Abstract

Significance Working memory (WM) is a core cognitive function thought to rely on persistent activity patterns in populations of neurons in prefrontal cortex (PFC), yet the neural circuit mechanisms remain unknown. Single-neuron activity in PFC during WM is heterogeneous and strongly dynamic, raising questions about the stability of neural WM representations. Here, we analyzed WM activity across large populations of neurons in PFC. We found that despite strong temporal dynamics, there is a population-level representation of the remembered stimulus feature that is maintained stably in time during WM. Furthermore, these population-level analyses distinguish mechanisms proposed by theoretical models. These findings inform our fundamental understanding of circuit mechanisms underlying WM, which may guide development of treatments for WM impairment in brain disorders.