Direct Brain Recordings Identify a Causal Subsequent-Memory Effect

Abstract

Endogenous variation in brain state and stimulus-specific evoked activity can both contribute to successful encoding. Previous studies, however, have not clearly distinguished among these components. We address this question by analyzing intracranial EEG recorded from epilepsy patients as they studied and subsequently recalled lists of words. We first trained classifiers to predict recall of either single items or entire lists and found that both classifiers exhibited similar performance. We found that list-level classifier output--a biomarker of successful encoding--tracked item presentation and recall events, despite having no information about the trial structure. Across widespread brain regions, decreased low- and increased high-frequency activity (HFA) marked successful encoding of both items and lists. We found regional differences in the hippocampus and prefrontal cortex, where in the hippocampus HFA correlated more strongly with item recall, whereas in the prefrontal cortex HFA correlated more strongly with list performance. Despite subtle differences in item- and list-level features, the similarity in overall classification performance, spectral signatures of successful recall, and fluctuations of spectral activity across the encoding period argue for a shared endogenous process that causally impacts the brains ability to learn new information.