What is an engram? How memories are stored
How researchers tag the cells that hold a memory, what changes inside them, and why a stored trace can still be updated.
Answer to the question
What is an engram? How memories are stored
An engram is the lasting physical change an experience leaves in a specific group of brain cells, so that reactivating those cells brings the memory back. Richard Semon proposed the idea in 1904. In mice, researchers can now tag the cells active during learning and switch them on with light to bring back the learned behaviour.
Key takeaways
- Richard Semon proposed in 1904 that an experience activates cells, leaves a lasting change in them, and that reactivating them later drives recall; the modern engram concept traces to that proposal (Josselyn and Tonegawa, 2020).
- In mice, amygdala cells active during fear learning were reactivated at retrieval, and more reactivated cells went with a stronger fear response (Reijmers and colleagues, 2007).
- Light shone on tagged dentate gyrus cells made mice freeze in a different setting, and three control conditions did not (Liu and colleagues, 2012).
- Robert Sapolsky's Behave describes long-term potentiation as more receptors on the spine, more sensitive receptors and stronger release from the sending side, with new synapses forming over weeks.
- A 2025 review argues that engram representations are more dynamic and flexible than previously thought, taking in updating, reconsolidation and forgetting (Zaki and Cai).
In 2012, light switched on over a small group of tagged brain cells made mice freeze in a different setting from the one where they had learned fear.
The authors read the freezing as a recall of the fear memory triggered by light, and the cells they lit up are an example of what neuroscientists call an engram: the lasting physical change that an experience leaves in a particular set of brain cells. Where such a change sits, and what it is made of, has been argued over for more than a century.
What does engram mean?
Where is a memory when nobody is thinking about it? Richard Semon proposed an answer in 1904: an experience activates cells, the cells undergo a lasting change, and reactivating them later drives recall. Sheena Josselyn and Susumu Tonegawa trace the modern engram concept to that proposal in a 2020 review. It covers how new tools now let researchers image and manipulate the cells involved, and it reports evidence that engrams depend on the excitability of neurons and on synaptic plasticity.
The harder question was always where. Karl Lashley's 1950 paper, In Search of the Engram, drew together 33 years of his research on memory and the brain. Darryl Bruce's later history of the paper reports that Lashley argued memories are spread across functional areas of the cortex and are not kept in one spot, and that a memory trace is not an isolated link between an input and an output. The paper carried such authority that this view became deeply embedded in neuropsychology.
Robert Sapolsky's Behave meets the same question when it asks where a learned fear is kept. He describes the hippocampus as the place where short-term knowledge becomes long-term memory, with the trace itself most likely held in the cortex, and he treats the amygdala's role as a store, and not only a way in, as a long-running debate in the field. The word engram therefore splits in two. One part is the claim that a memory has a physical form that persists when nothing is being recalled. The other is a claim about location, which has moved from a single spot to distributed networks and, in recent mouse work, to sparse groups of cells inside them.
How are engram cells found?
The first move is to mark the cells that are busy during learning. Leon Reijmers and colleagues asked whether learning and recall switch on the same neurons, and whether the number of reactivated neurons tracks the strength of the memory. To answer, they built mice in which recently active neurons could be tagged for a long time afterwards. Basolateral amygdala cells that were active during Pavlovian fear conditioning turned up again when the mice later retrieved the memory, and the more cells were reactivated, the stronger the fear response.
That match between learning and retrieval, and its link to the strength of the behaviour, made the tagged group look like a stable correlate of the memory. The authors present it as a correlate and note that the method opens genetic access to the molecular mechanisms that encode memory in distributed networks. Its value is practical. A vague phrase, the cells that hold a memory, becomes a countable set that can be compared across animals and across tests.
Timing is what makes the tag useful. A cell marked because it was active during learning is a cell with a claim on the memory, and the same cell can then be looked for at retrieval days later. Without that mark, activity at recall would be a sea of cells with no way to tell which had taken part in the original event.
The region matters too. In Behave, Sapolsky lists fear conditioning among the forms of learning that rest on long-term potentiation in the basolateral amygdala, so the cells that Reijmers and colleagues tagged sit where the cellular account points.
Can light recall a memory?
A correlation leaves a gap, and Xu Liu and colleagues closed part of it in 2012. They tagged dentate gyrus neurons that were active during fear learning with a light-sensitive protein, then shone light on those cells in a different setting. The mice froze more with the light on.
Three control conditions did not produce freezing: mice that carried the protein in a similar number of cells but had not been fear conditioned, conditioned mice whose cells carried a marker protein in place of the light-sensitive one, and mice whose labelled cells came from a context unconnected to fear. Each control removes a different alternative explanation, and the unconnected context shows that the freezing depends on which cells were tagged and not on tagging in general.
The authors concluded that activating a sparse, specific group of hippocampal neurons that contributes to a memory trace is enough to recall that memory. Enough is a narrower word than only: a group of cells that can bring a behaviour back need not be the only place the memory lives.
Steve Ramirez and colleagues used the same toolkit in 2013 to ask a different question. They labelled dentate gyrus or CA1 cells that were active in one chamber, then reactivated them with light while the mice received a foot shock in another chamber. Back in the first chamber, where no shock had ever been given, the dentate gyrus group froze more. The authors described this as a context-specific false memory that engaged downstream regions similar to those used in natural fear memory recall, and concluded that a fear memory, represented internally and expressed in behaviour, can be created artificially.
What changes inside cells?
Behave gives the cellular detail through long-term potentiation, which Sapolsky describes as a lasting increase in how strongly a synapse responds after it has been driven hard. When a long train of signals arrives, calcium enters through a receptor that normally stays quiet, and three things follow.
- More copies of the receptor are pulled onto the membrane of the receiving spine.
- Receptors already in place become more sensitive.
- A signal travels backward across the synapse and raises how much the sending side releases.
Over a longer span, Sapolsky adds, a new spine can appear on the dendrite, a sending terminal settles beside it, and across weeks the pair becomes a working synapse, a process tied to the same calcium signal. He notes that individual receptor copies last only days, and that how the change carries over to their replacements is left open in his account. He is also explicit that long-term potentiation is not confined to the hippocampus, which is why the same account can cover the amygdala and the frontal cortex.
Josselyn and Tonegawa's two named ingredients line up with this account. Excitability describes how readily a neuron responds, and Behave describes long-term potentiation as a rise in the excitability of the synapse itself, while synaptic plasticity is the wider capacity of connections to change with use. A group of cells that fired together during learning and then changed in both ways would be easy to find again, and hard to confuse with its neighbours.
This is the physical meaning of a lasting change, and it is the same capacity discussed in Is Neuroplasticity a Selling Point, where the strongest human evidence is a structural change in the hippocampus after years of training. Within the engram idea, strengthened synapses among a tagged group of cells are what make the group a trace and not just a set of neighbours. The step in which a fragile short-term memory becomes a durable one is called memory consolidation, and these are among its physical changes.
Is a memory ever fixed?
Yosif Zaki and Denise Cai's review, published in 2025, covers how memories are encoded, consolidated and recalled in sparse groups of neurons. They argue that engram representations are more dynamic and flexible than previously thought, and they take in updating, reconsolidation, forgetting, schema learning, the linking of memories and representational drift.
Three of the review's terms are worth separating.
- Updating is the revision of a stored memory in the light of new information.
- Reconsolidation names the idea that a recalled memory is briefly open to change before it settles again.
- Representational drift names the slow shift in which cells carry the same memory over time.
Each describes a way the physical trace can stay in the brain while its details move, which is a different picture from a recording that is filed once and replayed.
That flexibility matches what people see in daily life. A trip is summed up afterwards from how it felt, with the ending weighing heavily, as Why You Remember the End of a Holiday sets out. The pull of unfinished stories, taken up in The Zeigarnik Effect at the Credits, concerns what we go back for. The engram idea works one level below both, at the cells that hold whatever is recalled.
In our view, the sturdy claim is narrower than the headline version. In mice, a specific set of cells can be marked at learning, found again at recall, and made to bring a behaviour back with light. That a human memory is stored the same way is the natural extension, and the studies above did not test it. The 2025 review adds the wider point. An engram is kept, and it is revised.
Common questions
What is an engram? How memories are stored
An engram is the lasting physical change an experience leaves in a specific group of brain cells. On this view a stored memory is a pattern of strengthened connections among those cells, and recall is their reactivation.
Where does the idea of the engram come from?
Richard Semon proposed it in 1904, and Karl Lashley's 1950 paper, In Search of the Engram, summed up 33 years of his research on where memories sit. Josselyn and Tonegawa's 2020 review traces the modern concept to Semon.
Have engrams been tagged in people?
The labelling and light studies described here were run in mice. They show that a sparse group of tagged cells can be reactivated to bring back a learned behaviour, and they do not test the same method in people.
Is a memory kept in one place in the brain?
Lashley argued that memories are spread across functional cortical areas. Later work tagged sparse groups of hippocampal and amygdala cells, and Sapolsky describes the trace as most likely held in the cortex, with the hippocampus as the route in.
Sources
Sapolsky, R. Behave. Seed book; vault hub Books/Behave.
Josselyn, S. A., & Tonegawa, S. (2020). Memory engrams: Recalling the past and imagining the future. Science, 367(6473), eaaw4325. doi:10.1126/science.aaw4325 https://doi.org/10.1126/science.aaw4325
Bruce, D. (2001). Fifty years since Lashley's In Search of the Engram: Refutations and conjectures. Journal of the History of the Neurosciences, 10(3), 308-318. doi:10.1076/jhin.10.3.308.9086 https://doi.org/10.1076/jhin.10.3.308.9086
Reijmers, L. G., Perkins, B. L., Matsuo, N., & Mayford, M. (2007). Localization of a stable neural correlate of associative memory. Science, 317(5842), 1230-1233. doi:10.1126/science.1143839 https://doi.org/10.1126/science.1143839
Liu, X., Ramirez, S., Pang, P. T., Puryear, C. B., Govindarajan, A., Deisseroth, K., & Tonegawa, S. (2012). Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature, 484(7394), 381-385. doi:10.1038/nature11028 https://doi.org/10.1038/nature11028
Ramirez, S., Liu, X., Lin, P.-A., Suh, J., Pignatelli, M., Redondo, R. L., Ryan, T. J., & Tonegawa, S. (2013). Creating a false memory in the hippocampus. Science, 341(6144), 387-391. doi:10.1126/science.1239073 https://doi.org/10.1126/science.1239073
Zaki, Y., & Cai, D. J. (2025). Memory engram stability and flexibility. Neuropsychopharmacology, 50(1), 285-293. doi:10.1038/s41386-024-01979-z https://doi.org/10.1038/s41386-024-01979-z