bioRxivNeurosciencePreprint
A Connectome of the Adult Drosophila Central Brain
C. Shan XuiD, Michał JanuszewskiiD, Zhiyuan LuiD, Shin-ya TakemuraiD, Kenneth J. HayworthiD, Gary B. HuangiD, Kazunori ShinomiyaiD, Jeremy Maitin-ShepardiD, David AckermaniD, Stuart BergiD, Tim M. BlakelyiD, John BogoviciD, Jody ClementsiD, Tom Dolafi, Philip M. HubbardiD, Dagmar KainmuelleriD, William T. KatziD, Takashi KawaseiD, Khaled A. KhairyiD, Laramie LeavittiD, Peter H. LiiD, Larry F. Lindsey, Nicole L. NeubarthiD, Donald J. OlbrisiD, Hideo OtsunaiD, , Ian A. MeinertzhageniD
Abstract
The neural circuits responsible for behavior remain largely unknown. Previous efforts have reconstructed the complete circuits of small animals, with hundreds of neurons, and selected circuits for larger animals. Here we (the FlyEM project at Janelia and collaborators at Google) summarize new methods and present the complete circuitry of a large fraction of the brain of a much more complex animal, the fruit fly Drosophila melanogaster. Improved methods include new procedures to prepare, image, align, segment, find synapses, and proofread such large data sets; new methods that define cell types based on connectivity in addition to morphology; and new methods to simplify access to a large and evolving data set. From the resulting data we derive a better definition of computational compartments and their connections; an exhaustive atlas of cell examples and types, many of them novel; detailed circuits for most of the central brain; and exploration of the statistics and structure of different brain compartments, and the brain as a whole. We make the data public, with a web site and resources specifically designed to make it easy to explore, for all levels of expertise from the expert to the merely curious. The public availability of these data, and the simplified means to access it, dramatically reduces the effort needed to answer typical circuit questions, such as the identity of upstream and downstream neural partners, the circuitry of brain regions, and to link the neurons defined by our analysis with genetic reagents that can be used to study their functions. Note: In the next few weeks, we will release a series of papers with more involved discussions. One paper will detail the hemibrain reconstruction with more extensive analysis and interpretation made possible by this dense connectome. Another paper will explore the central complex, a brain region involved in navigation, motor control, and sleep. A final paper will present insights from the mushroom body, a center of multimodal associative learning in the fly brain.
- organism: Drosophila
- assay: imaging
- assay: spatial transcriptomics
- research focus: database / resource
- research focus: cell type annotation
- research focus: clustering
- approach: deep learning
- approach: experimental assay
- approach: statistical model
In-depth summary
Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 1 (21 Jan 2020) · CC BY-ND 4.0
A dense electron-microscopy reconstruction of half the fly central brain, with nearly every cell typed and queryable online, turns months of tracing partners into a database lookup.
Earlier fly electron microscopy (EM) work traced only the circuits each lab cared about, leaving coverage patchy and cell-type catalogues incomplete. Here the whole volume is reconstructed densely in one animal, types are set by connectivity as well as morphology, and the result is served through a public browser and programming interfaces.
Key results
ShownThe hemibrain is described as the largest dense reconstruction to date, and its 26-teravoxel image volume reduces to a connectivity graph about a million times smaller.
Around 25,000 neurons and about 20 million chemical synapses; the graph has 25,000 nodes and 3 million edges and occupies 26 MB even when split by brain region, roughly a million-fold reduction from the raw data. Imaging produced 26 teravoxels at 8 bits per voxel; over 50 person-years of proofreading over 2 calendar years went into it. · 1 Introduction; 2 Connectome Reconstruction
ShownNeurons were typed by a pipeline that starts from cell-body fibre bundles and combines morphological with connectivity clustering, yielding a type for nearly every reconstructed cell, most of them previously unnamed.
18,478 cell bodies found, 15,084 of them in the right central brain, falling into 192 cell body fibre bundles of which 154 matched 102 known clonal units. 4,768 cell types were assigned to 20,607 neurons, over a thousand of them represented by a single cell. Clustering used NBLAST followed by iterative connectivity clustering and manual review. · 2.7 Methodology for assigning cell types and…; 2.8 Results of cell typing; 2.9 Assessing Morphologies and Cell Types
ShownAutomated flood-filling segmentation was deliberately run in an over-merging regime, trading a high false-merge rate for long run lengths that human proofreaders then repaired.
Flood-filling networks were trained on roughly five billion voxels of ground truth and applied coarse-to-fine at 32, 16 and 8 nm after CycleGAN normalization. Base segmentation: expected run length (ERL) 163 microns at a 0.25% merge rate; after agglomeration, ERL 585 microns at a 27.56% merge rate. On 500 olfactory projection and Kenyon cells, ERL was 825 microns at 15.92%. · 2.2 Automated Segmentation
SuggestedConnections are deliberately sampled rather than complete, so the authors give explicit guidance on reading synapse counts: strong connections are trustworthy, weak ones often are not.
More than 90% of T-bars sit in identified neurons, but post-synaptic densities are traced at completeness typically between 20% and 85% per region. A connection of strength 10 is argued to be recovered with over 99.9% probability if half its synapses are traced. The authors advise ignoring connections of 3 or fewer synapses, and report single-connection pathways correct over 76% of the time. · 2.11 Assessing connection completeness; 2.10 Assessing Synapse Accuracy; 2.12 Interpreting the connection counts
ShownBecause every compartment comes from the same brain, their wiring statistics can be compared directly, and they differ widely on most measures except average graph distance.
Across compartments fully inside the volume with at least 100 neurons: neuron count varies 74-fold, mean partners per neuron 36-fold, network diameter 4-fold, mean connection strength 5-fold and the fraction of reciprocal connections 5-fold, while average graph distance between neurons varies only 2-fold. · 5.1 Compartment statistics
Methods, data and reproducibility
One female fly brain was hot-knife sectioned into 20 micrometre slabs and imaged by focused ion beam scanning electron microscopy (FIB-SEM) at 8 nm isotropic voxels, then stitched, segmented with flood-filling networks on CycleGAN-normalized images, annotated by two synapse predictors and proofread by hand; typing used cell body fibre bundles plus morphology and connectivity clustering.
- Hemibrain EM volume from one five-day-old female wild-type Canton S (G1 x w1118) fly, sectioned into sagittal slabs · Drosophila · n = 37 slabs of 20 micrometres; thirteen consecutive slabs imaged
- Volumetric ground truth for training flood-filling networks, from two tabs of the hemibrain · Drosophila · n = roughly five billion voxels
- Hand-annotated dense cubes across brain regions for assessing synapse precision and recall · Drosophila
- Released connectome version served through the browser and interfaces · Drosophila · n = more than 25K traced neurons · hemibrain:v1.0
- JRC2018 Drosophila template brain used to register the EM volume to light microscopy and GAL4 line libraries · Drosophila
- Methods & tools
- FIB-SEM, flood-filling networks, CycleGAN, synful, NeuPrint, DVID, NeuTu / Neu3, NeuroGlancer, NBLAST, ANTs, BigWarp, neo4j / Cypher
- Reagents & environment
- Data are served as grayscale in Google buckets, voxel segmentation, point annotations for synapses, SWC skeletons and a neo4j graph queried in Cypher; versions are named (hemibrain:v1.0) and a DOI is promised per milestone version.
- Assay conditions & compute
- Raw image data exceed 20 TB and take about 2 full days to download at 1 gigabit per second; reconstruction took over 50 person-years of proofreading across 2 calendar years on two customized FIB-SEM systems.
- Reproducibility gaps
- No URLs survive in this text for the NeuPrint browser, the interfaces or the data buckets, and readers are told to contact the authors to copy the underlying data and software. There is no institutional archive for data of this size; the authors pledge availability for at least the next 10 years.
Limitations (6)
- Post-synaptic densities are only partially traced (20% to 85% completeness by region), so reported synapse counts are representative rather than complete and weak connections may be spurious or missed. · stated by the authors
- Cell typing is acknowledged as inexact and still in progress; names are expected to be revised, and only body IDs are stable. · stated by the authors
- One hemibrain from a single female fly: stereotypy across animals, sex differences, the ventral nerve cord and the optic lobes are all outside this dataset, and neurons crossing the boundary are truncated. · stated by the authors
- EM here does not reveal whether a synapse is excitatory or inhibitory, and synapse count is used as connection strength although it is not a true coupling strength. · stated by the authors
- Proofreading effort was concentrated on regions of existing interest such as the central complex and mushroom body, so completeness and accuracy are uneven across the volume. · stated by the authors
- Accuracy was assessed largely in-house, against the project's own proofreaders, a second segmentation and a second synapse predictor, with only limited external comparison (an earlier alpha-lobe reconstruction and groups tracing a separate TEM volume), since no independent full-brain connectome exists. · noted in this summary
Read this if you study fly neural circuits and want to look up a neuron's partners, or if you build EM reconstruction pipelines and want the segmentation, synapse-prediction and proofreading trade-offs that made this scale feasible.
- Citations
- 144
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
- A Connectome and Analysis of the Adult Drosophila Central Brain
Louis K. Scheffer, C. Shan Xu, … Gregory S.X.E. Jefferis·eLifePeer-reviewed·Sep 3, 2020
- FlyBrainLab: Accelerating the Discovery of the Functional Logic of the Fruit Fly Brain in the Connectomic/Synaptomic Era
Aurel A. Lazar, Tingkai Liu, … Yiyin Zhou·eLifePeer-reviewed·Feb 22, 2021
- Network Statistics of the Whole-Brain Connectome of Drosophila
Albert Lin, Runzhe Yang, … Mala Murthy·NaturePeer-reviewed·Oct 2, 2024
- Neuronal wiring diagram of an adult brain
Sven Dorkenwald, Arie Matsliah, … Jansen Seguido·NaturePeer-reviewed·Oct 2, 2024
- Whole-brain annotation and multi-connectome cell typing quantifies circuit stereotypy in Drosophila
Philipp Schlegel, Yijie Yin, … Gregory S.X.E. Jefferis·bioRxiv·Jun 27, 2023
- Circuit Analysis of the Drosophila Brain using Connectivity-based Neuronal Classification Reveals Organization of Key Communication Pathways
Ketan Mehta, Rebecca Goldin, Giorgio A. Ascoli·Network NeurosciencePeer-reviewed·Oct 13, 2022