Structural basis for the specific cleavage of core-fucosylated N-glycans by endo-β-N-acetylglucosaminidase from the fungus Cordyceps militaris

N-Linked glycans play important roles in various cellular and immunological events. Endo-β-N-acetylglucosaminidase (ENGase) can release or transglycosylate N-glycans and is a promising tool for the chemoenzymatic synthesis of glycoproteins with homogeneously modified glycans. The ability of ENGases to act on core-fucosylated glycans is a key factor determining their therapeutic utility because mammalian N-glycans are frequently α-1,6-fucosylated. Although the biochemistries and structures of various ENGases have been studied extensively, the structural basis for the recognition of the core fucose and the asparagine-linked GlcNAc is unclear. Herein, we determined the crystal structures of a core fucose-specific ENGase from the caterpillar fungus Cordyceps militaris (Endo-CoM), which belongs to glycoside hydrolase family 18. Structures complexed with fucose-containing ligands were determined at 1.75–2.35 Å resolutions. The fucose moiety linked to GlcNAc is extensively recognized by protein residues in a round-shaped pocket, whereas the asparagine moiety linked to the GlcNAc is exposed to the solvent. The N-glycan–binding cleft of Endo-CoM is Y-shaped, and several lysine and arginine residues are present at its terminal regions. These structural features were consistent with the activity of Endo-CoM on fucose-containing glycans on rituximab (IgG) and its preference for a sialobiantennary substrate. Comparisons with other ENGases provided structural insights into their core fucose tolerance and specificity. In particular, Endo-F3, a known core fucose-specific ENGase, has a similar fucose-binding pocket, but the surrounding residues are not shared with Endo-CoM. Our study provides a foothold for protein engineering to develop enzymatic tools for the preparation of more effective therapeutic antibodies.

GH18 is one of the best-characterized families in CAZy. Although ϳ90 enzymes and proteins in GH18 have three-dimensional structures available in the Protein Data Bank, the majority of them are chitinases (EC 3.2.1.14), which endolytically cleave the ␤-1,4-glycosidic bonds of the chitin homopolymer. Crystal structures of GH18 ENGases from Elizabethkingia meningoseptica (former name Flavobacterium meningosepticum) (Endo-F1 and Endo-F3) (17,18), Streptococcus pyogenes (Endo-S and Endo-S2) (19 -21), Streptomyces plicatus (Endo-H) (22,23), Bacteroides thetaiotaomicron (BT3987 and BT1044) (24), and Trichoderma reesei (Endo-T) (25) have been reported. However, most of them are ligand-free (apo) structures. Structures of Endo-F3, Endo-S, and Endo-S2 only contain the octasaccharide of the biantennary complex glycan or high-mannose (Man 7 -GlcNAc) glycan as a cleavage product complex (18,20,21). These structures provide a structural basis for the enzymatic specificity for the leaving part of the glycan recognized at the "minus" subsites, according to the nomenclature defined in Ref. 26. However, complex structures of both GH18 and GH85 ENGases occupied at the "plus" subsites are not available, and recognition for the protein-side part of the substrates by ENGases is still unclear despite extensive structural studies. Because FUT8 (␣1,6-fucosyltransferase) cannot directly attach a core fucose on full-size N-glycans (5,27), protein engineering to produce ENGases that can efficiently transfer core-fucosylated glycans has been conducted (28,29). Therefore, there is a high demand for a structural basis for the recognition of the "plus" subsites and the Asn-linked peptide by ENGases.
We previously discovered and characterized GH18 ENGases that are specific for core fucose-containing N-glycans from two fungal species (Beauveria bassiana and Cordyceps militaris) and the bacterium Sphingobacterium sp. strain HMA12 (Endo-SB-ORF1188) (30). C. militaris is the ascomycete fungus that forms sexual fruiting bodies on mycosed pupae ("pupa grass" in China) and is used for traditional oriental medicine (31). The ENGase from C. militaris (hereafter called Endo-CoM) as well as the ENGase from B. bassiana (Endo-BB) and Endo-SB-ORF1188 exhibit activity exclusively on core ␣-1,6-fucosylated biantennary complex-type oligosaccharides (30). Endo-CoM and Endo-BB were active on fucose-containing glycans on rituximab (IgG) but not on the high-mannose glycans on RNase B. Endo-CoM, Endo-BB, and Endo-SB-ORF1188 form a distinct clade from other GH18 ENGases in a phylogenetic tree (Fig. 1B).
In this study, we report the crystal structures of Endo-CoM in the apo-form and complex forms with L-fucose and fucosecontaining ligands occupying the plus subsites. In addition, a mutational study and structural comparison with other ENGases provide structural insights into the core fucose-specific or nonspecific cleavage of N-glycans.

Crystal structure of Endo-CoM
A recombinant enzyme of the mature Endo-CoM protein with a C-terminal His 6 -tag was expressed in Escherichia coli. The purified enzyme migrated in SDS-PAGE as a single band with an estimated molecular mass of ϳ30 kDa, in agreement with the theoretical molecular mass of 34,216 Da. The molecular mass of the nondenatured recombinant Endo-CoM protein estimated by calibrated gel-filtration chromatography was ϳ20 kDa, suggesting that it is a globular-shaped monomer in solution. The crystal structure of Endo-CoM was solved by the single-wavelength anomalous dispersion (SAD) method using sulfur atoms contained within the native protein (Table 1). Our trials of molecular replacement using the crystal structures of GH18 ENGases as a search model failed due to low homology (Ͻ38% amino acid sequence identity). A ligand-free structure and complex structures with Fuc, Fuc-␣1,6 -GlcNAc, and Fuc-␣1,6 -GlcNAc-Asn were determined ( Table 1). The complex structures with the latter two were determined using a double mutant at the catalytic residues (D154N/E156Q), which were unambiguously identified based on the complete conservation among GH18 enzymes (Fig. 1C). The crystals contained one molecule in the asymmetric unit. A molecular interface analysis using the PISA server predicted that the protein is a monomer in solution (data not shown). The final models contain protein residues 22-315. The four structures of Endo-CoM are virtually the same, and the root mean square deviations (RMSD) for the C␣ atoms between them are less than 0.37 Å. Endo-CoM consists of a single (␤/␣) 8 barrel, which is a typical catalytic domain fold of GH18 enzymes ( Fig. 2A). There is an additional N-terminal ␣-helix, near which a disulfide bond is formed by Cys-24 and Cys-76.

Complex structures of Endo-CoM
Soaking of the Endo-CoM crystals in a solution containing L-fucose resulted in observation of clear electron density for ␣-L-fucopyranose (Fig. 2B). The ␤-anomeric form of the GlcNAc was also clearly observed in the active site when a disaccharide, 2-acetamido-2-deoxy-6-O-(␣-L-fucopyranosyl)-D-glucopyranose (Fuc-GlcNAc), was used for crystal soaking (Fig. 2C). We also used N-9-fluorenylmethyloxycarbonyl-Asn (Fuc-␣1,6 -GlcNAc-␤-)-OH (Fuc-GlcNAc-FmocAsn) for a soaking experiment. Although the electron density was relatively ambiguous, we could confidently place atoms of the molecule up to the Asn moiety (Fig. 2D). The Fmoc group attached to the amino group of Asn was disordered, probably because it was exposed to the solvent ( Fig. 2A). All of the sugar rings of Fuc and GlcNAc are in the stable chair conformation. The torsion angles of the Fuc-␣1,6 -GlcNAc glycosidic bond are as follows: ϭ Ϫ83.8°, ϭ 69.8°, and ϭ 173.0°for Fuc-GlcNAc, and ϭ Ϫ80.0°, ϭ 77.6°, and ϭ 161.1°for Fuc-GlcNAc-Asn with a definition of (O5-C1-Ox-Cx), (C1-Ox-Cx-Cxϩ1), and (O6 -C6 -C5-C4). Therefore, the C5-C6 -O6 exocyclic group of GlcNAc is in a gt conformation. This conformation was not a prevalent one among the results of molecular dynamic simulations on various forms of fucosylated complex Structure of fucose-specific endo-␤-N-acetylglucosaminidase biantennary N-glycans, but high flexibility at this bond was also observed in the same study (34). Protein structures of the three complex forms superimposed very well with no significant movement of the surrounding protein atoms (Fig. 2E). However, the side chain of the catalytic acid/base residue (Glu-156) of the apo (gray in Fig. 2E) and Fuc complexes (cyan) was significantly displaced compared with the complex with Fuc-GlcNAc (yellow) and Fuc-GlcNAc-Asn (green), in which the O4 atom of GlcNAc forms a hydrogen bond with the side chain of Gln-156. Fig. 3 shows the active-site structure of Endo-CoM complexed with Fuc-GlcNAc-Asn. The biantennary complex glycan bound to Endo-F3 (Fig. 3A) (18) and the chitin pentamer bound to ChiB (Fig. 3B) (35) are overlaid as blue sticks. The GlcNAc moiety of Fuc-GlcNAc-Asn is bound to the subsite ϩ1 that is located next to the catalytic residue (Glu-156). The O4 atom of the GlcNAc forms a hydrogen bond with the side chain of Gln(Glu)-156, supporting the implication that this residue is the acid/base catalyst. Asn(Asp)-154 is located near the nitrogen atom of the GlcNAc N-acetyl group at the subsite Ϫ1 position, confirming that this residue is the stabilizing residue for substrate-assisted catalysis (15).

Interactions with Fuc-GlcNAc-Asn and the active-site structure
The Fuc moiety is located in a side pocket, which we designate as subsite ϩ1Ј and is recognized by Endo-CoM with extensive interactions. The O2 and O3 hydroxy groups form hydrogen bonds with the main-chain amide and carboxyl of Tyr-216. Arg-218 and Asn-193 are hydrogen-bonded to the O3 and O4 hydroxy groups and the sugar ring O5 atom, respectively. In addition, the side chain of Trp-253 forms a hydrophobic interaction with the sugar ring of Fuc, the nearest carbon-carbon distance (Trp C3-Fuc C3) being 4.1 Å. The extensive and specific interactions at this site clearly explain the exclusive preference of Endo-CoM for core-fucosylated substrates (30). The residues forming the fucosebinding site (Asn-193, Tyr-216, Arg-218, and Trp-253) are completely conserved in other core fucose-specific ENGases (Endo-BB and Endo-SB-ORF1188) that were identified in our previous study (Fig. 1C) (30).
The Asn moiety is located at a displaced position from the linear homopolymer of the chitin oligosaccharide (subsites ϩ1 to ϩ3 in Fig. 3B), and there is no specific interaction with the protein. The side-chain residues in loop 7 (discussed below) are located near the Asn moiety, but the distances are Ͼ3.5 Å and Ͼ4.3 Å for Trp-253 and Thr-251, respectively. The Fmoc-linked amino group and the free carboxyl groups of the Asn moiety are exposed to the solvent, suggesting that the ENGase can act on various protein N-glycans without serious steric hindrance.

Mutational analysis
Based on the crystal structure, we performed a mutational analysis on the residues forming the catalytic and fucosebinding sites (Table 3). Mutations at the catalytic residues (E156Q and D154N/E156Q) resulted in significant 2 orders of magnitude loss of activity. Mutations at the fucose-binding site basically reduced the activity on pyridylamino (PA)fucosyl sialobiantennary, especially for the Y216A and R218A mutants, whereas the W253A mutant retained activity. Unexpectedly, mutation of N193A increased the activity,  (65). The Fuc-GlcNAc-Asn moiety bound to the Endo-CoM structure and the octasaccharide of biantennary complex glycan observed in other ENGase structures are boxed by dashed lines. Subsites of ENGases are indicated by blue characters. B, phylogenetic tree of GH18 ENGases. The fucose-specific ENGases and structure-known ones are selected. Bar, 5% sequence divergence. C, partial amino acid sequence alignment of GH18 ENGases. The catalytic residues and key amino acid residues for the Fuc-GlcNAc recognition of Endo-CoM are indicated above the sequences with red and blue arrows, respectively.

Structure of fucose-specific endo-␤-N-acetylglucosaminidase
suggesting that the interaction between Asn-193 and the O5 atom of fucose is not critical for core-fucose specificity. The activities of WT and most mutants (except for Arg-218) toward nonfucosylated PA-sialobiantennary substrate were basically very low (Ͻ1%) compared with the fucosylated substrate, consistent with our previous report (30). It was interesting that the R218A mutation slightly increased the activity against the nonfucosylated PA-sialobiantennary substrate (ϳ2.3-fold). Removal of the long and positivelycharged side chain of Arg-218 may reduce the steric assistance at subsite ϩ1Ј specificity for Fuc and endowed a broader substrate specificity to the enzyme.

Core fucose-binding pocket of ENGases
The low sequence similarity and frequent indels in the potential subsite ϩ1Ј region among ENGases (Fig. 1C) have been hampered in facilitating reliable prediction of their core fucose tolerance and specificity. Here, we compared the core fucose-binding site of Endo-CoM ( Fig. 4A) with other structure-known ENGases. Endo-F3 exhibits more than 300fold higher hydrolytic activity to core-fucosylated glycopeptides compared with nonfucosylated ones (36). In the Endo-F3 structure (Fig. 4B), several residues, including Tyr-148, Tyr-152, Asp-190, and Arg-219, surround the putative subsite ϩ1Ј, and the latter two potentially make hydrogen bonds with Fuc. Molecular surface presentation illustrates that there is a round-shaped pocket for the core fucose in Endo-CoM and Endo-F3 (Fig. 5, A and B).
Endo-S can cleave both core-fucosylated and nonfucosylated glycans and thus is described as "tolerant" to core-fucosylation (11,37). The putative fucose-binding site of Endo-S is relatively

Structure of fucose-specific endo-␤-N-acetylglucosaminidase
open and shallower than those of Endo-CoM and Endo-F3 (Fig.  5C). A side chain conformational change of Gln-308 appears to be possible (Fig. 4C), and it may give slight preference to corefucosylated glycans. Endo-S2 is also tolerant to core-fucosylated glycans (38,39) and has an open ϩ1Ј subsite similar to Endo-S (Fig. 5D).
A study on Endo-H and Endo-F1 demonstrated that the former exhibited over 50-fold higher activity to core-fucosylated substrates compared with the latter (40). The molecular surface presentation of these ENGases revealed that Endo-H has an open ϩ1Ј subsite (Fig. 5E), whereas the side chain of Asp-198 of Endo-F1 exerts severe steric hindrance on the overlaid fucose at its O2 hydroxy (Fig. 5F). Briliūtė et al. (24) reported that BT1044 is tolerant to core-fucosylated glycans. In this study, release of the Fuc-GlcNAc disaccharide from glycans liberated from human serum IgG, human serum IgA, and human colos-trum IgA was detected by HPLC and MS, and the activity was not quantitatively described. The molecular surface of BT1044 indicates that it has a very narrow pocket for subsite ϩ1Ј and appears to form a small steric clash with the overlaid fucose that can be alleviated by slight movement of the sugar side chain (Fig. 5G). Endo-T was shown to prefer high-mannose-type glycans over complex-type ones (41), but we could not find any literature about its activity on core-fucosylated glycans. The molecular surface presentation indicated that Ser-196, Phe-198, and Leu-171 of Endo-T may form a steric clash with the fucose (Fig. 5H).

Implications for glycan specificity
Structural basis of the specificity for the leaving part of N-glycans has been previously described for Endo-F3, Endo-S, and Endo-S2 (18,20,21). Here, we used the loop-displaying system that was used to describe the structural features of Endo-S and Endo-S2, in which the eight loops connecting the ␤-strands and ␣-helices of the (␤/␣) 8 -barrel (TIM-barrel) protein scaffold are separately colored (20,21). A ribbon model of Endo-CoM (Fuc-GlcNAc-Asn complex) colored with this system is shown in Fig. 6A, and the G2 octasaccharide of the biantennary complex glycan bound to Endo-S2 (thin blue sticks; Gal-

Structure of fucose-specific endo-␤-N-acetylglucosaminidase
In terms of molecular surface presentation (Fig. 6B), Endo-CoM has a clearly Y-shaped (forked) cleft with a protrusion of loop 2 (yellow) that appears to be suitable for accommodating biantennary glycans. Endo-F3 hydrolyzes both biantennary and triantennary complex-type N-glycans (42). Trastoy et al. (20) predicted a potential binding area for the Gal-GlcNAc-branch of triantennary complex-type glycans on the Endo-F3 structure (orange circle in Fig. 6C) (20). The corresponding area in Endo-CoM (Fig. 6B) as well as in Endo-S (Fig. 6D) and Endo-S2 (Fig.  6E) is blocked by protrusion of loop 7 (cyan). This observation is consistent with the previously determined substrate specificity of Endo-CoM (30); the PA-fucosyl asialotriantennary substrate was not hydrolyzed despite the presence of the core fucose.
Recently, Klontz et al. (21) succeeded in determining the crystal structure of Endo-S2 complexed with an octasaccharide part of a high-mannose glycan (salmon sticks in Fig. 6E; Man 7 -GlcNAc). The high-mannose glycan binds to the cleft in a similar manner with the complex biantennary glycan (blue sticks in Fig. 6E). However, a pocket for accommodating the ␣1,3-branched antenna, which is surrounded by loop 3 (orange) and loop 4 (magenta), was found (red circle in Fig. 6E). Although Endo-F3 and Endo-S do not have a corresponding pocket, Endo-CoM equips a similarly-shaped pocket at a corresponding position (Fig. 6B). Endo-CoM did not exhibit activity against nonfucosylated high-mannose-type PA substrates (M5-M9) (30). Unfortunately, we could not test the activity against core-fucosylated high-mannose-type glycans due to the unavailability of substrates.
Our previous study also demonstrated that the presence of the terminal sialic acids on the biantennary substrate (PA-fucosyl sialobiantennary) increased the activity of Endo-CoM by 2.7-4.6-fold compared with substrates without sialic acids (PA-fucosyl asialobiantennary and PA-fucosyl agalactobiantennary) (30). Interestingly, we found several Arg and Lys residues (Arg-58, Lys-67, Arg-113, and Lys-141) at both nonreducing ends of the Y-shaped cleft (blue in Fig. 6, A and B). These positively-charged residues probably contribute to the high activity with sialo-glycans via electrostatic interactions.

Concluding remarks
In this study, we provided a structural basis for the strict specificity of a GH18 ENGase for core-fucosylated N-glycans that are frequently present in mammal glycoproteins. However, the enzyme used in our study was isolated from a fungus that can infect caterpillars (C. militaris). The CAZy database defines 24 GH18 genes in the genome of C. militaris ATCC 34164, and Zheng et al. (31) reported that the genome of C. militaris strain

Structure of fucose-specific endo-␤-N-acetylglucosaminidase
CM01 contains 20 GH18 genes. The presence of such a large number of GH18 genes has been explained in the context of pathogenicity to infect across the insect cuticle via chitinase activities. In particular, the Endo-CoM gene (locus tag ϭ CCM_08020) was annotated as a chitinase by the genome project (31). According to our survey, the C. militaris genome does not have a fucosyltransferase gene (data not shown), suggesting that Endo-CoM is not dedicated to the degradation or transfer of intrinsic glycoproteins. Cordyceps is a genus of parasitic filamentous fungi whose main target is insects. Because insects have many N-glycans that are core-fucosylated by ␣1,3-modifications as well as ␣1,6-modifications (43,44), efficient degradation of the host glycoproteins accompanied by the action of chitinases may be vital for the invasion and energy intake of the pathogenic fungi. In contrast, another fungal GH18 ENGase, Endo-T, is specific for high-mannose N-glycans, which are observed in fungal and yeast glycoproteins, and apparently it does not have the ability to cleave core-fucosylated N-glycans (41). This is probably because the source organism of Endo-T (T. reesei) mostly relies on the degradation of plant cellulose.
The bacterium E. meningoseptica is a nosocomial human pathogen (45) and has multiple ENGases (Endo-F1, -F2, and -F3) (40). The strong preference of Endo-F3 for core-fucosylated N-glycans has been recognized for a long time, and our study provided a structural explanation for this observation.
Interestingly, Endo-CoM and Endo-F3 have a similarly-shaped pocket for fucose (Fig. 5, A and B), but the residues responsible for the recognition are not conserved (Fig. 4, A and B), suggesting that the core-fucose recognition of these ENGases has emerged independently as convergent molecular evolution.
Utilization of ENGases for chemoenzymatic glycosylation engineering is a promising strategy to achieve glycan-defined glycoproteins (46). Therapeutic antibodies with homogeneously-modified glycans produced by this method indeed exhibited enhanced antibody-dependent cellular cytotoxicity and complement-dependent cytotoxicity (47,48). Recently, Katoh et al. (29) succeeded in producing a mutant GH85 ENGase (Endo-M W251N) that is tolerant to core-fucosylation. In addition, an elegant glycosynthase strategy that enables high-transglycosylation ability has been developed for GH85 enzymes (Asn-175 mutants of Endo-M and Asn-322 mutants of Endo-D) (49 -51). These works were performed based on the ligand-free crystal structures of GH85 ENGases (16,52). Although only 11 ENGases are currently listed as "characterized" enzymes in GH85 of the CAZy database, GH18 contains 22 characterized ENGases. Therefore, our structural information on the core fucose-binding site of GH18 ENGases will support and facilitate future protein engineering of enzymes from wider origins with more precise structural tuning.

Structure of fucose-specific endo-␤-N-acetylglucosaminidase
The bacterial GH18 ENGase, Endo-S, has been successfully developed as a therapy to inhibit various experimental autoimmune disorders of mammalian systems (53). Endo-CoM was isolated from a eukaryotic origin and has a minimum size of ENGase (ϳ300 amino acids) that is much smaller than Endo-S (ϳ1,000 amino acids). Therefore, the new fungal ENGase with a concrete structural basis can be a good alternative for the therapeutic usage to mammalian IgGs because Endo-CoM possibly has an advantage on the recombinant protein production in eukaryotic cells. In the case of GH85, fungal ENGases, such as Endo-M and Endo-CC from Coprinopsis cinerea, are used as commercially available reagents (54,55). Exploring the usage of fungal GH18 ENGases that has low amino acid sequence identities to bacterial ENGases (Ͻ38%) will expand the options for future therapeutic use.

Protein production and purification
The expression plasmid for the mature Endo-CoM protein (residues 19 -315 without signal sequence) with a C-terminal His 6 -tag was constructed using pET-32b (Novagen, Madison, WI), as described previously (30). Recombinant proteins of the WT Endo-CoM and its mutants were expressed in E. coli BL21-CodonPlus (DE3)-RIL (Agilent Technologies, Santa Clara, CA). The cells were grown at 37°C in 1.5 liters of Luria-Bertani medium (1% tryptone, 0.5% yeast extract, and 1% NaCl) containing 100 g/ml ampicillin and 34 g/ml chloramphenicol until the absorbance reached 1.0 at 600 nm. After cooling the culture in a refrigerator at 4°C for 40 min, expression was induced using 0.4 mM isopropyl ␤-D-thiogalactopyranoside and continued at 15°C for 20 h. The cells were harvested via centrifugation at 8,000 ϫ g for 15 min and suspended in 100 mM Tris-HCl (pH 7.5), 500 mM NaCl, and 0.1 mM phenylmethylsulfonyl fluoride with a concentration of 0.1 g of wet cells per ml. The cells were disrupted via sonication (Branson Sonifier250D; Branson Ultrasonics Division of Emerson Japan, Kanagawa, Japan), and the supernatant was collected via centrifugation at 15,000 ϫ g for 30 min. Fine particles in the supernatant were removed using a syringe filter (Minisart hydrophilic 0.45 m; Sartorius Stedim Biotech, Göttingen, Germany). The protein was purified to homogeneity using Ni-affinity chromatography (HisTrap FF 5-ml column; GE Healthcare, Buckinghamshire, UK) and gel-filtration chromatography (HiLoad 16/60 Superdex 200 pg column, GE Healthcare). Solutions of 15 and 500 mM imidazole in 50 mM Tris-HCl (pH 7.5) and 150 mM NaCl were used for the wash and elution buffers for Ni-affinity chromatography, respectively. A solution of 20 mM Tris-HCl (pH 7.5) was used for gel-filtration chromatography, and the relative molecular mass of the protein was determined using molecular standards of thyroglobulin (669 kDa), ␥-globulin (158 kDa), ovalbumin (44 kDa), and myoglobin (17 kDa). The protein solution was concentrated using an ultrafiltration centrifugal membrane unit (Vivaspin Turbo 15, MWCO 10,000; Sartorius Stedim Biotech) in a solution of 20 mM Tris-HCl (pH 7.5). The protein concentrations were determined using the BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA) with BSA as a standard. Measurements of the protein concentrations based on absorbance at 280 nm and a theoretical extinction coefficient calculated from the amino acid sequence (68,995 M Ϫ1 cm Ϫ1 ), which was calculated using the ProtParam server (https://web.expasy.org/protparam/) 3 (57), were also used for purified protein samples to check the consistency.

Crystallography
Crystals were grown at 4°C using the sitting drop vapordiffusion method by mixing 1.0 l of an 8.8 mg/ml protein solution with an equal volume of a reservoir solution containing 0.1 M CHES-NaOH (pH 8.6) and 20% PEG3350. The apo crystals were cryoprotected in the reservoir solution supplemented with 25% PEG300. Complex crystals were prepared by soaking the crystals for 30 min (WT ϩ Fuc) or Ͻ1 min (other complex structures) in a reservoir solution supplemented with 25% PEG300 (cryoprotectant) and 100 mM L-fucose, 20 mM Fuc-GlcNAc, 10 mM Fuc-GlcNAc-FmocAsn, or 20 mM OMP-Fuc-GlcNAc. However, no interpretable electron density for OMP-Fuc-GlcNAc was observed. The crystals were flash-cooled by dipping into liquid nitrogen. X-ray diffraction data were collected at 100 K on beamlines at the Photon Factory of the High Energy Accelerator Research Organization KEK (Tsukuba, Japan) and SPring-8 (Hyogo, Japan). The data set was processed using XDS (58) and Aimless (59) for the sulfur-SAD phasing dataset or HKL-2000 (60) for other datasets. Diffraction data for phasing were collected at beamline BL-1A of the Photon Factory, which is designed for long-wavelength experiments (61). Initial phase calculation, phase improvement, and automated model building were performed using PHENIX (62). Manual model building and refinement were carried out using Coot (63) and Refmac5 (64). Molecular graphic images were prepared using PyMOL (Schrödinger, LLC, New York).
The activity assay for the WT and mutant enzymes was performed as described previously (30). Either 2 pmol of PA-fucosyl sialobiantennary or PA-sialobiantennary was mixed with a given amount of the protein (3.44 ng of WT, 354 ng of E156Q, 115 ng of D154N/E156Q, 4 ng of N193A and R218A, 36 ng of Y216A, or 11 ng of W253A). The assay was carried out in 10 l of 100 mM Na-acetate buffer (pH 3.0) at 30°C for 20 min, after which the reaction was stopped by incubating the mixture at 99°C for 10 min. The resultant samples were analyzed by HPLC (GL Science, Tokyo, Japan), which was equipped with a Wakosil 5C18 column (Wako Pure Chemicals), set at 40°C. The HPLC analysis was performed using 50 mM ammonium acetate buffer (pH 4.0) containing 0.15% 1-butanol at a flow rate of 1.5 ml/min. Fluorescence emitted from PA (excitation at 320 nm and emission at 400 nm) was monitored, and the relative hydrolytic activity of the WT enzyme was determined from the peak area of the hydrolyzed PA-fucosyl-GlcNAc or PA-GlcNAc. Experiments for each enzyme were repeated independently at least twice to ensure consistent results.