Interactions Between Hippocampal Areas CA3 and CA1 During Slow-Wave Sleep

Abstract

Many lines of evidence suggest that the hippocampus plays a critical role in memory formation. The predominant hypothesis is that new memories are initially stored within hippocampal circuits during awake behavior, and are subsequently consolidated across neocortical networks under the influence of hippocampal activity during sleep. The hippocampal memory trace is conjectured to reside within the recurrent circuits of area CA3, which is believed to function as an autoassociative memory. Area CA3 projects almost exclusively to area CA1, one of the major output stages of the hippocampus. How is CA3 activity transformed in CA1, and what is the function of the CA1 subfield that intermediates between CA3 and the neocortex, the presumed long-term storage site of memories? Here we characterize the relationships between CA3 and CA1 activity during slow-wave sleep (SWS), a stage of sleep conjectured to be important in memory consolidation. Activity in SWS is marked by the presence of short-lived (~100 ms) population bursts that are believed to be spontaneously generated within CA3 and that cooccur with high-frequency oscillations (~200 Hz ripples) in area CA1. We demonstrate that: 1. CA1 amplifies transient increases in CA3 activity levels, while attenuating background fluctuations. 2. The fraction of co-active neurons is higher in CA1 than in CA3, while the firing intensity of active neurons is higher in CA3 than in CA1. 3. The above dichotomy is particularly pronounced during the population bursts associated with ripples. 4. In comparison to isolated spikes, bursts of action potentials by CA3 neurons are particularly effective at triggering large CA1 responses and predicting the onset of CA1 ripples. These results show that CA1 acts as a selective filter and amplifier of CA3 activity patterns, and that bursting of individual CA3 neurons plays a special role in this CA3-CA1 transformation. We hypothesize that coordinated bursts in CA3 reflect convergence to attractors, each representing a stored pattern in the auto-associative network. Our observations suggest that these stored patterns are preferentially amplified by CA1 and transmitted to downstream targets, while activity representing intermediate states in-between attractors are less likely to be transmitted.