To the Editor:

Chan and colleagues report a kindred in which a specific haplotype at the MSH2 locus is associated with mosaic promoter hypermethylation and cancers suggestive of hereditary nonpolyposis colorectal cancer (HNPCC)1; they describe this as an example of germline epimutation similar to that which we and others have reported in another DNA mismatch repair gene, MLH1 (refs. 2,4,5,6). We have followed the terminology of Holliday7 in defining epimutation as epigenetic (transcriptional) silencing of a gene that is normally active3; epimutation may occur in the soma or the germline and is often accompanied by hypermethylation (the term may also be applied to activation or hypomethylation of a normally silent sequence8). The phenomena described by us and by Chan et al. seem to share an association between promoter hypermethylation in normal tissues and the risk of developing cancers with defective DNA mismatch repair. However, we suggest that what Chan and colleagues have found is distinctly different from the germline epimutation of MLH1, although no less interesting. Our conclusion rests on two features of the MSH2 epimutation: its mosaicism with prominent tissue specificity and its strong heritability.

In the MLH1 germline epimutations reported thus far, the epimutation appears to be present in nearly all somatic cells; we described the epimutation as mosaic only because it was absent from occasional somatic alleles3. In the kindred reported by Chan and colleagues, however, the mosaicism is prominently tissue specific: only the rectal mucosa shows a high proportion of cells carrying the epimutation. The authors propose that the MSH2 epimutation is a germline event and that somatic reversion has produced the tissue-specific pattern, but the evidence seems insufficient for this conclusion. It would be helpful to know more about the tissue distribution of the MSH2 epimutation—in particular, whether it is found in cells derived from all three germ layers. The tissue specificity of the epimutation in this unusual kindred seems consistent with somatic acquisition of a mosaic epigenetic state.

It is striking that the MSH2 epimutation appears to be inherited in a mendelian pattern: all individuals carrying the at-risk haplotype also have the epimutation. This is in sharp contrast to the MLH1 epimutation, which has occurred on multiple haplotypic backgrounds and seems to be only weakly heritable2,3,4,5,6. In the ten cases known to us, there is only one example in which the epimutation was transmitted to a proband's child6 (in another case, the epimutation was present in a low proportion of spermatozoa3). In all other cases where the affected allele was transmitted, it no longer carried the epimutation. Thus, whereas the MLH1 epimutation is not linked to a specific genotype and is clearly prone to reversion in the germline, the MSH2 epimutation appears to be inherited in lockstep with the underlying haplotype. We suggest that the MSH2 haplotype in this kindred predisposes in cis to somatic hypermethylation; the pattern is mosaic and tissue specific perhaps because of specific chromatin characteristics of the MSH2 locus in some tissues.

We have proposed that the MLH1 epimutation is an accident conditioned by the presence of epigenetically silent material (heterochromatin) in the germline3 and that any genomic sequence is potentially subject to this process, which can create the equivalent of a temporary loss-of-function mutation. The MSH2 epimutation may be something quite different: a haplotype that is prone to somatic hypermethylation in certain cell types (quite likely not limited to rectal mucosa). In essence, Chan et al. may have found an allele prone to spontaneous somatic epigenetic silencing similar to that which occurs in the FMR1 promoter when it carries an expanded triplet repeat9 and similar to other examples such as position-effect variegation. This view explains the strong heritability and tissue variability of the effect: it may be conditioned by the genetic structure of the allele and also by tissue-specific chromatin structure at the MSH2 locus.

The study of mammalian epigenetic inheritance is in its infancy (if it can be said to have been born at all). Compared with the routine predictability of mendelian inheritance, epigenetic effects are diverse, because epigenetics involves not only DNA sequence but also a variety of often unstable accretions to the DNA. We suggest that the phenomenon described by Chan et al. is distinct from germline epimutation as seen in MLH1. This does not make it any less significant, but the distinction illustrates the difficulties that will inevitably arise as we attempt to classify and analyze a complex and poorly understood system of inheritance.