The protein sequence modulates terminal GalNAc incorporation during N-glycan biosynthesis

Abstract

The factors governing protein-specific N-glycosylation remain incompletely understood despite extensive knowledge of the biosynthetic machinery involved in glycan assembly. In particular, the contribution of the protein sequence itself to the final glycosylation pattern has received comparatively little attention. Here, we employed isotope-assisted NMR spectroscopy to investigate the N-glycosylation of the receptor-binding domain of the SARS-CoV-2 spike glycoprotein (RBD-SCoV2) produced in HEK293F cells. Previous analyses revealed the unexpected presence of terminal GalNAc-containing epitopes, including LacdiNAc (LDN), sulfated LacdiNAc (4S-LDN), sialylated LacdiNAc (6Sia-LDN), and fucosylated LacdiNAc (LDNF). Inspired by earlier studies showing that certain β4GalNAc-transferases recognize basic peptide motifs, we generated an RBD variant carrying four substitutions of solvent-exposed basic residues located near the N-glycosylation sites. Comparative NMR analysis demonstrated that the mutant protein exhibits a marked reduction in GalNAc-containing terminal epitopes, accompanied by a corresponding increase in Gal-containing structures. The overall protein fold remained unchanged, indicating that the observed glycosylation differences arise from altered glycan processing rather than major structural perturbations. These findings provide direct evidence that local protein sequence features modulate the incorporation of terminal glycan epitopes during Golgi processing. More broadly, this work highlights the power of quantitative NMR spectroscopy for the structural characterization of glycoproteins and for detecting subtle changes in glycan populations that are directly linked to protein sequence.

Introduction

Protein N-glycosylation is one of the most abundant and functionally important post-translational modifications in eukaryotes. N-glycans influence protein folding, stability, trafficking, serum half-life, and molecular recognition processes, thereby affecting numerous physiological and pathological events (He et al. 2024). Although the biosynthetic pathway responsible for N-glycan assembly has been extensively characterized (Stanley et al. 2022) the factors that determine the final glycosylation pattern of a particular glycoprotein remain incompletely understood.

N-glycan biosynthesis begins in the endoplasmic reticulum (ER) (Aebi 2013), where enzymatic processing generates an unfolded nascent glycoprotein carrying a largely homogeneous population of high-mannose glycans that participate in protein folding and ER quality control. Once the protein has folded properly and is transported to the Golgi apparatus, these homogeneous glycans undergo extensive remodeling by glycosidases and glycosyltransferases (GTs), generating a diverse repertoire of complex structures (Stanley 2011). Thus, whereas ER-resident enzymes act on unfolded protein, Golgi-resident enzymes act on folded proteins. Moreover, unlike glycan processing in the ER, Golgi glycosylation involves multiple competing enzymatic pathways whose outcome depends on factors such as enzyme expression and localization, donor substrate availability, and cellular metabolic state. However, these factors alone do not fully account for the glycan heterogeneity observed among glycoproteins produced in the same cellular environment.

Increasing evidence indicates that the protein substrate itself contributes to shaping glycosylation outcomes. Site accessibility (Lee et al. 2014) (Quintana et al. 2024) and specific amino-acid motifs (Huang et al. 2017) (Ballard et al. 2025) (Saito et al. 2022) (Shakin-Eshleman et al. 1996) have been proposed to influence the action of Golgi GTs. A notable example is provided by the β1,4-N-acetylgalactosaminyltransferases β4GalNAc-T3 (Sato et al. 2003) (Tokoro et al. 2024) and β4GalNAc-T4 (Gotoh et al. 2004) which generate LacdiNAc (GalNAcβ1–4GlcNAc, LDN) motifs and have been shown to recognize peptide determinants containing clusters of basic residues in selected glycoproteins (Dharmesh et al. 1993) (Miller et al. 2008) (Fiete et al. 2012). Nevertheless, direct experimental evidence demonstrating that subtle modifications in protein sequence can alter terminal glycan epitopes remains limited.

The receptor-binding domain of the SARS-CoV-2 spike glycoprotein (RBD-SCoV-2) constitutes an attractive model to investigate this question. Previous NMR studies of isotopically labelled RBD-SCoV-2 produced in HEK293F cells revealed the unexpected presence of high levels of terminal GalNAc-containing epitopes, including LDN and its sulfated, sialylated and fucosylated derivatives (Lenza et al. 2020). Here, taking advantage of the quantitative nature of NMR spectroscopy, we examine the effect of mutating four solvent-exposed basic residues located near the N-glycosylation sites of the RBD-SCoV-2. Our results demonstrate that these substitutions markedly reduce the incorporation of terminal GalNAc-containing epitopes, providing direct evidence that local protein sequence influences glycan processing during N-glycan biosynthesis.

Results

We previously described the NMR-based characterization of the glycans of RBD-SCoV2 produced in HEK293F cells cultured in the presence of ¹³C-glucose, enabling selective ¹³C enrichment of the glycan moieties for structural analysis by NMR spectroscopy (Lenza et al. 2020). One of the most unexpected observations in that study was the presence of GalNAc containing terminal epitopes, in particular, LDN, its 4-sulfated (4S-LDN) and 6-sialylated (6Sia-LDN) versions, and its fucosylated derivative LDNF, which could be unambiguously identified through 3D-NMR experiments. In the ¹H,¹³C-HSQC-NMR anomeric fingerprint of the glycoprotein (Fig. 1), the anomeric correlations of the GalNAc residues in the 4S-LDN, 6Sia-LDN and LDNF epitopes appear at relatively isolated positions, whereas that of LDN exhibits partial overlap with other resonances. In addition, the H5/C5 and H6/C6 correlations of the fucose residue in the LDNF epitope, which resonate in well-isolated spectral regions (Aeschbacher et al. 2017), have been previously assigned and serve as readily identifiable markers of this epitope. (see below).

Figure 1.

¹H,¹³C-HSQC spectrum of RBD-SCoV2 with ¹³C-enrichment on glycans. Anomeric region annotated with cross-peak assignments derived from the 3D NMR assignments reported in reference (Lenza et al. 2020).

GalNAc containing epitopes are rarely found in N-glycans of human glycoproteins (Hirano and Furukawa 2022) (Tokoro et al. 2024). Their occurrence has been reported in only a small subset of N-glycosylated proteins and associated with specialized biological functions. Importantly, their presence on glycoproteins has been associated with fast clearance, mediated by their recognition by the lectins Asialoglycoprotein Receptor (ASGP-R, which bind terminal GalNAc) and Mannose Receptor (MR, which bind sulfated GalNAc). Although some studies have reported high levels of LDN on HEK293 cells (Uhler et al. 2021), our results on the characterization of several glycoproteins overexpressed in this system do not reflect this. In fact, the comparison of the ¹H,¹³C-HSQC-NMR anomeric fingerprint across different glycoproteins produced in our laboratory under the same conditions (Unione et al. 2019) (Lenza et al. 2020) (Quintana et al. 2024) highlights the exceptionality of these epitopes on the small RBD-SCoV2 glycoprotein fragment. This observation suggests that specific features of the RBD protein scaffold might favor the incorporation of terminal GalNAc residues during glycan processing.

LDN and 4S-LDN glycan epitopes were initially discovered in the N-glycans of a pituitary hormone (Green et al. 1985). Subsequent work demonstrated that (Miller et al. 2008) (Fiete et al. 2012) the activity of the two enzymes responsible for the incorporation of GalNAc on N-glycoproteins, the GTs β4GalNAc-T3 and β4GalNAc-T4, depend critically on the presence of specific peptide motifs. In particular, the presence of a cluster of basic residues close to the oligosaccharide structure was needed for these enzymes to recognize the glycoprotein, and essential for catalysis. Building on these precedents, we inspected the amino-acid sequence of RBD-SCoV2. The analysis revealed the presence of several positively charged residues, Arginines and Lysines, in the vicinity of both N-glycosylation sites, N331 and N343 (Fig. 2). Whereas some participate in intramolecular interactions that are likely important for maintaining the protein fold and were therefore excluded from mutagenesis, others, such as residues K356, R357, K386 and K529, are solvent-exposed and form positively charged patches adjacent to the glycosylation sites. These residues were therefore replaced with alanine to assess their potential role in modulating glycosylation, under the assumption that the mutations would not significantly perturb the overall protein fold.

Figure 2.

Amino acid sequence of RBD-SCoV2. N-glycosylation sites (N331 and N343) are highlighted in blue, and basic residues mutated to alanine in the mutant (K356, R357, K386 and K529) are shown in magenta. The 3D models below are based on crystal structure pdb 6xr8. Left: N-glycosylation sites and mutated residues are indicated. Right: The same structure is shown as an electrostatic surface representation, with annotations marking the N-glycosylation sites.

The tetramutant RBD-SCoV2 (RBD-SCoV2_mut) was thus produced exactly under the same conditions as the WT version. Size Exclusion Chromatography showed that RBD-SCoV2_mut eluted at the same time as the WT version (Fig. S1), suggesting a similar fold. This was further confirmed by comparing the 1D ¹H-NMR spectra of both proteins, which showed the same profile (Fig. S2).

Next, the ¹H,¹³C-HSQC spectra of WT and tetramutant proteins were compared. The overall spectra were very similar. However, careful inspection revealed clear differences in the relative intensities of some cross peaks. Figure 3 shows the anomeric region of the ¹H,¹³C-HSQC spectra of RBD-SCoV2 and RBD-SCoV2_mut. Both spectra were acquired and processed with the same parameters. Considering the anomeric signal of the core βMan as a reference (that should be the same amount in both samples), the anomeric correlations corresponding to GalNAc containing epitopes were less intense in RBD-SCoV2_mut than in the WT. This effect was readily apparent for the LDNF, 4S-LDN, and 6Sia-LDN epitopes. For LDN, however, the difference was less evident, although partial overlap with other resonances precludes a detailed analysis. In the particular case of LDNF, this intensity reduction was further demonstrated by comparing the signals of H5/C5 and H6/C6 of Fucose in LDN and LeX (Aeschbacher et al. 2017). These epitopes differ exclusively in the identity of the terminal monosaccharide, with GalNAc present in LDNF and Gal present in LeX. This difference gives rise to distinct H5/C5 and H6/C6 correlations in the ¹H,¹³C-HSQC-NMR spectra, enabling their unambiguous discrimination. Figure 4 clearly shows that the population of these epitopes is basically inverted in WT versus mutant proteins, highlighting the reduced presence of LDNF in the RBD-SCoV2_mut. Interestingly, at the same time, the intensities of signals corresponding to Gal containing epitopes were more intense in the RBD-SCoV2_mut compared to WT. This observation indicates that the enzymatic incorporation of Gal and GalNAc are competing processes.

Figure 3.

Anomeric region of the ¹H,¹³C-HSQC spectra of RBD-SCoV2 and RBD-SCoV2_mut. The signal of the core βMan residue, used as a reference is indicated, along with the assignment of the signals associated with GalNAc-containing epitopes.

Figure 4.

¹H,¹³C-HSQC spectra of RBD-SCoV2 and RBD-SCoV2_mut showing the H5/C5 and H6/C6 correlations of Fuc in LDNF and LeX epitopes.

Discussion

The mechanisms governing protein glycosylation remain poorly understood. In particular, how protein structure constrains and directs glycan biosynthesis remains largely unclear, despite growing evidence supporting this connection (Kellman et al. 2024).

Comprehensive characterization of protein glycosylation remains analytically challenging. Mass spectrometry (MS) is the method of choice for glycosylation profiling and has transformed glycoscience through high throughput glycoproteomic analyses. However, limitations remain, including the need for sample manipulation and the limited ability of standard workflows to unambiguously define glycosidic linkages and characterize highly complex glycan structures.

The development of metabolic labeling strategies for incorporating NMR-active isotopes into glycans (Kato et al. 2010) (Barb et al. 2019) (Rogals et al. 2021) has established NMR spectroscopy as a powerful complementary tool for glycoprotein characterization (Jiménez-Barbero and Millet 2025). NMR provides direct structural information on glycan epitopes and core architectures, including the unambiguous assignment of glycosidic linkages. Moreover, NMR is a quantitative technique because, under appropriate acquisition conditions, signal intensities are proportional to the concentration of the corresponding nuclei in the sample. Nevertheless, for protein-bound glycans, differences in molecular dynamics can lead to differential relaxation, which in turn modulates signal intensities, as previously reported (Barb et al 2012) (Unione et al. 2019). Consequently, denaturing conditions are generally more suitable for rigorous quantitative analyses (Schubert et al. 2015). In the protein studied here, however, the N-glycosylation sites are located at the N-terminus and, according to molecular modeling (Lenza et al. 2020), are highly dynamic, minimizing relaxation-induced intensity biases, especially when comparing terminal glycan motifs, as performed in this study, and thereby enabling reliable comparisons of their relative abundance. Additionally, because the mutant and wild-type proteins adopt the same fold, the observed glycan-associated NMR intensity changes are unlikely to result from major alterations in the protein structure. Importantly, once these considerations are taken into account, NMR is particularly well suited to detecting subtle changes in glycosylation associated with variations in protein sequence.

Here, the quantitative nature of NMR has allowed to reveal significant changes in glycan epitope populations following the substitution of only four amino acids in the glycoprotein sequence, providing direct evidence that local sequence features influence glycan processing. These observations support the existence of protein-encoded determinants of glycosylation and highlight the need to consider the protein substrate, together with the cellular glycosylation machinery, as an active contributor to glycan biosynthesis.

Conclusions

Using isotopically enriched glycoproteins and quantitative NMR spectroscopy, we demonstrate that the amino-acid sequence of a glycoprotein can influence the terminal glycan epitopes incorporated during Golgi processing. Replacement of four solvent-exposed basic residues in the SARS-CoV-2 RBD led to a clear reduction in GalNAc-containing motifs, including LDNF, 4S-LDN and 6Sia-LDN, while concomitantly increasing the abundance of Gal-containing epitopes.

These observations are consistent with previous reports showing that certain β4GalNAc-transferases recognize specific peptide determinants and support the concept that local protein sequence contributes to shaping glycan biosynthesis. The results further suggest that incorporation of terminal GalNAc and Gal residues represents a competitive process whose outcome can be modulated by the protein scaffold itself.

More generally, these findings support the emerging view that glycan biosynthesis is influenced not only by the cellular glycosylation machinery but also by information encoded within the protein substrate itself. Finally, the study highlights the unique value of NMR spectroscopy as a quantitative and structurally informative approach for investigating glycoprotein glycosylation and uncovering subtle sequence-dependent effects that may otherwise remain unnoticed.

Materials and methods

Protein expression and purification of RBD-ScoV2 and RBD-ScoV2-Mut

The coding sequence of the receptor binding domain (RBD) of the Spike glycoprotein of SARS-CoV-2 virus (amino acid residues 328–533, RBD-SCoV2) and the R356A, K357A, R386A and R528A tetramutant (RBD-SCoV2-mut) were synthesized by Genscript Biotech by insertion into the Agel and Kpnl restriction sites of pHLsec vector (Aricescu et al. 2006). 6× His tag was added to the C-terminus of the protein sequence for purification purpose. The proteins were produced by transient transfection of HEK293F suspension cells (Thermo Fisher Scientific) at 0.8–1.0·10⁶ cells/ml using FectoPRO (Polyplus Transfections) in a plasmid/transfection agent ratio of 1:1. Cells were grown in FreeStyle media supplemented with 3 g/L of uniformly ¹³C6-glucose at 37 °C, 180 rpm, 8% CO₂ in a Minitron Pro shaker (Infors HT) incubator during 6 days. Cells were harvested by centrifugation at 10.000 rpm at room temperature and the supernatant (containing the soluble protein) was filtered through a 0.22 μm Polysterene filter (Corning). The resulting solution was loaded into a HisTrap Ni-NTA column (GE Healthcare) and the protein was eluted using a step-gradient solution of Imidazole (from 0 to 500 mM) in phosphate saline buffer (PBS 1×) pH 7.4. Next, the fractions containing either the RBD-SCoV2 or the RBD-SCoV2_mut proteins were concentrated using 10 KDa cut-off Amicon concentrators (Merck Millipore) and further purified via size exclusion chromatography using a Superdex 75 Increase 26/600 column (GE Healthcare). Both the RBD-SCoV2 and the RBD-SCoV2-mut proteins eluted at the same retention time (Fig. S2), suggesting a similar folding.

NMR sample preparation and analysis

RBD-SCoV2 and of the RBD-SCoV2-mut proteins were buffer exchanged to deuterated phosphate saline buffer (PBS 1×) pD 7.4 and concentrated to 250 μM protein concentration in 300 μL. The resulting protein solutions were transferred to 5 mm Shigemi NMR tubes. 1D ¹H- and 2D ¹H,¹³C-HSQC-NMR spectra were acquired at 310 K on an 800 MHz BRUKER AVANCE III spectrometer, equipped with a TCI cryo-probe with z-gradient coil and TopSpin 3.2.7 (BRUKER) software was employed for data acquisition and processing.

1D ¹H-NMR spectra were acquired with 128 scans and 32 K points, 16.0 ppm of spectral width, which resulted in 0.8 Hz fid resolution (Fig. S2). Residual H₂O NMR-signal suppression was achieved using excitation sculpting with gradients.

High-resolution 2D ¹H,¹³C-HSQC spectra were acquired as previously described (Lenza et al. 2020) with an in-house coded pulse sequence using States-TPPI sampled constant t1(¹³C) time evolution during 22 ms and 47 Hz F1(¹³C) digital resolution, 3.7 ms ¹J_CH evolution and d1 = 1 s total interscan delay; this spectrum encodes signals with an odd or even number of ¹³C neighbors (excluding the decoupled ¹³CO groups) with a negative or positive sign, respectively. Spectra were acquired with 16 scans and 1536 points in the F2 dimension and 984 points in the F1 dimension, and with a digital resolution of 14.5 Hz and 51.6 Hz in F2 and F1 dimensions, respectively.

Author contributions

Luca Unione (Conceptualization [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Writing—review & editing [equal]), Sandra Delgado (Methodology [equal]), Jesus Jimenez-Barbero (Funding acquisition [equal], Resources [equal], Writing—review & editing [equal]), Ana Arda (Conceptualization [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Supervision [equal], Writing—original draft [lead]).

Conflict of interest

All authors declare no conflicts of interest.

Funding

None declared.

Data availability

The authors declare that the data supporting the findings of this study are available within the article and its supporting information file. Additional data can be provided from the corresponding author upon request.