Emergence of cortical network motifs for short-term memory during learning

• Main Author: Chia, Xin Wei
• Other Authors: Hiroshi Makino
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Medicine, Health and Life Sciences; Neuroscience
• Online Access: https://hdl.handle.net/10356/175902
• Institution: Nanyang Technological University

Learning of adaptive behaviours requires refinement of coordinated activity among neurons in multiple brain areas. Brain-wide studies are limited to wide-field imaging to study population-level neural interactions. However, single-cellular interactions across multiple regions and how they emerge remain unknown. Using a two-photon random access mesoscope, we simultaneously recorded calcium activity of layer 2/3 excitatory neurons across eight regions of the mouse cortex during learning of a motor task involving working memory. Using an encoding model, we identified functional coupling between neurons, which revealed cellular network motifs distributed in multiple brain regions. Over learning, while functional connectivity became globally sparse, there emerged a functional subnetwork composed of neurons in anterior lateral motor (ALM) cortex and posterior parietal cortex (PPC). Neurons in the PPC-ALM subnetwork showed coordinated activity on a trial-by-trial basis. Furthermore, PPC and ALM neurons sharing a similar choice code during the delay epoch formed recurrent functional connectivity to generate persistent activity, a neural substrate of short-term memory. This could be attributed to a learning-dependent refinement where choice-relevant functional couplings were selectively retained while choice-irrelevant couplings were lost. We further confirmed the importance of PPC-ALM subnetwork by inactivating PPC via optogenetics and designer receptors exclusively activated by designer drugs (DREADD), which led to a significant reduction in task performance. Improvement in task performance may be due to an enhancement in the robustness of choice- related attractor dynamics. We show that recurrent neural networks (RNN) reconstructed from neural activity of ALM supplemented with PPC-ALM activity rendered choice-related attractor dynamics more stable. In conclusion, we show that learning creates cortical network motifs with specific inter-areal communication channels. This is achieved through selective refinement of choice-related functional connectivity. We propose that these refined cortical network motifs enhance choice-related attractor dynamics to improve task performance.