What can neuroimaging of neuromodulation reveal about the basis of circuit therapies for psychiatry?

Abstract

Neuromodulation is increasingly becoming a therapeutic option for treatment resistant psychiatric disorders. These non-invasive and invasive therapies are still being refined but are clinically effective and, in some cases, provide sustained symptom reduction. Neuromodulation relies on changing activity within a specific brain region or circuit, but the precise mechanisms of action of these therapies, is unclear. Here we review work in both humans and animals that has provided insight into how therapies such as deep brain and transcranial magnetic stimulation alter neural activity across the brain. We focus on studies that have combined neuromodulation with neuroimaging such as PET and MRI as these measures provide detailed information about the distributed networks that are modulated and thus insight into both the mechanisms of action of neuromodulation but also potentially the basis of psychiatric disorders. Further we highlight work in nonhuman primates that has revealed how neuromodulation changes neural activity at different scales from single neuron activity to functional connectivity, providing key insight into how neuromodulation influences the brain. Ultimately, these studies highlight the value of combining neuromodulation with neuroimaging to reveal the mechanisms through which these treatments influence the brain, knowledge vital for refining targeted neuromodulation therapies for psychiatric disorders.

Introduction

Neuromodulation is an emerging therapeutic option for individuals with psychiatric disorders that are resistant to standard frontline treatments [1, 2]. When patients do not respond to cognitive behavioral therapy, pharmacotherapies, or even electroconvulsive therapy they become candidates for neuromodulation therapies that attempt to alleviate symptoms by targeting a specific area, circuit or group of circuits. The current neuromodulation approaches include deep brain stimulation (DBS) [2, 3], transcranial magnetic stimulation (TMS) [4], transcranial direct current stimulation [5, 6], and more recently focused ultrasound stimulation (FUS) [7]. When targeted to specific parts of the brain, and increasingly the exact location is personalized for each individual, these therapies have been associated with significant reductions in patient’s symptoms [8, 9]. In a few cases the level of response has been so great that the individuals enter remission [9–11]. Thus, neuromodulation therapies hold the promise of personalized medicine for psychiatric disorders.

Despite the progress that has been made in the past decades, not all individuals with psychiatric disorders respond to neuromodulation treatment. For example, a multi-site trial for DBS for depression targeting the subcallosal anterior cingulate cortex (ACC) produced variable clinical outcomes [12]. While many factors likely contribute to different clinical responses, at least part of the variability stems from the fact that neuromodulation approaches are applied to a specific brain circuit based on knowledge about the anatomy and functions of that brain area and brain circuit in health and disease. It is important to note that knowledge about the functions of specific brain areas and circuits is still actively being acquired and at present no single circuit that has been identified as being dysfunctional in either anxiety, schizophrenia, or depression. Thus, a key challenge in the field is to determine the biomarkers that are predictive of a specific circuit dysfunction and clinical response as well as the mechanisms through which neuromodulation has its beneficial effects on those circuits. Such knowledge is essential for guiding the application and refinement of targeted neuromodulation therapies for psychiatric disorders. For instance, while the multi-site clinical trial for subcallosal DBS for depression was stopped, the protocol used for this approach has since been extensively refined based on careful analysis of diffusion neuroimaging data from responders and non-responders [8, 13, 14]. Now the current protocol uses the confluence of specific fiber tracts that is individually defined for each patient to guide the placement of DBS leads [15]. This refined protocol is associated with greater than 70% response rates in patients that were previously treatment resistant [11].

One of the insights that helped motivate the use of approaches like DBS for depression came from neuroimaging work in healthy and clinical populations indicating that a specific brain area was involved in mood [3, 16]. While human neuroimaging investigations are essential for the further refinement of neuromodulation therapies [17], what is also needed are fine-grained insights into the neurophysiology and macro and micro-level anatomy of brain circuits that are impacted by targeted neuromodulation. This level of resolution cannot be obtained with current non-invasive neuroimaging methods in humans as they currently lack the spatial resolution to resolve single neuron responses or single axons. This necessitates the use of animal models, especially those animals with brains most similarly organized to humans such as nonhuman primates [18]. Indeed, in recent years several studies in nonhuman primates have combined neuroimaging methods with other modalities to investigate how neuromodulation of a specific brain area alters the function of that area as well as the circuits connecting to that area. Understanding the neural mechanisms engaged by neuromodulation therapies at levels of analysis not possible in humans will provide information essential for refining neuromodulation approaches, novel targets for neuromodulation as well as insights into the pathology of diseases that can be translated back to the clinic (Fig. 1). Here we will review studies that have combined neuroimaging with neuromodulation in nonhuman animals. To help provide a context for this work we begin by briefly reviewing the different types of neuromodulation approaches that have been used to treat psychiatric disorders.

Fig. 1. Cross talk between human and animal studies for the development of neuromodulation for psychiatric disorders.

Schematic of the development of human neuromodulation focusing on development of DBS for treatment resistant depression. Further refinements and novel approaches for neuromodulation will require crosstalk between studies in humans as well as those in animal models that allow access to the cellular level mechanisms that are engaged by clinically effective neuromodulation. Figure parts adapted with permission from refs. [3, 15, 18, 90, 106, 130, 156].

Neuromodulation therapies in humans for psychiatric disorders

As noted above, a number of neuromodulation strategies are being developed to treat a range of psychiatric disorders including depression [1], anxiety [19, 20], schizophrenia [21], autism spectrum disorder [22], substance use disorders [23], chronic pain [24, 25], and more. An exhaustive account of all the various approaches that have been advanced in recent years and their relative efficacy for providing symptom relief for each disorder is beyond the scope of the current review. Comprehensive reviews on neuromodulation and its clinical applications can be found here [1, 26, 27] but there are many more. Below we highlight some of the promising neuromodulation therapies that have been developed for two psychiatric disorders, depression and obsessive-compulsive disorder (OCD). We focus on the circuits that are being targeted, the specific neuromodulation methods, the clinical outcomes of those treatments, and the recent refinements that have been made to improve their effectiveness using insights from neuroimaging.

Depression

Major depression is the leading cause of disability world-wide [28]. Depending on the criteria used between 10 and 50% of people with depression are treatment resistant in that their symptoms are not appreciably reduced by standard front-line treatments [29, 30]. With the advent of functional neuroimaging over three decades ago, a series of seminal positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) studies found that major depression was associated with dysfunction within a network of areas and specifically within the subcallosal ACC [16, 31–35]. Notably, a role for subcallosal ACC in negative mood was not limited to people with depression but was also seen in healthy individuals experiencing transient sadness [16, 36]. Furthermore when depressed individuals are successfully treated with cognitive-behavioral, pharmacotherapies, or neuromodulation therapies there is a decrease in activity within the subcallosal ACC [34, 37, 38], indicating that hyperactive or dysfunctional activity with the subcallosal ACC is central to the pathophysiology of depression.

With this information about dysfunction within a circumscribed part of the brain and its relation to the etiology of a psychiatric disorder, Mayberg and colleagues specifically targeted the subcallosal ACC with high-frequency DBS in patients with treatment resistant depression [3]. In this first cohort, four out of six patients showed an improvement in symptoms. Longitudinal follow up studies of the original cohort and other patients that have received subcallosal DBS have shown that this treatment is effective and clinical responses are sustained many years after surgery [11]. In part because of the variability in treatment response during a large cohort study [12], the surgical placement of the DBS leads targeting subcallosal ACC has been further refined. Now the current treatment strategy targets the confluence of three white matter fiber tracts that is adjacent to the subcallosal ACC [8, 15]. Importantly this confluence of white matter tracts has to be individually defined as there is significant spatial variability in its location [15], and this detail may well explain why many people failed to show improvement when this approach was not taken [13]. Taking this approach as well as combining it with intra-operative biomarkers of treatment response, over 70% of patients being treated show a positive response and as many as 60% achieve remission [15, 39, 40]. Thus, DBS targeted to the white matter near the subcallosal ACC has proven to be a highly successful targeted neuromodulation therapy for treatment resistant depression with long term efficacy.

Of course, the white matter adjacent to subcallosal ACC is not the only DBS target developed for depression. Other targets include the nucleus accumbens, anterior limb of the internal capsule (ALIC), thalamic peduncle, medial forebrain bundle, lateral habenula, orbitofrontal cortex (OFC), and bed nucleus of the stria terminalis [2, 41]. Given the variability of presentations of the depression as well as of functional neuroimaging biomarkers of depression [42] one key challenge for future research will be determining which DBS target will likely have the best clinical response for each patient.

Despite DBS being a well-tolerated and safe therapeutic option, invasive neuromodulation approaches carry inherent risks such as those associated with the neurosurgical placement of DBS leads in the brain. Consequently, developing alternative non-invasive neuromodulation approaches has the potential to reduce some of these risks as well as make neuromodulation therapy more accessible. One FDA-approved non-invasive neuromodulation therapy for depression is high-frequency TMS delivered to the left dorsolateral prefrontal cortex [43]. The initial protocol for this approach was associated with variable clinical outcomes and relapse [44, 45]. Careful analysis of relationship between the functional connectivity of the locations where TMS was delivered and subsequent clinical response, however, revealed that the biggest reduction in symptoms occurred when the parts of the dlPFC that were connected to subcallosal ACC were stimulated [4, 46, 47]. Consequently, a recent double-blind randomized clinical trial used repeated theta-burst stimulation delivered to parts of the left dlPFC in each individual with high functional connectivity to subcallosal ACC. In this trial, more than 50% of participants exhibited a significant reduction in symptoms after 4 weeks of connectivity guided TMS [9]. While this is a lower rate of responding compared to the that reported in an open label trial using the same approach [48], the effects are nevertheless indicative that this targeted neuromodulation therapy can provide symptom reduction even if the long term efficacy and robustness of the effects is still to be determined.

As we highlight above, multiple lines of evidence point to the effects of dlPFC TMS on depression being mediated through altering connectivity between this area and subcallosal ACC. This highlights one of the limitations of TMS; it is unable to directly influence activity in deep brain structures that are located below the superficial cortex. To provide direct non-invasive neuromodulation of subcallosal ACC, Riis, Mickey, and colleagues recently used FUS to target the subcallosal ACC in a single patient with treatment resistant depression [7]. In this initial case, they reported that the patient experienced a decrease in depression symptoms that were sustained for over 6 weeks after treatment. What is needed now are prospective clinical trials to fully establish the clinical efficacy and robustness of this non-invasive approach. It will also be important to determine the specific locations within subcallosal ACC that produce the most robust effects. In particular this will require figuring out whether anatomical landmarks or functionally defined criteria such as connectivity between subcallosal ACC and some other part of the brain should be used to guide FUS. Irrespective of this, when targeted to the subcallosal ACC, FUS like DBS and TMS neuromodulation approaches has shown promise as a treatment for depression.

OCD

Approximately 2% of the world-wide population suffer from OCD [49]. While many individuals with OCD respond well to cognitive or pharmacotherapies, a small proportion do not and are described as treatment resistant or refractory [50]. As early as the 1950’s psychosurgeries, where lesions were placed in specific parts of the brain, were being trialed to manage OCD patients’ symptoms [51, 52]. Ablations of the striatum, cingulate cortex, or tractotomies of the capsule and internal capsule were all used to aim to reduce symptoms in patients with treatment resistant OCD [53]. While many patients undergoing these procedures experience improvements in their symptoms, the proportion of people showing a clinical response was often only around 50% of the total sample (for example, [54]). Subsequent efforts to alleviate symptoms of OCD including intrusive thoughts, agitation, impulsivity, and repetitive actions focused on using DBS neuromodulation therapies [19].

At this point it is worth noting the differences between lesion and neuromodulation therapies, such as DBS. While both approaches can be targeted to change the function of specific brain circuits, lesions made using electrolytic or gamma knife procedures are permanent, damage both white matter and gray matter, and also initiate a cascade of degenerative processes that lead to plastic reorganization of brain on different timescales [55]. Such lesion therapies continue to be used for OCD as they are effective and potentially better tolerated than DBS [56, 57]. By contrast neuromodulation, has the advantage of not permanently damaging the brain (aside from the insertion of DBS leads), stimulation can be dynamically adjusted based on the individuals’ response over time, and the location where stimulation is delivered can be varied to improve the clinical response.

Similar to the lesion approaches, the first target that was used for DBS for OCD was the internal capsule (IC) with one patient reporting symptom improvement [58]. Subsequent refinements of this approach have focused heavily on targeting of electrodes based on an individuals’ patterns of connections within the IC [10]. Further, it is theorized that the beneficial effects of IC DBS are the result of normalizing activity within distributed networks linking cortex, the subthalamic nucleus (STN), thalamus and brainstem [10, 59]. Despite this, the precise macro and microscale mechanisms that underlie successful DBS therapy for OCD are still to be fully revealed although recent reports in patients indicate that clinical efficacy are related to reducing periodic oscillations in the delta/alpha frequency bands of the LFP in striatum [60].

Similar to depression, TMS therapies have also been developed for OCD. Initial treatments for OCD delivered trains of TMS pulses to DLPFC [61] as well as other parts of the dorsomedial cortex [62]. More recently parts of the OFC and frontopolar cortex [63] have also been targeted. While effective for some individuals, large scale studies as well as meta-analyses indicate that these TMS approaches only offer moderate clinical efficacy [64, 65]. The specific brain areas being targeted were, however, guided by gross anatomical landmarks such as the distance from the central sulcus [64]. More recently, various groups have started to use individualized functional neuroimaging connectomic based targeting of TMS, selecting locations in OFC/frontopolar cortex that exhibit high resting-state functional connectivity with the ventral striatum [20]. Initial, open label studies show that combining this targeting with accelerated TMS delivery can ameliorate symptoms in some patients although further refinements for targeting and randomized control trials are needed to fully discern the efficacy of this non-invasive neuromodulation approach for OCD.

FUS has also been developed as a potential neuromodulation therapy for OCD, but the use of FUS in this case has mostly been limited to non-invasively making lesions in the internal capsule [66, 67] mirroring the approach taken in the aforementioned psychosurgeries. Alternating current therapies have also been developed for OCD and these have been reported to produce sustained reductions in compulsive behaviors when applied to OFC/frontopolar cortex [68]. The next few years will likely bring a rapid expansion of non-invasive approaches for OCD especially now that there is more consensus about which pathways are the ones that are associated with positive clinical responses [10].

As the previous review of the current invasive and non-invasive neuromodulation approaches for depression and OCD emphasizes, while some people respond to these therapies, others do not. Improvements in clinical outcomes will likely come from a number of sources including (1) refining which areas of the brain to target and (2) better understanding how neuromodulation approaches, both invasive and non-invasive alter neural activity within and the structure of distributed networks. Below we highlight research in both humans and animals that has begun to address these two points.

Combining neuromodulation and functional methods in healthy people and patients with treatment resistant psychiatric disorders

More than any other therapies for psychiatric disorders, neuromodulation approaches are critically dependent on information about the areas that are impacted in psychiatric disorders and how neuromodulation impacts local circuits and distributed networks. As we noted above, neuroimaging of intrinsic functional connectivity has been explored as an approach to establish biomarkers of distinct psychiatric disorders (for example, [42]). Further, based on functional connectivity between parts of lateral frontal cortex and targets of interest, one of the recent refinements for targeting TMS therapies for psychiatric disorders has been to use personalized intrinsic resisting-state functional connectivity to guide where stimulation is delivered (for example, [48]). Of course, an important piece of the puzzle to understand is how neuromodulation changes local and network-level activity in the human brain. This has been a focus of a number of recent reviews [69, 70] and below we highlight the key insights that the few studies in this area have revealed.

In their initial clinical trial on the effects of subcallosal ACC DBS, Mayberg and colleagues used PET imaging to identify the areas that were altered after successful treatment [3]. These analyses revealed that DBS caused a decrease in activity within the subcallosal ACC as well as other parts of the anterior medial frontal cortex. DBS-mediated symptom improvement was also associated with an increase in activity relative to baseline within the posterior cingulate cortex (PCC) and other parts of the default mode network (DMN). This pattern of effects has been replicated in subsequent studies where imaging has been conducted up to 6 months after the beginning of stimulation [71, 72]. Further, individuals that showed a clinical response to treatment exhibit a different of activation relative to those that didn’t. However, the use of PET imaging means that it is not possible to look at the relationship between changes in different brain areas, somewhat limiting what can be learned about how DBS alters functional networks.

Developments in MRI and electrode technology mean that it is currently possible to safely acquire structural and functional scans data from people with implanted neuromodulation devices. While this approach has been used extensively for movement disorders (for reviews see, refs. [69, 70]), fewer studies have been conducted for psychiatric DBS applications. One of the first to take such an approach, were Figee and colleagues who scanned a group of people that had been implanted with leads targeting the ALIC to control treatment resistant OCD [73]. When they compared the effects of having the stimulation on versus off, they found that clinically effective DBS delivered to the ALIC was associated with both local changes in activity in striatum as well as changes in functional connectivity between the frontal cortex and striatum. This use of neuroimaging therefore proved that successful ALIC DBS is associated with altering functional connectivity between striatum and frontal cortex. Follow up studies by this group highlight that modulating connectivity between prefrontal cortex and amygdala may also contribute to the clinical effectiveness of ALIC DBS [74]. The precise neural mechanisms associated with these changes in functional connectivity are less clear although the same study also reported that ALIC DBS changed local circuit oscillatory activity in frontal cortex. This highlights that one of the critical pieces that is missing from our understanding of how DBS works is the neural mechanisms – the patterns of single neuron and local field potential activity – that are engaged, a point we take up later.

More recently, Elias and colleagues obtained functional scans from a cohort of people who had been implanted with DBS targeting the subcallosal ACC, allowing DBS-associated changes in functional interactions between brain areas to be assessed [75]. Notably, resting-state functional neuroimaging data was acquired when stimulation was turned on at clinically effective lead contacts, on at a clinically suboptimal contacts, and when it was fully turned off. Clinically optimal stimulation was associated with decreased activity within PCC, dACC, and other parts of the midline cortex. Further, PCC was less functionally connected to the ventromedial frontal cortex indicating that DBS is decreasing functional connectivity with in the DMN. Note that this finding with fMRI is somewhat distinct from that reported from PET imaging of people receiving successful DBS stimulation; in the studies by Mayberg and colleagues, DBS was associated with increased activity in parts of DMN. Why the patterns of activity between PET and fMRI are different is unclear although these two techniques measure different aspects of brain activity. In addition, it is important to note that while Cha and colleagues [72] reported an increase in DMN activity after 6 months of stimulation, they actually found decreased activity in DMN shortly after DBS onset. Thus, the length of time stimulation has been delivered for may be a factor in how DMN activity is modulated.

The effect of neuromodulation for depression may not, however, be solely associated with altering activity within the DMN. Recently, studies using TMS to treat depression have revealed that this neuromodulation approach is also associated with altering activity within lateral frontal cortex that are associated with the central executive network [76, 77]. This point highlights that successful neuromodulation for disorders such as depression may require activity across multiple intrinsic brain networks to be affected.

While the aforementioned studies used PET and fMRI as a read out of neuromodulation, Scherer and colleagues looked at how subcallosal DBS altered brain-wide patterns of activity using magnetoencephalography (MEG) in responders, non-responders, and healthy controls [78]. Other groups have also used EEG to probe these effects as well [79] and these studies have the added advantage that they can begin to discern the specific patterns of activity associated with successful neuromodulation. Scherer and colleagues found that stimulation in responders was associated with a decrease in alpha-band (8–13 Hz) power in parts of posterior medial cortex including retrosplenial cortex as well as parts of lateral frontal cortex. The effect of DBS on activity in medial frontal cortex is notable as it somewhat matches the effects reported from PET and fMRI studies, and thus indicates that a critical component of successful DBS for depression is modulating activity in posterior parts of the medial cortex.

The white matter tract that connects subcallosal ACC to the posterior medial areas is the cingulum bundle [80] and it is this tract that is one of those targeted by some of the current approaches for DBS for depression [8]. Recently, in a small sample of people receiving subcallosal DBS for depression, it was found that the degree of myelination in cingulum bundle as measured using diffusion imaging before lead implantation was correlated with the degree of clinical recovery [40]. While only a small sample, this finding potentially highlights that while most hypotheses about the mechanisms of action of neuromodulation emphasize the functional effects of stimulation [81, 82], it may also be the case white matter in the brain may be impacted by stimulation. This possible role for DBS in remodeling of white matter is unproven but, potentially predicted by preclinical work in humans and rodent models [83, 84].

A role for neuroimaging in animal models to determine the neural basis and use of neuromodulation therapies

While the previously highlighted neuroimaging, intracranial recording, and MEG studies in humans have revealed the distinct networks and patterns of activity that are altered by neuromodulation therapies, these are unable to provide insight into the specific mechanisms of action at the level of brain cells and circuits. Obtaining cellular and subcellular-level resolution is, however, not possible in humans. In addition, research in animal models also has the potential to help to refine these treatments by enabling new brain targets and parameters to be tested before clinical use. Because their brains are more similar to humans, nonhuman primates provide a way to conduct more directly translational research into how neuromodulation therapies affect the brain, although we note that work in rodents will continue to provide fundamental insights [85]. Importantly, gray and white matter structure and organization are similar between humans and macaques [80, 86, 87] and intrinsic resting-state functional networks are similarly organized in both species [88]. Such networks are organized differently in rodent models [89] limiting the translation of MRI network-based findings from rodents to humans. Indeed, the similarity in resting-state functional networks between humans and macaques provides a unique opportunity to systematically test neuromodulation approaches and their parameter space while obtaining whole brain readouts of the treatment’s effects. Below we discuss how using functional and structural neuroimaging in the context of different neuromodulation approaches has helped to reveal how these treatments impact brain circuits. Before that we highlight the work that has combined neuroimaging in humans and macaques with tract tracing.

Characterizing the circuits of impacted by deep brain stimulation using comparative neuroanatomy

As mentioned above, current protocols for DBS for depression and OCD increasingly target specific fiber tracts in the brain on an individual subject basis [10, 15]. The location and arrangement of fiber bundles in the brains of each patient can be estimated by obtaining diffusion-weighted images and then conducting tractography. Using this approach is essential for prospective targeting of DBS leads in humans as well as improving basic understanding of which specific area-to-area connections need to be modulated to produce a clinical response. The limitation of diffusion tractography is, however, that it does not provide ground truth about the specific anatomical pathways that connect through a particular location in the brain, as the current resolution in humans is on the order of millimeters. To characterize the specific area-to-area connections that are likely being influenced by DBS, researchers have turned to comparative analyses of connections in nonhuman primates, specifically macaques.

In the context of neuromodulation for psychiatric disorders, Haber and colleagues has taken a combined MRI and anatomical tracing approach in macaques to discern the pathways that are impacted by DBS for OCD. First, they confirmed that macaques share similar connectivity and fiber tract organization with humans [80] and then they conducted tract-tracing experiments to determine the specific connections that course through different locations in the brain [90]. Specifically, they compared the paths of major fibers from the ventral frontal cortex and other parts of the brain through the ALIC in both humans and macaques [91]. Careful analysis of the ALIC has shown that fibers coursing through this area have a specific dorso-ventral and anterior-posterior arrangement that is conserved across macaques and human. They and others then looked at the connections that pass through the parts of the ALIC that are associated with the greatest reduction in OCD symptoms following DBS treatment [10, 92]. The greatest clinical response was seen when DBS was directed to fibers connecting both the ACC and ventrolateral prefrontal cortex (vlPFC) with parts of the thalamus and STN [10]. Such fine-grained detail is essential for (1) confirming that diffusion estimates of connectivity between brain structures are accurate, (2) establishing the pathways that need to be modulated by DBS to have a clinical response and (3) potentially provides insight into the parts of the brain affected in OCD.

Revealing the circuit-level impact of direct brain infusions of pharmacological agents with functional neuroimaging

Neuroimaging of animal models with targeted brain lesions can provide important insights into functional reorganization following loss of a brain region [55, 93]. The emerging neuromodulation approaches used to treat psychiatric disorders, by contrast, were specifically developed to move beyond chronic or permanent changes to the brain in order to be more selective in their effects. Direct brain infusions of a pharmacological agent can be used to transiently increase or decrease activity within an area or be used to probe the relationship between specific receptors within an area and a particular behavior or cognitive function. The majority of research using these methods have focused on the behavioral changes following direct brain infusions (for example, [94, 95]). This has provided unprecedented knowledge about the functions of the targeted region and in some cases the specific role of receptors within those areas (for example, [96]). More recently, a number of research groups have begun to combine these methods with neuroimaging to better understand the underlying neural circuitry that is structurally and functionally connected to the targeted brain regions. This work is essential if we are to understand how targeted delivery of neuromodulation or a pharmacological agent to one part of the brain impacts network function. Indeed, the possibility of non-invasive release of a drug in a brain area using neuromodulation in humans is close to becoming a reality [97–99]. Thus, the critical question becomes which areas and pharmacological agents should be used to treat depression or OCD.

The basal forebrain is the origin of many of the cholinergic and GABAergic projections to the cortex [100] and is heavily implicated in pathology of Alzheimer’s disease which is often comorbid with depression [101, 102]. Thus, understanding the role of this area in influencing activity or connectivity across functional brain networks is foundational to developing neuromodulation approaches for this disorder [103]. To understand the brain-wide impact of inhibition of the basal forebrain, Turchi and colleagues acquired functional neuroimaging data after infusions of either muscimol or saline were made into this structure in macaques [104]. Further, injections of muscimol were targeted to distinct subdivisions of basal forebrain to assess how this differentially modulated global brain signals. They found that regardless of location, inactivation of the basal forebrain had relatively little effect on the network organization of intrinsic functional resting-state networks. By contrast, what they did find was that activity within specific brain regions was suppressed depending on which subdivision of the basal forebrain was inactivated, and the pattern of suppression matched the known distribution of cholinergic and GABAergic projections to the cortex [100, 105]. Taken together, the results suggest that the basal forebrain is critical to control of global connectivity fluctuations in the brain but does not specifically modulate inter-regional functional communication that is seen in functional resting-state networks.

As the above study by Turchi and colleagues demonstrates, the effects of direct brain infusions are not seen across the whole brain. Instead, neuromodulation often only affects a circumscribed network of brain areas that are connected to the location that is modulated. Understanding how activity in one area influences activity across the brain is essential for revealing the specific network impacted by neuromodulation as well as revealing the signature of dysfunction within one part of the brain. Hyperactivity within the subcallosal ACC has been repeatedly found in people with depression as well as people experiencing transient sadness [16, 36]. To model this effect in primates, Alexander and colleagues used microinfusions of the glutamate transporter inhibitor dihydrokainic acid to over activate marmoset subcallosal ACC [106]. They reported that shortly after infusions were made, marmosets exhibited depressive-like behaviors. Further, using fluorodeoxyglucose PET, they showed that over-activation altered neural activity in a set of regions connected to subcallosal ACC that have been implicated in emotional response maintenance, including the insula, dorsal ACC, and dorsomedial prefrontal cortex. In addition, broadly increasing activity within the subcallosal ACC altered activity within the dorsal raphe, a nucleus that contains serotonergic projection neurons. Such an effect is notable given that many pharmacological treatments for depression target the serotonergic system [107]. Taken together, combining neuroimaging with direct pharmacological infusions revealed the networks of areas that are causally impacted by hyperactivity in subcallosal ACC and thus provides information about potential targets for the neuromodulation for disorders characterized by pathologically increased subcallosal ACC activity such as depression.

Chemo- and optogenetic manipulation of brain activity combined with functional neuroimaging

Now firmly established in basic neuroscience research, chemo- and optogenetic approaches allow activity within a specific set of neurons to be either decreased or increased with a high degree of precision [108–110]. Both methods work by using a viral vector to transduce non-endogenous receptors into neurons that can be activated by either a non-endogenous ligand in the case of chemogenetics or specific wavelengths of light in the case of optogenetics. While both approaches are virally mediated, their mechanisms of action on neural activity are quite distinct. Optogenetic stimulation is most similar to electrical microstimulation in that its effects are immediate and have a high degree of temporal precision [111, 112]. By contrast, chemogenetic stimulation is more similar to the effects of a direct drug infusion, with longer lasting effects, a slower time course, and can be used to influence a larger piece of brain tissue [108]. The main benefit of both techniques is that the viral mediated transfection has the promise of targeting specific cell types and pathways allowing for more precise control of neural activity as has been done extensively in mice (for example, [113]). The differences between the techniques mean that they can be used to provide different sights into brain function.

Optogenetic neuromodulation and fMRI

Optogenetic neuromodulation has become a gold-standard neuroscience technique in rodents, however its translation to primates has been slow mainly due to issues related to viral transfection and the amount of brain tissue that can be influenced relative to the size of the brain [109, 114]. Despite this, in recent years a number of groups have used optogenetics in combination with neuroimaging to understand how distributed circuits can be dynamically modulated in nonhuman primates [115]. For instance, two studies from about a decade ago used channelrhodopsin 2 (ChR2), a blue light-sensitive cation channel, to stimulate neurons in frontal eye fields during a visual attention task [112, 116]. Both studies found that saccades latencies were modified following FEF stimulation, and that activity in functionally relevant distal regions was also modified following stimulation. More recently, Ortiz-Rios and colleagues [117] used optogenetic stimulation of the primate primary visual cortex, V1 to determine how laminar activity in cortex was related to the network level effects of stimulation. When they optogenetically stimulated area V1, they found increased functional activity throughout the visual system. Importantly they were able to confirm through electrophysiology and histological analysis that the strongest optogenetic signals were in layer 4B of the transfected region of V1, and these neurons had direct projections to the regions that exhibited increased activity during functional neuroimaging. Taken together, these studies show the promise of combining of optogenetic stimulation with neuroimaging, to help reveal the precise contribution of a region, cell types within a region or even the patterns of activity to brain-wide networks. As we noted earlier, however, the full promise this approach has yet to be realized in macaques, but could be used to determine how specific patterns of activity differentially influence distributed brain circuits that are implicated in psychiatric disorders.

Chemogenetic neuromodulation and fMRI

Because they are activated by a pharmacological ligand, the effects of chemogenetic modulation tend to last for tens of minutes to a few hours. This profile means that, in time, they be amenable to use in humans to enable biologically mediated neuromodulation for psychiatric disorders [118, 119]. Chemogenetics, specifically designer receptors exclusively activated by designer drugs (DREADDs) are rapidly being adopted by many researchers working with nonhuman primates as they enable investigation of the causal relationship between a brain region or network of areas and particular behaviors [120]. While the use of DREADDs in nonhuman primates is still trailing rodent work, there has been considerable progress in the last decade towards gaining better transfection, activation, and specificity in nonhuman primates [121–125].

In part, this progress has due to the use of neuroimaging to help evaluate the effects of chemogenetic manipulation, and recently these tools have been used to begin to evaluate the efficacy of DREADDs as a possible treatment approach for psychiatric disorders. Much of this work has used PET imaging to evaluate the binding of DREADD actuators to both non-endogenous DREADD receptors and endogenous receptors [121, 126]. Of relevance to the advancement of chemogenetic in nonhuman primates has been the use of deschloroclozapine (DCZ) as a highly specific DREADD actuator, in comparison to the more generally used clozapine-N-oxide (CNO). This new actuator allows for more precise targeting of the non-endogenous DREADD receptors and has been shown in both PET [121] and now fMRI [127] studies to have minimal off-target effects at doses that are functionally effective. Additionally, recently orally administered DCZ has been shown to be an effective method for activating DREADD receptors [122]. Taken together, the specificity of DCZ and its ability to modulate both neural activity and behavior through oral administration increases the possibility of using DREADDs as a potential neuromodulatory therapy for neuropsychiatric disorders.

While chemogenetic approaches have been used to alter affective states by altering activity within specific pathways emanating from subcallosal ACC in nonhuman primates [128], as yet no studies have used DREADDs to alter pathological brain states related to psychiatric disorders. As a proof of principle for this, Miyakawa and colleagues used an inhibitory DREADD to alleviate the effects of cortical seizures in macaques [129]. They were able to show that after infusions of virus coding for the inhibitor DREADD receptor, administration of DCZ was able to suppress seizures. Additionally, using PET imaging they confirmed that DCZ only activated the tissue that was targeted. As such, this use of PET imaging confirmed that a biological neuromodulation approach is able to specifically alter pathological activity in a brain area in a translationally relevant way.

With relevance to the circuits impacted in depression and OCD, Elorette, and colleagues targeted the basolateral portion of the amygdala with an inhibitory DREADD in order to determine the basis of fMRI functional connectivity between this area and frontal cortex [130]. Single neurons in amygdala project widely across frontal cortex [131, 132] and dysfunction within these pathways has been repeatedly implicated in a variety of psychiatric disorders including depression, schizophrenia, and anxiety disorders [133–136]. They found that when the inhibitory DREADD receptors in the amygdala were activated with DCZ, resting-state fMRI connectivity between the amygdala and frontal cortex, particularly with the vlPFC increased.

To further probe the basis of such changes they used electrophysiological recordings to determine the neural signature of such DREADD induced increases in fMRI functional connectivity. Notably, increased functional connectivity between amygdala and vlFPC was closely related to increased coherence in the alpha band. No other frequency band showed such modulation. These results are potentially important for further developing neuromodulation therapies for psychiatric disorders for a number of reasons. First, they confirm that biomarkers of differences in fMRI resting-state connectivity in fronto-limbic structures have a specific signature in neural activity and could potentially be used as basis for selecting specific parameters for stimulation. The second is that the neural signature of functional connectivity reported here between amygdala and vlPFC (alpha band, 8–14 Hz) is different to those from sensory areas (beta band, 20–30 Hz)(for example, [137]). Such information is likely important for understanding how to modulate specific pathways connecting to the limbic system with invasive or non-invasive neuromodulation approaches, although we note that it is possible that each pathway in the brain could communicate in a specific frequency band. Third, the finding that altering activity in amygdala was specifically related to alpha-band coherence is in accordance with the previously discussed effects of DBS on alpha band activity changes in the DMN in individuals with depression [78]. This potentially points to towards this frequency band being a key target for neuromodulation. We note, however, that little other information exists concerning the network-level basis of experimentally induced dysfunction within one part of the brain at the level of single neurons and population activity. This highlights the need for further experiments on this issue. Finally, this work in macaques with DREADDs highlights that activating an inhibitory chemogenetic receptor in a brain area does not guarantee that there will be a concomitant decrease in activity or functional connectivity associated with that area. Thus, if chemogenetic approaches are to be translated to humans, careful analysis of the effects of biological neuromodulation in nonhuman primates will be required to ensure they are having the desired effects.

Electrical manipulation of brain activity combined with functional and structural neuroimaging

Electrical manipulation of the brain, both invasively and non-invasively, has long been a means of treating neuropsychiatric disorders [138, 139]. However, as is often the case, the underlying mechanisms through which these treatments work are either still being debated or only recently being discovered [81, 82]. Nonhuman primates represent a unique opportunity to reveal the mechanisms of neuromodulation as neural tissue can be stimulated and monitored using invasive methods to obtain both whole brain signals, as well as microscale data on the functional and structural effects of stimulation. Most relevant to neuropsychiatric disorders is the effects of DBS as it has increasingly become an alternative for patients with treatment resistant disorders.

Matching the neuroimaging work that has been conducted in humans with implanted DBS devices, in the last decade a number of groups have used fMRI in combination with other modalities in macaques to understand the effects of DBS. At present there is a paucity of data on the effects of DBS for either depression or OCD. Instead, initial insights into how DBS modulates neural activity have come from studying the effects of STN DBS, one of the targets that has been pioneered for the treatment of Parkinson’s Disease [140]. Taking this approach, a number of research groups have found that DBS stimulation of STN induces widespread increases in neural activity in motor related networks in macaques [141, 142]. Importantly, these studies revealed that the exact site of stimulation within the STN had a significant effect on fast-scan cyclic voltammetry measures of dopamine release in parts of the caudate and putamen distant to STN [141]. The information gained from these studies is both essential for understanding the pathophysiology of Parkinson’s disease as well as understanding the brain-wide mechanisms through which DBS can influence multiple brain systems. Obtaining this level of resolution for the DBS targets for depression and OCD is a key aim for future studies in animal models.

Essential to understanding how DBS impacts the brain will be determining how stimulation changes both functional networks using functional MRI as well as the structure of the brain using diffusion imaging. Combining diffusion weighed imaging with DBS or indeed other neuromodulation approaches has the potential to reveal whether the white matter structure of the brain is potentially altered by stimulation. Work in humans indicates that white matter can be plastically remodeled by training [143] as well as non-invasive neuromodulation [84] but no data are available as yet in patients with implanted DBS leads. Work in rodent models indicates that stimulation induces oligodendrogenesis and myelination in the stimulated pathway [83]. Determining whether and how DBS applied to the same parts of the brain targeted to treat depression and OCD alters white matter structure from voxels to individual oligodendrocytes in nonhuman primates has the potential to provide insights that would be highly challenging or impossible to acquire in humans. To that end, a recent study in macaques indicates that DBS targeted to the confluence of the forceps minor, cingulum bundle, and uncinate fascicle is associated with increased diffusion imaging estimates of white matter integrity as well as increased myelination in the cingulum bundle [144]. Further studies that assess the effects of DBS, and other non-invasive neuromodulation approaches, from the level of diffusion neuroimaging to cellular changes will be critical for understanding how neuromodulation impacts the structural organization of the brain.

Using neuroimaging to reveal the time course and impact of non-invasive neuromodulation

Due to their reduced spatial precision, noninvasive techniques are not often used in basic research with nonhuman primates. Despite this, a great deal can be learned about the basis of noninvasive neuromodulation, as it possible to obtain direct brain read-outs of the effects of neuromodulation at scales not possible in humans. Indeed, a key advantage to using nonhuman primates to investigate the mechanisms underlying neuromodulation is that it is possible to directly measure single neuron activity in virtually any part of the brain (Fig. 1).

TMS has been used extensively as a noninvasive treatment option for both depression and OCD. Over the last two decades, a number of studies in macaques have used a variety of neuroimaging techniques to get a better understanding of how TMS influences brain-wide networks [145–147]. As early as 2004, Ohnishi and colleagues used a combination of TMS and PET imaging to evaluate how stimulation of the motor cortex impacted extracellular dopamine concentrations throughout the brain [146]. They found that TMS induced an increase in dopamine in the ventral striatum, indicating that TMS delivered to distant cortical locations can have unintended effects on dopaminergic reward systems. More recently, Kadono and colleagues used a combination of functional neuroimaging and diffusion weighted imaging in macaques to investigate how repetitive TMS delivered to the motor cortex can help with recovery from strokes [145]. Following lesions to the posterolateral nucleus of the thalamus, they found that there was an increase in fMRI functional connectivity between the ipsilateral mediodorsal nucleus of the thalamus and amygdala. When TMS was applied to the motor cortex, this pattern of heightened functional connectivity was normalized back to baseline levels. What is needed now is studies to reveal the mechanisms of this effect at the level of single neurons or oscillatory activity.

Due to its ability to modulate neural activity deep in the brain FUS has a distinct advantage over other non-invasive approaches such as TMS. Unlike TMS where the effects of stimulation on neural activity are increasingly well understood [148, 149], much less is known about the mechanisms of action of FUS that is thought to alter brain activity through multiple different physical mechanisms [150, 151]. In part because of this, it was unclear how FUS neuromodulation targeted to one part of the brain might alter neural activity across distributed networks as well as how long neuromodulation would last for. In a series of studies, Verhagen, Sallet, Rushworth, and colleagues showed that brief delivery of FUS directed to both superficial and deep targets in the brains of macaques altered neural activity for over an hour [152]. Further, the functional connectivity of the targeted region with other areas of the brain was reduced following stimulation [153]. The effect of FUS on functional connectivity is potentially important for deciding when to use FUS for neuromodulation for psychiatric disorders as it reveals that FUS, at least with the parameters used in this study, decreases functional connectivity. Indeed studies by other groups in nonhuman primates using different parameters have shown that FUS delivered to deep brain targets can be associated with both increases and decreases in functional connectivity [154]. As such, this highlights that careful testing and selection of the parameters for FUS need to be considered when planning FUS neuromodulation of a specific circuit to treat psychiatric disorders.

Future research directions and conclusions

Neuromodulation for psychiatric disorders is an emerging option for individuals whose symptoms are not adequately managed by front-line pharmacological or cognitive-behavioral therapies. While refinements and progress continue to improve the efficacy of these therapies, they are not effective in all individuals, emphasizing the need for further research into their mechanisms and how targeting can be further refined. As we have highlighted here, studies into the basis of neuromodulation therapies have provided key insights into the mechanisms of action, but much remains to be determined. Indeed, in the discussion above we highlighted several key areas that require further investigation. First amongst these are neuroimaging and neurophysiology studies to establish how invasive neuromodulation approaches for psychiatric disorders such as subcallosal ACC or ALIC DBS alter brain-wide functional networks and potentially even brain structure. Another important avenue for research will be to determine how experimentally induced dysfunction within one area impacts brain wide circuits and the neurophysiological signature of such dysfunction. As we highlighted, pre-clinical studies in nonhuman primates have started to reveal the signature of localized dysfunction within specific brain areas [106, 130]. Systematically characterizing the fMRI functional connectivity fingerprint of localized dysfunction across many different areas is essential if we are to relate the patterns of fMRI functional connectivity in humans to specific circuit-level dysfunction that could then be used as a biomarker for targeted neuromodulation.

In addition to this, here we primarily concentrated on work that has only combined a single therapeutic approach with neuroimaging. Such an approach has begun to provide insights into the basis of neuromodulation therapies, but an important avenue for future research in both humans and animal models will be to discern the effects of combining multiple neuromodulation therapies at once. It may well be that the effects of one neuromodulation therapy when combined with another, pharmacotherapy, or even computational psychiatry modeling of behavior can be used to accelerate or augment the clinical response [155]. Here studies in animal models could be particularly important for determining how to target each approach as well as the timing of the different therapies relative to one another to reduce possible adverse events. With such knowledge it will be possible to translate what has been discovered to the clinic for the betterment of psychiatric care.

Author contributions

All authors wrote and edited the manuscript.

Funding

SF, KSC, BR, HM, and PR are supported by the grant from Hope for Depression Research Foundation and the grant from The National Institute of Mental Health (NIMH) (R01MH132789). SF, AF, CE, BR, and PR are supported by grants from NIMH and the BRAIN initiative (R01MH110822 and RF1MH117040). BR is supported by grants from NIMH (R01MH111439) and NINDS (R01NS109498). AF is supported by Overseas Research Fellowship from Takeda Science Foundation and a Brain & Behavior Research Foundation Young Investigator grant (#28979).

Competing interests

HM and KSC receive consulting fees from Abbott Neuromodulation. Other authors declare no competing financial interest.