Working memory requires large-scale coordination among widespread cortical and subcortical brain regions. Long-range coherence between the oscillatory activity of distant regions, arising in multiple frequency bands, may be a mechanism supporting information routing and integration across large-scale distributed multi-regional networks. In line with this hypothesis, the strength of long-range coherence links between frontal and parietal regions is consistently modulated by the specific identity of a visual stimulus held in working memory during a delay period (Salazar et al., 2012).
Here, we perform a systematic analysis of massively parallel Local Field Potentials recorded in non human primates performing a delayed match-to-sample task and characterize the dynamics in time of large-scale and coherence networks at frequencies ranging from 3 to 80 Hz. While some of the stronger links display a coherence level which is relatively stable and sustained through the entire delay period of the tasks, other links display marked fluctuations in the intensity of their coherence which is weaker on average but can rise very transiently to more sustained levels. Such weak links do not activate independently but in coordination with other links whose stochastic-like coherence fluctuations are mutually covarying, resulting in the emergence of bursting subnetworks, particularly in the high beta (~25-30 Hz) and gamma frequency bands (~60-80 Hz).
We then compute mutual information between the identity of the object held in working memory and: i) the fluctuating band-restricted powers of individual network nodes (brain regions); or, ii) the fluctuating coherence strengths of individual network links (pairs of brain regions). We confirm that a variety of distributed nodes and links can potentially encode for object identity. However, node-level encoding by fluctuating oscillatory power, prominent when the visual object is shown, rapidly fades away when entering the delay period. In contrast, link-based encoding by long-range coherence persists down to the latest stages of the delay period.
Remarkably, the pairwise coherence links which display the more mutual information with object identity are not the stronger, but, on the contrary, the more “entangled” ones, i.e. links whose fluctuating weak coherence is highly correlated with the fluctuations of other links, as quantified by brain-wide edge-centric measures of dynamic functional connectivity.
Analyses of the dynamics of large-scale oscillatory networks provide thus a useful window on collective coordination mechanisms, potentially shaping distributed neural computations in working memory.
Find here animated slides (mute video-talk) and a poster.