Uncovering neural architectures enabling robust and flexible motor control in Drosophila

Uncovering the neural architectures that enable robust and flexible motor control is a fundamental challenge in neuroscience. Understanding the neural circuitry organisation at a cellular level is crucial for advancing our knowledge and developing interventions for motor disorders. This thesis addresses this challenge using Drosophila melanogaster as a model organism, leveraging the relative simplicity of their nervous systemas well as the wealth of tools and datasets available. This work focuses on the descending neurons (DNs), which act as an information bottleneck between the brain and the ventral nerve cord (VNC)â the invertebrate equivalent of the spinal cord.

Far from being simple relays, DNs form a highly interconnected and structured processing layer essential for coordinating movements. We demonstrate that "command-like" DNs, previously thought to independently trigger behaviours, actively recruit networks of other DNs. Connectome analysis and functional recordings reveal that this recruitment is necessary to compose complex behaviours such as forward walking. These DN networks are organized into behaviour-specific clusters interconnected by predominantly inhibitory synapses, supporting a mechanism for action selection through mutual inhibition.

We identify inhibition as a core element of the network organisation in the VNC as well. Investigating the integration of descending signals within the VNC, we find that the circuitry is modular and relatively shallow. We hypothesise that the shallow nature of the VNC circuitâ where motor neurons are often just one interneuron away from DN inputsâ facilitates the direct influence of descending commands on movement onset while local VNC dynamics handle temporal patterning. As an example, we demonstrate that the moonwalker descending neurons, sufficient and necessary for backward walking, target leg-specific circuits within the VNC neuropils. Analysis of hind leg motor neurons reveals motor synergies, putatively controlled by inhibitory premotor interneurons acting as hubs. This suggests a model where targeted inhibition at the premotor level contributes to selecting and coordinating muscle groups for specific movements.

Overall, this research reveals key organizational principles of the Drosophila motor system: modularity across hierarchical levels ensures behavioural robustness and distributed control. A hierarchical architecture, potentially exhibiting scale-free network properties, enables the flexible composition of complex behaviours from simpler "motor units" or "atoms of movement" controlled by individual or small groups of DNs. This is facilitated by a shallow VNC structure that allows efficient translation of descending commands into motor patterns. These findings provide an updated framework for understanding how biological neural networks generate diverse and adaptive behaviours, offering insights applicable to the design of more distributed artificial control systems.