The nature of the association between 'raftophilic' proteins and rafts varies. Only a few proteins are permanent residents. These are important for maintaining a specific raft organization and include the caveolins (Galbiati et al., 2001; Harder & Simons, 1997; Monier et al., 1996). Many other proteins, however, are transient passengers or guests, taking advantage of the remoteness of the lipid island for intracellular targeting (Lusa et al., 2001) or specific signal transduction processes (Cheng et al., 2001; Drake & Braciale, 2001; Drevot et al., 2002; Dykstra et al., 2001; Langlet et al., 2000; Sharkey et al., 1990; van der Goot & Harder, 2001). The abundance of signalling molecules and immunoreceptors suggests a central role for rafts in modulating T-cell receptor function (Langlet et al., 2000).

Integral and peripheral membrane proteins use different strategies for raft association. Integral membrane proteins like CD36, SNAP-25 and caveolin-1 contain several palmitoylation (Buser et al., 1994; Rodgers et al., 1994; Sigal et al., 1994; Webb et al., 2000) sites adjacent to, or just within, the cytoplasmic leaflet of the membrane (Jochen & Hays, 1993; Monier et al., 1996; Veit et al., 1996a, b). Insertion of the palmitoyl chain into the lipid bilayer is energetically favourable and may provide the driving force for partitioning the modified protein into the raft environment (Schroeder et al., 1994).

Peripheral membrane proteins contact only one side of the raft membrane. Association with the inner leaflet of the bilayer is frequently mediated by multiple N-terminal acylations with saturated fatty acids such as myristate or palmitate. The binding energy they provide is approximately 10^-4 M K_d, which is not sufficient to anchor a protein to a membrane. Therefore, myristoylated proteins can only maintain an efficient membrane interaction when a second membrane-binding site is present. This second membrane-binding site can be created by a nearby stretch of basic residues or palmitoyl chains, as in the Src family of protein kinases or the subunits of heterotrimeric G proteins (Buser et al., 1994; Robbins et al., 1995; Rodgers et al., 1994; Sigal et al., 1994).

Glycosylphosphatidylinositol (GPI)-linked proteins such as CD55, PLAP and Thy-1 (Calafat et al., 1983; Harder et al., 1998; Marschang et al., 1995) are the most abundant proteins associated with the outer leaflet of the raft bilayer. They are synthesized initially as transmembrane proteins. The transmembrane region is then proteolytically cleaved and replaced by a pre-assembled glycolipid. Consequently, the modified polypeptide chains are anchored to the membrane only through the glycolipid modification. Most GPI-anchored proteins are raft associated through the long, saturated acyl and alkyl chains of their lipid anchors (Benting et al., 1999; Brown, 1994, 1998; Brown & London, 1998; Brown & Rose, 1992; Melkonian et al., 1999; Rodgers et al., 1994; Schroeder et al., 1994).

The presence of GPI-anchored proteins in the outer leaflet of the raft bilayer can be regulated by phospholipase C activity, resulting in the release of GPI-anchored molecules from the plasma membrane (Brown et al., 1994). Likewise, integral membrane proteins and proteins attached to the cytosolic face of the raft membrane can be controlled by reversible palmitoylation (Robinson et al., 1995). These mechanisms allow flexible control and suggest that raft microdomains are dynamic membrane assemblies.

In summary, rafts are dynamic structures that restrict their lipid and protein composition quantitatively but not absolutely. Once formed, rafts recruit other components by their affinity with the initial raft components. Exclusion appears passive. Components that lack affinity for the initial ones will not be concentrated in the raft region. This effect is cooperative; as the composition of the raft changes due to the incorporation of components, the strength of the attraction increases (Benting et al., 1999; Brown & London, 1998; van der Goot & Harder, 2001).

Similar modifications to those seen on cellular raft proteins are found on several viral proteins. Retroviral Gag proteins are myristoylated and the glycoproteins of retroviruses, filoviruses and influenza viruses may be palmitoylated (Ito et al., 2001; Schmidt, 1982, 1984; Yang et al., 1995). There is increasing evidence that a number of virus proteins, with or without modifications, are raft associated and that assembly and budding of virions may take place through rafts. The acid test of whether or not a virus assembles and buds through lipid rafts is whether or not the lipid composition of its membrane reflects the lipid composition of a raft and differs from the bulk cellular lipid composition. In practice, however, viruses are usually classified as raft viruses based on an assortment of more indirect methods that assay instead for raft-associated phenomena. They can be divided loosely into five general approaches: (1) co-flotation with DRMs and associated marker proteins in density gradients after cold detergent treatment (see above). This is the most widely used raft assay for both viruses and cellular proteins; (2) observing punctate co-localization with raft markers in the plasma membrane by immunofluorescence; (3) measuring the incorporation of raftophilic molecules into virions; (4) biophysical experiments to define the fluidity of the lipid environment in the viral membrane; and (5) looking for blocks to virus assembly and budding after raft disruption, normally induced by cholesterol depletion – it is important to remember the limitations of these methods when interpreting resulting observations.

The Triton X-100-extractability assay, while widespread, is not used with sufficient uniformity that comparison of results from different laboratories is straightforward. In the case of HIV there is evidence that the coarseness of commonly used density gradients disguises the fact that Gag–membrane complexes float at a slightly higher density than conventional rafts (Lindwasser & Resh, 2001). Other combinations of detergent and temperature can be used to isolate raft-like membrane domains, which may have different compositions to those preserved in Triton X-100 at 4 °C (Roper et al., 2000). It was reported recently that T-cell receptor-containing rafts could be isolated using polyoxyethylene ether detergents (Brij) at 37 °C (Drevot et al., 2002).

In the absence of an analysis of lipid composition, it is difficult to conclusively link flotation at a particular density to presence in membrane microdomains. Furthermore, when viral proteins are observed in association with detergent-insoluble fractions in infected cells, the possibility that this occurs during transport of the proteins prior to assembly should be considered.

The incorporation of raftophilic proteins into the viral envelope provides evidence of raft-mediated assembly only if relative concentrations in the plasma membrane are considered properly.

Biophysical results indicating reduced flexibility of lipid tails in the viral membrane may be difficult to interpret in the presence of highly ordered viral proteins, which may themselves induce ordering of lipids.

Interpreting the effects of raft disruption on virus assembly is also potentially dangerous. The depletion of cholesterol is likely to have widespread effects on cellular processes and it is difficult to attribute exclusively any deleterious effects on virus assembly to raft disruption. Alternatively, cholesterol may have a structural role in the budded virus beyond any contribution to raft formation, as is the case in SFV (Ahn et al., 2002; Lu & Kielian, 2000).

Despite these considerations, there is compelling evidence for the involvement of rafts in the assembly of a number of different non-icosahedral enveloped viruses.

Retroviruses

The majority of retrovirus studies are on HIV-1. Flotation assays have been carried out on the Gag protein and its cleavage product, MA, the Env protein and its cleavage products, TM and SU, and the Nef protein. Triton X-100 treatment ranged from 3 min at 0.5 % to 20 min at 0.25 % to 1 h at 1 %. Three-step gradients have been commonly used and varied from 7310 % gradients to 405 % gradients. This means that DRMs were isolated at interfaces ranging between 65 and 10 % and 30 and 5 %. HIV proteins were found throughout the gradients, but typically a number of fractions at the bottom of the gradient and around the interface were pooled to represent the soluble and raft components. Any attempt to quantify must therefore be critically dependent on the gradient used. Given that no two groups have used identical conditions, it is impossible to accurately compare their results. Nevertheless, some consistent patterns do emerge. Nguyen & Hildreth (2000) found 90 % of myristoylated Gag associated with DRMs and approximately 35 and 25 % of MA and TM, respectively. Ono & Freed (2001) found 25 % of Gag in the raft fractions, which was half of the membrane-associated Gag. Given that the DRMs represent only a small fraction of the plasma membrane, this represents a significant concentrating effect. Zheng et al. (2001) found that the concentration of Gag in the DRMs was 14 times higher than that in the soluble fraction and almost three times that in the plasma membrane. This concentrating effect was Nef dependent (Zheng et al., 2001). Nef itself is also enriched in DRMs (Wang et al., 2000; Zheng et al., 2001). Gp160 was found to be raft associated using both a flotation assay (Pickl et al., 2001) and a simple test for insolubility in cold Triton X-100 (Rousso et al., 2000). Rousso and colleagues observed that insolubility in cold Triton X-100 was dependent on the palmitoylation of at least one of the two cytoplasmic palmitoylation sites.

Lindwasser & Resh (2001) examined the flotation of Gag on multiple-step Optiprep gradients and found that while 6–8 % of Gag floats in the raft fraction (marked by GM1 and caveolin-1), about 30 % of Gag was found at a slightly higher density. A reduction in Gag multimerization relocates some of this Gag to the raft fraction. These authors (Lindwasser & Resh, 2001) termed these protein-containing domains 'barges'. Such observations highlight the limitations of three-step gradients. A recent article (Ding et al., 2003) explores this issue and shows that HIV-1 Gag rafts are more dense than classical lipid rafts.

Gag and Env co-localize in a punctate pattern in transfected Cos cells (Hermida-Matsumoto & Resh, 2000). A similar pattern of co-localization is seen between Env and GM1 (Pickl et al., 2001). Nguyen & Hildreth (2000) observed co-localization of an anti-HIV polyclonal antibody with a number of raft markers in infected Jurkat cells, but in broad patches rather than the small points observed by the other groups. Non-raft proteins were excluded from these areas. The same authors demonstrated that raftophilic proteins and GM1 were all incorporated into virions, despite being present in low concentrations on the cell surface. In contrast, the highly expressed CD45 was not incorporated into virions.

Cholesterol depletion leads to a decrease in virus infectivity (Ono & Freed, 2001; Zheng et al., 2001), which is Nef dependent (Zheng et al., 2001), as expected for a system in which assembly is mediated by rafts. Crucially, the viral membrane is known to be enriched in sphingolipids and cholesterol relative to the plasma membrane (Aloia et al., 1988, 1993). There is therefore a significant weight of evidence supporting a role for lipid rafts in the assembly of HIV.

Moloney murine leukaemia virus was also studied by Pickl et al. (2001). They observed flotation of Gag and Env, co-localization of Env and GM1, incorporation of raftophiles into virions and a reduction in virus titre after cholesterol depletion, implying that this retrovirus also assembles at raft domains. They also observed pseudotyping of virus particles with Env proteins from HIV, VSV and influenza virus.

Influenza virus

Neuraminidase (Barman & Nayak, 2000; Zhang et al., 2000) and HA (Pickl et al., 2001; Zhang et al., 2000) both float in the raft fraction of cell lysates. HA is also found floating in density gradients of Triton X-100-treated virus preparations. This behaviour is dependent on their cytoplasmic tails (Zhang et al., 2000). Co-localization of HA and raft markers in patches on the cell surface can be observed after cross-linking and GM1 is found incorporated into virions (Pickl et al., 2001). Electron paramagnetic resonance suggests the presence of two different domains with fast and slow oxygen collision rates (Kawasaki et al., 2001). The authors attributed the less mobile domain to a protein-rich raft domain. Diphenylhexatriene fluorescence polarization, used as a measure of acyl chain order, also suggests the presence of highly ordered domains within the viral envelope (Scheiffele et al., 1999).

Zhang et al. (2000) demonstrated that reducing the raft association of HA and neuraminidase, by deleting regions in their cytoplasmic tails, lowered the amount of cholesterol and sphingomyelin in the virion. Scheiffele et al. (1997) showed that HA expressed in a different lipid environment, namely the membrane of VSV, is not DRM associated (Scheiffele et al., 1997). They also demonstrated that the detergent resistance of the sphingomyelin within the influenza virus membrane is dependent on the presence of cholesterol. These results demonstrate the inter-dependence of both lipid and protein components in moulding the properties of the influenza virus membrane.

Vesicular stomatitis virus

Flotation assays with VSV give contradictory results. Pickl et al. (2001) found the VSV glycoprotein (VSV-G) associated with DRMs, whereas Scheiffele et al. (1999) found that it did not. The first group were working with extracts from cells expressing HA-tagged VSV-G, whereas the second group were working with purified virus. Although essentially the same gradient was used, the second group carried out a significantly shorter centrifugation step (2 h as opposed to 16 h), confounding comparisons between the findings. In the same work, Pickl et al. (2001) observed co-localization of HA-tagged VSV-G with GM1 in patches on the plasma membrane.

There is a significant body of work on the lipid composition of the VSV envelope. It is clear that the envelope is enriched in cholesterol and sphingomyelin when compared to the membrane from which it buds (Patzer et al., 1978; Pessin & Glaser, 1980). The high cholesterol content leads to an increase in membrane rigidity (Lisi et al., 1993). Glycoprotein (G) and M protein can induce the formation of domains with a similar composition to the viral envelope when incorporated into large vesicles in vitro (Luan et al., 1995).

There is evidence for incorporation of a wide range of foreign proteins into the VSV envelope. Pseudotyping of VSV has been observed with togaviruses, retroviruses, bunyaviruses, arenaviruses, paramyxoviruses, orthomyxoviruses, coronaviruses, herpesviruses and poxviruses (Altstein et al., 1976; Calafat et al., 1983; Dragunova & Závada, 1979; Lukashevich & Závada, 1982; Pastorekova et al., 1992; Schnitzer et al., 1977; Zajac et al., 1980; Závada, 1972b; Závada et al., 1972, 1978, 1983; Zavadova & Závada, 1980). CD4 is incorporated efficiently (Schnell et al., 1996). Confusingly, although murine leukaemia virus envelope proteins can pseudotype with VSV (Witte & Baltimore, 1977), HIV-1 gp160 is excluded (Johnson et al., 1998; Owens & Rose, 1993). Mutagenesis revealed that if all of the gp160 cytoplasmic domain is removed, the protein is incorporated efficiently into VSV virions, but truncations that leave 10 or 29 amino acids are still excluded. Strangely, replacement of the cytoplasmic domain with that of CD4 (which is incorporated efficiently into VSV) does not permit incorporation (Johnson et al., 1998). Failure to incorporate correlates with a different cell surface localization to that of VSV-G. Johnson et al. (1998) propose that the cytoplasmic domain of gp160 contains a signal inhibiting co-localization with VSV-G and suggest that a similar signal is generated in the CD4 tail by a change in folding when it is in combination with the gp160 transmembrane domain. If VSV-G and gp160 are expressed in the absence of other VSV proteins, the differences in localization become less dramatic but nevertheless persist. It appears difficult to interpret these observations in the context of raft association. All three of the truncated tail constructs lack both palmitoylation sites, at least one of which appears necessary for raft association of gp160 (Rousso et al., 2000).

These results may perhaps be reconciled best by suggesting that VSV proteins associate with raft-like membrane domains with a composition that differs to some extent from that of classical rafts. The properties of these domains would be influenced by the proteins within them, leading to the observed changes in localization of VSV-G and gp160 in the absence of VSV-M. The differences between these raft-like domains might also be blurred when raft-associated proteins were expressed at extremely high levels, such as during virus co-infections. Certain proteins, such as CD4, would be enriched in both types of domains, whereas others would be more specific to one or the other. This might provide a partial explanation for contradictory flotation results and also help to explain the exclusion of HIV gp160 from VSV virions.

The results of double infections in polar cells are informative. Rodriguez-Boulan et al. (1983) showed that influenza viruses, such as WSN, bud from the apical surface of polarized Madin–Darby canine kidney cells, while VSV buds from the basolateral surface. This has been demonstrated amply in singly infected cells (Fuller et al., 1984; Fuller, 1987; Pfeiffer et al., 1985; Roth et al., 1979; Simons & Fuller, 1985). During early infection in doubly infected cells, both viruses maintain their polar budding (Roth & Compans, 1981). At later times in infection, pseudotypic particles are formed in which WSN glycoproteins are incorporated into VSV virions (Roth & Compans, 1981). Up to three-quarters of VSV infectivity can be neutralized by anti-WSN antibody. The start of pseudotype formation correlates with the loss of tight junction integrity, and therefore polarity, as a result of the cytopathic effect of infection. Doubly infected non-polarized baby hamster kidney cells produce pseudotypes without any lag time. Hence, pseudotype formation requires that both glycoproteins be present in the same epithelial cell domain.

Other potential raft viruses

A number of measles virus proteins float in sucrose density gradient of Triton X-100-treated cells in a cholesterol-dependent fashion (Manie et al., 2000). Similar observations have been made for Ebola and Marburg viruses (Bavari et al., 2002). These filoviruses were also shown to incorporate GM1, but not the transferrin receptor. The viral glycoprotein and GM1 also co-localize on the cell surface. Respiratory syncytial virus has been shown to assemble within GM1-rich microdomains (Brown et al., 2002) and will form pseudotypes with VSV (Kahn et al., 1999, 2001).

It is becoming clear that rafts play a role in the assembly of a number of different non-icosahedral viruses. One way of addressing this function of rafts is to contrast the mechanism of assembly of membrane viruses that appear to employ rafts with those that do not.

SFV is a typical example of an icosahedral virus that does not appear to utilize lipid rafts in its formation. SFV and other alphaviruses such as Sindbis virus and Ross River virus (Cheng et al., 1995; Kielian & Helenius, 1986; Schlesinger & Schlesinger, 1986; Zhang et al., 2002) assemble in a well-defined manner (Kielian, 1995; Kielian & Helenius, 1986; Kielian et al., 1990; Schlesinger & Schlesinger, 1986; Simons & Garoff, 1980; Simons & Warren, 1984). A capsid is formed in the cytoplasm during incorporation of the single strand of genomic RNA (Simons & Warren, 1984). The capsid has a T=4 hexamer–pentamer arrangement involving 240 copies of the capsid protein surrounding the RNA (Fuller et al., 1995). The envelope proteins of the virus are transported through the trans-Golgi network, where they are processed to generate the fusion active mature spike, to the cell surface. There is reasonable evidence that the spikes form a hexagonal array at the cell surface (von Bonsdorff & Harrison, 1978), which presents an array of spike tails on the cytoplasmic side of the bilayer. This acts as an interaction site for the capsid. The icosahedral nature of the final structure allows the visualization of the geometry. The assembled virus shows complementarity between the array of spikes and the capsid that allows precise interaction over the icosahedral surface (Fuller et al., 1995; Mancini et al., 2000). The hexagonal arrangement of the spikes is spaced so that it can interact with the hexamer–pentamer arrangement of the capsid proteins. This results in the incorporation of precisely 80 copies of the trimeric spike complex to match the 240 copies of capsid protein. The arrangement is complementary, since individual trimeric spikes interact with two hexamers and one pentamer of capsid (Fuller et al., 1995; Mancini et al., 2000). This complementarity yields a very precise control of composition. Loss of a single spike would result in lack of complementarity over the array and loss of interactions over several spikes (Lescar et al., 2001; Mancini et al., 2000). There are no convincing reports of the incorporation of unrelated virus envelope proteins into SFV.

This model for SFV assembly differs from the one suggested by the hybrid retrovirus particles in Fig. 1. An important distinction is that the curvature of the membrane in a non-icosahedral raft virus is accomplished both by protein–protein interactions of the internal proteins of the virus, and their interactions with the inner leaflet of the lipid bilayer. Compare this with an icosahedral virus such as SFV, where the curvature of the membrane is dictated exclusively by the curvature of the two organized protein layers that sandwich it. This gives a raft virus greater flexibility in both the protein composition of the viral membrane and the geometry of the virion. In addition, assembly is dependent on the composition of the lipid bilayer. Retroviruses are an extreme example of this. Virus-like particles can form by expression of Gag alone without the presence of the Env proteins. The geometry of the particles formed does not show the consistent shape or size expected for a non-raft virus.

The steps leading to assembly of a raft virus take advantage of the dynamic properties of lipid rafts and the fact that the viral proteins are raftophilic. The cellular membrane never has the smooth, random distribution of lipids and proteins described in the early fluid membrane model (Singer & Nicolson, 1972). This model is too simple a description, since cellular proteins and lipids will tend to form microdomains of selected lipid composition. As shown in Fig. 3, pre-existing rafts are dynamic structures but they turn the smooth homogeneous fluid membrane of the early model into a landscape of locally varying lipid composition (1). Virus infection leads to the production of viral proteins and their expression at the cell surface (2). The affinity of the viral proteins for particular lipid populations will lead to recruitment of more of those lipids and their enrichment in the neighbourhood of the proteins (3). This will result in the formation of rafts enriched in lipids that have affinity for the viral protein and the recruitment of further lipids and proteins with similar affinity. This process would continue until the collection of viral proteins and their interaction with the inner leaflet of the membrane results in curvature and budding (4).

Fig. 3. A model for raft virus assembly. The proposed role of rafts in virus assembly is depicted. The plasma membrane contains domains of locally varying lipid composition (1). Virus infection leads to the production of viral proteins and their distribution to the cell surface (2). The affinity of the viral proteins for particular lipid populations will lead to recruitment of more of those lipids and their enrichment in the neighbourhood of the proteins (3). This will result in the formation of rafts enriched in lipids that have affinity for the viral protein and the recruitment of further lipids and proteins with similar affinity. This process would continue until the collection of viral proteins and their interaction with the inner leaflet of the membrane results in curvature and budding (4).