Cell biology beyond the diffraction limit: near-field scanning optical microscopy

Fluorescence microscopy has gained a firm position among the most important research tools in modern cell biology. It offers high chemically specific contrast, and is non-invasive, and sample handling is non-destructive and therefore allows the study of living cells. Moreover, optical detection is very fast, providing picosecond time resolution, and the very nature of light offers additional contrast mechanisms (e.g. based on polarization or phase discrimination) that yield additional information. Furthermore, imaging can be performed in three dimensions by multi-photon excitation, confocal detection or deconvolution techniques. Finally, a wide range of ultra-sensitive detectors – for example, cooled CCD (charge-coupled device) cameras and photon-counting devices – that make optical detection extremely sensitive, even down to the single-molecule level, are available.

The ability to tailor fluorescent fusion proteins, exploiting the strong autofluorescence of the green fluorescent protein (GFP) family (Patterson et al., 2001), has fueled interest in fluorescence microscopy even further. The most successful applications of these fluorescent proteins are in gene expression, protein targeting and trafficking, and protein-protein interaction studies (Tsien, 1998; Lippincott-Schwartz et al., 2001). Employing various forms of these fluorescent proteins, in most cases using fluorescence resonance energy transfer (FRET) as a read-out (Wouters et al., 2001), highly potent cellular indicators for calcium (Miyawaki et al., 1997; Miyawaki et al., 1999), pH (Llopes et al., 1998), cyclic AMP (Zaccolo et al., 2000), cyclic GMP (Honda et al., 2001), caspases (Harpur, 2001) and small G protein activity (Mochizuki et al., 2001), among others, have been developed.

The limit to the resolution that can be reached in optical imaging techniques is directly related to the wavelength of the light. This diffraction limit originates from the fact that it is impossible to focus light to a spot smaller than half its wavelength. In practice this means that the maximal resolution in optical microscopy is ∼250-300 nm. Since a large body of evidence indicates that dynamic cell-signaling events start by oligomerization and interaction of individual proteins (i.e. on the molecular scale), the need for imaging techniques that have a higher resolution is growing. Traditionally, high-resolution cell biology (Table 1) is the arena of electron microscopy, which offers superb resolution but lacks the above-mentioned advantages of fluorescence microscopy. The advent of scanning probe microscopy (Table 1), and especially atomic force microscopy (AFM), in which an atomically sharp probe attached to a cantilever is scanned over the surface of interest, has made nanometer resolution also attainable on living cells (Hansma et al., 1994; Putman et al., 1994). However, although AFM produces a high-resolution topographical picture of the sample, it lacks chemical specificity. Hence, although individual molecules can be seen, their identities cannot be defined. This seriously limits the usefulness of AFM for high-resolution imaging on cells. Initially, this contrast problem was tackled by the use of immunogold-labeling approaches (Damjanovich et al., 1995; Neagu et al., 1994). A promising new way around the problem comes from work on the specific labeling of the AFM probe with biomolecules (e.g. with antibodies or ligands). This introduces a contrast mechanism based on specific interactions between the probe and certain types of molecule in the specimen (Willemsen et al., 2000). Other attempts to enhance AFM contrast involve the modification of the probe by fluorescent molecules, which introduces an optical contrast mechanism (Vickery and Dunn, 2001). Currently, however, the combination of high-resolution scanning probe and fluorescence microscopy is the realm of another scanning probe technique: near-field scanning optical microscopy (NSOM).

In NSOM, as in the case of AFM, a sharp probe physically scans the sample surface (Fig. 1). Besides topography, NSOM also generates optical images. A typical NSOM configuration is shown in Fig. 2. The practical feasibility of this kind of NSOM was first demonstrated by Pohl et al., immediately following the advent of scanning probe microscopy and in fact before the introduction of the AFM (Pohl et al., 1984). The most generally applied near-field optical probe consists of a small aperture, typically 20-120 nm in diameter (i.e. much smaller than the wavelength of the excitation light), at the end of a metal-coated tapered optical fiber^* (Fig. 3A,B). This probe is the most critical element of the technique and is difficult to produce reliably and in large quantities (Van Hulst et al., 2000). In fluorescence mode^*, it serves as a constriction that funnels an incident light wave to dimensions that are substantially below the diffraction limit. This results in a light-source that has the size of the aperture. However, in contrast to common light sources such as lightbulbs and lasers, the light emitted by the probe is predominantly composed of evanescent waves rather than propagating waves. The intensity of the evanescent light decays exponentially and to insignificant levels ∼100 nm from the aperture. Effectively, the probe can excite fluorophores only within a layer of <100 nm from the probe – that is, in the ‘near-field’ region (Fig. 3B). Sample fluorescence can subsequently be collected by conventional optics (Fig. 2) and transformed into an optical image of the sample surface in which the resolution is now primarily dictated by the aperture dimensions rather than by the wavelength of the light.

Since the near-field intensity decays so rapidly with distance from the probe, for efficient excitation it is essential to have accurate control of the probe-sample distance during scanning. As in AFM, this can be accomplished by using a (force) feedback loop. However, in contrast to regular AFM, which exploits the bending of a cantilever attached to the probe as a direct measure of the probe-sample interaction force, in NSOM an indirect method, based on shear-force damping, is commonly used. For this, the NSOM probe, or a piezoelectric tuning fork attached to it, is oscillated at its resonance frequency in a lateral vibrational mode (with a <1 nm amplitude); when in proximity to the sample, shear-forces dampen this motion and induce measurable changes in the oscillation amplitude and phase. An electronic feedback system, controlling the probe-sample distance directly through the piezoelectric scan stage, is subsequently used to maintain a constant oscillation amplitude/phase during scanning. In this way, a constant probe-sample distance of <10 nm is realized (Fig. 2). The feedback signal itself, as in AFM, is used to generate a topographic map of the sample surface (Fig. 1). Of course, unique to NSOM is the fact that a corresponding fluorescence map is simultaneously generated.

The optical detection sensitivity of NSOM depends largely on the extremely small excitation/detection volume set by the aperture dimensions as well as the depth of penetration of the near-field into the specimen. Together, these properties effectively reduce background fluorescence and thereby enhance detection sensitivity. Betzig and Chichester exploited this eight years ago, providing the first observation of single-molecule fluorescence under ambient conditions (Betzig and Chichester, 1993). Furthermore, employing the polarization characteristics of the near-field, they showed that it is possible to determine the full spatial orientation of fluorescent molecules by making use of polarization-sensitive fluorescence detection (see below). Subsequently, developments in instrumentation that greatly improved signal-to-background ratios allowed single-molecule fluorescence studies to come within reach of far-field methods such as total internal reflection, confocal and bright-field microscopy (see Table 2). The advantage of single-molecule studies is that they provide a way to monitor time-dependent processes and reaction pathways in non-equilibrated systems, which reveals the distribution of a given molecular property instead of a statistical average. This has already led to a whole new frontier in science, its applications ranging from basic photo-physics and material research to biology (Frontiers in Chemistry, 1999; Xie and Lu, 1999; Sako et al., 2000; Ishii and Yanagida, 2000). Nowadays almost all experiments in this field use far-field methods, mainly because they are relatively easy to use. However, the obvious disadvantage is that, because of the diffraction-limited system response, only scarcely labeled samples can be studied, but, in biology, molecules are usually present in close proximity. The combination of topographical information, optical super-resolution and single-molecule detection sensitivity therefore makes NSOM a unique tool for biological applications.

A large body of evidence indicates that (induced) lateral organization in the plasma membrane is crucial to trigger cell signaling. One example is the early events in signal transduction by the epidermal growth factor receptor (EGFR), dimerization and phosphorylation following EGF binding (Yarden and Schlessinger, 1987a; Yarden and Schlessinger, 1987b). Another is the formation of focal adhesions, large signaling complexes involved in cell-adhesion and migration, and their equivalent on leukocytes, ‘adhesisomes’. An important first step in the assembly of such complexes appears to be clustering of integrin receptors (Balaban et al., 2001; Van Kooyk and Figdor, 2000). Understanding of the signaling capacities of these adhesion sites requires insight into their assembly and spatial organization on the molecular level. The concept of ‘lipid rafts’ provides yet another level of organization in the plasma membrane at the submicroscopic level (Simons and Toomre, 2000). So far, these phenomena have been studied by use of far-field optical techniques, such as TIRF, IRM and confocal microscopy (Table 1). Clearly, higher-resolution/sensitivity would produce a major step forward in our understanding of the underlying mechanisms.

So far, most applications of NSOM in biology involve systems that are more or less isolated (Table 3) – for example, studies on fluorescently labeled chromosomes (Moers et al., 1996), DNA (Ha et al., 1996; Garcia-Parajo et al., 1998) and purified fluorescent proteins (Garcia-Parajo et al., 1999; Garcia-Parajo et al., 2000). Cell biological studies include fluorescence imaging of cytoskeletal components in 3T3 fibroblasts (Betzig et al., 1993) and colocalization of malarial and host skeletal proteins on malaria-infected erythrocytes (Enderle et al., 1997). Furthermore, sub-wavelength-sized membrane patches in human skin fibroblasts (Hwang et al., 1998) and activation-dependent receptor clustering on a human breast carcinoma cell line have also been studied (Nagy et al., 1999). Although these studies show a resolution well beyond that of a confocal microscope, to the best of our knowledge no study showing single-molecule detection sensitivity in a cell membrane by NSOM has been reported.

To demonstrate the potential of near-field optical microscopy to study the distribution and orientation of cell surface molecules, we investigated the distribution of LFA-1 (αL/β2 integrin) molecules in murine fibroblasts. Fig. 4A shows a bright-field image as well as a near-field ‘zoom-in’ on the surface of a fixed and critical-point-dried L-cell (a mouse fibroblast cell line) expressing human GFP-tagged LFA-1^*.

The near-field image shows a number of interesting findings. First individual fluorescent LFA-1 molecules (circles, squares and hexagon) as well as clusters of molecules can be recognized (arrows). Secondly, one can observe different orientations of the GFP linked to the LFA-1 β chain, as determined by the polarization of the emitted light. The latter is of great value to discriminate LFA-1 molecules that are closer than half the tip-aperture diameter. Although these molecules cannot be discriminated by their fluorescence intensity signal, the difference in polarization contrast reveals them as separate entities (Van Hulst et al., 2000). Finally, the figure shows examples of photodissociation (circles) and the characteristic blinking behavior (squares and inset) of single molecule GFP fluorescence (Garcia-Parajo et al., 2000). Fig. 4B depicts a line trace through the feature marked with a hexagon, showing a signal-to-background ratio of 7 and a 90 nm full width at half maximum (FWHM; arrows). From the latter, one can calculate that the lateral resolution in this image is ∼40 nm (Van Hulst et al., 2000). Taken together, the discrete photo-behavior, defined polarization characteristics and the number of counts per spot provide compelling evidence that individual molecules can be recognized on the cell surface by NSOM.

A direct comparison between the localization accuracy of cell membrane receptor molecules of NSOM and far-field imaging techniques is instructive (Table 2). In confocal fluorescence studies, using a 1.3 numerical aperture objective, two groups have reported a lateral positioning accuracy of 30-40 nm (Schmidt et al., 1996; Van Oijen et al., 1998). For 10⁴ collected photons and a 70 nm aperture probe, NSOM can be calculated to attain a localization accuracy of 6 nm – a level comparable to the FRET interaction-distance regime (0-10 nm). Given a resolution of ∼50 nm, NSOM should allow independent observation of several hundreds of molecules/μm², which is an order of magnitude more than that achievable by far-field methods. We expect the achievable level of sensitivity of near-field methodology to be sufficient to tackle questions related to the distribution of molecules that are packed at physiologically relevant densities on the cell surface. We therefore anticipate that NSOM will yield results comparable to those obtained by (immunogold) transmission electron microscopy but under physiological conditions.

These intrinsic advantages of NSOM might revolutionize the life sciences. But why is NSOM hardly used in cell biology, despite its development almost two decades ago? First, as an imaging tool, NSOM is a complex technique that requires well-trained operators. Second, as we point out above, the resolution and sensitivity of a near-field microscope depend strongly on the quality of the probe aperture and the accuracy of the feedback system, necessitating full control of the technology. Moreover, the top-grade near-field microscopes are still under development in experimental physics laboratories and are therefore not widely accessible. Perhaps most important is the fact that, despite great efforts of various groups including our own, the standards set by the initial work of Betzig and Chichester cannot yet be reached in a liquid and more physiological environment. This is the technical challenge that has to be faced in the coming years.

NSOM combines the high resolution of scanning probe microscopy with the contrast of optical microscopy. It can achieve single-molecule detection sensitivity on the cell surface and allows independent observation of molecules at physiologically relevant packing densities. The level of detail and sensitivity offered by this technique will be of particular value in studies of the spatial organization of the plasma membrane. Co-localization studies, a common application of far-field fluorescence imaging in cell biology, when performed by NSOM will provide unprecedented detail and accuracy, which are impossible to obtain by diffraction-limited imaging techniques. Potentially of great importance will be the application of NSOM in single-pair FRET studies combined with high-resolution imaging to study intermolecular interactions or intramolecular dynamics at the single-molecule level on cells. The first such examples have already been reported (Kirsch et al., 1999); however, although clearly showing resolution below the diffraction limit, single-molecule detection sensitivity has not yet been reached. Importantly, NSOM bridges the gap between the diffraction-limited response of normal light microscopy and the 5-10 nm distance sensitivity inherent in FRET and, as such, already provides important additional but otherwise unobtainable information.

In the past decade, especially the past two years, several other approaches to break the diffraction limit have been developed. These include interferometric microscopy methods such as 4Pi confocal microscopy (Hell and Stelzer, 1992), I5M (Gustafsson et al., 1999), standing-wave-total-internal-reflection fluorescence microscopy (Cragg and So, 2000) and harmonic excitation light microscopy (Frohn et al., 2000). These methods, however, do not have single-molecule detection sensitivity, and all require extended electronic post-processing of the images. Recently, a novel technique involving ‘point spread function (PSF) engineering’, which exploits stimulated emission depletion (STED), has been described (Klar et al., 2000). This method involves the induced quenching of fluorescence by stimulated emission at the rim of the diffraction-limited focal spot, thereby squeezing it to an almost spherical shape of ∼100 nm in diameter. Thus far, this is the only technique that seriously rivals the small excitation/detection volume and therefore sensitivity of NSOM. The maximal resolution obtained with aperture-type NSOM relates to the limited energy throughput of the near-field probe. This limits the minimum size of the aperture to be used and hence the resolution of the microscope to ∼20 nm. A possible way around this involves exploiting single-molecule emitters, attached to a scanning probe, which act as light source to excite molecules in the sample (Michaelis et al., 2000).

The most important technical challenge that remains is the construction of an NSOM instrument that operates under physiologically relevant conditions and allows the study of soft, rough and motile surfaces, such as the plasma membrane of living cells. When combined with single-molecule detection sensitivity and an optical resolution that is comparable to transmission electron microscopy, this will prove to be an invaluable tool in cell biology. These are truly exciting times.

We thank Jeroen Korterik, Frans Segerink, Marjolein Koopman and Ben Joosten for expert technical assistance. Frank de Lange is supported by grant FB-N/T-1a from the Netherlands Foundation for Fundamental Research of Matter (FOM). Bärbel de Bakker is supported by grant TTN4812 from the Dutch Technology Foundation (STW). The research by Maria Garcia-Parajo was made possible by a fellowship of the Royal Netherlands Academy of Arts and Sciences (KNAW).