Proteins play essential roles in cellular functions; to fully understand their roles, we must know where they are localized relative to subcellular structures. Previously, there have been two methods for localizing proteins to subcellular structures: immuno-electron microscopy and fluorescence microscopy. Immuno-electron microscopy localizes proteins at an ultrastructural level using gold-labeled antibodies [1–5]. However, it has many drawbacks, including a reliance on antibodies that can work on plastic sections, poor survival of epitopes, poor preservation of morphology, high nonspecific background, and a spatial displacement of up to 38 nm (the molecular length of two antibodies) [1, 2, 6]. Moreover, only a minority of the proteins are labeled due to the inaccessibility of antigens in the plastic.

Fluorescence microscopy, on the other hand, has become the method of choice for protein localization. In particular, genetic tagging of proteins with green fluorescent protein (GFP) [7–9] is very useful because every protein is labeled and can be imaged in living cells. However, two drawbacks limit its ability to locate proteins in a cell: resolution and cellular context. The light from a fluorescent protein can be only focused to a 450 nm spot on a camera. If there were only two proteins in an image, they could theoretically be distinguished if they were separated by greater than 250 nm, but typically many proteins are labeled and the image is really no more than an incoherent micron-sized ghost at high magnification. Thus, the diffraction limit means that proteins, typically on the order of 4 nm in diameter (the diameter of GFP), cannot be associated with cellular organelles, which are on the order of 40 nm in diameter (such as synaptic vesicles or Golgi cisternae). In the last few years, several super-resolution techniques have been developed that can resolve fluorescence at resolutions below the diffraction limit [10–16]. Single-molecule-based techniques include photoactivated localization microscopy (PALM) [10], fluorescence photoactivation localization microscopy (fPALM) [11], stochastic optical reconstruction microscopy (STORM) [13], direct stochastic optical reconstruction microscopy (dSTORM) [14], and ground-state depletion microscopy followed by individual molecule return (GSDIM) [12]. Despite the variations in acronyms, these techniques overcome the diffraction barrier by imaging single molecules one at a time [17]. The location of individual molecules is then calculated and mapped to reconstruct an image. In this chapter, we will use the terms PALM, for imaging individual fluorescent proteins, and dSTORM, for imaging organic fluorophores using ground-state depletion (see Note 1). These imaging techniques break the diffraction limit of conventional fluorescence microscopy and can localize a protein with a precision of 10 nm standard error of the mean (SEM) [18, 19] (see Note 2).

The second drawback with fluorescence microscopy is that it lacks cellular context. A precisely localized pinpoint of fluorescence in a black background is not useful. Combining these new subdiffraction fluorescence methods with electron microscopy merges the protein localization advantages of fluorescence microscopy with the structural definition of electron microscopy [20].

To localize GFP in ultrathin sections, one must maintain GFP fluorescence during the harsh osmium fixation steps, the dehydration steps, and the incubation and polymerization in acidic resins. Recently, we developed a method to preserve fluorescence by reducing fixative concentrations, maintaining hydration, and compromising on ideal polymerization conditions [20]. In short, transgenic worms expressing proteins tagged with fluorescent proteins are subject to high-pressure freezing and freeze-substitution. These animals are embedded into glycol methacrylate resin at −20 °C. Ultrathin sections are cut using a microtome and collected onto pre-cleaned cover glasses. These sections are first imaged on a super-resolution light microscope and then post-stained with uranyl acetate and imaged using backscattered electrons on a scanning electron microscope (SEM). Using this approach, we successfully localized histones, mitochondrial membrane proteins, and liprin-alpha to their proper organelles in a cell [20]. However, several improvements can be implemented to increase the sensitivity and the utility of our methods.

There are five problems associated with the original protocol we published. First, ~40 % of fluorescent proteins were lost during the fixation. Second, photoactivatable fluorescent proteins suitable for PALM imaging are not ideal. EosFP and its variants are the most reliable fluorophores for PALM. However, they emit green fluorescence before photo-conversion and red fluorescence after photo-conversion, covering two large spectra of visible light. Thus, multicolor imaging is extremely difficult. Third, the number of photons collected from each EosFP molecule before bleaching is very low (~300), reducing the calculated precision of the protein location by the computer. Fourth, the protocol was originally developed using the nematode Caenorhabditis elegans, optimal conditions for other biological samples such as cultured cells were not known. Fifth, despite the well-preserved morphology, the resolution of images generated by collecting backscattered electrons in a scanning electron microscope is limited by the size of the electron beam scanning the sample and leads to blurry images at high magnification (see Fig. 1a, c).

Therefore, since the original publication, we have improved our methods to solve these five problems. First, we fine-tuned our protocol to preserve up to 90 % of the fluorescence without compromising the fixation quality [21]. Second, to overcome the paucity of suitable fluorescent proteins with PALM imaging and demonstrate the potential for the method to localize multiple proteins at a time, we imaged the fluorescent protein, citrine, that is not photoactivatable or photoconvertible using a ground-state depletion scheme [21]. Third, we labeled with an organic fluorophore, Alexa-647, capable of generating many more photons and hence higher precision, and imaged using a dSTORM scheme although this approach is not quantitative (see Note 3 and Fig. 2a, b). Fourth, we demonstrated that this method could be readily applied to specimens other than C. elegans using the protocol (see Fig. 2). Fifth, we developed a method to image using a transmission electron microscopy to provide higher-resolution details of ultrastructure (see Figs. 1 and 2). Here, we describe the improved protocol in detail.