An integrated TMS-EEG and MRI approach to explore the interregional connectivity of the default mode network

Thirty-five subjects were enrolled and screened for their compatibility with MRI and TMS-EEG equipment through a screening questionnaire (Rossi et al. 2009; Sammet 2016). All subjects were right-handed and reported no history of neurological deficits. The study was composed of three sessions (Fig. 1), one for MRI acquisition, and approximately 10 months later (mean days 290.6 ± 42.4), the first of the two TMS-EEG sessions was recorded (see below for details). Even though the gap between the MRI and the two TMS-EEG sessions was quite long, there is evidence that the reliability of the resting-state networks, over time, is higher than that of other brain networks (see Noble et al. 2019). For the second and third sessions, the sample was reduced to a total of 18 and 16, respectively, due to subjects dropping out or technical problems. Therefore, the final sample, which underwent all the testing sessions, was composed of 16 young subjects (9 males, mean age 25.2 ± 2.2 years; 7 females, mean age 23.6 ± 3.3 years).

During the MRI session, structural, resting-state functional, and DWI data were recorded for each subject. During the acquisition of the imaging data, subjects were asked to lay still with their eyes open, while a white fixation cross was projected on the screen. The two TMS-EEG sessions were performed in an identical manner 2 months apart on average (mean days 72.3 ± 35.8). During the TMS-EEG sessions, TMS was applied over four different areas of the brain, the right and left prefrontal and parietal areas, while EEG was recorded concurrently (see below for details). For each subject, each TMS-EEG session was divided into six blocks: two 5-min adaptation blocks of resting EEG recorded at the beginning and the end of the session (data not analysed) and four TMS-EEG blocks. Subjects wore earplugs, kept their eyes open, and were seated in a comfortable armchair in an acoustically isolated room. In all blocks, participants were instructed to fixate on a white fixation cross on a computer screen at a distance of 70 cm. The stimulation blocks were interleaved by 5-min breaks, and participants were allowed to move during that time. During breaks, a short video was presented on the computer screen. These videos, which consisted of short parts of a documentary about the planet and universe, were intended to attract the subject's attention and enable a distinction between a state of cognitive involvement and a resting state. EEG signals were not recorded during these periods. The current study was approved by the Human Research Ethics Committee of the University of Trento. Written informed consent was obtained from all subjects before they became involved in the study. All safety MRI and TMS procedures and guidelines were respected (Rossi et al. 2009; Sammet 2016). Brain Imaging Data Structure (BIDS)-formatted data from this study are available on gin.g-node.org/CIMeC/TMS-EEG_brain_connectivity_BIDS.

MRI data were acquired at the Center for Mind/Brain Sciences (CIMeC—University of Trento, Italy) using a 4 T MRI system equipped with an 8-channel receive head RF coil (Bruker MedSpec Synco). Functional images were acquired with a single shot T2*-weighted gradient-recalled echo-planar imaging (EPI) sequence. A 30-slice protocol was employed, acquiring images in ascending interleaved order, within a repetition time (TR; the amount of time between successive pulse sequences applied to the same slice) of 2000 ms [voxel resolution, 3 × 3x3 mm³; echo time (TE), 28 ms; flip angle, 73°; field of view (FOV), 192 × 192 mm]. Two hundred consecutive brain volumes were acquired. To coregister the low-resolution functional images to a high-resolution anatomical scan, we acquired two T1-weighted anatomical scans [magnetization-prepared rapid gradient-echo (MP-RAGE) 1 × 1 × 1 mm³; FOV = 256 mm; 176 slices; generalized autocalibrating partially parallel acquisition (GRAPPA) with an acceleration factor of 2; TR = 2700/2500 ms; TE = 4.18/3.37 ms; inversion time (TI) = 1020/1200 ms; flip angle = 7°/12°]. For whole-brain tractography, the DWI scheme was acquired using a spin echo EPI sequence (TR = 7100 ms, TE = 99 ms) with a b-value of 1500 s/mm2. Ten volumes without any diffusion weighting (b0-images) and 30 diffusion-weighted volumes (isometric voxel 2.3 × 2.3 × 2.3 mm³) were acquired.

EEG signals were acquired using a TMS-compatible EEG system (BrainAmp DC Brain Products GmbH, Germany), with continuous recording from 61 scalp electrodes (FCz, FP1, FP2, AF7, AF8, F7, F5, AF3, AFz, AF4, F8, F6, F3, F1, Fz, F2, F4, FT7, FC5, FC3, FC1, FC2, FC4, FC6, FT8, T7, C5, C3, C1, Cz, C2, C4, C6, T8, Tp7, CP5, CP3, CP1, CPz, CP2, CP4, CP6, TP8, P7, P5, P3, P1, Pz, P2, P4, P6, P8, PO7, PO3, POz, PO4, PO8, O1, Oz, O2, and Iz) positioned according to the 10/10 International System. Additionally, the ground electrode was placed at FPz, the online reference was placed at TP9, and an electrode at TP10 recorded an online signal to “balance” data, in terms of electrode location, in the offline re reference process (see the TMS-EEG session). Horizontal and vertical eye movements were detected by recording electrooculogram (EOG) data. The voltage between the electrode located to the lateral canthus of the left eye and the reference recorded horizontal eye movements. The voltage between the electrode beneath the left eye and the reference, recorded vertical eye movements and blinks. All the electrodes were TMS-compatible Ag/AgCl-coated electrodes mounted on an elastic cap (BrainCap TMS, Brain Products GmbH, Germany). The EEG and EOG signals were bandpass filtered at 0.1–1000 Hz and digitized at a sampling rate of 5 kHz. The impedance was kept below 5 kΩ for all skin/electrode interfaces.

Single-pulse TMS was carried out using a biphasic magnetic stimulator connected with a figure-eight coil with a 70-mm diameter (Magstim, Whitland, UK). Each TMS-EEG session consisted of 120 single pulses, for each area, applied at a random interstimulus interval (ISI) of 2–10 s. Target areas were defined as the cortical regions representing the four main nodes for each single-subject DMN (see Characterization of Group and Single-Subject DMNs). The nodes were identified through independent component analysis (ICA) of the individual resting-state fMRI. The coil position and orientation were monitored continuously using a neuronavigation system (SofTaxic, E.M.S., Bologna, Italy) to ensure a high degree of reproducibility across neurophysiological assessments. TMS was applied over the four target areas in blocks in a counterbalanced order. TMS intensity, which was set at 100% of the resting motor threshold (rMT), was defined as the lowest intensity producing motor-evoked potentials with a peak-to-peak amplitude > 50 µV in five out of ten trials in the relaxed first dorsal interosseous muscle of the right hand (Rossini et al. 2015). The hot spot for the motor threshold was found by positioning the coil over the central sulcus and moving it on the scalp in steps of approximately 0.5 cm towards the left motor cortex. Before setting TMS intensity at 100% of the rMT, we evaluated sub- (90%) and supra-threshold (120%) values to investigate the ideal intensity to obtain stable and clear evoked responses in all stimulation conditions. This final setting was done considering: number of TMS pulses, TMS intensity, TEP response, and secondary effects of TMS that would be obtained by raising the stimulation intensity. Moreover, we maintained the same intensity for all stimulation sites (right and left frontal and parietal areas) so that TEPs could be comparable for the stimulation intensity variable. The mean TMS intensity was 66.3 ± 8.8 for the first TMS-EEG session and 66.4 ± 8.2 for the second TMS-EEG session.

Data were preprocessed using SPM12 (http://​www.​fil.​ion.​ucl.​ac.​uk/​spm/​) running in MATLAB R2017b (The MathWorks, Inc., Natick MA, USA) and MATLAB code developed in-house. The pipeline steps consisted of slice-timing correction, rigid-body realignment, and removal of movement-susceptibility interactions, subtraction of baseline fluctuations by fitting a fourth-order polynomial, low-pass filtering using a second-order Butterworth filter having f_−3 dB = 0.09 Hz, removal of covariance with the six first-order head movement vectors (translations and rotations) and with average white matter and cerebrospinal fluids derived from individual tissue masks derived from the structural scan. Nuisance regressors were temporally filtered as described above. Head movement magnitude was quantified as median frame-to-frame displacement. Prior to further analyses, the EPI volumes were spatially smoothed through a Gaussian kernel that had a full width at half-maximum value of 8 mm.

First, a group ICA using MELODIC (http://​fsl.​fmrib.​ox.​ac.​uk/​fsl/​fslwiki/​MELODIC) within FSL 5.0.9 (Jenkinson et al. 2012) was performed to extract the group DMN. The DMN was chosen based on a template. To this aim, all subjects’ preprocessed resting-state fMRI data were spatially normalized to the MNI template via linear (affine) registration (Jenkinson et al. 2002), rescaled at a resolution of 4-mm isotropic voxels, and then decomposed into ten independent components (Jovicich et al. 2016) using the multisession temporal concatenation procedure in MELODIC. More components were not extracted to prevent splitting the DMN (Abou-Elseoud et al. 2010; Jovicich et al. 2016). The components and the DMN template were thresholded at z scores > 2.3, P < 0.05. Subsequently, a dual regression was used to derive a single-subject DMN from the group DMN (Beckmann et al. 2009; Zuo et al. 2010). Single-subject DMN volume maps were thresholded at z > 2.3, P < 0.05. For each subject, the cortical DMN nodes, including the bilateral and symmetric lateral parietal and medial prefrontal cortex, were manually extracted and were the ROIs used to define the structural and functional connectivity measures.

DWI data were processed by the concatenation of a step for preprocessing, a step for voxel-based diffusivity model reconstruction, and a step for probabilistic tractography. The analysis pipeline was implemented using FSL and MRtrix (Jenkinson et al. 2012; Tournier et al. 2012). Data were corrected for eddy current distortions and head motion by registering the DW volumes to the first b0 volume. The gradient direction (b-vec) for each volume was corrected by applying individual rotation parameters (Leemans and Jones 2009), and nonbrain voxels were removed with the FSL Brain Extraction Tool (FSL-BET). The fibre orientation distribution function was computed using a constrained spherical deconvolution (CSD) model (Tournier et al. 2007). White- and grey-matter tissues were segmented with FSL using the T1-weighted MRI images associated with each individual brain and then resampled at the resolution of the diffusion MRI data. A final tracking step was carried out using a seed-based probabilistic strategy (Tournier et al. 2010) constrained by the white matter mask. We referred to the ROIs that emerged from the processing and analysis of functional data as seeds for tracking.

This process resulted in an estimate of the most likely pathways connecting each pair of ROIs, namely, the probability of connection measured as the number of streamlines successfully reaching a target voxel from a given seed. The parameter settings used to perform tracking were as follows: step size, 0.5 mm; maximum length, 250 mm; and minimum length, 10 mm. The fibre orientation distribution function (f_ODF) amplitude cut-off was set to 0.1, and for the minimum radius of curvature, we adopted the default value (90°  ×  step size  ×  voxel size). Considering the four ROIs (bilateral lateral parietal and medial prefrontal), the resulting connections corresponded to parts of the corpus callosum called the forceps minor and the forceps major (Fig. 1 left side). The former connects homologous regions of the anterior frontal lobe and the latter connects homologous parieto-occipital regions.

After the selection of the connectivity structures of interest, namely, the forceps minor and major (Voineskos et al. 2010), we evaluated some measures that could be used to carry out a quantitative analysis (Yeatman et al. 2012). In the field of tractography, it is possible to define several indexes as measurements of the white matter tracts (Catani et al. 2013; Deslauriers-Gauthier et al. 2019; Mangin et al. 2013). We selected two of the most commonly used indexes in the literature to explore how white matter tracts behaved in correlation with a physiological response.

The first index was the weighted number of fibres, which was defined as the ratio between the number of streamlines between each pair of regions and the total number of streamlines for all seed/target regions, normalized by the number of voxels in each pair to account for differences in the size of seed and target regions (Smith et al. 2015).

The second index was an extracted tensor-derived quantitative measure of the diffusivity metric for each connection, the FA. FA quantifies the directionality of diffusivity in a summative manner (Yeatman et al. 2012).

TMS-EEG data were preprocessed offline (Brain Vision Analyser 2.0 Brain Products GmbH, Munich, Germany). As a first step, cubic interpolation from 1 ms before to 6 ms after the TMS pulse was applied to remove TMS-induced artifacts. Afterwards, a high-pass filter at 1 Hz was applied to the continuous data. Data were then segmented into epochs starting 1250 ms before the TMS pulse and ending 1250 ms after the pulse. The signal was downsampled from 5000 to 1000 Hz. Next, all the epochs were visually inspected, and the EEG epochs containing artifacts or noisy signals were rejected. Physiological and TMS-related artifact components were detected using INFOMAX-ICA and removed based on their scalp distribution, frequency, timing, and amplitude. Subsequently, the low-pass filter at 70 Hz and a notch filter at 50 Hz were applied to the data. All data were rereferenced to the average of all scalp channels, and residual epochs containing artifacts were removed during a second visual inspection. The signal was imported into Fieldtrip (Oostenveld et al. 2011) and resegmented beginning 100 ms before the TMS pulse and ending 400 ms after. Baseline correction was applied using the pre-TMS interval from -100 ms to -1 ms.

First, we aimed to test whether the TEPs recorded in the two TMS-EEG sessions differed in amplitude over space and time. Then, we assessed the Pearson correlation between the TEPs for all the electrodes at each time point, from 7 to 60 ms, with the two indexes of structural connectivity of the forceps, i.e., the weighted number of fibres and FA (i.e., connectivity), in which t tests were run to evaluate the hypothesis that the correlation coefficient R is different from zero. We hypothesized that TEPs of the prefrontal stimulation conditions may be related to indexes in the forceps minor and that TEPs of the parietal stimulation condition may be related to indexes of the forceps major. For these analyses, we conducted nonparametric cluster-based permutation tests to correct for multiple comparisons as implemented by the ft_timelockstatistics function in Fieldtrip (Oostenveld et al. 2011) with MATLAB (2019b, MathWorks).

The comparisons involved all the electrodes for each stimulation condition for the time window of interest from 7 to 60 ms after the TMS pulse. The definition of the time window was based on evidence that the interval is crucial for signal spreading to homologous regions in the opposite hemisphere (Chung et al. 2015; Ilmoniemi et al. 1997; Parks et al. 2012) and to prevent artefacts, i.e., auditory evoked potentials at 100 ms (Conde et al. 2019; Nikouline et al. 1999; ter Braack et al. 2015). Based on the initial one-sample t tests, all t values above a threshold, corresponding to an uncorrected p value of 0.05, were grouped into clusters based on adjacent significant time points and electrodes, considered separately, for a sample with positive and negative t values (two-tailed test). For a significant sample to be included in a cluster, it was required to have at least one adjacent neighbouring significant sample in both space (electrodes) and time. The spatial neighbourhood of each electrode was defined as all electrodes within approximately 5 cm, resulting in a mean of 6.3 (min = 3, max = 8) and median of 7 neighbours per electrode. The t values within each cluster were then summed to produce a cluster statistic. Subsequently, this procedure was repeated across 2500 permutations by calculating Monte Carlo estimates of the significance probabilities (p < 0.05).

The statistical comparison performed on all electrodes in the time window from 7 to 60 ms after the TMS pulse did not reveal any spatiotemporal differences in the two TMS-EEG sessions (all ps > 0.05), as reported in S1. Due to this result, data from the two sessions were merged before the correlations with the structural connectivity measures were evaluated.

The statistical analysis performed between the TEPs and the FA values of the forceps major revealed significant correlations for the left (Fig. 2a) and right (Fig. 2b) parietal stimulation conditions, with a similar pattern of results, as described in the following section. Instead, the correlation between the TEPs and the FA values of the forceps minor was not significant (all ps > 0.05). Similarly, statistics did not reveal any significant correlation between TEPs and the weighted number of fibres in any stimulation conditions (all ps > 0.05).

The interpretation of the clusters of correlations depended on the voltage of the signal with which it was correlated. If the voltage was positive, a positive cluster indicated that the signal increased with increasing FA, and a negative cluster indicated that the signal decreased with increasing FA. On the other hand, if the voltage was negative, a positive cluster indicated that the signal decreased with increasing FA, and a negative cluster indicated that the signal increased with increasing FA. The results described below reflect these relationships.

The correlations revealed three significant clusters (Fig. 3); one was positive (p = 0.002) and two were negative (p = 0.022; p = 0.045).

The positive cluster included the signal from frontocentral electrodes (FP1, AF3, AFz, AF8, F3, F1, Fz, F2, F4, FC3, FC1, FCz, FC2, FC4, Cz, and C2) in the time interval of 37 to 60 ms (Fig. 3a and S2a). Considering that the signal of this cluster was negative on average, as shown in the topography in Fig. 3a, this result indicated that the TEP amplitude in frontocentral sites decreased for higher values of FA in the forceps major.

The first negative cluster included parieto-occipital electrodes (TP7, CP5, P7, P5, P3, P1, PO7, PO3, POz, O1, O2, Oz, Iz, and PO8) on both hemispheres in the interval of 40 to 60 ms (Fig. 3b and S2b). When the TEP voltage was positive, as was the case for the electrodes close to stimulation, this correlation indicated that the TEP amplitude decreased for higher FA in the forceps major. In contrast, when the TEP voltage was negative, as was the case for the electrodes of the contralateral hemisphere, this correlation indicated an increase in the TEP amplitude.

Finally, the second negative cluster comprised left frontocentral electrodes (AF3, F3, F1, FC5, FC3, FC1, FCz, C3, C1, Cz, C2, CP1, and P1) in the time interval of 20 to 37 ms (Fig. 3c and S2c). As shown in Fig. 3c, the voltage on these electrodes and in this time window was positive. Therefore, this correlation indicated a decrease in signal amplitude with higher FA values of the forceps major.

The pattern that emerged from the significant clusters demonstrated a general reduction in the voltage recorded from electrodes close to the stimulation site and over frontocentral areas and an increase in the voltage in posterior electrodes contralateral to the TMS target in correlation with the FA values of the forceps major.

Statistical analysis revealed two significant clusters (Fig. 4), one of which was positive (p = 0.003) and one of which was negative (p = 0.007), with a pattern similar to that observed with left parietal stimulation.

The positive cluster involved frontocentral electrodes (FP1, AF7, AF4, AFz, F7, F3, F1, Fz, F2, F4, FT7, FC5, FC3, FC1, FCz, FC2, T7, Cz, C3, and C1) for the time interval of 33 to 60 ms (Fig. 4a and S3a). Considering that the signal of this cluster was negative on average, as shown in the topography in Fig. 4a, this result indicated that the TEP amplitude in frontocentral sites decreased for higher values of FA in the forceps major.

The negative cluster included parieto-occipital electrodes of both hemispheres (C2, C4, C6, T8, CP2, CP4, CP6, TP8, P2, P4, P6, P8, PO8, PO4, Pz, O2, Oz, PO7, P7, and P5) for the interval of 30 to 60 ms (Fig. 4b and S3b), in which voltage was positive, on average, in electrodes near the stimulation site and negative, on average, in the contralateral site. Therefore, this correlation indicated a decrease in signal amplitude for the stimulation site and an increase in amplitude for the contralateral site.

For the left parietal condition, considering the frontocentral electrodes (positive cluster), the negative voltage decreased with higher FA. In the bilateral parieto-occipital electrodes (negative cluster), the positive voltage on electrodes ipsilateral to stimulation decreased as FA increased. Conversely, voltage on electrodes contralateral to stimulation increased with higher FA values for the forceps major.