A memory that is not written in the letters
By now you have met the layers of eukaryotic control: transcription factors that read promoters and enhancers, and chromatin that can bury a gene in tightly packed heterochromatin or lay it open. But one puzzle remains. When a liver cell divides, both daughters are liver cells — never a stray neuron. The cell has somehow *remembered* its identity and copied that memory along with its DNA. And here is the twist: the genome in every one of your cells is, letter for letter, the same. The memory cannot be written in the sequence itself. So where is it?
The answer is [[molbio-epigenetics|epigenetics]] — from the Greek *epi-*, 'on top of'. It is the layer of marks and states sitting on top of the DNA that tells the cell which genes to keep on and which to keep off, without rewriting a single base. The crucial word in the definition is *heritable*: an epigenetic change is one that survives cell division, so the daughter cells inherit the parent's settings. That is exactly the memory we were missing. Two carriers do most of the work — chemical tags on the histone tails, which you met as the histone code, and a tag placed directly on the DNA. This guide is about that second one, the most stable and best-understood mark in mammals.
A methyl stamp on the C of CpG
The mark is [[dna-methylation-cpg|DNA methylation]], and it is wonderfully simple. An enzyme attaches a tiny chemical group — a methyl group, just one carbon with three hydrogens — onto the base cytosine, turning it into 5-methylcytosine. Chemically it is still a cytosine; it still pairs with guanine exactly as before. The only change is a small methyl bump now poking out into the major groove of the helix, where reader proteins can feel it. Think of it as a stamp pressed onto a letter without altering a word of the text: the message reads the same, but a seal now sits on top of it.
In mammals the tag goes on almost only where a cytosine is immediately followed by a guanine on the same strand — a spot written CpG (the lowercase p just stands for the phosphate linking the two). This little detail is the whole trick of heredity. Because C pairs with G and G pairs with C, a CpG on one strand always sits directly across from a CpG on the other, reading in the opposite direction. The pattern is symmetric. So when the DNA copies, each new double helix has one old, methylated strand and one fresh, bare one — and a maintenance enzyme simply reads the methyl tag on the old strand and stamps the matching CpG on the new strand. The mark thus rides through every round of cell division, which is precisely what makes it heritable.
CpG is symmetric across the two strands:
5'- ... C G ... -3' methyl (m) on the C
| |
3'- ... G C ... -5' and on the C of the other strand
After replication, one strand still carries m;
a maintenance enzyme copies it onto the bare new strand:
old: -m C G- new pair: -m C G-
- G C - ---> - G C m- (now restored)
So the methylation PATTERN is inherited, base for base.Now, where does silencing come from? Many gene promoters sit inside CpG islands — short stretches unusually crowded with CpG sites. Normally these islands are kept bare of methyl tags, and the gene stays available to be switched on. But when an island becomes methylated, the meaning flips: reader proteins that specifically recognize methyl-CpG move in, drag along chromatin-compacting machinery, and the local DNA is pulled into closed heterochromatin. The promoter is now physically buried, transcription factors can no longer reach it, and the gene falls durably silent. Methylation does not cut the wire; it locks the door.
Two famous jobs: imprinting and the silenced X
Methylation earns its keep wherever a gene must be shut *durably and selectively*. The cleanest example is [[molbio-genomic-imprinting|genomic imprinting]]. You inherit two copies of most genes, one from each parent, and usually both work. But for a small set of genes the cell keeps a label of which parent each copy came from and obeys only one. The other is silenced — not because it is broken, but because of which parent it arrived through. The label is a methyl mark laid down during egg or sperm formation: a mother's eggs methylate certain genes, a father's sperm methylate a different set, and the embryo respects both decisions. This is a striking exception to the textbook rule that your two gene copies are interchangeable equals.
The second showcase is [[molbio-x-chromosome-inactivation|X-chromosome inactivation]]. Females carry two X chromosomes, males one, yet both need roughly the same dose of the hundreds of proteins those X genes make. Doubling the dose in females would be harmful, so the body's solution is drastic: early in a female embryo, each cell independently and at random shuts down *one entire X chromosome* — sometimes the mother's, sometimes the father's — and keeps it shut for life and in all its descendant cells. The silencing begins when a long non-coding RNA called Xist is made from the chosen X and spreads in a cloud along that whole chromosome, recruiting silencing machinery; DNA methylation then arrives to lock the off-state in for good. It is epigenetics writ at the scale of a whole chromosome.
The honest part: reset, hype, and evidence
Now for the question that fills magazine covers: can what your grandmother ate or endured leave an epigenetic mark that you inherit and pass on — [[molbio-transgenerational-inheritance|transgenerational epigenetic inheritance]]? This is one of the most exciting and most over-claimed ideas in modern biology, and getting it right means being strict about a powerful obstacle. Twice in the mammalian life cycle the genome is deliberately *wiped clean*: once in the germ cells that become eggs and sperm, and again just after fertilization, most epigenetic marks — DNA methylation included — are stripped off and reset. This reprogramming exists precisely to clear each generation's slate. Most marks are designed *not* to be inherited across generations.
There is also a trap that masquerades as inheritance. If a pregnant mother is exposed to something, her fetus is exposed too — and so are *that fetus's own* germ cells, which will become her grandchildren. So an effect seen in children and grandchildren can be plain direct exposure of three generations at once, not anything passed through a heritable mark. Genuine transgenerational inheritance demands a stricter proof: an effect in a generation that was never exposed even as a germ cell — typically the great-grandchildren on the female side — carried by a mark that survived *both* rounds of reprogramming. That is a high bar, and it is rarely cleared in mammals.
So where does the honest line fall? In plants and in simple animals like the worm *C. elegans*, real transgenerational epigenetic inheritance is well documented — they reprogram far less aggressively. In mammals, including humans, a few intriguing cases exist, but rigorous, exposure-free, multi-generation proof is rare, and most headlines about 'inheriting trauma' or 'ancestral diet' run far ahead of the data. The careful summary: epigenetic inheritance across generations is real in some organisms, plausible but largely unproven in mammals, and routinely overstated in the press. Treat dramatic human claims with caution — not because the idea is impossible, but because the bar of proof is high and seldom met.
Where methylation sits among the layers
Two more honest caveats keep methylation in proportion. First, the methyl mark usually *accompanies and reinforces* silencing rather than being the lone first cause — transcription factors and chromatin often make the initial decision, and methylation arrives to lock it in. Second, the rule 'methylation silences' has clear exceptions: in the bodies of actively transcribed genes, and in certain tissues, methylation correlates with *activity*, not silence. As with the histone code, a mark is never magic in itself; it matters only because some reader protein recognizes it and acts, and what that reader does depends on context. Same stamp, different meaning depending on where it lands.
Step back and methylation slots neatly onto the stack you have been building. Transcription factors and enhancers cast the votes; chromatin state opens or closes the booth; and DNA methylation is the part of the answer that gets *remembered and copied* into every daughter cell — the layer that turns a momentary decision into a lasting identity. This is [[molbio-combinatorial-control|combinatorial control]] stretched across time: many inputs converge on whether a gene is on, and the epigenetic marks make the verdict stick. That is how one genome and ~20,000 protein-coding genes build, and keep, a liver cell distinct from a neuron.
And the marks are not frozen. Erasers can strip methylation off, drugs can do the same on purpose — several approved cancer therapies work by removing aberrant methyl tags that had silenced tumour-suppressor genes — and the whole system is reset in the germ line each generation. That reversibility is double-edged: it is why epigenetic medicine is promising, and why permanent inheritance of acquired marks is so hard to achieve. The mark is real, the mechanism is precise, and its limits are exactly what make 'epigenetic' a word to use with care rather than awe.