Armed with a complete chemical composition of the membrane, obtaining the spatial organization of this information, preferably in real time and in a live cell, is the important next step. Matrix-assisted laser desorption ionization (MALDI) can locally vaporize material into ionized molecules or molecular fragments which are subsequently analyzed with a mass spectrometer. Raster-scanning the laser over a sample will generate an image with unprecedented chemical resolution, albeit at the rather low spatial scale of a few micrometers ³² .

At the expense of chemical bandwidth, magnetic sector secondary ion mass spectrometry (NanoSIMS) offers a typical spatial resolution of 100 nm on cell membranes ³³ . In this method, a focused primary ion beam sputters neutral and ionized molecular fragments from the sample surface. These ejected secondary ions are subsequently collected and analyzed in the mass spectrometer. An additional benefit is a shallow sampling depth of 5 nm, which is due to the small secondary ion escape depth, making this technique exquisitely sensitive to the plasma membrane ³⁴ . Because of the monoatomic and diatomic nature of the secondary ion component, identification is possible only if the molecules of interest contain distinct elements or isotopes ³³ . For different lipids, this can be achieved by culturing cells in the presence of isotope-labeled precursors leading to their metabolic incorporation into the lipid of interest, which would have the same chemical structure as its unlabeled analogue.

Chemical mapping of the plasma membrane displayed 200 nm domains showing a significant sphingolipid enrichment ³⁴ ( Figure 1c,d). These domains were further non-randomly assembled in patched regions that were about 3–10 μm in size ^{34, 35} . Simultaneous chemical imaging of isotope-labeled cholesterol revealed that cholesterol, in contrast to sphingolipids, distributed in an apparent homogeneous fashion on the dorsal/upper membrane ³⁵ analogously to a recent dynamic study ³⁶ . Despite the non-overlapping spatial distribution, cholesterol depletion did disperse the sphingolipid-enriched domains. Actin depolymerization had a more dramatic effect, suggesting a link between lipid organization and the actin architecture ^{35, 37} . With the possibility to include specific protein labeling along with the mapping of lipid components at the nanometer scale, this technique will continue to contribute to our understanding of the spatial distribution of chemistry in the membrane ^{38, 39} .

The benefit of label-free methods is avoiding the possible influence of an attached label on the behavior of the specific protein or lipid of interest; for any lipid, given the mass ratio of a fluorescent label to the mass of the lipid species, this perturbation is likely to be substantial. Nevertheless, to get more detailed knowledge of how a cell constructs complexes in a membrane, molecular recognition with a high signal-to-noise ratio in an aqueous scattering milieu, a feature that label-free methods still lack, is essential.

Increase of signal-to-noise ratio is attained by specifically targeting or labeling the molecule of interest with, for example, an antibody, genetically, or by chemically incorporating a contrast agent. If the labels are carrying electron-dense material ²² or conjugated with gold nanoparticles ⁴⁰ , they can be visualized with an electron microscope. In general, however, fluorescence light microscopy is used where individual targets of interest are coupled to fluorophores.

Focusing of light, however, is inherently diffraction-limited. With lens-based optics, light cannot be focused better than about 200–300 nm and individual objects that are spaced closer cannot be distinguished as unique objects anymore. To overcome this concentration limit ⁴¹ , several approaches have come up in the last decade to either (temporally) dilute the observed molecules (stochastic super-resolution microscopy ^{42, 43} ) or decrease the observation volume (targeted super-resolution microscopy ^{44, 45} ). Single-molecule imaging ⁴⁶ has opened up a major avenue not only to observe the localization of single fluorophores at very high spatial resolution but also to study the biochemistry of individual species to derive ensemble properties of molecules inside a cell.

Optically interrogating the dynamic behavior of single molecules in their highly concentrated presence on the plasma membrane is made feasible by isolating a fluorescently labeled representative. Although the membrane components still move in their natural environment, their dynamics can now be characterized by recording the motion of a number of such ‘single representatives’ on a camera. If the distance between the individual molecules in each image is larger than the diffraction limit, their positions can be determined with nanometer precision ^{47, 48} . This accuracy of determining its center-of-mass is essentially inversely proportional to the square root of the number of photons emitted ⁴⁷ . The positions of multiple fluorescent spots can be identified and related to their position in earlier images to build up their time trajectories ⁴⁹ . The number of molecules can be tuned via the concentration of externally added specific markers, photo-activation of only a subset of fluorescent molecules ^{50– 52} , or the photo-bleaching of a well-defined area followed by the sparse diffusion back in the observation area ^{53, 54} . At the other end, technical advances in hyper-spectral detection should increase single-particle discrimination allowing an increase in concentrations of single molecular representatives ⁵⁵ .

Recording a sufficient number of tracks or a single molecule for a sufficiently long time builds up the statistical behavior of the membrane components in terms of the diffusion coefficient or type of mobility ⁵⁶ . Individual trajectories pooled into a distribution of diffusion coefficients can then be related to the functional/affinity state of a receptor ^{57, 58} . Local changes in the individual trajectory can be mapped out on the cell to indicate the nature of the area traversed ^{59, 60} , in terms of diffusion ⁵⁰ , confinement regions ⁶¹ , or local energetic changes ⁶² . Examining the individual tracks of receptors as they diffuse in the plasma membrane revealed that they could become obstructed by lipid domains ⁶³ , protein-protein interaction ⁶⁴ , tetraspanin network ⁶⁵ , glycan structures ^{59, 66} , or the actin cytoskeleton ^{67– 69} ( Figure 2a,b).

Detection of changes due to interactions or confinement within boundaries of a compartment is reflected in changes in the molecular diffusion coefficient, which in turn depends on interaction strengths, acquisition speed, and signal-to-noise ratio (localization accuracy) ⁷⁰ . On the other hand, prior information on molecular mobility could facilitate teasing out a subset of molecules without effectively diluting the experiment. If the subset (of interest) moves significantly slower because of an activation event or substrate binding, it is possible to experimentally deconvolve out the contribution of individual players in the reduction of mobility ^{60, 71} . Using relatively long exposure/integration times, fast-moving fluorescent molecules blur into the background while slower moving molecules emit photons from the same diffraction-limited volume and therefore can be localized.

The stochastic cycling of dyes between fluorescent on-states and non-fluorescent off-states is an unfavorable property for single-molecule tracking because the single molecule might become undetectable in several frames and get lost during the trace reconstruction. Alternatively, if tuned properly, this cycling between states can be used to temporarily dilute highly concentrated samples. The challenge is to have at each given time only a subset of molecules in the on-state and determining the center-of-mass for each molecule before they switch off again. If this process is repeated many times, all of the calculated positions can be used to reconstruct a “super-resolution” image ⁷² . This indeed is the concept of the localization techniques called stochastic optical reconstruction microscopy (STORM) ⁷³ and (fluorescent) photoactivatable localization microscopy, or (f)PALM ^{42, 43} . Whereas STORM is essentially based on organic dyes that reversibly switch between on- and off-states ^{74– 77} , PALM is based on engineered fluorescent proteins ^{78– 80} .

Photo-localization-based super-resolution is excellent in determining sub-resolution complexes such as the endocytic clathrin-coated pits ^{81, 82} , microtubular structures ^{83– 85} , cytoskeletal structures ^{86, 87} , and adhesion plaques ^{88, 89} ( Figure 3). Quantitative analysis of super-resolved domains in the plasma membrane in terms of absolute number of molecules is challenging because of blinking ⁹⁰ and activation efficiency ⁹¹ . Nevertheless, analytical methods such the pair-correlation ⁹² and Ripley’s K function ^{93, 94} have allowed a certain degree of quantification, but this is strongly dependent on the assumptions employed in applying these statistical analyses to the data.

Since photo-localization-based super-resolution is based on a random spatial sampling of the structure, activating or switching light pulses are not necessarily required. In fact, it is possible to make use of the intrinsic trait of fluorophores to get temporarily trapped in a dark state, blinking. By engineering fluorophores that have longer dark states ⁹⁵ or exploiting the known on-off blinking of quantum dots ⁹⁶ , the chances that nearby emitters are both in an on-state decrease. On the other hand, one could tune incorporation rates of fluorescent species to the membrane (proteins); in a bath of freely diffusing fluorescent ligands, only the temporarily bound ligands will be detected, essentially taking advantage of mobility difference between bound and unbound ^{97– 99} .

Super-resolution imaging in the context of breaking Abbe’s diffraction limit has been achieved by decreasing the observation volume below the diffraction limit of light. This has been mainly accomplished with (a) use of near-field geometries or restricted physical apertures to confine the excitation volume or (b) the clever use of lasers to selectively deplete excited fluorophores in all except the very center of the optical volume to confine the emission volume.

(a) Near-field optics. By physically confining the light inside a very small aperture of 50–150 nm in diameter, light propagation is discontinued and the electromagnetic fields become restricted to the aperture (i.e., to the near field). The light intensity exponentially decays away from the aperture, producing essentially a nanoscopic excitation source. For imaging purposes, such a sub-wavelength aperture is created in a tapered aluminum-coated optical fiber. The image can then be built up by raster-scanning the aperture in close proximity to the sample and in fact this was one of the first approaches to obtain super-resolution images ⁴⁵ . Because the rendering of the image is independent of the photo-physical properties of the sample/fluorophore, it allows near-field scanning optical microscopy (NSOM) to obtain quantitative information at the nanometer scale. Additionally, the same physical aperture confines multiple wavelengths and the technique is therefore free from chromatic aberrations ^{100, 101} . Colocalization ¹⁰² among multiple chemical species (due to biochemical interaction), random scattering ¹⁰¹ (due to the lack of interaction), and segregation ¹⁰³ are not the only modes of organization. Indeed, multicolor super-resolution imaging revealed multi-domain proximity on the order of 50–150 nm as another tendency ^{101, 104, 105} . Diffraction-limited techniques would erroneously identify such proximity as colocalization, showing the merit of any super-resolution technique. Proximity does not preclude interaction, and quantitative analysis showed that integrin activation could bias the glycosyl phosphatidylinositol-anchored protein (GPI-AP) organization to a more clustered state ¹⁰⁵ . This more detailed quantification was granted by the possibility of having single-molecule sensitivity at the nanometer scale. Practical resolution of NSOM, however, is limited to approximately 50–70 nm, driving the field toward the use of optical antennas. Optical antennas, analogous to their radio frequency equivalent, convert freely propagating electromagnetic radiation into localized energy, and vice versa. Initial experiments have shown resolutions of 30–50 nm of proteins on a plasma membrane using a gold nano-particle ¹⁰⁶ or a sculpted monopole ¹⁰⁷ as photonic antenna. Recent advances in antenna design provided simultaneous multicolor localization accuracies well below 1 nm with low photon budgets ¹⁰⁸ , showing tremendous potential for nanoscale sensing or imaging ¹⁰⁹ . Near-field imaging or spectroscopy, however, is confined to sample surfaces that are accessible to the physical aperture/probes, making in vivo imaging inside cells a very difficult proposition.

(b) Stimulated emission depletion. A different strategy toward true super-resolution is to confine the emission instead of the excitation ¹¹⁰ . Its principle is based on the positionally deterministic switching of the fluorophore state in contrast to the stochastic switching for localization-based super-resolution ^{111, 112} . Stimulated emission depletion (STED) microscopy is founded on depleting the excited state of a fluorophore by stimulating the excited fluorophore to emit a photon of specified wavelength. By creating a highly intense donut-shaped emission depletion region around the confocal excitation volume, only the fluorophores in the center of the donut are spontaneously emitting in the detected wavelengths. By increasing the intensity of the depletion donut, the resolution of the microscope is increased. STED imaging has been used to identify and quantify cluster sizes in cell membranes ^{59, 113– 115} . The cluster sizes ranged from 50 to 160 nm, and STED experiments on membrane sheets indicated that the protein clusters were fine-tuned by electrostatic interactions and that these clusters are further assembled in relatively stable multi-protein assemblies ¹¹⁶ . Further development of the technique toward parallelization ^{117, 118} , multicolor acquisition ^{119– 121} , and different illumination schemes ^{122, 123} will definitely increase imaging capacity. Better spatial resolution, however, is accompanied by an increased on- off cycling load on the fluorophore during scanning, requiring further progress in fluorophore engineering.

Instead of demanding the heavy burden of photo stability from the fluorophore during multiple rounds of irradiation in the course of image build-up, one could allow the molecule itself to diffuse through the observation volume. During this passage, a fluorescent molecule will cause an intensity burst that is inversely proportional to the number of molecules in the observation volume. When sufficient numbers of molecules have passed, the detected intensity fluctuations can be autocorrelated, designating the technique as fluorescence correlation spectroscopy (FCS). The time at which this autocorrelation function decays to half its original value corresponds to the characteristic timescale at which the molecules move through the observation volume.

There is a linear relationship between the diffusion time and the area of observation for 2D diffusion in the plasma membrane. The slope of this relation is inversely proportional to the apparent diffusion coefficient, and the time-axis intercept, obtained from extrapolation, is indicative of confinement ¹²⁴ . According to this methodology, particles diffusing in the membrane can be divided in three major categories: (a) random diffusive, (b) domain interacting, and (c) meshwork constrained ^{125, 126} . Exploitation of the FCS diffusion-law methodology found that sphingolipid- and cholesterol-dependent nanoscale domains are crucial for signaling ¹²⁷ .

Similar methods of optically diluting the sample, as described above, can be employed for FCS ¹²⁸ . More powerful is the combination of super-resolution techniques that confine the observation volume with FCS since this allows the registration of dynamics at the nanometer scale ^{129– 132} . Mobility characteristics at the nanometer scale do not have to be extrapolated anymore ^{125, 126, 133} but can be directly measured ^{130, 134, 135} ( Figure 2c,d). In fact, extensive research using the tunable nanoscopic observation volume provided by STED indicated that fast-moving lipid analogues exhibit distinct modes of mobility that can be divided in three classes ^{130, 136– 139} : (a) weak interactions of phosphoglycerolipids, (b) cholesterol-assisted binding mediated by the ceramide group, and (c) hydroxyl headgroup-assisted cholesterol-independent binding. In the future, bridging length scales with, for example, camera-based FCS ^{140– 142} or spatio-temporal image correlation spectroscopy ^{143– 146} should allow the visualization of how these fluctuating nanoscale assemblies can be stabilized to coalesce into functional signaling platforms ⁷ .

With the advent of reproducible nanofabrication techniques, a whole new field lies open for exploration. Engineered substrates can provide aperture-based ^{134, 147– 150} or optical antenna-based ^{151, 152} nanofocusing of light on conventional microscopes. By virtue of the cells adhering to the substrate, the plasma membrane is brought in the near field of various nanoscopic excitation sources. Each of these excitation hotspots can now be addressed to locally probe membrane dynamics down to 20 nm ¹⁵² or in a multicolor fashion ¹⁵⁰ .