Selective changes in cortical cholinergic signaling during learning

Cognition relies on the function of local and long-range neural circuits in the neocortex, which are dynamically regulated by neuromodulatory signals including acetylcholine (ACh). Recent work suggests that the role of acetylcholine varies across cortical areas and behavioral states. In addition, the precise role of cholinergic signaling during learning and plasticity remains unclear. We performed simultaneous, dual-color mesoscopic imaging of ACh and calcium signaling in the neocortex of awake mice across multiple stages of visual task learning to identify learning-related plasticity in neural activity and cholinergic signaling. Using a novel Pattern Reconstruction and Interpretation with a Structured Multimodal transformer (PRISMT) model combining masked autoencoding with multimodal causal attention, we identified key cortical regions that reorganize with learning. We found that cholinergic signaling exhibited spatiotemporally selective plasticity in frontal cortical subnetworks during visual perceptual learning. Overall, our findings demonstrate the utility of transformer models for generating biologically interpretable insight into large multimodal datasets and link long-term functional reorganization of cortical networks to enhanced task performance.