Altered temporal stability in dynamic neural networks underlies connectivity changes in neurodevelopment

Matthew J. Brookes, Madeleine J. Groom, Lucrezia Liuzzi, Ryan M. Hill, Helen J. F. Smith, Paul M. Briley, Emma L. Hall, Benjamin A. E. Hunt, Lauren E. Gascoyne, Margot J. Taylor, Peter F. Liddle, Peter G. Morris, Mark W. Woolrich, Elizabeth B. LiddleNeuroImage 2018 · DOI

TL;DR

This study uses MEG to investigate how brain connectivity changes from childhood to adulthood, revealing that the way different brain regions communicate becomes more stable and coordinated with age. These findings offer crucial insights into healthy brain development and could help us understand neurodevelopmental disorders.

What Problem Does This Paper Address?

Understanding how the brain develops is fundamental to grasping both healthy function and atypical conditions like autism. While previous research has mapped brain networks, most studies have focused on adults, missing the critical period of change during childhood and adolescence. This gap leaves us with limited knowledge about how brain networks mature. This paper aims to fill this gap by tracking the development of electrophysiological connectivity in the brain from mid-childhood to early adulthood, using magnetoencephalography (MEG) to capture rapid, dynamic changes in neural communication.

Methods

The researchers employed a combination of advanced neuroimaging techniques and sophisticated analytical methods:

  • Magnetoencephalography (MEG): This non-invasive technique measures the magnetic fields produced by electrical activity in the brain, offering excellent temporal resolution to capture rapid network dynamics.
  • Structural MRI: Used for anatomical reference and to ensure accurate source localization of MEG signals.
  • Stationary Functional Connectivity Analysis: This involved calculating the Pearson correlation between the envelopes of oscillatory signals from different brain regions. To mitigate signal leakage (a common artifact in MEG analysis), a multivariate symmetric orthogonalization method was used.
  • Dynamic Network Analysis (Hidden Markov Model - HMM): A key innovation was the application of an HMM to identify distinct "brain states" – recurring patterns of neural activity that fluctuate on a timescale of approximately 100ms. This allowed the researchers to analyze how the brain transitions between different network configurations over time.
  • Statistical Modeling: Linear, non-linear monotonic, and quadratic models were used to assess the relationship between age and various connectivity and amplitude measures. Permutation testing was employed to determine statistical significance.

Datasets

The study utilized resting-state MEG data from 51 healthy participants aged 9 to 25 years. No new datasets or benchmarks were introduced.

Key Insights and Ideas

  • Connectivity Strengthens with Age: The study found that overall brain connectivity, particularly in the alpha and beta frequency bands, significantly increases from childhood to early adulthood. This suggests a maturation process where brain regions become more coordinated.
  • Dynamic Networks Evolve: The HMM analysis revealed 8 distinct brain states, and importantly, the temporal stability of networks associated with attention and cognition increased with age. This implies that the brain becomes more efficient at maintaining specific network configurations as it matures.
  • Dynamics Underlie Connectivity: A crucial finding is the significant overlap between brain regions showing age-related changes in stationary connectivity and those involved in dynamic network states. This strongly suggests that the observed increases in connectivity are driven by changes in the underlying dynamic network activity.
  • Attentional Networks Mature: Specifically, the "temporo-parietal/frontal" state, which is linked to attention and cognition, showed increased stability with age. This suggests that the protracted development of attentional systems is reflected in the evolving dynamics of these brain networks.

Key Results

  • Global Connectivity: Non-linear monotonic increases in connectivity were observed in the alpha (p=0.0006) and beta (p=0.0048) bands. A significant quadratic relationship was found in the theta band, with connectivity peaking around age 18.
  • Oscillatory Amplitude: A negative linear trend was observed for theta amplitude with age (p=0.018), and a quadratic trend for alpha amplitude (p=0.026), though these did not survive multiple comparison correction. Notably, amplitude changes did not consistently mirror connectivity changes, suggesting genuine increases in neural coordination rather than just signal strength.
  • HMM States: Four out of eight HMM states showed significant age-related changes in power, with states involving parietal and temporal regions showing the most pronounced effects. State 6, linked to attention and cognition, demonstrated significant increases in fractional occupancy and state lifetime with age.
  • Spatial Overlap: A significant relationship (p < 0.05) was found between the spatial signature of stationary connectivity change with age and state 6 of the HMM, reinforcing the link between dynamic network changes and overall connectivity maturation.

Limitations and Future Work

The authors acknowledge several limitations:

  • Sample Size: The study's sample of 50 participants is relatively small, suggesting the need for larger studies to confirm findings.
  • IQ Confound: While efforts were made to control for IQ, a weak but significant positive correlation between age and IQ existed, raising the possibility of a confounding effect. However, supplementary analyses indicated that age effects remained significant even when IQ was included as a covariate.
  • Frequency Bands: The use of relatively broad frequency bands (especially for beta and gamma) might be influenced by the lower end of the band due to the 1/f distribution of power. Future work could explore narrower bands, though this would be computationally intensive.
  • MEG Sensitivity: MEG has lower sensitivity to deeper brain sources, potentially leading to underestimated connectivity among these regions.
  • Multivariate Models: Future studies should employ multi-variate models to account for other individual differences (e.g., working memory) that can influence connectivity.
  • Age Range: The youngest participants were 9 years old, an age where many basic functions are already mature. Scanning younger children would provide a more complete picture of early neurodevelopment.
  • MEG Technology: Current MEG systems can face challenges with whole-head coverage and signal-to-noise ratio in very young children due to smaller head sizes. Future research could benefit from non-cryogenic MEG systems.

The authors suggest that future work should focus on understanding the relationship between state dynamics and behavior, and exploring the potential role of myeloarchitecture in mediating the observed changes in connectivity.

Summary updated

BibTeX
@article{brookes2018altered,
  author = {Matthew J. Brookes and Madeleine J. Groom and Lucrezia Liuzzi and Ryan M. Hill and Helen J. F. Smith and Paul M. Briley and Emma L. Hall and Benjamin A. E. Hunt and Lauren E. Gascoyne and Margot J. Taylor and Peter F. Liddle and Peter G. Morris and Mark W. Woolrich and Elizabeth B. Liddle},
  title = {{Altered temporal stability in dynamic neural networks underlies connectivity changes in neurodevelopment}},
  journal = {NeuroImage},
  year = {2018},
  doi = {10.1016/j.neuroimage.2018.03.008},
}
Figure