Source-linked AI summary
A mechanistic model of connector hubs, modularity, and cognition
Maxwell A. Bertolero, B. T. T. Yeo, Danielle S. Bassett, Mark D'Esposito
TL;DR
It was unknown whether hub connectivity could mechanistically explain cognition and whether one network organization supports diverse tasks. Using individual differences in hub connectivity and cognition, the study models this relationship across four tasks and finds that diversely connected connector hubs and modular brain networks consistently predict higher performance.
Problem
Direct evidence linking hub organization mechanistically to cognition, and whether one optimal hub structure supports diverse tasks, remained lacking.
Method
The study analyzes individual differences in hub connectivity, modularity, and performance, testing whether connector hubs tune neighboring connectivity to increase modularity.
Results
The model significantly predicted performance in all four tasks, while diverse connector hubs were associated with higher modularity and task performance across tasks.
Takeaways & Limitations
Connector hubs appear to help maintain an optimal modular architecture during integrative cognition by tuning neighbors’ connectivity while supporting information integration.
Takeaways & Limitations
The study does not establish how its findings interact with connectivity patterns optimized for a single task.
Abstract
from arXiv · showhide
The human brain network is modular--comprised of communities of tightly interconnected nodes. This network contains local hubs, which have many connections within their own communities, and connector hubs, which have connections diversely distributed across communities. A mechanistic understanding of these hubs and how they support cognition has not been demonstrated. Here, we leveraged individual differences in hub connectivity and cognition. We show that a model of hub connectivity accurately predicts the cognitive performance of 476 individuals in four distinct tasks. Moreover, there is a general optimal network structure for cognitive performance--individuals with diversely connected hubs and consequent modular brain networks exhibit increased cognitive performance, regardless of the task. Critically, we find evidence consistent with a mechanistic model in which connector hubs tune the connectivity of their neighbors to be more modular while allowing for task appropriate information integration across communities, which increases global modularity and cognitive performance.
Abstract · Main
Across 476 individuals, hub diversity and locality, network modularity, and connectivity predicted performance across four cognitive tasks. Diverse connector hubs were associated with more modular networks and higher cognition, consistent with a mechanism in which they tune neighbors’ connectivity to support integration across communities.
- Main: The analysis leveraged 476 Human Connectome Project subjects with networks constructed from fMRI during seven cognitive states.Each subject contributed networks based on 264 brain nodes, including resting state and task conditions.
- Main: A model of hub diversity, locality, modularity, and connectivity significantly predicted performance in all four measured cognitive tasks.The model was fit to task and resting-state networks; prediction remained strong when individual connections were omitted, retaining hub and modularity features.
- Main: Hub and network structures optimal for individual tasks were typically also optimal for positive subject measures and suboptimal for negative measures.Similarity across tasks was assessed by correlating how their optimal structures generalized to subject measures outside the scanner.
- Main: Connector hubs’ diversity-facilitated modularity coefficients exceeded those of other nodes across all seven cognitive states.Connector hubs were defined as the top 20 percent highest-participation-coefficient nodes; the Working Memory comparison was t(dof:262):7.182, p<0.001, Cohen's d:1.104, CI:0.062,0.117.
- Main: Connector hubs generally had higher diversity-facilitated performance coefficients than other nodes, while Language & Math was not significant after correction.The Language & Math result was p=0.0677 after Bonferroni correction and uncorrected p=0.0169; higher participation coefficient was associated with higher task performance.
- Main: Mediation analyses supported a mechanism in which diverse connector hubs tune connectivity to increase segregation between sensory, motor, and attention systems, thereby increasing global modularity.The analyses tested whether neighbors’ edge patterns mediated the relationship between connector-hub participation coefficients and network modularity.
- Main: Across individuals, the most diversely connected connector hubs were associated with the highest modularity in all seven tasks and the highest performance in the four measured tasks.The findings support a generally optimal hub and network structure in which connector hubs integrate information while tuning neighbors’ connectivity to be more modular.
Methods
The study preprocessed task-based fMRI, represented each participant’s brain as a 264-node Power-atlas graph, and quantified modular organization using community-based network measures. Analyses were conducted separately by task, including an alternative test of whether connector hubs modulate neighbors’ connectivity toward greater modularity.
- Data and preprocessing: Task-based fMRI was preprocessed with AFNI, including regression of cerebrospinal-fluid, white-matter, whole-brain, and motion signals.High-motion frames were scrubbed when frame-wise displacement exceeded 0.2 millimeters, including the adjacent frames.
- Graph theory analyses: The Power atlas defined 264 graph nodes because it combines high node homogeneity, functional and task-activation foundations, canonical communities, and broad anatomical coverage.The canonical community division supported interpretation, generalizability, and within- versus between-community edge-weight analyses.
- Graph theory analyses: For each subject and task, regional mean signals were correlated pairwise to form a 264 by 264 matrix, then Fisher z transformed.Both left-to-right and right-to-left encoding directions were used, with custom Python code based on iGraph.
- Graph theory analyses: Participation coefficients, within-community strengths, and modularity Q were calculated across costs, averaged per subject, and modeled separately for each task.Participation coefficient quantified how evenly a node’s edges were distributed across communities, whereas within-community strength measured relative connectivity within its community.
- Alternative connector-hub analysis: An alternative analysis correlated each node’s participation coefficients with neighbors’ edge weights across subjects to test whether connector hubs tune connectivity toward greater modularity.The corresponding local-hub and non-local-hub correlations were not robust (-0.1>r<0.1), supporting the study’s connector-hub interpretation.
Statistical Methods
The analyses used Human Connectome Project sample sizes available at study onset, without a power analysis, and applied standardized confidence-interval and multiple-comparison procedures. Pearson correlations focused on individual differences in connectivity and their relationships with cognition.
- Sample size: 475 Working Memory, 473 Gambling, 458 Relational, 475 Motor, 472 Language & Math, 474 Social, and 476 Resting State subjects were reported.The sample was determined by subjects released by the Human Connectome Project; no power analysis was computed.
- Confidence intervals: Confidence intervals used alpha=0.05; Pearson r intervals were Fisher-transformed to z and reverse-transformed to r.For t-tests, intervals represented the largest and smallest differences in means across the two distributions.
- Multiple comparisons: Family-wise p values were Bonferroni corrected, with 47 tests when comparing task hub and network optimality across subject measures.The correction assessed whether effects remained true for particular subject measures.
- Correlation analyses: Pearson r values quantified individual differences in functional connectivity and their relationships with individual differences in cognition.The same correlation framework was extended to compare nodes’ participation coefficients across subjects.
Data availability
The study used data provided by the Human Connectome Project, funded by NIH Blueprint-supported institutes and the McDonnell Center for Systems Neuroscience.
- Data availability: Data were provided by the Human Connectome Project, WU-Minn Consortium, with funding from 16 NIH Institutes and Centers and the McDonnell Center for Systems Neuroscience.The authors state that the content is solely their responsibility and does not necessarily represent official views of the funding agencies.
Code availability
The study’s custom graph-theory analyses used Python and the iGraph library, with all analysis code publicly available in two GitHub repositories.
- Code availability: Custom Python code using the iGraph library executed all graph-theory analyses.The code is available at www.github.com/mb3152/brain_graphs.
- Code availability: All analysis code is publicly available in the hcp_performance GitHub repository.The repository is identified as github.com/mb3152/hcp_performance/.