An analysis of an optomotor reflex using methods from control theory, cross-correlation analysis, and information theory
Miller, Clyde Steven, Jr
Glantz, Raymon M.
Doctor of Philosophy
This report describes a novel approach to the question of how spike trains encode information. Traditional methods rely on signal processing techniques based on point process theory and focus on the relationship between stimulus and spike train response. Broadly, these methodologies attempt to either: quantify changes in some set of spike train parameters on the assumption that this set expresses the neural code, or reconstruct the stimulus from the spike train. Both are attempts to read spikes directly. But only the biological decoder can actually read spikes. In this new approach, inferences about the code are made only after simultaneously observing both the spike trains and the responsiveness of the target neuron (i.e., the decoder) to the input spike train. The biological decoder in this study is a set of motoneurons that mediate a compensatory optomotor reflex after decoding visually-elicited spike trains from two classes of neurons that synapse onto them. The reflex consists of eyestalk counterrotation in response to apparent motion of the global visual field. The range of behaviorally relevant stimuli is established by characterizing the ratio of eyestalk movement to angular displacement of the stimulus (gain) as several stimulus dimensions are varied. These same stimuli are employed during electrophysiological studies to elicit spike trains under conditions which cause the behavior. Information theoretic calculations quantify the extent to which the behavior and the neuronal ensemble responses can distinguish stimuli and enable direct comparisons between the two types of response. Information theoretic comparisons show that an ensemble of 6 motoneurons can account for the measured behavior. These comparisons also indicate that other sources of visual input probably participate in the pathway. The motoneurons read a rate code from one class of neuronal inputs and an interval code from the other.