Proximity labeling technologies to illuminate glycan-protein interactions

Abstract

Glycosylation is a ubiquitous post-translational modification read by glycan-binding proteins (GBP) that encodes important functions, but a robust understanding of these interactions and their consequences can be challenging to uncover. Glycan-GBP interactions are transient and weak, making them difficult to capture, and glycosylation is dynamic and heterogenous, necessitating study in native cellular environments to identify endogenous ligands. Proximity labeling, an experimental innovation that labels biomolecules close to a protein of interest, has recently emerged as a powerful strategy to overcome these limitations, allowing interactors to be tagged in cells for subsequent enrichment and identification by mass spectrometry-based proteomics. We will describe this nascent technique and discuss its applications in the last five years with different GBP classes, including Siglecs, galectins, and non-human lectins.

Graphical Abstract

Scheme of general proximity labeling approach for glycan-binding proteins.

A proximity labeling agent (PLA) is targeted to a glycan-binding protein interacting with its endogenous glycoprotein counter-receptor in live cells. Upon activation of the PLA, proteins near the glycan-binding protein are labeled with a tag. This tag is usually biotin, which can be used to enrich labeled proteins for identification via mass-spectrometry.

Introduction

The case for proximity labeling

Cell surfaces are coated by the glycocalyx, the dense collection of proteins and lipids molecules decorated with carbohydrate moieties called glycans. These glycoconjugates can serve as interactors for glycan-binding proteins (GBPs). As glycoconjugates and GBPs are often interchangeably referred to as ligands and receptors, for clarity we will refer to the protein carrier that has been post-translationally modified with glycans as the counter-receptor, and the proteins that bind to the glycan units as GBPs. GBPs serve important biological roles, including immunological regulation, cell adhesion, and differentiation [1, 2]. Thus, a thorough study of the binding events between GBPs and their glycoconjugate counter-receptors can prove useful in understanding a myriad of cellular processes and disease mechanisms.

Despite the ubiquity of glycosylation and its importance, the study of glycan-GBP interactions has been thwarted by the inherent challenges of studying these binding events in cells. Glycosylation is a dynamic post-translational modification that is not directly templated by the genetic code. While putative sites of N-linked glycosylation can be computationally predicted from consensus sequences, glycan occupancy is not guaranteed [3, 4]. Further, glycan modifications are highly heterogeneous. There are multiple carbohydrate monomers that can be linked to each other at many positions with different stereochemistry, and with varying degrees of branching. These variations are critical considerations in glycan-GBP interactions, as GBPs show distinct binding preferences for the identity and linkage of glycans [5]. Furthermore, glycan-GBP binding is often transient, occurring with k_off < 0.01 s⁻¹ [6], and they are thermodynamically weaker than protein-protein interactions (K_D ~ mM) [7]. As such, multivalency (from either the GBP or glycan), where several binding interactions occur simultaneously, is often required to elicit biologically relevant functions [8]. This accumulation of multiple, non-covalent interactions results in high avidity events that enhance the overall apparent affinity of GBP-glycan interactions [9-11]. Whereas much of glycoscience has been occupied with understanding the glycan structural requirements involved in binding GBPs, there is reason to place the preservation of native multivalency at the forefront of designing binding assays. Standard biochemical approaches, like co-immunoprecipitation from disrupted cell lysates, results in identification of few, if any, GBP interactors [12]. Further, lysate-based methods like immunoprecipitation disrupt the natural organization of biomolecules and can result in the identification of false counter-receptors, as interactions usually inaccessible are artificially generated [13, 14].

The recapitulation of the endogenous, multivalent presentation and diversity of glycans in vitro is immensely challenging, but advances in chemical synthesis and new tools have emerged to systematically study the tolerances of GBPs towards various glycan architectures. For example, well-defined glycan structures can be printed onto arrays [15, 16] or biologically presented in genetically modified bacteriophages and mammalian cells to measure GBP binding [17-20]. The largest limitation of these formats, however, is the inability to study the glycoconjugate itself within a native cellular environment. Extracellular glycan-GBP interactions can elicit responses that vary with the identity of the interactor [21], and the presentation of a glycoconjugate, including its localization on the cell surface and the identity of neighboring proteins, can affect GBP binding by altering the clustering and multivalency of presented ligands [22]. Thus, there is a growing appreciation to study glycan-GBP interactions in endogenous environments to accurately assess binding.

Proximity labeling is a method that tags biomolecules (usually proteins) neighboring a protein of interest to identify putative interactors. This protein of interest is often termed the “bait,” as it delivers a proximity labeling agent (PLA) to its apprehended targets. Developed over a decade ago, and originally intended for the study of protein-protein interactions [23] or mapping the constituents of specific cellular compartments [24, 25], proximity labeling technologies have typically co-opted non-human enzymes to promiscuously biotinylate proteins within a short radius. These biotinylated proteins can then be enriched with streptavidin and identified via mass spectrometry (MS)-based proteomics.

Glycoscientists have recognized that this system is particularly apt as a tool to overcome the inherent challenges of studying glycan-GBP interactions. Proximity labeling can be used to convert the low affinity, transient interactions of glycans and GBPs to covalent biotinylation, effectively taking a snapshot of the dynamic interactions in the cellular milieu. Advantageously, native interactions are preserved with spatiotemporal accuracy. Importantly, proximity labeling can also be performed in live, non-genetically modified cells, where the native glycosylation of proteins and the endogenous presentation of glycoprotein counter-receptors are preserved. Thus, biologically relevant interactors can be probed in a cellular context with preservation of multivalency and cellular architecture. Glycan-GBP focused proximity labeling methods share a similar strategy: allow a GBP bait to bind to its glycoprotein counter-receptors and then employ a proximity labeling agent (PLA) to tag the interactome with a biotin moiety in situ for enrichment, followed by western blotting or mass spectrometry (MS)-based proteomic identification. In general, PLAs biotinylate proteins near a molecular bait, but they can vary in the catalytic moiety, which inherently affects the radius of biotinylation, duration of biotinylation, and the degree to which multivalency is preserved (Fig. 1).

Figure 1: Comparison of proximity labeling agents.

Proximity labeling agents (PLAs) vary in their size and labeling chemistries, inherently affecting the labeling radius, the reaction time for tag addition, and the identity of the tag deposited. These parameters should be carefully considered in proximity labeling experiments to preserve the native binding of glycan-binding proteins and to minimize promiscuity of labeling. (Abbreviations: horseradish peroxidase – HRP, engineered ascorbate peroxidase – APEX2, iron (S)-1-(p-bromoacetamidobenzyl) EDTA – Fe-BABE, adenosine monophosphate – AMP)

Proximity labeling approaches were first applied to illuminate glycan-GBP interactions in 2017, and in this Review, we will discuss such examples over the last five years, emphasizing the applicability of this approach towards various GBPs. We have organized our discussion based on unique classes of GBP, as they are known to engage glycoprotein counter-receptors in different ways, and proximity labeling strategies are usually adapted to minimize perturbations in these binding events. We also highlight the future of this field, as proximity labeling for glycan-GBP interactions is still in its infancy and promises to be a significant tool for discovering more precise glycan-mediated biology.

Sialic acid-binding immunoglobulin-type lectins (Siglecs)

Sialic acid-binding immunoglobulin-type lectins (Siglecs) are immunomodulatory proteins that preferentially bind to sialic acid glycans [26]. Most Siglecs possess an intracellular immunoreceptor tyrosine-based inhibitory motif (ITIM) that recruits phosphatases after ligand engagement to dampen immune responses. These interactions are known to contribute to cancer cell immune evasion and autoimmunity; however, a mechanistic understanding of Siglec-mediated immune suppression is lacking in many contexts, as the counter-receptors are still relatively unknown [26-28].

In 2017, two groups independently used horseradish peroxidase (HRP) as the PLA to investigate the binding of Siglecs in cells. HRP is a metalloenzyme with a heme active site that generates free radicals upon incubation with hydrogen peroxide (H₂O₂). Upon co-incubation with biotin tyramide and H₂O₂, HRP catalyzes the formation of biotin tyramide radicals. These are short-lived species that rapidly react with electron-rich residues with a labeling radius of approximately 20 nm. The catalytic efficiency of HRP allows for short labeling times (< 10 min) that are important consideration for achieving temporal resolution. Wu et al. examined the Sialoadhesion (Siglec-1) counter-receptors on human erythrocytes [29] using a tripartite fusion protein composed of Siglec-1, HRP, and a human Fc domain. This construct was pre-complexed with an anti-Fc antibody to induce protein oligomerization, mimicking the clustering of Siglecs to promote cell surface binding. The authors confirmed that a known interactor of Siglec-1, Glycophorin A, was biotinylated by western blotting [30], and they orthogonally confirmed their results with a second proximity labeling method where HRP was introduced as a conjugate to an anti-Fc antibody, directing the PLA to a Siglec-1-Fc construct lacking its own HRP. Biotinylation of Glycophorin A was still observed, suggesting that both the peroxidase chimera and the peroxidase-conjugated antibodies could be used for proximity labeling. This work established the foundation for proximity labeling for GBPs but was not extended to MS-based proteomics. Chang et al. concurrently employed a similar strategy for Siglec-2 interactors in B cells [31]. They used an HRP conjugated anti-FLAG antibody to localize the PLA to a FLAG-tagged Siglec-2-Fc protein. Again, an Fc domain was used to promote the multivalency required for Siglec binding. After proximity labeling, samples were enriched for MS-based proteomics, resulting in the identification of new and previously known interactors. In the same work, they also characterized the Siglec-15 interactome in macrophages via proximity labeling, identifying the first known interactors of Siglec-15, including CD44. Further functional validation was performed with CD44 knockdowns, which demonstrated decreased binding of Siglec-15 and hampered cell fusion, an essential feature of osteoclast differentiation. Importantly, this work demonstrated the potential of proximity labeling integrated with MS-based proteomics to uncover new counter-receptors for GBPs. An important consideration in proximity labeling approaches is the radius of labeling. Using a direct system where the bait GBP and HRP are fused together versus a secondary system where a subsequent HRP incubation step (e.g. using an HRP-conjugated antibody) is performed can yield different labeling radii, and the latter approach can extend the labeling radius to ~200 nm and increase the likelihood of labeling proteins that are not direct interactors or counter-receptors [32].

With these foundational data, other groups shortly adopted proximity labeling for Siglecs studies. HRP-antibody conjugates were used for proximity labeling of Siglec-7 counter-receptors on K562 lymphoblasts to understand natural killer (NK) cell immune responses [33]. As expected for glycan-GBP interactions, co-immunoprecipitations of Siglec-7 and putative interactors by Yoshimura et al. showed inconsistent pull-down of proteins of interest. Thus, proximity labeling was performed to capture these transient interactions, which confirmed that CD43 is a ligand of Siglec-7. Subsequent cellular studies showed that CD43 expressed on K562 cells bound to Siglec-7 on NK cells to attenuate NK cytotoxicity, again emphasizing the power of proximity labeling to identify functional GBP ligands. This finding was independently confirmed the same year through a CRISPR screen conducted by the Bertozzi group, providing confidence that proximity labeling can uncover relevant interactors of GBPs [34].

In the previous examples, exogenous Siglec proteins were added to live cells to identify interactors, but a critical consideration for the Siglec class of proteins is their ability to not only have interactions with external stimuli (trans interactors) but also ligands on their own cell surface (cis interactors). Indeed, previous work has shown distinct binding preferences for cis and trans ligands [14, 35]. To capture this nuance, a cis-interaction proximity labeling platform was generated using HRP conjugated antibodies directed to endogenously expressed Siglec-2 [36]. A 2,6-sialyltransferase knockout was used as a control to delineate glycan-mediated events and those that do not require sialic acid. Interestingly, Western blotting confirmed that CD45 and Siglec-2 itself was biotinylated regardless of sialic acid, a finding supported by previous work [37]. This finding emphasizes that proximity labeling tags neighboring proteins of bait GBPs, whether engaged in a glycan-mediated interaction or protein-protein interaction, necessitating the use of appropriate controls for discrimination.

Galectins

Our group applied proximity labeling to another class of GBPs called Galectins. Galectins are a class of soluble lectins that preferentially bind to β-galactoside motifs [38]. These GBPs are known regulators of immune responses [39], liver fibrosis [40], and myogenesis [41]; however, the molecular mechanisms underlying Galectin-mediated biology remain poorly understood. Unlike monomeric Siglecs with anchoring transmembrane domains, Galectins can oligomerize and can be secreted extracellularly [42]. Thus, some understanding of the effects of the PLA tag towards oligomerization and secretion is important for this class of GBPs. We generated fusion proteins of a peroxidase and Galectins of interest and showed that normal oligomerization was not perturbed by the addition of the PLA. Instead of HRP, we used another heme-containing metalloenzyme, the engineered ascorbate peroxidase APEX2, as it is uniquely tolerant to both the extracellular oxidative and the reducing intracellular environments [43], a feature that is important for studying Galectins which can exhibit both types of interactions [44]. Like HRP, APEX2 affords short labeling times (~ 1 min) for high temporal resolution, which we envisioned could be useful for examining how Galectin interactomes change as they are endocytosed. APEX2 chimeras can be overexpressed by transfection, allowing for a comparison of interactomes based on subcellular localization, simply by varying between exogenous incubation and transfection. This new method was first optimized with APEX2-fused Galectin-3 in hepatic stellate cells (Fig. 2A), [45] followed by APEX2-fused Galectin-1 in myoblasts [46]. In both instances, we used quantitative MS-based proteomics to confirm known interactors and discover new ones (Fig. 2B). During this work, we have found that it is applicable to many cell types. Importantly, however, we have found that the optimization of protein expression, protein concentrations, and competition during labeling are key towards achieving success, which can be optimized via microscopy (Fig. 2C).

Figure 2: Proximity labeling with APEX2 to capture the Galectin-3 interactome.

(A) A scheme of galectin-based proximity labeling strategy shows the incubation of the Galectin-3 APEX2 fusion protein on the cell surface of live cells, followed by incubation with biotin phenol and hydrogen peroxide. This method results in the biotinylation of cell surface proteins, which can be visualized with immunofluorescence microscopy or enriched for mass spectrometry-based proteomics. (B) The interactome of galectin-3 identified by mass spectrometry is represented as a plot of the proteins enriched by proximity labeling and those competed by addition of lactose. Putative galectin-3 interactors of interest are highlighted in orange. (C) Proximity labeling was performed and biotinylation was observed by microscopy using streptavidin-Cy5. Dose-dependent labeling is observed, as well as competition with lactose to show glycan-mediated interactions. (B) and (C) reprinted with permission from E. Joeh, T. O’Leary, W. Li, R. Hawkins, J.R. Hung, C.G. Parker, and M.L. Huang, Mapping glycan-mediated galectin-3 interactions by live cell proximity labeling, PNAS. 117 (2020) 44, 27329-27338. Copyright 2020 National Academy of Sciences.

Non-human lectins

Proximity labeling has recently been performed with non-human lectins that have been repurposed as scouting proteins. Unlike studies with Siglecs and galectins in which the primary goal is to elucidate unknown counter-receptors, proximity labeling with non-human lectins is being used to analyze native glycosylation patterns on glycoproteins. Herculean efforts have been made to thoroughly characterize the glycan binding preferences of lectins to glycans, often with computational studies, glycan arrays, or a combination of both [47-49]. Traditionally, these lectins are used to identify global changes in glycosylation, but their use in proximity labeling has proven useful to identify changes in glycosylation of individual proteins.

The first use of a non-human lectin for proximity labeling was performed with Maackia amurensis lectin (MAL II), a plant lectin known to bind α-2,3 sialic acids [31]. MAL II was biotinylated and incubated with erythrocytes followed by the addition of streptavidin-HRP to label nearby proteins with biotin upon introduction of H₂O₂ and biotin tyramide. This work laid the foundation for the application of non-human lectins to study other glycosylation motifs, including the plant lectin PHA, which recognizes the terminal galactose residues of complex N-glycans [50]. Biotinylated PHA was used in the SUGAR-seq method to elucidate changes in glycosylation of individual cell surface proteins [51]. Using streptavidin-HRP to perform proximity labeling with biotinylated PHA, Kearney et al. showed that T cells derived from healthy and tumor positive mice had distinct glycosylation profiles. Interestingly, cells could be sorted into groups of high and low PHA binding, which correlated with the degree of chromatin accessibility and differentiation state, underlining the importance of glycosylation in cell function and state. Further, glycoproteins implicated in adhesion, migration, and activation showed changes in their enrichment in high and low glycosylation groups, suggesting changes in glycosylation which could affect their ability to serve as receptors for GBPs.

Another method called GlycoID uses proximity labeling to detect O-GlcNAc modifications [52]. Unlike the peroxidase-based PLAs, an engineered biotin ligase called TurboID was used [53, 54]. This enzyme uses ATP to convert biotin into AMP-biotin, which is then appended to nearby, accessible lysine residues, covalently tagging proteins without H₂O₂. This facet renders TurboID-based methods less toxic than HRP- and APEX2-mediated approaches, with longer labeling times as a trade-off. TurboID requires minute to hours for labeling, as it does not use radical propagation like the peroxidase-based PLAs, but it is tolerant of multiple cellular compartments like APEX2. To detect O-GlcNAc sugars, the authors generated a fusion construct of GafD (a bacterial lectin that recognizes O-GlcNAc [55]) and TurboID, which was localized to the nucleus and cytosol to reveal different hubs of O-GlcNAc-ylation. Advantageously, this platform was used to assess physiological changes over time. For example, they subjected their cells to starvation or insulin stimulation and observed distinct changes associated with RNA splicing, metabolism, and signaling. The temporal and spatial resolution of this technique underscores the utility of proximity labeling to measure the dynamics of glycosylation.

Not all proximity labeling approaches use enzyme-based PLAs. Instead, the Lebrilla group developed Lectin PROXL, which uses click chemistry to functionalize lectins with a small molecule-based PLA. Lysine residues on lectins are functionalized with NHS-PEG-azide followed by conjugation to the PLA, a dibenzocyclooctyne (DBCO)-modified iron (S)-1-(p-bromoacetamidobenzyl) EDTA (Fe-BABE) moiety [56]. This motif mimics the heme active site of the metalloproteins HRP and APEX2 to oxide proteins in the presence of H₂O₂. The oxidation of proteins can then be detected by mass spectrometry, effectively labeling interactors for identification without depositing an enrichment tag like biotin. Because Fe-BABE is approximately 2 kilodaltons in size and catalyzes radical formation, the labeling time and radius are smaller than GlycoID. Lectin PROXL was used to probe the specificity of non-human lectins, but theoretically Lectin PROXL could be used to identify Galectin and Siglec interactors, as the lectin functionalization step only requires a free lysine residue. They found high correlation between oxidation site and putative glycosylation sites, bringing confidence to their identified targets and confirming that the smaller radius of Fe-BABE can yield higher resolution to functional hits. In general, they identified that sialic acid- and fucose-binding lectins had greater specificity and selectivity than other lectins, an important consideration for glycobiologists that commonly use these probes to analyze glycosylation patterns. One limitation of this method is that biotin is not used for enrichment, so cell membrane proteins are isolated by centrifugation to detection these relatively low abundance proteins compared to the whole proteome. It is possible that this separation introduces contaminating intracellular proteins and incomplete capture of the cell membrane, leading to less coverage than biotin-based enrichment strategies.

The future of glycan directed proximity labeling and considerations

In the past five years, the utility of proximity labeling has begun to be appreciated by the glycobiology community. GBP ligands once undetectable by traditional biochemical techniques due to weak, transient interactions unamenable to traditional biochemical approaches are now identifiable in their native cellular context with an array of proximity labeling agents (Figure 3). We can also use proximity labeling to characterize the heterogeneous glycocalyx, which could emerge as a powerful tool to map changes in disease contexts. For example, cancer cells are hypersialylated to evade immunological surveillance, but the proteins that gain this glycan modification are relatively unknown. A comparison of normal and cancerous cells using proximity labeling with sialic acid-binding non-human lectins could reveal what proteins are excessively sialylated, allowing for further study of their role in immune evasion. While the current strategies have been directed towards GBPs, there is also the potential to expand proximity labeling to glycoconjugates, like proteoglycans [57].

Figure 3: Representation of current proximity labeling approaches applied to live cells.

Glycan-binding proteins (GBP) of interest are allowed to bind to their glycoconjugate counterreceptors in live cells on the cell surface or intracellularly. A proximity labeling agent (PLA) is directed to this GBP through various strategies, including streptavidin-horseradish peroxidase (HRP), HRP-conjugated antibodies, and direct tethering of chimeric constructs with HRP, engineered ascorbate peroxidase (APEX2), TurboID, or iron (S)-1-(p-bromoacetamidobenzyl) EDTA (Fe-BABE). Activation of the PLA results in the tagging of proteins near the GBP, which can be identified by mass spectrometry-based proteomics.

We also note that, while not within the scope of this review, there is also a recent effort to capture GBP-glycoprotein interactions by derivatizing cell surface glycans. Such incorporation of derivatized sugars, called metabolic oligosaccharide engineering, is another method for identifying GBP interactions for which we will provide key examples but will not cover exhaustively, as there are other reviews focused on this topic [58, 59]. For example, the Paulson group has used azide-modified sialic acids to detect Siglec ligands by MS-based proteomics [14, 35]. Additionally, the Kohler group has used diazirine-modified N-acetylglucosamine (GlcNAc) to crosslink GBPs and their ligands for MS-based proteomic identification of galectin-1 ligands [13]. As a final example, the Lebrilla group used azide-modified monosaccharides to click on the Fe-BABE moiety to glycoprotein ligands for subsequent oxidation of neighboring putative GBP interactors [60]. As an extension of this work, GBPs were modified with cyclooctyne to crosslink GBPs directly to the azide-modified sugars on glycoproteins to identify counter-receptors [61]. These approaches depend on the successful incorporation of derivatized monosaccharides by the endogenous glycosylation machinery of cells, which varies by cell type. To overcome this limitation, the Kohler and Capicciotti groups concurrently published an exo-enzymatic strategy to introduce diazirine-modified sialic acids to the cell surface [62, 63]. Here, a neuraminidase cleaves sialic acids off the cell surface and diazirine-modified sugars are subsequently added with a sialyltransferase to photocrosslink GBPs and their counter-receptors.

There are important caveats to consider when performing proximity labeling experiments. First, iron and peroxidase-based labeling (while rapid and efficient) is toxic in vivo, restricting these approaches to cultured cells. The rise of enzyme evolution and engineering has yielded efficient biotin ligases (like TurboID used for GlycoID), but the true scope of glycan proximity labeling in live animals has yet to be explored. Second, a growing concern in the field of proximity labeling is the delineation between an interactor and a functional counter-receptor. Due to the labeling radius of PLAs, there is a significant probability that tagged proteins are not relevant binding partners. Shortening incubation time or changing the PLA can mitigate off-target labeling. The use of soluble glycan competitors, inhibitors of the glycosylation machinery, and lectins with active site mutations are standard controls for proximity labeling and are essential for parsing out non-glycan mediated interactors, but better controls (e.g. anti-epitope antibodies) can increase confidence that biotinylated proteins are indeed functional interactors. To further filter hits, we envision the coupling of glycan-GBP proximity labeling with peroxidase-based cell surface enrichment techniques to determine the specificity of a GBP interaction relative to the abundance of the protein on the cell surface [64]. Third, the method of proximity labeling can also affect the amenability of tagged proteins to various proteomics workflows. During peroxidase- and TurboID-based methods, biotin is covalently linked to proteins that are enriched via streptavidin beads. Elution of these proteins is usually performed by protease digestion, resulting in the release of peptides for MS identification. In this case, biotinylated portions of the protein still remain bound to the bead. Therefore, proteomics databases can be searched without needing to add the mass of the biotin tag to the expected masses of peptides. However, biotinylated portions of the protein are trapped to the beads, which theoretically includes the region of the protein closest to the PLA, likely including the glycosylation site. Thus, standard biotin tyramide is optimal for proteomics experiments, rather than glycoproteomics. For glycoproteomic experiments, cleavable biotin linkers (usually acid- or photo-labile) can be used to release whole proteins from the streptavidin beads before protease digestion, preserving glycan information. Database searches will have to use search parameters that include the mass of the tag remaining after cleavage. Alternatively, the Lectin PROXL method is amenable to proteomics and glycoproteomics because biotin is not used for subsequent bead-based enrichment. Here, the search parameters for raw MS interpretation are paramount to distinguish oxidized and non-oxidized proteins, as GBP interactors are identified by this oxidation. Finally, after producing a list of putative interactors via proximity labeling, a significant amount of biochemical leg work must be performed to validate this binding event and characterize its biological relevance. We posit that collaborations between glycobiologists and bioinformaticians can improve the identification of functionally relevant targets for validation studies.