Spindle-locked ripples mediate memory reactivation during human NREM sleep

The study investigates how coordinated non-rapid eye movement (NREM) sleep oscillations—cortical slow oscillations (SO/SOS), thalamocortical sleep spindles, and hippocampal sharp-wave ripples—support memory consolidation via reactivation of prior experiences. Contemporary models posit that SOs create windows of cortical excitability, spindles gate plasticity through dendritic Ca2+ influx, and hippocampal ripples (80–120 Hz in humans) orchestrate neuronal population dynamics to replay recent memories, often nesting within spindle troughs. Although hierarchical coupling among SOs, spindles, and ripples in human NREM sleep has been observed, direct human evidence that spindle-locked ripples mediate memory reactivation has been lacking. The research question tests whether targeted memory reactivation (TMR) during sleep elicits decodable reactivation of previously learned spatial-context information and whether such reactivation is specifically tied to spindle-locked medial temporal lobe (MTL) ripples.

Prior work shows hierarchical synchronization of SOs, spindles, and ripples during NREM sleep in humans and animals, suggesting a mechanism for hippocampo-cortical communication during consolidation. Spindles nest in SO upstates and are linked to synaptic plasticity; ripples are proposed markers and drivers of memory replay and transfer. TMR studies in rodents indicate auditory cues can bias hippocampal replay and cortical-hippocampal interactions in loop-like dynamics. Human scalp EEG studies have implicated SO–spindle complexes in reactivation, but skull filtering obscures high-frequency ripples, leaving their human role uncertain. Thus, while theory emphasizes synchronous SO–spindle–ripple coordination, empirical human evidence for ripple involvement in sleep reactivation remained a gap the present study addresses.

Two complementary experiments were conducted: (1) scalp EEG in 25 healthy adults (mean age ~25.2 years; 16 female) and (2) intracranial EEG (iEEG) in 10 epilepsy patients (mean age ~31.2 years; 7 female). Participants learned associations between 168 (EEG) or 144 (iEEG) object images and four specific head orientations (−60°, −30°, +30°, +60°), each orientation linked to an orientation-specific sound. Retrieval involved recognition followed by associative recall of the head orientation for recognized items. During one hour of NREM sleep, two learning-related sounds (one left, one right orientation) plus an unrelated control sound were presented as TMR cues at ~5.5 s intervals (up to 60 min; EEG study: ~136 repetitions/stimulus; iEEG: ~187 repetitions/stimulus), with stimulation paused upon arousal/REM. EEG acquisition: 65-channel scalp system at 1000 Hz; iEEG: depth electrodes (frontal, parietal, temporal; MTL contacts where available) at 1000 Hz. Preprocessing included artifact rejection, ICA for ocular artifacts (scalp), and exclusion of epileptiform activity segments (iEEG). Time–frequency analysis used sliding windows (1–25 Hz) with Hanning taper; power z-scored. Multivariate pattern analysis (MVPA) used LDA with PCA dimensionality reduction (first 30 PCs; ~96–98% variance), features as channels/contacts, temporal smoothing (150 ms), AUC as metric, fivefold cross-validation repeated five times. Retrieval decoding: classifiers trained/tested within retrieval trials to decode left vs right head orientations. TMR decoding: cross-classification where classifiers trained on retrieval data (−0.5 to 1.5 s) were tested on TMR epochs to detect reactivation of orientation-related patterns. Source-level searchlight decoding (scalp) used DICS beamforming and spherical ROIs around voxels. Ripple detection (MTL): band-pass 80–120 Hz; RMS over 20 ms window, threshold mean +2 SD over TMR data; events 25–300 ms with ≥3 cycles; ripple-locked analyses centered on peak. Trials were stratified by SO–spindle power (median split) to assess ripple timing and reactivation dependence on SO–spindle strength. Ripple-triggered classification applied retrieval-trained weights to MTL channels around ripple peaks within spindle windows (e.g., ~700–1400 ms post TMR onset) with surrogate baselines from 1000 iterations. Statistical testing used cluster-based permutation tests (two-sided, corrected for multiple comparisons across time/frequency/space) and correlations (Spearman/Pearson).

- Behavioral (scalp EEG; N=25): Memory performance declined from pre- to post-sleep (F(1,24)=19.24, p<0.001), with a significant interaction indicating a TMR effect (F(1,24)=5.48, p=0.028). - Retrieval decoding (scalp EEG): Left vs right head orientation decodable around associative prompt onset (~30–680 ms; peak ~270 ms; p=0.0009, cluster-corrected), with searchlight suggesting bilateral parietal/occipital and right medial temporal pole involvement. - TMR effects (scalp EEG): Learned vs control sounds evoked increased SO–spindle-range power (p=0.002). Cross-classification from retrieval to TMR showed above-chance decoding 930–1410 ms after TMR onset (p=0.023). Classification performance correlated positively with TMR-triggered power (Spearman r=0.50, p=0.01). - Confidence (exploratory; scalp): Pre-sleep confidence correlated with reactivation/recapturing score (rho=0.41, p=0.003; remembered+cued trials rho=0.43, p=0.028). High-confidence trials showed significant sleep decoding (p=0.025); low-confidence trials did not. - Retrieval decoding (iEEG; N=10): Significant above-chance decoding of left vs right orientations during retrieval (peak ~250–500 ms; p=0.019, cluster-corrected). - TMR spectral effects (iEEG): Learned cues vs control increased SO–spindle power (p<0.001). - Ripple dynamics (MTL): Across 199 TMR trials, 2,217 ripples detected in 231.14 s; ripple PSD peaks at SO/delta (1–4 Hz), spindle (~14 Hz), and ripple (~84 Hz). Ripple rates (events/s) diverged in timing between high vs low SO–spindle-power trials, peaking during elevated spindle activity; overall ripple counts did not differ (high: 66.57±10.75 vs low: 70.35±10.96; t(13)=−1.1, p=0.283), indicating coordination of timing rather than number. - Reactivation depends on SO–spindle power (iEEG): Cross-classification revealed stronger TMR reactivation for high vs low SO–spindle trials (evidence reported with significant cluster in figure: p=0.019, cluster-corrected). - Spindle-locked ripples coincide with reactivation: Memory reactivation during TMR was temporally locked to MTL ripples occurring within spindle windows (approximately 700–1400 ms after cue), supporting a specific role of spindle-locked ripples in human sleep reactivation.

The results directly support models of systems consolidation in which SO upstates enable spindles, and spindles in turn group hippocampal ripples that drive memory reactivation and hippocampo-cortical dialogue. Using TMR to bias reactivation content, the study demonstrates that orientation-related neural patterns re-emerge during sleep and that their detectability depends on elevated SO–spindle activity and the presence of spindle-locked MTL ripples. Cross-modal evidence from scalp and intracranial recordings shows convergent timing: TMR-induced reactivation emerges during SO–spindle activity and aligns with ripple occurrence within spindle troughs. This establishes human evidence that ripples, especially when coupled to spindles, are tightly linked to reactivation. The findings align with animal studies showing auditory cue-induced hippocampal replay and loop-like cortical–hippocampal interactions centered on ripples, suggesting that TMR may facilitate generation of memory-related ripples by biasing hippocampal state. Moreover, the paradigm’s naturalistic spatial-context manipulation (head orientation) adds ecological validity, indicating that reactivation mechanisms generalize to real-world bodily contexts.

Combining scalp EEG in healthy participants with iEEG in epilepsy patients, the study identifies spindle-locked MTL ripples as key oscillatory events accompanying human memory reactivation during NREM sleep. TMR cues re-engage orientation-specific neural patterns, with reactivation strength enhanced by SO–spindle activity and temporally coinciding with ripple events nested in spindles. These findings bolster systems consolidation accounts positing coordinated SO–spindle–ripple interactions for hippocampo-cortical information transfer and long-term storage. Future work should test causal mechanisms and directionality within putative cortical–hippocampal–cortical loops, resolve conditions under which TMR benefits vs. detracts from behavior, and further dissect how ripple–spindle coupling supports the reactivation of complex, naturalistic memory contexts.

- Causality and directionality: The data are agnostic about a cortical–hippocampal–cortical loop and cannot establish causal information flow. - Scalp EEG sensitivity: High-frequency ripples cannot be directly measured at the scalp due to skull filtering; ripple inferences rely on iEEG data. - Clinical sample constraints: iEEG patients have epilepsy with limited electrode coverage; fixed-effects analyses at the contact level and seizure-related exclusions may limit generalizability. - Sample size and exclusions: Small iEEG sample (N=10) and trial exclusions for epileptiform activity reduce power. - Behavioral effects: TMR yielded an overall decline in recall with an interaction effect; the detrimental direction complicates straightforward behavioral interpretation and may reflect cue–target competition. - Mixed p-values across descriptions: Some reported significance values differ across text/figures, indicating potential reporting inconsistencies. - Limited ripple coverage: Ripple analyses restricted to MTL contacts in a subset of patients; not all ripples may reflect memory-related events.