Clocks or clouds: movie functional connectivity as a combination of intrinsic and evoked components
Poster No:
1559
Submission Type:
Abstract Submission
Authors:
Ahmad Samara1, Samuel Nastase2, Tamara Vanderwal3
Institutions:
1University of British Columbia, Vancouver, BC, 2Princeton University, Princeton, NJ, 3Department of Psychiatry, University of British Columbia, Vancouver, BC
First Author:
Co-Author(s):
Introduction:
Human brain activity is driven in part by external stimuli and in part by intrinsic fluctuations. Can we decompose this activity into the relative contributions of external and internal events (1,2)? In this reductionist view, a timeseries is more like a clock that can be understood by examining its pieces than a cloud that defies a simple additive understanding. The same concept has been applied to FC by viewing its coupling pattern as arising from invariant constraints and dynamic properties encouraged by the task state (3,4).
We have previously hypothesized that movie-watching causes whole-brain shifts in signal-to-noise ratios (5), such that processing rich, dynamic, multimodal stimuli may yield a "brain state" that is more than the sum of its parts (i.e., it is more cloud than clock). Here, we ask if movie-watching FC could be modeled using a combination of resting-state intrinsic FC and stimulus-driven FC (via intersubject FC, or ISFC). We predict that much of the variance in movie FC would not be explained by this model, but that perhaps more primary visual and auditory regions could be captured l with stronger weightings of the ISFC components.
We have previously hypothesized that movie-watching causes whole-brain shifts in signal-to-noise ratios (5), such that processing rich, dynamic, multimodal stimuli may yield a "brain state" that is more than the sum of its parts (i.e., it is more cloud than clock). Here, we ask if movie-watching FC could be modeled using a combination of resting-state intrinsic FC and stimulus-driven FC (via intersubject FC, or ISFC). We predict that much of the variance in movie FC would not be explained by this model, but that perhaps more primary visual and auditory regions could be captured l with stronger weightings of the ISFC components.
Methods:
All data are from the minimally preprocessed 7T release of the Human Connectome Project (n=174, 104 females, mean age 29.4±3.3). Specifically, we used 4 movie-watching and 4 resting-state runs, each around 15 minutes. Subjects were scanned in 4 sessions over 2 days.
Vertex timeseries were averaged within Schaefer-1000 atlas (6) parcels for each condition (concatenated) and subject, and a functional connectivity (FC) matrix was computed for each condition and subject as the pair-wise Pearson's correlations of parcel timeseries. ISFC matrices were computed as the Pearson's correlation between the timeseries of one parcel from a given subject n and the timeseries of another parcel averaged across all subjects excluding subject n for all pairs or parcels. Since the resulting matrix is asymmetric, each pair of off-diagonal values was replaced by its average.
At each edge, a simple linear model was fitted to predict the edge strength during movie-watching using an intrinsic component (i.e., the rest edge strength) and an evoked component (i.e., the ISFC edge strength) as follows: Y = β0 + β1X1 + β2X2 + ε, where Y is movie edge strength, X1 is rest edge strength, X2 is ISFC edge strength, β0, β1, and β2 are the intercept, X1 weight, and X2 weight, respectively, and ε is a random error term. Additionally, two partial models were fitted at each edge to predict movie edge strength using intrinsic edge strength in partial model 1 or the evoked edge strength in partial model 2.
Vertex timeseries were averaged within Schaefer-1000 atlas (6) parcels for each condition (concatenated) and subject, and a functional connectivity (FC) matrix was computed for each condition and subject as the pair-wise Pearson's correlations of parcel timeseries. ISFC matrices were computed as the Pearson's correlation between the timeseries of one parcel from a given subject n and the timeseries of another parcel averaged across all subjects excluding subject n for all pairs or parcels. Since the resulting matrix is asymmetric, each pair of off-diagonal values was replaced by its average.
At each edge, a simple linear model was fitted to predict the edge strength during movie-watching using an intrinsic component (i.e., the rest edge strength) and an evoked component (i.e., the ISFC edge strength) as follows: Y = β0 + β1X1 + β2X2 + ε, where Y is movie edge strength, X1 is rest edge strength, X2 is ISFC edge strength, β0, β1, and β2 are the intercept, X1 weight, and X2 weight, respectively, and ε is a random error term. Additionally, two partial models were fitted at each edge to predict movie edge strength using intrinsic edge strength in partial model 1 or the evoked edge strength in partial model 2.
Results:
Linear modeling of movie FC using intrinsic resting state and movie-evoked connectivity values at each edge explained more than 50% of the variance at 26.4% of edges. The amount of R2 across all edges followed the first movie gradient hierarchy (7), with models at edges involving dorsal attention, frontoparietal, and default networks demonstrating the highest amounts of explained variance and sensorimotor network edges demonstrating the lowest. Intrinsic edge strength explained more variance in higher-order association cortex, whereas ISFC explained more variance in visual and auditory regions. The variability (i.e.standard deviation) of movie-FC edge strength itself increased along the cortical hierarchy and was strongly correlated with explained variance across subjects (r = 0.71, p < 0.001).
Conclusions:
- In 1/4 of the edges in the brain, a basic linear model combining rest FC and movie-based ISFC can predict more than half the variance observed in a movie FC matrix.
- Intersubject variability of movie-FC inherently constrains the variance explained, and both variability of movie-FC and R2 of the model follow hierarchical brain organization during movie-watching.
- Ongoing work is testing these findings across time within a movie, and against task-based FC matrices.
- Intersubject variability of movie-FC inherently constrains the variance explained, and both variability of movie-FC and R2 of the model follow hierarchical brain organization during movie-watching.
- Ongoing work is testing these findings across time within a movie, and against task-based FC matrices.
Emotion, Motivation and Social Neuroscience:
Social Neuroscience Other
Higher Cognitive Functions:
Higher Cognitive Functions Other
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
fMRI Connectivity and Network Modeling 2
Keywords:
Acquisition
Cortex
FUNCTIONAL MRI
Modeling
Other - Naturalistic
1|2Indicates the priority used for review
Provide references using author date format
2. Nastase, Samuel A et al. “Measuring shared responses across subjects using intersubject correlation.” Social cognitive and affective neuroscience vol. 14,6 (2019): 667-685. doi:10.1093/scan/nsz037
3. Buckner, Randy L et al. “Opportunities and limitations of intrinsic functional connectivity MRI.” Nature neuroscience vol. 16,7 (2013): 832-7. doi:10.1038/nn.3423
4. Lynch, Lauren K et al. “Task-evoked functional connectivity does not explain functional connectivity differences between rest and task conditions.” Human brain mapping vol. 39,12 (2018): 4939-4948. doi:10.1002/hbm.24335
5. Vanderwal, Tamara et al. “Individual differences in functional connectivity during naturalistic viewing conditions.” NeuroImage vol. 157 (2017): 521-530. doi:10.1016/j.neuroimage.2017.06.027
6. Schaefer, Alexander et al. “Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI.” Cerebral cortex (New York, N.Y. : 1991) vol. 28,9 (2018): 3095-3114. doi:10.1093/cercor/bhx179
7. Samara, Ahmad et al. “Cortical gradients during naturalistic processing are hierarchical and modality-specific.” NeuroImage vol. 271 (2023): 120023. doi:10.1016/j.neuroimage.2023.120023