Tools for connectomic reconstruction and analysis of a female Drosophila ventral nerve cord

Anthony W. Azevedo, Ellen Lesser, Jasper S. Phelps, Brandon Mark, Leila Elabbady, Sumiya Kuroda, Anne Sustar, Anthony Moussa, Avinash Kandelwal, Chris J. Dallmann, Sweta Agrawal, Su-Yee J. Lee, Brandon Pratt, Andrew Cook, Kyobi Skutt-Kakaria, Stephan Gerhard, Ran Lu, Nico Kemnitz, Kisuk Lee, Akhilesh Halageri, Manuel Castro, Dodam Ih, Jay Gager, Marwan Tammam, Sven Dorkenwald, Forrest Collman, Casey M Schneider-Mizell, Derrick Brittain, Chris S. Jordan, Michael H. Dickinson, Alexandra Pacureanu, H. Sebastian Seung, Thomas Macrina, Wei-Chung Allen Lee, John C TuthillbioRxiv 2022 · DOI

TL;DR

Researchers have developed automated tools to reconstruct the neural wiring diagram (connectome) of a female fruit fly's ventral nerve cord (VNC), a crucial step for understanding how flies control movement. By integrating this connectome with a detailed map of motor neuron connections to muscles, they've identified circuits that coordinate leg and wing movements during escape behaviors.

What Problem Does This Paper Address?

Understanding how the nervous system controls movement is a fundamental challenge in neuroscience. While much is known about motor neurons that directly control muscles, the intermediate circuits that integrate sensory information and descending commands remain less understood, especially in complex systems like the vertebrate spinal cord. The fruit fly, Drosophila melanogaster, offers a tractable model for studying these circuits due to its relatively compact nervous system. However, a major hurdle has been the inability to fully reconstruct the neural circuits and link them to their peripheral targets (muscles). This paper addresses this by providing a detailed connectome of the Drosophila ventral nerve cord (VNC) and, crucially, mapping the specific muscles innervated by its motor neurons.

Methods

The authors employed a suite of advanced computational and imaging techniques:

  • Automated Reconstruction: Deep learning-based approaches, specifically convolutional neural networks (CNNs), were used for automated neuron segmentation (identifying individual neurons) and synapse prediction (locating connections between neurons) within a large electron microscopy (EM) volume of the Drosophila VNC.
  • EM Volume Alignment: Self-supervised CNNs were used to refine the alignment of serial EM sections, a critical step for accurate reconstruction.
  • Cell Type Annotation: A combination of automated nucleus detection and manual proofreading was used to identify and classify neuronal and glial cells.
  • Connectome Visualization and Interaction: Tools like Neuroglancer and the Connectome Annotation Versioning Engine (CAVE) were utilized for visualizing, proofreading, and querying the reconstructed connectome.
  • Peripheral Target Mapping: To link neural circuits to behavior, the authors integrated multiple datasets:
  • X-ray Nanotomography (XNH): Used to visualize the fine anatomical details of the fly's leg and wing muscles.
  • Genetic Driver Lines: Sparse labeling of specific neurons using genetic tools allowed for tracing their morphology and identifying their targets.
  • Light Microscopy: Used to visualize the expression patterns of genetic driver lines.
  • EM Reconstruction: The detailed wiring diagram from the EM volume was compared with morphology from other methods.

Datasets

  • Electron Microscopy (EM) Volume: A serial-section EM dataset of the Drosophila Female Adult Nerve Cord (FANC) was used for reconstruction. This dataset is publicly available.
  • X-ray Nanotomography (XNH) Dataset: Used for high-resolution anatomical imaging of the fly's front leg and wing.
  • Genetic Driver Lines: A collection of Drosophila Gal4 lines from the Janelia database were used for sparse neuronal labeling.
  • Literature Data: Existing morphological descriptions and connectivity data from previous studies were leveraged for identification and validation.

Key Insights and Ideas

  • Automated tools revolutionize connectomics: The paper demonstrates that deep learning-based tools can achieve segmentation and synapse prediction accuracy comparable to manual tracing, dramatically accelerating the process of connectome reconstruction.
  • Bridging the gap between connectome and body: A major contribution is the creation of a comprehensive atlas mapping motor neuron outputs (which muscles they innervate), a critical step for interpreting the functional significance of the connectome.
  • Escape circuits revealed: By combining connectomic data with muscle targets, the authors identified specific neural circuits that coordinate leg and wing movements during escape takeoff, suggesting a mechanism for rapid, coordinated action.
  • Subverting recruitment hierarchies: The study suggests that descending signals from the giant fiber (a key escape pathway) can bypass the typical hierarchical recruitment of motor neurons, allowing for faster, high-force movements necessary for escape.
  • Polyneuronal innervation enhances flexibility: The finding of polyneuronal innervation in some muscles suggests a mechanism for increased motor flexibility and precision in insect limbs.

Key Results

  • Automated reconstruction accuracy: Proofreading automated reconstructions of leg motor neurons required significantly less time (e.g., 2 hours vs. 200 hours for manual tracing of a single neuron). The automated methods captured >80% of the dendritic cable and total cable for individual neurons.
  • Synapse prediction precision: Precision and recall for predicting presynaptic partners with more than three synapses were consistently above 80%.
  • MN-to-muscle mapping: A comprehensive atlas was created, identifying the specific muscle targets for 69 left T1 leg motor neurons and 29 wing motor neurons.
  • Escape circuit identification: The study identified premotor neurons (GFC2 and GFC4) that synapse onto specific leg and wing motor neurons, forming a circuit that likely coordinates escape behaviors.
  • Synapse counts: The fast tibia extensor MN (FETi) received the most input synapses (14,904) among front leg MNs.

Limitations and Future Work

  • Proofreading is still necessary: While automated tools accelerate the process, manual proofreading remains essential for ensuring accuracy, especially for fine dendritic branches and in synapse-dense regions.
  • Incomplete connectome: At the time of publication, less than half of the FANC dataset had been proofread, indicating that the full connectome is still under development.
  • Chemical vs. Electrical Synapses: The current EM reconstruction primarily reveals chemical synapses. The paper notes the inability to visualize electrical synapses (gap junctions) or sites of neuromodulation, which are important for sensorimotor processing. Future work could involve higher-resolution imaging to resolve gap junctions.
  • Sex differences: The study focused on a female VNC connectome, complementing ongoing work on a male VNC. Future research can explore sex-specific differences in neural circuits.
  • Functional validation: While the study identifies circuits and predicts their function, further experimental validation (e.g., optogenetics, electrophysiology) will be needed to confirm these hypotheses.
  • Comparative connectomics: The authors suggest that comparing this connectome with others (e.g., larval Drosophila, other species) will be crucial for understanding the evolution of nervous systems.

Summary updated

Abstract

Summary Like the vertebrate spinal cord, the insect ventral nerve cord (VNC) mediates limb sensation and motor control. Here, we apply automated tools for electron microscopy volume alignment, neuron segmentation, and synapse prediction toward creating a connectome of an adult female Drosophila VNC. To interpret a connectome, it is crucial to know its relationship with the rest of the body. We therefore mapped the muscle targets of leg and wing motor neurons in the connectome by comparing their morphology to genetic driver lines, dye fills, and X-ray nano-tomography of the fly leg and wing. Knowing the outputs of the connectome allowed us to identify neural circuits that coordinate the wings and legs during escape takeoff. We provide the reconstruction of VNC circuits and motor neuron atlas, along with tools for programmatic and interactive access, as community resources to support experimental and theoretical studies of how the fly nervous system controls behavior.
BibTeX
@misc{azevedo2022tools,
  author = {Anthony W. Azevedo and Ellen Lesser and Jasper S. Phelps and Brandon Mark and Leila Elabbady and Sumiya Kuroda and Anne Sustar and Anthony Moussa and Avinash Kandelwal and Chris J. Dallmann and Sweta Agrawal and Su-Yee J. Lee and Brandon Pratt and Andrew Cook and Kyobi Skutt-Kakaria and Stephan Gerhard and Ran Lu and Nico Kemnitz and Kisuk Lee and Akhilesh Halageri and Manuel Castro and Dodam Ih and Jay Gager and Marwan Tammam and Sven Dorkenwald and Forrest Collman and Casey M Schneider-Mizell and Derrick Brittain and Chris S. Jordan and Michael H. Dickinson and Alexandra Pacureanu and H. Sebastian Seung and Thomas Macrina and Wei-Chung Allen Lee and John C Tuthill},
  title = {{Tools for connectomic reconstruction and analysis of a female <i>Drosophila</i> ventral nerve cord}},
  year = {2022},
  doi = {10.1101/2022.12.15.520299},
  howpublished = {bioRxiv},
}
Figure