A Toaster in the Bathroom: Neural Correlates of Semantic Construction During Episodic Memory Recall

When recalling what you ate for breakfast last Wednesday, you might not remember the exact meal, but you may confidently select the items you typically eat. Here, semantic knowledge (i.e., what you usually eat) contributes to the reconstructive process of episodic memory retrieval (i.e., what you actually ate). In the current fMRI study, we used a highly realistic virtual environment to test this influence of semantic knowledge on episodic memory retrieval. During the task, 60 participants actively (task-relevant) or passively (task-irrelevant) encountered everyday objects that were either congruent (i.e., rubber duck in the bathroom) or incongruent (i.e., a toaster in the bathroom) with their expected location. Thereby, we created conflicting information between the episodic memory trace (toaster in the bathroom) and semantic information (toaster in the kitchen) during retrieval. Using multivariate analyses, we analyzed the neural basis of this semantic bias. Further, we administered cortisol, typically associated with impaired episodic memory retrieval, to half of the participants prior to retrieval, thereby manipulating the balance between correct episodic and incorrect semantic retrieval. In the lateral occipital cortex (LOC), incongruent task-relevant objects showed greater similarity to their congruent semantic counterparts than did task-irrelevant objects. Notably, spatial memory tended to be reflected in similarity patterns in the LOC. Strikingly, incongruent objects showed a higher pattern reorganization (i.e., pre-/post-encoding similarity) compared to congruent objects, reflecting a difference in neural representation for objects encountered in conflict with prior knowledge. In contrast to our hypotheses, cortisol prior to retrieval had no effect on semantic bias. However, cortisol influenced neural pattern similarity: we found higher pattern reorganization within the posterior hippocampus in the cortisol group. Similarly, we found higher confidence to be linked with similarity patterns in the LOC and lingual gyrus in the placebo, but not in the cortisol group. This indicates an effect of cortisol on memory trace reinstatement during retrieval. Our findings on incongruent object processing contribute to the understanding of how the human brain constructs past episodes from episodic memory traces, suggesting an influence of prior semantic knowledge, reflected in neural similarity patterns.