Scientists have tracked how the brain switches between "new" and "old" memory

In an experiment with mice equipped with tiny eye-tracking cameras and electrodes in the brain, researchers found that during short "sub-stages" of sleep without rapid eye movement (non-REM sleep), recently acquired information is replayed.

Source: 'Sleep micro-structure organises memory replay', Nature (2024). DOI: 10.1038/s41586-024-08340-w

The main sign is a change in pupil diameter. When the pupil narrows, fresh memories are activated, which the brain "replay" and consolidate. And when the pupil dilates again, the system shifts to updating older memories. This alternating mechanism prevents us from "forgetting everything" when learning something new: otherwise, each new piece of information could overshadow existing knowledge.

The researchers trained a group of mice to perform a series of tasks, such as looking for water or small treats in a maze. The rodents were then attached tiny "spy" cameras in front of their eyes and fitted with electrodes to record brain activity. When the mice fell asleep after learning a new skill, the scientists monitored how their pupils changed and recorded neuronal signals during sleep.

It turned out that flashes of "revision" of memories last for a fraction of a second - about 100 milliseconds. It is at such moments that the brain processes the information received during the day and combines it with the already accumulated experience. At the same time, novelty and past experience are "sorted out" by different time slots: when the pupil is narrowed, there is work with new data, and when it expands, old memories are included.

This discovery showed that sleep is more complex than previously thought. Mice, like humans, experience more fractional phases of non-REM sleep, during which the brain alternately processes new and old knowledge. If this brief but important period for consolidation is interfered with, memories (whether new skills or old information) may consolidate worse.

The findings will help not only in creating methods to improve memory in humans, but also in developing more effective learning algorithms for artificial neural networks. After all, the principle of "new-old-new-old" allows to optimise the process and not to lose the accumulated information while continuing to learn new things.