Oligomannosidic glycans at Asn-110 are essential for secretion of human diamine oxidase

N-Glycosylation plays a fundamental role in many biological processes. Human diamine oxidase (hDAO), required for histamine catabolism, has multiple N-glycosylation sites, but their roles, for example in DAO secretion, are unclear. We recently reported that the N-glycosylation sites Asn-168, Asn-538, and Asn-745 in recombinant hDAO (rhDAO) carry complex-type glycans, whereas Asn-110 carries only mammalian-atypical oligomannosidic glycans. Here, we show that Asn-110 in native hDAO from amniotic fluid and Caco-2 cells, DAO from porcine kidneys, and rhDAO produced in two different HEK293 cell lines is also consistently occupied by oligomannosidic glycans. Glycans at Asn-168 were predominantly sialylated with bi- to tetra-antennary branches, and Asn-538 and Asn-745 had similar complex-type glycans with some tissue- and cell line–specific variations. The related copper-containing amine oxidase human vascular adhesion protein-1 also exclusively displayed high-mannose glycosylation at Asn-137. X-ray structures revealed that the residues adjacent to Asn-110 and Asn-137 form a highly conserved hydrophobic cleft interacting with the core trisaccharide. Asn-110 replacement with Gln completely abrogated rhDAO secretion and caused retention in the endoplasmic reticulum. Mutations of Asn-168, Asn-538, and Asn-745 reduced rhDAO secretion by 13, 71, and 32%, respectively. Asn-538/745 double and Asn-168/538/745 triple substitutions reduced rhDAO secretion by 85 and 94%. Because of their locations in the DAO structure, Asn-538 and Asn-745 glycosylations might be important for efficient DAO dimer formation. These functional results are reflected in the high evolutionary conservation of all four glycosylation sites. Human DAO is abundant only in the gastrointestinal tract, kidney, and placenta, and glycosylation seems essential for reaching high enzyme expression levels in these tissues.

N-Glycosylation plays a fundamental role in a wide variety of biological processes (1). Correct and complete N-glycosylation is a critical and a seminal parameter in the production of biotherapeutics. Recombinantly produced proteins may display glycan patterns that alter their in vivo half-life and cause severe immune reactions (2). Glycoengineering becomes increasingly important together with the choice of production system and screening for suitable clones (3).
N-Oligosaccharide attachment occurs at asparagine residues within an Asn-Xaa-Ser/Thr consensus sequence with any amino acid except proline as Xaa (4). The process starts in the endoplasmic reticulum, where the pre-formed, lipid-linked tri-glucosylated high-mannose-type tetra-decasaccharide Glc 3 Man 9 GlcNAc 2 is transferred onto the nascent protein by the oligosaccharyltransferase complex (5). The two terminal glucose residues are removed by glucosidase I and II. Protein folding is facilitated by the two lectin chaperones calnexin (CNX) 3 and calreticulin (CRT), and the remaining glucose residue is removed by glucosidase II (6). UDP-glucose glycoprotein glucosyltransferase 1 (UGT1) serves as a folding sensor. Immature proteins are re-glucosylated and reenter the CNX/CRT cycle (7). Terminally misfolded proteins undergo endoplasmic reticulum (ER)-associated protein degradation (ERAD), whereas the correctly folded proteins are exported from the ER to the Golgi followed by further glycan trimming and processing (4,6).

Glycan analyses of DAO from different sources
The N-glycosylation profiles of native DAO purified from human amniotic fluid, human Caco-2 cells, and re-purified from commercially available porcine kidney rhDAO expressed in and purified from CHO-K1, HEK293, and HEK293-Gly-coDelete cells are shown in Fig. 1 and Figs. S1-S6. GlycoDelete cells are engineered to produce a simplified mono-antennary trisaccharide consisting of N-acetylglucosamine (GlcNAc), galactose (Gal), and sialic acid (Neu5Ac) (3). The percentage distribution of the terminal carbohydrate residues and of the antennarity among the four glycosylation sites of all samples except porcine kidney DAO (pkDAO) is presented in Table 1 and Table S1, respectively. The Man, Neu5Ac, Gal, and GlcNAc percent distribution for pkDAO at Asn-115 (equivalent to Asn-110) is 49, 20, 32, and 0%; at Asn-541 (equivalent to Asn-538) is 0, 0, 97, and 3%; and at Asn-749 (equivalent to Asn-745) is 36, 0, 50, and 15%. Asn-168 is not present in the porcine protein sequence. Porcine kidney DAO carried only bi-antennary branches at all glycosylation sites. Additional glycosylation sites in the pkDAO sequence (Asn-437 and Asn-724) are described below.
To address the question whether the high-mannose content of Asn-110 is specific to DAO or also present in another CAO enzyme, we analyzed the glycan pattern of hVAP-1 (AOC3) expressed in CHO cells. The Asn-110 equivalent site in hVAP-1 is Asn-137, and it was also predominantly oligomannosidic with Man6 as the most abundant glycan variant (Fig. 1C). Mannose residues terminate 88% of the glycans at Asn-137. Sialic acid or GlcNAc terminating residues (6% each) account for remaining glycans. All glycan structures were bi-antennary. The amino acid sequence identity and similarity (positives in BLASTp) in the vicinity of Asn-137 compared with Asn-110 (Ϯ9 amino acids) are 67 and 75% (Fig. 6A), whereas the overall identity and similarity values between the two enzymes are only 38 and 56%, respectively.

N-Glycosylation of human diamine oxidase
sessed only bi-antennary branches, whereas human amniotic fluid DAO showed a clear shift to higher antennarity with 80% tetra-antennary structures. The Asn-168 glycosylation site was 100% fucosylated in DAO from amniotic fluid, HEK293, Caco-2, and CHO cells. HEK293-GlycoDelete cells cannot fucosylate and pkDAO does not have Asn-168.
The terminal glycan composition, including all four sites, is statistically highly significantly different with a p value of 0.0001 (Table S2). All pairwise comparisons except for Asn-538 with Asn-745 are also significantly different with p values of 0.016, which is the smallest possible p value using an exact non-parametric permutation test with this data set (Table S2). The percent distribution of the antennarity at the four glycosylation sites in the different DAO samples is overall also statistically highly significantly different with a p value of 0.0039 (Table S2). In the pairwise comparisons, only Asn-168 is significantly different from the other three sites with a p value of 0.063 (Table  S2).
Glycans at Asn-538 and Asn-745 of all samples except for HEK293-GlycoDelete (0/20% Man for Asn-538 and Asn-745) and pkDAO (0/36% Man for Asn-541 and Asn-749, which are the equivalent sites of Asn-538 and Asn-745), showed less than 10% Man, about 50% Gal, 27% Neu5Ac, and 20% GlcNAc terminal carbohydrate residues (Table 1). This terminal glycan distribution between Asn-538 and Asn-745 is statistically not different between the two sites (p value ϭ 0.41; Table S2). The antennarity is also very similar between the two glycan sites in our samples, with only native hDAO from amniotic fluid showing tri-and tetra-antennary branches (Table S1). In all samples, glycans at Asn-538 and Asn-745 sites carried a lower degree of sialylation than Asn-168 (Table 1).
In the pkDAO sample, 97% terminal Gal residues were found at Asn-541, the equivalent site to Asn-538 in human DAO. The Asn-749 site of pkDAO carried 36% terminating Man but no Neu5Ac residues (Fig. S3). Sialic acid residues in this sample might have been hydrolyzed during the protein extraction and 23.) cells. B, glycans detected at Asn-115 of DAO from porcine kidney (corresponding to Asn-110 in hDAO). C, glycans detected at Asn-137 of rhVAP-1 from CHO-K1 cells (equivalent to Asn-110 in hDAO). Relative abundances of glycans were calculated for each N-glycosylation site from total peak areas. Structures of the most abundant glycoforms are depicted according to Symbol Nomenclature of Glycans (SNFG) (69). Because linkage analysis was not performed, this is only a schematic illustration of possible isoforms. Man or M, mannose; Na, N-acetylneuraminic acid ϭ sialic acid; A, galactose; Gn, N-acetylglucosamine; Hex, hexose. A detailed explanation of the glycan nomenclature is published in Ref. 23. purification from porcine kidney tissue. We identified five occupied N-glycosylation sites for pkDAO (Asn-115 equivalent to Asn-110, Asn-437, Asn-541ϳAsn-538, Asn-724, and Asn-749ϳAsn-745). Except for Asn-724, these sites are predicted to be glycosylated using the default threshold potential of 0.5 (http://www.cbs.dtu.dk/services/NetNGlyc/ (26)). 4 For Asn-724, the potential is 0.48, and NetNGlyc predicts no N-glycosylation. The two glycosylation sites Asn-437 and Asn-724 are not present in the hDAO sequence. Interestingly, Asn-437 carried solely oligomannosidic glycans (Fig. S3), and 91% of terminating carbohydrates at Asn-724 were Gal, 7% GlcNAc, and only 1.5% Neu5Ac residues. All glycoforms on pkDAO showed a low degree of sialylation with the highest Neu5Ac content at Asn-115. In all other samples, the ratio is reversed with low Neu5Ac content at Asn-110 (mean of 7%) but a mean of 31-85% sialylation at the other sites (Table 1). As already mentioned above, the low sialic acid content in the pkDAO might be an artifact from the commercial isolation procedure.

Mutational analysis of the four glycosylation sites of rhDAO
Although the glycan patterns on three N-glycosylation sites were mainly complex-type, Asn-110 was consistently occupied by oligomannosidic glycans. This indicates that the enzymes involved in the conversion of high mannose to complex-type glycans do not have access to the Asn-110 site. The glycans might be interacting with local amino acids, and these interactions might be critical for protein folding and consequently secretion. To test this hypothesis, we exchanged Asn-110 with Gln-110.
Culture supernatants from hDAO-WT cells contained readily measurable DAO activity, but no activity was detected in the negative control and rhDAO⌬110 samples (Fig. 2B). The signals in the supernatants of the mutants were at least 100-fold lower. Cells transfected with rhDAO-WT, rhDAO⌬110, and empty vector plasmids were also tested for secreted DAO using a low-temperature immunofluorescence assay, which detects extra-membranous protein during the secretion process. Only cells expressing rhDAO-WT plasmid generated positive immunofluorescence signals (Fig. 2C). Fluorescence of the negative control and mutant rhDAO⌬110 was comparable (Fig. 2C). In Western blot analysis of the culture supernatants, hDAO was only detected from the rhDAO-WT plasmid in all three cell lines (Fig. 2D).
Although the rhDAO⌬110 gene was properly transcribed and translated, secretion of the mutated protein seemed severely disturbed. Therefore, we performed endoplasmic reticulum (ER)-and Golgi-specific immunofluorescence staining of rhDAO-WT, rhDAO⌬110, and empty vector transfected CHO-K1 cells. Recombinant hDAO-WT co-localized with both the ER and the Golgi apparatus ( Fig. 3A), whereas rhDAO⌬110 was exclusively located in the ER with no efficient transport to the Golgi compartment (Fig. 3B). The oligomannose glycans at Asn-110 seem essential for DAO folding and secretion.
To study the functional consequences of deleting the other glycosylation sites, we constructed single, double, and one triple mutant for expression in CHO-K1 cells. Western blot analysis, DAO activity assay, and DAO ELISA of culture supernatants showed that all mutants were secreted to varying degrees (Fig. 4, A and B). Single mutations of Asn-168 and Asn-745 had minimal to moderate effects on enzyme secretion with mean combined reductions of DAO activity and DAO concentration by 13 and 32%, respectively. DAO activity and secretion of rhDAO⌬538 mutants was about 71% or 3.5-fold reduced. The three double mutants showed significantly reduced DAO secretion with the strongest effects for rhDAO⌬538/745. The combined mean of DAO activity and concentration was 85% or 7-fold lower compared with wildtype plasmid (Fig. 4B). The secretion of the triple mutant was 15-fold lower (94%) in mutant versus control cells. Nevertheless, the amine oxidase activities of all mutants were comparable with WT rhDAO normalized to the amount of DAO protein (Fig. 4B). The Asn-745

N-Glycosylation of human diamine oxidase
mutation might have some effect on DAO activity. Based on these data, the importance of the different DAO glycosylation sites on protein secretion can be ranked from Asn-110 (mutations are detrimental with no protein secretion), Asn-538 (reduction by 71%), and Asn-745 (reduction by 32%) to Asn-168 (reduction by 13%).
Combinations of these mutations lead to increased secretion deficiencies with severe reduction of DAO export to the culture supernatant in the triple mutant.

Evolutionary conservation of the four DAO glycosylation sites
Considering the consistent presence of high-mannose glycan structures at Asn-110 from various tissues and cells, its pres-ence also in hVAP-1, the catastrophic effects of mutations in Asn-110, and the 15-fold reduction of DAO secretion in the Asn-168/538/745 triple mutant, we were interested in the evolutionary conservation of the four DAO glycosylation sites. Is the ranking of the functional consequences of DAO mutants also reflected in the degree of evolutionary conservation? In other words, is the Asn-110 site the most and Asn-168 the least conserved glycosylation site?
We selected 23 representative vertebrate sequences, including DAO from elephant shark. Because the genome of the elephant shark (Callorhinchus milii; chondrichthyes) is the slowest evolving vertebrate genome known, the 23 sequences cover

N-Glycosylation of human diamine oxidase
about 450 million years of evolution (27). No DAO homologue was found in the superclass of jawless vertebrates (agnatha). Fig. S8 shows the phylogenetic tree of DAO from these 23 vertebrate species representing the whole phylum. The overall DAO sequences are highly conserved demonstrated by the consistently short horizontal arms in the bony vertebrate sequences. The amino acids sequences (Ϫ9 to ϩ9 with Asn as 0 based on hDAO sequence) of the four DAO glycosylation sites of the 23 species are shown in Tables S4 -S7. Percentage identity and similarity calculations have been added to these tables. Table S8 summarizes these calculations, including the amino acid region Ϫ6 to ϩ4. Petrescu et al. (28), analyzing 386 occupied and nonredundant sequons, stated that occupied glycosylation sites differ in the frequency of the presence of certain amino acids from random distribution of amino acids in the respective protein only at positions Ϫ6 to ϩ4. Therefore, we analyzed this region separately, as it might indicate whether a glycosylation site is highly conserved compared with the extended vicinity of Ϫ9 to ϩ9. The conservation percentages of identity and similarity at positions Ϫ6 to ϩ4 among the 23 vertebrate sequences in the four glycosylation sites are presented in Table 2. These values reflect glycosylation site conservation.
Asn-110 is 100% conserved over a few hundred million years, where the region Ϫ1 to ϩ4 is 100% identical, which is higher compared with the other sites. There is low identity and similarity conservation at position Ϫ2 (17/17%), Ϫ3 (22/26%), and Ϫ5 (17/48%). Asn-110 is still 100% conserved, when the analysis was extended to all 174 vertebrate DAO sequences identified in the NCBI database. The conservation of Asn-110 in all these species indicates that it has been conserved for about 450 million years (29).
Among the 23 vertebrate sequences, Asn-168 is only found in humans and gorillas. Within the primates Asn-168 is 100% conserved in the Old World monkeys (catarrhini species). This explains the relatively low amino acid identity in the region Ϫ6 to ϩ4 of 48% (Table 2). Nevertheless, the similarity is 78%. Based on sequence identity and similarity data, Asn-168 is the least conserved glycosylation site.
The remaining two glycosylation sites, Asn-538 and Asn-745, which are both occupied by similar complex-type glycans and antennarity, are 91 and 87% conserved. The similarity score for the Asn-538 region Ϫ6 to ϩ4 of 88% is the highest of the four sites. The only amino acid, which is totally conserved within the Ϫ6 to ϩ4 regions, is proline at position ϩ4. The 100% conservation of the Pro-542 residue fits to the  The percentages in parentheses do not represent the quantified band but the combined mean of DAO activity and DAO concentrations relative to wildtype rhDAO from B. The band above 130 kDa is due to nonspecific antibody binding. B, mean DAO activity and DAO concentrations of culture supernatants. DAO activity and concentrations of the glycosylation mutant variants were normalized to wildtype rhDAO. The means of the DAO activity data were derived from duplicates of more than 20 measurement time points over 120 min in a steady-state assay. DAO concentrations were determined in duplicates. The means Ϯ S.E. of two independent experiments are shown. S.E. ϭ standard error of the mean.
Asn-745 shows the highest identity score of 73% with the similarity score only marginally increasing by 3 to a total of 76% ( Table 2). It is interesting that the region Ϫ4 to ϩ3 is highly conserved among the 23 vertebrate sequences. The identity and similarity percentages are 87 and 90%, respectively. These are the highest values among the four glycosylation sites. The previous 4 amino acids Ϫ5 to Ϫ8 (Ser-737 to Pro-740) show low conservation with an average less than 24%. The Cys-736 (residue Ϫ9) is forming the highly conserved (100%) intermolecular disulfide bond (Table S7).
This evolutionary conservation analysis indicates that Asn-110 is the most and Asn-168 the least conserved glycosylation site. This ranking is in agreement with the phenotypes from the mutational analysis of the four glycosylation sites.

Structural analysis of the N-glycosylation sites in human and porcine DAO
To improve our understanding about the differences in terminal glycan distribution and antennarity of the four N-glycans, we structurally analyzed the interaction of amino acids and carbohydrates in the heart-shaped homodimer of hDAO. The four N-glycosylation sites are distributed around the surface (Fig.  5A). Visual inspection shows that Asn-168, Asn-538, and Asn-745 protrude from the DAO surface (Fig. 5B), whereas Asn-110 is located in a hydrophobic cleft on the edge of the structure (Fig. 5B). To potentially find a structural explanation for the high conservation and unique oligomannosidic glycan composition of the Asn-110 site in the D2 domain, we performed a comparison with the corresponding Asn-137 site in hVAP-1 (Fig. 6A). The sequence identity of hDAO and hVAP-1 in the area surrounding the site (Ϯ9 amino acids) is 67%, whereas the overall identity between both enzymes is only 38%. Furthermore, the sequence similarity in this region is 80% considering the three amino acid pairs Ile/Val, Phe/Leu, and Ala/Val as similar (Fig. 6A). In the 3D structures, this site is similarly sur-rounded by hydrophobic residues on one side of the cleft both in hDAO and in hVAP-1 (Fig. 6, B-E). The attached oligomannosidic glycans are stabilized by conserved hydrogen bonds between the amide oxygen of Asn and the hydroxyl group of Thr (hDAO) or Ser (hVAP-1) in the N-glycosylation motif, and between the conserved Gln residues and the first GlcNAc of the glycan (Fig. 6, F and G). Solvent accessibility calculated for all the glycosylation sites in hDAO shows that Asn-110 is the least accessible N-glycosylation site with arbitrary units below 30 (Fig. S9, A and B). Asn-110 is located in the ␤1.3 ␤-strand that precedes the polar loop linking two ␤-strands of the central four-stranded ␤-sheet in D2 (Fig. 7A). The N-acetyl methyl group of the second GlcNAc in the core ManGlcNAc-GlcNAc trisaccharide forms hydrophobic interactions with Phe-103 and Phe-114 in hDAO (Fig. 7A). The Phe residues are part of the hydrophobic core of the D2 domain, and therefore, the N-acetyl-methyl group of the second GlcNAc contributes to the stabilization of the D2 domain. The Asn-168 site is located on a flat surface in the ␣4 helix of the D3 domain (Fig. 7B) and is the most accessible glycosylation site with arbitrary unit values above 100 (Fig. S9, A and B). This site is readily accessible for the glycan-processing enzymes resulting in predominantly complex-type glycans with high antennarity and 100% fucosylation.
Glycan Asn-538 in the D4 domain is located in the ␤4.3 ␤-strand of ␤-hairpin Arm 1 (amino acid residues 529 -559), which stabilizes the hDAO dimer by extending from one monomer to the other (Fig. 7C). The tip of Arm 1 is formed by amino acids ϩ3 to ϩ7 of the evolutionary highly conserved Asn-538 region (Asn-541-Pro-542-Trp-543-Ser-544 -Pro-545 in Fig.  7C). This structure plays an important role in attaching Arm 1 into the D3 domain of the other monomer, in particular via the -stacking interaction between Trp-543 in the tip of Arm 1 and Trp-200 in the D3 domain (Fig. 7C). The Asn-538 glycan further contributes to the inter-monomeric interactions and stabilizes the dimer because the N-acetyl-methyl of the first GlcNAc forms hydrophobic interactions with Pro-606 and Trp-609 from the other monomer (Fig. 7C).

N-Glycosylation of human diamine oxidase
The fourth glycosylation site Asn-745 is located near the C terminus in the D4 domain (Arm 2 amino acid residues 427-446 and Arm 3 residues 712-730; Fig. 7D). Asn-745 is part of an extensive hydrogen-bonding network (Fig. 7D) that connects the C terminus to the core of D4 (Asn-745-Glu-384 -Tyr-744) and to Arm 3 from the complementing monomer (Asn-745-Asn-724(B)-Tyr-744). The N-glycosylation motif is stabilized by a hydrogen bond between the amide oxygen of Asn-745 and the hydroxyl group of Thr-747 in the motif. The first GlcNAc of the attached glycan makes a water-mediated hydrogen bond to Asn-724(B) in Arm 3 and is therefore also involved in intermonomeric interactions (Fig. 7D). As the tip of Arm 2 lies in the active-site channel, the Asn-745-attached glycan is located near the entrance to the active site ( Fig. 7D; Fig. 8C). The Asn accessibility analysis ranked Asn-110 clearly as the least accessible glycan site followed by Asn-745, Asn-538, and finally the highly accessible Asn-168 (Fig. S9, A and B).
We also created a 3D homology model for porcine DAO using hDAO as a template to study the structural role of the N-glycosylation sites in porcine DAO. The sequence identity between human and porcine DAO is 85%. The monomers of the porcine DAO model superimpose with a root mean square deviation (r.m.s.d.) of 0.19 and 0.18 Å with the respective monomers of hDAO. The high identity and low r.m.s.d. values are indicators of high-accuracy models (30). A sequence identity of Ͼ50% and an r.m.s.d. value of about 1 Å are threshold criteria for a good model. The model was manually inspected and subjected to quality assessment using different programs/servers. These data are described under "Experimental procedures." Three of the five N-glycosylation sites we found in pkDAO are conserved in hDAO. The properties and accessibility of the Asn-115 and Asn-541 sites are similar between porcine and human DAO (Fig. 8A and Fig. S9). The third conserved site Asn-749 in D4 near the C terminus is clustered together with Asn-437 and Asn-724 (Fig. 8A). Asn-437 is located in the tip of Arm 2 and Asn-724 in the tip of Arm 3 (Fig. 8B). Therefore, unlike hDAO, the porcine protein has a cluster of three N-gly-  (18,70,71). The molecular surface view of hDAO (white) (B) with the N-glycosylation sites (green) and hydrophobic patches (cyan) shows that Asn-110 is located in a hydrophobic cleft (dashed box), whereas the other N-glycosylation sites protrude out of the hDAO structure.

N-Glycosylation of human diamine oxidase
cans near the active-site entrance (Fig. 8C). Of these N-glycosylation sites, Asn-437 is the least accessible (Fig. S9) explaining the presence of 100% mannose-terminating carbohydrates with 0% fucosylation of the Asn-437-attached glycans in porcine DAO. The high rate of mannose-terminating glycan branches at Asn-749 in porcine kidney DAO with 36% compared with on average 3% in the other analyzed glycans of human DAO at the corresponding Asn-745 site might be explained by the presence of three glycan structures in close structural proximity. The most accessible site Asn-724 is located in the tip of Arm 3, and accordingly, more processed glycans are found at this site.

Discussion
Human DAO is an important enzyme in the catabolism of exogenous histamine. The role of DAO in the degradation of endogenously released histamine is not clear. Increased histamine concentrations contribute to the symptomatology in patients with mastocytosis, mast cell activation syndrome, urti-

N-Glycosylation of human diamine oxidase
caria, and life-threatening anaphylaxis among other diseases. The physiological role of several hundredfold increased plasma concentrations of DAO during pregnancy is not known. A "simple" protection mechanism from increased endogenous or exogenous histamine has been postulated for decades, but supporting data are limited. In non-pregnant individuals DAO is expressed at high levels only in the gastrointestinal tract and the kidney (17). Animal experiments clearly demonstrated a role for DAO in the protection from exogenous, food-derived histamine (13,14). No data have been published about any mechanisms by which high level protein expression of DAO might be accomplished. The functions of the individual glycosylation sites of DAO have also not been characterized.
In our previous studies the glycosylation profile of CHOderived rhDAO showed that Asn-168, Asn-538, and Asn-745 carried mammalian-typical complex-type glycans, whereas Asn-110 was consistently occupied only by oligomannosidic variants (23). Because mannose-type glycans are not common in mammals and the presence of high-mannose residues at Asn-110 in CHO cells could be cell line-specific, glycosylation profiles of native and recombinant DAO proteins from various sources were analyzed.
The Asn-110 residue consistently carried high-mannose glycosylation in all analyzed cell types. The other three glycosylation sites were occupied by complex-type glycans with some micro-heterogeneity. The pkDAO equivalent site Asn-115 and human VAP-1 equivalent site Asn-137 were also occupied mainly by high-mannose glycans. These results strongly imply that the consistent glycosylation pattern of hDAO at Asn-110 is site-specific, irrespective of the DAO source, whereas the glycans of the other three sites follow some tissue-and cell line-specific variable complex-type patterns characteristic for mammalian cells. Nevertheless, the glycans at Asn-168 are statistically significantly different compared with Asn-538/745, irrespective of tissue-and cell line-specific variance.
Why does Asn-110 of hDAO carry high-mannose glycans, whereas the other three sites are "normally" glycosylated? It is likely that the rate of the final cis-Golgi mannose trimming to Man5 is decreased at this glycosylation site. The formation of Man5 is crucial for the addition of GlcNAc by N-acetyl-glucosaminyltransferase 1 (GnT1). Interestingly, Asn-110 of Caco-2 cells and amniotic fluid-derived hDAO carried significant proportions of Man5, which were, however, not further processed by GnT1. Terminal GlcNAc residues serve as anchor for

N-Glycosylation of human diamine oxidase
␣-mannosidase II, which performs the final trimming of the mannose residues to Man3 (31). The following processing steps convert high mannose to complex-type glycans. Consequently, rate limitation of the ␣-mannosidase I and GnT1 reactions leads to the secretion of high-mannose glycans. Why are these endogenous glycan processing enzymes not able to convert Asn-110 glycans into complex-type variants?
Attempts to release the oligosaccharide from Asn-110 of hDAO using N-glycosidase F were not successful, indicating that the glycan core region is not accessible (data not shown). Both the analysis of the X-ray structures and the low Asn accessibility calculations support this result. The very low Asn accessibility numbers of Asn-110 in hDAO and Asn-115 in pkDAO predict 100% high-mannose-type glycans, 0% fucosylation, and 100% bi-antennarity, and this is exactly what we found (32).
Wyss et al. (33) described extensive inter-residue carbohydrate interactions between several Man and the two GlcNAc residues in the single high-mannose N-glycosylation site Asn-65 of the adhesion domain of human CD2. Several mannose residues folded back to and interacted with the core trisaccharide (33). These high-mannose glycans are important for stabilization of the correct fold of the CD2 domain, but it is likely that most of the effect stems from the core trisaccharide. The ManGlcNAc-GlcNAc glycan motif stabilizes the CD2 protein by increasing the folding equilibrium constant 200-fold, and this model might be generalizable (34). The interactions of Man residues with the core trisaccharide might explain the presence of Man5 at Asn-110 as these Man5 residues are not accessible for GnT1 anchoring and further processing.
Based on the structural analysis of the two human CAOs (hDAO and hVAP-1), the protein-glycan interactions in the Asn-110 site are extremely conserved positioning the core trisaccharide on top of the hydrophobic cleft. The attached glycan is also involved in stabilizing the hydrophobic core of the D2 domain. Considering that mutations in Asn-110 lead to a complete secretion stop of DAO, we postulate that the Asn-110attached glycan is important for the correct folding of the D2 domain and thus the folding of a functional hDAO dimer. The 100% evolutionary conservations of Asn-110 in 174 vertebrate CAO sequences, but also the 100% conservation of amino acids Phe-104, Pro-109, Thr-112, and Glu-113 over a few hundred million years, are in agreement with this interpretation. Based

N-Glycosylation of human diamine oxidase
on the interactions of the core trisaccharide residues with the hydrophobic amino acid cleft, it is possible that the glycans at Asn-110 are not only involved in folding but also in stabilizing the protein structure after secretion. The singly glycosylated (Man5 to Man9) RNase B in contrast to the unglycosylated but otherwise identical RNase A shows increased dynamic stability (solvent accessibility, unfolding of the protein, and protease resistance) throughout the molecule (35). This mechanism might be applicable to all four DAO glycosylation sites on each monomer, likely independent of glycoform type.
The micro-heterogeneity of glycans present at Asn-168, Asn-538, and Asn-745 is a common phenomenon with mammalian glycoproteins and can be attributed to the processing reactions in the Golgi apparatus, which are influenced by cell type, culture conditions, and the steric environment of the glycosylation site (36). Nevertheless, only glycan structures at Asn-168 carried a maximum of four antennae with the expected exception of DAO expressed in HEK293-GlycoDelete cells. Asn-168 was also the only glycosylation site with 100% fucosylation. These data are in agreement with the highly significant positive correlation of Asn accessibility with complex-type glycans, fucosylation and antennarity (32). The Asn-168 glycosylation site is only found in primates and might be considered a gain of function mutation without any indication of what the function might be.
The Asn-538 and Asn-745 local vicinities are also surfaceexposed, but why are the antennarity (mostly bi-antennary branches) and the terminal carbohydrate content significantly different compared with Asn-168? Asn-538 showed not only the highest evolutionary similarity conservation but also had the second strongest effect on DAO secretion (reduction by 71%) of the single mutants after Asn-110 mutations with no expression. The location of this site in the structure of hDAO suggests a structural role by aiding in the docking of Arm 1 into the D3 domain of the second monomer. Glycan-processing enzymes might not have unhindered access compared with Asn-168. The reduced Asn-538 mutant expression might be caused by reduced folding efficiency or increased ERAD-dependent degradation.
Single mutations of Asn-745 have a moderate effect on DAO secretion, but in combination with Asn-538 DAO expression is reduced by at least 85%. Interestingly, Asn-745 is structurally very close to the second loop of protruding Arm 2. In analogy to Asn-538, it is again tempting to speculate that glycans at Asn-745 play a structural role in efficient dimerization. The presence of mainly bi-antennary branches combined with on average 65% Gal-or GlcNAc-capped and only about 30% Neu5Ac-capped terminating glycans might be explained by somewhat reduced access of glycan-processing enzymes in the Golgi for both Asn-538 and Asn-745. Artifactual de-sialylation during the purification process of the different samples is not likely considering that the sialic acid content of Asn-168 is in all cases 2-5-fold higher.
Are these complex-type glycans high in Gal-terminating glycoforms at Asn-538 and Asn-745 also involved in the stabilization of secreted DAO? The protein conformation of human erythropoietin seems to be stabilized via hydrophobic interactions of complex-type but less high-mannose gly-cans with the hydrophobic protein-surface areas (37). Especially the N-acetyl-lactosamine units of complex-type glycoforms, in particular galactose residues, seem to contribute to the formation of hydrophobic surfaces allowing interaction with hydrophobic amino acid regions (38).
The Asn-110, Asn-538, and Asn-745 sites of hDAO are also present in pkDAO, but two additional occupied glycosylation sites, Asn-437 and Asn-724, have been identified. The Asn-115 glycosylation site is structurally very similar to Asn-110, and mutations are expected to have a comparable phenotype with complete loss of secretion. In the structural model for pkDAO, Asn-437 and Asn-724 are located in the tip of ␤-hairpins Arm 2 and Arm 3, respectively, in the four-stranded ␤-sheet, which is covered by the C terminus of the other monomer. The two additional glycans in pkDAO cluster together with the conserved Asn-749 site near the active-site entrance. These three glycans might significantly facilitate local folding and consequently dimerization efficiency. It might also be speculated that mutations in these glycans adjacent to the substrate channel lead to decreased enzymatic activity. The 100% high-mannosetype glycans without any fucosylation at Asn-437 can be readily explained by the very low Asn surface-accessibility scores (32).
Site-directed removal of N-glycosylation sites leads to a reduction in the catalytic activity in the majority of enzymes (39). Human DAO seems to be an exception. Although Asn mutations severely influence secretion rate, the catalytic activity of secreted DAO was not different between the various mutations. Even the triple mutant showed normal DAO activity based on ELISA-measured enzyme concentrations. It is possible, although not likely, that Asn mutants changed the affinity of the antibodies used in the ELISA and distort the activity/ concentration relationship. We consider this unlikely as the DAO ELISA data with the different mutant and wildtype protein are in agreement with protein determinations using absorbance measurements of purified mutants.
In this study, we have shown that glycosylation of DAO plays an essential role in protein folding and secretion. Future experiments should address two questions about N-glycosylation. Does it stabilize secreted hDAO and affect pharmacokinetic parameters? D'Agostino et al. (40) published that DAO from porcine kidney, human plasma, and placenta was rapidly cleared with a half-life of less than 15 min in isolated and perfused rat livers. Southren et al. (41) observed unexplained ϳ6-fold increased DAO plasma levels in three cirrhosis patients. These studies suggest that human DAO is likely cleared via hepatic uptake. In general, the liver is the main organ for glycoprotein clearance (42). Is DAO cleared via the mannose, the asialoglycoprotein receptor, both, or none (43)? No data have been published demonstrating the involvement of Asn-110 oligomannose residues or complex-type glycans at the other glycosylation sites in the half-life of DAO. The presence of about 90% terminal Man with no fucosylation at Asn-110, 100% fucosylation at Asn-168 with high Neu5Ac capping, and 50% terminal Gal, 20% terminal GlcNAc carbohydrates with at least 50% fucosylation at Asn-538 and Asn-745 are not helpful in addressing this question.
If recombinant human DAO might be used in patients to rapidly degrade exogenous or endogenous histamine, the role N-Glycosylation of human diamine oxidase of these carbohydrates in the half-life and clearance must be established. Decades ago, porcine kidney DAO was used in the treatment of several diseases, but limited success is readily explainable by high immunogenicity, low specific activity, and short circulatory half-life (44, 45). The pharmacokinetic issues might be circumventable with modern biotechnological glycoengineering methods, but first we need to know which glycan structures are involved in the clearance of hDAO in vivo.

Heterologous expression of hDAO in HEK cells
HEK293 cells were cultivated in FreeStyle TM 293 expression medium (Gibco), HEK293-GlycoDelete cells (3) in a 1:1 mixture of FreeStyle TM 293 expression medium ,and EX-CELL 293 serum-free medium (Sigma) in a shaker-incubator at 37°C, 7% CO 2 , and at 140 rpm. Both growth media were supplemented with 10 M CuSO 4 (Sigma) according to Ref. 23.
For the heterologous expression of hDAO, the cells were transfected with 20 g of rhDAO expression plasmid (23) using Amaxa Nucleofector TM I/program X-01 and the Amaxa cell line nucleofector kit V (Lonza Group Ltd.). 5 days post-transfection, the cultures were centrifuged for 10 min at 170 ϫ g, and the supernatants were sterile-filtered through Stericup filter units (Merck Millipore).

Expression of hDAO in Caco-2 cells
The human colorectal cancer cell line Caco-2 is known to express hDAO. On batch day 0, the confluent cells were seeded at a ratio of 1:6 in T175 flasks using DMEM, high glucose, pyruvate (Gibco) with 10% fetal calf serum (FCS; Sigma and incubated at 37°C in 7% CO 2 . On day 4, the culture supernatant was removed and replaced with a 1:1 mixture of DMEM, high glucose, pyruvate, and Ham's F12 (Gibco) with 10 units/ml high molecular weight heparin (Gilvasan, 1000 units/ml). On day 7, the culture supernatant was harvested by centrifuging for 10 min at 170 ϫ g and sterile filtration through Stericup filter units.

Purification of rhDAO from HEK culture supernatants
The culture supernatants of HEK293 and HEK293-GlycoDelete were ultra-and diafiltrated using the Labscale TFF System in combination with a Pellicon XL 50 Ultrafiltration Cassette-Biomax Polyethersulfone with a 100-kDa molecular mass cutoff (both Merck Millipore). The supernatants were concentrated 10-fold, and the culture media were replaced by a 10 mM potassium phosphate buffer, pH 7.4 (Merck). The samples were loaded onto a 1-ml HiTrap Heparin HP column at a flow rate of 0.2 ml/min using an Äkta purifier HPLC device (both GE Healthcare). Stepwise elution was performed with 0.25, 0.5, and 1 M KCl (Sigma) in 10 mM potassium phosphate buffer, pH 7.4. The 0.5 M KCl eluate was concentrated and desalted using Merck Millipore's Amicon Ultra-0.5-ml centrifugal filter units (10-kDa molecular mass cutoff) and a 10 mM potassium phosphate buffer, pH 7.4.

Purification of hDAO from human amniotic fluid
Amniotic fluid from one individual was centrifuged for 1 h at 8000 ϫ g and sterile-filtered through Stericup filter units (Merck Millipore). Ultra-and diafiltration (25-fold concentration), ammonium sulfate precipitation, and hydrophobic interaction chromatography were conducted as described previously (23). The 12.5% (w/v) ammonium sulfate eluate from hydrophobic interaction chromatography was concentrated and desalted using Merck Millipore's Amicon Ultra-0.5-ml centrifugal filter units (100-kDa molecular mass cutoff) and a 10 mM potassium phosphate buffer, pH 7.4 (see above). The sample was then loaded onto three HiTrap Heparin HP prepacked 5-ml columns connected in series at a flow rate of 1.0 ml/min. Stepwise elution was performed with 0.25, 0.5, and 1 M KCl (Sigma) in 10 mM potassium phosphate buffer, pH 7.4. The 0.5 M KCl eluate was concentrated and desalted using Merck Millipore's Amicon Ultra-0.5-ml centrifugal filter units (10-kDa molecular mass cutoff) and a 10 mM potassium phosphate buffer, pH 7.4.

Purification of hDAO from Caco-2 cells
Human diamine oxidase from Caco-2 cells was purified according to the protocol established for amniotic fluid. Caco-2 culture supernatant was concentrated 100-fold in the initial ultra-and diafiltration step.

Purification of DAO from porcine kidney
The purity of Sigma's "diamine oxidase from porcine kidney" was insufficient for glycan analyses. We estimated based on amine oxidase activity compared with rhDAO from CHO cells that our Sigma batch contained less than 0.5% porcine kidney DAO (data not shown). Therefore, the protein was dissolved in 50 mM Hepes buffer, pH 7.5 (Hepes buffer grade, AppliChem), and purified according to the protocol described for hDAO from amniotic fluid with one exception. DAO could not be detected in the 12.5% but in the 0% ammonium sulfate eluate from hydrophobic interaction chromatography. Hence, this fraction was used for further purification.

SDS-PAGE
Reducing and non-reducing SDS-PAGE analyses were performed as described previously (23).

N-Glycan analysis using liquid chromatography-electrospray ionization-mass spectrometry (LC-ESI-MS)
Because the degree of purity was not sufficient for direct LC-ESI-MS analysis of DAO from amniotic fluid, porcine kidney, Caco-2 and HEK293 cells, the respective DAO band was cut from an SDS-polyacrylamide gel. DAO from HEK293-Gly-coDelete was pure enough for direct use. rhVAP-1 from CHO cells was purified according to Ref. 46.
LC-ESI-MS analyses were performed as described in Ref. 23 with the following alterations. A Thermo Fisher Scientific Bio-Basic C18 separation column 5-m particle size of 150 ϫ 0.320 mm instead of 0.360 mm was used. A gradient from 97% solvent A and 3% solvent B was applied instead of 95% A and 5% B.
When evaluating the peptide MS/MS data using X! Tandem (www.thegpm.org/tandem/) 4 the following settings were changed. Fragment mass error of 0.05 Da and Ϯ7 ppm parent mass error instead of 0.1 Da and Ϯ 100 ppm. Missed cleavage sites allowed was set to 3 instead of 2.

N-Glycosylation of human diamine oxidase
Manual glycopeptide searches were done using Data Analysis 4.0 (Bruker). For the relative quantification of the different glycoforms, peak areas of extracted ion chromatograms of the first six isotopic peaks were summed. All observed charge states were considered, and adduct formation (ammonium, sodium) was taken into account. MS/MS spectra were used for the verification of the glycopeptides (based on the detection of oxonium ions HexNAc (m/z ϭ 204.1) and Hex ϩ HexNAc (m/z ϭ 366.1), as well as the unique Y1 ion (peptide ϩ HexNAc)).

Phylogenetic analysis
All putative CAO sequences from vertebrates and cephalochordata were extracted from the NCBI database. After a search with the Basic Local Alignment Search Tool (BLAST) using hDAO (NP_001082.2) as reference, all sequences with Ն40% query cover and Ն30% sequence identity were considered potential hDAO homologues. The obtained sequences were aligned using the Multiple Sequence Alignment by Log-Expectation (MUSCLE) (47) as implemented in MEGA version 6 (48). Further selection and analysis of the sequences were continued in MEGA6. Only complete sequences with a length between 700 and 800 amino acids showing the catalytic amino acids Asp-373, Tyr-461, and the three copper-binding histidines His-510, His-512, and His-675 (numbering according to hDAO) were considered functional CAOs. Sequences containing transmembrane domains (TMDs) were identified using the TMHMM server version 2.0 (49). Only sequences without a TMD were considered to be DAOs. The only exception is the class of chondrichthyes, where only one sequence was identified by BLAST, and although this sequence shows a TMD, it was included in the DAO sequence comparisons.
The phylogenetic tree was calculated with a selection of 24 DAO sequences (Table S3) representing the whole phylum vertebrata and the cephalochordata as outgroup. Maximum likelihood calculations were performed using PhyML (50) and the LG substitution model (51) with an estimated number of invariable sites, four substitution rate categories, and an estimated ␥ distribution parameter. The LG substitution model was determined to be the best amino acid substitution model by ProtTest (52) under the Akaike Information Criterion. Statistical support for each node was evaluated by performing 500 bootstrap repetitions on the dataset.

Quantitative real-time PCR
To identify potential differences in transcription of the rhDAO-WT and rhDAO⌬110 mutant gene, quantitative realtime PCR (RT-qPCR) was performed. Forty eight hours posttransfection, up to 2 ϫ 10 6 cells were harvested by centrifuging for 5 min at 170 ϫ g. RNA isolation was performed according to Ref. 54.
Messenger RNA levels were determined by RT-qPCR using SensiFAST SYBR Hi-ROX kit (Bioline, London, UK). Eight hundred ng of isolated RNA were treated with DNase I (Fermentas, Waltham, MA). cDNA was generated by using the high-capacity cDNA reverse transcription kit (Thermo Fisher Scientific, Waltham, MA), consisting of 10ϫ RT Buffer, 25ϫ dNTP Mix (100 mM), 10ϫ RT random primers, MultiScribe reverse transcriptase, and RNase inhibitor. The resulting 20 l of cDNA were diluted by adding 60 l of nuclease-free water. The 10-l RT-qPCR mixture consisted of 1 l of generated cDNA, 0.25 l of primer 1 (accaagtacctcgatgtcgg), 0.25 l of primer 2 (ggcatttcaaagaggcagag), 3.5 l of nuclease-free water, and 5 l of SensiFAST SYBR Hi-ROX kit master mix. Quadruplets of each cDNA sample were used for the PCR performed on the Rotor-Gene-Q (Qiagen). The expression levels of the DAO relative to the GAPDH gene, an endogenous con-N-Glycosylation of human diamine oxidase trol, were determined using the 2 Ϫ⌬⌬Ct method. Average fold differences in the transcript levels of the glycosylation mutant rhDAO⌬110 were determined by comparison with the wildtype gene. Ct (threshold cycle, number of cycles required to generate a fluorescence signal that exceeds background levels) values of the empty vector control samples were consistently comparable with the water no-template-control.

Organelle-specific staining
The immunofluorescence staining was done according to Ref. 57 with adaptation to suspension cells. Therefore, 1.8 ϫ 10 7 CHO-K1 cells were harvested 48 h post-transfection at 1000 rpm for 5 min, washed once with PBS, and fixed for 10 min with 4% formaldehyde solution in PBS ϩ Mg 2ϩ rotating end-overend. The cells were washed 6 times with PBS ϩ Mg 2ϩ and resuspended in 500 l of PBS ϩ Mg 2ϩ . Thirty l of the suspension were transferred to a poly-L-lysine-coated (0.01%, Sigma) IBIDI 15-m chamber 12-well glass slide and incubated for 10 min. The excess suspension was removed with a cotton swab, and the slide was air-dried. To each well 0.2% BSA in PBS ϩ 0.1% saponin was added and incubated for 1 h in a humid chamber. The cells were then incubated with an anti-DAO antibody solution (anti-ABP1 antibody produced in rabbit, SAB1410491, Sigma, 1:100 in 0.2% BSA in PBS ϩ 0.1% saponin) together with monoclonal mouse anti-PDI antibodies (Enzo Life Sciences, 1:500 dilution) or with monoclonal mouse anti-GM130 antibodies (BD Transduction Laboratories, 1:100 dilution) for 1 h. After washing, secondary antibody solution with Alexa Fluor 488 donkey anti-rabbit IgG antibodies (Molecular Probes, diluted 1:200 in 5% FCS in PBS ϩ 0.1% saponin) and Alexa Fluor 555 donkey anti-mouse IgG antibodies (Molecular Probes, 1:100 dilution) were added, and cells were incubated for 1 h in the dark. After washing and a final fixation step, the samples were mounted in Vectashield (Vector Laboratories), covered with a coverslip, and viewed in a Zeiss Axio Observer Z1 fluorescence microscope. The microscope was equipped with an LCI Plan-Neofluar 63 ϫ 1.3 NA glycerol objective, the Zeiss filter sets 38 HE (excitation BP 450 -490 nm/emission BP 500 -550 nm) and 14 (excitation BP 510 -560 nm/emission LP 590 nm), and a Zeiss Axiocam 503 mono CCD camera.

Western blotting
On day 5, the transiently transfected cultures were centrifuged for 10 min at 170 ϫ g, and the culture supernatants were sterile-filtered. The undiluted samples were mixed with NuPAGE LDS sample buffer (4 times) (Life Technologies, Inc.) ϩ 5% 2-mercaptoethanol (Sigma) and incubated for 10 min at 70°C. A detailed description of the protocol can be found in Ref. 23) A 1:5000 dilution of a serum IgG fraction from rabbits, immunized with purified rhDAO, was used as the primary antibody.

DAO activity assay
Sterile-filtered supernatants from transient cultures (day 5) were analyzed for DAO activity using a chemiluminescence based enzyme activity assay according to Ref. 58.

DAO ELISA
DAO-specific ELISA of culture supernatants was performed according to Ref. 59.

Homology modeling of porcine DAO and structural analysis of the N-glycosylation sites in the 3D structures of hDAO, hVAP-1, and porcine DAO model
Because no crystal structure of porcine DAO was available, we used homology modeling to create a 3D model for porcine DAO and to study its N-glycosylation sites. The porcine sequence was retrieved from UniProt database (ID:F1SSL3). In the BODIL modeling environment, MALIGN was used for the sequence alignment and for 3D visualization (60). Homology models were generated with MODELLER (61) using the template structure (PDB code 3HI7) (18). The model with the lowest MODELLER objective function was chosen for further analysis. Model evaluation was estimated with PROCHECK (62), Verify3D (63), and ProSA-web (64). PROCHECK listed 90.7% of residues in the most favored regions and 8.3% in the additionally allowed regions of the Ramachandran plot (62). Verify3D showed that 85.1% of the residues have an average 3D to 1D score of 0.2 or more (63). The threshold for passing the test requires at least 80% of amino acids with a score of 0.2 or more. ProSA-web showed a Z-score of Ϫ9.88, which is in the range of scores typically found for native proteins of similar size (64).
Visual inspection by superimposition with the template structure was also used to evaluate the 3D model of porcine DAO. The visual analysis of hDAO (PDB code 3HII) (18) and hVAP-1 (PDB code 4BTW) (65) X-ray structures and the porcine DAO model was performed with PyMOL (66). Blast searches (67) using hDAO (UniProt ID: P19801) and hVAP-1 (UniProt ID: Q16853) as a query were performed to find homologous sequences. All the figures were prepared with PyMOL (66).
Solvent accessibility of N-glycosylated Asn residues was determined using NACCESS (http://wolf.bms.umist.ac.uk/naccess) 4 with a probe size of 5 Å to simulate the accessibility of the glycosylating enzymes into the glycosylation sites. The absolute accessibility values ("arbitrary units") were compared between the N-Glycosylation of human diamine oxidase different glycosylation sites. All waters, sugars, ligands, and other molecules that are not part of the polypeptide chain were removed prior to the measurements.

Statistical analysis
Except for p value calculations in Table S2, standard descriptive statistical analysis was used. For p value determinations in Table S2, an exact non-parametric permutation test was run in R (version 3.3.2) with the two null hypotheses of "equal frequency distribution of terminal glycan residues or equal antennarity in the glycan chains at all four sites." The p value for the permutation test is defined as the proportion of permutations, which result in a value of the test statistic being greater or equal to the originally observed value. Ten thousand permutations were used to calculate the global test p value. For pairwise comparison, all possible permutations (128 for terminal glycan composition and 64 for antennarity) were utilized. The small number of possible permutations for pairwise comparisons limits the smallest achievable two-sided p value of 2*1/128 ϭ 0.0156 for terminal glycan composition or 2*1/64 ϭ 0.0312 for antennarity. Because Caco-2 cells show only 100% bi-antennarity, the lowest achievable p value is actually only 0.0625.