Rapid reconstruction of neural circuits using tissue expansion and light sheet microscopy

  1. Joshua L Lillvis
  2. Hideo Otsuna
  3. Xiaoyu Ding
  4. Igor Pisarev
  5. Takashi Kawase
  6. Jennifer Colonell
  7. Konrad Rokicki
  8. Cristian Goina
  9. Ruixuan Gao
  10. Amy Hu
  11. Kaiyu Wang
  12. John Bogovic
  13. Daniel E Milkie
  14. Linus Meienberg
  15. Brett D Mensh
  16. Edward S Boyden
  17. Stephan Saalfeld
  18. Paul W Tillberg
  19. Barry J Dickson

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This paper introduces a light microscopy pipeline for imaging and fast reconstruction of the synaptic connections of individual neuronal types in the fruit fly and for correlated investigation of circuit structure, function and behavior in the same animal. Because of its speed and accessibility, this approach enables mapping of selected neuronal circuits of multiple animals across different conditions and behavioral states, thus filling an important gap in brain research.

    (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.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Brain function is mediated by the physiological coordination of a vast, intricately connected network of molecular and cellular components. The physiological properties of neural network components can be quantified with high throughput. The ability to assess many animals per study has been critical in relating physiological properties to behavior. By contrast, the synaptic structure of neural circuits is presently quantifiable only with low throughput. This low throughput hampers efforts to understand how variations in network structure relate to variations in behavior. For neuroanatomical reconstruction, there is a methodological gulf between electron microscopic (EM) methods, which yield dense connectomes at considerable expense and low throughput, and light microscopic (LM) methods, which provide molecular and cell-type specificity at high throughput but without synaptic resolution. To bridge this gulf, we developed a high-throughput analysis pipeline and imaging protocol using tissue expansion and light sheet microscopy (ExLLSM) to rapidly reconstruct selected circuits across many animals with single-synapse resolution and molecular contrast. Using Drosophila to validate this approach, we demonstrate that it yields synaptic counts similar to those obtained by EM, enables synaptic connectivity to be compared across sex and experience, and can be used to correlate structural connectivity, functional connectivity, and behavior. This approach fills a critical methodological gap in studying variability in the structure and function of neural circuits across individuals within and between species.

Article activity feed

  1. Version published to 10.7554/elife.81248 on eLife
  2. eLife

    Author Response

    Reviewer #3 (Public Review):

    Lillvis et al present a new method for quick targeted analysis of neural circuits through a combination of tissue expansion and (lattice) light sheet microscopy. Three color labeling is available which allows to label neurons of a molecularly specific type, presynaptic and/or post-synaptic sites.

    Strengths:

    • The experimental technique can provide much higher throughput than EM
    • All source code has been made available
    • Manual correction of automatic segmentations has been implemented, allowing for an efficient semi-automatic workflow
    • Very different kinds of analyses have been demonstrated
    • Inclusion of electrical connections is really exciting, what a great complement to the existing EM volumes!

    Weaknesses:

    • Limitations of the method are not really discussed. While the approach is simpler and cheaper than EM, it's still important to give the readers a clear picture of the use cases where it's not expected to work before they embark on the journey of acquiring tens of terabytes of data. Here are just a few examples of the questions I would have if I wanted to implement the method myself - I am a computational person and can easily imagine my "wet lab" colleagues would have even more to ask about the experimental side:

    Please see our response to the Essential Revisions (for the authors) section above in addition to the responses to each point below.

    • It is not very clear to me if the resolution of the method is sufficient to disentangle individual neurons of the same type. It has been demonstrated for a few examples in the paper, but is it generally the case? Are there examples of brain regions/neuron types where it wouldn't be possible? If another column was added to the table in Figure 1, e.g. "individual neuron connectivity", EM would be "+", LM "-", what would ExLLSM be?

    Individual neuron connectivity is possible using this current version of ExLLSM either by labeling individual neurons genetically or by manually segmenting neurons in sparsely labeled samples. Of course, the exact answer to this question depends on labeling density and sample quality, and we have added a statement to address this.

    Lines 585-591: The difficulty of such manual segmentation can vary substantially depending on labelling density and signal quality. For instance, manually segmenting individual L2 outputs (Fig. 3) took ~10 minutes/neuron whereas segmenting a pair of SAG neurons from off-target neurons (Fig. 4) took 1-5 hours depending on the sample. Of course, more densely labeled samples will take more time. Finally, while it is possible to segment individual neurons from entangled bundles as shown here and elsewhere (Gao et al., 2019), the expansion factor will need to be increased by an order or magnitude or more and neuron labels must be continuous to approach EM levels of reconstruction density.

    • Similarly, the procedures for filling gaps in the signal could result in falsely merged neurons. Does it ever happen in practice?

    Because the gap filling process is not utilized until after semi-automatic segmentation this was not a concern (the gaps were filled on manually inspected neuron masks that should only include signals from the neuron(s) of interest). This would certainly be a concern if we were using this gap filling step – or the fully automated neuron segmentation approach – to segment individual neurons from samples in which off-target neurons are also labeled, but that was not the case here.

    • How long does semi-manual analysis take in person-hours/days for a new biological question similar in scope to the ones demonstrated in the paper?

    The statement discussed above (lines 585-591) and an additional statement (lines 581-583) aim to address this.

    Lines 580-582: As such, analyzing the DA1-IPN data, for example, required relatively little human time. The semi-automatic neuron segmentation steps required a maximum of one hour per sample and all other steps are automated.

    • How robust are the networks for synaptic "blob" detection? The authors have shown they work for different reporters, when are they expected to break? Would you recommend to retrain for every new dataset? How would you recommend to validate the results if no EM data is available?

    We expect that the network for blob detection is quite robust as it essentially acts as high signal detector for punctate signals, as opposed to classifying a high-level shape or structure. We have modified the text to suggest that the synapse and neuron segmentation models we include be attempted before automatically retraining.

    Lines 368-372: Furthermore, the convolutional neural network models for synapse and neuron segmentation are classifiers of high signal punctate and continuous structures, respectively. As such, the models may already work well for segmenting similar structures from other species or microscopes. If not, these models can be retrained with a suitable ground truth data set and the entire computational pipeline can be applied to these new systems.

  3. eLife

    Evaluation Summary:

    This paper introduces a light microscopy pipeline for imaging and fast reconstruction of the synaptic connections of individual neuronal types in the fruit fly and for correlated investigation of circuit structure, function and behavior in the same animal. Because of its speed and accessibility, this approach enables mapping of selected neuronal circuits of multiple animals across different conditions and behavioral states, thus filling an important gap in brain research.

    (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.)

  4. eLife

    Reviewer #1 (Public Review):

    This is a very timely and substantial advance in connectomics research that allows the fast reconstruction of selected neuronal circuits at synaptic resolution using tissue expansion and light sheet imaging. The authors describe this methodology in detail as applied to Drosophila brain, with multiple examples across different neuronal types and labeling strategies. The study is very rigorously done, methods are presented with important details, and the discussion is engaging and balanced. The paper is excellently written and very informative.

    The authors begin by introducing a workflow to detect and quantify presynaptic structures of specific neuronal types. This approach takes advantage of the T-bar protein Brp ubiquitously expressed at presynapses and the widely used nc82 antibody against it, as well as the fact that presynapses are larger neurites that are readily resolvable with light microscopy. Using three distinct neuronal types, the authors show that the number of presynapses obtained with the presented light microscopy method, matches well the synapse number quantified by the gold standard, electron microscopy.

    Next, the authors present two approaches to tackle a more difficult task - the quantification of the synaptic connectivity between 2 specific neuronal types. Compared to mammals, the identification of the postsynaptic site is more difficult in the Drosophila, because each presynapse contacts several different postsynaptic neurites that are right next to each other and are much smaller in size. No ubiquitous postsynaptic marker is currently available for the fly brain either. However when there is a postsynaptic marker available for specific connection, this makes the synaptic connection identification much more reliable, as shown with the example of the synaptic connections between the cholinergic SAG neurons and their postsynaptic target, the pC1 neurons, using the postsynaptic marker Drep2. Using this strategy the authors demonstrate that mated female flies have significantly less synaptic connections between SAG neurons and pC1 neurons, compared to virgin flies.

    In addition to chemical synapses, this study also shows a proof of principle that electrical synapses, gap junctions, can similarly be mapped using the same approach. This is very important, because these synapses are much more difficult to identify with electron microscopy and are not currently included in the available Drosphila connectomes. Definitive mapping of gap junctions however will require further work, outside the scope of this study, because there are different gap junction proteins and individual gap junctions may be heterotypic, composed of two different proteins.

    Finally, the authors extend this approach to address the important question of whether variations in behavior can be explained by differences in underlying synaptic connectivity. Using the neuronal circuit known to be responsible for the male fly courtship song, the authors show that the synaptic connectivity between pC2l and pIP10 neurons is correlated with a specific component of the optogenetically-elicited fly song.

    The developed imaging and analysis pipeline includes software for visualization of multi-terabyte images, automated neuronal segmentation, detection and quantification of pre- and postsynaptic sites. As the authors point out, these tools could be useful for circuit analysis in other species as well. The different imaging and analysis pipelines are presented very well, with multiple examples that cover different scenarios, and are well validated. While with this method it is not possible to directly correlate the fluorescence signal with the underlying ultrastructure as seen with EM, and thus it cannot be confirmed that the detected synaptic connections correspond to ultrastructurally defined synapses, the authors have convincingly demonstrated that the proposed approach is precise enough to detect a similar number of synapses as EM studies of the same neurons, and that it is sensitive enough to detect changes in synapse numbers in different experimental conditions.

  5. eLife

    Reviewer #2 (Public Review):

    The manuscript by Lillvis et al. presents a new pipeline for the light-microscopy-based reconstruction of synapse-level connectomes in the fruit fly. The data and analyses are of very high quality and the authors made all analysis tools and protocols available. There is a well documented github page to aid the installation and running of the pipeline by others.

    I would only suggest that the authors discuss in more detail the limitations of the technology and how transferrable it is to other species with larger brains or where there are no advanced genetic tools available.

    Overall, a very impressive technical feat and this new pipeline has the potential to advance experimental and comparative connectomics by the rapid analysis of multiple individuals.

  6. eLife

    Reviewer #3 (Public Review):

    Lillvis et al present a new method for quick targeted analysis of neural circuits through a combination of tissue expansion and (lattice) light sheet microscopy. Three color labeling is available which allows to label neurons of a molecularly specific type, presynaptic and/or post-synaptic sites.

    Strengths:
    - The experimental technique can provide much higher throughput than EM
    - All source code has been made available
    - Manual correction of automatic segmentations has been implemented, allowing for an efficient semi-automatic workflow
    - Very different kinds of analyses have been demonstrated
    - Inclusion of electrical connections is really exciting, what a great complement to the existing EM volumes!

    Weaknesses:
    - Limitations of the method are not really discussed. While the approach is simpler and cheaper than EM, it's still important to give the readers a clear picture of the use cases where it's not expected to work before they embark on the journey of acquiring tens of terabytes of data. Here are just a few examples of the questions I would have if I wanted to implement the method myself - I am a computational person and can easily imagine my "wet lab" colleagues would have even more to ask about the experimental side:

    -- It is not very clear to me if the resolution of the method is sufficient to disentangle individual neurons of the same type. It has been demonstrated for a few examples in the paper, but is it generally the case? Are there examples of brain regions/neuron types where it wouldn't be possible? If another column was added to the table in Figure 1, e.g. "individual neuron connectivity", EM would be "+", LM "-", what would ExLLSM be?
    -- Similarly, the procedures for filling gaps in the signal could result in falsely merged neurons. Does it ever happen in practice?
    -- How long does semi-manual analysis take in person-hours/days for a new biological question similar in scope to the ones demonstrated in the paper?
    -- How robust are the networks for synaptic "blob" detection? The authors have shown they work for different reporters, when are they expected to break? Would you recommend to retrain for every new dataset? How would you recommend to validate the results if no EM data is available?

  7. Version published to 10.1101/2021.11.14.468535v2 on bioRxiv
  8. Version published to 10.1101/2021.11.14.468535v1 on bioRxiv