Even if we take DRMs as a useful operational definition for membrane rafts (see Box 1), it is clear that such a biochemical criterion cannot provide information on the physical organization of its components. These concerns have led to a variety of physical methods of varying sophistication designed to study the nature of lateral organization of specific lipids and proteins at different scales both on the cell surface and in artificial membrane systems.
‘Rafts’ in artificial membranes
If we were to view functional heterogeneities on the cell surface as an extension of the ideas portrayed in the fluid-mosaic model, then we would continue to treat artificial multicomponent membranes as model systems describing the physical properties of the plasma membrane. Thus, a lot of work has concentrated on establishing the existence of domains in artificial membranes composed of specific lipids, for example a 1 : 1 : 1 proportion of DPPC : Sph : Chol, resembling those obtained from DRMs (reviewed in (2,23,24)). In this point of view, ‘rafts’ are preexisting structures on the cell surface which are spontaneously formed by equilibrium phase segregation in a multicomponent system or equilibrium thermal fluctuations resulting in transient small-scale domains even in the homogeneous mixed phase. These possibilities have been examined in numerous artificial membrane bilayer systems (25–29), and even in Monte Carlo/molecular dynamics simulations using simple model lipid potentials in two dimensions (reviewed in (30)).
Using the aforementioned lipid composition, freely suspended monolayers at the air–water interface, suspended lipid bilayers, lamellar stacks of lipid bilayers and artificial giant unilamellar vesicles (GUVs) have been prepared and subjected to a variety of techniques suited for assessing heterogeneities at different scales:
thermodynamic measurements such as differential scanning calorimetry (DSC), and surface pressure-area isotherms coupled with preferential partitioning of lipid probes;
diffusion measurements via intervesicular transfer rates of various lipids, fluorescence recovery after photobleaching (FRAP), fluorescence correlation spectroscopy (FCS) and single particle tracking (SPT);
spectroscopic measurements such as fluorescence quenching and fluorescence energy transfer (FRET);
direct visualization and imaging by confocal microscopy, scanning atomic force microscopy (AFM) and near-field scanning optical microscopy (NSOM).
Many of these techniques and the results obtained pertaining to micron and submicron scale structures on artificial membranes have been reviewed recently (31–33) and will not be gone into in detail here.
These studies have shown that while the binary lipid system of Sph : PC shows a liquid–gel coexistence at temperatures below the main transition of sphingolipids (Tm = 40 °C), the ternary mixture of Sph : PC : Chol shows a liquid–liquid coexistence within a range of compositions and temperatures (26). A range of domain sizes have been reported ranging from the nanometer to the micron scale (25,29,34,35). Using fluorescent probes attached to glycolipids such as GM1 and GPI-anchored molecules, several researchers have demonstrated preferential partitioning of these molecules into liquid domains enriched in Sph/Chol with differing diffusion properties (36,37).
Liquid–liquid coexistence in the ternary system has been interpreted as being a coexistence between the high temperature liquid disordered (ld) phase with a cholesterol-poor composition and a liquid-ordered (lo) phase enriched in sphingolipids and cholesterol. The difference between ld and lo phase is that the latter is characterized by a sharp reduction in the area per lipid as a result of stiffening of the acyl chains (24,38,39). Direct measurement of acyl chain stiffening in lo regions may be made by small angle X-ray scattering from oriented lamellar samples, or by measuring the torsional flexibility of labeled acyl chains using nuclear magnetic resonance or electron spin resonance. However, data from X-ray diffraction studies are not conclusive, presumably due to lack of registry of the components in different layers (40).
The interpretation of the ‘lo’ nature of this phase comes primarily from observations of reduced area per lipid obtained from surface pressure-area isotherms, and preferential partitioning of saturated long-chain fatty acids. A recent alternate proposal (reviewed in (23)) is that this new phase represents a liquid rich in condensed complexes; a chemical complex of cholesterol and sphingolipids formed in the reversible reaction p C + q S ⇌ (CS). Even in the absence of macroscopic phase segregation, equilibrium thermal fluctuations in the mixed phase of a multicomponent system may give rise to transient, small scale lo domains or, more significantly, condensed complexes whose lifetime could be enhanced by proximity to a phase boundary. This interpretation however, has not been as clearly validated for bilayer vesicles. To test this interesting proposal, one might need additional spectroscopic evidence to measure molecular complexation (for example see (29)).
What are the intermolecular forces responsible for the phase segregation that brings sphingolipids and cholesterol (raft components) together? This is a difficult question to address experimentally since in addition to two-body forces such as hydrogen bonding between the OH group of cholesterol and the amide group of sphingolipids (or even ceramides), weak dipolar interactions between sphingolipids, and van der Waals interactions between saturated acyl chain and cholesterol, there are many body interactions such as hydrophobic shielding or the ‘umbrella effect’ (wherein cholesterol may segregate into regions of the membrane with strongly hydrated phospholipid head groups due to steric considerations) (personal communication, P. Kinunnen, Helsinki, Finland). Any observed clustering on artificial membranes is most likely due to a combination of all these physical forces.
A closely related point of view is that the constituents of the cell membrane are in a mixed, equilibrated phase, poised close to a phase boundary. In this view, any slight perturbation drives the system across the phase boundary, inducing large scale segregation of specific lipid components, as observed in experiments involving the depletion of cholesterol in living cells, which gave rise to large scale segregation of probes preferring the ld phase (41).
At the very least, ignoring all active processes and the multitude of components present in the cellular context, any comparison of lipid organization in artificial membrane systems with cellular membranes can be made only when the composition and external thermodynamic parameters such as temperature and surface pressure are maintained the same. But there is a more fundamental criticism – thermodynamically predicated or thermally induced structures (phase segregated domains or transient fluctuations) cannot be effectively regulated and utilized for specific cellular function. The basic problem is that this route of investigation is firmly grounded in the fluid-mosaic picture. Actively maintained lateral compositional heterogeneity and transbilayer lipid and protein asymmetry contribute to holding the cell membrane in a state far from equilibrium. This immediately questions whether lessons obtained from the study of lateral lipid segregation under equilibrium conditions are likely to be relevant to understanding the structure of rafts or functional lipid assemblies present in living cell membranes.
‘Rafts’ on the cell surface
It appears that the only way to address the question, ‘what is the physical nature of ‘rafts’ in the cell membrane?', is to directly observe raft-assemblies in living cells (31). Fluorescence microscopy in living cells has consistently failed to reveal large-scale laterally segregated structures enriched in a major raft-component, GPI-anchored proteins (42,43). This suggests that any preexisting cellular rafts must be much smaller than those recently characterized in artificial systems and hence undetectable by the limited resolution of the fluorescence microscope (> 300 nm), and/or extremely dynamic (31,32). Their detection is also likely to be beyond the scope of conventional electron microscopy (42–44). Conventional optical microscopy fails to reveal any large-scale heterogeneities (32). At this scale the membrane is consistent with the fluid-mosaic picture. To face this challenge, a number of new methodologies for detecting membrane heterogeneity in cell membranes have emerged.
Probe partitioning methods
Recent studies examining the distribution of lipid probes capable of differential partitioning into lo or ld domains in living cell membranes have been interpreted in terms of a preexisting ‘mosaic of domains’ of varying size, composition, timescale and physical properties (45). This interpretation should be viewed with some caution, since studies on the molecular origins of differential partitioning in artificial membranes suggest a complex of interactions involving both the head (steric and dipolar) and the long-saturated acyl chains (free volume, van der Waals (46,47)). In light of this, and the ability of exogenously added detergents to significantly alter preexisting domains (see Box 1), one needs to carefully check that the lipid probes faithfully report on preexisting structures and not on structures induced by them. The absolute concentration of lipid probes is an important parameter in this regard, since at the probe levels high enough to be visualized (e.g. 1000 molecules/μm2 of a probe results in at least 0.1% probe to membrane lipid fraction), the probes may themselves need to be treated as a separate component. For a similar reason one should take care that the fluorescent markers used to tag specific lipids and proteins do not induce aggregation of the tagged molecules. However, multiphoton imaging with appropriate lipid probes such as Laurdan, capable of differing fluorescence properties (generalized polarization (GP)) in lo and ld domains (48) has recently revealed regions of the living cell membrane with fluorescence characteristics consistent with ‘lo’ domains (49). It remains to be determined whether this ‘lo’ characteristic is due to preexisting lipidic structures or protein interactions, since crosslinking of a ‘non-raft or DRM-associated protein’, the transferrin receptor, increases the extent of these domains.
Methods of detecting proximity
In native cell membranes, methods designed to detect proximity between molecules have observed inhomogeneous distributions of many molecular components of rafts, including GPI-anchored proteins.
Chemical crosslinking with short (1.1 nm) crosslinkers (50) suggest that cholesterol-sensitive complexes of GPI-anchored proteins exist at the cell surface containing anywhere from two to 14 molecules. These experiments were conducted using nonspecific cell-impermeable crosslinkers at low temperatures for an extended period of time. While this procedure facilitates detection of relatively long-lived preexisting structures, it is difficult to quantify the actual size or abundance of preexisting clusters in the membrane with this methodology.
Fluorescence resonance energy transfer (FRET) methods are designed to detect proximity between fluorophores at 1–10 nm scale (51). Earlier work from our laboratory monitoring FRET between identical fluorophores (homo-FRET (52)) had suggested that GPI-anchored proteins occur in cholesterol-sensitive, submicron-sized ‘domains’ at the surface of living cells. Recently, data from our laboratories have shown that a small but significant fraction (20–40%) of GPI-anchored proteins form extremely high density clusters of nanometer size (∼4–5 nm), each consisting of a few (≤ 4) molecules and different GPI-anchored protein-species (60). The high local density of GPI-anchored protein molecules was directly derived from the FRET-related fast anisotropy decay rates observed in time-resolved anisotropy measurements in experiments conducted on three different proteins, the human folate receptor (FR-GPI) labeled via a monovalent fluorescent folic acid analog, N-α-pteroyl-N-ε-(4′- fluorescein -thiocarbamoyl)-l-lysine (PLF), GPI-anchored Enhanced Green Fluorescent Protein (GFP-GPI) and variants of GFP, mCFP- and mYFP-GPI, in a variety of cell types. Using fluorescence photobleaching experiments and theoretical modeling of the resultant changes in anisotropy, in conjunction with a knowledge of the interprotein distances, we have been able to show that that 20–40% of GPI-anchored protein species are present in clusters on the scale of the Forster's radius R0 (i.e. < 4.65 nm). Interestingly, these results resolve the apparent discrepancy between the lack of detectable hetero-FRET from clustered GPI-anchored proteins (53,54) and the detection of robust homo-FRET (52) and significant chemical-crosslinking of diverse GPI-anchored proteins with a nanometer-sized spacer (50). These nanoscale structures are sensitive to cholesterol levels in living cells. On the other hand, sphingolipid depletion does not directly alter the structure of this organization, it instead makes these nanoscale structures more susceptible to cholesterol depletion. A particularly intriguing feature of this organization is that it exhibits a constant fraction of clusters and monomers over a large range (10–20-fold) of GPI-anchored protein expression levels. We believe that this methodology is most suited for the elucidation of nanoscale organization in living cell membranes in other contexts as well.
Single particle tracking (SPT)
Numerous SPT studies have been conducted to examine the diffusion characteristics of membrane components (14). Observations made at video rate (33 frames/s) of particles attached to potential raft-molecules have not provided any conclusive evidence of regions of the membrane that exhibit characteristics expected for lo domains as observed in artificial membrane experiments. Observations at this time-resolution from a variety of groups suggest ‘sizes’ ranging from zero to 26–500 nm, likely to be due to intrinsic differences in the protocol for making single particles and cell type variation (55). In a tour-de-force of precision experimentation, A. Kusumi and colleagues have collected SPT data at an extremely high time resolution (40 000 frames/ s) to measure the diffusion characteristics of GPI-anchored proteins and fluorescent lipids in living cell membranes at different spatial and temporal scales (33,55). These studies suggest that the membrane of living cells is predominantly compartmentalized via membrane skeleton fences at a cell type-dependent scale ranging from 30 to 230 nanometers, restricting the free diffusion of proteins and lipids; membrane constituents' display confined diffusion at short time scales and hop diffusion at longer (14). Their results also suggest that the raft-constituents attached to single antibody-bead conjugates diffuse as extremely small species consistent with monomers or small preexisting assemblies, but inconsistent with any large scale organization (> 100 nm) of stable rafts. An important note of caution emerging from the studies of Kusumi and coworkers is that even mildly crosslinked GPI-anchored protein species exhibit diffusion characteristics that are distinct from monomers in the membranes of living cells (55). Thus, probes with potential for crosslinking GPI-anchored proteins are likely to report anomalous diffusion characteristics for these molecules; the use of single fluorophore reporters would fix this experimental bottle neck. Data from single fluorophore tracking studies conducted on a GPI-anchored isoform of class II MHC molecules (56), albeit at much lower time resolution, are consistent with the SPT studies of Kusumi and coworkers. These studies report that most GPI-anchored proteins appear to exhibit fast diffusion consistent with the monomer species identified by Kusumi, whereas only a small fraction (between 6 and 20%) of the labeled species are likely to have a significantly slower diffusion coefficient consistent with larger oligomers or rafts. However, these studies were unable to characterize the size or origin of the slowly diffusing species.
‘Rafts’ in the inner surface of cells
Any functional organization at the outer leaflet of the plasma membrane is likely to be reflected in an organization at the inner leaflet so as to provide a connection between the two leaflets of the bilayer. A large number of inner leaflet molecules such as the Ras family of small molecule GTPases and non receptor tyrosine kinases are lipid anchored with modifications ranging from acylation to poly isoprenylation. Recent data on the size and structure of rafts at the inner leaflet of the plasma membrane, utilizing statistical analysis of the spatial distribution of H-ras and K-ras fusion proteins, detected via EM on fixed cells (57), supports the existence of 40-nm-sized structures covering 20–30% of the cell surface of separately clustered distributions of farnesylated H-Ras and K-Ras (tethered to the inner leaflet via polybasic-amino-acid stretches). Though the H-ras clusters are not correlated to non crosslinked GPI-anchored protein on the external leaf of the plasma membrane, they are disrupted by removal of cholesterol. Moreover, they are stabilized/expanded by crosslinking an intracellular lectin called Galectin. On the other hand, K-Ras clusters appear fundamentally different and are formed independent of cholesterol. In a separate study, using FRET microscopy, Tsien and coworkers have shown that multiply acylated proteins can co-cluster at the inner leaflet of the plasma membrane, providing evidence for a potentially different type of lipid organization at the inner leaflet (58). At this juncture it is important to obtain quantitative data about the size and composition of these inner leaflet structures and their relation to outer leaflet rafts in living cells at different spatio-temporal scales. Particularly important will be the combined study of structure of these inner leaflet proteins and their modulation by different signaling stimuli.