Efficient coding of future position in a brain area important for navigation

Prediction of future states is a fundamental computation in the brain and may be key to navigation in the absence of external landmarks. Neurons in the hippocampus are known to encode an animal's current position as well as future movements, but whether prediction is optimal is unknown. We analyzed large-scale recordings from the mouse hippocampus during navigation on a one-dimensional circular track (1,485 neurons; 33 ms bins). Using mutual information, we quantified how much information hippocampal spike patterns convey about the animal’s future position and velocity, and compared these values to theoretical limits defined by the information bottleneck. Both individual neurons and small ensembles operate close to these limits, suggesting a near-optimal representation of forthcoming spatial states. Moreover, highly correlated groups exhibit coordinated prediction, jointly carrying more predictive information than random groups. Population-level analyses further revealed that slow-varying principal components of the collective activity preserve high predictive information over extended timescales. Together, these results identify predictive information as a key organizing principle of hippocampal population codes and demonstrate that hippocampal neurons efficiently compress past activity to retain information relevant for future position and motion.