Site-specific glycosylation of Ebola virus glycoprotein by human polypeptide GalNAc-transferase 1 induces cell adhesion defects

The surface glycoprotein (GP) of Ebola virus causes many of the virus's pathogenic effects, including a dramatic loss of endothelial cell adhesion associated with widespread hemorrhaging during infection. Although the GP-mediated deadhesion depends on its extracellular mucin-like domain, it is unknown whether any, or all, of this domain's densely clustered O-glycosylation sites are required. It is also unknown whether any of the 20 distinct polypeptide GalNAc-transferases (ppGalNAc-Ts) that initiate mucin-type O-glycosylation in human cells are functionally required. Here, using HEK293 cell lines lacking specific glycosylation enzymes, we demonstrate that GP requires extended O-glycans to exert its deadhesion effect. We also identified ppGalNAc-T1 as largely required for the GP-mediated adhesion defects. Despite its profound effect on GP function, the absence of ppGalNAc-T1 only modestly reduced the O-glycan mass of GP, indicating that even small changes in the bulky glycodomain can cause loss of GP function. Indeed, protein-mapping studies identified a small segment of the mucin-like domain critical for function and revealed that mutation of five glycan acceptor sites within this segment are sufficient to abrogate GP function. Together, these results argue against a mechanism of Ebola GP–induced cell detachment that depends solely on ectodomain bulkiness and identify a single host-derived glycosylation enzyme, ppGalNAc-T1, as a potential target for therapeutic intervention against Ebola virus disease.

Ebola virus (EBOV) 2 poses a significant public health risk as evidenced by the severe outbreak in 2014 -2015 that spread to several continents and multiple emerging outbreaks detected in the years since. With mortality rates as high as 90% following infection, an effective drug intervention is sorely needed. As of now, no vaccine against EBOV has been approved for wide-spread use, and no therapeutic options exist to specifically combat Ebola virus disease. A deeper understanding of the pathogenesis of this disease will open new avenues for therapeutic development.
Ebola virus disease is partially characterized by uncontrolled inflammation and endothelial damage, often resulting in hypovolemic shock, multiorgan failure, and death. The EBOV surface glycoprotein (GP), the only viral gene product on the virus membrane, is instrumental in causing many of these effects, including a dramatic loss of cell adhesion in infected tissues (1), and induction of pro-inflammatory signals (2,3). Expression of GP alone in endothelial tissue explants results in loss of vascular integrity caused by reduced cell adhesion, recapitulating the symptoms observed in infected patients (1). Similarly, experiments in cultured cells show that expression of GP results in cell rounding and detachment, and the severity of rounding reflects the known virulence of different Ebola species (4,5). Thus, the cell culture adhesion effects are a proxy for a key aspect of the complex symptoms seen with human infection. Significantly, all of these deadhesion effects depend upon the presence of the extracellular mucin-like domain (MLD) of GP (1, 6), which contains dense clusters of sites predicted to be modified by mucintype O-glycosylation (7). The resulting large, bulky mass of the O-glycosylated MLD is believed responsible for deadhesion caused by steric blocking of surface adhesion molecules (8).
Mucin-type O-glycosylation, hereafter termed simply O-glycosylation, is a ubiquitous post-translational modification of proteins traversing the secretory pathway. O-Glycosylation begins in the Golgi apparatus with the addition of a GalNAc sugar onto a serine or threonine (occasionally tyrosine) residue by a polypeptide GalNAc transferase enzyme (ppGalNAc-T) (9). Humans express 20 distinct genes encoding the initiating enzymes (ppGalNAc-T1-T20). Although each of these isozymes adds an identical modification-a GalNAc sugar-the substrate can sometimes be uniquely targeted by a particular isozyme. Substrate selectivity is a product of the glycosite sequence, the presence of preglycosylated residues near the glycosite, and the tissue-specific expression of the isoenzymes (9). The nascent O-glycan is then further modified by addition of other sugar moieties to extend the glycan tree or terminally capped with a sialic acid (9). Based on experiments in mice, unique defects appear in animals deleted for individual ppGal-NAc-T isozymes, but for all those tested, no single isozyme is essential for life (9).

cro ARTICLE
The critical role of the Ebola MLD and the selective, yet nonessential, roles of individual ppGalNAc-Ts raise the possibility of targeting the enzymes that initiate O-glycosylation of the MLD for novel therapeutics. However, rather than blocking O-glycosylation directly, previous work used relatively large deletions within the 180 residue MLD (1,2,5). These may cause additional defects including improper protein folding, so direct evidence of the role of O-glycosylation in GP function is needed. Further, elucidation of the O-glycosylation code is an ongoing effort, and predictive models of both O-glycan acceptor sites as well as isozyme selectivity toward these sites are still unreliable (7,10). Thus, which, if any, individual ppGalNAc-T might be required for GP function must be determined experimentally.
Here we show that the O-glycans of GP are indeed crucial for GP-induced cell detachment and, furthermore, that ppGal-NAc-T1 plays a substantial role in producing functionally competent GP. Surprisingly, however, deletion of ppGalNAc-T1 caused only slight changes in GP mass. Further, we identified a small but functionally important region within the MLD for which point mutation of its glycosites abrogated GP function. These findings identify a functionally important discrete epitopewithintheMLDandsupporttargetedinhibitionofO-glycosylation as a potential therapeutic goal for treatment of Ebola virus disease.

GP requires extended O-glycans to inhibit cell adhesion
To test the hypothesis that O-glycans are required for GP function, we took advantage of a HEK293 cell line that has been genome edited to block expression of core 1 ␤3GalT-specific molecular chaperone (Cosmc) (11). Cosmc is an essential chaperone for glycoprotein-GalNAc 3-␤-galactosyltransferase (C1GalT1), an enzyme required for extension of most if not all O-glycans in HEK293 cells. Cosmc knockout (⌬Cosmc) cells lack extended mucin-type glycan trees beyond the initial GalNAc or GalNac with sialic acid (10,11). N-Glycosylation and other types of O-glycosylation are unaffected (12). Complete inhibition of O-glycosylation was not feasible because it would require deletion of all 20 of the genes encoding the ppGalNAc-Ts that initiate O-glycosylation. Although two inhibitors of ppGalNAc-transferases were recently identified, one is selective for ppGalNAc-T3 (13), and the other, though more broadly selective, also does not affect all ppGalNAc-Ts (14). A better-known O-glycosylation inhibitor, benzyl-␣-GalNAc, acts similarly to the ⌬Cosmc approach in that it prevents O-glycan extension by competitively inhibiting C1GalT1. In our tests, this compound proved too toxic for use (data not shown).
To begin, GP was expressed in WT HEK293 cells. As expected based on previous work (1,5), GP expression caused widespread cell rounding and detachment (Fig. 1A, top row). This effect depended on the mucin-like domain because GP with the mucin-like domain deleted (⌬MLD) showed little effect (Fig. 1A, top row). The ⌬MLD construct served as a baseline in our experiments to exclude contributions, if any, to cell rounding from parts of GP outside the MLD. Next, GP was expressed in ⌬Cosmc cells. Strikingly, this resulted in almost no cell rounding (Fig. 1A, bottom row). Indeed, the level of apparent cell rounding was indistinguishable from that of cells expressing ⌬MLD.
To provide a quantitative measure of attachment defects, an adhesion assay was conducted. Cells were mechanically After 30 min at 37°C, the floating cells in the media and the adherent cells were separately collected, stained with Trypan blue, and counted. C, shown for each construct expressed in the indicated cell type is the normalized percentage of cells that were floating. Normalization used ⌬MLD in the corresponding cell line (n ϭ 4 Ϯ S.E.; *, p Ͻ 0.05). D, WT and ⌬Cosmc (⌬C) lysates from the adhesion assay were immunoblotted using the indicated antibodies. E, intensity bands from D were quantified, and the GP/GAPDH ratio was plotted relative to the WT cell sample. F, untreated or PNGase F-treated WT and ⌬Cosmc (⌬C) cells transfected with GP were immunoblotted using anti-GP antibodies. The image is of a single membrane cropped to remove intervening lanes.

Ebola GP requires O-glycosylation by ppGalNAc-T1
harvested 48 h post-transfection and replated on a collagencoated surface for 30 min, after which floating and adherent cells were separately counted (Fig. 1B). This provided a sensitive test of the effect of GP expression on cell adherence to a substrate. For WT cells, GP significantly increased the unadhered cell fraction over that of ⌬MLD, whereas for ⌬Cosmc cells, the effect of GP was markedly reduced (Fig.  1C). To ensure that differences were not due to varied expression level, cell lysates of WT and ⌬Cosmc cells were immunoblotted using antibodies against the His 6 epitope appended to the GP constructs. Quantification using GAPDH levels as an internal loading control revealed no significant difference (Fig. 1, D and E). Although the expected gel shift was present for GP lacking extended glycans, it was more apparent with an increased SDS-PAGE run time (Fig. 1F). Because N-glycosylation accounts for much of the mass of GP, this shift was even more pronounced after digestion with PNGase F, which removes N-glycans but not O-glycans (Fig. 1F). In summary, preventing O-glycan extension protects against GP's inhibition of cell adhesion.

ppGalNAc-T1 activity is required for GP-mediated cell detachment
Encouraged by this result, we next tested whether the activity of any one individual ppGalNAc-T might be crucial for GP O-glycosylation. A panel of available HEK293 cell lines was used, each lacking a single major ppGalNAc-T. The ⌬T1, ⌬T2, ⌬T3, and ⌬T11 cell lines were previously characterized (15)(16)(17) and generously supplied by the Clausen group. Strikingly, whereas GP appeared fully active in the ⌬T2, ⌬T3, and ⌬T11 cell lines, it failed to cause significant rounding in ⌬T1 cells ( Fig.  2A). Background levels of rounding with ⌬MLD expression were minimal for all tested cell lines. Fig. 2B is provided to show additional representative fields illustrating the protective effect of ppGalNAc-T1 deletion and also showing detection of surface GP as indicated by immunostaining in the absence of permeabilization (Fig. 2B). GP must be delivered to the plasma membrane to reduce adhesion (6), and we detected no defects in GP surface presence in WT, ⌬T1, or ⌬T2 cells. To quantify the effect of deleting ppGalNAc-T1, we used the adhesion assay

Ebola GP requires O-glycosylation by ppGalNAc-T1
described above. In direct comparison with GP-transfected WT and ⌬T2 cells, the ⌬T1 cells showed significantly reduced levels of nonadherent cells (Fig. 2C). This was not due to reduced expression level because quantification of immunoblots showed indistinguishable GP expression between WT and ⌬T1 cells (Fig. 2D).

Only a subset of the O-glycans on GP are added by ppGalNAc-T1
As a complement to the adhesion assay, ⌬T1 cells were also compared with WT cells in a cell detachment assay. Here, plated cells were transfected with GP, and after 48 h, floating cells were collected and quantified separately from adherent cells. This is a stringent assay because, beyond rounding or readherence, the cells must break all pre-existing contact with neighbors and the substrate. Consistent with results from the adhesion assay, the absence of ppGalNAc-T1 protected against the effects of GP in promoting detachment (Fig. 3A). Previous work shows that HepG2 cells lacking ppGalNAc-T1 show changes to ϳ10% of the glycoproteome, compared with WT cells (18). To test whether these differences might affect ⌬T1 adhesion properties and artifactually account for the protection we observed, the adhesion assay was carried out in the absence of GP expression. No difference was observed (Fig. 3B). We also considered whether expression in ⌬T1 simply delayed the effect of GP. However, a time course of GP expression using the adhesion assay showed that loss of adhesion in WT cells was approximately linear over time following transfection, whereas it remained low in ⌬T1 cells (Fig. 3C). We also tested whether the stability of GP was altered in ⌬T1 cells. Using cycloheximide to inhibit GP synthesis and reveal its degradation rate, we found no significant difference in turnover rates between the two cell lines (Fig. 3D). Thus, the overall level, turnover rate, and localization of GP are comparable in ⌬T1 cells and cannot account for the reduced adhesion defects.
To confirm that ppGalNAc-T1 glycosylates GP, we tested for a gel shift for GP expressed in ⌬T1 cells. The samples were treated with PNGase F to remove N-glycans and maximize detection of a shift caused by O-glycans. A significant downward shift of 8.6 kDa was observed for GP from ⌬T1 cells, as compared with GP from WT cells (Fig. 3, E and F). Thus, some O-glycans on GP are in fact initiated by ppGalNac-T1, but the small magnitude of the shift indicates that these are relatively few in number (compare the shift in Fig. 3E with that in Fig. 1F). The large loss of function conferred by a small change in mass suggests that the bulk of the glycan mass on the MLD, still present in the absence of ppGalNAc-T1, is not sufficient for GP function.

Identification of glycosylation sites in the MLD required for GP deadhesion activity
To test whether discreet and functionally important glycosites could be identified within the MLD, we used a mutationally based mapping approach, building on the work of Simmons et al. (5). A set of nonoverlapping deletions was created in GP spanning the MLD (Fig. 4A). Each construct was expressed in WT cells and tested using the adhesion assay. Two regions were identified, residues 312-345 and 384 -411, whose deletion potently reduced the deadhesion activity of GP (Fig. 4B). Significantly, flanking regions of deletion did not show the same effect.
Although deletion of the segment 384 -411 impaired GP function, this was not due to its O-glycosylation. Point muta-tion of all six potential O-glycan sites within this segment to prevent their glycosylation yielded a construct, labeled x6, with full deadhesion activity indistinguishable from that of fulllength GP (Fig. 4, A-C). Based on this result, we focused on the 312-345 segment.
The 312-345 segment was bisected to further pinpoint a required region, resulting in two additional constructs: ⌬312-328 and ⌬329 -345. Although each deletion was of 16 residues, only 329 -345 blocked GP-mediated deadhesion (Fig. 4B). Note that neither the number of potential O-glycan sites (boxed in the figure) nor the length of the deleted segments correlated with the degree of defect. Because 329 -345 contained only five potential glycan sites, we introduced point mutations at all five sites. This final construct (labeled x5) therefore comprised the

Ebola GP requires O-glycosylation by ppGalNAc-T1
full length of GP with a loss of only 5 of the potential 56 O-glycosylation sites in the MLD. Remarkably, this construct was significantly impaired in its ability to induce adhesion defects (Fig. 4B). The expression level of this construct did not significantly differ from that of WT GP (Fig. 4C), nor did it show a loss of plasma membrane localization (Fig. 4D). Further, following PNGase F treatment, a clear gel shift was observed for the x5 construct when compared with full-length GP (Fig. 4E). This observation confirms that O-glycans are normally added at one or more of these functionally required sites.
As a test of whether the mutated sites in the x5 construct encompass all of the ppGalNAc-T1 sites in GP, the x5 construct was expressed in ⌬T1 cells. We reasoned that additional sites would be indicated by a further downward gel shift. Indeed, a further shift was observed, but it was only 2 kDa (Fig. 4F). Because this shift was smaller than the 9-kDa shift for fulllength GP expressed in ⌬T1 cells, the result suggests that ppGalNAc-T1 acts at several of the sites within the 329 -345 segment and only a few outside of it. Consistent with this finding, there was no further functional deficit for the x5 construct when it was expressed in ⌬T1 cells (data not shown). Although a small change may have been missed because of the statistical certainty of the assay, the 329 -345 segment appears to contain most, if not all, of the functionally important ppGalNAc-T1 sites.

Discussion
The MLD of the Ebola GP is critical for many of the virus' pathogenic effects, including hemorrhaging, which in turn is caused partly by a loss of cell adhesion. Here we show that without extended O-glycans, GP cannot modulate adhesion. Further, not all sites of O-glycosylation on GP are required equally. Rather, GP-induced cell detachment depends on glycans added by ppGalNAc-T1 and on glycosites found in a particular region of the MLD. These findings provide direct evidence of O-glycan contribution to GP function and suggest that GP-mediated cell deadhesion is due to selective MLD glycosites rather than its nonspecific O-glycan bulk. The findings also identify ppGalNac-T1 as a potential therapeutic target.
The requirement for O-glycosylation of GP, shown here by preventing glycan extension, substantiates previous work that had deleted the entire MLD (1, 2, 5). As mentioned above, deletions of the protein backbone may result in undesired effects on protein structure. A complete block of O-glycosylation might also affect structure, particularly rigidity, because clustered O-glycans can provide a basis for rigid conformations and for protein stability (19). Preservation of the initial GalNAc and attached sialic acid, as is the case for ⌬Cosmc cells (10, 11) may retain the structural characteristics of the MLD, while additionally preserving any functional role of the protein backbone itself. Therefore, we attribute the significance of the MLD in GP-induced deadhesion to its O-glycans.
There are two major hypotheses regarding how O-glycosylation regulates GP function. The prevailing view is that glycan bulk in the MLD sterically blocks cell adhesion interactions. Alternatively, the MLD's deadhesion activity may depend on specific glycan determinants, which require a full glycan structure to be recognized by host factors that regulate adhesion.
The favored hypothesis is partly based on the finding that antibodies against the cell adhesion molecule ␤1-integrin fail to bind their surface antigen in the presence of GP, and this masking effect is reversed with deglycosylation of the cell surface (8). Further, the trimeric umbrella-like structure of the GP ectodomain is suggestive, particularly given that its overhanging regions are comprised of the bulky densely packed MLD O-glycans (8,20,21). One would expect this mechanism of inhibition to be insensitive to the absence of just a few of the many possible glycans. Thus, our observation of functional deficit from removal of select glycans by mutation or by ppGalNAc-T1 deletion supports the idea that there may be specific glycan-based epitopes within the MLD that are needed for its deadhesion activity. In this light, it is noteworthy that deadhesion activity is retained by a chimeric construct of transmembrane-anchored MLD that is monomeric, rather than trimeric (6).
To cause cell deadhesion, GP may activate signaling pathways that modulate cell adherence. GP is known to cause signaling in pathways involved in the inflammatory response, and this activity is present in the shed form of the ectodomain (2, 3). Future work that precisely identifies the sites and structures of the functionally relevant GP glycans will be an important step in obtaining a mechanistic understanding of their role in GP-induced cell detachment. Recent advances in MS-based identification of glycosites unique to particular ppGalNAc-T isozymes will likely be critical in this endeavor (7,10,18,22,23).
Regardless of the mechanism of GP-induced deadhesion, it is significant that protection is conferred by blocking a single ppGalNAc-T isozyme. ppGalNAc-T1 is expressed across all tissue types and organs tested (9), suggesting that its inhibition could be effective across the broad range of host tissues susceptible to Ebola infection. Although inhibition of all mucin-type O-glycosylation would cause severe side effects (12), selective inhibition of ppGalNAc-T1 may be tolerable, especially if it is titrated by dose and time. Mice lacking ppGalNAc-T1 are viable and even fertile despite defects in blood clotting and bone development (24). Other functions that involve ppGalNAc-T1 are basement membrane deposition (25) and extracellular matrix remodeling (26). The best characterized ppGalNAc-T1-specific substrates are two heavily O-glycosylated proteins involved in bone deposition and remodeling, bone sialoprotein, and osteopontin (27). There are at least 55 other glycoproteins that specifically depend on ppGalNAc-T1 for their glycosylation (18), but the functional importance of their glycosylation is unknown. Although an inhibitor of ppGalNAc-T1 has not yet been characterized, recent identification of a ppGalNAc-T3 inhibitor demonstrates the strong potential for developing inhibitors against ppGalNAc-Ts that are effective and isozyme selective at nontoxic doses (13,16).
In summary, Ebola GP requires extended O-glycans to cause cell detachment, and surprisingly, the loss of relatively few of these glycans, either by mutation or by deleting ppGalNAc-T1, blocks GP activity. These observations point to the possible role of specific glycan-based epitopes in the mucin-like domain that may act separately or in conjunction with nonspecific steric interference. Additionally, the requirement for ppGalNAc-T1 reveals for the first time a single glycosylation enzyme that

Ebola GP requires O-glycosylation by ppGalNAc-T1
could be an effective therapeutic target during Ebola virus disease.

Rounding, adhesion, and detachment assays
For all assays, the cells were transfected using JETPei transfection reagent (PolyPlus Transfection, Illkirch, France, catalog no. 101) according to the manufacturer's protocol. Except for the time course, all cells were then cultured for 48 h. For the rounding assay, the cells were directly imaged using an EVOS FL cell imaging system (Invitrogen) at 10ϫ magnification. For the adhesion assay, transfected cells were mechanically suspended (repeated gentle pipetting with a P1000 in growth medium), and ϳ2 ϫ 10 6 cells/well were transferred to a precoated 12-well plate. The plates were coated by addition of collagen coating solution (Cell Applications, San Diego, CA, catalog no. 125) for 1 h at 37°C, two washes with PBS, and drying. After cells were transferred to coated plates, they were recultured for 30 min, and medium containing floating cells was collected. Adherent cells were trypsinized (Corning, catalog no. 25-053-Cl) without washing and collected. Aliquots of the cell solutions were diluted 1:1 with Trypan blue dye and counted with the Luna II automated cell counter (Logos Biosystems, Anyang-si, South Korea). Only dye-excluded cells were included, and the measured cell concentration was used to calculate the total number of cells per sample. Next, the percentage of unadhered cells/well was calculated (floating/(floating ϩ adherent)). For the detachment assay, the medium containing floating cells was collected, the remaining adherent cells were trypsinized and collected, and the two samples were analyzed as described.

Immunoblotting
To determine GP expression levels, aliquots of cells were collected from the corresponding adhesion assay, washed once with PBS, and lysed with nonreducing sample buffer. Following separation by SDS-PAGE (4 -16%) and transfer onto nitrocellulose, membranes were cut between the 55 and 70-kDa markers for GP and GAPDH detection. Both were blocked for 1 h at room temperature with mild agitation in 5% nonfat dried milk in PBST (PBS ϩ 1% Tween 20), followed by incubation with primary antibody diluted in block for 1 h. Antibodies were anti-His (see above, but used at 1:2000) and anti-GAPDH (Cell Signaling Technology, Danvers, MA, rabbit mAb GAPDH clone 14 C10, catalog no. 2118, 1:3000 dilution). Following five washes in PBST, membranes were incubated with horseradish peroxidase-coupled secondary antibodies, washed again, and developed using Clarity ECL (Bio-Rad, catalog no. 170-5060) according to the manufacturer's protocol. The membranes were visualized using the Bio-Rad ChemiDoc touch imaging system (catalog no. SCR-014210) and analyzed with Image Lab software. To measure GP turnover rates, cells were treated with 10 g/ml cycloheximide in media at 37°C for indicated times before cell lysis and blotting using anti-His antibody, as above. To assay for gel shifts, the cells were washed twice in flow wash buffer (PBS ϩ 1% fetal bovine serum ϩ 0.5% sodium azide) and incubated with 50,000 units/ml PNGase F (New England Bio-Labs, Ipswich, MA, catalog no. P0704) for 30 min at 37°C. The cells were washed twice in flow wash buffer and lysed using reducing sample buffer and immunoblotted as above, but with anti-GP antibody (see above, 1:1000 dilution). Molecular weights were assessed using Image Lab software (Bio-Rad) using semi-log regression analysis. Images of uncropped immunoblots can found in Figs. S1-S4.