Physiological symmetry of transcranial magnetic stimulation‐evoked EEG spectral features

Abstract Transcranial magnetic stimulation (TMS)‐evoked EEG potentials (TEPs) have been used to study the excitability of different cortical areas (CAs) in humans. Characterising the interhemispheric symmetry of TMS‐EEG may provide further understanding of structure–function association in physiological and pathological conditions. We hypothesise that, in keeping with the underlying cytoarchitectonics, TEPs in contralateral homologous CAs share similar, symmetric spectral features, whilst ipsilateral TEPs from different CAs diverge in their waveshape and frequency content. We performed single‐pulse (<1 Hz) navigated monophasic TMS, combined with high‐density EEG with active electrodes, in 10 healthy participants. We targeted two bilateral CAs: premotor and motor. We compared frequency power bands, computed Pearson correlation coefficient (R) and Correlated Component Analysis (CorrCA) to detect divergences, as well as common components across TEPs. The main frequency of TEPs was faster in premotor than in motor CAs (p < .05) across all participants. Frequencies were not different between contralateral homologous CAs, whilst, despite closer proximity, there was a significant difference between ipsilateral premotor and motor CAs (p > .5), with frequency decreasing from anterior to posterior CAs. Correlation was high between contralateral homologous CAs and low between ipsilateral CAs. When applying CorrCA, specific components were shared by contralateral homologous TEPs. We show physiological symmetry of TEP spectral features between contralateral homologous CAs, whilst ipsilateral premotor and motor TEPs differ despite lower geometrical distance. Our findings support the role of TEPs as biomarker of local cortical properties and provide a first reference dataset for TMS‐EEG studies in asymmetric brain disorders.


| INTRODUCTION
Transcranial magnetic stimulation (TMS) coupled with EEG (TMS-EEG) is a controlled perturbational approach that directly activates a target cortical area (CA) to assess its excitability (Ilmoniemi et al., 1997;Rosanova et al., 2009). TMS-evoked EEG potentials (TEPs) are complex waveforms generated by averaging segments of EEG recording ("trials") which are time-locked to the TMS pulses. Typically, TEPs have a duration of hundreds of milliseconds and are characterised by sustained increases of power in specific frequency bands that depend on the CA targeted (Rosanova et al., 2009). When not confounded by scalp muscle artefacts, auditory, or extraneous sensory activations, TEPs provide a reliable readout of the reactivity of cortical circuits (Hallett, 2007;Komssi & Kähkönen, 2006;Rogasch & Fitzgerald, 2013;Rosanova et al., 2012). TMS can be delivered with different pulse configurations, whereby current is either delivered with a bidirectional flow (biphasic stimulation), or dampened after the first quarter cycle, thus delivering a unidirectional current flow (monophasic stimulation) (Sakai et al., 1997).
In TMS-EEG experiments, either passive or active EEG systems configurations may be used (Mancuso et al., 2021). Active configurations are more sensitive to artefacts due to transient voltage changes; however, signal pre-amplification directly at the electrode level may allow for better signal quality at higher interelectrode impedance compared to passive systems, and enables faster preparation in the clinical environment (Laszlo et al., 2014).
Previous studies employing biphasic TMS and passive electrodes have shown that TEPs follow a rostro-caudal gradient in their main oscillatory frequency (i.e., natural frequencies) (Rosanova et al., 2009). This intrinsic frequency gradient at which CAs oscillate is thought to reflect the cytoarchitectonics, as well as the connectivity, of distinct thalamocortical modules (Ferrarelli et al., 2012;Rosanova et al., 2009) and can be altered in brain pathological conditions (Ferrarelli et al., 2012). Up-to-date, this gradient has only been described using biphasic TMS and passive electrodes. Furthermore, the interhemispheric symmetry of the natural frequency gradient has never been assessed.
In this work, we sought to determine the symmetry of TEP spectral features by examining the presence of comparable bilateral natural frequencies specific to each CA, to corroborate previous work investigating the similarity of TEPs recorded from homologous regions (Casula et al., 2020;Vallesi et al., 2021). We also assessed the use of monophasic TMS coupled with active high-density EEG (hd-EEG) to reproduce the natural frequencies previously determined in premotor areas, and for the first time measured the natural frequencies for the hand region of the primary motor cortex on the left and right hemispheres. Additionally, we assessed the interhemispheric dynamics by applying the interhemispheric signal propagation (ISP) and balance (IHB) as described elsewhere (Casula et al., 2020). We extracted the spectral features of TEPs recorded from 10 healthy individuals in whom we targeted both premotor and motor CAs, and compared TEP natural frequencies and voltages by applying time-frequency analysis and Pearson's correlation coefficient (R), respectively. We also identified the TEP components that were reliably reproducible across participants and specific to the targeted CA by applying the correlated component analysis (CorrCA). Our ultimate goal was to demonstrate that TMS may serve as a potential biomarker to assess cortical activity providing functional information on the underlying cytoarchitectonics, beyond the information provided by standard neuroimaging.

| Participants
The study was approved by the local ethics committee (REC ref 15/LO/1642). All participants gave written informed consent. Exclusion criteria were as follows: history of traumatic brain injury, neurological or psychiatric diseases, presence of intracranial metallic implants, drug release dispensers, metallic tattoos or cardiac pacemakers (Rossi et al., 2009). Ten right-handed healthy participants were recruited to participate in the study (six females; age: median 33.00 years, interquartile range [IQR] 5.00, all with higher education level, Table 1).

| Experimental setup
Experiments were conducted at the Chalfont Centre for Epilepsy, Buckinghamshire, UK, under the research governance of University College London (UCL). We devised a systematic approach for our TMS-EEG assessment (Figure 1, full details of the protocol are available in the Supplementary Material S1).

| Transcranial magnetic stimulation-click sound masking
To avoid the contamination of TEPs by auditory responses to the click produced by the TMS coil, we adopted the TMS-click sound masking toolbox TAAC  and participants wore noisecancelling in-ear headphones (Shure SE215-CL-E Sound Isolating, for the detailed methodology see Supplementary Material S2).

| High-density EEG recording
For all the experimental sessions, we used hd-EEG recording following international standards (Nuwer et al., 1998;Sinha et al., 2016). An ActiCHamp 63-channel amplifier was used, and TMS-compatible acti-CAP active electrodes (Brain Products GmbH, Germany) were placed in the international 10-20 montage referenced to the forehead. The impedance of all electrodes was kept below 5 kΩ (Ilmoniemi & Kiči c, 2010). EEG signals were band-pass filtered between 0.1 and 500 Hz and sampled at 5 kHz with 32-bit resolution. With this specific equipment, the magnetic artefact induced by the stimulation pulse does not last longer than 5 ms in the EEG recording. At the start of each experiment, the motor "hotspot" for the first dorsal interosseus (FDI) muscle was located over the hemisphere contralateral to the muscle being monitored and the resting motor threshold (RMT) was determined for both hemispheres, following Rossini and colleagues' guidelines (Rossini et al., 2015). RMT was considered as the TMS intensity required to elicit motor evoked potentials (MEPs) >50 μV in 5 of 10 trials delivering pulses at 0.2 Hz minimum, measured from the FDI muscle (Supplementary Material S3).

| Brain navigation, TMS targeting and brain stimulation
In this work, CA refers to any of the motor or premotor brain regions, whilst we define the region of interest (ROI) as the group of four channels related to each CA under study (F1-Fz-Fc1-Fcz for left premotor, F2-Fz-Fc2-Fcz for right premotor, C5-C3-Cp5-Cp3 for left motor, C4-C6-Cp4-Cp6 for right motor, Table S1). Two bilateral scalp targets were identified within the areas indicated by the four abovementioned ROIs. Individual structural magnetic resonance images (MRI, T1-3D sequence) were available for five participants (Table 1).
During the assessments on these subjects, we verified that the two anterior targets were overlapping the caudal frontal superior gyrus (i.e., the premotor-BA6) and that the two posterior targets were overlapping the precentral gyrus (i.e., the motor-BA4) areas. This selection of CAs was based on previous studies that assessed cortical excitability in vivo in humans (Ferrarelli et al., 2012). In these subjects,  Table S1 for individual intensities).
A real-time visualisation tool (rt-TEP) was used to guide coil orientation and to determine the stimulator intensity, minimising muscle artefacts and ensuring the presence of a detectable TEP . Stimulation intensity was determined differently depending on the CA under study. For TEPs evoked in the motor CA, intensity was the highest available below ≤90% of the RMT, to avoid possible sensory-feedback contamination, as previously described (Fecchio et al., 2017). For premotor TEPs, we used the stimulator intensity able to elicit at least a 10 μV-amplitude response in the average of the first 20 trials. This response was measured in the channels closest to the stimulation site (as per Casarotto et al., 2022), on the first peak-to-peak component between 10 and 50 ms after the TMS pulse, using an average reference montage. In this work, we also measured the first TEP component after pre-processing and averaging. We consistently found intra-session asymmetric amplitudes with the highest voltages in the channels under the coil, which we considered as a proof of direct cortical perturbation (Belardinelli et al., 2019;Casarotto et al., 2022). Wilcoxon matched pairs signed rank test was

| Data pre-processing
TMS-EEG data pre-processing was performed using Matlab R2016a (The MathWorks). First, the TMS artefact was removed from all the trials replacing the recording between À2 and 5 ms from the TMS pulse with the 7 ms of time interval before (between À9 and À2 ms from the pulses). All trials were segmented ±800 ms around the stimulus and high-pass filtered at 1 Hz (Makeig et al., 2002). Channels containing line-noise >50 μV lasting more than the 10% of the entire session duration were manually excluded. Subsequently, segments containing more than 50 μV of electrical activity were detected and rejected by visual inspection by two trained researchers (SDA, DJJ).
Channels were re-referenced to the average reference and rejected channels were interpolated using the EEGLAB spherical interpolation function. We used independent component analysis (ICA) to remove any residual artefacts caused by eye movements, scalp muscle activation or electrical interference of devices (EEGLAB runica function; Makeig et al., 2002). Finally, data were down-sampled at 1 kHz, lowpass filtered (45 Hz, notch 50 Hz) and segmented again in a time window of ±600 ms around TMS pulses (Fecchio et al., 2017;Makeig, 1993). After pre-processing, we included 10 participants with bilateral TEPs obtained from premotor and motor stimulations for further analysis (Table 1).

| Natural frequencies
We considered the natural frequency as the main (i.e., the most powerful) frequency of TMS-evoked oscillation globally across all channels, between 20 and 200 ms after the TMS pulse (Rosanova et al., 2009).
We assessed TEP spectral features by analysing the event-related spectral perturbation (ERSP). ERSP measures the average dynamic change in amplitude across all the bands of the EEG frequency spectrum in relation to a specific event (Makeig, 1993). We performed time-frequency decomposition analysis with wavelet transformation (Morlet, 3.5 cycles) between 8 and 45 Hz by computing the EEGLAB Newtimef function (Grandchamp & Delorme, 2011), within the Matlab-based public license toolbox EEGLAB (Delorme & Makeig, 2004). Absolute spectral normalisation was conducted on all trials by performing a full-epoch length correction. A pre-stimulus baseline correction between À500 and À100 ms from the pulse was applied to the pre-processed data, with the time window set 100 ms apart from the stimulus in order to avoid any possible post-stimulus contamination. The Newtimef function computes the surrogate distribution (i.e., creates timepoints with values randomly chosen from the baseline) at each frequency by permuting real baseline values.
Thus, it tests whether the original ERSP value points are present in the 99.5% tail of the surrogate distribution of any given frequency (Fecchio et al., 2017), in which case each specific time-frequency point is considered significant at α < .01 (Fecchio et al., 2017), after correction for multiple comparisons, and using the false discovery rate (FDR) procedure (Grandchamp & Delorme, 2011

| Correlation coefficient analysis
We measured the degree of correlation between the voltages of the Since we adopted two different approaches for tuning the stimulation intensity across different areas, we controlled for the possible association between the natural frequencies and the stimulation intensities used. To this aim, we calculated the correlation coefficient between the absolute values of natural frequencies and intensities used in each area under stimulation. We then conducted a linear regression analysis of the distribution (y = α + βx, where y is intensity and x is the natural frequency) to further analyse the association between intensity stimulation and natural frequencies. Values of p < .05 were deemed as significant.

| Correlated component analysis
We

| Interhemispheric signal propagation and inter hemispheric balance
To assess the interhemispheric transmission, we first analysed the TMSevoked response on the stimulated hemisphere and on the contralateral.  Figure 2). We obtained similar results when comparing natural frequencies calculated at the local level (full results available in the Supplementary Material S6, Figure S1, Table S2). Additionally, we found that variability in the natural frequencies was not affected by the use of individual MRI versus template (details in Table S3).

| Correlation coefficient analysis
The Pearson's coefficient (R) analysis showed higher correlation between contralateral homologous ROI comparisons ( Figure 3)

| Amplitude of the first TEP component after pre-processing and averaging
For this component, we found larger amplitude in premotor TEPs than There was a significant difference between left ipsilateral premotor and motor TEPs (W = À39.00, p = .0488) and between right ipsilateral premotor and motor TEPs (W = À53.00, p = .0039).

| Local mean field power
In order to control for the potential confound of the stimulation impact on the reactivity of each stimulated CA, we calculated and compared the local mean field power (LMFP) of each TEP (detailed methodology and results in Supplementary Material S9, Figure S3).

| Interhemispheric signal propagation and interhemispheric balance
We calculated the ISP from the left and right motor cortices. We found a consistent interhemispheric inhibition across all our participants (see Supplementary Material S11).

| TEP differentiation and symmetry
Here, we used single-pulse (<1 Hz) navigated monophasic TMS to stimulate bilateral cortical areas and simultaneously record hd-EEG with active electrodes. Using a different TMS pulse configuration and a different amplifier to those in previous reports, we first reproduced previous findings showing the presence of distinct frequencies evoked across different CAs perturbed by direct transcranial magnetic pulses (Sakai et al., 1997;Sommer et al., 2006). This result confirms the existence of a specific frequency tuning, called the natural frequency, that is intrinsic to the targeted circuit and independent of the specific stimulation and recording set-up (Ferrarelli et al., 2012;Rosanova et al., 2009). Most importantly, we find that such tuning is symmetric across the two hemispheres.
LMFP analysis showed no significant difference in the response to TMS between 20 and 200 ms across the different stimulation targets. Therefore, the natural frequencies observed are unlikely to reflect differences in the overall effectiveness of the TMS perturbation. Our results showed symmetrical natural frequencies and a high correlation coefficient between contralateral homologous ROIs (Figures 2 and 3). We found significant differences when comparing the distance between homologous contralateral motor areas (which was higher) and the other distances ( Figure 5). However, motor TEPs showed symmetrical spectral frequencies despite higher geometrical distance between the contralateral homologous motor targets, whilst ipsilateral motor and premotor TEPs were different despite their closer proximity. These findings confirm that differences in the TEPs features depend more on the cytoarchitectonic differences of the underlying cortex rather than on the geometrical distance between the stimulated areas. Cytoarchitecture and histology are known to be similar in homotopic contralateral areas but can differ substantially at small distances along the anterior-posterior axis (Zilles et al., 2015). In line with this, the fact that TEP spectral features diverged more amongst targets that were geometrically closer on the scalp but more different in terms of underlying cortical cytoarchitecture has important implications for the methodological debate on the origin of TEPs. Indeed, this result can be more parsimoniously explained as the effect of a direct cortical activation, rather than in terms of nonspecific sensory responses to scalp or auditory stimulation. These findings support the role for TMS-EEG in the study of electrophysiological correlates of the local structural and functional arrangement of cortical circuits. Importantly, our findings of contralateral TEP symmetry only relate to frequency spectra and not amplitude. TEP amplitude amongst contralateral homologous CAs has been previously shown to be rather asymmetric during the same stimulation session, and this has been considered as a proof of direct cortical activation (Belardinelli et al., 2019). We confirmed this after applying ISP analysis (Casula et al., 2020;Hui et al., 2021)  indices of ISP and IHB by using TMS-EEG, providing the first evidence of correlation between ISP and IHB. In their recent study, they carried out TMS-EEG in 50 healthy participants, stimulating the motor (M1) cortex bilaterally. The resulting TEPs were found to be symmetrical, as measured by ISP and IHB (Casula et al., 2020) whilst these measures showed to be asymmetric in patients with schizophrenia, as demonstrated by Hui et al. (2021). We replicated similar findings for ISP and IHB calculations applied to motor cortices in healthy subjects (Casula et al, 2020), enhancing the reproducibility value of these two indices. Lastly, most studies report stimulation of M1 at or above the RMT to date. Here, we used stimulation intensities below F I G U R E 5 Relative stimulation distances and locations. We used the coordinates from the brain navigation software to calculate the distances between ipsilateral and contralateral homologous target stimulations. (a) Topography of all the targets that we stimulated across participants. The interconnected coloured lines represent the distances we calculated between different targets. Blue represents the distance between left premotor and right premotor targets. Red represents the distance between left motor and right motor targets. The grey colour coding shows the distances between left premotor and left motor, and yellow represents the distance between right premotor and right motor. (b) Box plot of the average of the distances in all the targets we stimulated across all participants. The black line crossing the bars represents the median of each target. The cross inside the bars represents the mean of each target.
the RMT for each motor stimulation, in order to avoid eliciting MEPs, which can contaminate the TEP signal via indirect stimulation of the sensory system (Fecchio et al., 2017).

| Decomposing TEPs with CorrCA
CorrCA identified TEP components that were reliably reproducible across participants and specific to the stimulated region. CorrCA dif-  (Bisogno et al., 2021;Georgopoulos et al., 1982). Further studies are needed to better understand the putative mechanism responsible for this component.

| Monophasic stimulation and hd-EEGtechnical considerations and limitations
We showed that the use of active EEG coupled with a monophasic stimulator can replicate the high-quality TEPs which have been obtained using the traditional methodology reported in the literature, that is, passive electrodes and biphasic stimulator. The use of active electrodes has the advantage of a faster preparation time and is therefore potentially useful in a clinical setting where the duration of the experiment is a critical factor. The feasibility of TEPs recordings with active electrodes in comparison with passive electrodes has already been demonstrated (Mancuso et al., 2021). In this study, active electrodes facilitated fast preparation and reliable maintenance of impedances below 5 kΩ. However, we observed voltage fluctuations after long stimulation sessions, as previously reported (Laszlo et al., 2014).
The use of active electrodes has been associated with more electrical artefact than passive systems, causing a larger decay artefact after the stimulation (Laszlo et al., 2014). We limited the decay artefact by consistently maintaining impedances as low as possible in this study.
Notably, we used online brain navigation instead of standardised coordinates, enabling precise targeting of cortical areas (<2 mm of error). Furthermore, we adopted a real-time TEP visualisation approach , instead of the typical RMT-driven technique, allowing for better control of the necessary intensity, recording of genuine TEPs, and avoidance of muscle and electrical artefacts.
Another potential limitation is represented by the systematic adoption of lower intensities for motor stimulations. Saari et al. (2018) previously showed how TEPs change in relation to the intensity used.
We conducted a correlation analysis between stimulation intensity and natural frequencies across each stimulated area and we did not find any statistical significance (detailed results in Supplementary Material S79.2). This can be ascribed to three main reasons: (1) the difference between the intensities we used was too little; (2) the range of intensities we used (from $90% to $110% RMT) was within a window in which TEP differences are more subtle (in fact the higher differences in TEP shapes in the work by Saari et al. (2018) is found in respect to intensities below 80% of RMT); (3) spectral features are maintained despite changes in the intensities adopted. As such, further work is needed to better understand the impact of TMS intensities on TEP spectral features.  (Fleischmann et al., 2020).

| Future directions
We demonstrate that TEPs are symmetrical in contralateral homologous premotor and motor areas, but asymmetrical in the same areas ipsilaterally, with higher evoked frequencies in the premotor than in motor CAs. Neurophysiological properties measured by TMS-EEG could reflect the functional architecture of the underlying cortical structures and are possibly determined by the underlying neuronal networks. Our findings support the use of TMS as a potential biomarker to assess cortical function in people with lateralized brain pathologies, such as typically seen in many focal epilepsies or stroke.
Our methodology may offer a measure of underlying cortical dysfunction, even in the absence of obvious structural abnormalities, providing an in vivo dynamic assessment of subtle circuit alterations. A normative dataset of TEPs obtained from multiple bilateral cortical targets, with high temporal and spatial resolution, may provide the foundation for future studies investigating focal brain disease and the effect of treatment strategies on a longitudinal basis.