Temporal metastates are associated with differential patterns of time-resolved connectivity, network topology, and attention

James M. Shine, Oluwasanmi Koyejo, Russell A. PoldrackProceedings of the National Academy of Sciences 2016 · DOI

TL;DR

This study reveals that the human brain doesn't maintain a single, static functional connectivity pattern over time. Instead, it fluctuates between distinct "metastates," each characterized by unique patterns of neural communication, network topology, and even self-reported attention levels. These findings highlight the dynamic nature of brain organization and its link to cognitive function over longer timescales.

What Problem Does This Paper Address?

While we know that brain activity is dynamic on short timescales (milliseconds to seconds), how brain-wide functional connectivity changes over much longer periods (weeks to months) and how these long-term shifts relate to behavior remains largely unknown. Understanding this longitudinal brain dynamics is crucial for comprehending development, aging, and the fluctuating symptoms seen in neurological and psychiatric disorders. This paper aims to bridge this gap by investigating if distinct, long-term "states" of brain connectivity exist and if they are associated with observable changes in cognitive function.

Methods

The researchers employed a multi-faceted approach to analyze resting-state fMRI (rfMRI) data collected over extended periods:

  • Time-Averaged Connectivity & Clustering: They first calculated the average functional connectivity between brain regions for each scanning session. Then, they used affinity propagation clustering to group sessions with similar overall connectivity patterns, identifying distinct "metastates."
  • Time-Resolved Functional Connectivity (MTD): To capture rapid fluctuations within each session, they used the Multiplication of Temporal Derivatives (MTD) technique. This method, described as more sensitive to connectivity shifts than standard sliding window correlations, estimates how brain regions' activity patterns change together over short time windows (10 seconds).
  • Community Detection & Network Topology: They applied a multislice community detection algorithm to the time-resolved connectivity data to track how brain regions form and break into functional communities over time. This allowed them to assess:
  • Flexibility: How often a brain region switches between different functional communities.
  • Cartographic Profile: The balance between within-module (segregation) and between-module (integration) connectivity.
  • Global Efficiency: The overall ease of information transfer across the entire brain network.
  • Behavioral Correlation: They correlated the identified metastates with self-reported psychological states, specifically using questions from the Positive and Negative Affect Schedule (PANAS) that were collected on the days of scanning.
  • Null Models: To ensure their findings weren't due to chance or artifacts like head motion, they employed vector autoregressive null models (simulating stationary data) and phase-randomized null models (scrambling temporal relationships).

Datasets

  • Primary Dataset (Discovery Cohort): Resting-state fMRI data from a single individual (R.A.P.) collected over 18 months, comprising 84 usable scanning sessions. This dataset was collected as part of the "MyConnectome Project."
  • Replication Dataset: Resting-state fMRI data from a separate longitudinal study of a single individual ("Kirby"), with 138 usable sessions collected over 4.5 years using a different scanner and protocol.
  • Human Connectome Project (HCP) Dataset: Data from 100 unrelated individuals were used to establish the generalizability of regional flexibility measures.

Key Insights and Ideas

  • Brain Fluctuates Between "Metastates": The most striking finding is the identification of distinct, long-term brain states (metastates) that individuals naturally transition between. This challenges the idea of a single, stable brain configuration over weeks and months.
  • Metastates Have Unique "Personalities": Each metastate isn't just a random fluctuation; it's associated with a specific "signature" of dynamic connectivity, network organization, and even how the individual feels (e.g., fatigued vs. attentive).
  • Flexibility is Key: The degree to which brain regions dynamically switch their connections ("flexibility") is a critical differentiator between these metastates. One state is characterized by higher flexibility in sensorimotor and frontoparietal networks, while the other is more stable.
  • Brain Dynamics Link to Cognition: The observed differences in brain network dynamics are directly related to self-reported psychological states, suggesting that these long-term brain fluctuations have real-world cognitive consequences, such as affecting attention and fatigue.
  • Generalizability: The replication of these metastate patterns in a separate individual with a different dataset strongly suggests that this phenomenon is not an artifact of a specific individual or scanning protocol.

Key Results

  • Two Metastates Identified: Affinity propagation clustering of time-averaged connectivity revealed two distinct metastates that fluctuated significantly more than predicted by chance (P < 0.001).
  • Differential Flexibility: Metastate 2 was associated with significantly higher flexibility in visual, somatomotor, frontoparietal, and cingulo-opercular networks compared to Metastate 1 (FDR P < 0.05).
  • Altered Network Topology: Metastate 2 showed a shift towards higher integration and greater global efficiency (P = 0.002) compared to Metastate 1.
  • Behavioral Correlations: Metastate 1 was associated with self-reported fatigue (drowsy, sleepy, sluggish, tired), while Metastate 2 was linked to heightened attention (attentive, concentrating, lively). These associations were not explained by caffeine/food intake.
  • Replication: The replication dataset showed similar patterns of flexibility, cartography, and network topology differences between its identified metastates, with a moderate spatial correlation (r = 0.360) in flexibility differences.

Limitations and Future Work

  • Single-Subject Focus: While the study used two individuals for longitudinal analysis, the primary findings are based on a single individual. Future work needs to investigate these metastates across larger cohorts to understand individual variability and generalizability.
  • Causality: The study demonstrates associations between metastates and behavior, but it doesn't establish causality. Future research could explore interventions to manipulate these states and observe behavioral changes.
  • Sleep Architecture: The authors acknowledge that while their findings suggest fluctuations aren't simply due to drowsiness, definitive resolution would require simultaneous EEG/fMRI to track sleep architecture.
  • Clinical Implications: The authors suggest that understanding these dynamic shifts could be crucial for tracking disease progression and treatment response in neuropsychiatric disorders where attention and mood fluctuate. Future work should explore how these metastates are altered in clinical populations.

Summary updated

Abstract

Significance The brain is an inherently dynamic organ; however, the manner in which the brain changes over longitudinal time remains poorly understood. An understanding of these dynamic mechanisms is critical for understanding normal childhood development and aging as well as neurological and psychiatric disease states. Here, we leverage data collected in a single individual over the course of 2 y to investigate changes in brain organization over time. In doing so, we show intermittent fluctuations in brain network configuration that were associated with separable patterns of time-resolved interactions between neural regions. The states also directly relate to alterations in whole-brain information processing and self-reported attention. In addition, the patterns that we observed were replicated in a separate longitudinal dataset.
BibTeX
@inproceedings{shine2016temporal,
  author = {James M. Shine and Oluwasanmi Koyejo and Russell A. Poldrack},
  title = {{Temporal metastates are associated with differential patterns of time-resolved connectivity, network topology, and attention}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2016},
  doi = {10.1073/pnas.1604898113},
}
Figure