Topographic contrast of ultrathin cryo-sections for correlative super-resolution light and electron microscopy

Fluorescence microscopy reveals molecular expression at nanometer resolution but lacks ultrastructural context information. This deficit often hinders a clear interpretation of results. Electron microscopy provides this contextual subcellular detail, but protein identification can often be problematic. Correlative light and electron microscopy produces complimentary information that expands our knowledge of protein expression in cells and tissue. Inherent methodological difficulties are however encountered when combining these two very different microscopy technologies. We present a quick, simple and reproducible method for protein localization by conventional and super-resolution light microscopy combined with platinum shadowing and scanning electron microscopy to obtain topographic contrast from the surface of ultrathin cryo-sections. We demonstrate protein distribution at nuclear pores and at mitochondrial and plasma membranes in the extended topographical landscape of tissue.

Scientific RepoRts | 6:34062 | DOI: 10.1038/srep34062 platinum by electron beam evaporation, providing a quick, simple and consistent way to obtain correlative image information from tissue samples.

Results
Collection of cryo-sections on silicon wafers facilitates CLEM. 100 nm-thick Tokuyasu ultrathin kidney cryo-sections were collected on a 7 × 7 mm silicon wafer (Figs 1a and 2a). The flat silicon surface provides stability, facilitates the handling of the sections compared to EM grids, reduces the deformations in the tissue caused by drying or exposure to a vacuum, and minimizes folds and breaks 20 that are often found in sections collected on formvar film-coated grids 21 (Fig. 2b). This conductive support prevents charging artefacts and is therefore ideal for scanning electron microscopy (SEM) imaging 21 . Sections were subsequently processed for immunofluorescence staining (Fig. 1b). For light microscopy (LM), a thin layer of buffer and glycerin (1:1) was retained between the wafer and the glass-bottom petri dish. This mixture keeps the sections wet and close to the coverslip (Fig. 1c-e). Optical imaging was performed with confocal laser scanning microscopes using a closed pinhole to Airy unit 0.7 followed by deconvolution (Supplementary Fig. 1; Huygens, SVI, Netherlands). After confocal microscopy, a short postfixation with glutaraldehyde was followed by embedding in methylcellulose and centrifugation of the wafer to cover the ultrathin Tokuyasu section with a thin layer of dried methylcellulose ( Fig. 1f-h). This latter step protects the tissue from drying artefacts 16 and avoids the need of using solvents and embedding in resin, which contribute to major problems in tissue extraction and shrinkage. Nevertheless, shrinkage occurs to some extent during drying in methylcellulose 13 . The sections were platinum/carbon shadowed to generate topographic contrast for SEM by detecting secondary electrons (Fig. 1i,j).
The fluorescent nuclear stain DAPI (4′ ,6-diamidino-2-phenylindole dihydrochloride) was used as a first rough reference to correlate the confocal and SEM images (Fig. 2c-e). The nuclear pore complex protein was observed along the nuclear membrane in the confocal and in the scanning electron microscopy image in relation to the different cellular components (Fig. 2f). Mitochondria, vesicles and microvilli exhibit high topographic contrast. Mitochondrial cristae are clearly visible, demonstrating the level of resolution this novel technique can provide (Fig. 2f,g).
All SEM images were taken using either an Inlens SE (SE I) or side mounted SE II detector. Contrast changes, in particular in mitochondria, are much more prominent when using the Inlens SE detector. Areas between dense cristae do not contain platinum or only tiny amounts (due to low angle coating of 8°) and only few secondary electrons escape from this area providing a "dark" area. On the membranes of the cristae itself (sticking out), platinum is added in larger amounts by the low angle shadowing and produce a high contrast due to more platinum and etch effects. The reduced resolution with the SE II detector (virtually larger scanning spot) also leads  to an averaging of the signal and therefore less contrast between membranes of cristae and areas between cristae, "washing out" high contrast differences in the nm scale.
Conventional and super-resolution microscopy combined with SEM. The above methodology was used to determine the localization of the protein Tom20, a subunit of the translocase mitochondrial outer membrane complex 22 with confocal microscopy and SEM (Fig. 2h,j) and by super-resolution microscopy (ground state depletion microscopy followed by individual molecule return, GSDIM) and SEM (Fig. 3a,b). The confocal images exhibit elongated or ring-like structures which, when overlaid onto the matching SEM image, correspond to the outer membrane of the mitochondria in these thin sections (Fig. 2j). A similar correlation was possible using super-resolution microscopy (Fig. 3a,b). GSDIM images however exhibited a sparser signal along the outer membrane of the mitochondria. This appears a more accurate immunolabelling pattern since it corresponds with the reported cluster expression of this protein 23 . Finally, we located the phosphorylated form of the NaCl cotransporter (pNCC) at the apical plasma membrane of distal tubules 24 by super-resolution microscopy. Transversally cut microvilli were identified by super-resolution microscopy. Images display strong staining along the membrane below the diffraction limit of resolution (Fig. 3c-f). The direct immunogold method on Tokuyasu sections (Fig. 3g) was applied to compare the level of accuracy between these two approaches. Both methods showed similar structural resolution and labelling patterns. The immunogold method on silicon wafers could be used to obtain protein localization information on large areas. However, super-resolution microscopy with SEM allows a quicker overview signal detection and identification of rare events (Fig. 3c). Additionally, this method can be applied to multiple labelling approaches and in other tissues and cells ( Supplementary Figs 2 and 3).

Discussion
We provide a relatively quick, facile and reliable method for CLEM in cells and tissue. A number of correlative approaches are currently available to detect endogenous proteins 3,25 . Among those methods some identify proteins at subcellular level in tissue overcoming the challenges of 3D complexity, sample preparation, preservation of antigenicity and structure 12,26,19 . Tokuyasu sections from cells and tissues are hydrated, and remain so during the labelling procedure; this preserves the ultrastructure and antigenicity of most epitopes. Immunofluorescence and immunogold labelling are routinely used techniques 27 that allow imaging of wet samples at super-resolution followed by transfer to electron microscopes without resin embedding 10 and associated potential artefacts 28 . Tokuyasu sections treated with standard contrast agents often appear poor in contrast. Recent procedures however, have introduced steps to increase sample contrast 17 . In this study we propose a simple solution to this problem; the application of a thin (2 nm) layer of platinum/carbon on the tissue surface as an effective method for contrast generation without the need for heavy metal post-staining. This contrast corresponds to the roughness or topography of the tissue section [29][30][31] . In addition, the platinum/carbon layer facilitates image acquisition by reducing possible charging artefacts during imaging of biological specimens. We demonstrate that topographic contrast is able to precisely resolve mitochondrial cristae, vesicles, and microvilli in cells of kidney tissue.
The excellent intrinsic axial resolution (100 nm section thickness) of the Tokuyasu sections provides better resolution than optical sectioning of thick samples 32 . In addition, we have imaged these 100 nm thin sections with confocal laser scanning microscopes using a closed pinhole [Airy unit 0.7] followed by deconvolution (Huygens, SVI, Netherlands). This generates higher resolution images in the XY plane than those acquired with widefield microscopes 33 . This combination provides exceptional XYZ resolution for accurate protein localization.
Collecting thin sections on silicon wafers offers many advantages 21 , among them stability, flatness and ease in sample manipulation throughout the processing procedure. Image acquisition by light and electron microscopy is much more stable on silicon than imaging sections on electron microscopy grids 21 . Handling of wafers for incubation with antibodies is also easier and facilitates transfer between instruments without damaging sections. Treated glass coverslips are also used for CLEM methods 34,35 . The main advantage of silicon wafers compared with glass is that they are highly conductive and very clean without the need of any pre-treatment. We have developed this method so that once Tokuyasu sections are collected on the wafer, the procedure can be completed in one day with simple steps. We also took into account the equipment commonly accessible to many research groups and the use of open-source software 36 for the alignment of correlated images.
The complete substitution of standard heavy metal contrast agents by platinum deposition brings homogeneous contrast and stability to the sample. However, fine structures with reduced topographic details are difficult to resolve (ER, microtubules) and complex homogeneous regions (neuropil in brain tissue) are difficult to identify ( Supplementary Fig. 3). Platinum replica protocols for cells 37 and tissue 19 have been applied to obtain correlative super-resolution microscopy with 3D electron microscopy information. Further developments of the protocol could provide adequate conditions for the finest structural elements. Another limitation of the present technique is the application to serial sections 26 , obtaining serial Tokuyasu sections are challenging and often only with few sections possible.
Our approach incorporates the Tokuyasu technique with simple and quick steps to perform routine correlative imaging at subcellular resolution using standard equipment. The direct topographic contrast obtained by this method represents a novel approach; visualizing sections without need of heavy metal solutions or replica preparation.

Material and M ethods
All animal experiments were conducted according to Swiss Laws and approved by the veterinary administration of the Canton of Zurich, Switzerland. Adult mice (n = 3) were anaesthetized with isoflurane (Baxter) inhalation combined with intracutaneous Temgesic (Reckitt Benckiser) followed by perfusion with 4% formaldehyde in cacodylate buffer (0.1 M, pH 7.4). Kidneys were transversally sectioned to 500 micron thick using a vibratome (Pelco 101, Vibratome R Series 1000, Ted Pella, Inc.) and postfixed with 2% formaldehyde and 0.025% glutaraldehyde in cacodylate buffer for 16 hours at 4 °C. These sections were then cut with a scalpel into 1 mm long pieces and immersed in 2.3 M sucrose overnight at 4 °C and stored at − 20 °C.
Light microscopy imaging. Before imaging, DAPI (4′ ,6-diamidino-2-phenylindole dihydrochloride, Sigma, 1:250 dilution) was applied for 10 sec followed by washes in PBS (3 × 10 sec). To avoid drying of the sample, the sections were incubated in a solution of glycerin (80%) and PBS (1:1) for 10 sec and then transferred with the section facing down to a glass bottom (thickness 170 μ m + /− 5 μ m) petri dish (Ibidi). A pipette was used to remove most of the liquid underneath the wafer in order to bring it as close as possible to the bottom of the petri dish.
Confocal laser scanning microscopy was performed on Leica SP5 and SP8 inverted microscopes (Leica Microsystems) with a 63x/1.4 NA oil immersion objective. Image stacks were acquired (60 × 60 × 170 nm) with a pinhole closed to 0.7 Airy units.
Super-resolution imaging. Wafers were mounted section side down in a petri dish (glass bottom, thickness 170 μ m + /− 5 μ m, Ibidi) onto a drop of a 1:1 mixture of glycerin (80%) and an imaging buffer containing an oxygen scavenging system (200 mM Phosphate buffer containing 10% glucose, 0.5 mg/ml glucoseoxidase, 40 μ g/ml catalase, 15 mM beta-Mercaptoethylamine (MEA), pH 8.0). Excess buffer was removed until the wafer was brought as close as possible to the cover glass.
The sample was first illuminated with the 642 nm laser at maximum laser power to ensure an efficient transfer of the fluorescent molecules into the off-state. Images of 180 × 180 pixels were then taken with an integration time of 15 ms in epifluorescence mode. A total of 30,000 frames were collected for each reconstruction.
Image reconstruction and visualization via Gaussian fitting was performed with the LAS AF software (Leica Microsystems).
Shadowing. Once LM imaging was performed the wafers were "lifted" up by application of buffer solution close to the edges of the wafer. After few washes in PBS (3 × 10 min) the wafers were quickly washed with 2% methylcellulose (Sigma-Aldrich) (2 × 5 sec) and left in a drop of methylcellulose at 4 °C for 1-2 min. The wafer was then placed, as fast as possible, in an Eppendorf tube and centrifuged at 14100 × g for 90 sec, mounted on a SEM aluminum stub (Agar Scientific) using an adhesive carbon pad, left drying at room temperature for 5 min and transferred to an electron beam evaporator (MED 020, Leica Microsystems). The specimen was then coated with Platinum/carbon (Pt/C, 2 nm) by unidirectional or rotary shadowing at an angle of 8 degrees (Supplementary Fig. 4).

Scanning electron microscopy imaging. Sections were imaged with a Zeiss Supra 50 VP and Zeiss
Auriga 40 SEM (large images above 4096 × 4096 pixels were acquired with the FIBICS Nanopatterning and Visualization Engine, FIBICS Incorporated) at an acceleration voltage of 1.5 keV, with a 30 μ m aperture and an InLens SE detector at a working distance of 2 mm or an SE2 detector at 5 mm working distance. Images were acquired at 4-5 nm pixel size.

Registration of LM and EM images. Alignment of light and electron microscopy images was done with
TrakEM2 36 within the open-source platform Fiji 38 . Firstly, the LM image was aligned based on manually inserted landmarks from the nuclear DAPI signal and their corresponding signals on the SEM image. After this rough alignment, a finer alignment was performed by registering the centre of several (3 to 5) clearly identified labelled structures (example, round mitochondria expressing Tom20) and their corresponding signals on the SEM image. We used the elastic alignment algorithm in TrakEM2 36 . For data presentation, the light and electron microscopy images were merged with the merge channels plug-in from the open source image processing software Fiji.