We have established important methods including techniques for volumetric electron microscopy (EM), methods for dense reconstructions of neuronal circuits from 3D EM image data, and an improved virtual reality (VR) system for adult zebrafish. In WP1, we trained adult zebrafish in an odor discrimination paradigm and analyzed effects on the dynamics of neuronal population activity as planned. These results identified specific, learning-related changes in odor representations, some of which are not consistent with classical memory models. In particular, we found that telencephalic area Dp, the zebrafish homolog of piriform cortex, operates in a state of “precise synaptic balance” that does not support attractor dynamics as predicted by classical memory models. We thus developed data-driven computational models that extend previous models and account for the complex dynamics observed experimentally. Moreover, we found that Dp contains a joint map of odor identity and valence that is updated by experience. In WP2, we acquired two large volumetric EM datasets covering Dp and parts of adjacent forebrain areas from adult zebrafish that were trained in an odor discrimination task. In both fish, odor-evoked activity was recorded from >1000 neurons in Dp prior to volumetric EM imaging. The reconstruction of synaptically connected networks is ongoing. In one of the datasets, the reconstruction of odor-responsive neurons is almost complete. In parallel, we created network simulations constrained by data obtained in WP1 that uncovered new functions of inhibition in memory networks. Moreover, consistent with experimental observations in WP1, the results revealed that Dp-like memory networks do not exhibit discrete attractor dynamics but store information by constraining activity to local activity manifolds. In WP3, we analyzed inhibitory microcircuits and discovered a specific and instructive role of inhibition in the mapping of odor representations onto a low-dimensional representation of valence during associative learning. These results support the emerging concept that inhibition balances excitation with high precision and makes critical contributions to manifold-based representations in memory networks. In addition, we characterized telencephalic cell types and neuromodulatory systems based on gene expression (transcriptomics), and discovered cognitive maps of external environments in the zebrafish telencephalon by imaging of neuronal activity during behavior in a virtual reality.
