A ROLE FOR THE ANTERIOR CINGULATE CORTEX IN REINFORCEMENT-GUIDED LEARNING FOR COVERT ATTENTIONAL SELECTION

The anterior cingulate cortex (ACC) has been associated with a variety of functions including conflict monitoring, error detection and more recently reward based learning. In this study we recorded from the ACC, the ventromedial prefrontal cortex (vmPFC) and the dorsolateral prefrontal cortex (dlPFC) while macaque monkeys performed a variably rewarded spatial attention task. First, we found dynamic encoding of reward outcome and reward expectancy associated with attentional targets within the ACC, mPFC and dlPFC. These results expand the function of the ACC beyond merely action value associations and suggest this area serves a broader role in reinforcement guided learning and decision making. Secondly, analysis of outcome encoding relative to reward reversal revealed two distinct types of neurons: positive/negative prediction error neurons and positive/negative prediction certainty neurons. Prediction error neurons encoded outcome information only when reward associations had recently changed and thus new outcome information was most informative for establishing new reward expectations. Prediction certainty neurons on the other hand signaled the certainty of the reward prediction itself and encoded outcome information only later, when reward expectations had been built up. Prediction error neurons showed a correlation between reward selectivity during outcome periods and reward selectivity preceding subsequent reward predictive events. This finding could serve as a link between prediction error signals and behavioural adjustment. Finally, prediction error neurons predominated in the ventral ACC whereas prediction certainty neurons predominated in dlPFC area 9. Though not definitive this supports proposals that outcome predictions are developed and adjusted within the ACC and mPFC and these predictions are used by the dlPFC to determine the behavioural response