Attention - Synchrony in working memory

When a stranger at a party gives you their number, how are you able to remember it for a few minutes until you find time to fish out your phone and store it? When we pay close attention to important information, such information is able to gain access to working memory, and is thereafter actively maintained over short time-scales (seconds to a few minutes). A unique characteristic of working memory is that it maintains information robustly even in the absence of repeated, explicit reminders (stable persistence).

When humans (or monkeys) engage in working memory tasks (such as remembering a number or a location), neurons in an evolutionarily advanced region of the brain, the prefrontal cortex, fire persistently, suggesting that they are actively involved in maintaining task-relevant information. However, the brain is inherently noisy, and previous models of working memory have shown that even slight differences in neural properties cause the maintained information to deteriorate very quickly over time.

How then does information persist stably in an inherently noisy brain without deterioration, over timescales of seconds to minutes?

When information is maintained in working memory, another phenomenon is consistently observed: the background activity (local field potential, or LFP) goes into powerful oscillations in the gamma band (25-90Hz).

We hypothesize that these gamma oscillations contribute to information's stable persistence in working memory.

Using John Arthur's chip we have shown that background gamma oscillations generated by inhibitory neurons homogenize the firing output (rates) of the other neurons by periodic shunting. This background inhibitory rhythm temporarily irons out built-in ("frozen") inequities in the neural landscape, thereby facilitating stable maintenance of information in working memory.

Robust spatial working memory through inhibitory gamma synchrony. A 32x32 array (left) of excitatory (blue) and inhibitory (yellow) neurons, a scrolling raster display (top right) of their spikes, and a histogram (bottom right) of all the spikes is shown. In previous models with recurrent excitatory-inhibitory coupling [Renart et al, 2003], the stored spatial pattern (blue pixels) tends to drift. Upon synchronizing the inhibitory neurons' firing in the gamma range (60Hz), the stored pattern remains relatively more stable [Sridharan et al 2011].