Hippocampal CA3 output is crucial for ripple-associated reactivation and consolidation of memory

Abstract

A widely held memory consolidation theory posits that memory of events and space is initially stored in the hippocampus (HPC) in a time-limited manner and is consolidated in the neocortex for permanent storage. Although posttraining HPC lesions result in temporally graded amnesia, the precise HPC circuits and mechanisms involved in remote memory storage remain poorly understood. To investigate the role of the trisynaptic pathway in the consolidation process we employed the CA3-TeTX transgenic mouse, in which CA3 output can be specifically and inducibly controlled. We found that posttraining blockade of CA3 output for up to 4 weeks impairs the consolidation of contextual fear memory. Moreover, in vivo hippocampal recordings revealed a reduced intrinsic frequency of CA1 ripples and a significant decrease in the experience-dependent, ripple-associated coordinated reactivation of CA1 cell pairs. Collectively, these results suggest that the posttraining integrity of the trisynaptic pathway and the ripple-associated reactivation of hippocampal memory engram are crucial for memory consolidation.

Figure 1. Consolidation of contextual, but not cued fear memory, is impaired in CA3-TeTX mice

(A) Averaged freezing levels during the recent contextual memory test following 3 weeks Dox withdrawal (Control; blue bar, Mutant; red bar, n = 12 per genotype). Mutants exhibited significantly less freezing (P < 0.05). (B) Averaged freezing levels during the contextual and cued memory tests. Dox was withdrawn 1 week before training (−1 week) and recent and remote memories were tested (n = 24 per genotype). Mutant mice displayed significantly less freezing in the remote contextual memory test compared to the control littermates (P < 0.05), but not in the recent contextual memory test. In cued memory tests, mutants and controls showed similar freezing levels in both recent and remote memory tests. (C) Averaged freezing levels during the contextual and cued memory tests. Dox was withdrawn three weeks after training (+3 weeks). Mutants exhibited no deficit in either recent or remote contextual or cued memory. Protocols for Dox⁺ to Dox⁻ diet switches relative to the timing of training and memory tests are indicated at the top of each panel (Figure 1A to C). (D) Freezing levels of individual mice during remote memory tests were divided by those during the remote memory tests (retention index) for contextual fear conditioning (Control; blue circle, Mutant; red circle, P <0.05 for the −1 week protocol). Data shown as mean ± s.e.m.

Figure 2. Loss of CA3 output to CA1 results in a decrease in the intrinsic frequency of ripple oscillations

(A) LFP samples from Control (top) and Mutant (bottom) mice during SQ periods. Detected ripples are highlighted in either blue (Control) or red (Mutant). Below the LFP traces are individual ripple examples from six randomly selected Controls (blue) and Mutants (red). Note the clear decrease in the Mutant ripple intrinsic frequency. (B) The amplitude of ripples is not altered in Mutants, however, there was a significant decrease in the intrinsic frequency of Mutant (N=12) ripples compared to Controls (N=14; P = 4.31e−19). (C and D) Both phase preference and probability of firing of pyramidal cells (C) were consistent between Mutants and Controls. Although the phase preference of interneurons did not change in the Mutants, the probability of firing was significantly lower in the Mutants (D) (P = 0.006). Data shown as mean ± s.e.m.

Figure 3. Blockade of CA3 to CA1 transmission results in a loss of reactivation during post-behavior SQ periods

(A and B) Representative examples of Control (left) and Mutant (right) cell pair correlations during pre-run and post-run SQ periods. Heat maps (right of graphs) illustrate the spatial overlap between the two cells in the novel environment. Note that although the place fields of Mutant cells had significant overlap, little or no increase in their pair-wise correlation was observed during post-SQ periods. (C and D) Representative examples of cell pair correlations for cells with non-overlapping place fields. Note the lack of pair-wise correlation increase during post-SQ periodsfor both Controls and Mutants. (E) Quantification of all pair-wise correlations during ripple epochs(Control: 91 cells from 14 mice, overlapping pairs=98, non-overlapping pairs=229; Mutant: 94 cells from 12 mice, overlapping pairs=132, non-overlapping pairs=215). Both genotypes show a significant increase in correlation when their place fields overlap and when taking all ripples into account. Note the significant reduction in Δ correlation (post-SQ periodscorrelation – pre-SQ periodscorrelation) in Mutant (red line) animals compared to Controls (blue line two-way analysis of variance, genotype × overlap F(1,670=3.96, P=0.047; Bonferroni post-tests: Control non-overlapping × overlapping, P<0.001; CA3-TeTX non-overlapping × overlapping, P<0.01; Control overlapping × CA3-TeTX overlapping, P<0.01). (F) Evaluation of reactivation during ripples only falling into the average Control ripple frequency. When Mutant ripples were in the average Control ripple frequency range, the deficit in Δ correlation was not observed. Data shown as mean ± s.e.m.