[Paper Review] Inter-regional ECoG correlations predicted by communication dynamics, geometry, and correlated gene expression

Richard F. Betzel, John D. Medaglia+4 more|arXiv (Cornell University)|Jun 19, 2017

Electrochemical Analysis and Applications18 citations

TL;DR

This study constructs whole-brain, parcellation-based functional connectivity networks from multi-subject intracranial EEG (ECoG) recordings and demonstrates that inter-regional ECoG correlations are predictably shaped by anatomical connectivity, interregional distance, and gene co-expression patterns. The model accurately predicts ECoG networks across individuals, with a subset of genes—particularly those involved in ion channel and membrane potential regulation—strongly linked to functional connectivity, revealing a molecular basis for brain network organization.

ABSTRACT

Electrocorticography (ECoG) provides direct measurements of synchronized postsynaptic potentials at the exposed cortical surface. Patterns of signal covariance across ECoG sensors have been associated with diverse cognitive functions and remain a critical marker of seizure onset, progression, and termination. Yet, a systems level understanding of these patterns (or networks) has remained elusive, in part due to variable electrode placement and sparse cortical coverage. Here, we address these challenges by constructing inter-regional ECoG networks from multi-subject recordings, demonstrate similarities between these networks and those constructed from blood-oxygen-level-dependent signal in functional magnetic resonance imaging, and predict network topology from anatomical connectivity, interregional distance, and correlated gene expression patterns. Our models accurately predict out-of-sample ECoG networks and perform well even when fit to data from individual subjects, suggesting shared organizing principles across persons. In addition, we identify a set of genes whose brain-wide co-expression is highly correlated with ECoG network organization. Using gene ontology analysis, we show that these same genes are enriched for membrane and ion channel maintenance and function, suggesting a molecular underpinning of ECoG connectivity. Our findings provide fundamental understanding of the factors that influence interregional ECoG networks, and open the possibility for predictive modeling of surgical outcomes in disease.

Motivation & Objective

• To develop a method for constructing whole-brain functional connectivity networks from multi-subject ECoG data despite variable electrode placement and sparse coverage.
• To determine whether ECoG functional networks share topological features with fMRI BOLD networks.
• To identify neurobiological factors—specifically anatomical connectivity, distance, and gene co-expression—that predict inter-regional ECoG correlation patterns.
• To assess whether gene expression patterns, particularly those related to ion channels, are enriched in networks that predict ECoG connectivity.
• To evaluate whether models trained on single-subject ECoG data can generalize and predict individual-specific network topologies.

Proposed method

• Constructed parcellation-based, band-limited functional connectivity (FC) networks from multi-subject ECoG recordings using standardized cortical parcellations.
• Compared ECoG FC networks to fMRI BOLD networks using metrics such as connection weight correlation, distance dependence, and modular structure.
• Built a multilinear predictive model using anatomical connectivity, interregional Euclidean distance, and gene co-expression correlation matrices as predictors of ECoG FC weights.
• Optimized gene expression subsets by iteratively selecting genes with highest predictive power, then performed gene ontology (GO) enrichment analysis on the best-performing gene sets.
• Fitted the predictive model to individual subjects and evaluated its specificity and out-of-sample prediction accuracy.
• Used permutation testing and cross-validation to assess statistical significance and model robustness.

Experimental results

Research questions

• RQ1 To what extent do ECoG functional connectivity networks resemble fMRI BOLD networks in topological organization?
• RQ2 Can inter-regional ECoG correlations be accurately predicted using anatomical connectivity, interregional distance, and gene co-expression patterns?
• RQ3 Which specific genes are most predictive of ECoG network topology, and what biological functions do they share?
• RQ4 Do the predictive models trained on individual subjects generalize well to other individuals, or are they highly specific?
• RQ5 Is there a molecular basis for ECoG network organization, particularly involving ion channels and membrane potential regulation?

Key findings

• ECoG functional connectivity networks showed strong topological similarity to fMRI BOLD networks, including correlated connection weights, distance dependence, and modular structure.
• A multilinear model combining anatomical connectivity, interregional distance, and gene co-expression predicted ECoG FC with high accuracy, outperforming models using single or dual factors.
• The best-performing gene subsets were significantly enriched for ion channel activity, membrane potential regulation, and voltage-gated ion channel function, particularly in the 1–4 Hz and 4–8 Hz frequency bands.
• Gene ontology analysis revealed that the most predictive genes were enriched for terms such as 'cation channel activity' (p = 6.55×10⁻⁶), 'voltage-gated sodium channel activity' (p = 5.78×10⁻⁵), and 'regulation of membrane potential' (p = 7.74×10⁻⁴).
• Models trained on single-subject ECoG data achieved high specificity and out-of-sample prediction accuracy, indicating shared organizational principles across individuals.
• The inclusion of restricted gene expression subsets—particularly those related to ion transport and membrane potential—improved model performance, suggesting a direct molecular underpinning of ECoG connectivity.