Project Summary/Abstract Movement is the primary way in which animals interact with the world. To produce the incredibly adaptable behavior of mammals the brain must continually choose actions, then flexibly generate control signals for the body. The main objective of this project is to understand how ensembles of neurons and brain areas work together to control movement and make simple decisions. To understand how behavior is generated, we must know: How is a high-level decision (e.g., reach left vs. right) transformed into a time-varying command signal? And, how does this transformation and the generation of complex outputs exploit the precise biology of neural tissue to function reliably, despite the inherent noisiness of neurons? This goal is critical not only to better understand how neural tissue implements challenging computations, but also because deeper knowledge of these processes is likely to improve treatments for motor disorders such as Parkinson’s and disorders heavily affecting neural connections such as stroke and traumatic brain injury. To achieve this aim, we pair the power of experimental tools in mice together with dynamical systems analysis, which provides mathematical tools for investigating the function of neural ensembles. In the monkey, the dynamical systems approach has led to many insights. For example, we have learned that activity in motor cortex unfolds over time according to oscillatory dynamical “rules”; that much of motor cortical activity exists to support these dynamical rules and does not influence movement directly; and that there is a separation between signals for what movement will be made and when it will be initiated. In the mouse, we propose taking this approach several steps further by mapping the dynamical rules to specific biological features, such as cortical layers and projection pathways. In Aim 1, we will use two-photon calcium imaging to record neural activity during a simple reaching task that elicits variable movements from the mouse. We will then exploit this variability with our dynamical systems tools to identify the rules that govern the M1 pattern generator, and uncover how these rules map to cortical layers. In Aim 2, we will determine how information processing is divided into stages as signals are passed from visual decision areas to motor areas. This second Aim will employ a more complex visually-guided joystick task, together with optogenetic inhibition of specific pathways, calcium imaging, and retrograde tracing. This will allow us to compare activity in the neurons that connect areas with those that are engaged only in local processing. Finally, in Aim 3, we will record from identified projection neurons and apply powerful new machine learning techniques to test two competing theories of how the brain produces consistent outputs: whether the brain suppresses neural “noise” in the output neurons themselves, or suppresses only task-relevant noise according to a more population- orien...