Project Summary/Abstract: Overall
The overarching goal of this U19 program is to determine how neural computations across brain regions
produce two core cognitive processes, working memory and decision-making, and thus to derive fundamental
principles of brain function. This renewal application proposes to pursue powerful new themes that emerged
from our previous work and to broaden our scope substantially. To do so, the eight PIs plan a tightly integrated
set of experimental and computational studies of mice doing the accumulating towers task—in which they must
remember how many towers flash on each side as they run down a maze in virtual reality—and related tasks.
The first theme arises from the finding that neurons across the brain encode task variables and are
necessary for task performance. Almost all of these areas exhibit sequential activity, in which neurons are
active at different times in the task and, together, tile the trial duration. Project 1 will identify the task features
that drive sequential activity, use cooling to identify neural circuits that generate sequential activity, and
elucidate its anatomical basis by combining transmission electron microscopy and computational modeling.
A second theme is that manifold inference methods, applied to large-scale hippocampal recordings in
our task, reveal the geometry of a joint neural representation for an external variable (position) and an internal,
cognitive variable (accumulated evidence). Project 2 will extend our work on the geometry of neural
representations to other brain regions. We will examine how geometries and representations in these regions
interact with each other, and we will develop models to explain how the observed neural manifolds arise.
A third theme, fueled by our development of statistical methods to infer internal brain states, is that
animals’ brains occupy qualitatively different states from trial to trial during the same task block. Our data
suggests that behavior in each state requires different neural structures and circuits—in the same animal and
the same trial block. Project 3 will use multi-region recordings and perturbations to investigate whether states
are local to subcircuits versus global across the brain, extend our behavioral inference methods to neural data,
and examine to what extent our inferred states are linked to internal states of arousal, thirst, and hunger.
Elucidating how multiple circuits performing local computations combine into a brain in action is the
goal of Projects 4 and 5. Project 4 will probe functional interactions in multi-region recordings, including very
large-scale simultaneous electrophysiological recordings with next-generation silicon probes; and through
targeted experiments will test two hypotheses of how subcortical regions interact with neocortex.
Project 5 will generate a set of mechanistic models that instantiate specific hypothesized roles of
different brain regions. These local models will be combined into a single multi-regional model, informed by
data from all projects, that will enable us to dissect the roles of individual regions and their interactions in
performance of the many variants of our decision-making task.