Alternatively N-azidoacetylmannosamine (Ac₄ManNAz) and N-azidoacetylglucosamine (Ac₄GlcNAz) can be used

For quantitative analysis of generated localization data based on XY coordinates lists, customized algorithms implemented with programing languages such as Python (available at http://www.python.org), and Mathematica (Wolfram Research Inc., Champaing, Il, USA) were used.

Since the development of the Staudinger-Bertozzi ligation between azides and phosphines in 2000 (Saxon and Bertozzi, 2000), bioorthogonal “click chemistry” reactions allowed the visualization of different biomolecules (e.g., proteins, glycans, lipids, and nucleic acids) in cultured cells, tissues, and living organisms (Sletten and Bertozzi, 2009). To this aim, one functional group (the label) is introduced into the biomolecule of interest followed by exogenous addition of fluorophores bearing the reactive partner (the probe). For example, unnatural amino acids and monosaccharides containing an azide group can be used as metabolic surrogates of their native counterparts to visualize proteins and glycoproteins as well as glycolipids (Laughlin and Bertozzi, 2009a; Tom Dieck et al., 2012).

Two different approaches have been used successfully to introduce amino acid analogs into proteins: (i) genetic encoding, i.e., site-specific modification, and (ii) metabolic labeling, i.e., residue-specific modification. Whereas, the first method introduces unnatural amino acids into one particular protein, the second method allows labeling of a wide part of the proteome replacing a native amino acid (e.g., methionine) by its non-natural analog (e.g., L-azidohomoalanine, L-AHA). Due to its structural similarity, L-AHA is recognized and tolerated by the methionyl-tRNA synthetase (MetRS), and incorporated into newly synthesized proteins co-translationally in a residue-specific manner. Alternatively, azido sugars (e.g., peracetylated N-azidoacetylgalactosamine Ac₄GalNAz, N-azidoacetylmanosamine Ac₄ManNAz, and N-azidoacetylglucosamine Ac₄GlcNAz), can be incorporated into different types of glycoproteins and glycolipids (Laughlin et al., 2006; Laughlin and Bertozzi, 2009a). Upon cellular uptake and deacetylation, Ac₄GalNAz, Ac₄ManNAz, and Ac₄GlcNAz are converted into activated sugars, recognized by the glycan biosynthetic machinery, and incorporated into sialic acids and mucin-type O-linked glycans, as well as into O-GlcNAc-modified proteins. After metabolic incorporation of amino acids and monosaccharide surrogates, the azide groups introduced into newly synthesized proteins and glycans can be conjugated with alkyne fluorophores via azide-alkyne cycloaddition allowing their direct visualization.

Originally, the classic reaction between terminal alkynes and azides was shown to be efficiently catalyzed by copper(I) at room temperature enabling it to proceed within minutes under physiological conditions, opening the door for biological applications (Rostovtsev et al., 2002; Tornøe et al., 2002). Since then, this reaction, now termed as the Cu(I)-catalyzed azide-alkyne cycloaddition (CuAAC), has been used to visualize different metabolically labeled biomolecules (Sletten and Bertozzi, 2009). However, due to Cu(I) toxicity fluorescent staining by CuAAC has been restricted to fixed cells. To overcome this problem, two alternative strategies have been developed. In 2004, it was shown that azide-alkyne cycloaddition can be strain-promoted in the absence of copper(I) using cyclooctynes (Agard et al., 2004). Since then, different cyclooctyne molecules with enhanced efficiency have been developed for copper-free strain-promoted azide-alkyne cycloaddition (SPAAC) (Jewett and Bertozzi, 2010; Debets et al., 2011). On the other hand, the optimization of the CuAAC, by means of copper(I) ligands and further additives in the reaction buffer, preserves cell viability while live staining. For example, the use of THPTA in addition to sodium ascorbate allow efficient CuAAC bioconjugation within 5 min with low copper concentrations (e.g., 50 μM) minimizing Cu(I) toxic effects (Hong et al., 2009, 2010).

Standard fluorescence microscopy, combined with metabolic labeling and click chemistry, has been used extensively to visualize both proteins and membrane-associated glycoconjugates within different cellular contexts. For example, newly synthesized proteins have been imaged in mammalian cells and rat hippocampal neurons (Dieterich et al., 2006, 2010; Beatty and Tirrell, 2008), and different glycan populations in culture cells (Baskin et al., 2007), developing zebrafish embryos (Laughlin et al., 2008), and living C. elegans (Laughlin and Bertozzi, 2009b). Remarkably, these studies demonstrated the versatility of metabolic labeling for temporal profiling of dynamic changes in large protein populations and glycans. More recently, the same chemical reporter strategy has allowed direct visualization of different membrane components by super-resolution microscopy (Letschert et al., 2014; Saka et al., 2014). Stimulated emission depletion (STED) was used to image unnatural amino acids incorporated into membrane proteins in monkey kidney cell line COS-7, demonstrating protein confinement with reduced diffusion dynamics (Saka et al., 2014). On the other hand, dSTORM was used to visualize different glycan types, including glycoproteins, after metabolic labeling with Ac₄GalNAz, Ac₄ManNAz, and Ac₄GlcNAz in human osteosarcoma (U2OS) and neuroblastome (SK-N-MC) cells (Letschert et al., 2014). Moreover, due to its ability for single-molecule detection and position determination dSTORM measurements provided quantitative estimates of molecular densities and spatial distributions of membrane-associated glycoconjugates.

In this section we provide protocols to combine metabolic labeling and fluorescent staining via click chemistry for super-resolution imaging with dSTORM of membrane proteins with single-molecule sensitivity. The method comprises four steps:

During the last decades, fluorescence microscopy has allowed the direct observation of cellular processes in a relatively non-invasive fashion with high molecular specificity and temporal resolution. However, due to the wave nature of light the spatial resolution is limited to approximately half the wavelength of the light in the imaging plane (Abbe, 1873). Recently, super-resolution microscopy methods have circumvented this problem improving the optical resolution substantially. Localization microscopy exhibits the highest spatial resolution of less than 20 nm, as compared to other super-resolution techniques such as STED (Klar et al., 2000) or structural illumination microscopy (SIM) (Gustafsson, 2000). Moreover, due to their single molecule sensitivity, localization microscopy can potentially provide quantitative information about the spatial organization of proteins, as well as the number of molecules residing inside and outside of subcellular compartments including PM nanodomains involved in different cell functions. For example, PALM and dSTORM, in combination with genetically encoded fluorescent photactivable proteins and immunochemistry, respectively, demonstrated nanocluster organization of synaptic proteins (Bar-On et al., 2012; Ehmann et al., 2014), membrane receptors involved in cell growth, proliferation and differentiation (Gao et al., 2015), tumor necrosis (Fricke et al., 2014), or related to the immunological response (Williamson et al., 2011; Rossy et al., 2013). Comparative studies have also proven PM heterogeneity depending on protein membrane anchor types including the transmembrane protein Lat, the lipid-anchored protein Lyn, the vesicular stomatitis viral glycoprotein VSVG, and GPI anchored proteins (Sengupta et al., 2011, 2013). However, all these studies were restricted to a limited number of proteins at a given time and thus, it became obvious that a more general approach for visualizing simultaneously a large population of membrane proteins is required to inspect the global distribution of PM proteins at the nanoscale. Moreover, fluorescent staining with antibodies and genetically encoded fluorescent proteins can induce artificial clustering of membrane proteins (Tanaka et al., 2010; Magenau et al., 2015) and limit the localization precision due to their relatively large size, especially in high density labeled samples. Metabolic labeling fills both gaps by introducing small bioorthogonal chemical groups such as azides into newly synthetized proteins.

Metabolic labeling has been used during the last decade to visualize newly synthetized proteins with standard fluorescent microscopy in cultured cells, tissues, and living animals. The advantage of this staining strategy is two-fold. First, labeling proteins with small and bioorthogonal chemical handles either by co-translational incorporation of unnatural amino acids or by post-translationally modification with non-natural monosaccharides minimizes perturbation of proteins and likely resembles physiological conditions. Second, metabolic labeling constitutes a unique tool to visualize spatial patterns of wide parts of the proteome. Whereas, immunochemistry and genetically encoded fluorescent are useful to visualize one specific protein, metabolic labeling allows to stain simultaneously newly synthesized proteins in a less specific way. Because the azido amino acid L-azidohomoalanine (L-AHA) replaces endogenous methionine, all proteins containing natively at least a single methionine are prompted to be labeled. On the other hand, the peracetylated azido sugar N-azidoacetylgalactosamine (Ac₄GalNAz) is incorporated into specific subtypes of glycans such as mucin-type O-linked glycans and O-GlcNAc-modified glycoproteins (Laughlin and Bertozzi, 2009a). Further identification of which proteins incorporated successfully L-AHA or Ac₄GalNAz has been achieved using alkyne affinity-tags (e.g., biotin-FLAG-alkyne tag) instead of alkyne fluorophores, in combination with proteomics studies (Dieterich et al., 2006; Laughlin et al., 2006). It is important to remark that the incorporation of L-AHA and Ac₄GalNAz into PM proteins occurs during protein translation and post-translational glycosylation before they are delivered to the cell membrane. Therefore, different metabolic labeling conditions (e.g., changes in incubation time or concentration of the azido surrogates) can be used to study not only the spatial but also the temporal organization of newly synthesized proteins and glycans as shown previously by standard live-cell fluorescence microscopy (Baskin et al., 2007; Beatty and Tirrell, 2008; Laughlin et al., 2008; Laughlin and Bertozzi, 2009b; Dieterich et al., 2010).

When combined with super-resolution microscopy, metabolic labeling allows to inspect the overall distribution of membrane proteins at the nanoscale. This has recently been proven by STED and dSTORM imaging of membrane proteins containing unnatural amino acids and azido sugars respectively (Letschert et al., 2014; Saka et al., 2014). Although both techniques provide images with substantially enhanced spatial resolution, due to their peculiarities, they exhibit unique advantages and limitations. For example, dSTORM exhibits better spatial resolution than STED and has the potential to quantify molecular densities of membrane components as well as their spatial distributions. However, due to fluorophore photoswitching kinetics, the necessity of high photon yields, and slow camera frame rates, image acquisition typically requires few minutes (van de Linde et al., 2011). On the other hand, STED achieves much higher temporal resolution and therefore it is more suitable for dynamic studies. Remarkably, STED combined with fluorescence correlation spectroscopy (STED-FCS), where very small areas are scanned at frequencies in the order of a few kHz, can be used to measure diffusion dynamics of membrane proteins and lipids demonstrating molecular confinement with both high spatial and high temporal resolution (Eggeling et al., 2009; Saka et al., 2014).

The conditions presented in the given protocols constitute a robust recipe to stain and visualize large populations of PM proteins and glycans with super-resolution localization microscopy (Figure 2). Nevertheless, critical aspects, as well as limitations and future perspectives, with regard to obtain reliable quantitative data and avoid artifacts are shown in the next subsections. First, we highlight potential artifacts of dSTORM as well as the inherent problem of 2D super-resolution images due to projections of 3D structures such as membrane ruffling, filopodia, overlapping membranes, and vesicles located in close proximity to the PM. Then, we compare the fluorescence staining efficiency achieved by copper-catalyzed and copper-free click chemistry reactions for fixed metabolic labeling conditions. Finally, we show how quantitative information about the distribution of PM components can be percolated from dSTORM data using statistical spatial analysis approaches, such as pair-correlation and Ripley's K functions.

The intrinsic features of localization microscopy, i.e., reconstruction of super-resolution images from localization of single molecules, determine its accuracy, and reliability. The precision of position determination of single and well isolated fluorescent emitters is mainly determined by the number of collected photons, the signal-to-noise ratio, and the accuracy of the algorithm implemented in the localization software used to fit the point-spread-function (PSF) of detected fluorophores (Thompson et al., 2002; Mortensen et al., 2010; Sage et al., 2015). In contrast, other considerations must be taken into account to reconstruct reliable super-resolution images. For example, overlapping PSFs of multiple fluorophores residing in their on-state simultaneously within the same diffraction-limited area must be prevented, except specialized algorithms capable of fitting multiple emitters PSFs are used (Holden et al., 2011; Zhu et al., 2012), to avoid incorrect localizations and ensure artifact-free images reconstruction (van de Linde et al., 2010; Sauer, 2013; van de Linde and Sauer, 2014; Burgert et al., 2015). As a rule of thumb to avoid PSFs overlapping and ensure reliable spot finding and fitting, the density of fluorescent emitters has to be kept below 0.2 spots per μm² (Wolter et al., 2011). Therefore, appropriate measurement conditions in dSTORM imaging such as laser irradiation intensities high enough to transfer the majority of organic dyes to long-living off states as well as suitable buffer compositions are required to guarantee good image quality.

Besides the aforementioned experimental traits of dSTORM, inherent problems and limitations appear when studying membrane components with 2D localization microscopy. Without 3D information the ability to extract unbiased information about PM can be error prone. The existence of Z-projections of inherent cell membrane structures such as invaginations and vesicle-like structures, including fluorophore-filled endosomes in contact with or located near the PM, as well as overlapping membranes in the lamellipodia, might distort severely the quantitative analysis and interpretation of super-resolution images. For example, a sliding window analysis applied to dSTORM images of PM under the nucleus reveals half of the localization density compared to lamellipodia indicating a two-fold membrane structure (Figures 3A,B). Furthermore, circular clusters with apparent sizes ranging from a few tens to a few hundred nanometers can be visually identified from more homogeneous distributions, however it is difficult to discern weather they represent nanodomains enriched in membrane proteins or projections from fluorophore-filled vesicles in close proximity to the membrane. Whereas, a 3D-dSTORM measurement would reduce any information bias on PM organization due to vertical projections, instrumentation, and implementation for 3D-dSTORM is more complex and expensive compared to 2D-dSTORM, and they usually achieve a lower axial than lateral resolution (Klein et al., 2014). In contrast, consecutive imaging of the same cell with slightly shifted focal planes above the feature of interest constitutes a fast control to determine the two-dimensional projection contribution from inherent 3D structures as shown in Figures 3C,D for vesicle-like structures located right above the plasma membrane (yellow circles) or further up (blue circle), and membrane ruffles (green circle) (Burgert et al., 2015).

The first step of any fluorescent microscopy technique is the efficient staining of the protein of interest with a fluorophore. Moreover, in localization microscopy higher staining efficiencies, reflected as higher labeling densities, affects the maximum resolution in localization microscopy (Sauer, 2013). Whereas, imaging resolution is usually defined as the minimal resolvable distance between two emitters, the extractable structural information is also related to the sampling frequency, i.e., fluorophore labeling density, as described by the Nyquist-Shannon theorem (Shannon, 1949). In essence, the theorem states that the sampling interval, i.e., the mean distance between neighboring localized fluorophores, must be at least twice as fine as the structural details to be resolved. Therefore, higher labeling densities prevent under sampling and improve spatial resolution.

The conditions given here for click chemistry staining of membrane proteins and glycoconjugates lead to maximum labeling densities ranging from 400 to 2000 localizations per μm² (Figure 4). For the four bioconjugated systems inspected, we observed that fluorophore concentrations around 20–50 μM are required to maximize fluorescent signal. Moreover, copper-free strain-promoted azide-alkyne cycloaddition (SPAAC) is equally efficient as CuAAC to stain Ac₄GalNAz-derived glycoconjugates, and two-fold better to detect membrane proteins containing AHA. Thus, optimal conditions for click chemistry can also be achieved in absence of copper avoiding toxicity effects and simplifying the protocol.

In dSTORM measurements, localization densities in a certain area of the sample can be directly calculated from the coordinate lists exported by the localization software. Whereas, the number of localizations per unit area can be used to estimate the staining efficiency for different labeling conditions, it only provides relative information on the detected numbers of membrane proteins present. Since organic dyes undergo several photoswitching cycles during a dSTORM measurement, counting molecular numbers with localization microscopy requires further correction for multiple detections of the same molecule. The typical number of localizations recorded per fluorophore under the same optical and chemical conditions can be determined in diluted samples (Figure 5). If the blinking of isolated spots can be unequivocally assigned to single fluorophores, a conversion factor can be extracted to estimate the detected number of labeled membrane proteins (Table 1). For example, we estimate the density of PM proteins labeled with AHA during 4 h 30 min to be approximately ~50 μm⁻² and ~125 μm⁻² when stained via CuAAC and SPAAC respectively. On the other hand, we detected higher densities of glycans, in the range of ~345 μm⁻² and ~280 μm⁻², metabolic labeled with Ac₄GalNAz during 48 h. It is important to mention that dividing the number of localizations in a region of interest by the average number of localizations detected per isolated fluorophore in reference experiments represents only an average correction value. To prevent over-counting effects in highly dense sample areas, more sophisticated methods based on the temporal and spatial fingerprint of single fluorophore blinking, such as off-time gap (Zhao et al., 2014) and pair correlation function analysis (PCF) (Veatch et al., 2012; Sengupta et al., 2013), can be applied.

Beyond density determination, coordinate lists obtained by localization microscopy can be used advantageously to inspect spatial distributions of membrane proteins. Analysis based on pair-correlation function (PCF) (Veatch et al., 2012; Sengupta et al., 2013) or nearest-neighbor based algorithms (including Ripley's K function) (Owen et al., 2012) can indicate weather proteins are more aggregated forming clusters or more dispersed than they were under a distribution of complete spatial randomness. All analysis routines need to take into account local self-clustering induced by single fluorophore blinking. Moreover, quantitative estimation of cluster size and densities can be difficult to extract without prior biological knowledge (Coltharp et al., 2014). Nevertheless, comparison with simulated spatial distributions mimicking experimental data can alleviate these problems and avoid miss-interpretations (Kiskowski et al., 2009; Veatch et al., 2012; Letschert et al., 2014). Finally, clustering algorithms, such as K-Means, DBSCAN, and polygon-based tessellation methods, have been used for morphological analysis of membrane proteins (Bar-On et al., 2012; Ehmann et al., 2014; Löschberger et al., 2014; Levet et al., 2015; Andronov et al., 2016). In contrast to pair-correlation and nearest-neighbor based algorithms, these methods rely on segmentation of the super-resolution image and thus the size and shape of each cluster, as well as their XY position, can be directly visualized.

To characterize the spatial distribution of PM components, we calculated Ripley's h functions from experimental data and two different sets of simulated spatial patterns. In particular, we simulated XY coordinates according to (i) a Poisson process and (ii) a Neyman-Scott process within 5 × 5 μm² with similar density as the number of localizations per μm² obtained from dSTORM images. Whereas, a Poisson process resembles complete spatial randomness, it lacks to mimic individual fluorophore blinking inherent to dSTORM measurements. In contrast, data sets simulated according to the Neyman-Scott process (Neyman and Scott, 1952) account photoswitching cycles from single fluorophores by including Gauss distributed offspring events around each parent position. Number of the offspring events and the standard deviation of the Gauss distribution (σ) where set from experimental data, i.e., on average ~5 blinks per fluorophore and experimental localization precision ~8 nm, respectively.

Ripley's k function reveals possible combinations of homogeneous distributions on large scales and clustering on small scales (e.g., due to the repeated blinking of individual labels). Figure 6 shows direct comparison between experimental (blue line) and simulated data for a Poisson and Neyman-Scott process (black and red line respectively). For all the labeling schemes inspected, our data showed maximum clustering on a length scale similar to the estimated localization precision (i.e., d ~20–30 nm). Therefore, clustering might reflect single fluorophore photoswitching. Since the maximum value of Ripley's h function for a simulated Neyman-Scott process is close to that of experimental data, we conclude that single fluorophore blinking is the only significant clustering process on this length scale. In addition, all the data indicate a small but significant deviation from complete spatial randomness on length scales from 30 to 800 nm. It is important to note that there is no characteristic length scale above 30 nm for any clusters of a well-defined size that can be identified. The indicated deviations from complete spatial randomness can have their origin in the various PM deformations e.g., due to the onset of vesicle formation or membrane ruffling. Whereas, it is possible to find small areas with a distribution that perfectly resemble a Neyman-Scott process (with clusters originating only from single emitter blinking), Ripley's h function for data in areas of 5 × 5 μm² in well-labeled cells under the nucleus (excluding double membrane contributions) typically appear as presented.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.