Inhibition of SARS-CoV-2 viral entry in vitro upon blocking N- and O-glycan elaboration

The Spike protein of SARS-CoV-2, its receptor binding domain (RBD), and its primary receptor ACE2 are extensively glycosylated. The impact of this post-translational modification on viral entry is yet unestablished. We expressed different glycoforms of the Spike-protein and ACE2 in CRISPR-Cas9 glycoengineered cells, and developed corresponding SARS-CoV-2 pseudovirus. We observed that N- and O-glycans had only minor contribution to Spike-ACE2 binding. However, these carbohydrates played a major role in regulating viral entry. Blocking N-glycan biosynthesis at the oligomannose stage using both genetic approaches and the small molecule kifunensine dramatically reduced viral entry into ACE2 expressing HEK293T cells. Blocking O-glycan elaboration also partially blocked viral entry. Mechanistic studies suggest multiple roles for glycans during viral entry. Among them, inhibition of N-glycan biosynthesis enhanced Spike-protein proteolysis. This could reduce RBD presentation on virus, lowering binding to host ACE2 and decreasing viral entry. Overall, chemical inhibitors of glycosylation may be evaluated for COVID-19.


INTRODUCTION
SARS-CoV-2 is the virus causing the global pandemic 'coronavirus diseases 2019' . This is a zoonotic beta-coronavirus from bats that causes severe acute respiratory syndrome (1). Its effects on human physiology extends well beyond the lung as it unleashes a cytokine storm to alter immune response (2) and cause disseminated intravascular coagulation (3).
It also exhibits multiorgan tropism that impacts kidney, liver, heart and brain function (4). Among its structural elements, the SARS-CoV-2 Spike protein is critical for viral attachment, fusion and entry into host (5)(6)(7). The RBD region of Spike enables binding to its primary human cellular receptor, angiotensin-converting enzyme-2 (ACE2), which is ubiquitously expressed on epithelial, endothelial and blood cells (6,8). A role for heparan sulfates in viral entry has also been proposed (9). Once bound, viral fusion and entry depends on two proteolysis sites, the furin/RRAR-S site located between the Spike S1 and S2 subunits, and a second S2'/SKR-S site (10). A variety of enzymes including TMPRSS2 (transmembrane protease, serine 2) and cathepsin-L may aid viral entry (11)(12)(13). While inhibitors of viral binding, RNA transcription and protease activity are being tested to reduce viral load, this manuscript suggests an orthogonal approach based on glycan engineering.
Both the viral Spike-protein and ACE2 receptor are extensively glycosylated, with a majority of the 22 N-glycosylation sites of Spike and 7 N-glycosylation sites of ACE2 bearing carbohydrates ((14-16), Figure 1). The exact distribution of the oligomannose, hybrid, complex, sialylated and fucosylated structures in these macromolecules is likely dictated both by the protein structure and host expression system (17,18). O-linked glycans are also reported on both proteins (16,17). A variety of functions have been proposed for these glycans including immunological shielding (18), direct regulation of Spike-ACE2 binding (19), and control of Spike up/down conformation (20). Experimental data supporting these concepts is currently lacking and thus our understanding of the impact of glycosylation on SARS-CoV-2 function is incomplete.
We address the above gaps in knowledge in the current manuscript. Our results suggest that both the O-and N-linked glycans of the SARS-CoV-2 Spike protein, including sialylated glycan epitopes, have a relatively minor role in regulating direct Spike-ACE2 binding interactions.
However, these glycans control the rates of viral entry into HEK293T cells expressing human ACE2. Here, blocking N-glycosylation on SARS-CoV-2 pseudovirus at the high-mannose stage using CRISPR-Cas9 and also a small molecule inhibitor (kifunensine) resulted in extensive cleavage/shedding of the viral Spike protein at the time of production due to enhanced proteolysis at the S1-S2 interface. Our data also suggest that glycans may contribute to additional aspects of Spike proteolysis during viral entry. Due to these effects, viral entry into human ACE2 expressing cells was reduced by >95 % when virus was produced in the CRISPR-Cas9 MGAT-1 knockout cells lacking complex N-glycans, and by 85-90 % upon production in the presence of kifunensine.
These observations support the need to screen chemical inhibitors of glycosylation for the treatment of COVID-19.

Modest role for ACE2 sialic acids during Spike protein molecular recognition
In order to study the impact of N-and O-linked glycosylation on SARS-CoV-2 function, we engineered cells over-expressing full-length Spike-protein and ACE2 (Figure 2A). We also histidine-tag purified soluble, dimeric Fc-his fusion proteins for the Spike S1 region, RBD and extracellular portion of ACE2. The molecules were extensively glycosylated with their apparent molecular mass being greater than their theoretical mass based on peptide alone. Thus, RBD-Fc was ~60 kDa instead of a theoretical mass of 51 kDa, S1-Fc was ~150 kDa rather than 101 kDa, and ACE2-Fc was ~140 kDa instead of 110 kDa ( Figure 2B). RBD-Fc and S1-Fc readily bound to HEK293T cells upon human ACE2 over-expression, confirming that they are functionally active ( Figure 2C). ACE2-Fc also specifically bound Spike-protein expressed on 293Ts. Baseline RBD-Fc and S1-Fc binding to Spike expressing 293T cells (i.e. 293T/S) was even lower than that of wild-type 293T (red line, Figure 2C). This suggests expression of low amounts of Spike binding proteins on WT-293Ts. These basal Spike receptors may engage cell-surface Spike on 293T/S cell (via cis-interactions) causing RBD-Fc and S1-Fc binding to be lower in 293T/S cells compared to WT-293Ts.
Sialidase treatment of Spike-protein expressed on 293Ts did not affect ACE2-Fc binding ( Figure 2D). Sialidase treatment of ACE2 expressed on 293Ts, however, increased RBD-Fc and S1-Fc binding by 26 % and 56 % respectively. Control lectin binding studies confirmed the high activity of the sialidase enzyme used in this study ( Figure 2-figure supplement 1A). The relatively small effect of the sialidase in these studies was not due to the absence of sialic acid on either S1-Fc or ACE2-Fc. In this regard, independent glycoprotemics mass spectrometry (MS) data showed that ~80-100% of the N-glycans expressed at specific sites of ACE2-Fc and ~40-100% of the S1-Fc glycopeptides expressed complex-type glycans (K.C., et al., manuscript in preparation). Up to ~60% of the antennae on these complex structures on ACE2 and ~20% of the structures on S1-Fc were terminated by sialic acid.

Neither sialidase treatment of Spike nor ACE2 markedly impacted viral entry
To complement the above binding studies, we measured SARS-CoV-2 pseudovirus entry into stable 293T/ACE2 cells. This assay provides an aggregate measure of both molecular binding and viral fusion/entry in the context of the physiological Spike trimeric configuration (21). Here, the control VSVG (Vesicular stomatitis virus G-protein) pseudotyped virus displayed broad tropism both for wild-type HEK293T and stable 293T/ACE2 ( Figure 2E only entered 293T/ACE2 (22). This mutation results in reduced proteolysis at the S1-S2 interface, as discussed later, likely due to the presence of a single arginine site at the S1-S2 interface rather than the polybasic ('RRAR') sequence. To study the role of sialic acid on viral entry, we titered the Spike-WT and Spike-mutant virus using p24 ELISA (Figure 2-figure supplement 2). The virus were then treated with a pan-sialidase from Arthrobacter ureafaciens, and the same amount of Spike-WT and Spike-mutant particles were applied to stable 293T/ACE2 cells ( Figure 2F, Figure   2-figure supplement 1C). Here, Spike-mutant pseudotyped virus was ~5-10 times more effective at stimulating DsRed reporter expression compared to Spike-WT. Thus, the efficiency of cleavage at the S1-S2 interface can fine-tune SARS-CoV-2 infectivity (11,13,14). Sialidase treatment of virus did not impact viral entry, consistent with the notion that Spike sialic acid does not regulate molecular recognition. Additionally, sialidase treatment of 293T/ACE2 did not alter either VSVG or Spike-WT pseudovirus entry ( Figure 2G, Figure 2-figure supplement 1D). A partial increase in Spike-mutant viral entry was measured upon sialidase treatment (P=0.08), suggesting a minor inhibitory role for ACE2 sialic acid when viral infectivity is high. Glycoproteomics analysis of purified pseudovirus produced in HEK293T cells showed that ~45-100% of the viral N-glycans were complex-type, with up to ~55% of their antennae being terminated by sialic acid (K.C., et al. manuscript in preparation). Thus, the pseudovirus used in this study was processed by glycoenzymes that are typically found in the cellular medial and trans-Golgi compartments.
Together, the binding and pseudovirus data suggest that ACE2 sialic acids may modestly shield virus/Spike binding in some biological contexts and perhaps some cell systems.

ACE2 binding
To determine if other aspects of N-and O-linked glycosylation may impact SARS-CoV-2 function, we applied CRISPR-Cas9 technology to develop isogenic HEK293T clones that lack either the core-1 O-glycan forming galactosyltransferase C1GalT1 or the N-glycan branching β1,2GlcNAc-transferase MGAT1 ( Figure 3A, (23,24)). Here, knocking out C1GalT1 blocks Oglycan biosynthesis at the GalNAcα-Ser/Thr (+/-sialic acid) stage as it is not elaborated to form  Figure 3D). Similarly, either S1-Fc ( Figure 3E) or RBD-Fc ( Figure   3F) were applied to ACE2 bearing 293T or the N-/O-glycan knockouts. These Fc-proteins were applied at sub-saturation concentrations in the cytometry binding studies in order to quantitatively evaluate differences in receptor-ligand binding. Here, we observed a 40% decrease in ACE2-Fc did not impact either S1-Fc ( Figure 3E) or RBD-Fc binding ( Figure 3F). Here, RBD-Fc bound ~10-times more avidly to ACE2 compared to S1-Fc. Together, the data suggest a modest role for Spike-protein glycosylation on direct Spike-ACE2 binding. Our observation that the binding of RBD for ACE2 was substantially higher than S1-ACE2 interactions is consistent with a previous report (13). This highlights the need to carefully consider RBD presentation/conformation in the context of the full protein when quantifying molecular affinity to ACE2.

Blocking N-glycan elaboration on Spike abrogated viral entry
We evaluated the impact of ACE2 and Spike glycosylation on viral entry. To determine the impact of ACE2 glycosylation, we transiently expressed this protein on wild-type 293Ts, The Spike pseudovirus produced in [N]‾293Ts also appeared to have lower intact Spike protein.
To examine this more closely as the anti-S2 pAb may have preference for binding the S2-domain over full Spike, we measured the FLAG-epitope at the C-terminus of Spike-mutant using an anti-FLAG mAb (monoclonal antibody). This may provide more uniform recognition of both the full Spike and the S2-subunit ( Figure 4F). This blot suggests that N-glycan loss may trigger precipitous (95 %, based on densitometry) loss of intact Spike protein due to enhanced cleavage at the S1-S2 interface. Consistent with the viral entry assay, we also noted greater proteolysis of virus produced in [O]‾293Ts (52 %) compared to that produced in wild-type 293Ts (32 %). Together, the data suggest that both Spike N-glycans, and possibly also O-glycans, may play a role in regulating SARS-CoV-2 Spike protein stability.

Small molecule inhibitors of N-linked glycosylation drastically reduced viral entry
Remarkably, the same observations as seen in the viral N-glycan knockouts could be In order to determine if the enhanced S1-S2 site proteolysis of Spike protein upon N-glycan inhibition is the exclusive reason for reduced viral entry, we produced virus with a Spike variant ('Spike-delta') that contained the C-terminal FLAG epitope but that lacked the furin-site ( Figure   5F). The substitution of the 'RRAR' sequence with a single Alanine ('A') resulted in viral Spike that was not extensively cleaved both in the absence and presence of kifunensine. Robust viral entry into 293T/ACE2 cells was observed using this furin resistant virus, and this too could be partially blocked by kifunensine. Thus, in addition to proteolysis at the S1-S2 site, N-glycans may have additional roles during viral entry.

DISCUSSION
This manuscript undertook a series of studies in order to comprehensively describe the role of O-and N-linked glycans on SARS-CoV-2 Spike binding to human ACE2 ( Figure 6A), and related viral entry ( Figure 6B). It demonstrates only a minor role for carbohydrates in regulating Spike-ACE2 direct binding. In this regard, we observed some enhancement in Spike binding and viral entry upon treating 293T/ACE2 with a pan-sialidase. Based on the published crystal structure and molecular modeling, the critical glycans regulating this process are likely located at Asn-90 or Asn-322, proximal to the Spike binding interface ( (7), Figure 1). The effect was nevertheless small compared to that reported for other sialic acid dependent viruses like influenza (26) This observation is consistent with previous cryo-EM (14,15) and molecular dynamics simulation results (29), which show that the active/"up" form of Spike RBD is largely exposed with few glycans at the binding interface. It is now well appreciated that this lack of glycan shielding, makes the exposed RBD a prime target for vaccine development.
In contrast to the modest effect of glycosylation on receptor-ligand binding, we observed a much more profound role for glycans in regulating viral entry in part due to their impact on Spike proteolysis at the S1-S2 interface. In this regard, we observed that blocking N-glycan elaboration using both genetic approaches (i.e. MGAT1-KO/[N]‾293T) and the natural product mannosidase-I inhibitor, kifunensine, dramatically reduced viral entry into 293T/ACE2 cells. This was observed for two virus types: Spike-WT and Spike-mut. In the case of Spike-mut, enzymatic and small molecule inhibition resulted in enhanced proteolysis of Spike-protein during pseudovirus production. These data suggest an inverse relation between furin cleavage and viral entry into 293T/ACE2 cells. Consistent with this, Spike-mut displayed reduced S1-S2 proteolysis compared to Spike-WT, and higher viral infectivity. Others have also noted a reduction in viral entry, upon increasing furin cleavage potential by addition of more basic residues ('RRRKR' mutation) at the S1-S2 interface (11). Finally, a recent study observed that the introduction of the D614G mutation in Spike reduced furin cleavage, and this resulted in enhanced viral infectivity (30). SARS-CoV-2 containing the D614G mutation is currently the dominant strain worldwide (>80% prevalence, (31)), and the possibility that this is due to altered furin cleavage potential is hotly debated. Overall, our studies with MGAT1-KO, C1GalT1-KO and kifunensine suggest that both N-and O-glycans may modulate the rates of proteolysis at the S1-S2 interface, thus affecting viral entry ( Figure 6C).
While some S1 may be associated with Spike even after partial cleavage at the S1-S2 interface (10), this may potentiate a quantitative reduction in RBD presentation on viral surface, resulting in reduced ACE2 binding and lower viral entry. Indeed several N-glycans are located proximal to the furin site: N 603 and N 657 , and also N 61 (18). O-glycans have also been hypothesized to exist in this region at S 673 , T 678 and S 686 (32), and some of these have been experimentally verified (33).
How this site-specific glycosylation affects SARS-CoV-2 entry, remains to be determined.
In addition to its role in regulating RBD presentation, our data suggest that N-glycans may also have other roles in regulating viral entry. In this regard, we noted that even the robust entry of Spike-delta pseudovirus into 293T/ACE2 cells could be partially inhibited by kifunensine ( Figure 6C), and this is independent of S1-S2 proteolysis. Others have noted that the effect of 'Spike-delta' mutation may be cell type specific in that this mutation results in lower infectivity in some cell types like Calu-3 which rely on the enzyme TMPRSS2 for viral fusion, while this mutation does not affect entry into other cell types like VeroE6 which lacks TMPRSS2 (11). Based on these observations, we speculate that additional glycosylation sites proximal to S2' may also modulate viral entry, perhaps because some proteases like TMPRSS2 have both ligand-binding and protease activity. The N-glycans proximal to S2' include N 801 , and to a lesser extent N 616 and N 657 . Additional studies with more cell types and mutant virus are necessary in order to fully elucidate the molecular mechanism by which O-and N-glycans regulate host-cell specific viral entry response.
In addition to basic science understanding, our findings have translational potential as it suggests that both O-and N-glycan truncation may be used to reduce viral entry. In this regard, it would be attractive to test additional N-glycosylation inhibitors, like Swainsonine, which has a demonstrated safety profile in humans (34). Additional potential inhibitors that may modulate viral entry include α-mannosidase inhibitors (deoxymannojirimycin, mannostatin A), α-glucosidase inhibitors (N-butyl deoxynojirimycin, N-nonyl-deoxynojirimycin, castanospermine, celgosivir) (35), and finally compounds like SNAP (36,37) and 4F-GalNAc (38) which block various aspects of N-and O-glycan biosynthesis. These potentially represent off-the-shelf drugs and compounds that may be repurposed to reduce viral load and ameliorate SARS-CoV-2 related respiratory symptoms. Studies are currently underway in our laboratory to test these concepts, and extend the assay to authentic SARS-CoV-2 strains. Overall, while SARS-CoV-2 has developed a natural ability to enhance infectivity, we propose that glycoengineering may provide a strategy to take advantage of these evolutionary features to regulate viral entry and reduce disease transmission. Anti-FLAG clone L5 was from Biolegend (San Diego, CA). All secondary antibodies (Abs) were from Jackson ImmunoResearch (West Grove, PA). When necessary, lectins and Abs were labelled by addition of 25-fold molar excess succinimidyl ester coupled Alexa-405, Alexa-488, Alexa-555 or Alexa-647 dye (Fluoroprobes, Scottsdale, AZ) to protein suspended in phosphate buffered saline (PBS, pH 7.4) for 1 h at room temperature (RT). Following this, the reaction was quenched with 1/10 th volume 1 M Tris, and unreacted Alexa-dye was removed using 7 kDa molecular mass cutoff Zeba desalting spin columns (Thermo). Unless mentioned otherwise, all other reagents were from either Thermo-Fisher or Sigma Chemicals.

Molecular dynamics simulations:
The ACE2-RBD complex structure (6LZG, (40)) was superimposed to the trimeric spike protein state (6VYB (14)), in which one of the RBD domains is "open" so that ACE2 is associated with the RBD domain in the open conformation. Missing residues in the structure were modeled using PyRosetta (41). N-glycans were then added to the structure using Glycan Reader within CHARMM-GUI (42,43). The identity of the glycans was assigned based on a published work (18). The final structure contained glycan modifications at 54 spike Asn and 6 ACE2 Asn.
All simulations were performed using NAMD2 (44) using the CHARMM36 parameters (45). The complex was first placed in a water box of sufficient size to ensure a minimum 12 Å separation from the wall. Net charge in the molecule was neutralized by adding 841 K+ and 743 Cl-ions, resulting in 867,166 atoms total. Protein and glycan atoms were first fixed while water (TIP3) molecules were energy minimized for 10,000 steps. Next, the entire system was relaxed for another 10,000 steps, and the system temperature was gradually increased to 310 K in increments of 1 K with 120 fs of equilibration at each temperature. A time step of 2 fs was used. Constant temperature and pressure were maintained using the Langevin framework (46). Long range electrostatic interactions were treated using the particle mesh Ewald method (47). The simulation was continued for 26.5 ns. Intermediate structures were saved every 10 ps.  In some assays, cells were treated with sialidase (200 U/mL Arthrobacter ureafaciens α2-3,6,8,9-Neuraminidase, New England BioLabs) for 1 h at 37 ⁰C, and washed using HEPES buffer prior to the binding assay. Sialidase activity was verified based on SNA and ECL staining. In this case, control cells were treated identically, except that sialidase was withheld. SNA was directly conjugated with Alexa-555 and ECL was conjugated with Alexa-647 in order to simplify the implementation of dual color cytometry measurements.       Yang et al. Figure 6 Spike-ACE2 binding assay

Spike-ACE2 molecular recognition (finding):
• Minor role for ACE2 sialic acid in shielding binding • Small regulatory role for Spike N-glycans, but no role for O-glycans