If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
The reward circuit is important for motivation and learning, and dysregulations of the reward circuit are prominent in anhedonic depression. Noninvasive interventions that can selectively target the reward circuit may hold promise for the treatment of anhedonia.
We tested a novel transcranial magnetic stimulation intervention for modulating the reward circuit. A total of 35 healthy individuals participated in a crossover controlled study targeting the reward circuit or a control site with intermittent theta burst stimulation (iTBS), an excitatory form of transcranial magnetic stimulation. Individual reward circuit targets were defined based upon functional magnetic resonance imaging functional connectivity with the ventral striatum, yielding targets in the rostromedial prefrontal cortex (rmPFC). Reward circuit function was assessed at baseline using functional magnetic resonance imaging, and reward circuit modulation was assessed using an event-related potential referred to as the reward positivity, which has been shown to reliably track reward sensitivity, as well as individual differences in depression and risk for depression.
Relative to control iTBS, rmPFC iTBS enhanced the reward positivity. This effect was moderated by reward function, suggesting greater enhancements in individuals with lower reward function. This effect was also moderated by rmPFC–ventral striatum connectivity insofar as iTBS reached the rmPFC, suggesting that efficacy relies jointly on the strength of the rmPFC–ventral striatum pathway and ability of transcranial magnetic stimulation to target the rmPFC.
These data suggest that the reward circuit can be modulated by rmPFC iTBS, and amenability to such modulations is related to measures of reward circuit function. This provides the first step toward a novel noninvasive treatment of disorders of the reward circuit.
). Correspondingly, depression has been linked to reduced activation of the brain’s mesolimbic reward system with functional magnetic resonance imaging (fMRI) meta-analyses converging on the ventral striatum (VS) as a consistent locus of hypoactivation (
). Normalizing such hypoactivity may therefore hold promise in the treatment of anhedonic depression. Progress toward this end requires 1) measures that reliably track reward circuit function and 2) interventions that modulate these measures.
Toward the former, reward dysfunction in depression has been documented using the reward positivity (RewP), a relative positivity in the event-related potential following rewards compared with nonrewards (
). Collectively, these data suggest that the RewP tracks reward circuit (dys)function and could be a novel target for intervention and prevention efforts.
Transcranial magnetic stimulation (TMS) is a promising tool for modulating brain function. However, reward structures (e.g., the VS) cannot be targeted directly by TMS owing to their deep location. Nevertheless, a growing body of research has demonstrated that TMS to cortical areas can indirectly modulate neuroanatomically connected subcortical areas (
), TMS to the rmPFC may modulate the VS and reward circuit function (Supplement).
Here, we tested the hypothesis that TMS to the rmPFC modulates reward circuit function as reflected by the RewP. We used intermittent theta burst stimulation (iTBS), an excitatory form of TMS designed to upregulate neural activity (
). Relative to iTBS delivered to a control target, we hypothesized that rmPFC iTBS targeting the reward circuit would increase the RewP (Figure 1). Moreover, we examined whether rmPFC iTBS modulation of the RewP is moderated by reward circuit function and rmPFC-VS connectivity. We predicted that individuals with reward circuit hypoactivity may benefit the most from rmPFC iTBS and that the strength of rmPFC-VS connectivity may enhance propagation of TMS signals through the reward circuit, resulting in greater rmPFC iTBS modulations of the RewP.
Methods and Materials
A total of 35 right-handed participants (age 18–26 years; mean = 20 years; n = 24 female) without contraindications for fMRI and TMS and no self-reported history of neurologic or psychiatric disorders were recruited from the Tallahassee, Florida, area. Planned recruitment (N = 40) was truncated by COVID-19. Exclusions (Supplement) left 24 participants for iTBS/electroencephalography (EEG) analyses, 30 for fMRI analyses, and 23 for analyses combining iTBS/EEG and fMRI.
Participants performed three sessions. A baseline session included MRI scanning and TMS motor thresholding. iTBS/EEG sessions followed within 10 days and were spaced 1 week apart from one another (8 days for 1 participant). Each iTBS session consisted of iTBS to either the rmPFC or control target (see below) followed immediately by the Doors task with EEG recording. The order of stimulation targets was randomly counterbalanced across participants. This study was approved by the Institutional Review Board at Florida State University (IRB Nos. STUDY00000532, STUDY00000533, and STUDY00000973).
The Doors task (Figure 2A) was designed to engage the reward circuit (
). On each trial, two identical doors were presented on a computer screen. Participants were told that they would either win or lose money on each trial and to guess which door would result in monetary gain. Because monetary losses are experienced as twice as valuable as monetary gains (
), participants were told that they could either win $0.50 or lose $0.25 on each trial and that they would get to keep their winnings. The task consists of 30 gain trials and 30 loss trials, presented in pseudorandom order such that participant choices had no actual bearing upon outcomes. The task was administered using Presentation (Neurobehavioral Systems, Inc.) for the EEG version and E-Prime (Psychology Software Tools) for the fMRI version.
For the EEG version, participants were shown a fixation cross (500 ms) followed by an image of two doors, which remained until participants made a selection by clicking either the left or right mouse button. After another fixation cross (interstimulus interval: 1000 ms), feedback indicating monetary gain (green upward arrow) or loss (red downward arrow) was presented (2000 ms). Finally, a fixation cross (intertrial interval) was presented for 1500 ms, followed by the message “Click For Next Round,” which remained until the participant clicked a mouse button. The fMRI version had different timing to account for hemodynamic lag and used finger presses for responding: doors were presented for a fixed time of 4000 ms, the interstimulus interval was uniformly pseudorandomized between 2000 and 5000 ms in 1000-ms increments, and the intertrial interval was pseudorandomized between 1500 and 9000 ms in 2500-ms increments in a weighted fashion (24 trials at 1500 ms, 16 trials at 4000 ms, 12 trials at 6500 ms, 8 trials at 9000 ms).
Prior to performing the Doors task in the scanner, participants completed a 9-minute eyes open resting-state scan while maintaining fixation.
EEG was recorded using an elastic cap with 10 actiCAP slim electrodes positioned in accordance with the 10/20 system using a LiveAmp amplifier (Brain Products GmbH). Electrode FCz served as an online recording reference, with a ground electrode placed at FPz. Two electrodes were placed on the left (TP9) and right (TP10) mastoids. Electro-oculogram was recorded using four electrodes: two placed approximately 1 cm above and below the left eye and two at the outer canthi of both eyes. The remaining two electrodes were placed at Cz and Pz. The EEG signal was digitized at 500 Hz. Impedance was kept below 25 kΩ.
EEG data were analyzed using BrainVision Analyzer, version 2.1 (Brain Products GmbH). Data were re-referenced offline to the average of the left and right mastoids and bandpass filtered (0.1–30 Hz). Data were segmented into feedback-locked epochs from −200 to 1000 ms, with −200 to 0 ms serving as the baseline. Ocular artifacts were corrected (
). Epochs containing a voltage step >50 μV between consecutive sample points, a 175-μV change within a 400-ms interval, or a change of <0.5 μV within a 100-ms interval were automatically rejected. Feedback event-related potentials were then averaged separately for win and loss trials.
The RewP was initially assessed by contrasting the average signal between 250 and 350 ms at FCz following win and loss feedback as we have done previously (
). This measure of the RewP was made separately for each visit and was used in subsequent analyses.
MRI data were acquired on a Siemens Prisma 3T scanner equipped with a 32-channel head coil. A high-resolution T1 structural MRI (magnetization-prepared rapid acquisition gradient-echo; 384 × 384 × 256 matrix of 0.67 × 0.67 × 0.7 mm3 voxels; repetition time = 1.84 seconds; echo time = 2.9 ms; flip angle = 9) was collected along with a T2 (256 × 256 × 192 matrix of 0.9 mm3 isotropic voxels; repetition time = 3.2 seconds; echo time = 408 ms). T2∗-weighted echo-planar imaging images were collected during the Doors task and resting state described below (84 × 84 × 54 matrix of 2.5 mm3 isotropic voxels; repetition time = 2 seconds; multiband = 2; echo time = 29.2 ms; flip angle = 45).
Univariate Image Analysis
SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) was used for preprocessing and analysis unless specified otherwise. Images were converted from DICOM into NIfTI format. Origins for all images were manually set to the anterior commissure. Functional data were spike corrected using AFNI’s 3dDespike (http://afni.nimh.nih.gov/afni). Functional images were corrected for differences in slice timing using sinc-interpolation and head movement using a least squares approach and a six-parameter rigid body spatial transformation. Structural data were coregistered to the functional data and segmented into gray and white matter probability maps (
). Segmented images were used to calculate spatial normalization parameters to the Montreal Neurological Institute template, which were applied to the functional data. As part of spatial normalization, the data were resampled to 2 × 2 × 2 mm3. Functional images were isotropically Gaussian smoothed at 8-mm full width at half maximum. All analyses included a temporal high-pass filter (128 seconds) and an autoregressive AR(1) model to correct temporal autocorrelation, and images were runwise scaled to a global mean intensity of 100.
Subject-level models were fit with a general linear model including separate regressors capturing win feedback, loss feedback, left responses, right responses, and visual presentation of the doors. Each model also included linear motion parameters as well as framewise displacement to capture motion-related signals.
A win minus loss contrast was performed at the subject level and submitted to a second-level one-sample t test thresholded with a p < .001 height and 402 voxel extent to achieve familywise error cluster-corrected p < .05. To produce covariates for correlation and moderation analyses, 6-mm spherical regions of interest were centered at peaks of activation in the dataset in the left (−12, 6, −6) and right (16, 8, −8) VS and medial PFC (−10, 52, 0). Region of interest activations were used to correlate with the statistically independent RewP. The blood oxygen level–dependent signal is affected by magnetic field inhomogeneities and head motion. To remove these artifactual contributions, for each region of interest, temporal signal-to-noise ratio (mean divided by standard deviation) and head motion [root mean square of framewise displacement (
), we modified preprocessing to calculate connectivity: ANTs toolbox (http://stnava.github.io/ANTs/) was used for T1 N4 bias field correction and spatial normalization to the Montreal Neurological Institute template. The first four volumes of each functional scan were discarded to ensure MRI gradient stabilization. Montreal Neurological Institute tissue masks were generated for noise correction using fslmaths (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL). Intensity normalization was conducted on brain-masked functional data, followed by linear detrending of the functional images using the REST toolkit (https://www.nitrc.org/projects/rest/). We applied 6-mm full width at half maximum isotropic Gaussian smoothing using SPM to all functional images, and frequencies below 0.008 Hz and above 0.08 Hz were filtered. ICA-AROMA (
) to match the resolution of our preprocessed functional data. Voxelwise correlation maps with the VS were computed for each subject, transformed with an arc-hyperbolic tangent function, and submitted to a second-level one-sample t test. Significance was determined using voxel-level familywise error correction (p < .05) using SPM12. The rmPFC search space was defined by a 15-mm sphere around the rostral-most peak of connectivity with the VS (2, 56, −4). For each subject in this dataset, VS connectivity was similarly calculated, and connectivity maps and the rmPFC search space were inverse warped into native space. Individual connectivity maps were defined on either the rest (n = 11), task (n = 1), or combined (n = 12) data, wherein choice of dataset was determined by which connectivity map produced the greatest spatial similarity to the independent group-averaged map. An iTBS target was defined within the rmPFC search space using a manual procedure balancing maximal connectivity strength, minimal distance to scalp, and sufficient distance from the orbits so as to minimize discomfort.
To quantify individual connectivity with the VS, more rigorous procedures were used to facilitate merging of rest and task data. Systematic biases induced by task activations were removed from the task data using procedures described by Cole et al. (
). ICA-AROMA was then used to remove signal artifacts and roughly equate the variance of the rest and task data (Supplement). Rest and task data were then merged prior to computing functional connectivity. Group-averaged functional connectivity is depicted in Figure 1A.
A control iTBS target was localized to the hand-knob of the primary motor cortex (n = 15) or the dorsomedial-most tip of primary somatosensory cortex (n = 9) based upon visual inspection of the T1 image (Supplement). Control targets were matched to rmPFC targets in hemisphere. Targets are depicted in Figure 1B.
Simulated electrical fields (e-fields) were calculated using SimNIBS 2.1 software (
) (Figure 1C). E-field modeling indicated variability in how effectively iTBS reached our rmPFC targets. Therefore, we distinguished between the ideal target and the stimulated target. The ideal target was defined as the voxel within the rmPFC search space showing maximal connectivity with the VS for a given individual. The stimulated target was defined as the voxel within our rmPFC search space showing the maximum product of connectivity with the VS and norm E for a given individual. That is, the stimulated target weighs both connectivity and stimulation intensity. Individual connectivity averaged within a 6-mm sphere around the stimulated target was arc-hyperbolic tangent normalized and used for moderation analyses, along with the Euclidean distance between the stimulated target and ideal target.
TMS was delivered using a MagPro X100 stimulator equipped with a figure-of-eight MCF-B65 coil. Electromyography was recorded on the left dorsal interosseous muscle using a Delsys Trigno system and EMGworks Acquisition software (Delsys Inc.). A hunting procedure was used to determine the scalp location in the right hemisphere producing the maximal contralateral hand twitch at the minimal stimulation intensity. Next, the participant maintained voluntary contraction of the first dorsal interosseous muscle at approximately 20% of maximum contraction. Active motor threshold was determined as the minimal stimulation intensity needed to produce a motor evoked potential >50 μV in 5 of 10 pulses.
), iTBS was delivered in three pulse bursts at 50 Hz repeated at 5 Hz in 2-second trains repeated every 10 seconds at 80% of active motor threshold for a total of 600 pulses over 190 seconds. iTBS was guided using Localite hardware and software (Localite GmbH). For control iTBS, the coil was oriented perpendicular to the target gyrus. For rmPFC iTBS, the coil handle was oriented vertically toward the top of the participant’s head to minimize stimulation around the orbits. Prior to the start of iTBS, a single test pulse was delivered to the target at the same intensity of iTBS to acclimate the participant.
A contrast of win-loss in the fMRI data revealed significant reward-related activations in the bilateral VS and medial PFC (Figure 2B) consistent with previous reports (
). Correspondingly, following control iTBS, a robust RewP was observed such that more positive potentials were observed following wins relative to losses starting approximately 200 ms after feedback (mean difference between 250 and 350 ms = 3.4 μV; t23 = 3.31, p = .003) (Figure 2C). These data confirm the expected reward-related fMRI and EEG signals.
Previous data have indicated a relationship between VS fMRI activation and the RewP (
), suggesting that they reflect a common reward signal. Activation was also observed in the vicinity of our TMS targets, so we also examined the correspondence between medial PFC activation and the RewP (Supplement). Both the left (r = 0.32, p = .13) and right (r = 0.29, p = .18) VS showed a nonsignificant positive correlation with the RewP (Figure 3). While not significant, effect sizes were similar to prior data (
) (r = 0.30–0.34). However, reward-related activation in the medial PFC showed an opposite nonsignificant trend (r = −0.16, p = .46), which was significant when using robust regression to reduce the impact of an outlier (t21 = −2.43, p = .02). This negative relationship was striking given that medial PFC activations were positively correlated with the VS (left VS: r = 0.58, p = .003; right VS: r = 0.55, p = .007), suggesting that the RewP captures a signal more specifically related to the VS. After partialling out medial PFC activations, VS activation (averaged across hemispheres) was significantly positively correlated with the RewP (r = 0.51, p = .01). Robust regression verified that the positive relationship between VS activation and the RewP was not due to outliers (t20 = 2.23, p = .04). These data suggest that the VS and RewP reflect a common reward signal, consistent with prior data (
Next, we examined whether reward circuit–targeted iTBS modulates reward signaling. Following rmPFC iTBS, a significant RewP was observed (mean difference between 250 and 350 ms = 5.45 μV; t23 = 5.61, p < .00005). A mixed-effects analysis of variance with factors of feedback (win, loss), iTBS target (rmPFC, control), and iTBS order (rmPFC → control, control → rmPFC) revealed a trend toward a feedback × iTBS target interaction (F1,22 = 3.87, p = .062) (Figure 4). This trend was driven by numerically larger responses to win feedback following rmPFC iTBS relative to control iTBS (mean difference = 2.9 μV; t23 = 2.03, p = .054) with no corresponding difference to loss feedback (mean difference = 0.1 μV; t23 = 0.09, p > .9). These data are suggestive of a selective effect of rmPFC iTBS to reward, but not loss signals. However, the lack of statistical significance suggests that the effect is not uniformly observed.
We tested two moderations of the rmPFC iTBS effect on reward processing. First, we examined whether those individuals with less reward signaling would show greater reward upregulation as a function of rmPFC iTBS. To do so, we created a latent variable reflecting reward signaling by combining left VS, right VS, and medial PFC reward-related activations (win-loss) and the RewP following control iTBS using principal component analysis (Supplement). We then correlated reward signaling with the change in RewP as a function of rmPFC iTBS (rmPFC iTBS > control iTBS; ΔRewP). Consistent with the hypothesis, a significant linear negative relationship was observed (r = −0.46, p = .03) (Figure 5A), which held when using robust regression to reduce impact of potential outliers (t21 = −2.61, p = .02). Similar results were obtained when using fMRI activations alone as a moderator (Supplement). While those individuals with high reward signals (latent variable > 0) did not show an increase in the RewP following rmPFC iTBS (mean ΔRewP = −0.64 μV, t8 = −0.23, p = .81), those with low reward signaling (latent variable < 0) did (mean ΔRewP = 4.01 μV, t13 = 3.14, p = .008) (Figure 5B). These data suggest that those individuals with low reward signaling are those who are specifically amenable to rmPFC iTBS.
Second, we tested the hypothesis that the effect of rmPFC iTBS is moderated by rmPFC-VS connectivity. Stronger rmPFC-VS connectivity should reflect a stronger pathway by which TMS can affect the reward circuit. However, this is qualified by the effectiveness with which TMS affects the rmPFC itself. To incorporate these two factors, we identified the stimulated target by weighing both the connectivity of the rmPFC to the VS and the e-field generated in the rmPFC (see Methods and Materials). The stimulated target differed from the ideal target—the rmPFC voxel showing maximal connectivity to the VS (mean distance = 16.7 mm, range = 0.66–31.3 mm). Although rmPFC-VS connectivity at the stimulated target was nonsignificantly positively related to the ΔRewP (r = 0.35, p = .11; robust regression t21 = 1.62, p = .12) (Figure 6A), this was qualified by a significant inverse relationship of the distance between the stimulated target and ideal target and the ΔRewP (r = −0.49, p = .02; robust regression t21 = −2.42, p = .02) (Figure 6B). This suggests that the effectiveness of rmPFC iTBS is related to the efficacy with which TMS can be directed to the ideal target for an individual. A component reflecting the shared variance among these measures (Supplement) was significantly correlated to the ΔRewP (r = 0.47, p = .02; robust regression t21 = 2.29, p = .03) (Figure 6C). These data suggest that effectiveness of rmPFC iTBS to increase the RewP is moderated jointly by the connectivity of the rmPFC to the VS and the efficacy with which TMS can reach the appropriate rmPFC target.
We tested a novel intervention targeting the reward circuit to modulate reward processing. By delivering iTBS to an rmPFC area showing functional connectivity to the VS, a critical node in the reward circuit, we found evidence of modulated reward processing reflected by increased RewP relative to control iTBS. This effect was selective to reward rather than loss feedback (Supplement) and was moderated by reward signaling. The increase in RewP induced by rmPFC iTBS was also related to rmPFC-VS connectivity insofar as TMS could reach the rmPFC. Collectively, these data provide evidence that rmPFC iTBS modulates reward processing, particularly in those individuals with low reward signaling and strong VS connectivity in areas that can be reached by TMS.
We tracked changes in reward processing using the RewP. Consistent with data reported here, a larger RewP has previously been related to increased structural (
), low cost, and ease of obtainment make it a desirable general index of reward circuit function and target for interventions. The finding that rmPFC iTBS increases the RewP therefore holds promise for clinical interventions targeting the reward circuit.
TMS has been used for the treatment of drug-resistant depression. The most common treatment targets the dorsolateral PFC (
). An interesting future avenue would be to directly compare rmPFC and dorsolateral PFC TMS and contrast their effects on brain circuits and symptoms (see the Supplement).
Various limitations should be considered. First, the sample was truncated owing to COVID-19. Future studies using larger samples will provide more definitive conclusions regarding the efficacy and effect sizes of rmPFC iTBS. Second, we sampled healthy individuals who were unselected for increased anhedonia. The extent to which the effects extend to psychiatric populations and the effects of rmPFC iTBS on anhedonia itself remain to be elucidated. However, the fact that those individuals with the lowest reward-related activations showed the greatest responsivity to rmPFC iTBS provides some promise for populations characterized by reward circuit hypofunction. Third, although we individualized rmPFC targeting based upon connectivity, our data indicate that both connectivity and e-fields dictate the effectiveness of rmPFC iTBS. While we anticipated the importance of individual connectivity based upon past work (
), the incorporation of e-fields was done post hoc. Our data suggest that incorporating e-field modeling to select targets may facilitate the effectiveness of interventions. However, the post hoc nature of the finding merits replication before reaching strong conclusions. Finally, our experimental and control targets likely differed in sensation. Discomfort from TMS is known to increase at more ventral areas relative to dorsal areas (
). Notably, the fact that the efficacy of rmPFC iTBS varied as a function of reward circuit factors (i.e., reward sensitivity, connectivity) suggests that the effect owes to the targeted network rather than reflecting a difference between TMS targets. Nevertheless, better control for discomfort would be prudent for future investigations.
We have provided evidence that a novel TMS intervention targeting the reward circuit modulates reward processing. This intervention may hold promise for the treatment of dysfunctions of the reward circuit.
Acknowledgments and Disclosures
Portions of this study were funded by Florida State University Team Science for Translational Research Seed Grant, funded by the FSU Office of Research (DEN, GH).
Part of the results of the study were previously presented at the Southeast Regional Clinical & Translational Science Conference, March 2021, Emory University, Atlanta, Georgia, and Annual Meeting of the Society for Neuroscience, November 2021, Chicago, Illinois.
The authors report no biomedical financial interests or potential conflicts of interest.