Electron microscopy in combination with immunolabeling allows the localization of macromolecules at high resolution in the structural context of tissue and cells. The label can directly be bound to the antigen-containing structure. If one starts with a new antigen, this approach is definitely not the method one should use first. It can be tedious or even impossible to find the region of interest in the three-dimensional structure within the tissue by electron microscopy.

Light microscopy, on the other hand, offers the advantages of rapid screening a large area of a specimen, of determining the specificity of labeling, and of simultaneously detecting multiple antigens with ease. In bright field, phase or differential interference contrast light microscopy, the majority of cellular organelles cannot be directly identified due to the low-resolution power and to the discernment of the color reaction products of the label. In epifluorescence microscopy, the signal of the label is much higher because of the dark-field imaging mode. Unfortunately, there are no reference structures visible in epifluorescence microscopy, so the correlation between labeled antigen and a counterstain is indispensable [1]—e.g., DAPI to visualize the DNA or additional labeling for assumed compartments. Still, even using confocal microscopy, co-localization cannot be unequivocally proven due to the poor resolution along the z-axis. Furthermore, the point of focus of the different wavelengths, e.g., blue from DAPI and green from FITC, is not at the same z-height because of the different refraction indices.

Thus, ideally one should combine the advantages of both light and electron microscopic imaging using the identical specimen. The ultimate approach would be live-cell observation and selection of the region of interest in the light microscope and subsequent high-resolution imaging of the fluorescent signal of the identical region in frozen-hydrated sections in a cryoelectron microscope [2, 3].

With the development of a cryo-fluorescence microscope [4, 5], genetically labeled cells are imaged by fluorescence and its coordinates fed into a cryo-TEM for cryoelectron tomography [6]. In the meantime there is also an integrated light electron microscope (ILEM) that makes direct correlation in one instrument possible [7–10].

Cryo-applications are still restricted to relatively small objects and therefore more conventional approaches are needed for more routine applications. Electron microscopy preparation protocols have been designed that can preserve genetic labels, like GFP, YFP, RFP, and mCherry. The label distribution is imaged on a section by fluorescence microscopy and afterwards the ultrastructure of the same section can be correlated with the fluorescence image [11–16].

Another approach is the efficient conversion of a fluorescent signal in a living cell into an electron-dense signal for subsequent characterization on electron microscopic level [17–20]. One of the drawbacks of the photo-conversion method is the low contrast and the rather poor spatial resolution of deposited reaction pro-duct that may in addition obscure the structure of interest. Furthermore, although a region of interest is easy to find by light microscopy, the identical region is intricate to spot by electron microscopy. This problem is particularly troublesome if the signal is not both unequivocally identifiable and associated with a well-defined structure.

The preparation schemes of immunolabeling for light microscopy compromise the cellular ultrastructure, cause dislocation or even loss of components such as lipids and nonstructural proteins, and limit the admission of the detecting antibodies through steric hindrance [21–24]. For pre-embedding labeling, the samples have to be opened to provide access to the intracellular antigens either mechanically by freeze-thaw methods or chemically with solvents or detergents.

In the past four decades, we have learned a lot about preservation of the cellular architecture for high-resolution electron microscopy, which raises the possibility of applying this know-ledge to light microscopy. The aim of this report is to show the great potential of correlating light and electron microscopy data, which has developed from improved specimen preparation techniques in combination with post-embedding labeling for both immunofluorescence and immunogold (see Note 1).

The key observation in post-embedding labeling has been the fact that antibody binding is mainly restricted to the surface of a section of properly fixed material [25, 26]. Hence, there is no need to cut thick or semi-thin sections for on-section immunolabeling, as the signal would be similar to that of ultrathin sections (see Note 2). Particularly with resin-embedded material, the potential of ultrathin sectioning can be exploited to provide adjacent serial sections for light and electron microscopy [27–31]. The use of the identical specimen preparation via thin sections for both light and electron microscopy offers the possibility to examine the same sample at different orders of magnitude. Thus, identical sections of a specimen can be used to give an overview of labeling distribution in a total organ by fluorescence microscopy and to define precisely the fine details at the level of cellular compartments by electron microscopy. In principal, this approach is suited to Tokuyasu cryosections [32–37] and all methacrylate and epoxy resins, as long as antigenicity is retained.

The main advantages of Tokuyasu cryosections are twofold: the samples are not dehydrated by solvents—hence, the proteins remain in their natural aqueous environment until immunolabeling—and due to the open surface structure, more antigens are accessible than in resin sections (see Note 3). Moreover, the preparation procedure is fast and the results are obtainable within 1 day. The Tokuyasu cryosectioning technique can also be combined with cryofixation using high-pressure freezing [38–40]. Though the procedure is a bit more time consuming, its benefit is to better preserve membrane structures and, more importantly, to make difficult to fix biological samples like plants and insects accessible for the Tokuyasu technique [40]. With the current improvements of the technology and the instrumentations, Tokuyasu cryosections can as easily be cut as resin sections, e.g., sections parallel to the substrate [25, 41] or serial sections.

Embedding in resins such as cross-linked methacrylates (Lowicryl, London Resins, Unicryl) [42–46] or epoxies (Epon, Araldite, Spurr’s) [47–49] allow wet sectioning to produce large (millimeter sized) and ultrathin sections (>50 nm thickness) with excellent structural preservation. Furthermore, established histo-logical staining procedures, e.g., toluidine blue staining [50], can be applied and are advantageous to differentiate between the structures within the specimen. Last, but not least, large collections of resin-embedded specimens can easily be stored at room temperature, requiring only a small space. Somehow surprisingly, most epitopes keep the antigenicity, even after sectioning, for a very long time, in the range of years. Hence, libraries of samples can be established and serve as references for new mutants or constructs but also to be reinvestigated for new proteins and antibodies of interest. There are reports that antigen-retrieval protocols may increase label efficiency in “old” standard preparations [51–53].

Epoxy monomers cross-link with the cellular components during polymerization; hence, they form a copolymer of resin and molecules in the biological specimen [54]. Thus, biological material embedded in epoxy resins is easy to section. While this method provides excellent morphology, antigenicity is often affected. In contrast, methacrylates cures by a radical chain reaction, allowing only the monomers to polymerize [54, 55] so that the biological components are enmeshed in a network of resin polymers. Hence, the more open structure and its higher degree of hydrophilicity than Epon provides better labeling properties.

Resin embedding can be combined with a large variety of fixation and dehydration procedures. Fixation can be done chemically at ambient temperature or by cryofixation. Dehydration can be performed through a graded series of solvents at ambient temperature, by progressive lowering of temperature (PLT) [43] or by freeze-substitution [56].

The gold standard of preparation methods is cryofixation followed by freeze-substitution and resin embedding [56–59]. The cryofixation method of choice is high-pressure freezing [60–63]. To quote Dr. Martin Müller [64], “Even the best microscopes and the most sophisticated image analysis algorithms are unable to provide high density information from samples distorted during the sampling step (i.e., specimen preparation).” Therefore, improved specimen preparation using cryotechniques in combination with post-embedding labeling for both immunofluorescence and immunogold is a promising approach to bridge the gap between light and electron microscopy [28].

An intrinsic property of high-resolution objectives with high numerical apertures of light microscopes is the small depth of focus (see Note 4). This results in out-of-focus blurring of the image and loss of resolution in both bright field and phase contrast illumination settings as well as in fluorescence light microscopy. This is because biological objects, even cell monolayers, are much thicker. Therefore, only thin sections not thicker than the depth of focus of objectives with a high numerical aperture (0.5–1 μm) can exploit the much higher resolution. A further aspect is the z-resolution of a light microscope. With confocal, also 2-photon, microscopy at optimum conditions, the z-resolution is in the range of 400–500 nm. Even super-resolution microscopy has difficulties with the z-resolution. The maximum x, y-resolution of about 10 nm can only be achieved on ultrathin sections of 50–80 nm [65]. This observation strengthens our arguments to use ultrathin sections for fluorescence microscopy. For single images the z-resolution is not lower than the section thickness, e.g., array tomography, or even higher on resin sections where only a few nm are accessible for the antibodies. Also super-resolution microscopy cannot overcome the precision of the antibody detection of an antigen that is in the range of 5–30 nm as in electron microscopy.

Thus, for correlative microscopy, first, the region of interest is approached with the help of toluidine blue stained thin sections. Second, the signal is localized on ultrathin sections (50–100 nm) with fluorescence microscopy and then at high resolution with electron microscopy.

The widespread and successful application of confocal light microscopy has pushed tremendous development of a great variety of stable fluorochromes and detection systems, e.g., more precise optical filters (see Note 5). Due to the small size of the fluorochromes (a few hundred Daltons), they influence the binding properties of antibodies less than other marker systems [23]. Generally, if labeling with fluorescence markers fails, there will be no chance to detect the antigen with gold probes. Hermann et al. [66] have demonstrated the dramatic influence of even a very small gold particle (1 nm). They showed that Fab fragments directly coupled to ultrasmall gold particles detect only one copy of a homotrimer of proteins at distinct sites of the tail fibers of a T-even phage, whereas uncoupled Fab fragments detect two copies. Nevertheless, the smaller the size of the gold particle, the higher is the density of gold label [23, 26, 67]. This observation suggests that not all the primary antibodies are detected by the gold label. The larger the gold particle the stronger the electrical repulsion. Therefore a large number of primary antibodies can still be labeled with a smaller, uncharged detection system, i.e., fluorochrome markers after the gold labeling [31].

To permit optimum correlation of the results obtained by on-section labeling, the labeling conditions for immunofluorescence and immunogold have to be kept identical. The potency of this approach (Figs. 1, 2, 3, 4, and 5) has been demonstrated in a series of publications [28, 31, 68–74].

Based on this concept Micheva and Smith [75] developed the array tomography. They collect ribbons of serial sections on microscopy slides and perform immunolabeling of an antigen that is detected by a fluorescence marker. After imaging all sections in a fluorescence microscope, the antibodies can be removed and a second fluorescence labeling can be performed. This process can be repeated for several times resulting in an array of images of different antigens. Finally the ultrastructure is imaged in a scanning electron microscope using backscattered electrons. The result is a three-dimensional map of fluorescently labeled proteins that can be correlated to the cellular ultrastructure [53, 75–77].

Due to the fact that after the detection with gold conjugates still a large number of primary antibodies are accessible for the smaller fluorochrome-labeled antibodies, double labeling could be done in an inverted sequence. Sections are mounted on indium tin oxide (ITO)-coated glass slides or coverslips [78]. The antigen is labeled with a primary antibody, then first detected with a gold detection system, followed by a fluorescence-tagged antibody directed against the same primary antibody. After imaging the label in a fluorescence microscope, the slides can directly be transferred to a SEM to image the gold label of the same sections (Figs. 6, 7, and 8).

Our goal is to use the labels at the level of resolution the physics of the instruments allow. With the fluorescence label the cell or organelle of interest is identified in the tissue. Then the tissue block is trimmed such that the area becomes accessible for sectioning [79–82]. On the section further proteins can be labeled with colloidal gold particles for electron microscopy to study the high-resolution distribution in the tissue and the cell of interest in particular.