Ultrastructural readout of in vivo synaptic activity for functional connectomics

Anna Simon, Arnd Roth, Arlo Sheridan, Mehmet Fişek, Vincenzo Marra, Claudia Racca, Jan Funke, Kevin Staras, Michael HäusserbioRxiv 2021 · DOI

TL;DR

This paper presents a novel method to simultaneously visualize the ultrastructure of neural circuits and the functional activity of individual synapses in vivo. By combining fluorescent activity markers with advanced electron microscopy and machine learning, they can now map "who's talking to whom" and "how strongly" at an unprecedented level of detail, paving the way for more comprehensive connectomics studies.

What Problem Does This Paper Address?

Understanding how neural circuits function requires not only knowing which neurons are connected (the "wiring diagram" or connectome) but also how active those connections are (synaptic strength and activity). Traditional methods for mapping neural circuits at the ultrastructural level, like serial section electron microscopy, provide incredibly detailed wiring information but lack a direct readout of synaptic activity. Conversely, functional imaging techniques can show which neurons are active but don't offer the fine-grained structural detail needed to pinpoint individual synapses and their specific properties. This gap makes it difficult to link circuit structure to function. This paper aims to bridge this gap by developing a method that integrates both ultrastructural detail and functional synaptic activity.

Methods

The core of this work lies in a multi-step experimental and computational pipeline:

  • In Vivo Functional Labeling: The researchers used FM1-43FX, a fluorescent dye that is taken up by recycling synaptic vesicles during periods of neuronal activity. This dye was injected into the visual cortex of awake, head-fixed mice while they were presented with visual stimuli. This step labels active synapses with a fluorescent signal.
  • Photoconversion: After fixation, the fluorescent FM1-43FX signal was converted into an electron-dense product (diaminobenzidine, DAB) using light. This photoconversion process makes the previously active vesicles visible under an electron microscope.
  • Focused Ion Beam Scanning Electron Microscopy (FIBSEM): This advanced microscopy technique was used to acquire high-resolution, near-isotropic 3D image stacks of the labeled tissue. FIBSEM mills away thin layers of the sample while simultaneously imaging the exposed surface, allowing for detailed reconstruction of large volumes.
  • Machine Learning-Based Segmentation: Convolutional neural networks (specifically, 3D U-Nets) were trained to automatically identify and segment various cellular components within the FIBSEM data, including neurons, active zones, mitochondria, and importantly, synaptic vesicles. These networks were also trained to classify vesicles as either photoconverted (PC+, indicating recent activity) or non-photoconverted (PC-, indicating no recent activity).
  • Hough Transform for Vesicle Detection: A Hough transform with a spherical kernel was employed to precisely locate the PC+ and PC- vesicles within the segmented boutons, providing confidence scores for each detection.

What's new here is the integration of in vivo functional labeling with FIBSEM and sophisticated machine learning for automated analysis, allowing for the simultaneous assessment of ultrastructure and synaptic activity across large tissue volumes.

Datasets

The paper does not explicitly mention the creation of new benchmark datasets. Instead, they generated their own datasets from experiments on mice for training and evaluating their machine learning models. The training data for the segmentation models (neurons, mitochondria, active zones, and vesicles) was derived from manually annotated regions within the FIBSEM datasets acquired from their experiments.

Key Insights and Ideas

  • Bridging the Structure-Function Gap: The most significant insight is the successful integration of ultrastructural detail with functional synaptic activity. This allows researchers to move beyond simply mapping connections to understanding the functional state of those connections.
  • Lognormal Distributions in Synaptic Properties: The number of PC- vesicles, PC+ vesicles, and total vesicles within boutons showed a lognormal distribution. This aligns with existing theories suggesting that lognormal distributions are efficient for information processing in biological systems, such as firing rates and synaptic weights.
  • Variability in Release Probability: Neighboring synaptic boutons from the same axon, which experience the same presynaptic action potentials, can exhibit significantly different release probabilities. This highlights a substantial degree of local regulation and variability in synaptic function, even within the same presynaptic terminal.
  • Large-Scale Ultrastructural Analysis is Feasible: The method demonstrated that FIBSEM can be used to image tissue volumes deep enough (>100 µm) to be amenable to large-scale connectomics studies, overcoming limitations of previous techniques.

Key Results

  • The developed method successfully identified two distinct classes of synaptic vesicles: photoconverted (PC+, indicating recent activity) and non-photoconverted (PC-, indicating no recent activity).
  • Spontaneous activity (darkness control) resulted in approximately one-tenth the number of PC+ vesicles compared to visually stimulated conditions.
  • The number of PC+ vesicles scaled with the total number of vesicles, but the fraction of PC+ vesicles varied considerably between individual synapses.
  • Analysis of boutons from the same axon revealed that the ratios of PC+ vesicle numbers (a proxy for relative release probability) were often dominated by values of 0.4 or less, indicating substantial differences in release probability between neighboring boutons.
  • A model of lognormally distributed release probabilities accurately predicted the observed distribution of release probability ratios.

Limitations and Future Work

  • Limitations:
    • The study focused on a specific brain region (primary visual cortex) and stimulus (drifting grating). Generalizability to other brain regions or more complex stimuli may require further validation.
  • While the method allows for the assessment of relative release probability between boutons of the same axon, directly quantifying absolute release probability would require further refinement.
  • The current approach relies on the photoconversion of FM1-43FX, which might have limitations in terms of penetration depth or potential phototoxicity, although they showed it could penetrate >100 µm.
  • Future Work:
    • The authors suggest that variants of their approach could be used with all-optical stimulation methods to precisely control and measure action potential activity, allowing for direct estimation of release probability at individual terminals.
  • This method opens up possibilities for investigating how specific structural features of synapses and their environment predict synaptic strength and function.
  • The ability to quantify synaptic activity in conjunction with ultrastructure could be applied to study synaptic plasticity, disease mechanisms, and the effects of various neuromodulators.

Summary updated

Abstract

Abstract Large-volume ultrastructural mapping approaches yield detailed circuit wiring diagrams but lack an integrated synaptic activity readout which is essential for functional interpretation of the connectome. Here we resolve this limitation by combining functional synaptic labelling in vivo with focused ion-beam scanning electron microscopy (FIBSEM) and machine learning-based segmentation. Our approach generates high-resolution near-isotropic three-dimensional readouts of activated vesicle pools across large populations of individual synapses in a volume of tissue, opening the way for detailed functional connectomics studies. We apply this method to measure presynaptic activity in an ultrastructural context in synapses activated by sensory input in primary visual cortex in awake head-fixed mice, showing that the numbers of recycling and non-recycling vesicles approximate to a lognormal distribution across a large number of synapses. We also demonstrate that neighbouring boutons of the same axon, which share the same spiking activity, can differ greatly in their presynaptic release probability.
BibTeX
@misc{simon2021ultrastructural,
  author = {Anna Simon and Arnd Roth and Arlo Sheridan and Mehmet Fişek and Vincenzo Marra and Claudia Racca and Jan Funke and Kevin Staras and Michael Häusser},
  title = {{Ultrastructural readout of <i>in vivo</i> synaptic activity for functional connectomics}},
  year = {2021},
  doi = {10.1101/2021.07.07.451278},
  howpublished = {bioRxiv},
}
Figure