Neuromodulatory systems partially account for the topography of cortical networks of learning under uncertainty

Human learning in a dynamic and stochastic environment relies on computational variables such as confidence and surprise. If the learning process is shaped by neuromodulation, then the spatial distribution of receptors and transporters across the brain could put constraints on the spatial distribution of learning-related neural activity. Here, using fMRI data from four probabilistic learning studies and a Bayesian ideal observer model, we reveal a strong spatial invariance across tasks for the functional correlates of confidence, and to a lesser extent, surprise. Using 20 PET receptor/transporter density maps, we then show that this invariance could be partly explained by the chemoarchitecture of the cortex. We identified multiple receptors and transporters whose distribution aligned with the spatial distribution of neural activity in the cortex. While many of these receptors/transporters are in line with previous proposals of neuromodulation of learning, the results also revealed novel associations that can be targeted in experimental studies.