Prefrontal Network Mechanisms of Psychiatric Deep Brain Stimulation

Abstract

Deep brain stimulation (DBS) is an emerging treatment for otherwise treatment-refractory psychiatric disorders. It can produce remarkable clinical results in expert hands, but has not fared as well in controlled, multisite trials. That difficulty with scaling up arises in part because DBS’ mechanisms are poorly understood, meaning that it is difficult to objectively identify patients likely to respond and/or to customize stimulation to match individual patients’ needs. In the first part of this review, we overview converging anatomic and physiological evidence that psychiatric DBS acts by modulating distributed networks centered on the prefrontal cortex (PFC). We discuss potential physiological mechanisms of that modulation, including mixed excitatory/inhibitory effects, changes in local field potential oscillations, and neuroplastic changes. A major challenge is that mechanistic data from humans are limited and sometimes contradictory, in part because these studies can only be conducted at low N with limited technical replication. Animal models may overcome that challenge, but are challenging themselves because psychiatric disorders are not fully recapitulated in animal models and are defined primarily by self-reports that are unavailable in nonverbal model species. Thus, in the second part, we review paths to more reliable animal models of psychiatric DBS, including putative disease models and models based on cognitive and decision-making impairments. The latter class includes traditional cognitive assays and newer approaches based on computational modeling, both of which also implicate PFC-centric networks. Taken together, these approaches may yield new insights into DBS’ mechanisms that can in turn improve its scalability and clinical reliability.

Introduction

Deep brain stimulation (DBS) is an invasive surgical treatment involving implantation of electrodes into structures in or near the center of the brain (hence the name), followed by high-frequency (>100 Hz) electrical stimulation of those same structures (Krauss et al., 2021). It is an approved and widely used treatment for movement disorders and epilepsy and also holds promise for treating psychiatric disorders. There is clinical evidence for the use of DBS in major depressive disorder (MDD), obsessive-compulsive disorder (OCD; Sullivan et al., 2021; Widge, 2024), and to a lesser degree in addictions (Salling and Martinez, 2016) and eating disorders (De Vloo et al., 2021). DBS provides the ability to directly target circuit dysfunctions that produce neuropsychiatric conditions, with response rates above 60% in patients who have been failed by all other treatments. Despite its promise, clinical applications of DBS are impeded by a lack of mechanistic understanding of how or why DBS promotes symptom improvement (Ashkan et al., 2017; Sullivan et al., 2021). Though open-label studies demonstrate the success of DBS in individual patients and carefully selected cohorts (Mayberg et al., 2005; Malone et al., 2009; Greenberg et al., 2010; Huys et al., 2019; Sullivan et al., 2021; Widge, 2024), randomized controlled trials show variable efficacy (Sullivan et al., 2021; Widge, 2024). Given that open-label trials can be limited by investigator expectancy, passing double-blind clinical trials remains the gold standard for treatment efficacy and has yet to be achieved by DBS for neuropsychiatric conditions. Understanding mechanisms might allow better patient selection (ensuring that patients have biological impairments that match DBS’ mechanism of action) and/or better customization of stimulation parameters (Cuthbert, 2022; Widge, 2024). One clue to those mechanisms is that multiple studies have found a correlation between DBS’ clinical effects and stimulation of specific white matter tracts—predominantly those that communicate between the frontal cortex and basal ganglia (Riva-Posse et al., 2014, 2018; Baldermann et al., 2021; Widge et al., 2022). As we will discuss below, PFC activation is a common thread linking multiple efficacious DBS targets. As such, prefrontal cortex (PFC) modulation is a likely mechanism of psychiatric DBS. The next step is understanding how DBS changes PFC physiology. There are at least three plausible mechanisms, which we explore further below: (1) mixed excitatory/inhibitory effects, (2) oscillations, and (3) neuroplasticity.

These are not mutually exclusive mechanisms. For example, the timing of excitation and/or inhibition within a circuit can modulate the oscillations between two regions, therefore strengthening or weakening their connectivity (neuroplasticity). Still, there is strong evidence that DBS modulates the PFC in a combination of these three ways to treat neuropsychiatric conditions. The further challenge is understanding how these mechanisms lead to symptom relief. Psychiatric conditions are diagnosed and characterized by self-reports and subjective observations, usually with observation timescales of 1–2 weeks. It is difficult to link these infrequent, slow-changing observations to the much faster timescale of neural activity (Lisanby, 2020). Similarly, human self-report is difficult to link to behaviors we can measure in model animals (Monteggia et al., 2018), limiting our ability to study DBS’ mechanisms in PFC or elsewhere. Animal models are likely essential, however, given our limited ability to directly observe DBS’ effects in human PFC.

Here, we review the evidence for PFC as the common mechanistic factor of DBS in multiple psychiatric conditions and targets. We discuss evidence and open questions for the three mechanistic hypotheses raised above. We further consider how animal models of psychiatric DBS might be constructed through the lens of the National Institute of Mental Health Research Domain Criteria (RDoC) framework (Cuthbert and Insel, 2013; Cuthbert, 2022) and related initiatives. We discuss recent examples specifically related to DBS’ effects on cognition, as well as newer computational modeling approaches that are further clarifying mechanisms.

Deep brain stimulation for neuropsychiatric conditions likely works by modulating PFC

Existing DBS targets reveal the importance of PFC modulation

Many neuropsychiatric conditions, including OCD and MDD, are considered network disorders due to their association with disrupted connectivity between brain regions (Gordon, 2016; Senova et al., 2019; Baldermann et al., 2021; Durstewitz et al., 2021). For example, OCD is linked with disruptions in the corticothalamic–basal ganglial circuitry, and as such, DBS targets for treatment-resistant OCD are positioned to modulate connectivity within these networks (Senova et al., 2019; Haber et al., 2021). As we discuss below, there is evidence of physiological changes in PFC to support this idea. Typical DBS parameters preferentially stimulate myelinated fibers—axons from projection neurons that form major white matter bundles in the primate brain (McIntyre et al., 2004; Haber et al., 2021, 2023; Sretavan et al., 2024). Thus, even though the stimulating electrode is placed in the deep brain, its primary effect might be cortical modulation by retro- or anterograde activation of long-range axons. Supporting this, the primary DBS targets for OCD—the ventral internal capsule (IC)/ventral striatum (VCVS), bed nucleus of the stria terminalis, medial subthalamic nucleus, and ascending midbrain ventral tegmental area (VTA) fibers—all have a common ingredient in that the DBS electric field activates the internal capsule (Haber et al., 2021; Fig. 1). The internal capsule is a white matter tract that connects frontal cortical regions to subcortical, thalamic, and brainstem nuclei in a topographical, bidirectional manner. It is the central link of cortico-striato-thalamic circuits that are intimately linked with decision-making and action selection (Cox and Witten, 2019; Sharpe et al., 2019). Understanding the connectivity of effective stimulation targets can elucidate which circuits, when modulated, are important for symptom improvement.

Figure 1.

Review of connectivity between DBS targets in humans and rodents. A–E, Figure 5 from Haber et al. (2021): “Summary of connections through the 4 deep brain stimulation targets.” The topography of corticostriatal connectivity is represented by color-coding. Images in coronal section represent the left hemisphere. A, Summary depiction of prefrontal circuits—cortical and subcortical. The inset shows how different portions of PFC project through the internal capsule. B, Schematic illustration of the projections captured by DBS of the anterior limb of the internal capsule (ALIC, a component of VCVS), highlighting frontal connectivity. C, The VCVS electrode placement for ventral striatum DBS positioning also allows for modulation of multiple subcortical targets. D, The subthalamic nucleus DBS target likely modulates the cortical and basal ganglia pathways indicated by the asterisk. E, Cortical and subcortical pathways that the STN DBS target likely modulates. F–H, Figure 8 from Coizet et al. (2017): “Topographical organization of OCIP descending pathways.” Depiction of the homologous topographic conservation of corticostriatal projection in rodents, similar to those depicted in B and C. F, Schematic representation of corticostriatal topography through coronal slices from retrograde tract tracers injected (Inj) in PFC subregions, projecting through and to the striatum. G, Color-coded coronal section showing the topography of PFC projections in striatum. H, 3D view of the striatum and the topography of corticostriatal projections through the striatum. I, Figure 2 from Figee et al. (2022): “SCC stimulation location (left panel) and tractography of connections from the target (right panel).” Visualization of prefrontal pathways modulated by the SCC DBS target. These include projections to the same cortical and subcortical regions modulated from VCVS. Legend: dlPFC, dorsolateral prefrontal cortex; vlPFC, ventrolateral PFC; dmPFC, dorsomedial PFC; OFC, orbitofrontal cortex; dACC, dorsal anterior cingulate cortex; vACC, ventral ACC; mOFC, medial OFC; VP, ventral pallidum; GP, globus pallidus; Gpe, GP external segment; GPi, GP internal segment; SN, substantia nigra; STN, subthalamic nucleus; Thal, thalamus; ALIC, anterior limb of the internal capsule; dPFC, dorsal PFC; VS, ventral striatum; Hipp, hippocampus; NB, nucleus basalis; Amy, amygdala; Hypo, hypothalamus; BNST, bed nucleus of the stria terminalis; PPT, pedunculopontine nucleus; VTA, ventral tegmental area; H, fields of Forel; HD, hyperdirect pathway; IC, internal capsule; MD, mediodorsal nucleus of the thalamus; Pu, putamen; Str, striatum; TH, tyrosine hydroxylase; CRT, cerebello-rubro-thalamic tract; 3n, third nerve; 10, area 10; ZI, zona incerta; VOLO, ventral OFC lateral OFC; CG, cingulate gyrus; PL, prelimbic cortex; IL, infralimbic cortex.

Following that theory, multiple groups propose to improve DBS surgical targeting by mapping white matter connectivity. Diffusion tractography has emerged as a powerful noninvasive tool to estimate white matter organization and reconstruct pathway anatomical connectivity (Hartmann et al., 2015; Baldermann et al., 2021; Haber et al., 2023; Sretavan et al., 2024). Tractography has been used to map white matter pathways that may contribute to clinical response (Riva-Posse et al., 2014, 2018; Baldermann et al., 2021; Widge et al., 2022). Those maps suggest that the clinically effective targets for OCD are not themselves therapeutic, but are points of access to broader networks, networks that invariably involve multiple PFC regions (Senova et al., 2019; Baldermann et al., 2021; Haber et al., 2021; Sheth and Mayberg, 2023). The same is true in MDD. Retrospective analysis of tractography suggested that clinical response involved stimulation of a pericingulate white matter intersection connecting PFC with subcortical nuclei (Riva-Posse et al., 2014). Stimulation that did not fully engage these intersecting tracts (and thus their originating PFC regions) was not associated with clinical improvement, whereas specifically targeting stimulation to ensure PFC engagement led to remarkable response rates (Riva-Posse et al., 2014, 2018). Other studies found conflicting results (Ramasubbu et al., 2020), suggesting caution, but the point remains that across DBS targets and indications, clinical responses are correlated with PFC modulation. The mechanisms linking that PFC modulation to clinical improvement, however, remain unclear (Senova et al., 2019; Widge et al., 2019; Sullivan et al., 2021; Basu et al., 2023).

Electrophysiological studies confirm that DBS modulates PFC

While tractography can identify the anatomical substrates, ultimately DBS must act by changing physiology. Implanted or surface recording electrodes can measure local field potential (LFP)–pooled extracellular potentials from nearby groups of neurons. LFPs reflect electrical input from presynaptic connections and reveal correlated neural activity across neural populations/regions (Buzsáki et al., 2012; Friston et al., 2015). Human electrophysiology studies have revealed three potential mechanisms by which DBS may modulate PFC to produce therapeutic effects: (1) mixed inhibitory/excitatory effects, (2) oscillations, and (3) neuroplasticity.

Mixed inhibitory/excitatory effects

Initial theories of DBS suggested an inhibitory effect of high-frequency electrical stimulation—that DBS produced “virtual lesions” at the stimulated target (Herrington et al., 2016; Ashkan et al., 2017). This evolved based on human and animal studies of DBS for Parkinson's disease (Dostrovsky et al., 2000; Levy et al., 2001; Filali et al., 2004; Lujan et al., 2008; Agnesi et al., 2013; Chiken and Nambu, 2013), to a hypothesis that high-frequency stimulation modulates firing rate patterns in the stimulated tissue and that this firing rate pattern propagates throughout the respective network with little variation. In this framework, DBS is not strictly inhibitory, but prevents information transmission by forcing neural firing to be temporally regular, i.e., carrying no information in the theoretical sense (Grill et al., 2004; Skidmore et al., 2006; Agnesi et al., 2013). More recent evidence suggests that high-frequency stimulation has both inhibitory and excitatory effects, in both PFC and subcortical structures. These mixed inhibitory/excitatory effects may be a clinical mechanism. As we discuss below, recent animal models of psychiatric DBS corroborate this hypothesis (van den Boom et al., 2023).

The first evidence for mixed excitatory/inhibitory effects came from VCVS and nucleus accumbens (NAc) DBS for OCD. In both targets, fludeoxyglucose positron emission tomography (FDG-PET) revealed both excitation and inhibition of multiple connected structures, including PFC. Early change was correlated with clinical OCD improvement (Park et al., 2019). Furthermore, single-unit firing rates and multiunit activity in NAc decreased during “DBS on” periods, evidence of local inhibition despite distributed excitation (Park et al., 2019). The same was more recently shown for the subcallosal cingulate (SCC) white matter target used in MDD (Alagapan et al., 2023). Fibers passing through this target project to the ventromedial PFC, anterior hippocampus, insula, and dorsal anterior and posterior cingulate cortices (Alagapan et al., 2023). SCC DBS produced mixed excitatory/inhibitory effects, as inferred from local field potentials (LFPs; Alagapan et al., 2023). Acute DBS had an inhibitory effect, decreasing beta (15–30 Hz) power in SCC LFPs. Conversely, chronic DBS had an excitatory effect, increasing beta power (Alagapan et al., 2023). These mixed inhibitory/excitatory effects on SCC LFP were associated with stimulation voltage changes and may serve as an objective biomarker for DBS treatment efficacy (Alagapan et al., 2023). An earlier paper from the same group showed changes in the broadband 1/f slope of SCC LFP (Veerakumar et al., 2019), and that 1/f slope is believed to reflect an excitation/inhibition (E/I) balance (Gao et al., 2017).

Beyond this physiological evidence, cognitive–behavioral effects of DBS also suggest mixed inhibitory/excitatory effects. We have shown that DBS augments cognitive flexibility in humans, including those with treatment-resistant MDD and OCD (Widge et al., 2019; Basu et al., 2023). Augmentation of cognitive function would be difficult to achieve through a purely inhibitory mechanism. Rather, it is more likely that cognitive–behavioral effects of DBS are attained through network-wide changes in activity that can shift the E/I balance—a concept that is also evident in animal models (van den Boom et al., 2023).

Oscillations

The E/I balance is fundamentally linked to and reflected in the presence of LFP oscillations (Buzsáki et al., 2012; Voytek and Knight, 2015). Accordingly, changes in PFC oscillations are also associated with therapeutic efficacy in DBS for MDD and OCD. LFPs contain oscillations in various frequency bands, which are thought to reflect network communication (Voytek and Knight, 2015). In PFC, oscillations are thought to organize and control goal-directed behavior by coordinating distributed networks (Helfrich and Knight, 2016). For example, during rule-shifting tasks, the same neural ensemble in PFC may encode each distinct rule via frequency-specific connectivity patterns or oscillations, allowing flexible switching between rules (Helfrich and Knight, 2016; Miller et al., 2018). Theta (5–8 Hz) oscillations in medial PFC (mPFC) specifically increase with novel information, conflict, negative feedback, and errors, across species (Narayanan et al., 2013; Cavanagh and Frank, 2014). Theta oscillations may thus coordinate cortical–subcortical networks in the service of top-down executive control. There is a long-standing link between high-frequency DBS and changes in lower-frequency oscillations, best demonstrated through disruption of pathological beta (15–30 Hz) oscillations in Parkinson's disease (Cagnan et al., 2019). At the white matter targets commonly used in psychiatric illness, DBS can also enhance oscillations by an as-yet-unknown mechanism. For instance, VCVS DBS increases theta power in PFC across multiple studies, and that theta increase is accompanied by improved executive function (Widge et al., 2019; Basu et al., 2023). Similar PFC theta enhancement occurs from subthalamic nucleus stimulation and correlates with improved cognition—but requires a lower stimulation frequency (Kelley et al., 2018).

It remains unclear whether oscillatory enhancement is always therapeutic—in a study in patients with OCD, higher power of 2–5 Hz frontal oscillations was associated with higher OCD symptom severity (Figee et al., 2013). DBS of the NAc, which overlaps the VCVS target, attenuated those oscillations, and that attenuation was associated with improved OCD symptoms (Figee et al., 2013). At the SCC target, as noted above, clinical response was associated with short-term suppression and long-term enhancement of beta oscillations (Alagapan et al., 2023).

Neuroplasticity

By influencing PFC activity via mixed inhibitory/excitatory effects and oscillations, DBS may produce long-term changes in network/circuit functional and anatomical connectivity. For example, NAc DBS reduced elevated frontostriatal network connectivity in patients with OCD (Figee et al., 2013). While changes in connectivity alone do not prove neuroplasticity as a mechanism of DBS, connectivity changes can arise from plasticity as correlated activity is often a sign of synaptic weight (Schneider et al., 2021). In macaques, DBS increased myelination of long-range pathways believed to be involved in DBS response (Fujimoto et al., 2024). However, DBS also presents a plasticity paradox. Neuroplasticity must be involved in DBS for psychiatric disorders, because symptoms change slowly when effective stimulation is applied (Herrington et al., 2016; Widge, 2024). There are acute positive effects of DBS at multiple targets (Okun et al., 2007; Choi et al., 2015), but they do not correlate well with long-term clinical response (Widge, 2024). At the same time, that plasticity must be unstable, given that symptoms return when stimulation stops (Sullivan et al., 2021; Widge, 2024). One possibility is that the slow timescale of response does not reflect traditional plasticity, but rather the time needed for a change in brain function to reach conscious awareness. Both cognition (Widge et al., 2019; Basu et al., 2023) and neurophysiology (Widge et al., 2019; Alagapan et al., 2023; Basu et al., 2023) change very rapidly when DBS is applied, but patients may not be able to detect these subtle changes in brain states. Rather, if DBS improves a basic cognitive/emotional function, patients may then need to use and live with that improved function for weeks before they can begin to notice it. There are analogies to the “learned nonuse” phenomenon described in motor rehabilitation (Hirsch et al., 2021), as well as in DBS treatment for dystonia, which often takes several months to see effects as the brain requires time to learn that the muscles are able to move. This has, for instance, been proposed as a reason why mobile components of dystonia improve relatively quickly, but tonic/posture components require months (Tisch and Limousin, 2020). Similar to DBS treatment of dystonia or recovery from an injury, patients may need additional therapy to learn to benefit from the new capacities that DBS provides (Mantione et al., 2014; Graat et al., 2023). This idea is further corroborated by evidence that patients with effective DBS treatment of OCD report a greater sense of “affordances,” actions that they can realistically take to challenge their symptoms (van Westen et al., 2019). That is still a form of plasticity, but would explain the instability—if the new capabilities are then taken away by deactivating the DBS, patients might no longer be able to fully compensate for the pathological processes that led to their illnesses. That is, plasticity might occur in higher-level (PFC) circuits controlling action selection, but not in subcortical circuits that also contribute to symptoms.

Speculatively, this implies that DBS might be made more effective if more persistent plasticity could be induced directly in subcortical or cortical–subcortical circuits that directly cause symptoms. There is a precedent for this concept. In spinal cord injury, persistent neuroplasticity can be induced by using DBS-like implants (and in one case, an actual DBS implant) to deliver spinal stimulation that is triggered by motor cortex activity (Borton et al., 2014; Capogrosso et al., 2016; Lorach et al., 2023). These implants leverage the long-recognized principle of timing-dependent/activity-dependent plasticity (Hebb, 1949). Given that plasticity exists throughout the brain and that psychiatric disorders are often framed as problems of pathological circuit plasticity (Insel, 2010; Koob and Volkow, 2016; Durstewitz et al., 2021), the same principles might be leveraged in circuits currently targeted by high-frequency DBS (Widge, 2024). Furthermore, DBS may promote long-term changes in network/circuit functional and anatomical connectivity by altering other mechanisms of neuroplasticity such as protein expression, neurotransmitter release, and through non-neuronal changes (e.g., astrocytes and microglia; Jakobs et al., 2019). The main challenge is that techniques designed to probe different mechanisms of neuroplasticity, such as activity-dependent techniques, cannot easily be implemented on current clinical hardware. While this discussion focuses on evidence of DBS-induced neuroplasticity, the mechanism of that plasticity is an equally important question that remains understudied. Most of the extant motor literature relies on animal models—though animal models relevant to psychiatric DBS are difficult to create, preclinical studies of DBS for depression suggest that stimulation-induced neurochemical changes are associated with antidepressant-like DBS effects in animal models (Campos et al., 2023). This plasticity and the utility of and considerations for animal models of DBS will be further discussed below.

Alternative ways to study DBS mechanisms

The studies above highlight both how modern methods are uncovering the mechanisms of DBS and psychiatric illness and the limitations of those methods when applied in humans. Data from human studies are inherently complex and often limited by the realities of clinical treatment. The available methods tend to lack spatial or temporal resolution, and it is rarely possible to repeat the same test multiple times in each participant to reduce experimental noise. We might overcome those limitations through animal models, constructed based on both existing understanding of DBS’ mechanisms and our growing understanding of the computational “ingredients” underlying mental disorders.

Animal models

Taking advantage of anatomical circuit homologies, animal models may serve as an alternative approach to investigating DBS mechanisms. They have been useful in elucidating the mechanisms of DBS for Parkinson's disease (Vitek and Johnson, 2019). It should be possible to develop animal models of psychiatric DBS, particularly for the striatal and internal capsule targets. The internal capsule (IC) in primates contains topographically organized white matter fibers connecting the PFC to the striatum, thalamus, and brainstem. While rodents do not have a compact IC, they do have diffuse white matter fascicles with similar topography that travel from prefrontal regions to and through the striatum (Coizet et al., 2017; Fig. 1F–H) and eventually form an internal capsule much more posteriorly. The rodent medial PFC (mPFC) is segregated into the cingulate cortex, prelimbic cortex, and infralimbic cortex, as well as the orbitofrontal cortex, composed of ventral orbital and lateral orbital subregions (Coizet et al., 2017). The rodent mPFC is most homologous to primate ACC, but takes on additional functions (e.g., working memory) given the lack of rodent dorsolateral PFC (Heilbronner et al., 2016). Critically, primates and rodents share similar dorsal–ventral topography and organization of both mPFC inputs to the striatum and white matter fibers passing through the striatum to the brainstem and thalamic terminations (Heilbronner et al., 2016). These circuit homologies, coupled with the expanse of experimental technologies available for use in animal models, serve as an alternative approach to studying DBS mechanisms in human patients. Recent rodent work stimulating these homologs of human DBS targets showed similar activation of prefrontal regions (Pinhal et al., 2018; van den Boom et al., 2023; Reimer et al., 2024) and provided evidence for all three mechanisms above: (1) mixed inhibitory/excitatory effects, (2) oscillations, and (3) neuroplasticity.

Mixed inhibitory/excitatory effects

The mouse ventral internal capsule is a rodent homolog of the human posterior IC, with similar topographic fiber organization. To model DBS for OCD in mice, one study stimulated the ventral internal capsule in SAPAP3 knock-out mice, a model for OCD-related behavior (Burguière et al., 2015; Hadjas et al., 2019; Lamothe et al., 2023; van den Boom et al., 2023). SAPAP3 mice demonstrate a compulsive behavioral phenotype, excessive grooming. In an open field, SAPAP3 mice groomed four times more than wild-type (WT) mice (van den Boom et al., 2023). Ventral internal capsule DBS reduced grooming in a dose-dependent way (larger effects with higher pulse width and current) in SAPAP3 mice, but not WT mice. Excessive grooming resumed when DBS was turned off, just as symptoms recur in humans when DBS is stopped. Calcium imaging showed a similar dose-dependent suppression of activity in the dorsal cortex with varying current and pulse widths (van den Boom et al., 2023). Similar to the acute versus chronic DBS effects in (Alagapan et al., 2023), activity increased immediately following DBS onset in all studied regions, but after a few seconds of DBS, activity was reduced in frontal regions (van den Boom et al., 2023). These mixed inhibitory/excitatory effects were replicated at the single cellular level when imaging pyramidal PFC neurons. DBS elicited immediate inhibitory or excitatory effects in some cells and sustained inhibition or excitation in others. Both inhibition and excitation correlated with decreased grooming in SAPAP3 mice, implying that the mixed physiological effect is necessary for behavioral effects. Furthermore, the sustained activation evoked by DBS was not present when mice were anesthetized, indicating that this effect relies on indirect network activity (van den Boom et al., 2023). Overall, these results suggest that DBS’ behavioral effects, particularly on compulsivity, require both inhibitory and excitatory modulation at the network level.

Oscillations

As far as we know, modulation of PFC oscillations via DBS has yet to be investigated in animal models. Given the circuit homologies and ability of internal capsule DBS to modulate PFC activity (Pinhal et al., 2018; van den Boom et al., 2023; Reimer et al., 2024), animal models could further elucidate how DBS changes oscillatory activity in PFC.

Neuroplasticity

Similar to human studies, there is behavioral evidence that DBS evokes a neuroplasticity effect in animal models. Comparing the effects of ventral internal capsule DBS and dorsal ventral striatum DBS on excessive grooming in SAPAP3 mice, one study found that grooming remained below baseline levels 90–120 min following discontinuation of DBS (Pinhal et al., 2018). Note that this does conflict somewhat with other papers showing rapid rebound of behavior changes with DBS off (van den Boom et al., 2023; Reimer et al., 2024).

Modeling DBS for treatment-resistant depression, a recent study identified structural and functional changes evoked by SCC DBS in macaques. Six weeks of SCC DBS caused significantly improved white matter integrity specific to the midcingulate region (Fujimoto et al., 2024). As such, SCC DBS induced selective macrostructure remodeling of the white matter tract that connects the SCC with the cingulate cortex. This same location showed increased proliferation of myelinating oligodendrocytes and subsequently, a larger degree of myelination—a form of structural plasticity (Fujimoto et al., 2024). In that same study, SCC DBS decreased fMRI functional connectivity between the SCC and frontal regions, as well as between SCC and functional networks, including the default mode, central executive, and limbic networks, consistent with SCC DBS modulating activity in regions connected to the cingulum bundle. Interestingly, networks not directly connected to the cingulum bundle were also modulated by SCC DBS. Functional connectivity between the sensory motor network and SCC was increased, and SCC DBS changed the balance of connectivity among salience network nodes (Fujimoto et al., 2024). It is not yet clear which of these changes may mediate clinical effects, but this study still shows that DBS can induce a form of plasticity in a complex, gyrified brain.

While the specific mechanisms by which psychiatric DBS changes PFC network connectivity remain understudied in humans, animal models have made more progress in understanding which kinds of neuroplastic changes are associated with antidepressant-like behavioral effects. Campos et al. (2023) presented an in-depth review of the neurochemical and behavioral changes elicited by DBS delivered to the common targets for depression, in animals. Notably, direct modulation of PFC resulted in the most robust antidepressant-like DBS effects. These behavioral changes were associated with neurochemical changes including increases in neurotransmitter release, synaptic markers, and neurotransmitter receptor expression (Campos et al., 2023). DBS at other targets—the nucleus accumbens, medial forebrain bundle, ventral tegmental area, and lateral habenula—commonly elicited neurochemical changes (e.g., increased neurotransmitter release) in PFC (Campos et al., 2023). This further suggests that modulation of PFC at the synaptic level may be necessary for the symptom relief elicited by psychiatric DBS across targets. Though further research is necessary to identify the specific mechanisms of neuroplasticity involved, it is possible that DBS effects at the synaptic level lead to long-term changes in PFC network connectivity. The use of animal models, with a consideration for how we measure relevant behavioral changes, is likely to expand our understanding of DBS-evoked neuroplastic changes in PFC. Preliminary efforts are already seeking to induce this type of plasticity in addiction-related circuits (Asp et al., 2024).

Objective and measurable cognitive constructs (RDoC)

Despite useful circuit homologies that strengthen the translational power of animal models for DBS mechanisms, neuropsychiatric conditions are not well modeled in animals (Monteggia et al., 2018). For instance, although the SAPAP3 knockouts described above show a single compulsive behavior, human OCD is a complex and heterogeneous disorder involving not just motor changes but changes in metacognition, habit formation/override, and error detection (Robbins et al., 2024). The macaque results from SCC DBS are fascinating, but did not attempt to model the pathophysiological process that gives rise to treatment-resistant depression in humans, and as such may not relate to what DBS does in a brain affected by a chronic, treatment-refractory disease. The fundamental problem is that because human disorders are characterized almost entirely based on self-report and highly subjective observations, including fairly uniquely human processes of expression and vocal tone, they cannot be directly modeled in nonverbal species (Monteggia et al., 2018; Redish et al., 2022).

As an alternative, disorders might be modeled by assessing objective, measurable constructs/cognitive phenotypes that are commonly disrupted in neuropsychiatric conditions and readily quantified in experimental animals. This is a central thesis of the National Institute of Mental Health's Research Domain Criteria (RDoC) project and of related initiatives such as the “computational psychiatry” movement (Cuthbert and Insel, 2013; Monteggia et al., 2018; Huys et al., 2019; Redish et al., 2022). Focusing on cognitive phenotypes can also allow researchers to assess how DBS targets mechanistically improve symptoms that cut across disorders. For example, disrupted cognitive control—the ability to flexibly adjust attention, thoughts, and behaviors in response to changing environmental demands—is not only common in OCD and MDD but also depends on the same PFC regions and theta oscillations that are known to be modulated by DBS (Widge et al., 2019; Basu et al., 2023). Importantly, fibers from the mPFC, dorsal ACC, lateral PFC, and striatum—the same cortical–striatal circuits that we know to be involved in cognitive control (Sharpe et al., 2019; Menon and D’Esposito, 2022; Basu et al., 2023)—all pass through the VCVS target. Accordingly, VCVS DBS increases theta oscillations (known to be associated with cognitive control) when patients perform cognitive control tasks and improves performance on those tasks (Widge et al., 2019; Allawala et al., 2021; Basu et al., 2023). Furthermore, DBS-evoked increases in PFC theta power were positively correlated with improved cognitive control, and changes in PFC theta power correlated with clinical outcomes (Widge et al., 2019).

Cognitive control has also been used as a lens to compare the therapeutic efficacy of two different DBS targets in the same patients (Tyagi et al., 2019). In the same patients with OCD (n = 6, five males and one female), researchers compared changes in mood, cognitive control, and neural circuitry during VCVS and anteromedial subthalamic nucleus DBS. Given the placement of their electrodes, the anteromedial subthalamic nucleus target showed connectivity with lateral orbitofrontal cortex, dorsal ACC, and dorsolateral PFC, while the VCVS target connected primarily with medial orbitofrontal cortex (Tyagi et al., 2019). Overall, DBS at both targets showed significant therapeutic efficacy, but improved distinct symptoms. DBS at the VCVS target significantly improved mood, while anteromedial subthalamic nucleus DBS significantly improved cognitive control by decreasing errors during an intra-/extradimensional set-shifting task (Tyagi et al., 2019). This suggests that while these targets may modulate distinct circuits or networks to improve distinct OCD symptoms, they commonly modulate PFC regions, which may be critical for therapeutic efficacy.

This cross-diagnostic, circuit-driven improvement in cognitive control can also be used to develop animal models. Stimulating homologous cortical–striatal circuitry in rats performing an attentional set-shifting task (Set Shift) replicated the human VCVS DBS effect seen in previous studies (Widge et al., 2019; Allawala et al., 2021; Basu et al., 2023), improving cognitive control (Reimer et al., 2024). Specifically, in both humans and rats, stimulation reduced reaction times without increasing errors. Given the topography of prefrontal white matter fascicles traveling to and through the rodent striatum, the rodent midstriatal DBS target likely engaged axons from the prelimbic cortex (Coizet et al., 2017), which was confirmed by measuring c-Fos expression (Reimer et al., 2024). As noted above, the prelimbic cortex corresponds to human ACC, a central structure in cognitive control and the putative generator of theta oscillations (Cavanagh and Frank, 2014; Cohen, 2014). Not only did this study demonstrate the translational power that measuring objective cognitive constructs adds to animal studies of DBS mechanisms, but it opened the door to a further approach: computational modeling to further elucidate mechanisms.

Computational modeling

Knowing that DBS changes PFC activity and PFC-linked behavior does not elucidate how DBS changes cognition, in the sense of which decision processes are changed. Computational modeling is a useful tool to understand how DBS and psychiatric treatments in general change decision processes (Huys et al., 2016; Redish et al., 2022). In Reimer et al. (2024), a reinforcement learning drift diffusion model (RLDDM) was fit to rat Set Shift behavior (Pedersen et al., 2017; Reimer et al., 2024). The RLDDM revealed that DBS improved cognitive control in rats by decreasing the amount of evidence required to make a decision (boundary separation), driving the actor toward the correct choice boundary (increased drift rate), and influencing the actor's precommitment to a given choice based on their learned values (bias; Reimer et al., 2024). As an early insight, this finding suggests that multiple cognitive mechanisms may need to be engaged by DBS to produce therapeutic effects. Interestingly, these RLDDM parameters explain the behavioral effect of DBS load onto different PFC subregions (Reimer et al., 2024). This concept is consistent with human DBS findings as well. As discussed above, optimal DBS targets for OCD and MDD simultaneously engage multiple white matter tracts and influence many PFC regions at once (Riva-Posse et al., 2014, 2018; Baldermann et al., 2021; Sheth and Mayberg, 2023; Meyer et al., 2024). As such, DBS may work by simultaneously engaging multiple PFC subregions, which then modulate unique cognitive/computational mechanisms that synergize to improve symptom severity.

Other modeling frameworks may also apply to DBS mechanisms. For example, the explore–exploit balance is often disrupted in neuropsychiatric conditions where cognitive flexibility is disrupted (Grissom and Reyes, 2019; Kaske et al., 2023; Wyatt et al., 2024). Value-based decision-making in uncertain environments requires a balance between sampling options to learn their values (exploration) and repeatedly selecting a valuable option once it is known (exploitation). Explore–exploit imbalances (e.g., decreased exploration in MDD and excessive exploitation in OCD) are linked with symptom severity and prognosis (Grissom and Reyes, 2019; Kaske et al., 2023; Wyatt et al., 2024). As just noted, DBS, particularly of VCVS, modulates structures involved in the value computations that are central to explore–exploit tradeoffs. If the key to DBS therapeutic efficacy is indeed simultaneously modulating numerous PFC subregions to change multiple cognitive computations/mechanisms, variability in baseline individual differences in cognitive constructs (e.g., explore–exploit tradeoff and cognitive control) may be related to differential DBS outcomes. Using hidden Markov models to analyze behavioral data from value-based decision-making tasks, latent explore–exploit states can also be identified in animal models (Ebitz et al., 2018; Chen et al., 2021) and have revealed robust sex differences in cognitive–behavioral states in mice (Chen et al., 2021). Indeed, sex and gender biases in neuropsychiatric risk are common across disorders of flexibility. Taken together, it is possible that sex-biased DBS mechanisms exist as well. As such, evaluating how the explore–exploit balance in animal models or humans changes with DBS may further reveal the computational mechanisms underlying DBS treatment efficacy. Computational modeling, alongside more detailed cognitive assessments, may answer a further question: is DBS-driven enhancement in one cognitive domain accompanied by impairments in others? Thus far, no such effects have been reported or detected in our own work, but it seems plausible that any bias on brain function should be problematic in at least some contexts.

Conclusion

Converging evidence from human and animal studies suggests that DBS for psychiatric disorders works primarily by modulating distributed networks that involve multiple PFC regions. DBS appears to change activity in PFC via (1) mixed inhibitory/excitatory effects, (2) oscillations, and (3) neuroplasticity (Table 1). Our review of these mechanisms primarily focuses on the VCVS and SCC targets for psychiatric DBS due to their prevalence in the human literature, but additional targets, such as the lateral habenula (Germann et al., 2021) and medial forebrain bundle (Kilian et al., 2024), do exist (Figee et al., 2022). Currently, there are too few human DBS studies using these additional targets to know whether the above arguments still apply. However, many psychiatric DBS targets share structural and/or functional connectivity with PFC and with each other. For example, the lateral habenula and medial forebrain bundle are highly connected to the SCC and VCVS targets (Figee et al., 2022) and, subsequently, to prefrontal regions (Coenen et al., 2018; Groos and Helmchen, 2024). Despite the discussed evidence that prefrontal mechanisms are important for the therapeutic effects of psychiatric DBS, current literature in this area is limited. In part, this might be due to the limitations posed by human DBS studies—channel count, availability of recording regions, and capacity for repeated testing. Mechanistic clarity will likely come by augmenting those human studies with more detailed measurements in animal models, which in turn can be enabled by a focus on cross-diagnostic domains of function and computations associated with those functions. If we can better understand how DBS acts on PFC elements and which elements are central to specific decisional dysfunctions, DBS holds great promise as a precision treatment for psychiatric disorders.

Table 1.

Summary of key PFC mechanisms of DBS findings

DBS target	Condition	Species	Effect	Mechanism	Reference	Type
VCVS and NAc	OCD	Human	Inhibition and excitation of connected structures (FDG-PET)	Mixed I/E effects	Park et al. (2019)	Primary data
NAc	OCD	Human	Local inhibition despite distributed excitation (single and multiunit activity)	Mixed I/E effects	Park et al. (2019)	Primary data
SCC	MDD	Human	Inhibitory effect of acute DBS, excitatory effect of chronic DBS (LFP)	Mixed I/E effects	Alagapan et al. (2023)	Primary data
SCC	MDD	Human	Changes in broadband 1/f slope (LFP)	Mixed I/E effects	Veerakumar et al. (2019)	Primary data
VCVS	OCD and MDD	Human	Behavioral augmentation of cognitive flexibility	Mixed I/E effects	Widge et al. (2019) and Basu et al. (2023)	Primary data
Ventral internal capsule	Compulsivity model	SAPAP3 KO mice	Excessive grooming reduced in a stimulation dose-dependent way, accompanied by dose-dependent suppression of activity in the dorsal cortex (calcium imaging)	Mixed I/E effects	van den Boom et al. (2023)	Primary data
Ventral internal capsule	Compulsivity model	SAPAP3 KO mice	Activity increased immediately following DBS onset but reduced in frontal regions after a few seconds of stimulation (calcium imaging)	Mixed I/E effects	van den Boom et al. (2023)	Primary data
Ventral internal capsule	Compulsivity model	SAPAP3 KO mice	Mix of inhibitory and excitatory effects in individual PFC pyramidal neurons following DBS. Inhibition and excitation were correlated with reduced excessive grooming	Mixed I/E effects	van den Boom et al. (2023)	Primary data
VCVS	OCD and MDD	Human	Increased theta power in PFC, accompanied by improved executive function	Oscillations	Widge et al. (2019) and Basu et al. (2023)	Primary data
VCVS; SCC	OCD and MDD; MDD	Human	Rapid physiological changes	Oscillations	Widge et al. (2019), Basu et al. (2023), and Alagapan et al. (2023)	Primary data
VCVS	OCD and MDD	Human	Rapid cognitive changes	Oscillations	Widge et al. (2019) and Basu et al. (2023)	Primary data
Subthalamic nucleus	Parkinson's disease	Human	PFC theta enhancement, correlated with improved cognition	Oscillations	Kelley et al. (2018)	Primary data
NAc	OCD	Human	Attenuation of high-power frontal oscillations associated with improved OCD symptoms	Oscillations	Figee et al. (2013)	Primary data
SCC	MDD	Human	Short-term suppression and long-term enhancement of beta oscillations associated with clinical improvement	Oscillations	Alagapan et al. (2023)	Primary data
NAc	OCD	Human	Reduction of elevated frontostriatal network connectivity	Neuroplasticity	Figee et al. (2013)	Primary data
Various	Various	Human	Slow symptom changes with effective stimulation	Neuroplasticity	Herrington et al. (2016) and Widge (2024)	Review
SCC	N/A	Macaques	Increased myelination of long-range pathways	Neuroplasticity	Fujimoto et al. (2024)	Primary data
SCC	N/A	Macaques	Functional connectivity between SCC and frontal regions, as well as SCC and functional networks decreased (fMRI)	Neuroplasticity	Fujimoto et al. (2024)	Primary data
PFC	Multiple MDD-related behaviors	Rodents	Antidepressant-like behavioral changes associated with neurochemical changes	Neuroplasticity	Campos et al. (2023)	Review

Throughout this review, we identify multiple mechanisms by which DBS modulates PFC networks to produce therapeutic outcomes. This table provides a summary of evidence supporting the involvement of each mechanism—mixed I/E effects, oscillations, and neuroplasticity—and includes the target, condition, species, effect, and reference for each DBS finding discussed. As noted in the “type” column, this table includes data from primary and review papers for the sake of conciseness; the review papers point to further primary data.