Authors: Elizabeth M Sachse (th; 2University of Minnesota, Department of Neuroscience, 6-145 Jackson Hall, 321 Church Street SE, Minneapolis, MN 55455), Alik S Widge (th)
Categories: Article, Deep Brain Stimulation (DBS), Cognitive Flexibility, Cortico-striato-thalamo-cortical (CSTC) circuits, Decision-making, Non-invasive Neurostimulation
Source: Current opinion in behavioral sciences
Authors: Elizabeth M Sachse, Alik S Widge
Cognitive flexibility, the capacity to adapt behaviors in response to changing environments, is impaired across mental illnesses, including depression, anxiety, addiction, and obsessive-compulsive disorder. Cortico-striatal-cortical circuits are integral to cognition and goal-directed behavior and disruptions in these circuits are linked to cognitive inflexibility in mental illnesses. We review evidence that neurostimulation of these circuits can improve cognitive flexibility and ameliorate symptoms, and that this may be a mechanism of action of current clinical therapies. Further, we discuss how animal models can offer insights into the mechanisms underlying cognitive flexibility and effects of neurostimulation. We review research from animal studies that may, if translated, yield better approaches to modulating flexibility. Future research should focus on refining definitions of cognitive flexibility, improving detection of impaired flexibility, and developing new methods for optimizing neurostimulation parameters. This could enhance neurostimulation therapies through more personalized treatments that leverage cognitive flexibility to improve patient outcomes.
Cognitive flexibility, the ability to adjust behaviors in response to a changing environment, is impaired across the spectrum of mental illnesses [1–3]. For instance, depression, anxiety, addiction, and obsessive compulsive disorder (OCD) all involve “stuck” behavioral drives that patients want to ignore, but feel unable to [4–6]. There is emerging evidence that cognitive flexibility deficits drive disease symptoms, e.g. recent work where improvements in cognition appeared to mediate improvements in depressive symptoms [7]. Thus, improving cognitive flexibility may be a promising approach for treating a range of mental illnesses [8,9].
Importantly, for simplicity in this review we use “cognitive flexibility” as an umbrella term that can describe both “cognitive” and “behavioral” flexibility (commonly used in preclinical neuroscience research) as well as “cognitive control” (commonly used in clinical research). There are differences between these constructs, but they heavily overlap and all are believed to leverage the same neural systems (see below) [10]. Similarly, there are many ways to measure cognitive flexibility, but nearly all of them involve some kind of behavioral metric making it challenging to isolate a pure cognitive component [6]. Flexibility and control inherently require and interact with each other and their definitions widely vary and often overlap [10,11]. Many definitions of cognitive control draw on early work from Miller and Cohen, who defined cognitive control as “the ability to orchestrate thought and action in accordance with internal goals”, primarily driven by the prefrontal cortex [12]. This definition has evolved over time and depends on the research context in which it is being described. For example, cognitive control has been defined as “a class of mechanisms” [10,13], “a set of mental processes” [14,15], and as a synonym for executive functions/control [16–18]. Some works study control without giving a formal definition [19,20]. Likewise, cognitive flexibility definitions vary, with some equating it with cognitive control [20] and/or behavioral flexibility [5,21,22], while others define them as separate concepts and explicitly discuss the nuances [6,11,23].
In these newer bodies of work, flexibility is considered a “domain” or “component” of cognitive control and/or executive function. This classification is captured by work from the Cognitive Atlas project, which has sought to fix this problem of ambiguous terminology for cognitive phenomena [24]. The project was partially motivated by a large literature review that found the term “cognitive control” was not uniquely associated with any behavioral tasks or tests, and instead always co-occurred with at least one other cognitive working memory, response inhibition, response selection, or task/set-switching [14]. Similarly, Niendam et al., 2012, framed cognitive control as a network in the brain that supports four traditional domains of executive initiation, inhibition, flexibility, and working memory [19]. This classification of flexibility under the cognitive control umbrella is the most prevalent in recent literature, but that does not imply consensus. For instance, while the descriptions of executive functions usually reference specific behavioral measures, cognitive control also leverages engineering control theory frameworks, and thus includes concepts of internal brain states that do not yet map to a known neural substrate [10].
Common themes in the control/flexibility field are also at odds with how control is framed in a clinical framework, the NIMH Research Domain Criteria (RDoC) [25]. In RDoC, none of the executive functions cited above are considered subconstructs of cognitive control. In fact, working memory is its own separate construct, and other aforementioned executive functions could be classified under several different cognitive systems. This highlights the problem of strictly classifying flexibility as a subcomponent of cognitive control. While cognitive control requires flexibility, in terms of the ability to switch goals/responses, flexibility itself requires other domains of control, like response inhibition and goal updating. Therefore, while cognitive control and flexibility are not equivalent, they are highly interactive, and current definitions are not generalizable to all research contexts. One solution is to narrow definitions of “cognitive flexibility” to specific task/behavioral measures. For example, some researchers define flexibility explicitly as task/set-shifting associated with certain behavioral paradigms, but this may be too restrictive [6,26].
Cognitive flexibility and related constructs such as attention and action selection are strongly linked to cortico-striatal-thalamo-cortical (CSTC) circuits [10]. CSTC circuits broadly work to transfer information from cortical areas to the basal ganglia and cerebellum, i.e. to link motor plans to broader goals [27]. Critically, the anatomical terms used to describe cortical components of the CSTC are inconsistent between humans, rodents, and non-human primates. For excellent reviews on this topic, see [28,29]. Importantly, CSTC circuitry in both humans and animals contains sub-circuits with distinct, yet related functions. For example, the medial prefrontal cortex (mPFC) in rodents, analogous to the cingulate cortex in primates, is necessary for adaptive control of action selection [30,31], whereas action value maintenance and updating load more onto the orbitofrontal cortex (OFC) [32]. Lesions of these structures in rats thus impair different forms of flexibility, e.g. set shifting vs. reversal learning [33]. Anatomically, many cortical projections converge in the striatum, making it a “hub” for information integration for goal-directed behavior [34–36]. Further, these cortical inputs are topographically organized, creating specialized subregions of the striatum [37]. Generally, for rodents and primates, the dorsomedial striatum (DMS, caudate nucleus in primates) is thought to mediate behavior adaptation to environmental changes while the dorsolateral striatum (DLS, putamen in primates) helps execute habitual actions when the environment is stable. The ventral striatum is believed to evaluate outcomes and modulate motivation, reward, and learning pathways [38–40].
Cognitive flexibility is also linked to specific physiologic signatures in CSTC circuits. Local field potential oscillations can be measured via EEG or implanted electrodes in humans and animals and provide neural signatures of flexibility and other executive functions. For instance, theta (5–8 Hz) oscillations in the frontal cortex track action monitoring over short and long time scales [41] and this theta activity is thought to be required for cognitive flexibility [42,43]. Theta oscillations in particular CSTC regions are associated with specific components of flexibility. For example, OFC theta power tracks reversal learning [44]. Theta [15] and other brain activity measures [16] are disrupted in mental illnesses, leading to impaired flexibility [45]. Therefore, CSTC networks and their oscillations are primary targets for neurostimulation interventions that seek to treat this impaired flexibility, and the symptoms associated with it.
Mental disorders are increasingly being treated by delivering targeted and precisely “dosed” electro-magnetic energy to CSTC circuit components [46]. For instance, deep brain stimulation (DBS) of ventral striatum and internal capsule (VCVS) can treat OCD [47–49], MDD [47,50], addiction [51], and anorexia [52] with varying degrees of clinical efficacy. OCD is currently the only condition for which DBS is FDA-approved. Closely linked structures, such as the subthalamic nucleus and the amygdala, are also effective stimulation targets [49,53,54]. Non-invasive therapies such as transcranial magnetic stimulation (TMS) also modulate CSTC elements. TMS to the dorsolateral prefrontal cortex (dlPFC) has long been used to treat medication-resistant depression [55] and also shows potential in treating OCD [56,57] and patients with comorbid disorders, such as PTSD and MDD [58]. Like TMS, transcranial direct/alternating current stimulation (tDCS or tACS) of the PFC can improve symptoms of MDD [59,60], OCD [61] and other illnesses [62–64].
Because all of the above therapies act on CSTC circuits, and treat multiple disorders marked by inflexibility, they might collectively act by increasing flexibility. We have specifically shown this for VCVS DBS [47]. OCD and MDD patients performed a task that engages cognitive flexibility with their VCVS DBS turned on or off. This Multi Source Interference Task (MSIT) was specifically designed to engage cognitive flexibility by creating response interference on certain trials. It produces strong activation in CSTC circuits [65]. With their DBS turned on, patients were able to respond faster on individual trials without making more errors, i.e. they displayed enhanced flexibility. Further, VCVS DBS strengthened the theta oscillations highlighted above as a known signature of cognitive flexibility. The same cognitive flexibility enhancement and increased theta power were replicated in a second study [64] and independently by another group using a different task [66]. Our replication study showed that particular electrode locations in the VCVS yielded larger flexibility improvements, indicating a potential “sweet spot” of CSTC circuits to target for neurostimulation. It also demonstrated a method for measuring flexibility improvements in real time, which we have since shown could be used to optimize stimulation to ensure improvement for individual patients [67]. That is, because cognition and cognitive flexibility can be measured at much higher density and with greater fidelity than vague constructs such as “mood” or “distress”, they could be used as specific biomarkers to verify that a therapy was “dosed” correctly in each individual [3,67].
Non-invasive neurostimulation can also modulate behavioral and physiological measures of cognitive flexibility [68–74], evidence that stimulating cortical areas engages the broader CSTC circuitry. TMS of the PFC alters executive function, attention and perception, cognitive flexibility, and more in both healthy and patient populations (see [23] for recent review). For example, TMS of the dlPFC reduced symptoms of depression and improved cognitive flexibility as measured via reaction times and accuracy on the Stroop Task [74]. Although cognitive flexibility measures were not used as a metric of effective symptom treatment in this specific study, it suggests TMS’s therapeutic effects may be driven by improvements in flexibility. Conversely, patients with depression marked by cognitive rigidity responded poorly to standard medications, suggesting that they may disproportionately benefit from TMS’ flexibility effects [9]. In support of this, in one recent study, patients with flexibility deficits were more likely to benefit clinically from TMS, and that clinical benefit appeared to be mediated by a change in a putative fMRI biomarker of flexibility [7]. Similarly, tDCS of the dlPFC improves task-related measures of cognitive flexibility in healthy individuals [70] and patients with ADHD [63]. Other studies that measure neural activity during non-invasive neurostimulation observe similar changes [59,72,75–77]. While not conclusive, these works support the idea that neurostimulation of CSTC circuitry improves cognitive flexibility. And, although it is not yet standard to measure cognitive flexibility as a clinical outcome, these studies collectively suggest that doing so would be a useful improvement.
The next challenge is clarifying the neural and behavioral mechanisms of enhanced cognitive flexibility. Mechanistic understanding could help translate these early scientific results into therapies, by helping select patients who are likely to benefit [9,78] and by ensuring that stimulation could be tuned to produce the maximum cognitive benefit in each patient [3,67,79]. Progress in identifying those mechanisms has been limited, in part due to the challenge of doing circuit-level physiologic research in humans, especially clinical populations [80]. Animal models may offer a way forward. Due to the strong topographic homology in rodents and primates, there is high translatability of anatomical and functional roles of CSTC pathways [29,34,37]. Rodent models are especially useful with their wide-variety of cell-type-specific techniques, provided researchers address important differences in anatomy. For instance, VCVS DBS in humans targets the internal capsule, a compact bundle of axon fibers. Rodents do not have a true internal capsule. Instead, PFC projections are topographically arranged in diffuse axon fascicles throughout the striatum. However, the topographic map of PFC projections is conserved between species [34,37], meaning that electrical stimulation of rodent VCVS analogs should model similar stimulation in humans.
Further, behavioral tasks similar to those used to assess cognitive flexibility and decision-making in humans are well-established in non-human primate and rodent models [6]. Similar to the development of human neurostimulation therapies, many animal studies have investigated how pharmacological agents and/or lesions of CSTC circuits affect cognitive flexibility [33,81–84]). These studies confirmed that structures analogous to human CSTC areas, like the mPFC, also share the same functions.
With this knowledge, multiple studies have electrically stimulated CSTC circuits in rodents to understand the mechanisms of human neurostimulation [51,85–87]. For example, VCVS DBS may improve OCD symptoms by augmenting patients’ ability to engage in extinction learning (another form of flexibility) [88]. Consistent with this hypothesis, rats show improved extinction learning and recall when receiving VS DBS during an aversive conditioning/extinction paradigm [89]. We showed that there is strong translational homology for cognitive flexibility between humans and rodents. We tested rats with a cognitive flexibility task (set shifting [90]) while stimulating various striatal areas analogous to the different regions tested in [91]. Stimulation of mid-striatal circuits reduced reaction times without increased errors, replicating the cognitive flexibility enhancement observed in humans receiving VCVS DBS. With this animal model, we used other behavioral measures and computational modeling to show that stimulation improves proactive control and decisional efficiency (defined in detail in [90]), in both rats and in humans. This work highlights that cognitive flexibility is not just a mechanism – it can serve as a platform on which to create translationally valid animal assays. Animals cannot directly model psychiatric symptoms that depend on unique human experiences [92,93], but they can model behavioral/decisional phenomena, and thus be used to optimize treatments that engage those phenomena [10,17].
Animals, particularly rodents, also offer techniques like optogenetics that enable cell type and projection-specific modulation. Recent work has used those techniques to show that specific projections are important to DBS’ effects across targets, and that a relatively simple stimulation pattern can have complex, mixed excitatory-inhibitory effects [94,95]. Given the findings above that different types of flexibility may require different CSTC sub-circuits, circuit- and cell-type-specific tools could be valuable for identifying the best pathways to modulate in humans for a given set of symptoms. Animal models, both rodent and primate, might also be valuable in developing a next generation of closed loop, physiology-driven therapies [3]. Multiple papers have demonstrated decoding of cognitive flexibility and related constructs directly from physiologic signals [91,96]. A next generation of psychiatric neurostimulation might directly target and change those signals, just as neurostimulation for movement disorders is moving from empirical, clinician-driven titration to objective, marker-driven treatment [97].
Neurostimulation to improve cognitive flexibility is a promising therapy for mental illnesses, but at least three main challenges remain. First, we need a more robust definition of “cognitive flexibility” and a means of detecting patients with impaired flexibility. Much of the work cited above used cognitive tasks as measurements of flexibility. These may be useful for measuring within-individual change (e.g., pre/post neuromodulation), but tasks are particularly poor at measuring between-individual differences [98,99]. Newer approaches based on validated neuroimaging, and specifically engineered for test-retest reliability, may be more suitable [9,78].
Second, it is not yet clear whether cognitive flexibility is the causal element – whether flexibility deficits lead to maladaptive behaviors that produce persistent clinical symptoms, or whether instead the presence of persistent symptoms leads to more inflexible styles due to stress. If the latter is true, therapies that improve flexibility may not reliably improve symptoms. There is circumstantial evidence for flexibility as causal. As noted above, there is one pilot study suggesting that a flexibility-improving intervention helps patients who are known to have a deficit [7]. In prior work, we showed that some patients report improved subjective well-being when their flexibility improves [91]. Conversely, in disorders characterized by limited flexibility, such as OCD, rigid behaviors emerge before patients report higher-level symptoms such as anxiety [2]. More generally, improving cognitive functions (like cognitive and behavioral flexibility) is a core element of cognitive-behavioral therapies, which are evidence-based treatments for almost every mental disorder [5,100]. In addition, medications commonly prescribed to treat mental illnesses can alter cognitive functions like flexibility [101]. Therefore, it may not matter whether a given patient’s symptoms are directly caused by a flexibility deficit – if one is present, then improving flexibility should still give patients a greater ability to manage those symptoms.
Third, as noted above, we need to identify the underlying mechanisms of invasive and non-invasive neurostimulation to refine current treatments. Psychiatric DBS parameters were originally developed based on DBS for Parkinson’s disease, and TMS parameters were selected based on effects in motor cortex that may not generalize to PFC. The ideal stimulation parameters (frequency, intensity, waveform) may vary by patient, brain target, and disease state [102]. Animal models are critical to understand mechanisms and optimize these parameters to maximize neural and behavioral effects, as they have in movement disorders [95,103,104]. Given the recent development of translationally valid animal models of psychiatric neurostimulation [90], the outlook is bright.