Coordinated reactivation of hippocampal cell assemblies during learning
Loren M. Frank and Sen Cheng
W.M. Keck Center for Integrative Neuroscience and Department of Physiology, University of California, San Francisco, CA 94143-0444

The hippocampus is essential for the acquisition of new memories for places and events. During learning the hippocampal network undergoes rapid plastic changes that establish new representations in hippocampal cell assemblies. For more permanent storage of memories, newly formed hippocampal representations are thought to be consolidated into neocortex through coordinated reactivation of cell assemblies during ripples. However, no direct link has been established between ripple reactivation and memory formation. Here we examine dynamic changes in spatial activity and reactivation of hippocampal neurons that represent novel or familiar locations. In our experiment four rats performed an alternation task in a maze containing one novel and one familiar arm. Animals clearly distinguished between the novel and the familiar arm throughout days 1-3 of novel exposure, running more slowly in the novel arm. To study the formation of spatial memories, we compared the neural representations of the novel and familiar arms in simultaneously recorded CA1 neurons. Hence, we selected neurons with place fields in the novel and familiar arms (novel arm cells and familiar arm cells, respectively).

We found that novel arm cells were initially much more likely to be reactivated (i.e., to fire at least one spike) during any given ripple than were familiar arm cells on days 1 and 2. These differences were no longer significant on day 3. For example, on day 1 novel arm cells were about two and a half times more likely to be reactivated than their familiar arm counterparts. Since individual novel arm cells were initially more active during ripples, pairs of novel arm cell should have been more co-active than their familiar arm counterparts. Indeed, on day 1 novel arm pairs were about six times more likely to be co-active than familiar arm cell pairs. This significant co-activity remained across all three days of exposure to the novel arm. Furthermore, reactivation of novel arm cell pairs was initially more coordinated: pairs were more co-active than would be expected if the two cells were firing independently during ripples.

We also measured the width (root-mean-square (RMS) time lag) of the CCG between neuron pairs including only spikes that occurred within ripples. Novel arm cell pairs were significantly more precisely coordinated than familiar arm cell pairs on days 1 and 2, but not on day 3. Novel arm cell pairs were therefore more frequently co-active in a window compatible with synaptic plasticity. This highly precise reactivation of novel arm cells was present throughout the environment, and could not be explained by proximity to the novel arm.

If, as has been proposed, reactivation is a simple replay of experience, then the level of coordinated reactivation during ripples should reflect the level of coordinated spatial activity during experience. We therefore examined the evolution of spatio-temporal structure in single cells and cell pairs. Surprisingly, reactivation was strongest and most coordinated when spatial activity was least coordinated, and reactivation became weaker as place cells came to express reliable spatio-temporal organization during exploration. These results demonstrate that coordinated reactivation of cell assemblies is strongest in novel environments where the hippocampus must form new representations and weaker in familiar environments where a stable representation is already present. Ripples originate in hippocampal area CA3, and we propose that one important role for reactivation is to replay recently learned correlations into CA1, allowing for off-line learning within the hippocampus following experience.