A Phase-Optimized Closed-Loop TMS-EEG Framework Guided by fMRI Drug Cue Reactivity for Stimulant Use Disorder
DOI: Pending DOI
Synopsis
This study demonstrates the feasibility of fMRI-guided, phase-specific closed-loop TMS-EEG for targeting addiction-related brain circuits in psychostimulant use disorder. This framework enables multi-level personalization of TMS parameters, including stimulation target, frequency, phase, and intensity, and represents a promising step toward precision neuromodulation approaches for substance use disorders.
Background
Psychostimulant use disorders (pSUD), including cocaine, methamphetamine, and amphetamine, contribute substantially to rising overdose deaths and significant physical and mental health morbidity. Currently, there are no FDA-approved pharmacotherapies for its treatment, highlighting the need for novel therapeutic approaches. pSUD is associated with dysfunction in the mesolimbic dopamine system, a network involving frontal, limbic, and subcortical regions that support reward processing, motivation, and salience attribution. Key nodes in this circuitry include the ventral striatum, insula, and amygdala. Transcranial magnetic stimulation (TMS) is a promising noninvasive neuromodulation technique capable of modulating these circuits through cortico-subcortical connections. However, treatment response to TMS in substance use disorders shows substantial inter-individual variability. Individualized stimulation approaches may help reduce this variability by optimizing stimulation parameters for each individual. By accounting for brain states via electroencephalography (EEG), closed-loop TMS-EEG can reduce the variability of TMS effects. Prefrontal theta oscillations are a promising target, as these rhythms are linked to treatment-relevant functions such as cognitive flexibility. Therefore, we aimed to test the effects of repetitive TMS synchronized to prefrontal theta oscillations over individualized dorsolateral prefrontal cortex (DLPFC) targets on cognitive flexibility and DLPFC engagement in pSUD.
Methods
To date, seven participants with pSUD in the abstinent phase have completed the two experimental sessions of this ongoing study. In session 1, participants underwent baseline assessments followed by MRI acquisition, including rest, task-based fMRI, and structural scans. During a 7-minute task-based fMRI, participants completed a standard drug cue reactivity task consisting of blocks of drug-related and neutral cues, after which craving was rated (0–4). Task-based connectivity analysis identified individualized left DLPFC targets showing maximal robust connectivity with the amygdala. In session 2, participants were randomly assigned to receive 600 pulses of intermittent theta burst stimulation (iTBS) with individualized theta frequencies, over fMRI-guided targets, delivered at the peak or the trough phase of the ongoing EEG theta oscillations. In a control group, participants received conventional, open-loop (random phase) iTBS. Three minutes of resting-state EEG were recorded to estimate each participant’s dominant theta frequency and to train a Bayesian Temporal Prediction algorithm for real-time prediction of theta phase. Stimulation intensity was set to 120% of the resting motor threshold (RMT). Neural target engagement was quantified using TMS-evoked potentials (TEPs) recorded before and after stimulation using 270 single TMS pulses delivered at the peak, trough, or a random theta phase (90 each). Changes in cognitive flexibility were assessed using the accuracy and reaction time in a cognitive shift task administered before and after iTBS.
Results
All participants in this initial sample were male (age range 44 6.50 years) with no significant differences in demographic data between peak (n=2), trough (n=3), and open-loop (n=2) iTBS groups. Inside the scanner, craving ratings were higher following drug cues (2.33 ± 0.80, range 1.25–3.50) compared to neutral cues (1.23 ± 0.38, range 1.00–2.00). The seven stimulation targets defined using task-based connectivity deviated from the canonical F3 coordinate (MNI: −38, 44, 26) with a mean Euclidean distance of 24.6 ± 16.6 mm. The average offsets relative to F3 were Δx = 5.0 ± 10.8 mm, Δy = −12.0 ± 16.9 mm, and Δz = 2.5 ± 19.1 mm. Preliminary analyses indicate that the trough group descriptively showed the clearest improvement from pre- to post-session. Overall accuracy increased from 0.577 ± 0.043 to 0.682 ± 0.048 (Δ ≈ +0.105), while trials to criterion (how many trials a participant needs before they consistently respond correctly) decreased from 10.16 ± 0.50 to 8.49 ± 1.46, indicating faster rule acquisition. In contrast, the peak group showed minimal change in overall accuracy and required more trials to reach the criterion. There were no clear trends in the open-loop group. Regarding TEP amplitudes, single TMS pulses delivered at theta peaks tended to evoke the largest TEPs before iTBS (9.4 ± 6.2 µV). After iTBS there was a notable reduction of TEP amplitudes in the open-loop group (pre: 6.9 ± 1.1 µV, post: 3.7 ± 0.4 µV).
Discussion
Our findings demonstrate the feasibility of closed-loop iTBS with individualized stimulation parameters in pSUD. Despite the small sample size, preliminary results suggest phase-dependent differences in both neural and behavioral outcomes. These results highlight the potential of personalized neuromodulation approaches for pSUD and support further investigation of phase-specific brain stimulation strategies in substance use disorders.
Disclosures
This work is supported by the Medical Discovery Team on Addiction (MDTA) at the University of Minnesota