Lipid domain association of influenza virus proteins detected by dynamic fluorescence microscopy techniques

Authors


For correspondence. E-mail mveit@zedat.fu-berlin.de; Tel. (+49) 30 2093 6272; Fax (+49) 30 2093 6171; E-mail andreas.herrmann@rz.hu-berlin.de; Tel. (+49) 30 2093 8860; Fax (+49) 30 2093 8585.

Summary

Influenza virus is thought to assemble in raft domains of the plasma membrane, but many of the conclusions were based on (controversial) Triton extraction experiments. Here we review how sophisticated methods of fluorescence microscopy, such as FPALM, FRET and FRAP, contributed to our understanding of lipid domain association of the viral proteins HA and M2. The results are summarized in light of the current model for virus assembly and lipid domain organization. Finally, it is described how the signals that govern domain association in transfected cells affect replication of influenza virus.

Biochemical and biophysical properties of membrane rafts

Membrane rafts are defined as small (10–200 nm), heterogeneous, highly dynamic, cholesterol- and sphingolipid-enriched domains of the plasma membrane that compartmentalize cellular processes. Membrane rafts can selectively incorporate certain components, such as glycolipid-anchored and double-acylated proteins, which are attached to the outer or inner leaflet, respectively, and also a few often S-acylated transmembrane proteins (Levental et al., 2010). While these assemblies are generally very short-lived and small, they can be fused to larger and more stable structures under certain biological conditions. These coalesced raft domains can recruit additional proteins and perform a specific function for the cell or aid in the assembly and budding of virus particles. For a comprehensive treatment of raft association of influenza virus proteins and budding of viral particles, the reader is referred to several recent reviews (Nayak et al., 2004; 2009; Chen et al., 2008; Rossman and Lamb, 2011; Veit and Thaa, 2011).

Different models have been proposed to explain how proteins associate with membrane rafts and how domain formation and protein integration are regulated (Kenworthy, 2008).

Biophysical properties of membrane rafts have been characterized extensively in model membrane systems. In the cholesterol/sphingolipid-rich phase, the (mostly saturated) acyl chains of the membrane lipids are densely packed and restricted in mobility, but the lipids are able to diffuse and rotate, and form a ‘liquid-ordered’ (Lo) phase segregated from the ‘liquid-disordered’ (Ld), more fluid membrane phase (Simons and Vaz, 2004). The ‘Lo/Ld phase coexistence’ model predicts that cholesterol–sphingolipid interactions are sufficient to form rafts in the plasma membrane of the cell. Newly synthesized proteins possess an intrinsic affinity for either of these domains and passively partition into these preexisting structures (Simons and Ikonen, 1997). This is reminiscent of a simple phase distribution model, for example, the distribution of a compound between an organic and an aqueous phase. A closely related 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 and protein components (Simons and Gerl, 2010).

In contrast, the ‘lipid shell’ hypothesis postulates that raft proteins act as nucleation sites that recruit specific lipids, e.g. cholesterol and/or sphingolipids. This leads to the envelopment of each protein molecule by four to five lipid layers (approximately 80 molecules). These ‘lipid shells’ have an affinity for preexisting rafts and thus target the encased proteins specifically to these membrane domains (Anderson and Jacobson, 2002).

The actively maintained domains model postulates that raft proteins do not partition into preexisting nanodomains, but that some cellular activity, probably dynamics of cortical actin, actively clusters cell surface molecules (Gowrishankar et al., 2012). In addition, the cytoskeleton, which divides the plasma membrane into submicron-sized compartments (40–300 nm) by a meshwork of filaments, might be involved in the generation and/or maintenance of nanoclusters. The lateral diffusion of membrane constituents, even of lipids, is confined to such a compartment for a very short time range (milliseconds), which might facilitate interactions required for raft formation (Kusumi et al., 2011).

Despite the discrepancies in these models, all of them agree that to qualify as raft-associated, a protein must be clustered rather than being randomly distributed in the plasma membrane, and that this clustering depends on cholesterol.

Extraction with cold detergent to analyse raft association of influenza virus proteins

How is it determined experimentally whether a protein is a raft protein? Owing to its technical simplicity, extraction with cold detergent is the original and most popular biochemical approach to identify raft components. The procedure involves treatment of infected or transfected cells or of purified virus particles on ice with a non-ionic detergent, e.g. Triton X-100. Membrane parts enriched in cholesterol and sphingolipids will not be solubilized by this procedure and are called detergent-resistant membrane (DRMs). They will float to a low buoyant density upon gradient centrifugation. Proteins with the same behaviour are considered raft components (Brown and Rose, 1992).

The envelope protein haemagglutinin (HA) is required for cell entry of virus particles by catalysing receptor binding and membrane fusion. HA forms trimers and is a type I transmembrane glycoprotein with a cleavable N-terminal signal peptide, a large ectodomain, that is processed into two subunits, a single transmembrane region (TMR) of approximately 27 amino-acid residues, and a short cytoplasmic tail. HA was one of the first transmembrane proteins described as a component of DRMs. Hydrophobic amino acids in the outer leaflet of the TMR and S-acylation at cytoplasmic and transmembrane cysteine residues are required for partitioning of HA into DRMs (Scheiffele et al., 1997; Melkonian et al., 1999).

The second glycoprotein of influenza virus, the neuraminidase (NA), is also DRM-associated and transported apically. The signals for apical transport and raft localization are both situated in the TMR of NA, but overlap only partly (Nayak et al., 2004).

The matrix protein M1 forms a layer underneath the envelope in virus particles and is thought to bind to both viral glycoproteins and to ribonucleoprotein particles (RNP), which comprise the viral genome. M1 is not intrinsically targeted to the apical plasma membrane, but rather localizes to the nucleus, internal membranes and the cytosol. M1 associates with DRMs when either HA or NA are coexpressed. The matrix protein is also known to oligomerize, a property which is thought to drive the budding reaction (Nayak et al., 2004).

M2 is tetrameric proton channel activated by acidic pH, which is important for genome unpacking during virus entry. In each monomer, the first 24 amino acids form the unglycosylated ectodomain, the following 19 residues are the TMR and the remaining 54 residues build up the cytoplasmic tail. The sequence immediately following the TMR shapes a membrane-parallel amphiphilic α-helix, which possesses two possible raft-targeting features, S-acylation (Sugrue et al., 1990; Veit et al., 1991) and cholesterol binding, probably by ‘cholesterol recognition/interaction amino acid consensus’ (CRAC) motifs (Schroeder et al., 2005). It was proposed that acylation and cholesterol binding target the amphiphilic helix to raft domains, but that the relatively short TMR of M2 prevents complete immersion of the protein in the more ordered, hence thicker, raft domains (Schroeder et al., 2005). As a consequence, M2 is supposed to localize to the edge of the viral budozone, where it mediates pinching off of virus particles from the plasma membrane by the induction of curvature through wedge-like insertion of the amphiphilic helix into the membrane (Schroeder et al., 2005; Rossman et al., 2010a,b). Similarly to HA and NA, M2 is transported along the secretory pathway to the apical plasma membrane. Contrary to the other viral transmembrane proteins, however, M2 does not associate with DRMs and is largely excluded from virus particles (Zhang et al., 2000).

In summary, based on Triton extraction experiments NA and HA, but not M2, are supposed to be constituents of lipid rafts, whereas M1 binds to rafts only when NA and/or HA are present. However, concerns that DRMs do not represent intact biological membranes have incited controversy about the existence of rafts (Munro, 2003). Despite these reservations the cold extraction method has been successful as starting point to identify raft proteins, but more sophisticated methods should be applied to analyse raft association of proteins.

High-resolution fluorescence microscopy to study clustering of HA at the cell surface

Fluorescence microscopy in living cells has failed to reveal laterally segregated clusters of HA or other raft-associated proteins, indicating that rafts in undisturbed cells must be smaller than the resolution of the light microscope (< 200 nm). However, when antibodies against both HA and a glycolipid-anchored protein were applied, HA was found to cocluster with the established raft component. It was assumed that cross-linking of HA trimers by antibodies stabilizes small raft structures, which subsequently coalesce with other raft domains to form large, visible patches (Harder et al., 1998). However, this study could not establish potential clustering of HA in unperturbed cells.

In the last few years light microscopy has seen a tremendous progress in the development of high-resolution techniques not limited by the diffraction limit of visible light. Fluorescence photoactivation localization microscopy (FPALM) is a technique by which an image is reconstructed by calculating the exact centre of individual emission signals to get higher (apparent) resolution. FPALM revealed that HA is not randomly distributed at the plasma membrane, but forms irregular clusters ranging in size from a few nanometres up to many micrometres, which is larger than other raft-associated clusters (Hess et al., 2007). However, only clusters at the nanometre length scale could be disintegrated by extraction of cholesterol (Hess et al., 2005). Thus, FPALM confirmed former data obtained by quantitative immunoelectron microscopy indicating that the method is suitable to analyse clustering of HA (Hess et al., 2005; Leser and Lamb, 2005). Three-colour FPALM showed that the HA clusters are well separated from clusters formed by the transferrin receptor, a typical non-raft protein. Both the transferrin receptor and HA colocalize with actin, indicating that cortical actin is involved in maintenance of membrane domains (Gunewardene et al., 2011).

Fluorescence resonance energy transfer to study raft association of influenza virus proteins

Principle of the method

Förster's (or fluorescence) resonance energy transfer (FRET) is the radiationless transfer of energy from an excited donor fluorophore to a suitable FRET acceptor and can occur when the two fluorophores are in very close proximity (2–6 nm). FRET is therefore exceptionally well suited to demonstrate direct interactions between proteins. The molecules under study need to be tagged with suitable fluorophores, e.g. CFP (as FRET donor) and YFP (FRET acceptor), spectral variants of the green-fluorescent protein.

The FRET can be measured: (i) as acceptor emission intensity upon excitation of the donor, (ii) as an increase in the donor's fluorescence intensity after photobleaching of the acceptor or (iii) as reduced lifetime of the excited state of the donor. For the latter, FRET measurements are performed by fluorescence lifetime imaging microscopy (FLIM) and are more robust and quantitative since no corrections are needed for donor fluorophore emission bleedthrough in the acceptor emission channel. The lifetime of the donor's excited state will be reduced if FRET occurs to the acceptor in close proximity. From these measurements the FRET efficiency, defined as the fraction of donor excitation events that result in energy transfer to the acceptor, can be calculated.

However, analysing FRET in a two-dimensional system, such as the plasma membrane, requires a more systematic evaluation of the data. Since membrane proteins only show lateral mobility, energy transfer can simply occur by random collision of both molecules. In this case, the FRET efficiency increases linearly with increasing acceptor protein concentration at the membrane. In contrast, if FRET is due to clustering of the two proteins under study, FRET already occurs at very low expression levels and the FRET efficiency is largely independent of the acceptor protein concentration. To evaluate the data, the FRET efficiency of every cell coexpressing donor and acceptor protein is plotted against the fluorescence intensity of the acceptor at the plasma membrane of this cell. The data points are fitted according to an equation describing the binding of a ligand to its receptor (Zacharias et al., 2002). The fitting yields a dissociation constant KD as a parameter to assess the associative properties of donor and acceptor. If KD is very small compared to the acceptor intensities, clustering of acceptor and donor occurs, but if KD is in the same range as or larger than the acceptor intensities, FRET is due to random collision of both molecules.

Using acceptor-photobleaching FRET and this evaluation it was demonstrated that lipid-modified proteins cluster at the inner leaflet of membranes. CFP and YFP were fused to peptide sequences directing hydrophobic modifications, either dual acylation or geranylgeranylation. FRET independent of the expression level was observed between dually acylated peptides and between prenylated proteins, but not between prenylated and acylated peptides. Only FRET between acylated peptides was sensitive to cholesterol depletion and only acylated peptides coclustered with caveolin, a marker for caveolae rafts. Thus, both acylated and prenylated peptides are clustered, but only acylated peptides reside in cholesterol-maintained membrane rafts. The exclusion of prenylated proteins from rafts is in line with the assumption that the branched isoprenoid moiety does not fit into densely packed raft domains (Zacharias et al., 2002).

HA clusters with markers for the internal and external leaflet of membrane rafts

We attempted to analyse association of HA and M2 with established raft markers using FLIM–FRET. We first confirmed that the double-acylated peptide (Myr–Pal–YFP) forms cholesterol-dependent clusters (Engel et al., 2010). To analyse whether HA associates with this raft marker in living cells, the donor fluorophore Cerulean (Cer), a variant of CFP with improved quantum yield and a higher excitation coefficient, was fused via linker sequence to the cytoplasmic tail of H7 subtype HA, yielding HA–Cer (Engel et al., 2010; see Fig. 1A). Upon transfection of CHO cells, HA–Cer trimerizes, becomes S-acylated and acquires Endo-H-resistant carbohydrates with the same kinetics as HA without any attachments to its sequence (Engel et al., 2012). In addition, HA–Cer present at the plasma membrane binds to and fuses with erythrocytes indicating that the protein is functional (C. Sieben, M. Veit, and A. Herrmann, unpublished).

Figure 1.

HA constructs and raft markers used in FLIM–FRET studies.

A. Cerulean (Cer), a modified version of CFP, was fused via a linker sequence to the cytoplasmic tail of HA. As established marker of the rafts of the inner leaflet, we used a double-acylated peptide from the N-terminus of the Lyn kinase fused to YFP (MyrPalYFP).

B. To avoid tagging of the cytoplasmic tail of HA, which might interfere with its role in lateral organization, the ectodomain of HA was replaced by YFP. As established marker for outer leaflet rafts we used CFP linked to a glycolipid anchor (GPI).

C and D. YFP was fused to the cytoplasmic tail of M2 to analyse possible association with small rafts (C) and with the HA-containing viral budozone (D).

The cholesterol binding site in M2 and the residues VIL in the TMR of HA are marked with a black star. S-acylations at cysteine residues of HA and M2 are depicted as black lines. These features are thought to target HA to membrane rafts and M2 to the edge of the viral budozone. Localization of the viral constructs was analysed with FLIM–FRET (Scolari et al., 2009; Engel et al., 2010; Thaa et al., 2010).

When we measured FLIM–FRET of HA–Cer with Myr–Pal–YFP a very high FRET efficiency (≈ 50%) was observed, which was completely independent of the expression level of the acceptor (Engel et al., 2010). We performed several experiments to analyse whether clustering is due to partitioning of HA–Cer into membrane rafts. Extraction of cholesterol with methyl-β-cyclodextrin led to a statistically significant lowering of the FRET efficiency to 45% and to considerable increase in the KD value. The FRET efficiency was further reduced to 35% when either one of the two described raft-targeting signals of HA (S-acylation sites and very hydrophobic amino acids in its TMR) were removed. The HA mutant with both raft-targeting signals exchanged simultaneously revealed a slightly higher FRET efficiency (40%) indicating that the targeting signals do not work synergistically (Engel et al., 2010).

In extension of the just described, already published data (Engel et al., 2010), we also assessed whether actin is involved in the clustering of HA with the raft marker. To this end, we cotransfected cells with HA–Cer and Myr–Pal–YFP (raft marker) and treated them with cytochalasin D, which inhibits actin polymerization and disrupts microfilaments. The FRET efficiency was reduced to the lowest value measured for all experiments (28%) and the KD value was increased compared to untreated cells (Fig. 2A and B). If clustering of HA–Cer in membrane rafts were independent of clustering due to the actin meshwork, the disruption of both rafts and cytoskeleton (by methyl-β-cyclodextrin and cytochalasin D respectively) would lead to enhanced ‘declustering’ compared to removal of only one. However, we did not observe this (Fig. 2C). Likewise, when HA devoid of raft-targeting signals was analysed in the presence and in the absence of cytochalasin D, similar FRET efficiency and KD values were calculated (Fig. 2D; Engel et al., 2010). Thus, both actin filaments as well as association of HA with lipids govern clustering of HA with the raft marker.

Figure 2.

Clustering of HACer with raft markers dependent on microfilaments. CHO cells were cotransfected with HACer and MyrPalYFP (Fig. 1A) and FLIM–FRET measurements were performed. For each cell the calculated FRET efficiency (FRET E, in %) is plotted against the fluorescence intensity of the acceptor. Continuous line: data were fitted to the saturable one-site binding model [FRET E = Emax × A / (KD + A)]. The broken line is the 95% confidence interval. FRET E, mean FRET efficiency; A, fluorescence intensity of the acceptor; KD, dissociation constant describing the associative properties. A low KD relative to the acceptor intensity indicates high clustering.

A. Coexpression of HACer with MyrPalYFP, number of cells (n) = 32; KD = 6 × 10–8.

B. Coexpression of HACer with MyrPalYFP. Cells were incubated with 1 μM cytochalasin D (Cyto D) for 20 h prior to FRET measurements. KD: 1996 ± 923, n: 48.

C. Coexpression of HACer with MyrPalYFP. Cells were incubated with cytochalasin D for 20 h and treated with 5 mM methyl-β-cyclodextrin for 30 min prior to FRET measurements. KD: 370 ± 147, n: 26.

D. Coexpression of HACer mutant (exchange of both raft-targeting signals) with MyrPalYFP. Cells were incubated with 1 μM cytochalasin D (Cyto D) 20 h prior to FRET measurements. KD: 825 ± 453, n: 40. Cytochalasin had no effect on the diffusional mobility of HACer as determined by FRAP.

For another set of experiments we constructed an artificial FRET probe by replacing most of the ectodomain of HA by the fluorescent protein (Scolari et al., 2009; see Fig. 1B). This excludes that tagging of the cytoplasmic tail of HA interferes with its role in lateral organization. This construct, TMD–HA–YFP, consists of YFP, preceded by the signal peptide of HA and followed by 38 amino acids (including one glycosylation site) of the HA ectodomain and the complete transmembrane and cytoplasmic domain of HA. Upon transfection of CHO cells, TMD–HA–YFP is partially expressed as a disulfide-linked dimer, and only dimers receive Endo-H-resistant carbohydrates and are targeted to the plasma membrane. As raft marker of the external leaflet of the plasma membrane, we used GPI–CFP that forms small cholesterol-sensitive clusters in CHO cells (Sharma et al., 2004). In line with this, we found cholesterol-sensitive FRET between GPI–CFP and GPI–YFP by FLIM (Scolari et al., 2009).

The GPI–CFP also clusters with TMD–HA–YFP. FRET efficiencies between the external raft marker and truncated HA were in the order of 10% and were reduced to 5% upon cholesterol depletion and were almost absent (2%) after removal of all three acylation sites from HA. Exchange of fluorophores between HA and GPI yielded similar results. Clustering of GPI–CFP and TMD–HA–YFP was also observed in plasma membrane suspensions. We concluded that the TMR and cytoplasmic tail of HA contain the molecular signals for targeting of HA into rafts (Scolari et al., 2009).

Although the FRET efficiencies determined for the two described FRET pairs differed widely (50% versus 10%), this does not indicate that HA clusters more strongly with the marker for inner leaflet rafts (Myr–Pal–YFP) than with the marker for outer leaflet rafts (GPI–CFP). FRET values cannot be compared between different protein pairs, even if they are attached to the same donor and acceptor fluorophore. The FRET efficiency depends on the distance between donor and acceptor and on the relative orientation of the dipol moments of donor emission and acceptor absorption, parameters which cannot be measured inside cells. In addition, GPI–CFP and HA–TMD–YFP were expressed in similar amounts whereas the marker for inner leaflet rafts (Myr–Pal–YFP) was approximately 10-fold overexpressed relative to HA–Cer; such overexpression of the acceptor renders a FRET pair prone to high transfer rates.

M2 clusters with HA, but not with the raft marker in transfected cells

In virus-infected cells M2 localizes to the base of budding filamentous virus particles, where it is supposed to catalyse scission of virus particles (Rossman et al., 2010a,b). We used the FLIM–FRET approach to analyse whether in transfected cells M2 (fused at its cytoplasmic tail to YFP) associates with rafts and whether it coclusters with HA (Thaa et al., 2010; see Fig. 1C and D). M2–YFP is palmitoylated, forms oligomers and is efficiently transported to the plasma membrane of transfected cells. In addition, M2–YFP conducts protons from the medium (after acidification) into the cell's cytoplasm indicating that its function is not impaired [M. Veit, B. Thaa, and T. Friedrich (TU Berlin), unpublished]. Testing possible raft localization of M2 by FRET showed that the molecule does not interact with the double-acylated marker for inner leaflet rafts. The mean FRET efficiency was low (9%) and, most importantly, revealed a linear dependence on the expression level of M2–YFP. Thus, in contrast to HA, M2 does not cluster with a marker for inner leaflet rafts.

However, the results from FLIM–FRET experiments point to very close colocalization (or interaction) of M2 with HA. We obtained a high degree of clustering of HA–Cer and M2–YFP, evidenced by FRET efficiencies considerably independent of the acceptor intensity. The FRET efficiency was diminished from 16% to 10% when the HA probe without raft-targeting signals was assessed and to 6% in the presence of the cytoskeleton-disrupting drug cytochalasin D. Thus, if the formation of HA clusters at the plasma membrane is inhibited, M2 does not associate with HA (Thaa et al., 2010).

Fluorescence recovery after photobleaching to analyse the stability of HA clusters

It is expected that the diffusion rate of molecules in a large raft complex is reduced compared to diffusion of unclustered molecules. Diffusion within the plasma membrane can be assessed experimentally by FRAP (fluorescence recovery after photobleaching), where it is recorded how quickly a previously bleached spot is replenished with fluorescence from unbleached neighbouring areas.

Stepwise removal of the raft-targeting signal from HA–YFP successively increased its diffusion coefficient from 0.14 μm2 s−1 (wild-type HA) to 0.3 μm2 s−1 (Engel et al., 2010), which is in agreement with other studies (Shvartsman et al., 2003). More than 80% of all HA molecules proved to be mobile, indicating that the HA clusters are not static in the timeframe of FRAP experiments (minutes). In addition, since Myr–Pal–YFP revealed a much higher diffusional speed (0.7 μm2 s−1), HA does not diffuse together with the raft marker in a stable complex. Yet, the diffusional mobility is mainly determined by the type of transmembrane anchorage rather than raft localization: proteins anchored by lipid moieties (prenylation, S-acylation) diffuse quicker than transmembrane proteins (Kenworthy et al., 2004).

Photobleaching experiments performed at a higher spatial resolution, using FPALM as well as line scan fluorescence correlation spectroscopy (FCS), showed that HA is highly mobile within the microdomains and that they exchange with those from outside the microdomains on a relatively fast time scale (Hess et al., 2007; Itano et al., 2011). All these results are consistent with a model of dynamic partitioning of HA into and out of raft domains.

Model for assembly of HA and M2 at the viral budding site

The described experiments, together with the initial observation that HA is in DRMs, provide conclusive evidence that HA clusters at the plasma membrane and that rafts are involved in cluster formation (Fig. 3A). The FRET experiments also verified the importance of the TMR and S-acylation of HA for raft targeting. Interestingly, the removal of the raft-targeting signal in the outer part of the TMR retards transport of HA through the Golgi (Engel et al., 2012). This suggests that raft-like domains might also be involved in intra-Golgi transport (Patterson et al., 2008). However, these (hypothetical) domains are likely to be different from rafts of the plasma membrane, since removal of the second raft-targeting signal, S-acylation, has no effect on intracellular transport of HA (Engel et al., 2012).

Figure 3.

Model for assembly and budding of influenza virus.

A. HA clusters in large membrane rafts (grey), which also contain cellular proteins, such as the double-acylated peptide attached to the inner leaflet (black) or glycolipid-anchored proteins present in the outer leaflet (not depicted). The arrow indicates that HA is mobile within the budozone and can leave the boundary to diffuse outside of rafts. M2 is thought to be targeted to the edge of the budozone by cholesterol binding (black star) and S-acylation (wavy black line) of its amphiphilic helix. Similar symbols are used to indicate the raft-targeting signals in HA: very hydrophobic amino acids in the transmembrane region and attachment of two palmitates at cytoplasmic cysteines and one stearate to a transmembrane cysteine. Cortical actin (small black line) is required for the clustering of HA with rafts and for the association of HA with M2.

B. In the absence of raft-targeting signals HA is randomly distributed in the plasma membrane. It neither clusters with the markers for small, unstimulated rafts [double-acylated peptide, glycolipid-anchored protein (not depicted)], nor with M2. Likewise, M2 does not associate with small rafts, even if the raft-targeting signals are present.

C. Binding of the matrix protein M1 (red) to the cytoplasmic tails of HA (and NA, not depicted) stabilizes the viral budozone. M2 localizes to the edge of the budozone. M2 is also drawn to the budozone by interaction with M1 at the cytoplasmic tail.

D. Oligomerization of M1 induces curvature in the membrane. Budding of the membrane might be aided by the line tension. In addition, the virus particles also contain the viral genome in the form of eight RNPs (not depicted).

Presently, one can only speculate on the molecular mechanism by which the raft-targeting signals cause incorporation of HA into rafts. In principle, α-helical TMRs with their protruding amino acid side chains should rather disrupt the tight packing of lipids in a raft domain as they do not readily accommodate the rigid, bulky sterol ring of cholesterol (Kusumi et al., 2011). It could be imagined that flexible acyl chains fill the voids in the irregular and rough surface of the transmembrane domain and thus ‘lubricate’ the region for interactions with cholesterol. Whether all the acylations of HA, one stearate at the end of the TMR and two palmitates attached to the cytoplasmic tail (Kordyukova et al., 2008), are required for raft association of HA is not known, but from detergent extraction experiments, it was concluded that the cytoplasmic acylation sites are more important for raft association than the stearate attached to the transmembrane cysteine (Chen et al., 2005; Wagner et al., 2005).

The raft-targeting signal in the HA–TMR is not precisely defined. The exchange of the amino acid sequence VIL in the outer region of the transmembrane domain led to decreased DRM association (Takeda et al., 2003) and decreased clustering with raft markers in FLIM–FRET (Shvartsman et al., 2003; Scolari et al., 2009; Engel et al., 2010). It is presently not known whether each of the exchanged amino acids (VIL), two of which (IL) are conserved through all HA subtypes, is required for raft association. One might speculate that these amino acids mediate interactions with lipids that promote inclusion of HA into rafts. Interestingly, cholesterol-interacting motifs (Schroeder, 2010) contain amino acids that are also abundant in the outer region of HA's TMR, besides V, I and L also the residues W, Y, F and K.

The large size of the HA clusters, larger than other cholesterol-dependent clusters observed so far in unstimulated cells (5–50 nm), indicates that they are coalesced raft domains (Hess et al., 2005). Since the area of the HA cluster is large enough to cover the surface of a spherical virus particle (30 nm2), these structures can be regarded as correlates of the viral preenvelope (budozone). The finding that sphingolipids and cholesterol are enriched in influenza virus particles compared to the whole apical membrane substantiates that viruses bud through raft domains (Gerl et al., 2012). Highly purified virus particles also contain raft-associated cellular proteins, such as glycolipid-anchored CD59 and glypican as well as S-acylated tetraspanins (CD9, CD81). However, other typical raft proteins were not identified, indicating that the HA clusters contain only a subset of all cellular raft proteins (Shaw et al., 2008). Whether HA selects the budding site due to the localized concentration of these cellular proteins or whether HA induces the formation of the large raft domain is still an open question. Interestingly, it was shown that a peptide representing the TMR of HA induces highly ordered domains in model membranes, but only if the leucine of the raft-targeting signal is present, indicating an active role of HA in cluster formation (Ge and Freed, 2011).

Not only are raft domains supposed to cluster the viral components in the budozone, but they could also directly aid in the process of budding. Upon phase separation of Lo and Ld domains, there is a hydrophobic mismatch and a thickness difference between the two membrane phases, leading to the formation of a ‘line tension’ at their interface. Line tension leads to the formation of a curved raft phase due to the propensity of the system to minimize its free energy (Fig. 3D). This may initiate or support protein-based budding. In line with this assumption is the observation that HA, when expressed alone, leads to the release of virus-like particles, vesicles containing HA and having the same density as actual virus particles (Chen et al., 2007).

However, the HA clusters do not have perimeter-minimized (round) boundaries, as would be expected for Lo/Ld phase separation, but often show elongated shapes and narrow extensions (Hess et al., 2005; 2007). The HA-containing raft domains might be shaped by the presence of HA (having nine acyl chains per trimer). In addition, other mechanisms of compartmentalization might mediate HA clustering (Kusumi et al., 2004) – such as cortical actin, which colocalizes with HA clusters (Gunewardene et al., 2011) and is required for clustering of HA with the raft marker (Fig. 2) and with M2 (Thaa et al., 2010). The actin cytoskeleton has also been shown to be functionally linked to the production of virus particles: actin (and several actin binding proteins) is present in highly purified virus particles, and the disruption of the actin meshwork reduces budding especially of filamentous virus particles (Simpson-Holley et al., 2002).

Results from various FRAP experiments have demonstrated that the HA clusters are not stable entities. HA does not diffuse together with a raft marker for minutes, and HA is mobile within the HA clusters (Shvartsman et al., 2003; Engel et al., 2010). These observations are consistent with a model of dynamic partitioning of HA into and out of raft domains. Assuming that assembly and budding of influenza virus requires several minutes, as has been shown by real-time studies for the biogenesis of individual human immunodeficiency virus particles (Jouvenet et al., 2008; Ivanchenko et al., 2009), it is unclear how such unstable HA clusters can support the whole assembly process. However, it is reasonable to assume that in the context of virus infection, binding of M1 of the cytoplasmic tails of HA (and also of NA) and the subsequent oligomerization of M1 might further stabilize the HA clusters and serve as a nucleation site for the recruitment of viral RNPs (Fig. 3C).

The final step of virus budding, the scission of particles, is thought to be mediated by M2. In order to do so M2 must be targeted to the edge of the viral budozone. Our FLIM–FRET study showed that M2 clusters with raft-associated HA in transfected cells, but not with the double-acylated raft marker (Thaa et al., 2010), an observation that seems inconsistent at first glance. However, the raft marker, when expressed in the absence of HA, is probably present in small, unstimulated rafts, to which M2 has no access (Fig. 3B). If HA is expressed, it organizes the larger viral budozone, into which the raft marker can partition, and M2 apparently interacts with HA or with the membrane domain functionalized by HA – even in the absence of the matrix protein M1, which directly binds to M2 (Chen et al., 2008) and probably also to HA (Chen and Lamb, 2008).

Thus, M2 must have an intrinsic signal other than interaction with M1 that targets the protein to the viral budozone, most likely the amphiphilic helix in the cytoplasmic tail of M2, which comprises two potential raft association features: S-acylation and cholesterol binding mediated by CRAC motifs. We showed that the cytoplasmic tail of M2 associates with membranes, both in vitro and in cells, and that the membrane binding properties are modulated by exchange of the acylation site and the CRAC motif tyrosines. Furthermore, M2 partly associates with the coalesced raft phase in cooled giant plasma membrane vesicles (GPMVs), cell-derived model membranes, a property which is dependent on acylation, but not on intact CRAC motifs (Thaa et al., 2011). Thus, in principle M2 can interact with large raft domains stabilized by HA (FLIM–FRET in live cells) or cooling (in GPMVs). The exact molecular basis for this needs further experimental assessment. Yet, enrichment at the interface between the liquid-ordered and -disordered phase, as has been described for the ras protein (Vogel et al., 2009), was not observed for M2 in the GPMV system.

Effect of raft-targeting signals in HA and M2 on virus replication

The raft-targeting signals of HA and M2 clearly affect localization of the molecules in cells or in model membranes. To analyse whether they are also required for virus replication they were exchanged in the context of the viral genome. However, it was not possible to generate influenza virus mutants with more than one S-acylation site removed from HA, which implies that this modification is essential for virus growth. In addition, mutants with one acylation site deleted showed severe defects in virus replication and revealed, depending on the HA subtype, either a defect in budding or in HA-mediated membrane fusion (Chen et al., 2005; Wagner et al., 2005). This might be due to reduced raft association of underacylated HA: enrichment of HA in rafts is thought to provide both a platform for virus assembly and a sufficiently high density of HA to catalyse membrane fusion. The essential role of S-acylation for virus replication is further illustrated by the presence of at least three acylation sites in all HA sequences present in the NCBI database (Veit, 2012).

Likewise, a virus with a deletion of the raft-targeting signal in the HA–TMR also showed severe growth defects. Mutant HA was not clustered, but randomly distributed at the surface of virus-infected cells. As a consequence mutant viruses showed reduced budding, contained reduced amounts of HA and exhibited decreased virus–membrane fusion activity. Thus, raft targeting of HA clearly affects virus infectivity (Takeda et al., 2003).

In contrast, removal of the raft-targeting signals from M2 only marginally affects the production of virus particles. Recombinant viruses in which the acylated cysteine (Castrucci et al., 1997) or parts of the CRAC motifs (Stewart et al., 2010) were replaced grew similarly well as the corresponding wild-type virus, and deletion of both the CRAC motifs and the acylation site simultaneously did not affect virus production either (Thaa et al., 2012). However, attenuation of virus infectivity was observed in mice both for virus with non-acylated (Grantham et al., 2009) and CRAC-disrupted M2 (Stewart et al., 2010). Taken together, these results indicate that the acylation and cholesterol-binding motifs in M2 are not crucial for the replication of influenza virus. Note that these motifs are not strictly conserved. It is thus likely that M2's function for budding depends on the overall structure of the amphiphilic helix rather than the individual raft-targeting motifs (Stewart and Pekosz, 2011). Additionally, targeting of M2 to the budding site in infected cells is also mediated by the matrix protein M1, which binds to the cytoplasmic tail of M2 and bridges the viral components in the budozone (Chen et al., 2008). The fine detail of M2's interactions in the virus bud also determines shaping of the virus particle (spherical or filamentous) (Rossman et al., 2010a). Indeed, it is obvious that apart from protein–lipid interactions also protein–protein interactions are essential for formation of the viral budozone. This is indicated by efficient squeezing out of cellular host membrane proteins from the budozone. Thus, assembly and budding of influenza virus is a very robust and partly redundant process and it is the cooperative function of all components that governs budozone organization.

Acknowledgements

Our work was funded by the DFG through priority program 1175 ‘Dynamics of cellular membranes and their exploitation by viruses’ and partly through SFB 740, ‘From molecules to modules’. The authors declare no conflict of interest.

Ancillary