We describe the preparation of ribbons of serial sections and their collection on large transfer support for use as Array Tomography samples, along with automated imaging procedures in a scanning electron microscope. The protocol allows screening, retrieval, and targeted imaging of local, rare events, and the acquisition of large data volumes.
Electron microscopy is applied in biology and medicine for imaging of cellular and structural details at nanometer resolution. Historically, Transmission Electron Microscopy (TEM) provided insight into cell ultrastructure, but in the recent decade, the development of modern Scanning Electron Microscopes (SEM) has changed the way of looking inside the cells. Even though the resolution of TEM is superior when protein-level structural details are needed, SEM-resolution is sufficient for the majority of the organelle-level cell biology-related questions. The advancement in technology enabled automatic volume acquisition solutions such as Serial block-face imaging (SBF-SEM) and Focused ion beam SEM (FIB-SEM). Nevertheless, to this day, these methods remain inefficient when the identification and navigation to areas of interest are crucial. Without the means for precise localization of target areas before imaging, operators need to acquire much more data than they need (in SBF-SEM), or, even worse, prepare many grids and image them all (in TEM). We propose the strategy of “lateral screening” using Array Tomography in SEM, which facilitates the localization of areas of interest, followed by automated imaging of the relevant fraction of the total sample volume. Array tomography samples are conserved during imaging, and they can be arranged into section libraries ready for repeated imaging. Several examples are shown in which lateral screening enables us to analyze structural details that are incredibly challenging to access with any other method.
Despite the importance of EM-related techniques, the effort required to master them keeps the entire field restricted to a small number of specialists. One significant difficulty is the identification and retrieval of a Region of Interest (ROI) in the samples preserved for EM. The appearance of the same sample considerably differs when analyzed by optic microscopy and after processing for EM observation. The changes for chemically prepared samples include the anisotropic sample shrinkage after the dehydration steps (~10% in each dimension) and the loss of fluorescence when using osmium in the fixation and staining protocol (Figure 1A). For the ultrathin sectioning, the samples are embedded in epoxy or acrylic resins using different strategies (Figure 1B). For successful results of this preparation, the entire sample must be fractionated into pieces that do not exceed 1 mm x 1 mm. To meet standard Transmission Electron Microscopy (TEM) observation conditions' requirements, this tiny portion of the sample is further sectioned to 50-150 nm thick slices. The resulting grayscale images show tissue organization and organelle structure of a minute fraction of the entire sample at greater detail than any other microscopy technique (Figure 1C). A typical TEM dataset provides 2D information, theoretically extrapolated to understand the processes naturally occurring in a 3D space in cells and tissues. Figure 1D presents the challenge of ultrastructural volumes acquisition: if a cube of side 1,000 µm is sectioned at 50 nm thickness, 20,000 sections will be required to cover the entire volume; for a 500 µm side cube, it will be 10,000 sections. To cover a 50 µm x 50 µm x 50 µm volume, 1,000 sections "only" might be necessary. Obtaining this volume manually is practically impossible and extremely challenging to perform with automation. If, in addition to sample depth, we need to cover the entire surface of such hypothetical cubes, the coverage of 1 µm2 surface at reasonable resolution becomes a serious logistic issue (Figure 1E). While for the extraordinary large-scale projects, such as connectomics approaches, the large number of sections is crucial, for the majority of the "mundane" EM projects, generating more sections required for the observation presents a significant disadvantage.
There are several methods to acquire 3D ultrastructural information: serial sectioning Transmission Electron Microscopy (TEM), TEM tomography, Array Tomography (AT), Serial Block Face Imaging Scanning Electron Microscopy (SBF-SEM), and Focused Ion Beam Scanning Electron Microscopy (FIB-SEM). Principal differences between these methods are the sectioning strategy and whether the image acquisition is coupled to section generation1. In serial sectioning TEM, sequential sections are collected on slot grids, TEM images are generated from these sequences and aligned2,3,4,5. In TEM tomography, tilt series from 150-300 nm sections on a grid, and when coupled to serial sectioning, provide a very high resolution, though relatively small volumes6,7,8. The AT approach uses physical sectioning with diverse manual and semi-automatic manners of sections collection on relatively large support, such as glass coverslip, silicon wafers or a special tape. For image acquisition, the support is analyzed in SEM, with diverse image acquisition strategies are available9,10,11,12,13,14,15 . For SBF-SEM, physical sectioning is achieved using a mini-microtome with a diamond knife set directly inside the SEM chamber, with the SEM image generated from the surface of the resin block16,17,18,19. For FIB-SEM, ion source removes thin layers of the sample, followed by automatic imaging of the exposed surface by the SEM20,21. TEM-tomography and the AT generate physical sections, which can be re-imaged if necessary, while FSBF-SEM and FIB-SEM eliminate the section after imaging. A recent combination of physical sections imaged by a multi-beam SEM provides a combination of methods that solves the "bottleneck" issue of the speed of image acquisition22. Each of these techniques has revolutionized the way EM data can be obtained and analyzed, and each approach has its practical impacts related to a given research question.
Given the nature of the preparation and the scale of the ultrastructural dimensions, it is not straightforward to predict where a specific target structure is located in the sample block (Figure1D,E). One solution for the ROI localization is recording the images from the entire block at the desired resolution from the beginning. The structures of interest can be in the acquired data volume when away from the microscope. Acquisition time and data handling associated with this strategy are problematic. It is desirable to reduce the amount of recorded data, especially if the ROIs are much smaller than the tissue block, i.e., if the objects of interest are specific types of cells (not whole organs). Different correlative light and electron microscopy techniques (CLEM) can be successful when the fluorescence is preserved and localized before or after preparation within the same sample23,24,25,26,27,28,29. Nevertheless, many cellular structures are recognizable even without fluorescence correlation, only based on the known ultrastructure. For these cases, we believe that lateral screening Array Tomography provides a balance tradeoff between the effort invested in ROI localization and the ultrastructural information quality. Using this strategy, a sub-set of sections on the wafer are screened within regular intervals, which can be established based on the ROI's size and nature. Once ROIs are found, data acquisition is set up in a continuous series of sections starting before and ending after the anchor section, collecting the relevant information in a targeted manner.
We present protocols for AT that simplify and accelerate the acquisition of regions or events of interest in numerous sections and yield better-aligned image volumes. Lateral screening and multistep acquisition produce data with very high resolution in precisely targeted regions. The procedure we describe addresses several challenges of 3D EM data acquisition, as it provides for: compatibility with a broad range of specimens without fundamentally changing the sample preparation workflow; targeted localization for sectioning and SEM acquisitions; reduced time and effort during setup; imaging of regions in multiple sections with better alignment of the resulting volumes; and a smooth stitching and alignment procedure to compile different images into a stitched mosaic picture. We chose to demonstrate the strength of our method with several samples from published and ongoing projects. We believe that this approach can significantly facilitate the generation and acquisition of targeted EM data, even for investigators with limited EM experience.
Electron microscopy gives insight into the ultrastructure of the cells and organisms, for which it is often desirable to image structures of interest in their 3-dimensional context. Despite numerous EM tactics for ultrastructural analysis, there is still no "gold standard" solution. The main reason is the wide variety of samples, many biological questions, which frequently require a tailored approach. The proposed AT workflow is designed to minimize the time necessary for sample processing, data acquisition, evaluation, and storage. Furthermore, the modified knife provides a helpful tool to simplify array acquisition. The compact layout of sections on wafers is convenient, both for the observation and the subsequent storage of the samples. This arrangement enables "Lateral Screening" of samples by horizontally moving from ribbon to ribbon and scanning only one section on each, significantly reducing the time required to localize an ROI. Proposed data acquisition scenarios facilitate the targeting of small and randomly distributed areas. Once found, AT/SEM restricts high-resolution imaging precisely to the volume of interest, whether performed manually or with the help of automatic function. AT for limited volumes can be completed manually, with the operator navigating through the sample and defining imaging regions one by one. The automated module of the software provides a flexible image acquisition strategy for imaging of small areas on large sections. The automation in this software allows recording large high-resolution images on hundreds of sections, achieving similar volumes as SBFI. Recording overview images of all sections simplifies the ROI localization and reduces the time spent at the microscope. Since the sections are not harmed during the recording of overviews and higher resolution Previews, AT/SEM permits reusing the sample to collect further data of other ROIs or at a higher resolution.
Image acquisition time is one of the most important (and most expensive) aspects of 3D EM and should therefore be considered in the experiment design. While it is unsurprising that imaging large areas takes longer than imaging small areas, it is easy to under-estimate the impact: Depending on the imaging parameters selected, the acquisition time on each section can vary from seconds to hours. Critical imaging parameters include the size of the field of view, resolution, and dwell time. Assuming a target resolution of 10nm per pixel and 1 µs dwell time, imaging a field of 20 µm x 20 µm, 100 µm x100 µm, or 500 µm x500 µm takes 4 seconds, 100 seconds, or 2,500 s to record. We can multiply these per-section imaging times by the number of sections to estimate the time required for the complete imaging job. Long per-section imaging times can be acceptable if the number of sections is small or if microscope tool time is of no concern.
However, it is necessary to limit the recording time to an overnight job or a weekend job in most cases. An equally critical aspect of 3D EM that should be considered, is the amount and structure of the resulting image data. Recording the abovementioned imaging fields in 100 sections generates 400 mb, 10 gb, or 250 gb of image data, respectively; the 500 µm x 500 µm images pose the additional issue of being larger than 2 gb each. Many of the software programs used for data evaluation cannot open images of this size.
To reduce imaging time, it is important to choose pixel dwell time to meet the signal-to-noise ratio requirements for the subsequent data evaluation (e.g., reconstruction, tracing), and limit the recordings to defined ROIs. The AT extension of the software facilitates image acquisition in small areas in serial sections. The software supports manual and automatic workflows and many semi-automatic variants: imaging areas can be manually positioned and focused on each section, or the user can use the automatic section finder and position alignment features. Depending on the level of automation chosen and supported by the sample type or imaging goals, the time required for setting up image acquisition in hundreds of sections can take an entire workday (done manually) or only a few minutes. In principle, Array Tomography makes it more challenging than other 3D EM methods to acquire small ROIs; imprecise region placement on consecutive sections must be compensated by acquiring larger areas. For example, if the ROI is 20 µm x 20 µm in size and the section-to-section position variability of imaging fields is 10µm, one needs to acquire 40 µm x 40 µm images in order to be sure that the ROI is fully captured in each image, on every section. Real-world image position variability ranges from 100 µm to <10 µm depending on the availability or quality of software features for position alignment or user's patience. With this software, 10 µm can be achieved without too much manual intervention in most samples.
Like any technique, the AT has several weak points that can influence successful data acquisition, and many are similar to other sectioning-based methods.Lack of homogenous distribution of empty resin versus tissue can result in curved or broken arrays. In extreme cases, sections can detach from the support (Figure 6A). Variable compression or stretching during the cutting process can create folds that can disrupt the sample in variable regions on subsequent sections (Figure 6B). Knife marks can appear on the surface of the collected sections using a damaged knife (Figure 6C). Differences in sectioning conditions can induce occasional section compression and thickness differences. Dust or dirt particles can land on a section and partially obscure the zone of interest (Figure 6D). Image acquisition can fail due to imperfect auto-contrast, auto-focus, and auto-stigmator functions. The positioning of automatically created imaging areas can be variable and may fail to capture the ROI in all sections.
Several problems can arise at the stage of stitching and alignment. The automatic stitching of mosaic tiles acquisitions can fail, for example, because of the large empty space inside the sample. Due to a drastic change in shape in 3D, image stacks can be challenging to register. Specially developed programs (e.g., IMOD, Fiji, TrackEM2, MIB, or MAPS-AT) can facilitate semi-automatic alignment32,38,39,40. More challenging sections can be manually aligned using photo editing software. Unfortunately, some datasets may be impossible to align correctly.
Large samples are challenging for TEM serial sections fitting onto grids; on the other hand, many projects do not justify long automatic acquisition using FIB/SEM or SBF-SEM. The AT is a straightforward alternative to a tedious serial section TEM where collection and manipulation of serial sections on a wafer are more straightforward than with the slot grids. Several strategies were developed to facilitate the collection of the arrays, and we share our method to expand the existing toolkit. In cases where the identification of the ROI is challenging, AT-SEM provides a fundamental advantage, with the efficient screening of the samples where organelle-scale resolution across 50 to 500 sections is required. For larger volumes, the automatic collection AT strategies can be efficiently collected if more sections are required. AT samples can be re-imaged multiple times, facilitating targeted imaging of high-resolution areas based on previously-acquired overview images. We believe that the targeted analysis and reduced oversampling by AT/SEM proposed here decreases the labor and data storage requirements. Ultimately, libraries of sections can be collected and maintained for later reuse and consultation. For volume acquisition, FIB or SBF-SEM approaches offer an excellent solution whenever the ROI is easy to identify on the block face or if large 3D volumes are needed for the analysis. However, FIB/SBF-SEM are less efficient when the high-resolution stack image has to be collected from a defined ROI in a targeted manner. To conclude, the proposed methods for screening of AT samples and the use of medium resolution overview images allow limiting image acquisition to the relevant parts of the section array. Precise aiming of imaging regions speeds up time-to-data and simplifies data evaluation.
In summary, although the concept of AT/SEM is not novel, its use is still not as widespread as its merits would suggest. Overall, it provides a complementary procedure to other existing EM methods. AT/SEM is compatible with the broadest range of sample preparation protocols and imaging workflows and can be done on any FIB/SEM or SBF-SEM microscope as an accompanying technique. In this paper, we have concentrated on AT for recording ultrastructural data from samples that are less likely to be successfully addressed by other methods. We hope that the described procedure for convenient collection of sections and considerably automated acquisition strategies will aid in the first attempts for those who have never encountered the method and will help to perfect it for those who already have some experience.
The authors have nothing to disclose.
We want to thank the members of the EM facility of the University of Lausanne for their support during this development of different steps of the AT procedure. We would like to thank Gareth Griffiths, Marta Rodrigues, Urska Repnik, Christel Genoud, Helmut Gnaegi, Einat Zelinger, Paola Moreno-Roman, Lucy O'Brien and Lindsay Lewellyn for discussions during the preparation of the manuscript and critical reading. We want to acknowledge the groups that contributed the samples used to demonstrate the different scenarios: Matthias Lutolf, Michail Nikolaev, Devanjali Dutta, Till Matzat, and Fanny Langlet.
Cutting | |||
AT sectioning knife | Diatome | DUATS3530 | Diatome Jumbo knife |
Diamond knife for trimming 90° | Diatome | DTB90 | Diatome trimming 20° Glass knife |
Pattex contact adhesive | Pattex | PCL3C | |
Silicon wafer | Ted Pella | 16015 | Resistance: 1-30 Ohms Type P: (Boron) (1 primary flat) Roughness: 2 nm No SiO2 top coating TTV: = <20 µm Wafer is polished on one side |
Ultramicrotome | Leica UC6 | Alternative: Leica UC7 | |
Wafer cleaving kit | EMS | 7642 | EMF, Small Sample Cleaver, CatNo. 7652 |
Image acquisition | |||
FESEM | Thermo Fischer Helios | 1072419 | Alternatives: Zeiss, Jeol, Hitachi, TESCAN |
Maps 3 for SEM with Correlative Workflow & Array Tomography | Thermo Fisher Scientific | 1135932 | Maps provides automation of SEM imaging workflows and allows importing of 3rd party data for CLEM and navigation. |
Image analysis | |||
Amira x.y | Thermo Fisher Scientific | 1131599 | Amira is a 3D data visualization and analysis software with several practical functions for Array Tomography data reconstruction. |
Image processing | Open source | Fiji (http://fiji.sc/#download) | IMOD, MIB (See text for refferences) |