CANCELED - MBC Seminar Spring 2018

Marta Zlatic, Janelia Research Campus, Howard Hughes Medical Institute (HHMI), Group Leader

Dr. Zlatic’s presentation

Title: Multilayered recurrent architecture for computing teaching signals in the insect brain

Abstract: The ability to learn and update associations between stimuli and rewards and punishments is essential for survival in an ever-changing environment. Modulatory (e.g. dopaminergic) neurons convey teaching signals for associative memory formation and updates across the animal kingdom. Many theoretical models propose how the teaching signals are computed, but the actual circuit implementation of this computation is unknown. Despite the pivotal role that the modulatory neurons play in forming and updating memories, a synaptic-resolution connectivity map of the circuits that regulate their activity is lacking. We used electron microscopy reconstruction to comprehensively identify all neurons presynaptic to all modulatory neurons in an associative learning center, the mushroom body of Drosophila larva. We also characterized the way in which the modulatory neurons encode different punishment types, and mapped the afferent pathways all the way from nociceptive and mechanosensory neurons. We discovered previously unknown layers of the mushroom body network: many different types of feedback neurons that receive direct or indirect input from mushroom body output neurons (MBONs) and signal back to the modulatory neurons. Strikingly, many modulatory neurons received more than half of their total dendritic input from distinct feedback pathways. We confirmed the feedback pathways are functional and can influence memory formation. Our study provides a detailed view of brain circuits that compute teaching signals and reveals the multilayered and highly recurrent nature of these circuits. This architecture could support persistent activity during memory consolidation and memory transfer, comparisons between actual and expected outcomes, and more generally mediate adaptive reinforcement processing to ensure that what an animal learns depends on what an animal already knows.