Assaying Three-Dimensional Cellular Architecture using X-Ray Tomographic and Correlated Imaging Approaches

Much of our understanding of the spatial organization of and interactions between cellular organelles and macromolecular complexes have been the result of imaging studies utilizing either light- or electron-based microscopic analyses. These classical approaches, whilst insightful, are nonetheless limited by either restrictions in resolution or by the sheer complexity of generating multi-dimensional data. Recent advances in the use and application of X-rays to acquire micro- and nano-tomographic datasets offer an alternative methodology to visualize cellular architecture at the nanoscale. These new approaches allow for the sub-cellular analyses of unstained vitrified cells, three-dimensional localization of specific protein targets, and have served as an essential tool in bridging light and electron correlative microscopy experiments. Here, we review the theory, instrumentation details, acquisition principles and applications of both soft X-ray tomography and X-ray microscopy, and how the use of these techniques offers a succinct means of analyzing three-dimensional cellular architecture. We discuss some of the recent work that has taken advantage of these approaches, and detail how they have become integral in correlative microscopy workflows.


Introduction
Understanding the three-dimensional spatial organization of inter-and intra-cellular components is paramount in elucidating their function in both healthy and diseased states. Modern microscopy has evolved from a disparate array of isolated techniques to a cohesive toolbox of imaging approaches that routinely allow the visualization of proteins of interest, dynamic events, as well as how changes in cellular morphology and macromolecular structure affect physiology. However, despite the unique insights enabled by these approaches, limitations in each imaging modality, along with the increasing desire to generate data from a near-native state with limited perturbation, have led to the need for and development of new approaches and correlative techniques.
Light microscopy approaches are widely available, relatively simple, and have enabled significant insights into cellular organization. This vast array of imaging modalities, spanning from brightfield transmitted light microscopy to superresolution fluorescence nanoscopy, provide tremendous flexibility in the analyses of cellular and sub-cellular morphology. Three-dimensional light microscopy is routinely performed through the use of confocal (1-3), lightsheet (4), and multi-photon (5,6) approaches. These techniques are, by necessity, dependent on either fluorescent reporter probes or on the presence of intrinsic signals such as autofluorescence or harmonic signal generation (7). As such, signals obtained from these imaging experiments are limited to visualizing structures that are either "tagged" or which generate the intrinsic signals being measured, thus rendering the remainder of the cell invisible.
Electron microscopy (EM) is routinely used to explore the ultrastructural organization of cells and tissues at nanometer or less resolution. As the electron beam has a limited depth of penetration, volumetric analyses have often proven to be complex, difficult, and time-consuming. Typically, preparation of samples for transmission electron microscopy (one of the most widely used ultrastructural imaging techniques) involves a primary aldehyde fixation with a secondary heavy metal fixation, infiltration into a resin support matrix, and thin sections taken (50-100 nm), which can then be visualized under the microscope. This methodology can be extended to three-dimensions using tilt-series tomography (8,9) and serial sectioning (10,11), with such approaches being utilized to study a wide array of tissues and cell types. These tomographic techniques, however, are limited by sectioning precision and the arduous alignment and processing steps, which, can be very time consuming. More recently, scanning electron microscopy has been used to generate volumetric data. Specifically, serial block-face electron microscopy (SBF-SEM) (12), array tomography (13), and the use of a focused ion beam in conjunction with SEM (FIB-SEM) (14) have all been used in nanotomography studies (15)(16)(17)(18). In spite of improved acquisition speeds and automation options, SBF-SEM and array tomography are limited by nonisotropic voxels, having different dimensions in the xy and z planes and thus obfuscating volumetric analysis, while FIB-SEM is limited by the small sample volumes that may be acquired (10 -40 µm in the z-plane).
In contrast, tomographic techniques that utilize X-rays, including soft X-ray tomography (SXT) (19) and X-ray microscopy (XRM) have recently been shown to be an invaluable tool in studying the spatial architecture of individual cells, as well as the localization of specific proteins of interest within cells and tissues. These approaches offer significant gains when compared with electron microscopy, with hard X-rays capable of penetrating through the entire organism, and soft X-rays allowing single-cell visualization of minimally-perturbed native specimens, without the need for contrast-enhancing agents. X-rays, with their substantially shorter wavelengths, offer almost an order of magnitude better resolution as compared to light microscopy, with more recent advances resolving objects down to 5nm (20). Naturally, these resolutions are atypical of the vast part of the work that has been done with the approach, with much more objective values shown in Table 1. Nonetheless, the use of X-rays on vitrified samples, offers a nanotomographic technique with preparation ease only matched by Cryo-FIB systems (21-23), but with substantially higher acquisition speed. Analyses of the mesoscale architecture, and the ability to target specific regions of whole tissues offers a bridge from the micro-to the nanoscale, and X-ray-based microscopy has become a vital step in correlative studies, linking multi-fluorophore microscopy with ultrastructural analyses. The cryopreservation requirements of soft X-ray microscopy has expanded to include cryo-light microscopy (24), while XRM has been employed as a mapping tool to facilitate localization of regions of interest in 3D nanotomography studies (25).

Soft X-Ray Tomography Theory and Instrumentation
On the electromagnetic spectrum, soft X-rays reside between the extreme ultraviolet and X-ray bands, from 250eV to several thousand eV, corresponding to wavelengths of 5 nm to approximately 2Å (26). This range includes specifically well-defined L-and Kedges, for identifying biologically relevant elements, such as Carbon, Nitrogen, and Oxygen. The so-called 'water window' (Figure 1C), from the Carbon k-edge at 284eV (4.38nm) to the Oxygen k-band at 543eV (2.34nm), is of particular interest in biological studies as it offers intrinsic contrast between water and protein (27). Visualizing unstained cells and tissues at such energies would thus lead to strong X-ray absorption by macromolecules, such as lipid membranes or intracellular organelles, with only minimal attenuation by water molecules. Differences in the absorption relate to a parameter termed the linear absorption coefficient (LAC), whose constant values can be used to identify specific intracellular structures (27). This property presents the underlying principle of soft X-ray tomography (SXT), and it led to the development and refinement of the technique over the past thirty years (28). Over the past two decades, the use of SXT has expanded, and, along with concurrent advances in detection technologies, has allowed for structural studies of samples, preserved in their native states, which were previously inaccessible.
As in all microscopes, a soft X-ray microscope (SXM) is comprised of a source, a detector, and a means of driving the beam to the sample (condenser) and focusing it to the detector (objective). The X-ray source necessary for an SXM need not be bright, but it does require a large photonflux (29). These requirements have generally restricted SXM to facilities where synchrotron radiation can be passed to low-emittance storage rings, where a series of undulators or bending magnets allow soft X-rays to be filtered and utilized for high resolution experiments (30,31). The complexity and scale of synchrotron facilities have kept SXM systems for biological samples to less than a handful worldwide (32). At the moment, there are four such facilities, including the Advanced Light Source (ALS) at Berkley, the BESSY II in Berlin, the Diamond Light Source in the UK, and ALBA in Barcelona. Smaller soft X-ray sources have been developed over the past decade, using nitrogen or methanol plasma (33,34), with the aim of producing a soft X-ray source suitable for a small laboratory. These sources, however, have not seen wide adoption, mainly due to their lower flux and the fact that unlike synchrotron radiation, plasma sources are not continuous. As such, data generated by them results in lower resolution, and requires longer integration times.
SXM systems condense soft X-rays onto samples and collect emission through diffractive elements called zone plates (29,35). These plates are comprised of alternating concentric rings of transparent and opaque "zones", with the outermost zone width (∆rn) defining the maximum resolution of a system ( Figure 1B). Zone plate resolutions range from 60 nm to 25 nm (36,37), allow imaging depths of up to 15 µm. Whilst offering enhanced resolution, zone plate use in soft X-ray microscopes does pose some constraints on imaging experiments. Firstly, zone plates are very inefficient, with most of the radiation being absorbed or diffracted, allowing roughly 10% through to the sample (38). Secondly, acquisition geometry requires the objective plate to reside millimeters away from the sample, which limits the usable energies of the microscope. Finally, the use of zone plates has historically been an inflexible solution, as changing of optics to alter resolution or imaging depth necessitates the complete venting of the imaging chamber, and careful realignment of all optical elements (29). Recent efforts have shown the utility of multiple micro zone plates, thus allowing for a more flexible, multiresolution approach (39).
In contrast to the complex X-ray sources, soft X-ray microscopes have modest requirements for detection. As soft X-rays that pass-through zone plates can be directly detected, there is no need for scintillation material. Thus, simple, back-thinned CCD cameras are sufficiently fast for data acquisition and the ability to modulate on chip gain levels and binning make such devices amenable for SXT imaging ( Figure 1A).
Among the chief advantages of SXT is that samples do not need to be chemically fixed or stained with exogenous contrasting agents. Such fixation protocols have been known to cause numerous artefacts, including uneven fixation, formation of aggregates, and morphological changes, such as dehydration (40-42). Yeast in particular, as described below, are notoriously unreceptive to chemical fixation, which results to dramatic morphological changes when compared with vitrified cells (43). As such, cells can be imaged in as close to their native states as possible. X-ray radiation readily damages cells (44), and most biological samples have to be vitrified to maintain their integrity during imaging. It has been reported that reducing sample temperatures to the cryogenic regime permits the acquisition of more than 1000 images without any evidence of radiation damage (45). Vitrification is typically achieved via plunge freezing into a cryogen such as liquid ethane, much like the manner used in cryo-EM (46,47), or by a cryogenic gas stream using N2-cooled helium (48). SXT of cryo-preserved cells thus requires the use of a cryo-stage to maintain the samples in a vitrified state. Depending on the vitrification approach and the stage design, this may require the use of vacuum chambers (29). Sample mounting is typically achieved by the use of either canonical-style TEM grids, or custom glass capillaries, which provide the ability to visualize the samples from any rotational angle (19,37). To facilitate reconstruction, most experiments employ gold bead fiducials, embedded on the exterior surface of imaging capillaries (49).

Acquisition principles and parameters
A single SXM image is simply a projection image and has no z-depth. Thus, an SXT dataset requires the sequential acquisition of images over a range of angular increments. The size of the specimen, along with resolution requirements and system limitations, dictates the necessary parameters for an optimal tomographic reconstruction. As images in a tomographic acquisition are attained in the axis perpendicular to sample rotation, and each reconstruction plane is independent of the rest, the resolution of the final tomogram is equal to the resolution perpendicular to the planes (50). The minimum number of equally-spaced projection images required to generate a tomogram with isotropic voxels that is collected over an angular range is given by the Crowther criterion, N = πD/d, where D is the sample diameter, and d is the spatial resolution (9,50,51). In practice, this means that for a 180º rotation, images would be acquired at increments of 1-2º, equating to 90-180 individual projection images (48).
The manner in which the sample is mounted to the imaging stage plays a vital role in an instrument's ability to generate complete tomograms. SXM systems that use flat TEM-style grids are limited to tilts of ±70º (32) by the short working distance of the objective zone plate, which have several disadvantages. Firstly, the inability to rotate the sample to 180º results in the 'missing wedge' problem, where there is insufficient data to generate a complete tomogram (49,51). Flat grids, when rotated through an imaging beam, will have a variable cross section of vitreous ice, with thicker sections at the limits of tilt (49). This variance results in an inability to consistently measure and corelate the LAC to specific intracellular structures. In contrast, using the glass capillary approach allows for a full 360º rotation, which in turn resolves the 'missing wedge' problem (48). Unlike flat grids, however, the capillary mounting approach is generally incompatible with adherent cell types (52).
Among the main advantages of soft X-ray tomography is the sheer speed of data acquisition. Due to the high-intensity synchrotron sources used, and the relatively thin samples typically imaged, acquisition rates are on the order of 5-10 fps (19,48). This allows an entire tomography dataset to be acquired on the order of 20 minutes, significantly improving throughput, and facilitating the use of high degrees of automation in some systems (32).

Applications of Soft X-Ray Tomography
Soft X-ray tomography (SXT) has evolved into an important tool for studying sub-cellular structures which have otherwise been inaccessible or difficult to assay. One such structure is the eukaryotic nucleus, whose spatial organization has been recently linked to the modulation of gene expression (53,54) as well as the development of various cancers (55)(56)(57). The resolution requirements and central location of the nucleus have, however, made topological analyses difficult. SXT has allowed the exploration of the relationship between the spatial organization and function of the nucleus in a minimally perturbed manner. In one particularly impressive study, Lomvardas and colleagues illustrated that the preferential expression of a specific olfactory receptor gene over others was due to the protein's proximity to the nuclear envelope, and such spatial organization may underlie monogenic gene expression (58). A more recent study utilized the SXT approach to explore the organization of chromatin in primary hematopoietic cells, revealing that both the amount of euchromatin and the spatial organization of heterochromatin govern the transition between pluripotent and progenitor cell states (59).
The ultrastructure of yeast cells has been difficult to study, largely due to the thick cell wall which prevents effective chemical fixation. Stripping of the cell wall leads to denaturation, altering the subcellular ultrastructure from the native state. High pressure freezing and subsequent freeze substitution processing methods are effective, but timeconsuming, and are ultimately limited by the data acquisition approaches. Furthermore, the divergent imaging approaches employed have contributed significant amount of data regarding different yeast species yet have offered limited understanding of the three-dimensional architecture. S. cerevisiae, specifically, has been studied using SEM (Figure 2A (43)), TEM ( Figure 2B (43)), and FIB-SEM ( Figure  2C (60)). While SEM data is particularly useful to analyze the topology of the cell walls, it offers limited organelle data. TEM, conversely, offers the best imaging resolution, at the expense of understanding three-dimensional organization. FIB-SEM provides excellent resolution and the ability to generate nanotomographic data, but long processing times and imaging have hindered these efforts. SXT has proven to be instrumental in overcoming these barriers and elucidating the internal mesoscale organization of several yeast species, including S. cerevisiae and S. pombe. Larabell and colleagues at the ALS, who have been at the forefront of this area and pioneered the development of SXT, have, in a series of studies, beautifully shown the morphology, density and organization of nuclei, mitochondria and lipid droplets in yeast species at resolutions down to 60 nm, while imaging over the span of minutes, rather than hours ( Figure 2D) (19,(61)(62)(63)(64).
The ability to study the three-dimensional organization of cells at nanoscale resolution using SXT has also been leveraged to elucidate the manner by which certain pathogenic microorganisms interact with eukaryotic cells. Using this approach, the structure of enveloped vaccinia virions, and the localization of high density regions, such as membrane-bound compartments have been extrapolated (65). More recently, studies have identified ER membrane alterations that are associated with Hepatitis C, showing colocalization between ER protrusions and viral RNA (66), as well as demonstrated that the initiation of herpes simplex type 1 infection evokes the generation of channels within the host nucleus that facilitate viral egress (67). These, and other studies have demonstrated the ability of SXT to assay previously undetected or underappreciated cellular phenomena.

X-Ray Microscopy
Despite the obvious advantages of SXT in terms of resolution and the visualization and localization of complex intracellular structures in near-native state, the technique requires high photon flux, which limits it to synchrotron x-ray sources and low x-ray energies which are required to access the 'water window' which restricts the thickness of samples to 10 -20 µm. Imaging much larger samples, such as whole organs or an intact embryo are simply not feasible. Furthermore, while LAC measurements can be instrumental in identifying intracellular structures, labeling target proteins with specific stains allows for increased precision in identifying and localizing such targets. X-ray microscopy (XRM), which utilizes a table-top hard X-ray source (> 10 keV) along with different modes of magnification, offers the ability to perform such experiments.

Instrumentation
Micro-CT sources are relatively compact, comprising of an electron gun (cathode) focused on a metal target (anode) typically made of tungsten or copper, housed in an evacuated chamber. The X-ray intensity is related, in a linear fashion to the current output to the cathode, while the potential difference between the cathode and anode dictates the energy of the emitted X-rays (68). This architecture permits the independent adjustment of both parameters, thus providing an extremely flexible imaging source. The geometrically divergent beam passes through the sample, which is typically mounted on a stage that has a large three-axis translational range and can rotate ±180º, and is then detected (69). Standard micro-CT instruments ( Figure 3A) suffer from a fundamental limitation, in that due to the divergence of the beam, the magnification geometry is solely dependent on the ratio of the source-to-sample and sample-to-detector distances. This is significant, as high-resolution can only be achieved when the sample is located as close to the source as possible. This has the consequence of limiting the range of sample motion and restricts the size of a sample that can be easily imaged. Detection is performed either via a flat panel detector (which is typical in micro-CT) or via a set of high numerical aperture objectives, that incorporate a scintillation layer in their front elements to convert impinging radiation into visible photons which are subsequently detected via an electron multiplied CCD (EM-CCD) camera ( Figure 3B) (70). This offers a secondary, optical magnification factor, allowing for what is termed 'Resolution at a Distance, or RaaD). This allows for substantially greater flexibility in the geometrical setup of an experiment, while still maintaining high-resolution over a much larger range of sample sizes (71).
While soluble, non-specific stains offer ease of use and an ability to visualize all tissues within a given sample, directing labels to specific targets remains a fundamental goal in XRM tomography studies. A number of contrast agents, for example, have been utilized in an intravasal manner to study lymphatics and microvasculature. Among these are barium sulfate (85,86) and lead oxide, the latter of which has been described as the 'gold standard' of fine vasculature visualization (87), though some work has shown that these agents might not fill smaller vessels, thus leading to an underestimation of the total vasculature (88). More recent studies have employed different tracing agents, such as gold nanoparticles ( Figure 3E) (89), bismuth formulations (90,91), and tantalum oxide (92), the results of which have shown great promise as tools for XRM vasculature analyses.
More specific targeting of proteins of interest has also been demonstrated with XRM, offering a way to directly quantify aggregation. One such approach uses immune precipitation to deposit metal ions in close proximity to the target of interest. Specifically, this method uses a horseradish peroxidase-conjugated secondary antibody which binds to a primary antibody, and locally produces singlet oxygen. A silver metal-deposition enzyme metallography kit then interacts with the singlet oxygen, thus reducing silver ions from the solution, and causing its deposition near the site of the primary antibody. Metscher and colleagues used this approach to visualize and study acetylated-tubulin in a chick nervous system as well as localization of type-II collagen in developing limbs (93). A similar approach has used a nickel-enhanced 3,3-diaminobenzidine (Ni-DAB) deposition (94) in conjunction with OsO4 to explore changes in avian excretory system morphology in RET tyrosine kinase mutants ( Figure  3F) (95). A set of novel genetically-encoded targeting approaches, including enhanced ascorbate peroxidase (APEX) (96) and mini singlet oxygen generator (MiniSOG) (97), were originally designed for use in electron microscopy, but have also been utilized to study the organization and distribution of specific cell types and localize specific structures using XRM in correlative studies (25,98).
Regardless of the staining modality employed, a key feature in datasets obtained using XRM are that they offer not only the spatial localization of structures and specific proteins, but also the density of aggregated contrast agent. These densities present a distinct advantage over other imaging modalities, as they directly corelate to the expressed protein they target (93). The use of proteinspecific staining, in conjunction with silver deposition, can thus be utilized to quantify a target gene product within a given tomographic volume.

Correlative Approaches
Using divergent imaging approaches to study a given sample is a tremendously powerful technique, allowing for analyses of a region of interest over different spatial and temporal scales, thus maximizing efficiency in resources and precious sample utilization, and extending the precision and breadth of complementary techniques. Thus, correlative light and electron microscopy (CLEM) approaches have become indispensable in associating ultrastructural organization with protein localization and temporally finite events (e.g. gene expression) (99). The recent popularization of cryogenic electron microscopy (cryoEM) (100,101), has also led to the development of correlative workflows with cryo fluorescence microscopy (cryoCLEM) (102,103). Both SXT and XRM have been shown to extend and enhance these correlative approaches, offering a bridge between light and electron microscopy, thus linking physiological processes, such as motility, metabolic transport and ion exchange to cellular ultrastructure.
As SXT necessitates the vitrification of cells to prevent radiation damage, correlative light workflows require the development and use of light systems capable of imaging fluorescence while maintaining cryogenic temperatures. Such cryo fluorescence imaging has gained much interest, not only due to the emergence of tomographic cryoEM studies, but also for the fact that cryogenic temperatures offer significant improvements in fluorophore stability (104), fluorescence yield (105), and resolution (106). While cryoCLEM with SXT has only been in use over the past decade, and developments in the workflow and technology are ongoing (49,52,107), the resulting data thus far has demonstrated the incredible potential for this approach. Applications have included identification of inactive X-chromosomes in female transformed thymic lymphoma cells (108), localization of vacuoles in yeast strains (109), and analysis of autophagosome formation in the endoplasmic reticulum of a mammalian cell (110). Correlation of the fluorescence and x-ray volumetric data is however dependent on the use of fiducial markers, which are typically fluorescent polystyrene beads which are capable of being visualized by both modalities.
While correlative light and electron microscopy techniques have been developed and improved upon for over four decades (111), XRM offers a means to bridge the two modalities with increased speed, precision, and reproducibility. Fiducial-based image registration (112,113) has been successfully used in CLEM studies, but due to the three-dimensional nature of the samples in question, this approach requires correlating six fiducials for localization, thus imposing limitations on the density of the markers. Most EM preparations involve secondary fixation with heavy metal solutions, such as osmium, lead, and uranium (114,115), thus rendering the samples opaque. As X-rays are readily attenuated by these compounds, EM staining is especially advantageous in XRM, and the resultant tomogram can be used as a three-dimensional map to locate regions of interest for subsequent electron microscopy (116). This approach has been demonstrated with FIB-SEM nanotomography of materials (117), in bridging brightfield light microscopy to TEM in the assaying of adipose tissue (118), as well as through the use of geneticallyencoded tags in murine brain sections, which were subsequently imaged via serial block face EM (25). More recent work has utilized XRM-dependent correlative approaches to an increasingly wide range of targets, from exploring ultrastructural properties of parasitic worms in the small bowel (119), to correlating histological and TEM imaging of Araneae papal organs (120), and to link intravital microscopy studies of mouse cerebral vasculature to FIB-SEM nanotomographic analysis (121). An added advantage of the high-resolution, mesoscale imaging, and speed of XRM acquisition is the ability to employ the technique as a region of interest localization tool. As an example, in exploring osseous metastatic disease in whole bones via TEM, we employed XRM within our workflow to facilitate the location of metastases within long bones (Figure 4). This allowed for targeted sectioning of the site of metastasis which would be impractical to accomplish without the aid of the XRM data. It is apparent that since its recent incorporation into correlative studies, XRM has become an indispensable tool.

Conclusions
The application of X-rays to study various biological samples is not a new technique. Indeed some of the earliest X-ray imaging was performed over a century ago, and zone plate-based systems were being developed as early as the 1960's (122). The past 20 years, however, have seen a rapid expansion in the application, utilization, and integration of X-raybased microscopy into workflows aimed to assay three-dimensional biological structures. Use of soft X-rays has been proven capable of studying minimally-perturbed cells, with resolutions nearing that of electron microscopy, and it holds the exciting prospect of being able to thoroughly explore the cellular mesoscale architecture (123). The development and application of metallic stains for XRM, and the ability to direct such staining to specific targets, offers the promise of exploring the means by which the spatial organization of any protein of interest effects the spatiotemporal dynamic properties of cells. The physical properties of both SXT and XRM hold unique quantification opportunities, with linear absorption coefficients offering the ability to identify sub-cellular compartments (123) and stain densities in XRM presenting a unique means of measuring protein aggregation. Application of X-ray microscopy techniques in EM correlative studies have demonstrated the increased efficacy the approach offers and the resultant increases in sample throughput. Ongoing developments in targeted contrast agents, workflow integration, and system flexibility and expansion, make SXT and XRM exciting tools for future studies of cellular organization and function.

Conflict of Interest
The authors declare that they have no conflicts of interest with the contents of this article.      Samples are mounted to a rotating stage between the source and the detector. Magnification is solely dependent on geometric ratio of source-to-sample and sample-to-detector. B. An X-ray microscope is similar in design to a micro-CT but introduces a high-numerical aperture objective which utilizes a scintillator to detect X-rays and emit photons which are detected by the EMCCD. The objective adds a second magnification factor to an XRM system. C-F Examples of volume rendered XRM tomograms of biological samples stained using different contrast agents (each tomogram is digitally cut to illustrate stain penetration). C. Tetranychus evansi, stained using 10% Lugols Iodine. D. Lumbricus terrestris, stained with 0.3% phosphotungstic acid. E. Murine pulmonary vasculature filled with AuroVistÔ contrast agent. F. Metal-enhanced DAB staining of murine Wolffian ducts (white box and inset). Example workflow, showing the generation of tomographic data using an XRM, and using these tomograms to facilitate targeted sectioning and TEM imaging. A sample is first fixed in EM-grade fixatives, and subsequently post-stained with heavy metals. The processed sample is embedded in a resin block and trimmed to facilitate imaging. Trimmed blocks are imaged on an XRM system, generating tomographic data, allowing for identification of a region of interest. Thin sections are thus generated from the targeted region and are imaged on a TEM.