Network Structure During Encoding Predicts Working Memory Performance
Working memory operates through networks that integrate distributed modular brain activity. We characterize the structure of networks in different electroencephalographic frequency bands while individuals perform a working memory task. The objective was to identify network properties that support working memory function during the encoding, maintenance, and retrieval of memory. In each EEG frequency band, we estimated a complex-valued Gaussian graphical model to characterize the structure of brain networks using measures from graph theory. Critically, the structural characteristics of brain networks that facilitate performance are all established during encoding, suggesting that they reflect the effect of attention on the quality of the representation in working memory. Segregation of networks in the alpha and beta bands during encoding increased with accuracy. In the theta band, greater integration of functional clusters involving the temporal lobe with other cortical areas predicted faster response time, starting in the encoding interval and persisting throughout the task, indicating that functional clustering facilitates rapid memory manipulation.