S8-3

Epigenetics: heritable settings that do not change the sequence

16 min

In S8.2 you built a gene switch out of proteins. A transcription factor lands on a promoter or an enhancer, recruits the machinery, and the gene reads. That whole picture quietly assumed one thing: that the transcription factor could physically reach its binding site. This lesson is about the layer that decides whether it can. That layer is called epigenetics, and it is the reason one genome can run a liver cell and a neuron out of the exact same sequence, and remember which one it is through a lifetime of cell divisions.

The packaging is not just storage

Recall from S6.1 the physical problem: about two meters of DNA has to fit inside a nucleus a few millionths of a meter across. It does that by winding. The DNA spools around little protein cores called histones, a little under two turns per core, and each DNA-plus-histone bead is a nucleosome. Strings of nucleosomes coil and fold again into the bulk material called chromatin.

Here is the beat most beginners miss: that packaging is not inert. How tightly a stretch is wound decides whether anything can be read from it. The transcription factor from S8.2 is a physical object. If its target sequence is buried inside a dense, tightly coiled region, the factor simply cannot dock. The information is still there, letter for letter, but it is behind a locked door.

So there is a control layer sitting above the transcription-factor logic. Loosely packed, accessible chromatin is called euchromatin, and its genes are switchable and often active. Densely packed chromatin is called heterochromatin, and its genes are effectively silenced no matter what factors float by. Epigenetics is the set of chemical marks that push a region toward open or toward closed.

Two families of marks

The marks come in two main kinds, and the crucial fact about both is what they leave untouched.

DNA methylation is a small chemical tag, a methyl group (one carbon with three hydrogens, written CH3), attached directly onto a cytosine base, usually where a C sits right next to a G along the strand. A heavily methylated promoter tends to shut its gene off. Why? The methyl tags do two things: they physically block some transcription factors from binding, and they recruit reader proteins that pull the region into compact, closed chromatin. Methylation is the classic silencing mark.

Histone modifications are chemical tags added not to the DNA but to the tails of the histone proteins the DNA is wound around. Adding an acetyl group tends to loosen the grip between histone and DNA, opening the region up. Certain methyl tags on histones do the opposite and lock it down. Other proteins read these tags and recruit the machines that physically remodel the packing. It is a combinatorial code written on the spool rather than the thread.

Notice what neither kind does. A methylated cytosine is still a cytosine. An acetylated histone still sits over the same sequence. The letters of the genome are not edited. Strip the marks off and the gene reads exactly as it did before.

Set in development, maintained through division, nudged by the environment

Three properties make epigenetics powerful. Take them in order.

Set during development. A fertilized egg has one genome and becomes hundreds of cell types. As a stem cell commits to becoming, say, a neuron, it packs the liver genes and muscle genes it will never need into heterochromatin and marks them silent, while keeping its own neuron genes open. This is the physical answer to the puzzle S8.1 posed. Same DNA everywhere, different accessible subset in each cell type. The cell's identity is largely a pattern of marks.

Maintained through cell division. This is the part that earns the word heritable. When a cell divides, it copies its DNA (the replication of S5). But it also copies its marks. Methylation has a dedicated maintenance mechanism: right after replication the new strand is bare, while the old strand still carries the pattern, and a maintenance enzyme reads the old strand and writes the matching methyl tags onto the new one. So a liver cell divides into two liver cells, each remembering the setting. The memory survives the restart.

Responsive to the environment. Marks are not frozen. Diet, stress, and exposures like tobacco smoke can add or remove methylation at specific spots over time. This is real, and it is measurable, but keep the effect sizes honest: most such shifts are modest and local, not a wholesale rewrite of who you are.

Slide the promoter methylation up on the toggle below and watch the gene fall silent. Before you drag it, predict the shape: is expression a smooth dimmer as methylation rises, or a hard cliff? Then check what the model shows and ask which one biology actually does.

epigenetics.ts

Promoter (CpG-island) methylation represses transcription. Drag the slider: as more CpG sites gain a methyl mark, the gene crosses the silencing threshold and switches OFF.

Promoter methylation20%

MeCGMeCGCGCGCGCGCGCGCGCG

GeneON

Expression level80%

Promoter methylation is low (20% of CpG sites): the promoter stays accessible, so the gene is transcribed (ON).

This is a simplified threshold model

Real silencing is graded and messy. Whether a promoter shuts off depends on which specific CpG sites are methylated, on the histone marks around them, and on the readers and writers that maintain the state, not on one clean cutoff. Here we collapse all of that into a single number: the fraction of promoter CpG sites methylated, crossing a threshold near 50%. The direction is the real biology (promoter hypermethylation represses transcription); the sharp line is a teaching stand-in. The whole switch is the pure function expressionState(methylation, threshold) in lib/bio/epigenetics.ts, which this widget only renders.

The analogy, and where it breaks

Think of epigenetic marks as access-control flags and cached config layered over the source, without editing the source. The DNA sequence is the committed code. A methyl mark on a promoter is like a permission flag set to no-read on that file, or a feature flag flipped to off. Crucially, the config persists across restarts. Because the maintenance enzyme re-writes the pattern after every division, the setting behaves like state written to disk and reloaded, not like a variable lost when the process exits. That single property, config that survives the restart, is what makes the analogy click for the developmental memory.

Use it, but know its failure edge, because an analogy without its limit is a bug.

Then there is the tempting overreach the analogy invites: if config survives restarts, does an individual's acquired settings get passed to their children? This is the Lamarckian idea, that experiences in life rewrite the settings you hand down. Be very careful here.

Lamarckian inheritance in humans is mostly overhyped

Cell-to-cell inheritance through division (mitosis) is solid, that is the maintenance mechanism above. Parent-to-child inheritance of an individual's acquired marks is a different and far weaker claim. Between generations, mammals largely wipe the slate: methylation is erased in two big reprogramming waves, once in the germ cells and again in the early embryo, precisely so each new individual starts from a developmental baseline rather than a copy of a parent's life history. A handful of contested examples survive that erasure partially, and clean transgenerational effects are better documented in plants and worms than in people. So when a headline says your grandmother's famine or your workout habit epigenetically programs your kids, treat it as unproven and usually oversold. The germline reset is the mechanism that blocks it.

Marks drift with age, and that became a clock

Here is the property that turned epigenetics into a longevity tool. Methylation at specific cytosine positions changes with age in a direction that is consistent enough to be predictable. Some sites gain methyl tags as you get older, others lose them, and the drift is regular.

In 2013 Steve Horvath showed you can read the methylation state at a few hundred chosen positions, feed the numbers into a simple weighted model, and estimate a person's chronological age within a few years, across many tissues at once. That model is an epigenetic clock. Later clocks predict not just age but risk, and a person whose epigenetic age runs ahead of their calendar age tends, on average, to face higher mortality.

Forward link: this is a load-bearing tool in the longevity track

Epigenetic clocks are one of the most-used measuring sticks in aging research, and you will meet them again head-on in the LON track. Two honesty notes to carry there. First, a clock is a correlational model, not a proof of cause. It reads a fingerprint that tracks aging, which is not the same as showing that methylation drift drives aging. Do not let the word clock smuggle in causation. Second, the exciting result that partial reprogramming (nudging cells part way back toward a younger mark pattern with the Yamanaka factors) can reset the clock in lab settings is genuine and genuinely early. Doing it safely and durably in a whole living animal, without erasing cell identity or inviting cancer, is an open problem, not a shipped product. Hold the promise and the skepticism at the same time.

The one place the sharp line blurs

The clean rule is that a mark never changes a letter. There is a subtle exception worth knowing because it is where epigenetics and mutation actually touch. A methylated cytosine is chemically more prone to a spontaneous accident that turns it into a thymine, while an unmethylated cytosine that suffers a similar accident becomes a base the repair machinery spots and fixes easily. Over evolutionary time this makes methylated CpG sites genuine mutation hotspots. Read the causality carefully. The mark itself still did not edit the sequence. It only raised the odds that a separate, later chemical event would. The setting and the edit stay distinct, they just happen to interact.

Key terms

Epigenetics: Heritable changes in gene expression driven by chemical marks on DNA or histones, not by changes to the DNA sequence itself.
DNA methylation: A methyl group attached to a cytosine base, usually where a C sits next to a G, that tends to silence the nearby gene.
Histone: A protein core that DNA winds around, whose chemical tags help set how open or closed the surrounding chromatin is.
Nucleosome: One bead of DNA wound around a histone core, the basic repeating unit of chromatin packing.
Chromatin: DNA plus its packaging proteins, existing as open euchromatin that is readable or dense heterochromatin that is silenced.
Maintenance methylation: The mechanism that copies a cell's methylation pattern onto the new DNA strand after replication, so daughter cells inherit the setting.
Epigenetic clock: A model that reads methylation at chosen sites to estimate biological or chronological age, correlational rather than causal.

Check yourself

1. What is the defining difference between an epigenetic mark and a mutation?

2. You add heavy methylation to a gene's promoter. What do you predict happens to that gene's expression?

3. A liver cell divides into two liver cells that keep the same expression pattern. Which mechanism most directly explains that memory?

4. An epigenetic clock reads methylation at a few hundred sites and estimates a person's age within a few years. What is the most accurate reading of that result?

4 unanswered