Targeted neural network interventions for auditory hallucinations: Can TMS Inform DBS?

Abstract

The debilitating and refractory nature of auditory hallucinations (AH) in schizophrenia and other psychiatric disorders has stimulated investigations into neuromodulatory interventions that target the aberrant neural networks associated with them. Internal or invasive forms of brain stimulation such as deep brain stimulation (DBS) are currently being explored for treatment-refractory schizophrenia. The process of developing and implementing DBS is limited by symptom clustering within psychiatric constructs as well as a scarcity of causal tools with which to predict response, refine targeting or guide clinical decisions. Transcranial magnetic stimulation (TMS), an external or non-invasive form of brain stimulation, has shown some promise as a therapeutic intervention for AH but remains relatively underutilized as an investigational probe of clinically relevant neural networks. In this editorial, we propose that TMS has the potential to inform DBS by adding individualized causal evidence to an evaluation processes otherwise devoid of it in patients. Although there are significant limitations and safety concerns regarding DBS, the combination of TMS with computational modeling of neuroimaging and neurophysiological data could provide critical insights into more robust and adaptable network modulation.

1. Introduction

Invasive neural network interventions such as ablative surgery or deep brain stimulation (DBS) are among the most controversial in modern psychiatry. The ethical, scientific and clinical barriers to the development and implementation of these treatments are juxtaposed with an urgent need to treat patients who remain profoundly disabled despite comprehensive treatment strategies. (Bell and Racine, 2013; Naesstrom et al., 2016; Nangunoori et al., 2013; Saleh and Fontaine, 2015). Recent proposals for the development of DBS for medication-resistant symptoms of schizophrenia highlight this juxtaposition (Mikell et al., 2009; Mikell et al., 2015; Salgado-Lopez et al., 2016).

One of the primary challenges inherent to all psychiatric DBS endeavors is extricating a target symptom from its Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 2013) construct. The diagnosis of schizophrenia, for example, is a cluster of symptoms associated with a wide range of pathologies involving nearly every part of the brain. By contrast, individual symptoms within this cluster may have more objective or individualized network-specific representations (Boschloo et al., 2015; Braga and Buckner, 2017). Verbal auditory hallucinations (AH) are one example of this specificity (Jardri et al., 2011; Zmigrod et al., 2016). As others have pointed out, DSM and International Statistical Classification of Diseases and Related Health Problems (ICD) codes do not always correspond with findings from behavioral, genetic and systems neuroscience (Cuthbert and Insel, 2013). DBS selectively targets neural networks (white matter targeting) or singular nodes within such networks (gray matter targeting) (Williams et al., 2014). Thus, DBS may have a better chance at reducing a target symptom associated with a target network than it would an entire cluster of symptoms encapsulated by a clinical diagnostic construct. This idea is consistent with the Research Domain Criteria (RDoC) approach developed by the National Institute of Mental Health (Insel et al., 2010; Insel, 2014).

A second primary challenge relates to the translation of DBS strategies from animal subjects into human patients (Cuthbert and Insel, 2013). Potential DBS targets are often first identified and manipulated in animal models of psychiatric disease. It is often difficult to evaluate the validity and fidelity of these models, particularly given the heterogeneity of human psychiatric disorders as well as the clustering of symptoms into clinical diagnostic constructs. The specific biological manipulations made to render a model ‘of schizophrenia,’ for example, tend to recapitulate a suite of behaviors that are relevant to positive, negative and cognitive symptoms. There are a few examples of rodent models in which hallucinations and delusions have been modeled (Honsberger et al., 2015; McDannald and Schoenbaum, 2009). Such strategies often embrace an RDoC perspective on hallucinations as trans-diagnostic symptoms (Forrest et al., 2014; Robbins, 2017). Even with more nuanced animal models of behaviors or symptoms, several barriers to translating animal DBS findings into humans remain; there are few causal tools with which to predict response, refine targeting or guide treatment decisions in patients considering DBS (Alhourani et al., 2015; O’Halloran et al., 2016).

Transcranial magnetic stimulation (TMS) is an external or non-invasive form of brain stimulation that may provide a means by which to partially address both challenges. There is emerging evidence that TMS may reduce AH in patients with schizophrenia and related psychotic disorders (Freitas et al., 2009; Hoffman et al., 2003; Hoffman et al., 2013; Montagne-Larmurier et al., 2011; Otani et al., 2015; Slotema et al., 2014). TMS has been used as an investigational tool for neural circuit mapping as well as a therapeutic tool for neural circuit modulation (George et al., 2013a), but few have explicitly examined its potential in the development and implementation of invasive circuit-based interventions such as DBS (Pathak et al., 2016). In this editorial, we briefly review the data on TMS for AH and propose that TMS may be a causal tool with which to carefully advance the development of DBS for individual symptoms within psychiatric constructs.

2. Intractable Voices as a Target Symptom in Schizophrenia and Other Disorders

AHs are one of the most debilitating symptoms of schizophrenia and psychotic disorders, particularly in light of their negative valence and intrusiveness (Andreasen and Flaum, 1991; Carter et al., 1996; Chaudhury, 2010; Falloon and Talbot, 1981; Peters et al., 2012). Although aggression is multifactorial, there are established associations between AHs in schizophrenia and destructive behaviors such as assault, homicide and suicide (Cheung et al., 1997; Haddock et al., 2013; Hoptman, 2015; Keers et al., 2014; Swanson et al., 2006; Volavka et al., 1997; Wong et al., 1997). There are also data suggesting that AHs contribute to disability and encumber rehabilitation (Ferdinandi et al., 1998; Thomas et al., 2014). Patients affected by schizophrenia, along with their families and their communities, would benefit greatly from an intervention that could either change the negative content of AHs or reduce the frequency with which they occur.

3. TMS for AH: A Review

Hoffman and colleagues were the first to publish a double-blind crossover study showing that 1 Hz stimulation of the left temporoparietal junction (TPJ) reduced AH severity as measured by standard clinical rating scales (Hoffman et al., 1999). The site of stimulation was chosen based on prior imaging studies, particularly positron emission tomography (PET) (Fiez et al., 1996; Silbersweig et al., 1995). Since these early publications, cognitive neuroscience has transitioned from structure-function relationships at the level of single regions to a hodological approach wherein functions are distributed across regions and information is processed in the relationships between them. Connectivity analyses of functional magnetic resonance imaging (fMRI) or electroencephalograph (EEG) signals signify the correlations between regions (functional connectivity) (Chang et al., 2017; Curcic-Blake et al., 2017) or the directional relationships between regions (effective connectivity) (Grana et al., 2017; Li et al., 2017). These developments have added nuance and complexity to the basic framework of AH. For example, connectivity analyses have suggested that AH may arise from aberrant top-down influences of cognitive and belief-mediating regions on sensory regions (Powers et al., 2016).

Although the left TPJ remains a critical hub, it is now clear that complicated circuit interactions between Wernicke’s area and its right homologue, bilateral putamen and left inferior frontal cortex are critical to the pathophysiology of AH (Alderson-Day et al., 2016; Chang et al., 2017). Some studies focus at this level of regional connectivity, exploring connections based on a seed region of interest (Alderson-Day et al., 2015; Hoffman and Hampson, 2011; Hoffman et al., 2011). Others model large network-scale dynamics, correlating changes in the default mode network, the central executive network, and the salience network with conscious percepts of speech (Chen et al., 2016; Lefebvre et al., 2016). As data acquisition and analysis improves, so too will models of AH circuits and the options to intervene on them. The success of these approaches will depend on how closely the pathophysiology of AH can be linked to particular network nodes, edges or motifs.

Modulating the AH circuitry with TMS has yielded intriguing but mixed results. Dozens of studies have been conducted since Hoffman and colleagues published their pioneering pilot study, prompting several review articles and meta-analyses (Aleman et al., 2007; Freitas et al., 2009; Montagne-Larmurier et al., 2011; Otani et al., 2015; Slotema et al., 2012; Slotema et al., 2014; Tranulis et al., 2008). Despite variations in response rates and effect sizes, the majority of available data suggest that 1Hz TMS of the left TPJ significantly reduces AHs when compared to sham stimulation. The most compelling data involving TMS for AHs are derived from sham-controlled multimodal studies correlating circuit function changes with rating scale changes. Arterial spin labeling (ASL) data show that patients with AHs who received real TMS exhibited reduced blood flow in primary auditory cortex, left Broca’s area and cingulate gyrus. This decrease in cerebral blood flow in the primary auditory cortex correlated with a decrease in AH scale score (Kindler et al., 2013). Other ASL studies have attempted to identify brain regions that could be used to predict TMS response before it is administered (Homan et al., 2012).

BOLD signal analyses from fMRI studies show similar results to ASL studies, with decreases in AH rating scales correlating with decreases in primary auditory cortex signal following TMS (Giesel et al., 2012). Connectivity analyses confirm that the network of structures implicated in AHs is altered by TMS of the TPJ. Compared to sham, real TMS normalizes the functional connectivity between temporoparietal regions like the TPJ and the angular gyrus with frontal regions like the dorsolateral prefrontal cortex (Briend et al., 2017; Gromann et al., 2012).

Techniques such as fMRI, PET and EEG have been used to tailor targeting, predict response, optimize dosing, adjust parameters, investigate state dependence and monitor response (Guller et al., 2012; Hoffman et al., 2007; Moseley et al., 2015). Multimodal studies have also been used to explore the relationship between anatomical variation and TMS response (Nathou et al., 2015). Whereas research studies use these approaches to probe circuitry, translational and clinical studies use them to try to improve therapeutic response (Donaldson et al., 2015).

Many questions remain regarding the clinical utility of TMS for AH. Some of these questions are a consequence of study heterogeneity. Patient characteristics and selection, targeting technique (e.g. EEG systems, neuronavigation, anatomical vs. functional targeting, etc.), TMS dose and parameters, primary outcome measures and other factors vary significantly between studies. This heterogeneity makes it difficult to evaluate the evidence base as a coherent whole (Lefaucheur et al., 2014). Other questions relate to sham variability. Many early TMS studies of AH used a coil tilting technique that made it difficult to maintain a proper blind (Dollfus et al., 2016). Newer studies have begun to employ the E-sham system that was implemented in the trials leading to FDA clearance of TMS for treatment-resistant depression (Borckardt et al., 2008; George et al., 2010a; Taylor et al., 2013; Taylor et al., 2012). Additional area of concern include tolerability, feasibility and compliance, as TMS can be somewhat uncomfortable (this diminishes with time, see (Borckardt et al., 2013)) and often requires daily sessions for several weeks or months (George et al., 2013a). Moreover, some patients may require regular maintenance sessions.

4. Considering a Transition from Non-invasive to Invasive Neuromodulation

Patients with refractory symptoms of schizophrenia are often prescribed clozapine plus or minus augmentation with antipsychotics, anticonvulsants, NMDA agonists, cognitive-enhancing agents and other pharmacological agents (Sommer et al., 2012). A full review of the clozapine literature is beyond the scope of this manuscript, but there are data to suggest that clozapine is ineffective or intolerable for 40–70% of patients with refractory symptoms of schizophrenia (Arumugham et al., 2016). Such patients frequently end up with polypharmacy while others end up in long-term care facilities. Relatively few patients with refractory psychosis are referred for electroconvulsive therapy (ECT) (Dold and Leucht, 2014; McIlwain et al., 2011). Although ECT shows some promise as an acute intervention, there are limited data on its long-term feasibility and efficacy (Petrides et al., 2015). Similarly, TMS shows promise as an acute intervention for AH but has lower response and remission rates when compared to ECT. Furthermore, TMS is not yet widely available clinically, has yet to be robustly evaluated in multisite randomized control trials, and may present logistical challenges in terms of accessibility and compliance.

DBS remains a relatively unexplored intervention for intractable AH in the context of schizophrenia. Preliminary investigations of DBS in animal models of schizophrenia have generally yielded positive results (Bikovsky et al., 2016; Ewing and Grace, 2013; Klein et al., 2013; Perez et al., 2013). Many of the targets in animal studies have been proposed for human trials, including the hippocampus, ventral striatum and associative striatum (Mikell et al., 2009; Mikell et al., 2015). Additional targets have also been proposed, including the medio-dorsal thalamus, internal globus pallidus and subcallosal cingulate gyrus. A prospective, double-blind clinical trial entitled Deep Brain Stimulation in Treatment Resistant Schizophrenia (NCT02377505) has been launched in Spain (Salgado-Lopez et al., 2016) with target sites of the nucleus accumbens and the subgenual anterior cingulate gyrus. The first patient in this trial has exhibited a positive trend in Positive and Negative Syndrome Scale (PANSS) scores, with positive symptoms showing a more robust decrease than negative symptoms (Corripio et al., 2016).

Caution has rightfully been urged when considering invasive network modulation for schizophrenia given its heterogeneous presentation between patients (Bakay, 2009) and its possible dynamic evolution within patients (Krystal and Anticevic, 2015). The aforementioned DBS trial for schizophrenia does not specifically target AH or VH, symptoms that are difficult to study in animal models of a symptom cluster. Moreover, network interventions targeting one specific symptom within a cluster may not address concurrent symptoms. Targeting AH in schizophrenia, for example, may not alleviate negative symptoms. The combined effects of medications like antipsychotics and DBS would need to be carefully explored, both in terms of efficacy as well as safety (e.g. clozapine lowering seizure threshold) (Arumugham et al., 2016).

As careful explorations into invasive circuit modulation for AH continue, it is important to explore the benefits of TMS as an experimental therapy as well as a causal tool with which to externally or non-invasively test network modulation in prospective patients. The first step in this process could be to evaluate if TMS therapy offers the patient transient or sustained relief from AH that have been refractory to multiple pharmacotherapy strategies. Data from patient reports, clinician evaluations and rating scales could be used to determine if a DBS evaluation should be considered. Perhaps the most compelling scenario would be one in which the clinical response to TMS is insufficiently robust or durable. In other words, the effects of TMS may only provide partial relief or may wear off over time.

There are few ways to address this problem with non-invasive stimulation. TMS has the capacity to induce neurophysiological changes that persist after the stimulation paradigm ends, but durable network changes often require multiple treatment sessions (and sometimes maintenance sessions) over the course of weeks or months (George and Aston-Jones, 2010; George et al., 2010b). Moreover, TMS therapy is limited in terms of parameters and cortical targets.

Several strategies have been explored in order to enhance the neurophysiological and clinical effects of TMS, including but not limited to extended acute phase treatment (Yip et al., 2017), maintenance sessions (Haesebaert et al., 2016; Kedzior et al., 2015; Philip et al., 2016; Richieri et al., 2013), new stimulation paradigms (e.g. theta burst stimulation) (Berlim et al., 2017; Brunoni et al., 2017; Desmyter et al., 2016) and new coil designs that increase stimulation depth at the expense of stimulation precision (Brunoni et al., 2017; Levkovitz et al., 2009; Levkovitz et al., 2015). Despite some progress, there are many unresolved questions about optimizing TMS treatment strategies and addressing TMS limitations.

DBS offers several potential advantages over TMS despite its own set of unresolved questions and limitations. Most of these advantages relate to parameter adaptability as well as target accessibility and precision. Unlike TMS, DBS enables exploration and manipulation of a wide range of continuous or variable stimulation parameters in any targeted fiber tract or network node. These dynamic, real-time adjustments might yield more immediate, robust or durable therapeutic effects than the serial cortical stimulation sessions (and likely maintenance or booster sessions) necessary to engender neuroplasticity with TMS. Another potential benefit to DBS is direct and spatially precise stimulation of white matter tracts or gray matter nodes. TMS (Drysdale et al., 2017; Fox et al., 2012a; Fox et al., 2012c) and DBS (Choi et al., 2015; Deeb et al., 2016; Noecker et al., 2017) studies have shown that precision matters, and DBS electrodes offer more spatial specificity than any current clinical or experimental TMS coil. One additional advantage to note is that DBS offers the rare opportunity to record neurophysiological data from the human brain. This information could lead to a greater understanding of the pathophysiology of AH and, by extension, improved treatment options.

5. A Potential Role for TMS in DBS Development

The idea of TMS informing DBS is a relatively new concept. There are few published manuscripts that explicitly explore the extent to which TMS can be used to predict DBS response, guide treatment or individualize targeting. Most of the work in this area has focused on movement disorders, in part because the basal ganglia-thalamocortical circuits are well mapped (Alexander, 1994; Wichmann and DeLong, 2016) and because stimulation of network nodes often results in predictable neurophysiological changes or measurable behavioral outputs. To this end, TMS has been used for perioperative preoperative mapping (Paiva et al., 2012; Picht et al., 2012) as well as intraoperative neurophysiological monitoring (Frey et al., 2012). More recently, studies have shown that fiber tracking is more accurate when TMS is used in conjunction with traditional anatomical tractography (Forster et al., 2015). TMS has also been used in conjunction with subthalamic nucleus DBS in order to induce associative cortical plasticity in patients with Parkinson’s disease (Udupa et al., 2016).

Some authors have begun to discuss the possibility of using TMS to inform DBS in regards to mood disorders (Pathak et al., 2016), but strong data are lacking. Lessons learned from movement disorders may not necessarily translate into studies of mood disorders, in part because the circuits are less well mapped and in part because stimulation of network nodes does not result in predictable neurophysiological changes or measurable behavioral outputs. From our perspective, these challenges reinforce the importance of using non-invasive stimulation modalities TMS in conjunction with multimodal neuroimaging as a way to carefully advance invasive neuromodulation interventions.

Functional connectivity is one of the few methods that links TMS to DBS in mood disorders. For example, the subgenual cingulate cortex has emerged as a brain region that is strongly correlated with therapeutic responses to TMS and DBS (Berlim et al., 2014; Drysdale et al., 2017; Fox et al., 2012b). A number of recent TMS studies have examined the connectivity between this and other brain regions in the so-called default mode network in order to predict or explain TMS response (Baeken et al., 2014; Liston et al., 2014). Some studies have used other modalities such as magnetoencephalography in order to show that patients who respond to TMS for depression exhibit durable changes in networks that are often targeted by DBS (Pathak et al., 2016).

Taken together, data from movement disorders and mood disorders suggests many ways in which TMS may be useful as an investigational screening tool during the rigorous pre-surgical workup for DBS. A number of imaging techniques could be used to assess the neurophysiological effects of stimulating or inhibiting the network of interest, from “online” approaches such as interleaved TMS/fMRI to “offline” approaches that involve functional imaging after repeated stimulation (Siebner et al., 2009). Online studies would be useful for exploring transient effects of single pulses as they are relayed from the cortical node to subcortical regions within the network in real time (Hanlon et al., 2013). These subcortical regions may become potential DBS targets. By contrast, offline studies could explore cumulative neurophysiological effects of rTMS sessions that acutely diminish AHs. Such studies could add critical information to the decision-making process prior to surgery. Although negative results may not completely preclude a patient from receiving further DBS consideration, particularly since there are likely differences between indirect and direct subcortical stimulation, a positive result would add useful data to an evaluation process otherwise devoid of individualized causal evidence.

Stimulation frequency is one of many unique areas to explore with an external stimulation tool such as TMS. The stimulation frequency employed in DBS protocols (generally 120–130 Hz and rarely below 50Hz) (Williams et al., 2014) tends to be significantly higher than the stimulation frequency employed in TMS protocols (generally 1–20 Hz but up to 50Hz for theta burst protocols)(Di Lazzaro et al., 2008; George et al., 2013a; Huang et al., 2005; Oberman et al., 2011; Rossi et al., 2009). Employing an external tool such as TMS might be a way to identify patients who could benefit from DBS at lower stimulation frequencies (e.g. 1–20 Hz). Additional work will be needed to determine if response to high frequency TMS could potentially predict novel low frequency DBS response.

The combination of TMS with structural, functional and effective connectivity analyses could further clarify how a network responds to stimulation prior to DBS implantation (Table 1). For example, Wernicke’s area may be aberrantly active and hyper-connected in patients who hear voices. Engendering something similar to long-term depression (LTD) in Wernicke’s area via 1 Hz rTMS may diminish its connectivity and hyperactivity (Briend et al., 2017; Gromann et al., 2012). Functional and effective connectivity analyses could provide information about whether and how coupling to this region changes as a function of symptom alteration. Such information would help to optimize TMS and possibly DBS interventions.

Table 1.

SYMPTOM	NETWORK	POTENTIAL TMS TARGET(S)	POTENTIAL DBS TARGET(S)
Auditory hallucinations	TPJ, Wernicke’s, Angular gyrus, Putamen, L IFG, Insula, Parahippocampal gyrus, Cerebellum	TPJ, Wernicke’s, Insula (dTMS), Cerebellum (dTMS)	Insula, Cerebellum, Putamen (associative)
Visual hallucinations	AMG, Visual cortex, L temporal pole, IFG	IFG, Visual Cortex, L Temporal Pole	AMG
Delusions	R DLPFC, Striatum, Insula	R DLPFC, Insula (dTMS)	Striatum (associative)
Catatonia	L DLPFC, ACC, Striatum	L DLPFC, ACC	Striatum (limbic)

Abbreviations: AMG = amygdala, ACC = anterior cingulate cortex, dTMS = deep transcranial magnetic stimulation, (L) = left, DLPFC = dorsolateral prefrontal cortex IFG = inferior frontal gyrus, (R) = right, TPJ = temporoparietal junction.

An important caveat to the power of connectivity analyses paired with brain stimulation techniques is the simple fact that neural networks are complex. Even basic circuits comprised of a few neurons (e.g. crustacean stomatogastric ganglion) have high redundancy and thus the consequences of perturbing particular nodes (or cell types) can be varied and difficult to predict (Marder, 2012). Perturbing the function of one network node may have unanticipated consequences throughout the network. Some, but not all, of these consequences may be explored with non-invasive probes such as TMS prior to invasive neuromodulation techniques.

Additional explorations of neural networks should be performed as TMS and neuroimaging technology advances. Traditional TMS coils are arranged in a figure-of-eight design that focuses the magnetic field into a small focal point in the center of the coil. This coil shape has been shown to have the capacity to stimulate neurons up to a few centimeters into the cortex (George et al., 2013a; George et al., 2013b). Different coil arrays are also available, including figure-of-eight coils with steeper angles and novel designs. Deep TMS (dTMS), for example, has the capacity to directly stimulate deeper cortical structures. One version of dTMS recently received FDA clearance for treating depression (Levkovitz et al., 2015). Unfortunately, there is a tradeoff between depth and precision; dTMS coils stimulate deeper but also stimulate more diffusely. Despite this tradeoff, there are many intriguing research possibilities with various dTMS coils in terms of targeting deeper structures or network nodes relevant to AH. Excitatory cerebellar dTMS or inhibitory insula dTMS have both been proposed as targets in order to disrupt the overweighting of perceptual priors (Powers et al., 2017).

6. The Potential for Closed Loop or On-Demand DBS

Traditional DBS for psychiatric and neurological indications involves fixed stimulation parameters that are episodically adjusted by a physician based on patient feedback and examination findings. New developments in technology have started to transform “open loop” designs with static parameters into “closed loop” designs with dynamic parameters. These closed loop designs involve real-time adjustments to stimulation parameters based on neurophysiological data. In responsive neurostimulation (RNS) for epilepsy, stimulation is selectively employed to reduce or abort epileptiform activity before it can entrain and generalize. (Morrell and Halpern, 2016; Sun and Morrell, 2014). Utilizing neurophysiological feedback to create intelligent stimulation paradigms that take into account brain rhythms and brain state (similar to cardioversion) could revolutionize network-based interventions for intractable symptoms such as AH in schizophrenia.

Closed loop stimulation is a relatively new technique for external neuromodulation paradigms. Most of the current studies focus on cortical excitability (Cancelli et al., 2016; Karabanov et al., 2016; Kraus et al., 2016; Raco et al., 2016; Zrenner et al., 2016). Employing a closed loop TMS design would require neurophysiological data that predict or encode AH in real-time. There are currently limited data with which to create such a closed loop. In a small cohort of patients, Hoffman and colleagues were able to identify resting brain activity in the prefrontal cortex and Wernicke’s area that portended reported AH perception by approximately 4 seconds (Hoffman et al., 2011). Perhaps an online TMS-fMRI design could be employed whereby the emergence of this real-time pattern could trigger 1 Hz TMS inside of the scanner.

The implementation of a closed loop design for AH in DBS remains unexplored. There are simply not enough data showing the neurophysiological signature of an AH before it is perceived. An alternative arrangement to closed loop DBS for AH would be a version of on-demand DBS. Unlike the continuous nature of movement disorders and depressive symptoms, AH can sometimes be experienced in more discrete episodes. A predictive biomarker for the onset of episodic symptoms such as AH could significantly enhance the potential of neuromodulatory therapy. Such a biomarker could inform physicians or the device itself to activate or adjust settings. The concept of patient-controlled stimulation is intriguing but relatively unexplored. There are many challenges inherent to implementing such a system for AH or in schizophrenia, particularly in light of negative symptoms, delusions and cognitive impairment. Patients with DBS for movement disorders are sometimes given the choice between two different stimulation paradigms. Perhaps a similar approach could be considered for high functioning patients with AH. For example, patients who experience acute onset AH or acute worsening of baseline AH could work in conjunction with a supervising physician to activate the device or change the paradigm in an on-demand fashion. This process may be a means by which to selectively perturb the network instead of simply entraining a new rhythm that eventually becomes maladaptive and resistant to change.

Computational modeling could play an integral role in implementing closed loop or on-demand DBS. One goal of computational modeling would be to increase the extent to which a particular symptom is associated with regional and inter-regional signals within networks. Cognitive processes, and in particular computational analyses of these processes, may be particularly appropriate metrics with which to set up closed loop or on-demand systems. For example, the Research Domain Criteria approach is predicated on relating circuits to cognitive functions and symptoms (Widge et al., 2016). A computational analysis of those processes and functions may serve as a basis for future DBS endeavors. In the case of AH, there are data to suggest that top-down prior beliefs (perceptual expectations based on previous experiences) sculpt perception. These priors can be recovered by the inversion of formal computational models and related to responses in particular regions. A causal role in the region-process relationship could be confirmed with TMS. Once confirmed, DBS could be implemented in that region in such a way that top-down priors could be regulated within an acceptable range.

7. Safety and Ethical Concerns

There are several safety and ethical concerns that should be considered in conjunction with scientific rationale and clinical efficacy(Mikell et al., 2015). There would need to be critically evaluated processes by which potential DBS candidates would be screened for eligibility. Part of these processes would involve determining what sort of treatment protocol should be explored prior to referral. Another important issue would be consent to the intervention, particularly in light of concerns about the time- and context-dependent nature of consent. Careful explorations into the manner in which proposed DBS would selectively target AH versus properly functioning speech processing would also need to be conducted. Finally, the best way to approach this conversation with patients, families and clinicians would need to be discussed. The possibility of ECT for DBS provides an illustrative example in this regard. Whereas ECT represents a powerful therapeutic intervention for various psychiatric symptoms, it is often underutilized because of concerns related to stigma and invasiveness. These concerns would only be magnified by the prospect of DBS.

7. Conclusion

Several factors have limited the development and implementation of DBS for intractable symptoms of psychiatric disorders, including symptom clustering within psychiatric constructs as well as a scarcity of casual tools with which to predict response, refine targeting or guide clinical decisions. TMS may offer a means by which to partially address these challenges. Individualized TMS therapy based on multimodal studies may offer some patients relief from intractable symptoms like AH that are typically treated with systemic pharmacotherapy. TMS responders may consider DBS workup if TMS therapy becomes insufficiently robust or durable. This graded progression from non-invasive to invasive therapies would be informed by some causal evidence that selective network manipulation is clinically beneficial for a specific symptom in a specific patient. From this point, “online” and “offline” strategies as well as computational modeling may individualize targeting in the pursuit of more sustained and adaptable network modulation via DBS. Although there are significant limitations and safety concerns regarding DBS, the combination of external or non-invasive brain stimulation techniques with computational modeling of advanced imaging and physiological data should provide critical insights into the future of DBS in psychiatry and neurology.