Functional network integration mediates arousal effects on naturalistic recall
Poster No:
1705
Submission Type:
Abstract Submission
Authors:
Jadyn Park1, Jin Ke1, Kruthi Gollapudi1, Ioannis Pappas2, Yuan Chang Leong1
Institutions:
1The University of Chicago, Chicago, IL, 2Keck School of Medicine, University of Southern California, Los Angeles, CA
First Author:
Co-Author(s):
Introduction:
A consistent finding in memory research is that arousing stimuli are more likely to be remembered than neutral ones. Yet, the neural mechanisms underlying how arousal supports memory are not fully understood. Fluctuations in arousal have been found to covary with functional network integration across the entire brain (Shine et al., 2016), which, in turn, is associated with memory encoding performance (Keerativittayayut et al., 2018). Here, we tested whether increases in functional network integration is a mechanism by which events with higher arousal are more strongly encoded and better remembered.
Methods:
We used two publicly available fMRI datasets: "Sherlock" (n=17; Chen et al., 2017) and "FilmFest" (n=15; Lee & Chen, 2022). In both studies, participants watched hour-long movie clips immediately followed by a free verbal recall session in the scanner. For analysis, we identified 48 and 68 events from each respective dataset.
We parcellated the cortical and subcortical regions into 216 ROIs by combining the Schaefer and the Melbourne subcortical atlas. We then constructed an unweighted, undirected graph from functional connectivity matrices for each participant and event. Metrics of functional network integration (global efficiency, participation coefficient) and segregation (modularity) were extracted to examine how the dynamic reorganization of their structure during encoding supports memory. To obtain a continuous measure of recall performance, we converted movie annotations and transcriptions of participants' recall to text embeddings using Google's Universal Sentence Encoder. We then calculated the memory fidelity as the cosine similarity between the movie and recall embeddings for each event--the higher the fidelity score, the better the participants recalled the event with accuracy and detail.
To estimate the arousal level for each event, we used an open-access Large Language Model (LLM), StableBeluga-13B. We provided the model with detailed annotations on each event and prompted it to rate the arousal level of the event on a scale of 1 to 10. The model-generated ratings were validated against human ratings (n=30 for each experiment).
We parcellated the cortical and subcortical regions into 216 ROIs by combining the Schaefer and the Melbourne subcortical atlas. We then constructed an unweighted, undirected graph from functional connectivity matrices for each participant and event. Metrics of functional network integration (global efficiency, participation coefficient) and segregation (modularity) were extracted to examine how the dynamic reorganization of their structure during encoding supports memory. To obtain a continuous measure of recall performance, we converted movie annotations and transcriptions of participants' recall to text embeddings using Google's Universal Sentence Encoder. We then calculated the memory fidelity as the cosine similarity between the movie and recall embeddings for each event--the higher the fidelity score, the better the participants recalled the event with accuracy and detail.
To estimate the arousal level for each event, we used an open-access Large Language Model (LLM), StableBeluga-13B. We provided the model with detailed annotations on each event and prompted it to rate the arousal level of the event on a scale of 1 to 10. The model-generated ratings were validated against human ratings (n=30 for each experiment).
Results:
LLM-generated arousal ratings were correlated with human ratings (Sherlock: r=.37, p=.02; FilmFest: r=.41, p=.006), demonstrating the model's consistency with human subjective ratings. Across both datasets, global efficiency at encoding was positively associated with subsequent recall fidelity (Sherlock: β=.19, SE=.03, t(809)=5.66, p<.001; FilmFest: β=.16, SE=.03, t(1017)=5.26, p<.001), suggesting that events associated with greater functional network integration were more likely to be remembered with greater accuracy and detail (Fig.1). Arousal was associated with both global efficiency and recall fidelity, such that highly arousing events coincided with greater network integration (Sherlock: β=.22, SE=.04, t(796)=4.91, p<.001; FilmFest: β=.21, SE=.04, t(1004)=5.33, p<.001) and higher recall fidelity (Sherlock: β=.20, SE=.04, t(796)=4.57, p<.001; FilmFest: β=.19, SE=.04, t(1004)=4.65, p<.001).
Functional network integration mediated the effects of arousal on recall (Sherlock:β=.04, 95%CI=[.02, .06], p<.001; FilmFest: β=.03, 95%CI=[.02, .05], p<.001) (Fig.2). Importantly, the effects of brain-wide integration on recall fidelity were driven by inter-modular integration across multiple networks (e.g., DMN, attention, control) (Sherlock: β=.11, SE=.03, t(808.3)=3.16, p<.01; FilmFest: β=.10, SE=.03, t(1017)=3.14, p<.01). Conversely, modularity was not associated with recall fidelity, indicating that network segregation was not related to memory performance.
Functional network integration mediated the effects of arousal on recall (Sherlock:β=.04, 95%CI=[.02, .06], p<.001; FilmFest: β=.03, 95%CI=[.02, .05], p<.001) (Fig.2). Importantly, the effects of brain-wide integration on recall fidelity were driven by inter-modular integration across multiple networks (e.g., DMN, attention, control) (Sherlock: β=.11, SE=.03, t(808.3)=3.16, p<.01; FilmFest: β=.10, SE=.03, t(1017)=3.14, p<.01). Conversely, modularity was not associated with recall fidelity, indicating that network segregation was not related to memory performance.
Conclusions:
The study establishes arousal-dependent biases in memory to dynamic changes in the integration of functional brain networks. Combining the approaches from systems and affective neuroscience, our work contributes to building a theoretical framework that bridges affective states, ongoing cognition, and functional network topology.
Emotion, Motivation and Social Neuroscience:
Emotion and Motivation Other
Learning and Memory:
Long-Term Memory (Episodic and Semantic) 2
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Keywords:
Cognition
FUNCTIONAL MRI
Memory
Other - arousal; network neuroscience
1|2Indicates the priority used for review
Provide references using author date format
Keerativittayayut, R., Aoki, R., Sarabi, M. T., Jimura, K., & Nakahara, K. (2018). Large-scale network integration in the human brain tracks temporal fluctuations in memory encoding performance. eLife, 7, e32696. https://doi.org/10.7554/eLife.32696
Lee, H., & Chen, J. (2022). Predicting memory from the network structure of naturalistic events. Nature Communications, 13(1), 4235. https://doi.org/10.1038/s41467-022-31965-2
Shine, J. M., Bissett, P. G., Bell, P. T., Koyejo, O., Balsters, J. H., Gorgolewski, K. J., Moodie, C. A., & Poldrack, R. A. (2016). The Dynamics of Functional Brain Networks: Integrated Network States during Cognitive Task Performance. Neuron, 92(2), 544–554. https://doi.org/10.1016/j.neuron.2016.09.018