The authors draw together the results of a series of detailed computational studies and show how they are contributing to the development of a theory of hippocampal function. A new part of the theory introduced here is a quantitative analysis of how backprojections from the hippocampus to the neocortex could lead to the recall of recent memories. The theory is then compared with other theories of hippocampal function. First, what is computed by the hippocampus is considered. The hypothesis the authors advocate, on the basis of the effects of damage to the hippocampus and neuronal activity recorded in it, is that it is involved in the formation of new memories by acting as an intermediate-term buffer store for information about episodes, particularly for spatial, but probably also for some nonspatial, information. The authors analyze how the hippocampus could perform this function, by producing a computational theory of how it operates, based on neuroanatomical and neurophysiological information about the different neuronal systems contained within the hippocampus. Key hypotheses are that the CA3 pyramidal cells operate as a single autoassociation network to store new episodic information as it arrives via a number of specialized preprocessing stages from many association areas of the cerebral cortex, and that the dentate granule cell/mossy fiber system is important, particularly during learning, to help to produce a new pattern of firing in the CA3 cells for each episode. The computational analysis shows how many memories could be stored in the hippocampus and how quickly the CA3 autoassociation system would operate during recall. The analysis is then extended to show how the CA3 system could be used to recall a whole episodic memory when only a fragment of it is presented. It is shown how this recall could operate using modified synapses in backprojection pathways from the hippocampus to the cerebral neocortex, resulting in reinstatement of neuronal activity in association areas of the cerebral neocortex similar to that present during the original episode. The recalled information in the cerebral neocortex could then be used by the neocortex in the formation of long-term memories.

A fundamental property of many associative memory networks is the ability to decorrelate overlapping input patterns before information is stored. In the hippocampus, this neuronal pattern separation is expressed as the tendency of ensembles of place cells to undergo extensive 'remapping' in response to changes in the sensory or motivational inputs to the hippocampus. Remapping is expressed under some conditions as a change of firing rates in the presence of a stable place code ('rate remapping'), and under other conditions as a complete reorganization of the hippocampal place code in which both place and rate of firing take statistically independent values ('global remapping'). Here we show that the nature of hippocampal remapping can be predicted by ensemble dynamics in place-selective grid cells in the medial entorhinal cortex, one synapse upstream of the hippocampus. Whereas rate remapping is associated with stable grid fields, global remapping is always accompanied by a coordinate shift in the firing vertices of the grid cells. Grid fields of co-localized medial entorhinal cortex cells move and rotate in concert during this realignment. In contrast to the multiple environment-specific representations coded by place cells in the hippocampus, local ensembles of grid cells thus maintain a constant spatial phase structure, allowing position to be represented and updated by the same translation mechanism in all environments encountered by the animal.

The hippocampus has differentiated into an extensively connected recurrent stage (CA3) followed by a feed-forward stage (CA1). We examined the function of this structural differentiation by determining how cell ensembles in rat CA3 and CA1 generate representations of rooms with common spatial elements. In CA3, distinct subsets of pyramidal cells were activated in each room, regardless of the similarity of the testing enclosure. In CA1, the activated populations overlapped, and the overlap increased in similar enclosures. After exposure to a novel room, ensemble activity developed slower in CA3 than CA1, suggesting that the representations emerged independently.