4/15, LSBM Monday Seminar Series #27, Miho Nakajima, Elucidating the structure of prefrontal cortical network underlying cognitive flexibility

Title: Elucidating the structure of prefrontal cortical network underlying cognitive flexibility

Presenter: Miho Nakajima

Affiliation: Deputy Team Leader, RIKEN Center for Brain Science, Laboratory for Distributed Cognitive Processing

https://distributedcognitionlab.riken.jp/member-jp.html

Cognitive control, the ability to orchestrate behavior/thoughts depending on the internal goal, is essential for animals to survive in dynamically changing world. While this control must be stable for a given task, it must also be flexible enough to rapidly switch between tasks (cognitive flexibility). It is often considered that the prefrontal cortex (PFC) implements this stable/flexible cognitive control by sending top-down signals to many other brain regions to guide their information processing depending on the task demands. However, due to the complexity of PFC representations, it remains unclear how PFC can perform multiple tasks without interference. One major question is whether the specific network structures of the PFC, consisting of diverse PFC cell types (Nakajima et al., Cell, 2014; Nakajima Semin.Cell.Dev.Biol, 2022), are required for flexible PFC computations (Ullman Science 2019; Yang et al., Nat. Neurosci. 2019; Dubreuil et al., Nat. Neurosci. 2022). We approached this problem by identifying a specific class of cell types, ‘decoder neurons’, which act as fixed nodes of PFC network and examined whether there are any principles in how the PFC network reconfigures to recruit appropriate decoder neurons after switching tasks. I previously demonstrated that in a rule-dependent sensory selection task, PFC projection neurons targeting different territories of the striatum acted as decoders, providing outputs that control the relative gain of meaningful and distracting inputs (Nakajima et al., Neuron, 2019a, 2019b). By changing which stimuli are distractors or target signal for each task, we succeeded in setting up new task-switching paradigms that recruit different sets of decoder neurons after shifting task. By recording and manipulating neural activity in the PFC and the striatum of task-performing mice, we found that inhibitory neurons (cell types that inhibit other neurons’ activity) are involved in controlling which decoder neurons to recruit in each PFC task network. Overall, our study shows the dynamic reconfiguration of PFC task networks are not generated randomly but rather that they are structured by inhibitory neurons to flexibly recruit proper decoder neurons. This finding shows that PFC is critically different from artificial intelligence (AI) based on randomly connected networks and might provide explanation for why the brain is more flexible than most AI.