Hierarchical reorganisation of the hippocampal code during learning

The hippocampal representation of space can change on a time scale of minutes as a result of goal learning. However, the link between behaviour performance and the representation of other variables in the hippocampus is less clear, especially at longer time scales. To investigate this, we recorded the activity of hippocampal CA1 neurons of rats using in vivo electrophysiology, as they learned a space-context association task over the course of 5 days. To assess the representation of different task-related variables such as position, reward, movement direction or context in the population activity, we used linear decoders and principal component analysis. We observed that, over days, the hippocampal code underwent a reorganisation of population activity in principal component (PC) space. Despite position explaining the highest variance at all times, we found that decoding context from high-variance PCs improved with learning; this happened specifically at positions where it mattered for task performance and resulted in fewer dimensions being necessary for decoding. Variables less relevant to task performance (e.g. movement direction) did not show this position specificity nor a reduction in dimensionality with learning. Altogether, our results show that non-spatial variables unequally contribute to the variance in hippocampal activity; their contribution depends on position and adapts to behavioural demands during learning. Downstream from the hippocampus, this enables better decoding of environmental features learned to be relevant for behaviour. This suggests that hippocampal representations may play a role supporting behaviour not only over minutes or hours, but also over many days of learning.