Flexible foraging decisions by configuration of orthogonal integrator dynamics

Abstract

The capacity to adapt behavior and cognition across contexts is fundamental to intelligence. Context-dependent decision making is an exemplar of flexible cognition in animals, yet its biological basis remains poorly understood. At the neural level, a critical mechanistic question is whether context largely alters the inputs to a common decision process or reconfigures decision activity itself to create context-specific functional modes. To study this problem, we developed a patch foraging task in which mice forage in two environment contexts defined by distinct reward dynamics. Mice made patch leaving decisions across environments using different context-specific parameterizations of a drift-diffusion integrator process. Using high-density acute and chronic neural recordings, we find that foraging environments recruit separate activity subspaces which organize orthogonal coding of context-specific decision variables. Unlike the context-invariant decision coding used for perceptual choice problems, foraging decision variables were encoded in orthogonal population vectors within largely separate neural subpopulations - a subspace configuration mechanism which can support modular learning, independent readout, and flexible toggling of decision strategies across environments. In the dorsal frontal cortex, orthogonal integrators emerge with experience, are preserved across three distinct task settings, and are required for rapid and volitional strategy switching. Our findings establish orthogonal integrator dynamics as a general solution for context-dependent decision making and open a new avenue to investigate the cell- and circuit-level basis of flexible cognition in the mammalian brain.