Learning as a window into how internal states influence decision-making

Summary/Abstract, Project 4 This proposal’s overarching goal is to understand how internal states influence decisions and to identify the underlying neural mechanisms. The goal of this project is to conceptualize the change in internal state that explains why experts respond differently from novices in the same situation. Although learning is well known to produce long-lasting changes in diverse brain structures, many questions remain about the population-level changes within single areas and the changes in communication among areas. This project will address these questions by studying learning on two timescales: the long-term learning that defines animals’ gradual mastery of a perceptual decision-making task, and the within-session learning that animals undergo when reward statistics dynamically change. We will leverage a standardized behavioral task that has already been developed by the International Brain Laboratory. In the task, animals judge the spatial location of a visual grating and report it with a movement. First we will study behavior and neural activity as animals learn the basic version of this task in which left and right choices are rewarded with equal probability. The interpretation of this data will be informed by behavioral analyses from Core D that will characterize trial-by-trial engaged versus disengaged internal states, a balance that will likely change over learning. We will also track behavioral changes over learning using detailed video analysis. With these tools in hand, we will survey the cortex broadly using widefield imaging over many training sessions as animals transition from novice to expert status. We will use this large dataset to identify the areas that undergo the largest changes during learning and target them for simultaneous neural population recordings using Neuropixels probes. These measurements will allow us to determine how communication among areas changes in cortical and subcortical structures and how these changes parallel individual learning rates. This analysis will follow a similar approach from Project 1: we will measure the magnitude and orientation of a communication subspace that defines which single-trial fluctuations are communicated among areas. As we change the stimulus statistics over uncued task blocks, we will also study within-session learning of experts. Finally, we will use three mouse models of autism with distinct genetic and molecular anomalies that affect behavioral flexibility and specifically the ability to adapt to blocks of trials in which left and right stimuli are presented with unequal probability. We will establish which neural activity patterns are common among these models and different from controls and thus identify the brain states that animals require to adjust their decision- making priors as circumstances change. Taken together, these experiments will test the proposal’s overall hypothesis, that each internal state change is associated with a specific pattern of neural activity and communication within and across neural structures, on multiple timescales, from a few minutes to many weeks. In this way, we expect to extract general principles for how internal states govern information flow through the brain and, ultimately, decisions about what to do next.