P-10 Flexible sensorimotor linkage during decision-making in changing environments

Background: Humans and non-human primates can adapt to changing environments by flexibly switching between arbitrary mappings of sensory information to motor action. When the sensory-motor mapping rule is uncertain and can switch in a hidden fashion, online updating of an internal belief about the currently active task state (i.e., rule) is required to guide a higher-order decision: the selection of a sensorimotor mapping used to report a lower-order decision (e.g., stimulus discrimination). The computations and brain networks involved in such hierarchical decisions under uncertainty are not yet understood.

Objective: We aimed to (i) unravel the internal belief updating dynamics underlying the selection of sensory-motor mapping rules for a basic perceptual decision under uncertainty; and (ii) the cortical and subcortical areas involved in such computations.

Methods: We performed concurrent functional magnetic resonance imaging (fMRI) and pupillometry in 22 healthy human participants performing a hierarchical decision-making task. The task entailed discriminating the orientation of a full-contrast visual grating stimulus and reporting that judgment by using one of two possible rules governing the mapping from stimulus orientation to left or right hand button presses. In one condition (‘instructed rule’), that mapping rule was explicitly instructed. In the other condition (‘inferred rule’), it varied in a hidden fashion and had to be inferred from a noisy stream of visual cues around fixation. We fit a normative (Bayesian) belief updating process to participants’ behavior to infer the hidden belief dynamics and related computational variables. We used general linear modeling to relate these latent (model-inferred) variable to brain-wide neural activity and accompanying pupil responses.

Results: Participants’ behavior was well captured by an approximately normative belief updating process, and better so than simpler heuristics. fMRI-activity in several cortical regions during the lower-order decision (orientation judgment) was (i) stronger during the inferred rule condition than the instructed rule condition and (ii) covaried with computational variables constituting the belief updating in the inferred rule condition. This was specifically true for the inferred probability of a rule change, which was also tracked by transient dilations of the pupil. In several brainstem nuclei controlling cortical state and pupil-linked arousal (e.g., the locus coeruleus), we also found stronger evoked fMRI responses during the lower-order decision for (i) larger compared to smaller evoked pupil dilations and (ii) the inferred rule compared to the instructed rule condition.

Conclusion: Our study sheds light on the computations underlying, and the large-scale brain systems involved in, hierarchical decisions in volatile task states (sensory-motor mapping rules). Neuromodulatory brainstem systems controlling pupil-linked arousal are involved in the underlying belief updating process. These systems may facilitate the implementation of a change in belief about a current task state, and the resulting change in sensory-motor associations, in cortical networks.