Research Highlights

New Experimental Protocols for Correlative Stochastic Optical Reconstruction Microscopy and EM


Correlative 3D STORM and TEM images of immunolabeled mitochondria in a BS-C-1 cell. (A) Correlative 3D STORM and EM  mages of mitochondria in a BS-C-1 cell innumo-stained for TOM20. Left: STORM image. Right: SEM image. Middle: Overlaid image. Part of SiN mesh is visible on the upper left corner of the image. (B) Magnified views of the boxed regions in A. (C) xy cross-section of the 3D STORM image of mitochondria in the same region. Scale bars, 5 μm in (A) and 500 nm in (B, C).

15-04-2015 La Jolla

Fluorescence light microscopy (LM) and electron microscopy (EM) are among the most widely used imaging modalities to learn about cellular, subcellular, and supramolecular structures. Each modality has strengths and weaknesses that complement the other. Fluorescence LM can allow spatiotemporal localization of labeled biomolecules, such as proteins and nucleic acids, with high molecular specificity and sensitivity. Multi-color imaging using spectrally distinct fluorescent labels (one color per target) allows several molecular targets to be imaged simultaneously to learn about their interactions. Recently, various super-resolution fluorescence imaging techniques have enabled imaging molecular structures in cells at resolutions of ~10s of nanometers in lateral resolution.

Transmission and scanning EM methods (TEM and SEM, respectively) generally provide higher image resolution than light microscopy. EM also enables ultra-high resolution imaging of subcellular compartments. But, while several specific molecular structures can be identified from EM images based on their characteristic shapes, such molecular structures are relatively few compared to the great diversity of macromolecules known to be made and employed in forming complexes, organelles, cells, and tissues.

Several new approaches use genetically encoded tags to provide EM contrast for specific molecules. These approaches provide much higher labeling efficiency. However, multi-color EM imaging of more than one or two molecular targets remains challenging.

Thus, it is desirable to perform correlative fluorescence LM (including super-resolution LM) in combination with EM to enhance researchers’ ability to determine the spatial locations of specific molecules. This strategy allows fluorescence imaging, followed by EM to provide more information about the context of the cellular ultrastructure.

It remains challenging, though, to obtain optimal quality in fluorescence LM and EM images. For example, the light emission from fluorescent proteins and dyes can be compromised by the acidic and oxidizing conditions used in the protocol to embed samples in plastic. Strong fixatives and stains used to distinguish membrane structures in EM can quench the fluorescence signal. When relatively mild fixatives and stains are used, though, they can compromise ultrastructure preservation and membrane contrast in EM images. Image quality is also affected by whether the sample is embedded in acrylic resin or the harder epoxy-based resins more suitable for ultrastructure preservation and high-quality ultrathin sectioning. Correlative super-resolution fluorescence and SEM/TEM studies circumvent these problems by performing fluorescence imaging before staining for EM, but these approaches can only be applied to imaging structures relatively close to the cell/sample surface.

So new methods are needed to complement the above approaches and optimize the power of correlative super-resolution fluorescence and electron microscopy.

In response to this need, a research team from the Howard Hughes Medical Institute, Harvard University, the National Center for Microscopy and Imaging Research, and UC San Diego developed a new method that combines 3D stochastic optical reconstruction microscopy (STORM) with several EM imaging modes, including protocols to image unembedded and embedded/sectioned samples. They applied the resulting protocols to several cellular and viral structures.

For unembedded samples, the best results were obtained by doing the EM-specific sample treatment after STORM imaging so as to separately optimize the conditions for both. However, since multiple sample treatment steps occur between STORM and EM imaging, it was important to minimize the extraction or distortion of the samples between these two imaging steps to produce highly correlated images. If the samples were not fixed sufficiently or if the dehydration procedure was too harsh or too long, the cellular ultrastructure could be degraded between STORM imaging and subsequent processing for EM imaging.

For the embedded samples, because EM-specific fixation and staining have to be applied to the sample prior to the resin-embedding step and STORM imaging is performed after embedding and sectioning, it was important to optimize the fixation/staining conditions and embedding resin materials so that the emission property of the fluorophores used for STORM imaging was not altered substantially, while still maintaining sufficient preservation and contrast of the cellular ultrastructure in the EM image. Strong fixatives/stains can quench emission from fluorophores, but weak fixatives/stains do not preserve ultrastructure or provide sufficient EM contrast. Compared to previous studies, the team used stronger fixatives/stains to preserve cellular ultrastructure and enhance membrane contrast in EM and found that the fluorescence signal was largely preserved.

The choice of embedding resin can also affect fluorescence emission and ultrastructure preservation. Instead of using acrylic resins, the team used epoxy resins because they are better suited to preserving the cellular ultrastructure and serial sectioning. They also found the fluorescence signals of the dye molecules to be better preserved in epoxy. Finally, because resin embedding can prevent the switching agents from reaching the embedded dyes and compromise their photoswitching property, they added a chemical etching step to partially remove the resin material after sectioning and recover the dyes’ photoswitching behavior. Without this step, switching of the dyes was substantially inhibited, degrading the STORM image quality substantially.

The methods for this work included cleaning the substrates; culturing and infecting the cells; fixing, staining, and dehydrating the cells; embedding the cells in the resin; sectioning the samples to obtain ultrathin sections (70 nm thick); and etching; followed by 3D STORM imaging; preparation for EM imaging (which protocol depended on whether TEM, SEM, or back-scattered-electron [BSE]-SEM was to be used); then EM imaging.

Then the images from the two modalities needed to be compared. The brightness and contrast of the EM images were adjusted using Photoshop, and the EM image was roughly rescaled until the scale bar matched the size of the scale bar on the STORM image. Features visualized in both images were used as fiduciary markers. A refined alignment between the two images was performed, based on structures within the cells, by rescaling, translating, and rotating the images using Photoshop and Matlab. Finally, to estimate the accuracy of the correlation, the team calculated the normalized cross-correlation between STORM and EM images.

The first experiment was to apply a simple correlative STORM and TEM imaging method on unembedded samples. This experiment largely followed standard protocols for sample preparations for STORM and TEM. STORM imaging was performed prior to sample preparation for EM imaging. Therefore, as the fluorescence signal acquisition precedes the addition of any EM fixative or stains, the optimal fixative and stain could be used for EM and, hence, neither STORM nor EM images were compromised.

The team chose microtubules in cells to test this imaging approach. Individual microtubule filaments were clearly resolved by STORM. The hollow tubular shape of the stained microtubules was resolved, which showed two peaks separated by 36 nm in agreement with previous reported values. For correlative STORM and TEM, the STORM image could still resolve the hollow tubular shape, with 32 nm width. All microtubules observed in the STORM image were observed in the TEM image, and all the thick filaments, likely representing microtubule structures in the TEM image, were found in the STORM image.

Next, the team imaged mitochondria in cells with correlative STORM and TEM. They were able to observe the expected hollow shape of the mitochondria in the STORM image and individual mitochondria correlated with their counterpart in the TEM image.

The team also wanted to target structure exposed or near the sample surface. That need could be addressed by SEM. As a test of the STORM/SEM correlative imaging approach, the team imaged filamentous influenza viruses budding from infected cells. They first obtained STORM images of the virus filaments that showed that it was possible to resolve the hollow shape of the virus filament. Analysis of the x-y cross-section revealed a filament width of 100 nm in the STORM image, which agrees with the known diameter of a virus filament (80-120 nm). The SEM images correlated well with the STORM images with viral filament width measured at 85 nm.

The team also wanted to probe interior regions of thick biological samples using EM. Typically, such samples are embedded in resin and thin-sectioned using an ultramicrotome. To conduct this experiment, the team developed a correlative STORM and BSE-SEM imaging approach. There are two major differences between this method and the methods describe earlier. The embedding procedure and choice of embedding resin can affect the brightness and photoswitching properties of the fluorophores used for STORM imaging and preservation of cellular ultrastructure. And EM-related fixatives and stains need to be applied to the sample prior to embedding and sectioning and, hence, can also affect the properties of the fluorophores.

To minimize the effects of these factors on STORM and EM image quality, the team tested various types of resins, polymerization strategies, and EM fixatives and stains, and used chemical etching described above.

As the first test of the STORM/BSE-SEM method, the team again imaged filamentous influenza virus budding from cells. The STORM and EM images were well correlated and showed the hollow tubular structures of viral filaments with the expected width.

Then the team performed correlative STORM/BSE-SEM imaging of an intracellular membrane-bound organelle, the mitochondria. The STORM images showed the expected outermembrane staining, and the EM images showed the typical cristae structure of mitochondria. Interestingly, compared to photoactivatable fluorescent proteins, photoswitchable dyes were much more robust when exposed to membrane fixatives/stains, causing less of a compromise between STORM and EM imaging.

Correlative super-resolution fluorescence and electron microscopy is expected to become a valuable tool for ultrastructural studies of many cellular processes. Future work is needed to explore the power of correlative imaging, for example, through development of correlative live super-resolution fluorescence imaging and EM, and through fast fixation or freezing after fluorescence imaging or liquid-chamber EM imaging.

Funding Source: This project is in part supported by the National Institutes of Health GM068518 to X. Z. and P41GM103412 to M.H.E., which funds the National Center for Microscopy and Imaging Research (NCMIR) at UC San Diego. X.Z. is a Howard Hughes Medical Institute investigator.

Relevant Publication: Kim, D., Deerinck, T.J., Sigal, Y.M., Babcock, H.P., Ellisman, M.H., and Zhuang, X. (2015). Correlative Stochastic Optical Reconstruction Microscopy and Electron Microscopy. PLOS ONE, 10(4).