The research team, led by Jiayang Xiao, Jonathan Daume and Ueli Rutishauser, recorded single-neuron activity from 42 patients undergoing intracranial monitoring for drug-resistant epilepsy at Cedars-Sinai Medical Center, Toronto Western Hospital and Johns Hopkins Hospital. In the first task, patients performed a working memory exercise in which 140 novel pictures were shown, each belonging to one of five visual categories. After a delay of 10 to 30 minutes, they took a surprise recognition test with 400 images — 200 previously seen and 200 new — indicating for each whether it was old or new, together with a confidence rating on a six-point scale. Participants performed well, with average recognition accuracy of 70.3 percent, and their confidence ratings tracked their accuracy, indicating that they had formed genuine declarative memories of the images.

Across the recordings, the team isolated 1,871 well-isolated single neurons distributed across the amygdala, hippocampus, anterior cingulate cortex, pre-supplementary motor area, ventromedial prefrontal cortex and ventral temporal cortex. Using a statistical comparison of firing rates across the five picture categories, they identified neurons that responded selectively to one preferred category. Such category cells were found above chance levels in the ventral temporal cortex, where they made up 67.8 percent of recorded neurons, as well as in the amygdala (40.0 percent), hippocampus (27.2 percent) and ventromedial prefrontal cortex (21.3 percent). In the amygdala, these cells most often preferred images of people, food and animals.

The key result emerged when the researchers sorted encoding trials according to what happened later. Pictures that the patient subsequently recognized as old were labeled ‘later remembered’, and those judged new were ‘later forgotten’. During the encoding period, amygdala category cells fired significantly more for later remembered items than for forgotten ones — but only for pictures in the cell’s preferred category. For the preferred category, the effect was highly significant across 196 neurons, whereas for nonpreferred categories it was absent entirely. A mixed-effects model confirmed that firing rate significantly predicted later memory accuracy for preferred trials, and population decoding of remembered versus forgotten items was more accurate for preferred than nonpreferred categories.

Timing proved to be a striking feature of the effect. Amygdala category cells differentiated between later remembered and forgotten items of their preferred category starting around 350 milliseconds after stimulus onset, a window that overlaps almost exactly with the cells’ sensory response to the image. In the hippocampus, a similar subsequent memory effect appeared but began much later, around 1,300 milliseconds, and was significantly weaker in magnitude than in the amygdala. Notably, category cells in the ventral temporal cortex — the region that supplies the amygdala with its visual input — showed no subsequent memory effect at all, despite having the strongest and most prevalent category responses of any recorded area. This pattern suggests that memory-related modulation emerges within the medial temporal lobe rather than in the visual cortex itself.

Could the effect simply reflect emotion rather than memory? Because emotionally charged images tend to be better remembered, the team collected valence and arousal ratings for every image from an independent online sample and added these as predictors in their statistical models. The memory effect on amygdala firing remained robust after accounting for both emotional dimensions, and neither valence nor arousal significantly predicted firing rate on its own. The effect also held separately for positively and negatively valenced images, and for both high- and low-arousal images. The amygdala signal, in other words, tracked memory encoding itself rather than the emotional flavor of the pictures.

The strength of the neural response also tracked the strength of the resulting memory. Amygdala category cells fired more during encoding of images later remembered with high confidence than those remembered with low confidence, producing a graded pattern: highest firing for high-confidence memories, intermediate for low-confidence memories and lowest for forgotten items. Using a signal detection model fitted to the behavioral data, the researchers separated recollection — the episodic component of memory — from familiarity. The difference in amygdala responses between high- and low-confidence remembered trials correlated significantly with the recollection parameter but not with familiarity, indicating that these neurons are linked specifically to the formation of memories with a stronger episodic component.

To test whether the finding was specific to one task, the team repeated the analysis in a second, independent dataset of 5,744 neurons recorded from 95 participants performing a classic recognition memory task. The replication confirmed the core result: amygdala category cells fired more for later remembered items, the effect survived statistical control for emotional valence and arousal, and firing rates were higher for images remembered with high confidence. In this dataset, the hippocampus again showed no significant relationship between firing and later memory. Pooling both tasks also revealed that the subsequent memory effect was significant in both hemispheres but significantly stronger in the right amygdala, an asymmetry the authors note may relate to the use of visual, nonverbal stimuli and that they caution requires replication in other tasks and datasets.

Why would stronger firing produce stronger memories? One possibility the researchers investigated is that well-timed coordination between neuronal firing and brain oscillations supports the synaptic plasticity underlying memory. They found that spike-field coherence — the alignment of spike timing with local field potentials — was stronger within the amygdala during encoding of later remembered items in the 9 to 18 hertz range. More strikingly, spike timing of amygdala category cells synchronized more closely with hippocampal field activity during successful encoding, with a peak difference around 10.5 hertz, and hippocampal neurons likewise synchronized more with amygdala activity during remembered trials. This interregional coordination was specific to preferred stimuli, supporting the idea that episodic memories depend on the coordinated activation of assemblies of memory-content-selective cells.

The authors suggest their findings reframe the amygdala’s role in memory. Rather than merely boosting memory for emotionally salient events, the amygdala appears to engage stimulus-specific circuits at the very moment an experience unfolds, helping determine which single exposure becomes an episodic memory. Because the amygdala is affected early in Alzheimer’s disease and its volume has been linked to memory performance, the researchers propose that disrupted encoding within the amygdala may contribute to memory impairments in disorders such as Alzheimer’s disease and temporal lobe epilepsy. They point to category cells in the right amygdala as a potential target for future interventions aimed at treating disorders affecting episodic memory, while emphasizing that this possibility rests on the present correlational findings in human patients rather than on demonstrated therapeutic benefit.

Subject of Research: Single-neuron recordings in the human amygdala predicting episodic memory encoding success

Article Title: Stimulus-specific recruitment of human amygdala neurons predicts episodic memory encoding success

Article References: Xiao, J., Daume, J., Salimpour, Y., Kurilenko, N., Mosher, C., Anderson, W. S., Valiante, T. A., Mamelak, A. N., & Rutishauser, U. (2026). Stimulus-specific recruitment of human amygdala neurons predicts episodic memory encoding success. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02493-9

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02493-9

Keywords: amygdala, episodic memory, memory encoding, hippocampus, single-neuron recording, category-selective neurons, spike-field coherence, recognition memory, confidence, right hemisphere, neurosurgical patients, Nature Neuroscience