Retracing in Correlative Light Electron Microscopy

Abstract

Correlative light electron microscopy (CLEM) combines the strengths of light and electron microscopy in a single experiment. There are many ways to perform a CLEM experiment and a variety of microscopy modalities can be combined either on separate instruments or as completely integrated solutions. In general, however, a CLEM experiment can be divided into three parts: probes, processing, and analysis. Most of the existing technologies are focussed around the development and use of probes or describe processing methodologies that explain or circumvent some of the compromises that need to be made when performing both light and electron microscopy on the same sample. So far, relatively little attention has been paid to the analysis part of CLEM experiments. Although it is an essential part of each CLEM experiment, it is usually a cumbersome manual process. Here, we briefly discuss each of the three above-mentioned steps, with a focus on the analysis part. We will also introduce an automated registration algorithm that can be applied to the analysis stage to enable the accurate registration of LM and EM images. This facilitates tracing back the right cell/object seen in the light microscope in the EM.

Introduction

The combination of light and electron microscopy using a correlative light electron microscopy (CLEM) approach is a powerful tool for the study of dynamic intracellular membrane trafficking events at high resolution (Brown et al., 2012, Kukulski et al., 2012, van Weering et al., 2010). Light microscopy (LM) involves the visualization of both live and fixed cells at low resolution using fluorescent markers, while electron microscopy (EM) can examine fixed cells at near atomic-level spatial resolution, revealing the detailed ultrastructure of the cell and the three-dimensional intracellular architecture. When utilized together in a single CLEM experiment, these two imaging modalities provide greater insight into subcellular protein localization and trafficking than that, which is achievable by each technique alone.

In order to localize proteins, they usually have to be marked with a specific label. Fluorescent proteins such as GFP and probes such as the Alexa Fluor dyes can be employed in fluorescence light microscopy (FLM) to reveal the localization and trafficking of a particular protein of interest within the cell. The movement of this protein and its colocalization with other labeled proteins can be examined to reveal information about its function. However, only fluorescently labeled proteins and/or cellular compartments are visible using FLM and, therefore, this technique is unable to reveal details about the cellular environment in which the protein is located (van Weering et al., 2010). Furthermore, FLM and indeed many conventional LM imaging techniques are limited by the resolution capabilities of the microscope and thus cannot resolve structures closer than ~ 150 nm.

The diffraction limitations of conventional LMs can be overcome using super-resolution imaging techniques, such as photoactivated localization microscopy, stochastic optical reconstruction microscopy, and 4Pi microscopy (Betzig et al., 2006, Hell and Stelzer, 1992, Huang et al., 2010, Rust et al., 2006). The increased resolution afforded by these super-resolution techniques enable the study of single molecules and protein–protein interactions in live cells. However, even these techniques are incapable of visualizing process that occur at the nanometer scale and lack much needed information about the cellular environment and nature of the surrounding unlabeled structures (Betzig et al., 2006, Gustafsson, 2005, Schönle et al., 2000). In addition, although constantly improving, the live-imaging capabilities of most super-resolution LM techniques are currently limited.

By contrast, EM is a high-resolution technique capable of revealing the precise location of a labeled protein and the ultrastructure and morphology of the compartment in which the protein is localized (Slot & Geuze, 2007). However, due to sample preparation and properties of the transmission electron microscope (TEM), it is not possible to use this technique to study dynamic and/or transient trafficking events in live cells, other than to analyze many time-resolved snapshots and reconstruct the event retrospectively (Kukulski et al., 2012). In addition, due to the small sample size and thickness required for penetration by the electron beam, it is somewhat challenging and time consuming to identify rare events within a collection of cells.

CLEM thus bridges the gap between light and electron microscopy and can overcome many of the limitations of these individual techniques. It allows first for the identification of the protein of interest at the LM stage and for instance movement or approximate colocalization and then enables information about “reference space” to be gathered at the EM level. In general, CLEM experiments first involve the imaging of a cell labeled with a fluorescently tagged protein or probe using the LM followed by sample processing, embedding, and sectioning, before finally retracing and analyzing the same cell using the EM. These three steps: probes, processing, and analysis in the CLEM process are necessary for almost any CLEM experiment and we will briefly discuss the current status for each. For the analysis part, we will focus on CLEM using transmission EM. Relocation methods for the SEM are described in some of the other chapters in this book.

A variety of CLEM-specific probes and markers have been developed over recent years to study protein trafficking and localization in a range of different applications (reviewed in Giepmans, 2008) (Fig. 1.1). In general, a suitable CLEM probe consists of a fluorescent component that is visible by LM and an electron-dense particle that is detectable using EM.

When designing a CLEM experiment, it is important to establish which CLEM probe is the most effective for the specific scientific question being asked. Ideally, both the fluorescent and electron-dense components should be linked to a single molecule to provide the most accurate correlation between the light and electron microscopy images. However, this is not always possible due to issues concerning fluorophore quenching by the gold particle, probe aggregation, and mistargeting (Brown and Verkade, 2010, Kandela and Albrecht, 2007). Therefore, a combinatorial approach is sometimes taken using two separate probes that interact together, most commonly a fluorescently tagged antibody and a colloidal gold marker.

CLEM probes can be grouped into two broad categories: invasive and noninvasive probes. Noninvasive probes either contain both LM and EM visible components conjugated to an antibody or ligand or a sequence encoding a fluorescent moiety that can be converted into an electron-dense particle. By contrast, invasive probes often contain separate fluorescent and electron-dense elements and require permeabilization of the cellular membrane, post fixation, to enable penetration into the cell interior. However, permeabilization of cellular membranes can alter the ultrastructure of the cell when examined using the EM, therefore, noninvasive probes are more desirable.

Noninvasive probes containing both LM and EM visible elements, such as quantum dots (QDs) can be attached to a ligand or antibody that binds to an extracellular tag. QDs are an exciting noninvasive probe that have generated much interest as suitable candidates for use in live labeling CLEM studies (Giepmans, Deerinck, Smarr, Jones, and Ellisman, 2005). They consist of a fluorescent, electron-dense crystal core containing cadmium telluride or cadmium selenide that is visible at both the LM and EM stage (Giepmans et al., 2005). QDs can be internalized into the cell when coupled to an antibody or targeting molecule which is exposed to cell surface, and although care must be taken (and controls included), they traffic in a similar manner to the uncoupled endogenous protein (Giepmans, 2006).

Alternatively, noninvasive probes can be linked to an endogenous protein through a genetic tag or linkage, such as the FlAsH/ReAsH system and mini singlet oxygen generator (mini-SOG). The fluorescent flavoprotein, mini-SOG, engineered from the blue-light photoreceptor, Arabidopsis phototropin 2 is a useful noninvasive biosynthetic CLEM probe for labeling endogenous proteins (Shu et al., 2011). It functions in a similar manner to the biarsenical derivative of resorufin, ReAsH (Gaietta et al., 2002) by generating singlet oxygen species that catalyze the polymerization of diaminobenzidine (DAB), following photooxidation. Mini-SOG is genetically tagged to the protein of interest, thus providing a true correlation between the light and electron microscopy images.

Furthermore, endogenous proteins can be directly tagged with a fluorescent moiety, such as GFP, which can be converted into an electron-dense precipitate by photooxidation of DAB (Grabenbauer et al., 2005, Meisslitzer-Ruppitsch et al., 2008). Alternatively, ectopically expressed proteins can be fused to GFP and a tag, such as the peroxidase protein, APEX or the metal-binding protein metallothionein, which are detectable by EM through photooxidation of DAB or through binding to gold clusters, respectively (Martell et al., 2012, Risco et al., 2012). These noninvasive approaches are useful for correlative studies because the EM component is essentially generated from the LM-visible tag or the LM and EM components are directly linked to the protein of interest, thus enabling true correlation of LM and EM images.

As described above, probes are often linked to a ligand or antibody that recognizes a protein or exofacial tag that is exposed to the extracellular milieu and subsequently internalized into the cell. Alternatively, when the antigen or protein of interest remains intracellular and the probe is membrane impermeable, a noninvasive probe must be added following fixation, permeabilization, and often ultrathin sectioning. An antibody linked to FluoroNanoGold is often used following chemical permeabilization of the cellular membrane by detergents or inorganic solvents (Robinson & Takizawa, 2009). Following permeabilization, this probe penetrates through the membrane into the cell due to its small size, but requires silver enhancement in order to visualize the gold in the EM.

There are multiple different ways to perform a correlative experiment and a number of different microscopy modalities can be employed at both the LM and EM stages (Fig. 1.2). The experimental approach once again taken depends on the scientific question that is being asked and the CLEM probe that is being used.

If a transient or dynamic intracellular trafficking event such as the fusion of two compartments or the redistribution of a protein from one compartment to another is being studied, then a live-cell imaging approach may be taken at the LM stage. This involves visualizing and acquiring serial stacks of cells labeled with a fluorescently tagged protein, such as GFP, or a CLEM probe with a fluorescent moiety using a confocal microscope. Once the event of interest has been identified and recorded, the sample is then chemically fixed or immobilized by freezing, processed for EM and the same cell is relocated in the TEM.

When studying transient endocytic trafficking events, the rapid fixation of the specimen is imperative. Consequently, live-cell imaging combined with high-pressure freezing (HPF) or plunge freezing is desirable and retains good ultrastructure of the specimen (Brown et al., 2009, McDonald et al., 1998, Verkade, 2008). Chemical fixation is restricted by the speed of the diffusion of aldehydes and the formation of cross-links within the sample and may introduce artifacts into the sample (Kellenberger et al., 1992). It is, however, often used when further immunolabeling is required postembedding and sectioning.

An alternative CLEM approach may be taken when examining rare events within a population of cells. In this instance, the LM stage is required to identify the cell expressing the particular protein or phenotype and this can be done following EM processing and sectioning and does not require live-cell imaging. This is routinely used in combination with Tokuyasu immunolabeling of cryosections (Tokuyasu, 1973). The ultrathin sections are incubated with antibodies and gold markers and then imaged first by LM and then transferred to the TEM microscope (Hodgson, Tavare, & Verkade, 2014). The correlation of LM and EM images using this technique is much simpler as both imaging modalities are performed on the same sections and therefore any alterations in the appearance of the cells following embedding and sectioning will not interfere with the image registration. Furthermore, this approach allows for the examination of multiple cells and proteins within a sample. This technique has also been adapted to use in combination with live-cell imaging and to study cells in situ, using a flat embedding protocol (van Rijnsoever, Oorschot, & Klumperman, 2008).

The final stage of the CLEM protocol involves the analysis and relocation of the same cell identified using the LM, in the EM. This relocation is most frequently done manually, aided by finder grids, fiducial markers, and the surrounding cell features. The sample is first analyzed at low magnification in order to identify the particular cell of interest, before increasing the magnification to visualize the ultrastructure within that cell and the precise location of the CLEM probe. The LM and EM images are finally registered by manually creating image overlays as described previously (Keene, Tufa, Lunstrum, Holden, & Horton, 2008).

The manual relocation and registration of LM and EM images can be challenging and time consuming and, therefore, it is important that accurate automated registration tools are developed. Distortions in the section created during the processing and sectioning of the specimen can result in differences in cellular shape, orientation, and features making the manual image registration extremely difficult. In addition, the delay between LM imaging of the event of interest and fixation or freezing of the sample may also affect the correlation of the images.

Many of these problems can be circumvented by embedding and sectioning samples prior to LM imaging, as done during a number of different studies including sectioned light electron microscopy (Hodgson et al., 2014), cryo-EM (Chapter 10) and lowicryl embedding of HPF samples (Kukulski et al., 2011, Peddie et al., 2014). Recent advancements have led to development of automated image registration processes for some of these techniques, which are discussed below.

Sectioned light electron microscopy using the Tokuyasu technique for labeling ultrathin sections has been successfully used in a variety of CLEM studies (Koster and Klumperman, 2003, Polishchuk et al., 2000). In this technique, sections are picked up on coated finder grids, which are visible in both the LM and EM and immunostained with various primary antibodies, fluorescent secondary antibodies, and gold markers. The ultrathin sections are then examined by FLM and the position of particular cells of interest on the finder grid are recorded and relocated back in the EM. Alternatively, GFP-expressing cells can be labeled with an anti-GFP primary antibody and gold-conjugated secondary (Hodgson et al., 2014). This method overcomes the issues of artifacts or alterations introduced during specimen processing and sectioning interfering with the registration of the two images, but still requires manual searching for the cell of interest and correlation of the EM and LM images.

LM imaging following sample preparation has also been applied to study vitrified samples in cryo-LM and EM using a specialized cryo-stage attachment (Sartori et al., 2007, van Driel et al., 2009). In this method, cells are frozen and retained at cryogenic temperatures throughout imaging at LM and EM stages. This ensures that the samples remain unaltered between LM and EM imaging and no artifacts are introduced during the processing of the sample, such as during the fixation, dehydration or staining (Lucic et al., 2007). Furthermore, cryo-immobilization and processing for cryo-EM retains samples in a hydrated state preserves both fluorescence and ultrastructure (see also chapters 9 and 10). Recent studies have combined cryo-LM and EM with an automated 2D coordinate system in order to identify the cell of interest (Fukuda et al., 2013). This technique first involves the manual rough correlation of the LM image with a low-magnification EM image using grid bars to find the region of interest. Finally, reference points, such as holes in the support film and grid bar corners within the sample are identified and used to create a coordinate transform between the two images to identify the cell of interest in the EM.

Hydrophilic resins, such as lowicryl have been shown to allow preservation of GFP fluorescence following embedding and therefore are useful for FLM imaging after sample preparation (Keene et al., 2008). HPF samples are freeze-substituted and then embedded in lowicryl resin. This method has been successfully utilized to correlate fluorescence with ultrastructural information to gain insight into a variety of different biological systems and processes, including HIV particle attachment to mammalian cells, zebrafish embryo development, and endocytosis in budding yeast of HPF samples (Kukulski et al., 2011, Nixon et al., 2009). In these aforementioned studies, fluorescent microspheres that are visible in both LM and EM were used as fiducial markers in combination with an automated correlation procedure to identify the cell of interest from the LM images (Kukulski et al., 2011). The samples are incubated with the microspheres prior to imaging and the location of each fiducial marker is measured and used to calculate an optical transform between the EM and LM images. The optimal transform can then be applied to the areas of fluorescence in the LM image to identify the corresponding region in the EM. This automated process dramatically decreases the time taken to identify the cell of interest and provides near nanometer registration accuracy between the two images (Schorb & Briggs, 2013).

The techniques listed above overcome the issues of fixation and dehydration artifacts and sectioning distortions that can alter the appearance of the cell following LM imaging. However, they still require transfer of samples from LM to EM, which can lead to sample damage and increase time taken for relocation of the region of interest.

An integrated light and electron microscope (ILEM) allows for the imaging of LM and EM samples at cryogenic temperatures in a single microscope (Agronskaia et al., 2008, Faas et al., 2013). The integration of the two microscope modalities enables the imaging of the sample at LM followed immediately by the imaging at EM and removes the need for complicated correlation procedures. Regions of interest from the LM images can be used to calibrate the microscope allowing for precise and accurate location within the EM (< 1 nm) (Faas et al., 2013). The drawback of using an integrated system is that the user cannot choose or tailor the imaging modalities to the experimental question as seen for a two-step imaging approach, thus making this procedure unsuitable for some applications where a particular microscopy technique (e.g., live imaging to determine the origin of an event) is required. In this particular instance, the reduced resolution of the LM in the ILEM (around 500 nm) may limit the identification of smaller structures. But it is an ideal tool to find objects of interest in a cryo-TEM where limited electron doses are essential (chapter 10).

As such, integrated microscopy but also LM imaging following embedding are less well suited to study dynamic and transient intracellular trafficking events as neither technique allows for live-cell FLM imaging. Furthermore, if an event is rare within a population of cells, it is often easier to identify this event first at the LM stage before sample processing. In general, the cell of interest is imaged live at the LM stage, the sample processed and the location of that same cell of interest recorded and retained throughout the sample processing before relocation in the EM. This is somewhat challenging when you consider that the LM samples are generally considerably larger than those imaged in the EM, corresponding to whole monolayers of cells and ultrathin sections of a small region of those cells, respectively. Therefore, the LM samples must be trimmed and sectioned prior to the EM relocation. It is not possible to use finder grids or fiducial markers to relocate the cell of interest in the EM and register the LM and EM images following live-cell imaging and sample processing. The identification of the cell of interest is instead achieved by examining the surrounding cell shapes, orientation, and nucleus location and comparing these to the LM images acquired before embedding and sectioning (see Fig. 1.3). This can be an extremely challenging and time-consuming process.

To overcome some of the challenges described above, we have developed algorithms that aid with retracing. In the remainder of this chapter, we focus on a CLEM workflow for cultured cells to retrace (very) rare events (see Fig. 1.4). We describe an automated image registration process that can be utilized to enable the accurate registration of LM and EM images. In the last part, we will describe the current status of this method and how it can be developed further.

Glass-bottomed dishes (Mattek, P35G-2-14-C-GRID) or live-cell carriers (Brown et al., 2012, Verkade, 2008) containing gridded coverslips are routinely used to assist in the trimming and relocation of the cell of interest throughout sample processing. For chemically fixed samples, this involves growing cells in an imaging dish containing a glass coverslip embossed with a finder grid pattern (van Weering et al., 2010), while cells for HPF-CLEM are grown on sapphire discs inside a live-cell carrier containing a carbon-coated finder pattern (Brown et al., 2012).

HeLa cells are seeded into a glass-bottomed imaging dish containing an embossed finder grid pattern at 400,000 cells per dish and left to adhere overnight. Rare events of interest such as cell division or specific stages within that process may be identified by fluorescence alone if detailed information of the ultrastructure is not required. If in addition, the localization of a protein is required, an immunogold-labeling step is included. Dishes are chemically fixed in 3% PFA containing 0.05% GA, permeabilized in 0.1% saponin before immunostaining with a primary antibody followed by FluoroNanoGold (Polishchuk et al., 2012, Polishchuk et al., 2000). Nuclear stains such as DAPI can be added to the cells before FLM imaging to visualize the nucleus which will later aid in the registration of LM and EM images.

Alternatively, if the antigen of interest is exposed to the extracellular media then cells may be labeled live using a noninvasive CLEM probe containing both the fluorescent tag and EM visible particle before fixation.

Fluorescence LM imaging is performed using a confocal scanning light microscope. The cell of interest is first identified and then bright-field images of the position of the cell on the grid and fluorescence images of the nucleus and the localization of the CLEM probe are acquired (Fig. 1.5). Z-stack slices are then taken to reveal the localization of the fluorescent CLEM probe throughout the cell. Finally, an overview of the grid pattern is taken either by switching to a lower magnification lens or by acquiring multiple bright-field images of the area surrounding the cell of interest and stitching a mosaic together using the MosaicJ plug-in in FiJi imaging software. This final step is crucial when trimming the sample to size before sectioning and for retracing the cell of interest back in the EM.

Following imaging the sample is processed and embedded in plastic resin. However, first the FluoroNanoGold particle must be enlarged so it is more visible by EM. This involves the enhancement of the 1.4 nm FluoroNanoGold using commercially available Gold or Silver enhancement kits (Polishchuk et al., 2012). The sample is postfixed in 2% osmium tetroxide in phosphate buffer before washing with buffer and water. It is possible to store the samples overnight at 4 °C in H₂O at this point without significant alteration of the ultrastructure. Next, the samples are stained with 3% uranyl acetate in H₂O and dehydrated with a graded series of ethanol (70%, 80%, 90%, 96%, 100%, and 100%). The dish is then covered in EPON resin and left at room temperature for at least 2 h. The EPON is removed and replaced with fresh resin, which is left on the dish overnight at 60 °C. Finally, fresh resin is added to the center of the imaging dish and a resin-stub placed on top. The samples are then left at 60 °C for 24–48 h to allow the resin to polymerize.

Next, the coverslip and imaging dish must be removed before trimming and sectioning (Fig. 1.6). This is achieved by rapidly transferring the embedded dish between liquid nitrogen and boiling water to dissolve the glue holding the coverslip in place. The coverslip can then be removed using a razor blade, leaving an imprint of the finder grid in the hardened resin block. This imprint can act as a guide during trimming of the block when correlated with the LM images in order to relocate the cell of interest and reduce the size of the sample prior to serial sectioning (Polishchuk et al., 2000). The imaging dish is cut and excess resin is removed using pliers and then the resin block is trimmed to a pyramid, before cutting 70–300 nm serial sections and collecting the section ribbons on pioloform-coated EM slot grids.

Once the resin block has been trimmed and serial sections cut (either 70 nm for standard imaging or 300 nm for electron tomography), the sample is then examined in the TEM. The finder grid imprint is lost after the first section of the block is cut so the grid is no longer present to assist in the relocation of the cell of interest. The relocation is performed manually by comparing the orientation, position, and shape of the cells from the LM image, as demonstrated previously (see Fig. 1.3). The first few sections cut from the block should have retained the grid bar indent; therefore, it is often easier to find the cell of interest in these sections using the grid coordinates as a reference. Once the position of the cell of interest has been found in the first section on the grid, it is possible to use this as a guide to find the cell of interest in the other sections within the same ribbon. However, the imprint is not always obvious in the EM sections.

Thus, as discussed previously, finding back the cell and/or organelle of interest observed by LM in the EM can be a real challenge, especially when groups of cells have been imaged. Hence, we set out to develop automated solutions that would aid us in this process (Fig. 1.7). The method is based on the feature that when cells are seeded they distribute in unique patterns each time. Following image acquisition, the EM micrograph and LM images are automatically aligned using a feature-based image registration algorithm (Nam et al., 2014). This algorithm registers the EM image with the bright-field LM image, using the geometric centers (cell centroids) of cells in both images.

First, cell centroids are detected. Centroid detection is achieved by creating a voting image, using a cone-shaped voting kernel. The vertex of the kernel is oriented in the negative gradient direction (at specific key points from the image). The voting image is then a summation of kernel contributions from each key point (Qi, Xing, Foran, & Yang, 2012). Three parameters define the voting area: r_min, r_max, and θ. r_min and r_max define the length of the voting kernel. Ideally, the radius of the cell should be halfway between r_min and r_max. r_min is set to be half the length of the cell radius and r_max to be 1.5 the length of the radius. The distance between these two values can be reduced to give a sharper voting image, but only if there is a lot of consistency in the observed cell size, and the cells appear very circular. Likewise, if there are larger variations in cell shape and size, we recommend increasing the distance between these two values. The angle θ is the angle at the vertex of the cone. We used 30°, while this value is suitable for most cases, decreasing its value can result in a sharper voting image (but with decreased robustness). Importantly, this method accounts for variation in the contrast between EM and LM images and section artifacts—such as dirt particles that would otherwise skew the calculation of the cell centroid.

The registration algorithm next calculates the transformation parameters using the cell centroid positions so that the EM and LM images align resulting in an LM and EM image overlay that is comparable to manually aligned images, but has been achieved more quickly and easily (Fig. 1.5). This is achieved by fitting disks onto cell center positions—a little smaller than the size of a cell, to add robustness—and then rotating and shifting the EM image across the LM image. The position with best overlap is used to obtain the correct parameters. The process is also repeated for the flipped EM image. Most importantly, the method is also able to account for the appearance or disappearance of nuclei in serial sections of the cell. This is a common occurrence in cultured cell monolayers, as some cells are tightly attached to the surface, while others loosely adhere and thus appear in a higher plane.

Automatically aligned images were compared to manually registered images to test the accuracy of the method. The manually registered images were created using Fiji by manually selecting and subsequently aligning corresponding points in the LM and EM image. The differences between these manually aligned images and those done with our method were found to be negligible.

• 35-mm glass-bottomed imaging dishes containing embossed finder grid (MatTek Corporation, P35G-2-14-C-GRID).

• PHEM buffer: 60 mM PIPES, 25 mM HEPES, 10 mM EGTA, 2 mM MgCl₂ (pH 6.9).

• Fixative: 3% Paraformaldehyde and 0.05% glutaraldehyde (GA) in PHEM buffer (pH 6.9).

• Permeabilization solution: 0.1% saponin in PHEM buffer.

• Blocking solution: 1% bovine serum albumin, 0.1% saponin in PHEM buffer.

• FluoroNanoGold: Alexa Fluor® 594 FluoroNanoGold™ Fab’ Fragment goat anti rabbit IgG (Molecular Probes, #A24923).

• 5 μg/ml 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI) (Molecular Probes, #D1306).

• R-GENT Silver Enhancement-EM (Aurion, #500.044).

• 200 mM Phosphate buffer (PB), pH 7.4.

• 1% Osmium tetroxide (OsO₄) in phosphate buffer.

• 3% Uranyl acetate (UO₂(CH₃COO)₂⋅ 2H₂O) in H₂O.

• 812 EPON resin (TAAB).

• Confocal microscope: Leica SP5 confocal imaging system.

• Ultratome: Leica UC6 ultramicrotome.

• EM microscope: FEI 120 kV BioTwin Spirit TEM.