Trait-dependent effects of theta burst stimulation after psychosocial stress: a sham-controlled study in healthy individuals

(cid:1) This study examined the effects of intermittent (iTBS), continuous (cTBS) or sham TBS (sTBS) of the left DLPFC before a stressor. (cid:1) cTBS compared to sTBS and iTBS resulted in an attenuated stress-induced cortisol response in high compared to low trait ruminators. (cid:1) There were rumination independent TBS effects on mood and heart rate, but not on other psychophysiological measures of stress.

Rumination has been functionally linked to various brain regions including the dorsolateral prefrontal cortex (DLPFC; Zhou et al., 2020).The DLPFC plays a crucial role in inhibiting physiological and psychosocial stress reactivity, potentially resulting in reduced rumination (Arnsten, 2009(Arnsten, , 2015)).Poor stress regulation (resulting in higher cortisol levels, heart rates, and lower HRV) is linked to increased limbic activity and decreased DLPFC functioning (Arnsten, 2009; although see Cooney et al., 2010;Henze et al., 2023).As a result, this may cause an inability to exert inhibitory control over negative perseverative processes, such as rumination, leading to sustained negative affect (De Raedt and Koster, 2010).Neuroimaging findings further support this, showing decreased activity of the left DLPFC in high, compared to low, ruminators (Ferdek et al., 2016;Rosenbaum et al., 2018a;Rosenbaum et al., 2021).Hence, investigating the modulation of brain regions involved in stress-reactive rumination and the psychophysiological stress response, such as the DLPFC, in healthy individuals can provide crucial information for conducting clinical trials with depressive patients.
Repetitive transcranial magnetic stimulation (rTMS) is a noninvasive approach to modulate brain activity by altering the excitability of targeted brain areas (Maeda et al., 2000).Besides its (adjunctive) use in depression treatment (Baeken et al., 2019;Sathappan et al., 2019), rTMS is a useful tool for directly manipulating prefrontal cortical excitability to study its effects on rumination and the psychophysiological stress response (e.g., Baeken et al., 2014;De Smet et al., 2021;Pulopulos et al., 2020;Remue et al., 2016).Theta burst stimulation (TBS), a patterned variant of rTMS, has gained popularity due to similar outcomes with reduced stimulation duration and intensity compared to conventional rTMS protocols (Blumberger et al., 2018;Huang et al., 2005).TBS involves three gamma frequency pulses repeated every 200 milliseconds, and its effects can induce long-term potentiation (LTP) or long-term depression (LTD) based on the delivery method and stimulation duration (Huang & Rothwell, 2007;McCalley et al., 2021;Suppa et al., 2016).The left DLPFC is often targeted in TBS protocols (Fitzgerald, 2021) and is considered the most suitable region for modulating psychophysiological stress reactivity (Moses et al., 2023;Vignaud et al., 2023).Therefore, applying TBS to the left DLPFC might be a promising approach to modulate stress-reactive rumination and psychophysiological activation.
Despite its scientific and clinical relevance, few studies have examined the effects of TBS on psychological and physiological changes following acute stress in healthy individuals, yielding mixed results.For instance, Era and colleagues (2021) reported that cTBS of the left DLPFC before a rumination induction led to a sustained physiological stress response, indicated by increased heart rate and cortisol levels, and reduced vagally-mediated HRV.However, no TBS-induced changes were found on behavioral and self-report measures.De Witte and colleagues (2020) reported no direct effects of two consecutive iTBS sessions (versus sham) after psychosocial stress on cortisol recovery.Nevertheless, higher levels of depressive brooding, a subform of rumination, were associated with a reduction in cortisol secretion during stress recovery.Although the data suggested a buffering effect of active iTBS on increases in state rumination in high trait ruminators (De Witte et al., 2020), the relatively small sample size, mainly consisting of low trait ruminators, might have limited the power to detect (rumination-dependent) TBS effects on stress-reactive rumination (De Witte et al., 2020).Moreover, a systematic review indicated that applying stimulation before exposure to a stressful event, rather than after, is more effective in decreasing psychophysiological stress responses (Vignaud et al., 2023).These findings underscore the complexity of TBS effects on psychophysiological stress measures, and indicate that individual differences in trait rumination might have a crucial role in TBS outcomes on stress measures.Notably, previous research has shown that individuals with higher ruminative tendencies have aberrant mechanisms of prefrontal control (e.g., De Raedt and Koster, 2010;Vanderhasselt et al., 2017), implying that prefrontal stimulation might be particularly beneficial for this population.Consequently, studies investigating the effects of TBS on stress-reactive rumination and the psychophysiological stress response should account for individual differences in rumination.This informs us about the working mechanisms of TBS on stress for individuals with varying ruminative tendencies and how to reduce variability in stimulation outcomes.
Building upon these prior findings (De Witte et al., 2020;Era et al. 2021), this well-powered randomized sham-controlled study aimed to evaluate the effects of a single session of TBS (iTBS versus cTBS versus sham TBS) on stress-reactive state rumination and the psychophysiological stress response in healthy individuals.To ensure a comprehensive representation of the population, we recruited a stratified sample based on scores on the Ruminative Response Scale (RRS), which included individuals with varying levels of rumination tendencies.Focusing on the left DLPFC as the target region for stimulation due to its crucial role in stress regulation and rumination, we hypothesized that iTBS of the left DLPFC would would lead to reduced stress-reactive rumination and psychophysiological stress response (lower negative affect and higher positive affect, lower subjective stress, lower mean heart rate and cortisol levels, and higher heart rate variability) compared to cTBS and sham.Conversely, we hypothesized that cTBS would result in heightened stress-reactive rumination and stress response (higher negative affect and lower positive affect, higher subjective stress, higher mean heart rate and cortisol levels, and lower heart rate variability) compared to iTBS and sham.Additionally, we expected the effects of iTBS and cTBS to be more pronounced in individuals with higher levels of trait rumination.By addressing the role of individual trait rumination in TBS outcomes, this study seeks to contribute valuable insights to the existing literature, enhancing the understanding of prefrontal stimulation's potential for stress management and therapeutic applications.

Methods
This study was part of a multicenter research trial in healthy participants conducted at the University Hospital of Ghent (UZGent) and the University Hospital of Tübingen (UKT).All procedures were approved by the local ethics committees (UZGent: B67Q2021000115, UKT: 673/2019BO1) and are in line with the Declaration of Helsinki in its latest version.

Study sample
At both study sites, right-handed healthy volunteers aged 18 to 35 years were recruited via posts spread across the university hospital and social media platforms.All volunteers were asked to fill in an online screening to assess their eligibility to participate in the study.To ensure a proportional number of low, medium and high ruminators, a stratified random sampling method was used based on participants' scores on the Ruminative Response Scale (RRS, i.e., low trait ruminators: mean RRS score < 1.82, medium trait ruminators: mean RRS score !1.82 and < 2.55, high trait ruminators: mean RRS score !2.55, cut-offs based on the combined data of 715 participants from prior studies from our group; Rosenbaum et al., 2021;Rosenbaum et al., 2018a;Rosenbaum et al., 2018b;Treynor et al., 2003).For an overview of all inclusion criteria, we refer to the Supplementary materials.All eligible volunteers received an invitation to participate in the study.Based on a priori power calculations (see data plan), 138 participants were recruited for the study.However, for a variety of reasons (see Supplementary materials), eleven participants (3 in Ghent, 8 in Tübingen) were excluded from the final sample, resulting in a total study sample of 127 participants (62.2% female, mean age = 22.21, SD = 2.99).Participants were randomly assigned to one of three stimulation conditions: intermittent TBS (iTBS, n = 39), continuous TBS (cTBS, n = 43) or sham TBS (sTBS, n = 45), resulting in a betweensubjects study design.We refer to the Supplementary materials for an overview of the sample characteristics and description of the baseline questionnaires that were assessed.

Procedure
After an eligibility screening, participants were invited for the experimental session which took place in a well-controlled laboratory environment at the university hospital of Ghent or Tübingen.To control for the circadian rhythm of cortisol, the sessions started at 12:30, 13:30, 16:00 or 17:00 (counterbalanced, there were no significant differences in starting times between conditions, X 2 (6, N = 127) = 2.82, p = 0.831, u = 0.15).Participants were asked to sleep sufficiently and restrain from alcohol and intense physical activity 24 hours prior to the experimental session (Labuschagne et al., 2019).Moreover, they were asked to abstain from any caffeinated beverages, food or nicotine 2 hours prior to their appointment (Labuschagne et al., 2019;Lovallo et al., 1996;Kudielka et al., 2009).At the start of the session, participants gave written informed consent, and their resting motor threshold was determined (see neurostimulation section).Afterwards, participants were seated and all lab equipment was set-up (i.e., for the physiological measurements, see also Fig. 1).With exception of the period during stimulation and the stress task, participants remained seated on a chair positioning their knees at a 90-degree angle.During a 10-minute habituation period, participants were instructed to relax and remain quiet.Participants were allowed to read garden magazines during this period and, in the meanwhile, the stimulation target (i.e., the left DLPFC) was marked using a watersoluble pencil.Following the habituation, the magazines were taken away and a 10-minute resting-state (i.e., baseline) measurement was assessed during which participants were instructed to let their mind wander but keep their eyes open.Afterwards, a baseline assessment of mood, state rumination and subjective stress followed and a first salivary sample (approximately 25-30 min before the Trier Social Stress Test i.e., the TSST, T 1 ) was taken.Next, depending on the stimulation condition, participants received iTBS, cTBS or sham stimulation in another room.After another mood, state rumination and subjective stress assessment, participants were guided to another room for the stress induction.Please note that in Ghent there were three rooms (one for the habituation and recovery period, one for the stimulation and one for the TSST), whereas in Tübingen, due to practical limitations, there were only two rooms as the TSST was performed in the same room where the habituation and recovery period took place.After reading the instructions of the TSST, participants filled out the primary appraisal secondary appraisal (PASA) scale to assess stress appraisal (for a description and results of the PASA, we refer to the Supplementary materials).Then, participants had 3 minutes to prepare the speech and take notes (i.e., in combination with reading the instructions and filling in the PASA, this resulted in a 5-minute stress anticipation or preparation phase).Following the speech part of the TSST, participants rated their subjective stress before the jury members instructed them to perform the arithmetic task.After another stress rating, mood assessment and salivary sample (directly after the TSST, T 2 ), another 10 min resting-state measurement analogue to the first one was performed.Following this, participants completed another stress rating, state rumination and state emotion regulation inventory (SERI) (for a description and results of the SERI, we refer to the Supplementary materials).During the recovery period, salivary cortisol samples were taken every 15 min (T 3-6 ) while subjective stress was assessed.One hour post TSST, a last assessment of mood, state rumination and subjective stress as well as potential side effects of the stimulation followed.Importantly, all participants were naive to the stimulation condition and purpose of the study.As a manipulation check, we asked participants about their beliefs of the stimulation condition they were in at the end of the experimental session.Thereafter, participants were debriefed and received monetary compensation or course credit for participation.

Neurostimulation
All participants received a single session of TBS applied to the left DLPFC.To obtain optimal coil positioning, the left DLPFC was localized using the Beam F3 localization method (Beam et al., 2009;Mir-Moghtadaei et al., 2015).To obtain the individual protocol intensity, each participant's resting motor threshold was operationalized as the minimum TMS intensity necessary to yield a motor response in the right abductor pollicis brevis in 5 out of 10 successive attempts (Rothwell et al., 1999).The stimulation intensity was set at 80% of the resting motor threshold (Huang et al., 2005).There were no significant differences in stimulation intensity between the cTBS (M = 39.40,SD = 7.45), iTBS (M = 39.58,SD = 8.93) and sTBS (M = 38.73,SD = 7.28) condition, F(2,122) = 0.14, p = 0.872, g p 2 < 0.001.However, the overall mean stimulation intensity was significantly higher in Ghent (M = 43.95,SD = 7.13) as compared to Tübingen (M = 34.56,SD = 5.25), F(2,119) = 69.53,p < 0.001, g p 2 = 0.37.For differences between the study sites (e.g., in coils and devices), we refer to the Supplementary materials.For both sites, the stimulation parameters included a total of 1200 pulses at a frequency of 50 Hz and burst frequency 5 Hz, during cTBS there was an 80 s train of uninterrupted TBS including 400 bursts of 3 pulses, and the iTBS protocol consisted of 40 cycles of 2 s theta burst trains (10 bursts of 3 pulses each) followed by 8 seconds of rest (i.e., a total of 390 s, McCalley et al., 2021).Please note that the duration and end of both stimulations, cTBS and iTBS, were aligned as a waiting period was inserted prior to the cTBS.Sham TBS was applied with a coil that mimics the active stimulation.Stimulation was applied while participants were seated in a TMS chair in Fowler's position.All participants were naive to the stimulation condition, were blindfolded and asked to wear ear protection during the stimulation.To ensure double blinding, the stimulation was applied by a study nurse not further involved in the experiment.Possible side effects of the stimulation protocol were systematically evaluated following the recommendations of Rossi andcolleagues (2009, 2011).Specifically, participants were asked to rate on a five-point likert-like scale if they had any of the following experiences during or after the stimulation: transient feelings of hyperactivity, impulsivity or irritability, headache, local pain, neck pain, tooth pain, tingling sensations, changes in hearing, cognitive or neuropsychological changes and irritation at the stimulation site (responses ranging from ''1 = not at all" to ''5 = very much").

The stress task
The Trier Social Stress Test (TSST; Kirschbaum et al., 1993) was used to induce psychosocial stress in the participants (Allen et al., 2014(Allen et al., , 2017)).Specifically, participants were instructed to imagine having applied for a job and part of the job interview is to prepare for a speech about their personal strengths and qualifications.After a short preparation period, participants had to give the speech in front of two jury members (i.e., two experimenters) wearing white physician coats who were unresponsive to any signs of social interaction of the participants.Each interview lasted 5 minutes and if the participants' speech ended before this predetermined time limit, a standardized response was given by one of the jury members to encourage them to continue.Following the speech, participants were confronted with a surprise arithmetic task during which they had to count aloud backwards in steps of 13 from 2081 for 5 minutes.To maximize social evaluative threat, participants were instructed to keep eye contact with one of the jury members and calculate as fast and accurately as possible.If a mistake was made, participants were instructed to stop and restart from the starting number.Independent of their performance, participants were asked to improve their speed and accuracy during the task.Additionally, participants were informed that a video camera was recording their performance throughout the TSST, and that this data was going to be used for further analysis by behavioral experts.To maximize feelings of uncontrollability, participants were not informed of the total duration of the interview and arithmetic task.
2.5.1.2.Subjective stress.Throughout the session, participants were repeatedly asked to rate their current subjective stress level using Visual Analogue Scales (VAS).Each scale consisted of a 100millimeter straight line ranging from 0% (i.e., ''not stressed at all") to 100% (i.e., ''very stressed"), where steps of 10% were marked at steps of 10 millimeters.Higher scores indicate higher levels of subjective stress.

State rumination.
To assess state rumination during the session, a questionnaire validated by Rosenbaum and colleagues (2018aRosenbaum and colleagues ( , 2018bRosenbaum and colleagues ( , 2021) ) was used, showing excellent internal consistency (Cronbach's ɑ current study = 0.93).Participants rated the extent to which they agreed on 18 statements referring to thoughts that might have occurred during the past rest period.Responses were given on a 5-point Likert scale ranging from ''1 = not at all" to ''5 = very often".The questionnaire comprises items of the Perseverative Thinking Questionnaire (PTQ; Ehring et al., 2011), the Ruminative Response Scale (RRS; Nolen-Hoeksema and Morrow, 1991), the Amsterdam Resting-State Questionnaire (ARSQ; Diaz et al., 2013) and a questionnaire by de Jong-Meyer and colleagues (2009).We refer to Table 1 of the Supplementary materials for an overview of all items.Total scores were calculated by summing all items (items 7 and 13 reversed), with higher scores reflecting higher levels of state rumination.
2.5.2.Physiological measures 2.5.2.1.Heart rate and heart rate variability.Heart rate was measured with a one-channel electrocardiogram (ECG).In Tübingen, three Ag/AgCl ring electrodes with a diameter of 8 mm were attached to the participants' skin above the right collar bone, below the left costal arch and below the neck (i.e., for reference).The signal was recorded with a BrainAmp MR plus amplifier and Brain Vision recorder software (Brain Products, Munich, Germany) at a 1000 Hz sampling rate.In Ghent, two Ag/AgCl electrodes (42x24 mm) were attached under the left and right clavicle, and one ground electrode under the ribs (i.e., for reference).The signal was recorded with a Biopac MP160 system using a ECG100D smart amplifier, at a sampling rate of 1000 Hz, and the Biopac Acqknowledge software 5.5 (Biopac Systems Inc., USA).All ECG data were manually checked for artifacts and analyzed with Kubios HRV Premium version 3.5.0(Lipponen and Tarvainen, 2019;Tarvainen et al., 2014).Medium correction was applied to ensure the detection of erroneous (e.g., ectopic) beats, using cubic spline interpolation to replace misidentified or missing values (Tarvainen et al., 2014).HRV was assessed by calculating the Root Mean Squared of Successive Differences (RMSSD, in milliseconds) which, compared to other HRV indices, is less sensitive to movement artifacts and respiratory influences (Laborde et al., 2017;Penttila et al., 2001).As recommended (Malik et al., 1996), RMSSD was calculated for time epochs of 5 minutes resulting in a total of 16 epochs.Consistent with RMSSD, the mean heart rate (beats/minute) was also calculated in 5-minute epochs.Given our interest in cardiac changes over the different phases of the experiment (e.g., stress reactivity and recovery), the RMSSD and mean HR epochs during the baseline and recovery were averaged resulting in one measure per phase per participant (see also De Smet et al., 2021;De Smet et al., 2023).For instance, the baseline phase resulted in 2 epochs of 5 minutes that were averaged resulting in one mean value per baseline.For the recovery, each three (or two for the last phase) consecutive 5-minute epochs were averaged into one mean value resulting in 4 recovery phase values, i.e., one for 0-15 min post TSST, 15-30 min post TSST, 30-45 min post TSST and 45-55 min post TSST.
2.5.2.2.Salivary cortisol.Salivary cortisol (nmol/L) was assessed six times during the experiment using salivettes (Sarstedt AG and Co., REF 51.1534.500).Samples were collected according to the manufacturer's instructions by placing the swab in the mouth of the participants for 2 minutes.After being stored at À20 °C the samples were thawed and centrifuged for 4 minutes at 1000g before enzyme immunoassays (IBL International, Cortisol ELISA, REF RE52611) were performed in duplicate according to the manufacturer's instructions.Intra and inter-assay variation coefficients were below 10%.

Data plan
Past studies have shown moderate to high effect sizes in clinical samples for the increase in state rumination after the TSST (Rosenbaum et al., 2021; and a small effect size in healthy participants following TBS, see De Witte et al., 2020).Sensitivity analysis using G*Power 3.1 software (Faul et al., 2009) resulted in a required sample size of N = 120 to detect small differences between stimulation conditions between d = 0.2 (assuming r = 0.8) to d = 0.3 (assuming r = 0.5; within-between interaction of repeated measurements: ɑ = 0.05, power of 0.80, groups = 3, measurements = 3, non-sphericity criterion e ¼ 0:6).Considering possible losses and a dropout rate of 15%, 138 participants were recruited.The final study sample consisted of 127 participants.Due to technical issues during the session, there was less data for the analyses of the cardiac measures (n = 120).
All statistical analyses were performed with R version 4.1.2(R Core Team, 2021) in the R studio environment (RStudio Team, 2021).To fit models within the linear mixed effects framework, the 'lmerTest' (Kuznetsova et al., 2017) and 'lme4 0 (Bates et al., 2015) packages were used.Model contrasts were set using sum (i.e., effect) coding schemes.For each repeated measure, model selection was conducted by comparing the Akaike Information Criterion (AIC) for different distributions (i.e., normal, gamma, inverse-gaussian).For all (generalized) linear models, intercepts were allowed to vary randomly across participants and trait rumination (i.e., measured using the RRS) was mean centered and entered as a continuous predictor in all models.Given the results did not change when entering the study site (i.e., Ghent or Tübingen) as a covariate in the model, and to ensure the use of parsimonious models, this term was not included in the final models.Hence, all reported models were built as follows: values of the dependent variable $ time*condition*RRS + (1 | participant).The p-values of the mixed effects models were reported using type III Wald chi-squared statistics.Post-hoc tests for interaction effects consisted of pairwise comparisons of the estimated marginal means or pairwise comparisons of the estimated marginal means of linear trends using the ''emmeans'' and ''emtrends" functions of the 'emmeans' package (Lenth, 2022), respectively.The Tukey method was used to correct for multiple comparisons.Effect sizes were described using the partial eta squared (g p 2 ) and phi (u) for F and X 2 test statistics, and Cohen's d was used for t and z tests, respectively.The significance level was set at 0.05 for all statistical tests.
First, one-way ANOVAs (type III sum of squares), X 2 and F exact tests were used to examine differences between the three stimulation conditions in sample characteristics (e.g., baseline questionnaire), self-reported side effects to the stimulation protocol and task performance during the TSST (i.e., number of errors).To evaluate the effects of the different stimulation conditions on the psychological and physiological measures, and whether these effects were dependent on individual trait rumination levels, (generalized) linear mixed effects models were fitted for the repeated measures.Specifically, for mood (i.e., positive and negative affect), 4 (time: baseline, post stimulation, directly post TSST, 60 min post TSST) by 3 (condition: cTBS, iTBS, sTBS) generalized linear models were conducted.Whereas the positive affect data was best represented by a gamma distribution with an identity link function, the negative affect data showed a better fit with a gamma distribution with log link function.For subjective stress, an 8 (time: baseline, post stimulation, post speech, post math, 15 min post TSST, 30 min post TSST, 45 min post TSST, 60 min post TSST) by 3 (condition: cTBS, iTBS, sTBS) linear mixed effect model with gaussian error distribution was fit.For state rumination, a 4 (time: baseline, post stimulation, 15 min post TSST, 60 min post TSST) by 3 (condition: cTBS, iTBS, sTBS) generalized model was conducted with a gamma distribution and log link function.The cardiac data (i.e., mean heart rate and heart rate variability) was best represented by an 8 (time: baseline, preparation, speech, math, 0-15 min post TSST, 15-30 min post TSST, 30-45 min post TSST, 45-55 min post TSST) by 3 (condition: cTBS, iTBS, sTBS) generalized linear model with a log link function and gamma distribution.For cortisol, a 6 (time: baseline, directly post TSST, 15 min post TSST, 30 min post TSST, 45 min post TSST, 60 min post TSST) by 3 (condition: cTBS, iTBS, sTBS) generalized linear model with gamma distribution and log link function was fit.Additionally, a cortisol reactivity index was used to examine the change in cortisol from baseline to the maximum cortisol secretion after the TSST.As such, a 2 (time: baseline, max cortisol peak) by 3 (condition: cTBS, iTBS, sTBS) generalized model was fit with gamma distribution and log link function.Lastly, for the different indices of the PASA (i.e., primary appraisal, secondary appraisal and the stress index), one-way ANO-VAs were performed (type III sum of squares) and for the subscales of the SERI (i.e., reappraisal, distraction, brooding and acceptance), a MANOVA was performed using Pillai's trace test statistic (including condition and trait rumination as independent variables).
Non-significant interactions between time, condition and trait rumination and, non-significant interactions between time and condition, were further examined using Bayesian statistics.The Bayes Factor (BF01) was estimated to provide evidence for the lack of interaction effects and compared the likelihood of the data under the null hypothesis compared to the alternative hypothesis.The structure of the null model excluded the interaction term of interest but followed the structure of the corresponding alternative model for all other terms.The BF01 was computed based on the Bayesian Information Criteria (BIC) of the null and alternative model (for the formula see Wagenmakers, 2007), with larger values indicating more evidence in favor of the null hypothesis and reported according to Jeffreys (1961).For an overview of the results of the Bayesian analyses, we refer to Table 5 of the Supplementary materials.

Descriptive statistics
Importantly, no differences between the cTBS, iTBS and sham TBS groups were observed in trait rumination (p = 0.859) or other study sample characteristics (all p's > 0.081).An overview of the sample characteristics, and accompanying comparative statistics, can be found in Table 2 of the Supplementary materials.An exact binomial test showed that the overall success probability of participants guessing the correct stimulation condition was significantly different from chance level (64% correct, p = 0.002).In addition, the success probability was different between study sites, with the exact binomial test indicating successful blinding in Ghent (56% correct, p = 0.374) but not in Tübingen (72% correct, p < 0.001).Additional tests showed that, in Tübingen, participants in the cTBS condition guessed, above chance level, that the stimulation was active (versus sham), p < 0.001.Crucially, participants were unable to identify sham from real stimulation as correct guesses in the sham condition were not significantly different from chance in both study sites (57% correct sTBS Ghent , p = 0.664 and 54% correct sTBS Tübingen , p = 0.839).

Adverse effects
There was a marginally significant difference between conditions in self-reported irritation at the site of stimulation (i.e., left DLPFC), F(2,122) = 3.19, p = 0.045, g p 2 = 0.05.Specifically, participants in the cTBS condition (M = 1.90, SD = 1.62) reported significantly more irritation at the stimulation site as compared to participants in the sham condition (M = 1.22,SD = 1.29), t (78.48) = 2.16, p = 0.048, d = 0.47.There were no other significant differences in self-reported adverse effects between the three types of stimulation (cTBS vs iTBS vs sTBS), all F's <2.66 and p's >0.074.For an overview of the mean responses to each of the self-report items assessing side effects to the different stimulation protocols, and comparative statistics between conditions, we refer to Table 3 of the Supplementary materials.
In line with the results of positive affect, the analysis revealed no significant interaction between time, condition and trait rumination, X 2 (6, N = 127) = 7.30, p = 0.294, u = 0.24, for the generalized linear model with negative affect.There was a significant interaction between time and condition (see Fig. 2B), X 2 (6, N = 127) = 1 7.68, p = 0.007, u = 0.37.Post-hoc analyses showed that although there was no significant difference in the level of negative affect between baseline and post stimulation for the cTBS and sTBS condition (cTBS: z = 0.91, p = 0.801, d = 0.17; sTBS: z = 0.75, p = 0.876, d = 0.14), participants in the iTBS condition reported significantly higher negative affect following stimulation compared to baseline, z = 2.81, p = 0.026, d = 0.55.In all conditions, there was a significant increase in negative affect directly after the TSST (versus baseline, all z's >12.02, p's <0.001, d's >2.23), followed by a significant decrease 60 min post TSST (all z's >11.64, p's <0.001, d's >2.19).Overall, the level of negative affect 60 min post TSST was not significantly different as compared to baseline (z = 1.58, p = 0.391, d = 0.30).

Salivary cortisol
The results of the analysis revealed a significant interaction between time, condition and trait rumination for the generalized linear model with cortisol (see Fig. 4), X 2 (10, N = 127) = 19.53,p = 0.034, u = 0.39.Specifically, there was a significant interaction between time and trait rumination for the cTBS, X 2 (5, N = 127) = 25.57,p < 0.001, u = 0.77, but not the iTBS nor sTBS condition, X 20 s < 8.65, p's > 0.124, u's < 0.47.Post-hoc tests showed a negative association between trait rumination and the cortisol secretion 15 min post TSST as compared to baseline, z = 6.05, p < 0.001, d = 1.18, and as compared to directly post TSST, z = 6.61, p < 0.001, d = 1.28, indicating a lower stress-induced cortisol response for individuals with higher levels of trait rumination following cTBS (see also Fig. 4).Although not present in comparison to baseline or directly post TSST, for the cortisol secretion 30 min to 60 min post TSST (i.e., T 2 to T 4 ) compared to 15 min post TSST (i.e., T 1 ), there was a positive association with trait rumination, all z's > 4.88, p's < 0.001, d's > 0.95, with smaller decreases during the recovery period for higher levels of trait rumination.
Regarding the analysis of the cortisol reactivity index, the interaction between time, condition and trait rumination was not significant, X 2 (2, N = 127) = 4.94, p = 0.085, u = 0.20.During the TSST, 73.23% of the participants showed an increase in cortisol and 52.76% showed an increase higher than 1.25 nmol/L (Miller et al., 2013).For an overview of the raw (mean) cortisol levels for each of the stimulation conditions during the session, we refer to Table 4 of the Supplementary materials.

Discussion
The aim of this randomized, sham-controlled study was to investigate the effects of TBS of the left DLPFC on stress-reactive rumination and the psychophysiological stress response, considering individual trait rumination levels.The stress task elicited robust psychological and physiological changes, including increased subjective stress, negative affect, salivary cortisol, and heart rate, along with decreased heart rate variability, consistent with prior work (Allen et al., 2014(Allen et al., , 2017)).We also successfully replicated a significant increase in state rumination following the TSST (Rosenbaum et al., 2018a;De Witte et al., 2020).Regarding trait rumination, the study found that cTBS (compared to sTBS and iTBS) led to a reduced stress-induced cortisol response in high trait ruminators compared to low trait ruminators.However, there were no trait rumination-dependent effects of TBS on stress- reactive rumination and other psychophysiological measures of stress, contrary to the initial hypotheses.Regardless of individual trait rumination levels, cTBS showed a buffering effect on positive affect, indicating that positive emotions were less diminished following the stressor.Both cTBS and iTBS (versus sham) resulted in an enhanced heart rate recovery.However, no significant TBS effects were observed for other measures of interest.Nonsignificant findings were further corroborated by Bayesian analyses.
The study results for physiological measures showed mixed findings.High (versus low) ruminators showed increased HRV during the acute stress phase but worse HRV recovery, although Bayesian analyses did not support any (rumination-dependent) TBS effects.This differs from meta-analytical evidence showing increased HRV following rTMS (Makovac et al., 2017).Notably, only one of the included rTMS studies used a stress induction (Remue et al., 2016), and the intensity of our stressor might have masked group differences.In line with prior work (Makovac et al., 2017), both cTBS and iTBS (compared to sham stimulation) had positive effects on cardiac recovery, but within different temporal windows.Participants in the cTBS condition showed enhanced cardiac recovery within 15 minutes after the stressor, while for iTBS, the improvement was present 45 minutes post TSST.These findings suggest that different TBS modalities have different temporal effects on stress-induced changes in heart rate and may have specific impacts on the sympathetic nervous system.Crucially, we found trait rumination dependent TBS effects on stress-induced cortisol secretion.Contrary to our expectations, cTBS resulted in an attenuated stress-induced cortisol response in high trait ruminators compared to low trait ruminators, while no such effects were observed for the iTBS or sTBS condition.These cortisol findings align with the enhanced heart rate recovery and buffering effects on positive affect observed with cTBS.Our initial hypothesis that cTBS would mainly lead to inhibitory neural effects and heightened psychophysiological stress responses is most probably an oversimplification.The excitatory or inhibitory nature of TBS depends on various factors, including stimulation parameters (McCalley et al., 2021), baseline neural activity, cognitive factors such as attention (Nicolo et al., 2015), and affective states preceding or during stimulation (Schutter et al., 2023).Our findings emphasize the complexity of TBS effects and highlight the potential of exploring trait-dependent effects of TBS treatments in future research.
For the psychological measures, we replicated the well-known reactivity pattern of high trait ruminators, showing higher levels of negative affect, subjective stress and state rumination, along with greater decreases in positive affect during the acute stress and recovery phases (Aldao et al., 2014;Nolen-Hoeksema et al., 2008;Watkins and Roberts, 2020).However, contrary to our expectations, Bayesian analyses showed extreme evidence against any trait rumination dependent modulation of TBS on mood (i.e., positive and negative affect) and subjective stress.It is essential to note that the effects of trait rumination were medium to strong, but overall time-dependent changes due to the stress induction showed even larger effects.The intensity of the stressor might have masked existing group differences and any relatively small TBSinduced changes.Nevertheless, regardless of trait rumination, we observed a significant interaction between time and stimulation condition for positive affect.Specifically, positive affect significantly decreased after the TSST (versus baseline) in both iTBS and sTBS conditions, but no significant changes were found in the cTBS condition.This suggests that cTBS had a buffering effect on the detrimental impact of stress on positive affect, rather than leading to a heightened psychological stress response.On the other hand, we found no TBS effects on stress-induced changes in negative affect and subjective stress, with Bayesian analyses strongly favoring the null hypothesis.These findings align with previous studies indicating a lack of rTMS effects on self-report measures of mood or subjective stress (e.g., Pulopulos et al., 2020).Moreover, meta-analytic evidence suggests that a single session of rTMS might not be sufficient to induce robust, clinically relevant effects on stress-related emotional reactivity (Smits et al., 2020; see also Moses et al., 2023).
Regarding self-reported stress-reactive rumination, we did not find any significant TBS effects dependent on trait rumination.Bayesian analyses provided extreme evidence supporting the absence of TBS effects.Overall, consistent with previous research (De Witte et al., 2020;Era et al., 2021), our results indicate that TBS does not significantly influence self-reported stress-reactive rumination.While there were some discrepancies with the findings from De Witte and colleagues (2020), their trending results could potentially be attributed to the low number of high trait ruminators in their sample and the potential priming effects of administering the TSST before stimulation (Jackson et al. 2016; although see Vignaud et al. 2023).In contrast, our well-powered study sample, which included a balanced representation of low, medium and high trait ruminators, found no evidence supporting TBS effects on stress-reactive rumination.The complexity of TBSstress interactions warrants further investigation, and future research should take into account the nuances in experimental designs and sample characteristics.Although our study did not observe significant TBS effects on stress-reactive rumination, it is important to recognize that the influence of TBS on these stressinduced changes in rumination is likely multifaceted and context-dependent.Understanding these nuances will be crucial in advancing the potential therapeutic use of TBS for stress and rumination-related disorders.
This study has several strengths, including its rigorous design, well-powered sample size, and the use of a stratified sample based on trait rumination levels.However, some limitations need to be underscored.First, blinding was insufficient, as 64% of participants correctly guessed the stimulation they received (active vs sham).Notably, this was mainly driven by participants from Tübingen who received cTBS, correctly identifying the active stimulation.The difference in blinding success between the two study sites may be attributed to the use of different TBS devices, known to impact blinding effectiveness due to variations in Efield strengths and distributions in the brain for the coils used (Smith and Peterchev, 2018).Despite recruiting participants naive to rTMS, blinding was still less effective in Tübingen, and the reasons for this remain unclear.Importantly, the study findings were not driven by placebo effects, as only two high ruminators in the cTBS condition believed their performance improved due to stimulation.Additionally, a similar number of low and medium trait ruminators in the cTBS condition report the same expectations.Nevertheless, given the differences in blinding between study sites, it remains important to recognize the potential impact of participants' beliefs regarding the received stimulation and the possibility of bias in the results.Crucially, despite this observation, additional analysis of the (in)correct guesses as a moderator did not reveal significant influence on the study results.Second, this study is limited in the causal inferences it is able to draw.Due to the scarcity of literature on TBS in the context of state rumination, this study did not include neuroimaging assessments.Investigating neural correlates could shed light on whether the lack of effects is due inadequate neuromodulation of the underlying brain areas and network.Future research with neuroimaging assessments may further illuminate the neural mechanisms underlying these effects.
In conclusion, contrary to our a priori expectations, cTBS over the left DLPFC (versus iTBS and versus sham) demonstrated beneficial effects on certain psychological (i.e., positive affect) and phys-iological (i.e., mean heart rate and cortisol) measures of stress.One potential explanation for these findings is that cTBS may exert its beneficial effects by suppressing specific brain regions or circuits responsible for self-monitoring of stress, leading to an improved stress regulation and a potential positive emotional buffer.However, it is important to note that further research is required to fully explore the underlying working mechanisms of cTBS in this context.Notably, individuals with higher trait rumination appeared to benefit more from the cTBS effects in terms of the stress-induced cortisol response, highlighting the importance of considering individual differences in rumination when examining the effects of stimulation.To gain a comprehensive understanding of the link between TBS and stress, further research is needed to investigate the underlying neural mechanisms and the specific brain regions involved in mediating TBS effects on stress and emotional responses.Additionally, studying the effects of TBS in various stress-related conditions and populations may help identify its potential as a therapeutic tool for stress management and related disorders.

Fig. 2 .
Fig. 2. Raincloud plots for (A) positive affect, (B) negative affect, (C) subjective stress and (D) state rumination.Individual jittered raw data is represented by dots.Data distributions are depicted by split-half violin plots (on the right side of each series of boxplots).Each boxplot displays the stimulation condition median alongside the interquartile ranges (horizontal lines).Bold dots in the boxplots represent the estimated marginal mean of the fitted models.Abbrev.: cTBS, continuous theta burst stimulation; iTBS, intermittent theta burst stimulation; min, minutes; sTBS, sham theta burst stimulation; TSST, Trier Social Stress Test.

Fig. 3 .
Fig. 3. Raincloud plots for (A) mean heart rate, (B) heart rate variability and (C) salivary cortisol.Individual jittered raw data is represented by dots.Data distributions are depicted by split-half violin plots (on the right side of each series of boxplots).Each boxplot displays the stimulation condition median alongside the interquartile ranges (horizontal lines).Bold dots in the boxplots represent the estimated marginal mean of the fitted models.Abbrev.: cTBS, continuous theta burst stimulation; iTBS, intermittent theta burst stimulation; min, minutes; RMSSD, root mean square of successive differences; sTBS, sham theta burst stimulation; TSST, Trier Social Stress Test.

Fig. 4 .
Fig. 4. Raincloud plots for cortisol secretion during the session for (A) low trait ruminators, (B) medium trait ruminators and (C) high trait ruminators.Individual jittered raw data is represented by dots.Data distributions are depicted by split-half violin plots (on the right side of each series of boxplots).Each boxplot displays the stimulation condition median alongside the interquartile ranges (horizontal lines).Bold dots in the boxplots represent the estimated marginal mean of the fitted models.Centered trait rumination scores were entered as a continuous predictor in the model.However, for visualization purposes, trait rumination is here presented as a factor.Abbrev.: cTBS, continuous theta burst stimulation; iTBS, intermittent theta burst stimulation; min, minutes; sTBS, sham theta burst stimulation; TSST, Trier Social Stress Test.