The power of context: Enhancing repetitive transcranial magnetic stimulation effects through skillful behavior

Abstract

Introduction:

Non-invasive neuromodulation techniques like repetitive transcranial magnetic stimulation (rTMS) have shown promise in treating psychopathology. However, emerging evidence suggests that the brain’s functional state during stimulation may significantly influence treatment outcomes. Pairing neurostimulation with cognitively engaging tasks that activate targeted neural networks could enhance neuroplasticity and improve efficacy. Emotional dysregulation is a core feature of many psychiatric disorders. To explore whether combining behavioral interventions with neuromodulation enhances outcomes, we conducted a secondary analysis examining the effects of pairing rTMS with cognitive restructuring (CR), a skill-based intervention, on emotion regulation.

Materials and methods:

Forty-three adults with high emotional dysregulation completed an fMRI task involving aversive sounds to identify individualized neurostimulation targets. During a subsequent experimental visit, participants engaged in three randomized tasks involving either passive listening or active downregulation of distress from aversive or neutral sounds. Each task was paired with one of three stimulation conditions: sham rTMS, 10 Hz rTMS targeting the emotion regulation network, or 1 Hz rTMS targeting a non-congruent network.

Results:

Results revealed that 10 Hz rTMS combined with CR significantly reduced self-reported distress. Moreover, pairing CR with stimulation over the emotion regulation network led to lower distress and increased high-frequency heart rate variability (HF-HRV), a physiological marker of emotion regulation, compared to stimulation of a non-targeted network.

Conclusion:

These findings underscore the importance of contextual factors in neuromodulation and suggest that integrating rTMS with active cognitive strategies like CR may be a promising approach for enhancing emotion regulation in adults with psychiatric distress.

Noninvasive neurostimulation, such as repetitive transcranial magnetic stimulation (rTMS) or transcranial electric stimulation (TES), has beneficial effects in reducing psychopathology severity across a range of conditions, including obsessive-compulsive [1,2], anxiety and mood [3,4], stress [5], and substance use disorders [6–8]. TMS involves placing a coil on the scalp and inducing a magnetic field that passes through the scalp, skull, and tissue, creating a targeted electrical current in the neurons closest to the coil. TES involves the direct or alternate application of current to the brain via electrodes placed on the scalp, which serve as both input and output. This, and other approaches to neuromodulation, have initially been used experimentally to understand brain function causally [9], and have been discovered to have small to moderate therapeutic effects when applied rhythmically and repeatedly [3].

Recently, there has been increased interest in understanding the parameters of therapeutic neurostimulation, with the ultimate goal of optimizing its benefits. Several studies highlight that targeting neural networks versus regions [10–12], improving precision with MRI and neuro-navigation [3,13], and employing e-field modeling to optimize coil position and orientation [10,14] improve outcomes. In addition, adjusting the inter-trial interval [13,15], administering at least 1600 pulses [16], and spacing stimulation sessions by at least an hour [17] can also increase the beneficial effect of neuromodulation. Emerging research also suggests that low doses of the NMDA receptor partial agonist D-cycloserine may enhance the efficacy of TMS [18].

One parameter that has received relatively less attention is the state of the brain before or during neurostimulation. Many neuromodulation therapies, including the FDA-approved treatments for depression, are applied passively without directing the neural activity during stimulation [19,20]. Applying neurostimulation in a passive state may result in less targeted plasticity, potentially limiting its effectiveness in producing clinically meaningful improvements [21]. Furthermore, directing brain activity before [22] or during [23] neurostimulation has the potential to enhance outcomes. Therefore, understanding and harnessing the state of the brain in neuromodulation paradigms can also yield additional therapeutic benefits.

Context, encompassing what occurs before and during neurostimulation, has already been established as a critical parameter in experimental cognition and perception studies using neuromodulation [24]. A series of seminal studies [25–27] demonstrated that the brain’s functional state, and relative activity of functionally distinct neural populations during stimulation, can significantly influence perceptual outcomes. Before neurostimulation selectively alter the excitability of neuronal populations of interest. Priming refers to the brief (< 1 s) exposure to cues, which activates but does not fully engage the neural networks that respond to the presented stimuli. Priming prepares the brain to more easily react to already encountered stimuli by activating relevant neuronal populations above the baseline level, thus requiring a smaller additional input to fire [28]. In contrast, adaptation refers to a more prolonged stimulus exposure (>40 s), which both activates and engages the relevant neuronal population. Adaptation leads to a decrease in neural firing in response to the adapted stimulus, which represents a reduction in excitability achieved through lowering the resting potential of involved neuronal populations [29]. Thus, neurons that have undergone adaptation have the lowest excitability, followed by neurons at rest, followed by primed neurons, which have the highest resting potential.

Neurostimulation is most effective when directed at neuronal populations with reduced excitability, such as those that have undergone adaptation. Prior studies show that brief adaptation enhances the brain’s responsiveness to TMS: for example, exposure to a 30-second color stimulus altered TMS-evoked phosphenes and boosted detection of congruent stimuli [30], while TMS reversed adaptation-induced slowing in emotion recognition tasks [26]. Critically, these effects occur only when adaptation happens immediately before stimulation. For example, preconditioning iTBS with cathodal tDCS increases neuroplasticity in the motor cortex, but only when applied directly beforehand [31]. The consistent finding is that adapted (less excitable) neurons are more receptive to TMS input, leading to stronger and more targeted effects.

Priming inhibits neurostimulation effects because primed neurons are already highly excitable and therefore less responsive to TMS. Cattaneo et al. [25] found that primed letters are more easily recognized than non-primed letters. However, TMS stimulation of the left posterior parietal cortex (PPC) reverses this, leading to faster recognition of non-primed letters. This is because non-primed neurons, being at rest and less excitable, respond better to TMS. Silvanto et al. also demonstrated that TMS alters the perception of visual stimuli that are incongruent with the prime, indicating that primed neurons are less responsive to TMS [32]. Overall, these findings demonstrate that the brain’s pre-stimulation state determines which neurons are influenced by neuromodulation.RTMS exerts stronger effects on less active neuronal populations.

Several clinical neurostimulation paradigms administer a task paradigm to participants before they receive stimulation. For example, Carmi et al. demonstrated that prompting OCD obsessions before stimulation improves outcomes [33]. Similarly, the studies from the addiction field have shown that activating craving before neurostimulation enhances efficacy [22,34]. Others have also found that performing cognitively demanding tasks before rTMS boosted its antidepressant effects [35]. For example, Li and colleagues used a cognitive task to enhance frontal theta (θ) power before rTMS, which led to significantly greater reductions in depression scores compared to sham stimulation [36]. Finally, Luber and colleagues [37] demonstrated that neurostimulation targeting the left lateral occipital cortex during a working memory task after sleep deprivation reduced the negative impact of sleep loss on memory performance.

Across these studies, participants spent more than 40 s (3–5 min in the Carmi experiments, light and inhale from a cigarette in the Dinur-Klein experiment, 5-minute task in the Li experiments) engaging with the cue before the stimulation ended. Given the duration, these are likely adaptation paradigms. We can hypothesize that the reason these contextual manipulations before TMS improve outcomes is that they engage relevant regulatory neural networks, such as those involved in inhibiting compulsive behaviors, potentially making these networks more responsive to neurostimulation. With depression studies, demanding cognitive tasks are likely engaging the same cognitive control networks that psychotherapeutic cognitive regulation strategies do [38,39] making these critical networks for reducing depression more amenable to TMS input. Despite these promising results, the mechanisms by which adaptation enhances TMS outcomes remain underexplored in clinical applications. Furthermore, to date, no clinical applications have used priming to enhance the benefits of neuromodulation. Fig. 1 presents theoretical possibilities for using adaptation and priming to enhance clinical outcomes in neuromodulation.

Fig. 1.

Theoretical model of how neural context can be manipulated to enhance neuromodulation outcomes.

The brain's state during neurostimulation also significantly impacts outcomes. In a recent review [24], Bradley and colleagues discuss how consciousness, attention, and working memory—each inducing different brain states—interact with the outcomes of neurostimulation. According to their review, attention allocation may direct the network affected by neurostimulation when administered to a node that is part of multiple networks. Furthermore, when attention is voluntarily allocated toward the outcome, stimulation effects are amplified; paired associative stimulation combined with attention directed either toward or away from the limb being stimulated exemplifies this case. This experiment suggests that where participants direct their attention is critical in guiding the efficacy of neuromodulation. The review also highlights that increasing the difficulty of the cognitive task improves working memory performance, supporting the theory that the more participants engage a relevant network via a behavioral task, the more effective neurostimulation over that network will be.

Importantly, the studies summarized above demonstrate congruence between the network activated by the behavioral task employed and that stimulated by TMS or TES. This congruence, in theory, can enhance neuroplasticity and improve therapeutic response [40–46]. For example, participants who underwent Visual Rehabilitation Therapy (VRT) in addition to active tDCS have shown significantly greater visual field expansion [31] and better recovery of limb motor function following stroke [47] compared to control conditions employing either intervention modality alone. Furthermore, Neacsiu and colleagues also showed that administering a single session of self-system therapy (SST) while participants simultaneously undergo rTMS—using network-guided principles—helped to reduce depression [47]. In contrast, when the behavioral task and neurostimulation network do not overlap, no benefit from neurostimulation can be found [44], or in some cases, neurostimulation may impair performance [45]. These findings, along with others, suggest that neuronal excitation may propagate more effectively across neural networks when the connections are actively engaged by a congruent task [21]. The mechanism of this phenomenon may be that concurrent activation strengthens synaptic efficacy, leading to more durable and functionally meaningful changes [48,49]. Most studies on brain context during stimulation, similar to those on priming and adaptation, primarily focus on basic science and normative cognitive processes. Limited research exists on clinical applications. However, combining neurostimulation with targeted skill-based interventions can be a very promising strategy to enhance therapeutic efficacy.

Although combining neuromodulation with psychological interventions is theoretically promising, many studies report limited benefits when neuromodulation is paired with psychological interventions [for a review, see Dendoncker et al. [50]]. This may be due to the use of behavioral interventions that are not evidence-based [51], or a mismatch between the neural circuits engaged by the task and those targeted by stimulation. For instance, Isserles et al. [52] found that encouraging rumination during dlPFC stimulation worsened depression outcomes, likely due to the activation of maladaptive networks. Overall, directing brain activity during neuromodulation can influence therapeutic outcomes. We hypothesize that behavioral interventions must engage neural circuits overlapping with those targeted by stimulation to produce synergistic effects.

1. Conceptual model

We propose a unified conceptual model (Fig. 1) aimed to enhance neuromodulation efficacy by (A) altering the excitability of targeted or unwanted neuronal populations immediately before stimulation through neural adaptation (reduced excitability) or priming (heightened excitability); (B) manipulating brain activity during stimulation with a task that engages similar circuitry for congruent effects or different circuitry for additive effects; or (C) taking advantage of the heightened state of plasticity induced by neuromodulation with a behavioral intervention. This model generates several testable hypotheses, of which we focus the proof-of-concept study on Fig. 1B. Using Hebbian principles, we can hypothesize that when the brain activity induced by a behavioral task aligns with the network being stimulated, this coactivation amplifies network-specific propagation and yields greater functional gains than either modality alone. We also hypothesize that synergistic effects yield better outcomes than additive effects (i.e., congruent activation will be superior to incongruent activation).

2. Current study

This study aims to investigate whether aligning a behavioral intervention—specifically, cognitive restructuring to regulate intense emotions—with the neural network targeted by neuromodulation enhances clinical outcomes in reducing emotional distress. Neuromodulation is a promising tool for addressing clinical difficulties in emotion regulation and emotion dysregulation [53]; it has been shown to enhance voluntary emotion regulation, as measured by both subjective and physiological indicators [54]. Our prior work [55] also showed that rTMS boosts high-frequency heart rate variability (HF-HRV), a marker of effective emotion regulation [56], during cognitive reappraisal in a transdiagnostic sample. In nonclinical populations, dlPFC stimulation has been shown to influence emotion regulation by reducing threat vigilance [57], enhancing inhibitory control [58], and improving cognitive reappraisal [59]. Collectively, these findings suggest that neuromodulation can be utilized to alter emotional processes, providing an optimal platform to test whether a congruent behavioral task can enhance its effectiveness.

In this proof-of-concept secondary analysis, we used functional imaging to identify two neural networks that we stimulated in a clinical sample of 43 transdiagnostic adults with high emotion dysregulation. The cognitive regulation network was activated when the participant engaged in cognitive restructuring to reduce distress resulting from listening to unpleasant sounds. The emotional reactivity network was activated when the participants listened to distressing sounds without attempting to regulate their distress. In a subsequent visit, participants learned cognitive restructuring (a behavioral intervention that activates the cognitive regulation network) and then applied it while receiving high-frequency rTMS over the regulation network (congruent condition) or low-frequency rTMS over the emotional reactivity network (incongruent condition). At the initial imaging session, participants were briefly instructed to use cognitive restructuring without the depth of training that occurred at the subsequent intervention session because the goal of the imaging session was to identify a neurostimulation target within the cognitive regulation network, while the overall objective of the study was to examine the effectiveness of combining neurostimulation with a session of cognitive restructuring in reducing emotional distress. We hypothesized that congruent interventions would yield the highest reductions in distress compared to additive interventions or each intervention alone.

3. Materials and methods

3.1. Participants and procedures

This study is a secondary analysis of our previous work on identifying intervention targets for misophonia [61], a disorder characterized by decreased tolerance to specific sounds [60]. The study is registered under Clinical Trials ID NCT04348591 and was conducted between October 2020 and May 2022. Specifically, in this paper, we included only participants with high emotional dysregulation [operationally defined as a total score of 89 or higher on the Difficulties in Emotion Regulation Scale; DERS)] [61], including all the clinical participants from the original analysis and some participants with misophonia who also met criteria for high emotion dysregulation. All participants with misophonia who did not meet the threshold for clinical emotion dysregulation were excluded from this analysis.

Participants (M_age = 28.05, SD = 9.44) were recruited from the community via online platforms, flyers, and physician referrals. Interested individuals (n = 313) completed both online and in-person screenings (n = 112) before undergoing an imaging session (n = 60). The subsample analyzed for this study includes 43 transdiagnostic clinical adults with high emotional dysregulation.

The Duke University Health System Institutional Review Board approved all procedures, and participants provided written informed consent. Each participant received up to $250 in compensation. Participants were also excluded because of MRI contraindications, TMS contraindications, or if they were no longer interested or unreachable. Detailed demographic characteristics are provided in Table 1.

Table 1.

Demographics and clinical descriptive by group.

	Study sample (n = 43)
Female gender identity (%)	86.00
Latinx background (%)	16.30
Racial background (%)
Asian/Asian American	18.60
Black/African American	0.00
Native American, American Indian, or Alaskan Native	2.30
White/Caucasian	74.40
Middle Eastern	4.70
Total # of diagnoses, current (SD)	2.00 (1.76)
Total # of diagnoses, lifetime (SD)	3.77 (2.26)
Primary diagnoses (%)
Attention-Deficit/Hyperactivity Disorder (ADHD)	7.00
Anxiety Disorders	58.10
Depressive Disorders	11.60
Post-Traumatic Stress Disorder (PTSD)	4.70
Other Specified Obsessive Compulsive Disorder	2.30
No Current Disorder	16.30
Lifetime disorders (%)
Mood disorders	76.70
Anxiety disorders	88.40
Any pd (%)	41.90

Note: SD = Standard Deviation; PD = personality disorder

Depressive disorder = MDD, PMDD, and PDD

Anxiety disorders = social anxiety disorder, generalized anxiety disorder, other specified anxiety disorder, panic disorder

Following written informed consent, participants completed diagnostic assessments [62,63], a verbal intelligence test [64], and a questionnaire packet. Participants who met eligibility criteria completed an in-person auditory task in which they categorized 101 pre-selected 6 s sound stimuli. Of these, 30 were classified as positive or neutral, while the remaining were aversive, including 40 misophonic sounds (e.g., coughing, snoring, chewing). Following each stimulus presentation, participants provided valence and arousal ratings, which were used to identify individually tailored aversive and neutral stimuli. Based on these ratings, personalized sets of 12 aversive, 12 misophonic, and four neutral sounds were used for each participant in subsequent experimental sessions. Importantly, for participants characterized by high emotion dysregulation but without a diagnosis of misophonia, only misophonic sounds rated as aversive were included in their final stimulus set. Consequently, within this subgroup, both the “aversive” and “misophonic” categories comprised sounds perceived as aversive.

3.2. Neuroimaging procedure

Next, participants completed an imaging experimental session on average 8.65 days (SD = 11.87) after the eligibility intake session. After standardized training, participants entered a research-dedicated GE HD 3.0 T MRI scanner, where an anatomical image was acquired, followed by four runs of EPI functional imaging. During these runs, participants were exposed to the personalized sounds and were instructed to either downregulate their emotional responses to the sounds using cognitive restructuring or to passively listen to the sounds and experience their emotions without altering them in any way.

After the MRI session, fMRI analyses were performed to define each participant’s regulation and reactivity network from which neurostimulation targets were identified. Structural and functional data were examined using MRIQC and then preprocessed with fMRIprep v1.1.4 [65] and analyzed using FSL [66]. In addition, a psychophysiological interaction (PPI) analysis was conducted to assess functional connectivity with the left insula while participants listened to personalized aversive sounds compared to neutral sounds. Connectivity with the left insula was chosen based on prior findings indicating contralateral connectivity between limbic and prefrontal regions, and the tendency for affective processing to be lateralized to the left insula [67,68].

We defined the cognitive regulation network based on the contrast between “downregulating” and “passive listening” in response to aversive sounds. The emotional reactivity network used the “passive listening” aversive versus neutral sounds contrast. Both statistical maps (unthresholded) were transferred back to native space [69] and overlaid onto the anatomical images using neuronavigation software (BrainSight, Rogue Research, Canada). For each participant, the cluster within the right dorsolateral prefrontal cortex (dlPFC) showing the strongest positive z-statistic for the “downregulate” contrast was defined as the congruent stimulation target (average z-value = [2.73], SD = [0.78]). Similarly, the right mPFC cluster with the strongest connectivity to the left insula in the PPI analysis of the emotional reactivity contrast was defined as the incongruent stimulation target (average z-value = [3.11], SD = [1.60]). Refer to Fig. 2 for a visual representation of the personalized targets (converted into MNI space). Targets were manually selected, and coil tilt and orientation were set such that the induced E-field was perpendicular to the closest sulcal wall, as done in prior studies [70].

Fig. 2.

Personalized targets used for participants in this study. Blue targets represent the voxel within the right dlPFC with the highest activation during the [Downregulate vs. Listen] to aversive sounds contrast, used to map the cognitive emotion regulation network. Stimulation over these targets activated a network that was congruent with the behavioral intervention utilized simultaneously (congruent condition). Red targets represent the cluster in the right mPFC with the highest connectivity following PPI analyses to the cluster within the left insula with the highest activation during the [Listen to aversive vs. neutral] sounds contrast. This contrast mapped the emotional reactivity network. Stimulation over these targets was intended to activate a network that was incongruent to the behavioral intervention (incongruent condition). All targets were selected in native space and were converted to MNI space for this illustration only. Therefore, the exact locations of targets on the native brain may have varied slightly from what is presented in this figure.

3.3. Neurostimulation procedure

Participants returned for a 3.5-hour session combining intensive skills training and neurostimulation (see Fig. 3) on average 13.61 days (SD = 11.38) after their MRI session. The first 45 min were devoted to in-depth, one-on-one cognitive restructuring (CR) training delivered by a clinical psychologist with expertise in cognitive-behavioral therapy (the first author). All behavioral intervention procedures were delivered by the same provider. Following the skills training, participants’ resting motor thresholds (rMT) were established using standard procedures [71].

Fig. 3.

Neurostimulation intervention study procedure '.

Continuous psychophysiological data were collected throughout the session using the BIOPAC MP150 system via electrodes monitoring galvanic skin response (GSR) and heart rate (HR). Active and sham rTMS were administered using a figure-8 coil (A/P Cool-B65) connected to a MagPro X100 stimulator (MagVenture, Denmark), which delivered biphasic pulses. For the congruent stimulation, a 10 Hz protocol (HF-rTMS) was applied over the individualized right dlPFC target using 5 s of stimulation followed by a 15-second inter-train interval at 120% rMT. The incongruent stimulation consisted of 1 Hz protocol (LF-rTMS) delivered continuously over the individualized mPFC connectivity target at 90% rMT [72]. 1 Hz was selected for the incongruent condition because we attempted to inhibit emotional reactivity as an additive to promoting regulation via cognitive restructuring, needing an inhibitory neuromodulation paradigm. Sham stimulation was applied at the same intensity with the coil in placebo mode [73]. Coil position and orientation were continuously monitored via a stereotaxic neuronavigation system (Brainsight, Rogue Research, Canada).

All participants underwent the three neurostimulation conditions (sham rTMS, HF-rTMS, and LF-rTMS) in a counterbalanced order (i.e., the order of sham, HF, and LF conditions was randomly selected for each participant). Between each condition, participants had a 10-minute break to recover and reduce carryover and fatigue effects. During each neuromodulation condition, participants received 3 min of neurostimulation alone, followed by three listening blocks and two downregulation blocks, each lasting 2 min. Each block consisted of four 25-second presentations of either aversive, misophonic, or neutral sounds, with each sound presentation followed by a 5-second rating period. For this secondary analysis, the misophonic and aversive sounds were combined. In listening blocks, participants were instructed to listen to the sounds while receiving neurostimulation. In the downregulation blocks, participants were asked to use CR while listening to the sounds with the end goal of reducing their emotional distress. In this way, each participant received blocks of congruent interventions (HF-rTMS+CR), blocks of incongruent interventions (LF-rTMS+CR), and blocks of each intervention alone (HF-rTMS+Listen, LF-rTMS+Listen, and sham-rTMS+CR) while listening to personalized aversive sounds. Emotion regulation and arousal were indexed by high-frequency heart rate variability (HF-HRV) and subjective units of distress (SUDS). HF-HRV was extracted from each 120-second sound task block, while changes in SUDS from baseline were assessed for each 25-second sound presentation.

3.4. Neuroimaging data processing

Artifacts in fMRI data were removed using the non-aggressive variant of ICA-based automatic removal of motion artifacts (AROMA) tool in fMRIPrep [74]. Preprocessing of the T1-weighted anatomical images included intensity correction using N4BiasFieldCorrection [75], skull-stripping, and spatial normalization to the ICBM 152 Nonlinear Asymmetrical template v2009c [76]. Functional image data were preprocessed with slice-time correction, motion correction, and spatial registration to the high-resolution T1-weighted image. The fMRI data were then moved to standard space by applying the concatenated motion-correction transforms, fMRI-to-T1 transform, and T1-to-ICBM transform. Next, the fMRI data underwent high-pass temporal filtering with a 100-second cutoff (FSL’s fslmaths) and were masked with a standard MNI brain template to exclude outside-of-brain signals. Functional data were analyzed at the first level as individual runs, using a general linear model (GLM) implemented in FSL’s FEAT procedure [66], followed by a second-level fixed-effects model that combined all runs.

3.5. Psychophysiological data processing

High-frequency heart rate variability (HF-HRV) is an established indicator of emotion regulation [77–81] and was therefore examined as a physiological metric of regulation success. AcqKnowledge software was used to pre-process raw ECG data following established guidelines [82,83]. Before analysis, each raw ECG segment was visually inspected, and artifacts were cleaned using band-pass and comb-band-stop filtering and manual removal of noise [83].

3.6. Statistical analyses

Analyses were conducted using SPSS version 25.0. We conducted analyses examining ΔSUDS and HF-HRV. Planned covariates for all analyses included coil-to-cortex distance. Data-driven covariates, such as headache incidence post-stimulation, were also examined and included as needed. Effect sizes were computed using Feingold's formula [84] and interpreted using Cohen's specifications [85]. HF-HRV was measured in seconds² and transformed to ms² (HF-HRV*1000000) for ease of interpretation. The distribution of HF-HRV was transformed to normal using the function lg10 (HF-HRV*1000000). All other data were normally distributed.

4. Results

4.1. Subjective distress (SUDS)

A multivariate mixed-model analysis of variance (MMANOVA) using a Toeplitz covariance structure was conducted, controlling for group (misophonic and non-misophonic), race (white and non-white), headache during rTMS, and brain-to-scalp distance. Group, race, and headache were covariates included post-hoc because of observed significant effects on subjective distress. Two participants were excluded from the SUDS analysis because they provided a SUDS rating of 0 at all timepoints.

We found a significant main effect of target favoring HF rTMS (F[2248.35] = 12.49, p < .001), instruction (F[2450.67] = 13.36, p < .001) showing higher distress following listening versus downregulating instructions, and group (F[1,38.35] = 6.81, p = .013), showing higher distress for non-misophonic participants with high emotion dysregulation [this effect is discussed in more detail in the original data analysis [86]]. We also found a significant group × target × instruction interaction (F[12, 413.12] = 2.00, p = .023), which enabled us to explore the hypothesized differences. Specifically, pairing CR with congruent stimulation (HF rTMS over the right dlPFC) while listening to aversive sounds led to the lowest change in SUDS from baseline (Estimated marginal mean [EMM] = 0.19, SE =.32), a significant difference from HF rTMS alone (EMM = 1.43, SE = 0.36; p < .001), CR alone (EMM = 0.91, SE = 0.30; p < .001), and CR with incongruent stimulation (LF rTMS over the mPFC-insula connectivity; EMM= 0.83, SE =.32; p = .002). In other words, combining CR with congruent stimulation resulted in more successful regulation than combining CR with incongruent stimulation, and was more successful than either intervention alone (Fig. 4). The main effects for racial background, headache, and brain-scalp distance were not significant.

Fig. 4.

Changes from baseline Subjective Units of Distress (SUDS) using Estimated Marginal Means (EMM) during different within-subject experimental conditions. The panels show changes in distress during sham rTMS, high-frequency rTMS (10 Hz), and low-frequency rTMS (1 Hz). Within each experimental condition, participants either listened to neutral or aversive sounds or were instructed to downregulate their distress related to aversive sounds using cognitive restructuring.

4.2. High-frequency heart rate variability (HF-HRV)

A multivariate mixed-model analysis of variance (MMANOVA) using an unstructured covariance structure was conducted, controlling for baseline, race, headache, and brain-to-scalp distance. We found that higher HF-HRV during baseline predicted higher HF-HRV during the experimental task (F_baseline[1, 48.95] = 122.23, p < .001). We also found a main effect of target (F[2, 61.38]=8.59, p < .001) showing the highest HF-HRV following HF-rTMS. The interaction effect was marginally significant (F_{targetXinstruction} [4,67.70] = 2.37, p=.06). Pairing CR with congruent stimulation while listening to aversive sound led to a greater increase in HF-HRV (EMM = 2.07, SE = 0.08) than CR with incongruent stimulation (EMM = 1.97, SE = 0.07; p=.22). Combining CR and congruent stimulation led to marginally higher HF-HRV when compared to CR alone (EMM = 1.997, SE =.07, p = .09; Fig. 5). There were no significant differences in HF-HRV between CR with congruent stimulation and stimulation alone (EMM = 2.13, SE =.08, p=.13). The main effects for racial background, headache, and brain-scalp distance were not significant.

Fig. 5.

Change in high frequency Heart Rate Variability (HF-HRV), a measure of arousal, from baseline (estimated marginal means). The panels show changes in distress during sham rTMS, high-frequency rTMS (10 Hz), and low-frequency rTMS (1 Hz). Within each experimental condition, participants either listened to neutral or aversive sounds or were instructed to downregulate their distress related to aversive sounds using cognitive restructuring.

5. Discussion

Our results demonstrate that augmenting neurostimulation with congruent manipulation of the targeted circuit may be a promising approach for reducing clinical emotional dysregulation, specifically, and for optimizing neuromodulation interventions more broadly. In this proof-of-concept secondary analysis, we examined whether integrating cognitive restructuring, a skill-based behavioral intervention, with repetitive transcranial magnetic stimulation (rTMS) enhances emotion regulation outcomes in clinical adults with high emotional dysregulation. We also examined whether overlap in the networks activated by the behavioral and neurostimulation interventions is important for enhancing outcomes. Our hypotheses were partially supported. Consistent with our hypothesis, we found that combining neurostimulation with CR significantly reduces subjective distress when presented with personalized aversive sounds, above and beyond the effects of either intervention alone. Contrary to our hypothesis, an augmenting effect of CR when added to neurostimulation on heart rate variability was not observed. Consistent with our hypotheses, both self-report and psychophysiological analyses revealed that congruent stimulation is crucial when augmenting neuromodulation effects with a behavioral paradigm. Incongruent stimulation led to significantly greater distress and lower heart rate variability when listening to aversive sounds, compared with congruent stimulation. Building off evidence from our previous work [55,70,87], these findings reinforce the growing consensus that the therapeutic efficacy of neurostimulation may be improved by considering the functional context in which it is delivered.

Cognitive restructuring (CR) is a well-established emotion regulation strategy used in many evidence-based psychotherapies [88–90]. CR helps treat psychopathology by encouraging individuals to reinterpret situations in ways that alter their emotional experiences [91]. Meanwhile, neurostimulation is gaining widespread clinical use, with FDA-approved protocols now available for conditions such as depression, obsessive-compulsive disorder, and nicotine dependence [92]. While both CR and rTMS have demonstrated individual benefits, emerging evidence suggests their combination may yield synergistic effects.

We examined whether combining CR with rTMS could yield greater reductions in emotional distress than either intervention alone. Additionally, we selected two stimulation targets based on their established roles in emotional processing and regulation: low-frequency (LF) stimulation was applied to the cluster in the medial prefrontal cortex (mPFC) with the strongest connectivity to left insula activation during exposure to aversive sounds. Diffusion tensor imaging data show direct anatomical connections between the insula and the mPFC [93,94], and experimental findings indicate that the mPFC modulates emotional reactivity [95]. In addition, high-frequency (HF) stimulation targeted the cluster in the right dorsolateral prefrontal cortex (dlPFC) that had the highest activation when the participant was engaged in restructuring thoughts associated with exposure to unpleasant sounds. A substantial body of research links activity in the dlPFC to cognitive emotion regulation [38,96]. By applying low-frequency stimulation to the mPFC to reduce emotional reactivity and high-frequency stimulation to the dlPFC to enhance cognitive control, we aimed to investigate whether altering different emotional processes would yield differential effects on emotion regulation.

We propose that the enhanced effects of dlPFC stimulation, compared to mPFC stimulation when added to CR, are primarily attributed to the congruence between the networks targeted by the combined interventions. Specifically, the dlPFC stimulation aligned with the neural networks recruited by CR, whereas the mPFC stimulation was targeted towards a network that was functionally active when emotional distress was induced and may not have overlapped as much with the network activated by CR. Very few studies have examined the effects of using fMRI-guided tasks to inform stimulation targeting. In our prior work, we have demonstrated the feasibility of combining CR with neurostimulation using network-driven approaches [97,98].

HF-HRV, a physiological marker of emotion regulation, increased significantly during HF-rTMS. However, adding CR did not have any additional beneficial effects. Nonetheless, the elevation of HF-HRV in the HF-rTMS condition suggests that high-frequency stimulation over cognitive control regions may promote physiological states associated with effective emotion regulation [55,99]. In other studies, implementation of CR did lead to increases in HF-HRV [100]. The lack of augmentation effects may therefore be due to the low power to detect a difference between these interventions, or HF-HRV hitting a ceiling effect from rTMS stimulation, especially given the observed beneficial effects of CR addition to rTMS on self-report outcomes. It is also possible that a larger dose of each intervention, a more sensitive measurement, or changes in measurement timing (e.g., after, rather than during CR) are needed to detect psychophysiological differences. We measured HF-HRV in 2-minute blocks, which may be insufficient to capture effects [101]. Longer regulation blocks may bring additional sensitivity to detect enhancements from CR. Additional research is needed to determine the optimal overlap of these interventions.

Additionally, findings from multiple fields, including motor learning, memory training, language rehabilitation, and treatments for post-traumatic stress disorder and addiction, further underscore the importance of the behavioral context during neurostimulation. In the motor learning field, several studies have demonstrated that coupling TMS with concurrent motor skill practice enhances cortical plasticity, accelerates task acquisition, and yields greater performance gains than stimulation alone [102–105]. In the cognitive domain, Martin et al. [106] showed that neurostimulation delivered during working memory training led to greater long-term improvements than either intervention alone. Similarly, Fridriksson et al. [107] found that pairing neurostimulation with speech-language therapy significantly enhanced language recovery in individuals with aphasia. In the context of addiction, Kearney-Ramos et al. [34] demonstrated that rTMS applied during alcohol cue exposure more effectively reduced cravings than TMS alone.

These effects may arise from the strengthening of task-relevant neural circuits during periods of heightened plasticity, in line with Hebbian learning principles. We suggest that these principles apply across domains, with neuroplastic changes best induced when neurostimulation is paired with an actively engaged brain state. In our study, cognitive restructuring (CR) may act as a catalyst, activating prefrontal circuits during stimulation to improve outcomes. Ignoring the brain’s initial state limits the generalizability of findings to the exact conditions in which neurostimulation was applied [27].

Study results should be considered in light of limitations. The sample size was small and lacked a comparison with a non-clinical control group. Additionally, HF and LF rTMS paradigms were applied at different intensities (120% and 90%, respectively) and at different stimulation frequencies (10 Hz vs. 1 Hz) according to existing protocols at the time of study design. Nevertheless, this inhomogeneity in stimulation intensities and frequencies may also have contributed to the differential findings, and future studies should replicate our findings with matched intensities and frequencies. As the first author was solely responsible for designing and administering the behavioral intervention, the fidelity of delivery was not assessed, which represents an additional limitation. Post-hoc comparisons following significant omnibus effects were not corrected for multiple comparisons. Replication is needed to confirm our results.

Additional research to guide how brain context manipulation can influence outcomes can have important implications for clinical practice. For example, based on our findings, it may be indicated that clinical participants engage in a congruent task during neurostimulation. For example, if undergoing an FDA-approved depression protocol [20], participants may benefit less from listening to music (an incongruent task), and may benefit more if engaging in a regulation task, such as cognitive reappraisal. These skills have been shown to modulate the dlPFC-anterior cingulate cortex circuitry [96] considered critical in major depressive disorder [108]. The specific skill would need to be taught in advance by a trained provider, but could be implemented by a trained technician in the room, as we demonstrate in our study combining self-systems therapy with neuromodulation [47], or without human guidance, by following pre-established prompts on a screen [87]. Cognitive restructuring is frequently taught in Cognitive Behavioral Therapy [109], and, therefore, many participants may already be familiar with this skill and could easily apply it in neuromodulation sessions. Per our findings, such an adjustment to clinical practice would improve rTMS efficacy, though this remains a hypothesis needing empirical examination. Beyond depression, for any clinical application, our research encourages clinicians to understand which specific circuitry their protocol targets and to identify a congruent behavioral task that engages the same circuitry during stimulation to improve outcomes.

In conclusion, we propose a theoretical model for optimizing neuromodulation by altering the neural context before, during, or after stimulation. We examine one of the hypotheses derived from this model and present preliminary evidence that combining a psychotherapeutic behavioral skills practice with a neurostimulation intervention yields the largest improvement in self-reported behavioral outcomes. This occurs when the stimulated circuit aligns with the neural circuit activated by the behavioral skill within the context of targeting transdiagnostic emotion dysregulation, a feature of many psychiatric disorders. Future studies with adequate statistical power are needed to generalize this framework to other psychiatric symptoms and to clarify how self-reported effects can be extended to more objective, physiological indices of distress. Our proposed theoretical model (see Fig. 1) generates additional hypotheses that may enhance the effects of neuromodulation. For instance, we anticipate improved outcomes when preparatory tasks modify the target circuit just prior to stimulation, thereby biasing plasticity toward the intended populations. Similarly, engaging the brain in the acquisition of skillful behavior immediately after neurostimulation of a relevant circuit may lead to better clinical outcomes. Although these hypotheses were not tested in the current proof-of-concept study, future research can develop translational protocols to explore how to optimally conceptualize brain states before, during, and after neuromodulation to maximize benefits.