Purpose

We address the neurophysiological underpinnings of the fluctuations in the correlation between electroencephalography (EEG) and blood oxygen level dependent (BOLD) signals. We also studied whether these variations are linked to the functional connectivity strength, magnitude or noise characteristics within the default mode network (DMN).

Methods

Simultaneous EEG-fMRI (functional magnetic resonance imaging) was measured from 16 subjects. FMRI was scanned with Siemens 3T SKYRA with a 32 channel head coil. We utilize magnetic resonance encephalography (MREG) sequence (TR = 100 msec, TE = 36 msec) obtained from Freiburg University.^{1, 2} 3D MPRAGE images were used to register the MREG data into MNI-space. Resting state networks were separated using group probabilistic independent component analysis (ICA) as implemented in MELODIC³ with model order of 70. Subject-specific versions of spatial maps and associated time series were generated with dual-regression.^4,5 The dominant DMN_vmpf and DMN_pcc components were selected for further analysis.^6,7 EEG was recorded with BrainAmp system (Brain Products) using 32 Ag/AgCl electrodes (including electrocardiography). Sampling rate was 5 kHz, band pass from DC to 250 Hz and electrode impedances < 5 kΩ. EEG pre-processing steps included scanner and ballistocardiographic artifact correction with the AAS method^8,9 in Brain Vision Analyzer (Brain Products), ICA calculation with EEGlab (Matlab) with maximum number of components (32), detrending and resampling to 10 Hz by getting an average of each of the 0.1-s periods to match the MREG sampling frequency. Individual EEG and MREG IC time courses were FFT band pass filtered to very low frequency (VLF) band (0.008-0.09 Hz) and normalized in OriginPro (Fig. 1C). VLF EEG ICs were used as regressors in FSL FEAT analysis to get individual VLF_EEG-MREG correlation maps. Temporal correlation coefficients between IC time courses were calculated in Matlab with Pearson correlation coefficients. The highest spatiotemporal correlation showing EEG IC was selected from each subject (Fig. 1A) and the temporal and spatial dynamics of correlations between VLF_EEG and MREG were studied in one-minute time windows (50% overlap). Temporal correlation coefficients were calculated between the selected EEG IC and dual regressed MREG IC (Fig. 1B). Also the spatial dynamics of the intrinsic MREG correlation maps and intermodal VLF_EEG-MREG correlation maps were studied by calculating correlations between these and MREG voxel time courses (Fig. 1D) with FSL fslcc tool. To see on group level where the intrinsic MREG maps show activations when the temporal correlation is highest and lowest, the corresponding correlation maps of these time windows were averaged among subjects and statistical significance between their differences was calculated with the randomize tool in FSL. Statistically significant dependency of the correlation strength and the MREG RSN ICs amplitude and root mean square (rms, supposed to reflect noise) (Matlab) as well as the number of active voxels within time window MREG RSN maps (fslmaths and fslstats tools in FSL) was calculated with one-tailed paired t-test.

Results

The spatial maps as well as spatial correlation coefficients between these maps revealed notable variation on single subject level (Fig. 1 E and F). Subject-wise averaged, highest and lowest temporal correlation coefficients are presented in Table 1. It shows the high variability in the averaged correlation coefficients between the highest and lowest correlating time windows resulting in average correlations of 0.3-0.4. Group level averaged maps of time windows that had highest and lowest correlation coefficient showed statistically significant differences in VLF_EEG-MREG correlation maps in DMN_vmpf area and in cortical parts of DMN_pcc (Fig. 2). The intrinsic MREG correlation maps did not reveal any statistically significant differences, but their spatial distribution indicates that correlation is strongest when the intrinsic RSN connectivity is strong and close to the pial brain surface (Fig. 2). Low intermodal correlations occurred when the intrinsic connectivity of the RSN was less coherent, in deeper structures or spread over wide areas VLF_EEG-MREG DMN_pcc maps shows also an anti-correlating task-positive network activated when correlation to MREG is low. Temporal correlation coefficients were found to be stongest dependent of the number of activated voxels in RSN where a higher number of activated voxels relates to higher temporal correlation coefficient and vice versa (Fig. 3). The RSN amplitude or overall RSN noise characteristics did not explain the correlation dynamics between EEG and MREG BOLD.

Conclusion

EEG ISFs are dynamically correlated to MREG BOLD signal due to intrinsic variability of the RSN source distributions and functional connectivity strength. The results indicate that the correlations between EEG and BOLD signal sources are low when averaged over a long period, but can reach high momentary synchrony due to intrinsic RSN dynamics.

Acknowledgements

We gratefully acknowledge the Academy of Finland grants #111711 and #123772 and the Finnish Medical Foundation for their support of this work. The authors declare no competing financial interests.