17. Modelling spatiotemporal interactions in direction selective V1 neurons

Pamela Baker pamela.baker@dpag.ox.ac.uk Wyeth Bair wyeth.bair@dpag.ox.ac.uk

Dept. of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK

The spatiotemporal interactions underlying direction selectivity in V1 neurons remain obscure. Recent work using a type of 2D white noise stimulus has revealed intriguing patterns in the resulting 2D sequential interaction maps from direction selective neurons in macaque V1 (Livingstone and Conway, J Neurophysiol, 89:2743-2759, 2003). Suppressive and facilitatory subregions in interaction maps may be elongated along the axis of preferred orientation of the neuron or symmetric and round, with no elongation. Elongated subregions were often asymmetric, with facilitatory interactions more extensive than suppressive ones. Elongation was correlated with direction tuning bandwidth, i.e. narrowly-tuned cells had more elongated interactions. We examined the shapes of 2D interaction maps and the associated direction tuning curves obtained with models of direction selective (DS) V1 cells. We tested how parameters that affect direction tuning width contribute to the shapes of 2D interaction maps.

We used two spatiotemporal filter-based models of DS V1 cells, a motion energy model and a Reichardt detector. The motion energy model generated interaction maps with symmetrical elongated facilitatory and suppressive subregions. Decreasing the window of integration of the spatial filter broadened direction tuning and eliminated subregion elongation in the maps. The Reichardt detector model also showed round, symmetric suppressive and facilitatory map subregions. Decreasing the distance between detectors narrowed direction tuning, but did not elongate map subregions. Neither of these models generated asymmetry in the extent of facilitatory vs. suppressive interactions. To further investigate the neural circuitry underlying the shapes of 2D interaction maps, we are developing a network model with spiking inhibitory and excitatory V1 neurons with realistic LGN inputs. This model will allow us to test how network connectivity patterns and non-linearities associated with biophysical mechanisms, including synaptic dynamics, dendritic integration and somatic spiking, shape the 2D interaction maps. This will shed light not only on the generation of experimentally observed map shapes, but also reveal how DS receptive fields are built. Support: The Wellcome Trust.