Arousal modulates functional connectivity through structured and hemispherically asymmetric community architecture during wakefulness
- State Key Laboratory of Cognitive Neuroscience and Learning and IDG/McGovern Institute for Brain Research, Beijing Normal University, China
- Beijing Key Laboratory of Brain Imaging and Connectomics, Beijing Normal University, China
- Chinese Institute for Brain Research, China
Figures
Figure 1
Pipeline for estimating arousal–time varying functional connectivity (tvFC) coupling from fMRI and pupillometry.
(A) Concurrent 7T fMRI and eye tracking were collected during resting state and naturalistic movie watching. (B) tvFC was computed over sliding windows for each pair of brain regions. (C) Pupil diameter was preprocessed to obtain a continuous arousal time series. Arousal–tvFC coupling for each connection was then defined as the Pearson correlation between its windowed functional connectivity (FC) time course and the corresponding arousal fluctuations. (D) These edgewise coupling values yielded a dense arousal–tvFC coupling matrix per run, which served as the input for analyses of community structure, hemispheric asymmetry, and cross-paradigm consistency. (E) The connections were categorized based on the hemispheres of the connected regions: Left-Left (LL) and Right-Right (RR) represent intra-hemispheric connections; Left-Right (LR) and Right-Left (RL) represent inter-hemispheric connections. This classification enabled hemisphere-specific analyses of arousal-tvFC coupling.
Figure 2 with 1 supplement
Arousal–time varying functional connectivity (tvFC) coupling partitions the connectome into seven distinct connectivity communities.
(A) Unsupervised clustering of edgewise arousal–tvFC coupling identified seven stable communities, demonstrating that arousal-linked modulation is organized into low-dimensional structure rather than uniformly distributed across connections. (B) Mapping communities onto network-pair space revealed distinct and reproducible composition profiles (upper panel), with some communities dominated by heteromodal interactions and others enriched in H–U or U–U network pairs (bottom panel). (C) Community participation entropy varied systematically across connection modal types: U–U network pairs showed lower entropy, indicating concentrated engagement in a small subset of communities, whereas H-H and H–U network pairs exhibited significantly higher entropy (F(2,88)=12.24, p=4.38×10–6), reflecting broader distribution across communities. (D) Dominant-community assignments confirmed this organization, showing that heteromodal interactions load onto multiple arousal-sensitive communities, whereas U–U network pairs show more restricted community involvement.
Figure 2—figure supplement 1
Reliability of community architecture across different resampling strategies.
(A) Confusion matrix (left) showing the consistency between community labels from 500 iterations (aligned via the Hungarian algorithm) and the primary results. The right panel displays the distribution of Dice coefficients for each individual community label. (B) Stability of community architecture using participant-level resampling, following the same analytical framework and visualization as in (A).
Figure 3 with 1 supplement
Arousal-modulated communities architecture exhibit community-specific hemispheric asymmetry.
(A) Integration indices for each network pair revealed significant leftward or rightward deviations from a spatial permutation null, indicating that arousal differentially modulates between- and within-hemisphere interactions for specific network pairs. (B) Segregation indices identified network pairs showing hemisphere-specific strengthening of within-hemisphere connectivity, further demonstrating that lateralization is localized rather than global. (C–D) Community-averaged integration and segregation values showed that hemispheric biases vary across communities and are not determined solely by connection modal type, underscoring the community-specific nature of the asymmetry. (E) Aggregating all significantly lateralized network pairs, each community exhibited a distinct integration–segregation profile, revealing unique hemispheric signatures across communities. (F) Low similarity among communities’ integration and segregation patterns confirmed that arousal imposes multiple, community-specific forms of hemispheric asymmetry, rather than a single unified left- or right-dominant pattern.
Figure 3—figure supplement 1
Reliability of network-pair specific hemispheric asymmetry in community architecture across resampling strategies.
(A) The relative value was calculated as:, where represents the mean of 500 resampling iterations and， denotes the original findings. The top and bottom rows represent integration and segregation patterns, respectively. (B) Stability assessment using participant-level resampling, following the same calculation and visualization framework as in (A).
Figure 4 with 1 supplement
Characterizing the spatial distribution, entropy, and hemispheric divergence of regional community affiliation.
(A–B) Nodal-level community affiliation matrices, computed separately for Left-Left (LL), Left-Right (LR), Right-Left (RL), and Right-Right (RR) edges, showed substantial heterogeneity in how nodes distribute their arousal–time varying functional connectivity (tvFC) coupling across the seven communities, with distinct patterns emerging across canonical networks. For visualization purposes, panel A is restricted to displaying values where the proportion exceeds 0.2. (C–D) Region-level community affiliation entropy revealed a systematic network gradient, in which heteromodal systems displayed more selective participation, whereas unimodal networks showed broader, more distributed engagement across communities (t(798) = –23.81, p=4.24×10⁻95) (E) No integration bias was detected in any community. (F) Significant leftward segregation biases were identified within specific communities (communities 3, 5, 6, and 7). These asymmetries were primarily localized in regions belonging to the DMN, FPN, LIMB, and VAN. The color of each box corresponds to the community identity. Statistical significance: * p_FDR <0.01; ** p_FDR <0.001.
Figure 4—figure supplement 1
Reliability of regional affiliation asymmetry across resampling strategies.
(A) Stability analysis using split-half resampling. Boxplots represent the distribution of asymmetry indices across 500 iterations, where colored diamonds indicate the original findings. (B) The heatmap background represents the absolute deviation (Δ) between the discovery sample and the resampled mean. Numerical annotations indicate the relative error percentage (%). To prevent the statistical inflation of relative values in regions with minimal effects, these annotations are explicitly omitted where the is low (<0.05), focusing the assessment on robustly lateralized regions. (C–D) Stability assessment using participant-level resampling, following the same quantitative framework. The overall tight alignment between resampled distributions and original discovery values confirms that the spatial patterns of regional hemispheric bias are highly stable and not artifacts of specific sample selection.
Figure 5 with 1 supplement
Spatial heterogeneity—not mean shifts—drives arousal-modulated hemispheric asymmetry.
(A) Mean integration and segregation values showed no consistent hemispheric bias across individual communities or at the whole-brain (WB) level, indicating minimal imbalance in overall modulation strength. (B) Spatial heterogeneity, quantified as the slope of integration or segregation values ranked along a lateralization axis, was significantly steeper in every community compared with the whole-brain baseline (all p_FDR <0.0001). These effects demonstrate that hemispheric asymmetry arises from spatially patterned variation, not from uniform shifts in mean modulation. (C) Leave-one-out analyses revealed that this heterogeneity reflects distributed contributions from many network pairs, rather than being driven by a few extreme edges. (D) Contribution patterns for integration and segregation were strongly correlated (rho ≈0.72, p=2.35×10–26), indicating coordinated spatial organization across metrics. (E–F) Similarity in contribution patterns between communities was positively associated with similarity in communities’ intrinsic structure (integration: rho ≈ 0.79, p=1.35×10–11; segregation: rho ≈0.64, p=5.82×10–7), showing that spatial heterogeneity is constrained by each community’s underlying connectivity architecture. Statistical note: *p_FDR <0.01; **p_FDR <0.001; ***p_FDR <0.0001.
Figure 5—figure supplement 1
Reliability of arousal-functional connectivity (FC) coupling asymmetry across resampling strategies.
(A) Reliability assessment using split-half resampling. Boxplots illustrate the distribution of asymmetry indices for arousal-FC coupling across 500 iterations, with diamonds indicating the original discovery findings. (B) The heatmap background represents the absolute deviation (Δ) between the discovery sample and the resampled mean. Numerical annotations indicate the relative error percentage (%). To prevent the statistical inflation of relative values in network pairs with minimal effects, these annotations are explicitly omitted where the is low (<0.05), focusing the assessment on robustly lateralized patterns. (C–D) Reliability assessment using participant-level resampling, following the same quantitative framework. The consistent alignment between the resampled distributions and primary findings confirms that the hemispheric asymmetry of arousal-time varying functional connectivity (tvFC) coupling is a robust biological feature rather than a result of sampling bias or specific data partitions.
Figure 6
Community structure and hemispheric asymmetry of arousal–time varying functional connectivity (tvFC) coupling are preserved across resting state and movie watching.
(A) Communities derived from rest and movie data were aligned using the Hungarian algorithm, revealing robust correspondence between paradigms (average dice ≈0.46). (B) Network-pair composition profiles for each community were strongly correlated across paradigms (mean rho ≈0.58), indicating that the modular organization of arousal–tvFC coupling is stable across cognitive contexts.
Appendix 2—figure 1
Robustness of community detection across sliding-window parameters and temporal lags.
(A) Dice coefficients between the community topologies derived from different combinations of window size, window step, and temporal lag. The top heatmap illustrates the Dice coefficients for the overall community partition, while the bottom heatmaps show the coefficients calculated independently for each individual community. (B–D) Marginal consistency matrices for specific parameters. These panels display the Dice coefficients averaged across all other parameters while retaining only (B) window size, (C) temporal lag, and (D) window step.
Appendix 3—figure 1
Robustness of community architecture after controlling for motion and global signal artifacts.
(A) Dice coefficients between the original community templates and those derived after regressing out nuisance covariates (Framewise Displacement and Global Signal) across various window steps and temporal lags. The top heatmap shows the Dice coefficients for the overall community partition, while the bottom heatmaps display the coefficients for each individual community. (B) Marginal Dice coefficient matrix averaged across all parameters except for the temporal lag. (C) Marginal Dice coefficient matrix averaged across all parameters except for the window step.
Appendix 4—figure 1
Evaluation and selection of the optimal number of communities.
(A) Evaluation curves for four complementary clustering quality metrics across K=2–15: Within-Cluster Sum of Squares (WCSS), Davies-Bouldin Index, Calinski-Harabasz Score, and Silhouette Coefficient. (B) Objective determination of the ‘knee’ point using the L-method. The plot shows the total root mean squared error (RMSE) of the two-line linear regression fit as a function of K, with the minimum RMSE achieved at K=7 (indicated by the arrow). (C) Spatial distribution of community labels across k=5–9. The visualization demonstrates the hierarchical evolution and spatial stability of the community topologies as the number of clusters increases.
Appendix 5—figure 1
Robustness of pupil diameter time courses across different preprocessing pipelines.
Line plots illustrate the Pearson correlation coefficients between the pupil diameter time courses derived from the original pipeline and those from 18 alternative preprocessing combinations. These pipelines varied across three parameters: (i) smoothing window size (100 ms, 200 ms, and 500 ms); (ii) interpolation method (linear or cubic spline); and (iii) blink buffer duration (25 ms, 50 ms, and 100 ms). Each data point represents the mean correlation across all fMRI runs, with error bars indicating ±1 standard deviation (SD).
Appendix 6—figure 1
Participant-level data retention and run distribution after quality control.
(A) Summary of valid runs per participant. The bar chart shows the number of participants who contributed 1, 2, 3, or 4 fMRI runs to the final analysis (N=139 participants). (B) Individual run availability matrix. The heatmap illustrates the distribution of retained runs across the four sessions (REST1 to REST4) for each participant. Each row represents an individual participant, and columns represent the four scanning sessions. Colored cells indicate that the specific run passed all quality control (QC) criteria, while white cells indicate excluded or unavailable runs.
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Arousal modulates functional connectivity through structured and hemispherically asymmetric community architecture during wakefulness
eLife 15:RP110294.
https://doi.org/10.7554/eLife.110294.3
