|Home | About | Journals | Submit | Contact Us | Français|
Achieving a spatial resolution that is not limited by the diffraction of light, recent developments of super-resolution fluorescence microscopy techniques allow the observation of many biological structures not resolvable in conventional fluorescence microscopy. New advances in these techniques now give them the ability to image three-dimensional (3D) structures, measure interactions by multicolor colocalization, and record dynamic processes in living cells at the nanometer scale. It is anticipated that super-resolution fluorescence microscopy will become a widely used tool for cell and tissue imaging to provide previously unobserved details of biological structures and processes.
Much of our knowledge regarding biological processes at the cellular and subcellular level has come from the ability to directly visualize them. Among the various microscopy techniques, fluorescence microscopy is one of the most widely used because of its two principal advantages: Specific cellular components may be observed through molecule-specific labeling, and light microscopy allows the observation of structures inside a live sample in real time. Compared to other imaging techniques such as electron microscopy (EM), however, conventional fluorescence microscopy is limited by relatively low spatial resolution because of the diffraction of light. This diffraction limit, about 200–300 nm in the lateral direction and 500–700 nm in the axial direction, is comparable to or larger than many subcellular structures, leaving them too small to be observed in detail. In recent years, a number of “super-resolution” fluorescence microscopy techniques have been invented to overcome the diffraction barrier, including techniques that employ nonlinear effects to sharpen the point-spread function of the microscope, such as stimulated emission depletion (STED) microscopy (1, 2), related methods using other reversible saturable optically linear fluorescence transitions (RESOLFTs) (3), and saturated structured-illumination microscopy (SSIM) (4), as well as techniques that are based on the localization of individual fluorescent molecules, such as stochastic optical reconstruction microscopy (STORM) (5), photoactivated localization microscopy (PALM) (6), and fluorescence photoactivation localization microscopy (FPALM) (7). These methods have yielded an order of magnitude improvement in spatial resolution in all three dimensions over conventional light microscopy. The observation of previously unresolved details of cellular structures has demonstrated the great promise of super-resolution fluorescence microscopy in elucidating biological processes at the cellular and molecular scale.
In this review, we summarize the current state of development and future challenges of super-resolution fluorescence microscopy techniques, as well as their application to the study of biological problems.
Microscopes can be used to visualize fine structures in a sample by providing a magnified image. However, even an arbitrarily high magnification does not translate into the ability to see infinitely small details. Instead, the resolution of light microscopy is limited because light is a wave and is subject to diffraction.
An optical microscope can be thought of as a lens system that produces a magnified image of a small object. In this imaging process, light rays from each point on the object converge to a single point at the image plane. However, the diffraction of light prevents exact convergence of the rays, causing a sharp point on the object to blur into a finite-sized spot in the image. The three-dimensional (3D) intensity distribution of the image of a point object is called the point spread function (PSF). The size of the PSF determines the resolution of the microscope: Two points closer than the full width at half-maximum (FWHM) of the PSF will be difficult to resolve because their images overlap substantially.
The FWHM of the PSF in the lateral directions (the x–y directions perpendicular to the optical axis) can be approximated as Δxy ≈ 0.61λ / NA, where λ is the wavelength of the light, and NA is the numerical aperture of the objective defined as NA = n sinα, with n being the refractive index of the medium and α being the half-cone angle of the focused light produced by the objective. The axial width of the PSF is about 2–3 times as large as the lateral width for ordinary high NA objectives. When imaging with visible light (λ ≈ 550 nm), the commonly used oil immersion objective with NA = 1.40 yields a PSF with a lateral size of ~200 nm and an axial size of ~500 nm in a refractive index-matched medium (Figure 1) (8).
The diffraction limit of resolution in light microscopy does not affect most imaging at the organ or tissue level. However, when zooming into cells, where a large number of subcellular structures are smaller than the wavelength of the light, it becomes an obstacle for studying these structures in detail. It is, therefore, important to develop techniques that improve the spatial resolution of light microscopy without compromising its noninvasiveness and biomolecular specificity.
Because the loss of high-frequency spatial information in optical microscopy results from the diffraction of light when it propagates through a distance larger than the wavelength of the light (far field), near-field microscopy is one of the earliest approaches sought to achieve high spatial resolution. By exciting the fluorophores or detecting the signal through the nonpropagating light near the fluorophore, high-resolution information be retained. Near-field scanning optical microscopy (NSOM) acquires an image by scanning a sharp probe tip across the sample, typically providing a resolution of 20–50 nm (9–11). Wide-field imaging has also been recently demonstrated in the near-field regime using a super lens with negative refractive index (12, 13). However, the short range of the near-field region (tens of nanometers) compromises the ability of light microscopy to look into a sample, limiting the application of near-field microscopy to near-surface features only. This limit highlights the need to develop far-field high-resolution imaging methods.
Several far-field imaging methods are known to extend the diffraction-limited resolution to smaller values.
Among far-field fluorescence microscopy techniques, confocal and multiphoton microscopy are among the most widely used to moderately enhance the spatial resolution (14, 15). By combining a focused laser for excitation and a pinhole for detection, confocal microscopy can, in principle, have a factor of improvement in the spatial resolution. In multiphoton microscopy, nonlinear absorption processes reduce the effective size of the excitation PSF. However, this gain in the PSF size is counteracted by the increased wavelength of the excitation light. Thus, instead of improving the resolution, the main advantage of confocal and multiphoton microscopy over wide-field microscopy is the reduction of out-of-focus fluorescence background, allowing optical sectioning in 3D imaging.
Because microscope objectives collect light from only one side of the sample, a large amount of light is lost, leading to a PSF that is elongated in the axial direction. One can imagine that if the full spherical angle could be covered by the objective, the PSF would become symmetric with the axial size being the same as the lateral size. Two techniques, 4Pi and I5M microscopy, approach this ideal situation by using two opposing objectives for excitation and/or detection (16, 17). This effective increase in NA boosts the lateral resolution by a modest amount, but substantially improve the axial resolution to ~100 nm.
Another approach to increase the spatial resolution of optical microscopy is to apply a patterned illumination field to the sample. In this approach, the spatial frequencies of the illumination pattern mix with those of the sample features, shifting the high-frequency features to lower frequencies that are detectable by the microscope. Periodic illumination patterns can be created through the interference of multiple light sources in the axial direction (18), the lateral direction (19), or both (20). By acquiring multiple images with illumination patterns of different phases and orientations, a high-resolution image can be reconstructed. Because the illumination pattern itself is also limited by the diffraction of light, structured illumination microscopy (SIM) is only capable of doubling the spatial resolution by combining two diffraction-limited sources of information. A resolution of~100 nm in the lateral direction and ~300 nm in the axial direction has been achieved (19–21).
The advantage of these purely optical approaches is that they are not dependent on any special photophysics or photochemistry of the fluorophore; therefore, virtually any fluorescent probe can be used. The best achievable result using these methods would be an isotropic PSF with an additional factor of 2 in resolution improvement. This would correspond to ~100-nm image resolution in all three dimensions, as has been demonstrated by the I5S technique, which combines I5M and SIM (22). Albeit a significant improvement, this resolution is still fundamentally limited by the diffraction of light. To break the diffraction limit and obtain much higher image resolution would need additional factors, in particular the involvement of fluorophore photophysics or photochemistry.
One approach to attain a resolution far beyond the limit of diffraction, i.e., to realize super-resolution microscopy, is to introduce sub-diffraction-limit features in the excitation pattern so that small-length-scale information can be read out. We refer to this approach, including STED, RESOLFT, and SSIM, as super-resolution microscopy by spatially patterned excitation or the “patterned excitation” approach.
The concept of STED microscopy was first proposed in 1994 (1) and subsequently demonstrated experimentally (2). Simply speaking, it uses a second laser (STED laser) to suppress the fluorescence emission from the fluorophores located off the center of the excitation. This suppression is achieved through stimulated emission: When an excited-state fluorophores encounters a photon that matches the energy difference between the excited and the ground state, it can be brought back to the ground state through stimulated emission before spontaneous fluorescence emission occurs. This process effectively depletes excited-state fluorophores capable of fluorescence emission (Figure 2a,b).
In order to use STED to sharpen the excitation PSF, the STED laser needs to have a pattern with zero intensity at the center of the excitation laser focus and nonzero intensity at the periphery. However, this spatial pattern is also limited by the diffraction of light. Therefore, the effect of STED alone is not sufficient for sub-diffraction-limit imaging. The key to achieving super resolution is the nonlinear dependence of the depleted population on the STED laser intensity when the saturated depletion level is approached: If the local intensity of the STED laser is higher than a certain level, essentially all spontaneous fluorescence emission is suppressed. By raising the STED laser power, the saturated depletion region expands without strongly affecting fluorescence emission at the focal point because the STED laser intensity is nearly zero at this point. Consequently, the fluorescence signal can be observed only in a small region around the focal point, reducing the effective width of the PSF (Figure 2c). The size of this region is limited by the practical power level of the STED laser instead of the diffraction of light. Super-resolution images are then obtained by scanning this small effective PSF.
The pattern of the STED laser is typically generated by inserting a phase mask into the light path to modulate its phase-spatial distribution (Figure 2b). One such phase mask generates a donut-shaped STED pattern in the xy plane (Figure 2c) and has provided an xy resolution of ~30 nm (24). STED can also be employed in 4Pi microscopy (STED-4Pi), resulting in an axial resolution of 30–40 nm (25). STED has been applied to biological samples either immunostained with fluorophorelabeled antibodies (26) or genetically tagged with fluorescent proteins (FPs) (27). Dyes with high photostability under STED conditions and large stimulated emission cross sections in the visible to near infrared (IR) range are preferred. Atto 532 and Atto 647N are among the most often used dyes for STED microscopy.
Stimulated emission is not the only mechanism capable of suppressing undesired fluorescence emission. A more general scheme using saturable depletion to achieve super resolution has been formalized with the name RESOLFT microscopy (3). This scheme employs fluorescent probes that can be reversibly photoswitched between a fluorescent on state and a dark off state. The off state can be the ground state of a fluorophores as in the case of STED, the triplet state as in ground-state-depletion microscopy (28, 29), or the dark state of a reversibly photoswitchable fluorophore (30). The gain in spatial resolution is achieved in the same way as in STED, which uses a depletion laser to drive fluorophores at the periphery of the excitation into a dark state. With saturated depletion, the size of the effective PSF, Δeff, becomes:
where Δ is the diffraction limited size of the PSF, I is the peak intensity of the depletion laser and IS is the saturation intensity for the fluorophore. Super resolution can be achieved when the degree of saturation, I / IS, becomes much larger than unity. Unlike STED which features a high IS value (~107 W/cm2) and thus requiring an intense depletion laser (often > 109 W/cm2), an optical transition that has low IS can be chosen for RESOLFT, allowing super-resolution imaging at a much lower depletion laser intensity (3). For example, RESOLFT has been demonstrated using a reversibly photoswitchable fluorescent protein asFP595 which leads to a resolution better than 100 nm at a depletion laser intensity of 600 W/cm2(30).
The same concept of employing saturable processes can also be applied to SIM by introducing sub-diffraction-limit spatial features into the excitation pattern. SSIM has been demonstrated using the saturation of fluorescence emission, which occurs when a fluorophore is illuminated by a very high intensity of excitation light (4). Under this strong excitation, it is immediately pumped to the excited state each time it returns to the ground state, and the fluorescence lifetime becomes the limiting factor for the fluorescent emission rate. Consequently, as the fluorescence intensity of the fluorophore approaches a saturation level, it is no longer proportional to the excitation light intensity. In SSIM, where the sample is illuminated with a sinusoidal pattern of strong excitation light, the peaks of the excitation pattern can be clipped by fluorescence saturation and become flat, whereas fluorescence emission is still absent from the zero points in the valleys (Figure 3a). These effects add higher order spatial frequencies to the excitation pattern. Mixing this excitation pattern with the high-frequency spatial features in the sample can effectively bring the sub-diffraction-limit spatial features into the detection range of the microscopy (Figure 3b). A super-resolution image can then be reconstructed in a similar manner as in SIM. Like STED and RESOLFT, the spatial resolution of SSIM is no longer fundamentally limited by diffraction, but rather by the level of fluorescence saturation than can be practically reached. SSIM has demonstrated a 50-nm resolution in two dimensions (4).
A biological structure is ultimately defined by the positions of the molecules that build up the structure. Similarly, a fluorescence image is defined by the spatial coordinates of the fluorophores that generate the image. It is thus conceivable that super-resolution fluorescence microscopy can also be achieved by determining the position of each fluorescent probe in a sample with high precision.
Although the image of a single fluorophore, which resembles the PSF, is a finite-sized spot, the precision of determining the fluorophores position from its image can be much higher than the diffraction limit, as long as the image results from multiple photons emitted from the fluorophore. Fitting an image consisting of N photons can be viewed as N measurements of the fluorophore position, each with an uncertainty determined by the PSF (8), thus leading to a localization precision approximated by:
where Δloc is the localization precision and Δ is the size of the PSF. This scaling of the localization precision with the photon number allows super-resolution microscopy with a resolution not limited by the diffraction of light.
High-precision localization of bright light sources, such as fluorescent particles, is often used in biophysical experiments (31, 32) and has reached a precision as high as ~1 Å (33). Taking advantage of single-molecule detection and imaging (34, 35), nanometer localization precision has also been achieved for single fluorescent molecules under ambient conditions (36). When multiple molecules are present in close proximity, however, localization becomes inaccurate or impossible because the images of these fluorophores overlap. Separation of the fluorescence signal from a few molecules with overlapping images may be achieved in the spectral domain according to their distinct excitation or emission spectra (37–39) or in the time domain, utilizing the fact that different emitters independently undergo photobleaching (40, 41) or blinking (42, 43). It is, however, difficult to scale up the number of fluorophores that can be resolved within a diffraction-limited area using these methods.
A typical fluorescently labeled biological sample, however, contains thousands or even millions of fluorophores at a high density, making them difficult to resolve by the single-molecule localization approach. Using fluorescent probes that can switch between a fluorescent and a dark state, a recent invention overcomes this barrier by separating in the time domain the otherwise spatially overlapping fluorescent images. In this approach, molecules within a diffraction limited region can be activated at different time points so that they can be individually imaged, localized, and subsequently deactivated (Figure 4). Massively parallel localization is achieved through wide-field imaging, so that the coordinates of many fluorophores can be mapped and a super-resolution images subsequently reconstructed. This concept has been independently conceived and implemented by three labs, and it was given the names STORM (5), PALM (6), and FPALM (7), respectively. All three labs initially used photoswitchable fluorescent dyes or proteins that can be activated by light at a wavelength different from the imaging light that excites fluorescence and deactivates the fluorophores. Separating activation and imaging allows one to control the fraction of molecules in the fluorescent state at a given time, such that the activated molecules are optically resolvable from each other and precisely localized. Iterating the activation and imaging process then allows the locationsof many fluorophores to be mapped and a super-resolution image to be constructed from these fluorophore locations. In the following, we refer to this approach as super-resolution microscopy by single-molecule localization. Other variants of this approach have been subsequently reported using asynchronous activation and deactivation of fluorophores (44) or nonoptical processes, such as the stochastic binding of diffusing fluorophores (45).
The photophysical and photochemical properties of fluorescent probes are critical for super-resolution imaging by single-molecule localization. A large number of photoswitchable fluorophores have so far been reported for this application (Table 1). These probes range from organic dyes to FPs, allowing the labeling of biological samples with a variety of methods. The resolution of this technique is limited by the number of photons detected per photoactivation event, which varies from several hundred for FPs (6) to several thousand for cyanine dyes such as Cy5 (5, 46). These numbers theoretically allow more than an order of magnitude improvement in spatial resolution according to the √N scaling rule. In practice, a lateral resolution of ~20 nm has been established experimentally using the photoswitchable cyanine dyes (5, 46). Super-resolution images of biological samples have been reported with directly labeled DNA structures and immunostained DNA-protein complexes in vitro (5) as well as with FPtagged or immunostained cellular structures (6, 44, 46).
Recent advances in super-resolution fluorescence microscopy (including the capability for 3D, multicolor, live-cell imaging) enable new applications in biological samples. These technical advances were made possible through the development of both imaging optics and fluorescent probes.
Most cellular structures are 3D in nature, and the ability to resolve a 3Dstructure is in general limited by the dimension with the lowest resolution. Therefore, it is important to develop techniques that reach super resolution in all three dimensions.
Extending the STORM/PALM/FPALM approach to 3D imaging requires determining the 3D position of the activated fluorescent probes. A number of 3D localization methods previously developed for particle tracking (47–51) have been adapted for this purpose.
Super-resolution imaging by single molecule localization was first implemented in three dimensions using the astigmatism method for axial localization (52). This approach inserts a cylindrical lens in the imaging optics so that the image is focused differently in the x and y directions. Consequently, the image of a single fluorophore becomes elliptical, making it possible to simultaneously measure its axial position from the ellipticity and the lateral position from the centroid of the image. This method has achieved a lateral localization precision of ~25 nm (FWHM) and an axial localization precision of ~50 nm (FWHM) over the depth of several hundred nanometers in z with photoswitchable cyanine dyes (52). Combining with z-scanning allowed whole mammalian cells (several micrometers thick) to be imaged (53). The spatial organization of microtubules in cells, the semispherical shell structure of clathrin-coated pits, with a diameter of about ~150 nm, and the outer membrane structure of mitochondria has been resolved using this technique (52, 53) For thick sample imaging, spherical abberations caused by the refractive index mismatch between the sample and the imaging optics also need to be minimized or corrected (53).
Another implementation of 3D superresolution imaging (54) uses two-focal-plane imaging, in which the emission light is split and imaged to two regions of the camera with different path lengths, allowing the determination of z coordinates from the defocused shape of the single-molecule images. This method has achieved 75-nm axial localization precision with caged fluorescein over a depth of several hundred nanometers. Similarly z-scanning further enables the imaging of samples with thicknesses of several micrometers, and 4-µm microspheres were imaged as a demonstration of the technique.
As discussed above, STED can operate either in the xy mode to improve lateral resolution by applying a donut-shaped depletion pattern in the x-y plane or in the z mode to improve the axial resolution by using a depletion pattern with two maxima along the z axis. STED can also be employed in 4Pi microscopy (STED-4Pi) to achieve super resolution in z. 3D super-resolution imaging requires a combination of these schemes. For example, super resolution has recently been obtained simultaneously in all three dimensions by applying STED in both xy and z mode (55) or by combining xy STED with STED-4Pi (56). The major obstacle for combining two STED patterns is the interference between the two STED lasers, which can be eliminated either by introducing a time delay between the two STED pulses (55) or by using two independent lasers as the source (56). An isotropic spatial resolution of 40–45 nm has been demonstrated, allowing the observation of the hollow shape of the mitochondrial outer membrane, which has a diameter of 300–500 nm (56).
SSIM has only been demonstrated in two dimensions to date, while the nonsaturated SIM has provided ~100-nm lateral and ~300-nm axial resolution, allowing chromosomes and nuclear envelope structures to be better resolved compared to confocal microscopy (21). It is only natural to expect that SSIM can also be realized for 3D super-resolution imaging by using a structured excitation pattern at the saturation level along both lateral and axial directions.
Resolving a biological structure is important for understanding its functionality, but it is often not sufficient because many biological processes involve interactions between different structures or biomolecules. These interactions are often addressed using multicolor microscopy, wherein different components are labeled with distinctly colored probes. The colocalization between different colors in the image is used as the indicator of possible interactions, but the accuracy of colocalization is inherently limited by the resolution of the imaging method. Multicolor super-resolution imaging thus promises to significantly advance our understanding of molecular interactions in cells.
Multicolor STED can be performed using fluorophores with different excitation and emission wavelengths. Two-color STED has been realized using two pairs of excitation and STED lasers spanning the blue to NIR range (e.g., excitation at 488 nm and STED at 603 nm for Atto 532; excitation at 635 nm and STED at 750–780 nm for Atto 647N) (57). Another approach is to use dyes with extremely large Stokes shifts, so that two fluorophores can share the same STED laser but can be distinguished according to their different excitation wavelengths. As an example, DY-485XL excited at 488 nm and NK51 excited at 532 nm, both with emissions in the 570–620-nm range and compatible with a 647-nm STED laser, were used for imaging mitochondria outer membrane and matrix in three dimensions (56).
The key for multicolor imaging using single-molecule localization is to determine the identity of each activated fluorescent probe. In addition to the excitation and emission wavelength, the activation wavelength can also be used as a fluorophore signature, adding more versatility. The prerequisite for multicolor imaging, therefore, is the availability of photoswitchable probes with distinct excitation, emission, or activation wavelengths. A palette of such probes have been created based on photoswitchable cyanine dyes (46, 53, 58). Each of these probes consists of a reporter fluorophore, which can switch to a dark state when excited by the imaging light, and an activator fluorophore, which facilitates the activation of the reporter when illuminated by light that matches the activator absorption. Far red or near IR cyanine dyes, such as Cy5, Cy5.5, and Cy7, can function as the photoswitchable reporters, whereas the activator has a wide range of choices among violet- to yellow-absorbing dyes, such as Alexa Fluor 405, Cy2, and Cy3. Combinatorial pairing of the activators and reporters creates many photoswitchable probes that can be distinguished according to their activation or emission wavelengths. Three-color imaging of immobilized DNA molecules, two-color imaging of microtubules and clathrin-coated-pits in cells, and two-color 3D imaging of microtubules and mitochondria have been demonstrated using these probes by different activation wavelengths (46, 53). The Cy5, Cy5.5, and Cy7 dyes can also be spontaneously activated and deactivated by the same imaging laser (23) (for example 647 nm) without the need of activators but emit light of different wavelenths ranging from red to near IR, providing the possibility of multicolor imaging by distinguishing emission color only. More recently, another series of four photochromic rhodamine dyes have also been synthesized, which have the same activation wavelength but different emission wavelengths ranging from green to red. Three-color imaging of fluorescent microspheres and two-color imaging of microtubules and keratin intermediate filaments have been implemented with these probes (59).
Super-resolution imaging using FPs for multicolor imaging is more difficult because most photoswitchable red-emitting FPs have preactivation fluorescence emission that overlaps with the postactivation fluorescence emission of green-emitting FPs. To circumvent the problem, two-color imaging was initially obtained by using a reversibly switching green FP and a red FP that can only be activated once. In this scheme, the green FPs were only imaged after all red FPs were imaged and photobleached (60). In practice, depleting one probe before imaging the other, however, does not allow multicolor imaging of the same region at multiple time points and therefore is not compatible with recording dynamic processes. This problem was overcome by a recently developed Dronpa variant, bsDronpa, with a blue-shifted emission (61), and the dark-to-red switching PAmCherry, which is nonfluorescent prior to photoactivation (62). Pairing bsDronpa with the green-emitting FP Dronpa, or PAmCherry with other FPs having shorter emission wavelengths, allows simultaneous two-color imaging according to their different emission wavelengths, although the the relatively low number of photons emitted by existing short-wavelength FPs tends to limit image resolution severely.
The ability to image a live biological sample is one of the advantages of light microscopy. Fluorescence imaging of a live cell has two requirements: specific labeling of the cell and a time resolution that is high enough to record relevant dynamics in the cell. The super-resolution imaging techniques demonstrated to date, including the patterned excitation and single molecule localization approaches, all have a relatively slow imaging speed. The latter approach requires the introduction of photoswitchable probes into live cells, which adds an additional challenge. Nonetheless, many fluorescent proteins and organic dyes, including cyanine dyes (46) and caged dyes, have been shown switchable in live cells. In this section, we describe the recently demonstrated examples of live-cell super-resolution imaging but defer to the next section a discussion of the factors limiting time resolution and the methods developed for live cell labeling.
Because STED has a much smaller PSF than scanning confocal microscopy, STED would inherently take more time to scan though the same size of image field. By increasing the scanning speed and limiting the field of view to a few µm, Westphal and coworkers have observed Brownian motion of a dense suspension of nanoparticles with an impressive rate of 80 frames per second (fps) using STED microscopy (63). More recently, they have demonstrated video-rate (28 fps) imaging of live hippocampal neurons and observed the movement of individual synaptic vesicles with 60–80-nm resolution (64).
Super-resolution imaging by single-molecule localization is also relatively slow because the final image is in fact a map of fluorophore localizations accumulated over many activation cycles. Sub-diffraction-limit imaging of focal adhesion proteins in live cells has recently been demonstrated (65). Photoswitchable fluorescent protein, EosFP, was used to label the focal adhesion protein paxillin. A time resolution of ~25–60 seconds per frame was obtained, and during this time interval, approximately 103 fluorophores were activated and localized per square micrometer, providing an effective resolution of 60–70 nm by the Nyquist criterion (65). More recently, super-resolution imaging has also been demonstrated in live bacteria with photoswitchable enhanced yellow fluorescent protein (EYFP), allowing the MreB structure in the cell to be traced (66).
Photoswitchable fluorescent probes can also greatly facilitate particle tracking measurements. Contrary to conventional single-particle tracking experiments, in which only a small fraction of the targets are labeled and tracked, the use of photoswitchable probes allows a high density of target molecules to be labeled. Although only an optically resolvable subset is activated and tracked at a given time, accumulation over many activation cycles allows a high density map of particle tracks to be constructed. This approach has been used to test the existence of membrane microdomains by observing the diffusion and aggregation of hemagglutinin on the plasma membrane (67) and to monitor the spatial map of the diffusion properties of membrane proteins (68).
Microscopy techniques provide the ability to observe both the spatial structure of a specimen and how the specimen changes in time. The spatial and temporal resolution of the technique determine to what degree the fine details of the structure can be observed and how quickly a dynamic process can be followed. This section discusses the limits of resolvability in these aspects of super-resolution fluorescence microscopy and potential ways to overcome these limits.
Optical resolution is the intrinsic ability of a given method to resolve a structure and can be defined as the ability to distinguish two point sources in proximity. For the patterned excitation approaches, such as STED, SSIM, and RESOLFT, the optical resolution is represented by the size of the effective PSF. For the single-molecule localization approach, such as STORM/PALM/FPALM, the precision of determining the positions of individual fluorescent probes is the principal measure of optical resolution.
By using a spatially patterned excitation profile, this approach achieve super resolution by generating an effective excitation volume with dimensions far below the diffraction limit. Taking STED as an example, the sharpness of the PSF results from the saturation of depletion of excited-state fluorophores in the region neighboring the zero point of the STED laser (which coincide with the focal point of the excitation laser). With an increasing STED laser power, the saturated region expands toward the zero point, but fluorophores at the zero point are not affected by the STED laser if the zero point is strictly kept at zero intensity. Therefore, a theoretically unlimited gain in spatial resolution may be achieved if the zero point in the depletion pattern is ideal.
In practice, however, nonideal zero points (with finite intensity) can result from many factors, limiting the maximum achievable resolution. For example, a zero point with an intensity that is 1% of the peak intensity would only lead to a factor of ~7 improvement in optical resolution compared to confocal microscopy (according to simulations), whereas a zero point intensity that is 0.1% of the peak intensity could provide a ~20-fold increase in resolution. The zero point intensity of the structured illumination profile has a similar importance in the resolution achievable by SSIM.
Factors leading to nonideal zero points include imperfections in the laser mode, phase plate (for STED) or diffractive grating (for SSIM), and microscope objective, as well as spherical aberrations caused by refractive index mismatch between the coverglass and the medium in which the sample is mounted. In tissue sections where the scattering of light is strong, the gain in optical resolution can be further limited.
Another practical factor that limits resolution is the sample damage by the high laser intensity used for saturated depletion or excitation. The more general approach of RESOLFT microscopy utilizes a light-driven transition between fluorescent and dark states of switchable fluorophores other than stimulated emission to deplete fluorescence emission at the periphery of the zero point. Although it is possible to achieve super resolution at a much lower depletion power with low-saturation-level transitions, the resolution of this technique is more often limited by other factors, including the inability of the depletion laser to completely suppress fluorescence and the number of on-off transition cycles that a fluorescent probe can undergo.
The single-molecule localization approach achieves super resolution through high precision localization of individual fluorophores. The number of photons collected from a fluorophore is a principal factor limiting the localization precision and hence the resolution of the final image.
A rigorous way to characterize single molecule localization precision is through repetitive localization of a photoswitchable fluorophore. Alternatively, the localization precision can be obtained from the distribution of measured positions of multiple probes on a well-defined structure, e.g., a flat surface for quantifying the localization precision in z. In the example of the photoswitchable cyanine dyes, the localization precision has been characterized experimentally to ~20 nm (FWHM) in the lateral direction (5, 46). By contrast, a theoretical resolution of 6 nm (FWHM), or a corresponding standard deviation of 3 nm, would be expected based on 6000 photons detected per activation cycle from these dyes under the specified imaging geometries (8, 46). Possible causes for this discrepancy include the facts that the PSF of a fluorophore is not a perfect Gaussian function (L.S. Churchman & J.A. Spudich, personal communication); the image of a nonfreely rotating fluorophore is not symmetric owing to the dipole emission pattern (69); the camera pixels differ slightly in size, quantum efficiency, and gain (S. Chu, personal communication); and residual mechanical instability of the instrument remains after drift correction (5). One must, therefore, be cautious when using the number of detected photons to infer the optical resolution of a particular experiment.
Several photoswitchable fluorophores have been reported to give thousands of photons detected per activation event [e.g., 6000 from Cy5 (46)].With the PSF fitting procedure and the mechanical stability of the system optimized, the background signal suppressed, and the nonuniformity of camera pixels corrected, optical resolution of just a few nanometers could potentially be achieved, reaching the molecular scale. As in the case of the patterned excitation approach, the optical resolution here is also unlimited, in principle, given a sufficient number of photons detected from the fluorescent probes.
The optical resolution, as discussed above, defines the physical limit of the smallest structure it can resolve. When imaging a biological sample, the effective resolution is also affected by several sample-specific factors, including the labeling density, probe size, and how well the ultrastructures are preserved during sample preparation.
In super-resolution fluorescence microscopy, because the optical resolution often approaches the distance between adjacent fluorescent probes in a sample, the labeling density could become a limiting factor of the effective spatial resolution. This limitation applies to all super-resolution fluorescence microscopy methods, including STED, RESOLFT, SSIM, as well as STORM/PALM/FPALM. When the labeling efficiency (fraction of target molecules labeled) of a sample is not sufficient, artifacts can be observed in super-resolution images, for example, causing continuous structures to appear discontinuous.
The effect of labeling density on the effective resolution can be quantified by the Nyquist criterion, which states that structural features smaller than twice that of the fluorophore-to-fluorophore distance cannot be reliably discerned (65, 70). The smallest resolvable feature size thus follows:
where N is density of labeling, and D is the dimension of the structure to be imaged. According to this formula, to achieve a 20-nm 2D (or 3D) resolution, an extremely high labeling density of 104 per µm2 (or 106 per µm3) is required in general. In practice, however, a lower requirement on the labeling density is often sufficient when the geometry of the sample is taken into consideration. Biological structures are often so heterogeneous that a high local density can result from a relatively low overall density. For example, when taking a 2D image of clathrin-coated pits, which are~150-nm round structures distributed on the plasma membrane with a surface density of about 0.5 pits per µm2, an overall labeling density of 90 per µm2 would already correspond to a local density of 104 per µm2 within individual pits, which is sufficient to obtain a 20-nm lateral resolution.
For the single-molecule localization approach, too high labeling density can also be disadvantageous because the majority of photoswitchable fluorophores are not completely dark in their dark state. They either emit weak fluorescence or can spontaneously switch to the fluorescent state in the absence of activation light. This type of nonspecific activation can either be spontaneous or induced by the imaging laser. Therefore, a large number of nonspecifically activated fluorophores associated with very high labeling density could cause the overlap of single-molecule images, which prevents high-precision localization. In other words, photoswitchable fluorophores with low dark-state emissions and low nonspecific activation rates are desired for ultrahigh-resolution imaging.
The size of the probe fundamentally limits on how well the image reflects the actual structure. For example, with indirect immunofluorescence, the binding of the primary and secondary antibodies increases the apparent diameter of microtubules from 25 nm to about 60 nm, which has been observed in both EM (71) and STORM (46). Smaller probes include primary antibodies for direct immunofluorescence (~10–15 nm in size) (52), antibody Fab fragments (~6 nm in size), and FPs (~3–4 nm in size). The ultimate solution of the probe size problem is the direct labeling of target molecules with organic fluorophores (~1 nm).
Immunostaining with dye-labeled antibodies is often the method of choice for tissue imaging because genetic engineering of an animal with FP labels could be time consuming or even impossible (for example, human subjects). Being brighter and more photostable, organic dyes also tend to have more ideal photophysical properties for super-resolution imaging than FPs. For live-cell imaging, a FP is often a more convenient choice because it can be genetically targeted to the molecule of interest inside the cell. The sizes of FPs are also smaller than antibodies. To combine the benefits from both sides, several methods have been recently developed to directly couple an organic dye molecule to a specific protein through genetic coding (72, 73). These methods fuse the protein of interest to a specific peptide sequence, which either exhibits direct affinity to certain fluorescence dyes [such as the tetracysteine tag binding to biarsenic fluorophores (74) and the polyhistindine tag binding to the NTA group (75)] or can be covalently linked to dye molecules through enzymatic reactions [such as those catalyzed by biotin ligase (76), liopoic acid ligase (77), and sortase (78)].
Biomolecules other than proteins can also be fluorescently stained for super-resolution microscopy. Membrane structures can be probed with fluorescent lipids. Glycosylation patterns were detected using fluorescently labeled lectins or through metabolism of synthetic azidosugar (79). Fluorescence in situ hybridization using photoswitchable probes could enable mapping the localization of specific DNA sequences or RNAs in a cell. The further development of methods for efficient and specific labeling of cellular structures with small and bright fluorophores will be highly beneficial for super-resolution fluorescence microscopy.
Damage or alteration to the ultrastructures during sample preparation can lead to artifacts in super-resolution imaging, even though they are not normally apparent in conventional fluorescence microscopy. For immunostaining, suboptimal fixation methods could break molecular structures, and strong permeabilization reagents tend to distort membrane protein distributions. One example of such artifacts is that methanol or formaldehyde fixation, although often used for conventional immunofluorescence, is known to cause significant changes to the ultrastructures of cytoskeletons that are visible in EM. Sample preparation methods developed for EM, such as rapid freezing, freeze-substitution, and fixation with glutaraldehyde (80), could benefit super-resolution fluorescence imaging. For labeling with FPs, the commonly encountered overexpression of the protein tagged with FP can also change the ultrastructures of organelles and distributions of molecules. The side effects of FP tagging and overexpression should thus be carefully controlled in super-resolution imaging.
The time resolution of an imaging technique limits the dynamic information of a biological process that can be extracted through the method. Because of the trade-off between time and spatial resolutions intrinsically involved in all high-resolution imaging approaches, super-resolution fluorescence microscopy is in general slower than conventional fluorescence microscopy under similar instrumental conditions. Here, we provide a brief discussion of the factors limiting the imaging speed of super-resolution microscopy and a comparison of the two super-resolution microscopy approaches in terms of time resolution.
The most common forms of STED and RESOLFT microscopy are point-scanning techniques similar to scanning confocal microscopy. The higher spatial resolution of STED and RESOLFT, however, implies a smaller scanning pixel size and thus inherently slower acquisition speed when imaging the same sample. The ability to increase the acquisition rate by shortening the dwell time at each scanning pixel is limited by the photon emission rate of the fluorophore and, in the case of RESOLFT, the switching rate of the fluorophore. Therefore, in order to achieve relatively fast time resolution, a reduced area of the image field and an enlarged pixel size are typically used. For example, a 28 frame per second acquisition rate in 2D STED microscopy has been realized for live-cell imaging in a ~2.5 µm × 1.8 µm field of view by using 30-nm pixel size (for 62-nm lateral resolution) (64).
The imaging speed becomes significantly slower for a larger field of view and more so for 3Dimaging. Extrapolating from the parameters given above, the frame rate would be reduced to 0.14 fps when imaging a 30-µm × 30-µm area, which is comparable to the size of a mammalian cell. It will take a further reduction by a factor of 100 when imaging a 3-µm thick sample (roughly the thickness of a surface-adhering cell in culture) in three dimensions, leading to ~12 minutes per frame. The time resolution may be significantly increased by scanning with multiple excitation focal points, an approach that has been used in the wide-field type of RESOLFT microscopy (81).
This approach is intrinsically a wide-field imaging method, and 3D imaging has also been realized without scanning for a relatively thin sample (several hundred nanometers thick) (52). These properties give the single-molecule localization approach advantages when imaging a larger field of view. The imaging speed of this approach is, however, limited by the time required to accumulate a fluorophore localization density that is sufficient for a desired resolution given by the Nyquist criterion, as described earlier. Because different subsets of fluorescent probes are localized sequentially, the time resolution of super-resolution imaging is intrinsically slower than conventional wide-field light microscopy. For example, the observation of the focal adhension morphology, for example, employed a 30 s per frame rate to achieve a ~70-nm Nyquist-limited lateral resolution in a 30 µm × 25 µm field of view (65).
The rate of accumulating localization points is limited by the switching kinetics of fluorescent probes and the acquisition rate of the camera. Photoswitchable fluorophores have been reported to exhibit an off-switching time on the order of 1 ms (82). At this high off-switching/acquisition rate of 1 kHz, probe molecules can be activated at a rate of ~1000 µm−2 s−1 without causing substantial overlap between single-molecule images, giving a Nyquist-limited lateral resolution of about 60 nm at 1 s time resolution for homogenous structures in a 30 µm × 30 µm region. 3D imaging at 60 nm Nyquist resolution in a 30 µm × 30 µm × 3 µm volume can be performed at about 100 s per frame. As discussed above, for heterogeneous structures, which biological samples usually are, a lower number of localizations is typically needed to achieve the specified resolution, and the imaging speed will be even faster. This projected imaging speed is faster than that of STED for large area imaging, although STED is expected to offer faster imaging for sufficiently small sample areas because the image speed of STORM/PALM/FPALM would not increase as rapidly as STED with reducing image area.
In addition to biomolecular structures prepared in vitro such as DNA and DNA-protein complexes (5), numerous cellular structures have also been imaged using methods described above. Here, we list several examples, which are by no means exhaustive, to demonstrate the capabilities of super-resolution fluorescence microscopy for cell imaging.
The cytoskeleton of mammalian cells, especially microtubules (Figure 5a) (29, 44, 46, 52), is the most commonly used benchmark structure for super-resolution imaging. Other cytoskeletal structures imaged so far include actin filaments in the lamellipodium (6), keratin intermediate filaments (59), neurofilaments (26, 83) and MreB in Caulobacter (66). These images have shown the ability of super-resolution fluorescence microscopy to separate nearby filaments and overlapped and unresolvable in conventional fluorescence images (46, 52), as well as to reveal the 3D organization of the cytoskeleton in a mammalian cell (52).
Organelles, such as the endoplasmic reticulum (27), lysosome (6), endocytic and exocytic vesicles (46, 52, 64), and mitochondria (6, 53, 56), have also been imaged. For example, using the single-molecule localization approach, 3D STORM imaging has clearly resolved the ~150-nm diameter, hemispherical cage shape of clathrin-coated pits (46, 52), which only appear as diffraction-limited spots without any feature in conventional fluorescence microscopy (Figure 5a,b).Two-color 3D STED has resolved the hollow shape of the mitochondrial outer membrane (marked by the translocase protein Tom20), enclosing a matrix protein Hsp60 (56), even though the diameter of mitochondria is only about 300–500 nm (Figure 5c). The outer membrane structure of mitochondria and their interactions with microtubules have been resolved by two-color 3D STORM (53). The transport of synaptic vesicles has been recorded at video rate using 2D STED (Figure 5d ) (64).
Many plasma membrane proteins or membrane associated protein complexes have also been studied by super-resolution fluorescence microscopy. For example, synaptotagmin clusters after exocytosis in primary cultured hippocampal neurons (84), the donut-shaped clusters of Drosophila protein Bruchpilot at the neuromuscular synaptic active zone (85), and the size distribution of syntaxin clusters have all been imaged (86, 87). Photoactivation has enabled the tracking of the influenza protein hemagglutinin and the retroviral protein Gag in live cells, revealing the membrane microdomains (67) and the spatial heterogeneity of membrane diffusion (68). The morphology and transport of the focal adhension complex has also been observed using live-cell PALM (Figure 5e) (65).
The many different types of cellular structures imaged to date with various superresolution fluorescence microscopy techniques demonstrate the power and versatility of these methods.
Although still in its infancy, super-resolution fluorescence microscopy has shown great promise in studying biological structures and processes at the cellular to macromolecular scale. The rapid pace of development of all forms of super-resolution imaging techniques in recent years is expected to spur further development of novel fluorescent probes and new labeling methods as well as to extend the availability of such techniques to the wider research community. Images obtained from these new super-resolution approaches will enable scientists to directly visualize biological samples at the nanometer scale and will complement the insights obtained through traditional molecular and cell biology approaches, significantly expanding our understanding of molecular interactions and dynamic processes in living systems.
This work is supported by in part by the National Institute of Health (to X.Z.). X.Z. is a Howard Hughes Medical Institute Investigator.
The authors are not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.