Harvard Medical School, Neurobiology
Yvette is a postdoc at Harvard in Rachel Wilson’s lab, and has solved the problem of how the visual scene maps on to head direction cells (compass neurons) in the fly central complex, creating a sense of direction (see her manuscript that came out last week in Nature).
We maintain our sense of direction in the dark by keeping track of our own movements, but when visual landmarks are available, our sense of direction is even better. Moreover, we can learn new landmarks in new environments. What neural mechanisms reconcile self-movement information with ever-changing landmarks to generate a coherent sense of direction? In the Drosophila brain, heading direction neurons form an attractor network whose activity tracks the angular position of the fly using both self-movement and visual inputs. These heading neurons receive visual landmark information from a population of GABAergic neurons, called R-neurons, whose receptive fields’ tile visual space. Using whole-cell recordings and calcium imaging from heading neurons, we show that each heading neuron is inhibited by visual cues in specific azimuthal positions, with different visual maps in different individuals. Inhibition arises from R-neuron axons that form an all-to-all matrix of potential connections onto heading neurons. We show that matrix weights can reorganize over minutes when visual-movement correlations change in virtual reality. This reorganization causes persistent changes in the reference frame of the heading representation and can depress or potentiate visually-evoked inhibition in a manner that depends on visual-heading correlations. We propose that rapid associative synaptic plasticity between R-neurons and heading neurons keeps the heading representation aligned with the external world. Computational models of grid cells and head direction networks have proposed that, by combining associative plasticity of sensory inputs with attractor dynamics that integrates self-motion, a network can establish a stable map of the world by iteratively incorporating information about consistent sensory cues. We propose that plastic synaptic connections between R-neurons and heading neurons serve as this flexible anchor for the Drosophila heading system. Associative plasticity of sensory inputs, combined with attractor network dynamics, should make neural heading maps self-consistent and progressively more accurate during exploration.