Galectin binding to cells and glycoproteins with genetically modified glycosylation reveals galectin–glycan specificities in a natural context

Galectins compose a protein family defined by a conserved sequence motif conferring affinity for β-galactose–containing glycans. Moreover, galectins gain higher affinity and fine-tune specificity by glycan interactions at sites adjacent to their β-galactoside–binding site, as revealed by extensive testing against panels of purified glycans. However, in cells, galectins bind glycans on glycoproteins and glycolipids in the context of other cellular components, such as at the cell surface. Because of difficulties in characterizing natural cellular environments, we currently lack a detailed understanding of galectin-binding specificities in the cellular context. To address this challenge, we used a panel of genetically stable glycosylation mutated CHO cells that express defined glycans to evaluate the binding affinities of 10 different carbohydrate-recognition domains in galectins to N-glycans and mucin-type O-glycans. Using flow cytometry, we measured the cell-surface binding of the galectins. Moreover, we used fluorescence anisotropy to determine the galectin affinities to recombinant erythropoietin used as a reporter glycoprotein produced by the glycoengineered cells and to synthetic N-glycans with defined branch structures. We found that all galectins, apart from galectin-8N, require complex N-glycans for high-affinity binding. Galectin-8N targeted both N- and O-linked glycans with high affinity, preferring 2,3-sialylated N-acetyllactosamine (LacNAc) structures. Furthermore, we found that 2,3-sialylation suppresses high-affinity binding of select galectins, including galectin-2, -3, -4N, and -7. Structural modeling provided a basis for interpreting the observed binding preferences. These results underscore the power of a glycoengineered platform to dissect the glycan-binding specificities of carbohydrate-binding proteins.

Galectins compose a protein family defined by a conserved sequence motif conferring affinity for ␤-galactosecontaining glycans. Moreover, galectins gain higher affinity and fine-tune specificity by glycan interactions at sites adjacent to their ␤-galactosidebinding site, as revealed by extensive testing against panels of purified glycans. However, in cells, galectins bind glycans on glycoproteins and glycolipids in the context of other cellular components, such as at the cell surface. Because of difficulties in characterizing natural cellular environments, we currently lack a detailed understanding of galectin-binding specificities in the cellular context. To address this challenge, we used a panel of genetically stable glycosylation mutated CHO cells that express defined glycans to evaluate the binding affinities of 10 different carbohydrate-recognition domains in galectins to N-glycans and mucin-type O-glycans. Using flow cytometry, we measured the cell-surface binding of the galectins. Moreover, we used fluorescence anisotropy to determine the galectin affinities to recombinant erythropoietin used as a reporter glycoprotein produced by the glycoengineered cells and to synthetic N-glycans with defined branch structures. We found that all galectins, apart from galectin-8N, require complex N-glycans for high-affinity binding. Galectin-8N targeted both N-and O-linked glycans with high affinity, preferring 2,3-sialylated N-acetyllactosamine (LacNAc) structures. Furthermore, we found that 2,3-sialylation suppresses high-affinity binding of select galectins, including galectin-2, -3, -4N, and -7. Structural modeling provided a basis for interpreting the observed binding preferences. These re-sults underscore the power of a glycoengineered platform to dissect the glycan-binding specificities of carbohydrate-binding proteins.
Galectins are carbohydrate-binding proteins with regulatory functions in intracellular glycoprotein traffic (1), cell adhesion (2,3), migration (4), growth (5), and apoptosis (6), with consequent implications in immunity (7,8), inflammation (9), and cancer (10,11). Galectins are characterized by their ability to bind specific carbohydrate chains that contain ␤-galactosides resident on glycoconjugates. This interaction is facilitated through at least one structurally conserved carbohydrate-recognition domain (CRD) 2 comprising ϳ130 amino acids (12). The galectin CRD is a slightly bent ␤-sandwich, with a groove on the concave side long enough to accommodate a linear tetrasaccharide. The binding pocket can be divided into subsites A-D, where interaction with a ␤-galactose unit in site C is essential for glycan binding. Saccharides linked on either side of the ␤-galactose may either increase or decrease binding by interactions in subsite A, B, and D, giving each galectin CRD its fine specificity for larger saccharides. Based on similarities in the structural arrangement of their CRDs, galectins have been divided into three subgroups: the prototypical, chimeric, and the tandem repeat galectins. Prototypical galectins (galectins-1, -2, -5, -7, -10, and -13-15) contain a single CRD that may selfassociate to form homodimers. Galectin-3 is the only vertebrate member of the second chimeric galectin group and possesses a single C-terminal CRD and a unique nonlectin N-terminal domain, and both confer critical biological functions (13). The last subgroup, the tandem-repeat galectins (galectins-4, -6, -8, -9 and -12), carry two structurally and functionally distinct CRDs, bridged by a small linker domain that varies from 5 to 70 amino acids in length (14,15). Although most galectins have a shared specificity for ␤-galactose residues that are widely distributed on glycoproteins and glycosphingolipids, individual galectins require additional distinct structures for high-affinity binding. These specificities have been extensively analyzed by a variety of biochemical methods using panels of purified, naturally occurring glycans or fragments of them (16 -21). Using these techniques, it has been demonstrated that galectin specificity toward N-glycans is particularly dependent on branching and terminal modifications. In particular, poly-LacNAc extensions on tri-and tetra-antennary N-glycans have been proposed to be of major importance in terms of the specificity of galectin-1, -3, and -8 binding, whereas galectin-4 binding appears unaffected (17,19,(22)(23)(24)(25)(26)(27). Sialylation of terminal galactose residues can also be important for some galectins. For example, glycans capped with ␣2,6-sialic acid globally block galectin binding (17,19,27), whereas galectin-8N has a particularly high affinity for galactose residues with a terminal ␣2,3-sialic acid (28,29). It has also been suggested that the unique specificity of the N-terminal CRD of galectin-8 enables high-affinity binding to O-glycans (22,30). These investigations have contributed greatly to our current understanding of galectin specificity, but most are limited by their evaluation of synthetic glycans in the absence of any appropriate protein or cellular context. The specificity of glycan binding in a more physiologically relevant setting has yet to be addressed in a systematic manner.
The emergence of stable genetic engineering strategies has enabled us to knock out specific glycosyltransferases responsible for the major biosynthetic glycosylation pathway (Fig. 1A). This has allowed us to create libraries of cells that display an almost homogeneous set of glycans on both their surface and on expressed glycosylated molecules (31). Using the homogeneous presentation of defined cellular glycans, it is possible to profile galectin binding to glycoconjugates presented in their natural cellular context. Simultaneously, the genetically engineered cells can be used to produce a reporter glycoprotein to assess galectin binding to single molecules in solution using fluorescence anisotropy (FA). In this study, we investigated the specificity of 10 different galectin CRDs using a combination of flow cytometry to assess surface binding to glycosylation-altered CHO mutants with well-defined glycophenotypes, and single-molecule binding to erythropoietin (EPO) produced by the same glycosylation-altered cells using FA (Fig. 1B). Finally, we performed 3D molecular modeling of galectins bound to natural N-glycans. This allowed us to systematically dissect the importance of different glycan determinants for galectin binding in a physiological context. Consequently, our results provide increased insights into glycan recognition and how galectin-glycan interactions are altered by changes to the glycosylation machinery. Figure 1. A, schematic representation of the glycoengineering strategy used to produce CHO cells bearing altered glycosylation pathways. The strategy is illustrated using a common tetra-antennary N-glycan and a core-1 O-glycan. Arrows indicate glycosyltransferase genes with potential roles in biosynthesis at each step. mgat1 knockout results in high-mannose N-glycans (primarily man5). mgat4A/4B knockout results in triantennary N-glycans that lack ␤1-4-GlcNAc branches. mgat4A/4B/5 knockouts lack both ␤1-4-and ␤1-6-GlcNAc branching and result in the synthesis of biantennary N-glycans. Knockout of st3gal4/6 results in the lack of 2,3-sialylation of complex N-linked glycans and, combined with the st6gal1 knockin, causes exclusive capping of N-glycans with ␣2,6sialylation. Knockout of cosmc produces simple O-GalNAc structures (i.e. GalNAc or NeuAc␣2,3-GalNAc). B, schematic overview of the strategy used to assess galectin binding in this study. First, glycoengineered cells were used to monitor cell-surface binding with fluorescently labeled galectins, using flow cytometry with or without the galectin inhibitor lactose. Second, we used purified EPO as a reporter glycoprotein produced from the glycoengineered cells to study galectin binding to a single molecule in solution using fluorescence polarization. C, different glycosylation forms of EPO produced by genetically engineered CHO cells. Each EPO protein carries three N-glycans and one O-glycan.

Results
Recombinant human galectins-1 C3S , -2, -3, -4C, -4N, -8C, -8N, -9C, and -9N, together with mouse galectin-7, were produced in Escherichia coli and then purified by affinity chromatography on a lactosyl-Sepharose column as described previously (32). Galectins were conjugated to fluorescein, with cell binding measured by flow cytometry as described under "Materials and methods." The glycophenotype of wildtype (WT) and each individual glycosylation CHO mutant was assessed by MS (Fig. S1). Glycoprofiling on total-cell lysates showed that the major glycoforms released with peptide:N-glycosidase F was high-mannose structures, which is expected to stem from intracellular precursor glycans. The complex type N-glycans seen in WT CHO cells included bi-, tri-, and tetra-antennary N-glycans with a variable number of sialic acid capping. All complex N-glycans were abolished in mgat1 knockout cells leaving only high-mannose structures (Fig. S1). As expected, mutants with knockouts of branch initiator enzymes (mgat4A/4B and mgat4A/4B/5) eliminated tri-and tetra-antennary N-glycans leaving only bi-antennary N-glycans (Fig. S1). The number of terminally sialylated glycans was dramatically reduced in mutant cells lacking both st3gal4 and st3gal6 (Fig. S1). Glycoprofiling of O-linked glycans showed that knockout of the core-1 synthase chaperone cosmc resulted in the loss of simple O-GalNAc structures exemplified by the abolition of ST (Neu5Ac␣2,3Gal␤1,3GalNAc) and disialyl-T (Neu5Ac␣2, 3Gal␤1,3(Neu5Ac␣2,6)GalNAc) (Fig. S1). The glycoprofiles obtained from the total-cell lysates are in agreement with our previous characterization of glycans released from EPO produced in these cells (31). EPO is a 160-amino acid protein with three N-glycans at positions 51, 65, and 110, which mainly carry tetra-antennary complex N-glycans with a variable number of ␣2,3-sialic acid residues. In addition, EPO presents a single O-glycan at position 153 carrying ST structures (31). All complex N-glycans on EPO were abolished when produced in mgat1 knockout cells, whereas knockout of branching enzymes (mgat4A/4B and mgat4A/4B/5) eliminated tri-and tetra antennary N-glycans. Finally, EPO produced in cosmc KO cells only presented the initial O-GalNAc structure and lacked sialylated core-1 structures (31). With this established array of CHO cells and EPO presenting defined N-and O-glycans, we next determined the binding of each fluorescently tagged galectin to both cells and recombinantly produced EPO variants. We first analyzed the binding of fluorescently tagged galectins to the surface of WT cells. A range of concentrations (0 -8 M) of fluoresceintagged galectin was added to CHO cells at 4°C to prevent endocytosis, and bound galectin was quantitated by flow cytometry (cell binding data can be found in full in Fig. S2). All galectins bound WT CHO cells but with varying affinities ( Fig. 2A).  (Table S1). Galectins-2, -4C, -4N, -7, and -8C demonstrated no binding or very low binding to WT CHO cells, as well as to most of the tested mutant cells. Galectin binding could be inhibited by 100 mM lactose (Fig. S3) confirming that the galectin binding was carbohydrate-dependent. This was further verified by a lack of binding of the glycan-binding deficient galectin-3 mutant R186S (12) to WT CHO cells (Fig.  S4). Next, we tested the affinities of the different galectins to soluble EPO produced in CHO WT cells. The different EPO glycoforms used in this study can be found in Fig. 1C. The galectin affinities to EPO in solution was calculated from EPO's potency to inhibit interaction of galectin with a small-molecule fluorescent probe using FA analysis as described for other saccharides, glycoconjugates, or synthetic inhibitors ( Table 1, Fig.  2B, and Fig. S4) (18). Galectins-1 and -3 bound EPO with K d values of 1.5 and 0.2 M, respectively, similar to haptoglobin (33) and transferrin (34). Galectin-9 also bound EPO, with similar range of K d values of 1.3 and 0.5 M for the C-and N-terminal CRD, respectively. In contrast, galectin-8N had much higher affinity for WT EPO with a K d at about 60 nM. Similar affinities have been observed for galectin-8N binding to the simple core-1 O-glycan structure Neu5Ac␣2-3Gal␤1-3GalNAc (29). For the other galectins (-2, -4C, -4N, and -8C), no inhibitory activity was seen with EPO even at the highest concentration (2.1 M, Table 1), indicating K d values greater than ϳ20 M or no binding.
Hill slope )). The dashed line at 15% inhibition is to highlight the threshold for a measuring point to be included in the calculation of the EPO/galectin K d values presented in Table 1.

Galectin glycan-binding in a natural context N-Glycans versus mucin type O-glycans
Past studies have shown that both N-and mucin type O-glycans act as ligands for galectins, although the vast majority of the published data report complex N-glycans as the primary interaction partners (22,30,(35)(36)(37). To determine the contributions of N-and O-linked glycans as ligands for the individual galectins, we probed the binding of galectins to CHO cells and EPO with or without complex-type N-linked glycans and with and without core-1 O-linked glycans. Cells lacking the gene mgat1 cannot make complex N-linked glycans and only display high-mannose structures, whereas cells lacking the chaperone cosmc fail to make the core-1 (T-antigen) and core-1 extensions (e.g. sialyl-T) normally found in WT CHO cells ( Fig. 1C and Fig. S1). As described above, we identified the corresponding N-glycan and O-glycan structures on EPO produced by the different genetically engineered cells (Fig. S1). Complete abolition of complex N-glycans resulted in the elimination of binding for most galectins, except for galectin-8N, which instead bound with a reduced affinity both when using mgat1 KO cells (Fig. 3) and purified EPO from mgat1 KO cells (K d 60 nM for WT cell EPO versus 1,16 M for EPO from mgat1 KO cells) (Table 1 and Fig. 4). Among the weak binders, galectin-4C, -4N, and 8C also showed marginally increased binding on mgat1 KO cells, but given the weak binding of these galectins it is, however, difficult to make any firm conclusions about their specificities. Collectively, the data suggest that although galectin-8N can bind with high affinity to core-1 O-glycans, most of the other galectins require complex N-glycans for high-affinity binding. For galectin-8N, this was corroborated by the fact that truncation of core-1 O-glycans resulted in reduced galectin-8N binding to cells and EPO. Overall, targeted knockout of cosmc in CHO cells (resulting in truncated O-glycans) showed the limited contribution of core-1 O-glycans to the binding of most of the other tested galectins to both engineered cells and purified EPO. We did, however, find that apart from galectin-8N, the binding of galectin-1 and to a minor degree also galectin-9N, was reduced in the absence of core-1 O-glycans. This suggests that galectin-1, and to a minor degree also galectin-9N, binds to both core-1 O-glycans as well as complex N-glycan structures (Table  1 and Fig. 4). Hence, we can conclude that complex N-glycans are the main glycoconjugate ligand essential for high-affinity galectin binding to the cell surface and individual glycoproteins. However, galectin-8N, and to some extent galectin-1 and galectin-9N, can also bind with significant affinity to simple O-linked structures.

N-Glycan branching
Some studies have suggested that galectins exhibit a (high) preference for branched N-glycans (in particular ␤1,6-) (24 -26, 35). Experimentally, however, many studies show no such preference or only subtle effects of N-glycan branching (17,22,38), whereas others do, e.g. of glycoproteins in solution (33,34). Surprisingly, binding of galectins was not compromised for the most part of the galectins to cells in which tri-and tetra-antennary N-glycan branching was eliminated (mgat4A/4B/5 KO; Fig. S2) (Fig. 3A). In fact, some galectins (-3, -8N, and -9C) even had increased binding to cells that displayed N-glycans with reduced branching. This finding is contrasted by the data from binding to EPO where galectin-3, galectin-8N, and galectin-9C all demonstrated an ϳ2-fold reduction in binding to EPO without tri-and tetra-antennary N-glycans (Table 1 and Fig. 4). To further dissect the importance of N-glycan branching, we performed fluorescence anisotropy assay with fluorescein tagged N-glycans and CRDs from galectin-1, -2, -3, 8N, and 9C ( Fig. 5 and Table 2). N-Glycan probes, representing bi-antennary, ␤1,4-tri-antennary, ␤1,6-tri-antennary, and tetra-antennary N-glycans in their nonsialylated and nonfucosylated forms ( Table 2), were mixed with increasing concentrations of galectins and the FA was determined. The binding curves were consistent with 1:1 interactions, and the affinity (K d ) was estimated by nonlinear regression (Table 2). For the individual galectins, the highest affinity was measured for galectin-3. Galectin-1, galectin-8, galectin-8N, and galectin-9C bound with weaker affinities, whereas galectin-2 did not bind ( Table 2). All the tested galectins, apart from galectin-2, displayed only smaller differences in affinities to bi-antennary, ␤1,4-tri-antennary, ␤1,6-tri-antennary, and tetra-antennary N-glycans. ␤1,6-Triantennary glycans, which have been proposed as preferred functional ligands for some galectins, did not bind significantly better. It is clear, however, that all the N-linked glycan probes bound significantly worse (by a factor of 10 -100) when compared with intact EPO (Table 1). Thus, an additional mechanism of interaction may exist in the intact glycoprotein. Indirect support of this is the different maximum anisotropy values approached (A max ) for the different galectins in Fig. 5. A max is the anisotropy of the galectin-probe complex, and its value depends on the environment of the fluorescein moiety, besides rotation of the complex as a whole. The rather large variance in A max for the different galectins (high for galectin-3 and very high for galectin-9, but low for galectins 1) suggests differences in interaction with the fluorescein moiety posi- All galectins were tested against the same production batch of EPO. Purity of EPO was determined with SDS-PAGE (Fig. S6).

Galectin glycan-binding in a natural context
tioned at the end of the glycan where the glycan is normally linked to the protein.

Branch elongation with poly-LacNAc
Branched N-glycans can be further modified by ␤1,4-galactosyltransferases attached to variably elongated LacNAc repeats (poly-LacNAc). The mgat5 generated ␤1,6GlcNAcbranch is preferentially elongated with such poly-LacNAc structures, suggested to produce N-glycans with higher affinities for galectins (22,35). To evaluate the contribution of poly-LacNAc for galectin binding in our system, we analyzed galectin binding to CHO cells with a knockout of b3gnt2, the glycosyltransferase that initiates poly-LacNAc synthesis on N-linked glycans. For all the tested galectins (-1, -3, -8N, -9C, and -9N), binding was reduced to cells lacking LacNAc extensions on N-glycans (Fig. 6). In agreement with previous studies, binding of galectin-1, -3, and -8N was reduced about 30 -50%, and galectin-9C and -9N was reduced up to 60 -80%. To visualize the steric availability for simultaneous binding of multiple galectins to poly-LacNAc repeats on branched N-glycans, we performed structural modeling with galectins in both their dimeric and monomeric form. Structures of a galectin-1 homodimer, galectin-3 CRD, and a galectin-8N monomer CRD bound to each LacNAc repeat of a ␤1,6 tetra-LacNAc tetraantennary N-glycan (generated on Glycam-Web, www.glycam. org/cb 3 (39)) were modeled using available crystal structures bound to LacNAc (PDB codes 1W6P, 4R9D, and 5T7S). An initial structure of each galectin bound to the innermost Lac-NAc was created using Gly-Spec (www.glycam.org/gr 3 (40)), which first superimposes the tetra-antennary N-glycan onto the LacNAc in the crystal structure and then adjusts the glycosidic linkages of the larger glycan to relieve any steric overlap with the protein surface. A second galectin was modeled by superimposing it onto a neighboring LacNAc unit using UCSF Chimera (41). The models showed that steric overlaps prevent

Galectin glycan-binding in a natural context
simultaneousbindingoftwogalectinsineitherdimericormonomeric form to two adjacent LacNAc repeats in a single glycan branch (Fig. 7A, panels A and D). However, simultaneous binding to nonadjacent LacNAc repeats is possible in the case of the monomeric galectin-8N (Fig. 7A, panels E and F). Both the 3rd and 4th LacNAc motifs in a tetra-LacNAc glycan are available for binding by an additional galectin-8N monomer CRD when the 1st LacNAc motif is bound (Fig. 7A, panels E and F). For galectin-1, a homodimer, simultaneous binding of another galectin to the 4th LacNAc without steric overlap may be possible, but it would involve adjusting the glycosidic linkages in the poly-LacNAc branch. Such simultaneous binding may be a contributing factor in the enhanced affinity observed for structures containing poly-LacNAc structures (Fig. 6). This mechanism would work in concert with any additional contacts formed between the extended glycan and the galectin when an inner LacNAc repeat is bound (Fig. 7A).
We finally examined the possibility of simultaneous galectin binding to N-glycan branches by 3D modeling. The ability of a galectin-1 homodimer to bind to each arm of both the bi-and tetra-antennary N-glycan was assessed. Simultaneous binding of two galectin-1 molecules to both branches of a biantennary N-glycan was found to be possible (Fig. 7B). However, because of steric overlap, two galectins could not bind simultaneously to both outer branches of either the 3-or 6-branch in a tetraantennary N-glycan, regardless of which low-energy shape for the glycan was used (Fig. 7B). Alternate rotamers for the six linkages changes the orientation of the lactosamine on these branches; however, each low energy rotamer caused a steric overlap to form between the bound galectin homodimer and another bound homodimer (structures with other rotamers not shown). This suggests that any enhanced affinity from interaction with a tetra-over a bi-antennary N-glycan containing a single LacNAc per branch would not be due to simultaneous binding to multiple galectins.

Sialylation
The presence of a terminal sialic acid on glycans in either the ␣2,3 or ␣2,6 linkage to galactose is suggested to be a key determinant of the differential recognition of glycans by galectins (19,29). We therefore analyzed the binding of galectins to cells with and without ␣2,3-sialylation of complex N-linked glycans (st3gal4/6 KO CHO cells). We found that galectin-1 and -9C binding was preserved regardless of ␣2,3-sialylation of the underlying LacNAc structures, which is in agreement with previous findings (17,19,28,29). In contrast, galectin-3 exhibited increased binding to N-glycans without sialylation. Although the binding of galectin-2 was very low, we also observed an increase in galectin-2 binding when sialylation was abrogated, which is in agreement with previous reports (19). Interestingly, we found reduced binding of galectin-8N to N-glycans ( Fig. 3 and Table 1). This was confirmed by reduced galectin-8N binding to EPO from st3gal4/6 KO cells to almost the same level as seen in mgat1 KO cells lacking complex N-glycans (Table 1). Several studies have suggested that ␣2,6-sialic acid reduces galectin binding to ␤-galactosides (17,19,38,42,43). To evaluate the importance of differential sialylation, we introduced a st6gal1 knockin to a st3gal4/6 knockout background, which resulted in exclusive ␣2,6-sialic acid capping on N-glycans (31). As expected, the increased ␣2,6-sialylation did reduce cell surface binding of most of the moderate/strong binding galectins (Fig. 3A). A substantial decrease in binding was seen for galectin-1, -3, and -8N, whereas a minor decrease in binding was seen for galectin-9C. No effect was seen for galectin-9N. Unfortunately, limited levels of EPO generated from the st6gal1 knockin cells restricted our analyses and only allowed us to Solid lines represent best-fit curves, where Y ϭ Y min ϩ (Y max Ϫ Y min )/(1 ϩ (X Hill slope /IC 50 Hill slope )). The dashed line at 15% inhibition is to highlight the threshold for a measuring point to be included in the calculation of the EPO/ galectin K d values presented in Table 1. Calculations of IC 50 values for the EPO glycoforms have been added in Table S1.

Galectin glycan-binding in a natural context
test galectin-8N binding to produced EPO with increased 2,6 sialylation.

Discussion
This work demonstrates the use of a genetically glycoengineered cell platform to dissect the binding affinities of 10 different galectins to specific O-and N-linked glycans presented atthecellsurfaceoronsolubleglycoproteins.Ourgeneticdeconstruction of cellular glycosylation provides the first direct insight as to how alterations in the cellular glycosylation machinery translate into galectin binding. Several of these findings are in line with previous biochemical studies (17,28,44,45) and were predicted by functional assays (13, 34, 46 -49). Taken together, our results demonstrate that bi-antennary N-glycans are sufficient for high-affinity binding of galectins, with sialylation as one of the key checkpoints for galectin binding. In addi-  (Table 2). Here, for example, galectin-3 has most left shifted curve showing highest affinity (lowest K d ).

Table 2 K d (M) calculated from Fig. 5
K d values were calculated from non-linear regression least-squares fitting of the data points to the simplified formula Y ϭ A 0 ϩ (A max Ϫ A 0 )⅐(X/(X ϩ K d )), where Y is measured anisotropy, and X is galectin concentration (M). In some cases, the data points do not fit this simplified formula probably because a glycan may have multiple binding sites and/or galectin binds with cooperativity. NB ϭ no binding. Ϯ indicates S.E. There is an additional uncertainty of K d values for the galectins where a clear A max could not be determined (all except galectin-3) and was only extrapolated by the non linear regression. For these, the K d values given are an estimated lower limit. The relative affinities of the different N-glycans are still valid.

Galectin glycan-binding in a natural context
tion to N-glycans, we find that the N terminus of galectin-8 can also bind core-1 O-glycans with high affinity.
We used glycoengineered CHO cells as a cell model to assess galectin binding. CHO cells present a rather simple but varied glycosylation profile compared with HeLa cells (50). This variety makes CHO cells a suitable model with which to selectively dissect the contribution of complex N-glycans and core-1 O-glycans for galectin binding. In agreement with previous reports (51), we find that CHO cells synthesize a range of highmannose and complex N-glycans, with these predominantly capped with ␣2,3-sialic acid structures similar to those found on EPO produced in the same cells (Fig. S1). We also confirmed previous findings that CHO cells carry the simple core-1 GalNAc O-glycans, Neu5Ac␣2,3Gal␤1,3GalNAc and Neu5Ac␣2, 3Gal␤1,3(Neu5Ac␣2,6)GalNAc. In addition, it is known that CHO cells primarily synthesize the simple glycosphingolipid GM3 and lack more complex glycosphingolipids (52,53). This limited complexity of N-glycans, O-glycans, and glycosphingolipids makes CHO cells a suitable model to determine galectin binding to defined glycans in vivo. It should be noted that CHO cells also carry O-fucose, O-glucose, and O-mannose glycans, but these have not been reported to bind galectins (12).
Using the genetic engineered CHO cell model, we find that galectin binding to WT CHO cells falls into three categories: strong (galectins-3 and -8N), moderate (galectins-1, -9C, and -9N), and weak (galectins-2, -4C, -4N, -7, and -8C). Cells engineered to lack complex N-glycans support the general concept that N-glycans are of major importance for galectin binding. However, in our system, the contribution of N-glycan branching is moderate in terms of high-affinity galectin binding. Accordingly, elongation and branching beyond the di-antennary structures catalyzed by mgat4A/4B/5 were not critical control steps and only had partial influence on the binding of galectin-3 and galectin-9C, although some influence was seen for binding of galectin-1 and galectin-9N. This is interesting considering earlier studies with immobilized galectin-1 suggesting a preference for poly-LacNAc chains (24,54), which are often found on ␤1,6-branched N-glycans (35). Other measurements in solution (38) and on arrays (19,55), however, have not verified a strong preference of galectins for poly-LacNAc chains. Instead, studies have found that especially galectin-3 is able to bind internal LacNAc residues in poly-LacNAc chains and hence have more possible binding sites, whereas galectin-1 prefers terminal LacNAc residues, which in some assays cause increased binding to poly-LacNAc residues simply due to protrusion of the long carbohydrate chains (27).
To visualize the steric availability for simultaneous binding of multiple galectins to poly-LacNAc repeats on branched N-glycans, we performed structural modeling with galectins in both their dimeric and monomeric form. Mod-   7. A, models of simultaneous binding to ␤1,6-tetra-LacNAc tetra-antennary N-glycan by a galectin-1 homodimer (panels A-C) or a galectin-8N monomer (panels D-F). All four LacNAc repeats fit into a single galectin-1 homodimer (gray surface). Steric overlaps with one galectin-1 homodimer (gray surface) prevent a second galectin-1 (ribbon, panels A-C) from binding to the glycan, with the possible exception of an interaction involving the 4th (terminal)LacNAcrepeat(panelC).Incontrast,althoughtwogalectin-8Nmonomers would not be able to bind simultaneously to adjacent LacNAc units (panel D), they could bind to nonadjacent LacNAc units (panels E and F). In each panel, the second galectin is depicted as ribbons to show any overlap with the first galectin, which is depicted as a surface. B, homodimer of galectin-1 (PDB code 1W6P) superimposed onto the LacNAc motif in each branch of a tetra-antennary N-glycan (shown as 3D-SNFG symbols) with one LacNAc unit per branch. It is possible for simultaneous binding to occur when the LacNAc motifs are on the 3-and 6-branches (i.e. to a biantennary glycan) but not when the motifs are present on the same 3-or 6-branch, due to steric overlap (shown as red surfaces) between the backbone of the superimposed galectin-1 homodimers (colored uniquely and shown as transparent surfaces with ribbons).

Galectin glycan-binding in a natural context
eling the 3D structure of a galectin bound to each LacNAc unit in a tetra-antennary, mono-LacNAc N-glycan suggests that only two galectins can bind simultaneously despite the presence of four LacNAc units. This may explain the similar affinities observed for tetraversus bi-antennary N-glycans. The experimental findings from the cell assays were confirmed by the fluorescence anisotropy assay with purified EPO from the genetically engineered cells and fluoresceintagged N-glycans. However, for some galectins (galectin-1, galectin-8N, galectin-9C, and galectin-9N), we observed a slight decrease in affinity when binding was measured to EPO carrying bi-antennary glycans compared with tetra-antennary glycans. This suggest that there are subtle differences in binding properties for some galectins to glycans presented on a cell surface and a single protein in solution. Another possibility is that the population of glycoforms (e.g. the extent of sialylation) presented on the cell surface differs from those on the EPO protein. To further examine the importance of N-glycan branching for galectin binding, we evaluated the binding to synthetic N-glycan probes, representing bi-antennary, ␤1,4-tri-antennary, ␤1,6-tri-antennary, and tetra-antennary N-glycans.
Although it was clear that all the N-linked glycan probes bound significantly worse when compared with intact EPO, again only minor changes in galectin binding was observed to the various branched structures. Part of the previously reported importance of N-glycan branching for high-affinity galectin binding could be explained by the preference for poly-LacNAc biosynthesis on the ␤1,6 branch of N-linked glycans. Accordingly, we found that abrogation of poly-LacNAc biosynthesis through the targeting of the ␤3GnT2 gene caused a reduction in binding for most galectins, which is in agreement with previous studies (17, 24 -27). Furthermore, our modeling of galectin-glycan interactions also show that poly-LacNAc branch elongation beyond two repeats can accommodate binding of two galectin-8N or two galectin-3 CRDs. The fact that a LacNAc 3 branch would be able to facilitate the binding of an additional galectin CRD compared with a LacNAc 2 branch could also in part explain why the poly-LacNAc branch elongation is particularly important for galectin biology.
Previous studies have demonstrated the importance of ␣2,6linked sialic acid in preventing the binding of most galectins (17,19,42,43) and suggested that ␣2,6-linked sialylation function as a general "off switch" mechanism for galectin binding. An example is the protection of human and mouse T helper cells from galectin-1-induced apoptosis through ␣2,6-sialylation (42,56). It has also been suggested that a similar type of ␣2,6-sialylationdependent protection is found in mature medullary thymocytes (57,58). Interestingly, and in contrast, galectin-3-induced apoptosis is not blocked by ␣2,6-sialylation (42) unless the expression of ␣2,6-sialylation is induced by high expression levels of st6gal1 (59). A plausible explanation for this is the ability of galectin-3 to bind internal LacNAc structures, thereby avoiding the ␣2,6-sialylation off-switch. Only the prevention of LacNAc elongation by forced expression of st6gal1 (31) is able to block galectin-3 binding. In our system, sialylation differentially affected galectin binding. Removal of the terminal ␣2,3-sialylation had little or no effect on galectin-1 and galectin-9C, whereas galectin-3 had increased binding. For most of the weak binders (galectins-2, -4C, -4N, and -7), we found increased cell surface binding to nonsialylated N-glycans. This was also true for galectin-1, where removal of sialylation on EPO increased binding 4-fold. In contrast, ␣2,3-sialylation was important for galectin-8N binding in accordance with previous findings (22,29,46). When the terminal ␣2,3-sialylation was replaced with ␣2,6-sialylation, a substantial reduction in binding was observed for galectins-1, -3, and -8 and a 50% reduction for galectin-9C. The capping with ␣2,6-sialylation did not affect the binding of galectin-9N. In conclusion, our data are at large consistent with previous studies showing that the individual galectins display a large variation in terms of affinity toward naturally occurring glycans. For example, galectin-3 binds ϳ50% of all glycoproteins in serum, whereas galectin-8, -9, and especially galectin-1 bind a smaller fraction. A more restricted set of galectins (-2, -4, and -7) bind only trace amounts of serum glycoproteins (16). The assessment of galectins' glycan specificities has in many ways been limited by the artificial presentation and context of the glycan ligands in many of biochemical assays that have been used to date. To overcome this limitation, we have previously assessed galectin binding toward a glycoprotein purified from human serum (33,34). However, the identification of those glycan determinants important for galectin binding is hampered by the glycan complexity of natural glycoproteins. To simplify the glycan repertoire, we took advantage of a glycoengineered cell platform. Using this system, we effectively probed the binding preferences of the majority of human galectin CRDs toward both cell-surface glycans and individual glycoproteins in solution. The combination of the two approaches resulted in a direct comparison of the broad specificity of a galectin toward the sizable glycome presented on the plasma membrane, as well as to the fine-tuned specificity toward a single glycoprotein expressed in the same cell. We believe that our glycoengineered platform shows considerable potential in future studies to dissect the contribution of disease-associated alterations in glycosylation that could have dramatic effects on the expression of galectinbinding sites.

Cell lines
Glycogene targeting was performed in a parent CHOZN GS Ϫ/Ϫ (glutamine synthase) clone produced by ZFN KO (Sigma). Cells were maintained as suspension cultures in EX-CELL CHO CD Fusion serum-free media (Sigma), supplemented with 4 mM L-glutamine. Culture was in 50 ml of TPP TubeSpin Bioreactors at 200 rpm, maintained at 36.5°C, and with 5% CO 2 in air. All glycosylation CHO mutants were produced using the ZFN knockout strategy (60) with glycoprofiles subsequently verified by MS as described by Yang et al. (31).

Production of recombinant galectins
Recombinant human galectins (Gal-1 C3S , Gal-2, Gal-3, Gal-3 R186S , Gal-4C, Gal-4N, Gal-8C, Gal-8N, Gal-9C, and Gal-9N), and murine galectin-7, were produced in E. coli BL21 Star (DE3) cells (Invitrogen) and then purified by affinity chromatography on lactosyl-Sepharose as described previously by Salomonsson et al. (13). WT galectin-1 is known to be very sensitive to oxidative inactivation, so a C3S mutant was used in which the oxidation-sensitive Cys is substituted with Ser. This mutant has previously been shown to have the same affinity for carbohydrates and glycoproteins as WT galectin-1 (61). Protein size and integrity were determined by SDS-PAGE using 4 -20% Precise TM protein gels from Pierce (Nordic Biolabs, Täby, Sweden) in Tris/HEPES running buffer. Purified galectins were lyophilized and stored at Ϫ20°C until use. Protein concentrations were measured using the BCA assay (ThermoFisher Scientific), with spectrophotometric data analyzed with the Graphpad Prism software.

NHS-fluorescein labeling of galectins
NHS-fluorescein (ThermoFisher Scientific) dissolved in DMSO was added at a 10-fold molar excess to 2 mg/ml galectin solution in coupling buffer (20 mM HEPES). The solution was kept in the dark and incubated at room temperature with continuous mixing for 1 h. Labeled galectins was separated from unreacted dye by buffer exchange with PBS on a PD10 column (GE Healthcare). The degree of labeling was calculated according to the manufacturer's instructions.

Flow cytometry
All cells were grown in suspension culture to a density of 0.5-3 ϫ 10 6 cells/ml. Trypan blue staining was used to assess cell viability, and only populations with Ͼ90% viability were used for flow cytometry. Harvested cells were collected by centrifugation and then washed in 2-3 ml of PBS (with Ca 2ϩ and Mg 2ϩ , pH 7.4) with 200 mM lactose added (Sigma) to remove bound endogenous galectins. This was followed by an additional wash in 5 ml of PBS to remove excess lactose. An appropriate amount of PBS (0.4 ϫ 10 6 cells/200 l) was used to resuspend the cells, which were then transferred to 96-well round bottom microwell plates (ThermoFisher Scientific TM Nunclon TM Delta surface). From this point, the cells were kept on ice throughout the experiment. Cells were spun down in the plates and the PBS was removed. 100 l of PBS Ϯ 200 M lac-tose, and an additional 100 l of PBS supplemented with the required concentration of galectin (1, 2, 4, 8, and 16 M), was added to each well. The cells were then incubated at 4°C for 1 h with gentle agitation at a final galectin concentration of 0.5, 1, 2, 4, and 8 M Ϯ 100 mM lactose. Because of limited availability of human galectin-8C, this galectin was not included in the first flow cytometry experiment ( Fig. 2A). Fluorescence cytometry was performed using an LSRII cytometer (BD Biosciences), with data analyzed using the FlowJo software. Fluorescence data were collected using logarithmic amplification on 30,000 light scatter-gated events (cell counts). PI (Life Technologies, Inc.) was added to a final concentration of 40 -50 g/ml, immediately prior to data acquisition, to identify permeable cells. Negative controls comprised cells treated with PBS Ϯ 100 mM lactose.

Analyses of flow cytometric data
FlowJo software was used to determine the geometrical mean fluorescence intensity (GMFI) of sample cell populations. Permeable cells, as defined by high PI staining, were excluded from analyses together with cell aggregates identified from an FSC-A versus FSC-H plot. Individual data points for binding curves were calculated as GMFI values for cells without lactose, with subtraction of GMFI values for lactose inhibition controls. In this fashion, we could exclude any potential noncarbohydrate binding by the galectins.

3D molecular modeling
Glycam-Web (www.glycam.org 3 (39)) was used to generate low energy shapes for tetra-antennary mono-LacNAc N-linked , as well as a ␤1,6 tri-LacNAc tetra-antennary N- Multiple low-energy shapes were generated for each structure. These structures were added to the Gly-Spec Webtool (www.glycam.org/gr) 3 (40). Each glycan was modeled into the galectin CRD via the Gly-Spec webtool by uploading the crystal structures of a galectin-1 homodimer (PDB code 1W6P) and a galectin-8N monomer (PDB code 5T7S (62)). Models for galectins bound to each LacNAc unit in the ␤1,6 branch of the poly-LacNAc tetra-antennary N-glycan were generated by superimposing the galectin onto the initial structure of galectin-1 or galectin-8N bound to the innermost LacNAc. For this purpose we used Gly-Spec (www.glycam.org/gr (40)), which adjusts the glycosidic linkages to relieve any steric overlaps between the larger glycan and the protein surface. A second copy of the galectin was modeled onto the other LacNAc units by superimposition in UCSF Chimera (41). Figures of the 3D structures were created using VMD (63).

Galectin glycan-binding in a natural context Fluorescence anisotropy experiments to analyze binding of galectins to EPO glycoforms and synthetic N-glycans in solution
Fluorescence polarization (anisotropy) experiments to determine the affinity of EPO produced in different CHO mutants for galectin CRDs in solution were performed as described previously for endogenous galectin ligands and synthetic smallmolecule galectin inhibitors (18,32,64). In short, binding curves for the interaction of galectins with fluorescein-conjugated saccharide probes were determined (Fig. 5 and Fig. S4) in which increasing concentrations of galectin were titrated against a fixed concentration (0.02 M) of saccharide probe. In the presence of increasing amounts of galectin, anisotropy increases from the value for the free probe (A 0 ) to the point where the probe is entirely bound by galectin (A max ). These direct binding curves can be used to calculate affinity between the galectin and the fluorescein-tagged probe as described before and summarized in legends to Fig. 5 and Fig. S4.
For the fluorescein tagged N-glycans, we also analyzed the anisotropy values of the free probe (A 0 ) and galectin-bound probe (A max ) as follows. The measured FA for a particular molecule depends on a number (here, three were considered) of factors multiplied by each other (65). The first-called photoselection depends on properties of the fluorophore itself and is about 0.4 for fluorescein. This means that the maximum FA of a set of randomly oriented fluorescein molecules is 0.4 or 400 mA as expressed here. This is decreased by the next factor due to rotation of the molecule as a whole. This was calculated using the Perrin equation (1/(1 ϩ /⌽) from the fluorescence lifetime of fluorescein of ϭ 4 ns and the estimated rotational correlation times (⌽). For the free N-glycans, with estimated ⌽ of about 3 ns by the program hydro-NMR, the measured anisotropy would be about 180 mA. For the galectin-bound N-glycans, with ⌽ of the galectins estimated as about 12 ns (66) or by size of about 30 kDa, the measured anisotropy would be about 300 mA. Comparing this with the measured FA provides an estimate of the third factor due to local effects on the fluorescein moiety. Thus for free N-glycans, measured anisotropy (A 0 ) was about 40 mA, much lower than the estimated 180 mA, due to faster movement of the fluorescein moiety itself in addition to rotation of the tagged glycan as a whole. In the same way, the measured anisotropy of the galectin-probe complex (A max ) was much lower than 300 mA, but to a different degree and interpreted as described under "Results." To study the interaction of different EPO glycoforms with galectin CRDs, fixed concentrations of galectin and optimized small molecule probes were used (Fig. S4). EPO glycoforms were added at increasing concentrations. By measuring the anisotropy values for the different EPO concentrations, together with knowledge of A 0 and A max (obtained from the binding curves in Fig. S5), the K d values for the EPO-galectin CRD interaction can be calculated as described previously by Sörme et al. (18). An optimal probe was chosen for each galectin that permits use of the lowest possible concentration of galectin, while still achieving a (sufficient) rise in anisotropy that could be reliably measured (see below for details). The fluorescence anisotropy of the fluorescein-tagged probes was measured at the excitation and emission wave-lengths of 485 and 520 nm, respectively, using a PheraStarFS plate reader and PHERAstar Mars version 2.10 R3 software (BMG, Offenburg, Germany). All experiments were performed in PBS, with each EPO glycoform subject to serial 1:2 dilutions to generate seven test concentrations with which to test binding to each galectin. All fluorescence anisotropy measurements were performed (at least) in duplicate, in a 386-well plate, at a total volume of 20 l. K d values were calculated as weighted mean values from concentrations of EPO that generated between 15 and 85% inhibition, and where inhibition values of nosyl(1-4)-2-acetamido-2-deoxy-␤-D-glucopyranosyl(1-3)-␤-D-galactopyranosyl(1-4)-␤-D-glucopyranoside (29) at 0.02 M.
Galectin-7 affinities-These could not be tested with low enough galectin concentration, and due to lack of an appropriate fluorescent probe, the binding affinities for galectin-7 were thus not determined in this study.