Microscopy approaches to investigate protein dynamics and lipid organization

Abstract

The structure of cell membranes has been intensively investigated and many models and concepts have been proposed for the lateral organization of the plasma membrane. While proteomics and lipidomics have identified many if not all membrane components, how lipids and proteins interactions are coordinated in a specific cell function remains poorly understood. It is generally accepted that the organization of the plasma membrane is likely to play a critical role in the regulation of cell function such as receptor signalling by governing molecular interactions and dynamics. In this review we present different plasma membrane models and discuss microscopy approaches used for investigating protein behaviour, distribution and lipid organization.

• Membrane order
• microscopy techniques
• protein interaction

Introduction

Over the past decades, the lipid raft hypothesis has changed the way cell biologists view lipids and has postulated a role for lipid domains in protein organization and dynamics. These membrane domains are therefore often viewed as biophysically and biochemically discrete platforms. However, lipid domains remain controversial, mainly because biochemical assays are ambiguous and direct visualization has been proven difficult, suggesting that domains sizes are below the diffraction limit of optical microscopes. Therefore, single molecule and super resolution approaches are promising methods to investigate plasma membrane organization and they have the enormous potential to advance our knowledge of the lateral distribution of lipids and proteins in membranes. Here, we first summarize the models and concepts proposed for the organization of the plasma membrane, and then discuss the experimental evidence of heterogeneities in membrane order, the fluorescence techniques used to measure protein interactions and dynamics and finally outline the potential of super-resolution for membrane biology. In addition, we give an overview of the microscopy techniques discussed in Table 1.

Table 1. Summary of microscopy techniques.

Method	Principle	Advantage	Limitation	Applications	Reference
Phasor analysis to fluorescent lifetime microscopy (FLIM)	FLIM detects changes in the lifetime of a fluorescent dye or protein in each pixel of an image and the phasor approach transforms it into a fit free 2D coordinate system.	• Does not require a priori knowledge of the number of components that need to be unmixed • Applicable to live cell imaging	• Slow acquisition speed • Does not provided information on domain size and geometry	• Membrane order measurements to detect lipid phases • Useful for FRET measurements	(Digman et al., 2008) (Owen et al., 2012b) (Golfetto et al., 2013)
Fluorescent resonance energy transfer (FRET)	A donor fluorophore non-radiatively transfers energy to an acceptor fluorophore by means of intermolecular long-range dipole-dipole coupling.	• Can detect interactions <10 nm • Suitable for live cell measurements • Applicable to confocal microscopy • Can be combined with FLIM	• Photobeaching • Probe orientation • Dipole moment • Must incorporate the distance of spacers	• Measures protein-protein interactions • Protein association with domains • Receptor oligomerization	(Clegg et al., 2005) (Varma & Mayor, 1998) (Kenworthy & Edidin, 1998) (Kenworthy et al., 2000) (Siegel, 2000)
Fluorescent recovery after photobleaching (FRAP)	Fluorescent molecules are irreversibly photobleached by a high power laser in the region of interest of the cell and the recovery of fluorescence from surrounding non-bleached fluorescence molecules monitored.	• Easy to perform • Live cell measurements • Applicable to confocal microscopy	• Ensemble measurement • Cannot reveal different diffusion modes	• Monitoring the mobility of proteins and fluorescent lipid analogues	(Ishikawa-Ankerhold et al., 2012)
Fluorescent correlation spectroscopy (FCS)	Fluctuations in fluorescence intensity caused by the diffusion of fluorescent molecules through a small focal volume are observed. The corresponding autocorrelation function can be fitted to quantify diffusion coefficients and molecule concentrations within the focal volume.	• Single molecule technique (nM fluorophores concentrations) • Small observation volume (<10 µl) • High temporal resolution with large range (µsec-sec) • Live cell measurements • Can be applied to a single point measurement or line scan	• Does not provide spatial information • Long acquisition times • Data fitting and interpretation is complex	• Measures: • Protein–protein interactions • Diffusion mode and coefficient • Chemical kinetics and photo-physical properties	(Benda et al., 2003) (Hinde et al., 2013) (Clegg et al., 2005) (Enderlein & Gregor, 2005) (Bohmer, 2002) (Benda et al., 2005) (Benes et al., 2004)
Raster image correlation spectroscopy (RICS)	A time series of frames is acquired using a raster scan, where pixels in each image are measured at different times and therefore the spatial correlation of the whole image contains details about dynamic processes occurring on different time scales.	• Relative simple method for diffusion measurements • Can differentiate slow diffusion from binding events • Large temporal range (µsec-sec) • Live cell measurements • Applicable to confocal microscopy	• Averaging over pixels, lines and frames within one measurement limits spatial resolution	• Measures protein behaviour such as formation of protein complexes and binding dynamics to an immobile or mobile structure.	(Digman et al., 2005) (Digman et al., 2009)
Single particle tracking (SPT)	Position of sparsely labelled single molecules is precisely mapped in time and space	• High spatial and temporal resolution • Live cell measurements	• Labelling of individual molecules is required • Small number of molecules is measured • Requires complex analysis	• Molecular mobility and detection of transient confinement • Hop diffusion (requires ultrafast cameras)	(Kusumi et al., 2011)
Stimulation emission depletion (STED)	Sample is scanned with two overlapping lasers. Molecules located in the outer part of the diffraction-limited excitation area are depleted by stimulated emission (doughnut shape beam) so that only fluorophores from the central region of the excitation spot are able to emit fluorescence.	• Super-resolution technique with 30–150 nm resolution (depends on STED laser power) • Built on laser-scanning microscopy • Can be combined with FCS	• Limited choice of fluorescent probes • Photo-bleaching due to high laser paper • Specialized microscope set-up	• Imaging with spatial super-resolution	(Geumann et al., 2010) (Halemani et al., 2010) (Singh et al., 2012) (Wurm et al., 2011)
STED-FCS	Combines spatial resolution of STED with temporal resolution of FCS.	• High spatial resolution of 30–150 nm (due to reduced excitation volume) • High temporal resolution (<µsec) • Live cell measurements	• Limited choice of fluorescent probes • Photo-bleaching due to high laser paper • Specialized microscope set-up • Long acquisition times • Data fitting and interpretation is complex	• Diffusion of fluorescent lipid analogues • First detection of differences in lipid diffusion and lipid nanoclusters	(Eggeling et al., 2009) (Mueller et al., 2011)
Stochastic optical reconstruction microscopy (STORM) and photoactivated localization microscopy (PALM)	Localization microscopy relies on the stochastic activation of conventional fluorophores (STORM) or photoswitchable fluorescent proteins (PALM). During one cycle, fluorescent molecules are photoswitched/activated, imaged and then photobleached, and the cycle repeated several thousand times.	• Localization of individual molecules with 10–25 nm precision in lateral direction and 60–100 nm in axial direction • Single molecule data	• Long acquisition time • Not suitable for all fluorophores • Photo-physical properties of fluorophores need to be understood for molecular counting • Complex analysis since each molecules needs to be fitted	• Molecular quantification of protein distribution and clustering	(Williamson et al., 2011) (Rossy et al., 2013) (Sengupta et al., 2011)

Models of the plasma membrane

In 1972 Singer and Nicolson proposed the fluid mosaic model (Figure 1A) that describes the plasma membrane as a two-dimensional liquid in which proteins are homogenously distributed in the lipid bilayers. This model allows for rapid lateral diffusion of lipids and proteins in the membrane but essentially describes the membrane as a passive matrix (Singer & Nicolson, 1972). Cell membranes are however heterogeneous and have distinct and diverse membrane features and domains such as membrane protrusions, cell-cell junctions or coated pits (Simons & Ikonen, 1997). In addition, molecular crowding, the existence of lipid and protein complexes that themselves may not be randomly distributed and local membrane deformation due to differences in lipid fluidity and proteins rigidity give the cell membrane a rather more complex architecture than simple lipid bilayers (Engelman, 2005).

Figure 1. Different plasma membrane models. (A) Fluid mosaic model presents the plasma membrane as a two-dimensional liquid in which proteins (orange and blue shapes) are homogenously distributed in the lipid bilayers (dark blue). This model allows the possibility of rapid lateral diffusion of lipids and proteins within the membrane. (B) The lipid raft model proposes the existence of small regions (green shapes) enriched in cholesterol (red shapes) and sphingolipids. (C) The picket fence model suggests that transmembrane proteins act as “pickets” (orange pink and blue shapes) and the membrane skeleton as “fences” (dark orange net). The membrane skeleton confines the diffusion of membrane components due to collision. The membrane is depicted from cytosolic side. (D) The lipid shell model suggests that 25% of the cell membrane surface contains proteins (orange shapes) which are surrounded by lipid rings (purple and pink). (E) The models are exclusive and it is likely that different various membrane domains and compartments coexist in the plasma membrane.

Simons and colleagues modified the fluid mosaic model with the concept of lipid-lipid interactions that led to the lipid raft hypothesis (Figure 1B) to describe protein trafficking in epithelial cells in 1997 (Simons & Ikonen, 1997). Seven years later, Kusumi et al. proposed the so-called picket-fence model (Figure 1C) in which transmembrane proteins (“pickets”) and the membrane skeleton (“fences”) confined the diffusion of membrane components due to collision (Kusumi et al., 2004). Membrane skeleton-dependent hop diffusion was directly observed for various receptors such as the transferrin receptor, α2-macroglobulin receptor, G-protein-coupled receptor or μ-opioid receptor as test molecules (Akihiro & Sakot, 1996; Sako et al., 1998; Suzuki et al., 2005; Tomishige et al., 1998). When the lipid L-α-dioleoylphosphatidylethanolamine (DOPE) was tagged with a colloidal gold particle, it localized in the outer leaflet of the plasma membrane and underwent hop diffusion that was dependent on the actin-based membrane skeleton (Fujiwara et al., 2002). It has also been shown that microdomain formation and sub-diffusion within the membrane might be enhanced by the association of actin filaments with the membrane bilayers (Ehrig et al., 2011; Garg et al., 2009). While the proposed membrane models suggested spatial organizations for proteins, lipids the membrane skeleton, the size of the lipid microdomains remains unknown. The Keystone Symposium in 2006 added a spatial dimension: “Small (10–200 nm), heterogeneous, highly dynamic, sterol- and sphingolipid-enriched domains that compartmentalize cellular processes. Small rafts can sometimes be stabilized to form larger platforms through protein-protein and protein-lipid interactions” (Pike, 2006). The postulated small size and dynamic nature of rafts makes them challenging to observe directly, given that their dimensions are below the diffraction limit of light microscopy.

The model called the lipid shell model (Figure 1D) suggests that ∼25% of the cell membrane surface contains proteins (20 000–35 000 proteins/μm²), which are surrounded by lipid rings (Jacobson et al., 2007). The number of integral membrane proteins in the endoplasmic reticulum and Golgi apparatus has been estimated as ∼30 000 molecules per μm² (Quinn et al., 1984), the number of rhodopsin in the rod outer segment was assessed as ∼30 000 per μm² (Liang et al., 2003; Liebman & Entine, 1974), and as a final example, the number of GPI-anchored folate receptor in caveolae following the addition of crosslinking antibodies was calculated as 32 000 per μm² (Rothberg et al., 1990). Thus these estimations show that the plasma membrane contains a much higher protein concentration than artificial membranes. In 2011, Kusumi and colleagues also suggested that the organization of the plasma membrane occurs at different length scales. They distinguished three main compartments (a) “membrane compartments” which are 40–300 nm in diameter, (b) “raft domains” with a size between 2 and 20 nm in diameter, and the smallest (c) “dimers/oligomers and greater complexes of membrane-associated and internal membrane proteins” whose size is 3–10 nm (Kusumi et al., 2011). Similarly to co-existing membrane compartments, it should be pointed out that the above-presented models do not exclude each other (Figure 1E) and different domains may coexist in the plasma membrane.

Membrane order and the challenge to image lipid organization

A central aspect of the lipid raft hypothesis is that the cell membrane exhibits two co-existing lipid phases: The liquid ordered phase (Lo) that mainly consists of saturated lipids and sterols, and liquid disordered phase (Ld) that is mainly made up by unsaturated lipids. The difficulty in directly imaging lipids is that a bulky fluorophore conjugated to a lipid often affects the partitioning and diffusion of the lipid analogue. Sezgin et al. showed that model systems such as giant unilameral vesicles (GUVs) are characterized by limited complexity and have different physical properties compared to intact the plasma membrane or giant plasma membrane vesicles (GPMVs). They observed that fluorescent “raft” lipid analogues that did not partition into the Lo phase in GUVs did do so in GPMVs and concluded that the fluorophore geometry and chemical properties have a significant influence on the partitioning behaviour of lipid analogues (Sezgin et al., 2012).

An alternative approach is to use fluorophores that partition into both phases and have fluorescent properties that are sensitive to membrane order. The fluorescence properties of the environmentally-sensitive membrane dyes are, for example, dependent on the polarity of their local solvent (Parasassi et al., 1998). Since penetration of polar water molecules into the non-polar bilayer interior depends on the degree of lipid packing, polarity-sensitive dyes can sense the local molecular environment and thus help to distinguish Lo and Ld phases in the membrane (Figure 2A).

Figure 2. Spectral and lifetime methods of analysis to detect lipid phases. (A) The plasma membrane is depicted with the membrane dye Laurdan (orange shapes) in a random distribution between liquid disordered phase (Ld, blue lipids) and liquid ordered phase (Lo, green lipids). (B) Schematic emission spectra of lipid-phase sensitive dye, Laurdan, with green and purple spectra corresponding to Lo and Ld phases, respectively. (C) Typical fluorescence lifetime decays correspond to liquid ordered (dotted line) and liquid disordered (solid) phases, respectively. (D) Transformation of a spectrum or decay into a phasor (vector) plot. Spots represent data for Lo (lower) and Ld (upper) phases.

One of the most common fluorophores for imaging lipid domains is Laurdan (6-lauryl-2-dimethylamino-naphthalene). It is a polarity-sensitive dye created by Weber and Farris in 1979 as a derivative of Prodan (Weber & Farris, 1979). Laurdan’s fluorescent group is located ∼10 Å from the centre of the bilayer (Antollini & Barrantes, 1998) while the Laurdan dipole molecules are aligned with the phospholipid head groups and the first carbon atom of the acyl chains in the phospholipid bilayers (Bagatolli et al., 1999). Laurdan’s fluorescent emission properties are dependent on the extent of water penetration into the bilayer (Figure 2B), which is different between ordered and disordered membrane domains (Gaus et al., 2003; Owen et al., 2010; Parasassi & Gratton, 1995; Parasassi et al., 1997, 1998). Laurdan’s lifetime is also membrane order-dependent (Figure 2C) where the Ld phase is characterized by a ∼4 ns lifetime and the Lo phase by a much longer lifetime of ∼6 ns (Parasassi et al., 1993). Another fluorescent probe often used to probe membrane order is the styryl di-4-ANEPPDHQ dye, which was designed as a voltage-sensitive dye to monitor electrical activity in cells and tissues (Obaid et al., 2004) but can also be used to differentiate lipid phases in model membranes and intact cells (Jin et al., 2005, 2006; Owen et al., 2012a,b). However di-4-ANEPPDHQ differs from Laurdan in the manner of membrane attachment and fluorescent lifetime. Di-4-ANEPPDHQ is attached to the membrane with its acyl heads being inserted deep into the membrane and its two positive charges confer a lesser ability to move between the inner and outer leaflets than Laurdan (Hof, 2012). Additionally, the lifetime of di-4-ANEPPDHQ (1.8 ns) is shorter in the Ld phase, and longer in the Lo phase (3.5 ns) compared to Laurdan (Owen et al., 2006). As an alternative to the previously described solvent-sensitive dyes is a novel group of voltage-sensitive hemicyanine dyes which exhibit electrochromism. These dyes can also be used to quantify membrane order and in contrast to di-4-ANEPPDHQ and Laurdan, hemicyanine dyes have a wide one-photon excitation spectrum that ranges between 440–670 nm (Kwiatek et al., 2013; Yan et al., 2012).

The most common method to investigate membrane order is to quantify the shift in emission spectra of a membrane order sensitive dye (as depicted in Figure 2B for Laurdan) by Generalized Polarization (GP) analysis (Gaus et al., 2003; Parasassi et al., 1997, 1998). The Generalized Polarization function proposed by Parasassi et al. in 1990 is defined as the normalized fluorescence intensity ratio of the spectral channels that correspond to ordered and disordered phases. GP values range between −1 for very fluid membranes to +1 for the most ordered membrane and are a measure of the relative water content in the membrane, which is directly related to membrane order (Parasassi et al., 1997). GP values do not depend on the probe concentration (Parasassi et al., 1990). The main advantage of GP analysis is that the method is robust and data acquisition can be acquired on confocal and 2-photon microscopes.

The phasor approach to fluorescence lifetime imaging microscopy (FLIM) is an alternative method to the GP analysis of emission spectra for monitoring the physical environment of membrane order sensitive dyes. FLIM detects changes in the lifetime of a dye in each pixel of an image (as depicted in Figure 2C for Laurdan) and the phasor approach transforms FLIM data into a two-dimensional coordinate system (Figure 2D), which allows us to distinguish between a mixture of immiscible phases and a homogeneous environment with intermediate order. It does not require a priori knowledge of the number of components that need to be unmixed (Digman et al., 2008). The major limitation of FLIM detection is the acquisition time. To determine the lifetime of a fluorophore with accuracy requires many more photons than are required for spectral imaging. This disadvantage makes FLIM detection of dynamic processes more difficult than spectral-based methods of detection (Owen et al., 2012b).

Owen et al. used the FLIM phasor approach to unmix the lifetime from ordered and disordered phases and, by a linear combination of liquid ordered and disordered lifetimes, quantified the membrane coverage in live cell membranes stained with membrane order-sensitive dyes (Laurdan or di-4-ANEPPDHQ). From the lifetime changes that were acquired in one spectral channel, it was found that the majority of the plasma membrane is covered by the Lo phase (Owen et al., 2012b). More recently, Laurdan lifetime changes were measured with a modified two-channel detection scheme with the objective to spectrally separate detection of the Lo phase with channel 1 (460/80 nm) and the Ld phase with channel 2 (540/50 nm). Using the phasor analysis to analyze the lifetime recorded in each channel, this detection scheme resolved two different trajectories that were attributed to changes in polarity (channel 1) versus changes in cholesterol content (channel 2). This spectrally resolved approach to FLIM detection of Laurdan demonstrates an increasing dipolar relaxation effect cause by rotating water molecules in the membrane and the polarity of the environment changes the Laurdan lifetime. Based on these observations, Golfetto et al. claimed that Laurdan, when combined with spectrally resolved detection and the phasor FLIM approach can be used as a biosensor for cell membranes to detect and distinguish membrane fluidity from cholesterol content (Golfetto et al., 2013).

The capability of the phasor method was further developed in recent work by Fereidouni et al. (2012). The phasor plot (Figure 2D) was extended to spectral imaging by translating the phasor plot from the lifetime domain to a wavelength representation. The spectral phasor method is more robust than standard GP analysis because the entire emission spectrum is considered through the collection of photons in many spectral channels rather than in just two channels, as is done with the traditional GP analysis. In summary, improvements have been made to quantify membrane order in cells more precisely and it is now possible to unmix the fluorescence signatures from lipid phase-sensitive dyes with the aid of phasor plots. However, these methods lack spatial resolution so that individual lipid domains and hence domain geometry and lifetimes cannot be revealed.

Protein interactions and dynamics: Energy transfer, fluctuation methods and photo-bleaching

Given that most cell functions are controlled by protein behaviour (Rossy et al., 2014), one may wish to map protein distributions, interactions and dynamics within the membrane and infer the existence of domains from protein behaviour rather than image the membrane domains themselves. Prior to the advent of super-resolution microscopy (which is required to measure protein distribution and clustering on the nanometre scale, see below), advanced microscopy methods have been developed to quantify protein mobility, diffusion and interactions in cell membranes. The first method of detection discussed here employs fluorescence resonance energy transfer (FRET) (Figure 3A) as the readout of protein interaction. FRET is a non-radiative energy transfer from an excited chromophore (the donor, D) to another chromophore (the acceptor, A) by means of intermolecular long-range dipole-dipole coupling. The efficiency of the energy transfer critically depends on the Förster distance between the donor and acceptor and their fluorescence properties. The FRET efficiency of a donor-acceptor interaction is relative to the donor’s quantum yield and lifetime, the donor and acceptor orientation (kappa factor), the degree of donor and acceptor spectral overlap, and is inversely proportional to the sixth power of distance between donor and acceptor, which usually has to be between 1 and 10 nm (Figure 3A, right). When a FRET interaction occurs, the donor fluorescence emission and lifetime are quenched, and the acceptor lifetime lengthened. Any one of these changes – quenching of the donor lifetime, decrease in the rate of donor photobleaching or sensitized acceptor fluorescence (Lippincott-Schwartz et al., 2001) – can be used for FRET detection but detection of changes in the donor lifetime has emerged as the preferred property to monitor FRET. This is because FLIM-FRET has the advantage that non-interacting species (un-quenched donor and acceptor) can be easily distinguished from the quenched donor lifetime. This is not possible with fluorescent emission-based methods since the donor and acceptor fluorescence spectra are linear combinations with or without FRET interactions (Chen et al., 2003; Hof, 2012). The most commonly used donor-acceptor pair consists of Cyan Fluorescent Protein (CFP) and Yellow Fluorescent Protein (YFP), in which the emission spectrum of the donor (CFP) highly overlaps with that of the acceptor (YFP).

Figure 3. Fluorescent approaches for detection of protein interactions and dynamics. (A) Fluorescence resonance energy transfer (FRET) with the most commonly used fluorescent protein pair, CFP-YFP. Middle scheme: FRET does not occur because of a large distance between the CFP and YFP proteins. Right scheme: FRET occurs when the distance (R₀) between donor and acceptor is below 10 nm. (B) Fluorescence correlation spectroscopy (FCS). Middle scheme: The cross correlation function between two proteins is constant because the proteins are not assembled into complexes or the distance between them is too large. Right scheme: the cross correlation function indicates the presence of a protein complex. (C) Fluorescence recovery after photobleaching (FRAP). Middle scheme: a region of interest is photobleached region (black square). Right scheme: recovery of fluorescence intensity.

FRET is routinely used to investigate protein-protein interactions and membrane domain organization. For example, FRET was used to investigate Fas and how this cell surface receptor transduces an apoptotic signal upon binding to its trimeric ligand FasL. A FRET study revealed that Fas exists as a pre-assembled trimer on the cell surface, in contrast to previous assumptions that Fas is a monomer which assembles into trimers after FasL binding (Siegel, 2000). FRET was also used to investigate protein association with membrane domains. If a protein is clustered in a pre-existing lipid domain, the FRET efficiency between individual proteins would be controlled by the domain size and occupancy but relatively independent of the protein expression level, at least at high expression levels. However, if the protein is randomly distributed in the plasma membrane, the distance between proteins will strongly depend on the protein expression level. One out of three studies observed FRET efficiencies consistent with raft-induced clustering of the glycosylphosphatidylinositol-anchored protein (GPI-AP), folate receptor (Varma & Mayor, 1998), while two others claimed that the organization of GPI-AP and glycosphingolipids (GSL) was random (Kenworthy & Edidin, 1998; Kenworthy et al., 2000). The major advantage of FRET is that it can detect protein interactions because energy transfer is typically limited to distance of less than 10 nm. However, FRET is also limited by several photo-physical factors such as photobleaching, probe orientation and dipole moment (Clegg et al., 2005) and data acquisition is relatively slow so that short-lived interactions may be missed.

An alternative method to probe protein diffusion and protein-protein interactions in live cells is fluorescence correlation spectroscopy (FCS) (Figure 3B). FCS monitors fluctuations in fluorescence intensity caused by the movement of fluorescent molecules through a small focal volume (He & Marguet, 2011; Magde et al., 1972). The autocorrelation function of these fluctuations can be fitted to quantify diffusion coefficients and molecule concentrations within the focal volume. FCS also gives information on molecular mobility and photophysical and photochemical reactions, such as translational diffusion, rotational diffusion (Benda et al., 2003), chemical kinetics (Clegg et al., 2005; Hinde et al., 2013), photon anti-bunching (Bohmer, 2002; Enderlein & Gregor, 2005), molecular interactions (Benda et al., 2005) and conformational fluctuations (Benes et al., 2004). FCS has been employed as a tool to investigate the mobility of proteins. Recently Hinde et al. combined two-channel FCS along a line scan (to introduce a spatial component) with FRET detection to investigate how protein-protein interactions change intracellular diffusion. The authors found from this multiplexed approach that the mechanism by which Rac1 and RhoA are organized and distributed upon cell migration is highly dependent on time and space and demonstrated that cross fluctuation analysis can reveal where and when protein-protein interactions occur in agreement with FRET (Hinde et al., 2013). While FCS is a single molecule technique and thus does not require high fluorophore concentrations, this technique does not provide spatial information (unless applied to line-scanning) and is restricted to observing protein dynamics without considering how these proteins may interact with their environment (Digman et al., 2005). Further, fitting and interpreting autocorrelation functions is not always straightforward and in cellular data, it is difficult to distinguish slow diffusing from transient binding.

Digman et al. developed a fluctuation-based method of analysis called raster image correlation spectroscopy (RICS). This technique is based on a raster scan where pixels in an image are measured at different times (microsecond separation along a line, millisecond separation between lines and second separation between frames) and therefore the spatial correlation of the whole image contains details about dynamic processes occurring on different time scales. The RICS approach can be used for investigating protein behaviour such as diffusion in cells including measuring binding dynamics, interactions with subcellular structures and quantifying local concentrations. The typical result of a RICS analysis is a map of the spatial distribution of protein complexes in the cell and their diffusion and/or binding properties (Digman et al., 2005). The RICS approach has been extended to a cross-correlation analysis (ccRICS) analogously to cross FCS, where two different proteins are monitored in two separate detection channels to assess interaction. ccRICS can be used to measure the diffusion of protein complexes, the spatial and temporal distribution of these complexes, and to evaluate the fractions of interacting molecules. In a recent study, Digman et al. used ccRICS to show the differences between cytoplasmic diffusion and binding of adhesion molecules. Moreover, the authors created maps of molecular interactions and their dynamics in the cell (Digman et al., 2009). RICS distinguishes between diffusion and binding which is the main advantage compared to other fluctuation methods.

In the above-described techniques photobleaching was considered as a limitation, however, it can also be utilized to measure protein mobility in the membrane by an approach called fluorescence recovery after photobleaching (FRAP) (Figure 3C). In this method, fluorescent molecules are irreversibly photobleached by a high power laser in the region of interest (ROI) of the cell and the recovery of fluorescence from surrounding non-bleached fluorescence molecules monitored. Analysis of FRAP data provides bulk information about protein kinetics including protein mobile fraction, M_f, and diffusion constant, D, as well as transport rate of binding/dissociation rate from other proteins. Mobility describes the fraction of fluorescent proteins which diffuse into the bleached region during the duration of the experiment, and the diffusion constant is a measure of the rate of protein movement in the absence of flow or active transport (Axelrod et al., 1976; White & Stelzer, 1999). Yechiel and Edidin used FRAP to investigate the lateral diffusion coefficient and the mobile fraction of a fluorescent lipid analogue and membrane proteins reviling the proof of protein-rich lipid domains existence in the plasma membrane and showing mobility of lipids and proteins in the domains (Yechiel & Edidin, 1987). Unlike FCS, FRAP is an ensemble measurement and therefore cannot reveal different diffusion modes at the single molecule level. FRAP parameters are influenced by a range of membrane scenarios and environments (Feder et al., 1996; Lippincott-Schwartz et al., 2001) and FRAP applications include monitoring the diffusion of plasma membrane proteins and phospholipids (Ishikawa-Ankerhold et al., 2012). However, given the small size of membrane domains and protein complexes, ensemble measurements like FRAP are difficult to interpret.

Nano-scale organization of membrane proteins and super-resolution microscopy

One of the major limitations of optical microscopy was until recently the diffraction limit that restricts the resolution to ∼250 nm in lateral direction and ∼600 nm in axial direction in confocal microscopy with ∼500 nm excitation. The diffraction limit means that it is impossible to detect small domains or individual proteins in the crowded milieu of the plasma membrane because neighbouring domains and proteins are closer than the resolution limit. Given that the diffraction limit is a law of physics, it is not an overstatement that super-resolution microscopy that can overcome this limit and achieve spatial resolutions of ∼20 nm is anticipated to revolutionize cell and membrane biology. However, spatial resolution is only one limiting factor in mapping small and highly dynamic domains such as lipid rafts. Temporal resolution can also be a limiting.

Kusumi et al. considered the smallest raft to be a complex of at least three molecules including one cholesterol molecule. He suggested that time can be used to distinguish small rafts from their surrounding environment. The lipid complexes must be stable for at least 10 ns since this is the time needed for a lipid to relocate to a neighbour position (Kusumi et al., 2011). It has been shown that cholesterol tends to create clusters with a lifetime from 1–100 ns in unsaturated lipid domains (Pasenkiewicz Gierula et al., 1991; Subczynski et al., 1991). These insights were not achieved with super-resolution microscopy but with single particle tracking (SPT) where only a few molecules are labelled and tracked very precisely in time and space.

Suzuki et al. (2012) applied single-molecule tracking to study the organization of glycosylphosphatidylinositol-anchored proteins (GPI-APs), probably the most studied lipid raft marker in the plasma membrane. The authors described the dependency between the GPI-APs expression level and complex formation. At low expression levels, GPI-AP homodimers are formed, both through protein-protein interactions and stabilization by cholesterol/rafts. At higher expression levels, homodimer rafts are assembled into homo- or hetero-oligomer rafts of GPI-APs. It was also shown in this paper that actin filaments were not involved in homodimer raft formation. Sharma et al. used FRET to estimate the percentage of GPI-AP species that are presented in clusters as well as cluster sizes. Changes in anisotropy revealed that 20–40% of GPI-anchored proteins form clusters. Homo-FRET was used to reveal that on average, clusters contained four GPI-AP proteins while the remaining GPI-APs were monomers. Additionally the data showed that homodimers or homo-oligomers were preferentially formed over heterometric GPI-APs clusters. Interestingly this study also indicated that GPI-AP clustering was independent of the surface density of proteins (Sharma et al., 2004). A few years later, Goswami et al. (2008) used anisotropy to describe GPI-APs distributions at steady state. GPI-APs were organized at the nanoscale (<10 nm) and optically resolvable scales (≤450 nm). It was assumed that the large scale domains may present properties reminiscent of lipid rafts (Simons & Vaz, 2004). The authors also emphasized that GPI-APs nanoclusters are dependent on ATP and actin filaments (Goswami et al., 2008). Recently the same group suggested that GPI-APs proteins might form nanoclusters due to cortical actin asters (Gowrishankar et al., 2012). The coupling of actin asters to a GPI nanocluster requires a membrane protein that becomes actively or passively trapped in the aster.

New super-resolution microscopy techniques such as stimulated emission depletion (STED) (Figure 4B) (Hell & Wichmann, 1994), photo-activated localization microscopy (PALM) (Betzig et al., 2006) or stochastic optical reconstruction microscopy (STORM) can now resolve structures smaller than Abbe’s diffraction limit (∼250 nm) and thus new insights about membrane organization and protein behaviour may be revealed. In STED nanoscopy, the fluorescent dye molecules located in the outer part of the diffraction-limited excitation area are depleted by stimulated emission so that only fluorophores from the central region of the excitation spot are able to emit fluorescence spontaneously (Hell & Wichmann, 1994). Images of actin and tubulin strikingly reveal the difference in resolving power of confocal versus STED imaging (Bückers et al., 2011; Kasper et al., 2010; Willig et al., 2007). STED imaging can also detect protein aggregates. For example, STED images showed that the SNAP25 protein (which regulates membrane fusion) forms clusters independently of the functionality of the SNARE motif and binds to clustered syntaxin (Geumann et al., 2010; Halemani et al., 2010), and mitochondrial proteins Tom20, VDAC1 and COX2 have been shown to be distributed in nanoscale clusters (Singh et al., 2012; Wurm et al., 2011).

Figure 4. Super-resolution methods. (A) Features in the plasma membrane such as lipid rafts are below (green PSF) the diffraction limit (black PSF). (B) Stimulated emission depletion (STED). Left panel: PSF of confocal beam. Right panel: PSF of STED beam where in a doughnut shaped STED beam is overlaid onto a confocal beam (green) which reduces the focal volume. (C) Combination of STED and FCS. The autocorrelation function is shifted towards shorter time when STED is applied to FCS, which allows a precise quantification of the diffusion coefficient. (D) Photo-activated localization microscopy (PALM) or stochastic optical reconstruction microscopy (STORM). Single molecules are photo-activated so that there are a limited number of fluorescent events in each frame. The PSF of each fluorescent event is analysed to obtain the localizations of each molecule with nanometre precision. All localization are combined to produce a super-resolution image.

Recently, STED was combined with FCS to assess the diffusion of lipid molecules in nanoscale membrane domains. Reducing the focal volume with the STED beam (Figure 4C) revealed the mode of diffusion of phosphoethanolamine, which differed from that of sphingomyelin that becomes temporarily trapped in domains of <20–50 nm for <10 ms (Eggeling et al., 2009). Later work confirmed that short-chained unsaturated lipids diffuse more rapidly than long-chain saturated lipids, as predicted by the lipid raft hypothesis (Mueller et al., 2011). Since STED-FCS is an extremely sensitive approach; only with SPT with ultrafast cameras and STED-FCS is it possible to detect nanosized clusters of sphingolipids, cholesterol and proteins in cell membranes (Mueller et al., 2011) but STED-FCS is technically not easy to implement and the instruments are expensive. STED also used high laser power that could damage the cells under investigation and to date, only a limited number of fluorophores are available for STED imaging and STED-FCS.

PALM and STORM employ photoactivatable or photoswitchable fluorescent proteins (e.g. PA-GFP and mEOS) and photoswitchable dyes or dye pairs (e.g. Cy3–Cy5) to achieve super-resolution images of single molecules at high densities (Figure 4D). The major concept of these approaches is similar: Localization microscopy relies on the stochastic activation of fluorophores. During one cycle, fluorescent molecules are photoswitched/activated, imaged and then photobleached, and the cycle repeated several thousand times. Each fluorescent event is localized by fitting the point spread function (PSF) to the intensity profile. Merging all single molecule positions gives the final super-resolution image (Betzig et al., 2006; Rust et al., 2006). Our group used PALM to quantify the clustering of T cell signalling proteins (Rossy et al., 2013; Williamson et al., 2011) where clustering was mainly dependent on protein-protein interactions or caused by association with vesicles. Sengupta et al. provided a more complex picture of the protein organization in the plasma membrane over a wide range of spatial scales using pair correlation analysis of PALM data. They measured cluster size, density and variability of nanoscale clusters of GPI-AP, transmembrane proteins (Lat and vesicular stomatitis viral glycoprotein [VSVG]), and an inner leaflet lipid-anchored protein (Lyn) and found that clusters differed in size and number of proteins. The radius of an average GPI-AP cluster was smaller than 60 nm and contained on average ∼2–3 proteins. In contrast Lyn-PA-GFP molecules were formed into larger clusters (>60 nm) and included three or more proteins. An average of 30% of Lat-PA-GFP clusters had radii above 100 nm with 2–40 proteins. Finally the size of VSVG-PAGFP clusters were also above 60 nm in radius which concentrate on average of 2.8 ± 0.4 detected proteins (Sengupta et al., 2011). The major advantage of PALM and STORM is its ability to localize molecules with nanometre precision. This makes PALM and STORM very promising tools to investigate protein clustering below the diffraction limit but so far no consensus has emerged as to how proteins clusters in the plasma membrane are formed. Dynamic information is more difficult to obtain with single molecule localization microscopy and probes to map lipid organization are not yet available for these super-resolution microscopy approaches.

Conclusion

Newly developed microscopy methods with high spatiotemporal resolution will afford us a better understanding of the complexity of the plasma membrane. These methods can return information about the lipid phase organization (e.g. Lo or Ld phase), lipid and protein interactions (e.g. clustering) and dynamics (e.g. diffusion). Moreover it is possible to combine two methods such as STED-FCS to exploit the temporal sensitivity of FCS and the improved spatial resolution of STED. However, currently there are distinct probes for lipid organization and protein behaviour such as lipid phase sensitive probes (e.g. Laurdan), probes for protein-protein interactions (e.g. CFP-YFP FRET pair) or fluorescent proteins for single molecule localization microscopy (e.g. PA-GFP) and not every probe is compatible with each microscopy technique. This makes it difficult to use for example, the microscope method with highest spatial resolution, single molecule localization microscopy, for the detection of Lo and Ld phases in cell membrane. It also makes it difficult to compare and integrate data from different studies. There are still controversies and some discrepancies in the literature about fundamental parameters of lipid domains and how lipids influence protein dynamics. For example, the existence of hop-diffusion was suggested by Kusumi et al. using ultra-fast SPT (Kusumi et al., 2004) but was not observed by the Hell group with STED-FCS (Mueller et al., 2011). Similarly, whether protein dynamics is mainly regulated by protein-protein interactions (Douglass & Vale, 2005; Yokosuka & Saito, 2009) or also influenced by lipid domains and the lipid environment (Gaus et al., 2005; Holowka, 1997; Sohn et al., 2008; Tavano et al., 2006) is not clear at the moment. Ultimately, in cellular membranes, lipids and proteins must be studied together and this requires a better integration between fluorescent probes and advanced imaging technology.

Declaration of interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

The work was supported by the Australian Research Council, National Health and Medical Research Council of Australia (Program Grant 1037320).