Structural and biochemical studies of HCMV gH/gL/gOand Pentamer reveal mutually exclusive cellentry complexesClaudio Ciferria,1, Sumana Chandramoulia, Danilo Donnarummab, Pavel A. Nikitinc, Michael A. Cianfroccod,e,Rachel Gerreina, Adam L. Feirec,2, Susan W. Barnetta, Anders E. Liljaa,3, Rino Rappuolib,4, Nathalie Noraisb,Ethan C. Settembrea, and Andrea Carfia,4

aNovartis Vaccines, Cambridge, MA 02139; bNovartis Vaccines, 53100 Siena, Italy; cNovartis Institutes for Biomedical Research, Emeryville, CA 94608;dDepartment of Molecular and Cellular Biology, Harvard University, Cambridge, MA 02138; and eDepartment of Cell Biology, Harvard Medical School,Boston, MA 02115

Contributed by Rino Rappuoli, January 1, 2015 (sent for review September 8, 2014; reviewed by Ann M. Arvin and Gary H. Cohen)

Human cytomegalovirus (HCMV) is a major cause of morbidityand mortality in transplant patients and the leading viral causeof birth defects after congenital infection. The glycoproteincomplexes gH/gL/gO and gH/gL/UL128/UL130/UL131A (Pen-tamer) are key targets of the human humoral response againstHCMV and are required for HCMV entry into fibroblasts andendothelial/epithelial cells, respectively. We expressed and charac-terized soluble forms of gH/gL, gH/gL/gO, and Pentamer. Massspectrometry and mutagenesis analysis revealed that gL-Cys144forms disulfide bonds with gO-Cys351 in gH/gL/gO and withUL128-Cys162 in the Pentamer. Notably, Pentamer harboring theUL128-Cys162Ser/gL-Cys144Ser mutations had impaired syncytiaformation and reduced interference of HCMV entry into epithelialcells. Electron microscopy analysis showed that HCMV gH/gLresembles HSV gH/gL and that gO and UL128/UL130/UL131Abind to the same site at the gH/gL N terminus. These data areconsistent with gH/gL/gO and Pentamer forming mutually exclusivecell entry complexes and reveal the overall location of gH/gL-, gH/gL/gO-, and Pentamer-specific neutralizing antibody binding sites. Ourresults provide, to our knowledge, the first structural view of gH/gL/gO and Pentamer supporting the development of vaccines andantibody therapeutics against HCMV.

human cytomegalovirus | HCMV | Pentamer complex | gH/gL/gO |virus entry

Human cytomegalovirus (HCMV) is a member of the β-her-pesvirus subfamily with >60% seropositivity in adults world-

wide (1). HCMV infection is typically asymptomatic, but can causesevere disease or death in immunocompromised solid organ andhematopoietic stem cell transplant recipients. In addition, HCMVcan infect the placenta and cross this barrier to infect developingfetuses, causing severe birth defects (2). Given the severity andimportance of this disease, obtaining an effective vaccine is con-sidered a public health priority (3).The ability of HCMV to cause disease in a wide range of organs

and tissue types is reflected at the cellular level by the virus infectingepithelial cells, endothelial cells, fibroblasts, dendritic cells, hepa-tocytes, neurons, macrophages, and leukocytes (4). Similar to otherherpesviruses, the envelope glycoproteins gB and gH/gL form theconserved fusion machinery required for viral entry (5, 6). Recentstructural and mutagenesis analysis suggested that gB is responsiblefor mediating virus and host membrane fusion during viral entry (7,8). The role of gH/gL in fusion is less clear because crystal structuresof herpes simplex virus 2 (HSV-2), pseudo-rabies virus (PrV), andEpstein–Barr virus (EBV) gH/gL did not reveal any similarity toknown viral fusion proteins (9–11). It has been proposed that gH/gLis involved in the entry process through activation of gB (12). Inaddition to gB and gH/gL, most herpesviruses encode additionalglycoproteins that are able to interact with gH/gL and are capable of

either mediating binding to specific cellular receptors or regulatingthe activity of the gH/gL–gB complex (5, 6).HCMV entry into both epithelial and endothelial cells

requires a pentameric glycoprotein complex (Pentamer) formedbetween gH/gL and the UL128, UL130, and UL131A proteins(13, 14). Mutations in the UL131A–UL128 gene locus are suf-ficient to eliminate epithelial/endothelial tropism and occurspontaneously within only a few passages of wild-type (WT)HCMV in fibroblasts (15, 16). In addition, Pentamer cell surfaceoverexpression interferes with HCMV entry into epithelial cells,but not into fibroblasts, suggesting the presence of a cell-type-specific Pentamer receptor (17).HCMV entry into fibroblasts is mediated by the gH/gL/gO

complex at the cell surface at neutral pH (18–21). gO is a highlyglycosylated protein and has been shown to covalently interactwith gH/gL (22, 23). It has been proposed that gO might functionas a molecular chaperone to promote gH/gL incorporation, butnot gH/gL/gO, into the virion (21). However, it has been recently

Significance

Human cytomegalovirus (HCMV) is a major cause of morbidityand mortality in transplant patients and in fetuses after con-genital infection. gH/gL/gO and Pentamer are targets forneutralizing antibodies. We show that gO and UL128/UL130/UL131A bind to the same site on gH/gL through formation ofa disulfide bond with gL-Cys144. The alternative use of thisbinding site by either gO or the ULs may provide a mechanismfor cell tropism modulation. Our analysis reveals that gH/gLantigenic sites are conserved among gH/gL, gH/gL/gO, andPentamer, whereas gH/gL/gO- and Pentamer-specific neutral-izing antibody-binding sites are located in the gH/gL N termi-nus protrusion that contains the gO and the UL subunits. Thesedata support the development of vaccines and antibody ther-apeutics against HCMV.

Author contributions: C.C., S.C., E.C.S., and A.C. designed research; C.C., S.C., D.D., P.A.N.,M.A.C., and R.G. performed research; C.C., S.C., D.D., P.A.N., M.A.C., A.L.F., S.W.B., A.E.L.,R.R., N.N., E.C.S., and A.C. analyzed data; and C.C. and A.C. wrote the paper.

Reviewers: A.M.A., Stanford University School of Medicine; and G.H.C., University ofPennsylvania.

Conflict of interest statement: The sponsor is a full-time employee of Novartis Vaccines, asare S.C., R.G., S.W.B., N.N., E.C.S., and A.C.

Freely available online through the PNAS open access option.1Present address: Genentech, Inc., South San Francisco, CA 94080.2Present address: Novartis Institutes for Biomedical Research, Cambridge, MA 02139.3Present address: Hookipa Biotech AG, 1030 Vienna, Austria.4To whom correspondence may be addressed. Email: rino.rappuoli@novartis.com orandrea.carfi@novartis.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1424818112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1424818112 PNAS | February 10, 2015 | vol. 112 | no. 6 | 1767–1772

BIOCH

EMISTR

Y

demonstrated that gH/gL/gO and Pentamer are much moreabundant on the HCMV envelope than gH/gL alone (24).Highly potent HCMV-neutralizing monoclonal antibodies

were isolated from the memory B-cell repertoire of HCMV-immune donors and shown to bind the Pentamer. These anti-bodies were capable of neutralizing HCMV infection of epithe-lial/endothelial cells, but not fibroblasts (25, 26). In addition,several studies have demonstrated that the Pentamer is the maintarget of the neutralizing humoral response to HCMV infectionin epithelial/endothelial cells (27–29). Consistent with theseobservations, immunization with the Pentamer has been shownto elicit a strong neutralizing antibody response in mouse, rabbit,and rhesus macaque models (30–32). Together these data in-dicate that the Pentamer represents a key antigenic target forvaccine development against HCMV infection.Here we report the purification and biochemical character-

ization of HCMV gH/gL, gH/gL/gO, and Pentamer. In addition,we describe the architecture of these complexes by electronmicroscopy (EM) and characterize their interaction with MSL-109, a previously described HCMV-neutralizing antibody iso-lated from the spleen of a HCMV-seropositive individual (33,34). Our data provide new insights into the structure and func-tion of the HCMV gH/gL/gO and Pentamer complexes.

ResultsgH/gL, gH/gL/gO, and Pentamer Are Stabilized by Internal DisulfideBridges. To investigate the architecture of HCMV gH/gL, gH/gL/gO, and Pentamer, we reconstituted each complex by transientlytransfecting HEK293 cells with plasmids expressing the individualcomponents (Fig. S1). Protein complexes were purified from thecell medium by affinity followed by size-exclusion chromatography(SEC) and then analyzed by SDS/PAGE (SI Materials and Methods;Fig. 1). Consistent with previous reports (24, 32, 35, 36), this analysisrevealed that HCMV gH and gL are covalently linked by disulfidebonds and that additional disulfides form between two gH/gLcomplexes. For simplicity, these dimers of gH/gL heterodimers willbe referred to as gH/gL homodimers (Fig. 1A).The gH/gL/gO complex migrated in SEC with an apparent mo-

lecular weight of ∼220 kDa, consistent with the theoretical molec-ular weight of a 1:1:1 complex (Fig. 1B). The complex disassembledcompletely only upon addition of reducing agents, supporting a roleof disulfide bridges in stabilization (Fig. 1B). Importantly, the gO-containing complex did not dimerize, suggesting that gO interfereswith the gH/gL propensity to homodimerize.Finally, coexpression of gH, gL, UL128, UL130, and UL131A

generated a stable complex migrating in SEC with an apparentmolecular mass of ∼180 kDa (Fig. 1C), suggesting the presence ofone copy for each of the components of the complex. SDS/PAGEunder boiling conditions showed that gH/gL and UL128 are co-valently linked, and reducing agent addition dissociated the com-plex. Similar to gO, UL128/UL130/UL131A (from now on referredto as ULs) interfered with gH/gL homodimerization. Formation ofdisulfide-linked gH/gL/gO and gH/gL/UL128 complexes was pre-viously suggested from coimmunoprecipitation experiments usinggO- and UL128-specific antibodies, respectively (23).

gL–Cys-144 Forms a Disulfide Bond with gO–Cys-351 and UL128–Cys-162. We used mass spectrometry (MS) to investigate which cys-teines are involved in disulfide formation (SI Results). Analysis ofgH/gL showed that gL–Cys-144 on one monomer forms a disul-fide bridge with gL–Cys-144 on a second monomer, demon-strating that the gL subunit is involved in homodimerization (Fig.S2A). To confirm this finding, we reconstituted a gH/gL complexcarrying a Cys-144–Ser mutation. The purified complex appearsto be monomeric as revealed by SDS/PAGE (Fig. 2A) and SEC.In addition, MS analysis identified a disulfide bond between gL–Cys-47 and gH–Cys-95 (Fig. S2B), consistent with the covalentinteraction between gH and gL. Mutation of either cysteine

prevented gH/gL secretion, suggesting that this disulfide is es-sential for gH/gL complex folding (Fig. S3).MS analysis of gH/gL/gO showed that gL–Cys-144 forms a

disulfide bond with gO–Cys-351 (Fig. S2C). We also detected analternative disulfide bond between gL–Cys-144 and gO–Cys-226(Fig. S2D). Reconstitution of gH/gL/gO carrying the gL–Cys-144–Ser point mutation resulted in lack of covalent interactionbetween gO and gH/gL (Fig. 2B), in agreement with the MSdata. The gO-Cys351Ser mutation impaired gO expressionresulting in secretion of gH/gL only (Fig. S4).Finally, MS analysis of the Pentamer showed that gL–Cys-144

forms a disulfide bond with UL128–Cys-162 (Fig. S2E). SDS/PAGE analysis of Pentamer carrying the gL–Cys-144–Ser or

Fig. 1. Analyses of gH/gL, gH/gL/gO, and Pentamer complexes. (A) SEC ofaffinity-purified gH/gL shows two peaks indicative of gH/gL homodimer andgH/gL monomer. SDS/PAGE analysis shows that DTT is required to completelydisrupt the gH/gL homodimer, suggesting that disulfide bonds may cross-linktwo gH/gL heterodimers. (B) SEC of affinity-purified gH/gL/gO shows onemajor peak. Both heat and DTT are needed to dissociate the gH/gL/gO com-plex into gH, gL, and gO. (C) SEC of Pentamer shows one peak. SDS/PAGE hasthree distinct bands: one composed of gH/gL/UL128 held together by internaldisulfide bonds and two others composed of either UL130 or UL131A. Heatand reducing agents disrupt gH/gL/UL128 and free UL128 has a similar size toUL131A. Individual protein identity was confirmed by N-terminal sequencing.

1768 | www.pnas.org/cgi/doi/10.1073/pnas.1424818112 Ciferri et al.

UL128–Cys-162–Ser revealed a lack of covalent interaction be-tween gH/gL and UL128 (Fig. 2C). Therefore, the same cysteinein gL, gL–Cys-144, forms alternative disulfide bonds with gO andUL128 and, in their absence, is available to form disulfide linkedgH/gL homodimers.

Effects of UL128–Cys-162 and/or gL–Cys-144 Mutations on PentamerFunction. To investigate the role of the disulfide bond betweengL–Cys-144 and UL128–Cys-162 in HCMV infection we as-sessed the effect of the single and double Cys→Ser mutations in

Pentamer-mediated interference of viral entry in adult retinalpigment epithelial 19 (ARPE-19) cells (17, 21). Cells weretransduced with adenoviruses (AdVs) expressing the full-lengthPentamer proteins, followed by infection with HCMV (low-passage HCMV clinical isolate VR1814). In this assay, Pentamercan bind and sequester host molecules required for viral entry,thus protecting cells from HCMV infection (17). VR1814 has anaccelerated viral gene expression profile compared with variousother clinical isolates and laboratory strains tested (34). WTPentamer overexpression protected >80% of cells from infection,compared with cells transduced with the LacZ control or gB-gH/gL (Fig. 3A; paired t test; P = 0.0012). In contrast, cells express-ing gH/gL–Cys-144–Ser/UL128/UL130/UL131A (gL–Cys-144–Ser) or gH/gL/UL128–Cys-162–Ser/UL130/UL131A (UL128–Cys-162–Ser) were partially permissive for VR1814 infection, asrevealed by expression of HCMV immediate early genes 24 hafter infection (P = 0.0081 and 0.0251). Moreover, gH/gL–Cys-144–Ser/UL128–Cys-162–Ser/UL130/UL131A (gL–Cys-144–Ser/UL128–Cys-162–Ser) overexpression did not interfere with HCMVinfection, indicating that the mutations functionally impaired theability of the Pentamer to compete with the native complex foundon virions during virus entry (Fig. 3A; P = 0.4520).The same mutants were tested in a syncytia assay in ARPE-19

cells. Coexpression of the full-length VR1814 WT gB and Pen-tamer resulted in syncytia formation of nearly all cells (Fig. 3B),whereas gL–Cys-144–Ser or UL128–Cys-162–Ser did not affectcell-to-cell fusion. Of note, we detected a small, but reproducible,delay in syncytia formation for UL128–Cys-162–Ser that never-theless induced cell fusion comparable to WT 30 h after trans-duction (Fig. 3B). Strikingly, gL–Cys-144–Ser/UL128–Cys-162–Serwas not able to induce epithelial cell fusion (Fig. 3B, bottom row).We next assessed whether Pentamer mutants expressed on the cell

surface could be recognized by MSL-109, a gH-specific monoclonalantibody (33, 34); 10P3, a monoclonal antibody recognizingUL130/UL131A as part of correctly assembled Pentamer (26);and Cytotect, human IgGs from HCMV-positive individuals that

Fig. 2. Anti-gL and -UL128 western blot analyses of purified mutant com-plexes under nonboiling and nonreducing conditions. (A) In the presence ofSDS, purified WT gH/gL runs as a homodimer and monomer, whereas gH/gL–Cys-144–Ser mutant does not form homodimers. (B) The gH/gL/gO complexwas affinity-purified by means of the Strep tag at the gO C terminus. The gL–Cys-144–Ser mutation completely abolishes the covalent interaction betweengO and gH/gL. (C) Pentamer complexes purified by Strep affinity (Strep-taggedgH) showed that mutation of gL–Cys-144–Ser or UL128–Cys-162–Ser interfereswith the covalent interaction between gL and UL128.

Fig. 3. Effect of gL–Cys-144–Ser and UL128–Cys-162–Ser mutations on HCMV interference and cell fusion. (A) HCMV interference assay. HCMV IE expression inARPE-19 epithelial cells transduced with replication-incompetent AdVs encoding individual HCMV VR1814 Pentamer subunits. HCMV WT Pentamer (black bar)interferes with subsequent HCMV VR1814 infection, whereas introduction of gL–Cys-144–Ser (green) or UL128–Cys-162–Ser (purple) single mutations attenuateinterference. Transduction of gL–Cys-144–Ser/UL128–Cys-162–Ser double-mutant Pentamer (blue) completely abolishes HCMV interference. Bar graphs are nor-malized to LacZ and represent four independent experiments. Error bars represent SEM. *P < 0.05; **P < 0.01; ns, not significant. (B) Syncytia assay. Representativeimmunofluorescent images of ARPE-19 cells transduced with HCMV VR1814 glycoproteins are shown. Cytoplasm, whole-cell staining; DAPI, nuclear DNA stainingwith 4’,6-diamidino-2-phenylindole; merge, combined cytoplasmic and nuclear staining together. Images represent four independent transductions. (C) AveragePentamer surface expression measured by FACS using MSL109, 10P3, or Cytotect antibodies is plotted as mean fluorescent intensity (MFI) for cells from A. MFIvalues represent four independent transductions. Error bars represent SEM. Color scheme and abbreviations are the same as in A.

Ciferri et al. PNAS | February 10, 2015 | vol. 112 | no. 6 | 1769

BIOCH

EMISTR

Y

contain a high fraction of Pentamer-specific neutralizing anti-bodies (28, 34). We did not observe significant changes in surfacebinding of these antibodies by FACS (Fig. 3C and Fig. S5).We concluded that the mutations introduced do not affect cell-surface expression and overall Pentamer structure. Consistent withthese data, purified Pentamer containing both gL–Cys-144–Ser andUL128–Cys-162–Ser was antigenically equivalent to the WT protein(32). Together, the interference and syncytia data suggest thatmutations of both gL–Cys-144 and UL128–Cys-162 affect Pentamerinteraction with host factors and its function in cell entry.

Architecture of gH/gL, gH/gL/gO, Pentamer, and Their Complexes withMSL-109. Structures of gH/gL from EBV and HSV have beendescribed (10, 11). EBV gH/gL presents an elongated structurein which all of the gH domains are aligned in the same plane(Fig. S6A). HSV gH/gL, conversely, is organized in a character-istic “boot” shape with a ∼60° kink between domain III and IV atthe C terminus of the molecule (Fig. S6B).The architecture of HCMV gH/gL was investigated by nega-

tive-staining EM and single-particle analysis. To facilitate theEM studies, we analyzed the gH/gL homodimer alone and boundto the fragment antigen-binding (Fab) region of the neutralizingantibody MSL-109. Negative-staining EM 2D class averagesrevealed that gH/gL forms a symmetrical homodimer, with the Ntermini providing the main interface for dimerization (Fig. 4A,Left). The localization of the dimerization interface is consistentwith the involvement of gL–Cys-144 in HCMV gH/gL homo-dimerization and the location of gL close to the N terminus ofgH in the structures of other herpesvirus gH/gLs (9–11). Analysisof the gH/gL/MSL-109 2D class averages revealed the presenceof a kink between domain DIII and DIV, suggesting that HCMVgH/gL is more similar to HSV gH/gL than EBV gH/gL (Fig. S6 Aand B). Finally, a 2D difference map between gH/gL and gH/gL/

MSL-109 confirmed the location of the Fab at the gH/gL “heel”region (Fig. S7A).The same boot-like organization and site of binding for MSL-

109 was also observed in gH/gL/gO and Pentamer (Fig. 4 B andC). Interestingly, 2D analysis of gH/gL/gO and Pentamer aloneand in complex with MSL-109 showed that both gO and the ULsbind the N terminus of gH/gL (Fig. 4 B and C). However, thestructural comparison suggests that no major domain rear-rangements occur in gH/gL upon binding to either gO or the ULs(Fig. 4 A–C). This result is supported by surface plasmon reso-nance (SPR) data demonstrating similar binding affinities ofMSL-109 to both gH/gL/gO and Pentamer (Fig. S8).We also generated a 3D reconstruction of gH/gL/gO and Pen-

tamer bound to MSL-109 Fab (Fig. 5 A and B). These complexesadopted a strong preferential view, which required the use of therandom conical tilt (RCT) reconstruction technique to guide thefitting of HCMV gH/gL and MSL-109 into the electron densitymap. Additional electron density emerging from the gH/gL N-terminal region describes the overall shape of the gO and UL128/UL130/UL131A subunits, respectively (Fig. 5 A and B). Therefore,gH/gL/gO and Pentamer RCT reconstruction comparisons indicatethat the gO and UL subunits interact with a shared surface ongH/gL. This finding is consistent with gL-Cys144 forming an in-termolecular disulfide bond in each of the three complexes.

Hydrogen–Deuterium Exchange Coupled to MS Analysis of gH/gL/MSL-109 Complex.Hydrogen–deuterium exchange coupled to MS(HDX-MS) was used to confirm the EM results and deter-mine the MSL-109 epitope at the amino acid level. The degreeof deuterium incorporation into a deglycosylated gH/gL wasassessed for several peptides, covering ∼70% of the gH/gL se-quence (SI Results; Fig. S9). HDX-MS revealed a reduction indeuterium uptake upon binding of MSL-109 for residues 380–396, 400–419, and 400–423. The epitope was narrowed to resi-dues 380–396 and 418–423 by comparing the deuterium uptakeof peptides 400–419 and 400–423 with the uptake of peptides400–416 and 404–417 (Fig. S10). An HCMV gH/gL homology

Fig. 4. Architecture of gH/gL, gH/gL/gO, and Pentamer alone and bound toMSL-109 Fab. (A) Stained 2D class averages of gH/gL homodimer alone (Left),bound to MSL-109 (Center), and model of the complex (Right). (B) Class aver-ages of gH/gL/gO alone (Left), bound to MSL-109 (Center), and model of thecomplex (Right). (C) Class averages of Pentamer alone (Left), bound to MSL-109(Center), and model of the entire complex (Right). The 2D analysis shows thatboth gO and the ULs bind the N-terminal region of gH/gL, consistent with gL–Cys-144 forming an intermolecular disulfide bond in each of the three com-plexes. The boot-like architecture of gH/gL is conserved in these complexes.

Fig. 5. RCT reconstruction of gH/gL/gO/MSL-109 and Pentamer/MSL-109complexes and epitope mapping of gH/gL/MSL-109 by HDX-MS. (A and B) RCTreconstruction was used to determine gH/gL/gO/MSL-109 (A) and Pentamer/MSL-109 (B) complex structures. (A and B, Upper) For both samples, the 3Dmodel (Right) is shown alongside a 2D reference free class average (Left) and3D model projection (Center). (A and B, Lower) Both RCT structures show theinteraction between gO and ULs and the gH/gL N-terminal region. (C) HDX-MS analysis reveals MSL-109 gH/gL binding site at the amino acid level. Spe-cifically two peptides are identified in the gH C-terminal region.

1770 | www.pnas.org/cgi/doi/10.1073/pnas.1424818112 Ciferri et al.

model based on the HSV gH/gL structure identifies these pep-tides (380–396 and 418–423) in the gH/gL ”heel” region inagreement with EM analysis (Fig. 5C). Notably, these peptidesare in proximity of escape mutations (W168C/R, P171S/H, andD446N) that were raised by growing HCMV VR1814 virus inepithelial cells (ARPE-19) or fibroblasts (MRC-5) in the pres-ence of suboptimal MSL-109 antibody concentrations (37).

DiscussionWe have characterized gH/gL, gH/gL/gO, and Pentamer bio-chemically, functionally, and structurally and studied their in-teraction with the human HCMV-neutralizing antibody MSL-109. EM analysis showed that HCMV gH/gL resembles HSV gH/gL and that gO and the ULs form a protrusion extending fromthe gH/gL N terminus in gH/gL/gO and Pentamer, respectively(Fig. 6). Moreover, SDS/PAGE analysis of the purified com-plexes showed that gO and UL128 form covalent interactionswith gL. Surprisingly, MS analysis revealed that the same cyste-ine, gL–Cys-144, forms disulfide bonds with UL128–Cys-162in Pentamer and gO–Cys-351 in gH/gL/gO. gL–Cys-144 andUL128–Cys-162 are strictly conserved among all HCMV strains,suggesting a functional role for this disulfide. Together, thesedata are consistent with gO and the ULs binding to an over-lapping site at the N terminus of gH/gL.Both EM and MS analyses demonstrate that the gH/gL/gO and

Pentamer complexes are mutually exclusive. This finding explainswhy Pentamer and gH/gL/gO form distinct complexes on the virusand/or cell surface (23). The alternative binding of the ULs andgO to gL–Cys-144 may explain the increase in gH/gL/gO levels onthe virus envelope upon suppression of UL128/UL130/UL131A(24). In addition, our results suggest that HCMV gH/gL mightnot exist as an independent complex on the viral envelope but isassembled either as gH/gL/gO or Pentamer. This hypothesis is inagreement with recent HCMV strain analyses revealing that onlygH/gL/gO and Pentamer are present on the viral envelope (24).Previous studies demonstrated that, depending on the pro-

ducer cells used (i.e., epithelial vs. fibroblasts), the gH/gL/gO:Pentamer ratio on free virions differs, which affects HCMVtransmission and tropism (38). We speculate that HCMV maymodulate this ratio through competition between gO and UL128for gL–Cys-144 binding (Fig. 6). However, the exact factors(cellular, viral, or both) that influence gO vs. UL128 binding anddisulfide formation with gL remain to be determined.gL–Cys-144 mediates homodimerization of recombinant soluble

gH/gL. Formation of these covalent homodimers might be causedby the lack of a biologically relevant binding partner for gL–Cys-144 (i.e., UL128 or gO). It has been recently proposed that gH/gLmay form homodimers when on the cellular membrane and that

MSL-109 binding affects dimers assembly as judged by FRET(37). It is unclear at this stage whether these homodimers are thesame as those described here for soluble gH/gL.Pentamer overexpression specifically inhibits HCMV infection

of epithelial cells, but not fibroblasts, whereas overexpression ofgH/gL/gO interferes with HCMV entry in fibroblasts, but notepithelial/endothelial cells (17, 21). These data point to the ex-istence of cell-specific entry receptors for gH/gL/gO and Pen-tamer, respectively. Our low-resolution 3D structures ofPentamer and gH/gL/gO show that gO or UL128 interactionwith gH/gL does not cause large conformational changes in gH/gL. This finding suggests that gO and the ULs are directly in-volved in receptor binding (i.e., in contrast to new sites beingexposed in gH/gL when the latter is in complex with gO or ULs).Of note, overexpression of the Pentamer gL–Cys-144–Ser/

UL128–Cys-162–Ser double mutant failed to interfere withHCMV entry in ARPE-19 cells, despite normal surface expres-sion levels, whereas the single cysteine mutant behaved similarlyto the WT complex. The same double cysteine mutant was un-able to allow syncytia formation when coexpressed with gB,presumably through the same mechanism (i.e., impaired bindingto a cell surface receptor). Furthermore, the antibody-bindingdata did not reveal large structural changes in the cysteinemutants. These results together suggest that the mutated cys-teines (gL–Cys-144–Ser and UL128–Cys-162–Ser) might be lo-cated near a receptor-binding site, which is affected by theirsimultaneous mutation.A role for the N-terminal region of gH/gL in receptor binding is

consistent with what has been observed in other herpesvirus gH/gLs(39–43). In the alpha herpesvirus PrV, a fusion protein in which theectodomain of the receptor-binding protein gD was fused to theN-terminal region of gH was able to interact with gB for entry,substituting for gD, gH, and gL (44). A similar solution may havebeen adopted in HCMVwith gO and ULs binding to gH/gL to formcomplexes that assume the functions of receptor binding and fusionregulation. Similar to HCMV, it has been shown that in HHV-6A,the gQ1/gQ2 components of the tetrameric gH/gL/gQ1/gQ2 com-plex are directly involved in binding to the entry receptor CD46 (45).Highly potent HCMV-neutralizing monoclonal antibodies that

bind the Pentamer, but not gH/gL, have been isolated from thememory B-cell repertoire of HCMV immune donors (25, 26).These antibodies were capable of potently neutralizing HCMVinfection of epithelial and endothelial cells, but not fibroblasts.Instead, antibodies binding to both gH/gL and Pentamer havebeen shown to neutralize HCMV infection in all cell types.However, in epithelial and endothelial cells, antibodies specificto the ULs are much more potent (25, 26). Our structural datahelp to rationalize these data and to identify the relative location

Fig. 6. gH/gL/gO and Pentamer formmutually exclusive complexes exposing neutralizing sites. gH/gL complex interacts either with gO or with the ULs to form gH/gL/gO and Pentamer, respectively. Two major regions for neutralizing antibody binding can be identified: (i) the gH component (in light blue), which is targeted byantibodies neutralizing infection of both fibroblasts and endothelial /epithelial cells; and (ii) the UL region contains the binding sites for potently neutralizingantibodies of epithelial and endothelial cells infection. The latter region likely includes the receptor binding site for entry in epithelial/endothelial cells.

Ciferri et al. PNAS | February 10, 2015 | vol. 112 | no. 6 | 1771

BIOCH

EMISTR

Y

of two Pentamer regions that are targeted by these differentclasses of neutralizing antibodies. One region, which involves thegH/gL part of the complexes (in light blue in Fig. 6), is struc-turally conserved in gH/gL/gO, gH/gL, and Pentamer and mayplay a role in cell-independent steps of the viral fusion process,such as gB binding. The second region is formed by the ULsubunits and forms an extension at the gH/gL N terminus. Thisregion likely includes the binding site for the entry receptor forepithelial and endothelial cells (Fig. 6).We have recently shown that mice immunized with MF59 adju-

vanted Pentamer protein raise high neutralizing antibody titers (32).These neutralizing responses were comparable to those observed inseropositive subjects and were 4-, 10-, and 90-fold more potent thanthose obtained with adjuvanted gH/gL in mice, rabbits, and cottonrats, respectively. In addition, analysis of immunized mouse serarevealed that the Pentamer vaccine raised antibodies binding toboth gH/gL and to the ULs and that the latter were responsible formost of the neutralization potency of HCMV infection in epithelialand endothelial cells. Therefore, vaccination strategies aimed attargeting the immune response toward the UL-containing region of

the Pentamer may prove valuable in raising strong and protectiveantibody responses.In conclusion, we have demonstrated that UL128 and gO

covalently bind to the same site on gH/gL using the conservedgL–Cys-144 and thus leading to formation of separate complexeson the virus. Both complexes retain the same structural confor-mation in the gH/gL region as confirmed by antibody bindingand EM analysis. Together, these findings will guide the rationaldesign of a new generation of HCMV vaccines with the potentialto elicit potent and clinically relevant protective neutralizingantibody responses.

Materials and MethodsDetailed methods are provided in SI Materials and Methods. These describeconstruct generation and protein expression, purification and characteriza-tion, EM analysis, HDX, MS, and interference and syncytia assays.

ACKNOWLEDGMENTS. The authors would like to thank K. Matsuoka,Y. Aggarwal, A. Nandi, and K. Balabanis for technical assistance andM. Bottomley, E. Malito, and I. Ferlenghi for critical reading of the manuscript.

1. Ho M (1990) Epidemiology of cytomegalovirus infections. Rev Infect Dis 12(Suppl 7):S701–S710.

2. Britt W (2008) Manifestations of human cytomegalovirus infection: Proposed mech-anisms of acute and chronic disease. Curr Top Microbiol Immunol 325:417–470.

3. Arvin AM, Fast P, Myers M, Plotkin S, Rabinovich R; National Vaccine Advisory Com-mittee (2004) Vaccine development to prevent cytomegalovirus disease: Report fromthe National Vaccine Advisory Committee. Clin Infect Dis 39(2):233–239.

4. Sinzger C, Digel M, Jahn G (2008) Cytomegalovirus cell tropism. Curr Top MicrobiolImmunol 325:63–83.

5. Heldwein EE, Krummenacher C (2008) Entry of herpesviruses into mammalian cells.Cell Mol Life Sci 65(11):1653–1668.

6. Connolly SA, Jackson JO, Jardetzky TS, Longnecker R (2011) Fusing structure and function:A structural view of the herpesvirus entry machinery. Nat Rev Microbiol 9(5):369–381.

7. Heldwein EE, et al. (2006) Crystal structure of glycoprotein B from herpes simplex virus1. Science 313(5784):217–220.

8. Backovic M, Longnecker R, Jardetzky TS (2009) Structure of a trimeric variant of theEpstein-Barr virus glycoprotein B. Proc Natl Acad Sci USA 106(8):2880–2885.

9. Backovic M, et al. (2010) Structure of a core fragment of glycoprotein H from pseu-dorabies virus in complex with antibody. Proc Natl Acad Sci USA 107(52):22635–22640.

10. Chowdary TK, et al. (2010) Crystal structure of the conserved herpesvirus fusionregulator complex gH-gL. Nat Struct Mol Biol 17(7):882–888.

11. Matsuura H, Kirschner AN, Longnecker R, Jardetzky TS (2010) Crystal structure of theEpstein-Barr virus (EBV) glycoprotein H/glycoprotein L (gH/gL) complex. Proc NatlAcad Sci USA 107(52):22641–22646.

12. Atanasiu D, SawWT, Cohen GH, Eisenberg RJ (2010) Cascade of events governing cell-cell fusion induced by herpes simplex virus glycoproteins gD, gH/gL, and gB. J Virol84(23):12292–12299.

13. Wang D, Shenk T (2005) Human cytomegalovirus UL131 open reading frame is re-quired for epithelial cell tropism. J Virol 79(16):10330–10338.

14. Ryckman BJ, Jarvis MA, Drummond DD, Nelson JA, Johnson DC (2006) Human cyto-megalovirus entry into epithelial and endothelial cells depends on genes UL128 toUL150 and occurs by endocytosis and low-pH fusion. J Virol 80(2):710–722.

15. Sinzger C, et al. (1999) Modification of human cytomegalovirus tropism throughpropagation in vitro is associated with changes in the viral genome. J Gen Virol 80(Pt11):2867–2877.

16. Hahn G, et al. (2004) Human cytomegalovirus UL131-128 genes are indispensable for virusgrowth in endothelial cells and virus transfer to leukocytes. J Virol 78(18):10023–10033.

17. Ryckman BJ, Chase MC, Johnson DC (2008) HCMV gH/gL/UL128-131 interferes withvirus entry into epithelial cells: Evidence for cell type-specific receptors. Proc Natl AcadSci USA 105(37):14118–14123.

18. Compton T, Nepomuceno RR, Nowlin DM (1992) Human cytomegalovirus penetrateshost cells by pH-independent fusion at the cell surface. Virology 191(1):387–395.

19. Li L, Nelson JA, Britt WJ (1997) Glycoprotein H-related complexes of human cytomegalo-virus: Identification of a third protein in the gCIII complex. J Virol 71(4):3090–3097.

20. Huber MT, Compton T (1998) The human cytomegalovirus UL74 gene encodes thethird component of the glycoprotein H-glycoprotein L-containing envelope complex.J Virol 72(10):8191–8197.

21. Vanarsdall AL, Chase MC, Johnson DC (2011) Human cytomegalovirus glycoproteingO complexes with gH/gL, promoting interference with viral entry into human fi-broblasts but not entry into epithelial cells. J Virol 85(22):11638–11645.

22. Huber MT, Compton T (1997) Characterization of a novel third member of the humancytomegalovirus glycoprotein H-glycoprotein L complex. J Virol 71(7):5391–5398.

23. Wang D, Shenk T (2005) Human cytomegalovirus virion protein complex required forepithelial and endothelial cell tropism. Proc Natl Acad Sci USA 102(50):18153–18158.

24. Zhou M, Yu Q, Wechsler A, Ryckman BJ (2013) Comparative analysis of gO isoformsreveals that strains of human cytomegalovirus differ in the ratio of gH/gL/gO and gH/gL/UL128-131 in the virion envelope. J Virol 87(17):9680–9690.

25. Gerna G, et al. (2008) Prophylaxis followed by preemptive therapy versus preemptivetherapy for prevention of human cytomegalovirus disease in pediatric patients un-dergoing liver transplantation. Transplantation 86(1):163–166.

26. Macagno A, et al. (2010) Isolation of human monoclonal antibodies that potentlyneutralize human cytomegalovirus infection by targeting different epitopes on thegH/gL/UL128-131A complex. J Virol 84(2):1005–1013.

27. Freed DC, et al. (2013) Pentameric complex of viral glycoprotein H is the primarytarget for potent neutralization by a human cytomegalovirus vaccine. Proc Natl AcadSci USA 110(51):E4997–E5005.

28. Fouts AE, Chan P, Stephan JP, Vandlen R, Feierbach B (2012) Antibodies against thegH/gL/UL128/UL130/UL131 complex comprise the majority of the anti-cytomegalovi-rus (anti-CMV) neutralizing antibody response in CMV hyperimmune globulin. J Virol86(13):7444–7447.

29. Genini E, et al. (2011) Serum antibody response to the gH/gL/pUL128-131 five-proteincomplex of human cytomegalovirus (HCMV) in primary and reactivated HCMV in-fections. J Clin Virol 52(2):113–118.

30. Wussow F, et al. (2013) A vaccine based on the rhesus cytomegalovirus UL128 complexinduces broadly neutralizing antibodies in rhesus macaques. J Virol 87(3):1322–1332.

31. Fu TM, et al. (2012) Restoration of viral epithelial tropism improves immunogenicityin rabbits and rhesus macaques for a whole virion vaccine of human cytomegalovirus.Vaccine 30(52):7469–7474.

32. Wen Y, et al. (2014) Human cytomegalovirus gH/gL/UL128/UL130/UL131A complexelicits potently neutralizing antibodies in mice. Vaccine 32(30):3796–3804.

33. Nokta M, Tolpin MD, Nadler PI, Pollard RB (1994) Human monoclonal anti-cyto-megalovirus (CMV) antibody (MSL 109): Enhancement of in vitro foscarnet- andganciclovir-induced inhibition of CMV replication. Antiviral Res 24(1):17–26.

34. Manley K, et al. (2011) Human cytomegalovirus escapes a naturally occurring neutralizingantibody by incorporating it into assembling virions. Cell Host Microbe 10(3):197–209.

35. Loomis RJ, et al. (2013) Vectored co-delivery of human cytomegalovirus gH and gLproteins elicits potent complement-independent neutralizing antibodies. Vaccine31(6):919–926.

36. Kaye J, Browne H, Stoffel M, Minson T (1992) The UL16 gene of human cytomega-lovirus encodes a glycoprotein that is dispensable for growth in vitro. J Virol 66(11):6609–6615.

37. Fouts AE, et al. (2014) Mechanism for neutralizing activity by the anti-CMV gH/gLmonoclonal antibody MSL-109. Proc Natl Acad Sci USA 111(22):8209–8214.

38. Scrivano L, Sinzger C, Nitschko H, Koszinowski UH, Adler B (2011) HCMV spread andcell tropism are determined by distinct virus populations. PLoS Pathog 7(1):e1001256.

39. Omerovi�c J, Lev L, Longnecker R (2005) The amino terminus of Epstein-Barr virusglycoprotein gH is important for fusion with epithelial and B cells. J Virol 79(19):12408–12415.

40. Vleck SE, et al. (2011) Structure-function analysis of varicella-zoster virus glycoproteinH identifies domain-specific roles for fusion and skin tropism. Proc Natl Acad Sci USA108(45):18412–18417.

41. Atanasiu D, et al. (2013) Regulation of herpes simplex virus gB-induced cell-cell fusionby mutant forms of gH/gL in the absence of gD and cellular receptors. MBio 4(2):e00046–e13.

42. Chen J, Jardetzky TS, Longnecker R (2013) The large groove found in the gH/gLstructure is an important functional domain for Epstein-Barr virus fusion. J Virol 87(7):3620–3627.

43. Zhou W, Chen F, Klyachkin Y, Sham YY, Geraghty RJ (2014) Mutations in the aminoterminus of herpes simplex virus type 1 gL can reduce cell-cell fusion without af-fecting gH/gL trafficking. J Virol 88(1):739–744.

44. Klupp BG, Mettenleiter TC (1999) Glycoprotein gL-independent infectivity of pseu-dorabies virus is mediated by a gD-gH fusion protein. J Virol 73(4):3014–3022.

45. Jasirwan C, Furusawa Y, Tang H, Maeki T, Mori Y (2014) Human herpesvirus-6A gQ1and gQ2 are critical for human CD46 usage. Microbiol Immunol 58(1):22–30.

1772 | www.pnas.org/cgi/doi/10.1073/pnas.1424818112 Ciferri et al.