Visualization of Organelles In Situ by Cryo-STEM Tomography

Summary

Cryo-STEM tomography provides a means to visualize organelles of intact cells without embedding, sectioning, or other invasive preparations. The 3D resolution obtained is currently in the range of a few nanometers, with a field of view of several micrometers and an accessible thickness in the order of 1 µm.

Abstract

Cryogenic electron microscopy (cryo-EM) relies on the imaging of biological or organic specimens embedded in their native aqueous medium; water is solidified into a glass (i.e., vitrified) without crystallization. The cryo-EM method is widely used to determine the structure of biological macromolecules recently at a near-atomic resolution. The approach has been extended to the study of organelles and cells using tomography, but the conventional mode of wide-field transmission EM imaging suffers a severe limitation in the specimen thickness. This has led to a practice of milling thin lamellae using a focused ion beam; the high resolution is obtained by subtomogram averaging from the reconstructions, but three-dimensional relations outside the remaining layer are lost. The thickness limitation can be circumvented by scanned probe imaging, similar to the scanning EM or the confocal laser scanning microscope. While scanning transmission electron microscopy (STEM) in materials science provides atomic resolution in single images, the sensitivity of cryogenic biological specimens to electron irradiation requires special considerations. This protocol presents a setup for cryo-tomography using STEM. The basic topical configuration of the microscope is described for both two- and three-condenser systems, while automation is provided by the non-commercial SerialEM software. Enhancements for batch acquisition and correlative alignment to previously-acquired fluorescence maps are also described. As an example, we show the reconstruction of a mitochondrion, pointing out the inner and outer membrane and calcium phosphate granules, as well as surrounding microtubules, actin filaments, and ribosomes. Cryo-STEM tomography excels in revealing the theater of organelles in the cytoplasm and, in some cases, even the nuclear periphery of adherent cells in culture.

Introduction

Three-dimensional (3D) visualization of organelles is a paramount task in modern cell biology. Given the scales involved, ranging from tens of nanometers for secretory vesicles to many microns for the cell nucleus, it is challenging to find a single microscopy technique to fit all applications. While modern fluorescence microscopy can span much of the range in terms of resolution, only the labeled molecules appear. The cellular theater remains the realm of electron microscopy. Conventional methods of chemical fixation, plastic embedding, and staining with heavy metals are strongly invasive, so the results may depend on the details of sample preparation. Cryo-EM, on the other hand, is constrained by the need to vitrify the aqueous medium; ice crystals that form diffract the electron illumination, causing contrast artifacts of higher contrast than the organic material of interest.

The past decade has seen a proliferation of EM imaging techniques developed or adapted for cellular studies¹. High-pressure freezing combined with iterative focused ion beam (FIB) milling and serial surface imaging using the scanning electron microscope (i.e., FIB-SEM) is currently the method of choice for large specimens². Cryogenic soft X-ray tomography (cryo-SXT) is suitable for samples of several microns in size, limited by the characteristic absorption length of the soft X-rays in water^3,4,5. This scale includes many intact cell types, and the quantitative nature of the X-ray absorption contrast adds an aspect of concentration measurement⁶ or spectroscopy⁷. When combined with subtomogram averaging, cryo-transmission electron tomography (cryo-ET), based on phase contrast transmission electron microscopy (TEM), offers the highest resolution for macromolecules or complexes^8,9,10. However, it is rare that intact organelles are so regular that they can be averaged whole. Moreover, the conventional mode of wide-field TEM is limited for specimen thickness to a few hundred nanometers by the combination of inelastic scattering (involving energy loss) in the specimen and chromatic aberration in the magnetic objective lens^11,12. The large energy spread dictates the use of an energy filter to remove the resulting out-of-focus haze, but the sensitive specimen still suffers radiation damage while the image signal becomes exponentially weaker with increasing thickness.

The alternative imaging mode, scanning transmission EM (STEM), circumvents the need for energy filtering and retains the inelastically scattered electrons for image formation, albeit currently at a lower resolution than for TEM tomography (Figure 1). In fact, no real image is formed. Instead, as in a scanning EM, measurements are made point by point, and the image is assembled by the computer. The magnification is determined only by the size of the scan steps without changing the lens currents. When properly configured, the useful range of specimen thickness for cryo-STEM tomography (CSTET) can reach 1.5 or even 2 µm, though the comfort zone, where the useful signal intensity remains a significant fraction of the illumination, is around 600-900 nm^11,13. This is sufficient to see a large fraction of the cytoplasm, and occasionally an edge of the cell nucleus. In practice, we find that vitrification by the standard method of plunging into cryogenic fluid imposes a more severe constraint on thickness than STEM imaging. The goal of this video article is to facilitate the incorporation of CSTET into the tool chest for cell and organelle imaging in research labs and microscopy facilities.

The first challenge is that microscope operations in CSTET are not yet standardized for life science applications in the way that they have been for cryo-TEM tomography. STEM hardware has rarely (if ever) been targeted to the cryo-EM market. However, this is changing with the newest generation of microscopes, and many existing tools can be retrofitted. STEM as a technique has taken off and largely taken over in the materials sciences, where there is also budding interest in cryogenic and low-dose methods^14,15. The materials science literature abounds with acronyms BF-STEM, ADF-STEM, HAADF-STEM, 4D-STEM, DPC-STEM, etc., which add to the confusion. We offer here a recommended starting point that, in our collective experience at the Weizmann Institute of Science, provides the most general protocol for useful results based on bright field (BF) STEM imaging¹⁶. In no way does it exhaust or even explore the range of possibilities, but it will serve as a basis for further enhancements. While we emphasize cryo-STEM, most of the protocol is equally relevant for room-temperature STEM tomography of plastic-embedded sections.

The essence of STEM is to scan the specimen with a focused electron probe (Figure 1), the illumination cone, and to record signals from the diffraction (scattering) plane in transmission, pixel by pixel, to produce 2D images^17,18. Amorphous specimens, including most cellular materials, will produce a diffuse scattering pattern in transmission. The simplest practical STEM configuration is to place a circular detector to record the central disk (i.e., the probe illumination that would be transmitted without a specimen). The specimen scatters electrons away from this illumination cone to the extent that the signal decreases. This produces a BF image-the specimen appears dark on a bright background. An annular detector may also (or instead) be used to detect the scattering from the specimen outside the illumination cone. With the specimen removed, there is no signal. When a specimen is in place, objects appear bright on a dark background in the dark field (DF) image. The nomenclature for STEM (BF, annular dark field [ADF], high-angle annular dark field [HAADF], etc.) refers mainly to the ranges of collection angles for the detectors.

The convergence angle of the illumination represents an essential adaptation of STEM to cellular tomography. When the top priority is high resolution, the convergence angle should be as large as possible. (This is similar to confocal laser scanning microscopy; the resolution is determined by the probe diameter, which scales as the wavelength divided by the numerical aperture. Note that we refer to the half-angle or semi-convergence angle for EM.) When the priority is the depth of field, on the other hand, a compromise in resolution affords a great advantage, as the focused beam remains roughly parallel for a distance equal to twice the wavelength divided by semi-angle squared. Ideally, the entire cell volume remains in focus¹⁹. For example, at 300 keV, the electron deBroglie wavelength is 0.002 nm, so a convergence of 1 mrad yields a resolution of 2 nm and a depth of field of 4 microns. Under these conditions, tomography can be performed even without focusing during the data collection process, but only once at the beginning of the acquisition. A conventional tomography-capable STEM can reach a semi-convergence angle of 7 or 8 mrad; therefore, in principle, we could reach a resolution in the order of 0.25 nm, but then with a focal depth of only 62 nm. This is clearly too thin for cellular imaging. More advanced microscopes with three condenser lenses offer continuous adjustment of the semi-convergence angle over a considerable range. With the more traditional two-condenser configuration, the convergence is fixed discretely by the condenser (C2) aperture.

For robust, plastic-embedded samples, one can record a focal series at each tilt and combine them for high resolution²⁰, but for cryogenic specimens, the radiation budget is too severely constrained. Finally, in weighing the advantages of BF or DF imaging, for thick specimens, one should consider the effects of multiple elastic scattering in the specimen. The BF signal is less corrupted by multiple scattering and shows a higher resolution for thick specimens^16,21.

A useful rule of thumb has been to set collection angles several times larger than the convergence. The thicker the specimen, the larger the collection disk should be. Too small a disk will provide a low signal intensity; too large a disk will result in poor image contrast, as only the highest-angle scattering will contribute. The collection angles should be optimized for a given sample. The detector angles as a function of (diffraction) camera length must be calibrated independently. They may be displayed conveniently by the microscope software. In practice, a factor of two to five in the ratio of collection to illumination semi-angles, θ to α, respectively (Figure 1), is a recommended starting point for CSTET of cellular specimens.

The following protocol describes STEM tomography operation using the popular SerialEM software for microscope control^22,23. SerialEM is not tied to a specific manufacturer, and it is widely used in TEM tomography. Most of the operations in setting up for tomography can be carried over directly from TEM. The SerialEM strategy is to model the scanning system as a camera. This enables the simple crossover from TEM to STEM. One should keep in mind, though, that parameters such as magnification and binning are entirely artificial. The important parameters are the field of view in microns, the number of pixels in the field of view, and the exposure time. The pixel spacing, or sampling, is the linear field divided by the number of pixels, while the dwell time is the number of pixels divided by the exposure time.

The minimum configuration for STEM and CSTET involves three features on the microscope: a scan generator, a STEM detector, and tomography control software. The protocol refers to the nomenclature of FEI/Thermo Fisher Scientific (TFS), but the concepts are generic. The proprietary software of TFS has been described in JoVE for TEM²⁴, and the STEM operation is very similar.

We assume that the microscope has been aligned in advance by the service team or experienced staff and that a column alignment can be called up by loading a file. Minor adjustments are called direct alignments and can be stored in so-called FEG registers (TFS microscopes). Direct alignments include rotation center, pivot points, diffraction alignment, and compensation for condenser astigmatism. Adjustments have to be performed iteratively. Note that TFS microscopes implement distinct nanoprobe (nP) and microprobe (µP) modes; for a given condenser aperture, these provide a relatively narrow or wide field of view with parallel illumination in TEM and a more or less convergent (tightly focused) electron beam in STEM, respectively. Other manufacturers use different schemes to cover the range of convergence angles.

Before starting, the field of view, L, and the sampling (pixel width), l, should be chosen, depending on the sample under study. For example, for l = 1 nm/pixel, a 4,000 x 4,000 pixel image that will cover a field of view 4 µm² should be chosen. The resolution will be, at best, twice the spatial sampling, so 2 nm, and the probe diameter, d, should match that. Calibration of the probe angle is beyond the scope of this protocol, so we assume that a table or a screen reading is available. The probe diameter is approximately the electron wavelength divided by the semi-convergence angle (in radians): d = λ/θ. The wavelength, λ, is 0.002 nm for 300 keV and 0.0025 nm for 200 keV electrons, so θ of 1 mrad will provide a spot diameter of 2 or 2.5 nm, respectively.

The protocol is presented in a progression of increasing complexity. The first task is to produce a STEM image, which depends on the microscope manufacturer's software, and then a tilt series, for which we use SerialEM. Many readers will undoubtedly be familiar with SerialEM, so the more complicated tasks will come naturally. There is no need to follow the procedures strictly. Developments relating to automation may be implemented directly for STEM as well as for TEM. Experienced users will likely invert the protocol, beginning with correlative registration of fluorescence maps and continuing to set up batch tomography. Further details can be found in the extensive documentation and tutorial libraries for SerialEM itself, including a recent JoVE article on the latest developments in automation²⁵.

Protocol

1. STEM setup

1. Preliminary alignments: Load the column alignment file, and then open Column Valves. If using a side-entry cryo-holder, open Cryo-shield. Start in TEM mode. The beam should appear on the screen. If not, lower the magnification.
2. Bring the microscope to eucentric focus by pressing the button on the control panel.
3. Set the spot size to a convenient value (e.g., 6) for visualizing the fluorescent screen, either directly or with the built-in camera (depending on the microscope model).
4. Set the microscope to STEM mode and verify that focus uses the condenser lenses (intensity) rather than the objective. Set eucentric focus. Then, go out of diffraction mode for initial adjustments.
5. Make sure the beam is not blanked. Reduce the magnification until the beam appears on the screen. Adjust the beam shift to the center and increase the magnification in steps up to about 70kx while keeping the beam in the center.
6. Insert the desired condenser aperture, typically 50 µm. Check the aperture centering. When turning the focus knob slightly back and forth, the spot should expand and contract but remain in place, as if a plane cuts an imaginary vertical hourglass. If the aperture is not centered, the illumination will shift laterally, as if the hourglass were tilted.
7. Bring the beam to focus and press Intensity List Focus (if available) in the alignments tab, or return to eucentric focus as in 1.2. Readjust the beam position to the center.
8. Adjust the rotation center. Turn the focus step wheel to the minimum or one step above, so that the beam pulses gently, and ensure that it remains stationary as the focus moves up and down.
9. Select the pivot points and bring the two points together with X and Y adjustments.
   NOTE: If a phase plate is installed on a TFS instrument, only the X adjustment should be used.
10. Defocus the beam slightly, and adjust the condenser stigmators to make the disk round. Go up and down through focus to optimize; there should be no tendency to elongate in one direction or the other when passing focus.
11. Normalize the lenses. Then, increase the magnification progressively to about 240kx while using the beam shift to keep the spot centered, and repeat the rotation center and pivot point adjustments (steps 1.6-1.8).
12. Return to diffraction mode. The beam should appear as a uniform disk on the fluorescent screen. The camera length (CL) now effectively controls the optical distance to the detector, as in X-ray crystallography. Change it and watch as the disk contracts and enlarges, as if the screen location would move toward or away from the specimen. This cone represents the BF illumination.
    NOTE: For each CL, there is a diffraction alignment to be adjusted in order to bring the projected beam to the central axis. There is normally no need to refine them all, only the one (or few) to be used for detection.
13. Engage STEM mode at high magnification.
    1. If a BF STEM detector is available, start with that. Bring the stage to an area with an empty hole and adjust the diffraction alignment to center the beam using the desired CL.
    2. Many microscopes are equipped only with a HAADF detector; there is a trick to use this as a BF detector. First, retract the detector, center the beam as above in the direct alignments, insert an objective aperture, typically a small one such as 20 µm, and adjust its position to surround the beam uniformly. (The easiest way to do this is by inserting a specimen or a test grid to see a halo around the illumination). Then, use the diffraction alignment to move the beam off axis and onto the detector area. Insert and retract the detector while watching the beam on the screen in order to confirm that the beam is blocked by the detector.
    3. Start a scan in the microscope software and adjust the brightness and contrast (B/C) settings so that the signal is close to 0 when the beam is blanked, and close to saturation when the full illumination falls on the detector. Use the scope display to assist. The settings may be coupled in unexpected ways, so the adjustment should be iterated several times.
14. Return the microscope to a relatively low magnification in the high magnification register (SA mode), without going into low mag (LM) mode.
15. Note the screen current for reference later (see step 3.1.1). The current can be changed with the gun lens and spot size settings, as in TEM, with increasing numbers corresponding to a reduced current.
16. At this point, save a FEG register to facilitate a return to standard values.
    NOTE: The facility can prepare a default set of FEG registers so that users can start from a working configuration.

2. Placing the specimen

1. In LM TEM mode, ensure that the grid is not opaque. Find a grid square for making initial adjustments. It is convenient to find some empty area, a hole, or a torn grid square. Record the stage positions in order to return to them conveniently.
2. Bring the sample to eucentric height. There are several methods to do this. For example, use the stage wobbler to tilt the grid while moving the specimen height along the Z axis until the image stops shifting laterally. Alternatively, mark some features on the viewing screen and tilt the stage to 30°. The feature will move laterally. Adjust the specimen height to bring it back to the original position. Increase the magnification or stage tilt to refine. Return to 0° tilt.
3. Return to STEM mode and insert the STEM detector. Go to the lowest high magnification (SA) mode. Ensure that an image is observed on the computer screen. Scan rapidly in order to avoid unnecessary exposure; 1 µs/pixel or less would be typical. Note that the image may appear distorted at the left border when the scan is excessively fast (see Figure 2).
4. Press Eucentric Focus, increase the magnification, and refine the focus while scanning. It is convenient to use the focus loupe provided by the microscope software.
5. Tune the condenser astigmatism. This can be done most precisely by the Ronchigram method. With the beam over a thin sample area, focus on the point where the transmitted beam blows up, in between shadow images of the sample on either side. Then, adjust the condenser tuning to make the central disk round. This requires some practice, especially for cryogenic samples.
   ​NOTE: An alternative is to find some gold particles (often used as fiducial markers for alignment in tomography). Increase the magnification and focus up and down, tuning the astigmatism so that the particles remain round without elongation in either direction.
6. Go to LM STEM mode and continue scanning to find an interesting area. Readjust the detector B/C settings if necessary (step 1.13.3), roughly at this point.

3. Recording an image

1. Estimate the dose for recording. As a rule of thumb, aim for 100-150 e^-/Ų for the whole tomogram. Check the dose tolerance per specimen. The current in Ampere (A) divided by the charge per electron represents a count of e^-/s, which is multiplied by the exposure time, T, in seconds and divided by the field of view in Ų. For convenience, express the current in nanoamperes (as read from the screen), the width of the field of view in micrometers, and the frame exposure time in seconds, with suitable factors:
   
   This number should be multiplied by the number of tilts to get the total exposure in the series. For example, with a current I = 0.04 nA, a field L = 4 µm, and an exposure time of 12 s, we obtain 1.9 e^-/Ų per tilt, or about 110 e^-/Ų for a series of 61 projections.
2. Return the stage to a hole (or retract the grid) and adjust the beam current using spot size and/or gun lens settings to reach the desired screen current. If the measurement reads 0, insert a larger C2 aperture to increase the current. Then, correct by dividing the measurement by the square of the ratio of diameters, and finally return to the smaller aperture.
3. Refine the B/C settings according to step 1.13.3, and finally return the stage to an area of interest and recheck direct alignments and astigmatism under the imaging conditions.
4. Acquire a test image.

4. Tomography with SerialEM

1. Start the SerialEM server on the microscope computer (SerialEM is normally installed on a different one). Then, open SerialEM. STEM appears in SerialEM like a camera, with the same interface. Make sure that communications are running appropriately for the setup. For example, the microscope tab should display the correct magnification. If this does not work, stop here and troubleshoot.
2. Find the camera and script tab and open Setup. Ensure that the STEM camera is selected, and for each mode choose suitable binning and dwell times. Ensure that the appropriate detector (or detectors) appears in the "channels to acquire" box at the lower left. Press Acquire at the bottom to test. Again, do not continue if nothing happens, or if the beam does not unblank. Stop and check the communications.
   NOTE: The modes are used in a similar way to TEM tomography. Binning is an artifice for compatibility with TEM. The important parameter is the pixel count; different microscope systems use different binning to reach the same count.
   1. View and Search should be fast scans with relatively few pixels (e.g., 512 x 512 pixels at 1 s).
   2. Focus is generally a small area with fast recording. Thus, in low dose mode, choose Focus to be at a different magnification to Record mode, which can aid in accuracy. Leave the spot size unchanged.
   3. For Record mode, use the parameters chosen above in the exposure estimation. If available, also select Dynamic Focus, which corrects the focus line by line in a tilted image.
   4. For convenience later, choose the binning for preview to be identical to that for View (see Enhancement #1, step 5.5). The image size (number of pixels) need not be the same.
   5. Use Trial mode to keep the Record area centered during acquisition. It can be similar to View, but do take care that there are no scan distortions apparent on the left side of the frame (see Figure 2 for an example). Again, in Low Dose mode, the Trial can have a different magnification than in Record mode.
   6. Use Montage mode immediately for the grid scan. It should be sufficiently pixel-dense to find areas of interest; recommendation: 512 x 512 or 1024 x 1024. Make sure that the image is good, with a decent dynamic range (seen in image display controls), as this mode is used for finding the specimen areas.
   7. Save the settings file so that these choices are kept as default for the next session (which may occur soon in case the program crashes).
3. Test the image shift and stage shift calibrations. In the microprobe (or nanoprobe) mode, click on View Image, hover over it with the mouse, hold the button down, and drag diagonally by about a quarter of the field of view. Then, take another View. The images should overlap perfectly. Repeat by dragging again while keeping the Shift key pressed (to enforce a stage movement). The images should overlap well, though perhaps less precisely.
4. Go to LM mode and test the stage shift. If these tests fail, stop and troubleshoot. The rest will not work.
5. Make a montage LM map of the whole grid. This can be done in TEM as well.
   1. Go to the lowest magnification STEM mode where the image is not blocked by apertures, typically around 185x.
   2. Click Navigator menu > Open. Then, click Navigator menu > Nav. Options > deselect Convert Maps to Bytes.
   3. Select Navigator menu > Montaging & Grids > Setup Full Montage. A menu will open. The options to check are: Move stage instead of shifting image, Make map from each montage if Navigator open, and Use Montage Mapping, not Record parameters. The image grid should be roughly 6 x 6 or 7 x 7. Look for the total area to be recorded. If too many images are required, then the magnification may not be read correctly from the microscope (Figure 3). Stop and check.
   4. To start the Montage acquisition, press Start on the Montage Controls or Montage under Camera & Script. Another menu will open to choose the File Parameters: mrc, store as integers, and in yet another dialog, choose the filename for storage. Make sure to store data in the suitable data area and not under SerialEM's own user settings. It is recommended to not store large data on the C disk where the system resides.
   5. When the acquisition is finished the montage will appear in the main window (Figure 4). One can now navigate around the grid using the marker and points on the Navigator, just as in TEM.
6. Recheck the astigmatism correction. Usually, it is sufficient to scan a small area quickly and adjust together with the focus so that point-like features such as gold nanoparticles remain completely round as they pass in and out of focus (as on an SEM) at high magnification. More conservatively, one can repeat the direct alignments of steps 1.5-1.8 before fixing astigmatism.
7. Align the high magnification imaging to the LM montage.
   1. Go to a position on the LM montage with some easily recognizable feature. Add a point, click on Navigator window > Add Points, and click on the feature. Push again (stop adding) to deactivate.
   2. Take a View or Trial image at the lowest high mag (HM) magnification and find the indicated item. Place the marker there (green cross), and with the corresponding point highlighted in the Navigator window, in the Navigator menu > Shift to Marker…, in the dialog that opens, choose All Maps and Save Shift, as shown in Figure 5.
   3. Test by moving the stage to other positions and verifying that the HM image is centered at the same locations selected in the LM montage. If the feature is not visible in the HM image, the shift between LM and HM modes may be too large. In this case, make a polygon montage over a larger area. Click Navigator window > Add Polygon, choose some points, then click Add Polygon again to close. Click Navigator menu > Montaging and Grids > Setup Polygon Montage, and Montage Controls > Start. Then, find corresponding points as above and click Shift to Marker….
8. Set up low dose mode.
   1. Open the Low Dose Control tab and check the Low Dose Mode box.
   2. Check Continuous update and go to View. Choose the microscope magnification for this mode, typically the lowest or next to the lowest. Choose each of the next modes and set the appropriate magnification. Record should be set to achieve the desired pixel size, and Focus magnification should be high enough that fiducial markers or other high contrast features for focusing will be clearly resolved. Then, immediately uncheck Continuous update.
      NOTE: Trial mode may either be set similar to Record, in which case the tracking area will have to be displaced from the Record area in order to minimize exposure, or to low magnification, encompassing much of the surrounding.
   3. Take a View image and set the Trial position using Define position of area: Trial.
      NOTE: Recommendation #1: Given the change in the sampled area and time, the added exposure for a low mag Trial may be minimal. As long as the grid bar does not enter the field of view, this method of tracking is typically more reliable. Recommendation #2: It is also possible to update the spot size in order to adjust the dose for different modes, such as Focus. For stability in the microscope optics settings, we recommend not to do this, but rather to leave the spot size as is appropriate for Record and then to adjust exposure time if necessary for the faster scan modes.
   4. Click on Low Dose Control tab > Go to: Rec., then click on Navigator menu > Open Imaging States > Add Current State. Give it a name so that it can be recalled easily (e.g., µP STEM for microprobe STEM). Various states may be defined for different tasks.
9. Move the stage to a location of interest for tomography (more on the use of the Navigator below).
   1. Add a point in the Navigator by clicking Navigator window > Add Points.
   2. Refine the eucentric height by selecting Tasks > Eucentricity > Rough. Navigator window > Update Z.
   3. Set the areas for Focus and Trial. First, take a View image. Then, in the Low Dose Control tab, under Define position of area, select Focus or Trial. Adjust the positions for the two areas along the tilt axis. Focus should not overlap the Record area.
      NOTE: Keep in mind that there is some over-shoot, particularly at the left edge, so neither should be set immediately adjacent to the Record area. The extent of the over-shoot can be evaluated on a test area by repeatedly scanning at the Record magnification, and then reducing the magnification to get a broader view.
10. For good measure (optional with experience), tilt the stage to +60° and then -60°, each time taking a View or Preview image to ensure that the grid bar does not enter the Record area field of view shown in green.
11. Set up the tilt series.
    1. Open a new file to save the data by clicking File > Open New and choose a filename. Select Tilt Series menu > Tilt Series Setup (Figure 6). Troubleshooting: if the Open New button is gray, check that the previous tilt series has been terminated. If not, terminate by clicking Tilt Series > Terminate.
    2. Set the extreme tilt angles in the Tilt to and End at boxes, typically -60° and +60°, respectively, as well as the increment, typically 2°.
    3. For the dose symmetric mode (recommended - ensure that Low Dose mode is still active), check Run series in two directions from ___. The blank is normally 0, but may be used to compensate for pre-tilted FIB lamellae (e.g., at 20°).
    4. For simplicity, check Keep exposure time constant. Changing this requires some reworking of the exposure calculation and can also interfere with algebraic reconstructions (e.g., ART/SART/SIRT), which compare projected intensities with original data (see step 7.1).
    5. Check Skip Autofocusing. A small semi-convergence angle, such as 1 mrad, implies such a long depth of field that it may be unnecessary to focus. As a test, focus at 0°, tilt to 60° or -60°, and push Record. Maintaining focus is primarily a matter of precise eucentric height. For larger convergence, it may be necessary to focus during the series. In this case take a view, and in the Low Dose Control tab, define the position of area and focus, and adjust the focus area.
    6. Initial and final Actions: If eucentric height adjustment was performed accurately, there is no need to repeat it. Uncheck Close column valves at end of series unless there is good reason not to do so.
    7. For tracking the control parameters, there are many options. For unattended data collection, the top priority is to avoid the program stopping for user input. Use the recommended settings: repeat Record if the percentage of field lost is more than 5. Then, track after and check Align with Preview before getting new track reference and Get new track reference if the Record alignment differs by 2%. For attended data collection, apply more stringent parameters to avoid the risk of losing hours of instrument time.
    8. When ready to start, push GO at the bottom of the window. At the end, select Tilt Series > Terminate.

5. Batch acquisition at multiple points using the Navigator (Enhancement #1)

NOTE: The protocol is rudimentary because a) it is identical to TEM and b) new tools are becoming available that will likely render a full description obsolete.

1. Load the full grid montage to the main window.
2. Choose a number of areas that look promising. Click Add Points and add to the centers of selected grid squares. Click Add Points again to deactivate the mode.
3. With the low dose imaging state activated, select Navigator window > check Acquire and New File at Item for each of the selected squares. For the first, a dialog will open for the file properties (Figure 7) and then the filename. Choose Montaged images, and then in the Montage Setup dialog (Figure 8), check Use View parameters in Low Dose mode and make a grid large enough to cover the square (roughly 100 x 100 µm).
4. Click on Navigator menu > Acquire at items and select the parameters for mapping. Most importantly, in Primary actions, check Make Navigator Map, then Rough Eucentricity and Fine Eucentricity, and press Go (Figure 9). For a 200 mesh grid, this will result in a map of roughly the whole square (Figure 10)
5. Prepare for tomogram position selections. Open a new file by clicking File > Open New. Give it a filename (e.g., AnchorMaps.mrc).
   1. With the Low Dose mode active, enter Camera & Script > Setup and ensure that View and Preview are at the same binning (otherwise, they cannot be saved in the same file).
   2. Visit points in the square maps by clicking Navigator window > Go to XYZ.
   3. Choose a position for the first tomogram with the marker. Take a Preview image and refine the position by dragging with the right mouse button. Then, select Navigator window > New Map, and click yes to the dialog to save. Next, take a View image of the same area and again save as a new map. Make sure the View map is highlighted, and check For anchor state in the Navigator window at the top right.
   4. Choose the second position, again take a Preview and refine the location as above, and this time push Anchor Map in the Navigator window. This will save both the Preview and the View as new maps in the Navigator.
   5. Repeat step 5.5.4 for all the desired locations. Move from one grid square to another by selecting the point and pressing Go to XYZ in the Navigator window.
6. For reference focus, choose a clear area of the grid and focus carefully, by hand or by autofocus. Press Reset defocus on the microscope panel. Create a script in SerialEM by clicking Script > Edit > Edit 15 (or another free number) and enter two lines: ScriptName SetDefocus, and SetDefocus 0.
7. In the Navigator window, highlight the first location (either Preview or View). Press Shift-T and then highlight the last location and again press Shift-T. The File Properties dialog will open. Choose single-frame images and parameters to save including the filename. Edit the filename but leave the item label at the end.
   NOTE: Filenames ending in numbers will update automatically for successive tilt series. It is convenient to use the option Navigator menu > Nav. Options > Use Item Labels in Filenames.
8. Next, the Tilt Series Setup menu will open automatically. Fill as before (Figure 6), but do not choose Refine eucentricity. The Preview points in the Navigator will now be marked as TS. One can return to this menu using the button in the Navigator window above the items list.
9. Choose Navigator menu > Acquire at items. This time, select parameters for Final Data. Select Primary Action: Acquire tilt series and choose Manage Dewars/Vacuum, Realign to Item, Fine Eucentricity, and Run Script before Action, with Script to run: SetDefocus. Select general options: no message box when error occurs and close column valves at end, then hit GO (Figure 11). SerialEM will now visit all the Anchor maps and record a tilt series at each position (Figure 12).

6. CLEM map registration (Enhancement #2)

1. If it has not already been done, load the full grid montage into the main window, and then create a new window by clicking Window > New Window and give it a name. Note that the full grid montage appears in Registration 1.
2. Import the CLEM map image by clicking Navigator menu > Import map. It should go into Registration 2. It can also be assigned a new window as above.
3. Inspect the CLEM map to determine the relative rotation with respect to the full grid montage. It is easiest to do this on a map in which the grid bars appear. Note that the map may also be flipped. Align it roughly using Navigator menu > Rotate Map, with Flip if needed.
   NOTE: The following step can also accommodate a flip, but doing so in advance makes the following much easier. This step may be repeated until the alignment looks satisfactory, but it is not necessary to be precise.
4. Introduce registration points for precise alignment. Place a number of corresponding points on one and then the other window. For each such point, check Registration point at the very top of the Navigator window. Corresponding points will be labeled with an ascending index, followed by R1 or R2 for the grid montage or imported map, respectively.
   NOTE: Use of finder grids makes this step very simple. Otherwise, choose solid features such as the center marker or the corners of grid squares. In principle, three points are sufficient for a crude alignment, but more points help compensate for minor mismatches in the montage constructions.
5. If several channels of a CLEM map are desired, for example, different fluorescent colors or fluorescence without a bright field background, import those as in step 6.2 above, and change the registrations to 2. This way, they will inherit the registration points from the first map.
6. Impose the registration by Navigator menu > Transform items. The corresponding registration points will now appear on all maps, which will be listed in Registration 1. In order to align visually, click Navigator menu > Rotate Map. Note that when a map is loaded from the Navigator window, it is not automatically rotated unless the Rotate when load box is checked. It can always be rotated from the Navigator window.
7. It should now be possible to place the marker or a point on the fluorescence map and to move the stage to the corresponding location on the grid (Figure 13).

7. 3D reconstruction

1. This involves the same basic steps as TEM tomography: alignment of the tilt series followed typically by back-projection. Several software platforms are available, and the data can be treated similarly to TEM data, with the exception that CTF correction should be skipped.
2. Apply intensity normalizations to enhance the contribution of high tilt projections or to balance a change in illumination during the series, with the caveat that quantitative information is affected. IMOD²⁶ works well, for example, while AreTomo²⁷ is very useful for fiducial-less alignment; where fiducial particles are present, ClusterAlign²⁸ can follow them as clusters rather than individual spots; this is particularly useful for STEM data where spots may be lost or hidden by high-contrast features.
3. Follow reconstruction with 3D deconvolution²⁹ for contrast enhancement and denoising. Once reconstructed, represent the STEM volume data by any of the standard methods, such as orthoslices or isosurface segmentation. Notably, the intensity distribution is unipolar, without contrast inversions or Fourier fringes that appear in conventional phase-contrast TEM. An interesting means of representation of STEM tomograms, in general, is to invert the bright field (Figure 14). The effect is that of "X-ray eyes," seeing the dense objects of interest through the clarified cell³⁰.

Results

A full grid montage prepared in STEM shows the areas with cells of interest (Figure 4). Note that the image is in dark field, so empty holes appear dark. Cells appear partly bright, where the electrons are scattered toward the HAADF detector. At the thickest parts, typically the centers or near the grid bars, the contrast goes dark again. This is due to multiple scattering, whereby the electrons reach angles not captured by the detector. In practice, these areas will be too thick for tomography.

The next stage involves maps of intermediate magnification (Figure 10). These are often called medium montage maps, or square maps, referring to the grid squares rather than the shape. Even at the lowest microprobe STEM mode, they will also be acquired as montage images. Clicking on the item in the Navigator window loads the map image into the main window. Specific points for tomogram acquisition are chosen within this area. Use the Preview mode to locate the area at the recording magnification. Keep in mind that there may be a lateral shift between Preview and Record, depending on the scan speeds. A single tilt series may be recorded here. Mitochondria, for example, may often be identified in the Preview and Record image (Figure 12).

For batch tomography, the stage will travel around the grid and may not return to precisely the same location. Anchor maps are used to ensure that the desired Record areas are re-identified reliably by matching the Preview image to a lower magnification View, even if the stage movement has shifted. PyEM^23,24 offers a faster alternative that is certainly recommended, but is not yet part of the standard installation.

In the CLEM approach, a fluorescence map may be used to identify the areas of interest for tomography. The fluorescence is acquired externally and must be registered onto the full grid montage. It is useful to have the grid bars and holes visible, so a merged fluorescence + bright field composite may be prepared before import, for example in GIMP (https://www.gimp.org) or ImageJ³¹ software (take care that ImageJ inverts the vertical axis). When the dynamic range of the fluorescence is large, it may be difficult to create such a merge without saturation. In this case, the two maps may be imported separately and then registered together as described (Figure 13). This way, a point on the fluorescence map in the Navigator window, followed by "Go To XY", brings the stage to the corresponding position on the grid as mounted in the TEM. Take care that small inconsistencies in the creation of the montage (or the fluorescence map, if it too is a montage) will lead to small displacements; therefore, for optimal precision, the fluorescence should be re-registered specifically to the square map. It is often sufficient to do this registration by eye, on the basis of holes in the support film or features shared with the bright field light image.

Reconstruction of the tilt series results in a 3D volume (Figure 14). This example has a pixel size of 2.042 nm, which results in a field of view of ~4 µm. Calcium phosphate deposits (orange arrowheads) stand out, because of the higher atomic number compared to the surrounding. Microtubules (red arrowheads) can be traced throughout the whole field of view. Also, the inner and outer mitochondrial membranes (green arrowheads) can be clearly seen. Actin can be observed as bundles or as individual filaments. To achieve a more isotropic resolution, the reconstruction may be processed by deconvolution.

Figure 1: Comparison of TEM versus STEM. In TEM, the field of view is illuminated by a near-parallel beam, and the objective lens forms a magnified image on the camera. For cryo-TEM, the objective aperture is opened to pass the scattered electron waves (dashed red line), which, with defocus, produce phase contrast by interference with the unscattered (green) waves. STEM, on the other hand, rasters a focused beam across the specimen and collects the scattered electrons in a pixel-by-pixel manner. Multiple detectors may collect electrons scattered to different angles. This figure has been modified from³⁰. Please click here to view a larger version of this figure.

Figure 2: Example of image distortion on the left side due to the high scan rate. Note the distortion marked by yellow arrowheads, as well as, the distorted oval-shaped holes in the Quantifoil. Please click here to view a larger version of this figure.

Figure 3: Montage setup dialogue for the whole grid montage. Choose the magnification and the number of pieces in X and Y so that the total area reaches around 2,000 µm to get a map for most of the grid. Also choose Move stage instead of shifting image and Use Montage Mapping, not Record parameters. Please click here to view a larger version of this figure.

Figure 4: Full low magnification GridMap recorded in STEM mode. Broken areas appear completely black in the dark field image. Dark gray areas are likely too thick and poorly vitrified. Good candidate areas for tomography are white or light gray in this scan. Scale bar = 100 µm. Please click here to view a larger version of this figure.

Figure 5: Shift to marker process. A recognizable object is marked in the low resolution map by Add Points (38 in this image). Note the shift between the low resolution map (green circle) and the medium resolution map (orange circle). Then, the object is marked by the marker (green cross, red circle), and the Shift to Marker dialog is started. The chosen option moves all maps at 245x by the same amount. Please click here to view a larger version of this figure.

Figure 6: Tilt series setup dialogue for a STEM tilt series. The dose-symmetric scheme is used from -60° to +60° in steps of 2°. Autofocus is skipped due to the high depth of focus when using a low convergence angle. No need to refine eucentricity as this has been done before. Please click here to view a larger version of this figure.

Figure 7: File properties dialogue for medium resolution maps. Save as an MRC stack file, choose integers, and save extra information in a .mdoc metadata file. Please click here to view a larger version of this figure.

Figure 8: Montage setup dialogue for saving medium resolution maps. For a square on a 200 mesh grid, which is 90 µm wide, a total area of at least 90 µm x 90 µm is advisable. Choose Move stage instead of shifting image and Use View parameters in Low Dose mode. Please click here to view a larger version of this figure.

Figure 9: Recording medium resolution map. Acquire at Items dialogue for recording medium resolution maps. Choose Mapping, Acquire and save image or montage, and tick Make Navigator map. In the Tasks before or after Primary, choose Rough and Fine Eucentricity. When unassisted, optionally choose No message box when error occurs and Close column valves at end. Please click here to view a larger version of this figure.

Figure 10: Image of medium resolution map. Medium resolution map (Square Map) recorded in STEM mode by a montage of 3 x 3. The two medium resolution Anchor Maps for data collection are indicated by the blue squares. Scale bar = 10 µm. Please click here to view a larger version of this figure.

Figure 11: Batch tilt series data. Acquire at Items dialogue for collecting batch tilt series. For tilt series data collection, choose Final Data and Acquire tilt series. At Tasks before or after Primary, check Manage Dewars/Vacuum (choose appropriate settings in the Setup menu), Realign to Item, Fine Eucentricity, and Run Script before Action. In the General options, check No message box when error occurs and Close column valves at end (see 5.14). Please click here to view a larger version of this figure.

Figure 12: 0° tilt of a STEM tilt series of a mitochondrion. Orange arrowheads point to calcium phosphate deposits (matrix granules), green arrowheads to cristae, and red arrowheads to 15 nm gold fiducial markers. Scale bar = 500 nm. Please click here to view a larger version of this figure.

Figure 13: Registered cryo-CLEM maps. (A) Medium resolution STEM map (Square map, 26-A) is registered to the (B) low resolution STEM Map, (C) BF channel cryo-FM Map, and (D) GFP channel cryo-FM map. The nine registration points (labels 4-21) are shown in the (E) Navigator. Please click here to view a larger version of this figure.

Figure 14: Volume rendering. Volume rendering of (A) 60 nm and a (B) 40 nm thick sections through a SIRT-like (30 cycles) filtered tomogram at different depths. The pixel size is 2.048 nm, which results in a field of view of approximately 4 µm. Please note the inverted density compared to Figure 12, meaning that high-intensity features are bright. Orange, white, red, green, and light blue arrowheads point to calcium phosphate deposits, ribosomes, microtubules, the inner and outer mitochondrial membrane, and actin filaments, respectively. Scale bar = 500 nm. Please click here to view a larger version of this figure.

Discussion

The protocol should assist life science microscopists who are interested in obtaining a 3D view of intracellular organelles in regions of the cell that are not accessible to conventional TEM tomography. The same protocol may be also used for STEM tomography of plastic sections, with relaxed constraints on the exposure. The protocol should be regarded as a starting point rather than a set of hard rules. Indeed, the power of STEM is its flexibility; there is no one right way to operate it.

We emphasize that STEM, per se, refers only to the scanned probe and does not define the image formation. Contrast depends primarily on the configuration of detectors. The more straightforward methods employ detectors with axial symmetry, either a disk centered on the optic axis or an annulus surrounding it. In general, when the illumination impinges on the detector directly, we record a BF image where the (electron-scattering) specimen appears dark; conversely, when the detector collects only scattered electrons, we record a DF image where the dense specimen appears bright. When suitable hardware is available, SerialEM can acquire and record both these signals simultaneously. Still more sophisticated configurations are available, ranging from detectors with multiple segments to fully pixelated cameras. Phase contrast imaging can be achieved, for example, by evaluating the difference between off-axis detector elements^32,33. Collecting the entire 2D scattering (diffraction) pattern per pixel defines the method known as 4D STEM³⁴, which enables the reconstruction of multiple image contrasts from the same original data. Four-dimensional STEM enables electron ptychography, which provides another means to obtain phase contrast³⁵.

We have focused on the particular STEM modality that we consider to be most useful for the study of organelles and intact cells or micron-thick cell sections. This entails specifically the use of BF imaging with a small semi-convergence angle in the illumination and a relatively large collection angle at the detector. The small convergence provides a large depth of field so that the entire sample remains in focus throughout the tilt series¹⁹. In practice, with a good microscope stage, it can also eliminate the need for focus adjustment during acquisition. The price is a compromise in resolution, as described in section 3 of the protocol. We have suggested 4k images with a probe of ~2 nm, which with 1 nm pixel spacing reaches a field of view of 4 µm. However, the reader is strongly encouraged to experiment. The second consideration is on the side of the detector. When the illumination disk underfills the on-axis BF detector, phase contrast is suppressed and scattering (amplitude) contrast dominates; this condition has been called incoherent bright field contrast³⁶. The question is by what fraction to underfill, and the answer depends on the sample. A very thin sample will show best contrast with the detector completely filled (i.e., the collection angle matching that of the illumination), but a thicker sample will scatter all the intensity away, leaving a noisy signal with poor contrast. A useful rule of thumb is that the thicker the sample, the larger the outer cutoff angle of the BF detector should be²¹. The detector size and position are of course fixed, so the diffraction disk size is adjusted using the camera length, as described in section 1. If the detector amplifier settings are such that the signal fills the dynamic range but does not saturate under direct illumination (1.12.3), then the camera length can be increased until a reasonable signal intensity and contrast are reached. Again, the reader is encouraged to experiment. The art, so to speak, is in the angles.

Another parameter, not discussed in the protocol, is the microscope acceleration voltage. Interaction of the illuminating electrons with the specimen will be stronger at a lower voltage. With all else equal, we can expect higher contrast at lower voltage. On the other hand, it is the onset of multiple scattering within the specimen that limits the useable specimen thickness, so a higher voltage permits the use of thicker samples. These effects are rather subtle, however. Our experiences to date with 200 kV and 300 kV accelerations are similar.

Considering what can be expected of STEM in terms of resolution, this again depends on the specimen and the detector configuration. Using a single particle analysis approach, metal ions on ferritin could be localized by annular dark field STEM to a precision of a few angstrom³⁷. More recently, sub-nanometer resolution was achieved for virus and protein samples using images obtained by integrated differential phase contrast (iDPC) STEM³² as well as ptychography³⁵. For unique objects in thick cellular specimens, and with the methods described here, such high resolution is not realistic. The optimal resolution is the probe diameter, which relates to the semi-convergence angle as described. Other factors will degrade resolution, particularly a coarse pixel sampling to reach a large field of view, misalignments in the tilt series, and spreading of the probe beam in transmission. Images compare well with plastic section tomography. With the setup described here, for example, it should be easy to see the hollow core of the microtubules, but not the individual protofilaments.

To summarize, STEM methods and hardware are both in a development phase. We can expect that innovations in imaging will impact tomography as well, leading STEM into domains that have not been accessible by other techniques. We expect that a convergence with correlative cryogenic fluorescence imaging based on modern optical super-resolution methods will be especially fruitful. The scale of organelles, 100-1,000 nm, is an ideal target for these emerging methods.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
SerialEM	University of Colorado	Veriosn 4.0	SerialEM is a free software platform for microscope control and data acquisition.
STEM-capable transmission electron microscope			The protocol was written based on experience with several microscopes of Thermo Fisher Scientific: Titan Krios, Talos Arctica, and Tecnai T20-F. In principle it should be applicable to other manufacturers as well.