GCIB-SEM: A path to 10 nm isotropic imaging of cubic millimeter volumes

Kenneth J. Hayworth, D. R. Peale, Michał Januszewski, Graham Knott, Zhiyuan Lu, C. Shan Xu, Harald F. HessbioRxiv 2019 · DOI

TL;DR

This paper introduces Gas Cluster Ion Beam Scanning Electron Microscopy (GCIB-SEM), a new technique that overcomes the limitations of existing 3D electron microscopy methods for imaging large biological volumes. By combining wide-area ion milling with thick sectioning, GCIB-SEM achieves high-resolution, isotropic 10 nm datasets of cubic millimeter volumes, paving the way for unprecedented studies of neural circuits and other complex biological structures.

What Problem Does This Paper Address?

Imaging the 3D ultrastructure of large biological volumes, such as entire nervous systems, at nanometer resolution is crucial for understanding complex biological processes like learning and memory, which are thought to be encoded in neural connectivity. Existing techniques face a trade-off:

  • Focused Ion Beam Scanning Electron Microscopy (FIB-SEM): Offers high z-resolution but is limited to small fields of view, making it incompatible with high-throughput imaging of large volumes.
  • Section-based techniques (e.g., ssTEM, ATUM-SEM): Compatible with fast, multibeam SEMs (like MultiSEM) for large-area imaging, but struggle with achieving the necessary z-resolution for automated segmentation due to limitations in section thickness and consistency.

This gap prevents researchers from comprehensively mapping neural circuits in volumes large enough to address fundamental questions about brain function.

Methods

The authors introduce Gas Cluster Ion Beam Scanning Electron Microscopy (GCIB-SEM), a hybrid approach that combines elements of FIB-SEM and section-based techniques. The core process involves:

  1. Thick Sectioning: Instead of cutting extremely thin sections, GCIB-SEM utilizes thicker sections (up to 10 µm).
  2. Electron Irradiation: These thick sections are first irradiated with a high-voltage electron flood gun to increase their conductivity, mitigating charging artifacts during subsequent imaging.
  3. Wide-Area Ion Milling: A Gas Cluster Ion Beam (GCIB) is used for top-down, wide-area milling of these thick sections. GCIBs consist of clusters of argon atoms, allowing for precise control over the milling energy and a gentler, more uniform removal of material compared to focused ion beams. This milling is performed in small steps (e.g., < 10 nm).
  4. SEM Imaging: After each milling step, the surface is imaged using a Scanning Electron Microscope (SEM), capturing raw image stacks for each thick section.
  5. Flattening and Stitching: A crucial post-processing step involves computationally "flattening" the bottom of each section (where milling breaks through to the substrate) and then stitching these flattened volumes together to create a continuous 3D dataset spanning the entire original volume.

Datasets

The paper utilizes datasets from:

  • Drosophila brains: Used for initial demonstrations of the GCIB-SEM process, particularly in Figure 1.
  • Mouse cortex: Used extensively for testing and demonstrating the capabilities of GCIB-SEM, including experiments with different sectioning methods (ultramicrotomy and hot knife sectioning) and varying irradiation doses.
  • Zeiss MultiSEM: A test sample was sent to Zeiss for imaging on their MultiSEM system to compare with GCIB-SEM imaging.

The paper does not introduce new benchmark datasets for training or evaluation but rather demonstrates the application of GCIB-SEM to existing biological samples.

Key Insights and Ideas

  • GCIB offers a gentler, wider-area milling solution: Unlike focused ion beams, GCIBs provide a more uniform and less damaging milling process suitable for large-area surface preparation.
  • Electron irradiation is key for conductivity: Overcoming charging artifacts in thicker sections is critical for high-quality imaging, and electron flood irradiation proves effective.
  • Computational flattening is essential: The "bottom index map" and subsequent stretching algorithm are vital for correcting uneven milling and enabling seamless stitching of sections.
  • GCIB-SEM bridges the gap: This technique successfully combines the high z-resolution of block-face methods with the throughput of section-based methods, making large-volume, high-resolution 3D EM feasible.
  • Compatibility with multibeam SEMs: The developed GCIB-SEM workflow is designed to be compatible with existing and future multibeam SEM systems, promising significant scalability.

Key Results

  • 10 nm isotropic resolution: GCIB-SEM successfully generates 3D datasets with isotropic voxels of approximately 10 nm.
  • Cubic millimeter volumes: The technique is demonstrated to be capable of imaging volumes spanning cubic millimeters, a significant leap from previous methods.
  • Improved z-resolution: GCIB-SEM improves z-resolution by a factor of 3 or more compared to traditional section-based techniques.
  • Reliable connectome tracing: The achieved resolution and data quality are sufficient for automated connectome tracing, enabling the study of neural circuits at an unprecedented scale.
  • Reduced artifacts: Compared to FIB-SEM, GCIB-SEM exhibits fewer milling artifacts in the direction of the beam.
  • High throughput potential: The wide-area milling and compatibility with multibeam SEMs suggest a path towards "industrial-scale" 3D EM.

Limitations and Future Work

  • InLens-SE signal drop: A slight drop in InLens-SE membrane contrast was observed during milling, attributed to a thin layer of disrupted surface material. This effect is quantified and suggests a recommended irradiation dose to minimize it.
  • Durcupan embedding limitations: Durcupan resin was found to be insufficiently resilient for sectioning very thick samples, leading the authors to switch to Spurr's embedding for later runs.
  • Flat embedding contamination: The hot knife sectioning method, while enabling thicker sections, resulted in some contamination from the flat embedding procedure.
  • Future Work: The authors envision a future where acquiring automatically segmented EM volumes of large tissue samples is routine, with automated shuttling of wafers between GCIB-milling and MultiSEM-imaging stations, similar to semiconductor fabrication plants. They also highlight the potential for further optimization of GCIB parameters and processing pipelines.

Summary updated

Abstract

Abstract Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) generates 3D datasets optimally suited for segmentation of cell ultrastructure and automated connectome tracing but is limited to small fields of view and is therefore incompatible with the new generation of ultrafast multibeam SEMs. In contrast, section-based techniques are multibeam-compatible but are limited in z-resolution making automatic segmentation of cellular ultrastructure difficult. Here we demonstrate a novel 3D electron microscopy technique, Gas Cluster Ion Beam SEM (GCIB-SEM), in which top-down, wide-area ion milling is performed on a series of thick sections, acquiring &lt; 10 nm isotropic datasets of each which are then stitched together to span the full sectioned volume. Based on our results, incorporating GCIB-SEM into existing single beam and multibeam SEM workflows should be straightforward and should dramatically increase reliability while simultaneously improving z-resolution by a factor of 3 or more.
BibTeX
@misc{hayworth2019gcib,
  author = {Kenneth J. Hayworth and D. R. Peale and Michał Januszewski and Graham Knott and Zhiyuan Lu and C. Shan Xu and Harald F. Hess},
  title = {{GCIB-SEM: A path to 10 nm isotropic imaging of cubic millimeter volumes}},
  year = {2019},
  doi = {10.1101/563239},
  howpublished = {bioRxiv},
}

Loading related content...

Figure