Divergent Neural Mechanisms of Memory Updating Under Varying Degrees of Inconsistency

Humans can flexibly update pre-existing memories with new experiences to accommodate the ever-changing world. However, the degree of inconsistency between new input and pre-existing memory, a critical factor that impacts memory updating, remains insufficiently understood. Here, we manipulated three degrees of inconsistency (low, moderate, high) to investigate how inconsistency modulates neural integration and separation in memory updating and impacts memory outcomes. In an EEG experiment, a retrieval-based associative learning paradigm was employed, in which participants were required to retrieve existing AB associations (overstudied in Session 1) before learning new AC associations in Session 2, followed by item (B) and associative (AB) memory recognition tests in Session 3. Using representational similarity analysis, we found that low and moderate inconsistency favored neural integration whereas high inconsistency promoted neural separation during reconsolidation. During retrieval, associative memory showed a monotonic decline as inconsistency increased, whereas item memory exhibited an inverted U-shaped pattern, and the associative and item memories correlated with neural integration and separation in opposite directions. Further event-related potential results showed that moderate inconsistency elicited the lowest P3b amplitude and high inconsistency elicited the highest P600 amplitude. These results provide empirical evidence that varying degrees of inconsistency modulate memory updating through divergent neural mechanisms, yielding differential mnemonic outcomes.