Does pericentral mu-rhythm “power” corticomotor excitability? – a matter of EEG perspective

Background Electroencephalography (EEG) and single-pulse transcranial magnetic stimulation (spTMS) of the primary motor hand area (M1-HAND) have been combined to explore whether the instantaneous expression of pericentral mu-rhythm drives fluctuations in corticomotor excitability, but this line of research has yielded diverging results. Objectives To re-assess the relationship between the mu-rhythm power expressed in left pericentral cortex and the amplitude of motor potentials (MEP) evoked with spTMS in left M1-HAND. Methods 15 non-preselected healthy young participants received spTMS to the motor hot spot of left M1-HAND. Regional expression of mu-rhythm was estimated online based on a radial source at motor hotspot and informed the timing of spTMS which was applied either during epochs belonging to the highest or lowest quartile of regionally expressed mu-power. Using MEP amplitude as dependent variable, we computed a linear mixed-effects model, which included mu-power and mu-phase at the time of stimulation and the inter-stimulus interval (ISI) as fixed effects and subject as a random effect. Mu-phase was estimated by post-hoc sorting of trials into four discrete phase bins. We performed a follow-up analysis on the same EEG-triggered MEP data set in which we isolated mu-power at the sensor level using a Laplacian montage centered on the electrode above the M1-HAND. Results Pericentral mu-power traced as radial source at motor hot spot did not significantly modulate the MEP, but mu-power determined by the surface Laplacian did, showing a positive relation between mu-power and MEP amplitude. In neither case, there was an effect of mu-phase on MEP amplitude. Conclusion The relationship between cortical oscillatory activity and cortical excitability is complex and minor differences in the methodological choices may critically affect sensitivity.


2
Alpha oscillations ) are a distinct feature of human cortical activity. Initially 3 interpreted to reflect "cortical idling" [1] they are today assumed to have an active 4 role in neural processing: According to the "gating-through-inhibition" hypothesis, 5 alpha oscillations gate information by active inhibition of task irrelevant areas [2,3]. 6 This hypothesis is supported by studies that demonstrated an increase in alpha 7 power over task-irrelevant cortical areas with simultaneous decrease of alpha power 8 in task-relevant areas [2][3][4][5]. Instantaneous fluctuations also actively influence task 9 performance and several studies have shown that performance on perceptual tasks 10 increases, when stimuli are presented during periods of low alpha power [5][6][7]. 11 12 While alpha oscillations are most prominent in the occipito-parietal cortex, they are 13 also expressed in pericentral, sensorimotor cortex where they are called mu-rhythm 14 [8][9][10][11]. Invasive recordings in monkeys provided evidence that mu-oscillations in the 15 sensorimotor cortex follow the gating-through-inhibition hypothesis: Low 16 sensorimotor mu power predicted an increase in neuronal spiking and increased 17 performance on a tactile perception task [12]. seem to support the gating-through-inhibition hypothesis and report associations 24 between low mu-power and higher MEP amplitudes [13][14][15], but these findings 25 Laplacian montage centered above the M1 [23,24]. Conversely, post hoc-studies 1 that have shown a negative linear relationship or no relationship at all between mu-2 power and MEP amplitude have predominantly used source projections or the 3 averaged activity in a cluster of surface electrodes [13,14,16,18,19,22]. 4 To re-examine the relation between mu-power and cortico-spinal excitability, as 5 reflected by MEP amplitude, we employed mu-triggered spTMS using an individual 6 source projection to trace pericentral mu-activity. We applied brain-state informed 7 EEG-TMS for real-time power estimation and targeted the highest and lowest 25% of 8 mu-power in each individual. In addition, we also re-referenced our data using a 9 sensor level Laplacian montage centered over M1-HAND and used post-hoc trial 10 sorting, to test whether the EEG montage used to extract power influenced the 11 detected relationship between MEP amplitude and mu power. For both methods, 12 we also defined the pre-stimulus phase and the inter-stimulus interval between two 13 pulses and used these as covariates to explore their ability to predict MEP 14 amplitudes. 15 16

Subjects: 18
15 right-handed healthy participants took part in this study (7 female, average age= 19 24,1 ±2.8 years). The sample size was based on previous studies using state-20 informed brain stimulation to investigate cortico-spinal excitability [22,28]. We did 21 not perform any preselection based on individual TMS or EEG characteristics (e.g. 22 size of the single alpha-band peak or resting Motor Threshold (rMT)). Subjects were 23 allowed to enroll in the study if they were right-handed and between 18-40 years of 24 age and did meet the criteria specified in the TMS safety screening questionnaire 1 [29]. All subjects gave informed written consent. The study was approved by the 2

Regional Committee on Health Research and Ethics of the Capitol Region in Denmark 3
and was in accordance with the Helsinki declaration (Protocol H-16017716).  Electrophysiological recordings: Both EEG and EMG were recorded using a NeurOne 15 Tesla system (NeurOne Tesla, Bittium, Oulu, Finland). The amplifiers had a sampling 16 rate of 5kHz and we used a 2.5 kHz antialiasing low-pass filter with 24-bits resolution 17 per channel, across a range of +/-430 mV. To ensure that the latency of the data-18 delivery for the real-time processing stayed below 5ms, the data was sent directly 19 from the EEG amplifiers Field-Programmable-Gate-Array via a user-datagram 20 protocol (1kHz update rate over a 1Gb/s Ethernet link). The scalp EEG was recorded 21 using a TMS compatible EEG cap (Easycap M10 sintered Ag/AgCl multielectrodes 22 Easy Cap, Woerthsee-Etterschlag, Germany), with equidistant spacing between the 23 63 surface electrodes. All electrodes were prepared using high-chloride, abrasive 1 electrolyte gel (Easycap, Herrsching, Germany), until the impedance was below 5kW. 2 The EMG was recorded using self-adhesive, disposable surface electrodes 3 (NeuroLine, Ambu A/S, Denmark), and the EMG signal was transferred in the same 4 way as the EEG signal described above. The electrodes were applied in a belly-5 tendon montage on first dorsal interosseous (FDI) muscle of the right hand ( the 6 ground was placed on the right wrist), making sure that a clear muscular response 7 was measured and that 50Hz noise was below 20uV.  containing the latest 500ms of data was updated upon the arrival of each sample. To 6 limit the computational cost of the subsequent processing steps the sampling rate in 7 this buffer was reduced to 1kHz by averaging the 5 samples received in each UDP 8 package. A processing loop running in a separate process with real-time priority 9 performed source projection and a discrete Fourier transform of the 500ms of data 10 [33]. As an indicator of mu-band power we considered the fraction of the sum of the 11 absolute squared coefficients within the defined mu-band and the total signal 12 power, thereby obtaining a mu-band power fraction. The average update time for 13 the processing loop was below 0.5 ms, which ensured that only very occasionally the 14 estimate was not updated for each sample, meaning that the additional latency due 15 to signal processing was negligible. 16

17
Re-extraction of the pericentral my-rhythm using a Laplacian montage: To test the 18 influence of differences in electrode montages for extracting the mu-power we re-19 extracted the pericentral mu-rhythm from the raw scalp EEG for each subjects using 20 a sensor level five-channel Laplacian montage centred on the electrode above the 21 M1-HAND area [23]. This enabled the post-hoc determination of pre-stimulus mu-22 power and mu phase according to a C3-centered Laplacian montage and allowed the 23 direct comparison of source-extracted and surface extracted mu-oscillations. 24 1

Experimental sessions: 2
Structural MRI scans: On the day before the experiment each participant was 3 scanned using a structural T1-weighted MRI sequence (T1-w MRI). T1-w scans were 4 done using a magnetization prepared rapid gradient echo (MPRAGE) MRI sequence 5 with 0.85mm isotropic spatial resolution, TR = 6ms, TE = 2.7ms and flip-angle = 8° on 6 a Philips 3T Achieva scanner (Philips, Best, Netherlands). The field of view was 7 245x245x208mm, such that the scan covered the whole brain. The T1-w scan was 8 required for obtaining the head model for the source projection, as well as for 9 neuronavigation during the state-informed experiment. 10 11 TMS-EEG Experiment: Each experiment started with a range of pre-measurements to 12 determine the subject specific stimulation criteria. First, the subjects were co-13 registered with their T1-w scan, then the individual EEG electrode positions were 14 registered to the T1-w scan. The EEG and EMG electrodes were mounted as 15 described above, and the signal quality of both measures were visually monitored 16 throughout the whole experiment, and EEG channels were re-prepared in case the 17 signal quality dropped. The next steps included the determination of the M1-HAND 18 hotspot, the RMT and the stimulation intensity required to evoke MEPs of 1mV 19 (1MV-MEP) amplitude for the right FDI. Both RMT and 1mV-MEP were estimated 20 using the threshold hunting algorithm described in the experimental setup section 21 above. Finally, the source projection matrix was calculated. 22 To identify the individual mu-rhythm, we recorded 5 minutes of EEG while the 23 subject was resting with open eyes. The individual peak frequency was determined 24 as the peak of the mu-power spectral density (PSD) within 7-13Hz, based on the 1 resting EEG data and the estimated source matrix. The mu-frequency band was 2 defined as ±2Hz around the peak frequency. However, in cases where the lower 3 frequency limit would be below 7Hz, the limit was set to 7Hz, resulting in a narrower 4 frequency band for these subjects. Furthermore, the PSD maps were used to 5 determine the highest (q75) and lowest (q25) quantiles of the mu-power, which later 6 were used for thresholding of the power-informed stimulation. between stimulations and non-stimulations triggers was pseudo-randomized, such 20 that no more than 3 of the same kind could appear in a row. 21 Between each block, a short break of 1-5 minutes was allowed. For a few subjects, 22 we adjusted the stimulation intensity between blocks, such that the stimulation 23 always elicited an average MEP of around 1mV, however the total adjustment never 24 resulted in a change of more than 4% stimulator output. Average stimulator output 1 across participants was 70% ± 13%. To avoid any systematic interaction between 2 TMS pulses, the minimum inter-trial-interval (ITI) set by the algorithm was 2s. Due to 3 the constraints set in the power-detection algorithm the actual ITI was considerably 4 longer and had a mean of 10.3 s across all individuals. On a few occasions (less than 5 5% of trials) the ITI either exceeded 60 seconds or was undefined (for the first trial 6 for each block) these trials were removed from further analysis. all trials with an EMG activity > 50mV during the 100ms prior to stimulation, or with 12 MEP amplitudes more than 2.5 standard derivations away from the mean were 13 excluded from further analysis. On average 6.3% of trials were excluded from further 14 analysis due to these criteria. 15 Phase at Stimulation: The phase at the time point of stimulation was estimated from 16 the recorded data using a continuous Morlet wavelet transform. The transform was 17 done for 51 frequency scales across the mu-band and the phase of the optimal 18 frequency scale 100ms prior to stimulation was projected to the stimulation time 19 using the same algorithmic procedure as described previously [22]. Depending on 20 the estimated phase, all TMS trials were sorted in a post-hoc analysis into four 21 distinct phase bins (0°, 90°, 180°, 270°). 22 Statistics: To test the hypothesis that the cortico-spinal excitability is modulated by 23 mu power at the time of stimulation while simultaneously assessing the relationship 24 between power, phase and interstimulus interval (ISI) on the individual trial basis we 1 performed a mixed-effect analysis that included mu-power fraction (categorical -2 high, low), the mu-phase (categorical -0°, 90°, 180°, 270°) and the inter-stimulus-3 interval between two trials (continuous -ISI) as fixed effects and the participant 4 number as a random effect. Statistical analysis was performed using the statistical 5 software package R (https://www.r-project.org ). Mixed effects analysis was 6 performed using the lme4 package (Team RC 2018) and all continuous variables 7 were log-transformed before they were entered in the model. The significance 8 threshold for null hypothesis testing was set to p<0.05. 9 To directly test evidence for the null hypothesis we additionally used Bayesian 10 analysis of covariance as implemented in Jasp v. 0.11.1 with the MEP as dependent 11 variable, high/low power and phase discretized into four bins as a fixed-factors, 12 subject as a random factor and included the logarithm of the ISI as a covariate. Online power-triggered EEG-TMS: 3 We verified that the online power triggering worked by assessing the power fraction 4 estimate for the non-stimulated trials in a 500ms window centered at the intended 5 stimulation timepoint. The mean power fraction was 0.15 ± 0.07 in the low-power 6 triggered condition and 0.30 ± 0.12 in the high-power condition, indicating that the 7 pericentral mu-power was successfully targeted by brain-state informed spTMS 8 ( Figure 2A). The ISI during the low-power was 13.9 ± 4.2 ms and 10.5 ± 2.0 ms during 9 the high-triggered condition. A paired t-test (unequal variance) testing for 10 differences in the ISI indicated that duration was significantly influenced by the 11 triggering condition (t14 = 2.47; p= 0.02) ( Figure 2B). The accuracy of the phase detection algorithm was evaluated by comparing the 7 projected phase to a centered phase estimate for non-stimulated trials as also 8 described in a previous publications [22]. As expected, this revealed that the phase 9 estimation accuracy was lower for the low power condition (q25) with a mean 10 absolute error of 61°, whereas the mean absolute error in the q75 condition was 11 comparable to our previous work with a value of 47° [22]. 12 13 Mu-rhythm extracted by source projection: The average MEP amplitude was 1.16 ± 14 0.39 mV across both conditions. The mean MEP amplitude in the high-power 15 triggered condition was 1.14 mV compared to 1.17 mV in the low-power triggered 16 condition. The linear mixed-effects model, that treated mu-power and mu-phase 17 and ISI as fixed effects and participant as a random effect showed no significant main 18 effect for power (x 2 (1) = 0.32; p=0.57) ( Figure 3A). Also, the main effect of phase was 19 not significant (x 2 (3) =1.54; p=0.20) (Figure 3B), while ISI significantly modulated the 20 MEP amplitude (x 2 (1) = 5.14; p=0.02). None of the interaction terms were significant. 21 The main effect of ISI was caused by an increase of MEP amplitude at when pulses 22 where given at longer intervals ( Figure 3C).  between extracted power and cortical excitability, we re-referenced the EEG data 14 using a sensor level Laplacian montage centered on M1-HAND, similar to previous 15 power-triggered experiments [23,24] and Laplacian mu-oscillations for each TMS 16 pulse were extracted post-hoc. The linear mixed-effects model, that treated 17 Laplacian mu-power fraction and mu-phase and ISI as fixed effects reveal a 1 significant power-effect (x 2 (1) = 5.91; p=0.01) indicating that the MEP amplitude was 2 larger when Laplacian mu-power was high ( Figure 4A). For phase, the Laplacian mu-3 extraction agreed with the initial source-projected data and showed no significant 4 effect of mu-phase (x 2 (3) = 0.56; p=0.64) ( Figure 4B). Figure 4 displays the MEP as a 5 function of phase to further illustrate that no effect of phase could be seen. The ISI 6 effect was significant as in the analysis of the source projected data (x 2 (1) = 3.9; 7 p=0.04).  it is interesting to note the spatial differences between the tested montages. The 17 Laplacian montage was more posterior with an average location over the postcentral 18 gyrus (i.e. the somatosensory cortex) while the source projection was on average 19 located over the precentral gyrus (i.e. the primary motor cortex). These results 20 imply that not all sources of pericentral mu-activity in the sensorimotor hand areas 21 modulate cortico-spinal excitability. We cannot exclude that tangentially oriented Whatever the relationship between pericentral mu-activity and MEP amplitude may 22 be, any relationship will not be rigidly expressed regardless of the magnitude of 23 cortex stimulation (i.e., the TMS intensity) or the ongoing level of neuronal activity 24 (i.e. the firing rate) [24]. This notion is corroborated by a modelling study using a 1 simple non-compartmental threshold-crossing motoneuron model without dendrites 2 [34]. In that study, Matthews showed that the response of the model neuron to a 3 stimulus depended upon stimulus strength, synaptic membrane noise, and intrinsic 4 tonic firing rate of the neuron (induced by intrinsic background drive). The study 5 revealed a non-linear relationship between background firing rate and the strength 6 of a brief external stimulus: For weak stimuli, the response increased with increasing 7 intrinsic tonic firing rate but for strong stimuli, the response was maximal at a low 8 firing rate and then decreased for higher firing rates. Matthews concluded that 9 "transferring" these findings to corticospinal neurons makes it unlikely that the 10 magnitude of the descending volley elicited by a given cortical stimulus 11 ('excitability') will always increase with the initial level of cortical activity. 12 Transferring these observations to TMS-EEG studies, it is unlikely that the MEP 13 amplitude elicited by a TMS pulse will monotonically scale with the level of cortical 14 oscillatory activity (and the associated changes in neuronal firing rates). Instead, 15 strong TMS pulses can be expected to elicit larger MEPs during low levels of intrinsic 16 neural activity (e.g. low firing rate at high mu-power), while weak TMS pulses will 17 elicit larger MEPs during high levels of intrinsic neuronal activity (e.g. high intrinsic 18 firing rate at low mu-power). This flip in the sign of the relationship between mu-19 power and MEP amplitude is illustrated in Fig.7. Accordingly, a recent TMS-EEG study 20 demonstrated that high mu-power (e.g. low endogenous activity) was only 21 associated with larger MEP amplitudes during higher TMS intensities [24]. The non-22 monotonic interaction between intrinsic firing rate (produced by transsynaptic drive) 23 and stimulus response suggested by Matthews [34] also reconcile our present 24 findings (i.e., no relationship or positive relationship between power and MEP 1 depending on montage) with the negative power MEP relationship reported in our 2 recent study [22] in which we only examined the highest power quartile in that 3 work.

10
Our results have major implications for the use of power-informed TMS-EEG as a 11 tool to better control the cortical "state" at the time of stimulation and hereby, to 12 render the brain response to TMS less variable [35]. Overall, the effect size of the 13 MEP amplitude modulation was modest and sometimes not detectable using 14 conventional average statistics (e.g. ANOVAs)in those studies that reported a 15 relation between pre-stimulus mu-power and MEP amplitude [22][23][24]. This suggests 16 that the contribution of the pericentral "oscillatory state" to the overall trial-to-trial 17 variability of the MEP is generally low. As long as the interactions between EEG 1 montage, neural background noise, intrinsic firing rate and stimulus intensity are not 2 better understood, the potential of power-triggered TMS to significantly improve the 3 intra-individual variability of single-pulse MEPs remains limited. Additionally, it has 4 to be mentioned that other oscillatory rhythms also have been suggested to 5 modulate cortico-spinal excitability. Especially the sensorimotor beta rhythm but 6 also gamma oscillations and cortico-muscular coherence have also been proposed to 7 modulate cortico-spinal excitability [13, 16, 18-20, 36, 37]. More complex 8 interactions like cross-frequency coupling between cortical oscillation with between 9 difference frequency bands are also possible modulators that have not yet been 10 explored. The interval between subsequent TMS pulses is another underexplored aspect of the 20 TMS-EEG approach. In our study, we ensured that TMS was given at very low 21 repetition frequency with a large jitter. This decision was motivated by several TMS 22 studies showing that continuous quasi-repetitive, or jittered application of supra-23 threshold TMS in the 0.5 -0.3 Hz range can induce changes in the excitatory-24 inhibitory balance in M1 and cause change in the inhibitory/faciliatory balance in M1 1 [38] [39] [40]. Studies also showed that long ISIs (> 0.2 Hz) significantly improve the 2 reliability and lower the variability of intra-individual MEPs [41]. 3 Our experimental procedure resulted in long and highly jittered ISIs, and replicated 4 our previous finding that MEPs tend to be larger when long ISIs are used [22]. This 5 suggests that the preceding pulse may have some conditioning influence on the MEP 6 response produced by the next TMS pulse. These "recency" effects together with the 7 virtue of TMS to shape corticospinal excitability and cortical oscillatory activity, 8 underscore the need that TMS-EEG always need to consider whether or how much 9 the repeated administration of supra-motor threshold TMS pulses "actively" shapes 10 regional cortical excitability and oscillatory activity rather than "passively" probing it. 11 12 Taken together, both experimental and theoretical work suggests that the 13 relationship between regional neuronal activity and corticospinal excitability is 14 complex and may interact with a range of different experimental choices such as 15 stimulation intensity, ISI and the number of given stimuli that influence the signal to 16 noise ration and the intrinsic inhibitory/excitatory balance in the stimulated cortex 17 [21,22,34,42,43]. The relationship between cortical excitability, cortical activity 18 and basic experimental parameters has to be better understood and investigated 19 before brain-state triggered TMs can become a standard technique to reduce inter-20 trial variability and increase the effectiveness of TMS protocols. 21 22