Functional connectivity of EEG is subject-specific, associated with phenotype, and different from fMRI

Abstract

A variety of psychiatric, behavioral and cognitive phenotypes have been linked to brain ''functional connectivity'' -- the pattern of correlation observed between different brain regions. Most commonly assessed using functional magnetic resonance imaging (fMRI), here, we investigate the connectivity-phenotype associations with functional connectivity measured with electroencephalography (EEG), using phase-coupling. We analyzed data from the publicly available Healthy Brain Network Biobank. This database compiles a growing sample of children and adolescents, currently encompassing 1657 individuals. Among a variety of assessment instruments we focus on ten phenotypic and additional demographic measures that capture most of the variance in this sample. The largest effect sizes are found for age and sex for both fMRI and EEG. We replicate previous findings of an association of Intelligence Quotient (IQ) and Attention Deficit Hyperactivity Disorder (ADHD) with the pattern of fMRI functional connectivity. We also find an association with socioeconomic status, anxiety and the Child Behavior Checklist Score. For EEG we find a significant connectivity-phenotype relationship with IQ. The actual spatial patterns of functional connectivity are quite different between fMRI and source-space EEG. However, within EEG we observe clusters of functional connectivity that are consistent across frequency bands. Additionally we analyzed reproducibility of functional connectivity. We compare connectivity obtained with different tasks, including resting state, a video and a visual flicker task. For both EEG and fMRI the variation between tasks was smaller than the variability observed between subjects. We also found an increase of reliability with increasing frequency of the EEG, and increased sampling duration. We conclude that, while the patterns of functional connectivity are distinct between fMRI and phase-coupling of EEG, they are nonetheless similar in their robustness to the task, and similar in that idiosyncratic patterns of connectivity predict individual phenotypes.

Keywords: Brain–behavior relationships; Electroencephalography (EEG); Functional connectivity; Functional magnetic resonance imaging (fMRI); Imaginary coherence; Reliability.

Fig. 1.. Data availability.

A) For each subject only tasks that are available on the HBN Biobank and have a reasonable data quality (rating > 4) are concatenated. B) Number of subjects available for each level of data quality, ‘All EEG’: rating >3, ‘Not Bad EEG’: rating >4, ‘Good EEG’: rating = 5; Video 1: ‘Diary of a Wimpy Kid’, Video 2: ‘Fun with Fractals’, Video 3: ‘Despicable Me’, Video 4: ‘The Present’; C) Time of recordings per modality and task Ratings of data quality are described in section 5.5.

Fig. 2.. Clusters in the FC of fMRI and EEG for different frequency bands.

A) Top row shows the mean cortical fMRI connectivity matrix measured as Pearson’s correlations, as well as the EEG source space FC matrices measured as iCOH. Here FC is computed with combined data from resting state, naturalistic viewing and a visual stimulation task and averaged over all subjects (fMRI N = 770, EEG: N = 1330). The significance of the connections is reported in Fig. A1. The leftmost column of spatial patterns shows the clusters identified by Yeo et al. (Yeo et al., 2011) in fMRI data based on 200 parcels in the Schafer atlas (Schaefer et al., 2018). Using the EEG data in different bands, connectivity has been clustered using the same analysis conducted by Yeo et al. (Yeo et al., 2011). The corresponding spatial location of the clusters are shown below the connectivity matrices. The colorbars to the left and bottom of the connectivity matrices indicate the cluster assignment. Broadband FC is most similar to the alpha band and is omitted for brevity. Clusters of the EEG FC have been colored to match the areas in the Yeo atlas with largest overlap, but do not necessarily share the same functions. B) Individual confusion matrices comparing the clustering between different bands and modalities. Values on the diagonal depict the number of voxels assigned to the same clusters. Values on the off-diagonal depict the number of voxels assigned to different clusters in each modality. The colorbar codes for the number of voxels that are in the same or different clusters. C) Similarity of clustering of FC in different EEG frequency bands and modalities measured as the precision of the confusion matrix. The colorbar codes the ratio of voxels assigned to the same cluster over the total number of voxels. Clusters 1–7 are located on the left hemisphere, clusters 8–14 are located on the right hemisphere. ‘fMRI-Yeo’: Clusters defined in Yeo et al. (2011) (Yeo et al., 2011), ‘fMRI-Here’: Clusters defined on fMRI data from the HBN dataset. Resting state and video task data have been concatenated. Cluster 1/8 = Visual; Cluster 2/9 = ‘SomMotor’: Somatomotor; Cluster 3/10 = ‘DorsAttn’: Dorsal Attention; Cluster 4/11 = ‘VentAttn’: Ventral Attention; Cluster 5/12 = Limbic; Cluster 6/13 = ‘FrontPar’: Frontoparietal; Cluster 7/14 = Default.

Fig. 3.. FC for subjects engaged in different tasks is more similar than for different subjects engaged in the same task.

A) Violin plots depict the distribution of the Pearson’s correlation between FC (iCOH) matrices across subjects. Dashed lines indicate the median of the distributions. Pearson’s R for ‘Same Subject/Different Task’ measures the similarity between different tasks (resting state, videos viewing, flashing gradings) computed for each subject separately. Pearson’s R for ‘Different Subject/Same Task’ measures the similarity between different subjects completing the same task. The correlation values between different subjects in the same task have been averaged per subject across all possible subject pairs. Then, the correlation coefficients from different tasks have been averaged. B) ICC measures reliability across tasks relative to variability across subjects. Distribution of ICC of FC (iCOH) between each pair of brain regions in the Schaefer atlas. We compare connectivity matrices between three tasks (videos, resting state and inhibition/excitation paradigm). The distribution of ICC is displayed for connectivity in each frequency band. Corresponding matrices are shown in Fig. A2. C) Distribution of ICC of FC between brain areas in resting state increases with length of recording time. ICC is computed between connectivity matrices of different sessions. Dashed lines show the median of the distributions. Median difference of ICC between 2.5 and 5 min sessions: Δ_r = 0.094, p ≈ 0. Matrices of ICC for all pairs of regions are shown in Fig. A3A. D) ICC of FC during a video task (10 min of ‘Despicable Me’) increases with recording time. ICC is computed between sessions of the same movie task. Median difference of ICC between 2.5 and 5 min sessions: Δ_r = 0.16, p ≈ 0. Corresponding matrices of reliability are shown in Fig. A3B.

Fig. 4.. fMRI FC correlates with demographic and phenotypic measures.

MDMR analysis was performed for each demographic/phenotype information separately. The Pseudo F-Statistic measures how much of the total variance of the similarity of subjects can be explained by demographic/phenotype information. A) Results of the MDMR analysis using concatenated fMRI data and showing all phenotypes tested. Numbers indicate number of individuals for which fMRI of sufficient quality and specific demographic/phenotype information was available. B) Connectivity-phenotype relationship computed separately with data from each task. Sex: biological sex. Phenotypes not shown are not significantly related to connectivity in any task. SWAN: Strengths and Weaknesses of Attention-Deficit Hyperactivity-symptoms and Normal-behaviors. WISC: Wechsler Intelligence Scale for Children. Barratt: Barratt Simplified Measure of Social Status. IAT Parent: Internet Addiction Test assessed by a parent. CBCL: Child Behavior Checklist. MFQ Patent: Mood and Feelings Questionnaire assessed by a parent. SCARED self: Screen for Child Anxiety Related Disorders assessed by the minor. SDS: Sleep Disturbance Scale assessed by a parent. DTS: Distress Tolerance Scale. False discovery rate (FDR) control, at a level of ɑ = 0.05, was performed over the 12 variables tested here.

Fig. 5.. EEG FC in various frequency bands are associated to sex, age and IQ.

A) FC (iCOH) was computed after source localization. FDR correction was performed here across 12 measures (the same as in Fig. 4A), but not across bands. The 9 phenotypes not shown here, but shown in Fig. 4A, had no significant association with EEG FC in any frequency band. B) FC of EEG (source space, beta band) is associated with phenotype when FC is computed separately for each task C) The FC-phenotype association is stronger when FC is computed in source space as opposed to sensor space. Sex and age effect are shown for one frequency band as and example. Effects of IQ are shown for all bands that show a significant effect in either source or sensor space. The effect for sex and age is present for all other frequency bands (Fig. A8). FDR control, at a level of ɑ = 0.05, was performed over the 12 variables tested here.

Fig. 6.. Functional connections with significant sex effect.

A) Spatial pattern of connectivity - sex relationship in EEG source space and fMRI. The subject-by-subject distance matrix is computed for each brain region. The distance is the Pearson’s correlation between the vectors of connectivity for each region. Pseudo F-Statistic for EEG and fMRI connectivity - sex association is computed by MDMR (Shehzad et al., 2014). Significant pseudo F-statistic values are plotted on the surface of the Freesurfer fsaverage template after correcting for multiple comparisons (FDR correction at ɑ = 0.05). A high pseudo F-statistic in a brain region indicates that between-subject differences in connectivity patterns to all other regions correspond to differences in sex. B) Pseudo F-Statistic for the relationship between sex and the connectivity patterns in resting state networks of the Yeo parcellation (Schaefer et al., 2018; Yeo et al., 2011). Each square depicts the strength of the relationship of sex and the connectivity within (diagonal) or between (off-diagonal) networks. FDR correction at ɑ = 0.05 was performed. Vis: Visual, SomMot: Somatomotor, DorsAttn: Dorsal Attention, VentAttn: Ventral Attention, FrontPar: Frontoparietal.

Fig. 7.. Functional connections with significant age effect.

A) Spatially resolved EEG source space and fMRI FC - age association based on each Schaefer brain region as a seed. B) Relationship of connectivity in resting state networks to age. Correction for multiple comparisons as in Fig. 6.

Fig. 8.. EEG functional connections with significant effects of phenotypes.

A) EEG source space and fMRI FC - phenotype (IQ and SES) association based on each Schaefer brain region as a seed. B) Relationship of connectivity in resting state networks to IQ and SES. Correction for multiple comparisons as in Fig. 6.