The Role of Sialylated Glycans in Human Platelet Endothelial Cell Adhesion Molecule 1 (PECAM-1)-mediated Trans Homophilic Interactions and Endothelial Cell Barrier Function*

Platelet Endothelial Cell Adhesion Molecule 1 (PECAM-1) is a major component of the endothelial cell intercellular junction. Previous studies have shown that PECAM-1 homophilic interactions, mediated by amino-terminal immunoglobulin homology domain 1, contribute to maintenance of the vascular permeability barrier and to its re-establishment following inflammatory or thrombotic insult. PECAM-1 glycans account for ∼30% of its molecular mass, and the newly solved crystal structure of human PECAM-1 immunoglobulin homology domain 1 reveals that a glycan emanating from the asparagine residue at position 25 (Asn-25) is located within the trans homophilic-binding interface, suggesting a role for an Asn-25-associated glycan in PECAM-1 homophilic interactions. In support of this possibility, unbiased molecular docking studies revealed that negatively charged α2,3 sialic acid moieties bind tightly to a groove within the PECAM-1 homophilic interface in an orientation that favors the formation of an electrostatic bridge with positively charged Lys-89, mutation of which has been shown previously to disrupt PECAM-1-mediated homophilic binding. To verify the contribution of the Asn-25 glycan to endothelial barrier function, we generated an N25Q mutant form of PECAM-1 that is not glycosylated at this position and examined its ability to contribute to vascular integrity in endothelial cell-like REN cells. Confocal microscopy showed that although N25Q PECAM-1 concentrates normally at cell-cell junctions, the ability of this mutant form of PECAM-1 to support re-establishment of a permeability barrier following disruption with thrombin was significantly compromised. Taken together, these data suggest that a sialic acid-containing glycan emanating from Asn-25 reinforces dynamic endothelial cell-cell interactions by stabilizing the PECAM-1 homophilic binding interface.

Mutagenesis studies performed nearly 20 years ago revealed the importance of five amino acids within IgD1 for PECAM-1/ PECAM-1 homophilic interactions (7), including a critical lysine residue at position 89, as a Lys-89 3 alanine substitution abolishes PECAM-1-mediated homophilic interactions (7), localization to cell-cell borders (9), cytoprotection against proapoptotic stimuli (18), and the ability of PECAM-1 to contribute to the vascular permeability barrier (16). The recently solved crystal structure of PECAM-1 IgD1 and IgD2 revealed homophilic binding interfaces involving additional amino acids in both IgD1 and IgD2 (8); however, somewhat surprisingly, Lys-89 was not one of them. Thus, a molecular explanation for how mutation of this residue results in loss of homophilic binding remains a mystery.
PECAM-1 is heavily glycosylated, with ϳ30% of its molecular mass contributed by nine complex N-linked carbohydrate chains (3,19), including two within IgD1 at Asn-25 and Asn-57 and one within IgD2 at Asn-124. Interestingly, although each of these asparagine residues is located within a homophilic binding interface, as demonstrated in the recent crystal structure (8), the participation of their attached glycans in homophilic binding could not be determined because the IgD1-D2 constructs used for crystallization were produced in insect cells, which have simpler N-glycans consisting only of short truncated terminal mannose residues. Kitazume et al. (20) reported the presence of terminal ␣2,6-linked sialic acid residues on the glycans of murine endothelial PECAM-1 and found that deletion of ST6Gal-1, which encodes the ␤-galactosidase sialyltransferase that adds this sugar residue to the ends of glycan chains, resulted in loss of PECAM-1 from endothelial cell-cell borders. A more recent study by the same group (21) found that ␣2,6-sialylated oligosaccharides inhibit murine PECAM-1 homophilic adhesion, further implicating ␣2,6 sialic acids in PECAM-1/PECAM-1 interactions. Murine and human PECAM-1 differ in a number of important respects, the most notable relevant example being the absence of the Asn-25linked glycan in the murine molecule; it has a glutamine at this position. Finally, a large number of studies examining the homophilic binding properties of PECAM-1 have employed a CHO cell-secreted recombinant human PECAM-1/IgG chimeric protein that binds with high affinity to human PECAM-1 (5, 6) and has been used in vivo to block ischemia/reperfusion injury (22). Because CHO cells express only ␣2,3-sialyltransferases, which add terminal ␣2,3-linked sialic acids to the terminus of glycan chains, but lack ␣2,6 sialyltransferases (23-25), it is likely that human versus murine PECAM-1 differ in the molecular requirements necessary for supporting PECAM-1/ PECAM-1 homophilic interactions, a concept reinforced by the long-held observation that human and murine PECAM-1 cannot bind to each other (5,6).
The purpose of this investigation, therefore, was to examine the relative ability of ␣2,3versus ␣2,6-linked sialic acid residues to contribute to human PECAM-1 trans homophilic interactions. Given the species specificity of the glycan at Asn-25 of human IgD1, we also examined the role of this glycan in concentrating PECAM-1 at cell-cell borders and in regulating junctional integrity. Using glycan-specific recombinant PECAM-1/ IgG constructs containing only ␣2,3 sialic acid moieties versus both ␣2,3 and ␣2,6 sialic acids, we found that the presence of ␣2,6 sialic acid inhibits, rather than supports, homophilic binding of human PECAM-1. Unbiased molecular docking analysis revealed that ␣2,6-sialylated glycan binds across the face of IgD1 in such a way as to inhibit the ability of an Asn-25-linked glycan terminating in ␣2,3 sialic acid to interact with Lys-89. Taken together, these data emphasize the species-specific requirements for PECAM-1 homophilic adhesion and provide a molecular explanation for the role of Lys-89 in mediating PECAM-1 homophilic interactions.

Results
␣2,6-linked Sialic Acid Residues Inhibit PECAM-1-mediated Homophilic Interactions via an Intradomain Electrostatic Interaction with Lys-89 -PECAM-1 is predominantly glycosylated with hybrid and complex N-glycans (19), and previous studies have implicated terminal ␣2,6 sialic acid residues in supporting the ability of murine PECAM-1 to form homophilic interactions. Because human PECAM-1 contains an N-glycosylation site at amino acid 25 that is not present in murine PECAM-1, and because human PECAM-1/IgG produced in CHO cells, which lack ␣2,6-sialyltransferase activity, has been used for many years to characterize the homophilic binding properties of PECAM-1, we examined whether addition of ␣2,6 sialic acids to human PECAM-1/IgG might enhance its binding ability. To accomplish this, we transfected a cDNA construct encoding PECAM-1/IgG into a specialized CHO cell line that had been stably transfected with ST6Gal-1 (a generous gift from Ajit Varki, University of California, San Diego). As shown in Fig. 1A, Sambucus nigra (SNA) lectin, which binds selectively to proteins containing ␣2,6 sialic acids, bound to ␣2,6ϩ␣2,3sialylated PECAM-1/IgG but not to the ␣2,3-sialylated form of PECAM-1/IgG that had been generated from wild-type CHO cells. SNA lectin also bound to PECAM-1 expressed on REN cells (data not shown). Somewhat surprisingly, the ␣2,6ϩ␣2,3sialylated form of PECAM-1/IgG was completely unable to interact homophilically with WT PECAM-1-expressing REN cells (Fig. 1B) unless it was desialylated with neuraminidase ( Fig. 1C). In contrast, ␣2,3-sialylated PECAM-1/IgG bound in a dose-dependent manner to WT PECAM-1-transfected REN cells (Fig. 1B), whereas desialylation had only a minor deleterious effect on homophilic binding (Fig. 1D). The relative binding of each of these glycoforms, at a 100 g/ml concentration, to PECAM-1-transfected REN cells is quantified in Fig. 1E. Thus, in contrast to murine PECAM-1, ␣2,6-linked sialic acids significantly inhibit, rather than support, the homophilic binding of human PECAM-1, whereas ␣2,3-linked sialic acid-containing glycans play only a minor role in steady-state trans homophilic interactions.
To understand the mechanism by which ␣2,6-sialylated glycans might inhibit PECAM-1/PECAM-1 homophilic interactions, a molecular docking analysis was used to predict the binding of ␣2,6-sialylated lactosamine (comprised of sialic acid, galactose (Gal), and GlcNAc, coordinates obtained from PDB code 1JSI (26)) to the recently solved crystal structure of human PECAM-1 IgD1 (8). The coordinates of this ligand were energyminimized using PRODRG2 (27), and partial charges of ␣2,6sialylated lactosamine and human PECAM-1 IgD1 were generated using the Gasteiger module in AutoDockTools (28). A cubic grid encompassing the entire surface of IgD1 was used for blind docking analysis, and the ligand was allowed to interact with the entire surface. As depicted in Fig. 2A, the resulting lowest estimated free energy of binding computed using Auto-DockTools revealed that the ␣2,6-linked sialylated lactosamine binds across the lateral face of PECAM-1 IgD1 to form an electrostatic interaction with amino acid Lys-89, with a best estimated free energy of binding of Ϫ6.62 kcal/mol. The distance from carbon 1 of the GlcNAc residue of the docked ␣2,6 lactosamine residue to the nitrogen atom of Asn-57 is 16.4 Å, a spacing that easily accommodates the core N-linked antenna that can span 10 -17 Å (GlcNAc-GlcNAc-Man 3 , PDB code 1GYA (29)), shown as three green rings attached to two blue rings in Fig. 2A) emanating from Asn-57. Because the mannose core antenna is flexible, carbon 4 of the terminal mannose can be positioned next to carbon 1 of the GlcNAc residue of the docked ␣2,6-sialylated lactosamine. Moreover, the ␣2,6 lactosamine⅐PECAM-1 IgD1 complex was also fitted into the crystal lattice to observe the involvement of the ␣2,6-sialylated glycan in the interactions between the docked PECAM-1 IgD1 and the opposing PECAM-1 IgD1 molecule that forms trans homophilic interactions. In contrast to the distance between the nitrogen atom of Asn-57 to the GlcNAc residue of ␣2,6 lactosamine, that of Asn-25 residing on the trans homophilicinteracting molecule is 19 Å, which is too large to harbor the core N-glycan. Therefore, ␣2,6-sialylated is not likely to emanate from Asn-25 of the opposing IgD1 molecule. For this reason, Asn-57 in the ␣2,6ϩ␣2,3-sialylated form of PECAM-1/ IgG might carry an inhibitory ␣2,6 sialic acid residue that inhibits homophilic interactions involving Lys-89. This observation is consistent with the experimental observations shown in Fig. 1.
To determine whether ␣2,6ϩ␣2,3-sialylated PECAM-1/IgG is actually modified at Asn-57 by ␣2,6 sialic acid, the protein was digested with trypsin/Lys-C to produce a series of glycopeptides and peptides. ␣2,6-Sialylated glycopeptides were captured using SNA-agarose affinity chromatography, eluted, deglycosylated with PNGase F, and subjected to tandem mass spectrometry. As shown in Fig. 2B, Asn-57 is indeed ␣2,6 sialylated, thus explaining the inability of ␣2,6ϩ␣2,3-sialylated PECAM-1/IgG to interact homophilically with PECAM-1-expressing cells. To determine whether an ␣2,3-linked sialic acidcontaining glycan emanating from Asn-25 might contribute to the adhesive properties of human PECAM-1, we again performed unbiased molecular docking, this time analyzing the binding of ␣2,3-sialylated lactosamine to PECAM-1 IgD1 using the identical docking parameters of ␣2,6-sialylated lactosamine. As shown in Fig. 3B, ␣2,3-sialylated lactosamine is predicted to bind in a grove of IgD1 in such a way as to form a hydrogen bond between its carboxyl moiety and the ⑀-amino group of Lys-89 with an estimated free energy of Ϫ4.8 kcal/mol. Bound in this fashion, when fitted into the crystal lattice, the hemiacetal carbon atom (C1) of the GlcNAc residue of the ␣2,3-sialylated lactosamine is 11.3 Å away from the nitrogen atom of Asn-25 residing on the trans homophilic-interacting PECAM-1 IgD1, providing favorable space for the core N-glycan to occupy (three green rings attached to two blue rings in Fig. 3B). This observation is consistent with the notion that ␣2,3-sialylated lactosamine extending from the core glycan of Asn-25 is capable of forming an interdomain molecular bridge to the Lys-89 residue of an opposing PECAM-1 IgD1 molecule interacting in trans. These modeling data predict that an Asn-FIGURE 1. ␣2,6-linked sialic acid residues inhibit the homophilic interactions of human PECAM-1. A, ␣2,3-sialylated PECAM-1/IgG was purified from the culture supernatant of PECAM-1/IgG-transfected wild-type CHO cells, which express only ␣2,3-sialyltransferases. ST6Gal-1-transfected CHO cells were employed to express ␣2,6ϩ2,3-sialylated PECAM-1/IgG. Both types of PECAM-1/IgG were also desialylated with neuraminidase and subjected to immunoblotting and lectin blotting analyses. Note that only PECAM-1/IgG purified from ST6Gal-1-transfected CHO cells is SNA-positive. Also note that neuraminidase treatment results in the generation of species of lower apparent molecular weight and reduced reactivity with wheat germ agglutinin, which binds to sialic acid residues and, to a lesser extent, the GlcNAc residues that are exposed following neuraminidase treatment. B, concentration-dependent binding of PECAM-1/ IgG to PECAM-1-transfected REN cells. Note that the presence of ␣2,6 sialic acid on PECAM-1/IgG inhibits its trans homophilic binding down to background levels, as defined by the binding of normal human IgG. C, removal of sialic acid residues restores the binding of ␣2,6-linked sialylated PECAM-1/IgG while having little effect on the binding of ␣2,3 sialylated PECAM-1/IgG (D). E, quantification of PECAM-1/IgG binding from eight independent FACS binding assays of PECAM-1/IgG at a concentration of 100 g/ml. The p values were derived from a Student's t test. MFI, median fluorescence intensity.

Molecular Modeling of the N25-linked Glycan of Human PECAM-1 Reveals That It Forms a Molecular Bridge between
25-associated sialylated glycan might reinforce PECAM-1-mediated homophilic interactions.
The Asn-25 Glycan Supports Functional Dynamic PECAM-1-mediated Homophilic Interactions-Sabri et al. (30) recently reported that sialic acid residues have a more pronounced effect on dynamic, rather than static, cell adhesive interactions. To determine whether the glycan attached to Asn-25 plays a functional role in PECAM-1-mediated homophilic adhesion, we generated three REN cell lines, each expressing a different form of PECAM-1: WT PECAM-1, Lys-89, a homophilically crippled  is circled in red. The truncated GlcNAc-GlcNAc-Man 3 glycan emanating from Asn-25 in PECAM-1 molecule 2 is represented by orange-tipped yellow sticks. The amino acid residues in red boxes have been shown previously by Newton et al. (7) to be implicated in PECAM-1 homophilic interactions. Note that Lys-89 is not located on the binding interface. B, en face view of the likely full-length complex carbohydrate emanating from Asn-25 molecule 2. The first two GlcNAc residues emanating from this residue are shown in blue, with the Man 3 antenna shown in green. A trisaccharide comprised of N-acetylglucosamine, galactose, and ␣2,3-linked sialic acid (␣2,3-sialylated lactosamine) was subjected to unbiased molecular docking and found to bind with high affinity to PECAM-1 molecule 1 in such a way as to hydrogen-bond (black line) with the ⑀-amino group of Lys-89. Thus, this glycan forms an intermolecular bridge between the two opposing PECAM-1 IgD1 domains interacting in trans. C-ter, C terminus; N-ter, N terminus. PECAM-1, and an N25Q form of PECAM-1 that lacks its associated glycan residue. As shown in Fig. 4A, the N25Q mutant form of PECAM-1 immunoprecipitated from detergent-solubilized REN cells ran with a slightly smaller apparent molecular weight, consistent with its loss of a carbohydrate chain. Although PECAM-1 IgG binding to the K89A mutant form of PECAM-1 was less than 10% of that observed for its binding to WT PECAM-1 (Fig. 4, B and C, and Ref. 7), N25Q PECAM-1 supported normal steady-state binding of ␣2,3-sialylated PECAM-1/IgG (Fig. 4, B and C), concentrated normally at cellcell borders (Fig. 4D), and maintained baseline junctional integrity (Fig. 5, A and B) to the same degree as WT PECAM-1, all well described features of PECAM-1-mediated steady-state homophilic binding. In stark contrast, however, when the per-meability barrier was disrupted with thrombin, the rate of recovery was severely compromised in REN cells expressing N25Q PECAM-1 (Fig. 5, A and C). These data provide strong support for the notion that the glycan attached to Asn-25 plays an important role in supporting dynamic PECAM-1/PECAM-1 homophilic interactions.

Discussion
Nearly all vertebrate cell surface receptors that pass through the endoplasmic reticulum and Golgi on their way to the plasma membrane, including those that participate in cell adhesion and signaling, are glycosylated. Although the role of glycans in cell adhesion has been best characterized in the Selectin (31) and sialic acid-binding Ig-like lectins (Siglec) (32) families of glycan-binding proteins, other notable cell adhesion molecules in which carbohydrate residues have been shown to play a prominent role include E-Cadherin (33,34), neural cell adhesion molecule (35), ICAM-1 (36,37), vascular cell adhesion molecule (38), and junctional adhesion molecule (39). Despite the fact that the molecular mass of PECAM-1 is ϳ30% carbohydrate (3,19), there have been only a handful of studies examining the potential contribution of PECAM-1-linked glycans to its adhesive and signaling function, and most of these have been performed in murine cells. The purpose of this investigation, therefore, was to identify, in the context of its recently solved crystal structure, the specific glycans that emanate from the human PECAM-1 homophilic binding domain and deter-mine how they might contribute to PECAM-1-mediated adhesion and function.
One of the major findings of this work is that, in contrast to its murine counterpart, homophilic binding interactions of human PECAM-1 are supported by ␣2,3rather than ␣2,6linked sialic acid residues and that ␣2,6-linked glycans, rather, are strongly inhibitory (Fig. 1). The expression and specific linkage of sialic acids to the underlying glycoconjugate chain is known to vary in a cell type-and vascular bed-specific manner (47) and can also be strongly influenced by the metabolic state of the cell (48) and whether the cells have been subject to various inflammatory stimuli (49 -51). More important, however, is the observation that endothelial cell adhesion molecules from different species often differ in both the number and location of functional glycosylation sites. For example, human ICAM-1, vascular cell adhesion molecule, E-selectin, and P-selectin have 8, 6, 9, and 11 N-linked glycans, whereas their murine counterparts express 10, 7, 11, and 12, respectively (47). In the case of PECAM-1, murine PECAM-1 contains seven as opposed to nine N-linked glycosylation sites and is missing the major glycosylation site, Asn-25, in IgD1 that is present at the homophilic binding interface of human PECAM-1 (Fig. 3A).
A mechanistic explanation for the differential effect of ␣2,6versus ␣2,3-linked sialic acid moieties on the ability of human PECAM-1 to interact homophilically was provided by unbiased molecular docking studies in which ␣2,6 sialylated and ␣2,3 sialylated lactosamines were allowed to interact in silico with human PECAM-1 IgD1. These studies predict that, although ␣2,6 lactosamine interacts with higher affinity to IgD1 than ␣2,3 lactosamine, in agreement with the experimental observations of Kitazume et al. (21), it binds across the face of IgD1 in such a way as to block the ability of Lys-89 to participate in homophilic binding (Fig. 2A). In contrast, molecular modeling of the binding of ␣2,3 sialic acid emanating from Asn-25 reveals that this glycoconjugate is capable of forming an intermolecular bridge between two opposing PECAM-1 interacting in trans (Fig. 3B), thereby reinforcing homophilic binding interactions.
A close examination of the interaction of this sialylated glycan with IgD1 of an opposing PECAM-1 molecule is shown in Fig. 6 and reveals four highly conserved amino acids, Ile-7, Asn-88, Lys-89, and Lys-91, that bind directly with the glycan. Of  DECEMBER 9, 2016 • VOLUME 291 • NUMBER 50 these, mutation of Lys-89 has been shown previously to have a strongly deleterious effect on PECAM-1-mediated homophilic binding (7,9), whereas mutation of Lys-91, which is predicted to form only a single hydrogen bond with the glycan, had little effect (7). Interestingly, the ⑀-amino group of Lys-89 forms a salt bridge with the carboxylate moiety of the ␣2,3 sialic acid in much the same way as the carboxyl group of sialic acid forms a salt bridge with a conserved Arg-97 residue that is present in all Siglecs studied to date (32). The effects of mutating the other two glycan-binding residues of human PECAM-1 (Ile-7, which is linearly distant but conformationally close and Asn-88) have not been examined. Arg-73 (ϭ Arg-90 counting the 17-amino acid signal peptide present in murine PECAM-1), which was predicted by Kitazume et al. (20) to form a glycan-interacting site in murine PECAM-1, appears to have no role in glycan binding in human PECAM-1, as it is neither at the homophilic binding interface (8) nor does mutation of Arg-73 to Ala have any effect on human PECAM-1-mediated homophilic binding (7). PECAM-1 has been shown in a number of laboratories to be an important contributor to the maintenance of the vascular permeability barrier as well as to its restoration following thrombotic or inflammatory insult (15,16,52,53). The ability of PECAM-1 to localize to, and concentrate at, cell-cell junctions, where it carries out this function, is completely dependent on its ability to form trans PECAM-1/PECAM-1 homophilic interactions, an adhesive property of the PECAM-1 extracellular domain that was first proposed more than 25 years ago (54), shown to be due to diffusion trapping of the receptor at cell-cell junctions 10 years later (9) and most recently found to support endothelial cell junctional integrity in a potentially regulatable manner (17,55). Interestingly, Cioffi et al. (56) have shown recently that not only are sialic acids an important determinant of endothelial barrier integrity but that, although the arterial endothelium is likely to display both ␣2,3 and ␣2,6 sialic acid residues, the cell surface receptors of microvascular endothelial cells are primarily ␣2,3-sialylated. Taken together with the findings presented here that ␣2,3-linked sialic acid residues support, whereas ␣2,6 sialic acids inhibit, PECAM-1-mediated homophilic interactions, it is tempting to speculate that PECAM-1 might be a target for a novel mode of regulating endothelial cell permeability; namely, dynamic glycan modification. Whether changes in the expression of glycosyltransferases and neuraminidases that take place during bouts of thrombosis and inflammation result in changes in the linkage specificity of the sialic acids bound to PECAM-1 and how this might contribute to endothelial cell junctional integrity, leukocyte trans endothelial migration, wound healing, and repair should be fascinating topics of future investigation.

Experimental Procedures
Construction of IgD1 N-glycan Mutants of Human PECAM-1-pcDNA3 (Invitrogen) containing wild-type PECAM-1 was used as a template to generate Asn 3 Gln constructs at amino acid positions 25 and 57. Site-directed mutagenesis was performed using the QuikChange Lightning kit (Agilent Technologies, Santa Clara, CA) following the directions of the manufacturer. N25Q mutant PECAM-1 was generated using 5Ј-GTG CAA AAT GGG AAG CAG CTG ACC CTG CAG TGC-3Ј (forward) and 5Ј-GCA CTG CAG GGT CAG CTG CTT CCC ATT TTG CAC-3Ј (reverse) primers. The Asn-25 PECAM-1 mutant was verified by DNA sequence analysis before use.
Cell Lines-The human mesothelioma cell line REN (57) was maintained in RPMI 1640 supplemented with 10% FBS and 2 mM L-glutamine. REN cells are a human mesothelioma cell line that exhibits a number of endothelium-like properties, including a cobblestone-shaped, monolayer morphology with well defined cell-cell borders and expression of numerous endothelial cell adhesion molecules (58). REN cells are also PECAM-1negative, are easily transfectable, and have, as a result, been used for more than 20 years as endothelial cell surrogates to study PECAM-1 biology (9,16,59). Plasmids encoding WT PECAM-1, K89A PECAM-1, and N25Q PECAM-1 were transfected into REN cells using Lipofectamine LTX (Thermo Fisher Scientific, Waltham, MA) following the instructions of  Fig. 3B), the hydroxyl group of the bound sialic acid also forms a hydrogen bond (black dashed line) with Lys-91. Glycan interactions are additionally stabilized by a hydrogen bond between the ␣ amino group of Ile-7 and the hydroxyl group of galactose (yellow ring) and two hydrogen bonds between Asn-88 and GlcNAc (blue ring). All four of these glycan-binding amino acids are completely conserved in human, pig, cow, dog, and mouse PECAM-1. B, 90 o rotation view of A. C-ter, C terminus; N-ter, N terminus.

Function of PECAM-1 Glycans
the manufacturer. Transfected cells were selected in culture medium containing 500 g/ml G418 (Corning, Manassas, VA). Cell lines expressing comparable levels of each PECAM-1 isoform were obtained by FACS. PECAM-1 protein from each line was evaluated by immunoprecipitation of transfected REN cell lysates with mouse anti-PECAM-1 mAb 1.3.
Binding of PECAM-1/IgG to PECAM-1-transfected REN Cells-REN cells expressing wild-type and mutant PECAM-1 isoforms were incubated with varying concentrations of PECAM-1/IgG in PBS containing 10% FBS for 1 h at room temperature. Fab fragments of the mouse anti-PECAM-1 monoclonal antibody, mAb PECAM-1.2, detected with Alexa 647-conjugated goat anti-mouse IgG (HϩL) (Invitrogen), were included so that PECAM-1/IgG binding could be analyzed on cells of comparable cell surface expression levels of PECAM-1. After incubation, cells were washed with PBS, incubated for 30 min with DyLight 488-conjugated mouse anti-human IgG (HϩL) (Jackson ImmunoResearch Laboratories, West Grove, PA), washed again, and then subjected to a flow cytometric analysis using a BD Accuri TM C6 plus (BD Biosciences). Normal human IgG (Sigma) was used as a control of PECAM-1/IgG.
Immunoblotting and Lectin Blotting-PECAM-1/IgG isoforms expressing various linkages of sialic acid were subjected to SDS-PAGE and immunoblot analysis using 1 g/ml of biotinylated wheat germ agglutinin (Vector Laboratories, Burlingame, CA), which binds to both ␣2,6and ␣2,3-linked sialic acids as well as exposed GlcNAc residues, or SNA lectin (Vector Laboratories), which is specific for ␣2,6-linked sialic acid residues. The lectins were incubated with transfer membranes at room temperature for 1 h in Tris-buffered saline ϩ Tween 20 containing 0.1 mM CaCl 2 , washed, and incubated with streptavidin conjugated with HRP for an additional hour. PECAM-1 antigen was detected by immunoblotting with 3 g/ml of the anti-PECAM-1 mAb, PECAM-1.3, followed by HRP-conjugated goat anti-mouse (HϩL). The blots were developed with SuperSignal TM West Pico chemiluminescent substrate (Thermo Fisher Scientific).
Identification of ␣2,6-sialylated Glycan Site(s) in ␣2,6ϩ␣2,3sialylated PECAM-1/IgG-Lectin affinity capture was performed as described previously (60) with some modification. Briefly, ␣2,6ϩ␣2,3-sialylated PECAM-1/IgG was reduced with DTT (Sigma) at 56°C for 20 min and alkylated with iodoacetic acid (Sigma) at room temperature in the dark for 15 min. ProteaseMax TM surfactant (Promega, Madison, WI) was also added to facilitate denaturation of protein as described by the manufacturer. The reduced and alkylated protein was dialyzed in 50 mM ammonium bicarbonate buffer overnight at 4°C to remove DTT. After that, the dialyzed protein was digested with Trypsin/Lys-C mixture (Promega) following the instructions of the manufacturer. The resulting peptides and glycopeptides were subsequently incubated with SNA-agarose (Vector Laboratories) overnight at 4°C to select ␣2,6-sialylated glycopeptides. The SNA-bound glycopeptides were washed with 50 mM ammonium bicarbonate and eluted with 0.5 M lactose as described by the manufacturer. Then, the eluted glycopeptides were deglycosylated with PNGase F (Promega). The deglycosylated glycopeptides were desalted with ZipTip (EMD Millipore, Billerica, MA) and subjected to LC/MS-MS. Nanoflow HPLC was performed using nanoACQUITY 10 cm ϫ 75 m column (Waters, Milford, MA) in-house packed with Magic C18 3-m (New Objective, Inc., Woburn, MA). Solvent A consisted of 0.1% formic acid in deionized water, and solvent B consisted of 0.1% formic acid in acetonitrile. The effective flow rate was 300 nl/min. The gradients were as follows: solvent B, 5% 0 min, 30% 53 min, 90% 63 min, 90% 65 min, 5% 70 min, and 5% 90 min. The standard 60,000 (nominal) resolution scan on the LTQ Orbitrap Velos mass spectrometer (Thermo Fisher Scientific) was set up in data-dependent acquisition mode to perform the survey scan. The 10 most abundant ions were selected for subsequent fragmentation using collision-induced dissociation with a relative collision energy of 35%. Orbitrap-selected precursor ions were fragmented and then sequenced in the linear ion trap. All data analyses were performed in the MaxQuant environment version 1.4.1.2 (61), and all reagents used in this section were of mass spectrometry or equivalent grade.
Barrier Function Measurements-REN cells were plated on 8W10Eϩ electrode arrays precoated with 0.1% gelatin and allowed to grow to confluence for 2-3 days. Electric cell substrate impedance sensing (ECIS) measurements were performed in duplicate chambers using ECIS model Z (Applied Biophysics, Troy, NY) at multiple frequencies to evaluate barrier functions as described by Giaever and Keese (62) and as described previously by us (16). On the day of the experiment, the culture medium was replaced with 380 l of RPMI supplemented with 1% FBS. Cells were allowed to achieve a stable baseline and then stimulated with 20 l of thrombin (Sigma) at a final concentration of 5 units/ml. ECIS measurements were modeled using ECIS software as described previously (16).
Prediction of Sialic Acid Binding Modes-The recently solved crystal structure of human PECAM-1 IgD1 (8) was used in molecular docking analysis to predict binding modes of ␣2,3and ␣2,6-sialylated lactosamine, the three-dimensional structures of which were obtained from PDB codes 1JSN and 1JSI (26), respectively. The coordinates of these ligands were energyminimized using PRODRG2 (27). Partial charges of ␣2,3and ␣2,6-sialylated lactosamine and human PECAM-1 IgD1 were generated using the Gasteiger module in AutoDockTools (28). A cubic grid containment having 114 ϫ 126 ϫ 80 grid points per side with a spacing of 0.375 Å was constructed to cover the entire IgD1 for blind docking analysis, and the ligand was allowed to bind the entire surface of IgD1, where the side chains of the interacting amino acid residues are allowed to be flexible for optimization of the interactions with the ligand. Affinity maps of the grids were calculated using AutoGrid 4.2. Auto-Dock 4.2 (28) was employed to dock either ␣2,3or ␣2,6-sialylated lactosamine onto IgD1 using the Lamarckian genetic algorithm, consisting of 200 runs and 270,000 generations, with a maximum number of energy evaluation set to 2.5 ϫ 10 6 . The resulting docked conformations were analyzed using Auto-DockTools, and the conformation of either ␣2,3or ␣2,6-sialylated lactosamine with the lowest estimated free energy of binding obtained from semiempirical free energy force field in AutoDock 4.2 was selected for further analysis. Graphical representation was generated using the PyMOL Molecular Graphics System version 1.2 (Schrödinger, LLC).
Statistical Analyses-Student's t test with unequal variance was performed for all statistical analyses using Microsoft Excel and expressed as the mean Ϯ S.D.
Author Contributions-P. L. conducted experiments, analyzed the results, and wrote the paper. C. P. developed REN cell transfection, immunofluorescence staining, and ECIS protocols. D. K. N. suggested experiments and analyzed and interpreted data. J. Z. helped design and interpret molecular modeling studies. M. J. T. designed mass spectrometry protocols and analyzed mass spectra. P. J. N. designed experiments, analyzed and interpreted data, and wrote the paper. All authors reviewed the results and approved the final version of the manuscript.