ACC representations of reward-driven motivation over hierarchically-organized behavior

The neural mechanisms of reward-guided behavior have been researched extensively but how rewards affect the neural processes underlying the production of hierarchically-organized sequential behaviors remains unclear. In particular, anterior cingulate cortex (ACC) is said to motivate the successful completion of hierarchically-organized tasks by grouping together low-level actions according to superordinate goals, which are encoded across ACC neural ensembles as distributed, multivariate representations (Holroyd & Verguts, 2021; Holroyd & Yeung, 2012). Here we applied model-based representational similarity analysis (RSA) to investigate how rewards modulate these representations. We adapted an existing recurrent neural network (RNN) model of ACC activity (Colin et al., 2024) to predict the effect of rewards on ACC representations in a hierarchical sequence production task, and then recorded the fMRI-BOLD response from participants engaged in this task. As predicted, searchlight RSA revealed a cluster of MRI activity in ACC showing the highest second-order similarity with the model representation, relative to the whole-brain distribution. Closer examination revealed that ACC discriminates between rewarded vs non-rewarded sequences mainly on the first step of each trial, consistent with evidence that ACC encodes a reward prediction error at the onset of hierarchically organized action sequences. Moreover, reward-related multivariate ACC activity positively correlated with task performance, with more distinct ACC activity patterns predicting higher accuracy and faster response times. These findings were unique to ACC as other brain areas exhibited different task-related effects. We conclude that ACC monitors reward contingencies to motivate control over hierarchically-organized, goal-directed behaviors.