Memory in the amygdala: neurons that pick what lasts

You look at a hundred photographs in a row. An hour later you recognize some of them instantly and draw a blank on others, though nothing about them seemed different at the time. Somewhere in the second you spent looking, the brain made a decision you were not aware of. Last week, a team at Cedars-Sinai in Los Angeles led by the neuroscientist Ueli Rutishauser, with first author Jiayang Xiao, published in Nature Neuroscience one of the clearest single-cell views yet of where that decision shows up. The surprise is the location: not first in the hippocampus, the structure every textbook names as the gateway to memory, but in the amygdala, a structure the size of a walnut usually filed under fear and emotion.

The amygdala, beyond fear

The amygdala sits deep in each temporal lobe, just in front of the hippocampus. Its reputation comes from decades of work on threat: it reacts to danger, it drives fear conditioning, and it explains why an emotional event is remembered better than a dull one. In that classic picture, the amygdala acts as an amplifier. When something frightening or exciting happens, it boosts the storage processes running in the hippocampus next door, and the memory comes out stronger.

That account leaves out ordinary memory. Most of what anyone remembers is not emotional at all: a face in a hallway, the layout of a kitchen, a photograph of a bridge. If the amygdala only mattered for arousal, it should have little to say about whether a neutral picture of a bridge is kept or lost. The new study tests exactly that.

Listening to single neurons in 112 people

Recording individual neurons in a human brain is possible in one situation. People with epilepsy that resists medication sometimes have thin electrodes implanted for a week or two, so surgeons can locate where their seizures begin. Some of those electrodes carry microwires fine enough to pick up the electrical spikes of single cells. Patients who agree can take part in research tasks while they wait in hospital.

The team pooled data from 112 such patients. In the main task, participants viewed a series of images drawn from a handful of categories, then later took a memory test to see which images they recognized. The first dataset alone yielded recordings from 1,871 neurons across six brain regions, 575 of them in the amygdala.

Two ideas are needed to read the results. A category-selective neuron is a cell that fires mainly for one kind of thing, for example faces, or animals, and stays quiet for the rest. A subsequent memory effect is the comparison at the heart of the design: take the activity recorded while a person first sees an image, and sort it after the fact by whether that image was later remembered or forgotten. Any difference means the activity at the moment of viewing already carried information about the memory’s fate.

What the neurons did

Category-selective cells in the amygdala fired more strongly during the first viewing of images that were later remembered than of images that were later forgotten. The effect was strictly local to each cell’s preference: a cell tuned to one category predicted memory only for images in that category and said nothing about the others. It did not depend on how pleasant, unpleasant or arousing the image was, which is the point that breaks with the emotional-amplifier picture.

Two comparisons make the finding sharper. The ventral temporal cortex, the stretch of visual cortex that recognizes objects and feeds the amygdala, also contains category-selective cells, and they showed no subsequent memory effect. They registered what the image was, not whether it would stick. And in the hippocampus, the effect was weaker and appeared later than in the amygdala. The authors report the amygdala effect was strongest in the right hemisphere.

The last piece is coordination. A local field potential is the slow, summed electrical rhythm of thousands of cells around an electrode, the background hum against which individual spikes occur. Spike-field coherence measures how tightly one neuron’s spikes lock to that rhythm. When an image was later remembered, the spikes of amygdala category cells were more tightly aligned with the hippocampus’s field potentials. The two structures were, in a measurable sense, more in step at the moments that mattered.

The authors report that the core result replicated in two further large datasets, using two different tasks. For a single-neuron study in humans, where each patient contributes a few hours and a limited number of cells, that replication is the strongest part of the work.

Why it matters

The finding reframes the amygdala as something closer to a filter than an alarm. Visual cortex says what an image is. The amygdala, a step later, appears to flag which of those images will be kept, cell by cell and category by category, and then synchronizes with the hippocampus, which does the long-term storing. Emotion would be one input to that filter, not its whole purpose.

There is a clinical motivation too. Rutishauser has suggested that the memory problems of Alzheimer’s disease and temporal lobe epilepsy may come partly from disruption in this amygdala process, not only from damage to the hippocampus. Xiao has proposed that targeted stimulation of these amygdala cells could one day serve as a treatment for memory disorders. Earlier work on brief amygdala stimulation in epilepsy patients has hinted that memory can be nudged this way, so the idea has a basis.

What the study does not show

The study is correlational. Firing more strongly predicted later memory; it does not show that this firing causes the memory. Something upstream, such as momentary attention, could drive both the amygdala response and the encoding without the amygdala doing the choosing. The decisive experiment is causal: stimulate or quiet these cells during viewing and check whether memory for the preferred category rises or falls.

The participants all have epilepsy, and the electrodes sit where clinicians needed them, not where researchers would choose. Medication, seizures and years of abnormal activity may change how these circuits behave. Results from such patients have a good track record of generalizing, but it is an assumption that needs to be stated each time.

The material was visual images in a laboratory session. Rutishauser himself notes that whether the amygdala plays the same role for sounds or smells is untested, and the test came after a short delay. Whether these cells predict what is still remembered weeks later, after sleep has reshuffled and consolidated the day’s traces, is open.

Finally, the treatment idea is a hypothesis. No trial has stimulated these specific neurons to improve memory, and turning a population of category-tuned cells, scattered among others, into a stimulation target is an engineering problem as much as a biological one. The study gives a candidate location. It does not give a therapy.

Further reading

• Stimulus-specific recruitment of human amygdala neurons predicts episodic memory encoding success, the paper, Nature Neuroscience, 2026.
• Research reveals small brain area has a giant role in memory, Cedars-Sinai announcement.
• How the amygdala shapes memory formation, a longer explainer from the same institution.
• The bioRxiv preprint, with the full methods and figures.