Address correspondence and reprint requests to Christian Eggeling, Department of Nanobiophotonics, Max Planck Institute for Biophysical Chemistry, Am Fassberg 11, 37077 Göttingen, Germany. E-mail: firstname.lastname@example.org
Recent developments in fluorescence far-field microscopy such as STED microscopy have accomplished observation of the living cell with a spatial resolution far below the diffraction limit. Here, we briefly review the current approaches to super-resolution optical microscopy and present the implementation of STED microscopy for novel insights into live cell mechanisms, with a focus on neurobiology and plasma membrane dynamics.
Microscopy has revolutionized our understanding of life and matter, since it allows the observation of molecular organization in complex biological systems. Among all the different microscopy techniques, optical microscopy is one of the least invasive approaches: the use of light hardly influences the system under study. Optical microscopy of the living cell is usually applied in the far-field, where the use of a lens-based system and focused light allows the observation micrometers to millimeters away from the given optical element(s), preserving the non-invasiveness and the ability to observe deep inside living cells or tissues. Further, when combined with the fluorescence readout (e.g., fluorescent-labeled molecules), far-field fluorescence microscopy achieves the specific and highly sensitive detection of cellular constituents and thus permits the disclosure of molecular distributions and dynamics in the living cell.
However, a number of challenging issues in biology lie beyond the reach of conventional far-field optical microscopy, especially because of its limited spatial resolution. When using visible and focused light, light diffraction prevents the discernment of alike objects closer together than approximately 200 nm (Abbe 1873). A remedy to this physical limit is the reversible inhibition of fluorescence, ensuring that the measured signal stems from a region of the sample that is much smaller than 200 nm (Hell 2007). Starting in 1994 (Hell and Wichmann 1994) different far-field fluorescence super-resolution microscopy or nanoscopy approaches have emerged, all capable of achieving cellular imaging with – in principle – unlimited spatial resolution. Examples include optical Stimulated Emission Depletion (STED) (Hell and Wichmann 1994; Klar et al. 2000), Reversible Saturable Optical (Fluorescence) Transitions (RESOLFT) (Hell et al. 2003; Hofmann et al. 2005; Grotjohann et al. 2011) Ground State Depletion (GSD) (Hell and Kroug 1995; Bretschneider et al. 2007; Han et al. 2010) nanoscopy, Stochastic Optical Reconstruction Microscopy (STORM) (Rust et al. 2006), and (fluorescence) Photoactivated Localization Microscopy ((f)PALM) (Betzig et al. 2006; Hess et al. 2006). Detailed reviews of the different nanoscopy approaches, especially of their different technical aspects, can be found elsewhere (Hell 2003, 2007, 2009a, b; Hell et al. 2004; Moerner 2006; Heintzmann and Ficz 2007; Rice 2007; Bates et al. 2008; Dedecker et al. 2008; Fernandez-Suarez and Ting 2008; Chi 2009; Evanko 2009; Heilemann et al. 2009; Heintzmann and Gustafsson 2009; Huang et al. 2009, 2010; Lippincott-Schwartz and Manley 2009; Huang 2010; Patterson et al. 2010; Dempsey et al. 2011; Muller et al. 2012). After a brief overview of these approaches, we will restrict ourselves to the basic principles underlying recent developments in optical nanoscopy which make it a highly versatile tool for the observation of the living cell, taking STED nanoscopy as an example and focusing on neurobiology and plasma membrane dynamics.
The principle of fluorescence nanoscopy
Differing in their technical details, all of the aforementioned fluorescence nanoscopy approaches are based on the principle of reversibly inhibiting the fluorescence emission of the cellular labels (Hell et al. 2003; Hell 2004, 2007) (Fig. 1).
STED (Hell and Wichmann 1994; Klar et al. 2000), GSD (Hell and Kroug 1995; Bretschneider et al. 2007; Han et al. 2010) or more generally RESOLFT (Hell et al. 2003; Hofmann et al. 2005; Grotjohann et al. 2011) nanoscopes use a targeted or deterministic approach by inhibiting the fluorescence at pre-defined spatial positions, as outlined in the examples in Fig. 1a: a laser beam that excites fluorescence, that is, which switches fluorescence ON, is focused onto a small (diffraction-limited) spot. Scanning this spot over the sample and detecting the fluorescence for each scanning position permits the reconstruction of an image by providing the spatial distribution of the fluorescent-labeled molecules (as is for example realized in a confocal microscope). The spatial resolution is given by the so-called point-spread-function (PSF) of the microscope, which for the scanning microscope is mainly determined by the size of the laser focus. Unfortunately, the diameter of the spot onto which the light can be focused is determined by the diffraction of the light at the limited aperture of the microscope's (objective) lens – approximately 200 nm for visible light (Abbe 1873; Hell 2007). However, addition of a second laser beam that switches off the fluorescence emission permits the observation of spots of nanoscopic size. Featuring a focal intensity distribution with at least one local intensity minimum (a zero-intensity point, e.g., at the focal center) this ‘OFF’ light beam selectively inhibits fluorescence everywhere but at the zero-intensity point. Scanning fluorescent images therefore constitute a huge improvement in spatial resolution. Reversible inhibition of fluorescence is realized by, for example, stimulated emission (STED), photoswitchable fluorophores (RESOLFT) or transient transitions to the fluorescent labels’ metastable dark states (GSD).
In 2006, a new approach arose, referred to under a multitude of different acronyms such as STORM (Rust et al. 2006), (f)PALM (Betzig et al. 2006; Hess et al. 2006), Ground State Depletion Microscopy followed by individual molecule return (GSDIM) (Fölling et al. 2008), direct STORM (dSTORM) (Heilemann et al. 2008), or Single-Molecule Active Control Microscopy (SMACM) (Biteen et al. 2008). This approach relies on the stochastic on- and off-switching of the fluorescence emission of individual, well-separated labels combined with the determination of their exact spatial positions (Fig. 1b): usually, a large area of the sample is illuminated all at once and the spatial distribution of labeled molecules is determined by imaging the fluorescence onto a camera. Again, alike objects closer together than 200 nm cannot be distinguished, since because of the diffraction of light the image of a single point (PSF) appears blurred on the camera. By inhibiting the fluorescence emission of most of the labels, only single isolated molecules are allowed to fluoresce at a given time, and their spatial positions can be precisely determined from their blurred image spots on the camera using different localization approaches. By stochastically switching on and off different single isolated molecules in subsequent camera recordings, the final image with sub-diffraction sized spatial resolution is reconstructed from the summation of all localized spatial positions. It should be kept in mind that localization per se cannot provide sub-diffraction spatial resolution: separating alike objects at small distances requires a criterion to discern them from one another, such as switching them on and off in space and/or time (Hell and Wichmann 1994; Hell and Kroug 1995; Gordon et al. 2004; Qu et al. 2004; Lidke et al. 2005; Hell 2009b). This is why although it had been routinely applied for decades (see, e.g., Kusumi et al. 2005), specifically for spatio-temporal tracking of single isolated molecules, localization microscopy on its own did not provide images with sub-diffraction resolution.
The different nanoscopy techniques can record images with both multiple colors (e.g., Donnert et al. 2007; Bock et al. 2007; Bates et al. 2007; Schmidt et al. 2008; Huang et al. 2008; Juette et al. 2008; Shtengel et al. 2009; Pavani et al. 2009) and improved spatial resolution along all spatial directions (e.g., Klar et al. 2000; Harke et al. 2008; Schmidt et al. 2008). However, they each have their advantages and disadvantages with respect to for example photobleaching, set-up complexity, image acquisition speed and image reconstruction. For instance, a direct image of the spatial distribution of labeled molecules is formed in the scanning approach of the deterministic STED, GSD and RESOLFT nanoscopes, whereas a reconstruction algorithm based on the identification and localization of single isolated molecules needs to be employed to create the final image in the stochastic (d)STORM/(f)PALM/GSDIM approaches. Potential bias to the latter reconstruction process may arise for example from a lack of knowledge of missed or repeated recordings of molecules (Shroff et al. 2008; Small 2009). Furthermore, the determination of molecular positions may be biased for out-of-focus molecules (Enderlein et al. 2006; Engelhardt et al. 2011). Quite a number of improvements in the identification and localization algorithms have therefore been developed over the past years, especially aimed at image reconstructions for more densely labeled samples and/or those having low signal-to-noise levels (see, Dertinger et al. 2009; Cronin et al. 2009; Hedde et al. 2009; Smith et al. 2010; Larson 2010; Laurence and Chromy 2010; Henriques et al. 2010; Mortensen et al. 2010; Pertsinidis et al. 2010; Wolter et al. 2010; Endesfelder et al. 2010; Holden et al. 2011; Huang et al. 2011; Jones et al. 2011; Cox et al. 2012).
Photobleaching of the fluorescent labels is a more complex issue in deterministic-based nanoscopy: whereas in stochastic-based methods a molecule needs in principle to be switched on and off only once, the fluorescence has to be switched on and off multiple times during the scanning recording of the deterministic nanoscopy approaches, favoring labels and experimental conditions that supply this condition – we will come back to this issue later on. On the other hand, the image acquisition time of the stochastic (d)STORM/(f)PALM/GSDIM approaches is limited by the need to acquire sufficient molecular positions for the image reconstruction. Despite the aforementioned recent improvements in reconstruction algorithms there will potentially always be uncertainty as to whether the position of all molecules has been recorded, that is, of when to stop the acquisition, simply because these approaches are based on the stochastic on/off switching of fluorescence (see, e.g., Shroff et al. 2008; Small 2009). This is different for deterministic-based STED/GSD/RESOLFT approaches, where the scanning process defines the acquisition time. The use of fast beam scanners has established STED/RESOLFT nanoscopy as the currently fastest sub-diffraction imaging technique available, with up to 60–80 frames per second recording times for a few μm-sized observation areas (Westphal et al. 2008; Testa et al. 2012). In particular, the parallelization of the scanning procedure, that is, the simultaneous scanning of multiple observation spots (such as realized for a structured illumination), leaves room for further improvement in the image acquisition speed of deterministic approaches (Heintzmann et al. 2002; Hell et al. 2003; Gustafsson 2005; Schwentker et al. 2007; Rego et al. 2012). The basic nanoscopy setups are usually expansions of conventional wide-field or (confocal) scanning microscopes by addition of an extra laser for switching fluorescence [note that confocality is not a requirement for the STED/GSD/RESOLFT concepts (Hell 2007)]. An initial concern of STED nanoscopy was the requirement for large, complex laser systems combined with the elaborate experimental procedure required to create zero-intensity point(s). However, recent advances in optical design and laser technology have permitted considerable simplification of the setups (Willig et al. 2007; Wildanger et al. 2008, 2009; Schrof et al. 2011; Vicidomini et al. 2011; Gould et al. 2012; Honigmann et al. 2012a; Mueller et al. 2012).
Most importantly, as they are based on the same principle of switching fluorescence on and off, the very same sample can be used for both targeted- and stochastic-based nanoscopy (Bretschneider 2007; Fölling et al. 2008; Brakemann et al. 2011). In view of the increased sensitivity and the aforementioned potential bias, both approaches can consequently be applied for the validation of results. Other approaches such as structured-illumination microscopy (SIM) (Gustafsson 2000; Schermelleh et al. 2008) or 4-Pi microscopy (Hell 1992; Gustafsson et al. 1995) are also often referred to as super-resolution microscopies. However, their spatial resolution is still limited (twofold improvement in spatial resolution for SIM and five to sixfold improvement in axial resolution for 4-Pi), that is they push the spatial resolution to the limits of the diffraction barrier. However, only the combination with reversible fluorescence inhibition processes can break these limits (Dyba and Hell 2002; Heintzmann et al. 2002; Hell et al. 2003; Gustafsson 2005; Schwentker et al. 2007; Rego et al. 2012). The fact that all the microscopy approaches are complementary, whether they are diffraction-limited or with nanoscale resolution, promotes research environments with access to various kinds of microscopes and nanoscopes, depending on their suitability for the case in hand.
STED nanoscopy: the road to live cell imaging
STED nanoscopy was not only the first nanoscopy technique to be developed, it was also the first to achieve cellular images with sub-diffraction spatial resolution (see, Klar et al. 2000; Willig et al. 2006b). Although a multitude of experiments over the past years have rendered STED imaging of fixed cells with down to 20 nm spatial resolution a generic tool for cellular studies, the potential phototoxic effect of the added STED light was long believed to be non-compatible with the study of living cells. Whereas the light intensities of the excitation laser are the same as in conventional confocal microscopy (1–10 μW), stimulated emission for example requires quite high light intensities of 1–10 MW/cm2 at around 740–780 nm of the added STED laser (for example supplied by 100–300 mW average power of ~ 100–300 ps-long pulses of a titanium:sapphire laser), introducing the need to pay heed to heat absorption and light-induced toxic reactions. However, the involved peak intensities are smaller than those used in two-photon microscopy (Denk et al. 1990), a well-established tool in live-cell microscopy (Denk 1996). Further concerns about the addition of a STED laser arise from the increased photobleaching of the fluorescent label. This derives from the fact that the STED light acts on the excited fluorescent label. STED light per se is however not absorbed by the label and does not produce any photo-reactive and thus -toxic species (unlike two-photon microscopy). Rather, stimulated emission shortens the time the fluorescent label spends in its excited (reactive) state and can reduce photoreactions such as photobleaching [compare with (Eggeling et al. 2005)]. Although excited state absorption of STED light can unfortunately cause severe photobleaching (Eggeling et al. 1998; Hotta et al. 2010), it was shown that fast scanning and the right choice of STED wavelength can minimize these effects (Donnert et al. 2009; Hotta et al. 2010; Moneron et al. 2010; Rankin et al. 2011). This is especially the case when using STED light in the range of 600 nm as required for live-cell labels such as green fluorescent protein (GFP) and its variants (Moneron et al. 2010; Rankin et al. 2011). Further, the use of a continuous-wave laser in conjunction with a gated fluorescence detection scheme has enabled a reduction in the STED laser power and a simplification of the setups (Vicidomini et al. 2011; Honigmann et al. 2012a; Mueller et al. 2012). As a consequence, STED microscopy is nowadays considered a straightforward technique for the study of the living cell using genetically encoded markers such as fluorescent proteins (Willig et al. 2006a; Hein et al. 2008; Nagerl et al. 2008; Eggeling et al. 2009; Li et al. 2009; Moneron and Hell 2009; Morozova et al. 2010; Rankin et al. 2011; Tonnesen et al. 2011; Urban et al. 2011), tagging proteins such as SNAP-, HALO-, or CLIP-tags (Schröder et al. 2008; Eggeling et al. 2009; Hein et al. 2010; Pellett et al. 2011; Lukinavicius et al. 2012), or fluorogen-activating tags (Fitzpatrick et al. 2009) (which both covalently bind functionalized and membrane-permeable organic dyes), even using commercial instrumentation (Schröder et al. 2008; Fitzpatrick et al. 2009; Morozova et al. 2010; Friedemann et al. 2011). Furthermore, the quite high laser intensities of 1–10 MW/cm2 can be circumvented when switching from STED to RESOLFT nanoscopy, which employs laser intensities in the range of 1–5 kW/cm2 only (Hofmann et al. 2005; Brakemann et al. 2011; Grotjohann et al. 2011; Testa et al. 2012). However, in contrast to STED, which can in principle employ any label, RESOLFT nanoscopy relies on the use of special reversible photoswitchable molecules such as the reversible switchable GFP (rsGFP) (Grotjohann et al. 2011). Fortunately, the number of reversible switchable fluorescent proteins has significantly increased over the last years and their switching characteristics steadily improved (Grotjohann et al. 2012).
Live-cell STED microscopy in neuroscience
Confocal and two-photon microscopies have become common tools in Neurobiology, and have found wide application in the study of neuronal connections or processes, particularly those lying a few microns deep in, for example, cerebral cortex. It is important to study the neurons in living brain tissue and even the living animal, where all connections are preserved and functional. However, the diffraction-limited resolution of conventional far-field light microscopy is inadequate to resolve fine details, obliging researchers to complement the live-cell imaging with electron microscopy. This comes at a high cost, because the latter form of microscopy requires fixation, dehydration, embedding, thin cutting, and vacuum observation of the dehydrated nervous tissue. A case in point is the study of dendritic spines, the fine processes that form the post-synaptic compartment of central nervous system synapses. The majority of the excitatory synapses in the central nervous system are formed on dendritic spines. These micrometer/sub-micrometer-sized protrusions of the dendrite come in a variety of shapes: filopodia-like, stubby, cup-shaped, or mushroom shaped. Interestingly, dendritic spines show developmentally regulated and activity-dependent morphological changes. By adapting a STED microscope to the demands of in vivo imaging, the subtle dynamics of spines in the cerebral cortex of a living mouse were made visible (Fig. 2) (Berning et al. 2012). In living organotypic cultures of brain slices, it was shown that morphological parameters such as the neck width and curvature of the spine heads can be quantified to record changes, for example, induced by chemical long-term potentiation (cLTP) (Nagerl et al. 2008). The imaging of sub-cellular structures is also made possible with STED microscopy: actin, a key player in the morphological changes of synapses, can be visualized by STED microscopy. Using a fusion protein of Lifeact (a small peptide label for actin that does not disturb the actin polymerization or dynamics) and the yellow fluorescent protein (YFP), dynamic changes in actin molecules in synaptic spines located up to 120 μm deep in the cerebral tissue were observed (Urban et al. 2011). Establishment of either LTP or long-term depression (LTD) apparently requires a responsive F-actin cytoskeleton (Yuste and Bonhoeffer 2001). During neuronal development, a healthy cytoskeleton is essential for brain neurons to undergo the sequential steps of neurogenesis, cell migration and terminal differentiation (Ik-Tsen Heng et al. 2009). Nanoscopy studies of the actin cytoskeleton is gaining momentum in work on dendritic spines in brain (Tatavarty et al. 2011; Berning et al. 2012),(Izeddin et al. 2011) and other cells and tissues (Xu et al. 2012). The localization of Na+,K+-ATPase has recently been accomplished using STED microscopy (Blom et al. 2011). Insight into the organization of protein complexes in the presynaptic active zone and post-synaptic density has also been gained with the aid of other nanoscopy techniques such as STORM (Dani et al. 2010). Endosomal sorting of vesicles in the presynapse has been analyzed in one of the few studies combining STED and electron microscopy (Hoopmann et al. 2010).
For imaging deep inside living tissue, STED microscopy has natural advantages over stochastic-based nanoscopy such as (d)STORM/(f)PALM/GSDIM. It provides sectioning capability and superior timing resolution. And, as shown by the various examples presented here, it can be a valuable tool in understanding basic working principles of the brain. Furthermore, it can be combined with two-photon excitation (Ding et al. 2009; Li et al. 2009; Moneron and Hell 2009; Bianchini and Diaspro 2012) or adaptive optics (Gould et al. 2012) for an optimized fluorescence excitation in deep tissue.
Live-cell STED microscopy of plasma membrane organization
A prominent example of the need for sub-diffraction resolution is the observation of plasma membrane heterogeneity such as in lipid–protein interactions, which are considered to play a functional part in a wide range of membrane-associated functionally relevant processes (see, e.g., Lingwood and Simons 2010). Unfortunately, many structural characteristics of the cellular membrane such as protein clusters, areas of different molecular order (often denoted lipid ‘rafts’), or clathrin-coated pits are < 200 nm in size. Thus, their direct and non-invasive observation in the living cell is impeded by the resolution limit of > 200 nm of the conventional far-field fluorescence microscope. Over the last years, we have published several reports on the detection of nanoscopic membrane heterogeneities in the plasma membrane of living cells using the superior spatial resolution of STED nanoscopy. Besides directly imaging the organization of < 70 nm large protein clusters in the plasma membrane (Sieber et al. 2007), we could obtain new details of molecular membrane dynamics by combining a (tunable) resolution of down to 30 nm with tools such as fluorescence correlation spectroscopy (FCS) (Eggeling et al. 2009; Ringemann et al. 2009; Mueller et al. 2011; Honigmann et al. 2012b; Sezgin et al. 2012) (Fig. 3). In FCS, the transits of single molecules through a usually fixed observation spot are observed to determine dynamic parameters such as the average transit time (Magde et al. 1972; Ehrenberg and Rigler 1974; Haustein and Schwille 2003). Investigating the dependence of the average transit time on the size of the observation spot highlights hindrances in molecular diffusion such as transient molecular interactions (Wawrezinieck et al. 2005). STED nanoscopy is a perfect tool for these studies as it allows a straightforward variation of the observation spot size by, for example, variation of the power of the STED laser (Eggeling 2012). Our STED-FCS experiments revealed that sphingolipids and certain proteins are transiently (~ 10 ms) trapped on the nanoscale in often cholesterol-mediated and/or cytoskeleton-associated molecular complexes, whereas other molecules diffuse freely (Mueller et al. 2011). Different lipid structures show up distinctly, but so far at least such differences do not appear to be because of variations in the labeling. We observed strong differences in the molecular nature behind the nanoscopic molecular interactions as observed by STED-FCS and the separation of membranes into liquid-disordered and -ordered phases (Mueller et al. 2011; Sezgin et al. 2012). Specifically the latter phase is often referred to as being an appropriate physical model of lipid nanodomains, the so-called lipid ‘rafts’. Together with a suitable fluorescent lipid marker for the liquid-ordered phase, STED-FCS has the potential to shed new light on the mystery and functionality of these lipid nanodomains (Honigmann et al. 2012b).
STED-FCS is very similar to single-molecule or -particle tracking (SPT), which is also often applied to investigate diffusion characteristics of membrane components (e.g., Kusumi et al. 2005; Wieser et al. 2007; Manley et al. 2008; Sahl et al. 2010). While SPT allows the direct visualization of the diffusion path of individual molecules and thus of any (even less abundant) hindrances, it requires much longer recording times and more detailed data analysis than FCS to obtain average molecular characteristics of high statistical accuracy. Nevertheless, both STED-FCS and SPT, and more so their complementary applications, have the potential to revolutionize our understanding of membrane bioactivity (e.g., Kusumi et al. 2010; Eggeling 2012).
It is highly likely that the novel possibilities of far-field fluorescence STED nanoscopy described in this short review will contribute to our understanding of the living cell in general, and will shed new light on a whole range of cellular phenomena, including key neurobiological processes such as development, pruning, and stabilization of synapses, their subtle dynamic changes triggered by chemical or electrical stimulation, and the possible alteration of these processes in disease. Likewise, the new nanoscopies are ideally suited to address important issues in membrane biology which are bound to find answers at the nanoscale, such as lipid–protein interactions and nanodomains in membrane bioactivity and receptor signaling in neurobiology and cell biology at large.