Hippocampal signatures of awake targeted memory reactivation

Abstract

Dominant theories of episodic memory propose that a key mechanism of memory consolidation is replay-a process, whereby neural patterns of activation during learning are reinstated during offline post-learning periods. Here, we tested whether key signatures of replay defined by studies in rodents, such as recapitulation of specific memory traces, as well as sequences, are apparent in humans during post-encoding memory reactivation. Thirty participants underwent functional imaging that consisted of interleaved encoding and rest periods. During an offline period of wakeful rest, we biased reactivation towards some memories by presenting sound cues that had previously been associated with particular stimulus sequences. Results showed that targeted hippocampal reactivation was biased towards cued memory sequences and that reactivation signatures preserved the temporal order of particular sequences. Importantly, the biased reactivation was related to differences in subsequent memory, suggesting that preferential reactivation may be a mechanism by which specific memory traces can be strengthened for enhanced subsequent memory retrieval.

Keywords: Hippocampus; Memory consolidation; Memory reactivation; Representational similarity analysis.

Fig. 1. Experimental design.

MRI scanning occurred during the tasks within the shaded grey box. Order of the study sets (Study Set A vs. Study Set B) and rest sound types (cued vs. novel sounds) were counterbalanced across participants. Each encoding trial began with a fixation cross, followed by serial presentation of the three items in a triplet with a concurrent sound cue, followed by an odd/even judgment task. To index reactivation, multi-voxel patterns of activation were extracted from the hippocampus, and pattern similarity was computed between hippocampal activation elicited by a particular triplet at encoding and rest

Fig. 2. Schematic of between-triplet pattern similarity computations.

Unique triplet sequences were encoded across repeated presentations (left). Measures of between-triplet pattern similarity were computed for triplet presentations occurring across the cued encoding and cued rest periods. Between-triplet pattern similarity (right) was measured by computing pairwise correlations between hippocampal activation patterns associated with each unique triplet sequence across a given presentation time-point

Fig. 3. Neural signatures of targeted memory reactivation.

Pattern similarity scores computed between multi-voxel patterns of hippocampal activation associated with specific triplet presentations across the various encoding and rest periods. (A) Left hippocampus. (B) Right hippocampus. *** denotes p < .001, ** denotes p < .01, * denotes p < .05, † denotes p < .07, n.s. denotes not significant. Error bars represent standard error of the mean

Fig. 4. Relationships between hippocampal reactivation indices and order memory cueing benefit.

(A) Scatterplots depicting the relationship between presumed cued reactivation (cued encoding – cued rest similarity) in the left hippocampus and the cueing benefit index (memory for uncued triplets subtracted from memory for cued triplets) derived from performance on the immediate serial position test (left) and the relationship between the uncued reactivation index (uncued encoding – uncued rest similarity) and the immediate serial position test cueing benefit (right). (B) Scatterplots depicting the relationship between presumed cued reactivation in the right hippocampus and the cueing benefit derived from performance on delayed order memory recognition judgments (left) and the relationship between the uncued reactivation index and the delayed order memory recognition judgments (right)

Fig. 5. Category-specific reactivation.

Pattern similarity scores computed between multi-voxel patterns of FFA activation associated with face sequences (left) and scene sequences (right) across the various encoding and rest periods. (A) Left FFA. (B) Right FFA. ** denotes p < .01, * denotes p < .05

Fig. 6. Preservation of temporal sequences.

(A) Between-triplet pattern similarity measured across triplet presentations during the cued encoding and cued rest periods. Between-triplet similarity indexes representational similarity between the different triplets at each repetition. Presentation numbers 1 – 4 represent the cued encoding presentations, while presentation numbers 5 and 6 represent the cued rest presentations. Each data point reflects a particular participant. Left: left hippocampus. Right: right hippocampus. (B) Scatterplot depicting the relationship between the slopes fit to each participant’s between-triplet similarity data, reflecting the magnitude of decreasing between-triplet similarity across triplet presentations during the cued encoding and cued rest periods, and the cueing benefit index derived from the delayed serial position test