Enhanced CpG Mutability and Tumorigenesis in MBD4-Deficient Mice

See allHide authors and affiliations

Science  19 Jul 2002:
Vol. 297, Issue 5580, pp. 403-405
DOI: 10.1126/science.1073354


The mammalian protein MBD4 contains a methyl-CpG binding domain and can enzymatically remove thymine (T) or uracil (U) from a mismatched CpG site in vitro. These properties suggest that MBD4 might function in vivo to minimize the mutability of 5-methylcytosine by removing its deamination product from DNA. We tested this hypothesis by analyzing Mbd4 −/− mice and found that the frequency of of C → T transitions at CpG sites was increased by a factor of three. On a cancer-susceptibleApc Min/+ background,Mbd4 −/− mice showed accelerated tumor formation with CpG → TpG mutations in the Apc gene. Thus MBD4 suppresses CpG mutability and tumorigenesis in vivo.

Deamination of 5-methylcytosine (m5C) to T at CpG sites is probably the single most important cause of point mutations in humans, accounting for more than 20% of all base substitutions that give rise to genetic disease (1). Estimates based on the in vitro deamination rate of m5C (2) suggest that approximately four m5C residues deaminate per diploid genome per day in the germ line. It is likely that many of the resulting mismatches are repaired, but appropriate repair must take into account that the incorrect base at the T:G mismatches is invariably T, not G. Two mismatch-specific T glycosylases that accomplish this discrimination in vitro have been discovered. Both thymine DNA glycosylase [TDG; (3, 4)] and MBD4 (5,6) can specifically remove T from a T-G mismatch within a CpG context without cleaving the DNA strand. This study concerns MBD4, which contains a COOH-terminal glycosylase domain (7, 8) and an NH2-terminal methyl-CpG binding domain [MBD; (7)]. The MBD binds to symmetrically methylated CpG sites but has a higher affinity in vitro for m5CpG/TpG mismatches, which arise due to deamination at methyl-CpG (5). Neither TDG nor MBD4 has been tested for its effect on mutations in vivo.

To investigate whether MBD4 suppresses the mutability of methylated CpG sites in vivo we generated mice bearing a mutated Mbd4 gene by means of targeted allele replacement in embryonic stem cells (9). The mutated allele contained a cassette within intron 1 containing a splice acceptor and polyadenylation signal. Mice homozygous for this allele were viable and fertile. Northern blot (10) and reverse transcriptase–polymerase chain reaction (RT-PCR) analyses confirmed that MBD4 expression was less than 0.1% of the wild-type level in tissues from homozygous-mutant mice (fig. S1). We refer to this allele as Mbd4 throughout this study.

To determine whether MBD4 deficiency leads to an increase in mutations at CpG sites, we crossed mice with the “Big Blue” reporter locus, comprising 40 head-to-tail copies of a recoverable lambda transgene (11), onto the Mbd4 −/− background and measured in vivo mutation frequencies at the λ cIIlocus (12). Bisulfite sequencing (9) of thecII locus from both wild-type andMbd4 −/− mice showed that, on average, 95% of the CpG sites in the cII gene are methylated (Fig. 1A). The frequency of cIImutations in liver and spleen of 105-day-old mice and in spleens of 183-day-old mice was determined by plating packaged genomic DNA (9, 12). The total mutant frequency inMbd4 −/− mice (6.8 × 10−5) was significantly higher than in wild-type animals (3.2 × 10−5) by a maximum likelihood ratio test [P < 0.0001; Fig. 1B; Table 1; (9)]. This difference was reduced, but remained statistically significant (P < 0.05), when a correction for the number of independent mutations was applied [Fig. 1B; (9)]. The true mutation frequency is likely to lie between the uncorrected and corrected values, as correction may underestimate the real number of independent mutational events.

Figure 1

MBD4-deficiency increases the frequency of CG → TA mutations at CpG sites. (A) The positions of the 22 CpG sites in the cII gene are indicated by vertical lines. The methylation status of 10 clones fromMbd4 −/− spleen DNA is shown below, with filled boxes denoting methylated sites and unfilled boxes denoting unmethylated sites. The number of sequenced clones with identical methylation patterns is indicated on the right, where it exceeds one. A similar level of methylation was observed inMbd4 +/+ mice, and in liver tissue from both genotypes (21). (B)Mbd4 −/− mice (hatched bars) have a significantly higher frequency of mutant plaques (P < 0.0001) and of individual mutations (P < 0.05) than wild-type mice (white bars). These data sum all time points and tissues. (C) Increased percentage of all point mutations that are caused by CG → TA mutations at CpG sites inMbd4 −/− mice, showing both raw and corrected data. (D) Mutational spectra on theMbd4 +/+ andMbd4 −/−backgrounds (corrected data). Mutations at CpG sites in all categories are shaded in black, and the total number of individual mutations found in each category for each genotype is indicated below the appropriate bar.

Table 1

Analysis of mutation frequencies at the cIIlocus of a bacteriophage λ transgene inMbd4 −/− and Mbd4 +/+mice. Numbers of “mutants” were derived from raw counts of mutant plaques. Numbers of “mutations” were deduced by correcting mutant numbers for likely clonal expansion of a single mutant cell (9). “Mutant” and “mutation” frequencies reflect the uncorrected and corrected data, respectively.

View this table:

A striking difference between wild-type andMbd4 −/− mice emerged when the spectrum of mutations was examined. The most abundant mutational change in both wild-type and mutant mice involved CG → TA transitions, but this category was more frequent in Mbd4 −/− mice than in Mbd4 +/+ controls (Fig. 1D; corrected data were analyzed). The difference between wild-type andMbd4 −/− mice could be specifically attributed to a highly significant 3.3-fold increase in CG → TA transitions at CpG sites (P < 0.0001; Fig. 1D). Other mutation categories, including CG → TA transitions at non-CpG sites, were not significantly altered by the absence of MBD4 (P > 0.1). The effect on the overall mutation spectrum was apparent from the fraction of all point mutations at CpG (Fig. 1C). In wild-type mice, 28 to 31% (this study) and 26.9% (13) of point mutations were found to be at CpG, which is similar to the fraction deduced from human genetic diseases (23%) (1). InMbd4 −/− mice, however, the fraction rose to between 56% (corrected) and 76% (uncorrected).

We next asked whether the increased mutability ofMbd4 −/− mice led to an increased incidence of cancer in an in vivo model of intestinal tumorigenesis. MBD4 is mutated in 26 to 43% of human colorectal tumors that show microsatellite instability (14, 15), suggesting that loss of MBD4 may contribute to genome instability (16). To test whether loss of MBD4 increases tumorigenesis, Mbd4 −/− mice were made heterozygous for the Min allele of the adenomatous polyposis coli gene (Apc Min), which predisposes mice to the development of spontaneous intestinal neoplasia (17). Littermates were divided into (Mbd4 −/−,Apc Min/+) and (Mbd4 +/−,Apc Min/+) cohorts and killed when they became symptomatic for intestinal neoplasia (9).Mbd4 −/− mice showed markedly reduced survival compared with Mbd4 +/– controls (P = 0.001, Fig. 2A). Moreover, whole-mount analysis showed a small but significant (P = 0.01) increase in intestinal adenoma burden at death in (Mbd4 −/− , Apc Min/+) mice (median = 28 tumors) compared with (Mbd4 +/– , Apc Min/+) mice (median = 14 tumors) (Fig. 2B). Histological analysis revealed no morphological differences between the Mbd4 +/–and Mbd4 –/tumors.

Figure 2

MBD4 deficiency accelerates intestinal neoplasia on an Apc Min/+ background. (A) Kaplan-Meier plot showing significant (P= 0.001, log rank) reduction in survival of (Mbd4 −/−, Apc Min/+) mice compared with (Mbd4 +/– , Apc Min/+) mice. Solid black line (Mbd4 +/− , Apc Min/+) mice (n = 17); broken line (Mbd4 −/− , Apc Min/+) mice (n = 36). (B) Tumor burden in (Mbd4 +/– , Apc Min/+) and (Mbd4 −/−, Apc Min/+) mice at necroscopy. The median value for each genotype is indicated by a horizontal line in the boxplots. There was a small but statistically significant increase (Mann-Whitney U test, P = 0.01) in tumor number in the (Mbd4 −/− , Apc Min/+) mice. (C) Apcmutations in tumors of the large intestine in (Mbd4 −/−, Apc Min/+) mice. The upper band (Min) is the Hind III-resistantApc Min PCR product, whereas the lower band (“wild type”) is the initially wild-type Apc allele, which can be cut by Hind III. Lanes a, nontumor control samples from intestines of Mbd4 +/– mice; lanes b, tumors from Mbd4 +/– mice showed 100% (n = 20) LOH; lanes c, tumors fromMbd4 −/− mice showed 62% (n = 39) LOH, which is significantly less thanMbd4 +/– mice (P < 0.001, chi-square test). The presence of a weak Apc +band in tumors that have lost this allele is due to unavoidable contamination with nontumor tissue (notably, blood).

To determine whether the loss of repair activity at CpG sites contributed to the accelerated tumor formation, we analyzed the nature of mutations causing loss of function of the wild-type Apcallele in intestinal tumors. On an Apc Min/+background in the absence of mutagenic challenge, the overwhelming cause of spontaneous tumors is complete deletion (loss of heterozygosity, or LOH) of the Apc + allele (18). In agreement with this, we found that all 15Mbd4 +/– tumors in the large intestine showed loss of the Apc + allele. In contrast, 15 of 39 tumors (38%) from (Mbd4 −/− , Apc Min/+) mice retained both Apc alleles (Fig. 2C). Sequencing of part of the Apc gene (9) in the latter tumors revealed eight mutations. Five mutations were CpG → TpG transitions, of which four would cause truncation of the APC protein (table S1). The complete absence of equivalent mutations in control mice indicates that the suppression of intestinal tumors by Mbd4 is due at least in part to suppression of CpG → TpG transitions at the endogenous Apc gene.

Our findings clearly implicate MBD4 in the repair of m5C deamination at methylated CpG residues, but is it solely responsible for this repair? If we assume that the in vitro deamination rate of m5C (5.8 × 10−13/second) (2) applies in vivo, we predict a C → T transition frequency of 1.25 × 10−4 at CpG sites in the cII locus in the absence of repair. The observed mutation frequency in wild-type mice is 4% of this calculated value, suggesting that repair of m5C deamination is about 96% efficient. In Mbd4 −/− mice, the CpG to TpG mutation frequency is 1.5 × 10−5, suggesting that repair efficiency has fallen to 88% but remains moderately effective. This approximate calculation indicates that repair of deaminated m5C is shared with factors other than MBD4, for example, TDG or equivalent activities. In theory, two independent repair activities that were each ∼80% efficient could give a combined efficiency of ∼96%.

Why is m5C a mutational hotspot in spite of the presence of at least one repair mechanism? One plausible explanation is that the existence of close to 109 T residues in the mouse genome, each one losing its base pairing with A many times per second (19), renders enzymes that remove unpaired T potentially mutagenic. With respect to C deamination, this problem is thought to have been solved by replacing U with T in the ancestral RNA-derived genome, so that the deamination product of C became easily distinguishable from a normal DNA base (20). All known U:G mismatch glycosylases are inert against T:G mismatches, despite the structural similarity between T and U, perhaps because of the danger of T removal. By methylating C, however, the problem is recreated, as deamination once more generates a normal DNA base whose excision may be problematic. Inefficient correction of the resulting T:G mismatches may represent a compromise between the benefits of repair and the damage that could arise by removal of legitimate T residues.

In conclusion, we have shown that murine MBD4 suppresses CpG → TpG mutations in both a bacteriophage-based transgene and at the endogenousApc gene locus. These findings suggest that human MBD4 plays a similarly important role in reducing inherited disease and cancer.

  • * These authors contributed equally to this work.

  • To whom correspondence should be addressed. E-mail: a.bird{at}


View Abstract

Navigate This Article