Social Reinforcement Learning and Its Neural Modulation By Oxytocin in Autism Spectrum Disorder
Objectives: We investigated the potential of OXT to compensate for deficits in socially reinforced learning in ASD and its underlying neural mechanism in a social learning task, which allowed for the differentiation of social feedback stimuli and social stimuli as the target of learning.
Methods: Using functional Magnetic Resonance Imaging we assessed brain activation during performance of a probabilistic reinforcement learning task in 24 typically developing controls (TDC) and 15 patients with ASD (18-26 years) in a double-blind placebo-controlled cross-over design. Participants indicated whether social or non-social stimuli belong to category A or B and social or non-social feedback with non-100% contingencies was provided. Data were analyzed using computational modeling according to the Q-Learning model. From the behavioral data, trial-by-trial reward-prediction error (RPE) values were calculated. We assessed the correlation of brain activation with RPE values during feedback and brain activation related to the anticipation of reward during choice. Based on previous studies of RPE learning and anticipation of reward, we focused on brain activation in the nucleus accumbens using an ROI approach (p<.05, voxel level corrected for ROI).
Results: In the ASD group, OXT enhanced the correlation of the RPE signal with activation in the NAcc during social feedback despite the learning target being non-social, whereas in the TDC group this effect was found in the placebo (PLC) condition. The learning target being social did not show a modulation by OXT during feedback in ASD whereas in TDC, an enhanced correlation was found for non-social learning targets during OXT. Behaviorally, subjects from both groups demonstrated significant learning during the task across conditions. Patients with ASD showed better learning when the learning target was social rather than non-social in the OXT but not PLC condition.
Conclusions: Our results demonstrate that in ASD, OXT selectively enhances the correlation of the RPE with brain activation during social feedback, but no selectivity was evident for social stimuli as a learning target. In the TDC group, OXT had a rather attenuating effect. Hence, on the neural level, OXT may disrupt a functioning system in TDC whereas OXT rather compensates for deficits in ASD. Behaviorally, OXT had a facilitating effect in ASD for social learning targets but not social feedback. This pattern suggests that the modulatory role of OXT for social learning is not specifically tied to the learning target or feedback being social, although differential neural effects can be observed.