Behavioral State—Getting “In The Zone”

Abstract

In this issue of Neuron, McGinley et al. (2015) investigate a classic observation from psychology linking arousal state with behavioral performance, demonstrating neural correlates of an “optimal” state for an auditory detection task.

Sensory processing is clearly affected by mental state. Everyone has had the experience of sitting in a lecture and spacing out—even though you are awake and sound waves have been hitting your eardrum, you have no idea of what was said in the past five minutes. But what is happening to those signals in your brain?

Numerous studies have examined the differences in ongoing neural activity and the interaction with incoming sensory input across different behavioral states from sleep versus wake to selective spatial attention. Recently, these studies have extended to the rodent, where motor states such as active whisking (Poulet and Petersen, 2008) and locomotion (Niell and Stryker, 2010; Schneider et al., 2014) have been shown to correlate with changes in sensory representations, even in the primary sensory cortices. Despite the variety of these states, they can all be seen to vary along the dimension of arousal.

Psychologists have encapsulated the effects of arousal on behavioral performance in what is known as the Yerkes-Dodson curve (Yerkes and Dodson, 1908), observing that performance on difficult tasks generally peaks at an optimal intermediate level of arousal (Figure 1). If your arousal level is too low, you are spaced out or disengaged, and if your arousal level is too high, then you are anxious and distracted—in both cases performance suffers. In this issue of Neuron, McGinley et al. (2015) deliver a tour-de-force exploration of the Yerkes-Dodson relationship in the auditory system of mice, finding it manifested across multiple levels from the resting membrane potential of individual neurons to sensory-evoked synaptic inputs to behavioral performance.

Figure 1. The Yerkes-Dodson Relationship and Its Neural Correlates, as Observed by McGinley et al.

Illustration credit: D. Piscopo.

The authors use a deceptively simple measure of behavioral state: pupil diameter. In addition to varying with ambient light level, it is also well established that pupil diameter in humans correlates with various aspects of arousal, from task difficulty to affective state; it is even possible to track the number of items a person is holding in memory based on their pupil diameter (Laeng et al., 2012). Two recent studies in the mouse visual system have used pupil diameter as a metric for arousal state (Reimer et al., 2014; Vinck et al., 2015), and McGinley et al. (2015) begin their study by confirming that pupil diameter agrees with an internal neural measure of arousal, the rate of hippocampal ripple waves. Notably, pupil diameter provides a continuous variable measuring arousal, rather than simply a discrete categorization into states. However, an important caveat is that a number of factors can contribute to arousal as measured by pupil diameter, from mental effort to anxiety to physical activity, so the precise nature of the arousal is not determined.

Using pupil diameter as their metric, the authors first examine neural correlates of arousal using whole-cell recordings of spontaneous activity in auditory cortex of awake head-fixed mice on a running wheel. Remarkably, they find that arousal, as measured by pupil diameter, predicts variations in membrane potential. In fact, on the timescale of seconds there is over 50% coherence between pupil diameter and membrane potential. In other words, pupil diameter can predict more than half the variance in membrane potential of neurons in auditory cortex.

The authors demonstrate that, just as the Yerkes-Dodson curve predicts, the relationship of membrane potential with arousal is U-shaped, as is the variability of membrane potential. At low levels of arousal, slow-wave oscillations begin, which increase variability and raise the membrane potential. At high levels of arousal, there is a tonic depolarization of membrane potential and high-frequency oscillations, which is often referred to as the desynchronized state. In between, at what might be expected to be the optimal point on the Yerkes-Dodson curve, the membrane potential sits relatively low and quiet. This is also reflected in spontaneous firing rate, which is lowest at intermediate levels of arousal.

Next, the authors studied sensory responses in the context of an auditory detection task, where pure tones must be detected against the background of a complex spectrotemporal auditory stimulus. Again, sensory-evoked responses, both at the level of synaptic potentials and multi-unit firing, show a U-shaped curve, now inverted with maximal evoked responses at the mid-point. Thus, a decrease in background variance and increase in evoked response work together to maximize the signal-to-noise ratio at intermediate levels of arousal.

To tie this all back to the initial Yerkes-Dodson finding, the authors find that performance of the detection task follows the U-shaped prediction. When animals are at low arousal, they often miss the stimulus, and at high arousal they often false alarm. Furthermore, the peak of the performance curve was at a similar level of arousal (pupil diameter) as the peak for neural encoding of the sensory input. Together these experiments provide a connection between baseline membrane potential, sensory-evoked responses, and task performance across three regimes of arousal (Figure 1).

An important aspect of their approach is that the mice, rather than being restricted to periods of high performance, were allowed to drift between behavioral states, which enabled the authors to map out the continuum of arousal. This provides a clear demonstration that, just because an animal is awake and performing, it is not in a specific well-defined state. In fact, just as one would measure depth of anesthesia in an anesthetized experiment, if one wants to compare across experiments it is necessary to either control (as best as possible) or measure (as best as possible) an animal’s behavioral state.

This study also provides insight into another recently used measure of behavioral state: locomotion. Like other studies, they find that locomotion is accompanied (and in fact preceded) by pupil dilation, suggesting it is often a consequence of arousal. Indeed, in the experiments here the effect of locomotion was not greatly different than high arousal alone, except for a few measures such as false alarm rate and evoked firing in MGN. However, in this study locomotion was always associated with arousal. Other recent studies in the visual system (Reimer et al., 2014; Vinck et al., 2015) have been able to segregate arousal from locomotion and found that while arousal accounts for many effects correlated with locomotion, there are distinct contributions of locomotion alone as well.

The different coupling between locomotion and arousal in these studies illustrates a limitation in using locomotor speed as a single scalar metric of behavior. An animal can run for many reasons—in some cases it may represent hyper-arousal, such as startle, whereas in others it may represent an optimal state, such as goal-directed navigation. Furthermore, locomotor speed itself is an important variable that the brain is likely to represent independently from arousal state, and in fact recent studies have demonstrated continuous encoding of locomotor speed in visual cortex (Saleem et al., 2013). Thus, locomotion likely represents both an internal state that is partially correlated with arousal, as well as a physical variable that is important for both navigation and processing sensory information relative to self-motion and self-generated noise.

The effects of state show a striking difference across sensory modalities. As shown here, in auditory cortex both high arousal and locomotion are coupled with a decrease in sensory-evoked responses, whereas other studies have shown that in visual cortex these are associated with an increase in response gain. These differences may be due to differing requirements for sensory processing. For example, in vision movement through the environment predictably interacts with the sensory input, whereas in audition movement can cause self-generated noise that needs to be canceled. On the other hand, these may also represent differences in when each sensory modality is engaged ethologically—vision may be most important when navigating, whereas audition may be most important for detecting predators.

The findings presented by McGinley et al. (2015), along with the diverse effects seen across sensory modalities, raise the question as to which neural circuits underlie the state changes. A likely candidate for global changes in state, such as arousal, is neuromodulation, and evidence points to both norepinephrine(NE) and acetylcholine (Ach). Pupil dilation is often thought to be associated with noradrenergic tone (Aston-Jones and Cohen, 2005), and a recent study by Polack et al. (2013) showed that NE was necessary for the elevation in baseline membrane potential with locomotion, similar to the correlated changes in membrane potential and pupil diameter demonstrated here. On the other hand, studies have demonstrated a role for cholinergic in puts to cortex in regulating the strength of response to a visual stimulus (Fu et al., 2014; Pinto et al., 2013). These findings may be reconciled by a model in which NE regulates a neuron’s “set point” in terms of resting membrane potential, whereas Ach regulates sensory-evoked responses, which is supported by pharmacological and lesion experiments in somatosensory cortex (Constantinople and Bruno, 2011).

The ability to self-regulate arousal state is an essential aspect of executive function that develops in childhood (Posner and Rothbart, 2000). Intriguingly, McGinley et al. (2015) found that even though the arousal state fluctuated, it was generally centered around the optimal state for maximizing neural responses and behavioral performance. This raises the possibility that these well-trained mice had learned to regulate their state for the task at hand, although this remains to be tested by measuring state fluctuations in untrained mice. If so, investigating the circuit mechanisms underlying the acquisition of this metacognitive control could have important implications for both educational and therapeutic interventions.

Interestingly, the initial study by Yerkes and Dodson (1908) was actually performed in mice, measuring the rate of learning in a visual task, and was subsequently adopted in human psychology. The work of McGinley et al. (2015) therefore represents a full circle, from mice to humans and back to mice. With new measures of behavioral state and powerful tools for observing and manipulating neural circuits, experiments in both mice and humans can now continue to explore how neural dynamics give rise to the ups and downs in our daily experience.