Cell-based glycan arrays for probing glycan–glycan binding protein interactions

Glycan microarrays provide a high-throughput means of profiling the interactions of glycan-binding proteins with their ligands. However, the construction of current glycan microarray platforms is time consuming and expensive. Here, we report a fast and cost-effective method for the assembly of cell-based glycan arrays to probe glycan–glycan-binding protein interactions directly on the cell surface. Chinese hamster ovary cell mutants with a narrow and relatively homogeneous repertoire of glycoforms serve as the foundation platforms to develop these arrays. Using recombinant glycosyltransferases, sialic acid, fucose, and analogs thereof are installed on cell-surface glycans to form cell-based arrays displaying diverse glycan epitopes that can be probed with glycan-binding proteins by flow cytometry. Using this platform, high-affinity glycan ligands are discovered for Siglec-15—a sialic acid-binding lectin involved in osteoclast differentiation. Incubating human osteoprogenitor cells with cells displaying a high-affinity Siglec-15 ligand impairs osteoclast differentiation, demonstrating the utility of this cell-based glycan array technology. Glycans, interaction platforms protruding from the surface of cells, are hard to study due to their diverse architecture. Here, the authors present a method to obtain cells carrying defined glycans, which can then be used to find proteins specifically recognizing these tags.

G lycans decorate the cell surface of both eukaryotes and prokaryotes, and in mammalian cells are involved in a variety of physiological processes, including angiogenesis, fertilization, stem cell development, and neuronal development [1][2][3] . Changes in glycosylation patterns have also been shown to mark the onset of cancer and inflammation 2,3 . In many cases, glycans execute these cellular functions by interacting with glycanbinding proteins (GBPs). Therefore, there is enormous interest in understanding the structural basis of these interactions for the dissection of the mechanisms of glycan-mediated biological processes and for the development of new therapeutic agents to treat glycan-regulated disease. Unfortunately, it is challenging to probe glycan−GBP interactions in vivo because glycosylation is a post-translational modification not under direct genetic control. The dynamic process of glycosylation orchestrated by glycosylation enzymes results in heterogeneous glycoconjugates found on the cell surface and on secreted proteins 3 .
Glycan microarrays were developed in response to the critical need for high-throughput methods to identify GBP interactions 4,5 . As highlighted in Transforming Glycoscience (section 5.1.1), these microarrays have been extensively employed to interrogate binding specificities of a diverse range of GBPs, determine dissociation constants, dissect binding energies, and assess multivalent and hetero-ligand binding 6 . Currently, most glycan arrays are constructed by coupling a chemically defined glycan to a solid support, such as a glass slide 4,5 . Such homogeneous glycans and derivatives are either synthesized 4 or purified from natural sources by multi-dimensional chromatography 7 . Several noteworthy drawbacks are associated with the current platforms. First, obtaining samples of pure, wellcharacterized oligosaccharides for the assembly of glycan arrays by chemical or chromatography-based purification is time consuming and can only be performed by a specialist. As such, glycosyltransferases are often employed in combination with chemical synthesis to facilitate the production of complex oligosaccharides 8 . However, only limited numbers of glycosyltransferases are present in carbohydrate chemists' toolbox. Therefore, many glycosidic linkages cannot be assembled in a straightforward manner. The second drawback is that the current glycan microarrays do not fully recapitulate the natural cellsurface environment on which glycans are presented. Indeed, Wong and co-workers have shown that the poor sensitivity of the conventional microarrays arises from their surface-generated pseudo-multivalent display 9 . To better mimic the natural multivalent presentation, several groups have developed creative strategies by attaching synthetic glycans to protein 10 or polymer scaffolds 11 . These approaches, however, also rely on the lengthy synthesis of complex glycans.
Here, we describe a method to chemoenzymatically install monosaccharides and their analogs directly on the cell surface to create in-solution, cell-based arrays displaying chemically defined peripheral glycan epitopes. The lectin-resistant Chinese hamster ovary (CHO) cell mutant Lec2 that expresses a narrow and relatively homogenous repertoire of glycoforms is employed as the foundation platform. With the conserved core glycan structures already expressed on the cell surface, the lengthy synthesis required to build complex carbohydrates is avoided. Using a handful of glycosyltransferases compatible with cell-surface glycosylation, sialic acid, fucose, and their analogs are introduced to these cells' peripheral glycans linkage specifically to form cellbased arrays displaying diverse glycan epitopes. We demonstrate the utility of these cell-based arrays to interrogate GBP specificities and ligand tolerance directly on the cell surface. This method is applied to high throughput screening for the identification of selective and high-affinity ligands of Siglecs, a family of sialic acid-binding immunoglobulin-type lectins that are differentially expressed primarily on immune cells. Using this approach, a high-affinity glycan ligand for Siglec-15 is discovered that can be used to modulate the differentiation of osteoclasts.
The cells displaying Le X were constructed by treating Lec2 cells with GDP-fucose (GDP-Fuc) (1) (500 μM) and α1-3FT in Hanks buffered salt solution (HBSS) for 10 min at 37°C (Fig. 2a). Lec8 CHO cells that do not express LacNAc due to a mutation in the UDP-Gal Golgi transporter were used as a negative control 19 . The modified cells were then probed with fucoside binding proteins or antibodies, including anti-stage-specific embryonic antigen-1 (anti-SSEA-1 or anti-Le X ), Aleuria aurantia lectin (AAL), and Lotus tetragonolobus lectin (Lotus A) as FITC conjugates 20 . Flow cytometry reveals that anti-SSEA-1, AAL, and Lotus A all exhibited strong binding to cells displaying the Le X epitope (Fig. 2b). In contrast, only background binding was observed with non-fucosylated Lec2 cells and control Lec8 cells.
The cells displaying Siaα2-3Galβ1-4GlcNAc and Siaα2-6Galβ1-4GlcNAc were constructed by treatment of Lec2 cells with CMP-Neu5Ac (6) (Fig. 2a) and ST3Gal4 or ST6Gal1, respectively, for 1 h at 37°C. Again, Lec8 CHO cells were used as a negative control. These two cells were first probed with FITC-conjugated Erythrina cristagalli lectin (ECL) that is specific for terminal galactose and N-acetylgalactosamine and as such, bound strongly to unmodified Lec2 CHO cells (Fig. 2c) 21 . Consistent with the specificity of ECL, the cells with installed sialic acid resulted in reduced binding (Fig. 2c). The cells were then probed with FITC-conjugated sialoside binding proteins, Sambucus nigra (SNA-I) and Maackia amurensis lectin (MAL-II), specific for α2-6and α2-3-linked sialic acids, respectively 20 . SNA-I strongly bound to cells displaying α2-6-linked sialic acid, whereas no binding was observed for cells with α2-3-linked sialic acid (Fig. 2c). For MAL-II, the opposite was observed (Fig. 2c). In these experiments, as little as 1 μg/mL of GBPs was used. Together, these results provide strong evidence that our approach for installing monosaccharides to construct ligands for probing their interactions with GBPs on cell surfaces is specific and sensitive.
Construct glycan arrays displaying unnatural epitopes. Screening glycan libraries with unnatural substituents serves as a powerful means to probe the binding pocket of a GBP especially when there is limited structural information available. To apply cell-based glycan arrays to profile the substrate tolerance of GBPs, we constructed three arrays displaying Le X , Siaα2-3Galβ1-4GlcNAc, Siaα2-6Galβ1-4GlcNAc and their structurally related derivatives using Lec2 CHO cells as the starting platform. The array displaying Le X and its derivatives was constructed using 500 μM GDP-Fuc (1) or GDP-Fuc analogs (2−5) (Fig. 3a). Our previous studies revealed that while α1-3FT is highly specific for acceptor glycans possessing LacNAc, it has relaxed donor substrate specificity and is able to accept GDP-Fuc donors with a wide variety of functional groups at the C5 position of fucose 15 . Kinetic analysis of the enzymatic transformation showed only minor differences (2.8, 8.4, 6.2, 24 μM for 2-5 vs. 16 μM for GDP-Fuc (1)) in K m values for GDP-Fuc donors with functional groups at C5. Furthermore, there was no appreciable influence on the V max ( Supplementary Fig. 1). Therefore, we chose 500 μM as the concentration for all donor substrates for surface fucosylation, which was well above the saturated substrate concentration.
The array displaying Le X and its derivatives was probed with fucoside binding proteins, SSEA-1, AAL, and Lotus A. The arrays displaying Siaα2-3Galβ1-4GlcNAc, Siaα2-6Galβ1-4GlcNAc and their derivatives were probed with SNA-I and MAL-II. Interestingly, we discovered that this panel of GBPs exhibited distinct binding tolerances to unnatural glycan derivatives. Anti-SSEA-1, also known as anti-Le X antibody, was able to tolerate an alkyne (2) and a hydroxyl group (4) at the C5 position of fucose, but not an azide (3) or methoxy (5) functional group, which abrogated interaction (Fig. 3c). All fucose analogs assessed resulted in a reduction in AAL binding but did not entirely abolish this interaction. Lotus A was very sensitive to C5 modification and the interaction was completely blocked by any modification to the C5 position of fucose. On the other hand, SNA-I and MAL-II were not sensitive to sialic acid modifications at C5, but demonstrated linkage-specificity (Fig. 3d). SNA-I bound to all sialic acid analogs when attached in an α2-6-linkage. Alternatively, MAL-II tolerated all analogs attached in an α2-3-linkage.
Screen high-affinity Siglec-15 ligands. Having demonstrated that our cell-based glycan array was capable of detecting GBP binding specificity and substrate tolerance, we next assessed the applicability of this method to identify high-affinity Siglec ligands. Siglecs comprise a family of 15 members of sialic acid-binding receptors that are differentially expressed on immune cells [24][25][26] . Because of the restricted expression of Siglecs to one or a few immune cell types, Siglecs are attractive targets for cell-directed therapies in immune-cell-mediated diseases [26][27][28][29] . Furthermore, Siglecs are endocytic receptors allowing efficient uptake of therapeutic agents conjugated to an antibody or glycan ligand [30][31][32][33] . High-affinity and highly selective glycan ligands represent attractive alternatives to antibody-based therapeutics with several notable advantages. Notably, nanoparticles bearing sialoside ligands for delivery of therapeutic cargo have shown promise for targeting Siglecs in vivo 34 . To date, high-affinity and selective glycan ligands have been discovered for approximately half of human Siglecs using conventional glycan microarray technology [35][36][37][38] .
We sought to determine the applicability of our cell-based array technique to discover high-affinity and selective glycan ligands for Siglecs. One Siglec of interest where high-affinity and selective ligands have not been discovered is Siglec-15. Siglec-15   Fig. 1 The workflow for constructing cell-based glycan arrays. a Distinct glycan epitopes are installed onto N-glycans of Lec2 CHO cells via sialyltransferase-mediated cell-surface in situ sialylation using the CMP-Neu5Ac donor (or its analogs) or fucosyltransferase-mediated cell-surface in situ fucosylation using the GDP-Fuc donor (or its analogs). b Cell-based glycan arrays displaying Le X or its derivatives. c N-glycans of Lec2 CHO cells. d Cellbased glycan arrays displaying α2-3or α2-6-linked sialosides, or their derivatives. X represents unnatural substituents introduced at fucose C-5 or NeuNAc acetyl position becomes expressed within days of RANKL (receptor activator of nuclear factor κβ ligand)-induced differentiation of osteoclast precursors into osteoclasts [39][40][41][42][43] . Studies have demonstrated that Siglec-15 positively regulates osteoclast differentiation and, consistent with this finding, loss of Siglec-15 results in impaired osteoclast differentiation and osteopetrosis in Siglec-15-deficient mice 40 . Therefore, Siglec-15 plays a critical role in bone remodeling and is a considerable potential therapeutic target for osteoporosis.
From what is known, Siglec-15 prefers α2-6-linked sialic acid residues in native ligands 43 . As several high-affinity Siglec ligands have been developed with functional modifications at the C5 position of Neu5Ac, we focused on this position on sialic acid to create chemical diversity [35][36][37][38] . To do so, we employed ST6Gal1 to install an α2-6-linked sialic acid bearing a bioorthogonal alkyne tag at C5 onto the cell surface of Lec2 cells (Fig. 4a). The alkyne tagged cells were then derivatized by reaction with a 40-member azide library (Supplementary Table 1) via biocompatible Cu(I)catalyzed azide-alkyne cycloaddition (CuAAC) (Fig. 4a) 44 to form a cell-based glycan array . The library was chosen to contain structures with a variety of different functional groups to cover a range of chemical space. The resulting cell-based glycan array was probed with a recombinant Siglec-15-Fc chimera pre-complexed to APC-conjugated anti-human IgG1. From this screen, several high-affinity glycan ligands for Siglec-15 were discovered (Fig. 4b).

High-affinity Siglec-15 ligands engage in cell−cell interactions.
With the identification that 10 could serve as the donor substrate to form high-affinity Siglec-15 ligands on the Lec2 cell surface, we then evaluated if these ligands could enforce interaction of modified cells with Siglec-15-expressing cells. Toward this end, we stained α2-3-10 and α2-6-10 displaying Lec2 CHO cells with CellTracker orange (CTO) as their identity marker. Unlabeled Lec2 CHO cells and cells labeled with the native ligand, α2-6-6, were used as controls. Subsequently, the Lec2 CHO cells stained with CTO were mixed with Siglec-15-expressing CHO cells that express extracellular Siglec-15 fused to intracellular GFP in a 100:1 ratio. The formation of cell clusters was determined by double staining and analyzed by flow cytometry and fluorescence microscopy (Fig. 5). Compared to the control non-sialylated Lec2 cells and cells modified with natural Neu5Ac, Lec2 cells modified with 10 induced significantly more cluster formation with Siglec-15-expressing cells.
While α2-6-10 was found to also bind to Siglec-2 weakly, osteoclasts and human osteoprogenitor cells (HOPs; CD14 + peripheral blood mononuclear cells) do not express Siglec-2. Therefore, after confirming that cells displaying α2-6-10 could engage in cell−cell interactions via Siglec-15 binding, we assessed if Lec2 CHO cells displaying α2-6-10 also could block osteoclast formation of RANKL -treated HOPs. RANKL, also known as osteoclast differentiation factor, is a cytokine secreted by osteoblasts that serves to activate osteoclasts, which are critically involved in bone resorption 42 . To determine if Lec2 CHO cells displaying α2-6-10 could block osteoclast formation from HOPs, HOPs were isolated and plated. Osteoclast differentiation was induced by the addition of human macrophage colonystimulating factor (hM-CSF) and RANKL 46 . Cells were allowed to differentiate for 7 days. On day 5, Lec2 CHO cells labeled with the high-affinity Siglec-15 glycan ligand, α2-6-10, or the native ligand, α2-6-6, were added to the plated cells. In these experiments, unlabeled Lec2 CHO cells were used as the negative control. On day 7, the formation of osteoclasts, which express abundant tartrate-resistant acid phosphatase (TRAP), was measured using the TRAP staining assay 47 . Osteoclasts were identified as TRAP-positive with 3 or more nuclei 46,47 . Remarkably, the addition of Lec2 CHO cells labeled with the high-affinity Siglec-15 ligand on day 5 resulted in over 50% reduction in osteoclast formation compared to the addition of unlabeled Lec2 CHO cells and HOPs treated with RANKEL and and hM-CSF only (Fig. 6). Furthermore, Lec2 CHO cells modified with α2-6linked natural sialic acid did not exhibit significant effects on osteoclast formation presumably due to the relatively low affinity of α2-6-6 for Siglec-15 compared to α2-6-10 ( Supplementary  Fig. 4).

Discussion
The use of molecular diagnostic techniques for the detection of biological markers in the genome or proteome has dramatically increased over the last several decades, which has stemmed from the advancements in molecular biology techniques and the miniaturization of high-throughput microarrays 48,49 . Microarrays are widely used to study disease-specific gene mutations or protein expression and to correlate these expression signatures with disease progression. As a result, clinicians are increasingly able to treat patients according to their individual gene and/or protein expression profiles, known as personalized medicine. While microarrays comprised of nucleic acid or peptides/proteins are well-established and accessible for common users, glycan microarray technology remains highly specialized 50 .
The in-solution, cell-based glycan array platforms described here are developed to address this unmet need. By combining STs, ST6Gal1 and ST3Gal4, and α1-3FT with Lec2 CHO cells possessing a narrow and relatively homogeneous repertoire of Nlinked glycoforms, cell arrays displaying sialosides, fucosides and their structurally related analogs can be assembled easily using instruments available in most biology labs. With the necessary glycosyltransferases and nucleotide sugar donors prepared in advance, it only requires 2-3 h to fabricate a cell-based glycan array displaying >30 distinct glycan epitopes in a 96-well plate. By contrast, it takes days to weeks to assemble a conventional glycan array using comparable chemoenzymatic chemistry because each individual glycan to be printed needs to be synthesized and purified 4,7 . Importantly, the diversity of cell-based glycan array platforms can be significantly expanded by including the recently developed CHO and HEK cell lines with simplified surface glycans via precision genome editing as foundation platforms 51 .
Using conventional glycan microarray technology, high-affinity and selective glycan ligands have been identified for six out of 15 human Siglecs, which involves the synthesis of hundreds of sialylated oligosaccharides bearing unnatural substituents [35][36][37][38] . The cell-based array technique described here has proven to be a simpler alternative for this endeavor. By using STs to install unnatural sialic acids directly onto the cell surface in a linkagespecific manner, the lengthy synthesis of sialylated oligosaccharides is avoided. Because sialylation reactions are performed on the cell surface, significantly fewer materials are required to produce the desired structures.
Recently, Boltje and co-workers reported an alternative method to discover Siglec ligands directly on the cell surface 52,53 . This method involved metabolic labeling using an alkyne functionalized ManNAc precursor followed by cellsurface CuAAC. Although a few high-affinity ligands were discovered by this approach, none were specific for a particular Siglec. This is likely due to the lack of linkage specificity of this approach. Most Siglecs differentially bind to sialosides based on their Neu5Ac linkage. Metabolic labeling leads to incorporation of unnatural sialic acids in both α2-3and α2-6-linkages, largely complicating the downstream Siglec binding studies. By contrast, the cell-based microarray described here utilized STs to install unnatural sialic acids directly on cell-surface glycans linkage specifically. Siglec-15 is constitutively expressed in osteoclasts. During the RANKL-induced osteoclast differentiation, Siglec-15 is considerably upregulated [39][40][41][42] . Several antibodies targeting RANKL are FDA-approved, or are in late stage clinical trials for the treatment of osteoporosis and cancer-induced bone loss, and there are keen interests in targeting Siglec-15 for the same therapeutic applications [54][55][56] . In fact, Tremblay and co-workers discovered a monoclonal antibody for Siglec-15 that impaired osteoclast differentiation both in vitro and in vivo 46 . Exposure of human HOPs during differentiation to Lec2 CHO cells displaying the high-affinity Siglec-15 ligand (α2-6-10) impaired osteoclast formation, suggesting that this ligand may serve as the lead compound to be further optimized for clinical evaluation. While our method serves as a useful tool for the high-throughput screening of high-affinity glycan ligands for Siglecs and other GBPs, cells may display a range of glycans containing the modified monosaccharide. For the future evaluation of its therapeutic potential, isolation, purification, characterization, and validation of the high-affinity Siglec-15 ligand are required. Currently, we are in the process of using the cell-based glycan array technique to identify specific and high-affinity ligands for other Siglecs.
Cell-surface lectins and antibody staining. Cells subjected to in situ fucosylation were assessed for binding of Le X lectins as follows. For SSEA-1, the cells were incubated with mouse monoclonal anti-SSEA-1 (5 μg/mL, R&D systems, Inc.) in 100 μL PBS with 2% FBS for 30 min followed by staining with PE-conjugated antimouse-IgM (5 μg/mL, Jackson ImmunoResearch Laboratories, Inc.) in 100 μL PBS with 2% FBS for 40 min. For AAL and Lotus A, the cells were incubated with Lotus A-FITC (0.1 mg/mL, EY Laboratories, Inc.), or AAL-FITC (10 μg/mL, Vector Laboratories) in 100 μL PBS with 2% FBS for 40 min. Cells were then washed three times with PBS, and resuspended in PBS with 2% FBS and 2 mM EDTA for flow cytometric analysis. Cells subjected to in situ sialylation were assessed for binding of LacNAc and sLacNAc lectins as follows. For ECA and SNA-I, the cells were incubated with ECA-FITC (1 μg/mL, EY Laboratories), or SNA-I-FITC (1 μg/mL, EY Laboratories) in 100 μL PBS with 2% FBS for 30 min on ice. For MAL-II, the cells were incubated with MAL-II-FITC (1 μg/mL, EY Laboratories) in 100 μL PBS with 2% FBS and 1 mM CaCl 2 for 30 min on ice. Cells were then washed three times with PBS, and resuspended in PBS with 2% FBS and 2 mM EDTA for flow cytometric analysis.
Cell-surface binding of Siglec-Fc chimeras. Siglec-Fc chimeras containing the Nterminal Ig domains fused to the Fc region of human IgG1 were prepared as described previously 35 . Labeled Lec2 CHO cells were resuspended in 100 μL of recombinant Siglec-Fc chimera supernatants pre-complexed with anti-human IgG APC (1:40 dilution, 200 μg/mL). Briefly, 2.5 μL of anti-Human IgG APC (200 μg/ mL) was added to 50 μL of recombinant Siglec Fc chimera supernatant and incubated in the dark at room temperature for 15 min followed by dilution with 50 μL of PBS containing 1% BSA. Cells resuspended in this solution were incubated for 30 min on ice before being washed three times with PBS, and resuspended in PBS with 2% FBS and 2 mM EDTA for flow cytometric analysis.
Cloning and expression of Siglec-15 in CHO cells. Siglec-15 contains a lysine residue in its transmembrane region that pairs with Dap-12. Since Dap-12 is not expressed in standard cell lines (e.g. CHO), we took another approach to expressing the extracellular portion of human Siglec-15 as a transmembrane portion of Siglec-15 by cloning a chimeric protein consisting of amino acids 1−260 of human Siglec-15 with the transmembrane and cytoplasmic tail of human CD22 fused to a C-terminal eGFP. This chimeric protein was cloned into pcDNA5/FRT/TOP/V5/ His (Invitrogen) using the NheI and AgeI restriction enzymes. CHO Flp-in cells (Invitrogen) cells stably transfected with this vector according to the manufacturer's protocol. After selection of cells with Hygromycin-B for 2 weeks, all cells were found to be expressing high levels of GFP by flow cytometry.
Lec2 CHO cell and Siglec-15-expressing CHO cell clustering. For flow cytometry analysis, Lec2 CHO cells were labeled with CellTracker orange (CTO) following the manufacturer's protocols (ThermoFisher Scientific), sialylated (as described above) and mixed with hSig15-expressing CHO cells that also express GFP in a ratio of 100:1 for 20−30 min on ice prior to analysis by flow cytometry. The GFP+ cell population was analyzed for CTO staining and interacting cells were defined as double stained. For fluorescence microscopy analysis, Lec2 CHO cells stained with CTO were sialylated as described above, mixed with hSig15expressing CHO cells that also express GFP in a ratio of 100:1 for 20−30 min on ice prior to plating and incubation at 37°C for 10 min to allow adherence. Plates were gently rinsed once with PBS and cell clustering was imaged by fluorescence microscopy. Images of each sample were analyzed for cell−cell interactions. The number of hSig15 CHO cells that were interacting with Lec2 CHO cells was counted and expressed as a percentage of the total number of hSig15 CHO cells analyzed.
Osteoclast differentiation. HOPs (CD14+ peripheral blood mononuclear cells) were isolated from normal human peripheral blood mononuclear cells using EasySep Human Monocyte Isolation Kit (StemCell Technologies) following the manufacturer's instructions. HOPs were plated at 0.5×10 6 cells/well in a 24-well plate with 1 mL media (αMEM supplemented with 1 mM sodium pyruvate and 10% FBS). To stimulate osteoclast formation, 25 ng/mL of human macrophage colonystimulating factor and 30 ng/mL of human RANKL (R&D Systems) were added. Cells were allowed to differentiate for 7 days with half media replaced every 3 days. On Day 5, Lec2 CHO cells (0.1×10 6 cells) sialylated as described above using glycosyl donor 10 were added to wells. Unlabeled Lec2 CHO cells were used as a negative control. On Day 7, cells were fixed and permeablized followed by staining for TRAP using the TRAP kit following manufacturer protocols (Sigma Aldrich). Osteoclasts were defined as TRAP+ with 3 or more nuclei.
Data availability. All data are available from the authors upon reasonable request.