The idea that some neural representations can “drift,” or change over time, even in the seeming absence of learning, is broadly accepted. But characterizing the phenomenon across the brain has proved challenging.
“The interesting part is what exactly seems to be stable and what exactly seems to be drifting. That’s not an easy question,” says Tobias Rose, a group leader at the University of Bonn Medical Center, who presented findings on drift in the mouse primary visual cortex earlier this month at the Computational and Systems Neuroscience (COSYNE) annual meeting.
Other new research adds nuance to the discussion: Neurons that code for head direction in the mouse post-subiculum show little drift, retaining their tuning for multiple weeks, according to a study published last month in Nature. And they differ from hippocampal place cells, which are also part of the spatial navigation system but have highly variable responses, as reported in previous research.
The new findings raise questions about how stable and flexible representations interact in the brain, given that signals from the post-subiculum ultimately feed into the hippocampus, says Rose, who was not involved in the work. “It’s a rather important study,” he says.
The relative stability of head direction cell tuning does not invalidate previous reports of drift elsewhere in the brain, says Adrien Peyrache, associate professor at the Montreal Neurological Institute, who led the head direction study. Instead, it may be that these invariant responses act as a “rigid backbone” onto which more flexible sensory and cognitive responses can be mapped, he says. “I find it reassuring.”
Still, the low drift reported in the new work may be partially due to the study’s methods, which eliminated cells that lost their response from one day to the next, says Timothy O’Leary, professor of information engineering and neuroscience at the University of Cambridge, who was not involved in the work.
Though the methods are valid, “I wouldn’t say that this paper demonstrates absence of drift,” he says. “If people just read the title and don’t read the paper, they reach the wrong conclusion.”
H
“It showed us that the system is hardwired somehow, at some level—meaning that it’s not just sensory inputs that shape the representation in this case,” Peyrache says.
That hardwired response persists for a given environment over time, Peyrache and his colleagues found in their latest study. When mice explored a circular enclosure five days a week for four weeks, head direction cells maintained stable tuning, both in terms of the pairwise relationship between neurons’ responses and the preferred direction that they coded for, calcium imaging revealed. That tuning also persisted, though with some drift, even after the animal explored different environments for up to six weeks and then returned to the circular enclosure, the team found.
When mice explored each of the four environments once a week for four weeks, an individual cell’s preferred direction did show additional drift, Peyrache and his colleagues found. That drift may stem from the increased cognitive load on the animal, Peyrache says. Still, pairs of head direction cells maintained the same relative relationship across all experiments.
If a mouse was placed in a new environment just once, its head direction cells oriented to a new preferred direction—and that orientation remained stable in that environment when tested four weeks later, the team found.
“It’s kind of an instant plasticity,” Rose says. “But then it remains rigid.”
P
On top of that, some studies may observe drifting representations because other factors shape cells’ activity across time. For example, drift in a mouse’s primary visual cortex may simply reflect changes in the animal’s behavior or internal state, according to a 2022 paper. That drift may also be due to changes in the animal’s gaze location, according to the work Rose and his colleagues presented last week at COSYNE.
“To be able to understand what happens exactly in the context of representational stability or representational drift, you need to really describe the code in a multidimensional space,” says Sadra Sadeh, a group leader at King’s College London and the Francis Crick Institute and an investigator on the 2022 paper. Only then can you truly separate the signal from the noise, he says.
That requires researchers to collect increasing amounts of data during an experiment: orofacial movements, pupil size, auditory input and body position from all directions, Rose says. “That’s the classic problem we have now in systems neurosciences: Effectively, what do you stop measuring? Do you have to measure bowel movements to be able to see what’s the cognitive variable?”
The neural response observed may also depend on the encoding variable being studied, says Michael Goard, associate professor of neuroscience at the University of California, Santa Barbara, who was not involved in the work. Because the head direction system seems to be based on an attractor, cells that have similar tuning could reinforce each other through local excitation, he says.
“I suspect different circuit architectures lend themselves to having more stable activity or less stable activity—both across areas and even within an area,” Goard says. “I think there’s really a continuum of how much neurons will change responses over time.”
Parts of the brain that more closely represent the world, such as head direction cells and primary visual neurons, may require responses that are more hard-coded, whereas other parts of the brain can be more flexible, Sadeh says. “Ultimately, we need to develop some computational models of the brain to be able to understand how we need to adjust our representations to be able to cope with the dynamic nature.”
