Introduction Brain processes of working memory involve oscillatory activities at multiple frequencies in local and long‐range neural networks. The current study addressed the specific roles of alpha oscillations during memory encoding and retention, supporting the hypothesis that multiple functional mechanisms of alpha oscillations exist in parallel. Method We recorded magnetoencephalography (MEG) in 25 healthy young adults, who performed a variant of a Sternberg working memory task. A sequential list of five consonant letters was visually presented and was followed after a 2.0 s retention interval by a probe of a pair of two letters from the study list. Participants responded whether the probe pair was in same or reversed order in the list. Result Reaction time (RT) was shortest for the first letters in the list, increased with increasing serial position, and shorter for the last position. RT was substantially longer for the probe in reversed order. Time‐frequency analysis of the MEG revealed event‐related desynchronization (ERD) of alpha oscillations during the encoding interval and an alpha power increase (ERS) during memory retention. Alpha ERD during encoding occurred at 10 Hz and ERS during retention at 12 Hz, suggesting different alpha mechanisms. Analysis of alpha coherence and alpha‐gamma cross‐spectral coupling, applied to MEG beamformer source activity, revealed connectivity across brain areas. Additionally, alpha‐gamma coupling identified centers of local computation. The connectivity between occipital and frontotemporal areas was correlated with alpha ERS during memory retention. Cross‐frequency coupling between alpha phase and gamma amplitude depicted a hierarchy of information flow from frontal to temporal and occipital brain areas. Conclusion Alpha decrease during encoding indicates an active state of visual processing, while subsequent ERS indicates inhibition of further visual input for protecting the memory, and phasic timing of temporal and occipital gamma oscillations is related to a long‐range working memory networks.

No abstract