henhaobio
Sign in

NatureNeurosciencePeer-reviewed journal article

Connectome-driven neural inventory of a complete visual system

Aljoscha NerniD, Frank LoescheiD, Shin-ya TakemuraiD, Laura E. BurnettiD, Marisa DreheriD, Eyal GruntmaniD, Judith HoelleriD, Gary B. HuangiD, Michał JanuszewskiiD, Nathan C KlapoetkeiD, Sanna KoskelaiD, Kit D. LongdeniD, Zhiyuan LuiD, Stephan PreibischiD, Wei QiuiD, Edward M. RogersiD, Pavithraa SeenivasaniD, Arthur ZhaoiD, John BogoviciD, Brandon S CaninoiD, Jody ClementsiD, Michael CookiD, Samantha Finley-MayiD, Miriam A FlynniD, Imran Hameed, , Michael B. ReiseriD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Vision provides animals with detailed information about their surroundings, conveying diverse features such as color, form, and movement across the visual scene. Computing these parallel spatial features requires a large and diverse network of neurons, such that in animals as distant as flies and humans, visual regions comprise half the brains volume. These visual brain regions often reveal remarkable structure-function relationships, with neurons organized along spatial maps with shapes that directly relate to their roles in visual processing. To unravel the stunning diversity of a complex visual system, a careful mapping of the neural architecture matched to tools for targeted exploration of that circuitry is essential. Here, we report a new connectome of the right optic lobe from a male Drosophila central nervous system FIB-SEM volume and a comprehensive inventory of the flys visual neurons. We developed a computational framework to quantify the anatomy of visual neurons, establishing a basis for interpreting how their shapes relate to spatial vision. By integrating this analysis with connectivity information, neurotransmitter identity, and expert curation, we classified the [~]53,000 neurons into 727 types, about half of which are systematically described and named for the first time. Finally, we share an extensive collection of split-GAL4 lines matched to our neuron type catalog. Together, this comprehensive set of tools and data unlock new possibilities for systematic investigations of vision in Drosophila, a foundation for a deeper understanding of sensory processing.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (1 Jun 2024) · CC BY 4.0

A FIB-SEM connectome of the male Drosophila right optic lobe sorts ∼53,000 neurons into 727 types, with neurotransmitter predictions and 577 matched split-GAL4 lines.

Earlier systematic EM surveys of fly visual neurons all used female brains, and roughly half of optic lobe cell types had never been systematically characterized or named. This work provides the first detailed male optic lobe connectome, with a curated type inventory, quantitative anatomy based on columns and layers, transmitter assignments, and matched genetic drivers for targeted experiments.

Key results

  1. ShownIn the right optic lobe of one male fly, ∼53,000 reconstructed neurons were sorted into 727 types, roughly half of them newly characterized and named. Most connectivity sits in a small number of populous, repeating types, and output to the central brain is heavily compressed.

    ∼53k neurons in 727 types (770 instances). 60 types account for over half of all synaptic connections and ∼75% of neurons. There are ∼49 million connections in the visual regions. VPNs: ∼4,500 neurons of 350 types relay signals from nearly 50,000 local neurons. · Abstract; Neurons of the Drosophila visual system; Communication with the central brain; Summarized inventory of visual neurons and…

  2. ShownReconstruction quality is higher than in the hemibrain: a larger share of synapses have both partners on traced neurons, and synapse detection improved. The partial lamina is the clear exception and is much less complete.

    Connection completeness was 52.7% overall (54.2% without the partial lamina) vs 37.5% for the hemibrain. Synapse detection was scored on 14 random 300×300×300-voxel cubes (287 T-bars, 2425 postsynaptic partners). It was substantially better than the hemibrain and similar to MANC. · EM volume synapse identification; EM volume Proofreading; Tables

  3. ShownConnectivity strongly constrains cell identity. The five strongest partner types of each cell type separate nearly all types, though this does not reliably classify single neurons. Morphology vectors built from columns and layers sort most medulla neurons nearly as well as connectivity clustering.

    >99% of cell types can be told apart by their top five connections. The 15 columnar types sort perfectly into connectivity clusters. On the same 68 medulla intrinsic types (80 clusters in the connectivity run), cluster completeness was 0.88 for morphology vectors vs 0.9 for connectivity clustering. · Cell typing visual neurons; Quantified anatomy and connectivity; Connectivity-based cell clustering (related to…; Figure legends

  4. ShownA per-synapse EM transmitter classifier was pooled to the cell-type level and checked against new expression data. Its assignments show about half of types cholinergic and half GABAergic or glutamatergic. GABAergic neurons have lower synaptic fan-out.

    Trained on 59 types (nearly 2 million synapses) for seven transmitters. Type-level accuracy was tested on 79 types not used in training; new expression data (mostly EASI-FISH) covered 66 additional types. 14/16 TmY types had >85% of synapses called as one transmitter, and all six assayed TmY types agreed. GABA fan-out was ∼13% lower (p < 0.001 vs ACh and vs Glu; ACh vs Glu p ∼0.64, ANCOVA). · Neurotransmitters in the visual system; Figure legends

  5. ShownA split-GAL4 collection gives genetic access to many EM-defined types. Almost all types match a type in the female FlyWire-FAFB connectome, which fits with little sexual dimorphism at the cell-type level. A few unmatched types are candidate dimorphic neurons.

    577 split-GAL4 lines matched to >300 EM-defined types. 98% of inventory types (99.8% of cells) matched to FlyWire-FAFB. Unmatched optic lobe types Tm26 and LoVP92 resemble previously reported male-specific cells. · Cell-type-specific genetic driver lines; Figure legends

Methods, data and reproducibility

The whole male CNS was hot-knife cut into 20 µm slabs, imaged by focused ion beam scanning electron microscopy (FIB-SEM) at 8 nm, segmented with flood-filling networks, then proofread. Neurons were typed iteratively by combining expert morphology review, cosine/Ward connectivity clustering, column and layer coordinates, and a VGG-style transmitter classifier.

  • Male CNS FIB-SEM volume; proofread right optic lobe connectome served as neuPrint dataset optic-lobe:v1.0 · Drosophila (male) · n = ∼53,000 neurons in 727 types; 66 slabs (13 contain the right optic lobe) · neuPrint optic-lobe:v1.0
  • Presynapses from cell types with known transmitters, used to train the synapse transmitter classifier · Drosophila · n = 59 cell types; nearly 2 million synapses
  • New transmitter expression data (mainly EASI-FISH) used to validate the predictions · Drosophila · n = 66 additional cell types (79 types in the validation set)
  • Split-GAL4 driver lines imaged by light microscopy (population and MCFO labeling) and matched to EM types · Drosophila (mostly female flies imaged) · n = 577 lines matched to >300 cell types
  • FlyWire-FAFB female EM connectome, used to match cell types across datasets · Drosophila (female)
  • Densely annotated optic lobe subvolumes used as synapse detection ground truth · Drosophila (male) · n = 14 cubes; 287 T-bars, 2425 postsynaptic partners
  • Cell Type Explorer web resource with per-type connectivity and coverage pages (downloadable archive) · Drosophila (male) · n = 727 cell types · DOI:10.5281/zenodo.10891950
  • Hemibrain neuron skeletons and the JRC2018M/JRC2018U templates, used for LM-EM registration · Drosophila
Methods & tools
neuPrint (Neo4j) with neuprint-python, Flood-filling networks (FFN) segmentation, render and hot-knife alignment/stitching (saalfeldlab), NeuTu, Neu3, DVID proofreading toolchain, VGG-style deep neural network transmitter classifier, EASI-FISH, Split-GAL4 drivers with MCFO stochastic labeling, scipy.hierarchy and fastcluster (cosine distance, Ward linkage), Elastix and BigWarp (registration to JRC2018M), Blender and neuVid, snakemake, neuroglancer
Software & environment
Python 3.10 or newer. The repository lists dependencies (Pandas, NumPy, SciPy, Jupyter, Plotly, snakemake, Trimesh) and has setup guides for Linux, Windows and Mac. Analyses use neuPrint's default synapse confidence threshold of 0.5.
Assay conditions & compute
35 brain and 31 VNC slabs were imaged on seven customized FIB-SEM systems (Zeiss Capella FIB column) running in parallel for almost a year. Rendering for >700 cell types was spread across compute cluster nodes with snakemake.
Reproducibility gaps
Proofreading of the central brain and VNC is still ongoing, so VPN/VCN central synapses are missing from optic-lobe:v1.0. The lamina is only partly inside the volume. Some driver lines are not kept as combined stocks; for these, most AD/DBD hemidrivers are at Bloomington.
Limitations (6)
  • Only part of the lamina lies in the volume (trimmed slab edges), and heavily stained photoreceptor axons segmented poorly. As a result, an estimated 2,777 lamina cells and 459 R7/R8 photoreceptors are undercounted. · stated by the authors
  • Transmitter predictions are less reliable in two cases: over 100 VPN types with few presynapses in the volume, and ∼50 low-count types predicted as octopamine, dopamine or serotonin. Peptidergic signalling cannot yet be predicted from EM. · stated by the authors
  • Central brain proofreading was incomplete, so the central inputs and outputs of VPNs and VCNs are missing; those arbors are present without synapses. Connectivity clustering therefore used only connections within the right optic lobe. · stated by the authors
  • Connectivity differences did not define types on their own. Clustering with preset cluster numbers split or merged some types, so morphology, mosaics or driver lines were needed to decide. About 350 bodies remain labelled '_unclear', mostly incomplete R7/R8. · stated by the authors
  • All data come from one optic lobe of a single male. Variability between individuals is not measured. Sexual dimorphism is inferred from matching to a single female dataset, not from replicate animals of each sex. · noted in this summary
  • With connection completeness at 52.7%, many synapses lack an identified partner on one side. This could bias type-level connection weights and top-five partner summaries toward the more completely reconstructed partner types. · noted in this summary

Read this if you study Drosophila vision or connectomics and need a reference catalog of optic lobe cell types, with their connectivity and transmitters, or genetic drivers to target specific visual neurons.

Citations
98
FWCI
31.88
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISMatsliah '24Hoeller '26Kim '26Garner '24Tanaka '22Ganguly '24
83% semantic matchNeuroscience
Arie Matsliah, Szi-chieh Yu, Krzysztof Kruk+40 more·Nature·Oct 2, 2024· 85 citations