My collaborator Nick Roy is presenting a poster on our work at the 2013 Society for Neuroscience meeting, on Monday, November 13, in the morning session. Our poster is titled: Dynamics and temporal stability of neural representations in the prefrontal cortex during encoding and maintenance of working memory. In this work, we applied population-level analyses to neuronal data recorded during two classic working memory tasks: the oculomotor delayed response task (developed extensively in the lab of Pat Goldman-Rakic), and the vibrotactile delayed discrimination task (developed extensively in the lab of Ranulfo Romo). Nick is an undergraduate at Yale University who has been working on this project for his Senior Thesis. Our abstract is below:
Dynamics and temporal stability of neural representations in the prefrontal cortex during encoding and maintenance of working memory
Nicholas A. Roy, John D. Murray, Ranulfo Romo, Christos Constantinidis, Alberto Bernacchia, Xiao-Jing Wang
Working memory requires the brain’s ability to convert a brief stimulus-driven signal into an internal representation that can then be maintained across a mnemonic delay of several seconds. Many experimental paradigms have demonstrated a neural correlate of working memory in the form of stimulus-selective persistent activity during the delay. Recent studies have highlighted temporal variations in the delay activity of single neurons and heterogeneous delay activity patterns across a neural population in the monkey prefrontal cortex. However, the relationship between the population level representations during the cue presentation and the delay has not been well characterized.
To elucidate the neural transformation from encoding to maintenance, we analyzed the activity of many single neurons in monkey prefrontal cortex during two classic working memory tasks used in primate electrophysiology: the oculomotor delayed response task and the vibrotactile delayed discrimination task. Both tasks demand working memory of an analog stimulus variable, but they differ in their respective stimulus properties, the type of neural tuning, and required behavioral responses.
In our analysis, we compared stimulus representations between the cue epoch and the delay epoch. For both tasks, population activity patterns change strongly from the cue to delay, such that activity during the delay is poorly correlated with activity during the cue. Using a demixed principal component analysis, we identified low-dimensional subspaces within the population activity that maximally encode the stimulus with minimal temporal dynamics. Despite the strong temporal dynamics across the epochs, the subspace of stimulus encoding is essentially preserved between the cue epoch and the delay epoch. These results were similar for both tasks, supporting the generality of these phenomena.
These findings suggest that although prefrontal networks undergo strong changes between and within the epochs of a working memory task, the stimulus is stably represented within a constant subspace shared across epochs. The properties of population dynamics across encoding and maintenance pose constraints for computational models of working memory networks.