fMRI

Preprocessing

Functional MRI data (I, II, III) were first converted to NIfTI (http://aedes.uef.fi/), then slice-timing corrected, motion-corrected, spatially smoothed (2 × 2 voxel full-width at half-maximum Gaussian kernel), and finally co-registered to a reference brain (previously acquired SE-EPI image) with SPM8 and Matlab (Version 2011a, The Mathworks Inc., Natick, MA, USA). The motion of the awake rats was evaluated by visual inspection of volumes from raw data and by mass-center displacement values obtained from SPM8. Data series with motion larger than the voxel size (0.391 mm) were excluded from the analysis. Motion scrubbed 10-min data were then preprocessed again to evaluate FC. In order to demonstrate the success of motion correction steps, translational (x, y, z) and rotational parameters from individual animals and average translational values were illustrated.

MB-SWIFT data (III) were first reconstructed with SWIFT package 2018 (https://www.cmrr.umn.edu/swift/index.php) using correlation, gridding, and three iterations of the FISTA algorithm (Beck and Teboulle, 2009). The intensity bias caused by intensity gradient in the MB-SWIFT images was removed from the images using an N4ITK bias correction (Tustison et al., 2010). Volumes were motion corrected using Advanced Normalization Tools (ANTs, http://stnava.github.io/ANTs/) (Avants et al., 2011), and three nuisance regressors for both translation and rotation were computed. For co-registration, anatomical

excluded from the analysis. In order to demonstrate the success of motion correction steps, translational (x, y, z) and rotational parameters from individual animals and average translational values were illustrated.

Additionally, body motion induced B⁰ changes were evaluated in isoflurane anesthetized rats from the raw or preprocessed images during MB-SWIFT or SE-EPI acquisition. Cortical signal variation between MB-SWIFT and SE-EPI were subsequently compared during the motion periods.

Functional connectivity

Whole brain functional connectivity was evaluated by a seed-based correlation (I, II, III, IV), voxel-wise correlation (I, II, III), and by independent component analysis (II, IV).

Before the analysis, the data were band-pass filtered at 0.01–0.15 Hz. Pearson correlation coefficients were calculated between time-series obtained from individual voxels or the selected regions of interests. For the seed-based correlation analysis, the 12 region of interests (ROIs) were as follows: medial frontal cortex (mFC), motor cortex (MC), somatosensory cortex (SC), visual cortex (VC), auditory cortex (AC), retrosplenial cortex (RC), nucleus accumbens (NAc), striatum (Str), hippocampus (HC), medial thalamus (ThM), ventrolateral thalamus (ThVL), and hypothalamus (HTH) (I-IV). The correlation was calculated from the time-series of 300 motion-free volumes (10 min) in each anesthesia or awake period (I-IV).

Additionally, 12 ROIs were divided into 92 smaller ROIs, and the correlations of ROI time-series were used in graph theory-based complex-network analyses (Brain Connectivity Toolbox, https://sites.google.com/site/bctnet/) (III). Average correlations were obtained from the calculated correlation coefficients by taking an average from all cortico-cortical, cortico-striatal, hippocampal-cortical and thalamo-cortical connections (I). Additionally, partial correlation coefficients were calculated by using motion correction parameters as nuisance regressors (II, IV).

Functional connectivity in awake rats was evaluated by group-ICA (ICA; FSL MELODIC, https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC) with manually drawn brain masks (II and IV). The number of components which had to be computed ranged from 30-50, according to previous awake rat studies (Becerra et al., 2011;

Liang et al., 2011). Computed components that were located only at brain surfaces, inside large vessels, or only unilaterally, or were anatomically poorly localized, were excluded from the analysis to minimize the inclusion of artificial non-neural components.

4.8.2 Electrophysiology

evaluated to study brain connectivity. Additionally, the EEG-fMRI cross-correlation was determined in order to study the correlation between isoflurane caused BS activity and the fMRI signal.

Correlation and burst suppression

In Study I, the LFP signal was first normalized, band-stop filtered at 49-51 Hz by using a notch filter, and band-pass filtered at 1-90 Hz by using a 2nd order Butterworth filter (Spike2). The inter channel correlation was measured from the LFP signal by determining LFP envelope amplitude of the full band signal between the electrodes (see Liu et al. 2013). First, the normalized full-band LFP signal was converted to the LFP envelope by taking the absolute value of a Hilbert transform (Hilbert function) in Matlab. The envelope was then convolved with a double-gamma hemodynamic response function (HRF) with a 4-s lag and a negative undershoot which was constructed accordingly to findings by Martin et al. (Martin et al. 2006, Neuroimage). The convolved signal was then low-pass filtered below 0.15 Hz and down-sampled to 0.5 Hz to match the sampling rate of MRI. Finally, the Pearson correlation (r) was calculated to compare the detrended signals during the 10 min isoflurane periods. Correlation matrices of control and treated groups were created separately, and groups were compared by subtracting the correlation values of the control group from the treated group (Figure 6B).

Isoflurane induced burst suppression (BS) occurrence rate (bursts/s), duration of suppression periods (s) and burst amplitude (µV) were also evaluated for each isoflurane concentration. First, BS activity was analyzed with an in-house modified

Matlab script FindRipples

(http://fmatoolbox.sourceforge.net/Contents/FMAToolbox/Analyses/FindRipples.ht ml). Burst occurrence rate and suppression durations from cortex and hippocampus were calculated by taking an average from the detected epochs from the left and right somatosensory cortices and from the DG and CA1 regions in hippocampus, respectively. When studying the amplitude of the burst activity, the suppression periods were first removed from the signal. Then, the root mean square (rms) function was used to obtain the amplitude for the remaining signal, for each isoflurane period.

EEG-fMRI

In unpublished data as a part of Study I, the correlation between EEG and fMRI

Denoised EEG data were filtered, Hilbert transformed, and convolved with a HRF similar that described above. The convolved signal was then low-pass filtered below 0.15 Hz and down sampled to 0.5 Hz to match the sampling rate of MRI. To extract the burst suppression activity from the fMRI signal, ICA was used, and the extracted BS signal was compared with the HRF-convolved EEG envelopes. Finally, the BOLD and EEG signals were detrended and normalized, and the cross-correlation (r) was calculated between the signals in 300-volume data. The correlations in the primary somatosensory cortex in the presence of 1.3% and 2.0%

isoflurane concentrations were calculated by taking an average of maximum cross-correlation values from the left and right hemispheres.

In Study IV, the cortical EEG signal was analyzed using similar methods as described above. MB-SWIFT and EEG signals were detrended, normalized, and the signals were aligned by a maximizing cross-correlation. A cross-correlation was calculated by using 50 volume windows with 10 volume steps. In addition, a gradient switching artefact on raw and denoised EEG, and electrode caused a susceptibility artefact on MRI image were evaluated for MB-SWIFT data and compared to SE-EPI data according to Publication IV.