Alpha oscillations track content-specific working memory capacity.

Though the neural basis of working memory (WM) capacity is often studied by exploiting inter-individual differences, capacity may also differ across memory materials within a given individual. Here, we exploit the content-dependence of WM capacity as a novel approach to investigate the oscillatory correlates of WM capacity, focusing on posterior 9-12 Hz alpha activity during retention. We recorded scalp electroencephalography (EEG) while male and female human participants performed WM tasks with varying memory loads (2 vs. 4 items) and materials (English letters vs. regular shapes vs. abstract shapes). First, behavioural data confirmed that memory capacity was fundamentally content-dependent: capacity for abstract shapes plateaued at around two, while the participants could remember more letters and regular shapes. Critically, content-specific capacity was paralleled in the degree of attenuation of EEG-alpha activity that plateaued in a similar, content-specific, manner. While we observed greater alpha attenuation for higher loads for all materials, we found larger load effects for letters and regular shapes than for abstract shapes - consistent with our behavioural data showing a lower capacity plateau for abstract shapes. Moreover, when only considering 2-item trials, alpha attenuation was greater for abstract shapes - where 2-items were close to the capacity plateau - than for other materials. Multivariate decoding of alpha-activity patterns reinforced these findings. Finally, for each material, load effects on capacity ( K) and alpha attenuation were correlated across individuals. Our results demonstrate that alpha oscillations track memory capacity in a content-specific manner and track not just the number of items, but also their complexity. SIGNIFICANCE STATEMENT Working memory (WM) is limited in its capacity. We show that capacity is not fixed for an individual but is rather memory-content dependent. Moreover, we used this as a novel approach to investigate the neural basis of WM capacity with EEG. We found that both behavioural capacity estimates and neural oscillations in the alpha band varied with memory loads and materials. The critical finding is a capacity plateau of approximately two items only for the more complex materials, accompanied by a similar plateau in the EEG alpha attenuation. The load effects on capacity and alpha attenuation were furthermore correlated across individuals for each of the materials. Our results demonstrate that alpha oscillations track the content-specific nature of WM capacity.