Episodic memory formation involves the integration of information about an experience's content and context into a unique memory trace. Each experience recruits different brain regions and temporal dynamics; encoding different experiences requires distinct communication patterns between brain regions, which dynamically change. While prior research has explored large-scale brain connectivity and its relevance to memory, examining the dynamic changes underlying memory formation and retrieval has been challenging due to limitations in temporal resolution. Studies using EEG, particularly intracranial EEG (iEEG), have begun to address this, linking dynamic sub-second changes in connectivity to memory. However, interpreting these dynamic modulations across different events remains unclear. This study aims to address this by examining how patterns of dynamic changes in connectivity are specific to individual events and reinstated during retrieval. The researchers hypothesized that specific dynamic fluctuations in functional connectivity occur during encoding, and these patterns are reinstated during successful retrieval.

Previous research has investigated large-scale brain connectivity and its role in memory, but examining the dynamic changes underlying memory formation and retrieval has proven difficult due to limitations in temporal resolution of neuroimaging techniques. Studies using scalp EEG and iEEG have offered insights into moment-to-moment connectivity fluctuations during encoding and retrieval, particularly in theta and high gamma frequencies. However, interpreting the dynamic modulations during individual memory events remains a challenge. The heterogeneity of these dynamic changes, requiring data aggregation across trials, obscures moment-to-moment fluctuations specific to individual memories. This study aimed to address this limitation by examining whether patterns of dynamic changes in connectivity are specific to individual events and whether these patterns are reinstated during memory retrieval. Prior work has shown that dynamic changes in oscillatory phase in local brain regions are reinstated during successful retrieval, suggesting the possibility of similar reinstatement of connectivity.

Twenty participants (8 female; mean age 33.40 years) with drug-resistant epilepsy, implanted with intracranial EEG electrodes, performed a paired associates verbal memory task. The task involved studying word pairs (encoding) and later recalling one word upon presentation of the other (retrieval). A windowed-scaled cross-correlation measure identified potential functional connections between electrode pairs exhibiting strong and consistent correlations with precise time delays across random recording segments. This selected electrode pairs demonstrating reliable functional connectivity. The coupling strength between these pairs was then calculated using 1000-ms sliding time windows across encoding and retrieval periods, normalized to a baseline period. To investigate reinstatement, feature vectors containing normalized coupling strength values for each connected pair were created for every time point during encoding and retrieval. Cosine similarity between these feature vectors quantified the extent to which dynamic coupling patterns were reinstated during retrieval. Spectral power reinstatement was similarly assessed using feature vectors containing oscillatory power from each electrode across five frequency bands (theta, alpha, beta, low gamma, high gamma). The relative contribution of each electrode to the reinstatement of both coupling and spectral power was computed using a leave-one-out approach. Finally, statistical analyses, including paired-sample t-tests, repeated-measures ANOVA, and resampling analyses (permutation tests with cluster-wise correction) were used.

The study found that successful memory formation involved dynamic sub-second changes in functional connectivity specific to each word pair. Electrode pairs showed dynamic increases or decreases in coupling strength relative to baseline, with similar dynamics during successful retrieval. The magnitude of changes in coupling strength was significantly greater during the encoding period of correct trials (400-3200 ms after stimulus onset) compared to incorrect trials (p < 0.05, cluster-based permutation test). Patterns of dynamic functional connectivity during encoding were significantly reinstated during successful memory retrieval (p < 0.05, cluster-based permutation test), peaking around 1360 ± 200 ms after study onset and -550 ± 170 ms relative to vocalization. This reinstatement was specific to individual word pairs, exceeding that observed with shuffled or adjacent correct trial labels. Importantly, reinstatement of dynamic functional connectivity and spectral power were found to be separable. Analyses comparing the relative contribution of electrodes to reinstatement of coupling versus spectral power showed no systematic correlation (average Fisher's transformed r = 0.021, p = 0.42). Dividing electrodes based on their contribution to coupling reinstatement significantly affected coupling reinstatement but not spectral power reinstatement, and vice versa. These findings demonstrate that the reinstatement of dynamic functional connectivity and local spectral power are independent processes.

The findings demonstrate that functional connections between brain regions are dynamically modulated to support the remembering of specific events. Every memory event involves unique associations, and representing these associations may involve distinct patterns of functional connectivity. Successful memory retrieval involves significant reinstatement of dynamic connectivity patterns, exceeding that of unsuccessful retrieval. The specificity of this reinstatement to individual word pairs indicates it's not a generic mechanism. The study complements previous findings on the reinstatement of local neural activity during memory retrieval. The results suggest that reinstating dynamic patterns of cortical connectivity efficiently recapitulates the initial neural representation of a past event, enhancing successful remembering. The observed patterns of dynamic connectivity provide specific information about the association being remembered, complementing the information contained in local oscillatory power. The study highlights the importance of considering both local and large-scale network dynamics in understanding episodic memory.

This study provides direct evidence that dynamic changes in specific patterns of functional connectivity are crucial for successful episodic memory formation and retrieval. The reinstatement of these patterns is specific to individual memories and separable from the reinstatement of local spectral power. Future research should investigate the specific mechanisms underlying these dynamic connectivity changes and explore how these findings can be translated to enhance memory function and treatment of memory disorders.

The study population consisted of epilepsy patients with implanted electrodes, potentially limiting the generalizability of the findings to the broader population. The specific placement of electrodes, determined by clinical needs, might influence the observed patterns of connectivity. The paired associates task, while a common paradigm, may not fully capture the complexity of real-world memory formation. Future studies should investigate a wider range of tasks and populations.