GalNAc-T3 and T7 Initiates Clustered O-glycosylation Covering the Multibasic Cleavage Site in S Protein of SARS-CoV-2
From recombinant spike proteins and authentic virus, previous studies have uncovered a few O-glycosylation sites, including T676, T678 and S686, near the multibasic furin site of the spike protein of SARS-CoV-222, 23, 29, 40–42 (Supplementary Fig. 1). Since there are in total 20 mammalian GalNAc-T isozymes, whose substrate specificity determines sites of O-glycosylation43, our first goal was to identify the specific GalNAc-Ts modifying near the furin cleavage site and to locate corresponding glycosites. We surveyed the in vitro activity of GalNAc-T1, T2, T3, T4, T5, T7, T11, and T18, representing different subfamilies of GalNAc-Ts43 (Supplementary Figs. 2 and 3), on a synthetic peptide encompassing the multibasic cleavage site sequence of S protein from the original Wuhan strain (Wuhan-Hu-1). Among all the tested GalNAc-Ts, we found that GalNAc-T3 demonstrated the most substantial activity on the synthetic multibasic substrate, adding a single GalNAc to the peptide (Fig. 1a). Using an electron transfer dissociation-based liquid chromatography coupled mass spectrometry (ETD-LCMS) method, we determined that GalNAc-T3 specifically modified the residue corresponding to T678 of the spike protein (Fig. 1b). In line with previous studies, our results indicated that GalNAc-T1, a “house-keeping” GalNAc transferase, could also modify the furin site-containing peptide in vitro, albeit with less efficiency38, 39.
All GalNAc-Ts except for GalNAc-T20 have a characteristic lectin domain besides a catalytic domain, which allows them to work in tandem by recognizing prior glycosylated substrates and creating additional glycosylation sites in close proximity16, 44. We therefore investigated which GalNAc-T could potentially pick up the GalNAc-T3 modified peptide to generate more glycosylations near the furin site. Interestingly, among the tested GalNAc-Ts, only GalNAc-T7 readily took the prior glycosylated spike peptide, adding upto two more GalNAc residues to the peptide (Fig. 1c). Using a similar ETD-LCMS method, we discovered that GalNAc-T7 mostly modified the residue corresponding to S686 of the spike protein (Fig. 1d), with a minor activity towards the residue corresponding to S689 (Supplementary Fig. 4).
Taken together, in vitro glycosylation results suggested that GalNAc-T3 and T7 could work in tandem to potentiate clustered O-glycan modifications near the furin cleavage site of SARS-CoV-2 S protein. Because the glycosylation by GalNAc-T7 at S686 is right next to the furin cleavage site, it can theorietically result in a great degree of inhibition against protease processing. Furthermore, determination of specific GalNAc-Ts and the modification order of the O-glycan clusters allowed us to test the impact of host cell O-glycosylation on SARS-CoV-2 S protein in vitro and on authentic viruses in human lung cell, as shown below.
GalNAc-T3 and T7 Inhibit the Furin Processing of the Spike Protein.
In order to test whether the glycosylation by GalNAc-T3 and T7 could inhibit the protease processing at the multibasic cleavage site of S protein, we constructed a luciferase-based biosensor, which encodes the S1/S2 boundary sequence sandwiched between a Gaussia luciferase and a membrane-anchored eGFP. Cleavage of the multibasic sequence by furin releases luciferase into the medium, while the membrane bound eGFP allows normalization of the luminescence signals due to variations in transfection and expression levels (Fig. 2a). Similar biosensors were once used to detect isoform specific activity of GalNAc-T2 and T3 in previous studies45. As demonstrated in Fig. 2b, annihilation of the furin cleavage site by a single mutation of the residue corresponding to R685 of the spike protein to alanine (R685A) reduced the luminescence signal to almost background, confirming that the constructed biosensor is a valid reporter of the protease processing activity in the multibasic cleavage site-containing sequence.
As expected, mutating either of T678 or S686 or both to alanine increased the luminescence signal by ~ 50%, which suggests those glycosylation sites are protective for the furin site in our biosensor, likely through O-glycosylation in HEK293T. Interestingly, the double mutant did not produce more luminescence increment than S686A or T678A mutant, which is in line with the dependence of glycosylation at S686 on prior glycosylation at T678 (Fig. 2b). Consistently, co-expression of the sensor with GalNAc-T3 and T7 greatly suppressed luciferase signals in the medium, whereas the suppression with GalNAc-T3 or T7 alone is not as significant (Fig. 2c). Together with the in vitro enzymatic assay, this result suggests that sequential glycosylations at the multibasic cleavage site of SARS-CoV-2 S protein by GalNAc-T3 and T7 inhibit furin cleavage.
To find out if the same applies to full length spike protein, we first examined protease processing of spike protein in HEK293T cells by comparing its overexpression and processing with or without mutations in the glycosylation sites. To our surprise, single mutation of T678A or S686A did not change the production of S2 fragment from the overexpressed spike protein significantly, whereas double mutation of these two potential glycosylation sites seemed to cause a slight increase in S2 production (Fig. 2d and Supplementary Fig. 5). This suggested to us there could be additional glycosylation sites in S protein that protect the multibasic cleavage site in HEK293T. Because GalNAc-T7 displayed promiscuity towards the peptide substrate in vitro, we speculated that GalNAc-T7 may have other modification sites on the full length spike protein, including the previously-discovered S689. We therefore stacked the mutation of T678A and/or S686A with genetic knockout (KO) of GALNT7 (Supplementary Fig. 6), the gene encoding GalNAc-T7. Indeed, in the absence GalNAc-T7, mutating either of T678 or S686 or both glycosylation sites significantly increased production of S2 fragments (Fig. 2d and Supplementary Fig. 5). Notably, even the native sequence of spike protein produced more S2 fragment when GALNT7 was knocked out, confirming that GalNAc-T7 is a key enzyme protecting the multibasic cleavage site of the spike protein.
As the occupancy of O-glycosylation is usually low in comparison to N-glycosylation22, 23, 29, we attempted to increase the degree of glycosylation near the multibasic cleavage site by site-specific knock-in (KI) of GALNT3, the gene encoding GalNAc-T3, and/or GALNT7 in HEK293T (Supplementary Fig. 7). As displayed in Fig. 2e and Supplementary Fig. 8, while KI of GalNAc-T7 partially protected the spike protein from being processed, double KI of GALNT3 and GALNT7 almost completely blocked the production of S2 fragment from the overexpressed spike protein. In contrast, KI of either GALNT3 or GALNT1 alone had no obvious effect on spike protein processing39. In all, our biosensor assay and western blot analysis of overexpressed S protein corroborated our in vitro finding that GalNAc-T3 and T7 work together to create tandem glycosylations near the multibasic cleavage site of the spike protein (Fig. 1), which protect the site from being processed by cellular proteases.
Because furin cleavage is necessary for the fusion activity of the spike protein of SARS-CoV-2 in Vero E6 cells2, 12, we next tested whether the glycosylation by GalNAc-T3 and T7 could suppress the spike protein-mediated syncytium formation in Vero E6 Cells. As expected, co-expression of the spike protein with GalNAc-T3 and T7 significantly suppressed the syncytium formation, displaying a phenotype resembling annihilation of the furin site (R685A mutation) (Fig. 2e).
In summary, the above evidence indicated that the coordinated glycosylation activity of GalNAc-T3 and T7 towards the multibasic cleavage site of S protein has functional consequence in the cell by inhibiting the furin activation of the spike protein of SARS-CoV-2.
GalNAc-T3 and T7 Inhibit Furin-Dependent Assembly of the Virus Like Particles (VLPs).
As most enzymes involved in O-glycosylation pathway reside in Golgi but the virons of corona viruses are assembled in the ER-Golgi intermediate compartment (ERGIC)46, 47, we wonder whether the glycosylation activity of GalNAc-T3 and T7 would affect furin cleavage of the spike protein in assembled virions. Therefore, we attempted to assemble virus like particles (VLPs) of SARS-CoV-2 in HEK293T with or without KI of GALNT3 and/or GALNT7. We chose to purify VLPs that contain co-expressed spike protein (S), membrane protein (M), envelop protein (E) and nucleocapsid protein (N). Similar VLPs systems have been previously used to mimic the virion of SARS-CoV5, 48, MERS-CoV49, and Ebola virus 50, and recently tested for SARS-CoV-251, 52. We therefore anticipated that the assembled VLPs would adequately report the inhibitory effect of GalNAc-T3 and T7 on the furin processing of the spike protein.
Unexpectedly, however, double KI of GALNT3 and GALNT7 substantially reduced the amount of spike protein packaged into the VLP pellets, despite little change in the expression level of the spike protein in total cell lysates (Fig. 3a). In contrast, single KI of GALNT3 or the universally expressed GALNT1 did not obviously affect the assembly of spike proteins into VLP, whereas single KI of GALNT7 also noticeably decreased the amount of incorporated spikes. We noticed that the spike protein packaged in our VLP samples were almost exclusively cleaved, even though there were abundant uncleaved spike proteins in the remaining cell lysate (Fig. 3a). It seemed to suggest that furin processing is necessary for the assembly of the spike protein into the VLPs of SARS-CoV-2. In agreement with this hypothesis, the R685A spike mutant devoid of furin cleavage site also failed to be incorporated into VLPs (Fig. 3b).
Since the outbreak of COVID-19, a number of groups have extensively studied the function of furin-mediated spike protein priming during SARS-CoV-2 infection9, 14, 37. We wonder why the involvement of furin processing in viron assembly was not previously discovered. One likely explanation is that those studies almost exclusively used pseudotyped virions to simulate the proteases-dependent activation process of SARS-CoV-2. However, previous studies found that the assembly of SARS-CoV virons is dependent on the interaction of spike proteins and membrane proteins48, 53. We speculate similar mechanisms may also exist in SARS-CoV-2 and pseudotyped virions may not accurately reflect the assembly of SARS-CoV-2. Indeed, as demonstrated in Fig. 3c, neither furin site mutation (R685A) nor the double KI of GALNT3 and GALNT7 had any impact on the S protein packaging into the HIV pseudovirus in HEK293T.
We next asked what could be the underlying mechanism of furin-cleavage dependent assembly of the spike protein into SARS-CoV-2 virions. Furin cleavage of the spike protein releases multiple arginine residues (-RRAR) from the multibasic cleavage site and exposes them on the loose end of S1 fragment, which conforms to a C-end rule (CendR) that interacts with host cell receptor neurophilin-154–56. Since the interaction between SARS-CoV S protein and M protein was previously found to be required for viral assembly47, 57, we hypothesized that the furin-cleaved, multibasic S1 terminal sequence could be involved in the interaction with M protein in the case of SARS-CoV-2. Interestingly, the sequence of the SARS-CoV-2 M protein contains a –EE– motif in its N-terminal, extracellular domain, which could potentially mediate charge-charge interactions with the cleaved S1 (Fig. 3d). We therefore tested if the acidic nature of the –EE– motif is required for the assembly of SARS-CoV-2 VLPs by alanine mutations. Notably, the –EE– motif mutant of the M protein failed to incorporate the intact sequence of the spike protein into VLP, mirroring the phenotype produced by R685A mutant of the spike protein with intact M protein (Fig. 3e).
In summary, the above results indicated that furin cleavage not only activates the S protein but also facilitates the assembly of the virion. Therefore inhibition of furin cleavage by host cell expression of GalNAc-T3 and T7 could have a significant impact on SARS-CoV-2 infection. Accordingly, the virus could in theory escape the host cell regulation through mutations affecting glycosylations by GalNAc-T3 and T7. We checked this hypothesis as shown below.
Mutations near the Multibasic Cleavage Site Alter Glycosylation Efficiency of GalNAc-T3 and T7.
As demonstrated in Fig. 4a, three out of the five WHO recognized Variants of Concern (VOC) carry a mutation at proline 681 (P681), three amino acids away from the glycosylation site T678. Among them, the alpha and omicron variants have a histidine (P681H) while the delta variant has an arginine (P681R) at this position. In addition, the omicron variant carries N679K mutation right next to the O-glycosylation site T678.
Due to its close proximity, we speculate that P681 mutations could affect the enzymatic activities of GalNAc-T3 and T7 in vitro. Indeed, as we swapped the P681 residue in our synthetic peptide substrate encompassing the multibasic cleavage site sequence with histidine (P681H), GalNAc-T3 and T7 could not efficiently glycosylate the peptide (Fig. 4b).
However, when the P681H mutation was stacked with N679K mutation, as is the case in Omicron, we surprisingly discovered that GalNAc-T3 and T7 regained significant activity toward the synthetic peptide in the enzymatic assay (Fig. 4b). As we showed that glycosylation by GalNAc-T3 and T7 inhibited furin processing of the spike protein with the native sequence (Wuhan-Hu-1), the in vitro results seemed to suggest that the currently prevalent Omicron, containing both P681H and N679K mutations, might be sensitive to host cell glycosylation, a regulatory mechanism the previous variants managed to escape.
To test this hypothesis, we next compared the effect of P681H and N679K mutations on the furin processing of S protein. As demonstrated in Fig. 4c, although the overexpressed spike protein with native sequence produced reduced amount of S2 in the presence of GalNAc-T3 and T7, the processing of the P681H spike protein was unaffected (S-P681H in Fig. 4c and Supplementary Fig. 9). In contrast, stacked mutations of P681H and N679K made the processing of S protein once again susceptible to the suppression of GalNAc-T3 and T7 (S-P681H/N679K in Fig. 4c). Moreover, when P681 was mutated to histidine, overexpression of GalNAc-T3 and T7 could neither suppress the spike protein-mediated syncytium formation in Vero E6 (S-P681H/GALNT3/T7 in Fig. 4d), nor affect the furin cleavage-facilitated incorporation of the Spike into VLP (S-P681H in Fig. 4e). These results were in line with the inefficient glycosylation of the P681H-substituted peptide by GalNAc-T3 and T7 (Fig. 4b, middle panel). However, when P681H was stacked with N679K, GalNAc-T3 and T7 regained control over those furin processing-dependent spike protein functions (S-P681H/N679K /GALNT3/T7 in Fig. 4d and S-P681H/N679K in Fig. 4e), which again echoed their efficient glycosylation activity on the P681H/N679K peptide substrate (Fig. 4b, bottom panel).
Taken together, our data suggested that mutations carried by SARS-CoV-2 variants near the multibasic cleavage site have dramatic impact on the glycosylation efficiency of GalNAc-T3 and T7. Single mutation at P681 carried by earlier variants (alpha and delta) of SARS-Cov-2 could have evolved to resist host cell glycosylation, albeit the most recent omicron variant reverted this resistance by introducing an additional N679K mutation.
Overexpression of GalNAc-T3 and T7 Suppresses Omicron in Human Lung Cells.
Since the omicron variant of SARS-CoV-2 has become one of the dominant strains globally, we were intrigued to test the susceptibility of the omicron variant to the suppression of GalNAc-T3 and GalNAc-T7 in human lung cells. To evaluate the impact of GalNAc-Ts on viral replication, we inoculated Calu-3 cells with an early omicron subvariant BA.1, the original Wuhan-Hu-1 and an alpha subvariant B1.1.7, and measured the viral titer with or without overexpression of GalNAc-T3 and T7, alone or in combination. As demonstrated in Fig. 5a, following a multiplicity of infection (MOI) of 0.1, the replication of the omicron subvariant BA.1 was significantly suppressed by co-expression of GalNAc-T3 and T7. At 48 hours post infection (HPI), the viral titer of the omicron virus was reduced by ~ 76% in Calu-3 overexpressing GalNAc-T3/T7 (Fig. 5a). Interestingly, expression of GalNAc T7 alone had a similar degree of reduction in the viral titer (Fig. 5c), which was in line with the dominant role of GalNAc-T7 in inhibiting the assembly of the spike protein into VLP (Fig. 3a).
To examine the inhibition effect of GalNAc-T3 and T7 on S protein processing and subsequent virion assembly, we next pelleted the omicron virions and analyzed both S and N proteins with western blot. Notably, either co-expression of GalNAc-T3 and T7 together or overexpression of GalNAc-T7 alone significantly reduced the amount of S protein incorporated into pelleted Omicron virions (Fig. 5b and Fig. 5d). This result mirrored the suppression of the assembly of S protein into VLP by GalNAc-T3 and T7 (Fig. 3a and Fig. 4e). Together with the in vitro glycosylation studies, this suggested that the suppression of Omicron by GalNAc-T3 and T7 could be a result of glycosylation towards the furin processing site of the Spike.
Surprisingly, however, the original Wuhan-Hu-1 virus appeared to less influenced by overexpression of GalNAc-T3 and T7, whereas the replication of the alpha variant (B1.1.7), containing P681H, was similarly suppressed by GalNAc-T3 and T7 as the omicron variant (Supplementary Fig. 10 and Fig. 11). Because GalNAc-T3 and T7 demonstrated the most enzymatic activity towards the native sequence from the Wuhan-Hu-1 virus and the least enzymatic activity towards the sequence harboring P681H only mutation in vitro (Fig. 4b), the results with authentic Wuhan-Hu-1 virus and alpha variant could suggest that Wuhan-Hu-1 was less dependent on furin activation than the newer variants. Alternatively, the many other mutations besides P681 and N679 of S protein in SARS-CoV-2 variants could alter the furin processing as well as the glycosylation in various ways, thus complicate the overall inhibition by GalNAc-T3 and T758, 59. Unfortunately, restricted by local law regulation, we cannot conduct reverse-engineering studies to further test those hypotheses. Nevertheless, it is clear that the glycosylation by GalNAc-T3 and T7 plays an important role in the regulation of viral infection and the current prevalent SARS-CoV-2 variant Omicron remain responsive to this regulation. Glycosylation at the multibasic site of the spike protein could be the exploited for future therapeutic strategies.