Hippocampal sharp-wave ripples correlate with periods of naturally occurring self-generated thoughts in humans

The study investigates how hippocampal sharp-wave ripples (SWRs) relate to naturally occurring self-generated thoughts in humans, such as mind wandering. Prior work indicates that self-generated cognition occurs frequently in daily life and is linked to individual differences in intelligence, neurodevelopmental and attentional traits, and well-being. Neuroimaging implicates default mode network regions, including posterior cingulate and medial prefrontal cortex, and structural variation in the medial temporal lobe, in these experiences. Contemporary theories suggest hippocampal mechanisms—especially memory replay reflected in SWRs—may support self-generated thought. SWRs occur during non-REM sleep and quiet wakefulness and decrease during active engagement. The research question asks whether fluctuations in hippocampal SWR rates track the content and form of ongoing thought in daily life and whether SWRs show diurnal patterns and relationships to physiological state.

Evidence links self-generated cognition to default mode network activity and medial temporal lobe structure and function. Hippocampal SWRs are established markers of replay, prominent during NREM sleep and quiet wakefulness, and implicated in memory consolidation, future planning, recall, and imagination. Prior human and animal studies show SWRs coordinate with slow oscillations and can influence default mode network activation. Lesions or atrophy in hippocampal regions reduce mind-wandering, and anterior parahippocampal volume relates to off-task and vivid thought features. Despite these insights, direct links between hippocampal SWRs and naturally occurring thought content in humans remain underexplored, motivating the present long-term intracranial EEG and experience-sampling approach.

Participants: 23 patients with drug-resistant epilepsy underwent intracranial electrode implantation between January 2020 and May 2022; 15 had depth electrodes in hippocampus or parahippocampal gyrus. After exclusions (e.g., hippocampal sclerosis), 10 patients (6 male, 4 female; 36.6 ± 14.2 years) with hippocampal depth electrodes consented to participate in experience sampling and wore a wrist sensor. Data collection occurred in hospital rooms under continuous video monitoring; analyses included EEG data from ≥4 days post-implantation. Recordings: Hippocampal local field potentials (LFPs) were recorded for 9–15 days, sampled at 10 kHz, preprocessed (0.016 Hz high-pass, 3000 Hz low-pass), downsampled to 2 kHz. Bipolar derivations were created between hippocampal contacts. Power-line noise at 60, 120, 180, 240 Hz (±1.5 Hz) was removed. SWR detection: Signals were bandpass filtered 70–180 Hz (zero-lag Hamming-window FIR, 5 Hz transition). Instantaneous amplitude via Hilbert transform was clipped at 4× grand SD, squared, low-pass filtered, and Z-scored. Ripple candidates were epochs exceeding 4× SD with durations 20–200 ms; peaks <30 ms apart were concatenated. Spectral decomposition (4–200 Hz) and inter-event interval distributions were computed. Manual exclusion removed epileptic activity periods (electrographic seizures, pre-/post-ictal segments with buffers) and movement artifacts, confirmed by an epileptologist. Physiological measures: Empatica E4 wristband captured electrodermal activity (EDA, 4 Hz), 3-axis acceleration (ACC, 32 Hz; RMS), blood volume pulse (BVP, 64 Hz), and interbeat interval (IBI) derived from BVP. Data were averaged in 1-min bins and day-wise Z-normalized. One patient with <2 days wearable data was excluded from physiological analyses. Experience sampling: Hourly between 08:00–20:00, patients completed a tablet questionnaire (multidimensional experience sampling; Likert 1–7). Items assessed thought content and affect (e.g., external-task focus, future, past, self, social, emotion valence, visually imaginable, words, vivid, detail, habit, unwanted, evolving, spontaneous/deliberate; plus mood items: happy, calm, excited). The external-task focus item was always first; others were randomized. Responses were standardized within-participant. TTL markers aligned response times with EEG. Scheduling and state classification: Mealtimes were defined as 07:00–08:00, 12:00–13:00, 18:00–19:00; nighttime 22:00–06:59; daytime 07:00–21:59. Video determined precise lighting and eating periods. Cortical delta power: For patients with cortical electrodes, bipolar ECoG pairs were used to estimate delta amplitude; relationships between hippocampal SWR rate and cortical delta power were assessed. Statistical analyses: One-way ANOVAs assessed diurnal variations. Correlations (Pearson/Spearman) examined associations (with Bayes factors reported for SWR–delta analyses). General linear mixed-effects models (random intercept for patient) related 1-hr standardized SWR rates to biometric variables and time of day, and related SWR rates in sliding windows (1-min bins) to thought and mood ratings (5-min averages across −15 min to +15 min around responses). Model selection used coefficients of determination with Bonferroni corrections; permutation tests (n=1000) assessed significance. Data and questionnaires were standardized per participant before modeling.

• SWR detection and dynamics: Detected SWRs exhibited canonical features (waveform peaks, dominant frequencies, inter-event intervals) consistent with prior reports. SWR rates showed pronounced fluctuations across days and increased during the night, decreasing upon waking.
• Diurnal modulation and sleep linkage: Across patients, normalized SWR rates positively correlated with cortical delta power measured from neocortical electrodes. Day–night comparison of correlations revealed no significant difference (R = 0.38 ± 0.25 daytime vs 0.36 ± 0.19 nighttime; paired t test t ≈ 0.93; P ≈ 0.78; Bayes factor ≈ 0.387), consistent with SWR coupling to sleep-related slow-wave activity. SWR rates exhibited clear 24-h variation with elevations at night and in early afternoon (13:00–15:00) and reductions around mealtimes.
• Seizure effects: Among 117 recorded seizures (mean duration ≈19 min), exclusion windows (±30–60 min) were applied. SWR event rates did not differ significantly in the 10 min before vs 30 min after seizures (paired t-test t475 = 0.564; P = 0.584), suggesting limited seizure-related confounding. No significant differences in seizure frequency were observed between mealtime and non-mealtime periods.
• Physiological correlates: Diurnal variations were observed in some biometric measures (e.g., IBI ANOVA P < 0.001; EDA, ACC, BVP not significant after correction in group-level ANOVAs). A mixed-effects model modestly explained SWR rate variance using physiological scores and hour (R^2 ≈ 0.204; P < 0.005, Bonferroni-corrected). At the best-performing time window, significant predictors included BVP (coef ≈ +0.11 ± 0.0035; P = 4.6×10^−6), ACC (coef ≈ +0.01 ± 0.0035; P = 0.037), and EDA (coef ≈ −0.029 ± 0.0053; P = 8.2×10^−3).
• Thought content associations: Using mixed-effects models on 5-min windows across −15 to +15 min relative to experience-sampling responses, SWR rates in the 5 min preceding reports were better explained by thought features than during the response period itself. Elevated SWR rates were associated with experiences that were more vivid and visually imaginable, less desirable (more unwanted), and less focused on the external task (i.e., more off-task). These associations were stronger than those with physical activity measures, indicating a tighter link between SWRs and cognitive content than with gross biophysical state.

Findings support the hypothesis that hippocampal SWRs relate to naturally occurring self-generated thought in humans. SWR rates tracked diurnal fluctuations aligned with sleep-related delta activity and decreased around mealtimes, paralleling known state-dependence of SWRs. Critically, SWR rates were elevated during periods when participants reported thoughts that were vivid, imagery-based, unwanted, and less externally focused, suggesting that hippocampal replay-like processes may underpin features of off-task cognition in daily life. The weaker and inconsistent relationships with peripheral physiological measures imply that cognitive content, rather than general arousal or movement, better explains SWR variability. These results align with literature linking the medial temporal lobe and default mode network to self-generated thought and with evidence that SWRs contribute to memory-related simulations such as planning and recall. The timing results—stronger associations prior to the questionnaire than during it—suggest that task engagement may transiently suppress the SWR–cognition coupling, consistent with reduced SWRs during active behavior.

This study demonstrates that human hippocampal SWR rates fluctuate diurnally and are linked to ongoing thought patterns during daily life, especially to experiences that are vivid, imagery-based, unwanted, and less tied to current tasks. By integrating long-term intracranial EEG, experience sampling, and peripheral physiology, the work bridges hippocampal network activity with naturally occurring cognition beyond laboratory tasks. Future research should test causality and generalizability: employing task manipulations, pharmacological interventions targeting neuromodulatory influences on SWRs, high-field imaging in healthy populations, and larger cohorts across naturalistic settings to delineate how hippocampal dynamics shape intrusive and self-generated thinking, with potential clinical implications for conditions characterized by unwanted thoughts.

• Patient and setting constraints: Data were collected from epilepsy patients in hospital rooms, limiting activity diversity and potentially shaping thought patterns; the number of responses varied across patients due to clinical schedules and condition.
• Correlational design: Associations between SWRs and thought content are correlational with broad temporal windows; causality and precise temporal precedence cannot be inferred.
• Epilepsy-related confounds: Although seizure periods and buffers were excluded and no significant peri-seizure SWR rate differences were found, contamination by pathological high-frequency oscillations (HFOs) cannot be fully excluded in epileptic hippocampus.
• Sleep-stage identification: Lack of full polysomnography (e.g., EMG) limits precise sleep stage classification; inferences rely on cortical delta power proxies.
• Medication and physiological factors: Anti-epileptic medication timing and metabolic factors (e.g., glucose) around meals may influence SWR rates; analyses suggest these do not fully account for effects but residual confounding cannot be ruled out.
• Sample size and generalizability: Small sample (n=10) and hospital context may limit generalization to healthy populations and natural environments; larger, diverse cohorts are needed.