Awake ripples enhance emotional memory encoding in the human brain

The study investigates how emotional experiences are preferentially remembered, focusing on the role of hippocampal sharp-wave ripples (80–150 Hz) and amygdala-hippocampal interactions immediately after encoding. Prior work suggests neuromodulation and amygdala-hippocampal interplay support emotional memory, and that memory reinstatement in the immediate post-encoding period predicts later performance. Hippocampal ripples are linked to synchronous neural activation and binding of distributed memory traces, with disruption impairing memory use. The authors hypothesized that ripples occurring immediately after stimulus encoding (post-encoding) facilitate emotional memory discrimination via coordinated amygdala-hippocampal reinstatement or working-memory retention, which would manifest as increased stimulus similarity during post-encoding ripples. They tested these hypotheses using intracranial EEG in epilepsy patients performing an emotional encoding and discrimination task, predicting increased post-encoding ripples for arousing stimuli, ripple-locked reinstatement in amygdala/hippocampus, and that these measures would predict later discrimination accuracy.

Emotional memory enhancement has been attributed to neuromodulatory influences and amygdala-hippocampal interactions (e.g., Talmi 2013; Yonelinas & Ritchey 2015; Dolcos et al. 2004). Immediate post-encoding activity in hippocampus can predict later memory (Ben-Yakov et al. 2013; Sols et al. 2017). Hippocampal ripples are transient high-frequency oscillations related to reactivation and consolidation (Buzsáki 2015); ripple disruption impairs memory (Jadhav et al. 2012). Awake replay during fear memory retrieval implicates ripples in emotional memory (Wu et al. 2017). Coordinated ripple-related activity between hippocampus and amygdala/cortex occurs in humans and animals (Logothetis et al. 2012; Skelin et al. 2021; Cox et al. 2020; Vaz et al. 2019, 2020; Dickey et al. 2022). Behavioral studies report enhanced mnemonic discrimination for emotional (especially arousing) items (Szöllösi & Racsmány 2020; Leal et al. 2014; Zheng et al. 2019). Reinstatement during post-encoding and sleep supports consolidation (Carr et al. 2011; Schreiner et al. 2021; Genzel et al. 2020). Together, these findings motivate testing whether awake post-encoding ripples and amygdala-hippocampal coordination mediate enhanced discrimination of emotional stimuli.

Participants: Seven iEEG patients (3 females; mean age ± SD = 33 ± 16) undergoing presurgical epilepsy monitoring at UC Irvine participated. Inclusion required ≥85% correct Novel-trial discrimination; one participant was excluded from behavioral analyses for low Novel performance, leaving 7 for behavior and 6 for ripple-based analyses.

Task: An Emotional Memory Encoding and Discrimination (EMOP) task with encoding and retrieval blocks. Encoding trials: fixation (1000 ms), stimulus encoding (2000 ms), then a self-paced post-encoding response (up to 2000 ms) to rate valence (negative/neutral/positive). Retrieval trials: fixation (1000 ms), stimulus (2000 ms) that was Repeat, Lure, or Novel, then self-paced discrimination (up to 2000 ms) classifying Old vs New. Stimulus valence, arousal (1–9), and lure-pair similarity (1–8) were rated by separate healthy cohorts and mapped to categories for patients (high correspondence ≥85%).

Recordings and preprocessing: iEEG recorded (Nihon Kohden/NeuraLynx ATLAS, 5000 Hz, high-pass 0.01 Hz). Signals downsampled to 2000 Hz, re-referenced to nearest white matter contact, line noise removed with FIR notch, high-pass 0.3 Hz. Power spectral density via multitaper. Interictal discharges and contaminated trials were marked/excluded by an epileptologist. Electrode localization used pre- and post-implant MRI/CT coregistration (ANTs) aligned to a medial temporal template; placements reviewed by a neurologist.

Ripple detection: Conducted on hippocampal channels after artifact exclusion using FMA toolbox. Bandpass 80–150 Hz (Chebyshev 4th order; filtfilt), Hilbert analytic amplitude, z-scored envelope. Events considered ripples if envelope > z=2 for 20–100 ms and peak > z=5; edge periods (±75 ms around trial onsets/offsets) zeroed. Control: synthetic signals matching channel PSDs underwent the same detection; only channels with ripple counts > synthetic and z-score > -2 were retained. The hippocampal channel with most ripples per participant provided ripple timestamps for both hippocampal and amygdala ripple-locked analyses. Ripple-like events in amygdala were also detected with the same algorithm. Only Lure trials were used for ripple-based analyses. Time-resolved ripple rates were computed with 1 ms bins, smoothed (Gaussian σ=150 ms), and compared by condition with non-parametric cluster-based permutation tests.

High-frequency activity (HFA) and time-frequency: HFA (30–280 Hz) extracted via Ensemble Empirical Mode Decomposition (EEMD) to derive intrinsic mode functions (IMFs); IMFs with center frequency >30 Hz were summed to reconstruct HFA. Instantaneous spectral power was computed using Morlet/superlet-style wavelet transforms across 30–280 Hz (1 Hz steps, cycles n=2–10), normalized by z-transform and baseline-corrected (-1000 to 0 ms pre-stimulus).

Representational Similarity Analysis (RSA): For each trial, power spectral vectors (PSVs) were computed for 100 ms time bins (10 ms step). Spearman correlations between encoding (0–2000 ms) PSVs and post-encoding response-period PSVs (0–RT) yielded trial-specific similarity matrices, Fisher-transformed and averaged per region and participant. Stimulus-specific similarity S_spec was defined by comparing same-stimulus vs different-stimulus similarity, normalized by their averages; significance assessed with Monte Carlo nulls by shuffling stimulus identities and non-parametric cluster-based permutation tests.

Ripple-locked similarity: Stimulus similarity time series were aligned to hippocampal ripple peaks (±250 ms) during the post-encoding period, separately for amygdala and hippocampus, avoiding overlap with encoding. Ripple-locked similarity significance was tested against null distributions generated by circular jittering of ripple times within ±500 ms of the post-encoding window.

Joint cross-structure similarity and directionality: Cross-structure joint ripple-locked similarity was computed as the outer product of amygdala and hippocampal similarity traces around each ripple, averaged across ripples for correct vs incorrect trials; significance via Monte Carlo jittering of ripple times. Directionality was assessed with time-lagged mutual information (MI) between amygdala and hippocampal similarity traces (200 ms bins, 10 ms step), binning similarity into 10 equiprobable bins. Directional influence was defined by MI(AMY→HPC) − MI(HPC→AMY), tested with Wilcoxon signed-rank per time bin and corrected using cluster-based permutation.

Statistics: Behavioral effects (valence, arousal, similarity) on Lure discrimination used logistic linear mixed-effects models with participant as random intercept. Ripple rate associations used Wilcoxon signed-rank tests and two-way ANOVAs; epoch dependence tested with two-way ANOVA and post hoc multcompare. Theta-power association with memory used logistic regression. Multiple comparisons controlled with Benjamini–Hochberg. Group comparisons of peak similarity timing used paired t-tests.

Behavior: Participants showed high accuracy for Repeat (89.4 ± 2.4%) and Novel (93.9 ± 1.4%) stimuli, with lower accuracy for Lure (61.5 ± 3.7%; t(6)=8.36, p=0.0002 vs Novel; t(6)=6.13, p=0.0009 vs Repeat). Lure discrimination negatively associated with lure-pair similarity (t(452)=-2.06, p=0.039) and positively with stimulus-induced arousal (t(452)=1.98, p=0.047); no significant effect of valence. In extended LME: arousal (t(448)=15.782, p=6.15×10^-5) and similarity (t(448)=50.562, p=2.99×10^-187) predicted correct Lure discrimination; valence did not (t(448)=1.020, p=0.308). Arousal × similarity interaction was significant (t(448)=10.327, p=1.44×10^-22). LDI showed no main effect of valence, but was higher for high-arousal stimuli (t(6)=-2.058, p=0.043, one-tailed).

Post-encoding ripple rate: In 6 participants (14 hippocampal, 20 amygdala electrodes), post-encoding hippocampal ripple rate was higher following arousing vs low-arousal stimuli (z(5)=-1.99, p=0.046) and was higher for subsequently correct vs incorrect Lure discrimination (z(5)=-2.20, p=0.028). No association with valence (F(2,15)=1.88, p=0.187). A two-way ANOVA showed independent main effects of arousal (F(1,20)=4.93, p=0.038) and later correctness (F(1,20)=8.32, p=0.009) with no interaction. Time-resolved analyses revealed distinct windows: higher ripple rates for correctly discriminated Lures from -400 to -50 ms relative to response (p=0.005) and for high-arousal Lures from -780 to -600 ms (p=0.035). Effects were specific to the post-encoding epoch; no conditional differences during encoding or retrieval. Epoch × discrimination interaction was significant (F(2,30)=10.97, p=0.0003), with post-encoding ripple rates higher for correct Lures (p<0.001) and no differences in other epochs. About 30.8 ± 7.4% of Lure trials contained post-encoding ripples. Ripple occurrence was higher during low-theta states; no overlap with broadband gamma increases. Post-encoding theta power did not predict correctness (t(261)=0.187, p=0.851).

Ripple-locked stimulus similarity: Significant post-encoding ripple-locked stimulus-specific similarity occurred in both regions with distinct timing: amygdala showed two intervals (-105 to -50 ms; 40 to 200 ms relative to ripple peak), while hippocampus showed -100 to 50 ms. Amygdala ripple-locked similarity positively tracked stimulus arousal (80 to -10 ms, p=0.035) but not later correctness; hippocampal ripple-locked similarity positively tracked later correct Lure discrimination (15 to 90 ms, p=0.008) but not arousal. A double dissociation showed stronger arousal association in amygdala (-70 to 20 ms, p<0.001) and stronger correctness association in hippocampus (-60 to 10 ms, p=0.046). Peak similarity occurred earlier in amygdala than hippocampus by 18 ± 11 ms (t(5)=-3.89, p=0.006). Trials without post-encoding ripples showed no significant similarity differences by arousal or correctness.

Joint cross-structure dynamics and directionality: Joint amygdala–hippocampal ripple-locked similarity was significant only for subsequently correct trials, with amygdala similarity preceding hippocampal by ~100 ms and maxima around ripple peaks. Time-lagged mutual information indicated a significant AMY→HPC directional influence before ripple peak (-70 to -30 ms, p=0.038) during correct trials; no effect for incorrect trials. Only 5.89 ± 1.82% of hippocampal ripples were coincident with amygdala ripple-like events within ±50 ms, indicating joint similarity did not require amygdala ripples.

Findings demonstrate that awake post-encoding hippocampal ripples are elevated following arousing stimuli and predict successful mnemonic discrimination of similar emotional items, supporting a role for ripples in prioritizing salient experiences. The effects are specific to the immediate post-encoding period required for valence rating, consistent with theories that ripples support both retrieval for ongoing decision-making and consolidation. Ripple-locked reinstatement (stimulus similarity) peaks around ripples and shows a regional double dissociation: amygdala similarity relates to emotional arousal and precedes ripple peaks, whereas hippocampal similarity aligns with ripple peaks and predicts later discrimination. Joint amygdala–hippocampus similarity during the same ripples, with amygdala leading and exerting directional influence (AMY→HPC), predicts correct discrimination, suggesting that amygdala-triggered processes recruit hippocampal ripple-associated plasticity to bind emotional and contextual information. These results align with rodent work showing ripple-associated reactivation supporting consolidation and extend them to human awake post-encoding processing, highlighting ripples as a mechanism underpinning emotional memory benefits.

This study provides electrophysiological evidence that post-encoding hippocampal ripples in humans are enhanced by emotional arousal and facilitate subsequent mnemonic discrimination of similar items. Ripple-locked reinstatement reveals temporally structured, coordinated representations in amygdala (arousal-related, leading) and hippocampus (discrimination-related, ripple-centered), with amygdala-to-hippocampus directional influence predicting successful memory. These findings support a model in which emotional content engages amygdala processes that trigger hippocampal ripple dynamics to strengthen memory representations. Future work could test causality via targeted stimulation or ripple disruption, assess generalization beyond emotional content and across sleep/wake states, increase sample sizes with broader electrode coverage, and examine clinical applications for enhancing memory consolidation.

• Small sample size (n=7 behavioral; n=6 for ripple-based analyses) due to rarity of simultaneous amygdala–hippocampus recordings outside seizure onset zones; limits generalizability.
• Epilepsy patient population and clinically determined electrode placements may affect neural dynamics and sampling; findings may not fully generalize to healthy populations.
• Ripple detection constraints: only hippocampal channels used for primary ripple detection; some trials lacked detected ripples (possible sub-threshold or contralateral events). Control synthetic-signal analyses mitigate false positives but cannot rule out all detection biases.
• Potential frequency overlap between broadband gamma and ripple bands; analyses suggest distinct phenomena, but residual confounds are possible.
• Debate on extrahippocampal “ripple-like” events; amygdala ripple-like activity had low coincidence with hippocampal ripples (~6%), leaving mechanisms in amygdala partly unresolved.
• Task-specific demands (valence rating) may influence post-encoding processes and ripple occurrence.
• Analyses focused on Lure trials; ceiling effects and limited incorrect Repeat/Novel trials precluded assessing ripple effects on other recognition types.