Memory's gatekeeper: The role of PFC in the encoding of congruent events

Abstract

Theoretical models conventionally portray the consolidation of memories as a slow process that unfolds during sleep. According to the classical Complementary Learning Systems theory, the hippocampus (HPC) rapidly changes its connectivity during wakefulness to encode ongoing events and create memory ensembles that are later transferred to the prefrontal cortex (PFC) during sleep. However, recent experimental studies challenge this notion by showing that new information consistent with prior knowledge can be rapidly consolidated in PFC during wakefulness and that PFC lesions disrupt the encoding of congruent events in the HPC. The contributions of the PFC to memory encoding have therefore largely been overlooked. Moreover, most theoretical frameworks assume random and uncorrelated patterns representing memories, disregarding the correlations between our experiences. To address these shortcomings, we developed a HPC-PFC network model that simulates interactions between the HPC and PFC during the encoding of a memory (awake stage), and subsequent consolidation (sleeping stage) to examine the contributions of each region to the consolidation of novel and congruent memories. Our results show that the PFC network uses stored memory "schemas" consolidated during previous experiences to identify inputs that evoke congruent patterns of activity, quickly integrate it into its network, and gate which components are encoded in the HPC. More specifically, the PFC uses GABAergic long-range projections to inhibit HPC neurons representing input components correlated with a previously stored memory "schema," eliciting sparse hippocampal activity during exposure to congruent events, as it has been experimentally observed.

Keywords: HPC–PFC network; inhibition; memory consolidation; pattern separation; schemas.

Fig. 1.

Encoding and consolidation of a memory in a naive neural network. Top: The HPC–PFC network encodes a pattern A. The network goes through two stages: an awake stage, where the network receives pattern A, and a sleeping stage, where the network evolves autonomously according to its intrinsic dynamics. (A) During the awake state, the hippocampus (HPC) and prefrontal cortex (PFC) network receive a pattern A represented by ones (1; red entries) and minus ones (−1; blue entries), targeting the first 10 neurons in the HPC and PFC network. The recurrent connections are plastic. The two regions are coupled through fixed one-to-one HPC-to-PFC excitatory, and PFC-to-HPC inhibitory connections ($W_{\text{HPC‒PFC}}=0.5,W_{\text{PFC‒HPC}}=-1$). Each circle represents a neuron of the network, with the color and the height of the corresponding bars representing its activity (Top and Bottom, respectively). At the end of the awake stage, the HPC and PFC show the same pattern of activation, with the HPC units more strongly activated (a1), and the HPC connectivity has formed an engram of pattern A (a2). (B) The sleeping stage is characterized by a REM phase, when the two regions are uncoupled ($W_{\text{HPC‒PFC}}=W_{\text{PFC‒HPC}}=0$), and a NREM phase, when the two regions are coupled through excitatory HPC-to-PFC and inhibitory PFC-to-HPC connections ($W_{\text{HPC‒PFC}}=1,W_{\text{PFC‒HPC}}=-1$). During the sleeping stage, the system cycles through the REM and NREM phases seven times. Every time the network enters the REM phase, the HPC and PFC networks are reset to a noisy random state, from which it evolves according to its intrinsic dynamics. In this case, the HPC converges to memory pattern A and the PFC decays to its naive state (b1; first sleep cycle). At the end of the sleeping stage, the memory engram A is consolidated in the PFC connectivity (b2).

Fig. 2.

Consolidation of novel memory pattern relies on hippocampal replay during sleep. Top: Hippocampal and prefrontal cortex activity are analyzed during the awake stage, when the network receives a pattern B whose representation does not overlap with the previously consolidated pattern A (overlap 0%), meaning that it targets a different neural ensemble. Encoding of pattern B happens after consolidation of pattern A in the PFC, and decay of its engram in HPC, i.e., when the recurrent hippocampal connectivity is back to its naive state. (A) During the awake state, the hippocampus (HPC) and prefrontal cortex (PFC) network receive a pattern B targeting 10 HPC and PFC units uncorrelated with the units encoding for pattern A (0% overlap). At the end of the awake stage, the HPC and PFC show the same pattern of activation, with the HPC units more strongly activated (a1). The hippocampal network has encoded pattern B in its connectivity, $W_{\text{HPC}}$, forming a memory engram B. The PFC connectivity, $W_{\text{PFC}}$ remains unaltered, i.e., it only encodes the memory engram A (a2). (B) During the sleeping stage, the HPC network converges to the memory pattern B at during the REM stage (b1, first sleep cycle). At the end of the sleeping stage, the memory engram B is consolidated in the PFC connectivity (b2).

Fig. 3.

Congruent pattern is quickly stored during wakefulness. Top: Hippocampal and prefrontal cortex activity are analyzed during the awake stage, when the network receives a pattern B whose representation overlaps with previously consolidated pattern A by 90%. Encoding of pattern B happens after consolidation of pattern A in the PFC, and decay of its engram in HPC. (A) The HPC network shows sparse activity, while the PFC units targeted by input B are strongly activated (a1). At the end of the awake stage, the HPC network connectivity remains unaltered (i.e., in its naive state). The PFC network, on the other hand, has integrated the uncorrelated components of pattern B in its connectivity with the memory engram A (a2). The circle highlights the nonoverlapping, i.e., uncorrelated, components of pattern B. (B) During the sleeping stage, the HPC network converges to its naive state (b1). At the end of the sleeping stage, the memory engram B is consolidated in the PFC connectivity but not in HPC (b2).

Fig. 4.

Differential roles for HPC and PFC in the encoding of congruent inputs. (A) The HPC–PFC network receives an input B overlapping with memory A by 90%. Coupling between the two regions is only mediated by inhibitory PFC projections ($W_{\text{HPC‒PFC}}=0$, $W_{\text{PFC‒HPC}}=-1$). At the end of the awake stage, the PFC network did not encode for the uncorrelated components of input B in its connectivity, indicating that the input was not consolidated. (B) The HPC–PFC network receives an input B that overlaps with memory A by 90%. Coupling between the two regions is only mediated by excitatory HPC projections ($W_{\text{HPC‒PFC}}=0.5$, $W_{\text{PFC‒HPC}}=0$). At the end of the awake stage, both the hippocampal and PFC network have encoded input B in its connectivity. However, we no longer have the sparse hippocampal activity observed during the encoding of congruent memories (–28).

Fig. 5.

Examining influence of degree of congruentity of new information in plasticity and HPC–PFC network activity. Top: HPC and PFC activity and changes in connectivity are analyzed during the awake stage, when the network receives a pattern B whose representation overlaps with the previously consolidated pattern A by 0 to 90%. (A) Mean absolute changes of the PFC (blue line) and HPC (green line) connections between neurons encoding pattern B, estimated at the end of the awake stage, for patterns B overlapping by 0 to 90% with A. For each degree of overlap, we considered 10 randomly generated patterns B, and the average of the mean connectivity changes obtained for each pattern. (B) Mean absolute activity of all the HPC and PFC neurons (green and blue line, respectively). Once more, for each degree of overlap, we considered 10 randomly generated patterns B.

Fig. 6.

Congruent inputs are linked in PFC whereas uncongruent stimuli exhibit pattern separation. Top: Testing PFC ability to perform pattern separation and memory linking at the end of the awake stage and at the end of the sleeping stage. (A) Testing the ability of PFC to recall pattern B and pattern A at the end of the awake stage. If pattern B overlaps by 10% with pattern A, the PFC will not be able to recall engram B or engram A (a1). The PFC pattern of activation 90 time steps ($t_1$) and 7,600 time steps ($t_2$) after activating 9 out of 10 engram B units is the same (Left and Right, respectively). If pattern B overlaps by 90% with pattern A, activating 3 engram B units results in the recall of both engram A and engram B, indicating that the two patterns are linked (i.e., activation of engram A plus the uncorrelated components of engram B; a2). (B) Testing the PFC ability to recall pattern B and pattern A at the end of the sleeping stage. If pattern B overlaps by 10% with pattern A, activating a subset of engram B units results in recall of engram B but not engram A, indicating pattern separation (b1). If pattern B overlaps by 90%, activation of a subset of engram B units recalls engram A and B, similar to what was observed at the end of the awake stage (b2). (C) Classifying pattern separation and memory linking at the end of the awake stage (Before sleep) and at the end of the sleeping stage (After sleep) for patterns B overlapping by 0 to 90% with pattern A. Pattern separation is defined as recall of engram B without recall of A. Congruent inputs (overlap >40%) are encoded in PFC during awake stage and are linked to previously consolidated overlapping representations. Novel inputs are separated and consolidated during sleep.