Validation of the number of pulses required for TMS-EEG in the prefrontal cortex considering test feasibility

A B S


Introduction
In 1997, Ilmoniemi et al. established the technique for combining transcranial magnetic stimulation (TMS) and electroencephalography (EEG) simultaneously (Ilmoniemi et al., 1997), and since then the simultaneous TMS-EEG methods have been applied to the primary motor cortex (M1) in numerous studies.Thereafter, this measurement method has been extended and applied to cortical regions other than M1, and now TMS-EEG studies on the dorsolateral prefrontal cortex (DLPFC) in psychiatric disorders have gradually been conducted (Tremblay et al., 2019;Noda, 2020;Arai et al., 2021;Li et al., 2021).
In previous TMS-EEG studies on M1, it has been suggested that to obtain a reliable TMS-evoked potentials (TEP), the number of epochs should be adjusted to around 100 trials, after removing epochs with obvious noise contamination (Lioumis et al., 2009;Kerwin et al., 2018;Hui et al., 2020;Ozdemir et al., 2020;Bertazzoli et al., 2021;Mancuso et al., 2021;Hernandez-Pavon et al., 2023).However, it has not yet been fully validated whether the reference values are directly applicable to brain regions other than M1, including the DLPFC.As such, the reliability and validity of TMS-EEG measurements in the prefrontal cortex are still relatively limited compared to those in M1, and the appropriate number of stimulus pulses required to elicit TMS-evoked responses as reliably as those in M1 have not yet been fully evaluated.In other words, in TMS-EEG experiments, no sufficient consensus has yet been reached on the number of TMS pulses that is both necessary and reasonable to obtain reliable TEPs and TMS-evoked oscillations, especially from the DLPFC (Hui et al., 2019;Cao et al., 2021;Ferrarelli and Phillips, 2021).
In the theory of signal analysis, it is known that the minimum number of trials required depends on the ratio of the magnitude of the signal (S) to the magnitude of the noise (N), or S/N (SN ratio: SNR), which also depends on the square root of the number of trials (T) (Goldenholz et al., 2009).In addition, in the case of an averaged response such as TEP, the total SNR is equal to (S/N)*sqrt(T) (Hernandez-Pavon et al., 2023).On the other hand, in practice, the actual number of trials required also depends on the quality criteria set (which are also qualitative), i.e., the required total SNR.Hence, if the total SNR is unknown or does not take a constant value, its change is non-linear, and thus, apart from the theoretical estimate, the minimum number of trials required needs to be validated by actual experimentation.Furthermore, since the total SNR of TEP changes follow a power law, removing noisy epochs a posteriori (e.g., 10 % of the total epochs) does not affect the overall result as long as a sufficient number of trials have been measured.
Moreover, since the amplitude of the TEP depends on the stimulus intensity applied and generally exhibits sigmoidal nonlinear changes (Noda, 2021), the lower the stimulus intensity, the higher the number of trials to obtain the required total SNR (Saari et al., 2018).Conversely, there is a dilemma in that the higher the stimulus intensity, the stronger the influence of activation by TMS-elicited muscle contraction of cranial muscles as well as contamination by peripheral somatosensory evoked potentials, resulting in a relatively larger magnitude of N. Note that Rosanova et al. advocate, as an expert opinion, that the number of trials required to obtain a high total SNR is in the range of 150 to 300 trials, depending on the stimulus intensity (Rosanova et al., 2012).However, since cortical responses elicited by TMS depends on the stimulation site, verification of the number of TMS pulses required to obtain appropriate TEP, especially from the DLPFC site, requires demonstration through actual experiments and analysis (Lioumis et al., 2009;Casarotto et al., 2010;Kerwin et al., 2018).That is, different target sites may require different numbers of trials for TMS.In particular, stimulation of the prefrontal cortex tends to produce more myoelectric noise due to blinking and eye movements that are not synchronized with TMS as well as muscle contractions at the stimulation site such as the frontal muscles and parts of the temporal and orbicularis oculi muscles that are synchronized with TMS compared to those by M1 stimulation (Hernandez-Pavon et al., 2023).Therefore, it is necessary to remove more epochs containing such noise in the offline analysis.
There have been a few studies of spTMS to the DLPFC that have examined the number of stimulation pulses.The study by Kerwin et al. was designed to examine the reproducibility of the TMS-EEG withinsubject test-retest condition in terms of pulse number, and showed that 60-100 pulses was reasonable (Kerwin et al., 2018).Gogulski et al. also quantified the intra-session reliability of the DLPFC-TEP after TMS to the left DLPFC, showing that reliable TEP can be obtained with as few as 25 pulses if the target and analysis parameters are chosen correctly (Gogulski et al., 2023).Thus, in their study, no consistent number of stimuli was presented for the DLPFC stimulation, although there were differences in conditions in each study.
The present study aimed to determine the minimum number of stimulation pulses required for the TMS-EEG experiments at the DLPFC, with the same subject as well as the same experiment, beyond the theoretical estimation of SNR.We also compared our results with previous studies and proposed a reasonable number of stimulation pulses based on the actual experimental measurements.To this end, we compared the differences and similarities in neurophysiological responses between the four main stimulation pulse conditions: (1) 40 pulses (trials), (2) 80 pulses (trials), (3) 160 pulses (trials), and (4) 240 pulses (trials) of spTMS under a constant 120 % resting motor threshold stimulus intensity.Since most previous TMS-EEG studies targeting the DLPFC have been performed with around 100 trials, we hypothesized that even condition (2) would satisfy the number of TMS pulses needed to obtain the TEPs and TMS-elicited oscillations we originally expected to see, and that the results would be neurophysiologically and statistically comparable to those obtained from conditions (3) and ( 4), but would differ significantly from those obtained from condition (1).

Participants
Data from TMS-EEG experiments in which spTMS was performed on the left DLPFC of young healthy subjects (28.0 ± 5.4 years old; 10 males and 10 females) were used.The eligible criteria for the healthy subjects in this study were as follows: (i) age between 20 and 35 years at the time of consent; (ii) no history of psychiatric illness at screening as assessed by board-certified psychiatrists of Japanese Society of Psychiatry and Neurology; (iii) normal cognitive function with a Mini-Mental State Examination score of at least 27 points; (iv) no substance-related disorders in the 6 months prior to study participation; (v) no contraindications to TMS and magnetic resonance imaging, including magnetic metal implants, pacemakers, or claustrophobia; (vi) no serious or unstable physical condition; (vii) no history of seizures or epilepsy; and (viii) no prescription for central nervous system agents including psychotropic drugs.In particular, in this study, we extracted data from 10 people each to equalize the male to female ratio.The research protocol was reviewed and approved by the Ethics Committee of Keio University School of Medicine (approval number: 20170152) and all participants provided informed consent.The experiments were conducted in accordance with the Declaration of Helsinki.

Data setting
The number of spTMS pulses was allocated to four conditions: (1) 40pulse, (2) 80-pulse, (3) 160-pulse, and (4) 240-pulse, and for each data set, TEP (amplitude and waveform), TMS-evoked oscillations (time--frequency analysis), and the similarity analyses (Lin, 1989;Kerwin et al., 2018;Ozdemir et al., 2021) were conducted to examine statistical difference and similarity in the data quality (the total SNR and precision of TEP) and neurophysiological characteristics under each condition.

Neuronavigation
The stimulus site on the left DLPFC was identified by designating the Montreal neurological institute coordinates (x = -38, y = 44, z = 26) at the individual level using neuronavigation system (Brainsight, Rogue Research Inc., Montréal, Canada) based on each subject's structural magnetic resonance imaging data.

TMS-EEG setup
Stimulus intensity was determined by the following process.First, with the EEG cap on, a hot spot on the left M1 that contracted the right first dorsal interosseous muscle was identified, and the resting motor threshold (RMT), which is the intensity at which TMS to the site elicits motor-evoked potentials of 50 µV or greater 50 % of the time, was identified.Then, based on the RMT, the intensity of 120 % RMT was used as the stimulus intensity of spTMS.
We used an EEG system (TruScan RE; DEYMED Diagnostic s.r.o., Hronov, Czech Republic) compatible with the TMS-EEG experiment.EEG signals were recorded at a sampling rate of 3 kHz, and the impedance between the scalp and the electrodes was kept below 5 kΩ throughout the experiments.Furthermore, we used a monophasic TMS stimulator (DuoMAG MP: DEYMED Diagnostic Ltd., Hronov, Czech Republic) as the TMS stimulator and a figure-of-eight butterfly coil with a diameter of 70 mm × 2 windings as the TMS coil (DuoMAG 70BF; DEYMED Diagnostic s.r.o., Hronov, Czech Republic).The coil was placed tangential to the target site on the scalp and the handle was directed backward at a 45-degree angle to the midline.The stimulus interval of TMS was based on 5-second intervals, with 500 ms of jittering applied to the base interval.In the present study, TMS-EEG data from three sessions of 80 spTMS applied to the left DLPFC of the same subject on the same day were extracted and used in the analysis.
Y. Noda et al.Preprocessing of EEG data TMS-EEG data were preprocessed using EEGLAB v2022.1,TESA 1.1.1,(TESA user manual.https://nigelrogasch.gitbook.io/tesa-user-manual/; Rogasch et al., 2017;Rogasch et al., 2022), and customized scripts on MATLAB (R2023a, the MathWorks Inc., Natick, MA, USA).To minimize the effects of subject time and fatigue as much as possible, we first collected three 80-pulse sessions of the DLPFC-spTMS experiments performed on the same subject on the same day, concatenated these data into one continuous EEG data set.
First, EEG data was epoched on the TMS-EEG data with respect to the trigger signal, and baseline correction was performed using data from the − 500 to − 150 ms interval.Next, automatic rejection of bad electrode data due to poor contact, etc. was performed.Specifically, for electrode removal, the z-score within the interval 30-300 ms after TMS was calculated for each electrode over all epochs, z-score of 3 was set as the threshold value, and electrodes with a median z-score for all epochs exceeding the threshold value were automatically removed.In addition, when removing bad channels in the preprocessing, our settings were set to preserve the five electrodes (F1, F3, AF3, F5, and F7) corresponding to the DLPFC region without automatically removing them.For epoch removal, an amplitude of 1,000 μV was used as the threshold, and epochs containing large noise exceeding the threshold were automatically removed.Afterwards, data in the − 5 to 25 ms interval was cut off to exclude TMS artifacts, and then zero padding was performed.At this point, however, the number of epochs was set to be over 210 out of 240 epochs.
Second, independent component analysis (ICA) was applied twice for noise removal in the preprocessing of TMS-EEG data.In the first ICA (EEGLAB FastICA 2.5), we identified and removed the ICs corresponding to TMS decay artifacts (mean ± S.D.) ((1) 3.2 ± 1.0 ICs; (2) 3.9 ± 1.7 ICs; and (3) 4.6 ± 1.7 ICs).Here, the − 5-25 ms interval from which TMS artifacts were removed was spline-completed before filtering, and the same time segment was cut again before applying ICA.Then, we applied a bandpass filter (forward-backward 4th order Butterworth filter) from 0.5 to 100 Hz and a notch filter between 48 and 52 Hz.The EEG data were downsampled to 1000 Hz before the second ICA (EEGLAB infomax (runica)), and spherical completion was applied to the electrode portions of the data that had previously been removed.In the second ICA, the remaining artifacts (ICs: mean ± S.D., (1) 20.4 ± 6.5 ICs; (2) 23.2 ± 6.2 ICs; (3) 25.0 ± 6.8 ICs) such as eye movement/blink, myoelectric noise, electrode noise, TMS-induced muscle noise, and heartbeat noise were removed.Finally, the EEG data of − 5-25 ms interval was again spline-completed and re-referenced to the average of all electrodes.Subsequently, the EEG data were randomly reconstructed into datasets for four stimulation pulses conditions: (1) 40-pulse, (2) 80-pulse, (3) 160-pulse, and (4) 240-pulse conditions.Thus, for this analysis, we assumed that approximately 12 % of all epochs in each condition were removed as noisy epochs during the preprocessing stage.The number of epochs to be removed was set to be larger from the preprocessing of this analysis.For example, (4) 240 pulses were actually unified to 210 epochs for the calculation.

Time-frequency analysis
For the time-frequency analysis, we used a MATLAB custom script based on the FieldTrip toolbox (Oostenveld et al., 2011).Total power was calculated for each stimulus-frequency condition.For time--frequency analysis, a continuous Morlet wavelet transform (time window was dynamically scaled from 3 to 7 Hz, with frequencies displayed as logarithms) was used for the range of 4 to 48 Hz.The average power of all trials at each frequency and each time point at each electrode site was calculated by dividing by the average power of the baseline interval (− 300 -− 100 ms).Finally, the average power of the 5 electrodes (F1, F3, AF3, F5, and F7 electrode sites) corresponding to the stimulation sites on the left DLPFC was calculated.

Analysis of similarity index of TEP waveforms
To evaluate the similarity of the TEP waveforms between each pulse number condition, the following two indices were calculated: the concordance correlation coefficient (CCC) and Similarity Index (SI) (Lin, 1989).Indeed, the CCC and SI have been previously applied as quantitative indices of similarity of TEP (Kerwin et al., 2018).The CCC is a similarity index that accounts for both the covariance and absolute distance between two data distributions (Kerwin et al., 2018;Mancuso et al., 2021;Rocchi et al., 2021), while the SI can calculate the degree of similarity in the direction of the amplitude vectors regardless of the magnitude of the waveforms (Ozdemir et al., 2021).Note that both the CCC and SI are indices that range in value between − 1 and 1.
The CCC was calculated using the following formula: where, x is defined as the number of stimulation pulses condition, including conditions (1), (2), and ( 3), increased at 20-pulse intervals from 20-to 220pulse.Then, y is the condition (4).σ xy denotes the covariance of x and y, σ x 2 and σ y 2 denote the variance of the distribution of x and y, and μ x and μ y denote the mean of the distribution of x and y, respectively.
The CCC at each time point for each electrode (F1, F3, AF3, F5, and F7 electrode sites) was calculated with 30-200 ms as the time of interest (TOI) for each pulse number condition.Here, we calculated the CCC over the entire analysis interval 30-200 ms (the entire interval that includes the late as well as the early component), which is often used in TEP analyses.These values were then averaged over the above 5 electrode sites corresponding to the region of interest (ROI), the left DLPFC, and compared to over time for (1) 40-pulse condition, (2) 80-pulse condition and (3) 160-pulse condition, based on (4) 240-pulse condition, respectively.In addition, TOI was divided into an overall interval, the early component (30-60 ms), and the late component (150-250 ms) and averaged for comparison of the number of stimulation pulses condition in units of 20-pulse.For conditions (1), ( 2) and (3), the CCC was calculated based on 1,000 random samplings, and the average value of these calculations was used as the final output.
The SI was calculated using the following cosine similarity formula: where, A is defined as the number of stimulation pulses condition, including conditions (1), (2), and (3), increased at 20-pulse intervals from 20-to 220-pulse, while B as the condition (4), A it and B it are the i-th vectors of all electrodes at time t for each pulse number comparison A and B, and n is the number of electrodes.

√
The SIs of the comparison between A (each number of stimulation pulses condition, that is, (1), (2), and (3)) and B (condition (4)) were Y. Noda et al. calculated, respectively, with the TOI set to 30-200 ms and the ROI as the 5-electrode average corresponding to the stimulation site on the left DLPFC.As with the CCC calculation, the SI was calculated based on 1,000 random samplings for the number of stimulation pulses condition from 20 to 240 pulses with 20-pulse intervals, and the average value of these calculations was used as the final output.The within-subject similarity between the two conditions was then calculated by averaging the diagonals of the SI matrix.

Statistical analysis
First, the five characteristic peak or trough values for TEPs were compared among the four conditions of 40-, 80-, 160-, and 240-pulse using one-way analysis of variance.The Bonferroni method was used to correct for multiple comparisons (α = 0.05/5).We also created topoplots corresponding to each peak or trough in the TEP waveform for each number of stimulation pulses condition.A cluster-based permutation test (α = 0.05) of the electrode region was performed for differences in potential distribution between conditions (1), (2), and (3) and condition (4).FDR correction was applied to correct for multiple comparisons.Next, we calculated the total power in each time range (early component: 30-60 ms, late component: 100-200 ms, and overall range: 30-200 ms) and each frequency band (theta: 4-7 Hz, alpha: 8-13 Hz, beta: 14-30 Hz, and gamma: 31-48 Hz) for each number of stimulation pulses condition, and compared the differences in time-frequency results between each number of stimulation pulses condition in a multivariate analysis of variance.Then, as for the actual TEP waveforms and TMS-elicited oscillations that were finally extracted, we assumed that the neurophysiological profile that we aim for in the TMS-EEG testing may not change significantly among the four conditions, although theoretically, the increase in the total SNR of TEP with increasing number of pulses will increase the accuracy and precision.In other words, we assumed that when the 80-pulse condition is regarded as the benchmark, the CCC and SI between the 80-pulse condition as the reference and the 160-pulse or 240-pulse condition would fall within a statistically equivalent range.In accordance with previous studies, the CCC of less than 0.6 was defined as weak correlation, the range of 0.7 to 0.8 as moderate correlation, and the range of 0.8 or higher as strong correlation (Lin, 1989).For the SI, on the other hand, since there is no explicit cut-off value for high similarity, we set the cutoff at 0.7, which is the criterion for determining a high correlation in a general correlation analysis.

TEP waveforms and their topoplots across the conditions
Butterfly plots for each of the four conditions and TEP waveforms at the DLPFC site are shown in Fig. 1.TEP waveforms were visually similar between the four conditions.In fact, statistical comparison of the five characteristic peak or trough values revealed no significant differences among the four pulse conditions (F(3, 76) = 0.11, p = 0.955; F(3, 76) = 0.02, p = 0.995; F(3, 76) = 0.04, p = 0.996; F(3, 76) = 0.16, p = 0.992; F (3, 76) = 0.12, p = 0.948).In particular, the TEP waveform under the 80-pulse condition did not appear to be qualitatively degraded compared to the waveform under the 160-pulse or 240-pulse conditions.The topoplots of the overall potential distribution of the TEP for each electrode are shown in Fig. 2.There were no significant cluster-based differences in the electrode distribution of TEP potentials among the number of stimulation pulses conditions.

Time-frequency maps for each condition
The total power in each TOI and in each frequency of interest (FOI) is presented as mean ± S.E.(see Fig. 3).The TEP power at three TOIs for the overall interval, early component, and late component, and four FOIs for the theta, alpha, beta, and gamma bands in the four conditions of 40, 80, 160, and 240 pulses were compared by multivariate analysis of variance.No significant differences were found between each stimulus pulse condition (Wilks' lambda = 0.747, F(3, 76) = 20.71,p = 0.886).

Results of the similarity index (SI)
Three SI matrix diagrams comparing the condition (4) 240-pulse with the conditions (1) 40-pulse, (2) 80-pulse, and (3) 160-pulse are shown in Fig. 5(a), (b), and (c).The within-subject correlations for the same subject on the diagonal in the SI matrix diagram for both comparisons between (2) 80-pulse and (4) 240-pulse conditions and between (3) 160-pulse and (4) 240-pulse conditions almost consistently exceeded 0.7, indicating that the SIs for each condition comparison were strongly correlated.In contrast, the SI for the comparison between (1) 40-pulse and (4) 240-pulse conditions was less than 0.7.Considering only the average value of SI, Fig. 5(d) shows that the 60-pulse condition exceeded the SI criterion.However, the 80-pulse condition or more robustly exceeded the SI criterion, including the error range.

Discussion
This study demonstrated the following results.There were no obvious differences in both the peak or trough values of TEPs and time-frequency powers for the 40-pulse, 80-pulse, 160-pulse, and 240pulse conditions.Furthermore, equivalence was confirmed for each index with respect to the similarity analysis for the comparison of the 80-pulse vs. 240-pulse conditions and for the comparison of the 160pulse vs. 240-pulse conditions, but not for the comparison of the 40pulse vs. 240-pulse conditions.
The CCC is a similarity index that incorporates the magnitude and gradient of change in the TEP waveform (Kerwin et al., 2018;Mancuso et al., 2021;Rocchi et al., 2021).Thus, the finding that the CCC was consistently greater than 0.8 between each of the pulse number conditions in the TEP waveform observed in the interval from 30 to 200 ms immediately after TMS indicates that the 80-pulse condition is comparable to the 160-and 240-pulse conditions not only in the qualitative aspect of visual appearance, but also in the quantitative analysis of the degree of similarity, indicating that a reliable TEP waveform can be obtained.A similarity analysis of TEPs between each number of stimulation pulse conditions, indexed by CCC with respect to our experimental data, indicated that the mean CCC in the overall time interval and the late component exceeded 0.8 at least in the 40-pulse condition; however, the mean CCC did not exceed 0.8 for the early component until at least 60-pulse condition.Nevertheless, our results of higher mean CCC in the overall time interval compared to the early component were also consistent with the results of a previous test-retest validation study of single-pulse TMS to the DLPFC recently reported by Gogulski et al. (2023).Since the late component is more robust and stable than the early component, it may be possible to extract the late component of TEP, including N100 and P180, even with a relatively small number of stimulation pulses, if appropriate experimental measures for noise and substantial denoising in the data analysis are implemented.In fact, the TEP components most reproducibly reported in TMS-EEG studies to date are N100 and other late components (Kerwin et al., 2018).
In addition, the similarity analyses in this study used a cutoff value of 0.8 for CCC, which is slightly more stringent than the 0.7 criterion used by Gogulski et al. (2023), to explore more rigorous stimulation pulse conditions.Consequently, Gogulski et al. reported that a minimum of 25 pulses of TMS stimulation could generate a reliable TEP (Gogulski et al., 2024).In this regard, our analysis showed similar results to Gogulski et al. for the overall time interval and the late component when the cutoff was set at 0.7 (Gogulski et al., 2023).However, our analysis showed that even with a CCC cutoff of 0.7, for the early component, 25 pulses of TMS stimulation would likely not generate a sufficiently stable TEP.Thus, taken together, we found that the TEP waveform can be stabilized with a minimum of 80 to 100 pulse stimulation trials per session at all TOIs for the overall time interval, early component, and late component when using the CCC as an index (Fig. 4(b)).Moreover, our observation was also consistent with the findings of Kerwin et al. who showed that TEP waveforms become stable and reach optimal CCC values after 60 to 100 pulse stimulation trials per session, and that the stability of TEP saturates even as the number of pulses is increased beyond these numbers (Kerwin et al., 2018).
On the other hand, the SI evaluates the similarity of orthogonal component vectors to the time axis (Ozdemir et al., 2021).Thus, the range of the measured SI, which represents the mean of the withinsubject correlation coefficient ±2 standard errors, clearly exceeds 0.7 for the stimulation conditions with 80 or more pulses, indicating that the 80-pulse condition is quantitatively comparable to the 160-and 240pulse conditions in terms of similarity of TEP waveforms.Indeed, the SI matrix maps exhibited high similarity within the same subject across the conditions (i.e., 80-pulse, 160-pulse, and 240-pulse conditions) but not between the 40-pulse and 240-pulse conditions.Taken together with the CCC and SI results, we could demonstrate that in TMS-EEG experiments with single TMS pulses to the DLPFC, at least 80 pulse trials per session are conservatively required to extract a reliable and stable TEP waveform.However, if the objective is to look at the late components of the TEP waveform or rough findings in the overall time interval, even 40-pulse trials may provide the minimal findings.Moreover, given the feasibility of the TMS-EEG test itself, including the physical demands on the subjects and the manpower required by the investigators, as well as the potential for widespread use and generalizability of TMS-EEG in the future, our results suggest that it is not necessary to be overly conservative in the number of stimulation pulses, e.g., to always deliver more than 100 stimulation pulses per session.
This study has several limitations.First, we classified the number of stimulation pulses into four conditions ((1) 40-pulse, (2) 80-pulse, (3) 160-pulse, and (4) 240-pulse) and compared them to demonstrate the validity of the 80-pulse condition from a neurophysiological perspective, but did not identify the optimal number of pulses for the DLPFC TMS-EEG testing.For the purpose of identifying the optimal number of pulses, it is necessary, for example, to plot a dose-response curve from 40 to 320 pulses in 40-pulse steps and identify the number of pulses at the point where the accuracy reaches a plateau.However, we did not conduct such analyses in this validation study because the main objective was to confirm whether the pulse number condition of around 100 pulses, which is used in many laboratories worldwide, is valid as a TMS-EEG neurophysiological testing for the DLPFC, as a more realistic problem set up.Therefore, we will consider this issue in the future.Second, since the TEP response obtained by TMS depends on the stimulus intensity, it is necessary to systematically examine the changes in reliability, validity, and accuracy of TMS-EEG with stimulus intensity as a variable in the future.Third, since we used the DLPFC-spTMS data obtained from young healthy subjects, this result may not directly apply to the elderly healthy subjects or neuropsychiatric patients.That is, when tested under the same conditions, the precision of TEP may be somewhat lower in the elderly subjects or patients with psychiatric disorders than young healthy people.Moving forward, it is necessary to examine what clinical-epidemiological background factors may be responsible for disturbing the DLPFC TMS-EEG neurophysiological index.In the field of TMS-EEG research, the most of experiments have customarily been conducted with around 20 samples, thus we also followed that size in this study.
The 80-pulse condition, which we originally envisioned as the number of pulses needed to extract reliable and valid TEPs in this spTMS for the DLPFC, was found to be particularly satisfactory from a neurophysiological perspective compared to the 40-pulse, 160-pulse, and 240pulse conditions.This finding would contribute to the feasibility of applying the DLPFC TMS-EEG examination to vulnerable populations, including those with neuropsychiatric disorders.

Role of funding source
This research received no external funding.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Butterfly plots of all participants' means in the four pulse conditions and TEPs for all subject means in the left DLPFC.Fig. 1(a), (b), (c) and (d) show butterfly plots for each pulse condition ((1), (2), (3), and (4)).(e) Mean TEP (n = 20) in the left DLPFC (F1, F3, AF3, F5, and F7 electrode sites) and a comparison of four stimulation pulse number conditions across five characteristic peaks or troughs of TEP.The solid lines (green line: (1) 40-pulse condition; red line: (2) 80-pulse condition; blue line: (3) 160-pulse condition; orange line: (4) 240-pulse condition) represent the mean TEP for each condition and the shaded areas depict the range of ±S.E.The violin plots for each stimulation pulse condition corresponding to the five components characteristic of TEP (P30, N45, P60, N100, and P180) are shown.Black vertical bars indicate line segments connecting the first and third quartiles of peak potentials.White circles indicate the median peak potentials while the horizontal bars in colors (green, red, blue, and yellow) for each pulse number condition indicate the mean peak potentials.Footnotes: TEP: transcranial magnetic stimulation-evoked potentials; DLPFC: dorsolateral prefrontal cortex; S.E.: standard error.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2 .
Fig. 2. The topoplots for 40-, 80-, 160-, and 240-pulse conditions at the peaks and troughs of the average TEP waveform.The topoplots for the 40-, 80-, 160-, and 240-pulse conditions are shown from the top to bottom.Each column corresponds to each peak or trough in the TEP waveform in our study.

Fig. 3 .
Fig. 3.The total power in each time of interest (TOI) and in each frequency of interest (FOI).The mean total power in the time-frequency analysis (n = 20) at the DLPFC is depicted for each pulse condition ((1), (2), (3), and (4)).For the TMS-evoked oscillations across the conditions, similar degrees of oscillations were observed in the early component (30-50 ms), centered in the beta frequency range.(e) The violin plots for the four pulse conditions are shown.Specifically, green violin plots represent (1) 40-pulse conditions, red violin plots represent (2) 80-pulse conditions, blue violin plots represent (3) 160-pulse conditions, and yellow violin plots represent (4) 240-pulse conditions.The overall time interval, early component, and late component are displayed from the top to bottom, respectively, while the columns show the frequency components in the theta, alpha, beta, and gamma bands from left to right.Black vertical bars indicate line segments connecting the first and third quartiles of peak potentials.White circles indicate the median peak potentials while the horizontal bars in colors (green, red, blue, and yellow) for each pulse number condition indicate the mean peak potentials.Footnotes: DLPFC: dorsolateral prefrontal cortex; TMS: transcranial magnetic stimulation.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4 .
Fig. 4. Comparative analyses for the concordance correlation coefficients (CCCs) between each number of stimulation pulses.The figure (a) shows the results of the similarity analysis using the mean CCC index (n = 20) for the 240-pulse condition as a benchmark compared to the 40-pulse condition (green line), 80-pulse condition (red line), and 160-pulse condition (blue line), which all consistently showed values above high degree of similarity threshold, 0.8 (dot line).Here, the CCC quantifies the similarity of the amplitude and gradient components of the TEP waveforms.The shaded area indicates ± S.E. of CCC for each stimulation pulses condition.The figures (b) plot the changes in the CCCs between each pulse condition when the number of stimulation pulses is increased from 20 pulses to 240 pulses, in 20-pulse intervals.These graphs display the CCC changes calculated by dividing the TOI into three-time segments: overall interval, early component, and late component, respectively.Footnotes: CCC: concordance correlation coefficient; TEP: transcranial magnetic stimulation-evoked potentials; S.E.: standard error.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5 .
Fig. 5. Comparative analyses for the similarity indices (SIs) between each number of stimulation pulses.The figures (a), (b), and (c) display the results of similarity analyses using the mean SI (n = 20) for the 240-pulse condition as a benchmark compared to the 40-pulse condition (a), the 80-pulse condition (b), and the 160-pulse condition (c), respectively.The figure (d) plots the SI changes between each number of stimulation pulses condition when it was increased from 20 pulses to 240 pulses, in 20-pulse intervals.Here, the SI represents the similarity of the orthogonal vector component to the time axis of the TEP waveform.The bars in the figure indicate ±2 standard errors.Footnotes: SI: similarity index.