ABSTRACT
Here, we combine super-resolution fluorescence localization microscopy with scanning electron microscopy to map the position of proteins of nuclear pore complexes in isolated Xenopus laevis oocyte nuclear envelopes with molecular resolution in both imaging modes. We use the periodic molecular structure of the nuclear pore complex to superimpose direct stochastic optical reconstruction microscopy images with a precision of <20 nm on electron micrographs. The correlative images demonstrate quantitative molecular labeling and localization of nuclear pore complex proteins by standard immunocytochemistry with primary and secondary antibodies and reveal that the nuclear pore complex is composed of eight gp210 (also known as NUP210) protein homodimers. In addition, we find subpopulations of nuclear pore complexes with ninefold symmetry, which are found occasionally among the more typical eightfold symmetrical structures.
INTRODUCTION
Electron microscopy (EM) and electron tomography have been extensively used to elucidate cellular ultrastructure and architecture of large multiprotein complexes. An impressive example of the resolution power of these methods is the determination of the three-dimensional (3D) structure of the nuclear pore complex (NPC), a gigantic molecular machine assembled from ∼30 different nuclear proteins with a molecular mass of over 100 MDa, with a resolution of a few nanometers (Beck et al., 2007; Maimon et al., 2012). However, conventional fixation and staining protocols for EM and electron tomography are optimized to provide optimal preservation of cellular ultrastructure and not to provide specific molecular labeling. Immunolabeling EM with gold-conjugated antibodies can reveal localizations of cellular components with nanometer resolution but is limited by the quality of fixation and the inaccessibility of antigens, and it can obscure structural details beneath it. Even under optimized conditions (i.e. with the availability of high-affinity antibodies and mild fixation and staining protocols) only a subset of target molecules is likely to be successfully detected (Morphew, 2007). Hence, the determination of the molecular composition of large biological complexes, such as the NPC, remains challenging.
Correlative light and electron microscopy (Müller-Reichert et al., 2007; Kukulski et al., 2011) can be used advantageously for imaging of proteins or organelles within cells because fluorescence microscopy allows specific molecular labeling and, in combination with new super-resolution fluorescence microscopy methods, spatial resolutions well below the diffraction barrier. Super-resolution fluorescence imaging by single-molecule photoactivation or photoswitching and position determination (localization microscopy) (Betzig et al., 2006; Hess et al., 2006; Rust et al., 2006; Heilemann et al., 2008, Sauer, 2013) can localize fluorescently labeled molecules at virtually molecular resolution. Furthermore, super-resolution fluorescence imaging enables higher labeling efficiencies than immunogold EM using fluorophore-tagged antibodies, which facilitates structure determination by localization microscopy. Thus, localization microscopy and EM are complementary methods that can be combined to determine molecular positions in the context of the cellular ultrastructure provided by EM with nanometer resolution (Betzig et al., 2006; Watanabe et al., 2011; Kopek et al., 2012; Nanguneri et al., 2012; Suleiman et al., 2013; Sochacki et al., 2014; Perkovic et al., 2014). In order to perform correlative localization and electron microscopy, several methodological requirements have to be fulfilled. First, protein fluorescence damage and autofluorescent background due to fixation and staining required to preserve good ultrastructure has to be kept to a minimum (Tsien, 1998; Watanabe et al., 2011). Second, protein positions determined by localization microscopy have to be located within the EM image across large areas with a precision in the nanometer range. This demand is impeded by a lack of suitable alignment markers that are stationary at the nanoscale range and exhibit good contrast in the two imaging modes (Watanabe et al., 2011; Kopek et al., 2012; Sochacki et al., 2014; Perkovic et al., 2014). Even when such types of markers are available, structural deformations occurring between the two imaging modes can aggravate the overlay of the two images with molecular precision.
Finally, one should bear in mind that localization microscopy can provide quantitative molecular information about molecular distributions and, appropriate controls assumed, absolute numbers of proteins present. Therefore, it has to be guaranteed that each protein of interest is labeled with an ‘active’ fluorophore and that its fluorescence is detected above a certain photon threshold (Sauer, 2013; Lando et al., 2012). This task is particularly demanding when using fluorophore-labeled antibodies for specific labeling.
Here, we demonstrate the use of correlative scanning electron (SEM) and direct stochastic optical reconstruction microscopy (dSTORM) (Heilemann et al., 2008; van de Linde et al., 2011; Löschberger et al., 2012) to image proteins of NPCs in isolated Xenopus laevis oocyte nuclear envelopes with molecular resolution in both imaging modes. Instead of using alignment markers, we use the periodic molecular structure of the NPC to localize proteins in dSTORM and corresponding SEM images to overlay them with nanometer precision. Our results demonstrate quantitative molecular labeling of the NPC protein gp210 (also known as NUP210) (Favreau et al., 2001; Gerace et al., 1982; Gajewski et al., 1996) by standard immunocytochemistry with primary and secondary antibodies, and that the NPC contains eight gp210 homodimers. In addition, dSTORM reveals subpopulations of NPCs with ninefold symmetry that are found occasionally among the more typical eightfold symmetrical structures (Hinshaw and Milligan, 2003).
RESULTS AND DISCUSSION
Recently, the NPC has been introduced as a suitable reference structure for verification of super-resolution imaging data. Two-color dSTORM images recorded either consecutively or simultaneously by spectrally resolved imaging in combination with particle averaging demonstrated that the organization of NPC proteins can be visualized with nanometer resolution (Löschberger et al., 2012, Wolter et al., 2012). In order to shed light on the heterogeneity of NPC composition and quantify the labeling efficiency, we set out to localize the integral membrane protein gp210 (Favreau et al., 2001; Gerace et al., 1982) and N-acetylglucosamine-modified nucleoporins of the central channel of NPCs by correlative SEM and dSTORM. Isolated Xenopus laevis oocyte nuclear envelopes were prepared on chambered cover glass without correlative markers (Fig. 1). The nuclear envelopes were labeled using a primary antibody directed against an epitope in the luminal side of gp210 (Gajewski et al., 1996), and Alexa Fluor 647 (Alexa 647)-labeled F(ab′)2 fragments or Alexa-647-labeled wheat germ agglutinin (WGA) binding to modified nucleoporins of the central channel (Davis and Blobel, 1987).
At this point, it has to be indicated that correlative imaging of NPCs in isolated nuclear envelopes with molecular resolution is challenging because sample preparation generates a situation in which the cytoplasmatic NPC side, with the anchoring proteins, is oriented in the direction of the coverglass, but SEM imaging is restricted to the nucleoplasmatic, nuclear-basket-containing side (Fig. 1A). Accidentally, we realized sample areas where the nuclear envelope folded back during preparation, hence, also enabling SEM of the cytoplasmatic NPC side (Fig. 1B). However, it is impossible to identify such folded areas in fluorescence images. Therefore, in the majority of preparations the cytoplasmatic side of the NPC is invisible in the SEM images. By contrast, the central channel and the integral membrane protein gp210 can be visualized from the cytoplasmatic and nucleoplasmatic side by fluorescence imaging.
In localization microscopy, single-molecule coordinates are generally visualized as a histogram by using sub-pixel binning (Wolter at al., 2012; Wolter et al., 2010). To determine the influence of pixel binning on resolution, we reconstructed dSTORM images of the central NPC channel at different pixel size (Fig. 1C). The resulting super-resolution images clearly demonstrate that the central channel of the NPC with a diameter of ∼40 nm did not become visible until the pixel size was reduced below 20 nm. This finding demonstrates that we achieved a spatial resolution considerably better than 20 nm for imaging of the central channel in NPCs (Fig. 1C). The high spatial resolution is due to the large number of photons detected per Alexa 647, ∼3.500 photons/frame, and the small size of a WGA dimer, of ∼5 nm (Schwefel et al., 2010), used to label the nucleoporins of the central channel. In accordance with this finding, we determined a localization precision of ∼6 nm (s.d., in the lateral direction) using localizations of unspecifically bound isolated WGA–Alexa-647 molecules in the sample.
Next, we mapped the xy positions of gp210 labeled with Alexa 647 antibody and WGA–Alexa-647-labeled nucleoporins onto the SEM image (Fig. 2). Image alignment was performed without markers, exploiting the highly symmetric circular structure of the NPC (supplementary material Figs S1, S2). The xy positions of Alexa 647 in the dSTORM images matched the shapes of the NPC structures in SEM (Fig. 2). The majority of WGA–Alexa-647-labeled NPCs showed the central channel perfectly superimposed with the SEM structures (Fig. 2A–D). Even in large correlative dSTORM-SEM images, the overlay error remained smaller than the spatial optical resolution (supplementary material Fig. S3).
By contrast, gp210 proteins were labeled by an antibody complex, which limited the overlay accuracy to ∼20 nm. In addition, the dSTORM structures were generally larger than the corresponding SEM structures. However, because the overlay of the two images was not based on direct protein assignment but on the center of the NPC structures in dSTORM and SEM images, we also expect a smaller overlay error for the gp210 structure. Whether the simple overlay of the center of mass of localization microscopy and EM data also works for unsymmetrical structures needs to be tested.
The majority of gp210-labeled NPCs showed an eightfold symmetric arrangement surrounding the central channel (Fig. 2E–H). Some of the gp210 eightfold structures appeared slightly distorted, which arises from the spreading of the nuclear envelope on the coverglass. Only very rarely (<1%) could completely unlabeled NPCs be identified in dSTORM-SEM images (Fig. 2D). The same applies to the central channel labeled by WGA–Alexa-647. The infrequent appearance of completely unlabeled NPCs cannot be explained by labeling statistics but must be ascribed to incomplete compositions of NPCs or incomplete maturated or erroneously folded proteins.
Interestingly, we also discovered some (<0.1%) NPCs featuring a ninefold rotational symmetry of gp210 proteins (Fig. 2I,J). Subpopulations of ninefold and tenfold symmetrical NPCs with slightly larger diameter have previously been found in preparations of Xenopus oocyte nuclei by EM, suggesting that the assembly process of NPCs can be influenced by local nuclear envelope inhomogeneity, e.g. curved membranes that favor the formation of larger NPCs due to packing constraints (Hinshaw and Milligan, 2003).
Taking advantage of the specific and efficient labeling properties of the anti-gp210 antibodies used (Gajewski et al., 1996) and the high detection and localization probability of Alexa 647, we set out to quantify the number of gp210 proteins surrounding the central channel of the NPC. So far, it is unclear whether eight or 16 gp210 proteins are arranged as monomers or homodimers around the central channel (Favreau et al., 2001; Gerace et al., 1982). Analysis of localization data of 318 NPCs revealed values of 274±4 (mean±s.e.m.) localizations per NPC (median, 253) and 35.1±0.5 (mean±s.e.m.) localizations per protein domain (median, 32) forming the eightfold NPC structure (Fig. 3A,B). This demonstrates that on average 7.9 domains are labeled and identified per NPC. For isolated fluorescence signals detected within the circular NPC structure, we determined a value of 8.1±0.2 (mean±s.e.m.) localizations per Alexa-647-labeled F(ab′)2 fragment (median, 5.5) (Fig. 3C).
With the assumption that the isolated signals resulted from single Alexa-647-labeled F(ab′)2 fragments, we estimate that on average more than four Alexa-647-labeled F(ab′)2 fragments bind per gp210 domain. Given that the probability for binding of more than 4 F(ab′)2 fragments per primary antibody is negligibly low and it is known that the primary antibody X222 binds to a single epitope of the gp210 protein (Gajewski et al., 1996), our data provide evidence that each domain contains a gp210 homodimer labeled on average with two monoclonal X222 antibodies and four to six Alexa-647-labeled F(ab′)2 fragments. Control localization experiments with non-specifically adsorbed Alexa-647-labeled F(ab′)2 fragments on nuclear envelopes in the absence of the primary antibody directed against gp210 protein corroborated our finding that more than four Alexa-647-labeled F(ab′)2 fragments bound per gp210 domain (supplementary material Fig. S4).
To summarize, our data convincingly demonstrate that localization microscopy with standard fluorescent probes, such as fluorescently tagged antibodies, can map the position of proteins in the context of the ultrastructure of large multiprotein complexes provided by EM with nanometer resolution and can reveal subpopulations with slightly altered characteristics that are washed out by particle averaging methods (Löschberger et al., 2012; Szymborska et al., 2013). In addition, correlative dSTORM-SEM showed that NPCs were only rarely invisible in the fluorescence imaging mode but appear perfectly assembled in the EM imaging mode. These results emphasize the complementarity of the two imaging modes and also provide evidence for the reliability of dSTORM when applied in combination with highly specific antibodies for quantitative super-resolution imaging. In combination with correlative focused ion beam scanning electron microscopy (FIB-SEM), dSTORM can potentially be used also for 3D quantitative super-resolution imaging. Alternatively, the sample can be imaged with two-dimensional tiling and at multiple angles to create 3D tomograms in transmission electron microscopy (TEM) (Sochacki et al., 2014)
MATERIALS AND METHODS
Sample preparation for dSTORM
Nuclear envelopes of Xenopus laevis oocytes were isolated as previously described (Löschberger et al., 2012). They were fixed for 20 min with 2% paraformaldehyde in phosphate-buffered saline (PBS). After a short washing step in PBS, they were saturated with 0.5% BSA (Serva) in PBS for 5 min. The envelopes were labeled with X222 antibodies (Gajewski et al., 1996) directed against gp210 for 45 min, washed for 10 min in PBS and incubated with Alexa-647-labeled F(ab′)2 fragments [Life Technologies, A-21237; degree of labeling (DOL) of ∼3.5] of goat anti-mouse-IgG antibody for 30 min. Finally, a washing step of at least 20 min in PBS was performed. For WGA staining, the nuclear envelopes were incubated in 1 µg/ml WGA–Alexa-647 (Sigma) in PBS for 10 min. Samples were stored in PBS with 0.2% sodium azide.
Localization microscopy
dSTORM imaging was performed as described previously (van de Linde et al., 2011; Löschberger et al., 2012). Briefly, we used an inverted microscope (Olympus IX-71) equipped with an oil-immersion objective (APO N, ×60, NA 1.49; Olympus). Irradiation at 640 nm was provided by a Genesis MX 639-1000 laser (Coherent). Typically, 20,000 frames were acquired with a frame rate of 105 Hz at excitation intensities of 1–10 kW cm−2 using an inclined illumination scheme (van de Linde et al., 2011). dSTORM images were reconstructed and analyzed with the open source software rapidSTORM 3.2 (Wolter et al., 2010; Wolter et al., 2012). Only fluorescent spots containing more than 1000 photons were analyzed. By analyzing their ellipticity, multi-fluorophore events were discarded from further analysis (Wolter et al., 2010; Wolter et al., 2012). Typically, 3500 photons were detected per Alexa 647 molecule and frame. Localizations of 845 unspecifically bound, isolated WGA–Alexa-647 molecules were aligned to their center of mass and binned into one histogram. The localization precision (s.d.) was determined as 6.41±0.03 nm (s.e.m. of data fit) by fitting a two-dimensional Gaussian function to the data.
Scanning electron microscopy
After finishing the dSTORM analysis, nuclear envelopes were fixed for 24 h at 4°C with 2.5% glutaraldehyde (50 mM sodium cacodylate pH 7.2, 50 mM KCl, 2.5 mM MgCl2) and washed three times each for 3 min with 50 mM sodium cacodylate (pH 7.2). Specimens were then fixed for 2 h at 4°C with 2% OsO4 buffered with 50 mM sodium cacodylate (pH 7.2), washed with distilled H2O, dehydrated stepwise with acetone and dried using a critical point dryer (CPD 030; BAL-TEC, Liechtenstein). Dried specimens were carbon coated (MED 010 BAL-TEC, Liechtenstein) and analyzed with a JEOL field emission scanning electron microscope (JSM-7500F) at 1 kV using the modus gentle beam high at a working distance of 4.5–5 mm. Overview images were used to recognize the previously super-resolved areas (supplementary material Fig. S3).
Image alignment
Thanks to their highly symmetrical structure, NPCs can be used for intrinsic image alignment. The center of the gp210 or WGA ring structure in dSTORM and SEM images that matched were identified manually (supplementary material Fig. S4) and used as reference points (landmarks). bUnwarpJ was used for image registration (Arganda-Carreras et al., 2006). Typically, six landmarks in different parts of the dSTORM and SEM images were used for transformation of a 2.5×2.5 µm2 image area. Briefly, bUnwarpJ calculates a B-spline transformation for registering two images. GIMP 2.8.10 was used to make the dSTORM images transparent before correlative overlay with the SEM images.
Quantification of dSTORM data
First, single NPCs were separated by applying a canny edge filter to the dSTORM image. The vertices of the resulting mask were used to crop localization sets of single NPCs from the data provided by rapidSTORM (Wolter et al., 2010; Wolter et al., 2012). To allocate the whole NPC into single gp210 domains, a multilevel k-means algorithm was applied to the localization cloud. Using the known spatial features of the gp210 ring (i.e. the existence of a maximum of eight separated domains and their circular symmetry), starting points for the k-means clustering can be provided, minimizing the usual difficulties with k-means clustering. First, a mean component analysis (MCA) was performed on the x and y coordinates of all localizations of the single NPC candidate. NPCs exhibiting a ratio of the absolute values of the resulting eigenvectors larger than two were discarded to exclude overlapping pores and other artifacts from the canny edge filter. Afterwards an ellipse was fitted to the localization ring, again using the eigenvectors resulting from the MCA. Starting points were then set by distributing the intended number of domains, N, randomly but (angularly) equidistant along the ellipse. It appeared that lower starting values than five domains were not necessary. A second set of starting points was generated by turning the first set by N−1×α, where α is the angle between two neighboring domains. This procedure is repeated for N×[5,6,7,8]. Subsequently each candidate domain in the current subset was subjected to an Anderson–Darling test (Hamerly and Elkan, 2004) to evaluate whether the current separation was sufficient or not. Briefly, every candidate domain was tested for whether their spatial localization distribution in x and y, respectively, is Gaussian like. If this is not the case for at least one candidate, the current separation state is rejected, and the next parameter set is tried. If there is no successful test for the last parameter set, which is usually the ‘eight turned starting points’ configuration, the candidate is discarded. This might be the case for candidates that were cropped faulty from the canny edge detector or had insufficient spatial resolution, causing the domains to overlap significantly. Both cases were very rarely apparent. All further analyzing steps following the separation process, for example, production of histograms and localization counts, were conducted directly by additional PYTHON scripts or in OriginPro 8.5 (OriginLab).
Acknowledgements
We thank Christian Stigloher and Thomas Niehörster for help with sample preparation and discussion.
Author contributions
A.L., G.K., S.v.d.L., C.F. and M.S. conceived of and designed the experiments; A.L. and G.K. performed the experiments; C.F. and S.v.d.L. analyzed the data; M.S wrote the paper.
Funding
This work was supported by the Biophotonics Initiative of the Bundesministerium für Bildung und Forschung [grant numbers 13N11019 and 13N12781].
References
Competing interests
The authors declare no competing interests.