A cortical hierarchy of localized and distributed processes revealed via dissociation of task activations, connectivity changes, and intrinsic timescales

Takuya Ito, Luke J. Hearne, Michael W. Cole

TL;DR

This research proposes that the "global dimensionality" of a brain network—the degree to which its connections to the rest of the brain are diverse and non-redundant—is the key mechanism behind its ability to represent a wide array of cognitive tasks. By analyzing fMRI data, the researchers demonstrate that the frontoparietal network (FPN) possesses the highest connectivity dimensionality, which directly predicts its superior representational flexibility across various cognitive challenges.

What Problem Does This Paper Address?

Neuroscientists have long known that certain brain regions, particularly in the prefrontal cortex, are "flexible hubs" capable of participating in many different tasks. At the cellular level, this is often attributed to "mixed selectivity," where individual neurons respond to various combinations of stimuli. However, it hasn't been clear how this flexibility emerges at the scale of large-scale brain networks.

The authors address a critical gap: does the way a network is "wired" to the rest of the brain during rest determine its ability to process information during a task? They move beyond simple measures of "hubness" (how many connections a region has) to investigate the geometry of those connections. They hypothesize that for a network to be truly flexible, its various regions must have distinct, non-overlapping connectivity patterns to the rest of the brain, allowing it to integrate diverse information without interference.

The Geometry of Connectivity: Dimensionality vs. Centrality

To quantify this, the researchers developed new graph-theoretic measures: Global Dimensionality and Network Pattern Separation (NPS). These measures quantify how "pattern-separated" a network's out-of-network connections are. High dimensionality means the regions within a network "talk" to very different sets of other brain systems.

Conceptual model of high-dimensional vs. low-dimensional hub networks and their impact on information integration.

As illustrated in Figure 1, a high-dimensional hub network (a) can integrate signals from multiple local networks because its internal nodes have unique connectivity profiles. In contrast, a low-dimensional hub (b) suffers from "information interference" because its nodes share redundant connections, causing task-specific signals to overlap and become indistinguishable.

Methods and Data

The study utilized data from the Human Connectome Project (HCP), specifically the "100 Unrelated Subjects" release. This included resting-state fMRI to estimate intrinsic connectivity and task-state fMRI across seven diverse paradigms: emotion cognition, gambling, language processing, motor tasks, relational reasoning, social cognition, and working memory.

Key computational approaches included:

  • Activity Flow Mapping: A mechanistic model used to predict task-evoked activations by passing activity from the rest of the brain through resting-state connection weights.
  • Representational Flexibility: Measured using Multivariate Pattern Analysis (MVPA) to determine how distinct a network's activation patterns were across different task conditions.
  • Decoding: Logistic regression decoders were used to test if task information could be accurately extracted from specific network activation patterns.

Key Results

  • FPN Dominance: The Frontoparietal Network (FPN) stood out as having both the highest global dimensionality and the highest representational flexibility. It had significantly higher network dimensionality than all other functional networks (averaged t(99) = 20.12).
  • Resting-State as a Blueprint: A major finding is that the dimensionality measured while a person is doing nothing (resting state) significantly correlates with how flexibly their brain represents information during active tasks (mean rho = 0.33).
  • Predictive Power: Each network’s global connectivity pattern could be used to predict its representational flexibility. The activity flow-predicted flexibility scores highly correlated with actual scores (mean rho = 0.71).
  • Individual Differences: The relationship holds true across people. Individuals with higher intrinsic network dimensionality in their FPN tended to show greater representational flexibility across tasks, suggesting a potential neural marker for cognitive adaptability.

Limitations and Future Work

While the study provides a powerful framework, it relies on "activity flow mapping," which assumes a linear relationship between brain regions. Future work should incorporate non-linear models to better capture complex neural processing. Additionally, while the HCP dataset is extensive, the seven tasks are still a small slice of human cognition, and many relied heavily on visual stimuli. Testing on a broader array of sensory and cognitive domains is needed to confirm the universality of these high-dimensional principles.

Summary updated

Abstract

Abstract Many studies have identified the role of localized and distributed cognitive functionality by mapping either local task-related activity or distributed functional connectivity (FC). However, few studies have directly explored the relationship between a brain region’s localized task activity and its distributed task FC. Here we systematically evaluated the differential contributions of task-related activity and FC changes to identify a relationship between localized and distributed processes across the cortical hierarchy. We found that across multiple tasks, the magnitude of regional task-evoked activity was high in unimodal areas, but low in transmodal areas. In contrast, we found that task-state FC was significantly reduced in unimodal areas relative to transmodal areas. This revealed a strong negative relationship between localized task activity and distributed FC across cortical regions that was associated with the previously reported principal gradient of macroscale organization. Moreover, this dissociation corresponded to hierarchical cortical differences in the intrinsic timescale estimated from resting-state fMRI and region myelin content estimated from structural MRI. Together, our results contribute to a growing literature illustrating the differential contributions of a hierarchical cortical gradient representing localized and distributed cognitive processes. Highlights Task activations and functional connectivity changes are negatively correlated across cortex Task activation and connectivity dissociations reflect differences in localized and distributed processes in cortex Differences in localized and distributed processes are associated with differences in intrinsic timescale organization Differences in localized and distributed processes are associated with differences in cortical myelin content Cortical heterogeneity in localized and distributed processes revealed by activity flow mapping prediction error
BibTeX
@article{ito2020cortical,
  author = {Takuya Ito and Luke J. Hearne and Michael W. Cole},
  title = {{A cortical hierarchy of localized and distributed processes revealed via dissociation of task activations, connectivity changes, and intrinsic timescales}},
  journal = {NeuroImage},
  year = {2020},
  doi = {10.1016/j.neuroimage.2020.117141},
}
Figure