Connectivity principles underlying network dynamics and learning

PROJECT ABSTRACT If an organism performs an action that leads to a desired outcome, it is able to perform that action again in the future in order to obtain that same outcome. While work on the mechanisms of reinforcement learning has extensively studied how the brain learns certain actions are more valuable than others, there is little knowledge about how the brain actually re-enters neural states on-demand to produce the behavior that leads to the desired outcome. This is a central question in neuroscience which underlies learning, memory, and movement and has implications for therapies to restore these abilities including brain-machine interfaces. It is believed that connectivity between neurons gives rise to dynamics—rules for how the brain transitions between neural states—and that modification of connectivity enables learning to re-enter neural states. However, two main experimental challenges have impeded direct investigation: 1) measuring and manipulating connectivity between neurons in vivo, and 2) identifying the neurons and activity patterns generating a behavior. In this proposal, I will overcome these challenges using 1) 2-photon microscopy to measure and manipulate functional connectivity in vivo by photostimulating individual targeted neurons and measuring the network’s response, and 2) a brain-machine interface (BMI) paradigm to define how neural activity is transformed into behavior and reinforcement. Through experiments that apply these techniques based on novel models of network dynamics, my proposal seeks principles for how functional connectivity underlies network dynamics and enables learning in motor cortex, a critical region for generating movement. In the first Aim (K99), I will determine whether a model of network dynamics predicts functional connectivity and how patterned photostimulation propagates through connectivity to modify the network state. In Aim 2 (K99/R00), I will design a BMI to study whether functional connectivity constrains learning. The BMI will test whether it is easier to learn network states that can be entered through photostimulation propagation. I will also determine whether changes in functional connectivity support learning by testing whether photostimulation more easily propagates to enter learned network states. Finally, in Aim 3 (R00), I will reveal principles for how network activity can change network connectivity and dynamics. I will test different protocols for stimulating spatiotemporal patterns and reveal principles of stimulation protocols that change the network. During the K99, this work will be conducted in the collaborative Zuckerman Institute for Brain and Behavior at Columbia University with the mentorship of Dr. Rui Costa - expert in the neurobiology of action and Dr. Liam Paninski – expert in computational modeling, and with the collaboration of Dr. Darcy Peterka – expert in optics and 2-photon microscopy with photostimulation. I believe their technical and professional mentorship will position me to lead an independent group studying principles for how networks generate and learn dynamics driving behavior. This work will have important therapeutic applications, including for brain-machine interfaces.