to BioTechniques free email alert service to receive content updates.
High-resolution three-dimensional reconstruction of a whole yeast cell using focused-ion beam scanning electron microscopy
Dongguang Wei1, Scott Jacobs2, Shannon Modla2, Shuang Zhang3, Carissa L. Young4, Robert Cirino2, Jeffrey Caplan2, and Kirk Czymmek2,5
Full Text (PDF)
Supplementary Material

Figure 1. Enhanced FIB resolution. (Click to enlarge)

Figure 2. 3D segmentation. (Click to enlarge)

Results and discussion

To achieve optimal morphology and contrast by EM, S. cerevisiae cells were high-pressure frozen, freeze-substituted and en bloc stained in lead and uranyl acetate prior to epoxy resin infiltration. To our knowledge this is the first demonstrated report of FIB-SEM resin infiltrated 3D cellular imaging using cryo-preservation without conventional pre-fixation/fixation. While en bloc lead staining (26) has been employed for serial block face imaging (30) our procedure demonstrated, as expected, excellent contrast via FIB-SEM. A comparison between TEM (Figure 1a) and BSE images from FIB-SEM (Figure 1b) showed the images to be very similar, but with somewhat reduced image detail and increased contrast of yeast cell walls with FIB-SEM. We then acquired serial FIB slices of yeast cells at 15 nm, 5 nm, and 3 nm z-intervals, which were carefully chosen to assess the practicalities, limitations and impact these parameters had on image quality and structural integrity. While the 15 nm z-interval produced sufficient detail for many studies and the 3 nm z-interval provided the maximum spatial resolution of cellular components, 5 nm z-interval data was a reasonable balance of resolution, data size and the time required to collect data for an entire cell. The data set acquired at the 3 nm z-interval resulted in an XZ image (Figure 1c) that compared favorably to the XY FIB-SEM image (Figure 1b). A close inspection of the XZ image (Figure 1d) showed that relatively small structures, such as nuclear pores (arrows), were easily identified.

We report here, the ability to image through large cell volumes at reduced z-intervals down to 3 nm isotropic voxels, which is a 6-fold reduction in z-interval from several previous reports using FIB-SEM in cells (6, 7, 15) and comparable to a recent 5 nm isotropic FIB-SEM data set acquired from brain tissue (12). Our data extended upon this previous work by reducing the isotropic resolution to 3 nm and segmenting and quantifying numerous major cellular organelles from a single cell at 5 nm isotropic resolution using cryo-preservation. Since it was not possible for us to directly measure the 3 nm milling as defined via software acquisition settings, we used three primary methods to validate the approach. First, as a rough visual inspection, the data showed no conspicuous distortions in aspect ratio when viewing the XY versus Z axis (Supplementary Video 5). Second, in order to empirically confirm the z-interval of ~3 nm, we used spherical vesicles within the resin embedded cell as an internal and sample specific control with a previously described cylindrical diameters method for measuring mean section thickness of TEM (31) and 3D FIB data (6, 11). Our results demonstrated a mean slice thickness of 3.603 nm, (SD = 0.606, N = 30). Consistent with this result, using 3 nm × 3 nm × 3 nm isotropic voxels, a 61 nm vesicle spanned 18 slices (Figure 1e). In comparison, the 15 nm z-interval data set required 6 slices to image section through a 74 nm diameter vesicle (Figure 1f). Third, we assessed milling consistency via continuous playback of the stacks of images (Supplementary Video 6) and while not flawless, overall the data showed relatively smooth transition from section-to-section for the data set. Furthermore, the gradual change in vesicle diameter at each step (Figure 1e) suggested that the BSEs that reached the EsB detector were restricted to a few nanometers from the surface. When low energy electrons hit a specimen, the penetrated distance depends roughly linearly on the electron's energy. The same is true with BSEs (8, 29) exiting the specimen surface which have a wide distribution in angle and energy since they may have experienced multiple scattering events and originated from different depths in the specimen. Conventional BSE detectors mounted below the objective lens typically collect nearly all BSE exiting the specimen surface and as a consequence x-, y-, and z-resolution are reduced. To overcome these limitations, an in-column EsB detector was employed (Supplementary Figure 1) similar to a previous report (12). BSEs with large scattering angles or that originated deep within the sample were restricted from reaching the EsB detector either geometrically or by applying a bias energy at a filtering grid before the detector (Supplementary Figure 2). In this study, for 3 nm and 5 nm z-interval data, the primary beam energy was set to 1500 eV and the energy filtering grid bias was set to 1000 eV, thus only BSEs having lost less than 500 eV were allowed to pass through the grid and reach the EsB detector.

Modeling electron-sample interactions of biological systems is highly complex due to local chemical heterogeneity, including variations in structure and composition and the lack of a priori knowledge of the detailed cellular composition of our samples. Though the BSE radius, depth of emitted BSEs and BSE coefficients have been modeled previously in two pure metals (carbon and osmium) for 2 keV beam with tilt (9), it is far from the real conditions of resin infiltrated biological cell systems. Here, we employed a simple and conservative composition, epoxy resin and osmium fixative [C21H25ClO5) + 1% (atom)Os], to estimate the extent of electron penetrations within the epoxy embedded yeast. Monte Carlo simulation of electron-sample interactions showed that the BSEs for imaging were predominantly 3–4 nm below the surface (E0 = 1500eV, Egrid = 1000eV, Tilt = 54°) (Figure 1g) as compared with over 20 nm in the case of conventional BSE detection for FIB without energy selection (E0 = 1500eV, Tilt = 0°) (Figure 1h). This simulation confirmed the empirical results (Figure 1e) where the images of 3 nm-interval slices consisted of mainly BSEs from those respective slices. The addition of atomic lead and uranium as part of simulations would only serve to further reduce the BSE signal collection volume (i.e., less than 3 nm). Considering the complexity of biological samples and our inability to reliably determine the local heavy metal stain concentration, simulations with lead and uranium were not included, thus we emphasize our model estimates of electron penetration under the conditions employed were intended to be conservative.

As part of our characterization of the FIB-SEM high-resolution cell imaging approach, we sought to acquire, render and compare yeast volumetric data sets with 15 nm (Figure 2a), 5 nm (Figure 2b, c) and 3 nm (Figure 2d-f) z-intervals. FIB-SEM acquisition produced very large data sets and in the case of the 5 nm experiment, the resulting 3D volume was a ~10×7.5×7.5 µm3 and contained fine structural details of several yeast cells represented by about 5 billion voxels. This large amount of information was examined interactively with state-of-art graphics processing unit (GPU) accelerated volume rendering technology (Supplementary Figures 3, 4 and 5). Whole cells were extracted from the volume by creating a mask that isolated region of interests (Supplementary Figure 4). Detailed biological structures within the cell were segmented using a combination of automatic and interactive software algorithms prior to reconstruction. The 15 nm z-interval data set of a budding yeast cell with selected segmented organelles revealed excellent structural and positional detail; however, the z-axis stepping artifact of a spherical nucleus (light blue) and other structures was prominent due to the anisotropic resolution (Figure 2a, Supplementary Video 1). A 5 nm isotropic voxel data set of an entire yeast cell (Figure 2b, Supplementary Video 2) showed significant improvement in overall structural detail (i.e., mitochondria, endoplasmic reticulum, nucleus) with concomitant reduction in z-axis stepping artifacts (i.e., nucleus) in this camera perspective. Furthermore, the spatial relationships of organelles with each other as well as sites of endoplasmic reticulum and mitochondrial interconnectivity (32) were readily documented (Figure 2c, Supplementary Video 3). To further push the systems capabilities while dramatically extending the total cell volume visualized, we acquired a data set at 3 nm isotropic voxel settings (Figure 2d) which permitted high-resolution 3D reconstructions of the nucleus, nuclear pore position and nucleoplasm/heterochromatin (Figures 2d and 2e), cisternae and vesicle distribution (Supplementary Video 4) and plasma membrane sights of invagination termed eisosomes (33) (Figure 2f). Images used in Figure 1c-f were derived from this same 3 nm volumetric data set and the raw x-y aligned but unsegmented stack showed overall structural detail in the x-y and x-z axis with our approach (Supplementary Video 5). The accurate reconstruction of organelles in an entire cell allowed us to quantify targeted cellular structures, including but not limited to volume, volume percentage and surface area of various organelles (Table 1).

  1    2    3