Declarative Memory Through the Lens of Single-Trial Peaks in High-Frequency Power

Adam J. O. Dede, Zachariah R. Cross, Stephen Gray, Joseph P. Kelly, Qin Yin, Parisa Vahidi, Eishi Asano, Stephan Schuele, Joshua M. Rosenow, Joyce Y. Wu, Sandi Lam, Jeffrey S. Raskin, Jack J. Lin, Olivia Kim McManus, Shifteh Sattar, Ammar Shaikhouni, David King‐Stephens, Peter B. Weber, Kenneth D. Laxer, Peter Brunner, Jarod L. Roland, Ignacio Sáez, Fady Girgis, Robert T. Knight, Noa Ofen, Elizabeth L. JohnsonbioRxiv 2025 · DOI

TL;DR

This study reveals that successful memory recall isn't just about sustained brain activity, but rather about precisely timed, transient "brain states" that coordinate local processing (high-frequency power) with global network organization (theta oscillations). By analyzing intracranial EEG data, researchers found that these fleeting states, anchored to internal brain events rather than just external stimuli, are crucial for forming and retrieving memories.

What Problem Does This Paper Address?

Declarative memory, our ability to recall facts and events, is thought to rely on both localized brain processing and broader network coordination. High-frequency broadband (HFB) activity is linked to local processing, while slower theta oscillations are associated with global network organization. The puzzle is how these two types of brain activity, which often occur at different times, communicate to support memory. Specifically, how do the results of local processing get communicated across the network when their timing isn't perfectly synchronized? This paper tackles this by investigating the precise temporal relationship between HFB peaks and theta oscillations during memory tasks.

Methods

The researchers used intracranial EEG (iEEG) data from epilepsy patients performing a recognition memory task. A key methodological innovation was their dual approach to time-locking analyses:

  • HFB-locked analysis: This approach anchored analyses to the precise moment of HFB peak activity in specific brain regions. This allowed them to investigate how other brain signals (like theta oscillations) behaved relative to these internal, localized processing events.
  • Image-locked analysis: This is a more standard approach, anchoring analyses to the presentation of the visual stimulus.

By comparing these two approaches, they could disentangle stimulus-driven effects from internally generated brain states. They analyzed data from several brain regions of interest (ROIs), including the hippocampus (Hip), parahippocampal gyrus (PHG), anterior cingulate cortex (ACC), dorsolateral prefrontal cortex (dIPFC), and polar prefrontal cortex (pPFC). Advanced techniques like inter-trial phase clustering (ITPC) and pairwise phase consistency (PPC) were used to measure theta phase organization and connectivity, respectively. Graph theory was employed to analyze the overall network state.

Datasets

The study utilized intracranial EEG (iEEG) data from 36 participants (adolescents and early adults) undergoing monitoring for epilepsy. These participants performed an image recognition memory task. The data was collected at multiple hospitals, including Northwestern Memorial Hospital, Ann & Robert H. Lurie Children's Hospital of Chicago, and others. No new datasets were introduced.

Key Insights and Ideas

  • Internal Timing is Key: Many memory-related brain signals were only detectable when analyses were anchored to internal brain events (HFB peaks), not just the external stimulus presentation. This suggests that memory processing is significantly driven by internally organized brain states.
  • Transient Brain States: Successful memory encoding and retrieval are supported by a series of short-lived, "sparse" network states, rather than sustained, highly connected networks. This implies a dynamic, moment-to-moment coordination.
  • HFB Peaks as Anchors: HFB peaks serve as crucial temporal anchors, allowing for the precise coordination of theta oscillations and connectivity, which in turn predict memory performance.
  • Network Tuning: When specific connections between brain regions strengthen for successful memory, the overall brain network becomes less connected, suggesting a "tuning" mechanism that focuses resources.

Key Results

  • Regional Timing Differences: HFB peak activity occurred earlier in the medial temporal lobe (MTL) regions (like the hippocampus) than in the prefrontal cortex (PFC) during encoding. During retrieval, PFC activity became earlier.
  • Theta-HFB Coordination: Stronger theta phase clustering, time-locked to HFB peaks, was observed for successful memory trials. This coordination was particularly evident in the PFC.
  • Predictive Connectivity: Connectivity between the PFC and MTL regions, especially when time-locked to HFB peaks, predicted individual memory performance.
  • Sparse Networks for Success: Graph analysis revealed that during successful memory recall, the overall brain network was sparser, with specific connections being strengthened while others were reduced. This pattern was absent when memory failed.
  • Individual Performance Predictors: Phase clustering in the hippocampus and dIPFC, and connectivity between the dIPFC and PHG, predicted individual memory performance.

Limitations and Future Work

The authors acknowledge several limitations:

  • Age and Sex Differences: The study focused on an adolescent and early adult sample. Future research should examine how these mnemonic effects vary across the lifespan and by sex.
  • Limited Brain Regions: The study focused on a specific set of brain regions. Future work should explore interactions with other brain areas.
  • Phase-Amplitude Coupling (PAC): PAC was not analyzed, though the authors argue that their focus on phase clustering at high-power HFB moments effectively captured the relevant relationships.
  • Anatomical Specificity: The analysis used relatively large regions of interest (ROIs) and was agnostic to hemisphere. Finer anatomical distinctions and hemispheric differences could reveal further nuances.

Future research could explore these limitations to gain a more comprehensive understanding of the dynamic neural mechanisms underlying memory.

Summary updated

Abstract

Summary Declarative memory depends on the coordination of local processing, indexed by high-frequency broadband (HFB) activity, with global network organization, indexed by theta oscillations. However, theta and HFB exhibit asynchronous timing, raising the question of how results of local processing are communicated throughout the network. Using intracranial EEG in patients performing a recognition memory task, we examined this coordination across the medial temporal lobe (MTL) and prefrontal cortex (PFC). HFB peak activity was earlier in the MTL than PFC. Anchoring analyses of theta phase clustering and connectivity to HFB peaks revealed strong phase clustering locked to HFB peaks in the PFC, as well as connectivity between the PFC and MTL that predicted individual memory performance. Graph analysis revealed specific connections amidst sparse network connectivity during memory success. This study demonstrates that transient brain states linked to internal physiological events support memory and refines our understanding of local and network-level process interactions. Highlights Memory-linked theta activity is time-locked to internal brain events Network connectivity changes dynamically during memory processing Sparse network connectivity supports successful memory Specific sequences of transient states may be critical for declarative memory
BibTeX
@misc{dede2025declarative,
  author = {Adam J. O. Dede and Zachariah R. Cross and Stephen Gray and Joseph P. Kelly and Qin Yin and Parisa Vahidi and Eishi Asano and Stephan Schuele and Joshua M. Rosenow and Joyce Y. Wu and Sandi Lam and Jeffrey S. Raskin and Jack J. Lin and Olivia Kim McManus and Shifteh Sattar and Ammar Shaikhouni and David King‐Stephens and Peter B. Weber and Kenneth D. Laxer and Peter Brunner and Jarod L. Roland and Ignacio Sáez and Fady Girgis and Robert T. Knight and Noa Ofen and Elizabeth L. Johnson},
  title = {{Declarative Memory Through the Lens of Single-Trial Peaks in High-Frequency Power}},
  year = {2025},
  doi = {10.1101/2025.01.02.631123},
  howpublished = {bioRxiv},
}
Figure