The pupil offers a window into processes that reflect neuromodulatory brain state. For example, changes in pupil diameter are thought to correlate with exploitatory and exploratory behavioural states. While many studies acknowledge that arousal-based pupil responses are important in controlling behaviour, another important source of information is a signal that carries information about the value of the current behavioural state. Here, we investigated to what extent pupil responses reflect dynamics of behavioural control related to value-based learning. Thirty participants performed an implicit reversal learning experiment where we manipulated the probability of receiving reward or punishment, while recording their pupil size. Coloured cues signalled an upcoming reward or punishment with 80% validity. The contingency between colour and reward/punishment reversed unpredictably, prompting value-based learning of cue-reward contingencies in the participant, who was required to signal these reversals by keypress. Using ridge regression deconvolution, we found pupil signals that indexed reward-based learning differently on separate timescales. Perception of value contingency reversals was tracked by slow pupil signals on a timescale of minutes and correlated with behavioural responses, whereas fast trial-based pupil signals tracked instantaneous prediction errors. These data show that slow pupil responses carry functional information about value-based learning processes over and above fast trial-based pupil responses. Our results hint at the utility of slow pupil size fluctuations as a proxy for brain states relevant to behavioural control.
展开▼
