NYU scientists discover a brain "switch" that keeps old memories while forming new ones

04.06.2026 | Science and discoveries

A team at NYU Langone Health has discovered a small group of neurons in the hippocampus that act as a biological "memory switch," allowing for the creation of new memories without erasing old ones, with potential implications for treating dementias and developing AI.

Снимка от Jean-Etienne Minh-Duy Poirrier from USA, Wikimedia Commons (CC BY-SA 2.0)

Scientists from "NYU Langone Health" reported that they have discovered a small subset of neurons in the hippocampus that act as a biological "switch," allowing the brain to form new memories without overwriting existing ones. The study results were published in the journal "Nature" on May 13, 2026, and offer a new explanation for one of the oldest mysteries in neuroscience – how the brain manages to be both plastic and stable.

A cell core that "multiplexes" memory

The research shows that about 25% of neurons in the "CA1" region of the hippocampus act as common hub points. They receive fast flows of information from the adjacent "CA3" region and transmit them to the retrosplenial cortex – a part of the brain involved in spatial navigation and memory recall.

A characteristic feature of these "hub neurons" is that they use different activation patterns for incoming and outgoing signals, thereby forming separate communication channels within the same structure – similar to an electronic switch that routes several conversations along different lines without mixing them.

The brain does not look for new cells for every memory

"Instead of triggering new neurons for every new experience, the brain alters the activation patterns of a stable cell core, effectively organizing information and protecting already encoded memories," explains Dr. "Joaquin Gonzalez," a postdoctoral fellow and one of the lead authors of the study.

This approach allows the same group of cells to serve multiple memories, maintaining a balance between memory flexibility and the stability of already accumulated information.

Active during sleep as well

The researchers also found that these same hub neurons in "CA1" remain active during sleep, when they participate in so-called "sharp-wave ripples" – brief bursts of neural activity long associated with memory consolidation.

Because the same group of cells is responsible for both daytime information processing and nighttime replay, the signal transmission channel from the hippocampus to the cortex remains open and supports the strengthening of long-term memories.

Experiment with mice and high-density electrodes

The experiment involved six mice trained to navigate a linear track. While the animals move, high-density electrode arrays simultaneously record the activity of hundreds of individual neurons in several connected brain regions.

This approach allows scientists to track in real time how signals move between the "CA3," "CA1," and the retrosplenial cortex and how the specific subgroup of neurons acts as a "distribution center" for memory.

Implications for understanding memory disorders

Dr. "Gyorgy Buzsáki," one of the lead authors of the study, suggests that the discovered mechanisms could shed light on the early stages of memory impairment in neurodegenerative diseases, such as Alzheimer's disease.

"The 'memory switch' we discovered in the hippocampus may provide key insights for understanding the early mechanisms of memory loss in such diseases," he states, emphasizing that dysfunction in this system could disrupt both the formation and preservation of memories.

Connection to artificial intelligence and "catastrophic forgetting"

The authors note that the results also have potential significance for the development of artificial intelligence. Existing systems often suffer from so-called "catastrophic forgetting" – the loss of previously learned information when training on new tasks.

Understanding how the brain preserves old memories while simultaneously forming new ones could serve as a basis for creating more resilient AI architectures that "multiplex" knowledge in a way similar to this biological "memory switch."