Maintaining effective sensory coding in the face of inter-neuronal variation

How can neurons have consistent properties to allow effective sensory coding, in the face of developmental noise and inherent inter-neuronal variability? This problem occurs across species, and we address it in Drosophila, where ~2000 neurons called Kenyon cells encode olfactory associative memories. To accurately distinguish learned associations for different odours, Kenyon cell population responses to odours must be decorrelated, i.e. different odours activate non-overlapping subsets of Kenyon cells. Inter-odour decorrelation requires Kenyon cells to be roughly equally excitable: if some Kenyon cells are more excitable than others, these same cells tend to dominate all odour responses, which increases overlap between odour representations. Yet recent work shows that Kenyon cells receive extremely variable amounts of excitatory input. Our computational models suggest that this variability impairs odour decorrelation unless Kenyon cells compensate for variability along one parameter (e.g., amount of excitatory input) with counteracting variability along another parameter (e.g., spiking threshold). We will test whether such compensatory variability occurs and computationally model how it would affect circuit function.