MUTYH-associated polyposis (MAP; MIM #608456) is an autosomal-recessive inherited familial colorectal cancer (CRC) and polyposis syndrome [Nielsen et al., 2010]. Biallelic mutations in MUTYH genes (MIM #604933) predispose patients to the development of polyps, generally between 10 and a few hundred polyps. The MUTYH gene encodes a base excision repair (BER) glycosylase, which is involved in the repair of the major base lesion 8-oxo-guanine (8-oxoG), caused by oxidative damage and prevents G:C to T:A transversion [David et al., 2007]. The MUTYH protein recognizes oxoG:A mismatches and excises the undamaged adenine. MUTYH consists of different functional domains (Fig. 1A). The N-terminal domain includes a catalytic region with a helix–hairpin–helix (HhH) motif, as well as a pseudo-HhH region and iron–sulfur cluster loop (FCL) motif, which function in the recognition and excision of adenine moieties opposite 8-oxoG [Guan et al., 1998; Lukianova and David, 2005]. The C-terminal domain is known as a MutT-like domain that shares the homology with the nudix-type motif 1 and functions in the recognition of the 8-oxoG lesion [Noll et al., 1999].

A biallelic variation p.Y179C and/or p.G396D has been reported in up to 70% of patients with MAP in Caucasian populations [Nielsen et al., 2010]. However, these mutations have not been identified in Asian populations (e.g., Japanese and Korean), suggesting the existence of founder mutations and ethnic differentiation. Indeed, the p.A359V is a common and likely founder variant in Japanese and Korean patients [Yanaru-Fujisawa et al., 2008]. Other locally common variants have been identified (e.g., p.P405L in Dutch and p.E480del in Italian populations) [Gismondi et al., 2004; Nielsen et al., 2005].

The Leiden Open Variation Database (LOVD) (http://www.lovd.nl/3.0/home) is a well-designed, open-access database of MUTYH variants [Out et al., 2010]. The majority of MUTYH alterations are missense variants. Currently, more than 200 unique missense variants have been registered in LOVD. Because the significance of these missense variants with respect to protein function is difficult to evaluate, functional assays are required to understand the pathogenesis of MUTYH variants.

One of the most common assays is the in vitro glycosylase assay, which examines the enzymatic activity of recombinant MUTYH proteins to excises adenines opposite 8-oxoG on the oligonucleotide substrates [Ali et al., 2008; D’Agostino et al., 2010]. Another common assay is the in vivo E. coli complementation assay, which examines the abilities of exogenously expressed MUTYH proteins to suppress MutY-deficient E. coli mutation [Kundu et al., 2009].

Functional analyses of two major variants, p.Y179C and p.G396D, have been performed using cells from several species with the equivalent variants. Furthermore, the structural resolution of Bacillus stearothermophilus MutY has provided a structural basis for these variants [Fromme et al., 2004]. Y179 is located on the DNA-binding interface, where it recognizes 8-oxoG:A mispairings and stabilizes protein–DNA complexes. G396 also contributes to 8-oxoG:A mispair recognition and imparts flexibility to the MUTYH conformation. In addition to the two major variants, functional assays have been performed for a number of MUTYH variants. However, the results have been inconsistent [Bai et al., 2005; Ali et al., 2008; Goto et al., 2010]. Moreover, almost all studies have assayed relatively small numbers of variants.

To evaluate the functional significance of many variants comprehensively, we assayed 47 variants differentiated by only subtle variations such as missense mutations or one amino acid deletion. The present study will provide useful information to understand the pathogeneses of these 47 MUTYH variants.

In this study, we evaluated the functional significance of 47 MUTYH variants by an analysis of E. coli complementation, protein stability, and subcellular localization in mammalian cells to enhance our understanding of the pathogenic effects in MAP.

Several functional assays were performed using a few different methods, of which the most frequently used is the in vitro glycosylase assay. This assay evaluated several variants to be functionally defective (p.Y179C, p.R182H, p.R241W, p.R245H, p.R245L, p.V246F, p.G286E, p.P295L, p.L388P, p.P405L, p.A473D, and p.E480del), partially defective (p.I223V, p.M283V, p.R274Q, and p.G396D), or retained (p.V22M, p.V75E, p.R185W, p.R309C, p.A373V, p.Q338H, and p.S515F) [Wooden et al., 2004; Bai et al., 2005, 2007; Ali et al., 2008; Kundu et al., 2009; Goto et al., 2010]. Our data were consistent with those previously published data for almost all variants, with the exceptions of p.V246F and p.G286E; these were evaluated as retained in our study, but defective by the in vitro study. p.V246F has been shown to be defective in terms of in vitro glycosylase activity, but only slightly defective in the E. coli complementation assay performed in the same study. This suggests that the discrepancy could be attributed to the partial activities assessed by the different methods. The reason for the discrepancy in p.G286E is uncertain, because both an in vitro glycosylase assay and analysis of a knock-in mouse harboring the equivalent variant have shown that this variant is defective; however, it was found to be retained in our study with the clear low mutation rate (0.5-fold) other than border-line

Eight variants (p.Y179C, p.R241W, p.R245L, p.V246F, p.Q338H, p.Q338R, p.G396D, and p.P405L) were examined in previous studies using a different E. coli complementation method than that used in our study [Shinmura et al., 2000; Bai et al., 2005; Bai et al., 2007; Kundu et al., 2009]. We assayed the variants according to complementation ability in MutY-deficient E. coli; in contrast, all the previous studies used MutY and MutM double-deficient E. coli to enhance the mutation rates. Among the eight previously examined variants, seven were assayed in the present study. Both studies found p.Y179C, p.R241W, p.G396D, and p.P405L to be defective and p.Q338H and p.Q338R to be retained. However, p.V246F was evaluated as retained in the present study but defective or slightly defective in the previous studies. These discrepancies are possibly due to the different E. coli backgrounds. Nevertheless, our functional evaluations are consistent with the previous studies for the majority of variants. This corroboration suggests that the functional evaluation for variants newly analyzed in the present study was reliable.

We analyzed a large number of MUTYH variants to comprehensively estimate the pathogenesis of each variant in a simple E. coli assay as a stand-alone method. However, the clinical application of our results might be limited by the difficulty in establishing an appropriate cut-off value, as we could not estimate how intermediate or subtle functional defects would contribute to the pathogenesis of MAP. p.Y179C and p.G396D have the highest incidence among Caucasian patients with MAP. In our assay, p.Y179C exhibited a loss of function, whereas p.G396D demonstrated only a 1.7-fold higher mutation rate relative to the wild type. Several functional assays have shown that p.Y179C induces a more severe functional defect than p.G396D does [Parker et al., 2005; Ali et al., 2008; Kundu et al., 2009]. Recently, the phenotype of patients with MAP harboring a biallelic p.G396D variation was found to be less severe than that of patients harboring a biallelic p.Y179C variation in terms of the hazard ratio for CRC development or the age at the diagnosis [Nielsen et al., 2009]. These genotype–phenotype correlations support our hypothesis that a comprehensive functional assay of a larger number of variants will provide insight into the clinical features, such as the relative risk of CRC or the age of onset. The most common polymorphism is p.Q338H, which has been reported in various countries, although the allelic frequency varies from 4.8% to 64% for a heterozygous allele [Ali et al., 2008]. According to the HapMap project, three common variants have been found in normal populations: p.Q338H (allelic frequency, 30%), p.V22M (13%), and p.S515F (3%). According to our data, p.Q338H, p.V22M, and p.S515F retained similar BER ability to wild type in E. coli. Considering the mutation rates of these five variants (p.Y179C, p.G396D, p.Q338H, p.V22M, and p.S515F), we set the two cut-off lines at 1.7- and 4.8-fold, respectively. These cut-off values were tentatively set for convenience to analyze the data. The appropriate method of categorization should be investigated by further research.

Clinically, the pathogenesis of two common variants (p.Y179C and p.G396D) in MAP is certain, whereas the pathogenesis of three putative polymorphisms (p.V22M, p.Q338H, and p.S515F) is less likely. However, rare variants are difficult to interpret from clinical data. Although biallelic variations predispose patients to the development of polyps and CRC (genotype relative risk: GRR of 117), monoallelic variation carriers have only a slightly increased risk of CRC (GRR of 1.27) [Tenesa et al., 2006]. Accordingly, we could not conclude that monoallelic variations found in a normal population are not pathogenic. Hence, clinical data from monoallelic mutation carriers are not suitable for comparison with functional data. By examining the LOVD database or literature concerning the MUTYH variants, rare variants were categorized into two groups: those reported in patients with only familial adenomatous polyposis, attenuated familial adenomatous polyposis, or multiple polyps; and those reported in healthy controls or patients with sporadic colon cancer (Table1). p.Y128H, p.G189E, p.R245C, p.R245H, p.G286E, p.P295L, p.P405C, p.A473D, and p.V493F were identified in the former group and detected in homozygous states. Our functional assay revealed eight of the nine above-mentioned variants except p.G286E to be defective. Therefore, our functional assays tended to correlate with the clinical features. However, further analysis will be required to understand the significance of these variants. We cannot exclude the possibility that the variant might still be associated with sporadic carcinogenesis, even though a high BER activity will likely not contribute to the pathogenesis of MAP. Even the most common putative polymorphism p.Q338H has been shown to be possibly involved in sporadic colorectal carcinogenesis in a population-based analysis [Picelli et al., 2010]. It is assumed that the reduced ability of this MUTYH variant is caused by a reduced binding capacity for the RAD9–RAD1–HUS1 complex, because the mutation at Q338 residue occurs in the interconnecting domain involved in the interactions with the HUS1 component of that complex [Turco et al., 2013]. Applications of functional information with genetic and clinical features such as the individual incidence of cancer will provide useful information regarding the validity of these functional analyses. Segregation analysis that is usually useful is unsuitable because of the low penetrance of monoallelic MUTYH variations.

Amino acid substitutions in MUTYH are expected to affect both the expression levels and functions of protein. However, our data showed that each variant did not lead to a significant difference in the amount of protein expression in the HCT116 cell lines. Therefore, there was no clear correlation between the MUTYH protein level and BER activity (Table1). A previous study has reported lower levels of endogenous MUTYH in cell lines harboring biallelic p.Y179C/p.G396D variants established from patients with MAP [Parker et al., 2005]. In another study, exogenous MUTYH protein variants in MutY knock-out MEFs were expressed at various levels [Molatore et al., 2010]. It is possible that endogenous protein levels differ among variants and the host cell influences that protein stability. Our results indicate that the variants exert slight effects on protein stability, and the effect of protein stability could almost be excluded from the evaluation of our functional data.

Subcellular localization of MUTYH differs from that of the isoforms generated by alternative splicing [Takao et al., 1998]. Isoforms 2 and 4 are the two major well-described isoforms [Takao et al., 1999]. Isoform 2, which features a mitochondrial targeting signal in the N-terminus, is predominantly located in the mitochondria. Isoform 4 has a nuclear localization signal in the N- and C-termini and is the most abundant nuclear isoform. Our results revealed that variants of MUTYH had no significant effects on the subcellular localization of isoform 4. Although a putative nuclear localization signal was shown to be located in C-terminal 55 residues, five variants (p.G503E, p.S515F, p.P516L, p.L526M, and p.R534Q) that affected this domain did not impair nuclear distribution. Our data suggest that the main reason for the functional defects caused by MUTYH variants is not altered distribution, but rather the loss of protein function itself.

According to the alignment, human MUTYH shares a well-conserved central region with E. coli MutY and features extended segments toward both the N- and C-termini [David et al., 2007]. The exclusive distribution of the functionally defective MUTYH variants in the central region suggests that this homologous region is important for protein function. Mapping of the secondary structure of MUTYH showed that the functionally defective variants were distributed widely in the two functional domains (i.e., the N-terminal catalytic domain and the C-terminal MutT-like domain) without a clear hot spot (Fig. 1A). In a previous study, the homologous structure of the B. stearothermophilus MutY provided a molecular basis for the functional defects of human MUTYH missense variants [Fromme et al., 2004]. p.W131R, p.Y179C, p.R182C, p.R182H, and p.R185Q can disrupt DNA-binding ability because these are near the equivalent site of the oxo-G or adenine-recognition surface of B. stearothermophilus MutY. p.C290W and p.P295L are in the FCL motif, and p.R241W resides in the equivalent site of the interface of [4F-4S] cluster. Our mapping of variants on the predicted MUTYH structure based on homology modeling indicated that functionally deficient variants surrounded the [4F-4S] cluster. This cluster is well conserved among DNA glycosylases such as E. coli endo III and is believed to play a structural role by searching damaged DNA for base removal activity [Boal et al., 2007]. Our homology modeling provides a structural basis for the functional deficiencies with some variants.

These sets of functional data also allow us to compare the properties of SIFT and PolyPhen-2 predictions. By comparing the BER complementation assays with these algorithms, we could evaluate the abilities of these programs to make functional predictions. Both SIFT and PolyPhen-2 were highly predictive of BER activity in MUTYH. Among the 17 variants deemed as retained and 13 deemed defective, consistent predictions using our functional assay were made for 22 variants (73.3%) with SIFT and 26 variants (86.7%) with Polyphen-2. Based on the data from our functional assay, the false-positive and false-negative rates were 15.4% and 35.3% for SIFT, respectively, and 0% and 23.5% for Polyphen-2, respectively. Our data indicate that the accuracy of in silico predictions can exceed 70%. This high level of accuracy was also supported by our previous study, which compared functional assay data with the SIFT predictions of 101 variants of the mismatch repair gene MLH1 [Takahashi et al., 2007].

In summary, we examined 47 MUTYH variants in functional complementation assay with MutY-disrupted E. coli, and characterized the functional alterations of MUTYH variants. We confirmed that the majority of functionally inactive MUTYH variants were located around the DNA-binding domain and [4Fe-4S] cluster. The results described herein can be applied to evaluate the risk of cancer in individuals or families harboring MUTYH variants, and might provide insight into the functions of MUTYH.