COMPUTING HUBS IN THE HIPPOCAMPUS AND CORTEX

<strong>Neural computation, which relies on the active storage and sharing of information, occurs within large neuron networks in the highly dynamic context of varying brain states. Whether such functions are performed by specific </strong><strong>sub</strong><strong>sets of neurons and whether they occur in specific dynamical regimes remain</strong><strong>s</strong><strong> poorly understood. Using high density recordings in the hippocampus, medial entorhinal and medial prefrontal cortex of the rat, we identify computing </strong><strong>sub</strong><strong>states, or </strong><strong>discrete </strong><strong>epochs, in which specific computing hub neurons perform well defined storage and sharing operations in a brain state-dependent manner. We retrieve a multiplicity of distinct computing </strong><strong>sub</strong><strong>states within each global brain state, such as REM and nonREM sleep. Half of recorded neurons act as computing hubs in at least one </strong><strong>sub</strong><strong>state, suggesting that functional roles are not firmly hardwired but dynamically reassigned at the second timescale. We identify sequences of </strong><strong>sub</strong><strong>states whose temporal organization is dynamic and stands between order and disorder. We propose that global brain states constrain the language of neuronal computations by regulating the syntactic complexity of these </strong><strong>sub</strong><strong>state sequences.</strong>