Distributed Encoding of Reinforcement in Rat Cortico-Striatal-Limbic Networks. (10th August 2019)

• Record Type: Journal Article
• Authors: Donovan, Clifford H.; Badenhorst, Cecilia A.; Gruber, Aaron J.
• Is Part Of: Neuroscience. Volume 413(2019)
• Page Start: 169
• Page End: 182
• Publication Date: 2019-08-10
• Subjects: decision-making -- deep learning -- field potential -- information flow -- machine learning -- reward; Neurochemistry -- Periodicals; Neurophysiology -- Periodicals; Neurology -- Periodicals; Neurochimie -- Périodiques; Neurophysiologie -- Périodiques; Neurochemistry; Neurophysiology; Electronic journals; Periodicals; Electronic journals; 612.8
• DOI: 10.1016/j.neuroscience.2019.06.019
• Languages: English
• ISSNs: 0306-4522

Abstract

Abstract: Decision-making in the mammalian brain typically involves multiple brain structures within the midbrain, thalamus, striatum, limbic system, and cortex. Although task specific contributions of each brain region have been identified, neurons responding to reinforcement have been found throughout these structures. We sought to determine if any brain area, or cluster of areas, are the source of information, and if the fidelity of information varies among the areas. We recorded simultaneous field potentials (FPs) in rats from seven brain regions as they completed a binary choice task. The FPs of a 0.5 s window following reinforcement were given as input to a classifier that attempted to predict whether or not the rat received reward on each trial. The classifier correctly categorized reward on 77% of trials. Any region-specific signal could be omitted without lowering accuracy. Frequencies above 40 Hz and signals recorded later than 0.25 s following reinforcement were necessary to achieve this accuracy. Further, the classifier was able to predict reinforcement outcome above chance levels when using FPs from any single recorded brain region. Some combinations of structures, however, were more predictive than others. Analysis of FPs prior to reward revealed most regions reflected the prior probability of reward. Lastly, analyses of information flow suggested reinforcement information does not originate within a single structure of the network, within the resolution afforded by FP recordings. These data suggest reward delivery information is rapidly distributed non-uniformly across the network, and there is no canonical flow of information about reward events in the recorded structures. Highlights: We recorded field potentials from seven brain regions of the cortico-striatal-limbic circuit as rats performed a binary choice task. Using a neural network classifier, we sought to determine which areas encoded reinforcement outcome. Using data from all seven regions, the classifier correctly categorized reinforcement outcome on 77% of trials. We found field potentials from any individual recorded region could classify the outcome of trials above chance levels.