Can the same information remain readable as neural activity changes?

If the combination of neurons responding to the same thing changes between yesterday and today, can the brain still treat it as the same thing?

Neuroscientists use the term representational drift for changes over time in activity patterns associated with the same stimulus or task. This is intriguing if we think of memory or recognition as fixed activity in a particular set of cells. If the way information is represented changes, what allows the receiving circuit to keep reading its meaning?

In a study published on October 6, researchers used flexible electrodes to follow neurons in the mouse visual cortex and hippocampus over several months. A challenge in long-term recording is that movement between an electrode and tissue can result in different cells being measured. By reducing this problem, the researchers investigated changes in the responses of the same cells. Original paper ↗

Even when mice repeatedly watched the same movie, neural activity changed across days. However, when the researchers represented the activity of many cells using a smaller number of coordinates and aligned those coordinates between recording days, they could improve the decoding of temporal information—identifying which part of the movie had elicited a response.

Individual cells did not need to respond identically for relationships within the population activity to remain alignable across days.

Importantly, this was not a case of taking the readout from the first day and using it unchanged indefinitely. The published analysis code uses responses from both an earlier day and a new day, aligning them by corresponding time points in the same movie. Information was recoverable not by ignoring the change, but by adjusting the readout to it. Authors' analysis code ↗

That distinction matters when considering how the brain works.

One possibility is that the receiving circuit gradually relearns how to read its input as the source changes. In this account, consistency in memory or recognition depends less on a fixed state at a particular location than on maintaining correspondence between circuits.

Another possibility is that the changes are not uniform. Some components of activity may change substantially while distinctions needed for behaviour remain intact. A change may look prominent in the cells we observe or the analysis we use, even if the information the brain actually relies on has changed much less.

These are not the same explanation. The first places importance on updating the reader. The second asks where the relationships that remain usable are preserved.

In the experimental analysis, replaying the same movie provided a basis for aligning activity on two days. What common reference might the brain itself use? Repeated experiences, the consequences of actions and rewards are possible candidates. If both the representation of information and its readout change, some form of learning is needed to keep them from diverging.

In our article on artificial hibernation and memory ↗, we asked what must remain for a memory to count as the same memory after a substantial change in the state of the nervous system. The present work lets us reconsider that question not only in an unusual state, but in terms of using the same information amid everyday changes in activity.

The information decoded here concerned a stimulus; the study did not directly demonstrate a mechanism for preserving long-term memory. Even so, it helps clarify what to investigate next: not just whether a cell responds as it did before, but how distinctions within a population are maintained and how the circuits using them keep up.

When a memory appears to have been lost, the loss of the information itself would mean something different from a breakdown in correspondence with the circuit reading it. To understand what remains stable in a changing brain, we can look beyond the states of individual cells to the relationships through which information is passed.