Connectomic Modeling of Orientation Selectivity in the Drosophila Optic Lobe

Abstract

The completion of the Drosophila melanogaster connectome through the FlyWire project has presented exciting possibilities for predicting the functional behavior of a neural system purely from its anatomy. This thesis explores whether the wiring pattern of a small Drosophila optic lobe circuit, the Dm3-TmY circuit, contains enough information to perform the orientation selective computations predicted by Seung. I simulate a rate-based model based on a point neuron architecture with connectivity and weights derived directly from the FlyWire connectome. The network receives input via Tm1 feedforward input that includes both difference of Gaussians spatial filtering and bandpass temporal processing. In order to validate my model, I repeat the stimulus protocol used in an in vivo calcium imaging experiment done by Heng Wu at the Clark Lab. The model displays clear orientation selectivity in response to a drifting sinusoidal grating stimulus for all six cell types within the circuit, with a preferred orientation spacing of approximately 30° to cover a full 180° range in orientation sensitivity, in agreement with Seung's predictions. The Dm3 tuning curves produced by my model closely resemble those generated by Heng Wu's experimental data. Through selective ablation and amplification experiments, I probe the effect of individual type-to-type connections, and determine how feedforward receptive fields, cross orientation inhibition, and Dm3-Dm3 lateral competition each contribute to the circuit's selectivity. Through these explorations, it is found that tuning of orientation in this circuit does not result from a singular source, but rather from the combination of feedforward geometric filters and recurrent inhibition, including the surprising discovery that Dm3-Dm3 competition enhances tuning in TmY cells through a disinhibition pathway.