Computational analysis reveals non-consensus N-glycosylation sequons in antibody Fab region

ABSTRACT

Protein glycosylation at asparagine typically occurs at a consensus motif. However, recent studies have reported instances of N-glycosylation at non-consensus sites, though the mechanisms and implications of these atypical modifications remain unclear. In this study, we identified novel non-consensus N-glycosylation motifs with low glycosylation occupancy in the Fab region of human antibodies. We developed a computational workflow to predict the interaction between non-consensus peptides and the eukaryotic oligosaccharyltransferase (OST) complex. This model was validated through site-directed mutagenesis around the asparagine residue and glycosylation quantification via mass spectrometry. Our results show that glycan occupancy at non-consensus sites can be modulated by mutations that influence OST binding affinity. Pharmacological inhibition of OST activity reduced non-consensus and consensus glycosylation in both Fab and Fc regions. Additionally, we identified new non-consensus glycosylation sites in natural human antibodies, revealing the sequence preferences governing these modifications. These findings provide mechanistic insights into OST sequence specificity and establish a computational and analytical framework for assessing atypical N-glycosylation, aiding glycan profile control in therapeutic antibody development.

Introduction

Asparagine-linked glycosylation (N-glycosylation), a post-translational modification (PTM) in proteins where an asparagine (Asn) residue is covalently attached to an oligosaccharide, plays a crucial role in protein folding, solubility, trafficking, and function. N-glycosylation typically occurs at a consensus sequence, N-X-S/T, where X is any amino acid except proline, followed by a serine or threonine. However, it can also occur at non-consensus sites, albeit less frequently. Several studies have reported unique non-consensus glycosylation motifs, including N-X-C, N-X-V, and the reversed consensus motif S/T-X-N in various human glycoproteins.^1–3 Glycosylation at non-consensus sites can significantly impact the structure and function of proteins. For example, Sda/Cad synthase (B4GALNT2) contains a conserved glycosylation site at the N-X-C motif within its Golgi-luminal domain. Mutation-induced depletion of this glycosylation site alters Golgi localization, reduces expression levels, and impairs enzymatic activity.² At non-consensus sites, glycosylation is less predictable and can be influenced by factors such as the local protein structure and cellular environment. The underlying mechanisms and effects of non-consensus glycosylation remain elusive.

In the context of antibodies, non-consensus glycosylation can lead to variations in their therapeutic properties, including changes in structure, binding affinity, and pharmacokinetics. Human antibodies are classified into five isotypes (IgM, IgD, IgG, IgA, and IgE) based on the sequence and structure of their heavy chains, with IgG being the most abundant isotype in human plasma. IgG antibodies contain a conserved consensus N-glycosylation site in the Fc region, where glycan variations modulate Fc receptor binding and downstream antibody-dependent cellular cytotoxicity (ADCC).⁴ Glycosylation can also occur in the Fab region, with approximately 10% of human IgG molecules containing Fab glycosylation. This is thought to arise from the incorporation of consensus motifs during somatic hypermutation.⁵ Several non-consensus motifs have been identified in the Fab region of antibodies, including an N-S-G motif in the C_H1 domain of the heavy chain and an N-E-N motif in the complementarity-determining region (CDR) of the light chain.^6–8 Although Fab glycans occur infrequently, they can influence antigen recognition as well as antibody folding and stability. The consensus glycan site within the variable domain of anti-citrullinated protein antibodies has been shown to modulate antigen binding and immune response.⁹ Moreover, lung squamous cell carcinoma-derived IgG contains an N-S-G glycosylation site within the C_H1 domain that is involved in CIgG-integrin-FAK signaling and oncogenic functions.¹⁰ Furthermore, N-glycans can become critical quality attributes in therapeutic antibodies, influencing product safety and efficacy.^11,12 Understanding the patterns and mechanisms of non-consensus glycosylation is essential for optimizing antibody engineering and improving their therapeutic characteristics.

N-glycosylation is initiated in the endoplasmic reticulum (ER) by oligosaccharyltransferase (OST), which catalyzes the transfer of a precursor oligosaccharide from a lipid-linked dolichol-PP donor to the acceptor Asn on the nascent polypeptide chain.¹³ The eukaryotic OST has two distinct forms (OST-A and OST-B) and modifies glycosylation sites co-translationally or post-translationally.^14,15 The oligosaccharide is subsequently processed in the Golgi apparatus, where it undergoes glucose residue trimming by glucosidases and the addition of other sugar moieties by glycosyltransferases.¹⁶

Eukaryotic OST is a multi-subunit protein complex that contains multiple domains responsible for transmembrane localization, ligand binding, and catalysis. A recent cryo-EM structure revealed the binding of a ligand peptide within the STT3B catalytic subunit of human OST-B complex.¹⁷ A proposed enzymatic mechanism suggests that an aspartic acid (Asp) residue in the active site of STT3 forms a hydrogen bond with the acceptor Asn, facilitating the nucleophilic attack of the oligosaccharide. This Asp residue is highly conserved across species (Asp103 in human STT3B, Asp56 in bacterial OST, and Asp47 in yeast STT3).^17–21 A separate Trp-Trp-Asp (WWD) loop interacts with the threonine at the +2 position of the consensus sequon (N-A-T), aiding in ligand binding and positioning.¹⁷ Several studies have demonstrated that OST has a higher binding affinity for N-X-T motifs compared to N-X-S motifs.²² The amino acids at the −2, −1, +1, and +3 positions influence glycosylation efficiency and occupancy, as observed in bacterial OST-catalyzed glycosylation.^23,24 Another study revealed that introducing an aromatic residue at the −2 position of the consensus sequence increases glycan occupancy in human cells.²⁵ These findings indicate that the glycan occupancy at a specific sequence is associated with OST binding specificity and affinity.

In this study, we investigated how amino acid sequence influences enzyme binding and glycosylation efficiency at non-consensus N-glycosylation sites. Using a non-consensus motif found in the Fab region of a human IgG1 antibody (mAb-1) as a template, we performed computational docking and residue scanning to predict peptide binding with the human OST-B complex. We then generated mutant antibodies to examine the impact of sequence variation on glycan occupancy. Our results indicate that enhanced fit within the OST active site promotes glycosylation at non-consensus sites. Notably, introducing an aromatic residue at the −1 position and an Asn at the +2 position significantly increased glycosylation. Analysis of natural antibody sequences revealed a similar pattern, reinforcing our understanding of non-consensus glycosylation mechanisms. Although the glycan occupancy at consensus sites were much higher, the non-consensus motifs may still qualify as critical quality attributes in therapeutic molecules and require careful evaluation and manufacturing control. This study establishes a predictive computational workflow for assessing glycan occupancy at non-consensus motifs, offering valuable insights for antibody sequence design and therapeutic optimization.

Results

Peptide docking with the OST complex

Due to the essential role of oligosaccharyltransferase (OST) in initiating glycan transfer, we investigated the binding of non-consensus sequences with eukaryotic OST. We obtained a published OST-B structure with a reference peptide bound in the STT3B catalytic subunit from the Protein Database (PDB: 6S7T).¹⁷ The reference ligand is a 7-amino acid peptide containing a consensus N-glycosylation sequon (AANATAA). Using the Molecular Operating Environment (MOE, v2024.06) platform, we analyzed the docking score, binding affinity, and thermostability of various non-consensus sequences to evaluate OST sequence preference.

First, we docked the consensus sequence (QYNSTYR) found in the Fc region of the mAb-1 molecule as a control. Due to the high conformational flexibility of peptides, we performed a conformational search within a defined energy window to generate a conformation library for the peptide.^26,27 Molecular docking was then performed using existing peptide conformations and using the triangular matcher placement and the rigid receptor refinement methods. This process generated 200 poses ranked by their docking scores.²⁸ We selected the pose that had the most negative docking score and formed hydrogen bonding between Asn and the catalytic Asp103, which facilitates nucleophilic attack during glycan transfer.¹⁸ The 2D-interaction revealed that the Thr formed hydrogen bond with the receptor Trp604, consistent with previous studies showing that the WWD loop in STT3B facilitates ligand binding (Figure 1(A)).^19,20

Figure 1.

Docking conformation of 7 amino acid-long peptide in STT3B active site. The docking conformations were shown for A. consensus sequence (QYNSTYR) and B. non-consensus sequence (XTNYGXX). The 2D-interaction between the peptide and the receptor was shown on the right. The arrow is pointing from the H-bond donor to acceptor.

Next, we docked a non-consensus motif (N-Y-G) identified in the Fab region of mAb-1. The docking of the non-consensus peptide resulted in a negative docking sore of −10.1332 kcal/mol (Figure 1(B)). In comparison with the consensus control sequence that had a docking score of −12.1643 kcal/mol, the non-consensus sequence showed slightly reduced binding affinity to the OST complex. However, the Asn residue within the N-Y-G motif formed strong interactions with Asp103, and the backbone nitrogen of Gly formed hydrogen bond with Asp606 in the WWD loop. These results suggest that the N-Y-G motif can fit into the STT3B catalytic site that facilitates N-glycosylation.

Evaluating sequence variation on OST binding

To further evaluate OST’s sequence preference for recognizing non-consensus sequons, we used the residue scan module to computationally generate point mutations around the Asn residue in the non-consensus peptide and calculated ligand properties including binding affinity and thermostability. This high-throughput analysis served as an orthogonal approach to docking analysis, enabling the identification of stable, high-affinity peptides. Each residue at the −2, −1, +1, and +2 positions was individually substituted with the remaining 19 natural amino acids. Heatmaps were generated to display relative binding affinity (ΔAffinity) and thermostability (ΔStability) compared to the wildtype peptide (Figure 2(A), Supplementary Figure S1). Negative ΔAffinity and ΔStability values indicate stronger binding affinity and increased peptide stability.

Figure 2.

Sequence preference prediction using residue scanning and peptide docking. A. Substitution with 19 natural amino acids was introduced at the −1, +1, and +2 position around the Asn in the non-consensus peptide (XTNYGXX). The relative binding affinity, thermostability, and docking score of the mutant peptide compared to the wildtype peptide were presented as ΔAffinity, ΔStability, and ΔS in the heatmap. The unit is kcal/mol. B. The relative docking score (ΔS) versus relative binding affinity and thermostability were plotted. Each dot represents a mutant peptide. n = 57.

Substitution of Thr and Gly at the −1 and +2 positions substantially improved the binding affinity and stability, while substitution of Phe at the −2 position decreased the peptide properties (Figure 2(A), Supplementary Figure S1). Docking analysis of mutant peptides revealed a positive correlation between docking score (ΔS) and peptide properties (ΔAffinity and ΔStability) (Figure 2(B)). Fourteen mutations were selected for in vitro antibody generation and glycosylation analysis. Notably, mutants such as T30F, G33N, T30R, and T30Y were predicted to exhibit stronger interactions with STT3B, supported by their negative ΔS, ΔAffinity, and ΔStability values. Overall, this computational workflow provides a predictive model for evaluating OST binding to non-consensus sequences.

Characterization of glycan occupancy at non-consensus site

To evaluate low-abundant glycosylation at non-consensus sites, we developed a deglycosylation method using PNGase F followed by trypsin digestion and mass spectrometry based-peptide mapping. PNGase F digestion removes N-glycan and leaves a deamidation modification on Asn with a mass shift of +0.988 Da. Deglycosylation using PNGase F is a widely used approach for glycoprotein analysis.^29,30 Due to the limited surface exposure of Fab glycan, we assessed PNGase F efficiency when added at different stages of sample preparation (Supplementary Figure S2A).³¹ Deamidation levels were quantified by mass spectrometry, and glycan occupancy was determined by the difference in deamidation in samples with or without PNGase F treatment. We found that the addition of PNGase F after disulfide bond reduction and alkylation effectively removed N-glycans in both Fab and Fc regions (Supplementary Figure S2B). We further evaluated the duration of PNGase F treatment on peptide deamidation and glycosylation quantification. We found no significant deamidation growth after longer treatment time, which suggests that this method can effectively detect the glycosylation site without introducing artificial deamidation during sample preparation (Supplementary Figure S2C).

To further confirm glycosylation in the Fab region of mAb-1, we performed size exclusion chromatography (SEC) and intact mass analysis. After IdeS digestion and reduction, the Fc, Fd, and light-chain fragments of mAb-1 were separated by SEC. A shoulder peak was observed adjacent to the Fd main peak, suggesting potential glycan modifications in the Fd region (Suplementary Figure S2D). We isolated this shoulder peak fraction and identified species with higher molecular weight by intact mass analysis (Suplementary Figure S2E). Peptide mapping further confirmed the presence of glycans containing sialic acid on the heavy chain N-Y-G peptide (Supplementary Figure S2F). Together, these results confirmed the glycosylation site within the Fab region of mAb-1.

Modulating glycan occupancy at non-consensus sites by mutagenesis

To validate the computational predictions of non-consensus sequons found in Figure 2, we introduced single-point mutations around the Asn residue at the Fab glycosylation site and quantified N-glycan occupancy in mutant antibodies. Both wildtype and mutant antibodies were expressed in ExpiCHO cells and purified using Protein A column (Figure 3(A)). In total, 16 antibodies were generated, and glycosylation levels were assessed using PNGase F-mediated deglycosylation coupled with peptide mapping analysis (Figure 3(B)). A positive control mutant (G33T) that introduced a threonine at the +2 position to create a consensus sequon showed high glycosylation (99.1%) in Fab region, suggesting that this Fab glycosylation site is accessible to glycan modification. In contrast, the negative control mutant (N31A) showed no detectable glycosylation. Among the remaining variants, 7 exhibited increased Fab glycosylation, with T30F and G33N showing statistically significant increases compared to the wildtype. The Fc glycosylation remained unaffected by mutations in the Fab region (Figure 3(B)).

Figure 3.

Modulating glycan occupancy at non-consensus sites by mutagenesis. A. the workflow of antibody expression, purification, and characterization. B. the levels of glycan occupancy in Fab and in Fc measured by mass spectrometry. Data is shown as mean ± SD of 2 technical replicates. One-way ANOVA: p < 0.0021 (**), p < 0.0001 (****). C. the docking score was compared between the Fab glycosylated mutants (n = 7) and non-glycosylated mutants (n = 7). Data is shown as mean ± SD. Unpaired t-test: p < 0.0021 (**). D.Cells were treated with 5 µM NGI-1, 20 µM NGI-1, or DMSO for 14 days post transfection of G33N and G33T mutants. Glycosylation in antibody was determined through mass spectrometry. Data is shown as mean ± SD of 2 technical replicates. Two-way ANOVA: p < 0.0001 (****).

We further evaluated the docking profiles of the mutant sequences. We found that the glycosylated variants exhibited more negative docking score compared to non-glycosylated variants, suggesting stronger binding to STT3B (Figure 3(C)). This strongly supported our computational analysis and suggested that the docking score may be used to predict glycosylation frequency in non-consensus sequons (Figure 2). Specifically, in the G33N mutant, the introduced Asn at the +2 position formed hydrogen bonds with the WWD loop, mimicking interactions seen in consensus sequons. In the T30F mutant, the Phe residue at −1 position formed arene-H interaction with Glu101, enhancing receptor binding (Suplementary Figure S3).

To investigate the dependency on OST complex during non-consensus glycosylation, we treated cells with an OST complex inhibitor, NGI-1, during antibody production.³² In the G33N mutant, the non-consensus glycosylation in the Fab region was abolished by NGI-1 treatment, while the consensus glycosylation in the Fc region was reduced by NGI-1 treatment in a dose-dependent manner (Figure 3(D)). In the G33T mutant, both Fab and Fc glycosylation at consensus sites were reduced by NGI-1 treatment, with Fc glycan showing stronger reduction. Together, these findings support that glycosylation at non-consensus site is sequence-dependent and influenced by interactions with the OST complex.

Fab and Fc regions contain distinct N-glycan structures

To investigate differences between the Fab and Fc glycans, we analyzed the glycan structures and their abundancies in the G33T and G33N mutants using mass spectrometry. In the G33T mutant that showed comparable glycan occupancy at the Fab and Fc sites (99.1% and 96.3%, respectively), we detected 10 Fab-specific glycans, 1 Fc-specific glycan, and 9 shared species, suggesting higher glycan diversity in the Fab region (Figure 4, Supplementary data S1). Notably, common biantennary glycans including G1F and G2F were significantly enriched in the Fab region, while the Fc region was enriched by G0F and high mannose species such as Man5.

Figure 4.

Data is shown as mean ± SD of 3 technical replicates. Multiple unpaired t-test using false discovery rate approach: p < 0.05 (*).

In the G33N mutant, due to the low glycan occupancy at the Fab glycosylation site, some glycan species were not detectable (Supplementary Figure S4, Supplementary data S1). We found higher level of G1F in the Fab region compared to Fc, consistent with the G33T glycan profile. These data suggest that Fab glycans have a distinct glycoform distribution from Fc, likely due to the local protein structure that influence the accessibility of different glycan processing enzymes.

Profiling naturally occurring non-consensus sequons in natural antibodies

To further evaluate sequence preference for non-consensus N-glycosylation in natural antibodies, we analyzed human intravenous immunoglobulin (IVIg), which contains IgG from a large pool of healthy donors and is widely used in immunomodulatory and anti-inflammatory therapy.^33,34 We first enriched Fab-glycosylated antibodies using Sambucus nigra lectin (SNA)-binding chromatography.^5,35 SNA binds to sialic acids that were found in approximately 15% of total N-glycans in IVIg, particularly in the Fab region.³⁶ We found that 9.88% of IVIg was enriched by SNA-binding (Figure 5(B)). Due to the lack of sequence information for IVIg, we performed peptide mapping on enriched antibodies and analyzed glycopeptide sequences using PEAKS DeepNovo (Figure 5(A)), which is a deep learning-based tool for de novo peptide sequencing and PTM analysis that does not require reference databases.^37,38 PNGase F digestion was used to remove N-glycans, allowing for the detection of deamidation modifications and simplifying the de novo sequencing process. Identified sequences were further validated through glycan analysis and database searches.

Figure 5.

Profiling glycosylation sites in human IVIg. A. The workflow summarizing the enrichment of Fab glycosylated antibody in human IVIg and deep learning-based glycopeptide identification. B. 5 mg IVIg was incubated with 1.5 mg SNA agarose beads. The percentage of antibody relative to total antibody amount in the bound and unbound fraction is shown. Data is shown as mean ± SD of 3 biological replicates. Unpaired t-test: p < 0.0001 (****). C. The docking score was compared between the non-consensus sequences (n = 10) and consensus sequences (n = 7). Data is shown as mean ± SD. Unpaired t-test: p < 0.0001 (****). The docking score (S) versus binding affinity is plotted. Each dot represents a peptide. n = 17. Linear regression: y = 3.863 * x − 22.43, R² = 0.3118.

This workflow resulted in the identification of 10 non-consensus sequences and 7 consensus glycosylation sequences (Table 1). We were able to detect previously reported glycopeptides such as SWNSGAL, SGNSQES, and QYNSTYR, confirming the reliability of this method.^6,7 In addition, we discovered new glycosylation motifs in the heavy and light chain variable domains that were not previously reported. We validated these sequences in human protein database (PeptideAtlas) and performed docking analysis.³⁹ We found that the consensus peptides exhibited significantly more negative docking scores than the non-consensus peptide, indicating that consensus sequences are preferred by OST in natural antibodies (Figure 5(C)). The docking score positively correlated with the binding affinity score, consistent with our previous data (Figure 5(D)). Hydrophobic residues (e.g. Tyr, Phe, Trp, Ala, Val) were frequently observed at the −1 position of N-glycosylation motifs. These results provide further insights into the sequence preference involved in non-consensus N-glycosylation in natural antibodies.

Table 1.

List of non-consensus and consensus N-glycosylation sequences in IVIg.

Non-consensus peptide	Protein ID	Site	Glycan modification	Glycan Cartoon
ATINCK	P06312	LC-V	Man1
ASQSISSYLNWYQQK	A2IPI0	LC-V	Man3
GVMTNAFDLWGQGTR	A0A7S5BXD3	HC-V	GlcNAc
VGLGYMNVWGK	A0A5C2GEB2	HC-V	GlcNAc2 F
QVQLVESGGNLVKPGGSLR	A0A7S5C4Y8	HC-V	G2GlcNAc
DDSNNTAYLQMNSLK	A0A5C2G7N4	HC-V	Man2
VDNALQSGNSQESVTEQDSK	P01834	LC-CH1	GlcNAc2
			G0F
DYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHK	S6BAN1	HC-CH1	Man2
FNWYVTAWR	S6BAN1	HC-CH3	G1F-GlcNAc SA
Consensus peptide	Protein ID	Site	Glycan modification	Glycan Cartoon
LSCAASGFNFSDYYMSWV	A0A7S5C687	HC-V	GlcNAc2
VNMSVDTS	A0A5C2GPN9	HC-V	G0F
NGSGSGTDFTLK	A0A5C2GZB1	LC-V	G0F
SGSGAGTNFTL	A0A5C2H0U9	LC-V	G1F
LLIYNASNR	A0A5C2GV37	LC-V	Man3
PALEDLLLGSEANLTCTLTGLR	P01876	HC-CH1	G2 SA
EEQYNSTYR	A8K008	HC-CH2	G0F

To evaluate the prevalence of non-consensus motifs in natural immune repertoires, we searched the Observed Antibody Space and analyzed a published dataset containing antibody sequences derived from naïve B cells of healthy cohort.⁴⁰ The sequences in this study were obtained using RNA sequencing and the protein glycosylation was not assessed. We calculated the frequency of previously identified non-consensus motifs in natural antibodies and found that YLNWYQQ had the highest occurrence (12.3%) (Supplementary Figure S5A). Next, we searched for these non-consensus motifs in the therapeutic antibody database Thera-SAbDab, and found that their frequency ranged from approximately 0–3% (Supplementary Figure S5B). Due to low glycan occupancy at non-consensus sites, glycosylation may not be easily detectable in these therapeutic antibodies. Collectively, this data could inform future efforts to evaluate non-consensus glycosylation during antibody development.

The glycopeptide was identified using PEAKS de novo sequencing and database search validation. The protein ID and location of the glycopeptide within the antibody structure were obtained from PeptideAtlas. The modified Asn is underlined. The glycan species were determined based on the mass difference between modified and unmodified peptides.

Ab breviations: HC-V, heavy chain variable region; LC-V, light chain variable region; C_H1/2/3, constant region 1/2/3.

Discussion

In this study, we developed a computational modeling workflow to predict OST binding and N-glycosylation frequency at non-consensus sequons. We identified a positive correlation between molecular docking scores and peptide properties during computational analysis, which facilitates the prediction of peptide binding and glycosylation by eukaryotic OST complex. Using a monoclonal antibody mAb-1 as a model, we studied the impact of sequence variants on glycosylation in the Fab region. Our findings reveal that glycosylation at non-consensus sites is sequence-dependent and associated with the recognition by OST complex.

The sequence preference of OST at non-consensus sites is not fully understood. Recent studies focusing on consensus sites have improved our understanding of OST structure and ligand interactions. A study on human OST revealed that sequences containing aromatic residues at the −2 or −1 position had significantly increased glycan occupancy.²⁵ Consistent with these findings, our study identified a non-consensus glycosylation site containing an aromatic Phe residue at the −2 position. This Phe residue is oriented toward the hydrophobic patch within the STT3B binding pocket and may facilitate peptide binding (Figure 1(B)). The residue scan analysis revealed that mutating this Phe into other amino acids decreased the ligand properties. In addition, we found that mutating the −1 position amino acid into an aromatic residue (Phe or Tyr) enhanced glycosylation at the non-consensus site. We further profiled naturally occurring glycosylation sites in human immunoglobulins and found frequently present aromatic residues (Phe, Tyr, and Trp) in consensus and non-consensus sequences with stronger binding with OST complex. Meanwhile, mutating the −1 amino acid to a positively charged arginine residue also increased glycosylation in the T30R mutant. This may be due to the interaction of Arg with the negatively charged protein patch surrounding the −1 amino acid within the STT3B binding pocket. Our study provides understanding of the rule behind non-consensus glycosylation sites that may facilitate glycosylation prediction and modulation in antibodies.

Moreover, we identified the sequence N-Y-N with the highest glycan occupancy compared to other variants. Notably, the predicted ligand interactions of this peptide closely mimicked the consensus sequence, where the Asn at the +2 position formed hydrogen bonds with the WWD loop, facilitating ligand binding and positioning. This is consistent with a previous study where an N-X-N glycosylation motif was identified in a monoclonal antibody.⁸ We further found that pharmacological inhibition of OST using NGI-1 abolished the non-consensus glycosylation at the N-Y-N motif, suggesting that the glycosylation mechanism at this atypical site is dependent on the OST pathway. Collectively, these findings suggest that glycosylation at non-consensus site follows an enzymatic mechanism catalyzed by OST similar to the consensus site, and the glycan occupancy is influenced by OST binding affinity. Thus, the occurrence of non-consensus N-glycosylation may be attributed to the broad substrate specificity of eukaryotic OST.

Current manufacturing strategies to control N-glycosylation and glycoform diversity in antibodies include cell-line selection, genetic modulation of glycosylation pathway, and optimizing culture media and condition.^41,42 These strategies primarily focus on controlling glycan diversity in the consensus sequence located in the Fc region, which plays a major role in ADCC.^43,44 However, due to their low frequency, non-consensus glycans are not well characterized. Recent studies have highlighted that Fab glycosylation often involves non-consensus sites and can influence antibody function, highlighting the importance in understanding and controlling non-consensus glycosylation.^35,45 In our study, we identified new non-consensus motifs in human IVIg and evaluated their prevalence in antibody repository derived from human naïve B cells. These motifs may contribute to critical quality attributes of therapeutic antibodies and require careful evaluation in antibody discovery and development.

To facilitate the application of our findings in antibody engineering, we propose the following actionable guidelines. First, analyze protein sequence to identify non‑consensus N‑glycosylation motifs, particularly N-X-N motif or those with hydrophobic residues (such as Tyr, Leu, or Phe) at the −1 position relative to Asn. Second, perform computational docking analysis to predict OST binding with identified non-consensus motifs. Third, for antibodies containing high-affinity motifs, conduct analytical studies – such as intact mass analysis, peptide mapping, or released glycan profiling – to experimentally evaluate the presence and extent of glycosylation at these sites. During mass spectrometry analysis, it is important to incorporate non-consensus glycan-modification into data analysis to ensure accurate detection and quantitation of these glycans, including those carrying sialic acid. Finally, where glycosylation at non‑canonical sites is undesired, substitute the asparagine residue or alter the surrounding sequence to disrupt the motif and eliminate glycosylation. Together, these steps may help mitigate the risk of unintended glycosylation during manufacturing, thereby ensuring the quality of therapeutic products, which contribute to their safety and efficacy.

By uncovering new non‑consensus glycosylation motifs, our study highlights questions and directions for future research. We found that Fab glycans exhibited higher structural diversity compared with Fc glycans, and that OST inhibition by NGI‑1 more strongly blocked Fc glycosylation (Figure 3(D) and 4). These differences may result from local structural variations between the Fab and Fc regions, which could alter their affinity for OST during post‑translational glycosylation, as well as for different glycosyltransferases during glycan processing.⁴⁶ To better understand these mechanisms, future studies could include structural analyses of OST binding and investigate the role of other glycosyltransferases in shaping Fab glycan diversity. Additionally, we identified several novel non‑consensus motifs in human IVIg after lectin enrichment (Table 1). While our analysis confirmed glycan modifications at the peptide level, further studies are needed to determine the extent of glycosylation within individual antibody that bear these motifs.

In summary, our study characterized the sequence specificity of OST in binding non-consensus motifs using computational modeling and in vitro mutagenesis. Our results demonstrate that aromatic and positively charged residues at the −1 position enhance binding with eukaryotic OST, contributing to increased glycosylation. This study addresses the gap in understanding glycosylation at unexpected sites and provides tools to predict and characterize non-consensus glycosylation. These findings contribute to improved sequence design and glycosylation prediction during biologics development.

Materials and methods

Computational analysis

Peptide docking

Computational modeling was performed in MOE software (v2024.06, Chemical Computing Group). The OST receptor structure was obtained from protein database (PDB: 6S7T). The structure was loaded in MOE as biomolecule assembly and prepared using default setting to assign bond orders and protonation states. The reference ligand AANATAA was used as a template to design 7 amino acid-peptide containing non-consensus sequence using the protein builder module. Designed peptide underwent side chain repack and minimization. Conformational search was performed using rigid peptide backbone and using the LowModeMD method with 0.5 RMS gradient and energy window of 50, while the other parameters were set as default. The generated conformational database was loaded in the docking module as existing ligand conformations. The receptor atoms were used as docking receptor and the reference ligand pocket was used as the docking site. Docking was performed using the placement method of triangular matcher with 200 poses generated using the London dG score and the refinement method of rigid receptor with 200 generated using the GBVI/WSA dG score. The resulted docking database was screened for receptor-ligand interactions at the acceptor Asn and Asp103 to select the pose with highest docking score.

Residue scan

The original FTNYGMN peptide was build using the reference peptide backbone and energy minimized. The amino acids at the −1, +1, and +2 positions were individually mutated into 19 natural amino acids. The binding affinity and thermostability was calculated for the ligand atoms.

Antibody modeler

The structure of mAb-1 was built from the protein sequence using the antibody modeler module. The sequence was imported into MOE and the full IgG structure was modeled using the IMGT numbering scheme and 1 maximum number of CDR and framework as default.

Peptide mapping analysis

Sample preparation

Protein was reduced in solution containing 4 M guanidine-HCl (Thermo Fisher Scientific 24,115), 80 mM Tris, 9.3 mM dithiothreitol (Thermo Fisher Scientific A39255) for 30 min at 37°C. Then 44 mM of iodoacetamide (Thermo Fisher Scientific A39271) was added to alkylate cysteines for 30 min at room temperature. The mixture was buffer exchanged using a spin column (micro Bio-Spin p-6 gel columns; BioRad 7,326,221) and buffer containing 20 mM sodium acetate at pH 5.2. Protein concentration was measured to calculate the amount of Trypsin/Lys-C (Promega V5071) to achieve an enzyme: protein ratio of 1:5 (w/w). The enzyme digestion mixture was incubated at 37°C for 5 hr. Then 0.5% trifluoroacetic acid (TFA) was added to the mixture to deactivate the enzyme.

For deglycosylation, three methods were evaluated (Supplementary Figure S1A). In the first method, rapid PNGase F (New England Biolabs P0711S) was added before reduction/alkylation at 3 µL enzyme per 100 µg protein and incubated for 1 hr at 37°C. In the second method, PNGase F was added after buffer exchange step at the same ratio. In the third method, digested peptides were dried in SpeedVac vacuum concentrator, redissolved in 50 mM Tris at pH 7.5, and digested by PNGase F.

Instrumentation

20 µg of peptides were loaded on Q Exactive HF Orbitrap (Thermo Fisher Scientific). Peptides were separated on a reverse-phase ACQUITY UPLC HSS T3 column (2.1x150 mm, 1.8 µm, Waters) set at 40°C. Mobile phase A contained 0.1% formic acid in water and mobile phase B contained 0.1% formic acid in 90% acetonitrile and 10% water. The fraction of mobile phase B was set as: 0–5 min: constant 0.1%; 5–91 min, from 0.1% to 40%; 91–95 min: 40% to 60%; 95–95.1 min: 60% to 90%; 95.1–99 min: constant 90%; 99.1–120 min: 90% to 0.01%. Peptides were analyzed by UV absorption at 214 nm followed by MS/MS analysis as previously described.⁴⁷ Data was acquired at defined isolation windows of 2 m/z in a range of 200 to 3,000 m/z. Source voltage was set to 3 kV for positive mode. Ion transfer tube temperature was set to 300°C and capillary temperature was set to 275°C. Gas flow rates were set at 40 µL/min, 10 µL/min, and 1 µL/min for sheath gas, auxiliary gas, and sweep gas, respectively. The full width at half-maximum (FWHM) of chromatographic peak was set as 15 s. Data was acquired in data-dependent mode and one full MS scan was followed by MS/MS scan of the top 10 most intense ions detected at a minimum threshold count of 1.5e⁵. Default charge state was set as 2. Full MS scan was collected using a resolution of 60,000, AGC target of 5e⁶, maximum injection time of 60 ms, and 1 microscan. The MS2 scan was collected using resolution at 30,000, AGC target of 5e⁶, maximum injection time of 200 ms, and 1 microscan. Stepped normalized collision energy was set to 28%. Dynamic exclusion time was set to 5 s for repeated precursor ion.

Data analysis

Raw data was processed using the PTM module in Byosphere (Protein Metrics). Instrument parameters were set as 10 ppm for precursor mass tolerance and 20 ppm for fragment mass tolerance. Peptide cleavage site was set as the C-terminal of lysine and arginine residues with fully specific digestion and no missed cleavage. Post-translational modifications used are listed in Supplementary data 2. The maximum number of common modifications was set to 2 and rare modification as 1 for each peptide. The other parameters were set as default. Sequence coverage achieved >95% for all samples.

Size exclusion chromatography

The antibody fraction enriched for the shoulder peak was analyzed by SEC using a Waters Acquity H-Class UPLC system and an ACQUITY UPLC Protein BEH SEC column (200 Å, 4.6x300 mm, 1.7 µm, Waters) at 25°C. Separation was achieved via isocratic elution with mobile phase containing 5% IPA, 400 mM sodium chloride, 55 mM sodium phosphate, pH 6.8. UV detection was performed at 280 nm.

Intact mass analysis

Reverse phase-mass spectrometry was used to locate the glycosylation site in the antibody. IdeS digestion was performed using IdeS enzyme (Genovis A0-FR1-020) to cleave the Fab and Fc portions of the molecule. The molecule was further reduced to release the light chain from the Fd portion of the Fab. Subunits were separated by reversed phase on a Waters Acquity H-Class UPLC system using an AdvanceBio RP-mAb C4 column (2.1x150 mm, 3.5 µm, Agilent) heated at 80°C. Mobile phase A was 0.1% TFA in water and mobile phase B was 0.1% TFA in 60/40 (v/v) n-propanol/acetonitrile. MS analysis was carried out on the attached Waters Synapt G2-Si mass spectrometer in positive ion mode, and the raw data was deconvoluted using maximum entropy algorithm embedded in MassLynx software (version 4.2).

SNA-binding

The method was adopted from a previously reported protocol using SNA-lectin.⁴⁸ In brief, 1.5 mg SNA agarose beads (Vector Laboratories, AL-1303–2) were placed in a Spin-X centrifuge tube (Millipore Sigma-Aldrich, CLS8161) and washed 6 times with lectin buffer containing 10 mM Tris, 140 mM NaCl, and 0.1 mM CaCl₂, at pH 7.4 to remove lactose in the storage solution. 5 mg IVIg (GAMUNEX-C, Grifols) was diluted to 500 µL and mixed with SNA beads and incubated overnight at 4°C. The filter was centrifuged at 1000 xg for 2 min and the flowthrough containing the unbound fraction was saved in a new tube. The beads were washed twice with 500 µL lectin buffer, and all flowthrough was pooled with the unbound fraction. To elute the bound fraction, 400 µL elution buffer containing 0.2 M lactose and 0.2 M acetic acid in lectin buffer was incubated with the beads for 20 min at room temperature, then the filter was centrifuged in a new collection tube containing 150 μL of 1 M Tris at pH 9. The elution step was repeated once, and the flowthrough was pooled with the bound fraction.

PEAKS de novo sequencing

Raw mass spectrometry data obtained from peptide mapping of deglycosylated IVIg was imported in PEAKS studio (v11, Bioinformatics Solutions Inc.). The DeepNovo workflow was used to process the data. The error mass tolerance was set as 10 ppm for precursor and 0.5 Da for fragment ion. The PTM list contained fixed carbamidomethylation and variable decarbamidomethylation at cysteine, variable ammonia loss and deamidation at asparagine, and variable oxidation at methionine. The maximum allowed variable PTM was set as 2 per peptide.

The output data from PEAKS contained identified peptide sequences and corresponding PTMs. All peptides with asparagine deamidation were extracted and converted as Fasta sequence file. This sequence file was used in Byosphere to process raw data obtained from peptide mapping of IVIg to identify glycosylation at asparagine without sequence constraint. The output peptide list from Byosphere was manually verified through PeptideAtlas search (http://www.peptideatlas.org/) to identify the full antibody sequence and the location of the glycosylation site within the antibody structure.

Database analysis

A natural antibody sequence dataset obtained from RNA sequencing of human naïve B cells from 48 healthy individuals was downloaded from the Observed Antibody Space database (https://opig.stats.ox.ac.uk/webapps/oas/)⁴⁰. This dataset contained unpaired light chain and heavy chain sequences from all antibody isotypes. Each non-consensus motif was searched against either the light chain or heavy chain dataset, depending on the motif’s location. The percent frequency was calculated by dividing the number of sequences containing the motif by the total number of sequences in the respective dataset.

To evaluate therapeutic antibodies, we used the sequence similarity search tool within Thera-SAbDab (https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/therasabdab/) to identify therapeutic molecules containing non-consensus motifs. The minimum sequence identity threshold was set to 60%, and the search results were manually reviewed. The percent frequency was then calculated by dividing the number of therapeutic antibodies containing the motif by the total number of therapeutic antibodies in the database.

DNA cloning

Single-site mutagenesis was performed following a previously described protocol.⁴⁹ In brief, the pcDNA vector containing the heavy-chain sequence was used as the template. Separate PCR reactions were set up using the Phusion High-Fidelity DNA polymerase kit (New England Biolabs E0553) for the forward and reverse primer. PCR products of the forward and reverse amplification were mixed and annealed by heating to 95°C and then gradually cooling to 37°C. Final dsDNA was digested using DpnI (New England Biolabs R0176) for 1 hr at 37°C. Primers are listed in Supplementary data 3.

Cell culture

ExpiCHO-S (Thermo Fisher Scientific A29127) cells were maintained and cultured following the manufacturer recommended protocol. Cells were cultured in ExpiCHO Expression Medium (Thermo Fisher Scientific A2910001) and shook at 125 rpm at 37°C in a humidified chamber with 8.0% CO₂. During antibody production, culture temperature and CO₂ concentration were reduced to 32°C and 5.0%. OST inhibitor, NGI-1 (MedChemExpress HY-117383) was dissolved in dimethyl sulfoxide (DMSO) to get a stock solution at 10 mM before adding to cell culture media to prevent precipitation. NGI-1 was added at 24 hr post transfection. The control group was added with DMSO to a final concentration of 0.2%.

Transfection

Transfection was achieved using the ExpiFectamine CHO Transfection Kit (Thermo Fisher Scientific A29129) and following manufacturer’s protocol. In brief, cells were seeded at 6 million cells per mL at a total volume of 25 mL in shaking flask. In 2 mL of OptiPRO SFM medium, 20 µg of heavy chain plasmid and 30 µg of light chain plasmid were added and mixed well. Then 80 µL of ExpiFectamine Reagent was added to the DNA solution and the final mixture was added in each cell flask. To achieve maximum titer, 150 µL Enhancer and 4 mL Feed were added 24 hr after transfection, and 4 mL Feed was repeatedly added on day 5 post transfection. Cells were harvested on day 14 post transfection by centrifuging at 5,000 xg for 30 min. The culture supernatant was collected and filtered through 0.2 µm filter (Thermo Fisher Scientific 564–0020). Antibody titer was measured using the Cedex Bio HT analyzer (Roche).

Protein purification

Antibody was purified from the culture supernatant using the Protein A resin (PreDictor MabSelect SuRe prepacked 96-well plates; Cytiva 28,925,825). Culture media was loaded to the plate and incubated for 20 min at room temperature. The maximum load density for antibody was 36 mg/mL per well. After loading, resin was washed three times using 0.04 M phosphate-buffered saline at pH 7.4. Antibody was eluted using 0.1 M citrate at pH 3.5 and further neutralized using 3 M Tris-HCl at pH 8.5. Protein concentration was measured using nanodrop.

Statistical analysis

Statistical analysis was performed in Prism 10. All data shown are representative data of at least two biological or technical replicates otherwise specified in the figure legend. Data are plotted as mean ± SD.

Funding Statement

This work was funded by Gilead Sciences Inc..

Author contributions

B.Q. performed all computational analysis, in vitro antibody production and mass spectrometry analysis. E.C. and B.Q. performed protein purification. Experimental design and execution were overseen by S.L. Manuscript was drafted by B.Q. and reviewed and edited by S.L., T.F., and B.Q.

Disclosure statement

E.C., T.F., and S.L. are current employees and shareholders of Gilead Sciences Inc. B.Q. is a former employee of Gilead Sciences Inc.

Data availability statement

All raw data is available upon request to corresponding author.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2025.2574406