Jaime de la Rocha, Ph.D., Principal Investigator in Neuroscience, IDIBAPS (Barcelona)
Abstract:
Working memory (WM) is central for cognition but it is prone to failure, especially when long-lasting memory maintenance or repeated engagement over long sessions are required. Persistent neural activity is associated with WM maintenance and attractor dynamics are viewed as the underlying network mechanism. However, a direct assessment of network dynamics during WM failure is still lacking. In this talk, we address this question in a two-alternative delayed-response task in which mice maintain a prospective response during a variable delay. We model the behavior using a Hidden Markov Model that switches between a task-engaged state described by attractor dynamics, and a task-disengaged state with responses purely determined by previous choices. Population recordings in the anterolateral motor cortex (ALM) show the hallmarks of attractor dynamics during task-engaged phases: delay persistent activity stably encoded impending choices in correct trials, whereas the encoding reversed during long delays in error trials, consistent with fluctuation-driven transitions between attractors. In contrast, in task-disengaged phases ALM does not show clear traces of attractor dynamics and instead exhibits enhanced spiking synchrony at ~8 Hz. Our results show that ALM activity switches between distinct dynamics during memory and non-memory guided choices, revealing the neural basis of forgetting errors caused by transitions between memory states, and non-memory errors caused by complete task disengagement.