The largest brain map ever shows fruit flies’s brain activity in detail
A comparative study of the KC activity in the mushroom bodies between FlyWire and the hemibrain (Extended Data Fig. 7 and 4), and a comparison of different populations
The values in the matrix are a combination of 0 and 1 and and. This helps to calculate analytical results for the KC activities. However, it is unrealistic as the connectomics data give the number of synaptic connections between the ALPNs and the KCs.
The number of inputs that each KC gets from the neural networks was used to generate theoretical predictions of K. 5k). KC activity is modeled after others.
The main goal of the KC analysis is to compare different populations ofKCs, and we did not want this bias in the analysis. For example, we examined the fraction of the APL output that is spent on each of the different KC types. For individualKCs, we used the budget for the wholeKC population as a reference point.
KCs, due to the high variability in numbers and synapse densities in the mushroom body lobes between FlyWire and the hemibrain (Fig. 5 and Extended Data Fig. 7).
Annotation Systematics in the FlyWire System for Central Brain Segment Quality Measurements: I. Comparison of 826 Neurons
A group of expert central proofreaders evaluated the quality of 3,106 segments in the central brain. These specific neurons were chosen based on certain criteria such as significant change since being marked complete and small overall volume. An additional 826 random neurons were included in the review pool as well. Proofreaders were unaware which neurons were added for quality measurement and which ones because they were flagged by a metric. We compared the 826 neurons before and after the review and found that the initial reconstruction scored an average F1-Score of 99.2% by volume (Extended Data Fig. 2a,b). The F1-Score refers to the ratio of true positives to true negatives, and the mean of recall among positively classified elements and all true elements.
When writing, the general FlyWire system operates in a read only mode in which users have the ability to make more annotations, but are not allowed to remove or modify existing annotations. Furthermore, the annotations consist of a single free-form text field bound to a spatial location. Many FlyWire users contributed a wide range of community annotations, which are reported in our companion paper, but are not considered in this study. As it became apparent that a complete connectome could be obtained, we found that this approach was not a good fit for our goal of obtaining a structured, systematic and canonical set of annotations for each neuron with extensive manual curation. Records for each neuron can be edited andcorrected over time through a web database set up by us that allowed columns with certain values to be added.
We only used the 1:1 and 1:many, not the many:1 matches. We used 2,954hemibrain type for the analyses.
We compared the cell types on the left and right of the hemibrain and there were some that were only on the left hemisphere.
A multi-hemisphere clustering study of FlyWire neurons using skeletons and dotprops in the L2 cache
For this NBLAST, we first generated skeletons using the L2 cache. The data structure underlying FlyWire is an octree, meaning that level 0 represents supervoxels, which are then agglomerated over higher levels. The second layer in this octree is larger than the first, and it is detailed enough to be used in the project. Precomputed information for the L2 chunk, including a representative x/Y/z coordinate, is in the L2 cache. We used the x/y/z coordinates and connectivity between chunks to generate skeletons for all FlyWire neurons (implemented in fafbseg; Table 1). Side branches less than 5 m were removed from the skeletons. We created the dotprops from those skeletons.
We ran the analysis across one, two, or three hemispheres to check the robustness of the groups. This treatment sometimes gave slightly different results. Some groups of neurons co-cluster across the different hemispheres; this is called apersistent clusters. Early-born neurons, which are often morphologically unique, frequently failed to participate in persistent clusters, and were omitted from further analysis. We used two- and three-hemisphere clustering to link persistent clusters across hemispheres, for example when clustering FlyWire left and FlyWire right together, the TuBu neurons from both sides would fall. Morphological groups are therefore defined by consistent across-hemisphere clustering. When neurons of a given hemilineage were sufficiently contained by the hemibrain volume, all three hemispheres (two from FlyWire and one from hemibrain) were used; otherwise, the two hemispheres from FlyWire were used. A small percentage of neurons with hemilineage annotations do not have a morphological group, because we prioritized consistency across 1. 2 and 3 hemisphere clustering. For example, if neuron type A clusters with type B in one-hemisphere clustering, but clusters with type C (and not B) in two-hemisphere clustering, then type A will not have a morphological group annotation.
We also provide a neuroglancer scene in which both FlyWire and hemibrain data are displayed in the correct orientation: https://tinyurl.com/flywirehbflip783. The standard convention says that the hemibrain and FAFB can be seen to the left of the screen. The scene shows that the right asymmetric body for FlyWire and the hemibrain has been superimposed, similar to what is seen in the Extended Data fig. 1a).
The NBLAST algorithm is implemented in both navis and the natverse (Table 1). We altered the jins implementation for more efficient parallel computation to scale to more than 100,000 neurons. For example, the all-by-all NBLAST matrix for the full 139,000 FlyWire neurons alone occupies over 500 GB of memory (32 bit floats). The large NBLASTs were run on a single cluster with 112 processors and 1 tb of RAM and took between 1 and 2 days to complete.
Hierarchical annotation and multi-connectome cell typing of Drosophila: Flow, superclass, class and cell type in the central brain
The central brain contains various superclasses: visual projection neurons, ascending neuron, and sensory neurons.
Hierarchical annotations include flow, superclass, class and cell type. The flow and superclass were generally assigned based on an initial semi-automated approach followed by extensive and iterative manual curation. See Supplementary Table 3 for definitions and the sections below for details on certain superclasses.
Biological outliers range from small additional/missing branches to entire misguided neurite tracks, and were typically assessed within the context of a given cell type and best possible contralateral matches within FlyWire and/or the hemibrain. When biological outliers were suspected, careful proofreading was undertaken to avoid erroneous merges or splits of neuron segmentation.
(2) Some neurons are missing large arbours (for example, a whole axon or dendrite) because a main neurite suddenly ends and cannot be traced any further. This occurs in commissures where many neurites cross the brain’s midline. In some but not all cases, we were able to bridge those gaps and find the missing branch through left–right matching. When the neurons weren’t complete, we labeled them as outliers.
Source: Whole-brain annotation and multi-connectome cell typing of Drosophila
Virtual FlyBrain (VFB): Enhanced box plots of Drosophila neuron reconstructions from the hemibrain
Enhanced box plots—also called letter-value plots125—in Fig. 5h and Extended Data Fig. 7f are a variation of box plots better suited to represent large samples. They replace the whiskers with a variable number of letter values where the number of letters is based on the uncertainty associated with each estimate, and therefore on the number of observations. The ‘fattest’ letters are the (approximate) 25th and 75th quantiles, respectively, the second fattest letters the (approximate) 12.5th and 87.5th quantiles and so on. The width of the letters is unrelated to the underlying data.
VirtualFlyBrain gathers and extracts information from all the publications about Drosophila neurobiology. There are many neuron reconstructions, including those from the hemibrain, that can be examined in the VFB. The persistent ID is known as VFB_xxxxxxxx, and it is the form for the individual neuron. Where cell types have been defined, they have an ontology ID (for example, FBbt_00047573, the ID for the DNa02 DN cell type). If different terms were used, VFB cross-references neurone cell types across publications. It identifies driver lines. In this paper, we generate an initial mapping providing FBbt IDs for the closest and fine-grained ontology term that already exists in their database. For example, a FlyWire neuron with a confirmed hemibrain cell type will receive a FBbt ID that maps to that exact cell type, while a DN that has been given a new cell type might only map to the coarser term ‘adult descending neuron’. It is currently being worked upon by the VFB to assign both of the two IDs to all FlyWire cell types as well as all individual FlyWire neurons.
Statistical Analysis of FlyWire Datasets: A Multi-Level Comparison between Theoretical Predictions and Implementations of the Python and FAFB Data Sets
Unless otherwise stated, statistical analyses (such as Pearson R or cosine distance) were performed using the implementations in the scipy123 Python package. We used t-tests for normally distributed samples to determine statistical significance.
The number of connections in each mixing layer is kept at a constant K. It is definitely a simplification that can be corrected by introducing a distribution P(K) but this requires further detailed modelling.
The global inhibition provided by APL to all of the mixing layer neurons is assumed to take a single value for all neurons. The level of inhibition varies depending on the number of APL and mixing layer neurons in the brain.
From Fig. 5k, the theoretical values of K that maximize dim(h) in this simple model demonstrate the consistent shift towards lower values of K found in the FlyWire left and FlyWire right datasets when compared with the hemibrain.
For consistency with visualizations and datasets obeying the standard convention (fly’s right on viewer’s left), FlyWire data can be mirrored. The tools we provide are the natverse nat.jrcbrains and the Python flybrains. 1c), through the
The post-processing of the FAFB dataset was extensive, and it resulted in a plethora of 3D visualization tools. Instead, we break a convention of presentation: usually, frontal views of the fly brain place the fly’s right on the viewer’s left. The view one has of oneself while looking in a mirror is similar to the view one sees from the fly brain in this paper. This maintains consistency with past publications. However, note that all labels of left and right in the figures in this paper, our companion papers, the supplemental annotations and associated digital repositories (for example, https://codex.flywire.ai, FAFB/FlyWire CATMAID) have been corrected to reflect the error during data acquisition. In these resources, a neuron labelled as being on the left is indeed on the left of the fly’s brain.
By screening all the axon profiles of a given nerve we identified all the non-visual sensory and motor neuron entering and exiting the brain.
The criteria for identifying ANs included that they have no soma in the brain, and main branch entering through the neck connective.
To identify DNs described in ref. 107 in the EM dataset, we transformed the volume renderings of DN GAL4 lines into FlyWire space. The same space gave us the ability to show closely related neuron in a certain order. We created space for the GAL4 or Split GAL4 line stacks by transforming candidate EM matches by eye and putting them on top of them. 107 for that type) in FIJI for verification. All but two of the 16 cell types in FAFB/FlyWire were identified using these methods. All other unmatched DNs received a systematic cell type consisting of their soma location, an ‘e’ for EM type and a three digit number (for example, DNae001). A detailed account and analysis ofDNs has been published.
Besides the canonical root point, the soma position was recorded for all neurons with a cell body. This was either based on curating entries in the nucleus segmentation table (removing duplicates or positions outside the nucleus) or on selecting a location, especially when the cell body fibre was truncated and no soma could be identified in the dataset. The soma locations were important for a lot of analyses, and they allowed a consistent side to be defined for each neuron. This was done using a cutting plane at the midline to define the left and right soma positions. However, all soma positions within 20 µm of the midline plane were then manually reviewed. The goal was to define a consistent logical soma side based on the examination of the cell body Fibre tracts entering the brain; this ultimately ensured that cell types were always annotated so that one neuron was identified as the left and the right. In only a few cases, we assigned the side as central for the bilaterally symmetric unpaired medial neurons.
Side is used for sensory neurone, if they enter the brain through the left or right nerve. In a small number of cases we could not unambiguously identify the nerve entry side and assigned side as ‘na’.
We also note that our annotations include a number of non-neuronal cells/objects such as glia cells, trachea and extracellular matrix that others might find useful (superclass not_a_neuron; listed in Supplementary Data 2).
Visual sensory neurons (R1–6, R7–8 and ocellar photoreceptor neurons) were identified by manually screening neurons with pre-synapse in either the lamina, the medulla and/or the ocellar ganglia93.
Johnston’s organ neurons in the right hemisphere were characterized based on innervation of the major AMMC zones (A, B, C, D, E and F), but not further classified into subzone innervation as shown previously104. Other sensory neurons (mechanosensory bristle neurons, taste peg mechanosensory neurons and gustatory sensory neurons) in the right hemisphere were identified through NBLAST-based matching of their mirrored morphology to the left hemisphere and expert review. Olfactory, thermosensory and hygrosensory neurons of the antennal lobes were identified through their connectivity to cognate uniglomerular projection neurons and NBLAST-based matching to previously identified hemibrain neurons40,106.
There were differences that were striking. The FlyWire fly had almost twice as many neurons in a brain structure called the mushroom body, which is involved in smell, compared with the fly used in the hemibrain-mapping project. The discrepancy could be because of the hemibrain fly’s brain development being harmed by starving.
Using Flywire and hemibrain to verify that VCNs are all typed according to the Hartenstein nomenclature
The majority of VCNs were assigned to specific types. Only 29 and 9 were able to be confidently assigned cell types, leaving them untyped.
We usedFlyWire and hemibrain to run clustering on 25 cross-identified hemilineages that aren’t truncated in the hemibrain to verify a subset of the new cell types. The procedure was otherwise the same as for the double-clustering.
The term ‘neuropil’ is used where it refers to regions innervated by VPNs and C/T means columnar versus tangential organization.
Similar to cell typing in the central brain, a significant proportion of VPN (61%) and visual centrifugal neuron (VCN) (60%) types are derived from the hemibrain (see the ‘Hemibrain cell type matching’ section). The Supplementary Data has a hemibrain_type column.
By comprehensively inspecting the hemilineage tracts originally in CATMAID and then in FlyWire, we can now reconcile previous reports. Specifically, new to refs. 33,34 (ItoLee nomenclature) are: CREl1/DALv3, LHp3/CP5, DILP/DILP, LALa1/BAlp2, SMPpm1/DPMm2 and VLPl5/BLVa3_or_4—we gave these neurons lineage names according to the naming scheme in refs. 3,34. New to ref. 31 (Hartenstein nomenclature) are: SLPal5/BLAd5, SLPav3/BLVa2a, LHl3/BLVa2b, SLPpl3/BLVa2c, PBp1/CM6, SLPpl2/CP6, SMPpd2/DPLc6, PSp1/DPMl2 and LHp3/CP5—we named these units according to the Hartenstein nomenclature naming scheme. The following clones weren’t taken from ref. There’s a reason why 33 is taken into account for the number of lineages, they aren’t clearly shown in the larva.
Light-level clones from refs were notable. Sometimes, clones with the same name only match part of the time. For example, the AOTUv1_ventral/DALcm2_ventral hemilineage seems to be missing in the AOTUv1/DALcm2 clone in the Ito collection33. It appears that there is a similar situation for the BM, CM, DALv2, and LHl3 lineages. Clones are preferred when there is a conflict. There is 34.
This pipeline is implemented in the coconatfly package (Table 1), which provides a streamlined interface to carry out such clustering. For example the following command can be used to see if the types given to a selection of neurons in the Lateral Accessory Lobe (LAL) are robust:
The optional interactive mode is an efficient way of exploring a web browser. For further information, see the example at https://Natverse.org/coconatfly/.
Multihemibrain Types and Connection-Maximization of Biological APL Inhibitions in the Three Biological Datasets [Revised]
In rare instances, the hemibrain types were double checked and corrected if there were more than one hemibrain type.
It affects the number of input connections. In other words, what are the numbers of connections K made to individual KCs that maximize their responses so as to prevent a global inhibition?
In a previous report122, there are more detailed calculations. The row with 1 and N K elements are included in the populated portion of (bfW). The parameter α represents a homogeneous inhibition corresponding to the biological, global inhibition by APL. The number of KCs in each of the three datasets was set to determine the value-inhibition. The dimension of the activities defines the primary quantity of interest.
The FlyWire Project: Quantifying the Proofreading Time of the Second Public Release of the Neuronal Wiring Diagram of an Adult Brain
The professional proofreading team received additional proofreading training. Correct proofreading relies on a diverse array of 2D and 3D visual cues. Proofreaders learned about 3D morphology, resulting from false merger or false split without knowing what types of cells they are. The ultrastructures give valuable 2D signals and are reliable guides for accurate tracing. Before Professional Proofreadings were admitted into Production, they practiced on an average of200 cells in a test dataset with additional feedback. In this dataset, we compared the accuracy of test cells to ground-truth reconstructions. To improve proofreading quality, peer learning was highly encouraged.
Any quantification of the total proofreading time that was required to create the FlyWire resource is a rough estimate because of the distributed nature of the community, the interlacing of analysis and proofreading and the variability in how proofreading was performed. The second public release, version 783, required 3,013,513 edits. We measured proofreading times during early proofreading rounds that included proofreading of whole cells in the central brain. We collected and number of edits after removing outliers with more than 500 edits. From these data, we were able to calculate an average time per edit. We saw that proofreading times were much higher for tasks that only needed minimal edits. The second round of proofreading used all segments with over 100 speaches, and our measurement was not representative of that. It takes 4–5 edits to do these. We adjusted for that by limiting the calculations to a few timed tasks and using estimates from the first round. The average time per edit in our proofreading time dataset, (round 2) the average time of tasks with 1–5 edits. We averaged these times for an overall proofreading time because the number of tasks in each category were similar. The result was an average time of 79 s per edit which adds up to an estimate of 33.1 person-years assuming a 2,000 h work year.
FlyWire uses CAVE50 for hosting the proofreadable segmentation and all of its annotations. The PyChunkedGraph is one of CAVE’s proofreading systems.
Source: Neuronal wiring diagram of an adult brain
How Many Neurons are in a Neuropil Mesh? A Comparison of the Central and Intrinsic Brains from Neuprint (v1.2.1)
The neurosciences were assigned to synapses based on their presynaptic locations. We used ncollpyde (https://pypi.org/project/ncollpyde/) to calculate whether the location was within a neuropil mesh and assigned the synapse accordingly. Some synapses remained unassigned after this step because the neuropils only resemble rough outlines of the underlying data. We then assigned all remaining synapses to the closest neuropil if the synapse was within 10 µm from it. The remaining synapses were left unassigned.
We calculated a volume for each neuropil using its mesh. The brains of the paper were given to the medulla plate in the aggregated volumes. The remaining neuropils but the ocellar ganglion were assigned to the central brain.
,,,,,,.
We retrieved the latest completion rates and synapse numbers for the hemibrain from neuprint (v1.2.1). In some cases, neuropil comparisons were not directly possible because of redefined regions in the hemibrain dataset. We excluded these regions from the comparison.
We try to understand whether a neuron is intrinsic to a region based on its synaptic locations. In other words, the neurites of an intrinsic neuron are allowed to exit the region, provided that they do not make synapses after leaving. The diagrams for White et al.31 have information about the C. elegans sphinx locations.
It should be noted that the estimate has error bars due to definitional ambiguities. It is unclear if motor neurons are classified as intrinsic neurons, which would potentially remove them from the list. Or the brain could be enlarged by moving the posterior border further behind the excretory pore, which would add 10 neurons (RIF, RIG, RMG, ADE and ADA). 35 10 intrinsic neurons is the estimate that we make explicit. Of the 302 CNS neurones, 180 make connections in the brain. The brain and the central nervous system comprise between 15 to 25% and 15% of the total neurons.
We calculated cell volumes and surface areas using CAVE’s L2Cache50. Volumes were computed by counting all voxels within a cell segment and multiplying the count by the voxel resolution. Area calculations were more complicated and were performed by overlap through shifts in voxel space. We shifted the binarized segment to different dimensions and obtained overlap of false and true voxels. For each dimension, we counted the extracted voxels and multiplied the count by the voxel resolution of the given dimensions. Finally, we added up the area estimates. smoothed measurements were too compute intensive and overstated the area slightly.
The synapse classifier by Buhmann et al. was trained on ground truth from the CREMI challenge (https://cremi.org). Three 5 5 5 m cubes from the calyx are contained in the three CREMI datasets. While the classifier from Buhmann et al. was trained and validated on only this dataset, they evaluated its performance on multiple regions (calyx, lateral horn, ellipsoid body and protocerebral bridge). It should be noted that performance varies by region.
The Buhmann et al.10 team used a machine learning model to predict their neurotransmitter identities. The images were based on electron microscopy alone. Each synapse was assigned a probability for one of six neurotransmitters: acetylcholine, glutamate, GABA, serotonin, dopamine and octopamine. They used neurotransmitter identities published for individual neuronal cell types and built a dataset with 3,025 neurons with known transmitter type assuming Dale’s law applies. The per-synapse accuracy was reported as 87% and the majority vote accuracy was reported as 94%.
For each neuron, we calculated the fraction of presynapses in the left and right hemisphere. The hemisphere opposite its dominant input side was named the contralateral hemisphere. We excluded neurons that had either most of their presynapses or most of their postsynapses in the centre region.
We used the information flow algorithm implemented by Schlegel et al.26,128 (https://github.com/navis-org/navis) to calculate a rank for each neuron starting with a set of seed neurons. The graph of the neurones is probabilistically crossed by the algorithm. The likelihood of a neuron being added to the traversed set increased linearly with the fraction of synapses it receives from already traversed neurons up to 30% and was guaranteed above this threshold. We used the same calculation for all set of afferent and sensory neurons. We used a group that included olfactory receptor neurons, gustatory receptor neurons, head and neck bristle mechanosensory neurons, taste peg neurons, thermosensory cells, hygrosensory cells, and visual photoreceptors.
Additionally, we created input seeds by combining all listed modalities, all sensory modalities, and all listed modalities with visual sensory groups excluded.
For each modality we performed 10,000 runs, which were averaged. We took the rank of the neurons, and ordered them according to their location in the order. To compute a reduced dimensionality, we use the UMAP129 and are able to use n_components and min_dist.
The FlyWire Project: Investigating the Connectivity of Brain Brain Neurons in the COVID-19 Epidemic and Future Directions
The wiring diagram needed to be checked for errors because these tools aren’t perfect. The scientists spent a great deal of time manually proofreading the data — so much time that they invited volunteers to help. In all, the consortium members and the volunteers made more than 3 million manual edits, according to co-author Gregory Jefferis, a neuroscientist at the University of Cambridge, UK. He notes that much of this work was done in 2020, when researchers were working from home during the COVID-19 epidemic.
The team was surprised by the ways in which the different cells communicate. For instance, neurons that were thought to be involved in just one sensory wiring circuit, such as a visual pathway, tended to receive cues from multiple senses, including hearing and touch1. “It’s astounding how interconnected the brain is,” Murthy says.
The map was still being annotated, and the researchers and volunteers had to describe each neuron as a specific cell type. Humans would need to check the results of the software they are trained to use if they want to recognize lakes or roads in the satellite images. The researchers identified far more types of neuron than they had anticipated. New research directions are going to be created by 4,581 newly discovered. “Every one of those cell types is a question,” he adds.
The map1 is described in a package of nine papers about the data published in Nature today. Its creators are part of a consortium known as FlyWire, co-led by neuroscientists Mala Murthy and Sebastian Seung at Princeton University in New Jersey.
A neuroscience professor at the Allen Institute for Brain Sciences in Seattle, Washington who is not involved in the project but has worked with one of the team members, says it is a huge deal. “It’s something that the world has been anxiously waiting for, for a long time.”
A team of international researchers has shared data on the structure of brain neurons in a FlyWire fly species, which had almost twice as many neurons as the fly used in the hemibrain-mapping project. They said the fly had almost twice the neurons in a brain structure called the mushroom body compared with the fly used in the hemibrain-mapping project.