Endo-F3 Glycosynthase Mutants Enable Chemoenzymatic Synthesis of Core-fucosylated Triantennary Complex Type Glycopeptides and Glycoproteins*

Chemoenzymatic synthesis is emerging as a promising approach to the synthesis of homogeneous glycopeptides and glycoproteins highly demanded for functional glycomics studies, but its generality relies on the availability of a range of enzymes with high catalytic efficiency and well defined substrate specificity. We describe in this paper the discovery of glycosynthase mutants derived from Elizabethkingia meningoseptica endoglycosidase F3 (Endo-F3) of the GH18 family, which are devoid of the inherent hydrolytic activity but are able to take glycan oxazolines for transglycosylation. Notably, the Endo-F3 D165A and D165Q mutants demonstrated high acceptorsubstrate specificity toward α1,6-fucosyl-GlcNAc-Asn or α1,6-fucosyl-GlcNAc-polypeptide in transglycosylation, enabling a highly convergent synthesis of core-fucosylated, complex CD52 glycopeptide antigen. The Endo-F3 mutants were able to use both bi- and triantennary glycan oxazolines as substrates for transglycosylation, in contrast to previously reported endoglycosidases derived from Endo-S, Endo-M, Endo-D, and Endo-A mutants that could not recognize triantennary N-glycans. Using rituximab as a model system, we have further demonstrated that the Endo-F3 mutants are highly efficient for glycosylation remodeling of monoclonal antibodies to produce homogeneous intact antibody glycoforms. Interestingly, the new triantennary glycan glycoform of antibody showed much higher affinity for galectin-3 than that of the commercial antibody. The Endo-F3 mutants represent the first endoglycosidase-based glycosynthases capable of transferring triantennary complex N-glycans, which would be very useful for glycoprotein synthesis and glycosylation remodeling of antibodies.

Protein glycosylation is one of the most prevalent posttranslational modifications, found in almost all living organisms ranging from bacteria to eukaryotes (1). Glycosylation can profoundly affect a protein's intrinsic properties, such as folding, stability, intracellular trafficking, and immunogenicity. In addi-tion, the oligosaccharide components of glycoproteins can participate directly in a number of important biological recognition processes, including cell adhesion, signaling, hostpathogen interactions, and immune responses (2)(3)(4)(5)(6)(7)(8). It is well documented that subtle changes in glycosylation can lead to a significant impact on the biological functions and, in the case of therapeutic glycoproteins, such as monoclonal antibodies, the in vivo stability and therapeutic efficacy (7, 9 -12). Natural and recombinant glycoproteins are usually produced as mixtures of glycoforms that differ only in the structures of pendant glycans, from which pure glycoforms are extremely difficult to isolate by current chromatographic techniques. As a result, synthetic homogeneous glycopeptides and glycoproteins emerge as indispensable tools for functional studies and for drug/vaccine discoveries. Many elegant chemical and biochemical strategies have been explored for making homogeneous glycoproteins and mimics, including total chemical synthesis with native chemical ligation (13)(14)(15)(16)(17), chemoselective ligation (18,19), chemoenzymatic synthesis (20 -24), and glycosylation pathway engineering in host expression systems (25)(26)(27)(28). As part of these efforts, we have attempted to develop a chemoenzymatic method for construction of complex N-glycopeptides and glycoproteins that is based on the transglycosylation activity of a class of endo-␤-N-acetylglucosaminidases (the endoglycosidases that hydrolyze N-glycans of glycoproteins) for convergent native ligation of preassembled glycans and GlcNAc-peptide/ protein (20,23,24). The success of this approach relies on the availability of novel endoglycosidase-derived glycosynthase mutants that are devoid of hydrolysis activity but can use the highly activated glycan oxazolines as substrates for transglycosylation. Several endoglycosidases have been successfully converted into glycosynthases via site-directed mutation of a critical residue that appears to be essential in promoting the formation of the sugar oxazolinium ion intermediate at the catalytic site during enzymatic hydrolysis. These include Endo-A 3 (from bacterium Arthrobacter protophormiae), Endo-M (from fungus Mucor hiemalis), and Endo-D (from bacterium Strepto-coccus pneumoniae) of the glycoside hydrolase (GH) family 85 as well as Endo-S (from Streptococcus pyogenes) of the GH family 18 (29 -34). These enzymes and related mutants showed remarkably different substrate specificity in transglycosylation. For example, the glycosynthase mutants derived from Endo-M (such as Endo-M N175Q) are efficient for transferring biantennary complex type and high mannose type N-glycans to GlcNAc-peptide/protein acceptors but are unable to accept ␣1,6-fucose-substituted (core-fucosylated) GlcNAc-peptide/ protein as acceptor substrates; nor can they efficiently glycosylate Fc domains or intact antibodies (29 -31, 35-38). The glycosynthases derived from Endo-S (such as the Endo-S D233A and D233Q mutants) appear to be highly specific for glycosylation of the Fc domain of antibodies (33,39,40), but they show only marginal activity for regular GlcNAc-peptide/protein acceptors. So far, no glycosynthases in this category have been reported that are able to transfer highly branched N-glycans, such as triantennary N-glycans, for glycoprotein synthesis. In addition, although Endo-D mutants (such as Endo-D D322A) have been shown to accept Fuc␣1,6GlcNAc-peptide as acceptor for transglycosylation, they are limited to transferring only the N-glycan core and are unable to accommodate common complex type N-glycans for transglycosylation (34).
Core fucosylation, the attachment of a fucose moiety to the innermost GlcNAc moiety of the N-linked glycans in an ␣1,6glycosidic linkage, is a modification frequently found in natural and recombinant glycoproteins. It has been reported that core fucosylation affects the conformations of N-glycans and modulates glycan-lectin interactions and antibody-Fc receptor recognition (41)(42)(43)(44)(45). Moreover, core fucosylation is associated with cancer progression, suggesting that core-fucosylated N-glycans and glycopeptides could serve as novel biomarkers for cancer diagnosis (46,47). Despite remarkable advances in synthetic methodology, synthesis of sialylated and fucosylated complex glycopeptides and glycoproteins is still a challenging task because of the acid-labile nature of the ␣-glycosidic linkages of fucose and sialic acid during multistep synthetic manipulations (48 -51). In contrast to sialylation that could be performed by sialyltransferases after the construction of full-size N-glycans, a late stage enzymatic introduction of a core fucose by the mammalian FUT8 (␣1,6-fucosyltransferase) would not be quite feasible because FUT8 cannot directly glycosylate fullsize N-glycans but can only accept an N-glycan core with a free GlcNAc residue at the ␣1,3-mannose branch as the substrate (47,52). Thus, new glycosynthase mutants that can transfer highly branched N-glycans and can accommodate core-fucosylated GlcNAc-peptide/protein acceptors are required to expand the scope of the chemoenzymatic method.
We have recently reported that several endoglycosidases from bacterium Elizabethkingia meningoseptica of the GH18 family, formerly named as Flavobacterium meningosepticum or Chryseobacterium meningosepticum (53)(54)(55), possess significant transglycosylation activity at neutral pH (partly due to the decreased hydrolytic activity at the less acidic pH) (56). Of these endoglycosidases, Endo-F3 has shown transglycosylation activity on core-fucosylated GlcNAc-peptides. However, the wild typeenzymealsopossesseshighhydrolyticactivityonthetransglycosylation product. In this paper, we report the generation of glycosynthase mutants from Endo-F3 by site-directed mutagenesis of the putative catalytic residue, Asp-165, which is assumed to play a role in promoting the formation of sugar oxazolinium ion intermediate in hydrolysis. Our experimental data reveal that the Endo-F3 mutants, including D165A and D165Q, are typical glycosynthases that are able to efficiently transfer both bi-and triantennary complex type N-glycans (in the form of glycan oxazolines) to fucosylated GlcNAc-peptide acceptors to form core-fucosylated complex glycopeptides with high efficiency. The Endo-F3 mutants D165A and D165Q represent the first endoglycosidase-based glycosynthases capable of transferring triantennary complex type N-glycans for glycopeptide synthesis and glycosylation remodeling of core-fucosylated glycoproteins, including therapeutic antibodies.

Experimental Procedures
Materials-Fmoc-Asn(Fuc␣1,6GlcNAc)-OH and core-fucosylated CD52 antigen were synthesized following our previously reported procedures (56). Monoclonal antibody rituximab (rituxan, Genetech Inc., South San Francisco, CA) was purchased through Premium Health Services Inc. (Columbia, MD). Asialo and sialoglycan complex type oxazoline was synthesized following a modified reported procedure (57). Endo-S WT and Endo-S D233A from S. pneumoniae were overexpressed and purified following our previous procedure (33). Fetuin was purified from the fetal bovine serum (Sigma) using a modified procedure of Spiro (58). Porcine fibrinogen was purified from porcine plasma (Sigma) using a modified procedure (59).
Cloning, Expression, and Purification of Endo-F3-The cDNA fragment encoding the Endo-F3 gene without the signal peptide (nucleotides 118 -987; amino acids 40 -329) was amplified by PCR from the genomic DNA of E. meningoseptica (ATCC number 51720D). The forward primer was 5Ј-GGAATTCCATATGGCTACTGCTTTGGCGG, and the reverse primer was 5Ј-CGCGGATCCCAGGTTCTTCACTG-CATCTC. An NdeI site was added to the forward primer and a BamHI site was added to the reverse primer (underlined). Both amplified DNA fragments were cloned into pCPD-Lasso (a pET22b-CPD derivative) after digestion with NdeI and BamHI. The constructed plasmid, pCPD-Lasso Endo-F3, was transformed into BL21 (DE3). The transformants were cultured in 1 liter of terrific broth medium supplemented with 50 g/ml carbenicillin. Cultures were grown at 37°C until the cells reached an A 600 nm of 1-1.5. Then 0.1 mM isopropyl ␤-D-1-thiogalactopyranoside and 100 ml of glycerol were added to the culture to induce protein overproduction at 16°C. After 30 h, the cells were harvested by centrifugation. The cell pellets were lysed using B-PER protein extraction reagent (Pierce) following the manufacturer's instructions. The cell lysis was applied to His-Trap HP (GE Healthcare) and washed with phosphate-buffered saline (PBS) with 30 mM imidazole (pH 7.4). Bound His-tagged protein was eluted with a gradient from 0 to 500 mM imidazole in PBS buffer. The purified Endo-F3 was desalted and concentrated with PBS using centrifugal diafiltration with an Amicon Ultra filtration unit (10 kDa; Millipore). The protein purity was confirmed using SDS-PAGE, and the concentration was quan-tified using measurement of the absorption at 280 nm on a NanoDrop 2000c.
Liquid Chromatography Mass Spectrometry (LC-ESI-MS) of Free Glycan and Glycopeptides-The LC-MS was performed on a LXQ Linear Ion Trap with an Agilent Poroshell 300SB-C8 column (5 m, 75 ϫ 1 mm). The analysis was performed at 30°C, eluting with a linear gradient of 0 -10% acetonitrile containing 0.1% formic acid within 10 min at a flow rate of 0.40 ml/min.
LC-ESI-MS of IgG-The LC-MS analysis was performed on a LXQ Linear Ion Trap or Exactive Plus Orbitrap (Thermo Scientific) with an Agilent Poroshell 300SB-C8 column (5 m, 75 ϫ 1 mm). The IgG antibody samples were treated with 50 mM tris(2-carboxyethyl)phosphine and heated at 37°C for 20 min and then subjected to LC-MS analysis. The analysis was performed at 60°C, eluting with a linear gradient of 25-35% acetonitrile containing 0.1% formic acid within 10 min at a flow rate of 0.40 ml/min. The LC-MS analysis of PNGase F-treated antibody glycoforms was performed in the same manner but included a 1-h incubation with PNGase F prior to tris(2-carboxyethyl)phosphine treatment. Raw data were deconvoluted using MagTran (Amgen). The molecular mass of the commercial rituximab was calculated using the amino acid sequence described in drugbank.ca with the following known modifications: N-terminal pyroglutamic acid formation on the light and heavy chain, C-terminal lysine cleavage of the heavy chain, and full reduction of intrachain disulfide bonds (two in the light chain and four in the heavy chain).
NMR-The purified N-glycans were deuterium-exchanged using 99% D 2 O (three times) and dried by lyophilization. The dried glycans were dissolved in 500 l of deuterium oxide (100%) and transferred to an NMR tube. Two-dimensional HSQC 1 H/ 13 C NMR spectra were recorded in a Bruker 950-MHz NMR system with a cryo-probe. All chemical shifts were assigned in ppm.
Site-directed Mutagenesis of Endo-F3-Endo-F3 mutants, D165A and D165Q, were generated using the QuikChange sitedirected mutagenesis kit (Stratagene). Briefly, primers containing the desired mutation were used to perform PCR using pCPD-Lasso Endo-F3 as a template. After PCR, the mixture was digested with DpnI, and then the mixture was transformed into DH5␣ maximum efficiency cells (Invitrogen). Mutations were confirmed by DNA sequencing. Mutants were expressed and purified by the same method as the wild type proteins with similar yields.
Semisynthesis of Triantennary Oxazoline-A semisynthesis of the triantennary oxazoline was achieved using bovine fetuin N-glycans as the starting material. The N-glycans of fetuin, which include both biantennary and triantennary complex type glycans, were cleaved using Endo-F3 and purified using the acetone/methanol-selective precipitation method (60). The purified glycans, which contain small amounts of impurities, were first desalted on a Sephadex G25 (Sigma) column and further purified using ion exchange chromatography on a HiTrap DEAE-Sepharose FF (GE Healthcare) column using an NaCl gradient of 0 -100 mM over 60 min at a flow rate of 1 ml/min. Each peak was separately desalted on a Sephadex G25 column (Sigma), and the purity of each fraction was determined by HPAEC-PAD. The fraction containing trisialylated triantennary glycans 5 was used as the starting material for the semisynthesis. The STCT glycan 5 was treated with neuraminidase (New England Biolabs) in 50 mM sodium citrate buffer, pH 6.0, at 37°C for 16 h. The reaction progress was monitored by HPAEC-PAD and ESI-MS, which confirmed complete desialylation. After digestion, the solution was subjected to gel filtration on a Sephadex G-15 column eluted by water. The purified asialo triantennary glycan 6 (25 mg) and 2-chloro-1,3-dimethyl-2-imidazolinium chloride (46.8 mg) were dissolved in 100 l of water at 4°C. Then 80 l of Et 3 N was added, and the mixture was stirred at 4°C for 30 min. The reaction was monitored by HPAEC-PAD, which confirmed the completion of the reaction. The reaction solution was subjected to gel filtration on Sephadex G-15 column eluted by water containing 0.1% triethylamine. The fractions containing the asialo triantennary complex type glycan oxazoline (TCTox) were combined and lyophilized to give the TCTox as a white powder (23.9 mg). The TCTox 7 was resuspended in a 20 mM solution of NaOH, ensuring an alkaline solution for the stability of the oxazoline. 1  Transglycosylation of Core-fucosylated CD52 with TCTox by Endo-F3 D165A; Synthesis of 9-A solution of TCTox (6) (450 g, 6.24 mM), CD52-GlcNAc-Fuc (100 g, 1.56 mM) (8), and Endo-F3 D165A (14 g) was incubated at 37°C in 100 mM Tris, pH 7.4, in a 40-l total volume. The reaction was monitored using reverse-phase HPLC analysis (method B). Completion conversion of 8 to 9 was observed within 2 h. The product (9) was confirmed with LC-MS. TCT-CD52 calculated M ϭ 3384.3 Transglycosylation of (Fuc1,6)GlcNAc-Rituximab with CTox, SCTox, and TCTox by Endo-F3 D165A; Synthesis of 12, 14, and 15-A solution of (Fuc␣1,6)GlcNAc-rituximab (1 mg, 69 M) (11) and SCTox (700 g, 3.5 mM) (1) was incubated with Endo-F3 D165A (35 g) at 37°C in 100 l of 100 mM Tris buffer (pH 7.4) for 45 min. Then another 550 g of SCTox (1) was added, and the reaction was monitored by LC-MS of aliquots. After incubation for a total of 1.5 h, LC-MS analysis indicated that a completion transglycosylation of the deglycosylated rituximab was achieved. The product (12) was purified using protein A chromatography. LC-MS: calculated for the heavy chain of 12 carrying the fully sialylated biantennary N-glycan, M ϭ 51,421 Da; found (m/z), 51,412 (deconvolution data).
Surface Plasmon Resonance-The binding between glycoengineered IgG glycoforms and galectin-3 was measured by surface plasmon resonance (SPR) using a Biacore T100 instrument (GE Healthcare). Galectin-3 was immobilized on a CM5 biosensor chip (GE Healthcare) using amine coupling chemistry, pH 5.0, to reach 1000 response units. Each glycoform of IgG was injected at 40 l/min. After each cycle, the surface was regenerated by injecting 3 M MgCl 2 .
The binding between glycoengineered IgG glycoforms and Fc␥RIIIa was measured by SPR using a Biacore T100 instrument (GE Healthcare). Protein A was immobilized at 6000 resonance units on a CM5 biosensor chip (GE Healthcare) using amine coupling chemistry at pH 4.5. Each glycoform of IgG in PBS was injected at 10 l/min over the protein A surface to a capture level of 150 resonance units. A serial dilution of Fc receptor was injected at 30 l/min. After each cycle, the surface was regenerated by injecting a glycine HCl buffer (10 mM, pH 2.5).

Cloning, Expression, and Characterization of Endo-F3-
Endo-F3 is a bacterial endoglycosidase from the GH18 family consisting of 329 amino acid residues (35.8 kDa) with an N-terminal signaling peptide (amino acids 1-39). Tarentino et al. (61) have previously reported that expression of Endo-F3 in Escherichia coli resulted in low yields (Ͻ1 mg/liter) of soluble enzyme due to the formation of insoluble inclusion bodies. Overexpression and refolding enabled higher yields (15 mg/liter) but required arduous solubilization with acid and detergent followed by time-consuming exhaustive dialysis (62). We initially chose to clone Endo-F3 (amino acids 40 -329) into both pET41b and pET22b for expression in E. coli. Overexpression from both vectors yielded inclusion bodies of Endo-F3, but purification and refolding of the inclusion bodies following a procedure similar to one reported previously (62) gave a very low overall yield (Ͻ1 mg/liter) of soluble Endo-F3. This situation prompted us to investigate alternative methods to overexpress Endo-F3 WT in a soluble form. Previously, Shen et al. (63) have demonstrated that expression with a novel tag, the Vibrio cholera MARTX toxin cysteine protease domain (CPD) was able to enhance the solubility and stability of recombinant proteins (63,64). To examine whether CPD could enhance the expression of the soluble Endo-F3, we cloned the gene encoding Endo-F3 (amino acids 40 -329) into a pET22b-CPD vector. This plasmid encodes the CPD domain and also includes a His 10 tag. We found that the fusion protein, Endo-F3-CPD (hereafter called Endo-F3), could be efficiently expressed as a soluble protein in E. coli and was readily purified using immobilized metal ion affinity chromatography to give the soluble enzyme with a yield of Ͼ15 mg/liter.
We tested the deglycosylation activity of the recombinant Endo-F3 using several substrates, including porcine fibrinogen (a natural glycoprotein carrying core-fucosylated biantennary N-glycans) (65), rituximab (a therapeutic monoclonal antibody), and the sialoglycopeptide (SGP) isolated from chicken egg yolks (Fig. 1). As revealed by SDS-PAGE and LC-MS analysis, the recombinant Endo-F3 showed potent hydrolytic activ-ity and was able to deglycosylate porcine fibrinogen and monoclonal antibody rituximab quickly even without the need of denaturing the glycoproteins (Fig. 1, A and B). The recombinant Endo-F3 could also hydrolyze the sialoglycopeptide, SGP ( Fig. 1C), but it was observed that the Endo-F3-catalyzed hydrolysis of SGP was much slower than that of porcine fibrinogen and rituximab, which carry core-fucosylated N-glycans. Our experimental data are consistent with the fact that commercially available Endo-F3 prefers core-fucosylated complex N-glycans and hydrolyzes them hundreds of times faster than it does the corresponding non-fucosylated N-glycans (65). Examination of the pH dependence of the Endo-F3-catalyzed hydrolysis using SGP as the substrate showed that the optimal pH for Endo-F3-catalyzed hydrolysis was around 4.5 at 37°C, which is in good agreement with previous reports of the naturally isolated Endo-F3 (65).
Generation and Characterization of Glycosynthase Mutants of Endo-F3-We have previously generated glycosynthases from endoglycosidases of both the GH85 and GH18 family by site-directed mutation at a key residue that is responsible for promoting the formation of the oxazolinium ion intermediate during hydrolysis, which proceeds in a substrate-assisted mechanism (29,30,(32)(33)(34). These include a key asparagine residue for the GH85 endoglycosidases Endo-A (Asn-171), Endo-M (Asn-275), and Endo-D (Asn-322) or a key aspartic acid residue for the GH18 family endoglycosidase Endo-S (Asp-233). A sequence alignment of Endo-F3 and Endo-S revealed that the Asp-165 of Endo-F3 was the residue equivalent to the Asp-233 of Endo-S essential for promoting oxazolinium ion formation in hydrolysis (Fig. 2). The crystal structure of Endo-F3 also suggested that the Asp-165 was critical for the hydrolysis (66). Thus, we generated two mutants, D165A and D165Q, using site-directed mutagenesis to create potential glycosynthase mutants of Endo-F3. The two mutants were also expressed as soluble proteins in the pET22b-CPD vector and produced similar yields as the wild type Endo-F3 enzyme. We then tested the potential hydrolytic activity of the Endo-F3 mutants using rituximab as the substrate under the conditions used to confirm the hydrolytic activity of Endo-F3 WT. Under the same conditions, the Endo-F3 mutants D165A and D165Q had no detectable hydrolytic activity on rituximab after 4 h when the reaction was monitored by LC-MS analysis. These data verify that the mutation eliminates the hydrolytic activity.
We have previously discovered a transglycosylation activity of wild type Endo-F3 using SCTox (1) as the donor substrate and Fmoc-Asn(Fuc␣1,6GlcNAc)-OH (core-fucosylated) (2) or Fmoc-Asn(GlcNAc)-OH (3) as the acceptor substrates (56). In this study, the same reactions were applied to test the transglycosylation activity of the Endo-F3 mutants (Fig. 3A). We found that Endo-F3 D165A and D165Q both showed apparent transglycosylation activity with the core-fucosylated GlcNAc-Asn acceptor (2) to give the transglycosylation product (4) (Fig. 3B), which was identical to the authentic compound characterized previously by ESI-MS (56). Interestingly, no transglycosylation was detected when the non-fucosylated GlcNAc-Asn (3) was used as the substrate (Fig. 3). The results clearly indicate that the Endo-F3 glycosynthase mutants are highly specific and active for ␣1,6-fucosylated GlcNAc acceptors for transglycosylation. In contrast to the wild type Endo-F3, which had transglycosylation activity but gradually hydrolyzed the product, the D165A and D165Q mutants were able to use the sugar oxazoline for transglycosylation without product hydrolysis, thus confirming them as typical glycosynthases. We also observed that the Endo-F3 D165A mutant was superior to the D165Q mutant in glycosynthase activity (Fig. 3B). Under the conditions, the Endo-F3 D165A-catalyzed reaction could reach 100% transglycosylation quickly without product hydrolysis, as monitored by reverse-phase HPLC and confirmed with ESI-MS analysis.
Semisynthesis of Asialo Triantennary Complex Type Glycan Oxazoline-Previous studies have shown that Endo-F3 can efficiently cleave fucosylated triantennary complex type N-glycans (65). The potent hydrolytic activity on branched complex N-glycans led us to investigate whether the glycosynthase mutants derived from Endo-F3 are able to transfer triantennary N-glycans. For this purpose, we carried out a semisynthesis of TCTox starting from the isolation of the N-glycans from bovine fetuin, a natural glycoprotein that carries sialylated bi-and triantennary N-glycans. The synthetic scheme was summarized in Fig. 4. Previously, the components and structures of N-glycans from bovine fetuin have been well characterized by chromatographic separation, HPAEC-PAD analysis, and detailed NMR studies (67)(68)(69)(70). The majority (Ͼ80%) of the N-glycans were triantennary N-glycans with varied degrees of ␣2,6and/or ␣2,3-sialylations at the terminus. For a relatively large scale preparation, we first purified bovine fetuin on the gram scale from fetal bovine serum using a previously reported method with some modifications (58). The N-glycans were released from bovine fetuin using Endo-F3 and were partially purified by acetone precipitation and extraction with 60% methanol (60,71). The crude N-glycans showed multiple peaks on HPAEC-PAD with four major peaks as the triantennary N-glycans carrying 2-3 sialic acids in isomeric ␣2,3and ␣2,6-linkages (Fig.  5A). After desalting on a Sephadex G-25 column and then further purification by ion exchange chromatography using a HiTrap DEAE FF column, fractions that contained the trisialylated triantennary glycans were desalted using a Sephadex G-25 column giving two major trisialylated triantennary complex type N-glycan isomers differing only in the linkage of the sialic acid on the 6Ј-mannose arm (5) (Fig. 5B). The assignment of the peaks was based on a detailed HPAEC-PAD analysis of fetuin N-glycans reported previously (67,69). The isomeric nature of the sialylated N-glycans was confirmed with ESI-MS analysis showing m/z species at 893.8 and 1340.01, respectively, corresponding to the [M ϩ 3H] 3ϩ and [M ϩ 2H] 2ϩ species (calculated M ϭ 2677.39). Desialylation of 5 with sialidase gave TCT (6) as a single species with Ͼ95% purity (Fig. 5C). The only minor (Ͻ5%) contamination could be attributed to a triantennary N-glycan isomer that contains an isomeric ␣1,3-galacto-

JOURNAL OF BIOLOGICAL CHEMISTRY
(7) in a single step by treatment with an excess of 2-chloro-1,3dimethylimidazolinium chloride and triethylamine in water, following the procedures reported previously (37,57,72). The reaction was monitored by HPAEC-PAD, and the product formation was found to be essentially quantitative (Fig. 5D). HPAEC-PAD analysis clearly showed a shift of 3 min in retention time as the oxazoline derivative eluted before the free sugar (Fig. 5D). The product was purified from the reaction mixture by gel filtration on a Sephadex G15 column. The identities of the triantennary N-glycan (6) and the corresponding glycan oxazoline (7) were further characterized by two-dimensional 1 H/ 13 C HSQC NMR analysis (Fig. 6). The NMR signal assignment (Fig. 6A) of glycan 6 was in good agreement with a similar triantennary derivative with two GlcNAc moieties at the reducing terminus (70), and the results further confirmed the homogeneity of the asialylated triantennary N-glycan (6). The twodimensional 1 H/ 13 C HSQC spectrum (Fig. 6B) of glycan oxazoline 7 clearly showed the expected downfield shift of the reducing end anomeric proton to 6.1 ppm with a J 1,2 coupling constant of 7.3 Hz, which is a signature for the anomeric proton of sugar oxazoline.
Endo-F3 Mutant D165A Is Able to Transfer Triantennary N-Glycan to Form Triantennary Complex Type Glycopeptides-With the synthetic triantennary complex glycan oxazoline in hand, we selected the CD52 glycopeptide, a small glycosylphosphatidylinositol-anchored glycoprotein found on the surface of mature lymphocytes and sperm cells, as a target to test the Endo-F3 mutants and glycosynthase mutants from other endoglycosidases. We performed the transglycosylation with Endo-F3 D165A using the TCTox (7) as the donor and the Fuc␣1,6GlcNAc-CD52 (8) as the acceptor substrate, the synthesis of which was described previously (56). The reaction progress was monitored by HPLC analysis, and the product was analyzed with LC-MS. We found that the Endo-F3 D165A mutant could efficiently use the triantennary glycan oxazoline (7) as substrate for transglycosylation to form the corresponding triantennary complex type CD52 glycopeptide (9) (Fig. 7). HPLC monitoring showed an m/z species of 1129.80 corresponding to [M ϩ 3H] 3ϩ (calculated molecular mass, M ϭ 3385.09). We found that a quantitative conversion of the Fuc␣1,6GlcNAc-CD52 (8) to the glycopeptide (9) could be achieved within 2 h when a 3-fold excess of the glycan oxazoline (7) was used, demonstrating the efficiency of the Endo-F3 D165A-catalyzed transglycosylation. We have also tested several glycosynthases described previously, including the D233A mutant of Endo-S, the N322A and N322Q mutants of Endo-D, the N175Q mutant of Endo-M, and the N171A mutant of Endo-A (29,30,(32)(33)(34). We found that none of those glycosynthases were able to transfer triantennary complex glycans from the glycan oxazoline. Thus, the Endo-F3 mutants generated here represent the first endoglycosidase-based glycosynthases capable of transforming branched triantennary complex type N-glycans for complex glycopeptide synthesis. The Endo-F3 Glycosynthase Mutant Enables the Glycosylation Remodeling of Intact Monoclonal Antibody-The ability of the Endo-F3-derived glycosynthases to transfer the triantennary glycan for the synthesis of complex CD52 glycopeptide encouraged us to test whether the Endo-F3 mutant would also be efficient at transferring triantennary N-glycans to intact antibodies. We have previously used mutant endoglycosidases to remodel the N-glycans on the Fc region of the monoclonal antibody rituximab using a pair of endoglycosidase/endoglycosynthase derived from Endo-S (33). The major Fc glycans of commercial rituximab are core-fucosylated biantennary com-plex type oligosaccharides carrying 0 -2 galactose moieties named the G0F, G1F, and G2F glycoform, respectively. The general glycosylation remodeling approach is presented in Fig.  8, and the reactions and reaction products were assessed by LC-MS analysis (Fig. 9).

Novel Endo-F3 Glycosynthases for Glycoprotein Synthesis
endoglycosidase and excess glycans by affinity chromatography with a protein A column to give the Fuc␣1,6GlcNAc-rituximab intermediate (11). We found that Endo-F3 D165A was able to efficiently transfer sialylated and asialylated biantennary N-glycans from the corresponding glycan oxazolines (1 and 13) to the Fuc␣1,6GlcNAc-rituximab acceptor (11) to form the respective new glycoforms of rituximab (12 and 14). Likewise, the Endo-F3 D165A mutant was also efficient to glycosylate the Fuc␣1,6GlcNAc-rituximab (11) using the triantennary glycan oxazoline (7) as the donor substrate to give the triantennary complex glycoform of rituximab (15). The reaction could be readily pushed to completion when a 40 -50 molar equivalent (i.e. 20 -25 molar equivalent per monomeric Fc domain) of the glycan oxazoline was added in two portions. The reaction yield was estimated by LC-MS analysis to be over 95% because almost no starting material was detected, which confirmed the completion of the transglycosylation. The rituximab remodeling progress was first monitored by SDS-PAGE analysis. In all cases, the products showed a single band for the heavy chain that was ϳ2 kDa larger than the heavy chain of the Fuc␣1,6GlcNAc-rituximab, whereas the light chain was a single band with the same size before and after the reactions. These data suggest that a single N-glycan was attached to each of the heavy chain of the intact antibody without modification on the light chain. LC-MS analysis of the transglycosylation products (12 and 14) carrying biantennary N-glycans revealed that the heavy chain of 12 and 14 appeared as a single species at 51,412 and 50,835 (deconvolution data), respectively, which are in good agreement with the calculated molecular mass (M ϭ 51,421 and 50,839 Da) for the heavy chains carrying a sialylated or asialylated biantennary N-glycan (with core fucose), respectively (Fig. 9, C and D). In the case of the triantennary glycoform (15), the deconvolution data of the ESI-MS revealed a single peak at 51,204 for the heavy chain of 15, which matches well the calculated molecular mass of the heavy chain carrying a single triantennary N-glycan (M ϭ 51,199 Da) (Fig. 9E).
To further confirm that the N-glycan was attached specifically to the Asn-297 N-glycosylation site of the Fc domain, instead of other sites of the polypeptide backbone that might occur by non-enzymatic reactions, we treated the transglycosylation products with PNGase F and examined the protein portion and N-glycan portion by mass spectrometry analysis. PNGase F was highly specific and could release the N-glycans only when they were attached to the Asn side chain in an N-glycosylamide linkage at the conserved glycosylation site in the glycoproteins. LC-MS analysis of the heavy chain of the PNGase F-treated transglycosylation product (12) gave a species of 49,066 (deconvolution data), which matches well with  APRIL 22, 2016 • VOLUME 291 • NUMBER 17 the calculated molecular mass of the polypeptide backbone of the heavy chain without any additional modifications (M ϭ 49,069 Da) (Fig. 9F). On the other hand, the light chain of the transglycosylation product (12) appeared as a single species at 23,038, which matches the calculated molecular mass of the light chain of rituximab (M ϭ 23,039 Da) (Fig. 9G). Taken together, these results clearly indicated that a single sialylated biantennary N-glycan was conjugated to the antibody heavy chain, and the intact N-glycan was attached at the conserved N-glycosylation site without any non-enzymatic conjugation of N-glycans. We also tested the glycosynthase mutant, Endo-S D233A, which was previously shown to be capable of transferring biantennary N-glycan to intact antibody acceptor. We found that Endo-S D233A was unable to transfer the triantennary glycan oxazoline (7) to rituximab because no transglycosylation product was detected even after a prolonged incubation (over 5 h).

Novel Endo-F3 Glycosynthases for Glycoprotein Synthesis
Binding of the Glycoengineered Rituximab to the C-type Lectin Galectin-3-The affinity of the remodeled glycoforms of rituximab for galectin-3 was examined by SPR analysis. Galectin-3 was immobilized on the CM5 chip, and the various rituximab glycoforms were injected as analytes, following our previous procedures (73). The commercially available rituximab showed no detectable binding under the concentrations tested (Fig. 10A). However, both the biantennary and triantennary glycoforms, G2F and G3F, showed strong binding to galectin-3 (Fig. 10, B and C). The K D values for the bi-and triantennary glycoforms were estimated to be 130 and 64 nM, respectively, indicating that the terminal galactosylated triantennary glycoform of antibody has higher affinity for galectin-3 than that of the terminal galactosylated biantennary antibody glycoform.

Binding of the Glycoengineered Rituximab to the Fc␥IIIa
Receptor (Fc␥RIIIa)-The affinity of the remodeled glycoforms of rituximab for Fc␥RIIIa was examined by SPR analysis. Protein A was nonspecifically bound to the CM5 chip and used to capture the various rituximab glycoforms. After capture of antibody glycoforms, the Fc␥RIIIa was injected over as the analyte to obtain the SPR responses (Fig. 11). Fitting the data with the 1:1 binding model did not give reliable kinetic parameters. Then the data were analyzed by using the steady state fitting (Fig. 11). The K D values for the binding of Fc␥RIIIa to the commercial rituximab and the G2F and the G3F glycoforms were estimated to be 245 Ϯ 1, 162 Ϯ 5, and 168 Ϯ 8 nM, respectively. The G2F and G3F glycoforms showed similar affinity for Fc␥RIIIa, but their affinity for the Fc␥RIIIa is slightly higher than that of the commercially available rituximab.

Discussion
In this study, we have reported a successful and high yield expression of soluble recombinant Endo-F3 as a CPD fusion protein, along with the generation and expression of glycosynthase mutants that possess potent transglycosylation activity with glycan oxazolines but are devoid of product hydrolysis activity. We have also established an efficient semisynthesis of triantennary complex type N-glycan and the corresponding glycan oxazoline, which will be useful substrates for assessment of new glycosynthase mutants. It should be mentioned that previous attempts to express soluble Endo-F3 have met with little success, because of inclusion body formation (61,62). The new Endo-F3 CPD construct makes it possible to express soluble and active Endo-F3 and its mutants on a large scale with high expression yield.  (10); B, heavy chain of the Endo-F3 deglycosylated rituximab (11); C, heavy chain of the transglycosylation product (12); D, heavy chain of the transglycosylation product (14); E, heavy chain of the transglycosylation product (15); F, heavy chain of the PNGase F-treated transglycosylation product (12); G, light chain of the transglycosylation product (12).
We have found that the two glycosynthase mutants generated, D165A and D165Q, are highly specific for ␣1, 6-fucosylated GlcNAc acceptor. The significant difference in transglycosylation activity with the core-fucosylated and nonfucosylated GlcNAc acceptors appears consistent with the known substrate specificity of Endo-F3 for N-glycan hydrolysis. It was reported previously that wild type Endo-F3 hydrolyzed core-fucosylated glycopeptides more than 300-fold faster than the corresponding non-fucosylated glycopeptide substrates (65). An important discovery in this study is that the Endo-F3 glycosynthase mutants are able to transfer triantennary complex type N-glycan to fucosylated GlcNAc-peptides and deglycosylated monoclonal antibodies, which could not be achieved by previously reported glycosynthases, such as those derived from Endo-A, Endo-M, Endo-D, and Endo-S. This highly convergent chemoenzymatic assembly of complex glycopeptides provides an alternative to pure chemical synthesis of biologically important glycopeptides, such as tumor-associated glycopeptide antigens useful for development of vaccines and diagnostic tools (74). The usefulness of the Endo-F3 mutants for complex glycopeptide synthesis was exemplified by an efficient synthesis of the core-fucosylated triantennary complex type CD52 glycopeptide antigen (74).
On the other hand, the ability of the Endo-F3 mutants to transfer both bi-and triantennary complex type glycans to the deglycosylated antibodies, such as rituximab, significantly expands the scope of the chemoenzymatic approach for glycosylation remodeling of antibodies (33). It should be pointed out that the Fc triantennary glycoform obtained by the glycosylation remodeling represents a novel antibody glycoform that is rare for natural and recombinant antibodies, probably due to restricted access of glycosyltransferases for further processing (75,76). It was reported previously that some amino acid mutations within the Fc domain could alter the conformation of the antibody, enabling the production of triantennary glycoforms as minor fractions (77).
We also measured the affinity of the terminal fully galactosylated triantennary antibody glycoform with galactin-3, a C-type lectin. Galectin-3 recognizes ␤-galactoside-terminating glycoconjugates in biological systems and plays important roles in a number of biological processes, including cell adhesion, autoimmunity, and cancer metastasis (78). The fact that the new triantennary complex type glycoform demonstrates much higher affinity for galectin-3 than that of the commercial rituximab suggests that the glycosylation-remodeled antibody may confer novel properties to the antibodies in vivo. Moreover, a recent report has demonstrated that IgGs with terminal galactose residues in the Fc glycosylation interact with Dectin-1 and Fc receptor to elicit an anti-inflammatory response (79). It would be interesting to test whether the newly obtained, fully galactosylated Fc glycoforms of antibodies will show enhanced anti-inflammatory activities in animal models.
In summary, the Endo-F3 mutants (D165A and D165Q) described in this study represent the first endoglycosidasebased glycosynthases capable of accepting triantennary complex type N-glycans for transglycosylation. The ability of the Endo-F3 mutants to transfer triantennary complex type N-glycans to ␣1,6-fucosylated GlcNAc-polypeptides and deglycosylated intact antibodies significantly expands the scope of the chemoenzymatic method, which could not be achieved by the previously reported glycosynthases, such as the Endo-S and Endo-M mutants. It is expected that the new Endo-F3 glycosynthases will find useful applications for convergent chemoenzymatic synthesis of core-fucosylated complex type glycopeptides and glycoproteins as well as for glycosylation remodeling of intact antibodies to obtain novel glycoforms for functional studies.