Experimental Induction of Epigenetic Inheritance

In my last post on epigenetics, I argued that transgenerational epigenetic inheritance in mammals is weak to nonexistent. This is because epigenetic marks are erased between generations, both in the early embryo and prior to gamete development.

But, even if transgenerational epigenetic inheritance is too weak to be important in a natural setting, it might be possible to induce experimentally with a very strong epigenetic perturbation. Such an experiment would provide an upper bound on what we might expect to see in nature.

That’s what a recent paper (Takahashi et al., Cell 2023) tried in mice. Basically, they were able to induce transgenerational epigenetic inheritance, but the mechanism is still quite unclear: DNA methylation was erased during gamete formation, but then mysteriously re-established around the time of gastrulation.

Here’s what Takahashi et al. did, and what happened.

Inducing DNA methylation

The first step of the experiment was to make mouse embryonic stem cells (mESCs) with extra DNA methylation at particular sites in the genome. To do this, the authors took an interesting approach. I probably would have tried something like dCas9-DNMT3, where a DNA methyltransferase enzyme is fused to a Cas9 protein that is targeted to a specific site in the genome using a guide RNA. Instead, the authors used CRISPR/Cas9 with homologous recombination to integrate a piece of DNA entirely lacking CpG sequences.¹ This caused nearby CpG sequences to become methylated. It’s still not understood how this induces methylation, but the effect is reliable.²

Afterwards, they removed the integrated DNA using an excision-only variant of PiggyBac transposase, which restored the original sequence, but now it was methylated. It also had a small edit (TTAA instead of TTCT), which the authors used to trace the allele.³ The authors were able to achieve an increase in methylation from ~0% to >85% in some of the mESC lines they generated.⁴ They targeted the promoters of the genes Ankrd26 and Ldlr, which are important in lipid metabolism. Methylating the promoters caused these genes to be silenced. The methylation and silencing were stable in the mESCs, and were specific to those genes.

Generating mice

Next, the researchers injected the methylated mESCs into 8-cell stage mouse embryos to generate mice.⁵ The recipient embryos were from white mice, whereas the mESCs were from agouti mice, so the researchers could use coat color to identify mice where most or all of the cells came from the mESCs. They identified two mice where this was the case. These mice had ~65% and ~100% methylation of the Ankrd26 promoter in their tail-tip cells, indicating that the methylation in mESCs largely persisted in somatic tissues. The methylated mice also had less Ankrd26 mRNA expressed, and higher body weight than wild type mice.⁶ At the Ldlr site, this method was less successful, and the methylation was only about 20%.

Epigenetic inheritance

The researchers next mated the mouse that had the most Ankrd26 promoter methylation with a wild-type female. The resulting offspring (F1) still had the methylation, and this persisted all the way through the F4 generation! The presence of methylation was associated with obesity (similar to what was seen in the F0 mice), and the methylation could apparently be transmitted through both eggs and sperm.

The researchers also did this with mice that had Ldlr promoter methylation. Here, the effect was less stable, resulting in mosaicism where different tissues had different levels of methylation. However, the researchers were able to observe increased cholesterol levels in F2 and F3 mice that had Ldlr promoter methylation.

Interestingly, the researchers did not observe the same levels of methylation in primordial germ cells (PGCs). When they isolated the PGCs from the mice, the Ldlr methylation was erased, as would be expected. The Ankrd26 methylation was also partially, but not fully, erased. Likewise, in sperm and eggs, the methylation was not re-introduced. But this seems to contradict their previous finding that the methylation was heritable! So, why were the alleles in the offspring methylated again???

To determine the developmental stage at which the methylation was re-established, the authors collected embryos from wild-type female mice that had been mated with Ldlr-methylated males. Importantly, neither the sperm nor the eggs had methylation at Ldlr. Neither did the blastocysts, after 3.5 days of development. But by 6.5 days, at the epiblast stage, the methylation was present again.

WTF is going on?

In accordance with Figure 7K from the paper, I propose to call this effect “trensgenerational epigenetic inheritance”

This is very weird. According to the authors, the hypermethylation causes some change in the eggs and sperm, that causes the hypermethylation to be re-established in the developing embryo even after it is partially or fully erased. It’s still a mystery what this persistent epigenetic change is. I do appreciate the investigation of methylation levels in PGCs and gametes, since most epigenetic inheritance papers don’t do this.

But at this point, the study raises more questions than it answers. One important point, which the authors mention in the discussion section, is whether this will generalize to other genes. The effects were most pronounced at Ankrd26, and weaker at Ldlr. Perhaps the authors just got lucky in studying Ankrd26. My biggest question, though, is about the mechanism of this effect.

In the discussion, the authors write,

when we only induced de novo DNA methylation at Ankrd26 and Ldlr CGIs using the dCas9-DNA methyltransferase (DNMT) systems, the acquired DNA methylation was not stably maintained in mouse ESCs (data not shown).⁷

So, their method of introducing DNA methylation by inserting CpG-free DNA might actually be acting through another type of epigenetic mark.

Overall Conclusions

This paper did a better-than-usual job of inducing DNA methylation and carefully tracking it over different stages of mouse development. There is probably something interesting going on here. However, at this point I’m still not ready to take back my claim that transgenerational epigenetic inheritance in mammals is, for practical purposes, negligible. Indeed, this paper confirmed that DNA methylation is largely erased during gametogenesis.

I hope that these researchers, or others, follow up on this study and elucidate the mechanism of the reported inheritance. When that happens, I’ll be sure to post about it here.

I thank Niko McCarty and Devon Stork for feedback on a draft of this post.

1. These should really be called “CG” sequences, it’s unnecessary to write “p” for phosphate. We don’t write TpTpApA for TTAA PiggyBac recognition sites. But almost everyone else (except my undergrad biochem textbook) writes CpG and I’ll follow the same convention here.

2. It might be due to disruption of histone marks such as H3H4me3, see this previous paper by the same researchers. See also this methods paper.

One advantage of using this over a dCas9-DNMT3, is that dCas9-DNMT3 actually inhibits methylation where it binds (since the Cas9 physically blocks DNMT activity) and only methylates nearby sites.

3. This edit wouldn’t be expected to have any effect, but I don’t think the authors did enough characterization to prove this. In most of their experiments, they are comparing methylated mice with the TTAA edit, to unmethylated mice with the wild-type sequence. This would all be invalidated if it turned out the TTAA sequence was causing the phenotypes. They did do a few experiments with seamless methylation edits, but I think this control should have been included for all of their experiments.

4. Although the efficiency wasn’t uniform, the authors were able to generate many mESC lines and pick the ones where it worked the best.

5. This is known as the Velocimouse method.

6. I have an objection to this, though. The control group for this experiment was the strain of mice used for the recipient embryos (ICR), which has a different genetic background to the mESCs (C57BL/6;129S hybrid). They should have compared the mESC-derived mice with C57BL/6;129S hybrid mice. I don’t know if the strain differences are big enough to invalidate the experiment, but it’s a bit weird. Also they only used three mice per group.

7. Too bad they got away with “data not shown”, I’d really like to see that data.