Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

For more than 400 years, optical microscopy has been utilized to study the microcosm. Hans and Zacharias Janssen in Middelburg in the Netherlands were probably the inventors of the compound microscope in the late sixteenth century. After gradual improvements during the following centuries, the resolution of conventional light microscopes was pushed to its physical limits in the second half of the nineteenth century, especially due to the work of Ernst Abbe, Otto Schott and Carl Zeiss in Jena, Germany. Abbe, among others, realized that in order to resolve a structure, e.g., a regular grating, at least the first-order diffraction has to be captured by the objective lens, resulting in a fundamental limitation to the resolution, known as the famous Abbe resolution law [1],

$$ d = \frac{\lambda }{{2n\,\sin \,\alpha }}. $$
(4.1)

Consequently, two objects can only be resolved in an image if their lateral separation is larger than a minimal distance, d, which depends on the wavelength, λ, and the numerical aperture, n sin α, with refractive index, n, of the medium between the object and the objective lens, and half-angle α of the objective lens aperture. For visible light, this popular relation yields a minimal distance of 200 nm. For a long time, it was widely believed that smaller structures cannot be resolved by using far-field optical microscopy with visible light. Various other techniques have been developed to resolve biological structures even down to atomic resolution. Kendrew [2] and Perutz [3] were the first to determine the molecular structures of proteins by x-ray crystallography [4] after solving the phase problem. For high-resolution protein structure determination of two-dimensional protein crystals [5], electron microscopy [6] is an excellent method. More recently, structures of large biomole-cular aggregates have been unraveled by electron tomography [7]. Atomic force microscopy [8], although limited to surface measurements, has nevertheless proven useful for structural studies especially of membrane proteins [9]. Nuclear magnetic resonance (NMR) spectroscopy is a spectroscopic method that only indirectly probes spatial scales via their effects on magnetic transitions of nuclear spins, but has become an important tool for protein structure determination [10], providing – in addition to structural data – a wealth of information on the dynamics [11].

However, X-ray diffraction, electron microscopy and NMR all require special sample preparations and are mainly restricted to in-vitro studies. Light microscopy, by contrast, is easy to use, fast and non-invasive, and thus ideally suited for visualization of biological structures and processes inside living organisms. These advantages have already been appreciated in the seventeenth century by famous scientists including Hooke, Malpighi and van Leeuwenhoek. Later, in the nineteenth century, Schleiden, Schwann, Flemming and others laid the foundations of cell biology using optical microscopy.

Optical microscopy with fluorescence detection is particularly powerful for cellular imaging [12]. By specific attachment of fluorophores, structures of interest can selectively be made visible due to their reemission of the absorbed light. Fluorescence light can be well separated from the incident and elastically scattered light since it is delayed in time and red-shifted in wavelength (Stokes’s shift), leading to excellent image contrast. Most often, organic dyes and fluorescent proteins are used as fluorescence markers. Organic dyes are molecules of comparatively small (1–2 nm) size that can be attached to cellular structures in various ways, most commonly by using antibody staining. A great variety of fluorescent dyes with excellent brightness, resistance to photobleaching and widely varying absorption and emission wavelengths have been developed. Fluorescent proteins are a family of small proteins (∼3 nm) that contain a fluorescent chromophore in their interior [13]. These marker proteins can be genetically encoded and, therefore, produced by the cell itself; no further staining steps are necessary prior to the experiment [14]. They can be attached to the protein under study by fusing the DNA encoding the fluorescent protein to the gene of the protein. Consequently, the cell expresses a so-called fusion protein, which has an additional fluorescent protein domain that often does not interfere with the biological function of the original protein. The photophysical and photochemical properties of fluorescent proteins are, however, still inferior to those of the best synthetic dyes.

In recent years, novel fluorescent biomarkers of ∼5–10 nm diameter have been introduced that are based on nanocrystals, including metal [15], semiconductor [16] and nanodiamond [17] quantum dots. They show great promise as fluorescent markers due to their excellent brightness and photostability. Yet, their specific attachment to biological structures is more difficult and, therefore, these markers have not yet found as wide-spread application as organic dyes in the life sciences.

Approximately 20 years ago, exciting new developments took off in the field of light microscopy, driven by both progress in fluorescence labeling technology and tremendous advances in key microscope components, e.g. detectors performing close to their physical limits (CCD cameras, avalanche photodiodes), powerful laser light sources (pulsed and cw lasers), optoelectronics and nanomechanical devices for beam steering and sample positioning. Furthermore, the field has also benefited greatly from the availability of affordable, high-performance computers capable of storing vast amounts of data and processing of images in reasonable amounts of time.

Evidently, diffraction is a fundamental physical property of the wave nature of light, and so the resolution of an optical image in a light microscope will always be limited. Nevertheless, in recent years, researchers have been able to find clever strategies that exploit the non-linear response of fluorescent dyes to light irradiation to retrieve spatial information beyond the Abbe barrier. In these ingenious microscope implementations, the image resolution is no longer limited by diffraction. A survey of those methods will be presented in this contribution.

2 Microscope Designs, Linear Widefield and Confocal Techniques

There are two basic far-field fluorescence imaging modes, widefield and confocal, which both use a set of lenses, namely the objective lens, to image an object onto a detector. The process of image formation can mathematically be described by two Fourier transforms in succession. The Fourier transform is limited to a range of spatial frequencies given by a filter function termed the optical transfer function (OTF) because the objective lens cannot capture all the information, as shown by Abbe [1]. By applying the convolution theorem, we find that the image is blurred by the point spread function (PSF),

$$ {{\Phi }_{\text{I}}}\left( {X,Y} \right) = {{\Phi }_0}\left( {x,y} \right) \otimes PSF\left( {x,y} \right), $$
(4.2)

where Φ0 describes the object with spatial coordinates x and y, and ΦI describes the image with spatial coordinates X and Y, respectively. Thus, image formation can be described by repainting the image using a broad brush, with its line width determined by the extension of the PSF.

2.1 Widefield Epifluorescence Microscopy

In standard widefield epifluorescence microscopy, the excitation light reaches the sample via the objective lens. All planes along the optical axis are illuminated equally (for optically thin samples) and thus contribute to the fluorescence image. The objective collects the reemitted fluorescence, and the entire field of view is imaged with an area-sensitive detector (frequently a CCD camera). A sharp image is only formed from the signals emanating close to the focal plane. Because there is no rejection of light from planes beneath and above the focal plane, a blurred background is always present that obfuscates the sharply imaged features in the focal plane. To avoid the strong background in widefield microscopy, researchers often resort to very thin sample slices (z < 2 μm).

2.2 Total Internal Reflection Fluorescence

Alternatively, total internal reflection fluorescence (TIRF) can be used for selective excitation [18]. In this microscopy mode, the excitation light is reflected off the interface between a medium of high refractive index, n, (e.g. glass, n ∼ 1.5) and a medium of lower n (e.g. water, n ∼ 1.3), resulting in an evanescent wave in this medium that decays exponentially in axial direction with a decay parameter, z,

$$ I(z) = {{I}_0} \cdot {{e}^{{ - d/z}}} $$
(4.3)

Here, I 0 and d represent the intensity of the incident light and the penetration depth, respectively. The decay parameter depends on the wavelength of the incident light, the angle of incidence, and the refractive indices of the media forming the interface. Typically, only fluorophores within 100 nm from the surface are efficiently excited. Thus, TIRF microscopy features an excellent axial selectivity but lacks 3D capability.

2.3 Confocal Microscopy

Confocal microscopy is a raster-scanning technique. A point source is imaged to a point in the focal plane of the sample by means of the microscope objective lens. A pinhole that is geometrically confocal to the point source is used to collect the fluorescence emanating from this point. A dichroic mirror separates excitation and emission light paths. The pinhole blocks light emanating from out-of-focus planes, only light from the focal plane is efficiently transmitted, resulting in an axial resolution of ∼500–800 nm with this method. By scanning either the sample or the laser laterally (x-, y-directions) as well as in the z-direction, 3D images can be acquired. The point spread function (PSF, Fig. 4.1a), i.e., the image of a point-like object, is the product of the PSFs for excitation and emission in confocal microscopy,

$$ {{\Phi }_{\text{I}}}\left( {X,Y} \right) = {{\Phi }_0}\left( {x,y} \right) \otimes \left[ {PS{{F}_{\text{exc}}}\left( {x,y} \right) \times PS{{F}_{{\det }}}\left( {x,y} \right)} \right], $$
(4.4)

so the method features a slightly improved lateral resolution over standard microscopy.

Fig. 4.1
figure 1

4Pi microscopy. (a) By using a single objective lens, only fluorescence photons from less than one half the total solid angle (2π) can be collected. Therefore, the PSF of a standard confocal microscope is elongated in the axial direction. With two objectives in an interferometric arrangement, coverage of the full solid angle (4π) is approximated more closely, leading to an interference pattern with a sharp maximum in the axial direction. (b) Side lobes appear due to the limited aperture angles of the objectives but can be removed by mathematical deconvolution. (c) Confocal image of membrane-stained (DiI) E. coli bacteria; scale bar, 500 nm. (d) Raw 4Pi image of the same specimen; the effects of the side lobes are visible as ghost images. (e) 4Pi image after deconvolution

Full size image

In spinning disk confocal microscopy, an array of pinholes is used for multiplex imaging of ∼1,000 confocal spots. Because the transmission of light through the pinhole disk is low, the pinhole disk is rotated in tandem with another disk carrying an identical array of microlenses to enhance the light throughput. The spots are moved by rotation of the disks, and light from all pinholes is collected simultaneously by using a large area detector (CCD camera), resulting in much faster image acquisition than with single-spot confocal scanning.

2.4 Two-Photon Excitation

Two-photon excitation is an alternative way to achieve depth discrimination. Instead of exciting fluorescence with a single photon transition, two photons of half the energy, or twice the wavelength, are used in a non-linear process. To achieve a sufficiently high yield, two-photon excitation requires both a high spatial and temporal density of photons. The temporal density can be provided by using powerful pulsed laser sources, often titanium-sapphire (Ti:Sa) lasers with pulse durations in the picosecond region. Even then, the spatial density is only sufficient close to the focus of the high numerical-aperture objective but not in neighboring axial planes due to the quadratic dependence of the fluorescence emission on the excitation intensity. Therefore, out-of-focus excitation is suppressed, leading to optical sectioning in the axial direction [19]. An additional benefit of two-photon excitation is the improved imaging depth in scattering media due to the weaker Rayleigh scattering of the long-wavelength infrared excitation light, allowing for deep tissue imaging [20].

2.5 Interferometric Microscopy

An interferometric arrangement of two opposing objectives is implemented in 4Pi confocal microscopy [21]. This design further improves the axial resolution and the collection efficiency of the incoherent photons emanating from the sample in all directions, i.e., into the entire solid angle, 4π. The observation volume is sharpened in the axial direction by a two-source interference pattern caused by two counterpropagating wavefronts that are coherently superimposed at the excitation spot, at the detector, or at both (Fig. 4.1b). The side lobes of the interference pattern lead to ghosting in the 4Pi image but can be suppressed by two-photon excitation and confocal detection. Finally, side lobes are removed by applying a mathematical deconvolution algorithm to the raw data. As an example, confocal, 4Pi raw and 4Pi deconvolved images of Escherichia coli bacteria are compared in Fig. 4.1c–e.

An axial resolution of ∼100 nm is typically achieved with 4Pi confocal microscopy [22, 23]. Even though the lateral resolution is not affected, structures are imaged more sharply because of the excellent sectioning in axial direction. A related interferometric approach has also been implemented in a widefield design called I5M [24]. The 4Pi and I5M techniques have been employed in studies of subcellular organelles in 3D [24–28]. Both concepts are still diffraction-limited, but they have been combined with super-resolution techniques including 4Pi-STED [29] and I5S [30], as will be discussed below.

2.6 Structured Illumination Microscopy

Structured-illumination microscopy (SIM) takes advantage of the fact that local information can be transformed into global information by using a Fourier transform [31]. The absence of high spatial frequencies in Fourier space, i.e., the OTF cutoff of the objective lens, results in a lack of detail and, thus, the resolution limitation in real space. By illuminating the sample (Fig. 4.2a) with a regular pattern, e.g., a line grid that still can be optically resolved (Fig. 4.2b), a beat pattern (Moiré fringes) with lower spatial frequency than the original structures emerges in the resulting image (Fig. 4.2c). The image in Fourier space, ψI(k), is given by the Fourier transform of the convolution of the object, Φ0(r), with the illumination pattern, Φill(r), finally convolved with the OTF,

$$ {{\Psi }_{\text{I}}}\left( {\vec{k}} \right) = FT\left[ {{{\Phi }_0}\left( {\vec{r}} \right) \times {{\Phi }_{\text{ill}}}\left( {\vec{r}} \right)} \right] \times OTF\left( {\vec{k}} \right). $$
(4.5)
Fig. 4.2
figure 2

Structured illumination microscopy (SIM). (a) Structures of a specimen that are too close to be resolvable by conventional widefield microscopy. (b) The sample is illuminated with a periodic pattern having a lattice constant close to the resolution limit. (c) In the overlay of the two patterns, the unresolvable sample structures become visible as a Moiré pattern with lower spatial frequency, which can be resolved by the objective lens. (d) Conventional TIRF image and (e) TIRF-SIM image of the microtubule cytoskeleton in a single S2 cell, scale bars, 2 μm. The intensity profile along the yellow lines reveals two microtubules at a distance of 150 nm in the SIM image, whereas this feature cannot be resolved in the conventional image. The images in panels d and e were reprinted from [34] with kind permission

Full size image

The beat pattern contains information about higher spatial frequencies, which can be extracted from the acquired image since the structure of the illumination pattern is known. Patterned illumination is used in combination with widefield or TIRF imaging. To be able to reconstruct a full image, the pattern has to be translated at least three times. Moreover, to enhance the resolution in all lateral directions, the pattern has to be rotated, and images have to be taken in 120° separations. Altogether nine images are, therefore, required using two-beam interference illumination. Because the spatial frequency of the illumination pattern is limited by diffraction, the resolution enhancement of SIM in the linear mode is restricted to a factor of two. Similarly, the resolution can be improved in confocal microscopy, but there, the pinhole size becomes so small in practice that the light throughput is no longer sufficient. SIM also offers axial resolution by using three coherent beams (3D-SIM) [32]. By combining SIM with the interferometric approach I5M, an axial resolution of ∼100 nm has been achieved. This design has been termed I5S [30]. A key advantage of SIM is that the technique works with all types of fluorophores, uses only a single excitation wavelength and can thus easily be extended to multicolor applications [33]. As an example, Fig. 4.2d shows a conventional TIRF image in comparison with a TIRF-SIM image (Fig. 4.2e) of the microtubule cytoskeleton of a S2 cell [34]. Another advantage of the SIM principle is its relatively fast image acquisition. A frame rate of up to 11 Hz has been reported [34], making SIM an excellent tool for live cell imaging experiments [35].

3 Non-Linear Techniques, Super-Resolution Microscopy

All microscopy techniques discussed in Chap. 2 are – despite their finesse – still diffraction-limited and provide an enhancement in resolution by at most a factor of two over the conventional Abbe limit. True super-resolution techniques have been introduced that provide theoretically unlimited image resolution. The common theme in all these techniques is to utilize the intrinsically non-linear response of the fluorescence markers to light irradiation to circumvent the Abbe barrier.

3.1 Photoactivatable Fluorescence Markers

For many fluorescence-based microscopy techniques (except for STED and SSIM, for which regular dyes can be used), photoactivatable fluorophores are key to super-resolution imaging.

3.1.1 Photoactivatable Fluorescent Proteins

Fluorescent proteins are ideally suited as markers for optical imaging of living cells and organisms. For super-resolution imaging, photoactivatable fluorescent proteins, which contain a fluorophore with emission properties that can be controlled by light, have become very popular. Two different modes of photoactivation are presently known. Irreversible photoactivation, also known as photoconversion, involves a permanent photochemical modification of the fluorescent protein, whereas reversible photoactivation, also known as photoswitching, involves an isomerization between a bright cis and a dark trans state [36].

The first photoconvertible fluorescent protein, photoactivatable green fluorescent protein (PA-GFP), was generated by mutagenesis of the original GFP [37]. It is initially non-fluorescent but acquires a green fluorescence after intense irradiation with 400-nm light. The diffusion coefficient of hemagglutinin in live fibroblasts was determined using PA-GFP and fluorescence photoactivation localization microscopy (FPALM, see Sect. 4.3.2.2) imaging [38]. Later, a monomeric red fluorescent protein PA-mRFP1-1 has also been introduced [39]. However, both of these photoactivatable proteins are not ideal for super-resolution imaging because of their low contrast ratio between the active and inactive states. Another class of photoconverting fluorescent proteins features a photoactivation mode in which the emission color is irreversibly altered from green to red upon irradiation with 400-nm light, such as in Kaede [40], KikGR [41] and EosFP [42]. Photoconverters feature a high dynamic range and are excellent tools for most localization based super-resolution microscopy approaches.

For reversible saturable optical fluorescence transitions (RESOLFT, see Sect. 4.3.3) imaging, however, reversible photoswitchers are required that can be turned on and off repeatedly [43]. The first application of RESOLFT utilized asFP595 [44] followed by Dronpa [45] and rsEGFP [46]. Dual-color super-resolution imaging was shown with the Dronpa/EosFP and PS-CFP2/EosFP pairs [47], and the first monomeric photoswitchable red fluorescent protein was rsCherry [48]. IrisFP is a fluorescent protein that combines two photoactivation modes [49, 50]. It is reversibly switchable in the green and, after photoconversion with 400 nm light, turns into a reversibly switchable red fluorescent protein. With the combination of two photoactivation modes, entirely new experiments have become available, such as pulse-chase imaging with superresolution.

3.1.2 Photoactivation of Organic Dyes

In addition to fluorescent proteins, organic dyes can also be reversibly photoswitched and used for super-resolution imaging, including photochromic rhodamine [51] and cyanine dyes, which were employed in the first stochastic optical reconstruction microscopy (STORM, see Sect. 4.3.2.2) experiment [52]. Photoswitchable dyes are usually a lot brighter than fluorescent proteins. Unfortunately, their application in living cells is compromised by their rather limited ability to permeate the plasma membrane. Moreover, thiol reagents at millimolar concentrations are typically required in the sample to induce photoswitching, which may not be compatible with the live cell environment. Irreversibly photoactivatable dyes for super-resolution imaging have also become available, e.g. Q-rhodamine [53, 54].

3.1.3 Nanocrystalline Quantum Dots

In comparison with synthetic dyes and fluorescent proteins, semiconductor nanocrystalline quantum dots exhibit superior photostability and brightness, and are therefore commonly used in single-molecule imaging including single-particle tracking. Recently, by using direct light driven modulation of Mn-doped ZnSe quantum dot fluorescence, RESOLFT imaging has been reported [55]. Often regarded as a nuisance in single-particle tracking, the intermittency of quantum dots has been taken advantage of in a further super-resolution technique termed super-resolution optical fluctuation imaging, SOFI (see Sect. 4.3.5) [56]. Small noble metal nanoparticles also appear promising as fluorescence labels [57]. Their quantum yields are still comparatively low, but they can be excited at a high rate to emit bright fluorescence.

3.2 Localization-Based Super-Resolution Microscopy

From single-particle tracking experiments, researchers have long been aware that the position of a single fluorescence emitter can be determined with a precision significantly exceeding the width of the PSF, which governs the resolution in standard imaging. Actually, using single-molecule optical imaging assays, individual steps of motor proteins along filaments were analyzed with an accuracy of 1 nm [58, 59]. Single-molecule, localization-based super-resolution microscopy exploits this fact and uses photoactivatable fluorochromes to disperse spatial information in the time domain.

3.2.1 Principle of Localization-Based Super-Resolution Imaging

A high density of fluorescence markers is required to faithfully image fine structures in conventional fluorescence microscopy. Because all fluorophores contribute simultaneously to the image, their PSFs overlap and their positions thus cannot be determined individually. However, using photoactivatable fluorophores as markers, the emission properties can be controlled externally by light, thus allowing localization of individual fluorophores with high precision. At the beginning of localization-based super-resolution image acquisition, all markers are maintained in their inactive (off) state. Subsequently, they are sparsely activated, so only a few appear in each image frame taken with a dwell time of typically 1–100 ms. Since the PSFs do not overlap, each marker can be localized individually. A large number of frames (102–104) are taken by repeated acquisition and are all analyzed individually. Green-to-red photoconvertible fluorescent proteins such as EosFP have been shown to be very useful for localization-based imaging [36, 42, 60]. By an appropriate adjustment of the 400-nm activating laser intensity, only a small number of molecules are converted to red emitters and registered in the red color channel during acquisition of a CCD camera frame of 1–100 ms integration time. Consequently, photoconversion and photobleaching are kept in balance until the supply of fluorescent markers is depleted. Sophisticated software has been developed to determine the positions of the fluorophores in each individual image. Finally, a density map depicting the distribution of emitters is produced with a resolution in the range of a few tens of nanometers. Figure 4.3a illustrates the entire process. The two-dimensional localization precision, i.e., the standard error of the mean,

$$ {{\sigma }^2} = \frac{{\sigma_{\text{PSF}}^2}}{N} + \frac{{{{a}^2}/12}}{N} + \frac{{8\pi \sigma_{\text{PSF}}^4{{b}^2}}}{{{{a}^2}{{N}^2}}}, $$
(4.6)

is governed by the number of photons, N, detected from each fluorophore; σPSF is the standard deviation of the PSF, a is the camera pixel size and b the background noise [18]. The localization precision consists of three contributions: the first term represents photon statistics, the second term results from the finite size of the detector pixels, and the last term arises from background noise. As is evident from this equation, the resolution in localization-based super-resolution microscopy is theoretically unlimited because the uncertainty in the determination of the mean of the distribution approaches zero with increasing number of collected photons. The only limitation is the finite size and labeling accuracy, and the number of photons collected from individual fluorescent markers.

Fig. 4.3
figure 3

Photoactivation localization microscopy (PALM). (a) Schematic depiction of the principle. By weak illumination with light of a specific wavelength, a small subset of spatially well separated fluorophores is photoactivated to a fluorescent state. The activated molecules are imaged by using a laser that excites their fluorescence until photobleaching occurs. Because the individual PSFs do not show significant overlap, each marker can be localized with a precision in the range of a few 10 nm. Typically, up to several thousand image frames are recorded sequentially so that a sufficient number of fluorophores can be localized. The final high resolution image is reconstructed from all fluorophore locations, depicting the density of fluorophores within the specimen. (b) Conventional widefield image and (c) PALM image of a live SW13 cell expressing a desmin-mEosFPthermo fusion construct, scale bar, 1 μm

Full size image

3.2.2 Concepts of Localization-Based Super-Resolution Imaging

In 2006, three variants of localization-based super-resolution imaging were introduced independently, photoactivated localization microscopy (PALM) [61], FPALM [62] and STORM [52]. In general, synthetic dyes provide a higher localization precision because they are more photostable than fluorescent proteins. As an example, ∼6,000 detected photons were reported per molecule per switching cycle using the photoswitchable fluorophore pair Cy3-Cy5 [63, 64], whereas tdEosFP, arguably the brightest photoactivatable fluorescent protein for PALM imaging [47], emits ∼2,600 photons per molecule. Monomeric FPs often yield only a few hundred photons [50], which, however, still leads to a ten-fold increase in resolution. FPs are expressed by the cell itself and are, therefore, easy to use and ideally suited for in vivo experiments [38, 65]. Synthetic dyes require specific buffer conditions involving high concentrations of reductants for photoswitching and additional labeling procedures, and so they are predominantly used for studies of fixed specimens [63, 66]. Direct STORM (dSTORM) is a simplified version of STORM that works without a second dye molecule for activation [67], and PALM with independently running acquisition (PALMIRA) [68] is a simplified version of PALM that uses only a single laser line for readout as well as photoactivation. In blink microscopy (BM), the off- and on-times of single oxazine dyes are controlled using oxidizing and reducing agents for continuous switching, respectively [69]. The speed of PALM imaging was increased by introducing a simultaneous two-color stroboscopic illumination (S-PALM) for fast switching of variants of the fluorescent protein Dronpa [70]. The triplet state of dye molecules can also be employed as a dark state: In a method called ground state depletion followed by individual molecule return (GSDIM), dye molecules are shelved in the triplet state from which they return spontaneously [71]. In point accumulation for imaging in nanoscale topography (PAINT), fluorescent molecules switch to the on-state only upon binding to the structure to be imaged and are in their non-fluorescent off-state while diffusing freely [72].

Due to the popularity of localization-based super-resolution microscopy, various new developments have emerged in fluorescence marker design, especially in the field of photoactivatable fluorophores. Dual color PALM experiments using different emission spectra were performed [47, 64, 73–76]. Cellular dynamics has been studied on the nanoscale [65]; high resolution pulse-chase experiments have also been implemented [50]. A key advantage of localization-based super-resolution methods is that they do not require any specialized microscopy hardware. Equipped with suitable lasers for excitation and photoactivation of the fluorophores and a fast, sensitive CCD camera, any existing widefield microscope can be used with these techniques. Additionally, special image analysis software is required to localize individual fluorophores and to reconstruct the final, high-resolution images. Frequently, a TIRF microscope is preferred for its improved contrast in comparison to standard widefield microscopy as a result of its axial sectioning capability.

Localization microscopy has also been extended to the third dimension in various ways. In the astigmatism-based method, a cylindrical lens is placed into the detection path to distort the PSF, and the distance to the focal plane can be extracted from the shape of the distorted PSF [77]. Another way is to introduce a spatial light modulator (SLM) in the Fourier plane of the imaging system [78]. The light is modulated such that every object point is convolved with two double-helical lobes, with the angular orientation of the lobes depending on the axial location of the object above or below the focal plane. In biplane PALM (BP-PALM) two image planes of slightly different focus are captured simultaneously, and the z-position of a fluorophore between the two focal planes is computed by fitting a three-dimensional model of the PSF to the two images of the fluorophore [79]. Axial sectioning can also be achieved by using two-photon photoactivation [51] or interferometric designs (iPALM) [80]. In iPALM, as in 4Pi microscopy, two opposing, interferometrically coupled objectives collect photons emitted by the fluorophores. The distance of the emitter from the focal plane can be determined with a very high precision of 10–20 nm, because of photon self-interference for equal path lengths. As an additional benefit, the number of photons collected is twice as large, improving the lateral localization precision as well. The vertical composition of focal adhesions was studied with this technique in great detail [81]. Finally, yet another method involves positioning of a tilted mirror close to the sample, so that a side view is captured in addition to the front view, thereby yielding isotropic 3D resolution termed virtual volume super-resolution microscopy (VVSRM) [82].

Recent years have also witnessed enormous improvements in PALM image analysis. Reconstruction of excellent high resolution images by localization of single molecules requires sophisticated image analysis algorithms. Until very recently, image reconstruction took much longer than data acquisition because image analysis is computationally demanding and, therefore, researchers were only able to assess the quality of their data long after the actual measurement. By using a single step triangulation algorithm instead of an iterative fitting procedure, localization can be performed during data acquisition [83], yielding an instant high-resolution preview and making the data acquisition procedure more intuitive and interactive [84]. Modern graphics processing units (GPUs) have also been employed instead of conventional central processing units (CPUs) to parallelize single-molecule localization so that image reconstruction can be accelerated by orders of magnitude while maintaining a high localization precision [85, 86]. Further improvements will be necessary to cope with the data flow from faster cameras that will soon become available. Localization software has been made available as an ImageJ plugin for an out-of-the-box application [87], including even a 3D reconstruction algorithm. Beyond increasing camera frame rates [88], faster imaging can also be achieved by analyzing images with a higher density of fluorophores, which reduces the total number of frames necessary to reconstruct an image fulfilling the Nyquist theorem. However, the fraction of overlapping PSFs will be significant, and the analysis software needs to be capable of reliably decomposing these signals [89, 90].

3.3 Reversible Saturable Optical Fluorescence Transition-Based Super-Resolution Microscopy

The earliest super-resolution technique, stimulated emission depletion (STED) microscopy [91] or, more generally, RESOLFT microscopy [43, 46, 92] is based on a targeted, point-scanning approach in a raster-scanning confocal microscope. STED microscopy is closely related to saturated structured illumination microscopy (SSIM), in which regular illumination patterns are used instead of point-scanning [93, 94]. All these techniques utilize non-linear responses of the fluorophores to light irradiation, i.e., switching between dark and bright states.

In a STED microscope, stimulated emission is induced with a depletion beam that has an annular shape in the focal plane, with zero intensity in the center. By spatially overlaying the exciting focused spot of a confocal microscope with the depletion beam, fluorophores that do not reside close to the center are efficiently deexcited by stimulated emission of photons in the direction of the depletion beam and thus do not reach the detector. Spontaneous fluorescence photons are only emitted by fluorophores near the center. Consequently, application of the STED beam yields a smaller effective size of the excitation PSF than the usual diffraction-limited PSF (Fig. 4.4a). The higher the intensity of the depletion beam, the more the fluorescence is confined to the central region [95]. The achievable resolution is given by

$$ d = \frac{\lambda }{{2n\,\sin \,\alpha \sqrt {{1 + {{I}_{\text{STED}}}/{{I}_{\text{s}}}}} }}, $$
(4.7)

where λ is the wavelength, n the refractive index, α the half-angle under which the fluorescence is collected, and I S and I STED are the characteristic saturation intensity of the fluorophore and the intensity of the STED beam, respectively. No special switching capabilities of the fluorescence markers are needed since STED exploits stimulated emission as a general physical principle of light-matter interaction. However, the fluorophores are required to undergo a large number of excitation-depletion cycles while both excitation and STED beams are scanned across the sample. Thus, excellent photostability of the fluorophores is a prerequisite (Fig. 4.4b). Confocal and STED images of β-tubulin in HeLa cells labeled with primary and secondary antibodies are shown in Fig. 4.4c, d.

Fig. 4.4
figure 4

Reversible saturable optical fluorescence transition (RESOLFT) microscopy. (a) A fluorophore can be optically switched between a dark state and a bright state. An illumination pattern featuring at least one intensity-zero can be applied to reduce the effective size of the point spread function (PSF), for example, by stimulated emission depletion (STED). (b) Sketch of the scanning procedure. A donut-shaped depletion beam with zero intensity in the center is overlayed with the excitation beam. Thus, all markers further away from the center (red dots) are efficiently deexcited by stimulated emission before they fluoresce; only markers close to the center emit fluorescence (blue dots). (c) Confocal image of β-tubulin in fixed HeLa cells labeled with primary and secondary antibodies. (d) STED image of the same region. The cross section marked by blue arrows features a width of 58 nm, scale bar, 1 μm

Full size image

The use of continuous-wave lasers for stimulated depletion avoids the need for temporal synchronization of the excitation and the depletion beam that is required with pulsed laser sources [96, 97]. The rather complex design of a STED microscope can be drastically simplified with a combination of wave plates as beam shaping devices [98]. Thus the excitation and depletion beams are intrinsically overlayed and do not require any additional alignment.

To improve the axial as well as the lateral resolution, STED microscopy can be further combined with 4Pi microscopy [29]. Another way to enhance the resolution in axial direction is to shape the pattern imprinted on the wavefront of the depletion beam in a way that depletion also confines the emission along the z-axis, which avoids the considerable complexity introduced by using two objectives [99]. A variety of STED applications to biological specimens have been published. Live cell imaging was shown to be feasible using photostable variants of fluorescent proteins [100, 101]. Fast beam-scanning was applied to study the movement of synaptic vesicles in living neurons at video frame rates [102]. Two-color STED at a resolution down to 30 nm has also been employed to study synaptic proteins in neurons [103] as well as the distribution of proteins in the mitochondrial membrane [104]. Moreover, because STED microscopy is a confocal technique, fluorescence correlation spectroscopy (FCS) can also be performed with a significantly reduced observation volume [105]. STED-FCS was applied in the direct observation of the nanoscale dynamics of membrane lipids in the plasma membrane of a living cell [106]. Fluorescent nitrogen-vacancy centers in diamond were imaged with a resolution of 8 nm using STED microscopy, and these defect centers were even localized with sub-nanometer precision [107]. Yet, effective stimulated depletion requires high laser intensities at the focal spot (MW-GW/cm²) that may damage biological samples. The key problem is the short excited-state lifetime, during which the depletion has to occur. Mechanisms other than stimulated emission can be used to deplete the active state of the fluorescent markers around the center of the excitation beam more efficiently. Those include pumping of a triplet state [108, 109], which permits a reduction of the laser intensity to ∼1 kW/cm² or exploiting a photoswitching mechanism, e.g., cis-trans isomerization of fluorophores between bright and dark states [43, 110, 111], for which ∼1 W/cm² is sufficientf.

3.4 Saturated Structured Illumination Microscopy

Recently, the SIM method has also been extended to take advantage of a nonlinear fluorophore response to push the image resolution beyond the Abbe limit. Indeed, saturated structured illumination microscopy (SSIM), like STED and localization microscopy, provides theoretically unlimited resolution. A nonlinear response of the fluorescence signal to the excitation intensity is achieved by optically saturating the fluorescence markers in the sample. Consequently, the periodic excitation pattern created in the sample contains higher harmonics of its fundamental frequency, which is key to resolution enhancement beyond the diffraction limit [112]. For SSIM, a lateral resolution of ∼50 nm was demonstrated using fluorescent beads [113]. A problem of SSIM is the fast photobleaching that occurs under saturation conditions. However, similar to RESOLFT, saturation may also be achieved using a photoswitching mechanism, enabling application of this technique to live-cell imaging.

3.5 Super-Resolution Optical Fluctuation Imaging

SOFI relies on higher-order statistical analysis of temporal fluctuations, e.g., fluorescence blinking of quantum dots, to obtain subdiffraction optical resolution in all three dimensions [55]. The fluorescence signal of a distribution of markers can be described as

$$ F\left( {\vec{r},t} \right) = \sum\limits_k^N {U\left( {\vec{r} - {{{\vec{r}}}_k}} \right) \cdot {{\varepsilon }_k} \cdot {{s}_k}(t)}, $$
(4.8)

where \( U\left( {\vec{r} - {{{\vec{r}}}_k}} \right) \)is the PSF of marker k, ε k the molecular brightness and s k (t) the temporal fluctuation of the marker signal. A second-order autocorrelation can be applied to each pixel of a series of images resulting in

$$ {{G}_2}\left( {\vec{r},\tau } \right) = \sum\limits_k^N {{{U}^2}\left( {\vec{r} - {{{\vec{r}}}_k}} \right) \cdot \varepsilon_k^2 \cdot \left\langle {\delta {{s}_k}\left( {t + \tau } \right) \cdot \delta {{s}_k}(t)} \right\rangle } $$
(4.9)

The intensity value assigned to each pixel of the SOFI image is then given by the integral over the correlation function, yielding a smaller effective PSF. Figure 4.5 shows a conventional widefield image (a,b) of QD625 labeled 3T3 cells versus the second-order SOFI image (panels c, d). Using a camera pixel size well below the diffraction limit, the effective PSF can be further reduced by a factor √n using n-th order cumulants. However, we note that the signal in a SOFI image does not represent the emission intensity but its temporal fluctuations. As an additional benefit, background from non-fluctuating sources such as autofluorescence is efficiently removed.

Fig. 4.5
figure 5

Super-resolution optical fluctuation imaging (SOFI). The intensity value assigned to each pixel of the SOFI image is given by the integral over the autocorrelation function of fluorescence fluctuations from individual emitters. (a) Conventional widefield image of QD625 labeled 3 T3 cells generated by time averaging over all frames recorded, scale bar, 2 μm. (b) The image in panel a deconvolved. (c) Second-order SOFI image. (d) The image in panel c deconvolved. (e–h) Magnified views of the boxed regions, scale bars, 500 nm. The images were reprinted from [56] with kind permission

Full size image

4 Considerations and Limitations

It is evident that the higher the desired image resolution in microscopy the higher the demands on the fluorophores. Linear approaches such as 4Pi, I5M and SIM do not rely on special dye properties and are, therefore, not more demanding on the dyes than conventional fluorescence microscopy techniques. However, it should be evident that, to achieve a higher resolution with concomitantly reduced pixel/voxel size, more photons are required to reach a certain signal-to-noise level for mere statistical reasons.

4.1 Limitations of Localization-Based Techniques

For localization-based techniques, the precision of fluorophore localization and, consequently, the achievable resolution, depends on the number of photons emitted by the individual emitter [18]. A more subtle issue is the lifetime of the dark and bright states of the fluorescent markers employed. For best possible localization accuracy, the PSFs of the single emitters should not overlap. Therefore, the fluorophores must be kept in their inactive, dark state considerably longer than in their active, bright state [114]. While this condition can easily be satisfied with photoconverting fluorophores, care has to be taken when using photoswitchers. Another important criterion is the dynamic range of the photoactivation mechanism, i.e., the intensity ratio between bright and dark statesf. To achieve a reasonable contrast for single molecule localization, the probability that a fluorophore emits a photon while in its inactive state, which is ideally completely non-fluorescent, must be sufficiently low.

4.2 Limitations in Resolft Microscopy

In RESOLFT microscopy, contrary to localization microscopy, the brightness of the individual molecule is less important. In principle, a single detected photon per molecule is enough for image formation as long as the density of emitters is sufficient to produce a strong signal above background. However, for localization-based approaches, only a single switching cycle per molecule is required, whereas in RESOLFT microscopy, the resolution is determined by the number of switching cycles that a molecule can undergo. With decreasing pixel size, the fluorescence markers have to undergo increasingly more transitions between the fluorescent and non-fluorescent states before contributing appreciably to the image. Therefore, they have to be exquisitely resistant to switching fatigue. Additionally, in STED microscopy, the high intensity of the depletion beam (>1 GW/cm²) may cause photodamage to living specimens [115].

4.3 General Limitations

A general requirement of all types of fluorescence microscopy, regardless of a particular type, is that the labeling density has to be at least twice as large as the desired resolution [116]. Therefore, the ultimate resolution limitation in super-resolution optical microscopy is posed by the ability to precisely label the structure of interest. Evidently, each fluorescent marker has a certain size and cannot reside where the structure itself is located. Frequently, linkers of considerable lengths connect the fluorophores specifically to a target, which further adds to uncertainties in marking the structure. In live-cell imaging, too dense labeling may interfere with the processes under study. Conditions are even more demanding when imaging dynamic processes. In localization-based techniques, a single high-resolution image is finally reconstructed from a large number of individual frames, each with a dwell time of 1–100 ms using current CCD camera technology. Thus, acquisition of super-resolved images presently takes a few seconds at the very minimum, and processes on the sub-second timescale are too fast to be studied. STED microscopy of live cells at video rate has been demonstrated, yet it is confined to micron-sized regions of interest [102].

5 Conclusions

A multitude of super-resolution light microscopy techniques have emerged in recent years that are likely to be extremely helpful for unraveling the molecular details underlying biological processes in cells and tissues. These techniques have been designed to address specific problems in specific ways. For example, they may provide multi-color, 3D, or video-rate imaging, but not all of these features will be available simultaneously in one and the same microscope. There are strong efforts toward simplified designs that hopefully will allow for a widespread application of these sophisticated techniques even by non-specialists. Apart from stable and easy-to-operate microscope designs, efficient and easy-to-handle labeling systems, featuring brighter, more photostable and smaller fluorescence markers are of crucial importance. The first commercial implementations of STED, SIM and PALM microscopes have become available, and exciting further developments are likely to soon appear on the horizon.