Light-microscopy based dense connectomic reconstruction of mammalian brain tissue

The brain consists of a complex network of neurons and other cells. To understand its spatial organization and connectivity, developing advanced imaging techniques is crucial. While light microscopy has potential and its resolution can be improved by various methods and applications [1,2], it has not been used for connectomic studies. In contrast, electron microscopy (EM) offers very high resolution, and is the instrument of choice for detailed connectomic analysis. However, EM also has its limitations e.g. in molecular labelling and relies on light microscopy for this information [3].

In this study, a technology called LICONN (Light Microscopy based Connectomics) is introduced, combining high-fidelity hydrogel expansion, protein-density staining, and deep-learning-based segmentation, enabling synaptic-resolution reconstruction of brain circuitry with light microscopy. LICONN bridges the gap between the resolution of electron microscopy and the molecular specificity of light microscopy, offering a powerful tool for understanding brain circuitry in greater detail.

Key Findings

Improved resolution by expansion microscopy and automated segmentation

The technology presented in this preprint uses expansion microscopy, a method which relies on expanding a tissue using hydrogel to enhance resolution for light microscopy [2]. Through iterative expansion steps, the authors achieved high-fidelity tissue preservation and traceability of neuronal structures. By integrating fluorescent labelling of molecules, the method allowed comprehensive visualisation of cellular ultrastructure at nanoscale resolution.

After validating the traceability of the neuronal structures by manual tracing, the authors aimed to analyse larger volumes using deep learning-based segmentation algorithms. Flood-Filling Networks were trained on manually traced images and applied to mouse hippocampal CA1 region data. Through an iterative process of model predictions and manual proofreading, the automated segmentation produced accurate reconstructions of axons, dendrites, and nearly all correctly attached spines.

The authors were able to demonstrate LICONN’s capability for precise connectomic analysis comparable to what can be achieved by using EM.

Identification of structures across diverse brain areas with molecular labelling

The authors next sought to use LICONN to integrate structural and molecular information, with a particular focus on synaptic connections. By applying immunolabelling, they were able to visualise molecular components within the tissue’s 3D architecture. This approach allowed for the identification and characterisation of synaptic proteins, gap junction proteins, and subcellular structures.

Ultimately, this study demonstrates the power of molecularly-annotated connectomics in providing detailed insights into brain organisation and function across multiple spatial scales, cell types, and communication modalities.

Deep-learning based synapse detection

To enhance circuit analysis efficiency and overcome microscopy limitations, the authors adopted a deep learning approach predicting locations of the synaptic molecules rather than immunolabelling and imaging them directly. Convolutional neural networks were trained to predict localization of pre-synaptic and post-synaptic proteins using the correlation between molecular features and local architecture in the structural channel. The evaluation of independent datasets showed high consistency with ground truth immunolabelling data, demonstrating the fidelity of synaptic location predictions. Overall, connectivity patterns were analysed in a volume without immunollabelling, revealing dense synaptic connectivity mapped onto individual neurite segments.

Conclusion

LICONN enables dense 3D reconstruction of neurites, synapses, and molecular annotations in large tissue volumes, combining manual and automated methods for accurate structural mapping with light microscopy images. It achieves reliable, accurate detection of chemical as well as electrical connectivity and other subcellular structures using either immunolabelling and/or deep-learning based approaches, surpassing the limitations of traditional electron microscopy-based connectomics.

What I like about this preprint

Connectomics is changing the way neuroscientists explore the brain’s intricate networks. As a neuroscientist myself, I find it incredibly exciting to witness advancements in connectomic techniques. This preprint stands out by presenting a novel method that achieves synaptic-resolution reconstruction using standard confocal microscopy rather than electron microscopy. This method leverages equipment that is widely available in many research institutions, potentially improving access to high-resolution connectomic analysis.

Questions for the authors

• Have you tested the LICONN technology on brain tissue from organisms other than mice?
• How do you envision LICONN being integrated into current connectomic studies, and what impact do you think it will have on the field?
• What criteria did you use to select the antibodies for immunolabelling, and would the size of the antibody be an important factor as you are working at nanoscale resolution?

References

1. Lothar Schermelleh et al., Super-resolution microscopy demystified. Nat Cell Biol 21, 72–84 (2019). https://doi.org/10.1038/s41556-018-0251-8
2. Fei Chen et al., Expansion microscopy. Science 347, 543-548 (2015). doi:10.1126/science.1260088
3. Pascal de Boer et al., Correlated light and electron microscopy: ultrastructure lights up!. Nat Methods 12, 503–513 (2015). https://doi.org/10.1038/nmeth.3400

Tags: connectomics, expansion microscopy, mouse brain, neuroscience, synaptic resolution

doi: https://doi.org/10.1242/prelights.37623

Read preprint (No Ratings Yet)