Neuronal encoding of natural imagery in dragonfly motion pathways

Abstract

Vision is the primary sense of humans and most other animals. While the act of seeing seems easy, the neuronal architectures that underlie this ability are some of the most complex of the brain. Insects represent an excellent model for investigating how vision operates as they often lead rich visual lives while possessing relatively simple brains. Among insects, aerial predators such as the dragonfly face additional survival tasks. Not only must aerial predators successfully navigate three-dimensional visual environments, they must also be able to identify and track their prey. This task is made even more difficult due to the complexity of visual scenes that contain detail on all scales of magnification, making the job of the predator particularly challenging. Here I investigate the physiology of neurons accessible through tracts in the third neuropil of the optic lobe of the dragonfly. It is at this stage of processing that the first evidence of both wide-field motion and object detection emerges. My research extends the current understanding of two main pathways in the dragonfly visual system, the wide-field motion pathway and target-tracking pathway. While wide-field motion pathways have been studied in numerous insects, until now the dragonfly wide-field motion pathway remains unstudied. Investigation of this pathway has revealed properties, novel among insects, specifically the purely optical adaptation to motion at both high and low velocities through motion adaptation. Here I characterise these newly described neurons and investigate their adaptation properties. The dragonfly target-tracking pathway has been studied extensively, but most research has focussed on classical stimuli such as gratings and small black objects moving on white monitors. Here I extend previous research, which characterised the behaviour of target tracking neurons in cluttered environments, developing a paradigm to allow numerous properties of targets to be changed while still measuring tracking performance. I show that dragonfly neurons interact with clutter through the previously discovered selective attention system, treating cluttered scenes as collections of target-like features. I further show that this system uses the direction and speed of the target and background as one of the key parameters for tracking success. I also elucidate some additional properties of selective attention including the capacity to select for inhibitory targets or weakly salient features in preference to strongly excitatory ones. In collaboration with colleagues, I have also performed some limited modelling to demonstrate that a selective attention model, which includes switching best explains experimental data. Finally, I explore a mathematical model called divisive normalisation which may partially explain how neurons with large receptive fields can be used to re-establish target position information (lost in a position invariant system) through relatively simple integrations of multiple large receptive field neurons. In summary, my thesis provides a broad investigation into several questions about how dragonflies can function in natural environments. More broadly, my thesis addresses general questions about vision and how complicated visual tasks can be solved via clever strategies employed in neuronal systems and their modelled equivalents.