A central goal of systems neuroscience is to relate behavior to its underlying circuit dynamics. This task is complicated by the complex and circuitous paths along which information flows as it is encoded and processed in the many steps between sensory inputs and motor outputs. Currently, we understand little regarding the organization and dynamics of interactions between brain areas. For example, we do not know the degree to which specific brain areas have separate representations versus when information is encoded jointly across brain areas. Moreover, it is unclear how the joint state of multi-brain area activity relates to behavior. For instance, when animals are trained to discriminate sensory stimuli we know that neural activity differs across multiple areas in trials when an animal makes a behavioral error, but it is not clear whether all the relevant brain areas have such error-related signals on a single error trial or perhaps if the erroneous pattern occurs only in a subset of brain areas, which is then sufficient to generate a behavioral mistake. Recent experimental and theoretical advances have rendered these questions tractable. On the experimental front, new recording and perturbation technologies allow simultaneous monitoring of population activity in multiple brain areas as animals perform a behavioral task with the ability to interfere with activity in specific regions while observing the effect of these perturbed dynamics on other brain areas. From a theoretical perspective, we have begun to make progress distilling dynamic, heterogenous population activity recordings into a quantification of time-varying interactions between brain areas. Here, we will make use of three high quality datasets of multiple brain area recordings and perturbations collected by our experimental collaborators to design analytical approaches to interpret this data. We will first develop statistical metrics of the coordination of activity between brain areas and relate it to behavior. We will then design neural network based dimensionality reduction approaches that capture complex data in more interpretable form and also denoise the data by assimilating and enforcing consistent population dynamics. Finally, we will devise and adapt modeling approaches that turn the statistical descriptors of population activity into mechanistic models of how circuit structure supports these dynamics. In summary, this proposal will reveal a quantitative description of multi-brain area encoding and communication between brain areas as well as lay down testable, predictive hypotheses for the circuit structures that generate these dynamics. Beyond the basic science importance of these questions, the approaches and tools we will develop and disseminate will be broadly applicable across many paradigms in neuroscience and beyond.
Sensory information passes through multiple brain areas as it spreads through the brain, interacts with current information processing, affects decision making circuits, and finally leads to behavior. We currently lack both a holistic view of this process and a detailed understanding of the nature of how multiple brain areas interact. Here, we propose to develop analytical techniques that will facilitate interpreting multi-brain area recording and perturbation datasets, which will reveal how brain areas interface with each other and how these interactions shape behavior.
