variants with difficulty in clear detection as loss of protein expression by immunohistochemistry. Among them, three variants, P28L, F80V, and P648L, actually have been shown to retain their protein levels by immunohistochemical analyses (8, 37). Functional assays are especially useful for these variants because the abnormality will not be detected by immunohistochemistry until subsequent sequence analysis and functional assays.

Several lines of biochemical investigations have shown that the heterodimerization with counterpart proteins such as PMS2 and conformational change by ATP binding are important in the MLH1 function (1, 3, 7–10). Mapping 101 MLH1 variants on MLH1 cDNA indicated that the majority of functionally defective MLH1 variants (showing DME−, DME1+, or DME2+ and/or MMR−) were located within two functional domains, the NH₂-terminal ATPase and the COOH-terminal PMS2-interactive domains. In particular, almost all variants involved in the ATP-binding pocket were functionally defective (Fig. 4B and C). The homologous structure of the NH₂-terminal E. coli MutL provided molecular basis for functional defects of human MLH1 missense variants (9, 10). P28L, M35R, and S44F can disrupt ATP binding and hydrolysis; G67R, I68N, I107R, T117M, and T117R degraded ATP binding pocket; and other variants (A128P, V185G, R226L, G244D, and V326A) can alter or destabilize the overall protein folding. Our functional assay showed that 13 out of these 14 variants were functionally defective and supported the functional prediction based on the protein structure. In contrast, some variants probably do not change the protein structure because the changes are conservative (I25F, A29S, S44A, I68V, V213M, and I219L; refs. 9, 10). Recent study resolved crystal structure of COOH-terminal MutL and identified dimerization interface (11). The alignment and homology modeling show that L653R, R654L, and R659P resided on the equivalent to the dimmer interface of MutL, which is the putative interface of MLH1 with PMS2. T662P is equivalent site of MutL critical for binding with DNA (Fig. 4D). The homology modeling provides the structural basis of functional deficiency for some variants, although the functionally deficient variations of MLH1 distributed through the whole COOH terminus.

The previous studies have shown that MSH6 mutations cause a partial MMR deficiency and related with atypical HNPCC families (42). This suggests the possibility that intermediate functional deficiency is associated with the atypical HNPCC kindred with weak family history or late onset. Based on this hypothesis, one of the methods to validate our functional assay is investigating the relationship between functional evaluations with clinical feature such as family history. The median MMR activity was significantly lower among the AC+ families (38.9%) than the AC− families (65.1%), whereas the median level of relative protein expression has no significant difference between these two groups (Fig. 5). Therefore, AC status has correlation with our functional data at least stronger than with protein level. Functions in the assays well correlated with clinical features, whereas it can be said that the Amsterdam criteria links with functions well and is a very useful clinical diagnostic criteria. Our functional data have consistency with previous functional assays in a majority of the 45 functionally known variants, including some variants evaluated for pathogenicity by both cosegregation study and functional assay. Precise family information was described even in the limited number among some of the variants newly analyzed in this study. For example, S193P was detected in three affected members and two other members among 12 individuals in an AC+ family (24). This variant showed DME−/MMR− in our assays and then indicated to be pathogenic from both family history and functional assay. R687W was found in three affected members but not two healthy members in the AC+ family, suggesting that this variant is likely pathogenic (20). However, our functional data indicated DME− but relatively high MMR activity (57.2%) even below 60% (defined to be MMR− in this study). Yeast assay detected pathogenicity of this variant more clearly than MMR in this case. Although L607H was found in one family member with colon cancer but not in two healthy members in the family, this colon cancer did not show any microsatellite instability, low staining by immunohistochemistry (29), or functional defect in our assay. Then, pathogenicity of this variant is supposed to be in question but functional data correlates with microsatellite instability and immunohistochemical results. Thus, in the several cases in which detailed clinical information is available, the functional phenotype well correlated with clinical features for families carrying the corresponding variants. Based on this, it may be expected that the pathogenicity can be well predicted by the functional phenotype in the assay for many of the other variants, especially those showing clear phenotype such as DME−/MMR− and DME3+/MMR+. However, we have to accumulate more data on clinical and functional characteristics and evaluate accurate sensitivity and specificity of functional data for predicting pathogenicity, to use for practical purposes in clinics, for example, to decide whether surveillance of a mutation noncarrier should be continued.

In this study, we analyzed functional significances of 56 MLH1 variants that have never been evaluated in any functional assay. Without functional analyses, pathogenicity of missense variants are estimated usually based on amino acid property, conservation among species, or computed prediction like SIFT as a recent way. The SIFT has a good accuracy for the functional prediction; however, ∼20% of variants could fail to be predicted. A similar tendency has been observed in another analysis (8) and in other proteins such as p53 tumor-suppressor protein (data not shown). Therefore, it is still desirable to carry out biochemical assays to predict the pathogenicity for variants newly reported if available.

In summary, we examined a large number of MLH1 variants using both yeast and in vitro functional assays and characterized the functional alterations of the variants. We confirmed that the majority of functionally inactive variants were located in the NH₂-terminal and COOH-terminal domains, especially around the ATP-binding pocket and the region responsible for heterodimerization with other MutL homologues. The results corresponded well with observations by the structural analysis of E. coli MutL crystals. The study described here should be useful for evaluating cancer risks in individuals or families carrying MLH1 variants and may provide clues for better understanding MLH1 functions.

Grant support: Ministry of Education, Science, Sports and Culture 17015002 (C. Ishioka) and 16390122 (H. Shimodaira), the Sagawa Foundation for Promotion of Cancer Research (C. Ishioka), the Gonryo Medical Foundation (C. Ishioka), NIH Grant GM50006 (R.D. Kolodner), and the Swiss Cancer League (R. Iggo).

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.

We thank Dr. Thomas A. Kunkel (Laboratory of Molecular Genetics and Laboratory of Structural Biology, National Institute of Environmental Health Sciences, NIH, Department of Health and Human Services, Research Triangle Park, NC) for kindly providing bacteriophage M13mp2, E. coli strains, and human cells; Dr. Alan B. Clark and Yuka Fujimaki for their technical support for MMR assays; and Michael F. Kane for sequencing a part of the MLH1 variant expression plasmids.