Flexible, intelligent behaviour requires the maintenance and manipulation of incoming information over various time spans. For short time spans, this faculty is labelled ‘working memory’ (WM). Working memory (WM) provides the stability necessary for high-level cognition. Influential theories typically assume that WM depends on the persistence of stable neural representations, yet increasing evidence suggests that neural states are highly dynamic.

In this study, we apply multivariate pattern analysis to explore the population dynamics in primate lateral prefrontal cortex (PFC) during three variants of the classic memory-guided saccade task (recorded in 4 animals). We observed the hallmark of dynamic population coding across key phases of a working memory task: sensory processing, memory encoding, and response execution. Throughout both these dynamic epochs and the memory delay period, however, the neural representational geometry remained stable.

We identified two characteristics that jointly explain these dynamics: (1) time-varying changes in the subpopulation of neurons coding for task variables (i.e., dynamic subpopulations); and (2) time-varying selectivity within neurons (i.e., dynamic selectivity). These results indicate that even in a very simple memory-guided saccade task, PFC neurons display complex dynamics to support stable representations for WM.

Eelke Spaak, Kei Watanabe, Shintaro Funahashi and Mark G. Stokes
Journal of Neuroscience 30 May 2017, 3364-16; DOI: https://doi.org/10.1523/JNEUROSCI.3364-16.2017

http://www.jneurosci.org/content/early/2017/05/30/JNEUROSCI.3364-16.2017