The Location of Asparagine-linked Glycans on West Nile Virions Controls Their Interactions with CD209 (Dendritic Cell-specific ICAM-3 Grabbing Nonintegrin)^*

The C-type lectins CD209 (DC-SIGN)³ and CD209L (DCSIGNR; L-SIGN) have generated considerable interest for their abilities to enhance viral infection in vitro (1, 2). These tetrameric type II integral membrane proteins serve as “attachment factors” that tether glycosylated virions to the cell surface (3). Stable attachment can result in greatly increased infection of the lectin-bearing cell or efficient transfer of bound virions to permissive target cells (4). A broad array of enveloped viruses is able to exploit CD209 and CD209L for increased infection in vitro. These viruses include members of the Retroviridae (human immunodeficiency virus and simian immunodeficiency virus), Filoviridae (Ebola and Marburg), Flaviviridae (Dengue virus, West Nile virus, and hepatitis C virus), Togaviridae (Sindbis), Coronaviridae (SARS coronavirus), and Herpesviridae (human cytomegalovirus) (5-13).

Increased viral infection mediated by CD209 and CD209L requires interactions between these lectins and asparaginelinked viral glycans (1). Biochemical and crystallographic studies have revealed that both lectins bind tightly to the outer branched trimannose structure common to both high mannose glycans and certain hybrid N-linked glycans (14-16). Recognition occurs via an extended binding pocket that makes contacts with multiple mannose groups in this outer branch point structure, explaining the high overall affinity of the interaction (16). Currently, mannose-rich glycans are the only reported ligands for CD209L. CD209 likewise binds tightly to these sugars but has an additional ability to bind certain bacteria and parasites bearing fucose-containing glycans such as the Lewis^x antigen (15, 17).

Little data are available regarding the importance of these well studied biochemical interactions to infections in vivo, largely due to a lack of a suitable animal model for studying these lectins (18-20). However, laboratory and genetic studies have supported an important role for these lectins in Dengue virus (DENV) disease in humans. CD209 is expressed by macrophages and dendritic cells (21-24), which are known cellular targets of DENV infection in vivo (25, 26). In vitro, antibodies to CD209 prevent DENV infection of human monocyte-derived dendritic cells (12, 13). Recently, a polymorphism in the CD209 promoter region associated with altered transcriptional activity in vitro was linked to a decreased risk of severe Dengue fever in three independent cohorts in Thailand (27). CD209L may also be involved in DENV replication, as it is expressed by endothelial cells in the liver and lymph nodes, both of which are known sites of DENV replication in vivo (25, 28).

We recently examined the interactions of CD209 and CD209L with West Nile virus (WNV), a mosquito-borne flavivirus related to DENV (6). We observed that CD209L promoted WNV infection much more efficiently than did CD209, particularly when the virus was grown in mammalian cells. This was unexpected since CD209L is thought to bind only a subset of the carbohydrate ligands of CD209 (15), and because preferential binding of a pathogen to CD209L had not been reported previously. Additional studies using virus-like particles (reporter virus particles (RVPs)) produced by trans-complementation of a subgenomic WNV replicon with the three flavivirus structural proteins (29-33) also showed enhanced infection of CD209L-expressing cells but not CD209-expressing cells. However, RVPs made using the structural proteins of DENV were able to use either CD209 or CD209L for increased cellular infection (6). Thus, when produced in mammalian cells, the structural proteins of DENV, but not those of WNV, have the ability to promote CD209 utilization.

DENV and WNV virions have very similar structures (34, 35). Their surfaces consist of a regular array of 180 envelope (E) protein subunits arranged in an icosahedral lattice (36). The small membrane (M) protein, generated following furin-mediated processing of pre-membrane protein (prM), is also present on the virion surface but is mostly buried in the viral membrane. Although it is not visualized in cryo-electron microscopic (cryo-EM) reconstructions of flavivirus particles, some level of uncleaved prM is generally present in infectious particles (37, 38), where it can also play a role in CD209L binding (6). The major structural differences between DENV and WNV virions stem from the number and location of N-linked glycosylation sites in the DENV viral E proteins (35). Most DENV isolates contain glycosylation sites at residues 67 and 153, although the site at 153 may not always be utilized (39). In contrast, WNV E proteins only contain an N-linked glycan at asparagine 154, although this is absent in many virus strains. The presence of N-glycosylation on the WNV E protein has been linked in some studies to increased neuroinvasiveness in mice (40-44) and to altered cellular tropism in vitro (45).

We hypothesized that the glycosylation site at residue 67 of the DENV E protein might be responsible for the ability of this virus to interact with CD209. Indeed, a recent cryo-EM study demonstrated that the isolated carbohydrate recognition domain (CRD) of CD209 binds preferentially to a subset of glycans present at position 67 on DENV particles (46). We therefore constructed a series of mutant WNV E proteins containing N-linked glycosylation sites at position 67 or at several novel positions. We studied the abilities of RVPs made with these glycosylation mutants to infect CD209- or CD209L-expressing cells. We also examined the contributions of specific N-linked carbohydrates to interactions with these two lectins by treating RVPs with specific glycosidases or by manipulating carbohydrate processing in the RVP producer cells. We show that certain glycosylation locations, including position 67, allow WNV to interact with CD209. These interactions are strictly dependent on the incorporation of “mannose-rich” sugars (Man_5-9 high mannose or Man₅ hybrid glycans). We also show that, surprisingly, CD209L can recognize flaviviruses bearing either complex or mannose-rich glycans. Binding of CD209L to complex glycans appears to entail the recognition of particular terminal sugar moieties, particularly N-acetylglucosamine.

EXPERIMENTAL PROCEDURES

Cells and Reagents—K562 control, K562-CD209, K562-CD209L, Vero, and BHK WNIIrep-REN cells were cultured as described previously (6). Note that we previously referred to K562-CD209 and K562-CD209L cells as K562-SIGN and K562-SIGNR cells, respectively. Lipofectamine 2000 was obtained from Invitrogen, and Renilla luciferase substrate was purchased from Promega. Polyclonal rabbit antibodies against prM were generated by ProSci (San Diego) by immunization of rabbits with peptides spanning the following three regions in prM: the N-terminal “pr” region, the furin cleavage junction, and the C-terminal “M” region. Antibodies were affinity-purified over peptide columns. Pooled antibodies from immunized rabbits were used for prM/M staining. Deoxymannojirimycin was purchased from Calbiochem. Endoglycosidase H_f (endo H) and protein N-glycosidase F (PNGase F) were purchased from New England Biolabs. A polyclonal rabbit serum against the WNV E protein was generated by immunization of rabbits with two recombinant His₆-tagged thioredoxin fusion proteins containing residues 146-249 and 34-137 of the NY99 WNV E protein. β-N-Acetylhexosaminidase (HEXase I) and β(1-4)-galactosidase were purchased from Prozyme. These exoglycosidase enzymes were originally cloned from Streptococcus pneumoniae, whose glycosidase enzymes are active at neutral pH values (47, 48). It is also worth noting that, unlike other commercially available β-N-acetylhexosaminidase enzymes that act upon both GlcNAc and N-acetylgalactosamine (49), HEXase I removes only GlcNAc residues (50).

Plasmids—We used the following plasmids: furin expression plasmid pcDNA3.1furin (51); DENV serotype 1 (Western Pacific strain) prM-E expression vector pDV1 prM-E VAX (provided by Drs. Wellington Sun and Robert Putnak); DENV serotype I capsid expression vector pDV1Cap (6); and WNV lineage II capsid expression vector pWNIIcap (29). The WNV strain NY99 prM-E expression plasmid pCBWN (52) and a derivative of this plasmid lacking the N-linked glycosylation site at E protein residue 154 (NY99-N154Q) (45) were used as templates for the introduction of novel N-linked glycosylation sites into the WNV E protein by site-directed mutagenesis. The following amino acid changes were introduced into NY99-N154Q: (i) Ala-54 to Thr (A54T) adds an N-linked glycosylation site at Asn-52; (ii) D67N adds a site at Asn-67; (iii) K84T adds a site at Asn-82; (iv) A173N and P174G (AP173NG) add a site at Asn-173; (v) Glu-182 to NGS (E182NGS) adds a site at Asn-182 by mutating Glu-182 to Asn and inserting two amino acids (Gly-Ser) to complete the sequon; (vi) S230N and V232T (STV230NTT) add a site at Asn-230; (vii) V279T adds a site at Asn-277; (viii) T301N and G303S (TYG301NYS) add a site at Asn-301; (ix) T330N adds a site at Asn-330; (x) K370T adds a site at Asn-368; (xi) G389N and Q391T (GEQ389NET) add a site at Asn-389. Mutants A54T, D67N, K84T, AP173NG, V279T, T330N, and K370T were generated by QuikChange mutagenesis. For mutants E182NGS, STV230NTT, TYG301NYS, and GEQ389NET, the entire mutant prM-E coding sequence was amplified by overlap-extension PCR and TOPO cloned into pcDNA3.1V5/HIS (Invitrogen).

Reporter Virus Particles—3 × 10⁶ WNV BHK WNIIrep-REN cells (containing a subgenomic WNV replicon expressing the Renilla luciferase gene) were seeded into wells of a 6-well plate. The following day, the cells were transfected using Lipofectamine 2000 with 2.6 μg of pWNIIcap, 1 μg of pcDNA3.1furin, and 0.4 μg of prM-E plasmid to produce RVPs. This ratio of capsid, furin, and prM-E vectors was found to allow full conversion of prM to M and to give the highest RVP titers for all WNV prM-E mutants tested in this study (data not shown). After overnight transfection, medium was replaced with 2 ml of RVP production medium (90% Dulbecco's modified Eagle's medium, low glucose, 10% fetal calf serum, 25 mm HEPES, pH 7.5), and RVP-containing supernatants were harvested and filtered with a 0.20-μm filter 48 h post-transfection. Cells, plasmids, and Lipofectamine were scaled up proportionally for larger RVP preparations. For RVPs made without furin, pcDNA3.1furin was replaced with empty pcDNA3 vector.

Western Blot—1 ml of RVP-containing supernatant was pelleted through 20% sucrose by centrifugation at 28,000 rpm for 4 h in an SW 28 rotor. Virus particles were subjected to SDS-PAGE on a 10-20% Tris-HCl gradient gel, transferred to polyvinylidene difluoride, and analyzed by Western blot with rabbit anti-prM/M polyclonal antibodies or rabbit anti-E polyclonal serum at a dilution of 1:3000.

Deglycosylation of Wild-type RVPs—Freshly harvested, undiluted RVP-containing supernatants were incubated overnight at 37 °C with various glycosidases and used for infections the following day. 100 μl of RVP stock was incubated with 1 μl of water (mock treatment), 1 μl of endo H (1000 New England Biolabs (NEB) units), 1 μl of PNGase F (500 NEB units), 2 μl of HEXase I (80 Glyko milliunits), 2 μl of sialidase A (10 Glyko milliunits), or 2 μl of β(1-4)-galactosidase (4 Glyko milliunits). For PNGase F protection studies, 100 μl of RVPs were incubated for 3 h at 37°C with 1 μl of endo H or 1 μl of water, followed by addition of 1 μl of PNGase F or water, and incubation overnight at 37 °C.

Deglycosylation of RVPs Containing Glycans at Novel Locations—RVPs glycosylated at position 67 had high titers and survived overnight treatment at 37 °C comparably with wildtype RVPs. Therefore, PNGase F and endo H treatment of these particles was done overnight as for wild-type particles. RVPs glycosylated at positions 182, 330, and 370 showed a dramatic loss in overall titer when incubated overnight at 37 °C in the absence of enzyme. Therefore, for these RVPs, digestion was shortened to 4 h. Time course experiments demonstrated that 4 h was sufficient for endo H treatment to exert maximal effects on RVP infectivity on K562-CD209 and K562-CD209L cells (data not shown). However, 4 h was not sufficient for maximal PNGase F activity or exoglycosidase activity.

Infections—5 × 10⁴ K562 control, -CD209, or -CD209L cells were seeded in 96-well plates and infected with RVPs diluted in K562 growth medium in a total volume of 200 μl. Cells were incubated in the presence of RVPs for 48 h followed by lysis and enumeration of luciferase activity as described previously (6). For antibody inhibition studies, cells were preincubated for 1 h at 37 °C with 10 μg/ml mAb 120526 (for K562-CD209 cells), 10 μg/ml mAb 120604 (for K562-CD209L cells), or 10 μg/ml control mouse IgG antibody (clone DC72) before addition of RVPs diluted in medium containing the same concentration of mAb. We have demonstrated previously that mAbs 120526 and 120604 are highly effective inhibitors of WNV infection via CD209 and CD209L, respectively (6).

RESULTS

Inclusion of a Furin Expression Plasmid during Transfection Results in RVPs Devoid of prM—In this study, we wished to focus solely on the effect of E protein glycans on the interactions of flaviviruses with CD209 and CD209L. Immature flavivirus particles contain two envelope glycoproteins, prM and E. During particle egress, prM is cleaved by the cellular enzyme furin, releasing the glycosylated, N-terminal portion of prM from the particle and leaving behind the small, nonglycosylated membrane protein M (36). However, furin-mediated cleavage of prM does not always go to completion for WNV, and thus virions may contain both E and prM glycans (38). We have demonstrated previously that these prM glycans may mediate interactions with CD209L, albeit less efficiently than E protein glycans (6).

To avoid prM-mediated effects in this study, we modified our normal procedure for producing WNV RVPs. When WNV replicon-containing BHK cells were transfected with plasmids encoding the flavivirus structural proteins, an expression plasmid for human furin was also included. Using an optimized ratio of the individual plasmids, this procedure resulted in the production of RVPs devoid of prM as assessed by Western blot (Fig. 1A).

WNV Pseudotypes Produced in the Presence of Excess Furin Interact with CD209L but Not CD209, whereas DENV Pseudotypes Interact with Both Lectins—Using this protocol, we produced RVP stocks with the structural proteins of DENV or WNV and used these stocks to infect K562 cell lines expressing CD209, CD209L, or a control message. As we observed previously (6), RVPs pseudotyped with the DENV structural proteins infected both K562-CD209 and K562-CD209L cells much more efficiently than control cells (Fig. 1B), whereas WNV-pseudotyped RVPs showed a strong preference for infecting K562-CD209L cells over K562-CD209 or control cells (Fig. 1C). RVPs made with a “glycosylation knock-out” WNV E protein lacking the single N-linked glycosylation site at residue 154 infected all three K562 lines with equally low efficiency (Fig. 1D), confirming that RVPs made using our optimized transfection procedure require E protein glycosylation to interact with CD209 or CD209L.

Addition of an N-Linked Glycosylation Site to Position 67 of the WNV E Protein Allows CD209 Utilization—We next investigated whether the N-linked glycosylation site at position 67 in the DENV E protein could explain why DENV-pseudotyped RVPs were able to infect CD209-expressing cells efficiently, whereas WNV-pseudotyped RVPs were not. We added an N-linked site at position 67 to both the wild-type WNV E protein and the glycosylation knock-out E protein. This resulted in mutant proteins containing N-linked glycans at both residues 67 and 154 or solely at residue 67. Pseudotyped RVPs made with either of these mutants infected both K562-CD209 and K562-CD209L cells more efficiently than controls (Fig. 1, E and F). This suggests that glycosylation at position 67 is sufficient to allow a WNV virion to interact with either CD209 or CD209L.

Infection of K562-CD209 cells was somewhat more efficient for RVPs containing glycans at both position 67 and position 154 than for RVPs glycosylated only at position 67. This difference suggests that glycosylation at residue 154 may increase the efficiency of CD209 usage when the glycan at position 67 is also present.

Removal of Mannose-rich Glycans from RVPs by Endoglycosidase H Blocks Their Interactions with CD209 but Not CD209L—CD209 and CD209L are both known to bind strongly to high mannose N-linked glycans on viruses (1, 10). We have shown previously that wild-type WNV particles produced in the presence of DMJ infect K562-CD209 and K562-CD209L cells with similar efficiencies (6). DMJ, an inhibitor of Golgi mannosidase I, increases the incorporation of high mannose glycans into proteins by inhibiting glycan maturation beyond the Man₈GlcNAc₂ stage (53). We hypothesized that RVPs glycosylated at position 67 might contain increased levels of mannose-rich (high mannose or Man₅ hybrid) glycans, which would explain their ability to interact with CD209. To test this, we treated RVPs with endo H, which specifically releases high mannose and Man₅ hybrid N-linked glycans by cutting between the two GlcNAc residues in the N-acetylchitobiose (GlcNAc₂) core (54). We also treated RVPs with PNGase F, which removes virtually all N-linked glycans by cutting the asparagine-glycan linkage (54). After overnight digestion at 37 °C with these enzymes or a water control, serial dilutions of the RVPs were used to infect K562 control, K562-CD209, or K562-CD209L cells. The results of a representative digestion/infection experiment are shown in Fig. 2.

As expected, treatment of wildtype RVPs (containing glycans at position 154) with PNGase F or endo H had essentially no effect on their (already low) ability to infect K562-CD209 cells (Fig. 2A). However, removal of glycans from wildtype RVPs with PNGase F caused a drastic decrease in their ability to infect K562-CD209L cells (Fig. 2B); the residual level of infection was similar to that seen in K562 control cells infected with the digested or mock-digested RVPs (data not shown). In contrast, endo H treatment of these RVPs had essentially no effect on infection of CD209L-bearing cells, suggesting that mannose-rich glycans were not present and/or not required for the interactions of these particles with CD209L.

To ensure that endo H was active under our native digestion conditions, we treated RVP producer cells with 1 mm DMJ to obtain wild-type RVPs containing high mannose glycans. As seen previously (6), DMJ-treated RVPs infected K562-CD209 cells much more efficiently than nontreated RVPs (compare Fig. 2C and Fig. 2A) but showed no change in their infectivity for K562 control cells (data not shown). Endo H treatment abrogated this increased infection (Fig. 2C), demonstrating that high mannose glycans at position 154 are accessible to endo H and required for infection via CD209. Surprisingly, endo H treatment caused only a modest (2-fold) decrease in the ability of DMJ-treated RVPs to infect K562-CD209L cells (Fig. 2D), a finding which we later investigated in more detail (see below). As expected, full removal of all glycans by PNGase F treatment substantially reduced infection of both cell lines.

We next treated mutant RVPs containing N-linked glycans solely at position 67 with endo H and PNGase F. As seen for DMJ-treated RVPs, endo H treatment had only a modest (2-fold) effect on the infection of CD209L cells (Fig. 2F). PNGase F treatment did not alter the infectivity of these RVPs for either cell type. Under native conditions, this likely indicates that PNGase F was unable to access the connection between the N-glycan and asparagine 67, whereas endo H was still able to access the N-acetylchitobiose core, which lies farther out from the peptide backbone. This explanation is consistent with the known sensitivity of PNGase F to protein conformation (55). Endo H treatment largely eliminated the ability of these particles to infect K562-CD209 cells (Fig. 2E), indicating that glycosylation at position 67 leads to the incorporation of high mannose or endo H-sensitive hybrid glycans that are required for interactions with CD209. Finally, RVPs containing glycans at both positions 67 and 154 were digested with PNGase F and endo H and used for infection of the K562-CD209 (Fig. 2G) and K562-CD209L cells (Fig. 2H). Both enzymes caused a modest (3-6-fold) loss of infectivity on K562-CD209 cells and had little effect (2-fold or less) on infection of K562-CD209L cells. We were unable to determine whether the addition of a second glycan at position 154 reduced the ability of endo H to remove carbohydrates from these particles (e.g. because of steric interference by the added glycan) or whether endo H-resistant structures were able to interact with CD209 when sites were present at both positions 67 and 154.

Levels of High Mannose Glycans below the Range of Detection of Gel-shift Assays Are Sufficient to Enable Interactions with CD209—A surprising finding of our digestion experiments was that DMJ-treated RVPs were still capable of preferential infection of K562-CD209L cells after treatment with endo H. The first possibility we considered was that 1 mm DMJ did not fully inhibit mannosidase trimming in the RVP producer cells, so that complex glycans were still incorporated. These complex glycans might be responsible for CD209L binding after treatment of the particles with endo H. Therefore, we increased the concentration of DMJ used to 5 mm and repeated the digestion/infection experiments. However, even RVPs made with 5 mm DMJ still efficiently infected K562-CD209L cells after endo H treatment (Fig. 3A).

To determine whether N-linked glycans were being fully removed by endo H treatment, we digested RVPs produced in the presence of 1 mm DMJ, 5 mm DMJ, or no DMJ, using the same native conditions used for our digestion/infection experiments but in larger scale. These RVPs were then pelleted and analyzed by Western blot for E protein (Fig. 3B). Treatment with PNGase F resulted in a slight, but noticeable increase in E protein mobility for all three RVP stocks. This small shift is expected because of the fact that the 55-kDa WNV E protein contains only a single N-glycan (average size ∼2.5 kDa (56)). RVPs made without DMJ showed no clear increase in E protein mobility following endo H treatment, suggesting that most of the N-linked carbohydrates in these particles were not mannose-rich. RVPs made with 1 mm DMJ likewise showed little mobility shift following endo H treatment, suggesting that the majority of the glycans in these RVPs is of the complex type. Evidently, however, sufficient high mannose glycans were incorporated into these particles to allow interactions with CD209 (Fig. 2).

The Single GlcNAc Residue Left by Endo H Digestion Is Able to Mediate Interactions with CD209L—Particles produced in the presence of 5 mm DMJ showed an increase in mobility when treated with endo H, similar to what was seen for PNGase F treatment (Fig. 3B). Thus, endo H removed essentially all glycans from these particles. Surprisingly, however, even after digestion these RVP stocks still showed a substantial capacity for CD209L-mediated infection (Fig. 3A).

We envisioned two possible explanations for the continued ability of these RVPs to utilize CD209 for infection even after removal of essentially all N-glycans by endo H. First, trace numbers of endo H-resistant N-glycans, sufficient for mediating interactions with CD209L but insufficient for visualization by Western blot, may have been retained on the particles after endo H treatment. Second, CD209L interactions may have been mediated by the single GlcNAc left attached to asparagine 154 after endo H digestion within the GlcNAc₂ core of the high mannose glycans.

To test the latter possibility, we took advantage of a previously published study that investigated the effect of prior cleavage of a high mannose glycan on the subsequent accessibility of the remaining GlcNAc to PNGase F (57). That study found that, under native conditions, PNGase F was able to fully remove the high mannose glycans of invertase and RNase B. However, if these glycans were first trimmed to a single GlcNAc by endo H, PNGase F was functionally unable to remove these residual sugars, because of impaired enzyme kinetics for this substrate.

We hypothesized that single GlcNAc residues attached to the WNV E protein might be protected from PNGase F activity as well. We therefore performed a sequential digestion experiment in which DMJ-treated or -untreated RVP stocks were first incubated with endo H or mock-digested and then incubated with PNGase F. We found that when DMJ-treated RVPs were first digested with endo H, PNGase F treatment no longer blocked their abilities to infect K562-CD209L cells (Fig. 3C). This protective effect was not seen for RVPs produced in the absence of DMJ, which would have fewer, if any, mannose-rich glycans and thus would not provide a substrate for endo H to act upon.

GlcNAc-terminated Glycans Play an Important Role in the Interactions of Wild-type WNV RVPs with CD209L—This protection experiment suggested that, at least under certain conditions, a single asparagine-linked GlcNAc residue could play a vital role in the interactions of WNV with CD209L. Given the surprising efficiency of these interactions, we then asked whether, under normal conditions, GlcNAc residues at the nonreducing termini of complex glycans might allow wild-type WNV RVPs to interact with CD209L. We therefore digested RVPs under native conditions with various exoglycosidases and infected K562-CD209L cells or Vero cells with these digested stocks (Fig. 3D).

Treatment of wild-type RVPs with HEXase I, which removes β-linked GlcNAc from the nonreducing termini of glycans, caused a reproducible >2-fold decrease in their ability to infect K562-CD209L cells, suggesting at least a partial role for GlcNAc in WNV-CD209L interactions. β(1,4)-Galactosidase treatment moderately increased the infectivity of wild-type RVPs for K562-CD209L cells (about 2-fold). Removal of β(1,4)-linked galactose from a nonreducing terminal lactosamine moiety would leave behind GlcNAc at the new nonreducing terminus of the N-glycan. Thus, the increased infectivity of β(1,4)-galactosidase-treated RVPs for CD209L-expressing cells may reflect the increased exposure of terminal GlcNAc residues on these particles. Treatment with sialidase A, which cuts all α-sialic acid linkages, had no effect on CD209L-mediated infection (data not shown). However, this lack of effect must be interpreted with caution as we do not have positive data demonstrating that this enzyme was active under our native digestion conditions. Neither HEXase I nor β(1,4)-galactosidase treatment altered the infectivity of DMJ-treated RVPs for K562-CD209L cells. This was expected, because these particles contain high mannose glycans, which are able to mediate binding to CD209L but are not cut by these enzymes. Furthermore, none of the exoglycosidases dramatically altered the ability of wild-type RVPs to infect Vero cells. Thus, although glycosidase treatment alters the interactions of RVPs with lectins such as CD209L, it does not cause global changes in particle infectivity.

Different N-Linked Glycosylation Locations on the WNV E Protein Have Varying Abilities to Promote RVP Interactions with CD209 but All Sites Allow Interactions with CD209L—Having established that glycosylation at position 67 allowed WNV to interact with CD209 and that glycosylation at either position 154 or 67 allowed interactions with CD209L, we wished to determine whether these sites were unusual in their abilities to promote these interactions. We therefore added N-linked glycosylation sites at 10 other locations in the WNV E protein. The crystal structure of the DENV E protein (58) was used as a model to select 10 locations predicted to lie in surface exposed loops on the protein (Fig. 4). All 10 sites were added to the glycosylation knock-out WNV E protein so that they would represent the sole N-linked sites in E. Based on our preliminary results with these 10 mutants, we selected four sites (positions 52, 230, 277, and 370) to be added to the wild-type WNV E protein containing the native N-linked site at position 154.

These 14 constructs were used to generate pseudotyped RVPs. Mutants that contained single glycans at positions 82, 173, and 301 failed to produce infectious RVPs (data not shown) and were not examined further. K562 control, -CD209, and -CD209L cells were infected with serial dilutions of RVPs made using the 11 remaining constructs. A representative titration experiment is shown in Fig. 5.

All RVP stocks made with constructs containing at least one N-linked site infected K562-CD209L cells much more efficiently than control K562 cells. Thus, although WNV RVPs must be glycosylated in order to interact with CD209L, there is no apparent requirement for glycosylation at a specific position in the E protein. In contrast, RVP stocks made with the different glycan add-in constructs displayed widely different abilities to infect K562-CD209 cells. RVPs containing glycosylation sites at positions 52, 230, or 277 showed little to no enhancement of infection on K562-CD209 cells. However, when these sites were combined with the native site at position 154, they showed an increased ability to interact with CD209. RVPs glycosylated at position 389 also showed no ability to interact with CD209. However, the overall titers of these RVPs were so low that we cannot rule out the possibility of a moderate increase in infection on K562-CD209 cells relative to controls that was still insufficient to raise the overall infection level above the limit of detection. RVPs glycosylated at positions 182 or 330 showed moderately enhanced infection of K562-CD209 cells, whereas RVPs glycosylated at position 368 (or both 368 and 154) showed strongly increased infection on these cells. For all RVP stocks (including those glycosylated at position 67), the enhanced infection observed on CD209L- and CD209-expressing cells was lost if cells were preincubated with blocking antibodies to these lectins (data not shown).

Regardless of the Location of N-Linked Glycans on RVPs, Endo H-sensitive Structures Are Required for Interactions with CD209 but not CD209L—Our studies identified three novel glycosylation sites on the E protein that individually could facilitate the interaction of RVPs with CD209: positions 182, 330, and 368. When RVPs glycosylated at these positions were digested with endo H, they showed a dramatic decrease in their ability to infect CD209-expressing cells but largely retained their abilities to interact with CD209L (Fig. 6A). Thus, mannose-rich glycans were required for the interaction of these RVPs with CD209, similar to what was seen for RVPs glycosylated at position 67 or DMJ-treated wildtype RVPs (Fig. 2). We also glycosidase-treated the RVPs containing more than one glycosylation site per E protein (i.e. one at position 154 and one at a novel site). However, neither PNGase F nor endo H significantly affected their abilities to infect K562-CD209 or K562-CD209L cells. This is similar to results obtained when we digested RVPs glycosylated at both positions 67 and 154 (Fig. 2). This may reflect a generalized resistance of these RVPs to glycosidase digestion under native conditions.

Regardless of Glycan Location, Forced Incorporation of High Mannose Structures into RVPs Allows Them to Interact with CD209—We had already determined that wild-type particles produced in the presence of DMJ and thus bearing high mannose glycans at position 154 could interact with CD209. To determine whether high mannose carbohydrate structures could support interactions with CD209 regardless of their location on the E protein, we produced our panel of mutant RVPs in the presence of DMJ and infected K562-CD209 and K562-CD209L cells with these DMJ-treated stocks. RVPs made with most of the glycosylation mutant E proteins showed an approximately equal ability to infect K562-CD209 and K562-CD209L cells. A representative infection using RVPs glycosylated at E protein residue 230 is shown in Fig. 6B. Similar results were seen for DMJ-treated RVPs glycosylated at positions 52, 182, 277, 330, or 368, and for DMJ-treated RVPs made with the dual glycan constructs (not shown). Thus, as a general rule, when glycan processing was blocked at the high mannose stage, it resulted in RVPs that were able to interact efficiently with either CD209 or CD209L, regardless of the location of the high mannose glycans on the particle. An exception to this rule was that particles glycosylated at position 389 still preferentially infected K562-CD209L cells compared with K562-CD209 cells (Fig. 6C). RVPs produced in the presence of DMJ but lacking any glycosylation sites on E infected K562 control, -CD209, and -CD209L cell lines with similar efficiencies (Fig. 6D). This demonstrates that the effects of DMJ require the incorporation of high mannose N-linked glycans into viral structural proteins.

DMJ treatment blocks the processing of high mannose glycans at Man_8-9 stage (53). As Man_8-9 high mannose glycans are bound more efficiently by CD209 and CD209L than shorter Man₅ high mannose glycans (15), we wished to determine whether DMJ treatment was enhancing CD209 interactions by skewing the length of the high mannose glycans present as well as their number. To test whether glycan processing had to be stopped at the Man_8-9 stage to allow all glycosylation sites to promote CD209 utilization, or whether any type of high mannose glycan would suffice for this, we also produced RVPs in CHO-lec1 cells (59). These cells fail to express GlcNAc transferase I, and as a result, glycoproteins derived from these cells lack complex or hybrid glycans. The processing defect in these cells causes an accumulation of Man₅ high mannose glycans rather than the Man_8-9 forms made in the presence of DMJ. All glycosylated RVPs grown in these cells that were released at detectable titers (all mutants except 182, 330, and 389) were able to utilize CD209 for infection (data not shown) suggesting that high mannose glycosylation, rather than Man_8-9 glycans in particular, is all that is required for CD209 utilization.

DISCUSSION

In this study, we used infection of the CD209- or CD209L-expressing cells as a surrogate for flavivirus binding. Studies such as this one complement traditional biochemical approaches by focusing on those interactions between glycans and lectins that lead to a biologically interesting outcome, infection. The ligand specificities of CD209 and CD209L have been examined in detail previously. These studies have used traditional biochemical techniques (14) as well as newer technologies such as glycan array profiling (15). Such studies are able to identify the glycans that bind these lectins with the highest affinity constants; however, such studies do not identify which interactions will have important biological outcomes, nor can they recapitulate the polyvalent interactions that occur between a viral particle and lectin receptors at the cell surface. Thus, glycans that bind with low monovalent affinity constants but contribute to strong polyvalent interactions may be missed.

Pure biochemical studies that characterize the glycans present on pathogens that interact with CD209 and CD209L have their own set of limitations. The glycan species that mediates attachment to these lectins may be rare and may therefore be missed. We have shown in this study that virion preparations produced in the presence of 1 mm DMJ are able to interact with CD209 via endo H-sensitive glycans. Despite this, we failed to detect these endo H-sensitive glycans by gel-shift assay, suggesting that they represent only a small proportion of the total glycans present on these particles. Infection-based assays can identify the glycans relevant to infection, regardless of whether these glycans represent the majority of the glycans present in a viral preparation or only a small fraction. Rare glycan species may be particularly important in the case of WNV, which contains 180 surface glycans per virion. Although the stoichiometry of the interaction between WNV and CD209 and CD209L is not yet known, it is conceivable that only a small fraction of the 180 glycans needs to be able to interact with these lectins in order to mediate attachment.

An unexpected finding in this study was the identification of GlcNAc-terminated complex glycans as important carbohydrate ligands for the binding of WNV to CD209L. To date, comparisons of the ligand specificities of CD209 and CD209L have demonstrated their shared abilities to bind mannose-rich (high mannose or Man₅ hybrid) glycans (16, 17) and the additional ability of CD209, but not CD209L, to bind certain fucose-containing carbohydrate structures (15, 17). To our knowledge, no high affinity carbohydrate ligands for CD209L other than high mannose or hybrid glycans have been identified by traditional biochemical approaches. Here we show that CD209L can interact efficiently with a flavivirus that lacks these mannose-rich, endo H-sensitive structures. Our PNGase F protection experiments suggest that glycans consisting of a single GlcNAc attached to asparagine 154 allow WNV to interact with CD209L. Although single GlcNAc residues are not normally attached to asparagine, this result suggests that the nonreducing terminal GlcNAcs on the intact, complex glycans normally found on WNV should also be able to mediate infection via CD209L. In agreement, treatment of RVPs with β(1-4)-galactosidase, which exposes GlcNAc by removing galactose from terminal lactosamine groups, increased CD209L-mediated infection. In contrast, decreasing the number of terminal Glc-NAcs by treatment with HEXase I reduced this infection. As HEXase I did not fully prevent WNV from interacting with CD209L, it may indicate that other terminal sugars besides Glc-NAc are capable of mediating interactions with CD209L. For example, removal of a GlcNAc linked to the trimannosyl core of an N-glycan would leave mannose at the new nonreducing terminus. These terminal mannose residues should also bind to CD209L in much the same way as GlcNAc, as they possess the equatorial 3′- and 4′-OH groups required for binding to “mannose-specific” C-type lectins (60).

The involvement of GlcNAc in WNV binding suggests a mechanism for the observation that WNV grown in mammalian cells binds to CD209L but fails to bind to CD209 (6). Certain monosaccharides, particularly GlcNAc, more efficiently block the interactions of neoglycoproteins with CD209L than with CD209 (14), suggesting that CD209L may bind these monosaccharides more efficiently than CD209 does. By extension, CD209L might bind the terminal GlcNAc residues on complex glycans more efficiently as well. As the interaction of these monosaccharides with CD209L is quite weak (with dissociation constants in the millimolar range), it is likely that multivalent interactions between WNV and CD209L must occur to account for the high overall strength of the interaction. These could involve either the binding of a WNV particle to multiple subunits within a CD209L tetramer or interactions between a single WNV particle and multiple tetramers at the cell surface.

Given that glycans present at nine separate positions on the E protein are able to interact with CD209L, it may seem paradoxical to suggest multivalent binding of a particle to a single CD209L tetramer. However, it should be noted that the orientations of individual carbohydrate recognition domains within a CD209L tetramer are somewhat flexible (61). Thus, a CD209L tetramer may be able to accommodate changes in inter-glycan spacing. Next, no matter where an N-glycan is added to the E protein, certain inter-glycan spacings will remain constant on the virus particle. The icosahedral E protein shell surrounding a WNV virion consists of 30 copies of a six-subunit E protein “raft” structure (36). Within each raft, E is organized into three roughly parallel dimers. Because of this parallel orientation, regardless of the location of an N-glycan on the E protein, there will be another N-glycan located ∼60 Å away on an adjacent E protein dimer. This 60-Å spacing matches well with the distance between calcium-dependent binding sites across the 2-fold symmetry axis of the CD209L tetramer (61).

The results of this study also explain our previous observation that DENV and WNV have differing abilities to interact with CD209 (6). We demonstrate that adding an N-linked glycosylation site to WNV at the same position (residue 67) as a unique N-linked site in DENV allows the glycosylation mutant WNV particles to infect CD209-expressing cells with efficiencies similar to those of DENV. Our data complement a recent study by Pokidysheva et al. (46) who found that CD209 CRD monomers bound to N-linked glycans at position 67 on DENV but not at position 153. Together, the data strongly suggest a key role for the DENV glycan at position 67 in CD209-binding and infection of CD209-expressing cells.

We also identified several other locations besides residue 67 where the addition of an N-linked glycan allowed CD209-mediated infection. However, the majority of the glycosylation sites introduced in this study (including positions 52, 182, 330, 368, and 389) was associated with a 10-fold or greater reduction in infectivity when RVP titers were measured on nonlectinbearing cells, including K562 control cells, Vero, 293T, and HeLa cells (data not shown). These mutant RVP stocks contained lower concentrations of E protein and fewer copies of the WNV genomic RNA compared with wild-type RVPs (data not shown), suggesting that a defect at the level of protein expression, folding, or particle assembly was responsible for their lower titers. RVPs containing no glycans or glycans at positions 67, 230, or 277 were produced at titers comparable with wild-type particles. Notably, position 67 was the only glycosylation site that allowed CD209-mediated infection without compromising the overall RVP titer. We hypothesize that the presence of this conserved site in DENV might reflect an important role for CD209 in the natural history of DENV infection. Glycosylation at this location may have been selected in DENV evolution because of its unique ability to allow the virus to infect CD209-bearing macrophages, while maintaining its infectivity for other cell types.

Why do certain glycosylation sites on the E protein, but not others, facilitate WNV interactions with CD209? We can envision the following two possible explanations: either certain sites are more predisposed toward high mannose or hybrid glycosylation patterns than others, or the spatial location of particular sites facilitates their interactions with CD209. The evidence presented in this study is most consistent with the first hypothesis. RVPs that normally exhibited enhanced infection of K562-CD209 cells lost this ability when their mannose-rich glycans were removed with endo H. Nearly any single glycosylation site on E could promote interactions with CD209 if the RVPs were produced in the presence of DMJ, which forces the incorporation of high mannose glycans at all sites. One possible explanation for the selective ability of particular N-linked sites to promote interactions with CD209 is that high mannose and/or hybrid glycans are efficiently incorporated into these sites even when RVPs are grown in the absence of DMJ.

Several factors influence which types of sugars are incorporated onto the surface of a virus. The cell type infected plays an important role, as different cells express different glycan-processing enzymes and produce different levels of the precursors needed for particular glycan modifications (62). Indeed, we have shown previously that the cell type in which WNV is grown affects its incorporation of high mannose sugars (6), with mosquito cell-derived virus showing a preferential incorporation of high mannose structures at position 154. This is similar to what has been shown previously for alphaviruses (10, 62) and is because of a defect in complex glycan synthesis in mosquito cells. Here we suggest that in a mammalian cell type that normally adds complex glycans to WNV E, variation in the location of the glycan will control the carbohydrate structure present on the mature virion. Selective incorporation of mannose-rich glycans into particular N-linked sites has also been documented previously for the envelope proteins of alphaviruses (62). All N-linked glycans are synthesized as high mannose forms that are subsequently converted to complex glycans by carbohydrate-processing enzymes present in the Golgi apparatus. It has been suggested that high mannose glycans might be retained at particular sites that are poorly accessible to Golgi mannosidases (62). We hypothesize that the glycosylation sites that promote RVP interactions with CD209 are located on portions of the particle that are poorly accessible to Golgi carbohydrate-modifying enzymes during particle egress. This idea is consistent with what is known about the structure of the immature flavivirus particle (63).

In the immature particle, individual E proteins exist as heterodimers with prM. Three prM-E heterodimers associate to form trimeric spikes that project away from the surface of the virus. The distal end of each protruding E protein segment is thought to interact with and be covered by prM. If the four glycosylation sites that promoted CD209 utilization when added to the WNV E protein are mapped onto the structure of the immature DENV particle, three of the sites (residues 182, 330, and 368) fall in the region of E that lies close to the viral membrane and is overshadowed by the projecting E-prM spikes. The fourth site (residue 67) lies in a portion of domain II thought to be covered by prM. Thus, all four sites are in areas of the immature particle that are predicted to be less accessible to Golgi mannosidases. When the sites that failed to promote CD209 utilization (residues 52, 154, 230, 277, and 389) are mapped onto the immature DENV particle, four out of five are in the region of E that is neither overshadowed by the E-prM spikes nor covered by prM. Thus, four of five sites that allowed CD209L usage only are in areas of the particle that should be easily accessible to Golgi enzymes and thus prone to mannosidase trimming and the addition of complex sugars. One site (residue 389) that allowed interactions only with CD209L is in the overshadowed region of E; however, even when RVPs glycosylated at this site were grown in the presence of DMJ, they were only poorly capable of promoting CD209 utilization. Therefore, a lack of mannose-rich glycans at this position may not be responsible for its failure to promote CD209 use.

One limitation of this study is that although we were able to demonstrate a requirement for mannose-rich glycans in all RVP interactions with CD209, we were unable to demonstrate biochemically an increased incorporation of these glycans at the sites that promoted CD209 usage. The limiting factor was the low concentration of E protein found in RVP preparations made using the optimized transfection protocol described here (data not shown). Our own experiments with 1 mm DMJ-treated wild-type RVPs demonstrated that intermediate levels of DMJ treatment lead to sufficient incorporation of high mannose glycans to mediate interactions with CD209, yet these glycans are not detected by endoglycosidase treatment and analysis by Western blot. Therefore, we cannot exclude the presence, at some level, of mannose-rich glycans on particles that failed to interact with CD209.

In their recent cryo-EM reconstruction, Pokidysheva et al. (46) found that CD209 CRD monomers did not bind to all glycans at position 67 on DENV particles. Each icosahedral asymmetric unit within the mature DENV particle contains three E protein subunits. The CRD was bound to two of the three Asn-67 glycans, specifically those immediately adjacent to each other in adjoining E protein subunits. These glycosylation sites are unique in that they represent the most closely spaced oligosaccharide structures on the surface of the virion. However, in the case of WNV, CD209 can interact not only with carbohydrate at position 67, but also with carbohydrates at nearly any position provided they are in a high mannose state. Thus, a specific spacing between N-glycans is not required for productive CD209 binding, at least in the context of WNV.

Does this mean that there are fundamental differences in CD209 binding to WNV and DENV carbohydrates? More likely, the differences are a result of experimental approach; we measured functional CD209 interactions as judged by virus infection of cells expressing the native tetrameric CD209 molecule, whereas Pokidysheva et al. (46) assessed the binding of soluble CRD monomers to particles and visualized them using cryo-EM. The latter approach may only detect the highest affinity interactions. In the context of an infection, lower affinity interactions between a virus and CD209 might suffice to allow viral attachment. Furthermore, weaker binding to a single monomer might be amplified by multivalent binding to multiple CRDs within a single CD209 tetramer or to multiple CD209 tetramers on the surface of the target cell. Further studies will be required to clarify the precise mechanisms of binding that facilitate attachment and infection.