Modulation of evoked alpha oscillatory activity in the frontocentral region during analgesia by non-invasive brain stimulation

Abstract

Background

Repetitive transcranial magnetic stimulation (rTMS) to the posterior superior insula (PSI) can produce analgesia, although the underlying mechanisms require further exploration. While alpha oscillations have been implicated in rTMS-induced plasticity, research has largely focused on resting-state metrics, whereas alterations in perturbation-evoked alpha activity remain unknown.

Objective

To examine the effect of PSI rTMS on perturbation-evoked alpha activity within a capsaicin-induced tonic pain model.

Methods

Twenty healthy adults (10 females) completed two randomised, counterbalanced rTMS sessions (active, sham). Capsaicin was applied to the forearm for 90 min and pain intensity recorded at 5 min intervals on a numerical rating scale. Left PSI rTMS (10 Hz, 100 pulses/train, 15 trains) was delivered ∼50 min after capsaicin application. Evoked electroencephalographic responses to primary motor cortex (M1) TMS were assessed at baseline, during capsaicin-induced pain pre-rTMS, and post-rTMS. The frontocentral event-related spectral perturbation (ERSP) and parietal-occipital inter-trial coherence (ITC) were extracted.

Results

Baseline-normalised frontocentral ERSP showed a more positive pre-to-post change in M1-TMS-evoked alpha power following active rTMS relative to sham (p = 0.013), though no significant effect on ITC was observed. Exploratory analysis showed that lower pre-rTMS ITC predicted a greater reduction in pain with active rTMS (r = 0.68, p < 0.001).

Conclusions

PSI-rTMS modulates evoked alpha oscillatory power in frontocentral regions, advancing our understanding of the mechanisms underlying rTMS-induced analgesia. Evoked frontocentral alpha power may index a change in pain state, whereas exploratory findings suggest that alpha-band parietal-occipital ITC may predict individual differences in the rTMS analgesic response, though larger studies are needed to confirm these findings.

Introduction

High-frequency repetitive transcranial magnetic stimulation (rTMS) is a promising treatment for chronic pain [1]. While the primary motor cortex (M1) is a common target, converging evidence suggests the analgesic effects of M1-rTMS are mediated by alterations in distributed pain networks [[2], [3], [4]]. Notably, the posterior superior insular (PSI) cortex is functionally linked to motor and premotor regions, suggesting M1 connectivity states may covary with parieto-opercular/insular activity [5,6]. Recent studies directly targeting PSI with rTMS report reduced pain and heat sensitivity in both healthy volunteers and people with chronic pain [[7], [8], [9], [10], [11]], motivating further work into the mechanisms underlying the analgesic effects of PSI rTMS.

Combined TMS–electroencephalography (EEG) enables characterisation of rTMS-induced changes across distributed cortical networks during analgesia. Several studies have focused on TMS-evoked potentials (TEPs) such as the N45 and N100, which index GABAergic inhibitory activity [[12], [13], [14], [15]]. In a recent capsaicin-induced tonic pain study, PSI-rTMS was delivered ∼50 min after capsaicin application and EEG responses to M1-TMS were recorded at baseline, during pain (pre-rTMS) and post-PSI-rTMS. PSI-rTMS was shown to reduce the frontocentral N45 peak, with this reduction correlated with analgesia [16]. This not only suggests that PSI-rTMS may recruit GABAergic inhibitory processes, but also raises the possibility that alterations in pain-relevant networks extend beyond the stimulation site and may be detectable with TMS-EEG to M1. In this sense, M1-TMS-EEG may provide a useful, albeit indirect, probe of distributed cortical state change.

TMS-evoked alpha-band (∼8–12 Hz) oscillatory activity may offer further insight into mechanisms underlying PSI-rTMS analgesia. Alpha activity supports thalamo–cortical gating and top-down control of sensory input [17]. Resting-state EEG studies have linked lower alpha power or slower peak alpha frequency to higher experimental [[18], [19], [20], [21], [22], [23], [24]] and chronic pain [[25], [26], [27], [28], [29], [30]], and neuromodulation shown to increase alpha power or peak alpha frequency and reduce pain [24,[31], [32], [33]]. However, resting-state alpha metrics may miss key aspects of the neuro-modulatory effects of rTMS. Perturbation-based metrics assessed with TMS-EEG, such as evoked event-related spectral perturbations (ERSP: induced power changes) and inter-trial coherence (ITC: phase consistency) can quantify alpha reactivity and phase resetting directly, and serve as mechanistic markers and predictors of response to rTMS [34,35]. In a recent study, heat pain reduced alpha-band ERSP at frontocentral electrodes and reduced alpha ITC at parietal–occipital electrodes following M1-TMS [34]. Given these widespread effects, one possible interpretation is that pain is associated with altered alpha-related gating or coordination across sensorimotor/allostatic and posterior sensory-attentional networks. This interpretation remains indirect in the present context, but is consistent with the idea that M1-TMS-EEG may capture state-dependent changes beyond the stimulation site [34]. What remains unknown is whether rTMS-induced analgesia is associated with an increase in alpha ERSP and ITC. While prior work has shown distributed changes in beta and gamma ERSP (probed with M1-EEG) after 10 Hz M1-rTMS [14,36], these studies did not examine alterations in alpha ERSP/ITC during a pain state.

This gap justifies further exploration into how PSI-rTMS influences alpha-band ERSP and ITC, as probed with M1 TMS-EEG during ongoing pain. If PSI-rTMS analgesia is associated with changes in network-level pain processing, one possible manifestation could be a relative increase in alpha-band ERSP and ITC in the frontocentral and parietal-occipital regions previously shown to be altered by thermal pain [34]. Accordingly, the aim of this study was to determine the effect of PSI-rTMS on alpha-band ERSP and ITC, using pre-specified regions of interest (frontocentral ERSP, parietal–occipital ITC) assessed previously. We hypothesised that, during capsaicin-induced pain, PSI-targeted rTMS would result in a more positive pre to post change in frontocentral alpha ERSP and parietal-occipital alpha ITC relative to sham stimulation.

Methods

Participants

In accordance with the Declaration of Helsinki, the protocol received approval from the local ethics committee (Videnskabsetiske Komite for Region Nordjylland: N-20210047). The present secondary analysis uses original data from a study in which effects of PSI-rTMS on cortical excitability, quantified from peaks in TMS-evoked potentials (TEPs) were reported [16]. Twenty healthy adults (10 females; mean age 26.5 ± 4.6 years [mean ± SD]) were recruited via online advertisements. Individuals were excluded if they had current pain, any history or presence of chronic pain, neurological, musculoskeletal, psychiatric, or other major medical disorders, were pregnant and/or lactating, or met any contraindication for TMS (e.g., metal implants in the head), as determined with the Transcranial Magnetic Stimulation Adult Safety Screen questionnaire [37]. Sample size calculations are detailed in the original work [16] based on effect sizes reported in prior studies examining the impact of rTMS on TEP measures and pain outcomes [10,38].

Experimental protocol

This study used a randomized, sham-controlled, cross-over design. Each participant completed two sessions, separated by approximately 2–3 weeks, receiving active or sham PSI-rTMS in a counterbalanced order. In each session, tonic pain was induced using topical capsaicin applied to the right volar forearm for 90 min (Fig. 1). Left PSI-rTMS was administered ∼50 min after capsaicin application and stimulation lasted 7.5 min. Assessment of TEPs using combined TMS-EEG targeting M1, and of thermal pain sensitivity, were collected at 3 timepoints: baseline, during capsaicin-induced pain but before rTMS (pain pre-rTMS: ∼30–45 min after capsaicin application), and after rTMS (pain post-rTMS: ∼65–80 min after capsaicin application). Participants were told that both sessions were identical except for the type of repetitive stimulation (active vs sham). After completing the second session, they were asked to indicate which session they believed contained active stimulation and which was sham.

Fig. 1.

Schematic of the experimental protocol, adapted from the parent study [16], and regions of interest.

Combined transcranial magnetic stimulation and electroencephalography

Participants were seated comfortably and asked to fixate on a cross on the wall in front of them. Single biphasic TMS pulses were delivered using a magnetic stimulator (Magstim Ltd., Whitland, United Kingdom) and a 70 mm figure-of-eight flat coil. EEG was simultaneously recorded using a TMS-compatible amplifier (g.HIamp EEG amplifier, g.tec medical engineering GmbH, Schiedlberg, Austria) at a sampling rate of 4800 Hz. Signals were acquired from 63 passive electrodes mounted in an elastic cap (EASYCAP GmbH, Etterschlag, Germany) according to the 10–5 system. Data were referenced online to the right mastoid, with ground on the right cheekbone, to minimise artifacts from TMS applied over the left hemisphere. Both abrasive and conductive electrode gel were used to keep impedances below ∼5 kΩ. To maintain low impedances for the duration of testing, 2 net caps (GVB-geliMED GmbH, Ginsterweg Bad Segeberg, Germany) were placed over the EEG cap, which was further stabilised with a plastic stretch wrap handle film. To reduce auditory responses to the TMS coil click, an auditory masking toolbox [39] was used and participants used noise-cancelling headphones (Shure SE215-CL-E Sound Isolating, Shure Incorporated, IL, United States).

Neuronavigation (Brainsight TMS Neuronavigation, Rogue Research Inc., Montréal, Canada) with a template MRI (MNI ICBM 152 average brain) was used to register each participant's head and track TMS coil position in 3D. Surface Ag/AgCl electrodes (Ambu Neuroline 720, Ballerup, Denmark) were placed over the right first dorsal interosseous (FDI) muscle, aligned with the muscle fibers, with the ground electrode positioned over the right ulnar styloid. The coil was held at 45° to the midline to induce a posterior–anterior current. The left M1 “hotspot”, defined as the scalp site that produced the largest motor evoked potential (MEP) in the FDI muscle, was located and marked. Resting motor threshold (RMT) was then established using maximum-likelihood parametric estimation by sequential testing. This adaptive algorithm iteratively estimates the relationship between MEP amplitude and stimulator output and converges on the intensity predicted to elicit an MEP of 50 μV in 50% of trials [40]. This approach provides accuracy comparable to classical methods (e.g., Rossini–Rothwell) [41] while requiring fewer pulses [42]. The test stimulus intensity for TEP acquisition was set to 90% RMT to limit reafferent contamination of EEG from muscle activation [13].

A real-time TEP visualization tool [43] was used to confirm that artifacts (e.g., muscle, auditory) were minimal and that, given the coil orientation and 90% RMT stimulation, early site-specific peaks (<100 ms; P30–N100) were clearly present [[43], [44], [45]]. Coil location and data quality were continuously monitored using neuronavigation and the visualization tool across all timepoints and both sessions. For each TEP block (baseline, pain pre-rTMS, pain post-rTMS), ∼150 TMS pulses (∼7 min total) were delivered with a jittered interstimulus interval of 2.6–3.4 s [43,45].

Thermal pain sensitivity assessment

Cold and heat pain thresholds were recorded at each timepoint (immediately after each TEP block) following prior procedures [13]. A 30 × 30 mm thermode (Medoc Pathway ATS device; Medoc Advanced Medical Systems Ltd, NC) was placed on the right thenar eminence. Starting from a baseline temperature of 32 °C (neutral skin temperature), participants completed 2 threshold tests in fixed order: [1] cold pain threshold (CPT), defined as the moment a decreasing temperature first became painful, and [2] heat pain threshold (HPT), defined as the moment an increasing temperature first became painful. Three trials of each threshold were collected and averaged, with a 6-s interstimulus interval. Participants indicated threshold during each trial via a handheld response button (left hand) connected to the thermal stimulator. Temperature increased or decreased at 1 °C/s and returned to baseline at 2 °C/s after the button press.

Capsaicin-induced tonic pain

After completion of the baseline TEP and thermal sensitivity measures, an 8% topical capsaicin patch (Transdermal patch “Qutenza,” Astellas, 4 × 4 cm) was applied to the volar right forearm (5 cm proximal to the wrist) to evoke ongoing cutaneous pain. Pain intensity was then recorded every 5 min using a numerical rating scale (NRS) from 0 (“no pain”) to 10 (“worst imaginable pain”).

Repetitive transcranial magnetic stimulation

Active or sham rTMS was delivered over the orthogonal projection of the PSI using a double cone coil (D110, Magstim Ltd, UK). Stimulation site and intensity were determined between capsaicin application and the first TEP block (i.e., within the first ∼30 min after capsaicin). Stimulation intensity was defined by administering single TMS pulses with the double cone coil over the left motor hotspot of the tibialis anterior (TA) muscle, chosen because TA lies at approximately the same cortical depth as the PSI [7]. The TA hotspot and resting motor threshold (RMT) were identified by visually observing leg muscle responses to stimulation, and RMT was then established using the same maximum-likelihood algorithm described above [40].

The fast PSI method was used to locate the PSI projection without requiring MRI-guided neuronavigation, substantially reducing targeting time [46]. rTMS was then delivered using a protocol recommended for analgesic effects: 1500 pulses at 10 Hz, delivered as 15 trains of 10 s each with 20-s intertrain intervals (total 7.5 min) [47]. Stimulation intensity was set at 80% of the TA RMT and the coil was oriented so that the main phase of the biphasic pulse induced a posterior–anterior current [48]. To implement blinding, in both active and sham conditions we mounted a D70 figure-of-eight coil (Magstim Ltd, UK) orthogonally against the double cone coil using an adjustable mechanical arm. During active stimulation the double cone coil was triggered and during sham stimulation the figure-of-eight coil was triggered instead [9]. Prior work indicates that this active–sham configuration effectively maintains blinding even in cross-over designs [49,50]. Neuronavigation (Brainsight) was used to keep coil placement consistent throughout rTMS delivery.

Data processing

TEP pre-processing was carried out in MATLAB (R2021b, The MathWorks, MA) using EEGLAB [51] and TESA [52], following previously published procedures [13,[52], [53], [54]]. Continuous data were epoched from −1000 ms to +1000 ms around each TMS pulse and baseline-corrected using the −1000 ms to −5 ms pre-stimulus interval. Channels showing large TMS decay artifacts were removed. The −5 ms to +12 ms window around the TMS pulse was excised and then interpolated using a cubic fit. Noisy epochs were automatically flagged using EEGLAB's auto-trial rejection routine [55] and then manually verified. Eyeblink and muscle artifacts were removed using fastICA with autocomponent rejection [52]. We then applied the source-estimation noise-discarding (SOUND) algorithm [53], which models the most likely cortical current distribution from the multichannel data and suppresses channel-specific noise accordingly. After denoising, the data were re-referenced to the average reference. A 1–100 Hz band-pass and 48–52 Hz band-stop Butterworth filter were applied. Any channels removed earlier were then interpolated back in.

Time–frequency analysis

A priori, hypothesis-driven regions of interest aligned with De Martino et al. (2024) [34] were used: FCz, Cz, FC1, FC2, C1, C2 for ERSP (frontocentral) and POz, Oz, PO3, PO4, PO7, O1, PO8, O2 for ITC (parietal-occipital). Time-frequency decomposition was performed using newtimef, an EEGLAB function that applies wavelet-based time-frequency analysis to estimate both event-related spectral perturbation (ERSP) and inter-trial coherence (ITC).

For decomposition, we used complex Morlet wavelets with 3.5 cycles across 8–45 Hz sampled at 100 frequencies. These settings were chosen to provide sufficient resolution across the alpha range while still capturing broader post-stimulus oscillatory responses. Baseline correction in newtimef used the −1000 to −5 ms pre-stimulus interval, providing a stable reference period immediately preceding the TMS pulse and TMS artifact contamination from the pulse itself. The 15–300 ms post-stimulus window was selected a priori because the same window, together with the same alpha band and electrode clusters, was previously shown to capture pain-related reductions in perturbation-evoked alpha ERSP and ITC following M1-TMS in a similar experimental pain model [34]. Retaining these parameters in the present study moderated analytic flexibility and allowed a direct, hypothesis-driven test of whether PSI-rTMS modulated the same perturbation-evoked alpha features. Accordingly, all inferential analyses were conducted on ROI-averaged alpha-band values extracted from these pre-specified frontocentral and parietal-occipital clusters.

ERSP quantifies stimulus-locked changes in oscillatory power relative to a pre-stimulus baseline. It was calculated as power relative to the pre-stimulus baseline at each time-frequency bin, and trial-level ERSP maps were then averaged within each participant.

ITC quantifies the consistency of oscillatory phase alignment across trials at each time-frequency bin. It was calculated from the phase of the wavelet transform as the magnitude of the mean unit-norm complex vector across trials, such that higher ITC values indicated greater trial-to-trial phase consistency.

Following De Martino et al. (2024) [34], ERSP and ITC were summarised in two complementary ways: first, as frequency-wise curves (8–45 Hz) averaged over the 15–300 ms post-stimulus interval; and second, as time courses in the alpha band (8–12 Hz) averaged across frequency.

Statistical analysis

Statistical analyses were performed using JASP 0.16.4.0. The primary dependent variable was the baseline-normalised ERSP and ITC at the pre-rTMS and post-rTMS timepoints quantified as the difference from that session's baseline, therefore controlling for potential between-session differences. The primary analysis for both ΔERSP and ΔITC was the Time × Stimulation interaction in a 2 × 2 within-subjects design (pre-rTMS and post-rTMS; active and sham rTMS).

Planned Analysis. Modulation of alpha ERSP and ITC was tested using a 2 (Time: pain pre-rTMS vs pain post-rTMS) × 2 (Session: active vs sham) repeated-measures ANOVA (α = 0.05, two-sided). Bayes factors were computed for each analysis with default priors; a BF₁₀> 3 was taken as moderate evidence for an interaction. For interpretability, we quantified the interaction with a difference-in-differences (DiD) contrast: ΔActive = (pain post-rTMS − pain pre-rTMS in the active session) and ΔSham = (pain post-rTMS − pain pre-rTMS in the sham session), and tested DiD = ΔActive − ΔSham against zero using a one-sample t-test, reporting Cohen's dz. Robustness checks were performed using log-transformed ERSP/ITC values and Wilcoxon signed-rank tests. Where a significant Time × Stimulation interaction was observed, follow-up simple-effects analyses were conducted to decompose the interaction. These analyses did not re-test the difference-in-differences contrast; rather, they examined the within-session pre-to-post change (post − pre) separately for the active and sham sessions using paired t-tests, alongside Wilcoxon signed-rank tests. We conducted additional analyses, reported in the Supplementary Material, to assess the robustness of the data, including pre-TMS-pulse ERSP/ITC differences, pre-rTMS ERSP/ITC differences between active and sham sessions, and overall post-pulse changes in ERSP/ITC.

Exploratory Analysis. First, to characterise physiological coupling between ITC and ERSP, the correlation between ITC and ERSP values was calculated at baseline.

Second, it was determined whether ERSP/ITC at the pain-pre-rTMS could serve as a predictor of the analgesic response to rTMS. Indeed, in a recent study [11], we found that ERSP and ITC values prior to the rTMS intervention could predict treatment response: patients with chronic pain receiving rTMS were more likely to respond (>30% reduction in pain intensity) if they had lower ERSP and ITC values prior to rTMS. To determine whether the same relationship occurred within the current experimental pain model, the correlation between the ERSP and ITC values was extracted at the pain, pre-rTMS timepoint with the change in capsaicin pain NRS ratings (at 90 min timepoint minus the 45 min timepoint) and change in heat pain thresholds (pain, post-rTMS – pain, pre-rTMS).

Third, the change in ERSP and ITC were correlated with the change in pain NRS ratings (90-min timepoint minus 50-min timepoint) and heat pain thresholds. For all correlations, the Pearson product–moment correlation (r) was reported with a two-sided p-value. All correlation analyses were repeated separately within the active and sham subsets and on the pooled data to make the contribution of session explicit. Statistical significance was set at p < 0.05.

Results

As reported previously [16], all participants completed the experiment. In the parent study, pain NRS decreased from 50 to 90 min after active PSI rTMS (BF₁₀= 14.30, d = −0.76), with only anecdotal evidence for change after sham (BF₁₀= 1.70, d = 0.50). Heat pain threshold change was larger after active stimulation (pain pre to pain post: BF₁₀= 145,080.16, d = 1.8), whereas sham showed only anecdotal evidence of a small decrease (BF₁₀= 1.60, d = −0.40).

End-of-study blinding assessment showed that 8/20 were able to correctly identify the sequence of the active and sham sessions, suggesting blinding was successful. Moreover, 4 of these 8 participants had active rTMS in the first session, while the other 4 had sham rTMS in the first session, suggesting the order of the sessions did not influence accuracy.

Frontocentral event-related spectral perturbation

Fig. 2 shows the time frequency maps across timepoints (baseline, pain pre-rTMS and pain post-rTMS) and sessions (active and sham). Fig. 3 shows frontocentral ERSP as a function of frequency (8–45 Hz), averaged over the 15–300 ms post-stimulus window, separately for each experimental timepoint and for the active and sham sessions. Note robustness analyses examining pre-TMS alpha power and baseline-normalised pain-pre-rTMS ERSP differences between active and sham sessions were non-significant (all p > 0.05; see Supplementary Material). A two-way repeated-measures ANOVA of alpha-ERSP revealed a significant Time × Stimulation interaction (F [1,19] = 7.49, p= 0.013, partial η²= 0.283), with the Bayesian analysis also supporting the interaction (BF₁₀= 9.94). Quantifying this interaction as a difference-in-differences contrast (ΔActive − ΔSham) yielded a significantly more positive value following active than sham stimulation (mean difference = 0.484 ± 0.791 dB, t [19] = 2.74, p= 0.013), with convergent results for log-transformed alpha-ERSPs (p= 0.013) and the Wilcoxon signed-rank test (p= 0.017). Simple effects analysis indicated a trend toward increased alpha-ERSP in the active session (mean change = 0.253 dB, 95% CI −0.019 to 0.526, t [19] = 1.94, p= 0.067; log-transformed p= 0.067; Wilcoxon p = 0.076) and a corresponding trend toward decreased alpha-ERSP in the sham session (mean change = −0.231 dB, 95% CI −0.470 to 0.008, t [19] = −2.02, p= 0.057; log-transformed p= 0.058; Wilcoxon p= 0.048).

Fig. 2.

Time frequency maps for frontocentral region of interest across timepoints (baseline, pain pre-rTMS and pain post-rTMS) and sessions (active and sham). rTMS: repetitive transcranial magnetic stimulation.

Fig. 3.

Frontocentral event-related spectral perturbation (ERSP) as a function of frequency (8–45 Hz), averaged over the 15-300 ms post-stimulus window, shown separately for each experimental timepoint (baseline, pain pre-rTMS, and pain post-rTMS) and for active and sham posterior superior insula repetitive transcranial magnetic stimulation (PSI-rTMS). Shaded areas represent the 95% confidence intervals.

Parietal-occipital inter-trial coherence

Fig. 4 shows intertrial coherence maps across timepoints (baseline, pain pre-rTMS and pain post-rTMS) and sessions (active and sham). Fig. 5 shows the ITC values for each timepoint and session as a function of frequency (8–45 Hz). A two-way repeated-measures ANOVA of alpha-ITC revealed a non-significant Time × Stimulation interaction, F [1,19] = 2.08, p= 0.165, partial η²= 0.02, with the Bayesian repeated-measures ANOVA also suggesting inconclusive evidence (BF₁₀= 0.91). Quantifying this interaction as a difference-in-differences contrast (ΔActive − ΔSham) yielded a positive value, indicating a larger increase in ITC following active than sham stimulation, though this did not reach significance (mean difference = 0.028 ± 0.086, t [19] = 1.44, p = 0.165) with similar results for log-transformed ITCs (p= 0.165) and the Wilcoxon signed-rank test (p= 0.202). Fig. 6 summarises the change in alpha-ERSP and alpha-ITC values across timepoints for active and sham sessions.

Fig. 4.

Time coherence maps for parietal-occipital region of interest across timepoints (baseline, pain pre-rTMS and pain post-rTMS) and sessions (active and sham). rTMS: repetitive transcranial magnetic stimulation.

Fig. 5.

Parietal-occipital inter-trial coherence (ITC) as a function of frequency (8–45 Hz), averaged over the 15–300 ms post-stimulus window, shown separately for each experimental timepoint (baseline, pain pre-rTMS, and pain post-rTMS) and for active and sham posterior superior insula repetitive transcranial magnetic stimulation (PSI-rTMS). Shaded areas represent the 95% confidence intervals.

Fig. 6.

Frontocentral event-related spectral perturbation (ERSP) and parietal ITC values (baseline normalised) at the pain, pre-rTMS and pain, post-rTMS timepoints for active and sham sessions. Error bars represent the 95% confidence interval. Asterisks indicate a significant interaction between time and condition.

Exploratory analyses

Exploratory correlations are shown in Supplementary Table 1 with selected associations shown in Fig. 7. Pooling sessions (n = 40), the correlation between alpha parietal-occipital ITC and frontocentral ERSP (Fig. 7A) was positive (r = 0.64, p < 0.001). Within the active session (n = 20), this association remained significant (r = 0.46, p = 0.041), as it did in sham (n = 20, r = 0.84, p < 0.001).

Fig. 7.

Correlations between alpha-band EEG measures and pain outcomes (overall and by session). (A) Association between baseline alpha event-related spectral perturbation (ERSP) and inter-trial coherence (ITC). (B) Association between pain, pre-rTMS alpha ITC (baseline-normalised) and change in pain numerical rating scale (NRS) from 50 to 90 min post-capsaicin. (C) Association between pain, pre-rTMS alpha ITC (baseline-normalised) and change in heat pain threshold (HPT) from pain pre-rTMS to pain post-rTMS. (D) Association between change in alpha ITC from pain pre-rTMS to pain post-rTMS and change in pain NRS from 50 to 90 min post-capsaicin. (E) Association between change in alpha ITC from pain pre-rTMS to pain post-rTMS and change in HPT from pain pre-rTMS to pain post-rTMS.

Lower pre-rTMS alpha ITC (baseline-normalised) was associated with greater post-rTMS pain NRS reduction (Fig. 7B) and larger HPT (Fig. 7C) increase when pooling sessions (n = 40; pain NRS: r = 0.44, p = 0.004, HPT: r = −0.42, p = 0.007). The association with pain was numerically stronger in the active than sham session (active: r = 0.68; sham: r = 0.32; Fig. 7B), whereas the association with HPT was numerically stronger in sham than active (sham: r = −0.59; active: r = 0.07; Fig. 7C).

Across sessions (n = 40), greater increases in ITC were associated with greater pain reduction (r = −0.31, p = 0.049, Fig. 7D) and larger increases in HPT (r = 0.45, p = 0.004, Fig. 7E). Session-specific analyses showed no association between ITC change and pain change in either active (r = −0.05, p = 0.84) or sham (r = −0.33, p = 0.16) sessions. For HPT, ITC change was positively associated with HPT change in the sham session (r = 0.54, p = 0.013), whereas no association was present in the active session (r = −0.02, p = 0.92).

Pain pre-rTMS alpha ERSP (baseline-normalised) and change in ERSP from pre to post rTMS were not significantly associated with post-rTMS change in pain NRS or HPT when pooling sessions (all p > 0.05).

Discussion

The present study aimed to determine the effect of PSI-rTMS on perturbation-evoked alpha brain activity during capsaicin-induced tonic pain in healthy participants. PSI-rTMS resulted in a more positive pre-to-post change in frontocentral alpha ERSP in the active compared with sham rTMS condition, whereas parietal-occipital alpha ITC did not change significantly at the group level. Exploratory analyses indicated that neither pre-rTMS ERSP during pain nor ERSP change from pre-rTMS to post-rTMS predicted the analgesic response. By contrast, alpha-band ITC (both pre-rTMS and change to post-rTMS) predicted changes in pain intensity. In the active rTMS condition, lower pre-rTMS ITC during experimental pain was associated with greater subsequent pain reduction by the rTMS. In the sham rTMS condition, where pain worsened over time in the absence of active stimulation, lower pre-rTMS ITC normalised to baseline was associated with larger decreases in HPT (greater sensitisation), and greater ITC reductions from pre-rTMS to post-TMS, predicting larger decreases in HPT. Taken together, our findings provide insight into the mechanisms underlying PSI-rTMS-induced analgesia.

Modulation of evoked alpha oscillatory power following brain stimulation

The present analysis suggests PSI-targeted rTMS delivered during tonic capsaicin pain results in analgesia and a more positive pre to post change (medium to large effect size) in frontocentral TMS-evoked alpha-band oscillatory power than sham. Within the context of the primary study [16], which focused primarily on TEP peak amplitudes, these findings extend the impact of PSI rTMS into the TMS-evoked oscillatory (alpha) domain. The specificity of this effect to the TMS-evoked response is further supported by the absence of differences in alpha power during the pre-TMS period across conditions, and the lack of ERSP differences between active and sham sessions at the pain-pre-TMS timepoint.

Previous TMS-EEG work using a tonic heat pain model reported reduced frontocentral alpha ERSP during pain compared with baseline, suggesting that tonic pain can suppress alpha reactivity in these regions [17]. In the present study, 10 Hz PSI-rTMS resulted in a more positive pre-to-post change in frontocentral alpha ERSP during pain relative to sham. As the within-session increase in ERSP in the active condition and the decrease in the sham condition did not individually reach significance, we cannot conclude whether rTMS causes a robust increase in ERSP, a relative preservation of ERSP compared with sham, or both. Moreover, no significant reduction in frontocentral alpha ERSP was observed from baseline to the pre-rTMS pain timepoint in this dataset, possibly because we did not sample TMS-EEG outcomes at the time of peak capsaicin pain. As a result, we cannot conclude that PSI-rTMS “restored” a pain-suppressed alpha signal to a normal baseline state. Instead, the present data more cautiously indicate that perturbation-based measures of alpha reactivity are modifiable by PSI-rTMS during ongoing pain and may track aspects of its analgesic effects [34].

A distinctive aspect of this work is the rTMS site, TMS-EEG site and ROI configuration. Repetitive TMS was applied to a PSI target, TMS-EEG was used to probe M1, and readouts were obtained over frontocentral and parietal–occipital electrodes. The presence of an ERSP effect in this configuration raises the possibility that M1's broad extra-motor connectivity [[2], [3], [4]] may allow it to serve as a practical indirect readout of distributed cortical state, such that informative rTMS-evoked measures can be obtained without stimulating and probing the same cortical site. The observed ERSP interaction, measured over the frontocentral region during PSI stimulation, may be interpreted from a network-based perspective of rTMS-induced analgesia, in which rTMS could influence distributed thalamocortical or corticocortical dynamics [56]. As entrainment of ongoing oscillations has been proposed as a key mechanism of rTMS [31], one possible interpretation of the present findings is that PSI-targeted rTMS may alter synchrony within a distributed alpha-generating network [[57], [58], [59]], expressed over frontocentral electrodes likely reflecting distributed cortical activity relevant to pain processing [13,15]. It may therefore be hypothesised that insular stimulation shifts frontal cortical engagement away from nociceptive-related processing involving parietal and deeper insular regions [60,61].

One methodological question arising from this study concerns the extent to which the posterior superior insula (PSI) was specifically targeted, particularly in the absence of individualized MRI-guided neuronavigation. In the present protocol, stimulation intensity was based on activation of the tibialis anterior representation, which has been proposed to approximate the cortical depth of the insula [7]. Prior work using this general approach has reported analgesic effects consistent with a role of the PSI in thermal pain processing [62], and has shown effects when deep but not superficial stimulation was applied [10], and has demonstrated current reaching the insular region in postmortem models [63]. In addition, the Fast-PSI method has been reported to show good consistency with neuronavigation-informed targets [46]. Nonetheless given the inter-individual variability in insular anatomy, cortical depth, and the proximity of adjacent opercular and somatosensory regions, it remains less certain whether stimulation consistently engaged the PSI specifically in every participant. The present findings should therefore be interpreted as arising from stimulation of a PSI-proximal/insular-opercular target rather than as definitive evidence of selective PSI modulation.

Preliminary evidence of an individual association between intertrial coherence and pain

Despite robust modulation of frontocentral ERSP, we did not observe group-level changes in parietal–occipital ITC following PSI rTMS. This suggests that PSI-rTMS acts more strongly on M1-related amplitude than on the phase consistency of alpha oscillations during ongoing pain. However, the sample size adopted from a parent trial powered for TEP peaks and pain outcomes may have been underpowered to detect ITC effects, especially for subtle phase-based changes. An alternative explanation is that ITC is more closely related to individual differences in pain intensity or sensitivity than to average group changes. Indeed, the exploratory analyses may help clarify the distinct roles of TMS-evoked power and phase-consistency metrics; while alpha ERSP was modulated by active rTMS, neither baseline ERSP nor ERSP change showed meaningful associations with pain intensity or pain threshold changes. This suggests limited value for ERSP as an individual-level marker and potentially a role as a general indicator of pain-related or network-state change, of which analgesia is one downstream consequence. In contrast, ITC behaved more like an individual-difference marker. First, lower pre-rTMS ITC was associated with a greater subsequent analgesic response, consistent with our recent study [11] in chronic pain patients, where lower baseline pre-rTMS ITC predicted a higher likelihood of being a responder. Second, changes in ITC were correlated to changes in pain and pain thresholds, particularly in the sham condition, where greater decreases in ITC were linked to greater sensitisation. This pattern aligns with prior work suggesting that parietal-occipital phase-based metrics measured during M1 TMS-EEG reflect individual pain sensitivity [34] and are reduced in fibromyalgia, where lower values correlate with symptom severity [11]. Nevertheless, these associations should be interpreted with caution, given the exploratory nature of our analyses.

Set against the broader literature on resting-state EEG markers of pain [18,[24], [25], [26], [27], [28], [29], [30], [31], [32], [33],[64], [65], [66]], our present findings have implications for the alpha-pain-biomarker framework. First, TMS-EEG perturbation-based measures are a useful complement to resting-state alpha EEG markers in understanding rTMS mechanisms, as evoked oscillatory dynamics can reveal pain-relevant information that is not apparent in static spectral measures. Second, exploratory analysis suggests that rather than a unitary “alpha marker” of pain severity, different alpha-based metrics might capture distinct facets of pain processing, e.g., evoked alpha power (ERSP) may track a general pain intensity state and its potential modulation by rTMS, whereas alpha phase consistency (i.e. ITC) is more tightly linked to individual differences in pain vulnerability and treatment response. Within a translational pipeline, such perturbation-based alpha markers can be refined and further validated in standardised healthy pain models designed for purpose, then subsequently evaluated for prognostic and treatment-response utility in clinical cohorts [67].

Strengths and limitations

A key strength of the present study was its hypothesis-driven nature. The same time window post-TMS (15–300 ms), same frequency band (8–12 Hz), and same frontocentral and parietal-occipital electrode clusters were targeted as previously shown to exhibit pain-related ERSP/ITC reductions [34], therefore significantly reducing bias in the selection of parameters. A further strength is the replication of the ERSP directional effect using a notably different pre-processing pipeline compared to De Martino et al. (2024) (e.g., the use of SOUND, altered order of filtering and ICA steps), suggesting that ERSP changes during pain and rTMS may be robust to reasonable variations in pre-processing. This is particularly important given that differences in pre-processing pipelines are often cited as a key source of replication and reproducibility failures in TMS-EEG research [68].

Several limitations should be acknowledged. First, this was a secondary analysis of a parent study powered for TEP peaks and behavioral pain outcomes rather than oscillatory ERSP/ITC metrics. Accordingly, both the significant ERSP finding and the non-significant ITC finding should be interpreted cautiously. Although the ERSP result was hypothesis-driven (based on previous literature) and supported by convergent robustness analyses, the observed effect size may have been inflated. Likewise, the absence of a significant ITC effect cannot be taken as strong evidence of no effect, as a smaller phase-based effect may have gone undetected. These findings therefore require replication in studies powered specifically for these oscillatory outcomes. Exploratory correlational findings should also be regarded as preliminary and hypothesis-generating. Second, while the mixed configuration of rTMS site (PSI), TMS-EEG probe site (M1), and distributed electrode readout conceptually supports “network-based” changes following rTMS, it also imposes interpretive constraints. Because TMS-evoked oscillations were not recorded directly from the stimulation site, it is not possible to determine whether the observed effects reflect local changes within the PSI or a more distributed network-level effect. Finally, the work examined the effects of a single rTMS session over a relatively brief time window within a short experimental pain model. The effects of multiple sessions on ERSP/ITC are still to be understood, and the relationship between changes in ERSP/ITC and pain over longer-term pain trajectories, as in other pain models which induce pain over multiple days [12]. As a result, while the present findings support evoked alpha dynamics as promising mechanistic markers, they do not establish clinical utility for guiding multi-session rTMS treatment in chronic pain populations.

Conclusion

PSI-rTMS delivered during tonic pain results in a more positive pre-to-post change in frontocentral alpha ERSP relative to sham, extending our understanding of the mechanisms of rTMS-induced analgesia from resting-state studies to perturbation-based TMS-EEG measures. Exploratory findings suggest ERSP might index a general change in pain-related network state, whereas ITC might be tied to individual pain vulnerability and treatment response. Together, these results position TMS-EEG perturbation-based measures as a promising next step in understanding rTMS mechanisms and in the development of alpha-based pain biomarkers.

Authorship contributions

N.S.C.: Conceptualization, data curation, formal analysis, investigation, methodology, software, validation, visualization, writing (original draft), and writing (reviewing and editing).

S.K.M.: Investigation, methodology, and writing (reviewing and editing).

E.D.M.: Methodology, investigation, resources, software, and writing (reviewing and editing).

D.B.L.: Conceptualization, project administration, and writing (reviewing and editing).

S.M.S.: Conceptualization and writing (reviewing and editing).

D.A.S.: Conceptualization and writing (reviewing and editing).

D.C.A.: Supervision, validation, visualization, writing (original draft), and writing (reviewing and editing).

T.G.N.: Funding acquisition, supervision, validation, visualization, writing (original draft), and writing (reviewing and editing).

Disclosures

Center for Neuroplasticity and Pain (CNAP) is supported by the Danish National Research Foundation (DNRF121). TGN receives funding from the Lundbeck Foundation (R441-2023-232). The authors have no conflicts of interest to declare.

Declaration of competing interest

The authors declare that they have no competing interests.

Appendix A. Supplementary data

The following is the Supplementary data to this article: