epicme эпигеном · цитозин · метилирование
Field notes · methylation series

One molecular mark, three kinds of memory

CpG methylation turns up wherever a cell needs to remember something — a synapse that fired, an antigen it fought off, a body that once went through something hard. The same chemistry, a methyl group on a cytosine before a guanine, running three completely different memory systems.

neuron · synapse
T cell · antigen
HPA axis · stress
filled, orange — methylated open, blue — unmethylated
Scope: everything here happens inside one lifetime, in somatic cells. Whether any of it crosses into the next generation is a separate question, on a separate page.
01 · Cognitive memory

How a synapse writes something down

Methylation was long treated as a developmental lock — set early, then static for the life of the cell. That view does not survive contact with the adult hippocampus, where marks are added and removed within hours of a learning event.

Quotes are short excerpts in the original English, one per source, each linked to the paper. Everything else is a paraphrase.

◆ Mouse Dnmt1 · Dnmt3a

Dnmt1 and Dnmt3a maintain DNA methylation and regulate synaptic function in adult forebrain neurons

Feng, Zhou, Campbell, Le, Li, Sweatt, Silva & Fan · Nature Neuroscience 2010

FoundDeleting either methyltransferase alone in forebrain neurons left learning largely intact. Deleting both impaired spatial learning and memory and weakened the late phase of long-term potentiation at Schaffer collateral synapses — the first clear demonstration that the adult brain needs an active methylation machinery to work, not just to develop.
nature.com · 10.1038/nn.2514 ↗
◆ Mouse Dnmt3a2 · ageing

Rescue of aging-associated decline in Dnmt3a2 expression restores cognitive abilities

Oliveira, Hemstedt & Bading · Nature Neuroscience 2012 · 15:1111–1113

“Dnmt3a2 is an activity-regulated immediate early gene”— Oliveira et al., Nat Neurosci 2012

FoundOf the two DNMT3A isoforms, Dnmt3a2 switches on within minutes of neuronal activity, like Fos or Egr1. Hippocampal levels of it fall with age — and restoring them in old mice reversed learning and memory deficits, while lowering them in young mice impaired memory formation. The enzyme is not background maintenance; its abundance sets cognitive capacity.
CaveatA viral overexpression rescue in mice shows sufficiency in that system, not that declining Dnmt3a2 is the cause of human cognitive ageing.
nature.com · 10.1038/nn.3151 ↗
◆ Mouse · neurons Topo IIβ

Activity-induced DNA breaks govern the expression of neuronal early-response genes

Madabhushi, Gao, Pfenning et al. (Tsai lab) · Cell 2015 · 161(7):1592–1605

“DSB formation is a physiological event that rapidly resolves topological constraints”— Madabhushi et al., Cell 2015

FoundStimulating neurons triggers deliberate double-strand breaks in the promoters of early-response genes including Fos, Npas4 and Egr1. Cutting those promoters on purpose switches the genes on with no external stimulus at all, and the breaks depend on topoisomerase IIβ. Writing an experience down involves physically unknotting the DNA at specific promoters.
CaveatThis paper is about topology, not methylation — the mechanism runs through Topo IIβ. It sits here because it is the same set of loci, on the same timescale, and because it shows how far the genome will go to make a transient signal durable. Treat the connection to demethylation as an open question, not a demonstrated pathway.
cell.com · 10.1016/j.cell.2015.05.032 ↗

Two enzyme families, two directions, at once

DNMT3A adds marks while the TET enzymes (TET1–3) oxidise 5-methylcytosine back toward an unmethylated state. Learning is not methylation going up; it is both families working simultaneously at different loci. That also means hippocampal methylation looks like a short-lived working step in consolidation — a scratchpad — while the more permanent copy is thought to shift to cortex over time. It is one molecular echo of why very old memories survive hippocampal damage and recent ones do not.

02 · Immune memory

The memory your T cells already have

This is the version closest to my own work. A naive CD8⁺ T cell that meets its antigen rewrites large parts of its methylome on the way to becoming an effector — and what happens next depends entirely on whether the antigen ever goes away.

naive

Unwritten

PDCD1 methylated, effector loci closed. Nothing has happened yet.

PD-1 locus: methylated
memory

Filed away

The antigen cleared. Effector-phase demethylation gets reversed — the cell goes back and re-methylates. Poised, not active.

PD-1 locus: re-methylated
exhausted

Stuck open

The antigen never cleared. Demethylation locks in and the inhibitory programme becomes constitutive.

PD-1 locus: unmethylated
● Mouse + human PDCD1 / Pdcd1

Chronic virus infection enforces demethylation of the locus that encodes PD-1 in antigen-specific CD8⁺ T cells

Youngblood, Oestreich, Ha et al. (Ahmed lab) · Immunity 2011 · 35:400–412 · acute vs chronic LCMV, plus human virus-specific cells

“Exhausted virus-specific CD8+ T cells retain an unmethylated Pdcd1 regulatory region”— Youngblood et al., Immunity 2011

FoundDemethylation of the PD-1 locus tracks the strength and duration of T-cell receptor signalling during the effector phase, in both fates. Only cells that go on to become memory re-methylate it. Exhausted cells keep it open even after virus titres fall — the mark, not the antigen, becomes the record.
cell.com · 10.1016/j.immuni.2011.06.015 ↗
● Mouse DNMT3A cKO WGBS

De novo epigenetic programs inhibit PD-1 blockade-mediated T cell rejuvenation

Ghoneim, Fan, Moustaki et al. (Youngblood lab) · Cell 2017 · 170(1):142–157

“fully exhausted during prolonged antigen exposure remain refractory to ICB-mediated rejuvenation”— Ghoneim et al., Cell 2017

FoundWhole-genome bisulfite sequencing across the effector-to-exhaustion transition found heritable de novo methylation programmes that limit how far a T cell can expand under PD-1 blockade. Deleting Dnmt3a, or pre-treating with the hypomethylating agent decitabine, let cells keep effector function through chronic stimulation. This is why checkpoint blockade alone often only partly rescues exhausted cells: you can block the signalling, but the methylation programme underneath is still there.
CaveatMouse infection and tumour models. Decitabine is genome-wide, not targeted — the observed benefit does not tell you which of the thousands of affected loci matter.
cell.com · 10.1016/j.cell.2017.06.007 ↗

The same lever I pull on CD55

The machinery behind CRISPRoff and ZFPoff — a KRAB repressor fused to a DNMT domain — is mechanistically the same lever a T cell pulls on itself while differentiating. The donor T cells in my Day 6 and Day 35 experiments already carry this history in their methylome before I touch them. Editing CD55 is writing one deliberate mark into a genome that has been writing its own for years.

03 · Biological embedding

When an experience becomes a cell

The most interesting and the most contested corner of the field. The claim is not that hardship changes your DNA — the sequence is untouched. It is that psychologically significant experience shifts the methylation state of specific stress-regulatory genes, measurably, decades later.

▲ Human FKBP5 intron 7 rs1360780

Allele-specific FKBP5 DNA demethylation mediates gene–childhood trauma interactions

Klengel, Mehta, Anacker et al. (Binder lab) · Nature Neuroscience 2013 · 16(1):33–41

“allele-specific, childhood trauma-dependent DNA demethylation”— Klengel et al., Nat Neurosci 2013

FoundFKBP5 is a co-chaperone that damps down glucocorticoid receptor signalling — part of the brake that shuts a cortisol spike back off. Its intron 7 carries glucocorticoid response elements, and demethylation there was found only in carriers of a particular risk variant who had also experienced childhood trauma. Less methylation means more FKBP5, which means a weaker brake on the stress response, present decades after the exposure. Genotype and environment are not additive here; the mark appears where both are present.
CaveatThe mark is not necessarily permanent — intervention studies have reported FKBP5 methylation shifting alongside symptom improvement, which cuts both ways: it makes the mark look responsive rather than fixed, and it makes causal direction harder to pin down.
nature.com · 10.1038/nn.3275 ↗
▲ Human · postmortem NR3C1 NGFI-A site

Epigenetic regulation of the glucocorticoid receptor in human brain associates with childhood abuse

McGowan, Sasaki, D'Alessio, Dymov, Labonté, Szyf, Turecki & Meaney · Nature Neuroscience 2009 · 12(3):342–348 · postmortem hippocampus

“increased cytosine methylation of an NR3C1 promoter”— McGowan et al., Nat Neurosci 2009

FoundThe human counterpart of Meaney's rat maternal-care work. Hippocampus from suicide victims with a history of childhood abuse showed lower glucocorticoid receptor mRNA and more methylation at the neuron-specific NR3C1 promoter than hippocampus from suicide victims without that history, or from controls. Constructs methylated to mimic that state bound less NGFI-A and drove less transcription — the same transcription factor site as in the rat.
CaveatPostmortem, cross-sectional, small, and childhood abuse is confounded with everything else that accompanies it. What the paper establishes cleanly is that the rat finding has a human analogue at the same site — not that abuse caused the methylation.
PMC2944040 · 10.1038/nn.2270 ↗

Reading blood as a proxy for brain

The caveat researchers keep repeating: tissue matters. Whole-blood methylation is a proxy, not a readout of cortex or hippocampus, and for most loci the correspondence between the two is weak. FKBP5 is treated as one of the more defensible cases, which is part of why it keeps appearing in this literature — but a blood measurement is a hypothesis about the brain, not a measurement of it.

Worth holding onto

Most of this literature is correlational and cross-sectional. Effect sizes at individual CpGs are typically small, findings often fail to replicate cleanly across cohorts, and causal direction is genuinely open: does adversity cause the methylation change, does baseline methylation shape vulnerability to adversity, or are both downstream of something else — genotype, earlier stress, current symptoms? Biological embedding is an active research programme, not settled fact at the level of any single gene.

04 · The common logic

Same mark, same trick, three systems

All three reuse one move: make a transient state durable by writing it into methylation, at a specific locus, in a specific cell type. What differs is how long the writing lasts.

System What gets written down Where Durability
Hippocampal neuron A moment of synaptic activity Fos · Npas4 · Egr1 · IEG promoters Hours to days — a scratchpad, rewritten constantly
Memory T cell An antigen that was cleared PDCD1 and effector loci Years — deliberately reversed, then held poised
Exhausted T cell An antigen that never cleared PDCD1 regulatory region Locked — survives antigen clearance and PD-1 blockade
HPA axis A period of adversity, often in childhood FKBP5 intron 7 · NR3C1 exon 1F Decades — but apparently still movable

The open question worth sitting with

What decides whether a mark stays reversible or locks? Hippocampal consolidation is reversible by design. FKBP5 seems to move under therapy. Exhausted-cell PD-1 does not move even when the antigen is gone and the checkpoint is blocked. That is not a neat parallel — it is a mechanistic question about chromatin stability, and it is one my own data speaks to directly: CRISPRoff writes a mark at CD55 on purpose, and nanopore then measures, molecule by molecule, how long it survives. Day 6 versus Day 35 is a durability experiment.

Related page

Does any of this reach the next generation? →

Everything above happens in somatic cells within one lifetime. FKBP5 and NR3C1 appear on both pages, but the question there is different: whether a mark survives the germline and shows up in a child who was never exposed. Short answer so far: for methylation, not demonstrated.
05 · Sources

Papers referenced on this page

  1. 1

    Dnmt1 and Dnmt3a maintain DNA methylation and regulate synaptic function in adult forebrain neurons

    Feng J, Zhou Y, Campbell SL, Le T, Li E, Sweatt JD, Silva AJ, Fan G · Nature Neuroscience 2010;13(4):423–430 · 10.1038/nn.2514
  2. 2

    Rescue of aging-associated decline in Dnmt3a2 expression restores cognitive abilities

    Oliveira AMM, Hemstedt TJ, Bading H · Nature Neuroscience 2012;15(8):1111–1113 · 10.1038/nn.3151
  3. 3

    Dnmt3a2: a hub for enhancing cognitive functions

    Oliveira AMM, Hemstedt TJ, Freitag HE, Bading H · Molecular Psychiatry 2016;21(8):1130–1136 · 10.1038/mp.2015.175
  4. 4

    Activity-induced DNA breaks govern the expression of neuronal early-response genes

    Madabhushi R, Gao F, Pfenning AR, et al., Kellis M, Tsai LH · Cell 2015;161(7):1592–1605 · 10.1016/j.cell.2015.05.032
  5. 5

    Chronic virus infection enforces demethylation of the locus that encodes PD-1 in antigen-specific CD8⁺ T cells

    Youngblood B, Oestreich KJ, Ha SJ, et al., Boss JM, Ahmed R · Immunity 2011;35(3):400–412 · 10.1016/j.immuni.2011.06.015
  6. 6

    De novo epigenetic programs inhibit PD-1 blockade-mediated T cell rejuvenation

    Ghoneim HE, Fan Y, Moustaki A, et al., Thomas PG, Youngblood B · Cell 2017;170(1):142–157.e19 · 10.1016/j.cell.2017.06.007
  7. 7

    Allele-specific FKBP5 DNA demethylation mediates gene–childhood trauma interactions

    Klengel T, Mehta D, Anacker C, et al., Rein T, Binder EB · Nature Neuroscience 2013;16(1):33–41 · 10.1038/nn.3275
  8. 8

    Epigenetic regulation of the glucocorticoid receptor in human brain associates with childhood abuse

    McGowan PO, Sasaki A, D'Alessio AC, Dymov S, Labonté B, Szyf M, Turecki G, Meaney MJ · Nature Neuroscience 2009;12(3):342–348 · PMC2944040 (open access)
  9. 9

    Synaptic control of DNA methylation involves activity-dependent degradation of DNMT3A1 in the nucleus

    Bayraktar G, Yuanxiang P, Confettura AD, et al., Kreutz MR · Neuropsychopharmacology 2020;45(12):2120–2130 · 10.1038/s41386-020-0780-2

Last checked August 2026. Open-access links are marked; the rest lead to the publisher's page.