TMS-EEG Shows Mindfulness Meditation Is Associated With a Different Excitation/Inhibition Balance in the Dorsolateral Prefrontal Cortex

Experienced mindfulness meditators and healthy controls between the ages of 18 and 65 were recruited to take part in the study. All participants gave their informed consent prior to participating in any study activities. Participants were recruited through community advertisements, posters and flyers. Advertisements were posted on social media platforms and flyers were distributed at meditation centres in Victoria, Australia, and within the broader community. Participants responded to the advertisements via phone or email and were screened for inclusion/exclusion criteria. Respondents were excluded from the study if they (1) had received a clinical diagnosis of a mental health disorder; (2) had a neurological condition or psychiatric illness (either previously or currently occurring); (3) were currently taking psychoactive medication and/or illicit drugs; or (4) had a history of concussion or traumatic brain injury in which they had lost consciousness for more than 10 min and/or presented to hospital. In addition, anxiety, depression, other psychopathologies and substance use disorders were screened for using the Beck Anxiety Inventory (BAI; Beck et al., 1988), the Beck Depression Inventory (BDI; Beck et al., 1996) and the Mini-International Neuropsychiatric Interview (M.I.N.I. version 5.0; Sheehan et al., 1998), respectively. Participants were excluded if they scored 20 or above on the BDI (moderate to severe depression range), 16 or above on the BAI (moderate to severe anxiety range) or screened positive to any neuropsychiatric conditions in the M.I.N.I.

Meditator respondents were asked to describe their meditation practice. Participants were eligible to participate if their self-described practice adhered to Kabat-Zinn’s definition of mindfulness as "paying attention in a particular way: on purpose, in the present moment, and non-judgementally" (Kabat-Zinn, 1994, p. 4). Additional screening questions were asked to ensure the participants’ meditation practice involved focusing on sensations in the body (e.g. paying attention to the breath or body sensations). To address common methodological issues in meditation research of inadequate control for meditation type and level of experience (Van Dam et al., 2018), the current study employed the following criteria for what could be considered an experienced mindfulness meditation practitioner: Meditator participants, at the time of testing, had to be practicing mindfulness meditation (as per Kabat-Zinn’s definition) for 2 or more hours per week and must have been doing so for a minimum of 2 years. These criteria have been used in previous research (Bailey et al., 2019; Payne et al., 2019). Uncertainties regarding meditation experience and quality of practice, where there was not a clear and obvious match between participant’s practice and the inclusion criteria, were resolved through informal discussion with the participant to seek further details about their practice, and direct consultation with the Principal Investigator (NWB), who is an experienced meditation researcher. In uncertain cases, the final decision to include or exclude participants based on their meditation practice was reached by consensus between the Principal Investigator and one other researcher, after discussion of how closely the details of each potential participant’s practice matched the inclusion criteria, with cases that were still uncertain after discussion being excluded. Healthy controls were included in the non-meditator group as long as they had completed not more than 2 cumulative lifetime hours of any form of meditation practice and met all other inclusion criteria. Meditator participants were instructed to not meditate during the testing session. In addition, participants were instructed to avoid caffeine or alcohol for 24 hr prior to administration of the TMS-EEG protocol. At the start of the session, participants completed a TMS safety screen (Rossi et al., 2009, 2011) and a questionnaire that obtained demographic data (age, years of formal education, gender, handedness, descriptions of their meditation practice including average duration, length and frequency of meditation sessions and current medication use), after which participants completed the BAI, BDI-II and the Five Facet Mindfulness Questionnaire (FFMQ). This was followed by verbal administration of the M.I.N.I. The current study was reviewed and approved by the Alfred Hospital and Monash University ethics committees, with informed written consent obtained for all participants in accordance with the Declaration of Helsinki.

Of the total 75 participants recruited, TMS-EEG data were not obtained for 37 participants. Of these 37 participants, data were not collected for the following reasons: 29 participants declined to receive TMS; 1 participant was excluded from receiving TMS due to safety concerns, as they reported a potential history of seizures; 2 were excluded due to scoring above the moderate to high anxiety range in the BAI; and for 5 participants, time restrictions precluded data collection. The high number of participants declining to receive TMS may have been due to the fact that participants were informed their data would still be useful to the other aspects of the study even if they declined to receive TMS, and that a high number of participants preferred not to receive brain stimulation when it was not clinically useful (although participant reasons for declining to participate in the TMS part of the study were not formally recorded, the research team recalls this as the primary anecdotally provided reason). For a further three participants, TMS-EEG data were excluded from analysis due to technical issues during recording. Data were excluded from analysis for a further single control participant due to reporting meditation experience above the 2-hr lifetime maximum limit (which was revealed after the testing session). The final sample of useable TMS-EEG data consisted of 19 controls aged between 20 and 57 years (10 females, M_age = 29.58 years, SD_age = 11.48 years), and 15 meditators aged between 22 and 64 years (5 females, M_age = 35.67 years, SD_age = 13.81 years).

Each participant took part in a 45- to 60-min single-pulse TMS-EEG data collection block as part of a more extensive session involving EEG recordings and three computerised cognitive tasks. The average total session duration was approximately 3-and-a-half hours and involved the following behavioural tasks: (i) a Go/Nogo task (Hill et al., 2024), (ii) an auditory oddball task (Payne et al., 2019) and (iii) an attention blink task (Bailey et al., 2023a). Following these tasks, the TMS resting motor threshold (RMT) was obtained, then each participant received 105 single pulses of TMS to the left and right DLPFC (with the order counterbalanced between participants).

RMT was determined using electromyography (EMG). Three electrodes were placed surrounding the First Dorsal Interosseous (FDI) muscle on the contralateral (right) hand to the site of stimulation (left M1) and continuous electrical activity of the muscle was observed using Scope V3.7 (ADInstruments, NZ). TMS pulses were delivered to the motor cortex with increasing intensity until a muscle response was observed. RMT was determined as the intensity at which a 1-mV peak-to-peak amplitude motor-evoked potential was evoked in the FDI muscle in at least three out of five trials (Enticott et al., 2012; Rossini et al., 2015).

After RMT was obtained, participants were provided with 105 single pulses of TMS (delivered at a rate of one pulse every 4 s \(\pm\) 10% jitter to control for expectancy effects) to the left- and right-DLPFCs (L-DLPFC, R-DLPFC), respectively. Stimulation intensity was set at 110% of the RMT. Monophasic stimulation was applied using a Magstim 200 stimulator, which delivers monophasic pulses (Magstim Company Ltd., UK) with a figure-of-eight magnetic coil (wing diameter of 70 mm). This approach was employed as prior research has indicated that it evokes larger TEPs than biphasic stimulation (Casula et al., 2018), while evidence also suggests that monophasic and biphasic stimulation both produce similar spatiotemporal distributions of cortical responses (Biabani et al., 2019). Pulse delivery was automated using Signal version 3.08 software (Cambridge Electronic Design Ltd., UK) connected to a stimulator trigger box. The coil was placed tangentially on the head and positioned with the centre of the coil above frontal EEG electrodes, namely F3 and F4, in accordance with the international 10–20 system for anatomical positioning of TMS stimulation of targeted brain regions. Electrodes F3 and F4 were chosen based on their spatial proximity to the L-DLPFC and R-DLPFC, respectively (Herwig et al., 2003). The TMS coil handle was pointed towards the rear of the participant and 45 degrees laterally from the mid-sagittal line (in accordance with current conventions; Enticott et al., 2012; Rossini et al., 2015). To control for auditory-evoked potentials due to loud clicking noises associated with delivery of each pulse, white noise was delivered using Neuroscan 10Ω ¼ stereo intra-auricular earphone inserts (Compumedics, Melbourne, Australia). The amplitude of white noise was increased gradually to a level that was loud enough to mask the click noises but still comfortable for the participant. Prior to administration of the TMS protocol, participants were asked to remain awake and to keep head movements to a minimum.

The FFMQ has been shown to provide reliable measures of overall trait mindfulness (McDonald’s ω > 0.89) (Lecuona et al., 2020). Cortical electrical activity was obtained using a 64 channel Compumedics Neuroscan Quik-Cap (Compumedics Ltd., Australia) with Ag/AgCl sintered electrodes in the standard 10–20 spatial format. The online reference electrode was located along the midline between CPz and Cz. Additional electrooculography recordings were obtained using a single supraorbital electrode positioned above the left eye in line with the pupil. ECI-Electrogel (Electro-Cap International, Inc., USA) was applied to each electrode and impedances were kept below 5kΩ throughout the duration of the session. EEG data were recorded using DC-coupled 1000 × amplification of the EEG signal, acquired using a SynAmps2 amplifier (SynAmps2, EDIT Compumedics Neuroscan, Texas, USA) and recorded using Neuroscan Acquire software V4.5 (Compumedics Ltd., Australia). As per standard TMS-EEG recording guidelines, EEG was recorded using a high sampling rate of 10 kHz (for review, see Farzan et al. (2016)). Amplifier saturation during TMS delivery was mitigated by using an operating range of ± 200 mV with a DC-2 kHz low pass filter applied.

EEG data were processed and analysed offline in Matlab R2018b (The MathWorks, USA) utilising the following toolboxes: EEGLAB (Delorme & Makeig, 2004), FieldTrip (Oostenveld et al., 2011), TESA (Rogasch et al., 2017) and Randomization Graphical User Interface (RAGU; Koenig et al., 2011). Prior to EEG data analysis, channels dedicated to muscle and eye artefact were removed. Initial processing involved removal of the large amplitude TMS pulse (− 5 to 15 ms), linear interpolation and the generation and concatenation of epochs (− 1 to 2 s around the TMS pulse). Data were downsampled to 1000 Hz and the removal of DC offset was achieved by baseline correcting each epoch to − 500 to − 50 ms prior to the TMS pulse. Large muscle artefacts, faulty channels and otherwise bad epochs were removed by visual inspection of the concatenated EEG trace (< 10% of total number of epochs selected for rejection). Two rounds of Independent Component Analysis (ICA; FastICA algorithm, "tanh" contrast function and semi-automated component) were implemented via the TESA toolbox in EEGLAB (Delorme & Makeig, 2004; Rogasch et al., 2017). The EEG signal was broken down into separate components via ICA. TESA then categorised each component as consisting of either neural or non-neural activity. The first round of ICA was implemented to identify and remove the large TMS pulse artefact and eye blinks. Once removed, data were bandpass filtered between 1 and 100 Hz (4th order Butterworth filter) and 50-Hz line noise was removed using a 50-Hz notch filter (band-stop filter between 48 and 52 Hz, 4th order Butterworth filter). Data were again visually inspected and bad epochs and channels were removed. A second round of ICA was performed using TESA’s semi-automated selection of neural and non-neural (eye blink/movement and muscle artefacts) components with visual inspection for conformation and rejection of artefactual activity. Missing channels were then interpolated using a spherical interpolation method (EEGLAB; pop_interp function). All channels were re-referenced to an average of all channels. All epochs were then collapsed into a single TEP by averaging each channel over all epochs for every individual and separately for both conditions (L-DLPFC and R-DLPFC). For further information on the TMS-EEG data analysis pipeline used in the current study, please see Rogasch et al. (2017). An example TMS-EEG analysis pipeline similar to the present study is provided at https://​nigelrogasch.​gitbooks.​io/​tesa-user-manual/​content/​example_​pipelines.​html (Rogasch, n.d.).

Average TEPs were generated for each subject and separately for each stimulated region of interest (L-DLPFC and R-DLPFC). Analysis focused on the a priori determined peaks of interest, namely P60 and N100, which have been suggested to measure excitatory and inhibitory neurotransmission, respectively (Rogasch et al., 2017). To measure the P60 and N100 components, the average amplitude over pre-specified time windows was used (P60, 55–75 ms; N100, 90–130 ms) using automated scripts in Matlab. An analysis was also performed on the ratio of the P60 to the N100. Due to variable voltage shifts across the TEP epoch, some P60 values were negative when measured as averaged amplitude within the window of interest, a feature of the data which would result in non-sensible comparisons of the P60/N100 ratio if not addressed. To address this issue, we visually inspected the data to mark the peak P60 and N100 latencies. We then computed the difference between the peak P60 and N100 amplitude and the average of the minimum value within 50 ms of the TEP of interest. This provided positive values for all P60 deflections, and negative values for all N100 deflections, representing the size of the deflection compared to the surrounding ongoing TEP.

To confirm adequate matching between control and meditator groups, the following statistical analyses were performed in SPSS (version 23): Independent samples t-tests were run to test for demographic group matching (age, years of formal education) and matching for scores on self-report measures (BAI, BDI and FFMQ). Chi-square tests were used to verify that groups did not differ in gender and handedness. A number of assumption checks and outlier checks were implemented prior to our statistical analyses. If the assumptions of a planned test were violated, we used a test that was robust to the assumption instead. Outliers were winsorised. The full details of these steps are reported in the Supplementary Information (assumption checks and outlier detection section).

To test whether the amplitudes of the P60 and N100 differed between groups, or whether group interacted with hemisphere (i.e. the site of stimulation), two separate mixed model ANOVAs were run in SPSS (version 23). The main effect of hemisphere was not included in our analysis plan, as a main effect of hemisphere did not offer valuable insight into the study’s primary focus, which aimed to determine the potential differences in neural activity associated with regular mindfulness meditation. Further analyses of the ratio of the P60 to the N100 (calculated by dividing each individual’s peak P60 deflection by their corresponding peak N100 deflection; \(\frac{\text{P}60}{\text{N}100})\) were performed using non-parametric Mann–Whitney U test statistics (Mann & Whitney, 1947). To control for multiple comparisons, the Benjamini-Hochberg False Discovery Rate (FDR) method was used (Benjamini & Hochberg, 1995), and p-values adjusted for the FDR are reported as p_FDR. Where null results were obtained, these were complimented by Bayes Factor (BF) analyses to quantify the probability of the null result. BF analyses determine the odds of observing the model describing the null hypothesis over that of the model describing the alternative hypothesis, expressed as an odds ratio with Bayes Factors > 1 favouring the null hypothesis model (Rouder et al., 2017).

To compliment the use of multivariate single electrode and time window analysis, the current study also performed statistical analysis using the Randomization Graphical User Interface software (RAGU; Koenig et al., 2011). The benefit of using RAGU to analyse EEG data is that it ameliorates several methodological and biasing issues by minimising the need for a priori selection of parameters (e.g. electrodes and time windows to analyse). In particular, RAGU uses non-parametric randomised permutation statistical techniques and whole EEG data (all channels, all epochs and all time points) to compare scalp field differences between groups at each time point along the averaged epoch. RAGU employs methods to control for multiple comparisons (Habermann et al., 2018; Koenig et al., 2011) by randomly permuting the data to generate distributions for which the null-hypothesis of no effect holds for each time point. The cumulative size of effects in the real data can then be compared to this distribution to obtain a ‘global count’ p-value, which is the probability of observing the real data at all timepoints in the epoch under the assumption that the null-hypothesis is true. This same process can also be applied to obtain a distribution of lengths of contiguous significant time periods in the randomised data termed ‘duration control’. Time periods of contiguous significance in the real data that exceed the 95% threshold in the distribution of the randomised data are then highlighted as effects with statistically significant durations in the real data. This duration control ensures the overall level of significance is maintained at the a priori determined level of significance (\(\alpha < 0.05\)).

To conduct the statistical tests, RAGU uses the Global Field Power (GFP) measure as a reference free measure of global neural response strength. GFP is equivalent to calculating the standard deviation from the average reference over all electrodes. The first step in the RAGU pipeline is to confirm that there exists a consistent activation of neural sources in response to TMS using the Topographical Consistency Test (TCT). This test ensures that significant results in subsequent RAGU analyses cannot be explained by extreme variation within a single group. The explanation of the TCT is provided in full in Supplementary Information.

Once there is confirmation of a consistent neural activation within groups and conditions using the TCT, further analysis is provided by two separate but related significance tests. The GFP test compared differences in overall neural response strength between groups (meditators and controls) and hemispheres (L-DLPFC and R-DLPFC stimulation) using the GFP as the measure of interest. The between-group/condition difference in the GFP of the real data is then compared to the null distribution of no effect; this distribution is generated by randomised shuffling of the individual observations 5000 times for each measured time point and drawing a distribution of the randomised GFP values. For each time point, the real data are compared to the randomised distribution and a p-value is determined. Values that exceed the 0.05 alpha-level threshold are considered significant, and the proportion of time points of the real data that exceed this threshold across the whole epoch determines the overall significance (i.e. the global count p-value). Contiguous periods of significance across the epoch are then controlled for using global duration statistics.

The GFP test measures differences between groups/conditions of neural response strength. To determine if the topographic distributions of active neural sources differ between groups, the topographical analysis of variance (TANOVA) test is performed. Importantly, this test does not require active sources to be localised to cortical regions to determine whether the topographical distributions differ. To achieve this, the TANOVA first applied the L2 normalisation of the data within groups/conditions by dividing the average group/condition voltage values by its scaling factor (the GFP). Any differences between groups/conditions can then be said to be independent of the global strength of neural activity. The within-group electrode space data is then averaged, and the average electrode space data from one condition/group is subtracted from the average electrode space data from another condition/group. The dGFP is then calculated from this between group/condition normalised difference data. If there are true between-group differences in active brain sources generating the scalp field, then values for the dGFP will be large. Statistical analyses are provided by 5000 randomised permutations and the generation of a dGFP null-distribution of no effect. The global count p-value and duration control statistics are then derived in the same manner as the GFP test. We have reported our outcome statistics to three decimal places to allow sufficient precision to enable our effect sizes to be accurately included in meta-analysis.

Most demographic variables were found to be normally distributed using Shapiro–Wilk’s test for normality and visual inspection of the histograms. However, scores for the BDI-II and BAI failed to meet normality assumptions (BDI-II: controls—W(19) = 0.836, p = 0.004, skewness = 0.854 (SE = 0.524), kurtosis = − 0.649 (SE = 1.014); meditators—W(15) = 0.259, p = 0.006, skewness = 0.976 (SE = 0.580), kurtosis = − 0.096 (SE = 1.121); BAI: meditators—W(15) = 0.832, p = 0.010 skewness = 1.347 (SE = 0.580), kurtosis = − 0.096 (SE = 1.121)). To mitigate the effect of these violations, comparisons between these data were conducted using non-parametric Mann–Whitney U tests, and they were found to be non-significant (BDI-II: U = 110, p = 0.271; BAI: U = 128, p = 0.817). No significant between-group differences were observed in age, gender, years of education or handedness using independent samples t-tests (all p-values ≥ 0.097; results provided in Table 1). However, a statistically significant difference was observed between groups on the FFMQ, with experienced meditators scoring an average 24.49 points higher compared to controls (t(32) = − 4.130, p < 0.001, 95% CI [− 36.57, − 12.41], d = 1.452). Demographic measures and results are presented in Table 1.

The group-averaged single-electrode TEPs are provided in Fig. 1. Butterfly plots of the group averaged TEPs are depicted in Fig. 2, revealing characteristic TEP deflections at N45, P60, N100, P200 and N280.

To investigate whether P60 and N100 differed between groups, mixed model ANOVAs were employed using data obtained from the single electrodes of the stimulated sites (F3 and F4). Preliminary analysis of the separate mixed model ANOVAs of P60 and N100 amplitudes indicated that most assumptions of normality and homogeneity of variance were met (F_max, Box’s M, Mauchly’s Test of Sphericity). However, the assumption of between-group equality of error variance in P60 amplitudes was violated (Levene’s test for equality of error variances: L-DLPFC, F(1, 32) = 0.014, p < 0.05; R-DLPFC F(1, 32) = 0.231, p < 0.05). Unequal group sizes, as in the present study, further exacerbate deviations away from between-group equality of variance. Despite this violation, the mixed model ANOVA F-test has been shown to be robust against deviations from unequal group sizes if the disparity between them are small (largest group size/smallest group size < 1.5; current study 19/15 = 1.27; (Stevens, 2012)). Corrections for violations of the equality of error variances are limited with respect to mixed model ANOVAs. Applying a lower alpha threshold (\(\alpha\) < 0.05) is one suggested method to control for the inflation of a type-I error rate (Allen & Bennett, 2008). However, the current (and subsequent) ANOVA analyses show that no group or factor measures met the requirements for statistical significance, and methods for correcting for this would favour the results towards the null hypothesis of no effect. The interpretations of these null results without correction are therefore warranted.

No significant main effect of group in P60 amplitudes was observed (Group; F(1, 32) = 1.712, p = 0.200, \({\eta }^{2}\) = 0.051). In addition, no significant interaction effect between group and hemisphere was observed (Group × Hemisphere; F(1, 32) = 0.303, p = 0.586, \({\eta }^{2}\)= 0.003). No significant differences in N100 amplitudes were observed between groups (Group; F(1, 32) = 1.128, p = 0.296, \({\eta }^{2}\)= 0.034). In addition, no significant interaction effect between group and hemisphere was observed (Group × Hemisphere; F(1, 32) = 0.130, p = 0.721, \({\eta }^{2}\)= 0.001). The results of the ANOVA tests are displayed in Table 2.

The P60/N100 ratios violated the required assumption of normality in parametric repeated measures ANOVAs using the Shapiro–Wilk test (all p < 0.017). To control for the violation of these assumptions, Mann–Whitney U test statistics (Mann & Whitney, 1947) were used to test for differences in the P60/N100 ratios (results presented in Table 3). Significant between-group differences were observed in the L-DLPFC condition (U = 270.50, p < 0.001, d = 1.571, BF₁₀ = 50.67, significant at FDR adjusted p_FDR= 0.004) and R-DLPFC condition (U = 252.5, p < 0.001, d = 1.447, BF₁₀ = 39.24, significant at FDR adjusted p_FDR= 0.004). Meditators were observed to have higher P60/N100 ratios on average compared to controls, showing mean P60/N100 ratio values of approximately − 0.8 compared to the means for controls which were approximately − 0.67. Since P60/N100 ratio values of − 1 would indicate the P60 and N100 amplitudes were equal in size (but with opposite polarities), this result indicates that meditators P60 deflections were on average more similar in size to their N100 deflections compared to the control participants (who showed N100 deflections that were larger than their P60 deflections, see Fig. 3).

To assess the potential driver of this effect, we performed a post hoc exploration of the absolute difference between the P60 and N100 deflections. These results showed no significant between-group difference in the L-DLPFC condition (U = 108.00, p = 0.242) or R-DLPFC condition (U = 125.00, p = 0.560), indicating that while the relative difference between the P60 and N100 differs between the groups, the result is not driven by a between-group difference in the absolute difference between the P60 and N100. While our groups did not significantly differ in age or sex, they were not directly matched in these factors. As such, we repeated this analysis after randomly excluding four control participants (out of a pool of 12 control participants who fell into a younger age bracket, who, if excluded, would improve age matching between the groups) to improve the age and sex matching between the groups. This control analysis showed the same pattern and significant result as our primary analysis, with significant between-group differences still observed in the L-DLPFC condition (U = 210.50, p < 0.001) and R-DLPFC condition (U = 196.50, p < 0.001). Finally, to confirm differences in the P60/N100 ratios were not simply the result of differences in either the P60 or N100 when these TEPs were measured as deflections (rather than as averaged amplitudes within the window of interest as per our amplitude analyses), we performed two confirmatory repeated measures ANOVAs comparing the P60 and N100 deflection values that were used in computing the P60/N100 ratios. These analyses showed no significant differences between the groups or interactions between group and hemisphere in the P60 or N100 deflections (all p > 0.18). This result aligns with our null result for the P60 and N100 amplitudes when measured averaged across their windows of interest and indicates that the P60/N100 ratio effect is driven by between-group differences in the relationship between the P60 and N100 ratio.

The TCT test showed that there was consistency in the active neural sources contributing to the scalp field throughout most of the epoch (− 100 to 50 ms) for both groups and conditions (see Supplementary Information, Fig. S2; overall p < 0.001). There were, however, short periods of inconsistent neural activity that were observed in all groups and conditions. Inconsistent activity can be observed in periods within the P60 time window used in the single electrode analysis (55–75 ms). Inconsistencies were also observed for meditators in the R-DLPFC condition between 280 and 287 ms, and for controls in the L-DLPFC condition between 312 and 347 ms. Inconsistencies in these periods indicate that significant results obtained in the TANOVA test (described in subsequent analyses) may be due to inconsistency in the topographical distribution of the neural response across individuals within a single group.

The between-group GFP tests for differences in neural response strength were non-significant (group main effect, global count p = 0.485, see Fig. 4). Additionally, no significant interaction between group and condition was observed (hemisphere × group interaction, global count p = 0.610). Further, the brief periods of significance did not survive multiple comparison duration control (global duration statistics: main effect of group = 39 ms; hemisphere × group interaction = 30 ms). Of note, no periods of significance were observed around the time windows of interest in the hypothesis driven analyses of P60 and N100 in agreement with the single electrode analyses. Results of the GFP test indicate that there were no differences between groups in the overall strength of neural response to TMS to the DLPFC for either hemisphere.

In contrast to the null results in the GFP test, significant group differences in the distribution of active neural sources were observed in the TANOVA test lasting from 269 to 332 ms post TMS pulse (global count p = 0.016; Fig. 5a). This period was the only period to survive the multiple comparisons duration control of 32 ms (the main effect of group averaged significance and effect size between 269 and 332 ms were as follows: p = 0.008, \({\eta }^{2}\) = 0.126). While our groups did not significantly differ in age or sex, they were not directly matched in these factors. As such, we repeated this analysis after randomly excluding four control participants to improve the age and sex matching between the groups. This control analysis showed the same pattern and significant result as our primary analysis (p = 0.0176, \({\eta }^{2}\)= 0.119). No significant interaction between group and condition survived the multiple comparisons duration control of 25 ms (hemisphere × group interaction, global count p = 0.655). The significant group effect between 269 and 332 ms post TMS pulse is evidence that there were different spatial distributions of active neural sources contributing to the scalp field during this time. However, as topographical inconsistencies were observed in the TCT (meditators between 280 and 287 ms; controls between 312 and 347 ms), at least some of this effect may have been due to within group variation. Although there was no interaction effect, the TCT showed a different pattern of topographical consistency for stimulation to the different hemispheres. As such, we examined whether the significant main effect of group replicated within a single hemisphere showed consistent topographical activation. The results of this analysis are provided in Fig. 5c and d. No between-group significance was present in the L-DLPFC condition (global count p = 0.289), and no significant periods survived the multiple comparisons duration control of 31 ms (Fig. 5c). A significant between-group difference was observed within the R-DLPFC stimulation condition (global count p = 0.029) which lasted between 270 and 332 ms post TMS (\({\eta }^{2}\) = 0.097) and survived the multiple comparisons duration control of 31 ms. Inconsistencies observed in the TCT with R-DLPFC stimulation were only observed in meditators between 280 and 287 ms (Fig. 5d). The remaining time window (287–332 ms) still exceeded the multiple comparisons duration control of 31 ms. This result suggests that the meditation group showed a different topographical distribution of active neural sources following R-DLPFC stimulation. Further characterisation of the topographical distribution of neural activity over the period of significance is provided in Fig. 5.

The topographical distributions of neural activity averaged across 269 to 332 ms post TMS pulse showed less fronto-midline negativity and greater posterior negativity in meditators (Fig. 6a) when compared to controls (Fig. 6b). The meditator group also showed greater central and temporal positivity compared to controls. In the control group, the positive voltages appeared to be shifted more posteriorly, focused over centro-parietal regions. In addition, controls appeared to show more equivalent centro-parietal positivity across both hemispheres compared to meditators (who exhibited greater positive voltages in electrodes over the left hemisphere). The t-map (Fig. 6c) illustrates the difference (controls subtracted from meditators) in voltage distributions during the significant period. Specifically, the t-map indicates that the differences between groups appear to be strongest around the general region of the left-sided site of stimulation (F3, F5 and AF3 electrodes) with meditators showing more positive neural activity compared to controls. In addition, meditators showed more negative voltages compared to controls in parieto-occipital regions (PO3 and Pz electrodes).

The functional interpretation of TEPs is controversial, with research suggesting that the P60 reflects excitatory synaptic activation and the N100 inhibitory processes, while other research suggests the TEPs are also influenced by auditory processing of the TMS click (Conde et al., 2019; ter Braack et al., 2015). The use of the sound masking protocol in the current study has been shown to reduce the influence of auditory processing on the N100, but residual influence is still suggested to remain (Biabani et al., 2019, 2021; Conde et al., 2019). While this controversy has implications for the interpretation of our results, it does not obviously invalidate our finding of differences in P60/N100 ratio, as it is not clear why auditory processing related effects would differ between meditators and non-meditators, and why any potential difference would influence the P60/N100 ratio. In particular, our previous research that examined neural activity in response to an auditory oddball using some of the same participants that were included in the current study indicated no differences between the meditator and non-meditator groups in auditory processing ERPs (Payne et al., 2019). Additionally, we are not aware of any existing data on the test–retest reliability of the P60/N100 ratio. As such, it is not clear whether the ratio would be consistent within an individual over time. However, given the consistency of the difference between meditators and non-meditators in the P60 to N100 ratio (Fig. 3), even if the measure varies to some degree between testing sessions, we suspect this random variation would be unlikely to provide an explanation for our results.

Another limitation is variability in the definition of the term mindfulness and the inclusion of heterogeneous samples of meditators from different traditions and with different levels of meditation experience are common methodological issues in meditation studies (Van Dam et al., 2018). These factors were controlled for in the current study to some extent by using an inclusion criteria that limited the meditation group to participants who practice breath/body focused meditation (Creswell, 2017; Norris et al., 2018; Tang, 2017; Tang et al., 2015). Additionally, some research has argued that distinctions between different mindfulness meditation practices may not provide much in the way of meaningful differences to neurophysiology, that the differences between meditators and non-meditators are likely to be considerably larger and that there is conceptual ambiguity in the definition of different meditation types and how these factors might affect neurophysiology (Bailey et al., 2023a; Schoenberg & Vago, 2019). Nonetheless, future research could reduce potential heterogeneity from the inclusion of a range of meditation practices by restricting recruitment of meditators to those who practice only a focused attention form of meditation and who have only practiced meditation for a specified number of years, for example.

Another common limitation of meditation research, including the current study, is the use of a cross-sectional study design, and the associated limitation in our ability to attribute causation in the differences associated with the meditator group. This limitation could be addressed by employing a controlled long-term prospective study. However, conducting such a study would be difficult given the effects observed in the present study are associated with an average of 6.23 years of practicing meditation. Nonetheless, structural and functional brain differences have been observed after mindfulness interventions ranging from 3 days to 8 weeks (Taren et al., 2017; Tomasino & Fabbro, 2016; Xue et al., 2011).

In addition to the cross-sectional nature of our study, sample selection factors might have affected our results. In particular, many participants of the broader study provided task-related EEG recordings but elected not to receive TMS. As such, our results may be affected by a self-selection bias. Similarly, although the two groups did not significantly differ in age, the meditator group’s mean age was older. This factor may have increased within group variability of the N100 amplitudes, and it is possible it may have influenced the null result for differences in N100 amplitudes. Future research could address this by directly age matching participants from each group.

Additionally, the TMS resting motor thresholds and single-pulse TMS-EEG measures were obtained after participants had completed a range of cognitive tasks. While it is conceivable that the completion of these tasks may have affected the RMT and/or TMS-EEG measures, both meditators and non-meditators completed the same tasks prior to the TMS measures. As such, this factor would only confound our conclusions if the effects of having completed cognitive tasks prior to the TMS measures were specific to only either the meditator or the non-meditator group, which we have no reason to believe is the case.

Finally, our study had a limited sample size. However, the use of Bayesian analyses indicated that our sample combined with the observed effect sizes provides moderate confidence in our conclusions for many of the null results, and strong confidence in our conclusions for the two positive results. However, future research with a larger sample size will provide more confidence and increase the potential generalisability of results.