Both STED and PALM, as well as localization microscopy techniques using standard fluorophores, allow us to also obtain a super-resolved image of a live cell [20]. However, the road from single “snapshots” of live cells to in vivo observations over a longer period of time is difficult, to say the least, due to the phototoxicity of the illumination intensities required for the “squeezing” of the fluorescent area (in STED) or for photoswitching (in PALM, [d]STORM and related approaches). The two fundamental limitations for imaging (not only live cell) are signal-to-noise (SNR) level and the structure integrity (of the sample), and for localization microscopy additionally the photon statistics of single molecule fluorescence emission [21], i.e., the absolute signal level (Figure 1). Localization accuracy and data acquisition speed are limited by the finite photon emission rate, the finite total number of photons emitted from a dye molecule, the noise level (of the detector and of the background signal), and in localization microscopy also the blinking rate. Super-resolution images are based on composition of thousands of scan points (in STED) or thousands of frames acquired (in single molecule localization microscopy); therefore, live cell experiments with their inherent mobility rely on sufficiently fast data acquisition. However, specifically in Localization Microscopy, in order to detect a fluorescent signal, a number of photons must be registered with the detector. The absorption cross-section, the quantum yield, and the life-time of the excited (fluorescent) state of a dye molecule, as well as the transition to a non-fluorescent state (e.g., the triplet state) fundamentally limit the photon emission rate. Quantum dots exhibit excellent photophysical characteristics for SRM [22, 23], but are challenging to interface with biological systems and structures.

The label size and solubility gives an indication for the accessibility (and hence the labeling efficiency) as well as for the distance between the measured position and the true location of the labeled target, while larger label sizes facilitate better targeting sensitivity. Additives such as the imaging buffer medium (which is used to control electronic transitions and photobleaching), or binding of the dye itself affect the structural integrity of the biological specimens, as does the incident light dose. In STED, circumventing the Abbe-limit means to apply STED intensities several orders of magnitude above the saturation intensity [3, 7]; in PALM (as described by Betzig et al. [14]), the low intensity illumination with UV light (405 nm) has to be repeated thousands of times (or used continuously); hence, with time also in this technique phototoxic doses of UV light are highly probable to be administered. While the disexcitation STED beam uses wavelengths that are much less phototoxic than the blue or UV wavelength used in PALM, the light doses are still sufficient to potentially inflict serious damage to the sample not only in biological samples, but also in the material sciences. Also in other localization microscopy approaches based on photoswitching of standard fluorophores using illumination wavelengths in the visible range only, phototoxicity remains a severe challenge for multiple frame live cell imaging. New dyes have to be developed with saturation/switching intensities sufficiently low for multiple time-lapse SRM imaging.

As an alternative way (or intermediate solution) to realize prolonged live cell SRM imaging, a combination of methods to circumvent the Abbe-limit of 200 nm should be highly useful. For example, various kinds of structured illumination excitation (SIE) [7] provide images at low or even very low illumination intensities. In structured illumination excitation, the sample illumination features high spatial variation (e.g., a sine-pattern or even a random pattern); thereby, fine object details which are inaccessible in classical microscopy are rendered accessible to the diffraction limited microscopic detection [7, 24–26], and dyes which are photostable can be applied. In this illumination mode, SIE techniques do not completely bypass the Abbe-limit (as the fluorescence response depends linearly on the illumination intensity), they can “stretch” it by a factor of about two to four in the object plane and also along the optical axis: But even a stretching of the 200 nm Abbe limit by a factor of two in all three dimensions can be of substantial importance in Biology and Medicine: In this case the observation volume is smaller by the factor 2³ = 8, i.e., the 3D sampling can be enhanced by about one order of magnitude. In many cases, this will already be sufficient to obtain multiple “nanoimages” of live cells; these high resolution live cell observations can be followed by a last “snapshot” image taken with one of the SRM methods indicated above [27, 28]. This is analogous to the observation of a moving animal at a distance, followed by a close look at the animal after it has been captured; peculiarities of movement can be understood by combining the two aspects, the lower resolution time lapse series and the final, further enhanced super-resolution snapshot. Similar considerations also hold for other advanced microscopy approaches replacing the SIE for in vivo observations, such as confocal microscopy or 4Pi microscopy in the low illumination intensity mode. In many cases such as for the size measurement of biomolecular machines (e.g., replication or transcription factories in a cell) under circumstances where they are optically well separated from one another, it is not required to realize either an enhanced optical resolution according to the Abbe criterion (smallest distance that can be measured between two adjacent point sources), or an enhanced structural resolution according to the Nyquist criterion (related to the density of individual point sources detected). In these cases, structured illumination excitation at very low intensities can be used to determine such sizes in the live cell; since structured illumination has been shown to allow size determination down to about 30 nm [29–31], for many biologically relevant problems structured illumination can provide an information equivalent to very advanced STED/PALM/(d)STORM/SPDM etc. techniques. The techniques of structured illumination can be combined with saturation effects [32–34] also known under the term “RESOLFT”; however, the strain on the fluorescent dyes and the lower SNR have rendered its applicability in live cell studies limited [6]; with the developments of novel, bleaching-resistant probes these difficulties may be overcome.

As a long term perspective, localization microscopy methods using a multitude of photostable spectral characteristics (e.g., absorption bands, emission bands, fluorescence life-times, etc.) should facilitate multiple time-lapse SRM acquisitions with an enhanced resolution substantially below 100 nm and very low illumination conditions; this is possible since in the case of photostable fluorophores, a linear dependency on the illumination intensity only is required. In contrast, STED/SMLM still require non-linear absorption or specific (often toxic) chemical environments. When using photostable spectral characteristics, in contrast to the previous SRM methods, scanning or multiple frame registration is not required to obtain an enhanced optical resolution, e.g., in case of different emission wavelengths all fluorescence signals can be registered simultaneously; in the case of different fluorescence life times, ranging, e.g., from 1 ns to 1 ms, only a short registration time are needed. Successful “proof-of-principle” experiments toward such a solution have been reported already in the 1990s [35, 36] and in the beginning of the 2000s [37], indicating an optical resolution down to few tens of nm. If N types of molecules need to be discriminated within the observation volume of the point spread function (PSF), N different spectral signatures allowing independent registration are required, without prior assumptions on the 3D molecule arrangement. In recent years, these approaches to “photostability” based localization microscopy have been neglected in favor of “blinking” based localization microscopy approaches because only these latter promised to allow a highly enhanced structural resolution according to the Nyquist theorem. As discussed below, using new detectors and dyes it should become possible to meet also this challenge: In case photostable 20 molecule types would become distinguishable in a diffracted limited region (Airy disc) of 200 nm diameter, theoretically up to 20/[0.2 μm × 0.2 μm] = 500 = 22 × 22 molecules could be localized per μm²; or 1 molecule per 1000 nm/22 = 45 nm in each direction in the object plane. In real samples, such maximum values will of course not be achieved. But even if the real structural resolution obtained should not be better than about 100 nm, the great advantage compared with structured illumination would be the single molecule resolution capability. This would for example allow to count the molecules and to locate them relative to each other with a localization accuracy down to the few nm range in time and at illumination intensities sufficiently low for multiple time lapse acquisitions of live cells.

Figure 1 has been limited to the fundamental challenges. Each of these problems may be addressed by a large variety of quantitative technical approaches.

As discussed in the context of live-cell experiments, the signals and noise levels, as well as the structural integrity fundamentally limit the achievable resolution. A particular sample preparation protocol might be necessary to preserve a specific nanostructure, but is not compatible with the imaging buffer protocol required for advanced optical microscopy methods. Furthermore, the high light doses in STED and various types of single molecule localization microscopy approaches lead to formation of chemical radicals and to fast photobleaching. Development of novel probes with low molecular weight and increased photostability in switching-based super-resolution microscopy is currently a major field of research. A combination of structured illumination excitation and other SRM techniques [38] such as “photostability” based localization microscopy at low intensities may be extremely useful also to further enhance substantially the optical resolution [39]: Computer simulations as well as proof-of-principle experiments have shown that the localization of a fluorescent point source by structured illumination excitation may enhance the localization accuracy and hence the optical resolution up to an order of magnitude, depending on the photon yield [40]. The underlying principle is similar to determining the position of the central maximum of the Airy disk from stochastic sampling of the diffraction pattern in 2D localization microscopy [41, 42]. Using an ideal optical system, in the localization microscopy mode a few thousand detected photons would be sufficient to obtain a subnanometer localization accuracy (and hence optical resolution), while under the same assumptions the resolution obtained by homogeneous wide field illumination would be 10 times worse. Since standard fluorophores presently used in localization microscopy may achieve even much higher photon yields, this goal can be realized. This should be possible not only in the object plane but even along the optical axis: Experimental observation using fluorescent point sources and axially structured illumination achieved an axial localization accuracy of 0.7 nm [43]; “virtual microscopy” simulations indicated that a detection of 10,000 photons should be sufficient for a localization accuracy of 0.3 nm (standard deviation) [40], corresponding to a two-point optical resolution around 0.7 nm (Full-Width-at-Half-Maximum/FWHM of a Gaussian shaped localization accuracy distribution). A technical challenge to realize such a microscope with “blinking” fluorophores will be the appropriately fast image acquisition in the structured illumination mode: Typically, the phase of the patterned excitation field has to be moved and rotated in a few milliseconds. This could be achieved already by present technology.

To summarize, from the point of Physics the construction of a combined structured illumination-localization microscope with an optical 3D resolution around 1 nm should become feasible. Such an “Avogadro” microscope (to honor the pioneer of Molecular Physics) would have a 3D resolution millions of times more enhanced than then conventional microscopy technology (V_obsconv/V_obsAvogadro = [200 nm × 200 nm × 600 nm]/[1 nm × 1 nm × 1 nm] = 2,4 × 10⁷. For example, this would make possible to use visible light microscopy to look directly into the nanostructure of the biomolecular complexes responsible for the functioning of the cell's machinery, in transcription, splicing, repair, translation, signaling etc.: If one assumes a biomolecular complex of about 20 nm in size, and a typical protein in the 3 nm range, then a resolution of 1 nm combined with appropriate “multispectral” labeling should be sufficient to resolve all its individual proteins and their relative location. Since at this level of resolution, antibodies and even F(ab)2-fragments would be too bulky, standard fluorophores engineered to bind to specific amino acid residues may be used [28], provided that the targeting specificity is not compromised. At the genome sequence level, a resolution of 1 nm would correspond to a very few base pairs/nucleotides only. Compared with X-ray microscopy or high resolution Electron Microscopy, the optical resolution of the envisaged “Avogadro” microscope may still be 10 times worse due to the real imaging conditions (background noise, long acquisition times, fixation artifacts, etc.); however, it would allow us to study such individual complexes and their formation on the single unit level under physiological conditions. Above all, these difficulties exist in the domain of cellular imaging. However, SRM methods may also be applied to isolated macromolecules/complexes, in case of which the imaging conditions can be highly optimized [44]. If fluorophore discrimination based on the triplet state [45] could be realized, e.g., by illumination of very few molecules only within the detection volume of the PSF and by use of photon counting detectors, this would potentially allow tracking of a high number of individual fluorescently labeled targets inside a live cell.

Even without combination of multiple super-resolution microscopy techniques into a single imaging device, strictly enhanced optical resolution might be obtained by defined swelling of the biological structure [46]. This technique termed “expansion” microscopy which acts on the sample side, rather than on the side of the instrumentation, may also be applicable to resolve subcellular or subnuclear biostructures.

The STED/PALM technology is presently limited to a very small number of different “colors” (i.e., fluorophore combinations); this means that only two or three different molecule types can be distinguished in a given object. In SMLM approaches, this problem is somewhat easier to overcome but remains difficult. The maximum number of compounds simultaneously resolved on the single molecule level by these has been 3–4. The cell is crowded, however, with targets composed of multiple types of macromolecules; their mutual spatial relations are of utmost importance for their function. For example, an important problem in present molecular biology is to study the activity of individual genes on the single cell level. In tissues, this can be done by sequence analysis combined with epigenetic annotation, such as the methylation status of associated histones; to perform this on the single cell level [47] (e.g., for 10,000 individual cells in a tissue section) is extremely difficult in terms of workload as well as expensive in terms of reagents required for sample preparation; but even if this problem should be solved, one would also like to study simultaneously details of transcription, splicing, protein synthesis, transport etc. on the single cell level. Since the products of genes interact with each other, one would like to know these parameters in the functional context on a cell to cell basis, not just the averaged bulk values. For the development of “multiplex” SRM, in particular localization microscopy techniques may be used which are based not on photoactivatable fluorescent proteins or on the very special and hence few molecule types useful for STED but on synthetic fluorophores with different spectral signatures (like differences in the absorption/emission spectrum, or the fluorescence life times).

New multichannel detector devices now allow us to measure the fluorescence life times of individual molecules with a time resolution of few tens of picoseconds [48]; hence many fluorescence life times should become distinguishable even at the same absorption/emission wavelength. If one assumes that 7 molecule types can be distinguished by their absorption/emission spectrum, and that for each of those wavelengths three different fluorescence times could be measured, then 7 × 3 = 21 different reporter molecule types might be discriminated on the single molecule/single cell level. In the case of DNA, for example, short fluorescently labeled oligonucleotides may be used in a combinatorial approach to label a given small section of the genome in an individual cell nucleus [49]. Since with n discrimination possibilities, N = 2ⁿ⁻¹ different combinatorial label schemes would be possible, already 10 different labeling possibilities should be sufficient to simultaneously analyze in an individual nuclei about 1000 different genomic sequence targets, or RNA species. Remaining spectral signatures could be used to label various protein types related to gene activity. The functional context provided in this way could very advantageously be combined with other single cell analysis methods, e.g., DNA/RNA sequencing, or proteomics. In the case of “optically isolated” biomolecular machines (i.e., two functional multi-protein complexes such as, e.g., transcription sites having a mutual distance of >200 nm, or nuclear pore complexes), even a few different spectral signatures would be sufficient to obtain highly relevant nanostructural information on the number, orientation, etc. of the constituting proteins [50]. On the other hand, multi- or hyperspectral detection of individual dye molecule emission—while providing great potential to extract additional information, e.g., about the nanoenvironment and multiplying the number of simultaneously usable fluorescent probes—is strictly limited by the number of photons emitted per dye molecules [51–53].

Another resolution challenge is the axial resolution, especially for imaging of a larger field-of-view: In thin specimens such as individual cells located on an object slide, present SRM methods using high Numerical Aperture (NA) objective lenses have achieved an axial optical resolution down to few tens of nm. However, the axial resolution and hence the 3D observation volume rapidly declines with decreasing NA (required for scanning larger areas). For example, at a NA of 0.7 (assumed refraction index 1) the best achievable (confocal) resolution at 500 nm excitation is about 360 nm, while the axial resolution is at best around 1 μm. Since in localization microscopy, the localization accuracy is proportional to the conventional resolution, the axial localization accuracy and hence the optical resolution along the optical axis is several times worse. This problem can be overcome using an “axial tomographic approach”: The image is taken at different angles (views), from all these angles, a 3D image can be calculated with an optical and structural resolution corresponding to the best (lateral) resolution obtained in each of the views [54].

To realize a “multiplex” super-resolution microscopy, a much more detailed experimental and theoretical understanding of the spectral characteristics and temporal dynamics of single organic molecules on the level of Molecular Physics is required. Such an advanced understanding will be instrumental to provide a rational basis for the synthesis of novel reporter molecules given the complex biochemical environment inside of cells. Typical fluorescent reporter molecules may undergo thousands of different electronic transitions, may modify their 3D structure, or change their composition. During this time, photoproducts may be formed. These can be a source for formation of radicals, for change in the microenvironment, and for toxicity, influencing both the cell as well as the fluorophore itself. An application scheme could be to develop combinations of reporter molecules which specifically alter the local nanoenvironment and hence have a specific effect on the time dependent spectral behavior of individual fluorophores in “blinking” based localization microscopy. As a second application scheme, for multiplex super-resolution in vivo studies it should be highly desirable to develop new tuples of fluorophores with similar absorption/emission spectra but with highly different fluorescent life times, and—where required—extremely low photoswitching intensities.

For the combination of various SRM techniques, reporter molecules should be extremely useful which are photostable at illumination conditions A, and photoswitchable at illumination conditions B [55]. Such molecules could for example be employed to perform prolonged time lapse nano-imaging of live cells and tissue sections using structured illumination schemes at low excitation intensities (conditions A), and then switch to Abbe-limit breaking super-resolution procedures like STED, PALM, STORM/(d)STORM and other “blinking” based localization microscopy techniques, using conditions B for final snapshots.

Once the specifications for photoswitching and fluorescence output required for super-resolution microscopy in the complex chemical environment of biological samples have been determined, corresponding synthetic fluorescent dye molecules need to be designed with high photon yield and high photostability. Achieving cleaner, narrower emission spectra (for more distinct colors) and controlled narrowed lifetime distributions could be additional benefits. Such enhanced dyes would be preferentially less sensitive to the nanoenvironment, or their sensitivity to the nanoenvironment could be tuned for SMLM in a way such as to use the resulting changes in the blinking behavior as novel contrasting agent [56]. Ideally, these dyes should be of low molecular weight, and carry molecular properties that provide control of their photoswitching behavior. A promising approach might be the direct conjugation of the fluorophore with a reducing group. These probes need to be coupled to the target binding site by means of highly selective, efficient, and stable (or transiently stable) binding. Furthermore, the linkers used to bind the fluorophores to the target binding site need to be small; click-chemistry or “nanobodies” are an important step in this direction [57].

One of the most challenging biological applications of novel labeling techniques will be to realize SRM imaging of specific genomic sites in living cells. To this end, it is envisaged that the recently developed imaging techniques using the bacterial CRISPR/Cas system in combination with green fluorescent proteins [58, 59] will be highly useful and significantly improve the capacity to study the conformation and dynamics of chromatin in living (even human) cells on the nanoscale.

Presently, a variety of evaluation programs is available for all kinds of present super-resolution microscopy approaches. Often even the developers do have difficulties to understand exactly which algorithm should be best for which problem. It is desirable to develop user friendly “plug-in programs” which are capable of automatically suggesting optimized evaluation algorithms based on the specific input dataset and on the quantitative analysis to be performed. As a long term perspective, one can imagine that for specific purposes simple, robust, and miniaturized (and therefore mobile) optical detection devices can be built. After acquisition, the data will be transferred to a remote computer system; which performs the detailed quantitative analysis, and returns the results to the local measurement device. Such a possibility is expected to be highly interesting in many application fields, e.g., the control of organic toxins or viruses and bacteria in the environment.

Sometimes it is reported in the literature that the exact localization of a single molecule from its fluorescence signal is “trivial.” This might be the case under ideal circumstances, especially under well-defined optical conditions and in a well-defined chemical environment; but in many practically relevant problems, extraction of exact molecule positions still remains a major challenge. For example, the background signal originating from the surrounding of a molecule, as well as blinking events above and below the z-plane of the molecule considered has to be taken into account.

In structured illumination approaches, one of the great challenges is to reduce the reconstruction time. For example, in our experience, a high precision multicolor structured illumination experiment may take up to several hours to reconstruct on a normal PC, and requires many Gigabytes of storage available. For use in high throughput analyses, or in medical diagnosis, such long times are unacceptable and have to be reduced by orders of magnitude.

Further challenges of SRM are remote processing and automatic calibration procedures. The first steps in SRM after fluorescence registration are the reconstruction of the super-resolved images, immediately followed by data visualization. STED and localization microscopy have less requirements and hence need less computation time. However, these are just the first steps: Quantitative evaluation and comparison with simulation data (computer models) has to follow. For example, localization microscopy of nuclear genome structure presently provides coordinates of millions of DNA sites per cell [55]; it is anticipated that in the near future, these coordinates can be refined to provide also the 3D coordinates of individual short DNA sequences, RNA sequences (including mi-RNAs and other oligonucleotides), and proteins. As in Astrophysics, this allows for the first time to compare the coordinates obtained and their dynamics with the coordinates obtained by numerical modeling; together with quantitative biochemical data, this will eventually result in truly mechanistic models of nuclear organization and gene regulation, and provide us with a deeper understanding of the emergence of complex biosystems from fundamental laws of Physics and Chemistry.

Presently, the thickness of an object which can be analyzed in 3D with an optical/structural resolution using high NA objective lenses is restricted to about 100 μm. This means that in most cases, individual cells arranged in monolayers on glass substrates are studied. It has been shown, however, that under these conditions cells may behave very differently from cells of the same type embedded in a 3D tissue surrounding. To overcome this problem, light sheet microscopy methods [60] have been developed; with Bessel beam shaped light sheets [61, 62], or in combination with axially structured illumination they already now provide a 3D resolution far beyond the conventional Abbe-limit. Such techniques are expected to be developed further and to be extended to still larger objects. An additional way to achieve this is the combination of structured illumination/localization microscopy with optical projection tomography approaches or with axial tomography (see above). In this way, it is anticipated to achieve single molecule resolution in an object close to 1 mm in thickness. For example, this would make possible to study the distribution of cancer related molecules (such as receptors, miRNAs, specific transcribed genes etc.) in individual cells of small tissue sections or 3D cell spheroids, and thus allow us, e.g., to identify metastatically competent cancer stem cells in their cellular context. Another highly relevant application would be to study on the nanoscale brain sections of unprecedented large thickness.

In many biomedical applications, a substantial problem of present super-resolution methods is their limitation to a relatively small field–of-view, typically in the order of 100 μm diameter. This allows a vast scope of applications in fundamental biological research. In many medical applications but also in developmental biology, a field-of-view many times larger should be highly desirable (e.g., 1 mm diameter, or even more).

The solution with high NA objective lenses would be to perform multiple acquisitions at different locations which may be very time-consuming and impractical as overlapping acquisitions cannot easily be realized due to irreversible bleaching. While this is doable in rare scientific cases (like the Nobel Prize winning PALM experiment of Betzig [14]), it should be difficult to implement this into cellular pathology, with thousands of samples to be screened in a given lab/year; time lapse imaging of the whole field of view would be impractical.

There are various approaches to overcome this problem. For example, one might use an objective lens with a lower magnification and NA. A more radical solution would be to scan the object by multiple beams, e.g., 10,000 scanning beams, each scanning a field of view of 100 μm; in this case imaging could be parallelized, corresponding to a total field of view of 1 mm². Such multiple beam scanning devices may be realized by using diffractive elements. In the case of STED beams, first steps in this direction have already been taken [63], and one can obtain even super-resolution; but also with diffraction limited beams like those used in confocal or 4Pi microscopy with laser excitation at constant excitation intensity, the combination with localization microscopy using photostable fluorophores has been shown to be possible [37, 50].

For quantitative analysis, it is essential to notice that in large field-of-view observations, optical aberrations (which can readily be corrected for in observation volumes of a few μm³) will have a more pronounced effect. For this, mapping of the refractive index inside the 3D sample could be a possible solution.

Present SRM techniques still offer an optical resolution 10 times worse than high performance Electron microscopy. For example, the distribution of proteins in a nuclear pore complex can now be studied optically, but the exact fitting of the molecules within the individual complex is still difficult due to the limited localization accuracy. Furthermore, the relation of individual molecule locations to structural landmarks is often difficult to establish, e.g., the question whether a given molecule adjacent to the nuclear envelope is located inside (i.e., in the nucleus) or outside the envelope (i.e., resides in the cytoplasm). Such questions are much easier to study in a combination of SRM techniques with EM. A number of attempts have been made to combine structured illumination with EM [64–67]. One way to do this is to perform first SRM and then transfer the specimen to the electron microscope. Another way would be to develop methods for SRM at cryo-temperatures in such a way that the EM image can be registered immediately afterwards of the same field-of-view [68].

UB and CC are also affiliated with LuciaOptics Forschungszentrum, a non-profit company based in Karlsruhe, Germany.