Spindle-locked ripples mediate memory reactivation during human NREM sleep

Abstract

Memory consolidation relies in part on the reactivation of previous experiences during sleep. The precise interplay of sleep-related oscillations (slow oscillations, spindles and ripples) is thought to coordinate the information flow between relevant brain areas, with ripples mediating memory reactivation. However, in humans empirical evidence for a role of ripples in memory reactivation is lacking. Here, we investigated the relevance of sleep oscillations and specifically ripples for memory reactivation during human sleep using targeted memory reactivation. Intracranial electrophysiology in epilepsy patients and scalp EEG in healthy participants revealed that elevated levels of slow oscillation - spindle activity coincided with the read-out of experimentally induced memory reactivation. Importantly, spindle-locked ripples recorded intracranially from the medial temporal lobe were found to be correlated with the identification of memory reactivation during non-rapid eye movement sleep. Our findings establish ripples as key-oscillation for sleep-related memory reactivation in humans and emphasize the importance of the coordinated interplay of the cardinal sleep oscillations.

Fig. 1. Experimental procedure, behavioral results, and retrieval locked reactivation of head orientations.

a During encoding, participants were consecutively presented with 168 images (EEG study) / 144 images (intracranial EEG study) of objects on four flanking screens (positioned at − 60°, − 30°, + 30° and +60° relative to the center screen). Participants turned their heads towards the relevant screen, cued by one of four orientation-specific sounds. Memory performance was tested via a recognition test followed by an associative retrieval (this procedure was used before and after sleep): First, participants made object-recognition judgments (old or new). Then, for recognized images only, participants indicated which of the four head orientations was associated with the item during the learning phase. During NREM sleep, two of the learning-related sounds (one related to left-sided and one related to right-sided head orientation) and one control sound, which was not part of the learning material, were presented for 60 min. b Behavioral results for both experimental sessions pre- (light gray) and post-sleep (dark gray), separated into cued and uncued trials. Bar graphs show the mean ( ± SEM across participants) percentage of recalled head orientations. Dots indicate individual memory performance of participants (N = 25). The star denotes the significant interaction (pre vs. post x cued vs. uncued) as derived from a repeated measures ANOVA (F_1,24 = 5.48; p = 0.028). c Later cued head orientations (left vs. right) could be reliably decoded (above chance) from the retrieval data, starting around the onset of the associative memory prompt (the black solid line indicates decoding performance ( ± SEM across participants)). The horizontal dashed line indicates chance level performance (i.e., 0.5). The vertical solid line indicates the onset of associative retrieval trials (time = 0). The lower horizontal gray line shows the temporal extent of significant decoding results as derived from a two-sided cluster-based permutation test (p = 0.0009, corrected for multiple comparisons across time, N = 25). The topographical insert illustrates the results of a “searchlight decoding procedure”, indicating that bilateral centro-parietal and occipital areas exhibited stimulus-category related effects (please note that statistical tests with regards to the searchlight procedure were done for illustrative purposes only, N = 25). Source data are provided as a Source Data file.

Fig. 2. Reactivation of head orientation-related activity during TMR in healthy participants.

a Power difference between learning-related TMR cues versus new control cues after statistical thresholding (p = 0.002, two-sided cluster-based permutation test corrected for multiple comparisons; N = 25) (b) Retrieval-related brain patterns (left vs. right head orientations) were decodable during TMR (p = 0.023, two-sided cluster-based permutation test corrected for multiple comparisons across time, contour lines indicate the extent of the cluster, N = 25); color range (blue to yellow) represents t values against chance level performance. c The source plots illustrate the results of a “searchlight decoding procedure”, indicating that frontoparietal networks and the right medial temporal lobe exhibited head orientation related effects (please note that statistical tests for the searchlight procedure were done for illustrative purposes only, N = 25). d Classification performance correlated positively with TMR-triggered power (Spearman correlation; r = 0.50, p = 0.01; N = 25). Source data are provided as a Source Data file.

Fig. 3. Intracranial EEG study: retrieval.

a Behavioral results for both experimental sessions pre- (light gray) and post-sleep (dark gray), separated into cued and uncued trials. Bar graphs show the mean ( ± SEM) percentage of recalled head orientations. Dots indicate individual memory performance of participants (N = 10). The star denotes the significant interaction (pre vs. post x cued vs. uncued) as derived from a repeated measures ANOVA (F_1,9 = 8.28; p = 0.018). b Later cued head orientation (left vs. right) could be reliably decoded (above chance) from the retrieval data, starting around 190 ms after the onset of the associate prompt. The black solid line indicates decoding performance ( ± SEM across participants). The horizontal dashed line indicates chance level performance (i.e., 0.5). The vertical solid line indicates the onset of associative retrieval trials (time = 0). The lower horizontal gray line shows the temporal extent of significant decoding results as derived from two-sided cluster-based permutation test (p = 0.019, corrected for multiple comparisons across time, N = 10). c Ripple-triggered grand average of all detected ripples (N_contacts= 14; locked to maximal negative amplitude, ( ± SEM across MTL contacts) during retrieval (− 0.5 to 1.5 s). d Time-frequency-representations of ripple-locked MTL data segments; N_contacts = 14). e Ripple rates (events per second) for remembered (red) and not-remembered (blue) trials. Ripple rates differed significantly between conditions (p = 0.047; two-sided cluster-based permutation test corrected for multiple comparisons across time), with MTL ripples peaking during remembered trials (0.4 to 0.5 seconds in relation to the associative memory prompt onset; ± SEM across MTL contacts; N_contacts = 14). Source data are provided as a Source Data file.

Fig. 4. Intracranial EEG study: TMR.

a Ripple-triggered grand average of all detected ripples (N_contacts= 14; locked to maximal negative amplitude) during TMR (− 0.5 to 1.5 s; 138.78 ± 21.72 ripples in 231.14 ± 19.94 trials). A zoomed version of the ripples is illustrated in the inset. b Power spectral density (PSD) averaged across all detected SWRs [± 300 ms] indicating distinct peaks in the SO/delta, spindle, and ripple range (i.e., 3 Hz, 14 Hz, and 84 Hz, ± SEM across MTL contacts; N_contacts = 14). c Power difference indicates that retrieval-related TMR cues triggered increased power in intracranial EEG recordings (p = 0.009, two-sided cluster-based permutation test corrected for multiple comparisons, N_contacts = 317 as compared to control cues. d Ripple rates (events per second, ±SEM across MTL contacts) for trials exhibiting high (red) and low power (blue) in the SO-spindle range, respectively. Ripple rate differed significantly between conditions (p = 0.027, two-sided cluster-based permutation test corrected for multiple comparisons across time, N_contacts = 14), with MTL ripples peaking during elevated spindle activity. e Head orientation-related brain patterns (left vs. right) were decodable during TMR when contrasting high and low SO-spindle activity trials (p = 0.019 two-sided cluster-based permutation test corrected for multiple comparisons; contour lines indicate the extent of the cluster, N = 10, color range (blue to yellow) represents t values against chance level performance). Source data are provided as a Source Data file.

Fig. 5. Spindle-ripple interactions and ripple-locked classification.

a Time-frequency-representations of ripple-locked MTL data segments (top; N_contacts = 14) and spindle-locked data segments from frontal, parietal, and temporal contacts (bottom; N_contacts = 317). (b, top) Unfiltered iEEG trace from single contact in the MTL with ripple band filtered signal (80 to 120 Hz) shown in blue. Blue shaded area highlights a representative ripple. (b, bottom) Unfiltered iEEG trace from single parietal contact with spindle band filtered signal (12 to 15 Hz) shown in blue. Blue shaded area highlights a representative spindle following a SO (green). c Assessing phase-amplitude coupling (PAC) using the Modulation Index revealed that the phase of cortical spindles influenced amplitudes in the ripple range in MTL contacts (~ 80–120 Hz; p = 0.005, two-sided cluster-based permutation test corrected for multiple comparisons, N_contacts = 14). In addition, the cortical delta/theta phase exhibited a significant effect on MTL ripple amplitudes (p = 0.007, two-sided cluster-based permutation test corrected for multiple comparisons, N_contacts = 14), while the spindle phase additionally modulated low gamma amplitudes in the MTL ( ~ 20–40 Hz; p = 0.0009; two-sided cluster-based permutation test corrected for multiple comparisons, N_contacts = 14). The top inset illustrates phases of the spindle-ripple modulation, indicating a clustering of ripples towards spindle troughs (corresponding to ±π; two-sided V test against ± pi: v  =  5.29, p  =  0.022 N_contacts = 14; mean coupling direction: −176.67 ± 16.61°, mean vector length = 0.21 ± 0.031). The bottom inset illustrates the temporal modulation of MTL ripple onsets by cortical spindle onsets (ripple percentage in relation to spindles ± 1 sec; mean ± SEM across MTL contacts, N_contacts = 14). A solid horizontal line indicates mean spindle duration (mean spindle duration: 0.75 ± 0.008 sec; peak latency: 0.37 ± 0.042 sec in relation to spindle onsets). (d) Head orientation-related brain patterns (left vs. right) were decodable during the presence of spindle-locked MTL ripples (p = 0.007, two-sided cluster-based permutation test corrected for multiple comparisons, N = 7; color range (blue to yellow) represents t values). Source data are provided as a Source Data file.