Neuronal morphology and projection analysis by multispectral tracing in densely labeled mouse brain

Douglas H. Roossien, John M. Webb, Benjamin V. Sadis, Yan Yan, Lia Min, Aslan S. Dizaji, Luke J. Bogart, Cristina Mazuski, Robert S. Huth, Johanna S. Stecher, Sriakhila Akula, Jeff W. Lichtman, Takao K. Hensch, Erik D. Herzog, Dawen CaibioRxiv 2017 · DOI

TL;DR

Researchers have developed "nTracer," an ImageJ plugin that enables rapid and accurate reconstruction of neuronal morphology from densely labeled mouse brains. This tool addresses the long-standing challenge of analyzing complex neural circuits by overcoming limitations of traditional sparse labeling techniques, paving the way for more comprehensive understanding of brain function.

What Problem Does This Paper Address?

Understanding the intricate wiring of the brain, from large-scale projections to microcircuits, is crucial for deciphering brain function. Traditional methods for mapping these neural circuits rely on light microscopy and monochromatic labeling. However, to avoid ambiguity when tracing individual neurons, this approach necessitates "sparse labeling," meaning only a few neurons are labeled at a time. This limitation means our current understanding of neuronal morphology is often based on averaged data from many samples, potentially missing fine details and individual variations. While newer "Brainbow" technologies allow for dense labeling of many neurons simultaneously with distinct colors, a significant gap existed in the tools available to efficiently and accurately analyze the resulting complex, multi-colored datasets. This paper tackles this problem by introducing nTracer, a software solution designed to bridge this gap.

Methods

The core innovation presented is nTracer, an ImageJ plugin designed for semi-automated tracing of neurons in densely labeled Brainbow images. Key features and methodologies include:

  • Post-acquisition Image Processing: nTracer includes functions to correct for common imaging artifacts that arise when imaging deep into brain tissue. These include:
  • Color/Intensity Normalization: Addresses signal decrease due to scattering and photobleaching by matching histograms across image slices, ensuring consistent intensity ratios between fluorescent proteins.
  • Channel Registration: Corrects for spatial misalignment between spectral channels caused by optical imperfections, ensuring accurate color representation.
  • 3D Stitching: Merges overlapping image tiles to create a single, large dataset, essential for analyzing extensive neural networks.
  • Semi-Automated Tracing: nTracer employs a user-guided approach that leverages the unique color signatures of individual neurons.
  • Mean-Shift Algorithm: Accurately repositions user-defined start and end points onto neurites, compensating for imprecise mouse clicks and ensuring accurate color sampling.
  • A* Algorithm: Generates a least-cost path to trace neurites between user-defined anchor points, considering both spectral and intensity differences. This algorithm is adapted from previous work on tracing monochromatic images but is extended for multi-spectral data.
  • Data Structure: nTracer utilizes a flexible JTree structure to store tracing data for multiple neurons simultaneously, allowing for efficient management of complex datasets. This structure differentiates it from tools focused on single-neuron tracing.

Datasets

The paper describes experiments using Brainbow AAVs injected into various mouse Cre lines to label specific neuronal populations. The specific datasets used for analysis and demonstration include:

  • Cholinergic neurons in the dorsolateral striatum (ChAT-Cre).
  • POMC neurons in the arcuate nucleus of the hypothalamus (POMC-Cre).
  • Basket cells in the visual cortex (PV-Cre).
  • Granule cells in the dentate gyrus (POMC-Cre).
  • Vasoactive intestinal polypeptide (VIP) expressing neurons in the suprachiasmatic nucleus (SCN).

The paper does not introduce a new benchmark dataset but rather demonstrates the utility of nTracer on these biologically relevant datasets, highlighting its ability to handle large, densely labeled samples.

Key Insights and Ideas

  • Color Signature as a Tracing Key: The fundamental insight is that the unique combination of fluorescent protein expression (color signature) in densely labeled Brainbow samples can be reliably used to distinguish and trace individual neurons, overcoming the limitations of sparse labeling.
  • Addressing Imaging Artifacts is Crucial: The paper emphasizes that raw Brainbow images are often degraded by scattering, photobleaching, and optical misalignments. Developing robust post-acquisition processing tools is essential for unlocking the full potential of dense labeling.
  • Semi-Automation Balances Efficiency and Accuracy: While fully automated tracing is ideal, the complexity and variability of neural structures make it challenging. nTracer's semi-automated approach, with user guidance and intelligent algorithms, strikes a balance, reducing user burden while maintaining high accuracy.
  • Comprehensive Workflow: The paper presents a complete workflow from tissue preparation and imaging to post-processing and tracing, demonstrating a holistic solution for analyzing complex neural circuits.

Key Results

  • Rapid Tracing: nTracer can generate tracing results at an average rate of a few hours per neuron, significantly faster than manual tracing.
  • Accurate Reconstruction: The software is capable of rapidly and accurately reconstructing whole-cell morphology of large neuronal populations in densely labeled brains.
  • Versatility Demonstrated: nTracer was successfully used to analyze diverse cell types across multiple brain regions, including cholinergic neurons, POMC neurons, basket cells, granule cells, and VIP neurons.
  • Morphological Analysis: The tool enabled detailed quantitative analysis of neuronal morphology, such as dendritic length, primary branches, and bifurcations, revealing differences between neuronal populations (e.g., granule cells in the hilus vs. granule cell layer).
  • Projection Pattern Analysis: nTracer was used to map projection patterns of SCN VIP neurons, quantifying their divergence into the PVN and sub-PVN and revealing that the majority of neurites bifurcate minimally outside the SCN.

Limitations and Future Work

  • User Burden: While semi-automated, nTracer still requires significant user time for tracing complete datasets. The authors acknowledge that this is a limitation compared to fully automated methods, though they argue that the "on-the-fly" proofing and editing in nTracer mitigates the bottleneck of post-tracing error correction found in other semi-automated tools.
  • Optical Resolution Limits: In densely labeled samples, color blending between closely spaced neuronal processes can still limit the ability to map connectivity or follow projections back to the cell body, especially at the limits of confocal microscopy resolution.
  • Future Work: The authors are actively developing additional semi-automated and automated tracing algorithms for nTracer. They also plan to adapt nTracer to leverage the more versatile data structure of the ImageJ2 platform. They envision combining nTracer with emerging super-resolution microscopy techniques (like Expansion Microscopy) to achieve higher spatial resolutions and enable more detailed synaptic analysis.

Summary updated

Abstract

Abstract Accurate and complete neuronal wiring diagrams are necessary for understanding brain function at many scales from long-range interregional projections to microcircuits. Traditionally, light microscopy-based anatomical reconstructions use monochromatic labeling and therefore necessitate sparse labeling to eliminate tracing ambiguity between intermingled neurons. Consequently, our knowledge of neuronal morphology has largely been based on averaged estimations across many samples. Recently developed second-generation Brainbow tools promise to circumvent this limitation by revealing fine anatomical details of many unambiguously identifiable neurons in densely labeled samples. Yet, a means to quantify and analyze the information is currently lacking. Therefore, we developed nTracer, an ImageJ plugin capable of rapidly and accurately reconstructing whole-cell morphology of large neuronal populations in densely labeled brains.
BibTeX
@misc{roossien2017neuronal,
  author = {Douglas H. Roossien and John M. Webb and Benjamin V. Sadis and Yan Yan and Lia Min and Aslan S. Dizaji and Luke J. Bogart and Cristina Mazuski and Robert S. Huth and Johanna S. Stecher and Sriakhila Akula and Jeff W. Lichtman and Takao K. Hensch and Erik D. Herzog and Dawen Cai},
  title = {{Neuronal morphology and projection analysis by multispectral tracing in densely labeled mouse brain}},
  year = {2017},
  doi = {10.1101/230458},
  howpublished = {bioRxiv},
}

Loading related content...

Figure