Mouse memories survive extensive synapse loss

Mice retained two learned memories after passing through artificial hibernation in which more than half of the synapses measured in their hippocampus disappeared. The study, published in Science on August 13, also recorded an approximately 70 percent decline in neuronal firing during the low-metabolism state.¹ ²

The animals still recognized a setting associated with an unpleasant stimulus and continued to locate a reward in a spatial task. Anatomy supplied a possible explanation: small clusters of synapses connecting neurons tied to the same memory were more likely to survive the remodeling. The authors found an association between those clusters and retention, not proof that the clusters caused it.¹ ²

The experiment uses mice prepared to enter a hibernation-like state and tests hippocampus-dependent memories. It does not place memory somewhere outside synapses or offer a way to preserve human recollections. The results indicate that persistence may depend on network layout as well as the strength or size of any one contact.

Hibernation became a controlled disruption

Mice do not undergo the prolonged seasonal hibernation seen in marmots or ground squirrels. In 2020, University of Tsukuba researchers identified a hypothalamic circuit that could induce a sustained period of low body temperature and metabolism in rodents. Their bodies continued regulating the state, and they recovered without external warming.⁴

Yu-Ju Lin, Kazumasa Tanaka, and their collaborators turned that circuit into an experimental tool. The mice learned their tasks before entering artificial hibernation. Electrical recordings tracked neural activity, while imaging and behavioral measurements compared conditions around the intervention.¹ ²

The hippocampus participates in forming and retrieving contextual and spatial memories. Neurons communicate at synapses. On the receiving side, small protrusions called dendritic spines change in size and number with activity. Frequently used connections can strengthen; quieter ones may shrink or disappear.

That plasticity makes learning possible and creates a problem for long-lived memory. Spines and even the cells contributing to a representation change over time, while recall remains available. Artificial hibernation accelerated the remodeling and let the researchers look for structure that stayed stable as many individual contacts vanished.

Spine size did not predict survival

A conventional account predicts greater stability for larger spines, which are associated with synapses strengthened during learning. Hibernation eliminated both large and small spines. Individual volume did not clearly separate the contacts that endured from those that disappeared.¹ ²

Behavioral tests showed that both memories were retained. Trained mice froze in the environment paired with the unpleasant stimulus and distinguished it from a neutral context. In the spatial task, they continued choosing the route to a reward. Dryad holds the data points behind those analyses, organized by animal and experimental condition.³

Recall was tested after the low-metabolism state ended, as connections were already returning. Two days after artificial hibernation, synaptic density had recovered, and more than 80 percent of lost spines reappeared in the same locations, according to the paper's results as described by Science News.¹ ⁵ The memory trace therefore survived a period of extensive removal and reconstruction. The experiment did not test behavior while half the measured synapses were absent.

Those tasks capture specific kinds of memory. They do not measure language, autobiography, or decades of experience. The study also uses separate groups and methods for synaptic density, electrical activity, and behavior rather than one sample that represents the whole brain.

The “more than half” figure has the same boundary. It refers to synapses observed in the regions and conditions examined, not half of every connection in a mouse brain.

Microscopy exposed an architecture that endured

The researchers labeled neurons that were active during learning to identify the engramA set of brain cells and connections involved in encoding and retrieving a particular memory. associated with a memory. They then combined fluorescence microscopy, which locates labeled cells, with electron microscopy, which resolves contacts far below the first method's scale.¹ ²

Two patterns stood out. One was the multisynaptic boutonThe transmitting end of a neuron that forms synapses with more than one dendritic spine., a transmitting terminal connected to more than one receiving spine. The other was a concentration of several engram synapses along nearby sections of a dendrite. These clustered contacts were overrepresented among the structures preserved through artificial hibernation.¹

Such a pattern could preserve relationships among cells while individual contacts turn over. A network may be able to lose redundant elements without erasing its more informative organization, and resistant connections could guide later rebuilding. That interpretation fits the observations; the experiment does not show the brain intentionally using redundancy as a backup system.

Electron microscopy adds another constraint. It captures fixed tissue at an endpoint and cannot watch the same synapse operate inside a living animal. The team aligned labels and reconstructions across samples to describe the resulting architecture. Numerical data for the figures and analysis scripts are in Dryad, while the large raw imaging, electrophysiology, and electron microscopy files were omitted because of their size.³

Synapse loss alone did not preserve recall

A comparison group underwent prolonged anesthesia and received a compound that blocks spine growth and stabilization. That treatment also caused extensive synapse loss. Memory retrieval deteriorated, and engram clusters were not preserved in the same way seen after artificial hibernation.¹ ⁵

The comparison indicates that a broad reduction in activity and connections was not sufficient to protect recall. Some feature of the induced state, or of the structure that survived it, accompanied the later performance.

Causality remains the missing experiment. Researchers would have to selectively remove the preserved clusters, or protect them under another condition, and then measure recall. Lin acknowledged that gap in OIST's account of the work.²

Sources

1. Artificial hibernation reveals synaptic engram architecture associated with memory retention · Science · https://doi.org/10.1126/science.aee7004 · Aug. 13, 2026
2. Artificial hibernation reveals secrets of long-term memory · Okinawa Institute of Science and Technology · https://www.oist.jp/news-center/news/2026/8/13/artificial-hibernation-reveals-secrets-long-term-memory · Aug. 13, 2026

Show 4 more sourcesHide sources

3. Data: Artificial hibernation reveals synaptic engram architecture associated with memory retention · Dryad · https://datadryad.org/dataset/doi:10.5061/dryad.m0cfxpphz · June 30, 2026
4. A discrete neuronal circuit induces a hibernation-like state in rodents · Nature · https://www.nature.com/articles/s41586-020-2163-6 · June 11, 2020
5. How memories may survive hibernation · Science News · https://www.sciencenews.org/article/memories-survive-hibernation-brain-cell · Aug. 26, 2026
6. Putting mice into hibernation causes a major loss of synapses · Ars Technica · https://arstechnica.com/science/2026/08/memories-stick-around-even-after-half-the-synapses-are-gone/ · Aug. 22, 2026