3.1 Super-resolution structured illumination microscopy

Super-resolution structured illumination microscopy (SR-SIM) (reviewed in [48]) is a wide-field microscopy approach employing coherent spatial modulation of the excitation light to provide a two-fold resolution improvement. In contrast to other super-resolution methods, it does not rely on special properties of the fluorophores as it only requires a linear, incoherent response to the excitation light intensity. Two-dimensional (2D) SR-SIM was first demonstrated by Heintzmann and Cremer [49] and Gustafsson [45]. In 2008, Gustafsson et al. [50] then expanded the technique to three dimensions (3D).

The SR-SIM process relies on mixing of the structured illumination pattern with a known frequency and the structure of the sample with unknown frequency, resulting in a pattern comparable to Moiré fringes. Higher frequency components become detectable as the difference-frequency mixing with the illumination pattern shifts them into the resolvable frequency band. Reconstruction algorithms retrieve this super-resolved information and compute an image with doubled spatial resolution.

To recover the original sample frequencies, the stripes of the illumination pattern have to be moved across the sample in evenly spaced phase steps for a given angle. An image is then taken at each phase position. That way, higher resolution information is collected in the direction of the phase translation. The acquisition of phase shifts is repeated for different evenly spaced angles by rotating the interference pattern to gain an isotropic resolution enhancement.

The SIM reconstruction process is best understood by studying images in Fourier (frequency) space. A 2D Fourier representation (power spectrum) of a real space image shows spatial frequencies within a circular pattern (Figure 3, wide-field image and corresponding Fourier transform). Fine details of the image will appear as high frequencies near the border of this circle, while coarse details result in lower frequencies close to the center. The maximum radius of the circular pattern (shown in yellow) is defined by the resolution limit [Eq. (1)], i.e. higher resolution leads to a larger circle and thus higher resolved frequencies.

Figure 3:

The principle of SIM.

Top row: micrograph images of U2OS cells stained for actin filaments (Phalloidin-ATTO488) acquired with a 3D-SIM in wide-field mode and SIM raw images at different angle orientation of the illumination pattern (1–3). Second row: Corresponding graphic representations of the Fourier transforms that show the frequency information. Bottom row: 3D-SIM reconstruction and the corresponding frequency space representation. In this visualization, low frequency components are centered, high frequency extends to the sides of the image. The strength and the cut-off of these frequency components indicate the increased resolution achieved by SIM compared to wide-field. Adapted from [51].

In SR-SIM raw images, higher frequency information is down-modulated to lower frequencies. As seen in Figure 3 (SIM raw data and Fourier space angles 1–3), this super-resolved information is entangled within the resolvable frequencies. Reconstruction algorithms extract these data and shift it back to their correct position in frequency space. The final image is then obtained after combining the information for each angle (Figure 3, Fourier space of SR-SIM reconstruction) and subsequent reverse transform from Fourier space to real space (Figure 3, SR-SIM reconstruction).

Linear SR-SIM can be utilized in 2D as well as in 3D. For 3D imaging, a structured illumination pattern is generated axially by three-beam interference [50]. This results in 2 times improved optical sectioning. To reconstruct 3D data sets, a total of 15 images per focal plane are needed. The quality of the final super-resolution image relies on the quality of the software implementation of the SR-SIM reconstruction algorithm. Reconstruction artifacts can be mitigated by careful instrument calibration and sample preparation. Recent open-source software tools permit additional quality assessment of the recorded raw data and possible different reconstruction parameter options [52], [53].

3.2 Single molecule localization microscopy

In 2006, three research groups independently developed similar optical super-resolution methods with a lateral sub-diffraction resolution of around 20 nm. While Betzig and collaborators [43] termed their method “photoactivated localization microscopy” (PALM), Hess and colleagues from the University of Maine called it “fluorescence photoactivation localization microscopy” (fPALM) [54]. Zhuang and collaborators [55] from Harvard University developed STORM: “stochastic optical reconstruction microscopy”. In 2014, the Nobel Prize in Chemistry was jointly awarded to Eric Betzig, Stefan Hell, and W.E. Moerner “for the development of super-resolved fluorescence microscopy”. As early as 1995, Betzig [56] proposed the general principle of isolating and localizing single fluorescent molecules in a crowded environment, one of the foundations of far-field optical super-resolution microscopy. All other methods of localization microscopy are therefore considered specific embodiments. These techniques have in common that they use effective means (temporal fluctuations, spectral distinctiveness, and photoswitching) to determine if indeed a single molecule is being probed. The position of the molecule can then be estimated very precisely through the carefully characterized PSF of the optical microscope. Multiple localizations of single molecules obtained in a series of images, usually hundreds to thousands, are then used to form a super-resolved image.

The efficacy of this method can be further improved by the use of photoactivatable or photoswitchable fluorescent proteins or fluorophores. Here, the molecules are switched between a fluorescent bright (“on”) and a non-fluorescent dark (“off”) state either upon illumination with light of different wavelengths (PALM/STORM) or photoactivation and photobleaching (fPALM). fPALM uses fluorescent proteins such as the photoactivatable green fluorescent protein (PA-GFP), while PALM additionally makes use of photoswitchable proteins, e.g. the green fluorescent protein Dronpa [57]. In turn, the first fluorophores that were used for STORM were coupled carbocyanine fluorophores [55], which exhibit photoswitching properties in the presence [58] or absence of a so-called “activator molecule” [59]. The need for coupling these organic fluorophores was soon thereafter removed by the introduction of special buffer systems that serve a similar purpose. The latter method is therefore called directSTORM (dSTORM).

3.2.1 direct Stochastic optical reconstruction microscopy

dSTORM utilizes the photoswitching properties of commercially available fluorophores, which can be achieved by irradiating the sample with either one [60] or two [61] wavelengths of light and an appropriate buffer condition [62]. The switching to the non-fluorescent (“off”) state is triggered by the same wavelength that also excites the molecule’s fluorescence for localization. The fluorescent (“on”) state can be reversibly recovered by the use of laser light at lower wavelength regions, such as 405 nm or 488 nm. For efficient photoswitching of fluorophores specifically engineered for single molecule detection (i.e. fluorophores exhibiting high fluorescence quantum yield, high photostability, and minimal intrinsic “blinking”), controlling the oxygen content in the buffer medium with an enzymatic oxygen scavenging system (OSS) as well as adding thiolated small molecules to the buffer medium permits almost seamless adjustment of the rates at which photoswitching occurs [63]. This leads to a dramatically improved reversibility of the photoswitching process that is required for significant high localization statistics.

In SMLM a small subset of all fluorophores attached to the structure of interest is activated at any time (Figure 4A and B), effectively confining the fluorescence emission of the activated fluorophores (Figure 4C). A fast and low-noise camera capable of exposure times in the millisecond range is used for image acquisition. High-precision localization with nanometer accuracy is then performed for each fluorescent spot detected in each individual image (Figure 4D) [64]. A sequence of several hundreds to thousands of fluorescence images (“image stack”) is recorded using illumination of an entire field of view (f-o-v). The specific illumination mode, [e.g. wide-field or total internal reflection fluorescence (TIRF) illumination, or the use of light sheets, to name just a few examples] is typically adapted to the specific experimental requirements [65]. All localized fluorophores are subsequently combined to form one super-resolved image that is able to resolve structures down to 20 nm (Figure 4E).

Figure 4:

The principle of dSTORM.

(A) Conventional image of the cellular microtubule network, (B) selective activation of a small subset of fluorophores attached to the sample, (C) diffraction limited image of a single molecule, (D) localization of single molecule position using 2D Gaussian function, and (E) comparison of conventional TIRF (lower left part) to dSTORM image (FWHM 21 nm).

3.3 Stimulated emission depletion microscopy

Stimulated emission depletion (STED) microscopy relies on Einstein’s concept of stimulated emission. The concept of using stimulated emission as a means of improving the spatial resolution of optical microscopy in the far field beyond Abbe’s diffraction limit was first proposed by Hell and Wichmann in 1994 [41]. It then took almost 6 years to experimentally demonstrate that this theoretical concept does, indeed, enable one to overcome the diffraction limit, when Klar et al. [74] were able to show a narrowing of the axial resolution from approximately 600 nm to about 100 nm by generating high intensity depleting laser spots just above and below the fluorescence excitation focus of a laser scanning confocal microscope. The concept relies on the fact that a molecule in an electronic excited state has several competing pathways through which it can relax back to the electronic ground state. The most frequently observed pathway depends on the spontaneous emission of a fluorescent photon, the rate of which is primarily ruled by the excited state lifetime. In the absence of external fields, molecules can also relax through triplet states and by thermal means, albeit at much lower rates. However, as Einstein showed and as realized in the laser, the presence of a strong, resonant electromagnetic field that fulfills several important attributes (i.e. wavelength, propagation vector, polarization, etc.), can also induce the transition of a molecule from the excited state to the ground state and leads to the emission of a photon with those same attributes. This mechanism has several important properties that allow it to be exploited in STED microscopy. First, the emission is induced, i.e. instantaneous, and does not depend on the excited state lifetime. Also, rather than the broad emission spectrum obtained by spontaneous emission, STED emission is monochromatic at the same wavelength used for the induction of the process. And, most importantly, the dependence of the process on the intensity of the laser beam inducing the transition is highly nonlinear [75]. Thus, by overlapping a laser beam with a wavelength within the emission spectrum of a fluorescent molecule with a confocal laser beam that excites fluorescence, a number of molecules in the excited state can be forced back to the ground state by stimulated emission. This introduces non-linearity as the emission begins to saturate above a certain intensity of the stimulating laser beam. Blocking the wavelength at which stimulated emission occurs, e.g. by a narrow notch filter, enables one to still collect the spontaneous fluorescence signal of the molecules remaining in the excited state. By shaping the STED beam, ideally in the form of a donut, where the centroid of the donut overlaps with the centroid of the confocal excitation beam, the nonlinear dependence of STED leads to a spatial narrowing of the molecules that remain in the excited state to a region in the very center of the donut beam. This concept has several important advantages: it enables “scaling” of the spatial resolution depending on the intensity of the STED beam. This has been advantageously exploited in STED fluorescence correlation spectroscopy, which has, e.g. allowed Eggeling and coworkers to determine diffusion rates of constituents of the cellular plasma membrane across multiple length scales well below the optical diffraction limit [76]. Also, another important advantage is that STED, implemented in a laser scanning microscope, produces “instant” super-resolution, meaning that no additional image processing is needed in order to obtain a super-resolved fluorescence image. A disadvantage, however, is the fact that in order to fully suppress spontaneous emission from the excited state, rather high STED powers are required, which is often detrimental to sample health. Here, the combination with photoswitchable fluorophores, as realized in “REversible Saturable Optical Linear Fluorescence Transition” (RESOLFT, i.e. photoswitching with low power), has significantly reduced this dependence at the expense of increased sample preparation [77].

3.4 Non-linear SR-SIM imaging

All of the applications of SR-SIM applied to LSECs, as well as the vast majority of SR-SIM applications related to a biological or biomedical research problem, utilize the linear implementation of SR-SIM. Linear SR-SIM means that the illumination pattern is purely a cosine-like interference pattern with gently rising slopes, leading to at most a doubled spatial resolution. As early as 2002, Heintzmann et al. had, however, already pointed out that SR-SIM is not conceptually or fundamentally limited to a mere doubling of its spatial resolution [78]. Gustafsson in 2005 first demonstrated nonlinear (NL-)SIM using the process of fluorescence saturation [79]. The source of fluorescence in organic fluorophores is the electronic excited state with a limited lifetime of a few nanoseconds. The excited state becomes more and more populated at high excitation intensities, because the probability for transferring an electron from the ground state to the excited state continues to rise linearly with the excitation intensity. The finite time that the electron spends in the excited state, however, prevents rapid depopulation. The fluorescence signal will therefore no longer linearly respond to the excitation power and begins to approach a maximum value. This saturation effect introduces a nonlinearity into fluorescence excitation, which could, in principle, be used for non-linear structured illumination microscopy (NL-SIM). A caveat of this mechanism is that the more time molecules spend in the excited state, the more of them will undergo irreversible photobleaching. Also, high excitation powers at visible wavelengths are typically detrimental to the health of biological samples, posing yet another limitation. While these issues are most relevant for live imaging, it should be pointed out that there is also a potential link between fluorescent staining and high excitation powers with regard to sample health as unstained samples are often more likely to survive exposure to high laser powers. This is related to the formation of free radicals by fluorescence excitation, which can often be minimized but never entirely avoided [80]. Nonetheless, by carefully controlling experimental conditions, i.e. by using a high power, Q-switched diode laser and adjusting its repetition rate to the camera exposure time, Gustafsson was able to keep these effects to a minimum, and he demonstrated the detection of 50 nm beads by NL-SIM with a resolution of 49 nm. In order to extract the additional higher order harmonics from the limited support of the optical transfer function (OTF), images at nine different phases and 12 different angles had to be collected, resulting in 108 images to be processed by the SR-SIM reconstruction algorithm. In this experiment, the same pattern was used to induce the nonlinear saturated excited state and to excite fluorescence. Other than the pulsed diode laser and requiring more and smaller phase steps and angles, no further modifications were made to the SR-SIM instrument [79]. In principle, this approach could be applied to imaging samples with more robust, non-bleaching fluorophores, such as nanodiamonds or quantum dots, but biological applications have, so far, not been demonstrated with this implementation of NL-SIM.

In the search for more biology-friendly approaches to NL-SIM, the group of Mats Gustafsson next set their eyes on other types of fluorophores that provide an intrinsic nonlinear optical response, which could be utilized at lower excitation power levels. This was demonstrated by Rego et al. in 2012, where they utilized a photoswitchable protein, combined with 2D TIRF-SIM to obtain similar spatial resolution as in the prior use of saturated fluorescence excitation [81]. Again, by carefully characterizing the photoswitchable protein Dronpa, a condition was found that allowed them to switch Dronpa between a nonfluorescent “off” state and a fluorescent “on” state between 60 and 70 times. Here, the NL-SIM image acquisition process required three consecutive exposures of the sample. First, all Dronpa molecules were switched to the “on” state by applying uniform illumination of the sample at 405 nm wavelength. Then, a sinusoidal excitation pattern was used to turn the molecules off – the duration and power of which was used to control the saturation of this process, such that only narrowly confined strips of molecules along the minima of the cosine pattern remained in the on state. Then, after shifting the phase of the excitation pattern by π, the fluorescence of the molecules remaining in the on state was collected. This process required only seven phase steps and nine different angles or 63 images in total to be collected in order to extract the information of two higher orders, resulting in the imaging of the cytoskeleton and nuclear pores in fixed cells with approximately 60 nm resolution across the entire f-o-v [81]. The authors already mention in this paper that as few as five phases and five angles, resulting in 25 images, could be used to detect these two higher orders.

More recently, Li et al. demonstrated an impressive improvement of NL-SIM by extending it to live cell applications. Several changes were made to the previous procedures: an increased NA of the objective lens readily extended the spatial resolution of TIRF-SIM to well below 100 nm. This enabled the rapid imaging of endocytic uptake processes and cytoskeletal changes in cells with 84 nm spatial resolution for up to 100 time points (i.e. final reconstructed, super-resolved images) [82]. By utilizing the green photoswitchable fluorescent protein Skylan-NS, NL-SIM was implemented in TIRF mode. Here, a 1.57 NA TIRF lens was used rather than the 1.7 NA TIRF lens, because the index oil used as immersion fluid for the 1.7 NA objective lens was found to absorb blue light at 405 nm. The image acquisition procedure was also changed, such that fluorophores were kept in the off state. Exposure of a small part of the sample to a sinusoidal activation pattern at 405 nm was then followed by a carefully controlled excitation pattern at 488 nm. Here, the two sinusoidal standing wave patterns had to be modified such that they precisely overlapped in TIRF mode across the entire f-o-v. This overlap between the patterns increased the nonlinearity of the process, allowed for low activation and excitation intensities, and kept the number of molecules present in the excited state to a minimum, all of which prolonged the lifetime of the sample and reduced phototoxicity. With this procedure, as few as 25 images (5 phase×5 angles), as well as the use of a high NA objective lens, were sufficient to reconstruct images of living cells approaching 60 nm spatial resolution. It should be noted, however, that utilizing this minimum number of images, as well as the low signal-to-noise ratio of this imaging procedure, results in stronger image reconstruction artifacts of the final images [83], [84]. Improvements to the image reconstruction algorithm or filters used to reconstruct NL-SIM images should, however, help reduce these effects in future implementations [85], [86]. Notably, by implementing the principles of NL-SIM to lattice light sheet microscopy, the resolution of this imaging method could also be improved in 3D, albeit not in all three directions at the same time [82]. Although the implementation of NL-SIM is more complex than standard SR-SIM, requiring the use of special, photoswitchable dyes imaged by a SIM microscope providing for precise pattern alignment, we expect this method to witness rapid growth in the near future.

4.1 SR-SIM applied to LSECs

Cogger et al. [87] first showed that super-resolved SIM offers the possibility to investigate LSECs. The spatial resolution of linear SR-SIM reaches about 100 nm (lateral) in the best case. Since SR-SIM utilizes standard fluorescent microscopy protocols or even non-bleaching, non-fluorescent surface-enhanced Raman scattering (SERS) particles [88], SR-SIM is ideally suited for biological systems that are difficult to use in conjunction with fluorescent proteins.

SR-SIM was used to first visualize fenestrations and their distribution in sieves plates in isolated rat LSECs. For this purpose the cell membranes were labeled with CellMask Orange after cellular fixation. This dye results in fairly uniform staining of the plasma membrane. On fixed samples, it will also incorporate into all intracellular lipid membranes.

The measured average fenestration diameter of 123±24 nm was similar to the sizes and structure obtained from scanning EM studies [87]. Diameters as small as 69 nm were measured. The distribution of the fenestrations in sieve plates corresponds to the very thin cellular areas distinct from the thicker cytoplasmic volume surrounding the nucleus.

Furthermore, the previously speculated tight interaction between the actin cytoskeleton and the fenestrations distributed into sieve plates was imaged [87], [89]. While thin actin filaments surround fenestrations, thick bundles and tubulin form rings around the sieve plates. Endothelial nitric oxide synthase, a potential key protein for the regulation of fenestrations, appeared to colocalize with the actin cytoskeleton surrounding but not within the sieve plates [87]. 3D-SIM was further used to assess the localization of lipid raft associated stain Bodipy FL C5 ganglioside GM1 in rat LSECs. Rafts appeared preferentially distributed in the perinuclear regions of LSECs while not being present in sieve plates. These SIM observations lead to the postulation that fenestrations form in non-raft regions and diminished actin filament supported regions of LSECs [90].

4.2 Direct stochastic optical reconstruction microscopy applied to LSECs

Recently, Mönkemöller et al. [15] applied dSTORM to LSECs to gain deeper insight into the structural details of fenestrations in LSECs in an aqueous environment. dSTORM enables 5 times greater spatial resolution than 3D-SIM for typical biological samples, making it a great choice for obtaining high structural detail.

The relative thinness of LSECs (150–300 nm) make them ideal candidates for TIRF/HILO (highly inclined and laminated optical sheet) imaging techniques that are typically used in SMLM. This resulted in imaging with low background as the sample thickness is within the z-resolution of single emitters.

Super-resolved imaging of LSECs is, however, a challenging task because multiple experimental parameters have to be considered. The complete membrane has to be stained to observe fenestrations because no specific markers for fenestrations are available. Generally speaking, imaging of tiny holes in biological samples (“negative imaging”) is much more demanding compared to, e.g. imaging of the fluorescently labeled cytoskeleton (“positive imaging”).

Negative staining requires a significantly lower background to visualize dark areas, which are the objects of interest. To avoid nonspecific binding of fluorophores to the coverslip surface, several washing steps using a mild detergent (0.1% Tween 20) are performed during sample preparation, which we found to have no effect on the membrane of fixed cells [15]. Additionally, the surrounding structure must be labeled with fluorophores as densely as possible. According to the Nyquist-Shannon sampling theorem the mean density of the fluorophores has to be twice as high as the desired resolution. This results in an apparent labeling density of 10⁴ molecules/μm² to obtain a spatial resolution of 20 nm in 2D. For the same resolution in 3D the density increases to 10⁶ molecules/μm³, which corresponds to a fluorophore concentration in the millimolar range.

As mentioned earlier, dSTORM applies a Gaussian fit function to a spatially separated fluorescence signal. With increasing labeling density, the probability for non-separated signals rises. This makes an accurate fluorophore position retrieval using only one Gaussian function impossible. In order to process data containing fluorescence signal emitted from connected fluorophores, fitting of multiple Gaussian functions can be performed. Although time-consuming, software packages capable of multiple emitter fitting are available [68], [91].

In the end, localization precision is also crucial, especially for a quantitative data analysis, e.g. to determine the diameter of fenestrations in LSECs. Since Poisson statistics limits localization precision, as many photons as possible have to be collected from each single molecule. Post-processing of the data is very effective at increasing the quality of super-resolved images. Taking only localizations with a rather high precision into account, i.e. rejecting localizations containing photons below a certain threshold value, lowers the total number of localizations. This leads to a reconstructed super-resolved image of lower quality, but longer acquisition times resulting in more raw data images to be processed can easily compensate for this.

The challenge lies then in finding a photoswitchable fluorophore that offers dense and evenly distributed plasma membrane staining to best observe the negative contrast. We found that by using an OSS in combination with cysteamine [62], an increased concentration of another variant of the CellMask stain family, CellMask DeepRed (5 μg/ml), can be used for SMLM as shown in Figure 6. An example for the dSTORM reconstruction of an entire isolated rat LSEC is shown in Figure 6A. In this super-resolved image the characteristic cellular substructures can be clearly identified. The cell contains a large number of sieve plates with diameters of up to several microns. Each sieve plate contains tens of well-defined fenestrations in the plasma membrane with a mean diameter of about 120 nm (Figure 6B). The wide-field fluorescence image, on the other hand, is almost featureless with dark regions embedded in brighter surroundings adumbrating the location of sieve plates. Only by verification with the complementary dSTORM image can these regions be identified as sieve plates.

Figure 6:

Single molecule localization microscopy reveals nanopores in the cellular plasma membrane.

(A) dSTORM image of a LSEC plated on a fibronectin-coated glass coverslip, stained with 5 μg/ml CellMask Deep Red, and post-fixed with 3.7% paraformaldehyde. The single frame exposure time was 29.55 ms, and a total of 13,483 frames were used in this reconstruction, resulting in a total acquisition time of about 7 min. The region highlighted in (A) is shown magnified in (B) and compared to the conventional wide-field microscopy image (C). Sieve plates and individual fenestrations can be seen as dark spots against the bright background from the plasma membrane. Taken from [15].

Since dSTORM can, in principle, achieve a localization precision of less than 20 nm, it enables observations of smaller structures than 3D-SIM which is limited to 100 nm. Fenestration diameters below 50 nm can thus still be observed. The quantitative analysis of the pore diameters of SIM and dSTORM measurements lead to similar results (129±33 nm for 3D-SIM and 120±38 nm for dSTORM) [15], with dSTORM exhibiting a more “hexagonal” inner structure of the pores, which appear rounded in 3D-SIM.

These results demonstrate that SMLM is able to replace the more elaborate and expensive procedures required to view fenestrations by SEM (Figure 2). dSTORM can now be performed even on standard fluorescence microscopes with super-resolution capabilities at very low cost [92]. Alternatively, SR-SIM also offers the possibility to analyze multiple structures simultaneously at high speed using different stains with high chemical specificity. This makes the analysis of LSECs and their nanosized substructures accessible to a much larger range of biomedical research laboratories around the world.

4.2.1 Correlative 3D-SIM and dSTORM imaging

The higher spatial resolution of dSTORM is bought by extensive image acquisition times, because it requires the collection of 10,000–50,000 frames compared to just 15 frames per z-slice (3D-SIM). Mönkemöller et al. [89] recently showed the advantage of combining two complementary super-resolution techniques if the targets of interest, such as the cellular fenestrations of LSECs in the study, are only visible at resolutions well below the optical diffraction limit. In this study a multi-modal approach for imaging LSECs combines the rapid search mode over large areas (up to hundreds of microns) by 3D-SIM that enables the identification of relevant features, e.g. fenestrations, followed by the more time consuming and higher resolving dSTORM imaging of regions of interest on a single imaging platform. Other groups combined SIM and SMLM [93], [94], [95]. However, such a combination of 3D-SIM and dSTORM, to take advantage of the specific strengths of each technique (rapid, multi-color 3D-SIM search and stitch mode to survey large samples), had not been utilized.

A correlative analysis of 3D-SIM and dSTORM requires a switching buffer system that is suitable for both super-resolution methods. The chemical environment is crucial to facilitate successful dSTORM reconstructions, and most measurements are therefore conducted using OSS buffers. In contrast, 3D-SIM imaging requires mounting media with well-adjusted index of refraction that enables aberration-reduced imaging of cells, e.g. Vectashield. Olivier et al. [96] recently characterized Vectashield as an excellent switching buffer for certain dyes such as Alexa Fluor 647. Shim et al. [97] describe DiI and DiD as plasma membrane stains with excellent photoswitching behavior under standard buffer conditions, while Mönkemöller et al. [89] showed the ability of using DiD in Vectashield for dSTORM imaging of negative structures (Figure 7C). In contrast to the plasma membrane stain CellMask (Figure 6), DiD has higher specificity for cell membranes and shows extremely low background.

Figure 7:

Comparison of cellular features imaged by different super-resolution microscopy modalities.

(A) Nine multi-color maximum intensity z-projection 3D-SIM images of fixed rat LSECs were stitched together to produce this overview image. The cells were stained for nuclei (DAPI, blue), actin (Phalloidin-AF488, green) and membrane (DiD, magenta). (B) Enlarged 3D-SIM view of the ROI shown in (A) highlighting how fenestrations are surrounded by actin fibers. (C) Is the corresponding dSTORM image of the DiD membrane channel. (D) Is an enlarged view of the ROI shown in (B). (E) Plot of the line section shown in (D) comparing the actin (green) and membrane (magenta) channels of the 3D-SIM image. (F) Overlay of the actin channel from 3D-SIM (D) shown in green and the membrane channel of dSTORM (C, outlined box) shown in gray. (G) Plot of the line section shown in (F) comparing the 3D-SIM actin (green) and dSTORM membrane (gray) channels. (G) The actin line (green) shows the same trend as the membrane dSTORM line (gray), which suggests that actin filaments support fenestrations. Taken from [89].

This multimodal imaging is shown in Figure 7 on fixed, freshly isolated rat LSECs, which are stained for nuclei (blue, DAPI), actin (green, Phalloidin-AF488), and membranes (magenta, DiD) to analyze the connection between the plasma membrane of the fenestrations and the actin cytoskeleton. 3D-SIM imaging offered the fast survey over a large area as shown in Figure 7A. A comparison of the 3D-SIM and dSTORM images of a region of interest (ROI) is depicted in Figure 7B and C. The cross-modality correlation of a sieve plate imaged by both super-resolution optical methods enables one to directly compare the association between the LSEC cytoskeleton and the plasma membrane (Figure 7F). The actin cytoskeleton colocalizes well with the membrane stain obtained by dSTORM (Figures 7G and 7F), while the 3D-SIM membrane signal lacks the resolution to truly make this conclusion (Figure 7D and E).

In conclusion, 3D-SIM and dSTORM have advantages that can significantly improve studies of subcellular ultrastructures of LSECs and other biological systems [98] by combining both techniques to reduce their respective disadvantages.

4.2.2 Correlative chip-based optical nanoscopy with dSTORM

A novel (and potentially game-changing) illumination method has recently been used for dSTORM. Diekmann et al. [99] used photonic chips as substrates for dSTORM imaging of LSECs. To avoid out-of-focus light, in conventional dSTORM a high NA high magnification TIRF objective lens is used to generate an evanescent field for exciting fluorescent molecules. The same TIRF objective lens is used to excite and to the collect the fluorescence signal from the sample. Therefore, in conventional dSTORM the f-o-v is limited (typically 50×50 μm²) due to the high magnification of the TIRF objective lens. Diekmann et al. [99], however, instead used a waveguide chip to hold and to illuminate the sample to obtain super-resolution images (Figure 8A). The light was coupled into an optical waveguide and was guided to the imaging area by total internal reflection [100]. The evanescent field present on top of the waveguide surface illuminated the sample, and the fluorescence signals from the sample were imaged using an ordinary objective lens (Figure 8B and C). High intensity (0.5–10 KW/cm²) in the evanescent field above the waveguide is desirable to induce blinking of single molecules and is a prerequisite for SMLM. This was achieved by using high-refractive index contrast waveguide material (such as tantalum pentoxide, Ta₂O₅) and by fabricating a thin waveguide geometry (100–200 nm). Different waveguide geometries (rib and strip, Figure 8A and B) were used for chip-based optical nanoscopy. LSECs were directly grown on top of waveguides and similar to how the samples are prepared on top of glass coverslips. The fenestrations in LSECs were imaged with a water immersion 1.2 N.A. 60X objective lens (Figure 8D and E) using an upright optical microscope setup. The optical resolution of 50 nm was reported using chip-based dSTORM setup. Multi-color chip-based dSTORM imaging was performed on LSECs to image the interaction between the actin network and fenestrations (Figure 8D and E). It was thus further confirmed that actin filaments colocalize with the membrane supporting both individual fenestrations and sieve plates.

Figure 8:

Chip-based nanoscopy.

(A) Schematic diagram of optical waveguide made of either rib or strip geometry. (B) The evanescent field generated on top of tantalum pentoxide (Ta₂O₅) is used to excite the sample. (C) The setup consists of an upright microscope for fluorescence detection and an illumination unit to provide coupling to the input facet of the waveguides. The illumination and collection light path is thus decoupled in chip-based nanoscopy techniques. (D) and (E) Dual color chip-based dSTORM reveals the interplay between actin (magenta) and the plasma membrane (green) in LSECs. (D) Groups of fenestrations form sieve plate superstructures that are surrounded by actin bundles. The inset shows that actin is present between neighboring fenestrations where it colocalizes with the plasma membrane (scale bar, 5 μm and the scale bar on inset, 1 μm), (E) The line profiles taken at different positions in the liver cell as shown in (E) reveal diameters of approximately 200 nm. Figures taken from [99] and used with permission.

Chip-based optical nanoscopy decouples the illumination and the collection light paths such that the generation of the evanescent field is independent of the collection objective lens (see Figure 8C). This enables the acquisition of super-resolved LSEC images over an extraordinarily large f-o-v using a lower magnification objective lens. Using an optical waveguide of 0.5 mm width and an objective lens of 20X 0.45 N.A, an f-o-v of 0.5×0.5 mm was imaged with an optical resolution of 140 nm. The use of waveguide illumination has the potential to revolutionize nanoscopy in that waveguide slides/chambers can be used to retrofit standard diffraction limited microscopes, turning them into nanoscopes for a fraction of the cost of commercial nanoscopy systems. And, although coupling lasers to waveguides is not (yet) a trivial exercise, they still represent a significant reduction in complexity compared to the bulky opto-mechanics of nanoscopes available on the market today.

5.1 One-photon excitation nanoscopy on thin tissue sections

5.2 Optical nanoscopy on thick tissues/tissue sections