A DIVISION OF LABOUR ACROSS FRONTAL–ENTORHINAL CIRCUITS FOR MAPPING SPACE AND BEHAVIOUR

To behave flexibly you have to know both the structure of the world and the structure of the problem you are trying to solve. Here, we test for neural signatures of two algorithms that can support this: a code where the two structures are factorised and then added at decision time, versus one where the two are multiplied and hence all possibilies are precomputed. We recorded neuronal activity in freely moving mice from 1000s of neurons in three brain regions implicated in abstraction: medial frontal cortex (mFC), medial entorhinal cortex (MEC) and lateral entorhinal cortex (LEC) of mice performing the ABCD task, which orthogonalises world and task structure. MEC primarily signalled current location; goal-progress tuning there was weak. LEC showed strong, task-general goal-progress signals that tiled the 0–100% progress range towards a single goal. mFC represented both space and progress, and organised into modules whose activity tracked the full four-state sequence relative to individual behavioural steps. Across regions we found no over-representation of absolute ABCD-state–preserving neurons and no globally-coherent population rotation, arguing against factorisation. Pairwise coherence in EC was limited to similarly tuned cells (consistent with spatial tuning), whereas mFC exhibited ring-like coherent rotations spanning many between-neuron angles. These findings favour the non-factorised code in mFC over factorisation and point towards a functional division of labour between association cortices: MEC = where you are now; LEC = progress to a single goal; mFC = space + progress, plus sequence-structured dynamics tracking progress along multiple goals.