Microarray analyses of closely related glycoforms reveal different accessibilities of glycan determinants on N-glycan branches

Abstract

Glycans mediate a wide variety of biological roles via recognition by glycan-binding proteins (GBPs). Comprehensive knowledge of such interaction is thus fundamental to glycobiology. While the primary binding feature of GBPs can be easily uncovered by using a simple glycan microarray harboring limited numbers of glycan motifs, their fine specificities are harder to interpret. In this study, we prepared 98 closely related N-glycoforms that contain 5 common glycan epitopes which allowed the determination of the fine binding specificities of several plant lectins and anti-glycan antibodies. These N-glycoforms differ from each other at the monosaccharide level and were presented in an identical format to ensure comparability. With the analysis platform we used, it was found that most tested GBPs have preferences toward only one branch of the complex N-glycans, and their binding toward the epitope-presenting branch can be significantly affected by structures on the other branch. Fine specificities described here are valuable for a comprehensive understanding and applications of GBPs.

Introduction

Glycan-binding proteins (GBPs), which bind specific glycan (or carbohydrate) structures but have no enzymatic activity, decipher the glycan code of cell surfaces and translate information obtained from glycan recognition into functions during various biological processes, including signal transduction, cell adhesion, immune recognition, microbial infection, etc. (Taylor et al. 2015) Given their specific glycan-binding properties, many GBPs (especially plant lectins) have been extensively employed in glycobiology as detection and analysis tools.(Hendrickson and Zherdev 2018) For example, Sambucus nigra lectin (SNA) and Maackia amurensis lectin I (MAL-I) are routinely used to identify α2,6- and α2,3-linked sialic acid (Sia), respectively, and concanavalin A (ConA) is a lectin commonly used to enrich tryptic O-mannosyl peptides. Lectin microarrays have also been developed for rapid analysis of protein glycopatterns and glycomes (Pilobello et al. 2005, 2007). On the other hand, immunotherapies targeting human GBPs and their ligands offer promising therapeutic strategies in combating tumor and other diseases.(Hudak and Bertozzi 2014; RodrIguez et al. 2018) Such applications rely on a complete knowledge of GBPs and their related glycan ligands.

Other than the minimal glycan-recognition motif that defines the primary specificity of a GBP, recent studies revealed that its binding could be significantly affected by the way the motif is presented, including the length of the underlining structures, the location of the branch, the modification and substituents on the neighboring structures, etc.(Taylor and Drickamer 2009; Wang et al. 2013, 2018; Wu et al. 2016) Such fine details are valuable for a comprehensive understanding of GBPs and their applications. While the primary binding feature can be revealed readily using a small number of simple glycoforms with the powerful glycan microarray technology, the fine specificity is harder to determine, which would require libraries of closely related structures. Fewer efforts have been put forth for this purpose. Haab and coworkers described an “outlier-motif” algorithm for automated extraction and systematic analysis of existing glycan array datasets, which uncovered substructure patterns that could provide more binding details of lectins (Maupin et al. 2012). They further reported another strategy to elucidate higher-level lectin binding preference by treating glycan microarray with exoglycosidases before binding assays (Klamer et al. 2017). Other valuable software/algorithms have also been developed to mine refined glycan substructures that influence protein-glycan interactions (Xuan et al. 2012; Kletter et al. 2013; Agravat et al. 2014; Zhao et al. 2015; Grant et al. 2016). All these algorithms are based on systematic analyses of existing microarray result datasets, which should be done with caution concerning diversity on array platforms, conjugation or immobilization strategies and processing protocols (Padler-Karavani et al. 2012; Wang et al. 2014).

Microarray profiling using glycans with similar structural features represents an alternative and more reliable strategy to uncover fine binding specificities of GBPs. For example, microarray analysis using a library of extended airway glycans containing a terminal N-acetylneuraminic acid (Neu5Ac) revealed that currently circulating human H3N2 influenza viruses had evolved to adopt a preference toward branched glycan structures with extended poly-N-acetyllactosamine (poly-LacNAc) chains (Peng et al. 2017). We previously synthesized a number of symmetric bi-antennary N-glycans with tandem epitopes, and microarray analyses revealed fine preferences of GBPs toward terminal and internal epitopes (Wu et al. 2016). N-Glycan positional isomers were also prepared recently for fine specificity mining (Echeverria et al. 2018; Gao et al. 2019). It was also shown that SNA recognized a minimal Neu5Ac-α2,6-LacNAc determinant presented at the terminus of the α1,3-Man branch but not the α1,6-branch of N-glycans (Smith et al. 2010). We recently synthesized a set of O-mannosyl glycans containing closely related structures, and microarray analyses again revealed fine details besides the primary features recognized by lectins and anti-glycan antibodies (Wanget al. 2018). Nevertheless, fine details obtained from all reported cases are limited, focusing on a narrow range of examples and are just the tip of the iceberg. Obtaining a large set of structurally related complex glycans will help to get a comprehensive view of the fine specificities of GBPs and to speed up their applications.

We sought to study the fine specificities of commonly applied GBPs toward glycan determinants that are presented on N-glycans. Specifically, in the present study, we report the preparation of a unique set of closely related glycoforms that differ at the monosaccharide level. Microarray profiling using these structurally well-defined glycoforms revealed fine specificities of GBPs, including their preference toward branch, modifications on underlying glycans, and more.

Results

Preparation of closely related glycoforms and construction of the glycan microarray

N-Glycans, which share a signature core pentasaccharide structure (Man₃GlcNAc₂), are covalently attached to proteins at asparagine (Asn) residues via an N-glycosidic bond. Naturally occurring N-glycans are inherently complex and diverse with variations on the number of antennae and glycan determinants. Interactions between GBPs and asymmetrically branched N-glycans follow a context-dependent recognition manner, (Wang et al. 2013) and glycan asymmetry was believed to play a regulatory role in such interactions (Shivatare et al. 2016). To identify fine glycan-recognition specificities of GBPs, we prepared 68 asymmetric and symmetric N-glycans presenting five major glycan determinants, (Li et al. 2015) including LacNAc (LN, Gal-β1,4-GlcNAc), Neu5Ac-α2,3-LacNAc (3’SLN), Neu5Ac-α2,6-LacNAc (6’SLN), Lewis X [Le^X, Gal-β1,4-(Fuc-α1,3-)GlcNAc] and sialyl-Le^X [sLe^X, Neu5Ac-α2,3-Gal-β1,4-(Fuc-α1,3-)GlcNAc] epitopes (Table I, Figure S1, glycans 1–68). This well-defined glycan library covers all possible bi-antennary N-glycans presenting the five epitopes, including 28 pairs of branch positional isomers. For example, as shown in Figure 1A, glycans 16, 22, 40, 44, 47, 66 and 67 share the 6’SLN epitope on the α1,3-Man branch, but differ on the α1,6-Man branch, with systematic increase of the number of monosaccharides from 0 to 5. In contrast, their corresponding branch isomers 28, 34, 56, 60, 63, 50 and 51 share the same epitope on the α1,6-Man branch but differ on the other branch. Such closely related glycoforms provide an ideal set of standards to probe fine specificities of GBPs.

Table I.

Structures of 81 out of 98 glycans presented on the microarray. The symbolic nomenclature for glycans is shown. A pair of branch isomers defined as two N-glycans (e.g., N1 and N2) where the α1,3-Man branch of N1 and the α1,6-Man branch of N2 share one glycan motif, meanwhile, the α1,6-Man branch of N1 and the α1,3-Man branch of N2 share another glycan motif

Fig. 1.

Examples of N-glycans that share one identical branch but differ in the other (highlighted in red rectangles) (A), and structures of multiantennary, core-fucosylated N-glycans and 8 linear oligosaccharides as control (B). The symbolic nomenclature for glycans is shown.

Additionally, 10 N-glycolylneuraminic acid (Neu5Gc)-terminated N-glycans were prepared to study the binding preferences of GBPs toward the two most abundant sialic acid forms: Neu5Ac and Neu5Gc. Among these, 69–71 were prepared previously, (Li et al. 2015) and 72–78 were enzymatically synthesized following similar one-pot two-enzyme systems (Supplementary Information I). The glycan library also contains four multiantennary and five core-fucosylated structures (Figure 1B, 79–87),(Calderon et al. 2016, 2017) as well as three additional N-glycans (Figure 1B, 88–90) to study the influence of corresponding glycan modifications on protein-glycan interactions. N-glycans 88 and 89 were enzymatically synthesized starting from 14 and 54, catalyzed by bovine α1,3-galactosyltransferase (Bα3GalT) (Fang et al. 1998) and Neisseria meningitidis α1,4-galactosyltransferase (NmLgtC), (Zhang et al. 2002) respectively. Interestingly, unlike Bα3GalT-catalyzed reaction that showed a 100% conversion, only less than 5% of 54 was converted to 89 in the NmLgtC-catalyzed reaction, suggesting that the large underlying N-glycan core structure is unfavorable to NmLgtC, consistent with our previous report (Ban et al. 2012).

All N-glycans, together with 8 ganglioside glycans that contain the five epitopes (Figure 1B), (Yu et al. 2016) are free reducing oligosaccharides. They were individually synthesized via chemoenzymatic approaches, purified by HILIC-HPLC, and characterized by NMR and HRMS (Supplementary information I) (Li et al. 2015; Calderon et al. 2016, 2017). The synthesis of the desired structures (e.g., linkages and anomeric configurations) was controlled by the well-established step-wise synthetic strategy as well as substrate specificities of corresponding glycosyltransferases. All glycosylation products were further tagged with 2-amino-N-(2-aminoethyl)-benzamide (AEAB), (Song et al. 2009) with high yield and further purified by a second porous graphitic carbon-HPLC to ensure homogeneity. After confirmed by mass spectrometry, AEAB-labeled glycans were printed by noncontact printing on N-hydroxysuccinimide (NHS)-activated glass slides at equal molarities (Figure S2). These steps ensured the comparability of the structures for microarray analysis. The guidelines set forth by the MIRAGE (Liu et al. 2016) were followed during sample preparation, microarray printing, analysis and data processing.

Fine specificity of Sia-binding lectins SNA and MAL-I

SNA and MAL-I have been extensively employed to probe Sia-containing glycans. SNA recognizes α2,6-linked Sia with a preference to the 6’SLN motif, whose interactions can be greatly influenced by the underlying structures, (Padler-Karavani et al. 2012) whereas MAL-I binds α2,3-linked Sia, with trisaccharide 3’SLN as the reported minimal recognition motif (Wang and Cummings 1988; Song et al. 2011). Our results showed that both lectins prefer Neu5Gc-containing glycans (Figure 2A, 70, 73, 76, 78 for SNA; Figure 2B, 69, 72, 75, 77 for MAL-I) over their Neu5Ac-containing counterparts (Figure 2A, 10, 16, 40, 4 for SNA; Figure 2B, 9, 15, 39, 3 for MAL-I), indicating that the additional hydroxyl group on C5 of Neu5Gc may play a role in the SNA/MAL-I recognition. On the other hand, SNA exhibited an apparent α1,3-Man branch preference as previously reported (Smith et al. 2010). For example, the relative fluorescence units (RFU) of SNA binding to N-glycans with 6’SLN on the α1,3-Man branch (16, 22, 40, 44, 47, 66, 67) are 3–6 folds higher than those with their branch isomers (28, 34, 56, 60, 63, 50, 51) (Figure 2A), whereas MAL-I has a strong binding preference to the 3’SLN on the α1,6-Man branch, reflected by much higher RFU with 27, 33, 47, 48, 55, 59 compared to those with their branch isomers 15, 21, 63, 64, 39 and 43 (Figure 2B). Another noticeable difference of the two Sia-binding lectins is that SNA prefers complex N-glycans over linear structures such as glycan 97 (Figure 2A), whereas MAL-I prefers linear structures (93, 94, 95, 96) over branched N-glycans (Figure 2B). Most recently, Gao and coworkers reported weak binding of MAL-I toward symmetric Gal-terminating N-glycans, (Gao et al. 2019) which is confirmed in this study (2, 79). Using this asymmetric N-glycan library, an additionally binding specificity observed is that MAL-I binds Gal-terminated epitopes only on the α1,6-branch (e.g., 15, 26, 32, 45, 54, and 86) but not any other branches (e.g., 14, 20, 38, 61, and 81–83) (Figure 2B). Most interestingly, MAL-I binds weakly to biantennary N-glycans presenting the same ligand (LN or 3’SLN) at both branches (2, 3), but the strongest binding to the asymmetric 59 (with LN on the α1,3-Man branch and 3’SLN on the α1,6-Man branch) among all N-glycans tested, suggesting that the lectin may have two different ligand-binding pockets that favor the epitope presentation in 59 (e.g., its branch isomer 43 showed only 1/3 of binding intensities for 59).

Fig. 2.

Selective recognition of N-glycans bearing terminal Sia by SNA (A) and MAL-I (B). The x-axis shows the glycans, and the y-axis shows the relative fluorescence. Readout by Cy5-streptavidin (5 g/mL), NC1 = printing buffer, NC2 = 100 M AEAB in printing buffer, PC = 0.01 mg/mL biotin-PEG-amine, M = Marker (0.01 mg/mL Cy3-conjugated anti-Human IgG + 0.01 mg/mL Alexs647-conjugated anti-Human IgG.

When comparing the bindings toward glycans that share a 6’SLN-containing branch, a gradually decreased influence by motifs on the opposing branch can be concluded (3’SLN > Le^X ≈ LacNAc > GlcNAc), for example, as shown in Figure 2A, RUF with 47 < 66 ≈ 44 < 40 (the glycans share the 6’SLN on the α1,3-Man branch) and RUF with 63 < 50 ≈ 60 < 56 (the glycans share the 6’SLN on the α1,6-Man branch). A similar binding pattern was observed at low concentrations of SNA (e.g., 1 μg/mL, Supplementary Information II). These results suggested that unlike β1,4-galactosylation or α2,3-sialylation, α1,3-fucosylation on the opposing branch may not affect the binding of SNA toward the 6’SLN-containing branch. However, bindings of SNA to 6’SLN on one branch of human milk oligosaccharides (HMOs) were found negatively affected by α1,3-fucosylation on the opposing branch, (Klamer et al. 2017) probably due to disparate spatial conformations of HMOs (β1,2-GlcNAc and β1,6-GlcNAc branches). In some cases, different branch configurations could abolish bindings, as we observed for both SNA and MAL-1 toward an array of 6’SLN- and 3’SLN-containing O-mannosyl glycans (Wang et al. 2018) (data not shown). Similarly, it is also observed that structures on the opposing branch could also affect MAL-I binding toward the 3’SLN-containing branch, but more complicated due to nonspecific recognition toward the LacNAc motif.

This microarray shares seven symmetric bi-antennary N-glycans (2, 3, 4, 5, 6, 77 and 78) with our previously reported platform, (Wu et al. 2016) on which glycans are linked to Asn instead of AEAB. Binding activities of SNA and MAL-I (as well as ECL, anti-CD15 and anti-CD15s antibodies) toward these glycans are consistent on the two platforms, with the only exception that MAL-I exhibited very weak binding toward 3 in this study, whereas a moderate binding activity toward the same N-glycan attached to Asn (BA-04) was observed previously (Wu et al. 2016). Such observations suggest that presenting formats can affect binding activities of certain GBPs toward certain glycans as concluded previously (Padler-Karavani et al. 2012). Thus, binding specificities concluded in current platform may not well applicable to other platforms.

Fine specificity of LacNAc-binding lectins ECL and RCA-I

Erythrina cristagalli lectin (ECL) and Ricinus communis agglutinin (RCA-I) recognize terminal Gal residues, with a strong preference to type II structures (LacNAc, Gal-β1,4-GlcNAc), and modifications on the Gal residue can significantly influence the binding (Itakura et al. 2007; Wu et al. 2007). Such features were confirmed by current microarray, as both lectins showed strong bindings to N-glycans with terminal LacNAc motifs (Figure 3, glycans 8, 14, 20, 26, 32, 38). Surprisingly, neither lectins exhibited significantly increased binding activity to multiantennary LacNAc-containing structures (2, 79, 81, 82), suggesting that the spatial conformation of these structures may only allow proper recognition of one LacNAc motif. In terms of branch preference, ECL slightly favors the α1,3-Man branch (Figure 3A, RFU with 8, 14, 20 > 26, 32), while RCA-I exhibited a stronger preference to the same branch (Figure 3B, RFU with 8, 14, 20 >> 26, 32). Such specificity was also observed against a library of branched O-mannosyl glycans (Wang et al. 2018). Neither GBP could tolerate α2,3-sialylation (3, 9, 15, 21, 27, 33, 39), (Gao et al. 2019) α1,3-fucosylation (5, 11, 17, 23, 29, 35, 41), or α1,3-galactosylation (88). On the other hand, while α2,3-sialylation could also block the binding of ECL, it exhibited a more complicated modulation for RCA-I, with limited influence toward LacNAc on the favorable α1,3-Man branch (RFU with 8, 14, 20, 38 are slightly higher than those with their α2,6-sialylated forms 10, 16, 22, 40, and 76), but significant negative effect on binding toward LacNAc on the unfavorable α1,6-Man branch (28, 34, 56) as well as a linear glycan 97 (Figure 3B).

Fig. 3.

Selective recognition of N-glycans bearing LacNAc disaccharide by ECL (A) and RCA-I (B). Readout by Cy5-streptavidin (5 g/mL).

Lastly, ECL and RCA-I exhibited different patterns concerning the influence of modifications on the opposing branch. A clear negative gradual effect (sLe^X ≈ 6’SLN ≈ 3’SLN > Le^X > GlcNAc) was observed for ECL (RFU with 62 ≈ 60 ≈ 59 < 61 < 38, Figure 3A), whereas RCA-I showed comparable binding activities toward N-glycans sharing one LacNAc-containing branch (e.g., glycans 8, 14, 20, 38, 59–62 share the α1,3-Man branch, and glycans 26, 32, 43, 45, 46 share the α1,6-Man branch, Figure 3B). Collectively, ECL has a very strict requirement toward unmodified the LacNAc motif on N-glycans, whereas RCA-I could well tolerate α2,6-sialylation on the α1,3-Man branch, as well as large structures on the opposing branch of N-glycans.

Fine specificity of man-binding lectins ConA and GNL and other lectins

Given the high affinity and a quite broad specificity to α-linked Man-containing glycans, ConA has been widely used as an analytical and enriching reagent in glycomics (Kuno et al. 2005; Hirabayashi et al. 2015; Zhang et al. 2016; Zou et al. 2017). The lectin is known to bind biantennary and high-mannose N-glycans and could accommodate nonreducing terminal α2,3/6-sialylation (Porter et al. 2010; Gao et al. 2019). Our results revealed Man-α1,6-(Man-α1,3-)Man trisaccharide as a minimum ligand for ConA, and the binding can be significantly affected by further glycosylation on Man residues. It could not recognize structures lacking either α1,3-Man (26–30) or α1,6-Man (14–18, 88) (Figure 4A). GNL recognizes the Man-α1,3-Man motif, (Shibuya et al. 1988) which is confirmed herein by apparent bindings toward glycans 13–17 but not 25–29 which contain a Man-α1,6-Man motif (Figure 4B). An additional α1,6-Man can slightly increase the binding of GNL (19–23). Besides, sialylation and/or α1,3-fucosylation on either α1,3-Man or α1,6-Man branch can reduce (in certain cases, abolish) the binding of GNL (Figure 4B, 19–24, 54–58). Comparing with ConA, GNL exhibited much lower overall binding activity toward N-glycans, with an optimal concentration of 10 μg/mL. As shown in Figure 4A, ConA exhibited much lower binding intensities toward 7 and 90 comparing with those toward 19 and 31, suggesting that the Man-α1,6-(Man-α1,3-)Man trisaccharide on the α1,6-Man branch of N-glycans cannot be recognized by ConA, and could significantly reduce its binding to the core Man-α1,6-(Man-α1,3-)Man trisaccharide. Another noticeable feature is that ConA could not tolerate galactosylation on an additional β1,4-GlcNAc branch (79, 80, 82), the same as a recent report (Gao et al. 2019).

Fig. 4.

Selective recognition of N-glycans by ConA (A) and GNL (B). Readout by Cy5-streptavidin (5 g/mL).

We also investigated fine specificities of the Fuc-binding Aleuria aurantia lectin (AAL) and the GlcNAc-binding Griffonia simplicifolia lectin (GS-II) using this array. As shown in Figure S3, AAL bound to all Fuc-containing glycans (α1,3-Fuc or α1,6-Fuc) with high binding activity at concentrations as low as 0.1 μg/mL. An interesting observation is that α2,3-sialylation on the same branch of the α1,3-Fuc can reduce the binding by 44–61% (Figure S3). In addition, glycans with multiple Fuc residues on separate branches (5, 6, 52, 68, 80) did not display apparent higher binding signals, which is different from the fine specificity of AAL toward HMOs, (Klamer et al. 2017). GS-II recognizes nonreducing terminal GlcNAc residues, (Zhu et al. 1996) as shown in Figure S4. A preference of GS-II toward the α1,3-Man branch was observed (even though with pretty weak bindings), as RFU with 13 and 19 are 4–5 times of those of 25 and 31. It is worth to note that core-fucosylation did not exhibit apparent influence in bindings of all tested lectins, probably because such a modification is spatially far away from their recognized ligands.

Selective recognition of (s)Le^X–containing N-glycans by anti-CD15(s) antibodies and CTB

The anti-CD15 antibody specifically recognizes terminal Le^X epitopes, and modifications on the Gal residues will completely block the binding (Wu et al. 2016). Our previous results using symmetric bi-antennary N-glycans arrays and array data published by Consortium for Functional Glycomics (CFG) highlighted possible negative effects of the core-pentasaccharide (Man₃GlcNAc₂) on bindings (Wu et al. 2016). Similar results were observed in this study, e.g., anti-CD15 antibody exhibited no or only neglectable bindings toward 5, 11, 41, 45, 48, 50, 57, 61, 64, 66 and 80 (Figure 5A). However, it exhibited strong bindings to glycans 23 and 35 (which contain Man₃GlcNAc₂ but lack further glycosylation) at the same level as the linear Le^X-containing glycan 98, suggesting that the negative effect may be caused by opposing branches instead of Man₃GlcNAc₂ of N-glycans. One Man residue on the opposing branch (23, 35) can be tolerated by the anti-CD15 antibody, while the attachment of another GlcNAc (41, 57) or more residues will greatly affect or abolish the binding. The anti-CD15s antibody recognizes sLe^X epitopes, but no binding was observed toward N-glycans with two sLe^X epitopes directly linked to the core-pentasaccharide (Wu et al. 2016). Our results showed that the antibody bound N-glycans lacking one branch (18, 24, and 30) with high activities, and certain bi-antennary N-glycans (36, 42, 46, 49, 51, and 52) with moderate to low activities (Figure 5B), again suggesting that it is the opposing branch that affected bindings instead of the core-pentasaccharide Man₃GlcNAc₂.

Fig. 5.

Selective recognition of N-glycans by anti-15CD (A) and anti-15CDs (B) antibodies. Readout by goat anti-mouse IgG–Alexa Fluor 647 conjugate (5 g/mL).

In addition, higher RFU with 17, 23, and 41 than those with their branch isomers 29, 35, and 57 (Figure 5A) were observed, suggesting that the anti-CD15 antibody slightly favors the α1,3-Man branch. The anti-CD15s antibody also exhibited a preference to the α1,3-branch, as RFU with N-glycans containing the sLe^X epitope on the α1,3-branch (24, 42, 46, and 49) are higher than those with their branch isomers (36, 58, 62, and 65), with the exception of 18/30 (Figure 5B). Nevertheless, our results proved that both antibodies prefer N-glycans with only one branch, and the extension of the opposing branch can significantly affect their bindings.

It was recently reported that the B subunit of Vibrio cholera toxin (CTB) could not only recognize ganglioside GM1 with high affinity, (Kim et al. 2012) but also bind fucosylated cell surface glycoproteins on normal human intestinal epithelia (Wands et al. 2015). As shown in Figure S5, at a high concentration of 50 μg/mL, CTB only exhibited moderate binding to LNFP III (98) among all glycans tested. Interestingly, it showed weak bindings to 4 N-glycans, 29, 30, 35 and 36, which share the same α1,6-Man branch (Le^X or sLe^X) but lack the α1,3-branch. It is possible that (s)Le^X or similar epitopes on glycoproteins may be responsible for CTB binding and play a role in cholera pathogenesis.

Summary

A comprehensive understanding of GBPs and interactions with their ligands can benefit both fundamental research and potential applications. We analyzed a panel of GBPs using a specific glycan microarray that contains 98 closely related glycoforms and branch isomers, revealing common features concerning their fine specificities as summarized in Table II. With the analysis platform we used, all tested GBPs except for AAL have preferences toward one branch of bi-antennary N-glycans, which were also observed previously toward branched O-mannosyl glycans (Wang et al. 2018) and HMOs (Klamer et al. 2017). SNA, ECL, RCA-I, GS-II and anti-CD15(s) antibodies prefer ligands presented on the α1,3-Man branch, whereas MAL-I and CTB prefer ligands on the α1,6-Man branch. Given the common multiantennae property of major mammalian glycans (e.g., N-glycans, O-glycans and HMOs), it may be a critical modulator of protein-glycan interactions which should be carefully considered in related researches. Worth noting that for all GBPs tested, the presence of multiple ligands on separate branches will not increase the binding activity, but in many cases confer negative impact. A possible explanation would be that one GBP molecule may be only able to recognize one specific ligand, and other ligands on opposing or adjacent branches do not serve as a binder, but instead behave as a steric effector.

Table II.

Summary of fine specificities toward N-glycans observed in this study. Gray shading indicates the minimum recognition motif (ligand) of corresponding GBPs. Branch preferences, ++ means apparent preference, + means slight preference. NA, not applicable. The symbolic nomenclature for glycans is shown

While it is well accepted that the underlying structures and modifications of the glycan ligand can modulate bindings, the influence by structures on opposing branches of N-glycans has been rarely discussed (Klamer et al. 2017; Wu et al. 2017; Gao et al. 2019). As summarized in Table II, one N-glycan branch can significantly affect (usually in a negative manner) GBP recognition of the ligand on other branches, depending on the branch length and the modifications it bears. Such a “masking effect” by the opposing branch is GBP-specific, as the binding can change slightly (e.g., AAL), moderately (e.g., SNA), significantly or even be abolished (e.g., anti-CD15 and anti-CD15s antibodies). It should be mentioned that the masking effect may not occur on glycans with elongated branches, (Peng et al. 2017) given likely extended and/or flexible distances between terminal ligands on different branches. Further analyses should be performed using such structures as they are frequently found on mammalian cells especially cancer cells (Pinho et al. 2015). As for same branch modifications, ECL and GS-II cannot tolerate any glycosylation, while AAL and GNL could accommodate α2,3/6-sialylation, α1,3-fucosylation and β1,4-galactosylation. Interestingly, RCA-I could only tolerate α2,6-sialylation when the ligand is located on the α1,3-Man branch. As reported previously, (Gao et al. 2019) core-fucosylation can be well-tolerated by all tested GBPs.

In summary, protein-glycan interactions contain much more details than their primary binding features, which have been largely ignored. Uncovering such details requires both experimental platforms with massive well-defined glycans to acquire meaningful datasets and algorithms to interpret them. Our glycan microarray with closely related structures and isomers provides valuable insights into fine specificities of GBPs. It is worth noting that all glycoforms are presented in one platform (with AEAB labeling) for comparability, and the specificities may not be well applicable to other platforms (Padler-Karavani et al. 2012). Further analyses using these glycans with other aglycones and on other microarray platforms will be performed in the future.

Materials and methods

Materials

Biotinylated Aleuria aurantia lectin (AAL), Concanavalin A (ConA), Maackia amurensis lectin I (MAL-I), Sambucus nigra lectin (SNA) and Ricinus communis agglutinin I (RCA-I) were purchased from Vector Laboratories (Burlingame, CA). Biotinylated Galanthus nivalis lectin (GNL), Erythrina cristagalli lectin (ECL) and Griffonia simplicifolia lectin II (GS-II) were purchased from E Y Laboratories (San Mateo, CA). Mouse monoclonal anti-CD15 antibody and biotinylated cholera toxin B subunit (CTB) were purchased from Sigma (St. Louis, MO). The mouse monoclonal anti-CD15s antibody was purchased from Santa Cruz Biotechnology (Dallas, TX). Streptavidin-Cy5 conjugate and goat anti-mouse IgG–Alexa Fluor 647 conjugate were purchased from Thermo Fisher Scientific (Waltham, MA).

Glycan preparation

The majority of N-glycans (1–71, 79, and 81–87) and the eight gangliosides were synthesized and purified previously by us (Li et al. 2015; Yu et al. 2016; Calderon et al. 2017). Neu5Gc-terminated N-glycans 72–78 were prepared as described previously using one-pot two-enzyme systems, starting with 2, 14 or 28, respectively. The systems contain Neu5Gc, CTP, MgCl₂, N. meningitidis CMP-sialic acid synthase (NmCSS) and α2,3-sialyltransferase 1 mutant E271F/R313Y from Pasteurella multocida (PmST1m) or α2,6-sialyltransferase from Photobacterium damselae (Pd2,6ST). Glycan 80 (0.3 mg) was synthesized via the reaction catalyzed by Helicobacter pylori α1,3-fucosyltransferase (Hp3FT) in the presence of 79 and GDP-fucose as described before (Li et al. 2015). Glycan 88 (0.1 mg) was synthesized via the bovine α1,3-galactosyltransferase-catalyzed reaction starting from 14 in the presence of UDP-galactose, as described previously (Wu et al. 2016). Glycan 89 (0.05 mg) was synthesized via the reaction catalyzed by N. meningitidis α1,4-galactosyltransferase (NmLgtC) (Zhang et al. 2002) in the presence of 54 and the sugar donor UDP-galactose. Newly synthesized N-glycans were purified by HPLC as described before, (Li et al. 2015) using a Waters XBridge BEH amide column (130 Å, 5 μm, 10 mm × 250 mm) under a gradient running condition (solvent A: water or 100 mM ammonium formate; solvent B: acetonitrile; flow rate: 4.5 mL/min, B%: 65–50% in 30 min), monitored by UV absorbance at 210 nm. Product-containing fractions were pooled and lyophilized for mass spectrum (MS) and NMR analysis to confirm structures. Glycan 90 (Man5) was purchased from Sigma.

Glycan derivatization and quantification

All glycans are with free reducing-end and were derivatized by reductive amination using 2-amino-N-(2-aminoethyl)-benzamide (AEAB) as previously described (Song et al. 2009). Labeled glycans were further purified by HPLC to homogeneity using a porous graphitic carbon column (5 μm, 4.6 mm × 150 mm) under a gradient running condition (solvent A: 0.1% TFA in water; solvent B: 0.1% TFA in acetonitrile; flow rate: 1 mL/min, B%: 15–45% in 30 min), monitored by UV absorbance at 330 nm. Product-containing fractions were pooled and lyophilized for MS characterization. To quantify labeled glycans, 5 mg of AEAB-labeled LNnT was dissolved into various concentrations (1, 5, 10, 20, 50 and 100 μM), and equal volumes (2 μL) of each concentration were loaded into HPLC using the porous graphitic carbon column (same condition as described above). A standard curve of peak area vs. concentration was drawn accordingly, and all purified AEAB-labeled glycans were quantified using the standard curve.

Microarray fabrication

The labeled-glycans were prepared at a concentration of 100 μM in the printing buffer (300 mM phosphate, pH 8.5), and printed on multivalent NHS-derivatized microscope-glass slides (Z Biotech, LLC), each for 400 pL in replicates of six, as described previously (Heimburg-Molinaro et al. 2011). Noncontact printing was performed at room temperature with a humidity of 60% by a sciFLEXARRAYER S3 spotter (Scienion) with two PDC 80 Piezo Dispense Capillary, and eight subarrays were printed on each slide. After overnight dehumidification under room temperature, the slides were washed with MilliQ water and subsequently blocked with 50 mM ethanolamine in 100 mM Tris buffer (pH 9.0) for 2 h. The blocked slides were then washed with MilliQ water twice, dried and stored desiccated at −20°C until use.

Microarray binding assays

Slides were fitted with ProPlate 8-well microarray modules to divide into subarrays and then rehydrated for 10 min with 200 μL of Buffer TSMTB (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM CaCl₂, 2 mM MgCl₂, 0.05% (v/v) Tween-20, 1% (w/v) BSA) at room temperature. Next, the buffer was aspirated and 200 μL of GBP samples were added into each subarray, sealed and incubated at room temperature for 1 h with gentle shaking. Slides were then washed with Buffer TSMT (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM CaCl₂, 2 mM MgCl₂, 0.05% (v/v) Tween-20) four times (3 min per time) and Buffer TSM (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM CaCl₂, 2 mM MgCl₂) four times (3 min per time). Next, slides were added with secondary reagents (200 μL per subarray), sealed and incubated at room temperature for 1 h with gentle shaking. Finally, slides were washed four times each with TSMT, TSM and MilliQ water, respectively, and dried by brief centrifugation. All samples and secondary reagents were dissolved to appropriate concentrations (as shown in Table SI) in Buffer TSMTB. Slides were scanned at10 μm resolution with a Genepix 4100A microarray scanner (Molecular Devices Corp., Union City, CA) using 500 or 600 PMT gains and 80% power, and image analyses were carried out with Genepix Pro 6.0 analysis software as previously reported (Padler-Karavani et al. 2012). Spots were defined as circular features with a variable radius as determined by the Genepix scanning software, and local background subtraction was performed. Microarray data resulted from optimal concentrations of each GBP were used to interpret their fine specificities. The optimal concentration of each GBP (Table SII) was chosen to make sure relative binding values are in the linear range of the scanning instrument.

Funding

This work was supported by National Institute of General Medical Sciences (U01GM125288 and U01GM116263) and National Heart, Lung, and Blood Institute (U54HL142019).

Declarations of interest

The authors declare that there are no competing interests associated with the manuscript.

Abbreviations

3’SLN

Neu5Ac-α2,3-LacNAc

6’SLN

Neu5Ac-α2,6-LacNAc

AEAB

2-amino-N-(2-aminoethyl)-benzamide

Fuc

L-fucose

Gal

galactose

GBP

glycan-binding protein

GlcNAc

N-acetylglucosamine

LacNAc

Gal-β1,4-GlcNAc

Le^X

Gal-β1,4-(Fuc-α1,3-)GlcNAc

Man

mannose

Neu5Ac

N-acetylneuraminic acid

Neu5Gc

N-glycolylneuraminic acid

NHS

N-hydroxysuccinimide

Sia

sialic acid

sLe^X

Neu5Ac-α2,3-Gal-β1,4-(Fuc-α1,3-)GlcNAc

TFA

trifluoroacetic acid