Coordinating structure and function for neuronal computations mediating context-dependent behavior - Project Summary
This proposal explores an emergent computational framework for understanding the neural population codes
that support flexible, context-dependent behavior. The current state of the field is based on two competing
views. According to the circuits view, fixed behaviors arise from specific anatomically or genetically defined cell
populations that serve specific functions. Alternatively, the network computation view instead holds that neural
activity provides mixed representations of task variables and can be understood only based on the joint
activation of many neurons. Currently, these competing views are pursued by different communities with
different tools, different behavioral paradigms and different model organisms. This has led to a disconnect
between neural computation and the underlying biological circuit mechanisms. Here we propose a unified
framework, in which the combinatorial activity of biologically-identified populations of neurons shapes the
computations through low dimensional dynamics. A new interdisciplinary team of investigators - Pesaran (NYU
- primate experimentalist), Johansen (RIKEN - rodent experimentalist) and Ostojic (Ecole Normale Superieure
- computational theory) will develop the computational theory and apply it to flexible input-output tasks in multiple
species - rats and non-human primates. To achieve these goals, we will analyze recurrent neural network models
trained to perform a sensory-motor context-dependent decision-making task and fit low-dimensional models to
experimental data. We will perform experiments in rodents and non-human primates to validate model predictions
that combinatorial coding can support context-dependent behavior and is behaviorally-significant (Aim 1). We will
analyze whether biologically-defined cell types map onto computationally-defined cell classes during context-
dependent behavior by determining how activity in combinations of genetically and anatomically identified PFC
cell types corresponds to low-dimensional dynamics (Aim 2). In parallel, we will determine how biologically-
defined cell types control context-dependent behavior during explicitly-cued and implicitly-signaled contexts by
performing optogenetic perturbations of anatomically-defined PFC cell types and testing behavioral performance
(Aim 3). Successful completion of these aims will link neural computation to biology across multiple species to
deliver a computational framework explaining how biological circuit mechanisms give rise to neuronal
computations that mediate context-dependent behavior.
