Abstract
The International Brain Laboratory (IBL) is a collaboration of more than 20 experimental and theoretical labs studying decision-making in the mouse. Using a visually-guided decision task, IBL has demonstrated robust, replicable, behavioral and electrophysiological data, along with a growing complement of tailored and richly validated computational models. This provides an unparalleled opportunity to study the activity, and ultimately the function, of neuromodulators, which play a special and important role in theories of learning and decision-making. Dopamine (DA) has been suggested to report a reinforcement prediction error, at least under some circumstances and to some target nuclei. Equally, there is evidence that serotonin (5-HT) modulates learning based on confidence or uncertainty, and in reporting an unsigned or state prediction error. To understand the differential activation of these neuromodulators in the IBL task, we are using fiber-photometry to record bulk Ca2+ activity using GCaMP6 sensors expressed in genetically-defined neural populations in key nuclei. We will compare the activity of the neuromodulators to theoretical predictions and to the IBL behavior and brain wide electrophysiology data sets. As a validation of the approach, we tested the hypothesis that ventral tegmental area DA neurons report a signed prediction error and dorsal raphe nucleus 5-HT neurons an unsigned prediction error. We used, respectively, 5 and 4 adult mice to target DA and 5-HT, using the isosbestic signal to control the GCaMP6 signal. During the first training stage, habituation, the mouse is being acclimatized to the rig and head-fixation, passively observing a visual stimulus that moves from the side to the center of the screen followed by a water reward delivery. Both DA and 5-HT neurons respond to the unexpected water. Next the mouse must learn to move the stimulus to the center by using a wheel. We observe a divergence in the DA signal associated with the outcome of the choice. DA activity increases upon water reward and decreases when the mouse makes the wrong decision and receives a noise burst. 5HT, in contrast, increases to both reward and noise burst, with a stronger response associated with incorrect trials, consistent with an unsigned prediction error signal. Currently, we are acquiring preliminary data for norepinephrine and acetylcholine, also proposed to play key roles in decision-making, with the goal of recording the activity of the four primary neuromodulators in the same task. In combination with the current IBL data sets, this study provides an unprecedented opportunity to investigate the role of neuromodulators in learning and decision-making.
