A unifying computational framework for stability and flexibility of arousal

Abstract

Arousal and consciousness flexibly adjust to salient cues, but remain stable despite noise and disturbance. Diverse, highly interconnected neural networks govern the underlying transitions of behavioral state; these networks are robust but very complex. Frameworks from systems engineering provide powerful tools for understanding functional logic behind component complexity. From a general systems viewpoint, a minimum of three communicating control modules may enable flexibility and stability to coexist. Comparators would subtract current arousal from desired arousal, producing an error signal. Regulators would compute control signals from this error. Generators would convert control signals into arousal, which is fed back to comparators, to make the system noise-proof through self-correction. Can specific neurons correspond to these control elements? To explore this, here we consider the brain-wide orexin/hypocretin network, which is experimentally established to be vital for flexible and stable arousal. We discuss whether orexin neurons may act as comparators, and their target neurons as regulators and generators. Experiments are proposed for testing such predictions, based on computational simulations showing that comparators, regulators, and generators have distinct temporal signatures of activity. If some regulators integrate orexin-communicated errors, robust arousal control may be achieved via integral feedback (a basic engineering strategy for tracking a set-point despite noise). An integral feedback view also suggests functional roles for specific molecular aspects, such as differing life-spans of orexin peptides. The proposed framework offers a unifying logic for molecular, cellular, and network details of arousal systems, and provides insight into behavioral state transitions, complex behavior, and bases for disease.

Keywords: arousal; computation; control; histamine; hypocretin; hypothalamus; narcolepsy; orexin.

Figure 1

Functional connectivity for making a stable-yet-flexible signal. (A) A general mechanism (feedback loop) for tracking a set-point despite disturbance (after Csete and Doyle, 2002). For simplicity, load/noise is modeled as entering at a final summing point, but it can enter the system at any point. (B) An equivalent representation of a neural circuit for producing appropriate levels of arousal-enhancing neurotransmitters. (C) System performance with and without integral control. Computational simulations of temporal dynamics of the system shown in (A), in the face of changes in load/noise or changes in desired set-point. Left, effect of escalating disturbance on system output with (green) and without (magenta) integration in the module R (k = 1 in both cases). Right, response of system output to changes in desired set-point for the two types of control. Integration here had a time window longer than the simulation time; effects of varying the integration window are described in the text and Figure 2A. See Supplementary Figure for more information on computational simulations.

Figure 2

Integrators and their hypothetical neural correlates. (A) Theoretical input-output dynamics of neurons that act as multipliers (top left graph), or integrators (top right graph) with different integration windows (bottom graph). W here is defined as the duration of integration window, i.e., given ${\displaystyle {\int }_b^{a}inputdt,w=a-b}$. Input is shown in magenta, and output in green. (B) Experimental input-output relations of neurons in brain orexin circuits. Top row, firing outputs of orexin/hypocretin neurons (OHNs ) in response to bath application of 1 µM (left) and 200 nM (right) orexin/hypocretin peptides (from Li and van den Pol, , reproduced with permission from The Society for Neuroscience). Bottom row, firing outputs of histamine neurons (HAN) in response to optogenetic stimulation of OHNs at 20 Hz (blue bar). Responses due to orexin transmission are in red (data from Schöne et al., 2014).

Figure 3

Diagnosing control roles of neurons from temporal activity pattern. (A) Firing of an orexin neuron in the hypothalamus of an awake rat experiencing sensory inputs (reproduced with permission from Cell Press from Mileykovskiy et al., 2005). (B) Top drawing, hypothetical mapping of control elements to neuronal types in brain orexin circuits. To protect the output (arousal drive) from instability, R could correspond to OXR2-expressing cells (e.g., histamine neurons), and e could come from orexin neurons driven by positive inputs (e.g., sounds, Mileykovskiy et al., 2005) and negative feedback inputs (e.g., serotonin and/or noradrenaline, Li et al., ; Yamanaka et al., 2003b). G could correspond to noradrenaline cells, that express OXR1 and are expected to be excited by HANs (Haas et al., 2008). The intermittent firing of orexin cells during wakefulness in vivo (Mileykovskiy et al., 2005) is broadly consistent with this placement in the feedback loop. Bottom plots, computational predictions of temporal variations in activity of the different neurons to changes in set-point or in load/noise (R is modeled as an integrator with a time window that exceeded simulation duration; C and G are modeled as algebraic summing points of the inputs they receive, DiStefano et al., 2012). See Supplementary Figure for more information on computational simulations.