Mammalian α-1,6-Fucosyltransferase (FUT8) Is the Sole Enzyme Responsible for the N-Acetylglucosaminyltransferase I-independent Core Fucosylation of High-mannose N-Glycans*

Understanding the biosynthetic pathway of protein glycosylation in various expression cell lines is important for controlling and modulating the glycosylation profiles of recombinant glycoproteins. We found that expression of erythropoietin (EPO) in a HEK293S N-acetylglucosaminyltransferase I (GnT I)−/− cell line resulted in production of the Man5GlcNAc2 glycoforms, in which more than 50% were core-fucosylated, implicating a clear GnT I-independent core fucosylation pathway. Expression of GM-CSF and the ectodomain of FcγIIIA receptor led to ∼30% and 3% core fucosylation, suggesting that the level of core fucosylation also depends on the nature of the recombinant proteins. To elucidate the GnT I-independent core fucosylation pathway, we generated a stable HEK293S GnT I−/− cell line with either knockdown or overexpression of FUT8 by a highly efficient lentivirus-mediated gene transfer approach. We found that the EPO produced from the FUT8 knockdown cell line was the pure Man5GlcNAc2 glycoform, whereas that produced from the FUT8-overexpressing cell line was found to be fully core-fucosylated oligomannose glycan (Man5GlcNAc2Fuc). These results provide direct evidence that FUT8, the mammalian α1,6-fucosyltransferase, is the sole enzyme responsible for the GnT I-independent core fucosylation pathway. The production of the homogeneous core-fucosylated Man5GlcNAc2 glycoform of EPO in the FUT8-overexpressed HEK293S GnT I−/− cell line represents the first example of production of fully core-fucosylated high-mannose glycoforms.

Glycosylation, the covalent attachment of glycans to proteins, can profoundly affect the intrinsic properties and biological functions of proteins (1,2). For the recombinant production of therapeutic glycoproteins such as monoclonal antibodies, it is particularly important to control glycosylation to achieve the optimal therapeutic efficacy (3)(4)(5). The biosynthesis of asparagine-linked (i.e. N-linked) glycoproteins involves multiple steps at different sites, which includes the initial attachment of a large precursor N-glycan, Glc 3 Man 9 GlcNAc 2 , to the nascent protein by the oligosaccharyltransferase at the endoplasmic reticulum, followed by the trimming of the precursor to Man 9 GlcNAc 2 and Man 8 GlcNAc 2 at the ER and then further trimming and processing to complex type N-glycans in the Golgi apparatus. A crucial step in this process is the transfer of a GlcNAc, in a ␤1,2 linkage, to the three-arm mannose residue of the Man 5 GlcNAc 2 intermediate by N-acetylglucosaminyltransferase I (GnT I). 2 This step is considered a prerequisite for two key further steps of N-glycosylation: the core fucosylation involving the transfer of an ␣1,6-linked fucose to the innermost GlcNAc by an ␣1,6 fucosyltransferase (FUT8) as well as the further trimming of the Man 5 GlcNAc 2 intermediate to form complex-type glycans (6 -8) (Fig. 1). Nevertheless, trace amount of core-fucosylated high-mannose glycans were detected from lysosomal proteins, like human ␤-glucuronidase (9), porcine cathepsin D (10), rat liver alkaline phosphatase (11), and bovine ␣-mannosidase (12). It was once proposed that those minor core-fucosylated, high-mannose glycans found in lysosomal proteins were derived from hydrolysis of core-fucosylated, hybrid-type N-glycans by N-acetylglucosaminidase ( Fig. 1). However, the discovery of core-fucosylated Man 5 GlcNAc 2 glycan from natural proteins of a GnT I knockout CHO cell lines by Lin et al. (13) provided the first evidence suggesting the presence of a GnT I-independent pathway for core fucosylation of Man 5 GlcNAc 2 and paucimannose glycans in mammalian systems. Later on, Crispin et al. (14) reported the detection of minor fractions (ϳ5%) of core-fucosylated, highmannose glycoforms in recombinant glycoproteins produced in GnT I knockout CHO and human HEK293S cell lines, confirming the presence of a GnT I-independent fucosylation pathway. Along this line, what remains to be characterized is the enzymes involved in the core-fucosylation of the Man 5 GlcNAc 2 and paucimannose glycans in the GnT I-independent fucosylation pathway. The failure of in vitro fucosylation of the Man 5 GlcNAc 2 or Man 3 GlcNAc 2 substrate, either crude extract (6,15,16) or purified mammalian fucosyltransferase (FUT8) (7,17), implicated the presence of enzymes other than FUT8 for core fucosylation of high-mannose N-glycans.
In this study, we demonstrate that core fucosylation of highmannose-type glycans is a common phenomenon when glyco-proteins were expressed in the HEK293S GnT I Ϫ/Ϫ cell line, but the degree of core fucosylation could vary significantly depending on the context of glycoproteins. We further show that expression of erythropoietin (EPO) from the FUT8-overexpressed and FUT8 knockdown cell lines yield fully core-fucosylated (Man 5 GlcNAc 2 Fuc) and completely non-fucosylated (Man 5 GlcNAc 2 ) glycoforms, respectively. Our data suggest that FUT8, the mammalian ␣1,6-fucosyltransferase, is the sole enzyme responsible for the GnT I-independent core fucosylation of high-mannose N-glycans in N-glycoprotein biosynthesis.

Materials and Methods
Cell Culture-Human HEK293S GnT I Ϫ/Ϫ (18) (a gift from Dr. Kelly Moremum, University of Georgia) and its derivative cell lines were used for the expression of EPO and other glycoproteins. They were cultured in suspension in serum-free medium of a 50:50 mixture of EX-Cell (Sigma-Aldrich) and FreeStyle TM 293 medium (Life Technologies), incubated at 37°C and 8% CO 2, and shaken at 140 rpm/min. The cells were passaged every 3-4 days with an initial seeding density of 5 ϫ 10 5 cells/ml.
Creation of EPO-S126V-and GM-CSF-expressing Constructs-The EPO cDNA sequence with a His 6 tag was synthesized and then cloned into the pcDNA3 mammalian expressing vector (Life Technologies), resulting in the pcDNA-EPO plasmid. To eliminate the O-glycosylation site, serine 126 was mutated to valine (S126V) by site-directed mutagenesis with a Quik-Change II mutagenesis kit (Agilent Technologies) according to the instructions of the manufacturer. The GM-CSF-expressing construct was created using similar procedures but without mutagenesis. The expressing construct for the ectodomain of Fc␥IIIa receptor was a gift from Dr. Eric Sundberg (University of Maryland School of Medicine). All glycoproteins contained a His tag to facilitate purification.
Transfection and Expression of EPO-S126V and Other Expressing Constructs in HEK293S GnT I Ϫ/Ϫ Cells-The endotoxin-free expressing plasmids for transfection were prepared with the NucleoBond Xtra Maxi EF kit (Clontech). Transient transfection was performed following a protocol modified from Hacker et al. (19) using PEI (1 mg/ml, 25-kDa linear PEI, Polysciences). One day before the transfection, the cells were seeded at a density of 5 ϫ 10 5 /ml. The next day, the cells were spun down and resuspended in FreeStyle TM 293 medium at a density of 2 ϫ 10 7 /ml. The expressing plasmid was added into the cell suspension to a final concentration of 25 g/ml and gently swirled, followed by the addition of PEI solution (62.5 g/ml final concentration). After gentle swirling to mix, the transfection was kept at 37°C for 3 h. Finally, the transfected cells were diluted with EX-Cell/FreeStyle TM 293 medium to 1 ϫ 10 6 cells/ ml, and the histone deacetylase inhibitor valproic acid was added to a final concentration of 2.2 mM. The transfected cells were kept in culture for another 3 days until harvest.
Purification of High-mannose-type EPO and Other Proteins-The His-tagged EPO and other proteins were purified by nickel-affinity chromatography. The supernatant of the cell culture (50 -150 ml) was centrifuged at low speed (4000 rpm for 15 min) and then filtered through a 0.22 M membrane to remove cell debris. The clarified supernatant was loaded on a HisTrap HP 1-ml column (GE Healthcare) that was equilibrated with PBS (pH 7.4) with 20 mM imidazole. The column was washed with 10 ml of PBS (pH 7.4) containing 50 mM imidazole and then eluted with 5 ml of PBS (pH 7.4) containing 200 mM imidazole.
Enzymatic Transformation-Purified high-mannose-type EPO was treated with several specific enzymes for further characterization. PNGase F and Endo H (New England Biolabs) treatments were performed under native conditions according to the instructions of the manufacturer. ␣1,6-Fucosidase from Lactobacillus casei was expressed and purified following the published methods (20). Treatment with Endo H and ␣1,6-fucosidase was carried out by incubation of 10 g of recombinant glycoproteins with the respective enzyme (enzyme/substrate, 1/100, w/w) in 20 l of PBS buffer (pH 7.4). The mixture was incubated at 37°C overnight.
Release of total N-Glycans-The total N-glycans of EPO and other glycoproteins were released from 10 g of glycoprotein by 100 units of PNGase F (New England Biolabs) in 20 l of 50 mM phosphate buffer (pH 7.5) with incubation at 37°C for 2 h. The released N-glycan was isolated with porous graphite carbon solid phase extraction as reported by Packer et al. (21).
Mass Spectrometry Analysis-LC electron spray ionization MS was used to analyze intact glycoproteins. The LC electron spray ionization MS was performed on an LXQ system (Thermo Scientific) with a C8 column (Poroshell 300SB-C8, 1.0 ϫ 75 mm, 5 m, Agilent). The released N-glycans were analyzed using a Bruker Autoflex III MALDI-TOF mass spectrometer (Bruker Daltonics) with 2,5-dihydroxybenzoic acid (10 mg in 10% ethanol, 10 mM NaCl) as the matrix in reflectorpositive mode.

FUT8 Is the Sole Enzyme for GnT I-independent Fucosylation
Generation of Stable Cell Lines-The two stable cell lines with either overexpression or knockdown of FUT8 were created with high-efficiency lentiviral transduction. For the overexpression cell line, FUT8 gene was cloned into a lentiviral vector, pLV-CT (a gift from Cellomics Technology, Halethorpe, MD), under the control of the CMV immediate early promoter. The vector contains a puromycin resistance gene for the screening of positive cell clones. For the knockdown cell lines, three validated shRNA lentiviral vectors that highly efficiently knock down the FUT8 gene and target different region of its mRNA were purchased from Sigma-Aldrich (TRCN0000229959, 97%; TRCN0000229960, 88%; and TRC N0000229961, 94% knockdown). Both overexpressing and shRNA lentiviral vectors were packaged with the pMD2.G envelope and psPAX2 helper plasmids (Addgene, Cambridge, MA). The lentiviruses were packaged and titrated according to Oberbek et al. (23). To generate the pooled stable cell lines, HEK293S GnT I Ϫ/Ϫ cells were transduced with either the overexpressing or shRNA viruses at a multiplicity of infection of 2 in the presence of 8 g/ml Polybrene (Clontech). Eight hours after transduction, the fresh medium with 1 g/ml puromycin (Clontech) was added to screen the puromycin-resistant clones. The pooled stable cells were expanded with 2 weeks of continuous culture.

Results
Characterization of EPO Produced from the HEK293S GnT I Ϫ/Ϫ Cell Line-As part of our project for producing homogeneous glycoforms of therapeutic glycoproteins via a chemoenzymatic approach (24 -26), we attempted to express EPO using a human GnT I knockout cell suspension culture system in the hope of producing the homogenous EPO-Man 5 GlcNAc 2 glycoform as the precursor. The natural human EPO has three conserved N-glycosylation sites and carries one O-glycan at the Ser-126 site. To focus on N-glycosylation in the production of well defined glycoforms of EPO, we eliminated the natural O-glycan by mutating Ser-126 to valine (S126V mutation). An early study also indicated that O-glycosylation was not essential for either the stability or bioactivity of EPO (27). As a result, all EPO glycoforms in this study are expressed as the S126V mutant. The expressed EPO was purified to homogeneity with nickel-nitrilotriacetic acid affinity chromatography. SDS-PAGE analysis showed that the purified protein migrated as a band around 27 kD ( Fig. 2A), which is a little bit larger than its expected molecular weight of 24 kDa, probably caused by the hydration of glycans. After treatment with PNGase F, the degly- cosylated protein shifted to a lower position of around 19 kDa ( Fig. 2A), as expected for the removal of all the three highmannose N-glycans. Unexpectedly, MALDI-TOF MS analysis of the N-glycans released by PNGase F showed two major molecular ions: one with an m/z of 1257.3 that matches N-glycan Man 5 GlcNAc 2 (calculated, molecular weight ϭ 1257.4 Da) and the other with an m/z of 1403.5 that is in agreement with a Man 5 GlcNAc 2 glycan plus a fucose moiety (Man 5 GlcNAc 2 Fuc; calculated, molecular weight ϭ 1403.4 Da) (Fig. 2B). The relative intensity indicates that the putative fucosylated Man 5 GlcNAc 2 species accounts for more than 50% of total N-glycans. For further quantification of the two N-glycans, the glycans released by PNGase F treatment were tagged in situ with an Fmoc group (by reaction with Fmoc chloride).The Fmoc glycans were isolated and quantified by HPLC analysis following a procedure reported previously (22). Based on the UV absorbance, the ratio between the fucosylated and nonfucosylated Man 5 GlcNAc 2 was found to be 58:42, which was similar to the relative intensity estimated from the MALDI-TOF MS analysis (Fig. 2B, inset). LC-MS analysis of the PNGase F-treated EPO gave a species of 19,366 Da (deconvolution data), which is consistent with the polypeptide backbone of EPO (calculated, 19,363 Da) (Fig. 2C). LC-MS analysis of the intact recombinant EPO revealed a mixture of EPO glycoforms carrying three Man 5 GlcNAc 2 glycans with varied (no to three) additions of a species of 146 Da (matching the molecular mass of fucose) (Fig. 2D).
To determine where the putative fucose moiety is attached, we performed a comparative LC electron spray ionization MS analysis of the intact recombinant EPO before and after treatment with Endo H as well as with mixtures of Endo H and an ␣-1,6-fucosidase from L. casei (20). Treatment of EPO with Endo H, which hydrolyzes high-mannose-type N-glycan at the chitobiose core to leave the innermost GlcNAc still attached to the protein, gave four ion species (m/z): 19,970,20,117,20,263, and 20,407 (Fig. 2E). The m/z 19,970 species is the molecular ion of EPO carrying three GlcNAc moieties (each at a respective N-glycosylation site). The next three ion species each had an extra 146 Da added, suggesting that the fucose moiety is attached at the innermost GlcNAc moiety of the N-glycans in the recombinant EPO. Finally, treatment of the Endo H-deglycosylated EPO with a recombinant ␣1,6-fucosidase from L. casei that is specific for hydrolyzing ␣1,6-fucosidic linkage to GlcNAc resulted in the removal of all fucose moieties in the Endo H-treated EPO, as reflected by the generation of a single molecular ion at m/z 19,969 that is corresponded to the EPO carrying three GlcNAc moieties (calculated, molecular weight ϭ 19,972 Da) (Fig. 2F). These results confirm that the new N-glycan is the Man 5 GlcNAc 2 carrying an ␣1,6-fucose moiety at the innermost Asn-linked GlcNAc residue.
Expression, Purification, and Glycan Analysis of Other Glycoproteins in the HEK293S GnT I Ϫ/Ϫ Cell Line-After the observation of more than 50% fucosylation of the high-mannose glycan of EPO in HEK293S GnT I Ϫ/Ϫ cells in serum-free suspension culture, we were wondering whether this was an observation unique for EPO or whether it would be a general phenomenon applicable to most glycoproteins expressed in this cell line and/or such a culture condition. To answer this question, we expressed two other glycoproteins, GM-CSF and the ectodomain of Fc␥IIIa receptor. GM-CSF contains two conserved N-glycosylation sites, whereas Fc␥IIIa potentially has five. As shown in the N-glycan analysis of the two recombinant glycoproteins (Fig. 3), GM-CSF also had a significant amount of Man 5 GlcNAc 2 Fuc species (estimated as 30% of total glycan), whereas Fc␥IIIa contained a relatively low level of fucosylated oligomannose glycan (ϳ3% total glycan). Taken together, these results suggested that core fucosylation of highmannose N-glycans in the recombinant glycoproteins expressed from the HEK293S GnT I Ϫ/Ϫ cell line appeared as a common phenomenon, but the extent of core fucosylation also varied depending on the nature of the proteins.

Generation of FUT8-overexpressing and Knockdown Cell Lines and Production/Characterization of EPO from These Cell
Lines-After confirming the core fucosylation of high-mannose glycan in the GnT I knockout cell line, we sought to identify the enzyme catalyzing such an unusually high transfer of fucose to high-mannose-type glycan. In previous publications, FUT8 was speculated for such a GnT I independent reaction, but no direct experimental verification was carried out. In our study, a significant amount of Man5GlcNAc2Fuc was produced, particularly in the case of recombinant EPO, which may suggest the involvement of other unknown, more active fucosyltransferases or cofactors. To investigate this possibility, we first tried to knock down the FUT8 gene in HEK293S GnT I Ϫ/Ϫ cells and then analyze the glycoforms of EPO produced from such a cell line. If a glycotransferase other than FUT8 was involved, then knockdown of FUT8 should not have a significant impact on the glycoforms of EPO. We first did a quick test by co-transfecting the EPO-expressing plasmid with three shRNA plasmids that highly knock down FUT8 expression. After glycan analysis of  MAY 20, 2016 • VOLUME 291 • NUMBER 21 purified EPO, the Man 5 GlcNAc 2 Fuc glycoform decreased from 58% to ϳ20%, as quantified by HPLC analysis of the PNGase F-released N-glycans with in situ Fmoc tagging (data not shown). Considering just 30% transfection efficacy (30% of cells were delivered in transfected plasmids, as determined by parallel transfection with the EGFP reporter plasmid), this result strongly suggests that FUT8 is at least the major transferase for the fucose transfer. To acquire conclusive evidence, we generated stable cell lines that knocked down its endogenous expression (HEK293S GnT I Ϫ/Ϫ FUT82). For the creation of stable cell lines, we first tried to use a conventional transfection method. However, the low efficiency of transfection could not generate enough seed cells (Ͼ1 ϫ 10 4 /ml) to effectively expand the cell population. Alternatively, we chose lentiviral transduction with a transfer efficiency of more than 90% to generate the stable cell line. For the knockout cell line, we used shRNA lentiviruses that would be highly efficient in knocking down FUT8 expression.

FUT8 Is the Sole Enzyme for GnT I-independent Fucosylation
Expression of EPO in the 293S GnT I Ϫ/Ϫ FUT82 cell line gave a homogeneous glycoform of EPO. LC-MS analysis of the intact recombinant EPO showed a single molecular ion of 23,009 (deconvolution data), which matches well with the glycoform of EPO carrying three Man 5 GlcNAc 2 glycans (calculated, molecular weight ϭ 23,011 Da) (Fig. 4A). MALDI-TOF MS analysis of the PNGase F-released N-glycan gave a single species at m/z 1257.2, which corresponds to Man 5 GlcNAc 2 (Fig. 4B). There was no fucosylated glycan detected. The complete exclusion of fucosylated Man 5 GlcNAc 2 in the FUT8 knockdown cell line suggests that FUT8 is the sole enzyme responsible for the attachment of fucose to the high-mannose glycans observed in the recombinant glycoproteins produced in the parent HEK293S GnT I Ϫ/Ϫ cell lines. Parallel to the creation of the knockdown cell line, we also created a FUT8-overexpressing GnT I knockout cell line (HEK293S GnT I Ϫ/Ϫ FUT81). The EPO expressed in this overexpressing cell line gave another homogeneous glycoform. LC-MS analysis showed a single molecular ion at m/z 23,448 after deconvolution, which matches an EPO carrying three homogeneous Man 5 GlcNAc 2 Fuc glycans (calculated, molecular weight ϭ 23,449 Da) (Fig.  5C). MALDI-TOF MS analysis of the N-glycan released by PNGase F confirmed that all N-glycans attached were the fucosylated Man 5 GlcNAc 2 (found, m/z 1403.8; calculated, molecular weight ϭ 1403.4 Da) (Fig. 5D). The complete fucosylation of the N-glycans in the recombinant EPO produced in the 293S GnT I Ϫ/Ϫ FUT81 cell line suggests that Man 5 GlcNAc 2 -EPO would be an efficient substrate for FUT8 in vivo. Taken together, our experimental data provide clear evidence that the FUT8 is the sole enzyme for such a GnT I-independent fucosylation pathway. Overexpression of the glycosyltransferase led to complete fucosylation of the EPO Man 5 GlcNAc 2 glycoform, whereas knockdown of this enzyme in the stable cell line totally eliminated the Man 5 GlcNAc 2 Fuc glycoform observed in the parent cell line. In addition, through this experiment, we successfully produced EPO with homogeneous core-fucosylated Man 5 GlcNAc 2 . To our knowledge, this is the first report of the production of recombinant glycoproteins with the pure Man 5 GlcNAc 2 Fuc glycoform.
In the creation of two stable cell lines, we also observed an interesting morphology change in the HEK293S GnT I Ϫ/Ϫ FUT81 cell line. As shown in Fig. 5, the knockdown cell line demonstrated a similar phenotype as the parent HEK293S GnT I Ϫ/Ϫ cell line. In contrast, the overexpressing cell line created with a FUT8-overexpressing lentivirus showed significantly enhanced cell-cell adhesion. It has been reported that overexpression of FUT8 in human colon cancer cells will enhance the core fucosylation of E-Cadherin, promoting cell-cell adhesion (28). The promotion of cell-cell adhesion in hepatoma and other types of cells by core fucosylation of adhesion molecules was reported previously as well (29).

Discussion
Previously, small amounts of Man 5 GlcNAc 2 Fuc were detected from some natural and recombinant proteins. Early reports indicated that core-fucosylated Man 4 -5 GlcNAc 2 could be detected in recombinant glycoproteins such as tyrosine phosphatase and ligand of inducible co-stimulator produced in GnT1 knockout CHO and HEK293 cell lines (13,14), but it appears as a minor component (Ͻ5%). Addition of the ␣-glucosidase I/II inhibitor N-butyldeoxynojirimycin hydrochloride (NB-DNJ) could increase the percentage of the Man 5 GlcNAc 2 Fuc glycoform in recombinant proteins produced from GnT I knockout CHO cell lines to 28% (14).
In this study, we observed unusually high core fucosylation of Man 5 GlcNAc 2 in recombinant EPO (more than 50%) and recombinant GM-CSF (roughly 30% core fucosylation). On the other hand, the core fucosylation in recombinant Fc␥IIIa receptor was detectable, but at a relatively low level (Ͻ3%). The detection of core-fucosylated Man 5 GlcNAc 2 glycoforms in all three recombinant glycoproteins expressed in the HEK293S GnT I Ϫ/Ϫ cell line, together with the previously reported detection of core-fucosylated high-mannose glycoforms in GnT I Ϫ/Ϫ CHO and HEK293S cell lines (13,14), strongly suggests that the GnT I-independent core fucosylation of high-mannose glycoforms produced in the GnT I knockout mammalian cell lines is a common phenomenon. However, the extent of core fucosylation varies and may depend on the microenvironment of the glycosylation sites, the nature of the expressed glycoproteins, and the cell culture conditions.
To examine whether FUT8 was the only fucosyltransferase responsible for the fucosylation of the Man 5 GlcNAc 2 glycoform or whether there were other enzymes or factors involved, we established stable cell lines with FUT8 overexpression or knockdown. Our experimental data demonstrate that knockdown of FUT8 completely eliminated the core fucosylation of the Man 5 GlcNAc 2 glycoform of EPO. On the other hand, overexpression of FUT8 gave full core fucosylation of the Man 5 GlcNAc 2 glycoform of EPO. These results provide direct evidence indicating that FUT8 is the sole ␣1,6-fucosyltransferase responsible for the GnT I-independent core fucosylation of high-mannose N-glycans. This conclusion is also consistent with the fact that FUT8, the mammalian ␣1,6-fucosyltransferase, appears to be the single gene so far discovered for core fucosylation of glycoproteins in mammalian systems (30). Nevertheless, at least one question remains to be answered. So far, all in vitro attempts to fucosylate high-mannose-type N-glycans by FUT8 have failed because the attachment of a terminal GlcNAc ␤1,2 linked to the mannose residue on the ␣1,3 branch of the N-glycan core appears to be a prerequisite for both the core fucosylation of Man 5 GlcNAc 2 and its further processing by ␣1,2-mannosidase (7,13,17).
The unsuccessful in vitro core fucosylation of high-mannose N-glycans by FUT8 seems contradictory to the suggestion that FUT8 was the sole enzyme to fucosylate high-mannose-type glycoforms in vivo. It is possible that other factors, such as molecular chaperones, might be involved in the in vivo systems. However, another possibility, that core fucosylation of Man 5 GlcNAc 2 is heavily dependent on the context of the protein, cannot be excluded. In fact, in vivo core fucosylation of highmannose glycan occurs in the context of proteins, whereas all in vitro FUT8 core fucosylations of Man 5 GlcNAc 2 or Man 3 GlcNAc 2 were examined either with free N-glycans or simple glycopeptides. The fact that significantly different levels of core fucosylation were observed among different glycoproteins when expressed in the GnT I Ϫ/Ϫ CHO cell line suggest the importance of protein contexts in core fucosylation. For example, the EPO Man 5 GlcNAc 2 glycoform appears to be an excellent substrate for FUT8, at least in vivo, as indicated by the fact that over 50% of the Man 5 GlcNAc 2 glycoform is core-fucosylated in the parent HEK293S GnT I Ϫ/Ϫ cell line, whereas expression of recombinant GM-CSF and Fc␥IIIa receptor under the same cell culture conditions gave 30% and 3% corefucosylated glycoforms, respectively. One explanation is that the N-glycosylation sites in EPO are easily accessible by either glycosyltransferase or glycosidase. This is supported by the fact that most of the N-glycans in recombinant EPO (produced from HEK293 cells) are well processed tri-and tetra-antennary glycans on all three glycosylation sites (31).
In the normal cells, when the N-acetylglucosaminyltransferase I and ␣1,6 fucosyltransferase are both present, the transfer action of GlcNAc by GnT I could go much faster than the attachment of a core fucose by FUT8 because Man 5 GlcNAc 2 is known to be an excellent substrate for GnT I. However, in GnT I knockout cells, the Man 5 GlcNAc 2 could be accumulated for fucose transfer by FUT8. Moreover, the easy access of the glycosylation sites of EPO is likely to make this transfer much more feasible than that of other proteins with less accessible glycosylation sites. It would be worthwhile to revisit the in vitro substrate specificity and activity of FUT8 with different N-glycans  MAY 20, 2016 • VOLUME 291 • NUMBER 21 in the context of different proteins. Understanding the details of the glycosylation pathways is essential for the precise engineering of recombinant glycoproteins to produce the desired homogeneous glycoforms for therapeutic applications without contaminations.

FUT8 Is the Sole Enzyme for GnT I-independent Fucosylation
The core-fucosylated Man 5 GlcNAc 2 glycoform of EPO produced from the FUT8-overexpressing cell line represents the first example of a successful production of fully fucosylated high-mannose-type recombinant glycoproteins. It has been reported that EPO glycoforms produced from tobacco showed a 2-fold increase in cytoprotective effect compared with that produced from CHO cells. Glycan analysis of the EPO expressed in tobacco cells revealed a significant amount of high-mannose EPO (32). In this perspective, the homogeneous Man 5 GlcNAc 2 and Man 5 GlcNAc 2 Fuc glycoforms of EPO obtained here may have potential values in the study of the cytoprotective effect of EPO.
Author Contributions-Q. Y. and L. X. W. designed the experiments, analyzed the experimental data, and wrote and edited the manuscript. Q. Y. performed the experiments. Both authors reviewed the results and approved the final version of the manuscript.