Cholinergic-mediated adaptive learning in cortical microcircuits

Abstract

With frequent exposure to novelty, it is imperative that the brain is capable of rapidly adjusting the appropriate synaptic connections to adapt its behavioural responses to new situations. Acetylcholine is a neuromodulator released during novelty and is important for synaptic plasticity and learning via its heterogeneous modulatory effects on different cell types. Here, we introduce a framework of cholinergic-mediated credit assignment (ChoCA) that models the cholinergic system (CS) based on adaptive methods used in machine learning. The CS serves as an adaptive module, integrating error signals to modulate synaptic updates derived from a backpropagation-like credit assignment mechanism in cortical microcircuits. The cholinergic signals are modelled at varying degrees of diffuseness to reflect the heterogeneity of the information transmitted. On standard pattern recognition tasks, our results show that diffuse cholinergic modulation is sufficient for effective adaptive learning. Moreover, the model postulates that cholinergic modulation of somatostatin (SST) interneurons is critical for credit assignment; specifically, it predicts that acetylcholine depresses the SST-to-pyramidal cell (PC) synapse. We also develop a novel plausible backprop-like model called Bursting Cortico-Cortical Networks which provides a mechanism for communicating error signals as bursting activity to the CS. Preliminary findings from \textit{ex vivo} slice electrophysiology experiments reveal that while the excitability of SST/OLM interneurons increases with bath application of a cholinergic agonist, the inhibitory transmission of these interneurons onto CA1 PCs is reduced in line with the model’s prediction. A dual optogenetics approach using the Chrimson and ChRger opsins was also developed to examine the effects of endogenous acetylcholine on this inhibitory transmission. Furthermore, optogenetic stimulation of these interneurons during a long-term potentiation (LTP) induction protocol prevents LTP indicating their importance in the regulation of synaptic plasticity. Overall, our work demonstrates the efficacy of coupling neuromodulatory signals with efficient credit assignment for adaptive learning in the brain.

Date of Award	20 Jun 2023
Original language	English
Awarding Institution	University of Bristol
Supervisor	Jack R Mellor (Supervisor) & Rui Ponte Costa (Supervisor)