REPRESENTATIONAL DRIFT IN HIPPOCAMPAL CA1 IS GEOMETRICALLY PRESERVED

The phenomenon of representational drift (i.e., changing neuronal tuning despite consistent behaviour), is a fundamental paradox in neuroscience that raises the question how stable behaviour can emerge from unstable neural representations. Place cells in CA1 of the hippocampus are crucial for spatial navigation and memories but show gradual changes in their preferred firing location over time when exposed to the same environment. This drift occurs despite the fact that the animal maintains the ability to navigate and to perform spatial tasks, suggesting a complex relationship between neural activity and behaviour. Here, we show that while the spatial tuning functions of individual CA1 neurons drift over time, stable space coding prevails at the level of population geometry. Representational drift is not random but can be described as a translation and rotation in population state space, which preserves the internal geometry of population activity over time. Compensating for the coordinated translation and rotation allows for drift correction and recovery of spatial tuning on future days. Moreover, the preserved internal geometry stabilises downstream readout under noisy conditions. We propose that the conserved population geometry might serve as a mechanism by which downstream reader networks achieve effective drift correction and, thus, ameliorate the readout of stable information.