A study in mice found that CA1 cells separate different information channels, even when the same cells participate in several memory processes.
One of the most intriguing problems in neuroscience is simple to formulate and difficult to solve: how the brain manages to learn something new without erasing what it has already learned. A new study by researchers at NYU Langone Health suggests a possible mechanism for this. According to the study, conducted in mice and published in Nature, some memory cells in the CA1 region of the hippocampus act as a kind of switchboard: they participate in more than one channel of information, but change their pattern of activity so as not to mix up memories. (NYU Langone Health)
The researchers found that about a quarter of the memory cells in the CA1 region serve as a common junction between signals coming from the CA3 region of the hippocampus and signals going out to the retrosplenial cortex, an area involved in navigation and scene reconstruction. These cells do not necessarily create a “new cell for each memory,” but use different firing patterns to maintain separation between pathways.
Stability and flexibility in the same system
The hippocampus is a brain region central to organizing new experiences into memories. CA3 is thought to provide rapidly changing information, while CA1 is in a position to integrate and pass on information. The question is how such a system can be flexible enough to absorb new information, yet stable enough not to lose previous information.
According to the study’s findings, the possible answer is not a complete separation of cells, but a separation of patterns. Some CA1 cells carry most of the messages coming from CA3. When those cells send information to the retrosplenial cortex, they operate in a different pattern. This allows the brain to use the same neural infrastructure, but maintain separate channels.
The image of a “switchboard” is appropriate here, but it should be used with caution. There is no mechanical switchboard in the brain, of course. What is meant is a network of cells that can transmit many messages without each message overriding the others. Biologically, these are different patterns of neural activity in those cells, not hard switches.
Experiment with moving mice
In the study, six mice were trained to run back and forth on a straight track, with water rewards at either end. As the mice moved, the researchers used densely packed electrodes to record activity from hundreds of neurons simultaneously. At the same time, they tracked the mice's position, matching each pattern of neural activity to actual behavior.
The researchers then examined how signals from the CA3 region pass through CA1 and reach the retrosplenial cortex. In further experiments, they recorded brain activity during sleep. Here, another important point was found: the same CA1 cells that were active in processing information during wakefulness remained active during brain events called sharp-wave ripples, which are thought to be associated with the “re-tuning” of activity patterns and the consolidation of memories.
Why is this important for artificial intelligence?
The researchers note that the finding also has a technological connection. Artificial intelligence systems sometimes suffer from a problem known as catastrophic forgetting: When a system is trained on a new task, it can lose some of the ability it acquired on previous tasks. If the brain can update knowledge without erasing old knowledge, understanding the mechanism may inspire more robust learning architectures.
However, it is important not to turn the research into a technological promise. There is no new, ready-to-use algorithm here, nor is there a proven solution to the problem of forgetting in AI systems. This is a possible biological principle: reusing cells while separating activity patterns. The implication for artificial intelligence is a research idea, not a product.
What is still unknown?
The study was conducted in mice, using a simple pathway and under laboratory conditions. The researchers themselves emphasize that it cannot be directly concluded from it about what is happening in more natural environments or in the human brain. They plan to test whether similar channels exist in other memory circuits.
Despite this limitation, the finding provides an interesting way to think about memory. The brain does not have to choose between retaining the old and absorbing the new. It may be able to do both with the same network, as long as its patterns of operation remain distinct. This is not a “breakthrough” in the simplistic sense of the word, but an important addition to our understanding of the delicate balance between learning and memory.
for the scientific article https://www.nature.com/articles/s41586-026-10481-z
Short FAQ:
Was the research done on humans? No. The study was done in mice, so the conclusions regarding humans are cautious.
What is the CA1 region? A region of the hippocampus that is involved in processing and coordinating information related to memory.
What does this have to do with artificial intelligence? The mechanism may offer inspiration for AI systems that learn new information without losing prior knowledge, but there is no ready-made technological solution here.
More of the topic in Hayadan:
