Computational functions of precisely balanced neuronal assemblies in an olfactory memory network

Abstract

ABSTRACTStructured connectivity in the brain organizes information by constraining neuronal dynamics. Theoretical models predict that memories are represented by balanced assemblies of excitatory and inhibitory neurons, but the existence and functions of such EI assemblies are difficult to explore. We addressed these issues in telencephalic area Dp of adult zebrafish, the homolog of piriform cortex, using computational modeling, population activity measurements, and optogenetic perturbations. Modeling revealed that precise balance of EI assemblies is important to prevent not only excessive firing rates (“runaway activity”) but also the stochastic occurrence of high pattern correlations (“runaway correlations”). Consistent with model-derived predictions, runaway correlations emerged in Dp when synaptic balance was perturbed by optogenetic manipulations of fast-spiking feedback interneurons. Moreover, runaway correlations were driven by sparse subsets of strongly active neurons, rather than by a general broadening of tuning curves. These results reveal novel computational functions of EI assemblies in an autoassociative olfactory memory network and support the hypothesis that EI assemblies organize information on continuous representational manifolds rather than discrete attractor landscapes.