Deep brain stimulation for refractory major depressive disorder: a comprehensive review

Abstract

Deep brain stimulation (DBS) has emerged as a promising treatment for select patients with refractory major depressive disorder (MDD). The clinical effectiveness of DBS for MDD has been demonstrated in meta-analyses, open-label studies, and a few controlled studies. However, randomized controlled trials have yielded mixed outcomes, highlighting challenges that must be addressed prior to widespread adoption of DBS for MDD. These challenges include tracking MDD symptoms objectively to evaluate the clinical effectiveness of DBS with sensitivity and specificity, identifying the patient population that is most likely to benefit from DBS, selecting the optimal patient-specific surgical target and stimulation parameters, and understanding the mechanisms underpinning the therapeutic benefits of DBS in the context of MDD pathophysiology. In this review, we provide an overview of the latest clinical evidence of MDD DBS effectiveness and the recent technological advancements that could transform our understanding of MDD pathophysiology, improve the clinical outcomes for MDD DBS, and establish a path forward to develop more effective neuromodulation therapies to alleviate depressive symptoms.

INTRODUCTION

Major depressive disorder (MDD) is a debilitating neuropsychiatric syndrome that affects an estimated 6% of the population per year worldwide and up to one in six adults over the course of a lifetime [1]. Women are approximately twice as likely as men to be diagnosed with MDD [2]. As of 2017, MDD was the leading cause of disability worldwide when measured by years lived with disability [3], and MDD has been associated with an estimated 60–80% increase in mortality when compared to the general population [4, 5].

The cardinal symptoms of MDD are depressed mood and diminished interest or pleasure (anhedonia), according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-V). However, each individual with MDD will likely experience a multifaceted combination of emotional, cognitive, and/or physical symptoms (Fig. 1). A patient must experience the core symptoms plus at least five of these symptoms over a two-week period to be formally diagnosed with MDD. The severity of MDD symptoms is predominantly measured using validated clinical rating scales, such as the Hamilton Depression Rating Scale (HDRS) [6] or the Montgomery-Åsberg Depression Rating Scale (MADRS) [7]. MDD is usually an episodic disease; up to two-thirds of patients experience at least one recurrent episode, with approximately 15% of patients not recovering (i.e., no episode-free years since the first episode) [8, 9]. Additionally, the burden of MDD can be further heightened by other comorbid psychiatric disorders, such as anxiety and psychosis [10, 11]. Suicide is a major concern in MDD [12], with a lifetime prevalence of 2.2–8.6% in depressed patients, compared to 0.5% in the general population [13]. The complex heterogeneity of symptoms, coupled with variability in episode chronicity and severity, makes MDD a challenging disorder to treat effectively.

Fig. 1. Symptoms of MDD.

The core symptoms (depressed mood and decreased interest or pleasure (anhedonia)), along with five other symptoms, are required for diagnosis of MDD according to the DSM-V. The combination of other symptoms is heterogeneous across patients and may include emotional, cognitive, and physical symptoms.

First-line treatments for MDD are psychotherapy, medications, or a combination of the two depending largely on the severity of the depressive episode [14]. Although psychotherapy and medications can be successful in the majority of patients with MDD, approximately 30% of patients experience refractory symptoms [15, 16]. Although definitions vary, refractory MDD is most commonly defined as having persistent symptoms despite undergoing at least two adequate trials (i.e., sufficient dose and duration) of first-line antidepressants [16]. For select patients with severe, refractory MDD, deep brain stimulation (DBS) can provide substantial antidepressant effects [17]. DBS is an invasive neuromodulation therapy that involves delivering electrical stimulation to deep structures of the brain using surgically implanted electrodes attached to a pulse generator. DBS for MDD remains experimental and has not received official approval for clinical use by any regulatory agencies worldwide. Open-label studies and meta-analyses have collectively reported beneficial outcomes. However, outcomes of randomized controlled clinical trials have been variable and several challenges remain preventing the widespread adoption of DBS therapy for MDD.

This review provides a comprehensive overview of the latest clinical evidence for DBS for MDD and outlines recent technological advancements that could transform our understanding of MDD pathophysiology, improve the clinical effectiveness of DBS for MDD, and establish more effective treatments. Other invasive and non-invasive neuromodulation therapies, including electroconvulsive therapy (ECT) [18], transcranial magnetic stimulation (TMS) [19], and vagus nerve stimulation (VNS) [20] are used to treat refractory MDD but are outside the scope of this review.

Pathophysiology of major depressive disorder

Much of the evidence to explain MDD pathophysiology has been derived from comparing patients with MDD to healthy controls using electrophysiology (e.g., EEG) and neuroimaging modalities that measure structural, functional, and metabolic alterations either at rest or during behavioral tasks [21–24]. Animal models have also been instrumental in understanding specific domains of MDD [25]. Although the pathophysiology of MDD is not fully understood, the collective evidence strongly suggests that MDD is a network-level disorder affecting intrinsically connected networks that mediate emotional and cognitive processing.

Various brain network-based models have been proposed to explain the complex pathophysiology of MDD [26–28]. To discuss the network-based rationale for MDD DBS, we focus on four main networks implicated in MDD (recently reviewed by Li et al. [29]): the affective network, the default mode network, the reward network, and the cognitive control network (Fig. 2A). Dysfunction in these networks is often summarized as either hyper- (increased) or hypo- (decreased) connectivity and has been hypothesized to underpin symptoms associated with MDD. For simplicity, we have classified each network as generally exhibiting hypo- or hyperconnectivity in MDD compared to healthy controls based on existing neuroimaging evidence. However, it should be noted that these networks are dynamic under various time scales and processes, and therefore the pathophysiology of MDD is likely more complex, with a mix of hypo- or hyperconnectivity, as well as other measures of connectivity (e.g., network density and global versus local connectivity), within or across defined networks. Hyperconnectivity of the affective network (also known as the ventral limbic network) is thought to underlie depressed mood and emotion dysregulation in MDD [30–32]. Notably, the insula is also often defined alongside the midcingulate cortex in the salience network, which has also shown hyperconnectivity in individuals with MDD and has been associated with heightened responses to negative stimuli in MDD when compared to healthy populations [33]. The default mode network, which is mainly active during passive self-referential processes, has also shown hyperconnectivity in patients with MDD and is thought to give rise to negative rumination [34–36]. In contrast, hypoconnectivity of the reward network (or fronto-striatal network) is thought to give rise to the loss of pleasure, interest, or motivation (anhedonia) in MDD [31, 37, 38]. Finally, hypoconnectivity of the cognitive control network, has also been demonstrated in patients with MDD [39–41] and has been associated with reduced cognitive control, including dysfunction in regulating thoughts, aligning actions with internal goals, and exerting control over emotional regulation. Additionally, the somatomotor network has also shown hypoconnectivity in patients with MDD and may underpin some of the psychomotor deficits in MDD [42].

Fig. 2. DBS targets to modulate pathophysiological networks implicated in MDD.

A Hyperconnectivity (pink arrows) of the affective network and the default mode network (left) and hypoconnectivity (blue arrows) of the reward network and the cognitive control network (right) have been linked to the pathophysiology of MDD. Differences in individual connections within these networks exist across studies; therefore, this figure is a simplification and represents the general agreement about hyper- and hypoconnectivity of these networks in MDD. Anatomical regions denoted by circles are in approximated locations. B Brain regions targeted with DBS for MDD relative to cortical and subcortical structures implicated in MDD pathophysiological networks. Regions are colored by whether they are involved in predominantly hyperconnected networks (pink), hypoconnected networks (blue), or a combination (blue and pink). Anatomical regions are in approximated locations overlaid on a single sagittal slice from an ultra-high resolution (500 μm) 7 T MRI atlas [133]. C Connectivity of DBS targets to cortical and subcortical regions implicated in MDD pathophysiology colored based on hyper- or hypoconnectivity to the networks in (A) (simplified depiction based on tract tracing studies in humans and animals summarized in [97]). Abbreviations: ACC anterior cingulate cortex, Amg amygdala, BNST bed nucleus stria terminalis, Caud caudate, dACC dorsal anterior cingulate cortex, dlPFC dorsolateral prefrontal cortex, DR dorsal raphe, Hip hippocampus, Ins insula, IPC inferior parietal cortex, ITP inferior thalamic peduncle, LHb lateral habenula, LC locus coeruleus, MFB medial forebrain bundle, mPFC medial prefrontal cortex, NAcc nucleus accumbens, OFC orbitofrontal cortex, PC parietal cortex, PCun precuneus, PCC posterior cingulate cortex, PFC prefrontal cortex, Put putamen, SCC subcallosal cingulate, Thal thalamus, VC/VS ventral capsule/ventral striatum, VTA ventral tegmental area.

The models of MDD pathophysiology collectively suggest imbalance among widespread, intrinsically-connected networks implicated in various emotional and cognitive processes. Thus, DBS has been proposed as a therapy for MDD in a deliberate effort to modulate the neural activity within these networks in order to improve depressive symptoms.

Deep brain stimulation for major depressive disorder

Targets to modulate network function in MDD.

Several brain regions to treat refractory MDD with known structural and/or functional connectivity to key nodes in the MDD brain networks have been investigated for DBS (Fig. 2B). The subcallosal cingulate cortex (SCC; also referred to as subgenual cingulate or Brodmann’s Area 25) has been most studied and was the first target used for human DBS for MDD [43]. The SCC is interconnected with the affective, salience, and default mode networks involving the PFC, ACC, OFC, insula, and subcortical regions. The second most commonly studied target is the ventral capsule/ventral striatum (VC/VS) [44] (also referred to as the ventral anterior limb of the internal capsule (vALIC) and includes the nucleus accumbens (NAcc) [45]). The superolateral branch of the medial forebrain bundle (slMFB) has been increasingly used [46]. The VC/VS and the slMFB are interconnected with the reward network, involving the striatum, PFC, and subcortical regions. Notably, the SCC, the VC/VS, and the slMFB are also all interconnected. Other targets are being explored, including the inferior thalamic peduncle (ITP) [47], the bed nucleus stria terminalis (BNST) [48], and the lateral habenula (LHb) [49]. All of the described brain targets have shown promising evidence of clinical efficacy in open-label trials in select patients, which suggests that DBS targeted to different brain regions can improve depression by modulating neural activity in pathophysiological networks implicated in MDD.

The targets investigated for MDD DBS exhibit different connectivity profiles to pathophysiological MDD networks but with considerable overlap and interconnectivity between targets (Fig. 2C). Neuroimaging studies have shown commonalities and differences in MDD-related networks modulated by different DBS targets. Stimulation in either the SCC or the VC/VS has been shown to decrease activity in the SCC [43, 50–52] and in the NAcc/VC/VS [52, 53], pointing to modulation of a common network across targets, although some studies report increased activity in these regions following DBS [54, 55]. Generally, a positive clinical response to SCC DBS has been associated with decreased activity in the SCC, medial PFC, OFC, and the insula following chronic DBS [43, 50, 56], which suggests DBS may improve depression by reducing MDD-related hyperactivity in the affective, salience, and default mode networks. The network effects of DBS targeted to the VC/VS or the slMFB are less clear; neuroimaging studies have reported either increased or decreased activity in the SCC, PFC, OFC, ACC, and basal ganglia regions following DBS [51–54, 57, 58], highlighting the complexities of understanding the modulatory effects of DBS within and across the pathophysiological networks of MDD.

Although previous studies have focused on simple decreases or increases in network activity following DBS, depression improvement likely involves complex interactions within and across networks that may change over time in response to stimulation. Emerging evidence in the SCC supports the hypothesis that the mechanisms of DBS for MDD likely involve dynamic neurophysiological changes that may reflect alterations in network activity associated with depression improvement [59]. Further studies are needed to understand the dynamic local and network mechanisms underlying depression improvement with DBS, especially in targets beyond the SCC.

Patient selection.

MDD is a highly heterogeneous disorder, and the MDD subpopulation most likely to respond to DBS is not well defined. Inclusion criteria for DBS for MDD have been established through clinical studies and have been generally consistent [60]. Typical inclusion criteria include: (1) severe MDD, defined by an HDRS score of >= 20 or MADRS score of >34 averaged over 4 weeks, (2) substantial functional impairment, defined by the Global Assessment of Functioning score [61] of >= 50, (3) current depressive episode lasting at least 12 months, (4) failure of at least four adequate antidepressant treatments, defined by >3 on the Antidepressant Treatment History Form [62], including documented ECT failure or intolerance, and (5) low risk of suicide, although patients exhibiting suicidal ideation have been considered eligible. However, given the heterogeneity of the disorder and comorbidities, particularly anxiety disorders and personality disorders, patient evaluation and medical record review by a multidisciplinary team is crucial to verify eligibility given the intensive and invasive nature of DBS; this often includes evaluation by two psychiatrists, a neuropsychologist, and a neurosurgeon. Additionally, many research teams include an independent medical ethicist to review medical records and screening data to confirm eligibility and identify any ethical concerns to be addressed prior to device implantation.

Clinical efficacy.

The clinical efficacy of DBS for MDD has been investigated in several open-label studies and in a few randomized controlled trials (RCTs). The largest meta-analysis to date, which summarized the results from 14 open-label studies and three RCTs (N = 233 patients) across multiple surgical targets, reported an overall statistically significant improvement in depressive symptoms (as measured by the HDRS or the MADRS) [63]. The meta-analysis reported that 56% (range 43–69%) of patients met the criteria for response (>50% reduction in symptoms) and 35% (range 27–44%) of patients were considered in remission, while only 14% (range 4–25%) experienced recurrence of symptoms. The meta-analysis also highlighted that although MDD DBS had a favorable safety profile, adverse events and suicide emerged as crucial considerations when monitoring this patient population.

Several long-term studies following patients many years after DBS surgery have also reported sustained clinical benefit, including reduced depression symptoms and improved quality of life (Table 1) with DBS targeted to the SCC [64–67], the VC/VS [44, 68, 69], the vALIC [70, 71], and the slMFB [72–75]. Examining data at the latest follow-up post-DBS, 32–80% of patients were considered responders and 20–73% of patients were in remission across all studies. Collectively, these studies demonstrate that DBS can induce robust and sustained antidepressant effects, which is especially striking in a population whose MDD symptoms were otherwise refractory to all other treatments.

Table 1.

Long-term follow-up studies evaluating the efficacy of DBS for MDD.

Target	Reference	Number Patients of	Latest Follow-up	Outcome at Latest Follow-up (% Responders / % In Remission)	Main Conclusions
Subcallosal Cingulate Cortex	Kennedy et al. (2011)	20	3–6 years	64.3% / -	SCC DBS showed long-term efficacy, which translated to improvement in quality of life
	Bogod et al. (2014)	4	42 months	50% / -	SCC DBS was associated with general stability in cognitive abilities over time in long-term follow-up
	Crowell et al. (2019)	28	8 years	76% / 50%	SCC DBS provided robust and sustained antidepressant effects in the majority of patients
	Aibar-Dúran et al. (2022)	17	5 years	65% / 35%	SCC DBS showed long-term efficacy in the majority of patients
VC/VS	Malone et al. (2009)	15	6 months to 4 years	53.3% / 40%	VC/VS DBS showed long-term sustained clinical effects on depression
	Bewernick et al. (2012)	11	12–24 up to 48 months	45% / -	NAcc DBS showed long-term efficacy, including improvements in quality of life and anxiety
	Hitti et al. (2021)	8	1.7–11.9 years	50% / 25%	VC/VS DBS provided meaningful and sustained long-term clinical benefit for several patients
vALIC	van der Wal et al. (2020)	25	2 years	32% / 20%	vALIC DBS showed continued efficacy in long-term follow-up
	Bergfeld et al. (2022)	25	6–9 years	44% / -	vALIC DBS showed long-term efficacy, which translated also to improvement in quality of life
slMFB	Bewernick et al. (2017)	8	1–4 years	75% / 50%	slMFB DBS induced acute and sustained antidepressant effects
	Bewernick et al. (2018)	21	5 years	73% / 73%	slMFB DBS induced sustained antidepressant effects and did not change patients’ personalities
	Fenoy et al. (2018)	6	1 year	66% / 66%	slMFB DBS induced acute and sustained antidepressant effects
	Fenoy et al. (2022)	5	5 years	80% / -	slMFB DBS induced acute and sustained antidepressant effects

Despite promising results from open-label and long-term follow-up studies, nine RCTs investigating the clinical efficacy of DBS for MDD have yielded variable conclusions (Table 2) [76–89]. Although all of the RCTs have reported statistically significant reductions in depression severity with DBS in open-label follow-up when compared to preoperative baseline, only half of studies reported greater symptom improvement with active stimulation when compared to sham. Two of the largest RCTs failed to reach their primary endpoints, including the BROADEN trial (N = 90 patients implanted in the SCC) [79] and the RECLAIM trial (N = 30 patients implanted in the VC/VS) [83]. The BROADEN trial showed that only 20% of patients receiving active SCC DBS for 6 months reached the criteria for response (defined as a 40% reduction in MADRS), which fell short of the primary endpoint of 40% of patients achieving response; however, after 24 months of long-term follow-up, the proportion of responders increased to 49% [79]. The RECLAIM trial similarly observed that only 20% of patients receiving active VC/VS DBS for 16 weeks reached the criteria for response (defined as a 50% reduction in MADRS), with a slight improvement to 23.3% responders in long-term follow-up for 24 months [83]. The variability in results across trials has made it challenging to generalize their findings or draw definitive conclusions about the efficacy of DBS for MDD.

Table 2.

Randomized controlled trials (RCTs) evaluating the efficacy of DBS for MDD.

Target	Reference	Number of Patients	Study Design	Study Design Description	Time to Primary Outcome	Latest Follow-up	Outcome at Latest Follow-up (% Responders / % In Remission)	Significant Difference Between Active vs. Sham?	Main Conclusions
Subcallosal Cingulate Cortex	Holtzheimer et al. (2012)	17 (N = 10 MDD; N = 7 bipolar depression)	SB / Cross-over	Patients received single-blind sham stimulation for 4 weeks, followed by open-label active stimulation for 24 weeks, then single-blind discontinuation of stimulation	4 weeks	2 years	92% / 58%	Not directly tested	Sham DBS significantly reduced depression symptoms compared to baseline, but not compared to post-surgery/pre-DBS scores; chronic DBS significantly reduced depression severity; discontinuation of DBS caused relapse in 3/3 patients tested
	Merkl et al. (2013)	6	DB / Randomized / Cross-over	Patients randomized to active stimulation or sham for 24 h, followed by open-label active stimulation for 24 weeks	24 hours	36 weeks	33% / 33%	No	No significant difference in antidepressant effects with active vs. sham DBS
	Puigdemont et al. (2015)	5	DB / Randomized / Cross-over	Patients with stable clinical response to DBS randomized to OFF-ON or ON-OFF (3 months each phase)	6 months	6 months	100% / 80% (only responders included in cross-over phase)	Yes	Continuous DBS was required to maintain antidepressant effects
	Holtzheimer et al. (2017)	90	DB / Randomized / Parallel	Patients randomized to active stimulation or sham (6-month delayed onset stimulation)	6 months	30 months	48% / 25% (active group)	No	No significant difference in antidepressant effects in active vs. sham groups; modest improvement with longer open-label follow-up
	Merkl et al. (2017)	8	DB / Randomized / Parallel	Patients randomized to active stimulation or sham (4 weeks delayed onset stimulation)	8 weeks	28 months	33% / 33%	No	No significant difference in antidepressant effects in active vs. sham groups; modest improvement with longer open-label follow-up
	Eltan et al. (2018)	9	DB / Randomized / Cross-over	Patients randomized to high frequency or low frequency stimulation; if non-responder, crossed over to the other frequency; if responder, maintained same frequency	6 months	12 months	50% / - (only 6/9 reached 12-month follow-up)	Not directly tested	DBS significantly reduced depressive severity overall; no significant difference in antidepressant effects with high vs. low frequency; long-term follow-up showed higher improvement with high-frequency DBS
	Ramasubbu et al. (2020)	22	DB / Randomized / Cross-over	Patients randomized to short PW or long PW; if non-responder, crossed to other group; if responder, maintained same PW	12 months	12 months	50% / 27%	Not directly tested	DBS significantly reduced depression severity; no significant difference in antidepressant effects with short vs. long pulse width
VC/VS	Dougherty et al. (2015)	30	DB / Randomized / Parallel	Patients randomized to active stimulation or sham (16-week delayed onset stimulation)	16 weeks	24 months	23% / 20%	No	No significant difference in response rates or antidepressant effects in active vs. sham groups; modest improvement with longer open-label follow-up
vALIC	Bergfeld et al. (2016)	25	DB / Randomized / Cross-over	A open-label stimulation parameter optimization phase (up to 52 weeks max); then, patients randomized to OFF-ON or ON-OFF (6 weeks each phase)	52 weeks (responder rate); 64 weeks (active vs. sham)	52 weeks	40% / 20%	Yes	Significantly higher antidepressant effects with active stimulation compared to sham
slMFB	Fenoy et al. (2016)	4	SB / Cross-over	Patients received single-blind sham stimulation for 4 weeks, followed by open-label active stimulation for 48 weeks	4 weeks	26 weeks	66% / -	Yes	Sham DBS reduced depression severity but the effect was not statistically significant; antidepressant effects were rapid (~1 week of stimulation); chronic DBS reduced depression severity
	Coenen et al. (2019)	16	DB / Randomized / Parallel	Patients randomized to active stimulation or sham (2-month delayed onset stimulation)	12 months	12 months	100% / 50%	Yes	DBS significantly reduced depression symptoms compared to baseline; no significant difference in response rates in active vs. sham groups during blinded randomized phase
BNST/IC and ITP	Raymaekers et al. (2017)	7	DB / Randomized / Cross-over	Patients underwent IC/BNST optimization, then randomized to OFF-ON IC/BNST or ON IC/BNST-OFF (1 week each phase). Then patients underwent ITP optimization, then randomized to a permutation of OFF-IC/BNST-ITP (2 months each phase).	1 week (OFF vs. IC/BNST); 2 months (OFF vs. IC/BNST vs. IT)	63 months	80% / 40%	No	No significant difference in antidepressant effects in active vs. sham DBS or across targets. Long-term outcomes were favorable, with the majority of patients preferring IC/BNST DBS over ITP DBS.

Although disappointing, these negative results of recent RCTs have highlighted the challenges of studying the effect of DBS in this population and have offered an opportunity to reevaluate how to approach DBS in psychiatric indications and how to design trials that properly control for potential factors impacting the patients’ clinical response [90–94]. As one recent commentary pointed out, “these [trials] are examples of failed studies and not failed treatments” [91].

Several themes have emerged regarding lessons learned from previous trials and how the field could potentially improve future trials of MDD DBS:

1. Refining and verifying lead placement relative to relevant neuroanatomical structures

   Given the numerous MDD DBS targets and potential variability in targeting approaches across centers, characterization of DBS lead placements across patients in each trial cohort will be crucial to verify that nonresponders’ leads are placed in the intended target. Additionally, we must understand how variability in lead locations relative to neuroanatomy and key fiber pathways may contribute to clinical response. This is a crucial step toward refining the optimal target for MDD DBS.

2. Determining appropriate patient selection criteria

   Patient selection criteria for DBS includes a history of persistent, severe symptoms despite multiple adequate treatment attempts. However, a subpopulation of patients with refractory MDD might be more likely to respond to DBS. For example, the degree of treatment resistance may be an important consideration; some trials for other neuromodulation therapies (e.g., TMS and VNS) have limited treatment resistance as an important criteria for study eligibility [95, 96], as outcomes of MDD therapies were generally worse for patients with higher treatment resistance [15]. The degree of response to ECT or TMS (or less commonly used VNS) may ultimately be useful to determine the patients’ likelihood to respond to neuromodulation therapies in general and to evaluate eligibility for DBS. Additionally, some symptom profiles may be better suited for DBS and for certain DBS targets, especially given the heterogeneity in MDD symptoms across patients.

3. Study design and when to assess symptoms for the primary outcome

   Previous trials have included a mix of parallel designs (i.e., patients randomized to active or sham groups) and cross-over designs (i.e., patients randomized to active or sham condition first, then crossed over to the other condition). The parallel design, however, relies on differences between individual patients at a group level, which can prove challenging in a population such as patients with MDD whose symptoms are highly heterogeneous. In contrast, the cross-over design allows for both within-patient contrasts (i.e., individual patient’s symptom severity during active versus sham stimulation) and across-patient differences (i.e., group-level symptom severity during active versus sham stimulation). However, cross-over trials comparing on/off DBS must be designed with ample washout time to account for any longitudinal effects induced by DBS that may result from neuroplastic changes in brain networks in response to stimulation [97]. The optimal washout time for MDD DBS is unknown and may vary across individuals and across DBS targets.

Thus, we would argue that alternative study designs are worth considering. One potential design is a double-blind stepwise discontinuation sham condition (in which optimized stimulation parameters are progressively decreased over several weeks) [98]. Another potential trial design could be comparing treatment as usual (TAU) to active stimulation while controlling for medications, as it is unknown if certain medications inhibit or amplify the effects of DBS. Regardless of study design, using a single time point as the primary outcome may over- or under-estimate the clinical benefit from DBS. One alternative statistical approach could be to analyze the outcomes over the course of the treatment as repeated measures or area under the curve [94].

The ultimate goal of RCTs is to establish clear evidence that MDD DBS is safe and effective for reducing depression symptoms; thus, future RCTs must be carefully designed while considering key lessons learned from previous trials and continuing to incorporate relevant research findings and technological developments.

Technology and data-driven approaches to guide DBS for MDD

Despite promising evidence of clinical efficacy and sustained benefits of DBS for patients with refractory MDD, several remaining challenges may hinder its widespread adoption as a therapy. MDD symptoms are inherently internal and personal to the patient, and therefore cannot be easily measured objectively by an external evaluator (in contrast to motor symptoms like tremor). Establishing methods to measure MDD symptoms objectively with both accuracy and specificity will be crucial for evaluating the therapeutic effects of DBS. Characteristics of the patient population most likely to respond to DBS for MDD are unclear, which limits clinicians’ ability to discriminate between patients who will respond well to DBS from patients who will not benefit. Although several brain regions have been investigated as targets for DBS for MDD, it is unclear how to select the best target for each patient. MDD is a highly heterogeneous disorder; each patient experiences a unique combination of symptoms which may fluctuate over time. This heterogeneity, coupled with an incomplete understanding of MDD pathophysiology, suggests that DBS for MDD must be tailored to address disease-specific, symptom-specific, and patient-specific needs in order to advance its widespread use.

Altogether, these challenges highlight the need for novel, objective approaches to guide DBS for MDD. In particular, approaches combining both technology (to drive objectivity and innovation) and neuroscience (to drive specificity to MDD pathophysiology and clinical translation) will be crucial to move the field of MDD DBS forward. Fortunately, many new lines of research in MDD and DBS for MDD have aimed to address these challenges using behavioral and biometric measures, neuroimaging, electrophysiology, and multimodal methods which may uncover the pathophysiology of MDD, personalize treatment, and develop approaches to optimize DBS for MDD (Fig. 3).

Fig. 3. Symptom markers, neuroimaging, and electrophysiology to advance DBS for MDD.

A Examples of candidate signals that show potential as markers of depression symptom severity, including facial expression (via video), speech analysis (via audio recordings), smartphone-based metrics, and wearable sensor-based metrics. B Top: Neuroimaging studies have revealed regions showing preoperative and/or postoperative alterations in regional volume, neurotransmitter levels, or brain activity (glucose metabolism, cerebral blood flow, or blood oxygenation level dependent (BOLD) activity) that were correlated with improvement in MDD symptoms with DBS. This is a simplified depiction of anatomical regions. Bottom: In an example case as part of a clinical trial, intracranial EEG and DTI using fiber tractography were used to identify a therapeutic network involving the amygdala (recording site for marker of depressive symptoms) and the VC/VS (stimulation site) to target with DBS for MDD [130]. DBS electrodes capable of directional stimulation may aid in steering current toward the therapeutic fiber pathways. C Top: Local field potential (LFP) recordings may be used to evaluate the neurophysiological effects of DBS and identify markers of therapeutic stimulation. In this example from Sendi and Waters et al. [125], left SCC DBS reduced beta power compared to OFF DBS, and this reduction in beta power was correlated with greater symptom decrease 1 week after surgery. Bottom: Evoked potentials elicited by stimulation in the VC/VS and recorded from the amygdala using intracranial EEG to identify functional connectivity within the therapeutic network for DBS and guide stimulation parameter selection to optimize network engagement [130]. Abbreviations: ACC anterior cingulate cortex, Amg amygdala, Caud caudate, LHb lateral habenula, OFC orbitofrontal cortex, PCC posterior cingulate cortex, PCun precuneus, PFC prefrontal cortex, SCC subcallosal cingulate, Thal thalamus.

Behavioral and biometric markers of depressive symptoms.

Current evaluation of MDD symptoms is performed using validated clinical rating scales, such as the HDRS or the MADRS; although inexpensive, these scales are administered by clinicians, rely on patients’ recall of their symptoms, and are designed to evaluate symptoms on the timescale of weeks, which will not detect day-to-day or moment-to-moment symptom fluctuations. These fluctuations will likely be important to evaluate the overall stability of patients’ symptoms to help distinguish depression worsening versus relapse, or depression improvement versus remission. Identifying markers which can accurately track longitudinal symptom severity at a finer temporal scale could supplement clinical rating scales by providing an objective window into the patient’s wellbeing in their day-to-day life and could be used to better understand and to predict patient responses to DBS.

Recent studies including patients who underwent DBS for MDD have focused on automated evaluation of facial expression, language, and speech as potential markers of symptom severity and treatment response (Fig. 3A). A recent study found that facial recognition, using videos at baseline and after up to 6 months of SCC DBS, could successfully classify remission and clinical response following treatment [99]. Nonverbal behaviors, including increased reactivity and fidgeting/engaging behavior, were also associated with depression severity and showed significant reduction following three months of SCC DBS therapy [100]. Additionally, attributes of language and speech detected from clinical interview recordings have shown potential as markers of response. Patients with refractory MDD showed changes in familiar speech (a measure of social connectedness) [101] and a shift from mainly sad speech to predominantly excited and neutral speech [102] following SCC DBS.

Smartphone devices and wearable sensors have been increasingly explored to track depression symptoms in patients with MDD (Fig. 3A). Smartphone-based markers that have been associated with depression include application use, overall screen time, metrics of social interactions (e.g., number of text messages), and geospatial data via GPS [103]. Promising wearable-based markers of depression have also been identified and include heart rate or heart rate variability, skin conductance or temperature, sleep and circadian rhythm patterns, and metrics of movement (e.g., number of steps) [104]. Most studies thus far have combined metrics into machine learning models to predict depression diagnosis or symptom severity. However, it is unclear how these models will perform in longitudinal prediction spanning multiple MDD episodes and in the context of DBS for MDD in a population with severe, refractory symptoms.

Although preliminary, these behavioral and biometric markers could serve as noninvasive, inexpensive, and objective methods to evaluate the longitudinal response to MDD DBS, which could help clinicians decide when and how to change stimulation parameters and could supplement clinical rating scales in future MDD DBS studies.

Neuroimaging: patient selection, correlates of clinical response, and DBS targeting.

Neuroimaging in the context of DBS for MDD has been used to noninvasively evaluate baseline structural morphology and brain activity, as well as to evaluate acute and longitudinal changes in response to stimulation. Neuroimaging could play a role in establishing markers to guide patient selection, identifying correlates of clinical response to stimulation, and delineating effective stimulation targets.

Several studies have investigated if preoperative neuroimaging could be used to identify candidates who are likely to respond to DBS for MDD, the majority of which focus on the SCC target. A key region linked to the response to SCC DBS is the target itself; responders to SCC DBS show increased baseline glucose metabolism [56] and regional volume [105] in the SCC region when compared to nonresponders. Another region closely related to the SCC is the rostral ACC, which has shown decreased glutamate concentrations in MR spectroscopy [106] and increased glucose metabolism in FDG-PET [55] in responders when compared to nonresponders with SCC DBS. Other potential preoperative predictors of response to SCC DBS include decreased regional volume in PFC, OFC, and precuneus/occipital cortex [55], increased cerebral blood flow in the PFC [53], and increased thalamic and amygdalar volumes [105]. A recent study reported a correlation between the response to slMFB DBS and the preoperative volume in a left fronto-polar/OFC region, which showed connectivity to the slMFB and the SCC target [107]. Collectively, these studies indicate pre-treatment alterations in structural morphology and brain activity in regions implicated in MDD pathophysiology may predispose some patients to responding better to DBS; however, prospective validation will be required for translation and more studies on other targets beyond the SCC are needed.

Neuroimaging has also been used to identify functional and structural correlates of the antidepressant response to stimulation. In addition to abnormal baseline activity, PET studies have shown a reduction in SCC cerebral blood flow with 3 months and 6 months of therapeutic SCC stimulation [43, 50]. Similarly, a study using simultaneous SCC DBS during fMRI found that DBS-induced decreases in activity in the dorsal ACC, posterior cingulate, and precuneus predicted long-term response [108]. However, correlations between response and the effects of DBS on SCC glucose metabolism have been inconsistent; some recent studies have reported decreases [56] and increases [50, 55] in SCC glucose metabolism in responders, which may reflect variability in follow-up scan time points or in the effects of DBS on cerebral blood flow and glucose metabolism. Recent studies suggest the habenula may also mediate the response to SCC DBS, with responders exhibiting increased habenula volume in longitudinal follow-up [109]. One recent study also evaluated the effects of slMFB DBS on glucose metabolism and showed a reduction in right caudate metabolism correlated with clinical response after 12 months of slMFB DBS [58]. Although findings vary across studies, DBS-induced changes in functional and metabolic activity measured using neuroimaging could potentially be used to objectively verify the clinical response over time and provide insight on the potential mechanisms of DBS for MDD. A summary of showing preoperative and/or postoperative alterations in brain activity is shown in Fig. 3B, highlighting the potential role of MDD-related brain networks in mediating the clinical response to DBS.

The use of diffusion-weighted imaging to delineate white matter fiber pathways associated with response has gained substantial recognition [110–112]. In a seminal study of patients who underwent SCC DBS for MDD [112], probabilistic tractography of pathways emanating from the site of stimulation revealed four critical bilateral pathways implicated in the antidepressant response: the forceps minor and the uncinate fasciculus (projecting to the mPFC), the cingulum bundle (projecting to the rostral and dorsal cingulate cortex), and subcortical pathways (projecting to the ventral striatum, putamen, hypothalamus, and anterior thalamus). Recent studies have further confirmed the role of these pathways in the chronic effects of DBS [113] and in the acute effects of intraoperative stimulation on mood and behavioral changes [114]. Surgical targeting for SCC DBS has since transitioned from anatomical landmark-based targeting to tractography-based targeting [115], which has yielded greater and more rapid antidepressant effects of DBS [116]. Another example is DBS targeted to the slMFB, which has incorporated tractography into surgical planning for target identification [117, 118]. The SCC and slMFB targets are translational success cases that highlight the potential benefits of neuroimaging-based target refinement. Other studies have investigated the VC/VS target [110, 111], but further validation is needed.

Electrophysiology: correlsates of response and stimulation parameter selection.

Electrophysiological recordings, both noninvasive and invasive, have been used to identify potential markers of target engagement, clinical response, and symptom severity to guide DBS for MDD.

Studies using noninvasive scalp EEG have revealed that oscillatory neural activity in the theta and gamma bands, particularly in the prefrontal cortex, may be associated with depression improvement with SCC DBS. Greater depression improvement with SCC DBS has been associated with increased prefrontal theta oscillations during a cognitive control task [119], increased left prefrontal theta-gamma coupling during a working memory task [120], and increased frontal theta cordance, an EEG-specific measure of regional brain activity [121]. Cortical evoked potentials in response to SCC DBS measured with EEG may be used to differentiate between stimulating electrodes on the DBS lead in individual patients [122, 123], suggesting that cortical evoked potentials in response to SCC DBS may be a marker of therapeutic pathway engagement and could be useful to guide stimulation parameter optimization. One study used EEG with VC/VS DBS and found DBS increased theta (5–8 Hz) oscillations in medial and lateral PFC during a cognitive task, and this theta increase correlated with clinical outcomes [119].

Invasive neural recordings acquired during DBS surgery and during chronic DBS have also been used to study the pathophysiology of MDD and to identify markers to guide DBS. Acute local field potential (LFP) recordings have revealed that reduction in theta power in the left SCC during stimulation has shown promise as a potential marker for therapeutic DBS contact selection [124], whereas reduction in SCC beta power during stimulation may be associated with greater reduction in depression symptoms at 1 week post-surgery [125] (Fig. 3C). A recent study employed a DBS device capable of chronic LFP recordings (Medtronic PC + S) and reported that responders showed increased 1/f activity in the right SCC compared to nonresponders [126]. This 1/f activity may reflect changes in neural activity linked to excitation-inhibition balance and circadian rhythms. Other preliminary findings using chronic LFP recordings also suggest beta power in the right SCC may be linked to the clinical response [127], and artificial intelligence methods may be useful for revealing the link between LFP features and the longitudinal response to SCC DBS [128]. Indeed, a comprehensive recent study employing LFP recordings in the SCC over the course of treatment used artificial intelligence to identify a multiband LFP pattern (including concurrent alpha, low and high beta, and gamma band power) that can differentiate early behavioral changes from stable recovery changes [59]. This LFP pattern, validated with facial expression changes and neuroimaging analysis, was sensitive to stimulation parameter adjustments and likely reflects dynamical network changes that mediate individual recovery trajectories in response to DBS. Altogether, invasive neurophysiological recordings point to oscillatory neural activity as potential markers of MDD pathophysiology and response to DBS.

Neurophysiology has shown promise to develop markers associated with MDD symptoms and clinical response to DBS and to guide data-driven, individualized approaches for identifying MDD DBS targets, facilitating programming, and tracking the long-term response to MDD DBS.

Multimodal approaches for personalized DBS.

Recent ongoing clinical trials of MDD DBS have introduced intracranial EEG (iEEG), an electrophysiology modality commonly used in the epilepsy monitoring unit to perform seizure localization. This technique involves temporarily implanting electrodes (typically for 7–10 days) in several brain regions (example shown in Fig. 3B) and performing stimulus response mapping, which provides information about the patient-specific effects of stimulation on MDD networks (examples shown in Fig. 3C). With a focus on personalized DBS, these studies employed multimodal approaches that combine neurophysiology, neuroimaging, and computational models of stimulation to guide patient-specific targeting and stimulation parameter selection.

One trial (ClinicalTrials.gov NCT04004169) has published case reports [129, 130] on a patient with MDD who underwent iEEG recordings, stimulus response mapping, and fiber pathway mapping with the goal of developing personalized closed-loop DBS. Closed-loop DBS automatically adjusts stimulation based on an electrophysiological biomarker of disease states, which may reduce stimulation-induced side effects and improve effectiveness. A patient-specific structurally and functionally connected depression subnetwork was identified, consisting of the amygdala and the VC/VS. Fiber pathways depicting structural connectivity are shown in Fig. 3B and stimulation-evoked potentials depicting functional connectivity are shown in Fig. 3C, highlighting multi-modal methods to refine the therapeutic network. Stimulating the VC/VS consistently reduced biomarker activity in the amygdala and improved MDD symptoms. Closed-loop DBS was subsequently implemented based on these findings, which significantly improved depression in this patient [130]. Notably, the iEEG testing phase highlighted that the patient’s mood/arousal state often affected the behavioral response to stimulation [129], and thus the patient’s symptom profile and the current mood/arousal state may be important considerations when evaluating the effects of DBS on mood and developing closed-loop MDD DBS paradigms.

Another trial (ClinicalTrials.gov NCT03437928) has published a case report [98] on a patient with MDD implanted with directional DBS leads bilaterally in both the SCC and the VC/VS, along with temporary iEEG electrodes in frontotemporal regions relevant to MDD pathophysiology. The patient underwent intensive acute stimulation testing from the SCC and VC/VS while recording from the iEEG sites. The stimulus response mapping during the iEEG testing phase and computational models of fiber pathway modulation with directional DBS were used to derive network-based therapeutic stimulation parameters, which were then tested chronically. The patient was reported to reach remission criteria during the open-label phase, and their symptoms steadily worsened during a double-blinded discontinuation phase, indicating that placebo effects were unlikely [98].

Initial results of these trials highlight the potential to advance MDD DBS through leveraging multimodal, individualized approaches to identify patient-specific MDD networks and symptom biomarkers in order to guide DBS targeting, stimulation parameter selection, and closed-loop stimulation paradigms. However, we must be cautious of case studies and case series results for long-term considerations of how DBS for MDD may be implemented practically in the future.

CONCLUSIONS AND FUTURE DIRECTIONS

DBS has shown promise for select patients with refractory MDD. Although open-label studies and meta-analyses have revealed evidence of its clinical efficacy, pivotal RCTs have fallen short in demonstrating significantly greater depression improvement with active DBS compared to sham. These failed studies have prompted forward-looking discussions [90–94] about how to design suitable RCTs in an effort to evaluate clinical efficacy and to improve future implementation. Accordingly, recent research has focused on using technology, neuroimaging, and electrophysiology in a deliberate effort to develop data-driven approaches to personalize the therapy and advance the effectiveness of MDD DBS.

Despite recent advancements, crucial steps in progressing MDD DBS have yet to be taken. These include characterizing the patient population most likely to respond to DBS for MDD and optimizing DBS targeting at an individual level. Preoperative evaluations using behavioral or biometric markers (e.g., wearable devices or video/audio), noninvasive electrophysiology (e.g., EEG), or neuroimaging could potentially be used to predict a patient’s likelihood to respond to DBS. Patient-specific symptom profiles and corresponding functional or metabolic brain network activity may also play a key role in target selection; for example, SCC DBS may be well-suited for patients with high negative-dominant symptoms (e.g., high sadness or anhedonia) [114], while VC/VS DBS may be well-suited for patients with low positive symptoms (e.g., low motivation or energy) [85]. Although research has been steadily increasing on the slMFB target over the past decade, the majority of studies have focused on the SCC, and further research is critically needed to characterize preoperative and postoperative markers of DBS response in other target regions. In patients who are deemed eligible for DBS, invasive methods such as iEEG have shown promise in initial case studies of ongoing clinical trials. These cases have revealed the potential benefits of pre-implantation testing in order to optimize targeting and postoperative stimulation parameter selection in a patient-specific manner.

Following DBS implantation, individualized stimulation parameters to optimally improve MDD symptoms must be better characterized. Neuroimaging, invasive or noninvasive electrophysiology, and markers of MDD symptoms could all potentially help clinicians titrate stimulation parameters based on successful modulation of critical MDD networks or brain activity (via neuroimaging or electrophysiology) or whether stimulation decreases the patient’s symptoms during follow-up at home (via wearable sensors, audio/video, or smartphone metrics). Objective markers of MDD symptoms may be important to expedite programming, especially when using DBS electrodes capable of directional stimulation, which provide higher stimulation specificity but dramatically increase programming complexity and time for the clinician. Invasive electrophysiology markers may be critical in long-term management strategies, including type and timing of rehabilitation based on ongoing functional and functional changes in plasticity [131, 132]. New devices capable of longitudinal LFP recordings may be advantageous for uncovering and validating chronic markers of MDD symptoms and predictors of the therapeutic response to DBS. Electrophysiology markers could lead to new patient-specific closed-loop paradigms for MDD DBS, however this remains speculative.

Improving the effectiveness of MDD DBS for each patient will ultimately require individualized, multimodal approaches. Combining imaging, electrophysiology, and markers of symptom severity, as demonstrated in the initial results of recent trials, may prove impactful in treatment-based decision making. Implementing multimodal approaches will likely lead to interdisciplinary collaboration between teams of engineers, neuroscientists, and clinicians. Multicenter collaborations which pool data from MDD DBS studies have the potential to generate hypotheses and to guide prospective research directions. Future research studies should aim toward integrating interdisciplinary, multimodal approaches in an effort to improve the clinical outcomes of MDD DBS.