Connectivity, Complexity, and Excitatory vs. Inhibitory Control.  For full article, click here.

Connectivity, Complexity, and Excitatory vs. Inhibitory Control.  For full article, click here.

Applying Control Systems Engineering Approaches to Human Neuroimaging Data.

Our general aim has been to develop a new approach to brain-based functioning and disease that is based upon identifying key points of failure in circuit regulation which, depending upon how it breaks, can lead to a wide variety of signs and symptoms that cluster as different diagnoses. Our specific approach has focused upon developing new ways of exploiting neuroimaging, by using it to identify subtle features of dysregulation that, over time, cause the brain to develop in different ways. As a first step in developing these methods, we focused our interest on understanding how the neural circuit that regulates emotion in the brain might correspond with variation across a spectrum of stress-resilience.  This led to a set of studies that included four cohorts:  those with pathological (clinical) anxiety, variation in trait anxiety, variation in endocrine/cardiovascular responses to an extreme stressor (first-time tandem-skydive), and U.S. Navy SEALs.  We pursued this research for two reasons.  First, we wanted to better understand regulation of the prefrontal-limbic system and how it might cleanly differentiate across a continuum of phenotypes for one well-defined symptom: fear.  Second, “stress,” when distilled of its emotional and cognitive baggage, is essentially a perturbation, affecting a large interconnected network of physiological control systems designed to maintain homeostasis. Research in neuroendocrinology and immunology has shown that emotional stress acts systemically throughout the body, lowering thresholds for illness, from schizophrenia to autoimmune disease to cancer[20].  Thus, by studying stress we might access hierarchical (nested) control circuits:  not only probing negative feedback loops in the brain, but also probing feed-forward and feedback interactions between the brain and other physiological control systems (e.g., HPA-axis, autonomic, immune). 

Emotional arousal in the brain is regulated by the prefrontal-limbic meso-circuit, a negative feedback loop with multiple excitatory and inhibitory components. Like all physiological control circuits, ideally it must be supple enough to respond to the environment, while having sufficient inhibitory feedback to return the response to baseline.  Dysregulation of a control circuit leads to a system’s inability to respond and/or to suppress response, a dynamic that can be found in many diseases, such as diabetes, heart disease, Graves, and Cushing’s.  Homologies between different physiological control systems suggested approaches for measuring neural control circuits.  The most diagnostically sensitive measures for risk of heart attack and cardiovascular disease have been entropic or “complexity” measures of heart rate variability (HRV).  In physics, the balance between order and chaos has been identified with critical states at which phase transitions most easily occur.  From a biological perspective, these states minimize the energy required both to excite as well as to inhibit a response.  By this logic, an insufficiently constrained control circuit should show outputs with greater complexity relative to that critical state, and vice-versa.

The application of systems-based analyses to neuroimaging led us in several new directions.  First, we optimized complexity analyses for neuroimaging, with resulting software development.  Functional MRI time-series are both shorter and sparser than those acquired with HRV; thus, many numerical requirements for HRV complexity measures are violated by neuroimaging data.  Moreover, while amplitude or connectivity-based analyses can rely upon averaging across trials or subjects to increase signal/noise, our use of complexity (analyzed as one long-time-series, rather than averaged across conditions) and diagnostic classification (based upon a single subject) meant that we needed to be able to analyze data without averaging.  We thus embarked upon optimization of neuroimaging-specific acquisition parameters, experimental design, and signal processing procedures for entropic measurements and their resulting application for neurodiagnostic classification with support vector machine.  The end result was development of two software packages:  the NIRS Analysis Package (NAP) and NeuroClass. 

Unlike HRV time-series acquired from only one node (the heart-beat), fMRI permits acquisition of time-series at all points in the circuit, both excitatory and inhibitory.  Our modeling and simulations demonstrated that control networks of sufficient size reliably produce power-law behavior, and that the distribution of frequencies at the critical state corresponds to a balance between excitatory and inhibitory components.  We additionally showed that deviations from this state vary as a function of input control and input density, providing a framework for interpreting results from multi-node circuits acquired with fMRI .

In doing so, our results from our subjects along the stress vulnerability-resilience continuum converged to a coherent symmetry, which disrupted some key assumptions about roles played within the circuit. Looking at a spectrum of threat assessment that included 137 individuals, across three independent studies (N=57 for our study of clinical anxiety, N=50 for our study of trait anxiety, and N=30 for our study of ‘reckless’ risk-takers), we observe a clear inverted-U pattern for Brodmann’s Area 45, a subset of the inferior frontal gyrus (IFG). Individuals at the center of the threat assessment spectrum (showing accurate perception of threat:  physiologically in anticipation of jumping out of a plane, behaviorally in classifying faces with ambiguous affect, as well as emotionally with self-reported low trait anxiety in non-dangerous contexts) showed IFG regulation in the critical (‘pink noise’) range, which our simulations show to occur when a control system includes optimal feedback, and which our dynamic causal modeling results linked with strong bi-directional connectivity to the ventromedial prefrontal cortex (vmPFC).  However, individuals at both ends of the spectrum (exceptionally anxious and exceptionally reckless) showed circuit-wide dysregulation localized most strongly to the IFG, with chaotic ‘white-noise’ dynamics.  Individuals at each end of the spectrum would appear to be opposites of one another (our clinically anxious sample identified threat where it did not exist, whereas our reckless sample failed to identify threat where it did exist), yet the most prominent feature that they both had in common was a failure to accurately assess ambiguous threat.  The fact that both ends of the spectrum exhibit a disconnect between the IFG and the rest of the prefrontal-limbic circuit suggests that the IFG’s role in fear inhibition is indirect, potentially by providing disambiguation of ill-defined stimuli, which informs the (inhibitory) vmPFC.  Our interpretation was consistent with our dynamic causal modeling results, which provided evidence for a fully connected (closed-circuit) model, with the altered IFG dynamics found in anxious patients most strongly reflecting interactions between the IFG and vmPFC—but not between the IFG and amygdala, or between the vmPFC and amygdala .  

Stochastic dynamic causal modeling (DCM) suggests the corticolimbic threat circuit consisting of the inferior frontal gyrus (IFG), ventromedial prefrontal cortex (PFC), and amygdala during fear generalization. (a) Model space. (b) In Bayesian model selection, Model 1 (a fully connected model) showed the greatest expected probability and exceedance probability. (c) Individual variability in bidirectional connectivity between the IFG and vmPFC significantly correlates with one in IFG complexity (power spectrum scale invariance) β-signatures.   For full article, click here.

Stochastic dynamic causal modeling (DCM) suggests the corticolimbic threat circuit consisting of the inferior frontal gyrus (IFG), ventromedial prefrontal cortex (PFC), and amygdala during fear generalization. (a) Model space. (b) In Bayesian model selection, Model 1 (a fully connected model) showed the greatest expected probability and exceedance probability. (c) Individual variability in bidirectional connectivity between the IFG and vmPFC significantly correlates with one in IFG complexity (power spectrum scale invariance) β-signatures.   For full article, click here.