Deep Brain Stimulation for Addictive Disorders—Where Are We Now?

In the face of a global epidemic of drug addiction, neglecting to develop new effective therapies will perpetuate the staggering human and economic costs of substance use. This review aims to summarize and evaluate the preclinical and clinical studies of deep brain stimulation (DBS) as a novel therapy for refractory addiction, in hopes to engage and inform future research in this promising novel treatment avenue. An electronic database search (MEDLINE, EMBASE, Cochrane library) was performed using keywords and predefined inclusion criteria between 1974 and 6/18/2021 (registered on Open Science Registry). Selected articles were reviewed in full text and key details were summarized and analyzed to understand DBS’ therapeutic potential and possible mechanisms of action. The search yielded 25 animal and 22 human studies. Animal studies showed that DBS of targets such as nucleus accumbens (NAc), insula, and subthalamic nucleus reduces drug use and seeking. All human studies were case series/reports (level 4/5 evidence), mostly targeting the NAc with generally positive outcomes. From the limited evidence in the literature, DBS, particularly of the NAc, appears to be a reasonable last resort option for refractory addictive disorders. We propose that future research in objective electrophysiological (e.g., local field potentials) and neurochemical (e.g., extracellular dopamine levels) biomarkers would assist monitoring the progress of treatment and developing a closed-loop DBS system. Preclinical literature also highlighted the prefrontal cortex as a promising DBS target, which should be explored in human research. Supplementary Information The online version contains supplementary material available at 10.1007/s13311-022-01229-4.


Introduction
Addictive substances pose significant socioeconomic impact on society and the healthcare system, costing hundreds of billion dollars annually in the USA alone [1,2]. Psychological and adjunctive pharmacological treatments are the current mainstay for addiction therapeutics.
However, the risk of relapse remains high, with some citing relapse rates as high as 75 to 98% within 1 year of treatment [3]. Investigations into new treatment modalities are urgently warranted. To develop novel therapies for addiction, researchers are focusing on the behavioral and neuroscience aspects in the biopsychosocial model ( Fig. 1). Deep brain stimulation (DBS) in particular is a neuroscience-based potential treatment for medically refractory addictive disorders, given its diminishing surgical risks. "Medically refractory" cases have no formal definition, although these are generally regarded when conventional treatments repeatedly fail [4].
Recent advances in neuromodulation techniques and devices led the indication of DBS to include movement disorders, such as Parkinson's disease, to neuropsychiatric conditions, such as depression, Tourette's syndrome, and obsessive-compulsive disorder (OCD) [5][6][7][8]. The use of DBS to reduce substance use has also been explored in both preclinical and clinical settings [9][10][11][12], offering a beacon of hope to patients suffering from medically refractory addictive disorders. This review aims to summarize and evaluate the potential therapeutic efficacy, mechanisms of action, and biomarkers involved in the use of DBS in the treatment of addictive disorders. We cautiously report that DBS is promising to treat certain refractory addictive disorders. Recommendations for the future directions are also provided.

Addiction Neurobiology: a Brief Overview
Addiction is a highly complex process involving multiple neurochemicals and brain structures. A highly simplified model highlights the central role of dopamine in mediating drug reward learning and drug seeking (compulsive) behaviors (Fig. 2). In essence, addictive drugs, such as the psychostimulants and opioids, enhance synaptic concentrations of dopamine in forebrain subcortical structures, such as the nucleus accumbens (NAc) [13]. These drugs of abuse significantly enhance dopamine concentrations over and above the ability of natural rewards, hijacking the system and encoding a more biologically salient reward association with surrounding environmental cues [13]. Such Pavlovian drug-environmental conditioning can trigger the desire to seek more of the drug. For example, the drug associated cues can elicit a surge of dopamine, which may manifest in drug craving, seeking and ultimately use [14].
Serotonin (5-HT) is another important neurotransmitter mediating neuroplasticity, hedonic tone, motivational, and reinforcement processes, as well as cognitive functions [15,16]. 5-HT 2A and 5-HT 2C receptors in the prefrontal cortex (PFC), NAc, and ventral tegmental area in particular appear to be critical. 5-HT 2A receptor increases, and 5-HT 2C receptor decreases, striatal dopamine release [15]. The serotonergic neuroadaptations may be accountable for emotional components such as anhedonia and depression during drug withdrawal [16]. Recent evidence in mice also suggest that serotonin signaling via 5-HT 1B receptor may attenuate transition from casual to compulsive cocaine use [17].
Chronic drug use ultimately alters the homeostatic neuroplasticity in the brain which not only results in a sensitized dopaminergic system [18], but also leads to changes in the glutamatergic system [13,19,20]. For example, calciumpermeable ionotropic glutamatergic AMPA receptors become upregulated relative to glutamatergic NMDA receptors [13,19,20]. The changes in AMPA and NMDA receptors have  [140]. Interactions between components in different domains lead to addictive behaviors. Created with Biorender.com Fig. 2 A simplified illustration of the neurotransmitters and neuroanatomical structures involved in the pathophysiology of addiction [13,19,20]. Other relevant areas are omitted for clarity. The effect of dopamine receptors D 1 -R and D 2 -R is shown in inset. D 1 -R mediates the direct pathway (positive reinforcement) and D 2 -R mediates the indirect pathway (negative reinforcement). Italic descriptions denote changes secondary to chronic drug use. 5-HT, serotonin; AMPAR, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; D1/2R, dopamine 1/2 receptors; DA, dopamine; GABA, gamma aminobutyric acid; NAc, nucleus accumbens; NMDAR, N-methyld-aspartate receptor; VTA, ventral tegmental area. Created with Biorender.com become a hallmark process of neuroplasticity that underpins behavioral changes following chronic drug use [21]. Another neurobiological adaptation that occurs with chronic drug use is the process of "dorsalization" [22]. Here, repeated exposure to drugs of abuse leads to recruitment of the circuit that projects from the orbitofrontal cortex (OFC) to the dorsal striatum (caudate and putamen), where synaptic potentiation occurs at both dopamine receptors 1 and 2 (D 1 -R and D 2 -R) on GABAergic medium spiny neurons (MSNs). Activation of D 1 -Rs receptors causes excitation in the postsynaptic cell via cyclic adenosine monophosphate production (cAMP) production [23] that ultimately leads to promotion of long-term potentiation [24]. D 1 -R-and NMDA receptor-dependent activation of extracellular signal-regulated kinase (ERK) in brain areas such as striatum is an important pathway for drug-induced locomotor response and sensitization [25][26][27][28]. For example, systemic injection of D 1 -R-or NMDA antagonist prevents ERK phosphorylation triggered by d-amphetamine in the mouse striatum [29]. In addition, cocaine-induced ERK phosphorylation and locomotor sensitization is prevented with mitogen-activated protein kinases/ ERK kinase inhibitor or in mutant mice with alanine-replaced Thr-34 residue of dopamine-and cAMP-regulated phosphoprotein (DARPP-32) [29]. D 1 -R MSN structural plasticity depends on NMDA and DARPP-32 signaling [24]. Psychostimulants appear to affect ERK signaling via DARPP-32based D 1 -R-and NMDA intracellular pathways [29].
Importantly, D 1 -R activation inhibits long-term depression, as revealed by D 1 -R antagonist unmasking long-term depression [30]. In contrast, D 2 -Rs can promote long-term depression and prevent the induction of long-term potentiation [30]. However, D 2 -Rs are particularly sensitive to dips in the tonic dopamine level, during which they activate protein kinase A (PKA) to induce long-term potentiation [30][31][32]. Taken together, D 1 -R and D 2 -R are critical for bidirectional plasticity of striatal MSNs when dopamine levels are high vs low to guide adaptive and maladaptive behaviors involved in reward learning [31,32]. Therefore, D 1 -R and D 2 -R signaling dynamically modulates cortico-striatal plasticity depending on the level of synaptic dopamine to facilitate transition into persistent drug seeking and use.
While DBS appears to target these processes involved in addiction to reduce drug seeking and taking [10,[33][34][35][36][37], the potential therapeutic efficacy and mechanisms of action of DBS in the treatment of addictive disorders are poorly understood. We address this gap in the present review.

Methods
A search was performed using keywords, such as "deep brain stimulation" and "addiction" in MEDLINE, EMBASE, and Cochrane Library database between 1974 and 6/18/2021. Details of search criteria and inclusion and exclusion criteria are given in Supplementary Information Appendix A. Search results are presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow chart (Fig. 3). This review was registered at the Open Science Framework [38]. The process yielded 47 papers (25 preclinical studies and 22 human studies), which were further reviewed in full text. The results of this literature search are summarized in the sections below.

Preclinical Models and Clinical Correlation
Animal DBS studies in the selected literature (Table 1) predominantly used conditioned place preference (CPP) or self-administration paradigms (Fig. 4) to model addiction. Self-administration models have high construct, content, and face validity relative to human addictive behaviors. Experimenter-administration models such as CPP can measure drug-induced behavioral sensitization, such as increased locomotion, and are easier to perform. Both models also allow modeling of withdrawal symptoms during abstinence and/or extinction phase, followed by relapse behaviors, such as cue or drug-induced reinstatement. Most common DBS brain targets in the selected animal studies (Table 1) were the NAc [33,[39][40][41][42][43][44][45], PFC [44,46,47], and subthalamic nucleus (STN) [10,37]. The following sections will focus on these three most studied targets.

NAc Electrical Stimulation
The NAc is divided into core and shell subregions with different histopathological compositions, projections, and functions [48]. Studies suggest that the shell is responsible for reinforcing properties of novelty, rewards (both drug-and non-drug-related), and drug relapse, whereas the core mediates spatial learning, conditioned responses, responses to motivational stimuli, and impulsive choices [14,49]. These features of motivated behaviors suggest NAc shell underlying extinction and reinstatement behavior [50], and NAc core underlying the initial acquisition of drug taking and cue-elicited drug seeking [48]. However, there is synergy between the two subnuclei, and both play a role in drug addiction.
Stimulation of either the core or shell subregions of the NAc using a range of low and high frequencies leads to reduced drug taking, seeking, or CPP in rats conditioned with alcohol [34,51,52], nicotine [53], opioids [9,37,39,41,46,[54][55][56][57], cocaine [10,33,35,40,42,43,47,58,59], or methamphetamine [45] (Table 1). Behavioral improvement was related to the timing of when electrical stimulation was given-that is, stimulation given before a self-administration session or CPP acquisition would reduce drug intake or prevent CPP, respectively [39,45]. When given during abstinence or reinstatement, stimulation would diminish drug or cue-induced drug seeking after extinction and/or forced abstinence phase [40][41][42][43]45]. Such efficacy of stimulation during abstinence to prevent relapse is particularly promising in the clinical context. While most animal and human studies ( Table 2) show success in modulating addictive behavior using NAc stimulation, there are two main issues. First, there is conflicting evidence in animal studies regarding the efficacy of NAc core vs shell stimulation. Both high-and low-frequency stimulation (HFS, LFS) of NAc (electrodes placed across core and shell) can reduce cocaine-primed drug seeking following 24 days of abstinence [43]. Other studies using core as the DBS target showed reduced morphine-induced CPP [39], ethanol self-administration [51], and heroin reinstatement [41]. The shell was also shown to be an effective target for morphine-induced CPP [55], cocaine reinstatement [40], ethanol self-administration [44,51,52], and methamphetamine reinstatement [45]. In contrast, Vassoler et al.  demonstrated that stimulation of the NAc shell but not the core attenuated cocaine-induced reinstatement [42]. The contradicting evidence above may be explained by the large activation volume of stimulation pulses, such that pulses centered on the core may also affect the shell, and vice versa. From human studies, the volume of activation was estimated to be between 30 and 120 mm 3 [60] and the human NAc volume is around 700 mm 3 [61]. However, for adult Sprague-Dawley rats, the total volume of NAc is approximately 5 mm 3 with the NAc core making up half of that [62]. Therefore, neighboring structures could be co-stimulated in rats. Notably, the core and shell divide is functionally present in humans while they are less distinct in rodents [63,64].
The second issue concerns laterality. While most studies used bilateral stimulation, some studies only used unilateral NAc stimulation [34,39,41,43]. One study used left, right, and bilateral NAc core stimulation, demonstrating only the latter two were effective in reducing heroin seeking behavior [41]. It is unclear why such asymmetry exists. Dopamine concentration in the NAc is reported to be higher on the side ipsilateral to the paw-dominance of a mouse [65]; however, morphometric NAc human studies based on laterality have been controversial [49].
Almost all of these published studies in humans show some improvement from DBS in the NAc in addictionrelated behaviors ( Table 2) as well as cognitive improvement, which raises the possibility of potential publication bias. With this precaution in mind, NAc DBS is likely the lowest hanging fruit for refractory addictive disorders. A recent systematic review supports NAc DBS to treat substance use disorders, demonstrating that NAc DBS in patients led to a relapse rate of ~ 39% with a reasonable safety profile, which is an improvement compared to the ~ 60% relapse rate reported in that population (selection bias factors notwithstanding) [66]. In addition, individual case reports revealed that those with comorbid psychiatric disorders, such as depression and OCD, showed improvement in these other symptoms with NAc DBS with or without anterior limb of internal capsule (ALIC) intervention (DBS or capsulotomy) [12,67,68]. DBS may provide additional benefit for patients with multiple psychiatric comorbidities.

Cortical Stimulation
The PFC is believed to provide top-down control of NAc in addiction (Fig. 2). DBS of the PFC (subregion not specified) significantly reduced cocaine seeking and motivation to take cocaine measured by progressive ratio test [35]. Stimulation of the infralimbic cortex of the PFC following a cocaine prime also reduced cocaine reinstatement [47]. Notably, mPFC stimulation in rats showed increased phasic dopamine release in the NAc, which might explain the stimulation effect [69]. However, stimulation of the infralimbic cortex starting 1 h or 4 days before alcohol prime did not affect alcohol-induced relapse following abstinence in rats [44]. The contrasting findings may be due to the different mechanisms of actions of the drugs being given (alcohol is a depressant whereas cocaine is a stimulant). Alternatively, infralimbic cortex stimulation may be more effective in reducing relapse-like behaviors following extinction rather than forced abstinence, consistent with the well-known role of the infralimbic cortex in extinction [70]. Infralimbic cortical DBS success is concordant with reduced craving by transcranial direct current stimulation (tDCS) of dorsolateral PFC in humans [71]. However, DBS of the prelimbic cortex following cocaine priming had no effect on cocaine reinstatement in rats [47].
In the OFC, DBS prevented rats from learning morphineinduced CPP, while DBS during extinction facilitated extinction acquisition and prevented morphine-induced reinstatement [9]. Chronic DBS of the anterior insular cortex either during 11 days of extinction or 14 days of abstinence reduced CPP, but CPP relapsed 10 days after chronic DBS [57]. This finding is consistent with how repetitive transcranial magnetic stimulation (rTMS) reduced heavy drinking only during the intervention but not after in alcohol-dependents [72]. The insula's superficial location and delicate anatomy may be better suited for non-invasive interventions such as rTMS rather than DBS.
Anterior cingulate cortex (ACC) dysfunction can disrupt ongoing processing and detection of erroneous outcomes relevant for reward learning [73,74]. ACC DBS following cocaine prime did not alter reinstatement following selfadministration in rats [47]. However, when people with alcohol dependence received bilateral DBS (6-or 10-Hz stimulation), craving was significantly reduced [75]. The reduced craving was associated with a reduction in beta-1 band (13-18 Hz) current density on electroencephalogram (EEG) post-stimulation in the rostrodorsal ACC [75]. Alcohol craving also correlated with beta activity in the dorsal ACC in a case report, although relapse was associated with increased gamma band activity [76]. Another case report showed that a patient with concomitant OCD and alcohol dependence was successfully treated with ACC rTMS [77]. Interestingly, NAc DBS reduced craving correlated with improvement in error-related negativity in LFP in the ACC in humans [78], suggesting the effect of NAc DBS may alleviate ACC dysfunction. In addition to DBS, rTMS of the dorsal ACC demonstrated improvement in alcohol craving and concomitantly reduced NAc activation on fMRI [76]. Taken together, DBS in ACC may be particularly relevant for alcohol use.
Most of these cortical regions project to the NAc, raising the possibility that their DBS effects are mediated via these connections [49,79]. Animal studies show not only strong projections from the cortex to NAc, but also a high degree of interplay between the two regions [80]. For example, c-Fos reactivity is observed in infralimbic cortex and NAc during DBS [42,56] Pharmacological silencing of cortical interneurons can reinstate drug seeking [42,81], which is consistent with silencing effects observed in the NAc [81]. Nevertheless, NAc is unlikely to be the sole driver of effects of DBS in the PFC considering that DBS in the prelimbic cortex does not affect drug seeking in rats, even though prelimbic cortex has strong connections to NAc [47].

STN Stimulation
Parkinson's patients who take dopaminergic medications for their movement disorder may chronically feel compulsion to over-consume such agents, even at the expense of dyskinesia and other side effects. Studies suggest that STN DBS in humans may improve movement symptoms as well as reduce compulsive use of medications [82,83].
STN DBS or lesions in rats have been shown to reduce motivation for cocaine taking, while increasing motivation to consume more common rewards such as food [10,37,59,84]. The lesion study in particular suggests that STN DBS may work via local inactivation [59,84]. Further evidence shows STN HFS in rats inhibiting substantia nigra, entopeduncular nucleus, and NAc shell measured with brain mapping analyses of immediate-early gene expression but not PFC, and silencing STN neurons measured using recording ex vivo [37].
Although HFS of STN in rats reduced both cocaine taking and heroin seeking [37,59], a conflicting study demonstrated only LFS (30 Hz) but not HFS modulates cocaine taking behavior in the presence of foot-shock punishment [10]. Abnormal alpha/theta and low-beta oscillatory activity during escalation of the cocaine intake phase also predicted the subsequent emergence of compulsive-like seeking behavior, where the animals were shock-resistant [10]. The discrepancy here may be explained by the absence/presence of punishment during drug taking, but further studies are warranted to understand STN LFS vs HFS outcomes.

Other Brain Regions
Electrical stimulation targeting other brain regions in animals-amygdala [47,85], hippocampus [47], periaqueductal gray matter [54], lateral hypothalamus [35], dorsal striatum [40,44], and lateral habenula [58]-has also been described in the literature (Table 1). However, due to the limited number of studies per brain region, meaningful conclusions are difficult to draw. Hence, these regions are beyond the scope of this review.
The effect of HFS in NAc shell may be short-lived, lasting less than 4 h in a cocaine experimenter-administered mouse model [33] or 1 day in an alcohol self-administration rat model [34]. However, reduction in cue-induced seeking behavior was maintained more than 4 weeks post-HFS (NAc shell) compared to the sham group in rats that self-administered methamphetamine [45]. Yet, LFS and HFS (mixture of NAc core and shell) effects lasted at least 2 weeks but less than 30 days in a cocaine self-administration rat model [43]. The different DBS parameters, targets, drugs, rodents, and tests across these studies (Table 1) pose a challenge to evaluate the best DBS protocol to reduce drug use. Fig. 4 Illustration of two most widely used experimental paradigms for the study of drug addiction in animals. On the left, conditioned place preference (CPP) is shown, where "non-contingent" drug-administration is associated with one side of the chamber during drug-side associated conditioning. If the drug is experienced as rewarding, the animal will spend more time in the drug-administered side relative to the other side, even in the absence of further drug administration. On the right, "contingent" drug self-administration paradigm allows the animal to regulate its own drug intake (e.g., via nose-poke or lever-pressing). A discrete light and/or tone cue may be provided in association with the drug intake to serve as a drug-associated cue. In extinction, the animal is placed back into the environment where CPP or self-administration is acquired but without drug availability. Extinction sessions lead to reduced place preference or drug seeking. During reinstatement, the drug, the drug-associated cue, and/or stress can be given, which can lead to "relapse" measured by preference or drug seeking. Created with Biorender.com Two remained abstinent for years and 3 showed a reduction of consumption

Neuronal Inhibition
Despite the promising results of DBS in addiction, the mechanism of action is poorly understood and likely highly complicated. Several hypotheses on the mechanism of DBS in addiction have been suggested, such as direct inhibition/ excitation of neural activity, information lesioning, and synaptic filtering where DBS blocks transmission of pathological low-frequency oscillations [86]. In particular, the inhibition hypothesis is supported by animal studies where both injection of GABA agonists (i.e., temporary inhibition) and HFS in the NAc was found to reduce alcohol use in rats [34]. Similar effects were also seen between STN lesion paradigms and DBS with cocaine conditioning [59,84]. It is likely the neurons and their respective synaptic activities respond in a combination of different ways depending on the stimulation frequency and the drug being tested.

Depotentiation of Excitatory Inputs onto D 1 -R Via Modulation of Glutamate Receptors
Evidence also suggests the interaction between dopamine and glutamate receptors is involved in the mechanism of DBS action. For example, NAc shell LFS (12 Hz), only when used in combination with a D 1 -R antagonist, could reverse the motor sensitization and cocaine-evoked plasticity changes in D 1 -Rs on GABAergic MSNs, in the form of increases AMPA/NMDA glutamate receptor ratio [33]. By using a metabotropic glutamate receptor (mGluR1) antagonist to block the therapeutic effect, the authors proposed that the mechanism of action of DBS is mGluR-dependent, the activation of which depotentiates excitatory synaptic inputs onto D 1 -Rs on GABAergic MSNs [33]. Such depotentiation was associated with a normalization in drug-adaptive behavior. The findings are also consistent with the upregulation of GluR1 seen in the NAc in other stimulation studies [35,36]. In contrast, both LFS (20 Hz) and HFS (160 Hz) stimulation of the NAc core attenuated cocaine seeking behavior in rats without the need of additional medications [43], similar to the outcome of PFC stimulation [35]. The contrasting findings between NAc core and PFC stimulations may be explained by the different levels of basal dopamine in these brain regions. Examining how DBS affects mGluR1 expression in different regions to affect D 1 -R signaling would be informative in future work.

Antidromic Activation of Cortex
Post-mortem brain c-Fos immunohistochemistry following in vivo NAc shell HFS showed both local activation and activation of the infralimbic cortex, suggesting that DBS of the NAc shell led to antidromic stimulation of the cortex via cortico-accumbal afferents [42]. Consistent with infralimbic cortex-NAc circuitry in DBS, electrophysiological studies showed NAc HFS but not LFS enhances both NAc-evoked LFP responses and spontaneous LFP slow oscillatory activity in OFC in an antidromic fashion [87]. These results further demonstrate the complex interplay between NAc and the cortex.

Dopamine Replacement
One study showed that HFS of the NAc shell led to a paradoxical augmentation in relapse behavior in a rat model of alcohol addiction [44], a contrary finding to other studies ( Table 1). The authors also demonstrated an increase in extracellular levels of dopamine as measured by microdialysis. DBS-evoked increase in dopamine is reminiscent of the intracranial self-stimulation model, where the subjects selfadminister electrical stimulation via implanted electrodes in the brain [88]. Reinforcing effects of electrical stimulation is presumably mediated by dopamine release in the reward circuitry [89]. In addition, medial PFC DBS evoked dopamine release in the NAc in a frequency-dependent manner in anesthetized rats measured by fast-scan cyclic voltammetry (FSCV) [69]. Taken together, DBS of PFC may operate partially via the replacement of dopamine, but whether this can reduce drug taking or seeking has yet been tested. Most of the DBS studies did not employ in vivo microdialysis techniques for direct quantification; hence, replacement of dopamine as a DBS mechanism is unconfirmed.

Limitations of Animal Models
Human subjects tend to require weeks and months of DBS to observe an improvement, whereas animal studies show an almost immediate effect. In animal studies, the conditioning occurs over days (rather than years). The short timeframe indicates that the neuroplasticity reversal may not be the dominating effect of DBS in animals and the dopamine replacement caused by stimulation may mediate the acute improvement in behavior. In humans, DBS likely operates via reversal of neuroadaptations that have been built up on a much longer time scale than animals subjected to conditioning. This interpretation is supported by case reports where subsequent explantation of the DBS device did not reverse the initial effect of abstinence [90,91]. In addition, a further case report showed that NAc DBS reduced risk taking behavior and increased activation of brain regions implicated in behavioral control arising from lasting neuroadaptations, measured by positron-emission tomography [92].
Some human studies also include subjects with comorbid OCD, depression, and anxiety [12,93,94]. In fact, addiction may be part of a maladaptive coping mechanism for other underlying disease/s. DBS may alleviate addiction by reducing the severity of an underlying comorbid disease, which is challenging to model in animals. For example, the higher cognitive and emotional component as well as the impact of social cues are difficult to assess in animals.

Alternatives to DBS
Given the invasive nature of DBS, other neuromodulatory modalities are worth considering. Of note, noninvasive techniques such as rTMS and tDCS have been of particular interest [95,96]. A detailed review of these techniques and outcomes is out of the scope of this article but in brief, rTMS operates by applying a stimulating coil at the scalp, which generates magnetic fields. The magnetic fields then pass through the skull and induce strong focal currents in the underlying brain tissues [97]. It is believed that long-term repetitive stimulation can lead to neuroplasticity, which can provide therapeutic effect to the addicted brain [97]. tDCS involves application of a direct or alternative electrical current to the scalp, such that current travels from a positive (cathode) to a negative (anode) electrode. Cortical neurons are suggested to be facilitated under the anode, whereas those under the cathode are suppressed [98]. Although rTMS and tDCS have been shown to be effective in preclinical studies and small clinical case series on reducing symptoms such as craving, more data are required to ascertain their long-term safety profiles, efficacy, and the ideal patient profile [98][99][100]. Although research shows that non-invasive neuromodulation can be successful in certain cases without the need for neurosurgery, the main limitations are the focality of these non-invasive treatments and the relatively low spatial resolution. These non-invasive interventions also require numerous frequent visits to facilities in a specialized, tertiary center. In contrast, DBS can be implanted as a permanent or chronic stimulation system, which can potentially be programmed remotely [101]. There was a recent success of semi-invasive MRI-guided focused ultrasound (MRgFUS) thalamotomy in the treatment of movement disorders and other psychiatric disorders (depression and OCD) [102,103]. Such a technique would avoid the complications associated with the implantation of devices, which may necessitate explantation for reasons such as lack of efficacy or infection. Alongside DBS, MRgFUS offers a novel therapeutic avenue that should be explored to treat refractory addictive disorders.
Invasive techniques can offer a more focal region of therapy. Ablation techniques in drug addiction targeting the NAc with promising results have long received attention [104]. Indeed, a recent systematic review showed that NAc DBS has similar efficacy and safety profile as radiofrequency ablation of the NAc [66]. Compared to ablation techniques, however, DBS allows adjustment of stimulation parameters tailored to maximize therapeutic efficacy while minimizing side effects. Nonetheless, a recent systematic review showed that NAc DBS has similar efficacy and safety profile as radiofrequency ablation of the NAc [66].

Potential Biomarkers
The hunt for a reliable biomarker serves three critical purposes: to facilitate patient selection, to monitor the progress of the disease (and therapeutic effect of treatments), and to progress towards a holy grail of DBS-a closed-loop therapeutic system. Not only can this potentially increase the battery longevity (hence reducing the need for battery replacement), it also can increase the efficacy of stimulation by tailoring to the patients' neural responses. Further, a "tolerance" effect can be observed with chronic open-loop DBS stimulation in movement disorder patients, with the stimulation effect wearing off after a period of time [105]. Such a tolerance is also likely in patients receiving DBS to treat psychiatric illnesses. Closed-loop system with intermittent, dynamic stimulation can potentially minimize tolerance. Such biomarkers need to be easy to measure and possess a relatively high spatiotemporal resolution of measurement that, in turn, correlates well with clinical/behavioral changes. In our opinion, the most promising candidates would be electrophysiological and neurochemical measurements, although they will need to be tailored towards individual addictive substances and patients.
In terms of electrophysiological measurements, LFP is the most widely studied potential biomarker among DBS studies. LFP's wide use is mostly because of its measurement using the stimulating electrodes themselves. In urethane-anesthetized rats, NAc HFS suppressed pyramidal cell firing and enhanced slow LFP oscillations in the OFC via antidromic activation of cortico-striatal recurrent inhibition [87]. In addition, prolonged NAc HFS and LFS (delivered over 90 min) have been shown to affect both spontaneous and evoked LFPs in the medial PFC, lateral OFC, mediodorsal thalamus (MD), and NAc sites [106]. HFS also produced widespread increases in spontaneous beta and gamma power and enhanced coherent beta activity between MD and all other regions. In contrast, LFS elevated theta power in the MD and NAc. Acutely, HFS increased and LFS decreased induced relative gamma coherence. Stimulation-induced changes in gamma coherence shows that NAc DBS may achieve therapeutic effects by restoring the synchronicity of a neural circuitry that is disrupted in neuropsychiatric diseases. The differences in the effect on LFPs between HFS and LFS may also explain the differences in behavioral effects described above.
In a study assessing 7 people with heroin use disorder, pre-DBS electrophysiology revealed prominent theta and alpha band activity in both the NAc and ALIC [107]. Additionally, a beta band peak was detected in the power spectra of ALIC LFPs, which, the authors hypothesized, may represent the activity of striatal bridge cells. Furthermore, there was a negative correlation between the power of the theta band of ALIC LFPs and subjective craving, and a positive correlation between the power of the alpha band of NAc LFPs and depressive symptoms. LFPs of the NAc and ALIC both exhibited higher coherence values in the theta and alpha frequency bands. No comparison with post-DBS recordings was given in this study. Together with the animal and human DBS studies on NAc [108,109], the most consistent biomarker frequency is yet unknown. Further studies are required for clarification. New technologies such as the Medtronic Percept Stimulator [110] may be able to provide more information by simultaneously recording from and stimulating brain targets in humans. In addition, trials investigating electrophysiological changes secondary to DBS in addiction (such as [111]) will provide more insights.
Unlike electrophysiological signals (i.e., action potentials) that naturally lead to neurotransmitter release into synapses, quantification of synaptic-related neurotransmitter extracellular concentrations directly measures the activation of relevant neural circuits (such as the ones depicted in Fig. 2). Most studies in addiction focus on the monoamines, particularly dopamine, glutamate, and 5-HT. One animal study reviewed here [44] employed in vivo microdialysis and found that local dopamine concentrations increased after NAc shell HFS. While in vivo microdialysis has long been the gold standard for sampling neurochemicals, the process necessitates an external analyzer and has low temporal resolution. In contrast, in vivo electrochemical techniques, such as FSCV, provide potential benefits such as higher temporal resolution, less tissue damage, and the ability to be chronically implanted [112][113][114][115][116]. However, compared to microdialysis, electrochemical techniques have other limitations. Often, they are restricted to measurement of a single electroactive amine at a time, such as dopamine or serotonin. Thus, future studies which aim to develop biomarkers for DBS efficacy will need to demonstrate suitability of neurochemical biomarkers for translation and may be limited by present detection technology. While microdialysis provides important measurements of the tonic (basal) extracellular levels of neurochemicals, voltammetry can provide measurement of phasic and tonic measures of neurochemicals [117][118][119][120][121][122].

Future Preclinical and Clinical Research Considerations
Other important considerations for future research include inclusion of both genders in both animal and human studies (currently most studies focus on males-see Tables 1 and 2), and the use of larger animals-preclinical studies are predominantly rodent-based, but also well suited to large animal models. In addition, we have focused on technologies that are currently clinically viable in this review, namely electrical DBS devices. However, the use of optogenetics, where a viral vector is able to transfect a specific neural population, is increasingly popular in the preclinical study of underlying pathophysiology of different neuropsychiatric disorders and relevant therapies. Optogenetics has provided new insights into DBS protocol optimization for Parkinson's disease with translational potential [123,124], which may be extended to psychiatric illnesses. In addition, recent optical imaging techniques allow combination of fiber photometry and electrophysiology to study specific cell population activity via genetically encoded calcium indicators, including during deep brain stimulation [125,126]. Cell-specific viral transfection in calcium imaging or optogenetics helps to study specific ion channels or neurotransmitters. However, the use of viruses in humans is very limited. Further technological advancements are necessary to harness optical methods in the clinical setting.
Currently, the use of DBS in addiction remains off-label, primarily due to the lack of high-quality clinical evidence. As seen in Table 2, most human studies consist of observational studies with a small sample size. However, more clinical trials are currently being conducted. The ClinicalTrials. gov database show that there are currently around 10 DBS trials underway trying to tackle this paucity of scientific evidence, targeting the NAc, ALIC, and STN [127]. As demonstrated in the present review, they are all very promising targets. In addition, we would add PFC to this list given the success from animal studies [47] as well as human imaging studies showing its dysfunction in subjects with substance use disorder [128].
Supporting the potential of DBS in alleviating addiction-related behaviors, DBS to treat obesity has also shown early successes [129][130][131]. Given the similarities between substance use disorders and obesity, the findings in DBS for substance use disorder presented so far may be translatable to DBS for obesity. With the increasing prevalence and comorbidities associated with excessive calorie consumption in the modern society, this can be hugely beneficial.
While it is exciting to perform cutting-edge patient surgeries, it is also important to recognize that DBS is not an immediate cure and requires careful patient selection, consent, and patient-based optimization (including treating other comorbidities); otherwise, DBS can lead to a risk of major adverse outcomes [132]. DBS also has cost, translational, and capacity limitations for a highly prevalent disorder, but has the advantage of elucidating fundamental neural biology capable of detecting new therapeutic targets.
In our literature search, all human studies that involve stimulation of deep structures were conducted exclusively with high-frequency stimulation, as commonly used in DBS for movement disorders (see Table 2). Therefore, future work to compare how stimulation at different frequencies may affect the biomarkers such as LFP in humans would be beneficial.
The duration of trials may also be important. One of the possible reasons attributed to the failure of early trials in DBS for depression is the short treatment duration, as trials with longer duration appeared to show significant efficacy [6,[133][134][135]. DBS duration effects will need to be assessed in larger clinical trials. Further, for neuropsychiatric conditions such as addiction, careful consideration of the ethical aspect is required.

Ethical Considerations
Although DBS is much safer than early forms of psychosurgery such as lobotomy, it is not a risk-free treatment [136]. Therefore, clinicians are obliged to always have the patients' best interest in their mind. The risks and benefits of the treatment need to be balanced appropriately. One of the main issues with human trials is the principle of autonomy, as patients with addictive disorders may not have the capacity to consent to such process. Furthermore, there are concerns that DBS (and indeed any forms of psychosurgery) may alter a patient's personality, which can lead to actions and consequences that may otherwise not be performed [137][138][139]. Academic societies and government institutions should also provide a medico-legal framework to ensure both medical professionals and patients are protected. A full ethical discussion is out of the scope of this review. Nevertheless, we recommend that the clinical team be multidisciplinary, with access to the resources and expertise to assess and manage the patients before and after the DBS surgery.

Conclusions
Preclinical and clinical studies suggest DBS is a potential treatment for refractory addictive disorders after other noninvasive options fail. Here, we reviewed these studies and their implications. However, given the uncertainty in its mechanism of action and a lack of large scale randomized controlled trials, clinicians and scientists should be cautiously optimistic. We also provided recommendations on the directions of future research, such as the need to identify reliable biomarkers, for which LFPs in the NAc and in vivo electrochemical measurement of monoamines so far appear the most promising options. With further research, we believe it can prove to be a highly effective treatment for a selected group of patients.