ITEM-SPECIFIC SELECTIVE MODULATION OF MOTORIC ASSOCIATIVE TRACES BY REWARD PREDICTION ERRORS: A FUNCTIONAL DISSOCIATION

Optimal cognitive control relies on the ability to update specific associative links while maintaining stable goal representations. While Reward Prediction Errors (RPEs) are fundamental learning signals, it remains unclear whether they drive a general state-level shift or an item-specific update of memory traces. We investigated the impact of phasic RPEs on the functional dissociation between item-specific motoric (Stimulus-Action, S-A) and conceptual (Stimulus-Classification, S-C) associative bindings. We employed an item-based task-switching prime-probe paradigm (N=25) featuring 480 unique stimuli. Prime trials manipulated RPE valence (Negative, Neutral, Positive) via performance feedback. In subsequent probe trials (variable lag 2-7), we independently manipulated Action-switching (S-A) and Task-switching (S-C) relative to the prime encounter. This design allowed us to isolate the long-term, stimulus-specific effects of RPE-driven encoding on subsequent associative retrieval. Repeated-Measures ANOVA confirmed a robust functional dissociation at the item level. RPEs selectively modulated motoric traces, significantly increasing S-A switch costs (p = .004). In contrast, conceptual S-C traces remained unaffected by RPE valence. Detailed analysis revealed that RPEs simultaneously strengthened the "old" S-A binding (increasing proactive interference) while demanding increased inhibitory control during item-specific action switching. Our findings demonstrate that phasic RPEs drive an item-specific updating mechanism within the striatal procedural system (S-A binding), while sparing higher-level prefrontal conceptual representations (S-C binding). This suggests that RPEs serve as a precision tool for recalibrating motoric associations, and provides a behavioral framework for future neuroimaging studies to explore the segregation of prediction error processing across cortico-striatal circuits.