Introduction

Substance use disorders (SUDs) are a growing public health concern of epidemic proportion, resulting in more than 600,000 deaths in the United States annually, with over 100,000 annual deaths in the United States from opioid use disorders (OUD) alone1. The greatest challenge to treatment success is preventing relapse, and relapse risk has been attributed at least in part to neural adaptations in the frontostriatal circuitry involving the nucleus accumbens region (NAc), an important component of the mesolimbic dopamine system2. Functional neuroimaging assays utilizing cue-reactivity paradigms to probe neural activity underlying drug desire (“craving”)3,4, compulsive drug seeking and drug-use behaviors have revealed hyperreactive responses to drug-related cues. These phenomena are associated with poorer treatment outcomes and increased relapse rates5,6. However, a well-recognized challenge with functional MRI is its limited spatial and temporal resolution, which precludes exquisite detailing of specific anatomophysiological involvement of the NAc region. Deep brain stimulation (DBS) is a long-term neuromodulatory treatment with the potential to rescue frontostriatal adaptations and reverse behavioral over-response to drug cues in severe cases7. However, the best way to engage the NAc network with DBS in OUD remains largely unknown8.

In this case study, we seized a serendipitous clinical opportunity to synchronize a personalized cue-reactive task to intracranial electroencephalographic (iEEG) recordings via an in-situ NAc DBS electrode. This afforded high spatial and temporal resolution to decipher the involvement of the NAc in drug-related cravings and behavior4,9. We report a proof-of-principle use of a specific drug-cue-reactive electrophysiological biomarker, alongside anatomical and behavioral correlates, to optimize DBS programming in a patient with OUD. Our work heralds the potential for personalizing DBS for OUD using an objective, physiologically grounded marker of drug-cue reactivity.

Results

The patient is a 25-year-old male with an existing NAc DBS to treat his severe OUD off-label (see Online Methods; Figs. 1a & S1a-c). DBS therapy improved his OUD and substance cravings, allowing him to abstain from drug use for several years, but required a prohibitively high dose of stimulation (14.5 mA) in a monopolar configuration that diffusely engaged a broad region including the ventral anterior limb of the internal capsule (vALIC) and adjacent dorsal NAc region (Fig. 1a). He suffered a wound erosion over the left subclavicular incision that warranted removal of the implantable pulse generator (IPG), which was followed by rapid recurrence of opioid cravings and drug-related thoughts (Fig. 1a). Due to suspicion of ongoing infection, the DBS system was externalized with an extension cable while cultures and antibiotic treatment ensued (Online Methods).

Fig. 1: A specific 1–6 Hz low-frequency electrophysiological biomarker can be identified with a personalized cue-reactivity paradigm task.
Fig. 1: A specific 1–6 Hz low-frequency electrophysiological biomarker can be identified with a personalized cue-reactivity paradigm task.
Full size image

a Timeline of events leading to subject presentation and evaluation for DBS replacement and optimization. b Schematic of personalized drug-cue-reactivity paradigm task based on pre-task interview of the patient’s drug use habits. In stimulation trials (right), 10 min of stimulation was delivered prior to start of the task. Between each cue, ~15 s intermittent stimulation bout (*) was delivered, followed by a brief pause during which stimulation was turned off to eliminate recording artifacts when the cue was presented to allow artefact-free iEEG recording. c A specific, low-frequency (1–6 Hz) biomarker associated with drug cravings was identified in the presumed nucleus accumbens (NAc) shell subregion. Left panel; Coronal (top) and sagittal (bottom) T1 MRI views of DBS electrode contacts in the NAc shell. Yellow dotted circle demarcates contacts 1-2 where bipolar recordings were obtained. Middle panels; Time-frequency analysis (−1 to 10 s; video starts at 0 s and ends at 6 s) showed a significant dissociation in power in the low-frequency range when drug-related cues were presented (Top), as compared to neutral cues (Bottom). Right panel; Cluster-based permutation test result of the spectral differences between drug and neutral cue trials. This was used to identify a spectro-temporal region of interest (ROI) for further analyses. Black box highlights the focus in the lower panel (Bottom), which contained the surviving cluster, showing a statistically significant increase in low frequency (1–6 Hz) power upon presentation of drug-related cues (p < 0.05, two-tailed permutation test). Similar analysis did not identify any significant difference in power in iEEG recorded in other anatomical regions. d Plot of power spectral density (PSD) differences in trials of drug vs. neutral cues showed a large dissociation in the low frequency range, only when measured in NAc shell, but not in other contact pairs, suggesting anatomical specificity. e Average power of 1–6 Hz frequency band across time, without stimulation (red) and with NAc shell stimulation (blue). Shaded areas represent the standard error of the mean (SEM) across trials within each condition. Red dotted box indicates the time window of interest when power of the 1–6 Hz signal was significantly reduced with stimulation relative to no-stimulation conditions (p = 0.032, two-tailed permutation test). This same ROI was used to extract trial-level power values for comparisons shown in (f). ‘Without stimulation’ trials were conducted while the DBS lead was recording only. ‘Stimulation’ trials were preceded by 10 min of continuous DBS (1-/2+ , 6 mA, 90 μs, 130 Hz) followed by brief (15 s) stimulation before each cue. f Violin plot of average 1–6 Hz band power measured in contacts 1-2 under four conditions (line denotes mean and white dot denotes median). Trial numbers: Drug cue (no stim): n = 7; Neutral cue (no stim): n = 10; Drug cue (stim): n = 12; Neutral cue (stim): n = 12. These values reflect the number of usable trials retained after artifact rejection. Power was significantly elevated when drug cues were presented (red) vs. neutral cues (pink) without stimulation (p = 0.029, two-tailed permutation test). With stimulation (1-/2+ , 6 mA, 90 μs, 130 Hz), 1–6 Hz power was significantly reduced when drug cues were presented (blue) (p = 0.032, two-tailed permutation test) and this was not statistically different from power in neutral cue condition (cyan). Stimulation of the NAc shell significantly reduced cravings levels when drug cues were presented (6.57 vs. 3.09, p = 6.16e-04, two-tailed permutation test), but did not affect ratings in neutral cues (dashed green line). Source data are provided as a Source Data file. AC = anterior commissure; DBS = deep brain stimulation; ED = emergency department; IPG = implantable pulse generator; MRI=magnetic resonance imaging; NAc=nucleus accumbens; OSH = outside hospital; OUD = opioid use disorder.

We hypothesized that prior high charge requirements were due to non-specific engagement of the NAc. To identify focal sites within the NAc region where cue-reactivity was most robust and specific to drug cues, three computer-based tasks were conducted for clinical purposes: a personalized drug-cue-reactivity task to assess for presence of drug-related cue-electroreactivity6 (Fig. 1b), the Monetary Incentive Delay, MID task10, and the Blink Suppression Task, BST11 (Online Methods). This battery of tasks allowed us to attempt to assess specificity of electrophysiologic drug-cue reactivity.

NAc low-frequency power is associated with high craving state

Presentation of personalized drug-related video cues evoked a robust elevation of low-frequency band (1–6 Hz) power in the iEEG recordings within the ventromedial-most region of the NAc, the presumed shell subregion12,13 (most ventral contacts 1-2) when compared to neutral cues (Fig. 1c). The signal was most prominent from 4.5 to 5.5 s post-cue onset, consistent with prior evidence that craving responses intensify with prolonged cue exposure3,14. This low-frequency power elevation was observed specifically during presentation of drug-related cues, in comparison to neutral cues, and was associated with an elevated craving state as assessed on a visual analog scale (VAS; range 1-9; task specific). It was also focally localized to the presumed NAc shell (Figs. 1d and S1d-e), consistent with anatomical specificity and known subregional physiology of the NAc15. Finally, this 1–6 Hz frequency power elevation was not elicited by non-drug incentive cues presented during the MID and BST tasks (Figs. S2 and S3). Rather, the MID task produced an elevation across a broader spectral power range (notably 3–21 Hz) during the anticipation window, and this was not anatomically specific to the NAc shell subregion (Fig. S2). The BST, which induces a strong urge to blink, showed no statistically significant low-frequency elevation during high-urge periods (Fig. S3). These well-validated tasks served as controls to assess possible domain specificity of the observed low-frequency signal. While they similarly engage reward anticipation and motor urgency, respectively, they differ from the cue-reactivity task in task design and phase of reward processing they probe. These differences limit direct comparability but reinforce the conclusion that the 1–6 Hz elevation observed during opioid cue exposure is at least not entirely attributable to general arousal, urge, or task engagement, and may represent drug-cue specific pathophysiology.

Stimulation modulates low-frequency power and lowers craving

Next, a systematic stimulation assessment was conducted across the DBS electrode contact pairs akin to standard clinical practice, but with additional focus on drug-related cravings in this subject (Table 1). The optimal stimulation site with the widest therapeutic window aligned with selection of the presumed NAc shell as the target, with reduction of cravings noted at the lowest dose required without adverse effects. Positive effects were also noted in more dorsal contacts in the vALIC (contacts 4 + /5-) where he was previously stimulated, although a higher current amplitude (8 mA) was required to elicit positive effects during the evaluation. Sham-controlled trials suggested the acute effects of stimulation observed were not placebo effects.

Table 1 Acute stimulation with real-time craving assessment
Full size table

We then investigated the effect of NAc shell bipolar stimulation during the drug-cue-reactivity task. As stimulation artifacts precluded monitoring iEEG signals simultaneous with stimulation, DBS was delivered for ten min (1-/2+ , 6 mA, 90 µs, 130 Hz) prior to administration of the task to establish a physiologically “stimulated” state. For each trial, an approximately 15 s stimulation bout was delivered before each cue presentation followed by a brief pause in stimulation during cue presentation to enable artifact-free iEEG recordings (Fig. 1b). This interdigitated approach allowed for the observation of stimulated effects without introducing simultaneous artifacts. Acute NAc shell stimulation attenuated the drug-cue evoked elevation of 1–6 Hz low-frequency band power towards a power suppression (p = 0.032, two-tailed permutation test), as seen in neutral cues, and this was associated with >50% reduction in mean subjective rating of opioid-related cravings (6.57 vs. 3.09, p = 6.16e-04, two-tailed permutation test) (Fig. 1e-f).

Biomarker-optimized stimulation sustained long-term effects

Based on these clinical and electrophysiological findings, we personalized the chronic DBS parameters (1-/2+ , 6 mA, 90 µs, 130 Hz) guided by convergent findings from behavioral testing and electrophysiological readouts evoked by the cue-reactivity task. The ventral NAc shell contacts exhibited both an electrophysiologic signal of craving and a robust acute behavioral and electrophysiological response to stimulation. This multi-modal approach enabled effective symptom control at substantially reduced current amplitude requirements compared to the original stimulation dose (14.5 mA). Six months following this personalization, the patient reported sustained reduction in drug cravings and continued abstinence from opioid use, meeting DSM-5 criteria for early remission (Fig. 2a). Notably, while the patient did not relapse to opioid use at any point, including during the DBS withdrawal phase, self-reported craving levels and inhibitory control improved progressively over the first several months following reactivation of DBS. These clinical improvements were specific to drug-related behavior, and did not affect his self-reported anxiety, mood and energy levels, as monitored via structured electronic questionnaires using VAS scales (range 0-10) during remote follow-up (Fig. S4). This progressive improvement suggests a possible cumulative or time-dependent effect of DBS on craving suppression and behavioral regulation (Fig. 2a). Such a trend may reflect ongoing plasticity within frontostriatal circuits or gradual behavioral integration of neuromodulatory effects. These observations underscore the value of longitudinal follow-up in future DBS studies to better characterize the trajectory and durability of therapeutic effects.

Fig. 2: Optimized NAc stimulation resulted in sustained clinical outcomes, with lower charge requirement.
Fig. 2: Optimized NAc stimulation resulted in sustained clinical outcomes, with lower charge requirement.
Full size image

a Longitudinal self-reported outcome measures following personalized DBS (1-/2+ , 6 mA, 90 µs, 130 Hz). Each bar represents a single standardized assessment (e.g., BSCS, OCDUS, craving VAS) at predefined follow-up visits. Different measures have different scoring ranges; see Online Methods section for scale descriptions. No error bars are shown, as values reflect single observations per time point. The subject remained abstinent from drug use throughout follow-up period. Source data are provided as a Source Data file. b–g Volume of tissue activation (VTA) modelling in green based on finite-element modeling (FEM) and structural connectivity analysis based on (b–d) new DBS parameters (1-/2+ , 6 m A, 90 μs, 130 Hz) and (e–g) original DBS parameters (C+/4-(60%)5-(40%), 14.5 mA, 70 μs, 130 Hz). Middle and right panels display coronal and sagittal views of structural connectivity of estimated VTAs to other brain regions. Blue streamlines are connectivity associated with NAc shell stimulation. Red streamlines are connectivity associated with ALIC stimulation. BSCS = Brief Substance Cravings Scale; DBS = deep brain stimulation; NAc = nucleus accumbens; OCDUS = Obsessive-Compulsive Drug Use Scale; VTA = volume of tissue activation. The notation “C+/4-(60%) 5-(40%)” indicates stimulation with the case as anode, and contacts 4 and 5 as cathodes, receiving 60% and 40% of the current, respectively.

Optimized NAc stimulation engaged SCG circuit and conserved charge

Structural connectivity analysis of the two stimulation sites revealed that the original stimulation setting had a much larger (25x) volume of tissue activation (VTA) (896 mm3 vs. 35.4 mm3), overlapping with the presumed NAc shell subregion, with additional, non-specific activation of other brain regions. Conversely, more apparently focal DBS delivery to the NAc shell appeared to engage a known circuit involving the subcallosal gyrus (SCG)12,13,16 (Fig. 2b-g). A basic modelling of charge requirement with this new stimulation paradigm predicted substantially improved battery lifespan (Energy Use Index 12.3 vs. 39.3) of the patient’s IPG, further emphasizing the value of precision targeting and personalization of DBS therapy. These results suggest that our personalized strategy delivered DBS with similar therapeutic efficacy, fewer off-target activations, and less device charge drainage.

Discussion

Here, we leveraged a rare clinical opportunity to optimize DBS therapy by understanding the electrophysiological underpinnings of opioid cravings. Patient-specific drug-cue-reactive low frequency (1–6 Hz) oscillatory activity localized to the presumed human NAc shell personalized a more sustainable DBS intervention. This serendipitous case establishes a foundation for developing individualized DBS approaches for OUD. While limited to a single patient, this study demonstrates proof-of-concept for a three-pronged framework combining neuroanatomical localization, cue-specific electrophysiological biomarker discovery, and acute craving responses to stimulation testing to guide DBS programming. This framework mirrors established paradigms in movement disorders DBS workflows17,18,19 and emerging applications in neuropsychiatric disorders20,21. It can be operationalized with temporary lead externalization or sensing-enabled IPGs in prospective studies. In this case, co-localizing the cue-evoked biomarker to the most stimulation-sensitive region of the NAc enabled a more focal and efficient stimulation strategy to be further evaluated in future studies.

There have been very few reports of iEEG recordings in individuals with OUD. In an intraoperative setting, prominent resting-state theta (4–8 Hz) and alpha (8–14 Hz) frequency activity of the anterior limb of the internal capsule (ALIC) and NAc were identified in patients with heroin addiction22. This was, however, not correlated to moment-to-moment craving states. Valencia-Alfonso et al reported an identification of a gamma-band (40–60 Hz) response in the right dorsal ALIC that was associated with drug-related stimuli in a stimulation-naïve patient with OUD, although this was not used to guide stimulation23. We present a paradigm pairing a personalized cue-reactivity task with NAc region iEEG recording, which enabled identification of an anatomically localized, drug-cue-specific electrophysiological biomarker associated with an elevated craving state, despite more than four years of abstinence. This discovery guided the re-direction and re-programming of DBS to the presumed NAc shell that was not initially selected through a standard monopolar assessment at the outside hospital. The convergence of this electrophysiological biomarker (i.e., power elevation across 1–6 Hz detected in the most ventral NAc contacts 1-2) with the most robust stimulation-induced attenuations in craving levels increased our confidence in selecting this alternative target along the same DBS lead. This NAc subregion is well-known to be involved in neural adaptations underlying drug-related behavioral sensitization24. The reprogramming resulted in restoration of effective suppression of drug cravings and sustained abstinence, all at a lower charge requirement. Similar electrophysiological approaches have been employed to successfully identify biomarkers of compulsion in subjects with obsessive-compulsive disorder and loss-of-control eating20,25,26,27. Notably, there appears to be shared low-frequency, electrophysiologic representation of reward-seeking behavior within the NAc transdiagnostically that appears to be conserved across species26,27,28,29.

The presence of a drug-cue-reactive electrophysiological response in the NAc is not unexpected, but the specificity of the identified biomarker to the ventral NAc (presumed shell subregion) presents an exciting therapeutic concept—that DBS would have relatively specific anti-craving effects despite the broader functionality of this region. The response to drug cues (1–6 Hz) was indeed quite focal in the NAc shell region, and emerged later in the cue-viewing window. This is in line with prior reports that craving intensifies with prolonged cue exposure3,14. In contrast, the MID task elicited an earlier, broader-band power increase (3–21 Hz) during reward anticipation that was more diffusely distributed. Although these tasks are not directly comparable, the convergence of frequency band that is spatially localized with relevant behavioral correlation suggests that the 1–6 Hz evocation represents a drug-cue-specific electrophysiological process. This specificity of effect further suggests that DBS effects on craving may also be relatively specific rather than a general reward-blunting effect, even though overlapping reward circuitry may be engaged in both.

The NAc is a central node in the mesocorticostriatal circuitry and is involved in all neurobiological stages of addiction30. Effects of disrupting the NAc activity with DBS have been demonstrated in rodent models of relevant endophenotypes of SUD31,32,33,34 and subsequently translated to off-label clinical application in humans7. Notably, the updated stimulation contacts were located in the ventral-most region of the NAc, consistent with the shell subregion, which has been strongly implicated in behaviors related to SUD in preclinical studies and corroborated by structural and functional neuroimaging work in humans12,13,35. Rodent work has demonstrated that modulation of the NAc shell (rather than the core) can attenuate drug-seeking, reinstatement, and cue-induced behaviors31,32,36,37,38. These findings support the translational relevance of our findings. Of course, it is important to recognize that the spatial resolution of electrical stimulation and functional imaging may not entirely resolve the NAc subregions with high confidence, speaking to the importance of electrophysiological assays such as reported here to decipher such functionally distinct nodes.

A recent systematic review reported that the majority of 26 human studies (71 subjects) targeted the NAc for the treatment of various SUDs7. While efforts have been made to improve targeting with advanced imaging techniques39,40, none of the studies reported subregion specificity, nor routinely utilized a biomarker-driven approach or cue-reactivity task to optimize DBS therapy. DBS for SUD remains investigational and reported outcomes are highly variable, emphasizing the need for objective data to both guide programming and to improve understanding of human addiction neurobiology8. We demonstrated the use of electrophysiological target engagement in a patient with OUD to optimize DBS therapy. Our case study also adds significant support to the potential feasibility of personalizing future OUD DBS therapy using craving-related electrographic biomarkers recorded in the NAc region to improve outcomes. While we could not determine if DBS suppressed the cue-reactive electrophysiologic signal over time due to the limitations of the IPG technology, the presence of a reproducible 1–6 Hz response to drug cues during a period of DBS discontinuation and its attenuation to targeted stimulation underscores its potential as a state-dependent marker of cue-reactivity. This unique case study supports the need for chronic sensing tools to track biomarker fluctuations over time, guide stimulation therapy titration, and potentially forecast relapse risk. Although most prior DBS studies in OUD have utilized bilateral NAc stimulation, our patient experienced sustained benefit from unilateral stimulation. Preclinical studies have demonstrated comparable behavioral effects between unilateral and bilateral NAc DBS in addictive behavior models28,41,42,43. These findings support that unilateral stimulation may modulate bilateral networks44. Studies in movement disorders and affective circuits have also demonstrated that unilateral stimulation can induce bilateral network modulation through inter-hemispheric or subcortical network connectivity45,46,47. These observations suggest the possibility that targeted unilateral stimulation may be sufficient in selected individuals and merit further investigation.

Based on personalized tractographic analyses (Fig. 2b-g), we found that the new stimulation site within the NAc shell exhibited strong structural connectivity with the SCG, suggesting preferentially engagement of affective and valuation networks in the medial prefrontal cortex15,16. These circuits have been reported to mediate interoception, emotional and reward valuation, and compulsive urge regulation, functions reported to be at least in part mediated by the broader medial prefrontal cortex15,16,48,49,50. While SCG has primarily been studied in the context of depression, emerging evidence suggests that the NAc–SCG circuit plays a broader role in the regulation of motivated behavior and impulsivity, as it relates to drug cravings12,13,16,25,41. This is supported by recent preclinical evidence that DBS of the infralimbic cortex (homolog of SCG) in rats specifically reduced cocaine and opioid-seeking and reinstatement51,52. Similar results were achieved with stimulation in the rat NAc shell, through local and antidromic activation of the infralimbic cortex53. Notably, we have previously reported that optogenetic activation of this circuit attenuated binge-eating behavior in preclinical models54. The current human case study provides convergent clinical evidence that NAc-SCG circuit modulation may underlie effective craving suppression in OUD, suggesting engagement of a broader network that may mediate limbic responses to cue-induced craving not limited to affective responses and valuation. In contrast, the original vALIC stimulation site likely engaged wider networks traversing cognitive and motor territories, potentially diluting specificity and requiring higher charge density for clinical effect. These findings support the hypothesis that precise engagement of the NAc-SCG axis may underlie more effective and efficient craving suppression in OUD and point to a promising therapeutic target for future studies, although there is currently no validated tractographic target or connectomic signature for OUD. Prospective studies combining imaging-based connectivity with behavioral and physiological markers across individuals may help establish such a target.

Despite the constraints of this clinically-driven scenario, we observed an association between the cue-reactive 1–6 Hz low-frequency band power detected in the NAc shell subregion and self-reported craving states. In the past decade, interest in brain signal-based closed-loop neurostimulation has gained significant traction in movement disorders55,56, epilepsy57, and neuropsychiatric disorders21,26,58. This mode of stimulation is particularly intriguing for conditions that exhibit dynamic symptomatology commonly considered for epilepsy47,59, but similarly, SUD60. Chronic and continuous DBS may in fact result in tolerance or undesirable consequences such as blunting of normal physiological functions that utilize the same pathway38,61. While the intent of this case study was not to facilitate a responsive stimulation strategy (in part due to limitations of the in-situ DBS system), the successful identification of a behaviorally relevant electrophysiological biomarker that can be modulated with acute stimulation heralds promise in utilizing electrophysiological localization of aberrant activity to guide DBS for OUD. Finally, the overall charge density delivered was substantially reduced, further providing support that this strategy may help reduce adverse and untoward effects of DBS and provide a more durable solution.

We acknowledge the limitations of this N-of-1 case study, for which the results may not be broadly generalizable to all SUD populations, or even to other individuals with OUD. It remains unclear how factors such as age of onset, route of drug administration, prior treatment exposure, and other clinical variables may influence the electrophysiological expression of drug-cue reactivity. In addition, iEEG recordings were constrained by the existing DBS electrode contacts. Based on our findings, the low-frequency band power was strongest in the ventral-most contacts located in the NAc shell subregion. It may be possible that this electrophysiological signal extends to other related brain regions not sampled by the in-situ electrode. While we observed consistent condition-dependent effects across trials, our statistical comparisons were based on individual trial responses within a single participant. As such, the resulting p-values reflect within-subject reliability rather than generalizable population-level inference. Additionally, despite efforts to minimize potential confounds such as carryover effects, fatigue and cue-habituation through task design and scheduling, we cannot fully exclude the possibility that these factors may have influenced the observed behavior or electrophysiological responses. Nonetheless, we outline a compelling utility of electrophysiological guidance to optimize DBS programming and potentially personalize this therapeutic. These considerations highlight a proof-of-concept that inspires future studies in larger, phenotypically diverse OUD populations to assess reproducibility, specificity, and potential clinical utility of the observed biomarker. We propose this cue-reactive electrophysiological signal in the NAc region as a potential candidate biomarker that warrants further investigation in ongoing and future studies of DBS in OUD. Future studies should also explore whether this low-frequency biomarker generalizes to other SUDs and relapse triggers, including stress-induced craving, and whether it can differentiate compulsive drug motivation from other reward-related states. Such work could refine its specificity and broaden its potential as a transdiagnostic marker of maladaptive reward processing.

Methods

Clinical history

At the time of the current presentation, the patient was a 25-year-old male with DBS implanted 4.5 years prior. He first developed an addiction to opioids from thirteen years old following rapid withdrawal of prescription opioid medication. He also had co-morbid depression, anxiety and attention deficit hyperactivity disorder, and was treated with multiple anti-depressants, mood stabilizers and anti-psychotic medications. He started intravenous use of opiates at age fifteen and experienced episodes of opioid overdoses. He also developed disabling tardive dyskinesia that disrupted all his activities of daily living and led to severe disability (Global Assessment of Functioning Score 40) and dropped out of school. In spite of chronic methadone maintenance therapy (maximum dose 320 mg/day), he had recurrent opioid cravings and required several inpatient hospitalizations for detoxification. After several years of medication optimization without success for his neuropsychiatric conditions, he was finally referred by his movement disorder specialist and psychiatrist for off-label bilateral DBS of the globus pallidus interna (GPi) to treat his severe tardive dyskinesia62. At the same time, DBS therapy for his severe, medication-refractory OUD was discussed. After discussion with the patient, with his parents and after careful consideration of potential risks and benefits, a consensus decision was made for bilateral GPi DBS (with right-sided IPG) to treat his tardive dyskinesia, as well as a unilateral left NAc DBS (left-sided IPG) targeted at treating his OUD (Figure S1). A unilateral NAc implant was chosen to mitigate the risk of this off-label surgery, recognizing an additional lead could be placed in the future if deemed necessary.

The DBS therapy was programmed based on standard clinical monopolar assessment used in movement disorders and psychiatric practice. Bilateral GPi DBS improved his dyskinetic movements significantly, which remained stable and managed at an outside facility with minimal adjustment required since initiation. The left NAc DBS electrode was activated based on in-clinic assessments of mood, anxiety and energy, but required a very high dose of stimulation (14.5 mA) between contacts 4 and 5, which were located in the vALIC (Fig. S1c). When stimulated at yet higher doses, he experienced diaphoresis and myoclonus. Unfortunately, he was unable to tolerate a rechargeable system, which destabilized his clinical condition whenever the IPGs were inadequately charged, leading to the team to recommend switching to non-rechargeable IPGs. The high dose requirement, however, led to rapid charge depletion of the IPG. Several months after clinical stabilization, he relocated to France, where a local psychiatrist managed his condition and DBS programming. Following DBS activation, methadone dose was tapered to 120 mg/day, and the patient remained abstinent for four years without requiring residential, intensive outpatient (IOP), or sober living programs. His clinical course suggested that DBS played a central role in suppressing craving and maintaining abstinence. In the interim, he developed episodes of hospital-acquired infections (pneumonia with bacteremia), complicated by infective endocarditis that required five cardiac surgeries for valve replacements. All these cardiac procedures occurred between ages 22–23, almost two years before his presentation to our team. There was no clinical or toxicologic evidence of intravenous drug use relapse during this period and abstinence was corroborated by urine toxicology and caregiver reports. He abstained from drug use for several years (sustained remission) while the DBS was on, and craving levels were minimal. He was able to regain function in life, completing high school education and was able to work.

Four months after his last cardiac surgery, he underwent replacement of his left IPG due to battery depletion. A subsequent wound erosion (eleven months later) over the left subclavicular incision warranted removal of the left DBS IPG (extension wires were retained) and wound debridement, which was followed by a rapid recurrence of opioid cravings and drug-related thoughts (Fig. 1a). Within the first several weeks of IPG explant, the patient condition destabilized, and he attended emergency department three times for symptoms of elevated cravings and fear of relapse. His medications were uptitrated and methadone had to be increased from 120 mg/day to 240 mg/day to cope with his symptoms. This led him to re-engage with addiction support services. Because no local surgeon with necessary expertise was available and given his complex medical history, his family and psychiatrist reached out to our institution to re-establish DBS therapy. Due to clinical suspicion for ongoing wound infection given ecchymosis at the subclavicular site, the retained extension wire was explanted and the DBS lead externalized with an extension cable while local and blood cultures and antibiotic treatment ensued. During this period, personalized computer-based cue-reactivity tasks were performed to identify and localize a potential electrophysiological biomarker of drug-cue-reactivity by recording brain activity from the DBS lead. We hypothesized that prior charge density requirements were due to non-specific engagement of the NAc region from vALIC stimulation site. To identify sites within the NAc region where cue-reactivity was most robust and specific to relevant drug cues, three computer-based tasks were conducted: a primary drug-cue-reactivity task (designed with his personalized drug use habits) to assess for presence of drug-related biomarker6 (Fig. 1b), the Monetary Incentive Delay, MID10 and the Blink Suppression Test, BST11 (Figs. S2 and S3). The two latter tasks were used as behavioral controls to assess specificity of electrophysiologic drug-cue reactivity.

There were several concerns in the management of this patient. First, a time-sensitive restoration of his DBS therapy was considered to be crucial, as he reported exacerbation in his cravings (though he had demonstrated sustained drug abstinence), a daily struggle for him. Second, the high charge requirement and intolerance of a rechargeable system was worrisome in view of his young age. Thirdly, it was critical to ensure complete eradication of the infective organism before re-implantation of the IPG due to risk to both the DBS device as well as his prosthetic heart valves. With these considerations in mind, the patient was hospitalized for treatment, with a motivation to restore his DBS therapy as soon as he was cleared by infectious disease and potentially optimize programming parameters. Due to suspicion of unresolved infection of the left IPG site, a wound exploration and swab were performed, while explanting the retained extension wire. This revealed persistence of Staphylococcus aureus. Upon consultation with infectious disease specialist, he was commenced on intravenous antibiotics and the DBS system was externalized percutaneously with an extension wire using previously reported surgical methods in an attempt to preserve the proximal system while antibiotic treatment ensued63. Using the externalized system64,65, we devised a plan to use personalized computer-based cue-reactivity tasks to attempt to identify potential electrophysiological biomarker of drug-cue-reactivity. The overarching purpose of this was to localize the most active node encompassed by the NAc electrode for clinical purposes. We hypothesized that prior high amplitude requirements were due to non-specific stimulation target engagement and that a more precise stimulation may help reduce charge requirement and extend device lifespan which would be critical in the clinical management of this patient. The patient provided informed consent to participate and to allow publication of relevant information for educational purposes. The DBS recordings were performed entirely for clinical purposes in this unique off-label case of a patient at imminent risk for relapse. The recording data collected added confidence to how the acute stimulation assessment was interpreted though did not change clinical practice. Moreover, the University of Pennsylvania maintains an Institutional Review Board-approved protocol for post-implantation recordings of electrophysiological data in DBS patients (IRB# 851872).

Computer-based tasks

  1. (a)

    Personalized cue-reactivity paradigm task3,6. The patient was clinically interviewed by an expert in cue-reactivity (A.R.C) for specific details related to his drug use habits and preferences prior to the task. The patient was presented with a series of 6 s-long videos in a quasi-random block design in two categories (drug cues and neutral cues). The video clips precisely reflected his paraphernalia, preferred opioid drug and injection preparation. This personalized approach seemed necessary to evoke a reliable electrophysiologic cue reactivity and an associated craving response in this patient, as he reported that standardized cue databases, developed primarily based on U.S. practice, did not reflect his experience and failed to elicit strong craving. Neutral cues consisted of emotionally neutral nature scenes to serve as a low-salience comparator. After each video presentation, the subject was asked to rate his level of craving from 1 (no craving) to 9 (extremely high craving), reflecting the patient’s momentary craving intensity in response to the preceding stimulus. Each session included six drug-related cues and six neutral (video) cues, presented in a quasi-randomized block design. The task was designed based on our team’s clinical judgement and cue-reactivity paradigm expertise of one of our senior authors (A.R.C) to balance the need for cue repetitions (for signal averaging) with risk of reduction of cue-induced responses with too many exposures (habituation or extinction effects). To help minimize the repeated, exposure-related reductions in responses, cue sessions can be less frequent or spaced widely apart. Therefore, in our design, the cue-reactivity tasks were conducted on alternate days, and tasks were not performed on days involving stimulation testing with craving assessment to allow sufficient stimulation washout. A total of four cue-reactivity task sessions were completed: two before any stimulation was initiated and two with active stimulation. Sham stimulation was not included in the cue-reactivity task to minimize exposure-related reductions in cue-triggered responding.

  2. (b)

    Monetary Incentive Delay (MID) task10. The MID task is a validated and widely-used cognitive task to identify neural substrates of monetary reward anticipation and outcome, and has been used by this team to activate cue-reactivity within the NAc using both functional magnetic resonance imaging and electrophysiology28,66. The patient was presented with a visual shape cue that signals the possibility of a reward (potential gain), loss (potential loss) or neutral (no gain or loss) trial. Following the visual shape cue presentation of two seconds, a delay phase of two seconds is imposed before a target appears to prompt a response by the participant with a button press as quickly as possible to gain the reward or avoid a loss. Finally, in the feedback/outcome phase, the participant receives performance feedback with cumulative earnings. The MID task was performed using real monetary rewards, with immediate feedback and actual earnings delivered based on trial outcomes, to ensure ecologically valid engagement of reward anticipation circuits67. This task is one of the most widely validated paradigms for probing reward-related neural activity.

  3. (c)

    Blink Suppression Task (BST)11. This task uses eyeblink suppression as a model for sensory-based urges, with blocks of free blinking (30 s) alternating with longer blink suppression periods (60 s). During the free blinking periods, the instruction ‘Normal’ is displayed and subject is allowed to blink as normal. During the blink suppression period, the instruction ‘Hold’ is displayed and the participant is told to withhold blinking as long as the instruction is present. After this block, there is a recovery period with instruction ‘Ok to blink’ where the participant is allowed to blink as much as he wished, and a visual-analog-scale (VAS) rating of the urge to blink during the ‘hold’ period is collected. Eight blocks of each period are presented.

Electrophysiological recording and signal analysis

After wound debridement and cultures, a Boston Scientific DBS extension wire (model NM-3138-55) was connected to the in-situ Boston Scientific DB-2201 8-ringed contacts DBS electrode, tunneled and externalized to allow intravenous antibiotic treatment of the wound infection and preservation of the remaining DBS system63. An in-house customized adapter was used to connect to the Boston Scientific DBS extension wire (Boston Scientific, Marlborough, MA) to the g.HIAMP PRO system (g.tec medical engineering GmbH, Austria) for continuous iEEG recordings of the NAc region during conduct of the computer-based tasks.

Raw iEEG data acquired at 1200 Hz sampling rate was exported post-hoc for offline analysis using custom MATLAB scripts, applying a high-pass filter (1 Hz) and notch filter at line noise frequency (58–62 Hz) and harmonics. A time-frequency analysis using multi-taper convolution methods was used to characterize spectral power components of the cue-reactive signals. The power was extracted by squaring the magnitude of the complex Fourier-spectra and transformed by the log function. The time window length was shortened (4 to 0.13 s) with increase in frequency (1–115 Hz) and time step of 0.02 s. The frequency resolution was dense in the low frequencies starting from 1 Hz and more sparse in the high frequencies up to 115 Hz using logarithmic space. The power spectrum was corrected using a common baseline from the interval 0.1-1 s before the cue was presented. Difference in mean power spectrogram obtained from each condition epoch (e.g., drug-cue or blink suppression) was compared to a neutral condition epoch (e.g., neutral-cue or normal blink) which served as the control condition. To identify significant time-frequency regions of interest without predefining time epochs and frequency bands, we applied a data-driven procedure across the full time-frequency spectrogram (1–115 Hz, log-scaled). For each time-frequency sample, we computed a two-tailed paired-sample t-test, to compare responses to drug vs neutral cues across trials. Samples exceeding an uncorrected threshold of p < 0.05 with the same sign were grouped into connected sets, i.e., ‘clusters’, based on spectral and temporal adjacency68. Clusters were identified by size (i.e., the number of supra-threshold samples), and a permutation test was performed to test for significance for each of the cluster (10,000 permutations, alpha=0.05). Within-pair sign-flip permutations were used to construct a null distribution of the largest cluster size observed by chance (two-tailed). Each observed cluster was compared against the null distribution to determine whether the cluster was unlikely to occur by chance, and only clusters that passed this comparison were considered significant. No conclusions were drawn from individual time–frequency bins. This cluster-correction process resulted in statistically significant clusters in time and frequency that can be used to define a region of interest (ROI) for further hypothesis-driven analyses. To compare responses of stimulation (e.g., ROI mean power or craving ratings during drug cues with vs. without stimulation), a non-parametric two-tailed permutation test (10,000 permutations) was applied. For each comparison, within-pair sign-flip permutations were performed to generate a null distribution of mean differences. The p-value was computed as the proportion of permuted differences greater than the observed difference (two-tailed). This analysis of the stimulation effect is distinct from the cluster-based permutation described above, and was chosen due to the small sample size and non-normal distribution of these data. Trials with excessive movement artifact or recording dropouts were excluded from analysis to ensure data quality.

Neuroimaging analyses

High resolution pre-operative MRI and post-operative computer tomography (CT) head were available for imaging analyses. T1, T2, and fast gray matter acquisition T1 inversion recovery (FGATIR) sequences were acquired at 1 mm thick slices on a 3 T GE scanner (GE Healthcare, Chicago, IL), as well as multi-shell 64-direction diffusion weighted imaging for tractography. The post-operative CT was co-registered to the pre-operative MRI using a 2-step linear and non-linear registration using Advanced Normalization Tools69. Probabilistic tractography was performed using FMRIB Software Library (FSL)’s Probtrackx2 using the NAc as a seed and the subgenual cortex as a target to subsegment the NAc into shell and core regions using clustering of streamline density according to our previously published methods13.

Volume-of-tissue activation (VTA) analysis was performed using Lead-DBS software based on active DBS parameters using its in-built finite element modeling (FEM)70. Structural connectivity analysis with streamlines was performed using the DSI-Studio71 using the estimated VTAs as seeds in the whole brain tractography derived from patient’s diffusion MRI.

Acute anti-craving stimulation assessment

Using the Boston Scientific external stimulator, a systematic stimulation assessment was performed across adjacent electrode contact pairs in bipolar fashion. For each contact pair, stimulation amplitude was gradually increased in 0.5 mA stepwise increments, with each level maintained for at least 1 min to observe acute effects and monitor for side effects. Acute stimulation effects were recorded for therapeutic as well as side-effect thresholds. The patient’s mood, anxiety and affect were recorded per standard limbic stimulation assessment. In addition, opioid craving levels were recorded using a VAS scale compared to baseline level.

To confirm anti-craving effects of stimulation, a short sham-controlled (patient and assessor blinded) stimulation session was also conducted to verify the effects of active stimulation. Sham stimulation was delivered using a dedicated and separate stimulation box that generated auditory outputs similar to active stimulation (e.g., clicks and device engagement) but did not deliver electrical current to the brain. This approach preserved sensory cues associated with stimulation while avoiding neuromodulatory effects. The patient was informed that some blocks would include stimulation and others would not, but was blinded to specific condition assignment during each trial. For the stimulation parameters notation, the configuration “C + /4-(60%) 5-(40%)” refers to the case acting as the anode, with current split between contacts 4 and 5 (60% and 40%, respectively) as cathodes.

Clinical outcomes

A composite self-report electronic questionnaire was administered at regular intervals throughout the follow-up period. This included items adapted from the Obsessive Compulsive Drug Use Scale (OCDUS), Brief Substance Craving Scale (BSCS) and Brief Addiction Monitor (BAM) questionnaires as well as VAS for anxiety, mood and energy72. VAS ratings of craving frequency and severity range from 0 to 10 unless otherwise noted. The composite was used to monitor the patient’s subjective levels of craving, substance use, mood and behavior over time. Scores from this tool are presented in Figs. 2a and S4 as part of the longitudinal assessment of clinical outcomes.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.