Palmitoylation Strengthens Cholesterol-dependent Multimerization and Fusion Activity of Human Cytomegalovirus Glycoprotein B (gB)*

Herpesviruses are a large order of animal enveloped viruses displaying a virion fusion mechanism of unusual complexity. Their multipartite machinery has a conserved core made of the gH/gL ancillary complexes and the homo-trimeric fusion protein glycoprotein B (gB). Despite its essential role in starting the viral infection, gB interaction with membrane lipids is still poorly understood. Here, evidence is provided demonstrating that human cytomegalovirus (HCMV) gB depends on the S-palmitoylation of its endodomain for an efficient interaction with cholesterol-rich membrane patches. We found that, unique among herpesviral gB proteins, the HCMV fusion factor has a Cys residue in the C-terminal region that is palmitoylated and mediates methyl-β-cyclodextrin-sensitive self-association of purified gB. A cholesterol-dependent virus-like particle trap assay, based on co-expression of the HIV Gag protein, confirmed that this post-translational modification is functional in the context of cellular membranes. Mutation of the palmitoylated Cys residue to Ala or inhibition of protein palmitoylation decreased HCMV gB export via Gag particles. Moreover, purified gBC777A showed an increased kinetic sensitivity in a cholesterol depletion test, demonstrating that palmitoyl-gB limits outward cholesterol diffusion. Finally, gB palmitoylation was required for full fusogenic activity in human epithelial cells. Altogether, these results uncover the palmitoylation of HCMV gB and its role in gB multimerization and activity.

Human cytomegalovirus (HCMV) 2 belongs to the Herpesviridae sub-family in the Herpesvirales order of enveloped viruses (1). The broad cell tropism and the relapsing life-long infection make this widespread virus associated with a large number of inflammation-related diseases, and a major cause of neuro-developmental disorders as well as of immunosenescence (2)(3)(4). Several studies report the remarkable complexity and diversity of the fusion machinery of herpesviruses, required to start the target cell infection. The herpesviral fusion mechanism evolved around two type I integral membrane glycoproteins, the conserved tri-protomeric fusogenic protein gB and the triggering heterodimer gH/gL (5). A number of additional species-specific envelope polypeptides, expressed either as independent transmembrane glycoproteins or as gH/gL complex subunits, complement the two herpesviral core elements providing receptor binding and cell tropism determinants (6 -10). HCMV expresses two gH/gL forms, the gO-and the pUL131A-pUL130-pUL128-bearing complexes. The gH pentamer in particular is essential to infect monocytes and endothelial, epithelial, and dendritic cells, as well as for the leukocytemediated spread of the infection. The ultimate consequence of the receptor-dependent lateral interactions between the envelope fusion players is the activation of the fusogenic refolding in the gB ectodomain (11,12). According to the current models of virion-mediated fusion (13), the rearranging gB is supposed to insert two hydrophobic loops per protomer into the target membrane while it folds back and pulls the two membranes close to each other (14 -16). This mechanical force is expected to go along with the heat released by the conformational shift, melting and joining together the two bilayers. Hence, the interaction with lipids is central to the activity of viral fusion proteins. Indeed, this is not only limited to membrane joining. For the maturation of the initial outer leaflet splicing (the hemifusion state) into a large pore, the fusion protein units have to multimerize (17,18). The quantitative rearrangement of virionic gB molecules observed during HCMV entry (11) suggested an uneven distribution of this protein onto the virion surface during the fusion execution. Strikingly, cryo-electron tomography of purified herpes virions has provided strong evidence for post-fusion gB clustering within the viral envelope (19). Although homotypic protein-protein contacts in the gB ectodomain are likely involved, the zonal restraint of gB seen by Grünewald and co-workers (19) in the virion envelope is reminiscent of partitioning within membrane microdomains. Coherently, virionic gB was previously found partially associated to detergent-resistant membranes, whereas cholesterol depletion from purified virions and target cells inhibits the infection (20). Moreover, purified post-fusion full-length HCMV gB stably associates with host cell-derived lipids (21). In the present work, the gB-lipid relationship has been investigated and the S-palmitoylation of HCMV gB endodomain is reported. A virus-like particle (VLP)-based assay showed that palmitoylation modulates gB targeting to cholesterol-rich membrane domains. Furthermore, acylation was responsible for the stable coalescence of purified gB bicelles in a cholesteroldependent fashion. Finally, palmitoylation affected gB cell surface expression and strongly contributes to gB fusion activity. The sequence comparison of gB protein homologs indicates that acylation is a distinctive feature of HCMV gB. Hypotheses on the virion-mediated fusion mechanism and for synthetic vaccine design are discussed.

Mutagenesis, Expression, and Purification of gB Protein
Variants-All of the PCR and cloning reactions were carried out with Phusion DNA polymerase (Thermo Scientific) and In-Fusion enzyme (Clontech), respectively. The primers used are listed in Table 1. Full-length HCMV UL55 ORF strain VR1814 with a Twin-Strep-tag TM (21) and with the dispensable Furin site ablated (i.e. changing Arg-456/459 to Ala; primer mutFur) was used as a wild type reference. C777A mutation was introduced to generate gB C777A -coding ORF. The recombinant ORFs were inserted into the pOET1 transfer vector (primers OFw and ORv), sequence was validated, and recombinant baculovirus vectors were generated with flashBACGOLD (Oxford Expression Technologies) following the manufacturer's guidelines. Recombinant gB proteins were expressed and purified as already described (21) with some modifications. Briefly, infections of High Five TM cells were performed in Optimum Growth TM 5-liter flasks (Thomson Instrument Company). Glucose and glutamine were kept at 30 and 7 mM, respectively. gB protein variants were purified onto a 5-ml Strep-Tactin HP column (GE Healthcare), omitting the use of dithiothreitol. Resin bound-proteins were extensively washed with 10 mM methyl-␤-cyclodextrin (M␤CD, Sigma-Aldrich) before elution. gB and gB C777A proteins were obtained as spontaneously formed 600-kDa bicelles, containing host-cell lipid carry-over as reported previously (21). Final yields were 0.8 mg/liter for either gB or gB C777A ; both proteins were stored in 100 mM Tris⅐HCl, 0.15 mM NaCl, 0.05% w/v cymal-5 (Anatrace), 20% v/v glycerol, pH 8 (protein buffer).
VLP Production and Quantification-A recombinant baculovirus vector encoding the HIV Pr55 Gag was generated as described above, with the DNA encoding HIV Pr55 Gag transferred from pMDLg/pRRE (22). Pr55 Gag was co-expressed with either gB or gB C777A by co-infecting High Five TM cells at 10 6 cell/ml with three plaque-forming units/cell of each baculovirus derivative. For cholesterol depletion/repletion experiments, cells at 45 h after infection were incubated 45 min with or without 10 mM M␤CD. M␤CD wash-out was followed by a 2-h incubation in medium supplemented with 0.3 mM cholesterol (Sigma-Aldrich) or vehicle alone (0.1% DMSO, 0.01% chloroform). In some experiments, cells were incubated for 18 h in culture medium containing 50 M 2Br-palmitate (Sigma-Aldrich). VLPs were harvested from Ն95% viable cell culture supernatants as described elsewhere (23). In brief, cell culture supernatants were 0.45-m-filtered and centrifuged at 125,000 ϫ g for 1 h. The pellet was resuspended, loaded over a 20% sucrose cushion, and centrifuged at 200,000 ϫ g for 1.5 h.

Name Sequence
OFw Final pellets were measured by nanoparticle tracking analysis (NTA) with NS500 (NanoSight, Malvern Instruments), and corresponding VLP titers were obtained (all the quantifications reported in this study are expressed as mean Ϯ S.D., n ϭ 3). Collected samples were also probed in immunoblots for Pr55 Gag or gB. Acyl-Biotinyl Exchange-An acyl-Biotinyl exchange assay (24) was performed according to Brigidi and Bamji (25) and adapted to Strep-Tactin chromatography. In brief, insect cells expressing gB or gB C777A were washed and lysed in the presence of 20 mM Tris [2-carboxyethyl]phosphine hydrochloride and 50 mM N-ethylmaleimide (both from Sigma-Aldrich) to quench unmodified cysteine residues. Clarified cell lysates were incubated with Strep-Tactin resin using an Amicon Pro purification device with a 100-kDa nominal molecular weight cut-off filter (Merck Millipore). The acyl-biotin exchange was initiated by washing the resin with 0.5 M hydroxylamine. Acylated Cys residues were labeled with 1 M BMCC-biotin (Pierce) in the elution buffer containing 2.5 mM biotin. Eluted and concentrated material were revealed by immunoblot with mAbs anti-gB or anti-biotin.
IRES sequence was amplified from pIRES2-EGFP vector (Clontech) with D7iLF1/D7iLR1 primer pair. The resulting amplimer was PCR-spliced with firefly luciferase cDNA, the latter amplified from pGL4 vector (Promega) using the D7iLF2/ D7iLR2 primers and further extended at the 3Ј end with the D7iLR2b primer. The resulting p7IRESLuc hybrid DNA molecule had the T7 promoter and terminator sequences to the 5Ј and 3Ј end of a IRES-luciferase transcription unit, respectively.
Capped and polyadenylated mRNAs for gB variants, gH pentamer sub-units, and T7 RNA polymerase were individually synthesized from the respective PCR products with the mMESSAGE mMACHINE T7 Ultra Kit (Ambion, Life Technologies) and purified according to the manufacturer's instructions. Each transcription product was individually complexed with the transfection vehicle for RNA cell transfection (TransIT-mRNA Transfection Kit, Mirus Bio), and mRNA-vehicle complexes were mixed afterward to give the combinations described under "Results" (we found this is to be a more efficient procedure for co-transfections). DNA transfection of p7IRESLuc PCR product was with FuGENE HD (Promega). In all of the experiments, mock transfections were performed for baseline correction.
Detergent-free Raft Flotation-The method developed by Persaud-Sawin et al. (26) was used. Briefly, 10 8 ARPE cells were mRNA-transfected in culture medium containing 10% dialyzed FBS to express either gB or gB C777A . Cells were harvested 18 h after transfection and lysed at 4°C in a Dounce homogenizer in 250 mM sucrose, 20 mM Tris⅐HCl, 1 mM CaCl 2 , 1 mM MgCl 2 , cOmplete protease inhibitor cocktail (Roche Applied Science), pH 8. The post-nuclear supernatant was mixed 1:1 (v/v) with the lysis buffer containing 85% sucrose, placed at the bottom of ultracentrifuge tubes, and overlaid with 35% and then 5% sucrose in the same buffer. Samples were centrifuged at 200,000 ϫ g for 18 h at 4°C. Fractions were collected from the bottom and probed in immunoblot for gB, flotillin 1, and the transferrin receptor. In some experiments, cells were cholesteroldepleted with 10 mM M␤CD or incubated with 50 M 2Brpalmitate as described above.
Analysis of gB Multimerization and Competition Dialysis-gB multimerization was induced by mixing purified gB or gB C777A , obtained in their spontaneous 600-kDa form, with 50 M cholesterol dissolved into the protein buffer for 1 h at 37°C, without further manipulations. Multimers were visualized by blue native protein electrophoresis (BN-PAGE) in a 4 -16% NativePAGE TM Novex Bis-Tris pre-cast gel system (Life Technologies) and analyzed in a ChemiDoc XRSϩ with Quantity One 1-D analysis software (Bio-Rad) within the 0.5-5-g linear range. gB monomers were quantified against cholesterolfree gB. Results were expressed as monomer fractional abundance and fitted with a regression analysis (GraphPad Prism 6.0, GraphPad Software).
For multimer dissociation experiments, 100 l of either gB or gB C777A cholesterol-induced multimers were loaded into a 2-kDa nominal molecular weight cut-off micro-dialysis device (Pierce, Thermo Scientific). The dissociation was started by placing the sample at 37°C against a reservoir of identical volume filled with buffer containing equimolar cymal-5:M␤CD concentrations as follows. For the steadystate equilibrium dialysis, each gB variant (1.7 M final concentration) was incubated for 1 h with the indicated M␤CD concentrations in the reservoir. For the time-course analysis, 1.7 M gB or gB C777A was incubated with 10 mM M␤CD in the dialysis reservoir, and the protein was sampled at increasing time points. The samples were analyzed by BN-PAGE and densitometry as above.
Analysis of gB Surface Expression-ARPE cells that were mock, gB, or gB C777A mRNA-transfected as above were processed 18 h after transfection for CELISA (procedure described in Ref. 11), and steady-state surface expression was measured with 2F12 mAb. gB surface dynamic trafficking was measured 16 h after mRNA transfection by incubating ARPE cells in 50 mM NH 4 Cland 2F12 mAb-containing medium for 30 min. Then cells were washed and lysed in 1% Triton X-100 PBS, and clarified cell lysate was loaded into polystyrene microplates (Nunc Max-iSorp, Thermo Scientific). Internalized anti-gB antibody was quantified in ELISA with anti-mouse horseradish peroxidase conjugate and o-phenylenediamine dihydrochloride (Sigma). Some experiments were performed in the presence of 2Brpalmitate as above. Values were plotted after mock signal subtraction.
Flow cytometry on mock, gB, or gB C777A mRNA-transfected ARPE cells was performed 16 h after transfection. Cells were detached by using TrypLE TM Select (10ϫ) reagent (Thermo Scientific), washed in cold 1ϫ Dulbecco's PBS, 0.5 mM EDTA, and resuspended in cold 3% formaldehyde at 10 5 cells/ml under continuous agitation. Cells were labeled without permeabiliza-tion with mAb 2F12, stained with R-phycoerythrin-CF594conjugate anti-mouse (BD Biosciences), and analyzed in a BD FACSCanto TM II (BD Biosciences). Plasma membrane protein fractionation from gB or gB C777A transfectant ARPE cells was performed according to Schindler and Nothwang (27) and used the plasma membrane protein extraction kit (Abcam), following the manufacturer's instructions.
Fusion Assay in Human Epithelial Cells-Parallel ARPE monolayers were transfected with gB or gB C777A , gH pentamer, and T7 RNA polymerase mRNAs (donor cells) or with p7IRESLuc (target cells). After 6 h, cells were extensively washed with culture medium containing 50 g/ml heparin to stop the transfection. Target transfectants were then co-seeded 1:3 with donor cells and incubated for an additional 12 h. In some experiments, the fusion assay was performed in the presence of 50 M 2Br-palmitate and 10% dialyzed FBS. Luciferase activity (Luciferase Assay System, Promega) was then quantified in a plate luminometer (Clarity TM , BioTek). To visualize cell syncytia, ARPE cells were mRNA-transfected as for the donor cells above, stained supravitally with DAPI and imaged under an inverted microscope (Axio Observer, equipped with an AxioCam MRm CCD camera, Zeiss).

Results
The recombinant 360-kDa tri-protomeric gB was previously reported to carry over lipids from host cells, forming a spontaneous bicelle of ϳ600 kDa (hereafter referred to as gB monomer) when purified under reducing conditions (21).
gB Associates with Rafts at the Plasma Membrane-To investigate whether the lipid domain surrounding gB was derived from membrane microdomains, and to study gB targeting to cholesterol-rich microdomains, a VLP-based assay was set up in insect cells. The HIV capsid assembles at plasma membrane lipid rafts (28), sampling and exporting raft-associated proteins during budding. At the same time, distantly related insect host cells allow us to focus on autonomous gB targeting elements, conserving the post-translational modifications.
In baculovirus-infected insect cells, Pr55 Gag is released as pleomorphic immature particles in a protease-and Rev-independent manner (29). The diameter of the particles obtained ranged between 150 and 250 nm, with a peak at 200 nm (Fig.  1A). The comparison with the control from Pr55 Gag -negative infected cells showed that contamination by Autographa californica multicapsid nucleopolyhedrovirus (AcMNPV) virions or structured cell debris was undetectable from samples collected at Ն90% cell viability (detection limit 10 6 particles/ml). Co-expressed gB co-sedimented with Pr55 Gag VLPs (Fig. 1B). A faint gB signal was present in the Pr55 Gag -negative control, likely due to a low degree of cell lysis or exosome secretion. Sterol depletion/repletion tests verified that Gag-mediated gB export was cholesterol-dependent (Fig. 2). Subtraction of sterols by 10 mM M␤CD effectively inhibited Pr55 Gag particle release by ϳ85% (from 3.2 Ϯ 1.8 to 0.54 Ϯ 0.23 ϫ 10 11 particles/ liter Ϫ1 in control versus sterol-depleted cultures, Fig. 2A). Supplementing stripped cells with exogenous cholesterol partially restored Pr55 Gag VLP concentration (1.9 Ϯ 1.2 ϫ 10 11 particles/liter Ϫ1 ). gB export displayed the same trend, with a significant reduction of its abundance in the supernatants from M␤CD-treated cells that could be amended by back-supplementation of cholesterol (Fig. 2B). Incorporation into Pr55 Gag VLPs is, then, a consistent assay for evaluating raft association of membrane proteins. Notably, the extent of the M␤CD effect suggested that membrane microdomains influence gB residence time on the cell surface. On the other hand, comparing VLP titers with immunoblot signals (Fig. 2,  A and B) suggests that cholesterol repletion shock caused the release of particles barely resistant to the VLP harvesting procedure. Indeed, Pr55 Gag immunoblot signals did not reflect the particle titers recorded by NTA, the latter having a 30 nm detection limit and possibly unable to sense fragmented particles.
Palmitoylation of gB Endodomain-HCMV gB has a single Cys residue in the endodomain (Cys-777), close to an Arg-rich stretch featuring the juxtamembrane region of this protein (Fig.  3A). Hence, Cys-777 resembles the transmembrane proximal palmitoylation sites found in other fusion proteins from unre-FIGURE 1. HCMV gB co-fractionates with Pr55 Gag VLPs. A, NTA of the particles recovered from culture medium of co-infected cells expressing gB with (Pr55 Gag ) or without (Mock) HIV Pr55 Gag , following the protocol described under "Experimental Procedures." Plots showing particle size (nm, on x axis), scattered light intensity (relative units, y axis), and concentration (10 6 /ml, z direction) were generated with NanoSight NTA 2.3 software (NanoSight, Malvern Instruments). The parental AcMNPV vector replaced Pr55 Gag -expressing baculovirus in co-infections of mock controls. B, immunoblot (IB) with anti-gB or -p24 antibody on the VLP fraction and mock control shown in A; the same volume of each preparation was loaded for blotting. lated enveloped viruses (30). 3 In an acyl-biotin exchange assay, gB was labeled with BMCC-biotin specifically after thioester bond cleavage with hydroxylamine (Fig. 3B). Replacing Cys-777 with an Ala residue abolished gB thio-biotinylation, demon-strating that the HCMV gB endodomain actually undergoes S-palmitoylation.
Palmitoylation does not necessarily correlate to protein localization into membrane microdomains (31), although palmitoylation dominates among raft proteins (32). Indeed, S-acylation turned out to enhance gB sorting with cholesterolrich microdomains (Fig. 4). Although the intracellular levels were similar to those of parental gB (Fig. 4A), the gB C777A mutant was inefficiently exported in the VLP trap assay (Fig.  4B). Nevertheless, gB C777A export was also sensitive to sterol   depletion by M␤CD. Inhibition of protein-acyl transferases with 2Br-palmitate confirmed that gB raft association was enhanced by S-acylation (Fig. 4C). VLPs produced from cells incubated with the palmitoylation inhibitor contained a very low amount of gB, whereas neither incorporation of gB C777A nor Pr55 Gag export was substantially affected. gB partition into lipid rafts was confirmed in flotation tests on extracts from mRNA-transfected human epithelial cells ( Fig. 5; see also "Experimental Procedures"). gB was present in the fractions positive for human flotillin 1, whereas most of the gB C777A co-migrated with the non-raft marker transferrin receptor. gB variants were absent in the top gradient fractions of the extracts from cells treated with M␤CD, whereas 2Brpalmitate reduced the amount of gB associated with rafts.
Both VLP trap and flotation assays showed that palmitoylation increases gB basal association with cholesterol-rich membrane domains, although none of the above approaches distinguish interaction strength from kinetic effects. Taking advantage of purified full-length HCMV gB (21), we further analyzed the influence of palmitoylation on the gB-cholesterol relationship.
Palmitoyl-gB Forms Stable Cholesterol-dependent Multimers-Both gB and gB C777A recombinant variants were obtained at homogeneity with similar yields, confirming that C777A mutation does not affect gB stability ( Fig. 6A; see also "Experimental Procedures"). To preserve palmitoylation, the use of dithiothreitol was omitted and monomeric gB was obtained by stripping host cell sterols in the chromatographic steps. As expected from previous observations (21), gB C777A was found embedded in a particle of the same size as the wild type protein (Fig. 6B). More interestingly, cholesterol subtraction did not affect the integrity of the bicelle surrounding gB, proving that neither cholesterol nor palmitoylation is required for gB monomer embedding. At the same time, palmitoylation does not confer generic stickiness to gB. The effect of cholesterol on gB was evaluated in a 2-fold dilution series of either wild type gB or the C777A mutant in protein buffer containing a final concentration of 50 M cholesterol. In this way, the same cholesterol:detergent equilibria were kept across the series. Cholesterol consistently triggered gB oligomerization at different molar ratios, with 2-and 3-fold gB multimers visible in BN-PAGE but undetectable gB monomers (Fig. 6C). Conversely, non-palmitoylated gB C777A displayed a weaker response to cholesterol, with ϳ25% of the protein remaining in the monomeric form (0.23 Ϯ 0.02 monomer fractional abundance). For both wild type and the C777A mutant gB, protein dilution and increasing cholesterol:protein molar ratio compensated for each other, ruling out that multimerization was merely due to a solvent effect.
Palmitoylation Reduces Cholesterol Diffusion from gB Multimers-gB-cholesterol interaction was analyzed by competition dialysis (see "Experimental Procedures"). Differential sensitivity to M␤CD of cholesterol-induced multimers with purified gB and gB C777A showed that the palmitoyl moiety increases gB clustering by stabilizing cholesterol-dependent coalescence (Fig. 7). In standard competition dialysis experiments, M␤CD reverted cholesterol-induced gB association in a dose-dependent manner after a 1-h incubation (Fig. 7A), although a low amount of oligomeric gB was still present at the highest M␤CD concentration. However, gB and gB C777A displayed similar sensitivity at the steady state with 4.2 Ϯ 0.012 and 2.5 Ϯ 0.015 mM M␤CD EC 50 , respectively. This finding argues against a role of palmitoyl-Cys-777 in putative cholesterol binding sites within the gB transmembrane domain. By contrast, a clear difference for multimer dissociation was detected at a kinetic level (Fig. 7B). When incubated with 10 mM M␤CD, gB C777A multimeric forms exhibited a fast decay curve, whereas gB dissociation followed a sigmoidal trend with a 6-fold longer multimer half-life (t1 ⁄ 2 3.9 Ϯ 1.2 and 22.2 Ϯ 0.3 min for gB C777A and gB, respectively).
In agreement with VLP trap and raft flotation experiments, multimerization tests confirmed that palmitoylation enhances the basal gB responsiveness to cholesterol. The steady-state and kinetic analyses jointly provide the evidence that diffusion of cholesterol out of the purified gB multimers was limited by the presence of the palmitoyl-Cys residue. In turn, the stiffness of such a milieu seems sufficient to prevent palmitoyl-gB multimers from splitting into monomers, as shown by the palmitoylation-dependent high tendency of gB to stay clustered in the presence of cholesterol.
Efficient gB Fusion Activity Requires Endodomain Palmitoylation-The observations collected pointed out that cholesterol-and acylation-dependent multimerization reflects the strength of raft association around the HCMV fusion factor. Measuring gB-mediated membrane fusion in HCMV-permissive epithelial cells highlighted that gB endodomain palmitoy- lation is also required for full gB activity (Fig. 8) The functional influence of palmitoylation on gB-mediated fusion was evaluated in a luciferase reporter assay, using RNA transfections. This experimental setup avoids membrane saturation and may better approximate the temporary nature of the virion-target cell interaction.
As for insect cells, gB and gB C777A had similar intracellular expression levels, with the palmitoylation-deficient gB C777A  mutant showing a 4-fold reduction at the plasma membrane of ARPE cells (Fig. 8A). The lower steady-state surface abundance was not due to defective gB C777A targeting. An anti-gB antibody was equally captured by cells expressing either wild type gB or the C777A mutant, demonstrating that similar protein amounts reached the plasma membrane. Along with the VLP trapping in insect cells, the latter result demonstrated that thioacylation affects cell surface gB residence time independently of the cell origin.
Fusion activity was quantified in a well established reporter assay. The assay is based on the cytoplasmic transcription by T7 RNA polymerase of a reporter transcription unit in target  A. U., absorbance units. B, gB or gB C777A mRNA was co-transfected with or without mRNAs coding for gH pentamer subunits (gH 5) and T7 RNA polymerase (T7RNAP) into ARPE cells and then mixed with cells transfected with p7IRESLuc DNA (see "Experimental Procedures"). gH 5-and T7RNAP-positive, gB-negative transfections were used as control. In one experimental group, the fusion assay was carried out in the presence of 50 M 2Br-palmitate (2Br-p). Luciferase activity was recorded and plotted after subtracting mock transfection signals. Error bars indicate mean Ϯ S.D.
transfectants, when membrane fusion takes place with polymerase donor cells. Luciferase activity was consistently above T7RNAP-negative control in a gB and gH pentamer-dependent manner (Fig. 8B). Importantly, the reporter signal was substantially lower when gB C777A replaced wild type gB as the fusion factor or in the presence of 2Br-palmitate; hence the assay output was sensitive to the level of gB surface expression and protein palmitoylation. Nevertheless, the extent of reduction was larger than the gB C777A deficit at the plasma membrane.
Palmitoylation turned out not to be a primary signal for targeting gB to the plasma membrane (Fig. 9). Flow cytometry, CELISA, and two-phase plasma membrane protein fractionation consistently showed that gB C777A is not prevented from reaching the cell surface. Actually, gB C777A expression at the cell surface was increased by transfecting four times the amount of the corresponding mRNA relative to the wild type protein. Moreover, the increase in the plasma membrane levels obtained by overexpressing gB C777A was poorly sensitive to the palmitoylation inhibitor (Fig. 9, B and D). Normalized cell surface expression showed a substantial dependence on S-palmitoylation by the gB ability to promote membrane fusion (Fig.  10). In fact, use of both 2Br-palmitate and gB C777A resulted in about 80% reduction of the luciferase signal (Fig. 10A). The fusion defect in the palmitoylation-deficient gB C777A mutant was also detectable as a reduced syncytiogenic ability with respect to wild type gB in transfected cells (Fig. 10B). Co-transfecting ARPE cells with gB and gH pentamer mRNAs resulted in the formation of large multinucleated cells made of several nuclei, whereas replacing gB with gB C777A led to smaller syncytia with few nuclei.

Discussion
Cholesterol-rich membrane microdomains play critical roles as self-assembling partitions of specific proteins. Recent advances in this field are demonstrating that raft integral mem- Error bars indicate mean Ϯ S.D. OD, optical density; A. U., absorbance units. C, transfected ARPE cell extracts were fractionated by the two-phase method (see "Experimental Procedures"), and partition of gB or gB C777A between total cell membranes (TM), cytoplasmic membranes (CM), and plasma membrane (PM) was analyzed in immunoblots loaded with 10 g of total proteins and compared with the plasma membrane marker ␣1 Na ϩ /K ϩ ATPase (NKA). D, 60 g of total proteins of plasma membrane fractions from ARPE cells transfected and processed as in C, with or without prior incubation in 50 M 2Br-palmitate (2Br-p), were probed in immunoblot for gB or ␣1 Na ϩ /K ϩ ATPase. FIGURE 10. Palmitoylation is required for full gB fusion activity. A, ARPE cells were transfected with the indicated relative amounts of gB-or gB C777Acoding mRNAs, along with gH pentamer subunits and T7 RNA polymerase mRNAs; the luciferase-reporter fusion assay was carried out as in Fig. 8B in the presence (light gray) or absence (dark gray) of 50 M 2Br-palmitate. Error bars indicate mean Ϯ S.D. B, ARPE cells were co-transfected with gH pentamer and the indicated relative amounts of gB-or gB C777A -coding mRNAs. Cell nuclei were stained with DAPI; multinucleated cells are indicated by white arrows.
brane proteins (rIMPs), rather than mere cargoes, stabilize the inherently short-lived assembly of cholesterol with companion lipids, by hydrophobic matching with their transmembrane domains (TMs). The raft phases surrounding rIMP TMs are proposed to grow to larger platforms by further supramolecular interactions, often involving protein acylation (33). One pair of cholesterol molecules was shown to pack within the hollow formed by two palmitoyl residues in the groundbreaking x-ray structure of the mono-palmitoylated ␤ 2 -adrenergic receptor (34). The cholesterol-palmitic interaction seemed to rely on steric packing only. Whether this arrangement in the depth of a non-polar environment could lead to a stable binding was not addressed. Based on data collected in this work, we propose that palmitoylation drives the growth of cholesterol-dependent protein multimers by increasing protein-cholesterol friction. We showed that HCMV gB features the intrinsic ability to localize into cholesterol-dependent microdomains and that the newly discovered palmitoylated Cys-777 considerably enhances a basal, low-grade gB partition (Figs. 1-5). The multimerization analysis of purified gB demonstrated that the residence time of cholesterol within multimers is palmitoylation-dependent (Figs. 6 and 7) and correlates with multimer stability, although it does not support a role for acylation in cholesterol-gB stable binding. Further work is needed to clarify whether gB TM protein backbone actually holds any cholesterol site. Notably, our results are reminiscent of the high occupancy site hypothesis raised from molecular simulations of G protein-coupled receptor-cholesterol interactions (35), in which it is suggested that cholesterol-protein interactions are highly dynamic and that they swiftly exchange between adjacent cholesterol sites. The clustering model by high exchange rate under limited interactor diffusion, proposed by Woolf and Linderman (36), explains the stability of gB multimers by cholesterol diffusion confinement. It is plausible that this phenomenon is strengthened by the three-fold symmetry of the gB monomer, and we suggest that palmitoylation generates cholesterol hot-spots between gB monomers. In turn, this may build a defined layer where cholesterol from the target cell membrane and the virion envelope can mix. Indeed, gB raft localization and clustering have been so far shown to be related to virion entry and post-fusion gB (19,20), and our experiments on multimerization also made use of post-fusion full-length gB, as characterized by electron micros- FIGURE 11. Alignment of gB protein homologs encoded by human herpesviruses. Overall sequence homology of gB conceptual translation from human herpesviruses (see Footnote 3) is shown schematically in grayscale (light gray, no homology; black, identity). Cys residues are highlighted with black squares above each sequence (asterisks mark conserved Cys); arrow points to HCMV gB Cys-777. The close-up view shows the alignment in the juxtamembrane-TM region. VZV, varicella zoster virus; HHV 6A, human herpesvirus 6A; HHV 7, human herpesvirus 7; KSHV, Kaposi's sarcoma-associated herpesvirus.
copy (21). The gB moiety, C-terminal to the ectodomain, displayed a compact fold in EM two-dimensional reconstructions. However, it is yet unknown whether pre-and post-fusion gB conformations differ in the relative orientation of TM helices or in their interaction with cholesterol. At the same time, ectopically expressed gB associates with the raft phase in both insect and human cells, independently from fusion triggers.
The evidence that inhibition of protein palmitoylation and palmitoylation site mutagenesis both impaired gB fusion activity in human epithelial cells (Figs. 8 -10) is significant. These data reinforce the idea that the clustering of virion fusogens provides both the mechanical force and the energy required for the pore formation.
The chance to observe a diffusivity effect in cholesterol-gB multimers makes the point for an interesting speculation about the energetics in virion-mediated membrane fusion. Under mechanical stress, viscous substances release heat as the consequence of internal forces. It is conceivable that the deformation in the hemifused membranes around the refolded gB multimers cooperates with the ectodomain in supplying the energy needed to the pore formation (see Ref. 37 for the correlations in membrane deformation, bending heat, and melting). This hypothesis is supported by the observed binding to the liquid order/disorder inter-phase by the HIV gp41 fusogenic peptide (38).
Palmitoylation also affects gB trafficking at the plasma membrane. Although this is likely not relevant per se in HCMV virogenesis (39), the data reinforce the idea that the supramolecular organization of rIMPs impacts membrane dynamics. Moreover, the virion assembly compartment observed in cells infected by herpesviruses shares similarities with the multivesicular body, whose lipid bilayer is contributed by the plasma membrane. Hence, the relationships that gB has with cell surface lipids could also occur during nucleocapsid envelopment at the end of the virus assembly process. In this view, the interplay of gB with the other two known HCMV palmitoylated proteins, gN and pp28, both linked to virogenesis (40), may deserve deeper investigation in the near future.
Intriguingly, gB palmitoylation seems to be specific to HCMV among herpesviruses, because only this species codes for a Cys residue in the endodomain of the fusion protein (Fig.  11). Elucidating the outcome of such a divergence will shed light on the supramolecular level of herpesviral fusion machineries. The question of whether evolving a palmitoylation site has added new features to HCMV gB, or whether non-HCMV gB TMs can provide analogous cholesterol hot-spots without S-acylation, is fascinating.
Finally and (to some extent) independently from the role that gB multimerization plays in the infection cycle, antigen clustering is a powerful tool in synthetic vaccinology for boosting an immune response. Along with artificial adjuvants, multimerization determinants can be exploited to push the efficacy of subunit vaccine candidates protecting against herpesviral infections.
Author Contributions-M. P., A. S. C., A. P. T., and P. M. A. conceived the study, analyzed the data and wrote the paper. M. P. designed, performed and analyzed the experiments.