Site‐specific O‐glycans influence lacritin structure and multimerization in tears

Abstract

Lacritin is an abundantly expressed glycoprotein in tear fluid and plays key roles in immune response, tear secretion, and bacterial killing. These biological functions are tightly regulated through several biochemical mechanisms including multimerization, proteolysis, and alternative splicing, especially within its C‐terminal domain. Given its critical role at the ocular surface, lacritin is currently under investigation as a diagnostic biomarker and therapeutic candidate for dry eye disease (DED). However, despite over three decades since its initial discovery, the functional significance of the O‐glycans that comprise more than 50% of its molecular weight remains largely unknown. To address this gap, we leveraged mass spectrometry (MS)‐based glycoproteomics, AlphaFold 3.0, and molecular dynamics (MD) to explore the structural role of site‐specific O‐glycans on C‐terminal lacritin. In doing so, we identified distinct glycosylation profiles between monomeric and multimeric lacritin, particularly at glycosites located near crosslinking residues (Lys101 and Lys104) that modulate multimer formation. Based on our glycoproteomics data, we performed MD simulations on monomer and multimer glycoforms and revealed that O‐glycans may participate in intramolecular glycan–protein interactions that influence its structure and the spatial arrangement of Lys101 and Lys104. Differences in the solvent accessible surface area (SASA) and root mean squared fluctuation (RMSF) of these residues further suggested that proximal O‐glycosylation could affect their ability to participate in crosslinking. To test these predictions, we performed in vitro crosslinking assays and demonstrated reduced TGM2‐mediated multimerization of glycosylated versus unmodified lacritin. Finally, we show that recombinant glycoforms bearing O‐glycan patterns similar to endogenous lacritin multimers are enriched after TGM2 crosslinking. Taken together, these findings underscore a central role for lacritin O‐glycans in affecting structural topology and multimerization with implications for its downstream biological activity.

1. INTRODUCTION

Lacritin is a ~23 kDa glycoprotein at the ocular surface that participates in tear production (Samudre et al., 2011; Sanghi et al., ²⁰⁰¹; Wang et al., ²⁰¹⁵), antimicrobial activity (McKown et al., 2014; Sharifian Gh et al., ²⁰²⁵), epithelial regeneration (Wang et al., 2006; Wang et al., ²⁰¹⁴), and protection against inflammatory stress (Wang et al., 2013). Although lacritin is highly abundant (~18 to 27 μM) (Willcox et al., 2017) in healthy tear fluid, its expression levels are known to be dysregulated in ocular pathologies such as dry eye disease (DED), an affliction experienced by over 400 million individuals worldwide (Willcox et al., 2017). Given its ubiquitous role in tear film, lacritin has been studied extensively in the past 30 years and is currently under investigation as a diagnostic biomarker and therapeutic candidate for DED (Tauber et al., 2023).

Recent advances in lacritin‐based diagnostics and therapeutics are largely driven by foundational biochemical discoveries that elucidated the mechanistic basis of lacritin function. For instance, Ma et al. first uncovered that the downstream biological activity of lacritin is mediated through its binding interaction with syndecan‐1 (SDC1), which initiates calcium signaling involving a G‐protein coupled receptor (GPCR) (Ma et al., 2006). Importantly, signaling through the lacritin‐syndecan‐1 axis is established through noncovalent interactions between the C‐terminal alpha helix of lacritin (residues 114–138) and the conserved GAGAL N‐terminal domain of SDC1 (Zhang et al., 2013). Additionally, 3‐O sulfation (generated following heparanase‐mediated deglycanation of SDC1) and chondroitin sulfate within the N‐terminal region of SDC1 are required for lacritin binding (Dias‐Teixeira et al., 2020; Ma et al., ²⁰⁰⁶). These findings have since informed the development of Lacripep™, a peptide‐based therapeutic candidate for DED derived from the lacritin C‐terminus which aims to reverse the pathological outcomes of dry eye (Georgiev et al., 2021; Tauber et al., ²⁰²³).

Several regulatory mechanisms influence the ability of lacritin to bind SDC1. For instance, lacritin forms dimers, trimers, and multimers through transglutaminase (TGM2)‐mediated crosslinking, where Lys101 and Lys104 act as donor residues and Gln125 acts as an acceptor residue (Francisco et al., 2013). Although the biological role of multimers in circulating tear fluid is still unknown, the process of multimerization significantly impairs lacritin binding to SDC1 and thus abrogates downstream signaling. Alternative splicing to generate lacritin isoforms A, B, C, and D is also known, where A represents the canonical sequence and is the only isoform known to bind SDC1 (Samudre et al., 2011). Until recently, isoform D was not predicted to exist at the protein level, and it still remains unknown whether isoform D can participate in multimerization (Chang et al., 2025). Finally, several studies have demonstrated that C‐terminal proteolysis of lacritin generates cleavage‐potentiated fragments which can engage SDC1 or participate in bacterial killing (Chang, Chen, et al., 2026; Georgiev et al., ²⁰²¹; McKown et al., ²⁰¹⁴; Sharifian Gh et al., ²⁰²⁵). Collectively, these processes contribute to ocular homeostasis through regulating lacritin effector function.

Crucially, lacritin is also highly decorated with glycans that contribute over 50% of its molecular weight (Chang et al., 2025; Ma et al., ²⁰⁰⁸). Glycosylation is a post‐translational modification (PTM) whereby glycans most commonly covalently modify Asn (N‐linked) and Ser/Thr residues (O‐linked) in a non‐templated fashion (Bagdonaite et al., 2022; Reily et al., ²⁰¹⁹; Steigmeyer et al., ²⁰²⁵). These glycan modifications can alter the physiochemical properties of the underlying protein to influence protein folding, confer proteolytic resistance, and shape protein–protein interactions (Goth et al., 2015; Goth et al., ²⁰¹⁷; Madsen et al., ²⁰²⁰; Reily et al., ²⁰¹⁹; Steigmeyer et al., ²⁰²⁵). However, despite the biochemical significance of glycosylation, its functional role in lacritin biology remains largely unexplored. This gap in understanding is primarily attributable to experimental difficulties in studying lacritin glycans. Most notably, low expression yields in mammalian systems have prevented analysis by controlled biochemical assays, while the analytical complexity of tear fluid has prevented its characterization by MS‐based glycoproteomics (Karnati et al., 2013; Ma et al., ²⁰⁰⁸).

Moreover, the substantial heterogeneity inherent to extensively glycosylated proteins and the high conformational flexibility of glycans have long hindered structural elucidation efforts by X‐ray crystallography or cryogenic electron microscopy (cryo‐EM) (Ince et al., 2022). Although molecular dynamics (MD) simulations have emerged as a powerful tool for predicting O‐glycoprotein structures, the high technical expertise and computational resources required limit the accessibility and throughput of these approaches. Recent efforts by the Fadda group have helped bridge these gaps through their development of GlycoShape, a publicly available modeling platform that enables rapid generation of glycoprotein models via a user‐friendly web interface (Ives et al., 2024). Similarly, the Moremen group recently explored AlphaFold 3.0 (Abramson et al., 2024) as a promising strategy for non‐specialists to perform high‐throughput structural prediction of glycoproteins and glycolipids (Huang et al., 2025). Although both approaches are accessible in their implementation, they offer different types of structural information. GlycoShape incorporates glycan conformational ensembles derived from MD simulations, whereas AlphaFold 3.0 predicts structural models without explicitly sampling protein dynamics. For highly flexible proteins, traditional molecular dynamics simulations can therefore provide complementary insight into glycosylation‐dependent changes in protein flexibility, hydrogen‐bonding interactions, and solvent‐accessible surface area (Lemkul, 2024).

To address current challenges in studying lacritin glycosylation, we recently developed a glycoproteomics workflow to uncover the tear fluid glycoproteome, providing the first in‐depth view of site‐specific glycans which modify this protein (Chang et al., 2025). Here, we characterized highly diverse glycan structures, including the Tn antigen (GalNAcα1‐Ser/Thr), sialylated core 1 (Galβ1‐3GalNAcα1‐Ser/Thr), type 3 H‐antigen (Fuc α1‐2‐Galβ1‐3GalNAcα1‐Ser/Thr), and sialylated and fucosylated core 2 O‐glycans (GlcNAcβ1‐6(Galβ1‐3)GalNAcα‐Ser/Thr). In a follow‐up study, we demonstrated that lacritin glycoforms harboring core 2 O‐glycans could serve as a ligand for extracellular galectin‐3 (Gal‐3), a glycan‐binding protein present in tear fluid (Chang, Lian, et al., 2026). Interestingly, we also found that Gal‐3 displayed higher affinity towards lacritin multimers compared to monomers, hinting at a potential difference in the glycosylation landscape between these two populations.

Herein, we hypothesized that multimeric and monomeric lacritin exhibit unique glycosylation profiles, and that these glycans could affect their structural conformation. More specifically, we asked whether glycans proximal to crosslinking residues (Lys101 and Lys104) could influence accessibility around these sites, with potential implications for multimerization. To investigate this, we developed a sample preparation method which could separate lacritin multimers from monomers for downstream glycoproteomic characterization. From MS data, we quantified the most abundant glycoforms in each fraction and found differences in O‐glycan heterogeneity at Ser86, Ser91, and Thr95. Building on this, we leveraged AlphaFold 3.0 and publicly available software (CHARMM‐GUI and GROMACS) to perform standard MD simulations on the most abundant glycoforms in our dataset. Here, structural analyses predicted that glycosylation within the two C‐terminal alpha‐helices may participate in intraglycan–protein interactions and alter the solvent accessibility (SASA) and conformational dynamics (RMSF) of Lys101 and Lys104 relative to unmodified lacritin. Given that Lys101 and Lys104 are established sites of TGM2‐mediated crosslinking, these structural observations prompted us to directly examine whether O‐glycosylation modulates lacritin multimerization. In vitro crosslinking experiments revealed differential TGM2‐mediated multimerization of glycosylated and unmodified lacritin and glycoproteomic analyses further identified specific glycoforms that were enriched after crosslinking. Taken together, our study explores site‐specific O‐glycosylation as a key structural determinant of lacritin function, revealing a possible regulatory role for lacritin O‐glycosylation by mediating multimerization. More broadly, our approach demonstrates how integrating glycoproteomics with accessible MD simulations can generate testable hypotheses linking site‐specific O‐glycosylation to glycoprotein structure and function.

2. RESULTS

2.1. Development and validation of a method to characterize lacritin monomers and multimers

Given the distinct molecular weights of lacritin monomers (~25 kDa) and multimers (≥50 kDa), we reasoned that GlycoFASP (Finn et al., 2025) (filter‐aided sample preparation) would provide an effective strategy for separating and characterizing these two populations. In brief, this workflow couples glycoprotease(s) with molecular‐weight cutoff (MWCO) filters, which retain proteins above a designated MW while lower MW species are separated into the filtrate. We first collected tear fluid (6–8 μL collected by microcapillary) from three different healthy donors for immediate reduction and alkylation before loading onto a 30‐kDa MWCO filter. After five rinses, monomers were recovered in the filtrate (<30 kDa) while multimers (>30 kDa) remained in the retentate (Figure 1a, top). Next, the filtrate was loaded onto a separate 10 kDa filter and both samples were processed identically using mucinase SmE to generate O‐glycopeptides (Chongsaritsinsuk et al., 2023; Mahoney et al., ²⁰²⁴). Next, O‐glycopeptides were collected through the filters and subjected to trypsin digestion and desalting before subsequent LC–MS/MS analysis (Figure 1a, bottom). Once MS raw files were generated, we manually validated MS spectra and performed label‐free quantitation (LFQ) using area under the curve (AUC) intensities to quantify glycoforms found in each fraction (Figure 1b, top). This data was then used to inform downstream in silico analysis where AlphaFold 3.0 and CHARMM‐GUI allowed us to build a glycosylated starting construct for triplicate MD simulations with GROMACS (Figure 1b, bottom). Finally, molecular structures and intraglycan–protein interactions were visualized with Pymol.

FIGURE 1.

Glycoproteomic and molecular dynamics workflow to characterize lacritin. (a) Tear fluid (3–4 μL) was collected by microcapillary tubes and resuspended in 20 mM Tris before being subjected to reduction and alkylation with dithiothreitol (DTT) and iodoacetamide (IAA). Next, the sample was washed five times on a 30‐kDa MWCO filter (containing multimers) and the filtrate (containing monomers) was collected and loaded onto a separate 10‐kDa MWCO filter. Next, O‐glycoprotease (mucinase) SmE was added to the top of each filter to generate lacritin glycopeptides which are collected as the flowthrough. Finally, tryptic digestion was performed before desalting and downstream LC–MS/MS analysis. (b) MS raw files were searched with Byonic and lacritin glycopeptides were manually validated before quantitation with LFQ using AUC intensities. The abundances of validated glycopeptides were used to inform the starting glycoform construct which was built using CHARMM GUI and AlphaFold 3.0. Lastly, molecular dynamics was executed with GROMACS and Pymol was used for molecular visualization.

Western blot analysis of control tear fluid (input) and the washed retentate across three biological replicates confirmed efficient separation of lacritin multimers from monomers, with the input containing both species and the retentate (after five washes) containing only multimeric lacritin (Figure 2a). While the filtrate was also collected for Western blot analysis, the concentration of lacritin collected after five sample rinses was too dilute to be detected by Western blot with our starting sample input. Nonetheless, we proceeded with downstream MS analysis and obtained near‐full sequence coverage of lacritin in both the 10 and 30 kDa preparations (Figure 2b). Further analysis of peptides generated from SmE and trypsin revealed that peptides containing Lys104 were only detectable in the monomer samples (highlighted pink in Figure 2b). Given that Lys104 is known to participate in TGM2‐mediated crosslinking, tryptic cleavage at this site would likely be hindered or result in a crosslinked peptide fragment (as previously reported (Francisco et al., 2013)) in the multimer fraction. Though the crosslinked peptide was not detected by our methods, the lack of a tryptic peptide containing Lys104 in our multimer fraction likely suggested crosslinking and provided further confirmation that these two populations could be separated with our workflow.

FIGURE 2.

Validation of FASP workflow by Western blot and MS. (a) Western blot analysis of lacritin in tear fluid input (3–4 μL) and the 30‐kDa filter retentate after five washes using a polyclonal N‐terminal anti‐lacritin antibody (“anti‐Pep Lac N‐term”). (b) Lacritin sequence coverage map from downstream MS analysis of the monomer and multimer fraction after glycoFASP processing. Multimer sequence coverage is shown in teal, monomer sequence coverage is shown in pink, and the shared sequence coverage is shown in purple. Coverage of one of the crosslinking lysines (Lys104) is highlighted in red and was identified only in the monomer fraction. (c) Quantification of N‐terminal mucin domain glycosylation in both fractions. Glycan structures and their corresponding colors in the pie chart are denoted in the legend on the bottom.

Next, we characterized the glycosylation landscape of the N‐terminal mucin domain (residues 23–77) in both fractions of a single patient. Interestingly, we observed highly similar glycan profiles with the Tn antigen as the most abundant structure detected (Figure 2c and Table S1a, Supporting Information). Monosialylated core 1 and sialylated and fucosylated core 2 O‐glycans were also detectable at lower relative abundances, consistent with our previous glycoproteomic analysis of lacritin.

2.2. In‐depth analysis of Ser86, Ser91, and Thr95 glycosylation

As previously noted, the C‐terminal helices of lacritin participate in SDC1 binding, multimerization, and antimicrobial activity to modulate ocular homeostasis (Francisco et al., 2013; Georgiev et al., ²⁰²¹; Ma et al., ²⁰⁰⁶). Given the biochemical significance of this region and the difference in biological activity between monomers and multimers (Francisco et al., 2013), we hypothesized that site‐specific O‐glycans in the C‐terminus (Ser86, Ser91, and Thr95) might differ between these two species. To investigate this, we first extracted ion chromatograms (XICs) for representative glycopeptides modified at these sites for both the monomer (Figure S1a) and multimer (Figure S2a) fractions across three different donors. To validate our replicates were comparable, we further examined glycopeptide retention times (Figures S1b and S2b), coefficient of variation of the total lacritin intensity (Figures S1c and S2c), and unique glycopeptide identifications (Figure S2d and Table S1b). Having established that our samples were comparable throughout these analyses, we next performed in‐depth quantitation using LFQ to compare the relative abundances of site‐specific glycoforms. Here, relative abundance is defined as the AUC intensity of each glycan composition expressed as a percentage of the summed AUC intensity of all glycoforms identified at that site.

To assess O‐glycan heterogeneity at Ser86, Ser91, and Thr95, we compiled the AUC intensities of all glycopeptides spanning these glycosites and averaged their relative abundances across the three donors (Tables S1c–S1e). At Ser86, the monosialylated core 1 O‐glycan (H1N1A1) was the predominant glycoform in both the monomer and multimer fractions (Figure 3a,b). However, multimers exhibited a significantly higher relative abundance of core 2 structures compared with monomers (18.36% vs. 7.78%, p < 0.01 using a two‐tailed t test). Most strikingly, we found that monomeric Ser91 was predominantly modified with H1N1A1 (81.70%), whereas in the multimer, Ser91 was largely unoccupied (44.63%) or decorated by a single GalNAc (27.01%) (Figure 3c,d). Further comparison of relative glycan abundances between the two populations revealed that H1N1A1 at Ser91 was significantly enriched in monomers (p < 0.0001), whereas the unmodified and Tn‐modified forms were significantly enriched in multimers (p < 0.001 and p < 0.01, respectively). A smaller but significant difference was also observed for H1N1F1 at Ser91 (p < 0.05). Finally, Thr95 was primarily modified by the Tn antigen for both monomers and multimers (58.31% and 44.11%, respectively), although differences in disialylated core 1 (p < 0.01) and core 2 (p < 0.05) O‐glycans were also detected (Figure 3e,f). We note, however, that the relative abundances of sialylated glycoforms across all three glycosites are likely an underestimate given that sialylated glycopeptides are known to ionize less efficiently in positive ion mode (Nishikaze, 2019; Ruhaak et al., ²⁰¹⁸). Nonetheless, these results highlight that the two lacritin populations exhibit distinct C‐terminal glycosylation profiles, which could contribute to their differing biological roles. Although site‐specific O‐glycan differences have been reported for recombinant SARS‐CoV‐2S monomer and dimer derived from insect cells (Bagdonaite et al., 2021), the extent to which O‐glycan patterns coincide with multimeric state in native biological contexts remains underexplored. To the best of our knowledge, this study represents the first glycoproteomic observation of endogenous O‐glycosylation variation between monomeric and multimeric species.

FIGURE 3.

Site‐specific O‐glycoproteomic analysis and glycan quantitation at Ser86, Ser91, and Thr95. (a, c, e) All identified glycopeptides containing Ser86, Ser91, and Thr95 from a matched monomer and multimer fraction of three individual donors were subjected to manual validation and LFQ using AUC intensities to generate corresponding bar graphs. Light blue bars correspond to the monomer and dark blue bars correspond to the multimer. Error bars represent the standard deviation of relative percent abundances across the three samples. Asterisks represent a statistical difference in means as calculated by a two tailed t test. p < 0.05, p < 0.01, p < 0.001, and p < 0.0001 are represented by *, **, ***, and ****, respectively. (b, d, f) Pie chart representations of O‐glycan heterogeneity at Ser86, Ser91, and Thr95 in both fractions averaged across the three samples. In the colored legend below, H denotes Hexose, N denotes HexNAc, A denotes Neu5Ac, and F denotes Fucose. Unmod describes no glycan occupancy at that site.

2.3. Structural analysis of C‐terminal lacritin with molecular dynamics

The presence of differential O‐glycosylation within the ordered C‐terminal domain prompted us to ask whether glycans might impart distinct conformational features. Notably, structural visualization of this region could provide insight into how O‐glycans affect SDC1 binding and/or multimerization at the C‐terminus of lacritin. To explore this, we modeled three C‐terminal constructs (non‐glycosylated, “monomer,” and “multimer”) spanning residues 79–137 of lacritin. As described in Figure 1b, we used the most abundant glycans at Ser86, Ser91, and Thr95 (Figure 3) to represent the monomer and multimer glycoforms. Ser86 was modified with a monosialylated core 1 O‐glycan and Thr95 was decorated with a single GalNAc for both species. At Ser91, a monosialylated core 1 O‐glycan was used for the monomer while the multimer was left unmodified (Figure 4a).

FIGURE 4.

C‐terminal lacritin structural analysis and potential glycan–protein interactions. (a) A 500‐ns production step was executed on GROMACS for three C‐terminal lacritin constructs (residues 79–137) and the 450‐ns frame was extracted for molecular visualization. Below each structure is the sequence and corresponding glycan occupancy at Ser86, Ser91, and Thr95 as previously determined in Figure 3 by taking the most abundant glycan structure at each site. The non‐glycosylated structure, monomer‐associated glycoform, and multimer‐associated glycoform are colored in red, teal, and blue, respectively. (b) Intraglycan–protein interactions (GPIs) in the monomer‐associated glycoform are shown. Asn82 participates in polar contacts with Neu5Ac (3.2 Å) and Gal (3.0 Å) of Ser86 (top). The Lys104 backbone forms a polar contact with GalNAc (3.1 Å) of Ser91 (middle). Ser121 and Ile92 each make a polar contact with GalNAc (3.0 Å and 3.4 Å) of Thr95 (bottom). (c) Intra‐GPIs in the multimer‐associated glycoform are depicted where the Leu81 backbone has a polar contact with GalNAc (2.7 Å) of Ser86 (top). Lys90 forms a polar contact with Neu5Ac (2.9 Å) of Ser86 (middle). Pro111 makes a polar contact with GalNAc (3.0 Å) of Thr95 while Phe116 participates in CH‐pi interactions (2.6 Å) with the same GalNAc (bottom). PDB‐format coordinate files of the extracted frame were visualized on Pymol and distances of polar contacts were traced on Pymol. The final structure was rendered and exported onto adobe illustrator for residue and glycan labeling.

To generate the initial structures for triplicate MD simulations, we leveraged AlphaFold 3.0 and CHARMM GUI which enabled us to build lacritin C‐terminal constructs (residues 79–137) with defined glycosylation. AlphaFold 3.0 predicted two α‐helices spanning residues 83–104 and 113–133, consistent with previous circular dichroism studies of synthetic peptides spanning residues 84–107 and 112–132 that support the helical character of these regions (Ma et al., 2008; Moore et al., ²⁰⁰⁸; Wang et al., ²⁰⁰⁶). We also evaluated the per‐residue confidence (pLDDT) scores reported by AlphaFold 3.0 and found that >60% of residues within these helices were predicted with high confidence (pLDDT > 90), while the remaining residues had moderate to confident (70 < pLDDT < 90) predictions (Figure S3). We note, however, that AlphaFold predictions generated using the full‐length sequence yield slightly lower confidence values within this region, potentially reflecting the influence of the intrinsically disordered N‐terminal domain. For the C‐terminal constructs used here, AlphaFold 3.0 generated five models (Conformers 0–4), and we selected the highest‐confidence model (Conformer 0) as the starting structure for each simulation. These structures were subsequently subjected to energy minimization, equilibration, and a 500‐ns production using GROMACS, with each condition performed in triplicate (see section 4 for additional details). The equilibrated structures used to initiate the production simulations are shown in Figure S4. Due to the computational cost, time, and resources required for these simulations, alternative AlphaFold‐derived conformers and a broader conformational ensemble were not explored.

The resultant structures shown in Figure 4a represent frames extracted at the 450‐ns time point of a given replicate, which were used to explore possible glycan–protein interactions (GPIs) as shown in Figure 4b,c. Inspection of these structures suggested that the proximity and positioning of glycans relative to nearby amino acids could permit intraglycan–protein contacts across the two alpha helices. Using the predominant glycoform of the monomer, we showed that Asn82 can participate in hydrogen bond interactions with the C2 hydroxyl on Gal of Ser86 (3.0 Å) and the N‐acetyl on Neu5Ac of Ser91 (3.2 Å). Additionally, the GalNAc N‐acetyl on Ser91 can coordinate with the Lys104 peptide backbone (3.1 Å) while Ile92 and Ser121 participate in polar interactions with the N‐acetyl and C3 hydroxyl of GalNAc on Thr95 (Figure 4b). These specific contacts were not observed in the representative multimer frame. Instead, the Ser86 glycan adopted a different orientation, with potential hydrogen bond interactions between the Leu81 peptide backbone and the GalNAc C6 hydroxyl (2.7 Å) as well as between Lys90 and the Neu5Ac C8 hydroxyl (2.9 Å). In the flexible loop connecting the two helices, Pro111 formed hydrogen bond interactions with the Thr95 GalNAc C6 hydroxyl (3.0 Å), which also participated in CH–π interactions with Phe116 (~2.6 Å) (Figure 4c). To assess glycan–protein interactions beyond these individual snapshots, we tracked intraglycan–protein hydrogen bonds throughout the 500‐ns production trajectories and found a similar number of interactions (Figure S5 and Table S1f).

While these observations highlighted possible glycan–protein interaction modes, additional conformational sampling could help establish the persistence of these contacts and identify other potential interactions. As enhanced sampling and simulations initiated from conformational ensembles were not explored in the present study, we interpret these as several possible GPIs within this sampling framework. Additionally, we note that glycan–protein hydrogen bonds represent only one component of the glycan interaction landscape, as O‐glycans can also engage extensively in hydrogen bonding with the surrounding solvent. Nevertheless, these results highlight a possible function for site‐specific O‐glycans in affecting the structural topology of C‐terminal lacritin, with possible implications for downstream binding events such as SDC1 recognition. Specifically, differences in glycan heterogeneity may result in distinct residue–glycan interactions, potentially affecting the spatial arrangement of the protein backbone. As such, we envision that these models may inform hypotheses surrounding lacritin C‐terminal binding interactions or protease processing of this domain.

2.4. O‐glycans affect C‐terminal backbone RMSD and the SASA of crosslinking residues

Given the prevalence of intra‐GPIs in our glycosylated constructs, we next examined whether glycosylation was associated with differences in the structural behavior of the modeled protein backbone (Figure 5a) over the course of the simulations. In particular, the observed intra‐GPIs could affect the relative conformation of the two helices through electrostatic interactions. To investigate this, we quantified the root mean square deviation (RMSD) of the protein backbone for all three constructs over the 500‐ns production step, performed in triplicate (Table S1g). Here, we showed that the unmodified construct displayed a higher average RMSD over the entire duration of the simulation, with a value of ~1.4 nm compared to ~0.6 nm when glycosylated (representative replicate shown in Figure 5b). This measurement was validated across three independent production replicates with randomized initial velocities for all three structures, where the average RMSD and standard error are plotted for each (Figure 5c and Table S1g). We note that these RMSD values may be influenced by the AlphaFold conformer selected as the starting structure and that simulations initiated from a broader conformational ensemble could help determine whether these trends persist across alternative starting conformations. Despite these limitations, our results suggest that O‐glycosylation alters the conformational behavior of C‐terminal lacritin, with the glycosylated constructs remaining closer to their starting conformations than the corresponding unmodified structure over the simulated timescale. More broadly, our observation that O‐glycans have projected effects on the protein backbone has been supported in previous MD studies (Chongsaritsinsuk et al., 2023; Kearns et al., ²⁰²⁴) and by atomic force microscopy (Ince et al., 2022; Kramer et al., ²⁰¹⁵).

FIGURE 5.

C‐terminal O‐glycans affect backbone RMSD and the SASA of crosslinking residues. (a) Protein structures after 450 ns MD simulations were labeled. (b) Protein backbone flexibility (RMSD in nm) was measured for a single replica over the 500‐ns production step for the nonglycosylated (pink), “monomer” (teal), and “multimer” (dark blue) glycoforms. (c) Three independent replica simulations for each construct were executed, and the average RMSD over the 500‐ns production step was taken for each for a total of nine data points. The bar graph shows the average RMSD value for the three structures, where each bar represents three replicate simulations. (d) Molecular visualization of multimer and monomer‐associated glycoforms and the impact of glycosylation on Lys101 and Lys104 orientation (using the 450‐ns frame). (e) The solvent accessible surface area (SASA) in Ų of Lys101 and Lys104 (averaged over the 500‐ns production step) was first recorded for each replicate and then averaged across three replicates. Next, the difference in SASA (ΔSASA) was taken between the multimer and the unmodified structure, the multimer and monomer, and the monomer and the unmodified structure for Lys101 (left three bars) and Lys104 (right three bars). Green bars represent a predicted increase in SASA while red bars represent a predicted decrease. The RMSF (averaged over the 500‐ns production step) of Lys 101 (left) and Lys 104 (right) side chains was measured and averaged across three replicates for each construct. Error bars represent the SEM across three replicates. See section 4 and Tables S1g–S1j for more details.

Building on these analyses, we asked whether specific glycoforms could provide structural insights into propensity for multimerization, where residues that participate in crosslinking might be affected by proximal glycans. As shown in Figure 5d, Lys101 and Lys104 of both glycoforms are anticipated to position in a manner that may promote access to TGM2 for multimerization. Interestingly, monosialylated core 1 O‐glycans on the monomer glycoform (Figure 5d, right structure) are positioned in closer proximity to Lys101 and Lys104 through possible hydrogen bonding interactions with the Lys104 backbone and orient in the same direction as these residues. Consistent with this observation, tracking the average glycan‐Lys distance throughout the simulations revealed that the monomer‐associated glycans maintained mean distances of 6.58 ± 0.87 Å and 8.91 ± 3.54 Å to Lys101 and Lys104, respectively, whereas the multimer‐associated glycoform exhibited larger corresponding average distances of 9.14 ± 0.76 Å and 13.51 ± 1.47 Å (Figure S6 and Table S1h). These findings suggested that the monomer‐associated glycans are spatially closer to Lys101 and Lys104 and could influence TGM2 access through steric effects.

Beyond these qualitative observations, we extracted the averaged solvent accessible surface area or SASA (per residue) from all simulations (three for each construct) and quantified their mean difference (ΔSASA) (Tables S1i and S1j). By calculating the ΔSASA of Lys101 and Lys104 across the three structures, we observed a general increase in solvent accessibility for the multimer‐associated glycoform relative to the monomeric and non‐glycosylated structures (Figure 5e and Table S1j). Most notably, the SASA of Lys101 and Lys104 was predicted to increase by 28.4 ± 14.7 Ų and 14.7 ± 6.1 Ų, respectively, when comparing the multimer to the unmodified C‐terminus. Similarly, the SASA of Lys104 was predicted to increase by 19 ± 10.7 Ų when comparing multimer and monomer glycoforms, though the monomer glycoform displayed similar or reduced SASA relative to the unmodified form. As an additional measure of local structural behavior, we also quantified the root‐mean‐square fluctuation (RMSF) of the Lys101 and Lys104 side chains across all simulations. Here we observed modest differences in RMSF among the three constructs (Figure 5e), suggesting subtle changes in local residue mobility over the simulated timescale. Given the conformational dependence of residue accessibility, additional sampling across a broader conformational ensemble could help determine whether these differences are also observed across alternative conformational states. Within the sampling performed here, a complete list of SASA and ΔSASA values and their corresponding standard errors across all nine simulations is provided in Table S1i. Collectively, these analyses support that site‐specific O‐glycans may contribute to differences in the structural landscape of lacritin, with implications for affecting multimerization.

2.5. O‐glycosylation affects lacritin multimerization by TGM2 in vitro

Based on our structural insights, we hypothesized that O‐glycosylation could affect TGM2‐mediated multimerization of lacritin, with distinct glycoforms exhibiting differential susceptibility to crosslinking. To test this, recombinant (His‐tagged) glycosylated and unmodified lacritin were subjected to in vitro TGM2 crosslinking at a 1:3 enzyme‐to‐substrate ratio over 2 h (Figures 6a and S7a). Initial visualization by total protein staining revealed greater retention of the glycosylated lacritin monomer relative to the unmodified form after 2 h, indicating reduced TGM2‐mediated multimerization. To more accurately quantify this difference, we performed three replicates of each condition followed by anti‐His immunoblotting and quantitative densitometry (Figures 6b and S7b and Table S1k). At the final time point, ~76% of unmodified lacritin was crosslinked (calculated as 100% minus the percent monomer remaining), whereas only ~40% of glycosylated lacritin underwent multimerization (Figure 6b and Table S1k). Interestingly, the amount of glycosylated monomer did not change significantly between 15 and 120 min, while the unmodified form steadily decreased.

FIGURE 6.

O‐glycosylation modulates TGM2‐mediated multimerization of lacritin. (a) In vitro TGM2‐mediated crosslinking of recombinant unmodified and glycosylated lacritin. His‐tagged lacritin (1 μM) was incubated with TGM2 (0.33 μM; 1:3 E:S ratio) for the indicated times, and crosslinking was visualized by REVERT total protein staining. Control samples were prepared identically without TGM2. Red arrowheads indicate TGM2, glycosylated lacritin, unmodified lacritin, and a protein impurity. (b) Quantification of TGM2‐mediated multimerization of unmodified and glycosylated lacritin across three replicates using anti‐His immunoblotting and quantitative densitometry. Percent multimerization was calculated from the fraction of monomer remaining at each time point and is shown as mean ± SD. A two‐tailed t test was performed on the final time point and statistically significant (p < 0.01, indicated by **). (c) Schematic of control vs. multimer lacritin populations which were characterized by LC–MS/MS. (d) Glycoproteomic comparison of lacritin glycoforms in the multimer fraction relative to the untreated control. Fold changes in normalized AUC intensities are shown for seven peptides encompassing unmodified and O‐glycosylated forms containing Tn, T, or sialylated T antigens. Positive values and green bars indicate enrichment in the multimer fraction, whereas negative values and red bars indicate depletion relative to the control.

From our anti‐His immunoblot, we also observed that glycosylated lacritin initially appeared as a broad, diffuse band centered at approximately 25 kDa, consistent with a heterogeneous population of glycoforms. As crosslinking progressed, the increased sensitivity from anti‐His staining revealed discrete bands persisting within the remaining monomeric population (Figure S8), suggesting that susceptibility to TGM2‐mediated multimerization may vary among glycoforms. To characterize the glycoforms that underwent multimerization, we used a 30‐kDa MWCO filter to isolate and compare the glycoproteomic profiles of the untreated control and crosslinked multimers (Figure 6c). A complete list of glycopeptides identified in the untreated control is provided in Table S1l. We note that attempts to similarly characterize the residual monomer fraction were unsuccessful, potentially due to sample complexity arising from the EDTA required for reaction quenching and the high salt content of the crosslinking reaction.

Given our previous identification of unmodified Ser91 as the predominant species in endogenous multimeric lacritin, together with MD simulations predicting potential steric effects from a monosialylated core 1 O‐glycan at this site, we expected that species with unmodified Ser91 would be enriched in the multimer fraction, whereas Ser91 bearing H1N1A1 would be depleted. We therefore compared normalized AUC intensities for seven glycopeptides encompassing unmodified and O‐glycosylated forms containing Tn, T, or sialylated T antigens in the control and multimers (Figure 6d). Consistent with our hypothesis, unmodified Ser91 and Tn‐modified Thr95‐containing peptides were enriched approximately two‐fold in the multimer fraction relative to the control, whereas the Ser91 H1N1A1 glycoform was depleted approximately 20‐fold. Together, these findings recapitulate glycosylation features identified in endogenous lacritin multimers and support a model in which C‐terminal O‐glycosylation modulates the susceptibility of distinct lacritin glycoforms to TGM2‐mediated crosslinking.

3. DISCUSSION

To date, the functional role of lacritin O‐glycosylation has remained a critical blind spot in our understanding of its biology. To be sure, the chemical complexity of O‐glycan heterogeneity has largely impaired biochemical, bioanalytical, and structural analyses of lacritin glycoforms over the past three decades. In our prior study (Chang et al., 2025), we leveraged recent advances in MS‐based glycoproteomics (Afshari et al., 2025; Chongsaritsinsuk et al., ²⁰²³; Mahoney et al., ²⁰²⁴; Movassaghi et al., ²⁰²⁵) to characterize the tear fluid glycoproteome, establishing the first in‐depth view of site‐specific O‐glycans which decorate lacritin. Subsequent analysis with AlphaFold 3.0 predicted that O‐glycosylation promotes a “bottlebrush‐like” secondary structure, in line with observations for other mucin‐domain glycoproteins (Chang et al., 2025; Kearns et al., ²⁰²⁴). From additional follow‐up studies (Chang, Chen, et al., 2026; Chang, Lian, et al., ²⁰²⁶), we demonstrated that the densely O‐glycosylated region of lacritin can confer proteolytic protection or serve as a ligand for Galectin‐3 (Gal‐3) in tear fluid. In this context, the mucin domain may modulate the extracellular half‐life of lacritin by sterically blocking protease access through its O‐glycans. Alternatively, engagement of Gal‐3 as an extracellular binding partner may likewise influence circulating half‐life, consistent with observations reported for other extracellular glycoproteins (Grazier & Sylvester, 2022; Johannes et al., ²⁰¹⁸).

Building on these insights, we hypothesized in the present study that O‐glycans affect the C‐terminal domain of lacritin, which has implications for its downstream biological activity and ocular homeostasis. Notably, multimerization stands out as a key regulatory mechanism that mediates lacritin effector function, since multimeric lacritin is incapable of binding SDC‐1 to trigger GPCR signaling. This led us to ask whether multimers and monomers differed in their O‐glycosylation patterns, particularly at glycosites proximal to crosslinking residues (Lys101 and Lys104). To investigate this, we modified our previously reported GlycoFASP method to isolate lacritin multimers and monomers for separate downstream MS analyses. Here, we discovered distinct glycoform distributions in each fraction and subsequently quantified site‐specific glycan abundances to inform the design of C‐terminal lacritin constructs for MD simulations.

From in silico analyses, O‐glycans were predicted to participate in intra‐GPIs between the two alpha helices, with some representative examples shown in Figure 4b,c. We expanded on these observations by measuring the protein backbone RMSD and revealed differences between the glycosylated and non‐glycosylated constructs, consistent with previous observations in other glycosylated proteins (Chongsaritsinsuk et al., 2023; Kearns et al., ²⁰²⁴; Kramer et al., ²⁰¹⁵). Further inspection of the C‐terminal structures suggested that the local conformation of Lys101 and Lys104 may be affected by a monosialylated core 1 O‐glycan at Ser91, which was also supported by SASA and RMSF measurements at these sites. To test whether specific glycoforms might influence TGM2‐mediated multimerization, we performed in vitro crosslinking studies on recombinant glycosylated and unmodified lacritin, followed by glycoproteomic analysis. From these experiments, we showed that glycans can influence multimerization, whereby distinct Ser91 glycoforms are more resistant or susceptible to crosslinking. Together, these results point to a biochemical role for O‐glycosylation in influencing multimerization, where intra‐GPIs and glycan‐imposed steric effects may play a key function. Consistent with this, a study by Wu and Robinson showed that trimeric tumor necrosis factor‐⍺ (TNF‐⍺) is stabilized by an O‐linked glycan at Ser80, while Karampini et al. demonstrated that von Willibrand factor multimerization is altered upon treatment of endothelial cells with an O‐glycosylation inhibitor (GalNAc‐O‐benzyl) (Karampini et al., 2024; Wu & Robinson, ²⁰²²). Although these studies highlight an emerging role for O‐glycan‐driven effects on multimerization, this phenomenon remains largely underexplored, with only these few examples reported to date. Our findings expand on the growing body of literature on O‐glycan‐mediated multimerization and underscore the importance of discovering these interactions in other O‐glycoproteins.

This study lays the critical groundwork for future investigation of lacritin biology mediated by its glycans. It currently remains unknown whether distinct lacritin glycoforms might exhibit varying affinities towards SDC1 or engage different protein ligands entirely. Similarly, the extent to which glycans might affect the antimicrobial activity of C‐terminal lacritin peptides will require future studies to elucidate. Finally, spliceoform‐specific multimerization has also been reported (Justis et al., 2020), where isoform C and (potentially) isoform D can also multimerize. Whether the canonical splice variant forms heterodimers with isoform C or D remains unclear, though it is likely that O‐glycans would play a role in this process as well. As glycosylation vastly expands the chemical space of lacritin, it will be necessary to account for this additional layer of complexity when defining new roles for lacritin at the ocular surface. From a diagnostic perspective, it is possible that lacritin glycosylation could be concomitant with ocular pathologies such as DED. Though it has not been established whether lacritin monomers or multimers offer more diagnostic potential based on their disease‐associated glycan changes, the methods described in this study now enable these two populations to be separated for independent glycoproteomic characterization. Further efforts to profile their glycosylation changes across ocular diseases will thus be crucial to inform glycan‐centric diagnostic strategies.

More broadly, this study highlights MS‐based O‐glycoproteomics and basic MD simulations as complementary analytical tools to explore fundamental biochemistry mediated by glycans. Until recently, the ability to identify and quantify O‐glycans in a site‐specific manner with high sensitivity has presented an enormous analytical challenge. This is evidenced by recent studies where MD modeling of O‐glycoproteins relies on released O‐glycan analysis from large quantities of a purified glycoprotein to obtain the glycan data needed to inform simulations (DeBono et al., 2025; Hintze et al., ²⁰²⁶). Though glycoproteomics experiments can be performed as an alternative to obtain site‐specific O‐glycan data, accurate and sensitive characterization of low abundance glycoproteins from complex samples remains a critical bottleneck in the glycoproteomics workflow (Bagdonaite et al., 2022; Chongsaritsinsuk et al., ²⁰²⁴; Rangel‐Angarita et al., ²⁰²³; Rangel‐Angarita et al., ²⁰²⁵). To address these limitations, recent advances in MS‐based glycoproteomics such as improved enrichment strategies (Riley et al., 2021), optimized gas‐phase fragmentation methods (Riley et al., 2020), and the use of O‐glycoproteases and mucinases have enhanced our ability to map O‐glycosylated proteins (Chongsaritsinsuk et al., 2024; Movassaghi et al., ²⁰²⁵). This is highlighted in a recent collaboration between our laboratory and the Amaro group (Chongsaritsinsuk et al., 2023), where we demonstrated that site‐specific glycoproteomics data can be coupled with MD simulations to achieve detailed models of densely O‐glycosylated mucins at the cell surface. However, the technical expertise required, reliance on specialized in‐house MD software, and substantial computational demands (several months on a supercomputer) decrease the accessibility of this approach to non‐specialists and laboratories without MD capabilities. As such, looking forward, we envision that the methods described in this study will provide an accessible framework for integrating high‐quality O‐glycoproteomics data with publicly available MD software to investigate structural effects driven by site‐specific O‐glycans.

4. MATERIALS AND METHODS

4.1. Tear fluid sample collection

Tear fluid was purchased from Innovative Research (obtained via microcapillary collection) and stored at −20°C until further processing. Three individual donors (56669‐ND0557‐CF35, 56668‐ND0350‐CF42, and 56667‐ND0143‐CF43) were used for western blot and downstream MS analysis. According to the vendor's collection documentation, all tears were collected under basal (non‐stimulated) conditions, and no products or procedures were used to stimulate tear production (no local anesthesia was used). All donors were classified by the vendor as “normal donors” after passing an ocular history and overall health risk‐factor screening questionnaire. Patients also provided signed clearance for collection. The ocular screening questionnaire evaluates any current or recent history of dry eye disease, eye irritation, ocular surgery, current redness or infection or inflammation, contact lens use and recency of wear, use of ocular medications/eye drops, allergy history and recent active allergic symptoms. Screening of general health status evaluated recent infection, chronic medical conditions, pregnancy/lactation status, prescription medication use (including immunosuppressants/steroids/biologics), recent antibiotic/antiviral use, infectious disease history and exposure risk (including HIV, HBV, HCV), recent transfusion, recent tattoos/piercings, recent active allergic symptoms, smoking/vaping status, recent alcohol intake, recent exposure to air pollution/chemicals/smoke, recent illness, recent vaccination, and recent COVID‐19 infection. See Table S1n for more details.

4.2. Mass spectrometry sample preparation with GlycoFASP

Freshly thawed tear fluid proteins from each patient were diluted to a final concentration of 0.2 mg/mL in 100 μL of 20 mM Tris. DTT was then added to a concentration of 2 mM and reacted at 65°C for 1 h followed by alkylation in 5 mM IAA for 15 min in the dark at RT. Subsequently, samples were loaded onto a 30‐kDa filter and the filtrate was collected over five washes (400 μL of 20 mM Tris for each elution). The filtrate was then loaded onto a 10‐kDa filter and denoted as the “monomer fraction” whereas the retentate of the 30‐kDa filter is the “multimer fraction.” For both fractions, mucinase SmE was added at a 1:3 (E:S) ratio and allowed to react for 12 h at 37°C. After glycoprotease digestion, O‐glycopeptides were filtered into a separate tube and subjected to an additional trypsin digestion for 3 h at 37°C. Finally, peptides were acidified by adding 4 μL of formic acid and desalted before injection into the mass spectrometer. Desalting was performed using 10 mg Strata‐X 33 μm polymeric reversed phase SPE columns (Phenomenex). Each column was activated using 500 μL of acetonitrile (ACN) (Honeywell) followed by of 500 μL of 0.1% formic acid, 500 μL of 0.1% formic acid in 40% ACN, and equilibration with two additions of 500 μL of 0.1% formic acid. After equilibration, the samples were added to the column and rinsed twice with 200 μL of 0.1% formic acid. The columns were transferred to a 1.5‐mL tube for elution by two additions of 150 μL of 0.1% formic acid in 40% ACN. The eluent was then dried using a vacuum concentrator (LabConco) prior to reconstitution in 10 μL of 0.1% formic acid. The resultant peptides were then injected onto a Dionex Ultimate3000 coupled to a Thermo Orbitrap Eclipse Tribrid mass spectrometer. We employed a higher‐energy collision dissociation product‐dependent electron transfer dissociation (HCD‐pd‐ETD) method; in some cases, we used supplemental activation in ETD (EThcD). The files were searched using Byonic, followed by manual data curation.

4.3. Mass spectrometry data acquisition

Samples were analyzed by online nanoflow liquid chromatography–tandem mass spectrometry using an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific) coupled to a Dionex UltiMate 3000 HPLC (Thermo Fisher Scientific). For each analysis, 4 μL was injected onto an Acclaim PepMap 100 column packed with 2 cm of 5 μm C18 material (Thermo Fisher, 164564) using 0.1% formic acid in water (solvent A). Peptides were then separated on a 15‐cm PepMap RSLC EASY‐Spray C18 column packed with 2 μm C18 material (Thermo Fisher, ES904) using a gradient from 0 to 35% solvent B (0.1% formic acid with 80% acetonitrile) in 60 min. Full scan MS1 spectra were collected at a resolution of 60,000, an automatic gain control target of 3e5, and a mass range from m/z 300 to 1500. Dynamic exclusion was enabled with a repeat count of 2, repeat duration of 7 s, and exclusion duration of 7 s. Only charge states 2–6 were selected for fragmentation. MS2s were generated at top speed for 3 s. Higher‐energy collisional dissociation (HCD) was performed on all selected precursor masses with the following parameters: isolation window of 2 m/z, 29% normalized collision energy, orbitrap detection (resolution of 7500), maximum inject time of 50 ms, and a standard automatic gain control target. An additional electron transfer dissociation (ETD) fragmentation of the same precursor was triggered if (1) the precursor mass was between m/z 300 and 1500 and (2) 3 of 8 HexNAc or NeuAc fingerprint ions (126.055, 138.055, 144.07, 168.065, 186.076, 204.086, 274.092, and 292.103) were present at m/z ±0.1 and greater than 5% relative intensity. Two files were collected for each sample: the first collected an ETD scan with supplemental energy (EThcD) while the second method collected a scan without supplemental energy. Both used charge‐calibrated ETD reaction times, 100 ms maximum injection time, and standard injection targets. EThcD parameters were as follows: Orbitrap detection (resolution 7500), calibrated charge‐dependent ETD times, 15% nCE for HCD, maximum inject time of 150 ms, and a standard precursor injection target. For the second file, dependent scans were only triggered for precursors below m/z 1000, and data were collected in the ion trap using a normal scan rate.

4.4. Mass spectrometry data analysis

Raw files were searched using Byonic (version 4.5.2, Protein Metrics, Inc.) against a FASTA file containing the canonical lacritin sequence (accession number: Q9GZZ8) and its spliceoforms (accession numbers: H0Y100 and F8W0V3). For all samples, we used the default O‐glycan database containing nine common structures. Files were searched with six missed cleavages and N‐terminal to Ser/Thr and C‐terminal to Arg/Lys. Mass tolerance was set to 10 ppm for MS1's and 20 ppm for MS2's. Carbamidomethyl Cys was set as a fixed modification. From the Byonic search results, glycopeptides were filtered to a score of >200 and a logprob of >2. From the remaining list of glycopeptides, the extracted ion chromatograms, full mass spectra (MS1s), and fragmentation spectra (MS2s) were investigated in XCalibur QualBrowser (Thermo) for selected glycopeptides. Glycopeptides were manually validated from the filtered list of Byonic's reported peptides (score >200 and logprob >2) according to the following steps: The MS1 was first used to confirm the precursor mass and chosen isotope was correct. This also allowed us to identify any co‐isolated species that could interfere with the MS2s and/or explain unassigned peaks. The HCD and EThcD fragmentation spectra were then investigated to identify sufficient coverage to make a sequence assignment. When possible, multiple MS2 scans were averaged to obtain a stronger spectrum. For HCD, an initial glycopeptide identification was confirmed if the presence of the precursor mass without a glycan present (i.e., Y0), along with coverage of b and y ions without glycosylation. For longer peptides, we required the presence of Y0 and fragments that were expected to be abundant (e.g., N‐terminally to Pro, C‐terminally to Asp). When the peptide contained a Pro at the C‐terminus, the b_n−1 was considered sufficient. Further, when the sequence contained oxidized Met, the Met loss from the bare mass was considered as representative of the naked peptide mass. We then used electron‐based fragmentation MS2 spectra for localization. Here, all plausible localizations were considered, regardless of search result output. We confirmed the presence of fragment ions in ETD or EThcD that were between potential glycosylation sites, if sufficient c/z ions were present then a glycan mass was considered localized. For glycopeptide manual validation, extracted ion chromatograms are evaluated at the MS1 level to determine the charge and m/z of the highest abundance precursor species. Mass spectrometry data files and raw search output can be found on PRIDE with identifier PXD076054.

4.5. LFQ and AUC analysis of glycopeptides containing Ser86, Ser91, and Thr95

Glycopeptides and unmodified peptides spanning glycosites Ser 86, Ser91, and Thr95 were compiled and manually validated as described above. After manual validation, we extracted ion chromatograms and took the area under the curve (AUC) for each glycopeptide, which is reported in Tables S1c and S1d. The AUC intensities of all glycopeptides which contained a glycan structure at a given site were summed and divided by the total intensity of all glycan structures found at that glycosite to give a percent relative abundance value. Relative abundance values were calculated for one three healthy donors to give three total measurements for each glycan structure on a glycosite, where the mean and standard deviation of the three values was used to generate Figure 3.

It is important to note that sialylated glycopeptides are known to ionize less efficiently when compared to their nonsialylated counterparts, which may lead to an underestimate of the relative abundance of sialylation reported in this study. Since other LFQ‐based methods such as the use of glycopeptide spectral matches (GSM) also suffer from this limitation, we opted to use AUC‐based quantitation [as in our previous studies (Afshari et al., 2025; Chang et al., ²⁰²⁵; Chongsaritsinsuk et al., ²⁰²³; Chongsaritsinsuk et al., ²⁰²⁴)] for relative quantitation of sialylated glycopeptides. Finally, we note that the presence of a Lys residue could also affect peptide quantitation in positive ion mode. In our analysis, the glycopeptide T[N1]EQALAKAGK contains Lys104 and was only detected in the monomer fraction. As such, the use of AUC‐based quantitation for this glycopeptide could increase the apparent relative abundance of N1 at Thr 95 for the monomer fraction. However, given that this glycopeptide accounts for only ~0.01% of the total intensity of Thr95‐containing glycopeptides in the monomer fraction, it is unlikely to significantly affect our quantitation. We were also unable to detect any other glycopeptides containing Lys104 which could affect AUC measurements.

4.6. MD analysis of lacritin C‐terminal constructs

The most abundant O‐glycan structures of lacritin were first calculated at each O‐glycosite using LFQ of AUC intensities of XICs of all lacritin glycopeptides identified. Once glycan structures were determined, the sequence of C‐terminal lacritin (Uniprot ID: Q9GZZ8, residues 77–137) was used to generate a pdb‐format coordinate file on AlphaFold 3.0 webserver. The “nonglycosylated construct” consisted of the base C‐terminal sequence without glycans. The “glycosylated construct” was generated by using 3‐letter Chemical Component Dictionary (CCD) codes for glycans which were manually input onto Ser/Thr. Here, NAG was used for a single HexNAc residue. Per‐residue confidence was assessed using the predicted local distance difference test (pLDDT) scores reported by AlphaFold 3.0. The majority of residues within the modeled α‐helices exhibited high confidence (pLDDT >90) and the remaining residues displayed moderate confidence (70 < pLDDT < 90) (Figure S3). Additional metrics such as predicted TM‐score (pTM), interface pTM (ipTM, applicable only to the glycosylated model), AlphaFold 3 ranking score, fraction of residues predicted as disordered, mean and maximum predicted aligned error (PAE, Å), and the percentage of residue pairs with contact probability >0.5 are also reported in Figure S3. For each construct, five structural models (Conformers 0–4) were generated, and the highest‐confidence model (Conformer 0) was selected as the starting structure for all downstream analyses. The resultant PDB‐format coordinate file with the predicted fold was input into CHARMM‐GUI for further glycan editing and to obtain a solvated and electrically neutral system in a rectangular periodic boundary water box. KCl ions were added via the Monte Carlo method at a concentration of 150 mM. CHARMM‐GUI output files (.pdb solvated structure files, carbohydrate_restraint.str files, equilibration.inp, and production.inp files) were then generated for downstream GROMACS analysis. Following system construction, initial geometry refinement was carried out in GROMACS using the steepest‐descent algorithm for 5000 steps, employing a maximum force tolerance of 1000 kJ/mol·nm⁻¹. The minimized structures were then equilibrated using the CHARMM‐GUI multistage protocol, consisting of NVT heating and NPT equilibration at 303.15 K and 1 atm. All equilibration steps were run with a 1‐fs integration time step. The final equilibrated configuration (step4.1_equilibration.gro and step4.1_equilibration.cpt) served as the starting point for the production simulations. Five‐hundred nanoseconds of production simulations were performed in the NPT ensemble at 303.15 K and 1 atm on an NVIDIA A5000 GPU node of the Yale Grace high‐performance computing cluster. Following production, solvent‐accessible surface area (SASA) and RMSD calculations were performed using GROMACS analysis tools (gmx rmsf, gmx sasa). To ensure adequate sampling of conformational space, three total production replicas were performed for each construct. Replicas were initiated from the same equilibrated structure but with independently randomized initial velocities generated immediately after the equilibration stage. All replicas were run under identical thermostat, barostat, and integration settings. In total, three 500 ns production trajectories were generated per construct, yielding 1.5 μs of aggregate sampling for a given structure. Replica consistency was assessed by comparing RMSD traces, SASA distributions, and qualitative inspection of alignment overlays across independent trajectories.

4.7. TGM2 crosslinking of recombinant lacritin

For each condition, 1 μM of His‐tagged recombinant glycosylated lacritin (Sino Biological, 15335‐H08H‐100) or unmodified lacritin (Antibodies Online, ABIN7318709) was incubated with 0.33 μM TGM2 (R and D systems, 4376‐TG‐050) in a final reaction volume of 80 μL (1:3 E:S ratio). All reactions were performed in 25 mM HEPES buffer with 20 mM CaCl₂ and quenched by adding 30 mM EDTA and boiling in LDS loading buffer for 10 min. Control samples are prepared identically without TGM2. A total of five time points (5 min, 15 min, 30 min, 60 min, and 120 min) were taken in triplicate for both glycosylated and unmodified lacritin. Proteins were run on a 4–12% Criterion XT BisTris gel (Bio‐Rad, 3450123) in MES XT buffer (Bio‐Rad, 1610789) at 180 V for 50 min, alongside Chameleon Duo Prestained Protein Ladder (LICORbio P/N: 928–60000). Proteins were then transferred onto a nitrocellulose filter paper with a Trans‐Blot Turbo Transfer System (Bio‐Rad, 1704150EDU) and stained with REVERT 700 total protein stain (LICORbio P/N: 926–11015) following manufacturer's protocol. After total protein stain imaging (imaged on a LiCOR Odyssey instrument), a 5% milk blocking solution was allowed to rock with the blot paper for 1 h at room temperature. After blocking, recombinant 6× His lacritin was stained with a fluorescent anti‐His antibody (IR‐800) at a 1:5000 ratio for 1 h followed by four washes with 0.1% PBST and once with PBS before imaging. Densitometry analysis for each replicate was performed on ImageJ using bands stained with anti‐HIS antibody to ensure band intensities corresponded to lacritin. Raw measurement values were exported to Microsoft Excel (Table S1k) and these were used to create corresponding line plots. Graphs were then exported to Adobe Illustrator for further font editing.

Recombinant glycosylated lacritin was also analyzed by LC–MS/MS using the GlycoFASP methods described above with several modifications. After a 60‐min crosslinking reaction, the sample was boiled and quenched with EDTA before dilution with 20 mM tris to a final volume of 400 μL. The sample was then loaded onto a 30‐kDa filter, and the multimer fraction was washed three times to remove CaCl₂ and EDTA before digestion by mucinase SmE (1:10 E:S ratio) overnight and trypsin (1:40 E:S ratio) for 3 h. Finally, peptides were acidified and desalted prior to downstream analysis as previously described.

AUTHOR CONTRIBUTIONS

Vincent Chang: Conceptualization; data curation; methodology; visualization; writing – original draft; writing – review and editing; validation; investigation. Ryan J. Chen: Data curation; formal analysis; investigation; methodology; visualization; writing – review and editing. Isaac Lian: Data curation; formal analysis; visualization. Madilynn Hamilton: Formal analysis; validation; data curation. Keira E. Mahoney: Methodology; supervision. Jeff Romano: Resources; writing – review and editing. Gordon Laurie: Resources; writing – review and editing. Stacy A. Malaker: Conceptualization; writing – original draft; writing – review and editing; funding acquisition; resources.

FUNDING INFORMATION

V.C. is supported by an NSF GRFP (DGE‐2139841). S.A.M. is supported by CRI Lloyd J Old STAR Award and a NIGMS R35‐GM147039.

CONFLICT OF INTEREST STATEMENT

S.A.M. is a co‐inventor on a Stanford patent related to the use of mucinases as research tools. G.W.L. is cofounder and C.S.O. of TearSolutions, Inc.; and cofounder and C.T.O. of IsletRegen, LLC. Other authors declare no competing interests.

Supporting information

Table S1a. Lacritin mucin domain glycopeptides.

Table S1b. CV, AUC, RT, UGP analysis. CV = coefficient of variation based off total intensity, AUC = area under the curve, RT = retention time, UGP = unique glycopeptide identifications.

Table S1c. Ser86, Ser91, and Thr95 glycosylation Donor 3 analysis.

Table S1d. Ser86, Ser91, and Thr95 glycosylation Donor 1 and 2 analysis.

Table S1e. Combined analysis.

Table S1f. Intraglycan–protein H‐bonds over time.

Table S1g. RMSD calculations for three replicates.

Table S1h. Mean glycan‐Lys distances.

Table S1i. SASA values for three replicates.

Table S1j. Lys101 and Lys104 SASA replicate analysis.

Table S1k. Densitometry for in vitro crosslinking.

Table S1l. Recombinant lacritin glycopeptides.

Table S1m. Glycopeptide quantification vs. multimer fraction.

Table S1n. Tear fluid donor information.

Figure S1. Lacritin monomer replicate analysis.

Figure S2. Lacritin multimer replicate analysis.

Figure S3. AlphaFold 3.0 predicted C‐terminal constructs of lacritin.

Figure S4. Starting C‐terminal constructs for MD simulations.

Figure S5. Intraglycan–protein H‐bonds over time.

Figure S6. Glycan‐Lys distances.

Figure S7. Uncropped blots for in vitro crosslinking studies.

Figure S8. Anti‐His staining of recombinant lacritin monomers.

Chang V, Chen RJ, Lian I, Hamilton M, Mahoney KE, Romano J, et al. Site‐specific O‐glycans influence lacritin structure and multimerization in tears. Protein Science. 2026;35(11):e70812. 10.1002/pro.70812

DATA AVAILABILITY STATEMENT

All mass spectrometry data and search results acquired for this manuscript have been deposited on the PRIDE repository. Reviewers can access raw data with the following login information: Project accession: PXD076054, Token: qrOsmudfTPhF. Alternatively, reviewers can access the dataset by logging in to the PRIDE website using the following account details: Username: reviewer_pxd076054@ebi.ac.uk, Password: 7mhpq1TvRHbN. The data that support the findings of this study are openly available in PRIDE at https://www.ebi.ac.uk/pride/, reference number PXD076054.