A frontal dopamine system for reflective exploratory behavior

•We examine exploratory behavior in relation to frontal dopamine gene COMT genotype.•We model participant’s behavior with reflective and reflexive computational models.•Met carriers perform better than Val/Val homozygotes under dual-task conditions.•Met carriers better maintain reflective behavior under dual-task conditions. The COMT gene modulates dopamine levels in prefrontal cortex with Met allele carriers having lower COMT enzyme activity and, therefore, higher dopamine levels compared to Val/Val homozygotes. Concordantly, Val/Val homozygotes tend to perform worse and display increased (interpreted as inefficient) frontal activation in certain cognitive tasks. In a sample of 209 participants, we test the hypothesis that Met carriers will be advantaged in a decision-making task that demands sequencing exploratory and exploitive choices to minimize uncertainty about the reward structure in the environment. Previous work suggests that optimal performance depends on limited cognitive resources supported by prefrontal systems. If so, Met carriers should outperform Val/Val homozygotes, particularly under dual-task conditions that tax limited cognitive resources. In accord with these a priori predictions, Met carriers were more resilient in the face of cognitive load, continuing to explore in a sophisticated manner. We fit computational models that embody sophisticated reflective and simple reflexive strategies to further evaluate participants’ exploration behavior. The Ideal Actor model reflectively updates beliefs and plans ahead, taking into account the information gained by each choice and making choices that maximize long-term payoffs. In contrast, the Naïve Reinforcement Learning (RL) model instantiates the reflexive account of choice, in which the values of actions are based only on the rewards experienced so far. Its beliefs are updated reflexively in response to observed changes in rewards. Converging with standard analyses, Met carriers were best characterized by the Ideal Actor model, whereas Val/Val homozygotes were best characterized by the Naive RL model, particularly under dual-task conditions.