Pupil diameter tracks changes in control state predicted by the adaptive gain theory of locus coeruleus function

Abstract

An important dimension of cognitive control is the adaptive regulation of the balance between exploitation (pursuing known sources of reward) and exploration (seeking new ones) in response to changes in task utility. Recent studies have suggested that the locus coeruleus-norepinephrine system may play an important role in this function and that pupil diameter can be used to index locus coeruleus activity. On the basis of this, we reasoned that pupil diameter may correlate closely with control state and associated changes in behavior. Specifically, we predicted that increases in baseline pupil diameter would be associated with decreases in task utility and disengagement from the task (exploration), whereas reduced baseline diameter (but increases in task-evoked dilations) would be associated with task engagement (exploitation). Findings in three experiments were consistent with these predictions, suggesting that pupillometry may be useful as an index of both control state and, indirectly, locus coeruleus function.

Figures

Figure 1

Relationship between tonic pupil diameter and baseline firing rate of a locus coeruleus (LC) neuron in the monkey (adapted with permission from Rajkowski, Kubiak, & Aston-Jones, 1993). Pupil diameter measurements were taken by a remote eyetracking camera at each instant in time at which the monkey achieved fixation of a visual spot during the target detection task (see the text for details). Note the close positive relationship between pupil diameter and the rate of LC activity.

Figure 2

Variable-width histograms constructed from Vincentized response time distributions (Ratcliff, 1979) in Experiment 1A for trials associated with small (top panel) and large (bottom panel) baseline pupil diameters.

Figure 3

Time courses of grand-averaged poststimulus pupil dilations for correctly detected targets (hits) and correctly rejected (CR) distractors. For Experiment 1A (top panel), dilations are plotted separately for trials that were preceded by large and small baseline pupil diameters. For Experiment 1B (bottom panel), dilations are plotted separately for the light and dark conditions.

Figure 4

Grand-averaged baseline pupil diameter (top panel) and pupil dilation magnitude (bottom panel) as a function of conflict (low vs. high) and feedback (negative vs. positive), the two independent variables in Experiment 2.

Figure 5

Illustration of a sample trial in the diminishing-utility task used in Experiment 3. All the stimuli were separated by delays of sufficient length to allow resolution of any pupillary response to the preceding event. Baseline pupil diameters were measured in the 1 sec immediately preceding the onset of the score/value screen. Dilations of the pupil were measured as peak deviations from this baseline in the 2.5 sec following onset of the comparison tone. See the text for further details.

Figure 6

Grand-averaged dependent measures for peri-escape trials in Experiment 3. Trial number “0” indicates the escape trial. Top panel: Response time (RT), accuracy, and expected accuracy. Middle panel: Baseline pupil diameter and pupil dilation magnitude. Note that no measures of RT, accuracy, and reflexive pupil dilation are available for escape trials, because, on these trials, no comparison tone was presented. Bottom panel: Trial value and its computed expected value. Expected value most closely tracks the pattern of the pupil data.

Figure 7

Schematic outline of adaptive gain theory (AGT) and its relationship to the present research. ACC, anterior cingulate cortex; OFC, orbitofrontal cortex; LC, locus coeruleus; NE, norepinephrine; PGi, paragigantocellularis nucleus of the ventral medulla. One central assumption of AGT is that the LC firing mode is adaptively driven by online assessments of task utility. This part of the theory received empirical support in Experiments 2 and 3, which showed the predicted effect of experimental manipulations of conflict (costs) and feedback valence (reward) on pupillary dynamics. The second core assumption of the theory is that the LC uses this information about utility to make compensatory adjustments in control state. This part of the theory received support in Experiments 1 and 3, which showed the predicted relationship between pupillary dynamics and behavioral indices associated with the two LC modes.