Biological serial block face scanning electron microscopy at improved z-resolution based on Monte Carlo model

Serial block-face electron microscopy (SBEM) provides nanoscale 3D ultrastructure of embedded and stained cells and tissues in volumes of up to 107 µm3. In SBEM, electrons with 1–3 keV energies are incident on a specimen block, from which backscattered electron (BSE) images are collected with x, y resolution of 5–10 nm in the block-face plane, and successive layers are removed by an in situ ultramicrotome. Spatial resolution along the z-direction, however, is limited to around 25 nm by the minimum cutting thickness. To improve the z-resolution, we have extracted depth information from BSE images acquired at dual primary beam energies, using Monte Carlo simulations of electron scattering. The relationship between depth of stain and ratio of dual-energy BSE intensities enables us to determine 3D structure with a ×2 improvement in z-resolution. We demonstrate the technique by sub-slice imaging of hepatocyte membranes in liver tissue.


Supplementary Figure 1 | Representation of image stack used for sub-slice reconstruction.
White rectangles indicate a continuous column structure that penetrates the sample. The slight shifts of columns in different cuts are due to residual charge-induced electrostatic beam deflection. Sub-slice reconstruction requires that image alignment be accurate to within a single pixel.
Supplementary Figure 2 | Backscattered images acquired from six successive 25-nm cuts of a liver sample that had been stained with heavy metals and embedded in epoxy resin. The blockface images are shown after fine alignment to correct for drift. (a), (c) BSE images acquired at beam energies of E low = 1.0 keV and E high = 1.4 keV; (b), (d) pre-processed BSE images in (a) and (c) after setting negative pixels intensities to zero. (e) (f) and (g) pre-processed BSE images and difference images of cut 1, 3 and 5 from (b) and (d) after normalizing the background. To display differences between BSE images acquired at dual landing energies, pairs of images in Figs. b1 / d1, b3 / d3, and b5 / d5 are shown at higher magnification in Figs. eI / eII, fI / fII, and gI / gII, respectively, and their differences are displayed in Figs. eIII, fIII, and gIII. White circles indicate typical differences between features, which appear bright or dark depending on whether they are located in the top or bottom subslice. Red triangles indicate homogenous region used for scaling image intensities in the E low and E high image stacks. Scale bar, 500 nm. Figure 3 | Four pairs of backscattered images at (a) E low = 1.0 keV and (b) E high = 1.4 keV from successive 25-nm cuts of a heavy metal-stained liver block, together with calculated structure of (c) top and (d) bottom sub-slices, each of thickness 12.5 nm. The numbers indicated in white represent the cutting order and correspond to the numbers in Supplementary  Fig. 2. The numbered arrows (red and blue) denote selected fine-scale features that change from sub-slice to sub-slice. Scale bar, 500 nm.

5
Supplementary Figure 4 | Energy-dispersive x-ray spectrum from thin section of tissue block used for SBF-SEM imaging, recorded at 120 keV beam energy using an FEI Tecnai TEM equipped with an Oxford Instruments X-Max SDD and Inca microanalysis system. X-ray spectra generated by the DTSA-II software were used as reference spectra, which were fitted to the tissue spectrum to quantitate the relative concentrations of the heavy atoms: Os, Pb, and U.
The x-ray spectrum shows that the sample contains uranium, lead and osmium, whereas the high copper peaks are due to x-ray generation in the copper grid used to mount the thin sections, and these are ignored in the analysis. There is strong overlap between the osmium L lines with copper K lines at ~ 8.9 keV, and between the lead L lines and the osmium L lines, which complicates quantification of these elements. We therefore used DTSA-II 1 to fit the spectrum, as were calculated at 1.0 keV with a stain composition from 1%-9%. (c) Intensity vs. stain composition profiles for cuboids shown in (a); simulation data were calculated at 1.4 keV with a stain composition from 1%-9%. (d) Intensity vs. stain composition profiles for cuboids shown in (a); simulation data were calculated at 1.0 keV with a stain composition from 10%-100%. (e) Intensity vs. stain composition profiles for cuboids shown in (a); simulation data were calculated at 1.4 keV with a stain composition from 10%-100%. Curves were fitted though the data using a second order polynomial regression. The red curve is based on the intensity of cuboid 3 in (a), and the blue curve is based on the sum of the backscattered signals from cuboid 1 and cuboid 2.
The curves in (a) and (c) show that the backscattered signal depends linearly on stain content for concentrations below 5 atomic percent, whereas the curves in (c) and (d) reveal a strong nonlinear dependence on stain content for concentrations above 10 atomic percent.

Number of incident electrons / pixel
10, 000 Electron beam diameter 12.5 nm Pixel size 12.5 nm Simulation noise Shot noise Backscattered detector efficiency (! !"# ) 1.0 Low primary energy for stained cuboids in epoxy (25 nm/cut) 1.0 keV High primary energy for stained cuboids in epoxy (25 nm/cut) 1.4 keV Low primary energy for stained cuboids in epoxy (50 nm/cut)