The role of top-down dendritic processing in credit assignment and cortical dynamics

Top-down regulation of cortical processing is critical for learning, attention, figure-ground separation, multisensory integration, contextual-modulation and many other processes associated with cognition. However, the biological mechanisms supporting top-down computation remain elusive. A popular and compelling hypothesis is that top-down computation is implemented via the engagement of apical tuft dendrites in layer 1. This hypothesis is indirectly supported by convergent anatomical and functional evidence, including observations in patients with disorders associated with disrupted top-down processing, but more direct evidence linking dendritic integration to top-down mechanisms is limited. To address this gap, we developed a novel imaging approach that allows simultaneous recording of somas and dendrites in large populations of neurons during learning, without signal contamination. We will combine this approach with two complementary behavioral tasks: a highly controlled Brain-Computer Interface (BCI) paradigm and a comparatively naturalistic virtual navigation task. Through the set of proposed experiments, we aim to test the overarching hypothesis that top-down computation is implemented via the engagement of apical tuft dendrites in layer 1, and that these signals are responsible for guiding learning in networks of neurons. We will do this by: (1) Establishing the relationship between single-neuron dendritic integration and circuit dynamics and studying the principles governing the changes in this relationship over the course of learning; (2) Interrogating the relationship between dendritic activity and behavioral variables, how this relationship is modified over the course of learning, and how dendrites instruct changes in their corresponding somas; and (3) Testing the hypothesis that dendrites receive vectorized error signals consistent with an efficient solution to the credit assignment problem. Results from these experiments will catalyze a new ways of thinking about cortical computation and learning principles in biological systems, propelling the field into new directions with impactful scientific and translational potential.