The role of REM sleep in neural differentiation of memories in the hippocampus

Abstract

When faced with a familiar situation, we can use memory to make predictions about what will happen next. If such predictions turn out to be erroneous, the brain can adapt by differentiating the representations of the cues that generated the prediction from the mispredicted item itself, reducing the likelihood of future prediction errors. Prior work by Kim et al. (2017) found that violating a sequential association in a statistical learning paradigm triggered differentiation of the neural representations of the associated items in the hippocampus. Here, we used fMRI to test the preregistered hypothesis that this hippocampal differentiation occurs only when violations are followed by rapid eye movement (REM) sleep. In the morning, participants first learned that some items predict others (e.g., A predicts B) then encountered a violation in which a predicted item (B) failed to appear when expected after its associated item (A); the predicted item later appeared on its own after an unrelated item. Participants were then randomly assigned to one of three conditions: remain awake, take a nap containing non-REM sleep only, or take a nap with both non-REM and REM sleep. While the predicted results were not observed in the preregistered left CA2/3/DG ROI, we did observe evidence for our hypothesis in closely related hippocampal ROIs, uncorrected for multiple comparisons: In right CA2/3/DG, differentiation in the group with REM sleep was greater than in the groups without REM sleep (wake and non-REM nap); this differentiation was item-specific and concentrated in right DG. Differentiation effects were also greater in bilateral DG when the predicted item was more strongly reactivated during the violation. Overall, the results presented here provide initial evidence linking REM sleep to changes in the hippocampal representations of memories in humans.

Keywords: REM sleep; memory consolidation; neural differentiation; predictive coding; statistical learning.

Figure 1.. Experimental design and methods.

(a) Study day timeline. Participants entered the MRI scanner at 10:00 AM and completed an incidental encoding task. Next, participants were randomly assigned to one of three EEG-recorded offline conditions: wake, a short 50-minute nap followed by podcast listening, or a long 90-minute nap. Participants re-entered the MRI scanner at 5:00 PM and completed one run of post-learning “B” scene snapshots, the reward association task, and one run of post-learning “X and Y” face snapshots. (b) Task design and analysis schematic. Participants viewed streams of scene and faces images presented one at a time. At the beginning of each run, the A and B members of each pair were shown once separately (i.e., B did not follow A) to obtain pre-learning snapshots of each item. During the learning phase, pairs in the violation condition followed a sequence of three initial learning repetitions followed by two cycles of violation and restudy trials (AB-AB-AB-AX-B-AY-B). The nonviolation control condition did not have any violation events, but did include two B restudy trials (AB-AB-AB-B-B). Each pair’s trials were interleaved with repetitions of other pairs (represented here as gray dots). In session 2, the B scenes were presented one more time in a random order to take post-learning snapshots. To measure neural differentiation, we correlated voxel patterns for the pre-learning snapshot of A and post-learning snapshot of B (“pairmate pattern similarity”) for all pairs. To measure the amount of “B prediction” on violation trials, we correlated the pre-learning snapshot of B and the pattern of activity evoked by the X and Y violation events, then averaged these values for one B prediction score per pair in the violation condition. Note: The scene images in this figure were sourced from the internet and not used as actual stimuli in our experiment; face stimuli have been replaced with black ovals.

Figure 2.. Neural differentiation.

Neural differentiation scores were calculated as the difference in preA/postB pattern similarity for the violation minus nonviolation task conditions, in each sleep group in (a) CA2/3/DG and (b) CA1. A planned contrast in our 6 ROIs revealed more violation-related neural differentiation in the REM group than the Wake and NREM groups in right CA2/3/DG (p = 0.03, uncorrected for multiple comparisons). Within the REM group, post-hoc tests showed that the neural differentiation score in right CA2/3/DG was significantly negative (p = 0.01, one-tailed) and reliably item-specific based on a randomization analysis (p = 0.02). Brain image shows segmented ROIs for one subject overlaid on their high-resolution T2w anatomical image. n = 23 in each group; *p< 0.05.