Theta rhythmicity governs human behavior and hippocampal signals during memory-dependent tasks

The study investigates whether theta oscillations that coordinate hippocampal encoding and retrieval manifest as rhythmic patterns in human behavior. Prior work shows human hippocampal activity during memory often concentrates in a slow theta band (~1–5 Hz), with successful encoding linked to higher theta power and theta–gamma coupling, and retrieval linked to theta synchronization between hippocampus and cortex. Theoretical accounts propose encoding and retrieval processes are segregated by theta phase, reducing interference, with evidence for phase-specific LTP/LTD and distinct phase relationships during encoding vs retrieval. While attentional processes can show rhythmic behavioral fluctuations, it was unknown whether memory-dependent behaviors do. The authors test whether timing of encoding and recall, reported via button presses, shows oscillatory modulation, particularly in the slow theta range, and whether corresponding hippocampal LFPs exhibit phase consistency aligned with behavior.

The paper synthesizes evidence that: (1) human hippocampal theta during episodic memory is slower (1–5 Hz) compared to rodents; (2) successful encoding correlates with increased theta power and theta–gamma coupling; (3) retrieval involves theta increases and hippocampal–cortical synchronization, with hippocampal recall preceding cortical reinstatement by about a theta cycle; (4) theoretical models suggest encoding and retrieval are locked to opposing theta phases, consistent with synaptic plasticity findings (LTP near theta troughs, LTD near peaks) and rodent phase-dependent information flow; (5) human and animal studies demonstrate phase-locked spiking to theta during successful memory; (6) attentional sampling can impose theta/alpha rhythmicity on behavior. Collectively, these motivate testing for theta-rhythmic behavioral signatures in memory tasks and corresponding hippocampal phase dynamics.

Participants: 226 individuals completed memory tasks (groups 1 and 2) and 95 a visual control task (group 3). Additionally, 10 epilepsy patients performed the memory task during intracranial EEG (iEEG); their behavioral data were pooled with healthy participants for behavioral analyses. Tasks: Memory tasks comprised encoding (cue then object; button press when association formed), distractor, and retrieval phases. Group 1 retrieval: participants pressed a button at subjective recall (retrieval press) and later answered catch questions (catch-after-retrieval press). Group 2 retrieval: answer options preceded cue; participants pressed when able to answer (catch-with-retrieval press), indexing retrieval time. Visual control task (group 3): participants answered perceptual/semantic questions about concurrently visible objects (visual press), eliminating episodic memory demands. Exclusions: Participants performing at chance on catch questions (binomial test) or with insufficient trial counts due to time-outs were excluded per phase. Included participants exhibited mean accuracies: memory 84.0% and visual 96.3%. Behavioral oscillation analysis (O-score): Button press times from correct trials were treated as event trains. After removing early/late outliers (retain middle 90%), the Oscillation score method was applied: compute auto-correlation histogram (ACH), smooth, remove central peak (use positive lags beyond peak), Fourier transform, identify peak frequency within adjusted bounds (min 0.5 Hz, max 40 Hz, further adjusted to ensure ≥3 cycles and sufficient responses), and compute O-score as peak magnitude divided by spectrum average. To control for response trends and frequency bias, a gamma distribution was fit to each participant’s response density to generate 500 Poisson time series with matched trends and counts; O-scores at the observed peak frequency were computed for each surrogate and used to Z-transform the observed O-score. Second-level tests across participants assessed significance per phase. Peak frequency distributions were also analyzed. Linear mixed-effects models tested differences between memory-dependent (encoding, retrieval, catch-with-retrieval) and memory-independent (catch-after-retrieval, visual) phases, controlling for time series length. Phase of responses: For participants with significant O-scores, a continuous oscillatory trace was derived by Gaussian convolution, narrow band-pass filtering (±0.5 Hz around individual peak), and Hilbert transform. Phases of incorrect presses were taken from the correct-trial-derived phase trace; phases of correct presses used leave-one-out traces. V-tests assessed non-uniformity around 0 rad. Differences between correct and incorrect phase locking were tested via permutation of trial labels and, for participants with ≥10 incorrect trials, by downsampling correct trials to match counts and paired t-tests on V-statistics. iEEG recordings and analysis: 42 Behnke-Fried microwire bundles in hippocampus across 10 patients recorded at 32 kHz, preprocessed (0.5–200 Hz bandpass, notch 50 Hz, downsampled to 1 kHz), artifact-rejected, and segmented into encoding and retrieval trials. Complex Morlet wavelets (1–12 Hz) yielded time–frequency phases. Pairwise phase consistency (PPC) across trials was computed per time–frequency point for correct and incorrect trials. PPC significance was assessed against pre-cue baseline using non-parametric tests and cluster-based permutation with time-shuffled trials; correct vs incorrect PPC differences used trial-label shuffles. Power changes were analyzed relative to baseline with FDR correction. Phase opposition between encoding and retrieval was tested at peak PPC (cue/stimulus-locked and response-locked) using V-tests around 180°, validated against time-shuffled references. Additionally, response-locked event-related potentials (ERPs) were computed, sign-corrected across electrodes, filtered at 1–5 Hz, and instantaneous phases compared in 200 ms sliding windows with FDR correction.

- Behavioral oscillations: Significant Z-scored O-scores for memory-dependent phases: Encoding t(181)=6.20, p<0.001; Retrieval t(68)=4.58, p<0.001; Catch-with-retrieval t(143)=5.08, p<0.001 (Bonferroni-corrected). Memory-independent phases showed no evidence: Catch-after-retrieval t(69)=1.69, p=0.240; Visual t(94)=-4.10, p=1.00. - Proportion with significant O-scores: Encoding 74.7%; Retrieval 69.6%; Catch-with-retrieval 76.4%; Catch-after-retrieval 64.3%; Visual 37.9%. - Peak frequencies: For memory-dependent phases, non-uniformly concentrated in slow theta (1–5 Hz) or harmonics (KS tests p<0.001). Memory-independent phases showed broad, uniform distributions (KS p≈1.0). Mixed model showed significantly lower frequencies for memory-dependent vs independent phases (coefficient ≈ -5.09; t(371)=6.55; p<0.001). - Incorrect vs correct phase locking: Correct trials showed strong phase concentration around the oscillation peak (V-tests p<0.001 across phases). Incorrect trials’ phase distributions were uniform (Encoding V(1270)=44.1, p=0.120; Retrieval V(954)=11.2, p=0.911; Catch-with-retrieval V(5328)=73.8, p=0.229). Permutation tests confirmed stronger phase modulation for correct than incorrect trials (p<0.002 in all phases). Within-subject downsampled analyses also showed higher phase locking for correct trials (Encoding t(51)=4.07, p<0.001; Retrieval t(38)=5.41, p<0.001; Catch-with-retrieval t(101)=7.25, p<0.001). - iEEG PPC: Correct trials exhibited significant increases in PPC after cue/stimulus onset, extending in time within 2–3 Hz band, persisting to responses, especially in retrieval. Incorrect trials showed no significant PPC increases. Correct > incorrect PPC differences were significant in 2–3 Hz clusters (cluster-based permutation, α=0.05). - Power: Retrieval showed increased power accompanying PPC increases; encoding and catch questions showed PPC increases without corresponding power increases, indicating partly independent phase and amplitude modulations. - Phase opposition: Encoding and retrieval phases differed by approximately half a theta cycle. At peak PPC, cue/stimulus-locked phase difference mean ≈ 250.7°±14.1° (Rayleigh Z=30.9, p<0.001; V-test vs 180°: V=33.1, p=0.005), and response-locked ≈ 162.4°±30.6° (Rayleigh Z=7.35, p=0.001; V=46.7, p=0.001). Response-locked opposition survived permutation control (p=0.042). ERP-based analyses showed widespread time windows (55.8% after FDR) with significant phase opposition leading up to responses.

The findings demonstrate that human behavioral responses that index the timing of memory encoding and retrieval are rhythmically modulated, aligning with slow theta frequencies characteristic of human hippocampus during episodic memory. Memory-dependent phases exhibited robust behavioral oscillations, whereas visually driven, memory-independent phases did not, indicating that rhythmicity is linked to hippocampal mnemonic processing rather than perception or motor output. Correct responses, but not incorrect ones, aligned to specific phases of the behavioral oscillation, underscoring behavioral relevance. Intracranial hippocampal recordings revealed temporally extended phase consistency (PPC) at 2–3 Hz during successful encoding and retrieval but not during failures, providing a neural mechanism for the observed behavioral rhythms. Moreover, encoding and retrieval peaked at opposite theta phases, supporting computational theories of phase-segregated encoding vs retrieval to mitigate interference. Together, the results suggest that task events reset hippocampal theta phase and that rhythmic hippocampal output modulates the timing of downstream processes culminating in button presses, producing detectable behavioral oscillations.

This study shows that theta rhythmicity (1–5 Hz) governs the timing of human behavior during memory-dependent processing and is mirrored by sustained hippocampal phase consistency. Behavioral oscillations emerge specifically during encoding and retrieval and predict successful performance; hippocampal PPC increases and phase opposition between encoding and retrieval provide a mechanistic substrate. The work establishes behavior as a practical readout of underlying rhythmic mnemonic processes. Future research should clarify causal links between hippocampal theta and behavior (e.g., via theta-frequency TMS or other neuromodulation), delineate hippocampal–cortical coherence contributions, and determine whether similar rhythmic behavioral signatures generalize across cognitive domains and working memory maintenance.

- Behavioral timing relies on subjective button presses, integrating multiple processes and potentially introducing variability. - Sparse trial counts in one-shot memory tasks limit oscillation detectability; although mitigated by large samples and simulations, sensitivity may vary across phases (e.g., encoding had lowest response density). - The catch-after-retrieval phase may involve working memory maintenance; iEEG could not disentangle retrieval- from maintenance-related theta. - iEEG sample size was modest (10 patients) and limited to epilepsy patients; generalizability to healthy populations rests on behavioral convergence. - Phase opposition at cue/stimulus-locked peak PPC may be partly inflated by channel counts or waveform asymmetry; only response-locked opposition remained robust under permutation control. - Multiple task versions and stimulus sets introduce heterogeneity, though qualitative patterns were consistent across experiments.