Rapid spatial cognition in mice, with and without neocortex and hippocampus

A groundbreaking study reveals that mice can learn, remember, and generalize quickly, even without the involvement of the neocortex and hippocampus. Traditionally, rapid cognition has been attributed to these brain regions, but their precise role remains uncertain. To investigate, researchers observed mice navigating a reconfigurable 3D maze called the Manhattan Maze.

Initially naive wildtype mice demonstrated remarkable ability, improving within just two rewards, adopting efficient paths after about 20 trials, retaining a nine-turn route overnight, and learning faster in new configurations.

Much of this rapid learning appears to stem from a rule-based forward bias that emerges before any reward is given. Intriguingly, in a maze with loops where that rule is less relevant, mice quickly learned to prefer one specific bottleneck corridor far from the reward location while maintaining flexibility in other areas. To explore the connection between these cognitive abilities and brain function, researchers tested mutant mice lacking the hippocampus and most of the neocortex.

These mice initially struggled, exhibiting repetitive scanning that made them about three times slower to obtain the first few rewards. However, once past this initial exploration stage, their learning, retention over weeks to months, and generalization to new mazes remained largely intact.

The swift learning observed here defies traditional trial and error models, which rely on backward value propagation from the reward. Instead, a neuromorphic circuit model provides a more plausible explanation. This model enables the mice to build a map of the environment without reward, and crucially, it does not require cortical or hippocampal circuit motifs. The authors conclude that structure learned before the first reward, rather than the reward itself, may be the key to facilitating few-shot learning.