1Carl Zeiss Microscopy, LLC, One Zeiss Drive Thornwood, NY, USA
2UD Bio-Imaging Center, Delaware Biotechnology Institute, University of Delaware, 15 Innovation Way, Newark, DE, USA
3Visualization Sciences Group, 15 New England Executive Park, Burlington, MA, USA
4Department of Chemical and Biomolecular Engineering, University of Delaware, 150 Academy Street, Newark, DE, USA
5Department of Biological Sciences, University of Delaware, 330 Wolf Hall, Newark, DE, USA
BioTechniques, Vol. 53, No. 1, July 2012, pp. 41–48
We developed an approach for focused gallium-ion beam scanning electron microscopy with energy filtered detection of backscattered electrons to create near isometric voxels for high-resolution whole cell visualization. Specifically, this method allowed us to create three-dimensional volumes of high-pressure frozen, freeze-substituted Saccharomyces cerevisiae yeast cells with pixel resolutions down to 3 nm/pixel in x, y, and z, supported by both empirical data and Monte Carlo simulations. As a result, we were able to segment and quantify data sets of numerous targeted subcellular structures/organelles at high-resolution, including the volume, volume percentage, and surface area of the endoplasmic reticulum, cell wall, vacuoles, and mitochondria from an entire cell. Sites of mitochondrial and endoplasmic reticulum interconnectivity were readily identified in rendered data sets. The ability to visualize, segment, and quantify entire eukaryotic cells at high-resolution (potentially sub-5 nanometers isotropic voxels) will provide new perspectives and insights of the inner workings of cells.
Elucidating the three-dimensional (3D) spatial distribution of structures and molecules within cells and tissues is critical for determining the nature of numerous cellular processes. Consequently, a multitude of technologies using photons and electrons (1-3) as well as x-rays (4) have been developed to permit the 3D spatial visualization of organisms, tissues, cells and sub-cellular constituents. In particular, focused ion beam scanning electron microscopy (FIB-SEM) has drawn significant attention as a powerful technology for creating 3D volumes of biological specimens at electron microscopy resolution (5, 6). To date, the majority of FIB-SEM approaches have employed variations in conventional chemical fixation that include an aldehyde primary fixation in combination with osmium tetroxide and uranyl acetate en bloc staining (6, 7) while others (8, 9) also included thiocarbohydrazide-osmium ligand treatment (10), potassium ferrocyanide (9, 11-14), and/or tannic acid (5, 14) for enhanced contrast. In one instance, pre-fixation in paraformaldehyde prior to cryo-preservation with high-pressure freezing was employed, but no differences were observed compared with conventional fixation, thus cryo-data data was not shown (15). FIB-SEM of biological samples has been demonstrated at ~50–60 nm (5, 9, 13, 14), 30–40 nm (11) and down to ~18–20 nm slice z-intervals in mammalian cells (6, 15, 16) and 100 nm z-interval specifically in yeast (7) which are comparable to serial sectioning methods or block-face imaging using diamond knives (1, 17). More recently, the limits of FIB-SEM has been even further pushed with FIB conditions set to acquire 5 nm isotropic voxels in brain tissue (12). For high-resolution 3D imaging of biological samples, transmission electron microscopy (TEM) tomography is the method of choice for acquiring cellular structures down to ~3 nm resolution (18), including yeast (19); FIB milling in conjunction with TEM tomography has been shown as a powerful enhancement to cryo-TEM tomography (8). However, TEM tomography is typically limited to 0.5 micron thick sections, making reconstruction of an entire eukaryotic cell technically very challenging. In an attempt to approach the resolution of TEM tomography using FIB-SEM for whole cell reconstructions, we pushed the limits of focused ion beam milling and backscattered electron (BSE) detection to obtain ca. 3 nm isotropic voxels. This significant z-resolution improvement enabled us to create high-resolution 5 nm x 5 nm x 5 nm isotropic renderings and quantification of an entire cell, including numerous subcellular structures, with the yeast model organism, Saccharomyces cerevisiae. The purpose of this work was to optimize cryo-fixation sample preparation, milling and imaging conditions to quantify and document the 3D spatial distribution of sub-cellular structures from an entire single cell at nanoscale resolutions.
Materials and methods
S.cerevisiae yeast strains BJ5464 and BY4742 grown in YPD were used in this study. Overnight cultures were grown to mid-log phase (0.8 ≤ OD600 ≤ 2.0) in 500 mLs – 1 L of YPD media (30°C, 275 rpm) (20, 21). YPD medium was prepared as described (22). To contrast the effects of cellular stress induced by chemical treatment (23), 8 mM of dithiothreitol (DTT) (Sigma-Aldrich, St. Louis, MO, USA) was added to BY4742 (5 nm isotropic data set only). Following incubation for 2 h at 30°C and 275 rpm, unstressed and stressed BY4742 cells were harvested. Cells were centrifuged at 2,500 rcf for 5 min at 4°C for subsequent cryo-preservation.
Electron microscopy sample preparation
A yeast paste was high pressure frozen in a Leica EMPact in 1.2 mm x 200 µm hats and freeze substituted at -80°C in 2% osmium tetroxide in acetone containing 1% water for 3–4 days as described previously (24). It must be noted, that even with freeze-substitution, some yeast membranes are difficult to contrast (25). Samples were warmed to -20°C over a 16 h period, to 4°C for 3 h, to room temperature for 2h and washed in acetone. Yeast were en bloc stained in saturated uranyl acetate in 100% acetone overnight, washed in acetone, transferred to 1:1 ethanol:acetone and en bloc stained with saturated lead acetate in 1:1 ethanol: acetone for 2–4 h (26). Samples were dehydrated in acetone and infiltrated with Embed-812 epoxy resin, embedded in BEEM capsules and polymerized for 48 h at 60°C. All samples were evaluated on a Zeiss LIBRA 120 Plus transmission electron microscope with a Gatan Ultrascan 1000 2k x 2k camera prior to FIB analysis.
FIB/SEM image stack acquisition
Cured resin blocks were manually trimmed to expose yeast cells and a glass knife was used to face the block. The final block size was trimmed as small as possible (about 1 mm in size) so that sample drift from continuous degassing of resin was minimized. The block was then attached to an aluminum SEM stub with silver paint. The block and stub was coated with Pt for 5 min resulting in a continuous conductive Pt layer about 100 nm thick. This continuous Pt layer served as a medium for dissipating charge and heat generated from the electron and gallium ion beams interacting with the sample block. A Zeiss Auriga FIB/SEM XBeam system that is equipped with a Cobra-focused gallium ion beam column, a Schottky field emission gun and a Gemini electron column was employed for all the milling and imaging work in this study. For imaging and milling, the sample was tilted to a 54° angle and placed at a working distance about 5 mm below the SEM final lens so that the sample surface was perpendicular to the ion beam where the electron beam and ion beam meet. The electron beam (SEM imaging) was used to locate a site of interest and then the ion beam (FIB) was used to prepare the block face for automated image stack acquisition. For FIB milling, we employed a 30 keV Ga ion beam with a 12 nA current to cut a trench so that the 3D imaging targeted volume was exposed and the surface was prepared to be milled and imaged. This also helped accommodate some re-deposited materials and reduced the possibility that neighboring materials would interfere with the passing of BSEs to the detector. To reduce the damage caused by large current FIB milling, the periphery of the volume of interest was cleaned with a 1 nA beam. A 600 pA beam was used for final 3D data acquisition.
Image stack acquisition was executed in an automated fashion where both ion milling parameters (i.e., probe current, dwell time and milling interval (z-interval)) and SEM imaging parameters (i.e., beam energy and current, dwell time, pixel size, detector selection and energy filtering grid bias, focus and stigmator settings) were set before the process started. Specifically, all images were taken with a beam energy at 1.5 keV, and probe current of 931 pA (60 µm aperture and high current mode), working distance at 5.0 mm and tilt angle at 54°. Line averaging was used for noise reduction (N = 30) at an image store resolution of 3073×2304 with a dwell time of 12.7 µsec/pixel, resulting in a total cycle time of around 1.5 min/image. Electron dosage per square nanometer was calculated at total electrons per square nanometer (30 line passes in “line average”), or electrons per square nanometer per pass. For the 3 nm data set, it was 8220 electrons/nm2 and 274 electrons/nm2/pass, the 5 nm data set 2958 electrons/nm2 and 99 electrons/nm2/ pass and the 15 nm data set 329 electrons/ nm2 and 11 electrons/nm2/pass. The energy filtering grid was set at 500 eV, 750 eV and 1000 eV for the data sets with 15 nm, 5 nm, and 3 nm z-interval, respectively. A typical stack of images with an x-y-z pixel size of 5 nm covered a volume of ~8 µm χ 10 µm χ 8 µm and took about 35 h.
Normal FIB artifacts such as curtaining were minimized by careful preparation of the block face and deposition of surface protection layer (Pt or C). Surface charging was also reduced by recoating the sample after a trench was FIB-cut (making the side walls of the cut block conductive) and by using the energy filter to block the lower energy BSEs (more susceptible to surface charges) from reaching the Energy Selective Backscattered electron (EsB) detector. We also noted that blanking the electron beam during FIB milling allowed us to obtain higher quality images. Through long acquisitions hours, image drift was an unavoidable phenomenon and was minimized by the following steps: (i) we cut the embedded sample block as small as possible (~ 1 mm cube) and pre-evacuated the sample for at least 24 h before starting FIB sectioning and imaging, (ii) we mounted the sample on a stub with silver paint around it, (iii) we coated the sample with about 10–15 nm thick Pt layer using a Cressington Sputter Coater to make the surface as conductive as possible in order to avoid surface charge induced image drift, and (iv) we allowed at least one hour for the sample to achieve equilibrium (mechanically and thermally) after the sample was inserted and the site of interest was found.
Stack alignment, segmentation, and 3D presentation
Once an image stack was acquired, alignment of the image stack (See Supplementary Material for additional details), noise reduction, reconstruction of a voxel image from the aligned image stack, segmentation based on intensity and/or shape of known organelle structure and the final presentation in 3D space was generated. For the cell FIB sectioned at a 10 nm isotropic pixel resolution, the ER was segmented with Avizo software application framework whereas all other organelles were segmented in IMOD (5, 27). All other data sets were processed exclusively using Avizo with a customized workflow supporting individual cell segmentation. Segmentations were based on intensity and known morphology of the target structure. Segmentation results were used to reconstruct the geometrical representations of the organelles and to extract quantitative information. Avizo was also used to generate all pictures and movies.
Monte Carlo simulations
When a primary electron beam penetrates into solid sample surface, backscattering occurs. The BSEs carry energy below the primary beam energy due to inelastic interactions between the incident electrons and the solid sample. The distance that the BSEs travel within the sample back toward the surface follows nearly a linear relationship with the energy they carry (28, 29) and can be calculated with Monte Carlo simulation using CASINO v184.108.40.206. In this work, the primary beam energy was set at 1500 eV and a bias energy was applied to the energy filtering grid thus preventing BSEs of energy below the set bias (for example, 1000 eV) to reach the detector (Supplementary Figure 2). Therefore, only BSEs carrying energy above the set bias energy reached the detector. We conservatively calculated the stopping range of electrons in the simulated yeast sample [Epoxy resin (C21H25ClO5) with 1% (atom) Os)] at energy filtering biases of 1000 eV and 0 eV (no filter, conventional BSE detection system) respectively, using median values of the selected energy windows, for instance, 250 eV for the set bias of 1000 eV (e.g., energy window width of 500 eV) and 750 eV for no filter (e.g., primary beam energy, 1500 eV). Figure 1g showed the electron trajectory at 54° tilt with 1000 eV bias and Figure 1h showed the electron trajectory when no filtering was applied at 0° tilt.