A computational and experimental study of the Drosophila larval locomotor system and its developmental critical period

Critical periods of nervous system development are phases of heightened plasticity during which neuronal network dynamics is adjusted through activity-regulated plasticity mechanisms. Because plasticity is elevated, these phases are also vulnerable, such that disturbances during this highly plastic phase, but not outside, can lead to lasting maladjustment. In humans, several neurodevelopmental conditions, including forms of epilepsy and schizophrenia, have been linked to suboptimal critical periods. How networks adjust during a critical period, or how they are changed following critical period perturbations, is not well understood. This is partly because of the complexity of the commonly used mammalian model systems. To gain new understanding of critical period-regulated properties of the network, I have used a much simpler experimental model with a clearly defined critical period, the locomotor network of the Drosophila larva. This experimental system presents a unique opportunity to study the role of critical period experience in functional circuit formation in an experimentally accessible and genetically tractable model organism. The aim of the presented thesis is to further our understanding of the central pattern generator underlying Drosophila larval locomotor network and how it is affected by transient manipulations present during its developmental critical period. First, I developed a computational model of the central pattern generator employing modified, interconnected Morris-Lecar units. This improves on existing computational models of this system, which failed to explain key properties revealed by nerve cord truncation experiments. By modelling segment-specific differences in oscillator dynamics, of posterior segments acting as pacemakers that generate rhythmic activity even in absence of input, and anterior segments as followers, I was able to simulate the experimental outcomes in silico: namely, fictive locomotion activity is generated in isolated, truncated nerve cords, but only when only posterior segments are included. Next, I used this model to explore the control points of this system. Initially, I focused on the frequency of repeating activity propagation along the network in presence and absence of sensory feedback. Specifically, experimental findings show that in isolated nerve cords (i.e. deprived of sensory feedback) the speed of activity propagation is dramatically reduced, as compared to in intact animals. Through in silico modelling I identified three components that are necessary for the production of these higher frequency patterns found in intact animals: a long range head-to-tail connection, sensory-derived inhibition of the next posterior segment, and additional excitation (from sensory feedback or descending drives). I then identified a potential pathway in the existing connectome for this model predicted sensory inhibition to the next posterior segment. Using the model I mapped out how cellular and synaptic parameters affect measurable properties of the network output, such as the frequency and duty cycle of activity waves, or the phase delay and duty cycle of segments. These findings should help explain different experimental phenotypes of locomotor network output. Aiming to put this to the test, I analysed functional imaging data recorded from isolated nervous systems (data collected by collaborators). In controls I identified that phase delay between neighbouring segments is a non-monotonic function of position along the body axis, suggesting segmental heterogeneity. Unexpectedly, in nervous systems of animals exposed to different manipulations during their critical periods showed little or no changes in network output. In intact animals, in contrast, many of these critical period manipulations do lead to changes in crawling behaviour. This demonstrates the importance of sensory feedback for the manifestation of those phenotypes. Lastly, I studied network output in intact animals. To be able to take long functional recordings of segmental muscle contractions in freely crawling larvae, I built a fluorescence tracking stereo microscope with a collaborator and I wrote code for tracking animals. Segmentation of video recordings and labelling of segments is usually done manually and is therefore a time consuming bottleneck. For the segmentation problem I trained a custom neural network using Cellpose. For labelling the segments I created a custom algorithm. Together, these two provide an automated solution for the analysis of larval crawling. Analysis of data from intact animals revealed that, similarly to fictive locomotion activity waves, the phase delay between neighbouring segments is a non-monotonic function of position along the body axis. Furthermore, I found that the average size of larval hemisegments during crawling also shows a similar spatially non-monotonic heterogeneity between segments. Finally I looked at the effects of transient embryonic heat stress, a critical period manipulation of the larval locomotor system, on late larval crawling behaviour and found that larvae crawl slower via both the elongation of the wave progression phase and the inter wave phase. I also found that the peristaltic wave progression did not slow down in each segment: wave progression speed decreased in posterior and anterior segments but it did not significantly change in central abdominal segments.