To induce de novo methylation at hitherto unmethylated CGIs, the authors inserted a stretch of foreign CpG-free DNA into the embryonic stem cell genome close to either of the two promoters. The use of the word “acquired” to describe the imposition of the methylated state at these loci recalls the Lamarckian concept of inheritance of acquired characteristics, but, unlike adaptations caused by behavior or the environment, this one is triggered by drastic genetic manipulation. Thus, although the authors provocatively chose to target two genes (Ankrd26 and Ldlr) that have been implicated in epigenetic transmission of metabolic phenotypes between generations [5], the initiating event in this case is not the environment, but a mutation. It is nevertheless remarkable that when the inserted DNA was removed, leaving a small genetic scar (TTAA instead of TTCT and GTAC, respectively), the methylated state appeared to persist, not only in cultured cells but across many generations of mice derived from these cells.

While DNA methylation is the key “phenotype” being inherited, several lines of evidence suggest that it is not bearer of this memory. Firstly, primordial germ cells essentially erase DNA methylation at these CGIs, only restoring it later in development. Secondly, the authors refer to unpublished data showing that artificial methylation of a CGI in embryonic stem cells (ESCs) via recruitment of a DNA methyltransferase (epigenome editing) is not maintained in the same way as their genetic modification approach. Thirdly, as the aberrant methylation is mosaic (i.e., present in a subset of cells only), it is not clear how a mosaic methylation pattern would be inherited, given that a specific parental sperm or egg can only be methylated or be unmethylated. In some mouse lines, the targeted CGIs are even unmethylated in all germ cells and at the blastocyst stage. Fourthly, a purely epigenetic effect would be expected to weaken and fade away after a few generations, whereas levels of DNA methylation at these CGIs remain constant.

Whatever is being memorized at the locus, it does not appear to be DNA methylation. One obvious possibility is that a persisting genetic change (TTAA) renders these loci susceptible to de novo methylation. The authors are at pains to address this objection, as there is extensive evidence that DNA methylation patterns are often affected by DNA sequence polymorphism, pointing to a genetic rather than an epigenetic effect [6]. However, for Ankrd26, the authors subsequently derived scar-free cell and mouse lines by targeting a TTAA site 45 bp upstream of the TTCT site and here again the methylated state of the Ankrd26 SL allele was observed in three generations. Evidently, the presence of the scars did not guarantee the methylated state as the allele reverted to the non-methylated state in some mouse lines. Moreover, extensive DNA sequencing in the surrounding genomic region detected no other legacy DNA sequence changes.

If neither DNA methylation nor genome sequence were responsible, what are the alternatives? There is a precedent where naturally acquired DNA methylation survives the global demethylation that accompanies early mammalian development (although the modification is erased prior to the next generation). This concerns imprinted genes, where the methylated alleles specifically recruit KRAB zinc finger proteins which, with the aid of the repressor KAP1, protect the methylated allele [7]. Persistent methylation of transposable elements can also involve KRAB zinc finger proteins, but in this case DNA methylation is transiently lost during differentiation of the germline and subsequently restored as development proceeds [8]. In both these cases, continuity of DNA methylation depends upon DNA sequence-specific repressor proteins. These mechanisms could only explain consistent re-establishment of CGI methylation as reported by Takahashi and colleagues if the newly methylated state itself inadvertently created a DNA binding motif for repressors of this kind. In this context, it may be of interest that methylation of the Ankrd26 CGI gives rise to a binding motif for ZFP57 (TGCmCGC), a mouse protein implicated in protecting methylated imprinted loci against loss of methylation during early development [9, 10]. Also a new ZFP57 binding site is formed close to the Ldlr CGI. Neither of these identical sequence motifs would be capable of recruiting ZFP57 in the absence of CG methylation, but whether a single site could contribute to sustained methylation of these entire CGIs is unknown.

An interesting alternative mechanism depends on evidence that transcriptional activity or silence drives DNA methylation states. In this regard, it is important to note that the vast majority of vertebrate genomic DNA is highly methylated at CpG sites. Not only is heterochromatin methylated, but also to a roughly equivalent extent euchromatin, including intergenic DNA and gene bodies. So while is often assumed that DNA methyltransferases (DNMTs) are actively recruited to discrete target loci, the ubiquity of CpG methylation resembles a default state. Evidence that chromatin modifications can influence de novo DNMT recruitment (e.g., histone H3K36me2 [11]) appears to contradict this assertion, but these interactions arguably modulate rather than switch local DNMT activity. It is possible that the primary determinant of whether a CpG becomes methylated may not be attraction, but repulsion of DNMTs. In other words, DNA methylation goes everywhere that it is not excluded, most often by chromatin that has been or still is involved in active transcription, such as CGIs [12]. Mechanistically, we know that H3K4me3—a characteristic histone mark of CGIs and active promoters—excludes DNMT3A and DNMT3B from chromatin (reviewed in [13]). It appears that insertional mutagenesis has permanently crippled the ability of the affected CGIs to exclude DNA methylation, perhaps due to irrevocable interference with promoter function and concomitant alteration of the histone modification landscape. If the methylation-free status of each CGI indeed depends on active transcription at critical stages of development, it is possible that the high levels of DNA methylation following insertional mutagenesis may deny access to one or more transcription factors. Conceivably, this may tip the balance in favor of locus silencing at critical developmental stages, leading to absence of H3K4me3 and effectively permanent DNA methylation.