A searchable image resource of Drosophila GAL4 driver expression patterns with single neuron resolution
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This study bridges the gap between connectomic data from the fly hemibrain and driver lines needed for functional experiments through a new freely available computational tool, NeuronBridge. It demonstrates that this software provides users with the ability to identify the same neurons within different driver lines, and the opportunity to match expression of neurons in a driver line with those in a connectomic database. Overall, this manuscript does a commendable job of describing an important resource for the community, which will hopefully be built upon via collaborative science of many groups as the field develops.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Neuroscience (eLife)
Abstract
Precise, repeatable genetic access to specific neurons via GAL4/UAS and related methods is a key advantage of Drosophila neuroscience. Neuronal targeting is typically documented using light microscopy of full GAL4 expression patterns, which generally lack the single-cell resolution required for reliable cell type identification. Here, we use stochastic GAL4 labeling with the MultiColor FlpOut approach to generate cellular resolution confocal images at large scale. We are releasing aligned images of 74,000 such adult central nervous systems. An anticipated use of this resource is to bridge the gap between neurons identified by electron or light microscopy. Identifying individual neurons that make up each GAL4 expression pattern improves the prediction of split-GAL4 combinations targeting particular neurons. To this end, we have made the images searchable on the NeuronBridge website. We demonstrate the potential of NeuronBridge to rapidly and effectively identify neuron matches based on morphology across imaging modalities and datasets.
Article activity feed
-
-
Author Response
Reviewer #1 (Public Review)
Overall the claims in the manuscript are clearly communicated and justified by the data. However, one of the features on NeuronBridge that was mentioned in the manuscript did not work intuitively and could use more description in the manuscript. This was the feature to upload a confocal stack to search for other Gal4 lines or the appropriate neurons in the EM hemibrain. When a known Gal4 was in the database, it was easy and intuitive to go from a driver line to an EM neuron or, alternatively if an EM neuron was known it was easy to go from that neuron to find a driver line. It was, however, difficult to upload a stack and find the neuron names or a driver line. The example on Neuronbridge was somewhat helpful but an accompanying brief 'How-to' for this process in the manuscript would be …
Author Response
Reviewer #1 (Public Review)
Overall the claims in the manuscript are clearly communicated and justified by the data. However, one of the features on NeuronBridge that was mentioned in the manuscript did not work intuitively and could use more description in the manuscript. This was the feature to upload a confocal stack to search for other Gal4 lines or the appropriate neurons in the EM hemibrain. When a known Gal4 was in the database, it was easy and intuitive to go from a driver line to an EM neuron or, alternatively if an EM neuron was known it was easy to go from that neuron to find a driver line. It was, however, difficult to upload a stack and find the neuron names or a driver line. The example on Neuronbridge was somewhat helpful but an accompanying brief 'How-to' for this process in the manuscript would be very welcome. If it's a possibility, I recommend adding this in as a 'box' or Figure in the revised paper. Further, the authors may want to provide a troubleshooting guide on the website for uploading a confocal stack onto Neuronbridge.
We are revising the text on the website for clarity and adding additional troubleshooting information. This, along with other updates to the website, will be available in the next release of NeuronBridge towards the end of 2022.
Reviewer #2 (Public Review):
- Figure 4 and its two supplements show the distribution of correct hits in the top 100 for a forward search, as well as illustrating the complementary nature of the 2 methods, with some correct hits found by one of the methods but not the other. Figure 5 shows the results for a reverse search. It seems that this does not correlate to neuron morphology. The manuscript does not mention however if any attempts were made to improve the scoring so that correct hits would be more highly ranked. It would be helpful to clarify this.
Development of CDM and PPPM search algorithms and associated pre- and post-processing optimizations has proceeded in parallel with the MCFO data release and NeuronBridge application described in the paper. Mais et al., 2021 describes in detail their work to optimize PPPM. CDM improvements since Otsuna et al., 2018 will be described in Otsuna et al., 2023, which isn't ready yet. While we view the search approach evaluations as showing that neuron matches can be found with CDM and PPPM, the evaluation can't be comprehensive across all neurons, datasets, and algorithm variations.
- Related to the point above, the examples used for the forward search are all visual projection neurons. In order to illustrate the usefulness and comprehensiveness of the searches, it would be helpful if some examples of central brain neurons, not truncated in Hemibrain, were also used.
We acknowledge the limited set of neurons examined in the evaluation of CDM and PPPM search, and tried to weight the claims accordingly in lines 305 and 309 of the submission. We agree more examples would be useful, but providing them hasn't proven feasible during the revision period. While the example neurons are truncated, it does not appear likely that searches with completely reconstructed neurons would generally produce worse results.
-
-
Evaluation Summary:
This study bridges the gap between connectomic data from the fly hemibrain and driver lines needed for functional experiments through a new freely available computational tool, NeuronBridge. It demonstrates that this software provides users with the ability to identify the same neurons within different driver lines, and the opportunity to match expression of neurons in a driver line with those in a connectomic database. Overall, this manuscript does a commendable job of describing an important resource for the community, which will hopefully be built upon via collaborative science of many groups as the field develops.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers …
Evaluation Summary:
This study bridges the gap between connectomic data from the fly hemibrain and driver lines needed for functional experiments through a new freely available computational tool, NeuronBridge. It demonstrates that this software provides users with the ability to identify the same neurons within different driver lines, and the opportunity to match expression of neurons in a driver line with those in a connectomic database. Overall, this manuscript does a commendable job of describing an important resource for the community, which will hopefully be built upon via collaborative science of many groups as the field develops.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
-
Reviewer #1 (Public Review):
The combination of near-completion of the Drosophila brain connectome and the simultaneous development of neurogenetic tools for manipulating neurons with high temporal and spatial specificity provides a new opportunity to understand the functional relevance and underlying molecular biology of circuits within the Drosophila brain with unprecedented coverage and resolution. A major challenge to this is matching neurons in connectomic datasets to those in known driver lines. NeuronBridge is a useful online search tool that builds on previous tools developed by the community (such as Neuron Basic Local Alignment Tool (NBLAST) and Color Depth Maximum Intensity Projection (CDM)) to link images from ~74000 fly brains to themselves so it's possible to find multiple lines that express in the same neuron, and to …
Reviewer #1 (Public Review):
The combination of near-completion of the Drosophila brain connectome and the simultaneous development of neurogenetic tools for manipulating neurons with high temporal and spatial specificity provides a new opportunity to understand the functional relevance and underlying molecular biology of circuits within the Drosophila brain with unprecedented coverage and resolution. A major challenge to this is matching neurons in connectomic datasets to those in known driver lines. NeuronBridge is a useful online search tool that builds on previous tools developed by the community (such as Neuron Basic Local Alignment Tool (NBLAST) and Color Depth Maximum Intensity Projection (CDM)) to link images from ~74000 fly brains to themselves so it's possible to find multiple lines that express in the same neuron, and to neurons in the FlyEM hemibrain connectomics data. This is an important resource for the Drosophila neuroscience community as it provides the ability to generate tools for manipulating neurons with unparalleled resolution and link high resolution anatomy and connectivity to function. Meissner et al is a very accessible manuscript which is written to provide detail and clarity for the expert reader, and includes enough information, resources and references for amateur and novice readers to follow. The authors did an excellent job of outlining their questions and problems, how these challenges were addressed, and the performance of the NeuronBridge software.
Overall the claims in the manuscript are clearly communicated and justified by the data. However, one of the features on NeuronBridge that was mentioned in the manuscript did not work intuitively and could use more description in the manuscript. This was the feature to upload a confocal stack to search for other Gal4 lines or the appropriate neurons in the EM hemibrain. When a known Gal4 was in the database, it was easy and intuitive to go from a driver line to an EM neuron or, alternatively if an EM neuron was known it was easy to go from that neuron to find a driver line. It was, however, difficult to upload a stack and find the neuron names or a driver line. The example on Neuronbridge was somewhat helpful but an accompanying brief 'How-to' for this process in the manuscript would be very welcome. If it's a possibility, I recommend adding this in as a 'box' or Figure in the revised paper. Further, the authors may want to provide a troubleshooting guide on the website for uploading a confocal stack onto Neuronbridge.
As a relatively minor point, could the authors also provide more clarifications on the known number of neurons in the adult Drosophila brain? On line 182, the authors cite that the adult central brain has ~30,000 neurons. The approximations I'm most familiar with for the adult brain with range between 100,000-200,000 cells with ~50-67% of cells being in the optic lobes and maybe 10-15% being glia. That being said, some of those numbers don't appear to have rigorous cell counts to back up the data although Raji et al (2021) recently found the whole adult brain has ~200,000 neurons with ~100,000 in the central brain and ~100,000 in the optic lobes. The authors should rewrite that statement in the introduction to provide clarity and accuracy on their numbers of neurons in the adult brain.
-
Reviewer #2 (Public Review):
The manuscript describes the generation of a large set of light-level (LM) images obtained by the stochastic method MultiColor FlpOut, covering both the brain and ventral nerve cord, males and females. The breadth of coverage as well as the size of the dataset (~74k images) make this a very significant resource. These images allow the identification of single neurons from the original Generation 1 GAL4 lines created by the FlyLight project and provide a very useful database from which potential line pairings to generate sparse split-GAL4 lines can be identified. By providing automated search tools in the new public website NeuronBridge users can match electron-microscopy (EM) Hemibrain neurons to these LM images, or vice-versa.
The website incorporates two different search methods previously published: color …Reviewer #2 (Public Review):
The manuscript describes the generation of a large set of light-level (LM) images obtained by the stochastic method MultiColor FlpOut, covering both the brain and ventral nerve cord, males and females. The breadth of coverage as well as the size of the dataset (~74k images) make this a very significant resource. These images allow the identification of single neurons from the original Generation 1 GAL4 lines created by the FlyLight project and provide a very useful database from which potential line pairings to generate sparse split-GAL4 lines can be identified. By providing automated search tools in the new public website NeuronBridge users can match electron-microscopy (EM) Hemibrain neurons to these LM images, or vice-versa.
The website incorporates two different search methods previously published: color depth maximum intensity projection (CDM) and PatchPerPixMatch (PPPM). Results are precomputed, making searches very fast. The user can filter and sort the ranked results lists and analyse each hit in more detail. The search methods are complementary, overlapping on some correct hits but not others. Ensuring a comprehensive search for correct hits can, as shown in the manuscript, demand the review of the top 100 hits, or even further ones.By generating the image dataset and the NeuronBridge public website incorporating the precomputed searches, this work provides Drosophila neuroscientists with a very useful resource which we expect will be used for years to come. It should also be noted that the image dataset is publicly available and can be downloaded.
Overall, the description of the resource and tools is accurate and well supported by the data. There are a couple of aspects that could be further clarified.
Figure 4 and its two supplements show the distribution of correct hits in the top 100 for a forward search, as well as illustrating the complementary nature of the 2 methods, with some correct hits found by one of the methods but not the other. Figure 5 shows the results for a reverse search. It seems that this does not correlate to neuron morphology. The manuscript does not mention however if any attempts were made to improve the scoring so that correct hits would be more highly ranked. It would be helpful to clarify this.
Related to the point above, the examples used for the forward search are all visual projection neurons. In order to illustrate the usefulness and comprehensiveness of the searches, it would be helpful if some examples of central brain neurons, not truncated in Hemibrain, were also used.
-
Reviewer #3 (Public Review):
Meissner et al. employ stochastic Gal4 labeling with MCFO to ease the identification candidate lines for split-Gal4 line generation to genetically target neurons of interest identified in EM traces. Data basis for the approach is a novel resource of 74k MCFO images aligned to the JRC18 template allowing the matching between EM and LM traces of single neurons. The resource is released in combination with data processing and query tools. In addition, an open web-based data portal to the released data collection and data mining tools is made available. This will allow broad access to this novel resource with the potential to create high impact in the community.
Strength:
The possibility to bridge between EM neuron traces and expression patterns in LM images is a key method to achieve and accelerate genetic …
Reviewer #3 (Public Review):
Meissner et al. employ stochastic Gal4 labeling with MCFO to ease the identification candidate lines for split-Gal4 line generation to genetically target neurons of interest identified in EM traces. Data basis for the approach is a novel resource of 74k MCFO images aligned to the JRC18 template allowing the matching between EM and LM traces of single neurons. The resource is released in combination with data processing and query tools. In addition, an open web-based data portal to the released data collection and data mining tools is made available. This will allow broad access to this novel resource with the potential to create high impact in the community.
Strength:
The possibility to bridge between EM neuron traces and expression patterns in LM images is a key method to achieve and accelerate genetic access to individual neurons. The proposed resource and tools contribute to this effort and provide open and easy access to it. This also includes the possibility to upload and analyze own data using the provided infrastructure, which is a great asset.
Weaknesses:
While the generation and analysis of the MCFO data is described in great detail and the overall technical approach seems feasible, the description of the technical part and its evaluation are lacking important implementation details and scientific rigor. Although this is primarily a life science paper introducing a new data resource it's the mining capability making this resource really valuable. The provided evaluation of the image mining capabilities however is currently insufficient to support the very general claims on effectivity and speed of the method.
-
-