Highly conserved cysteines of mouse core 2 beta1,6-N-acetylglucosaminyltransferase I form a network of disulfide bonds and include a thiol that affects enzyme activity.

Core 2 beta1,6-N-acetylglucosaminyltransferase I (C2GnT-I) plays a pivotal role in the biosynthesis of mucin-type O-glycans that serve as ligands in cell adhesion. To elucidate the three-dimensional structure of the enzyme for use in computer-aided design of therapeutically relevant enzyme inhibitors, we investigated the participation of cysteine residues in disulfide linkages in a purified murine recombinant enzyme. The pattern of free and disulfide-bonded Cys residues was determined by liquid chromatography/electrospray ionization tandem mass spectrometry in the absence and presence of dithiothreitol. Of nine highly conserved Cys residues, under both conditions, one (Cys217) is a free thiol, and eight are engaged in disulfide bonds, with pairs formed between Cys59-Cys413, Cys100-Cys172, Cys151-Cys199, and Cys372-Cys381. The only non-conserved residue within the beta1,6-N-acetylglucosaminyltransferase family, Cys235, is also a free thiol in the presence of dithiothreitol; however, in the absence of reductant, Cys235 forms an intermolecular disulfide linkage. Biochemical studies performed with thiolreactive agents demonstrated that at least one free cysteine affects enzyme activity and is proximal to the UDP-GlcNAc binding site. A Cys217 --> Ser mutant enzyme was insensitive to thiol reactants and displayed kinetic properties virtually identical to those of the wild-type enzyme, thereby showing that Cys217, although not required for activity per se, represents the only thiol that causes enzyme inactivation when modified. Based on the pattern of free and disulfide-linked Cys residues, and a method of fold recognition/threading and homology modeling, we have computed a three-dimensional model for this enzyme that was refined using the T4 bacteriophage beta-glucosyltransferase fold.

Mucin-type O-glycans expressed on the cell surface of leukocytes figure prominently in controlling cell adhesion events and, in this respect, have been shown to play a role during the initial course of the inflammatory cascade (1,2). These sugars serve as ligands for the selectins, a family of structurally related cell-surface glycoproteins that mediate tethering and rolling of leukocytes on activated endothelial cells (3,4). There are several types of O-glycans, which are classified based on their core structure (5). Those with a ␤1,6-GlcNAc branch, namely the Core 2-type, may be extended with poly(N-acetyllactosamine) and capped with the Lewis x antigen, an ␣2-3-sialylated, ␣1-3-fucosylated tetrasaccharide that represents the minimal carbohydrate epitope recognized by P-, E-, and Lselectins (6).
Three isoforms of C2GnT have been identified and cloned to date. Two of them, the widely expressed leukocyte-type (C2GnT-I) (17,18) and the thymus-associated enzyme (C2GnT-III) (19), exhibit exclusive core 2 acceptor specificity, whereas C2GnT-II, which is distributed in mucin-secreting tissues along the gastrointestinal tract, can form core 2, core 4, and I-branching structures (20,21). Cloning of C2GnTs from different sources revealed a high degree of sequence identity across species (Fig. 1). When aligned to rat and bovine C2GnT-I, the murine protein exhibits sequence identities of 92 and 79%, respectively. Similarly, there is significant homology with the human C2GnT isoforms, C2GnT-I (84%), C2GnT-II (54%), C2GnT-III (44%), and human I-GnT (37%). In addition, a bovine herpesvirus enzyme homologous to human C2GnT-II was identified based on 81.1% amino acid sequence similarity (22), with a 57% homology to murine C2GnT-I. Interestingly, all members of the ␤1,6-Core2/I-GnT family share nine conserved Cys residues, whose structural/functional role has not yet been elucidated.
A number of in vitro studies confirm a fundamental requirement for C2GnT-I activity in the biosynthesis of ligands with effective binding to P- (23)(24)(25)(26) and L-selectins (25,(27)(28)(29); conversely, it is not certain whether the enzyme plays a prominent role in the functional decoration of E-selectin ligands (26). In vivo studies using C2GnT-I null mice showed a severe, albeit incomplete, deficit in neutrophil recruitment following induction of peritonitis by thioglycollate (25), which is consistent with the sharp reduction in P- (30), E- (30), and L-selectindependent (31) leukocyte rolling that was observed in cremaster muscle venules via intravital microscopy. Based on these results, small molecule, orally available inhibitors of C2GnT-I activity can be expected to be of therapeutic use for the treatment of overzealous inflammatory responses that lead to pathological conditions such as reperfusion injury, rheumatoid arthritis, asthma, and inflammatory bowel disease. In this regard, the development of SLe x mimics as anti-inflammatory drugs has already become an active area of research (reviewed in Ref. 32), and soluble SLe x tetrasaccharide was shown to block neutrophil invasion and acute inflammation in animal models of injury (33,34). However, the short serum half-life and low affinity of the monomeric sugar ligand in solution has, thus far, limited success in discovering viable drug therapies (32).
In support of a program focused on the development of druglike C2GnT-I inhibitors, we initially computed two possible three-dimensional protein models of the catalytic domain of the enzyme for use in both virtual screenings and optimization of lead inhibitor candidates identified via high-throughput enzyme assay efforts (35). In order to select and refine the most likely protein fold, we conducted Cys analysis by LC/ESI-MS/MS in a murine recombinant form of C2GnT-I.
This study shows that of the nine highly conserved Cys residues shared by all members of the ␤1,6-Core2/I-GnT family, eight are engaged in disulfide bonds and one is a free thiol that, although not required for enzyme activity, is responsible for inactivation of the enzyme if modified. We also noted that the only unconserved Cys residue forms an intermolecular bridge under non-reducing conditions, thereby causing dimerization of the enzyme protein. Finally, our results show that the spatial constraints defined by the disulfide bonds, together with evidence for lack of divalent cations in the core protein, argue for a protein fold predicated upon that of T4 bacteria phage ␤-glucosyltransferase.

EXPERIMENTAL PROCEDURES
Materials-Materials used for methods in molecular biology, enzymology, and cell biology were routinely purchased from reputable suppliers (specified in the text for most items) and were of the highest quality available. For the radiometric enzyme assay, UDP-GlcNAc and BSA (essentially free of fatty acids and globulins, used as a stabilizer) were from Sigma; MES (free acid) was from Calbiochem; UDP-[ 3 H]Glc-NAc (12 Ci/mmol) was from Toronto Research Chemicals (Toronto, Canada), and the acceptor substrate Gal␤1-3GalNAc␣-pNp was from Rose Scientific (Edmonton, Canada). Sep-Pak C 18 columns (500 mg) were purchased from Waters, and scintillation fluid (Scintisafe™ 30%) was from Fisher. Protein quantitation was performed with the bicinchoninic acid (BCA) kit from Pierce.
Murine, Recombinant C2GnT-I-The procedures followed to obtain a murine C2GnT-I fusion protein have been described previously (35). A truncated cDNA fragment encoding amino acids 38 -428 was prepared by PCR and fused in-frame to pPROTA vector (36) for expression as a secreted NH 2 -ProteinA-C2GnT-I-COOH chimera (ProtA-C2GnT-I). The expression vector was co-transfected into Chinese hamster ovary cells, along with pSV2neo, in a 10:1 molar ratio, using the calcium phosphate method. Cells were cultured in the presence of 800 g/ml geneticin (antibiotic G418, Invitrogen), and resistant cell clones were selected and tested for C2GnT activity in culture medium. The representative clone 614 C2 showed stable expression of C2GnT activity and was selected for enzyme production. The cells were routinely propagated in minimum essential medium (Invitrogen) containing 5% fetal bovine serum (Invitrogen) and G418 (0.2 mg/ml). To partially purify the enzyme, IgG-Sepharose Fast Flow™ beads (Amersham Biosciences) were added in a ratio of 5 l of a 50% bead slurry, 2.5 l of 2 M Tris⅐HCl, pH 8.0, and 5 l of 10% Tween 20 per ml of culture medium. Following incubation on a rocking platform at 4°C for 20 h, the beads were collected by centrifugation, washed with 10 volumes of TNT buffer (50 mM Tris⅐HCl, pH 8.0, 150 mM NaCl, 0.05% Tween 20) and 2 volumes of 5 mM NH 4 Ac, pH 5.0. The recombinant ProtA-C2GnT-I enzyme was then eluted with 1 volume 0.5 M acetic acid, pH 3.4, and immediately resuspended in 3 volumes of 0.5 M MES, pH 7.5. The enzyme preparation was routinely passed through a 0.22-micron filter prior to storage at 4°C in 30 mM MES buffer, pH 6.7. Purity of the ProtA-C2GnT-I protein was ϳ95%, based on Coomassie Blue staining following SDS-PAGE.
The cDNA was mutated at base pair 650 where a G nucleotide was exchanged for a C nucleotide, thereby resulting in a segment encoding a mutant protein in which Cys 217 was replaced with Ser. Site-directed mutagenesis was performed using the QuikChange® kit (catalog number 200518, Stratagene). The mutated construct was transiently transfected into Chinese hamster ovary-S (CHO-S) cells (Invitrogen) using LipofectAMINE™ 2000 (Invitrogen) following the manufacturer's instructions. Briefly, CHO-S cells were grown in Dulbecco's modified Eagle's medium/F-12 (Invitrogen) supplemented with 10% fetal bovine serum and 0.1 mM non-essential amino acids (growth medium). cDNA (200 g) was added to 3 ml of reduced serum medium (Opti-MEM® I, Invitrogen) and then mixed with a second 3-ml aliquot of Opti-MEM® containing 130 l of LipofectAMINE™ 2000. After 20 min of incubation at room temperature, the cDNA/LipofectAMINE mixture was added to cells (90% confluence) that were seeded in flasks that had been treated previously for 24 h with 30 ml of growth medium. After 16 h of culture, medium was removed and tested to confirm enzyme activity. Cells were washed twice with Dulbecco's phosphate-buffered saline (Invitrogen) and then cultured in 50 ml of serum-free medium (CHO-S-SFM II, Invitrogen) for 24 h. Medium was harvested and stored at 4°C if not promptly used in the subsequent enzyme purification step. Cells were washed again in Dulbecco's phosphate-buffered saline, resuspended in 70 ml of CHO-S-SFM II, and cultured for an additional 36 -48 h, after which medium was again collected.
Purification of the FLAG-C2GnT-I (Cys 217 3 Ser) chimera was performed via affinity chromatography, using the ANTI-FLAG® M1 monoclonal antibody-agarose affinity gel (Sigma) according to the manufacturer's instructions (5 ml of gel beads per 400 ml of medium). The column was washed 4 times with 50 ml of a solution made of 50 mM Tris⅐HCl, pH 7.4, 150 mM NaCl, and 5 mM CaCl 2 prior to elution of the fusion protein in 30 ml of 50 mM Tris⅐HCl, pH 7.4, containing 2 mM EDTA and 0.05% Tween 20. Eluates were concentrated using Centricon plus-80 centrifugal filter units (30,000 molecular weight cut-off, Millipore), followed by dialysis in 30 mM MES buffer, pH 6.7. Each sample was routinely passed through a 0.22-micron filter prior to storage at 4°C. Purity of the FLAG-(Cys 217 3 Ser) C2GnT-I fusion protein exceeded 98%, based on Coomassie Blue staining after SDS-PAGE.
Determination of C2GnT-I Activity-A radiometric enzyme assay for C2GnT-I was performed essentially as described previously (9,37), using a mixture of 30 mM MES buffer, pH 6.7, 1 mg/ml BSA, 1 mM UDP-GlcNAc, 1 Ci of UDP-[ 3 H]GlcNAc (12 Ci/mmol), 0.5 mM Gal␤1-3GalNAc␣-pNp as the acceptor substrate, and 10 l of recombinant C2GnT-I enzyme (15-20 ϫ 10 Ϫ6 units), 2 in a total reaction volume of 30 l. Reactions were incubated for 2 h at 37°C, followed by C 18 Sep-Pak processing to separate the reaction product, which was then quantified based on radioactive emissions. In all instances, measurements were performed in duplicate; background was calculated under identical conditions except for the presence of enzyme and regularly subtracted from the signal.
Enzyme Treatment with Thiol-reactive Agents-A fresh preparation of ProtA-C2GnT-I was divided into identical aliquots, each with ϳ1 M protein. Samples were treated with varying concentrations of iodoacetamide, 5,5Ј-dithiobis-(2-nitrobenzoic acid) (DTNB), or N-ethylmaleimide. A control sample was prepared identically, except for the addition of thiol-specific reagents (replaced by the same volume of water). Each sample, in a total volume of 20 l, was preincubated at 37°C for various time points, after which 3-l aliquots were removed from each solution and diluted with 600 l of 30 mM MES, pH 6.7. Ten l of the diluted protein solution (ϳ3.2 ng) was added to 20 l of assay mixture to determine enzyme activity (38).
Substrate Protection Experiments-Aliquots of ProtA-C2GnT-I protein (1 M) were preincubated at 37°C for 5 min with various concentrations of either donor or acceptor substrate, prior to treatment with thiol-reactive agents. Controls were prepared in the same way, except for the addition of thiol reactants (replaced by an identical volume of water). The total volume in each sample was 20 l. At various time points, 3-l aliquots were removed from each tube and added to 600 l of 30 mM MES, pH 6.7. Ten l of the diluted solution, corresponding to ϳ3.2 ng of enzyme protein, was used as the source of activity in the enzyme assay.

Alkylation of Cys Residues for LC/ESI-MS/MS Analyses-A
ProtA-C2GnT-I preparation (0.14 g/l in 10 mM Tris⅐HCl, pH 7.5, containing 50 mM KCl) was split into 2 aliquots (200 l each), one of which was treated with 1 mM DTT for 8 h at room temperature. Both samples were subjected to alkylation in the presence of 10 mM polyethylene oxidemaleimide-activated biotin (M-biotin, Pierce), which was incubated for 60 h in the dark at room temperature. M-biotin-labeled ProtA-C2GnT-I samples were denatured with 8 M urea for 1 h, after which they were transferred to a Microcon YM-30 (Millipore) unit and ultrafiltered (8,000 ϫ g) three times with 200 l of 20 mM Tris⅐HCl, pH 7.5, to remove unreacted M-biotin. Prior to PNGase F digestion, the enzyme protein preparations were reconstituted in 80 l of 50 mM Tris⅐HCl, pH 7.2.
Peptide-N-Glycosidase (PNGase) F Digestion-Lyophilized PNGase F (ProZyme) was dissolved in 10 mM Tris⅐HCl buffer, pH 7.2, containing 15 mM NaCl and 1 mM EDTA, at a concentration of 1250 units/ml. Two l of this preparation was used to treat each enzyme sample overnight at 37°C.

Identification of Free Cys Residues and Disulfide Bond Pairs-Liquid
Chromatography/Electrospray Ionization-Tandem Mass Spectrometry (LC/ESI-MS/MS) analyses were performed using a LCQ™ Classical ion trap mass spectrometer (Thermo Finnigan) with a modified electrospray ionization source (39). A positive voltage of 1.8 kV was applied to the electrospray needle, and the temperature of the stainless steel heating capillary was maintained at 220°C. The voltages at the exit end of the heating capillary and the tube lens were held at 17 and 3 V, respectively, to minimize source-induced dissociation and optimize the ESI signal of the analyte. Ion injection was controlled by automatic gain control to avoid space charge effects. The full scan mass spectrum was acquired from m/z 300 to m/z 2000. MS/MS experiments were carried out with a relative collision energy of 38%. The LC/MS analysis was conducted using a Micro-LC system (Micro-Tech Scientific) coupled to the ion trap mass spectrometer. The mobile phase was subjected to splitting prior to injection, after which a flow rate of 0.3 l/min was established through the capillary C 18 column (75 m ϫ 90 mm). The enzymatically digested peptides were eluted from the column using 0.1% formic acid in water (mobile phase A) and 0.1% formic acid in acetonitrile (mobile phase B) using a three-step linear gradient, where phase B was 10 -25% during the first 35 min, 25-35% over the next 10 min, and 35-50% in the final 10 min. The LC/ESI-MS/MS analysis was conducted using an automated data acquisition procedure, based on a cyclic use of full, zoom, and MS/MS scan modes. The most intense peak (signal Ͼ1.5 ϫ 10 5 counts) in a full scan was selected as the precursor ion, upon which a high resolution zoom and an MS/MS scan were performed to determine charge state and structural fragment ions, respectively. The resulting MS/MS spectra were then screened against a protein data base using the program Sequest to confirm the sequences of the tryptic peptides.
Fold Recognition and Comparative Molecular Modeling-Two glycosyltransferases were used as templates for homology modeling of C2GnT-I: (i) bacteriophage T4 ␤-glucosyltransferase (T4BGT, Protein Data Bank code 1JIV), which transfers glucose from UDP-glucose to DNA, and (ii) rabbit N-acetylglucosaminyltransferase I (GnT I, Protein Data Bank code 1FOA), a medial Golgi enzyme that catalyzes the addition of N-acetylglucosamine in ␤1,2-linkage from UDP-GlcNAc to Man 5 GlcNAc 2 . Initially, the lack of homology with T4BGT and GnT I prompted us to align the amino acid sequence of murine C2GnT-I with that of both templates, using each fold for threading. Manual adjustments were performed based on conformational constraints dictated by pairing of Cys residues in disulfide bonds. Alignments were further adjusted by inspection of the Protein Data Bank structures, with insertions and deletions that were moved into proximal loop regions whenever they occurred inside secondary structure elements. The predicted secondary structure for C2GnT-I was obtained using the PHD server (University of Columbia) (40). The MODELER module in Insight II (Accelrys) was used for the homology modeling; the coordinates of homologous regions were transferred from each template structure to C2GnT-I following final alignments. Optimization was performed locally to remove steric side chain clashes. Three models were generated for each alignment, and the one with the lowest objective function was chosen for research applications.
Sample Preparation for Inductively Coupled Plasma Atomic Emission Spectrometry-One liter of stock solution of partially purified mouse ProtA-C2GnT-I in 30 mM MES, pH 6.7, was concentrated to a volume of 500 ml by ultrafiltration using a Prep-Scale TFF cartridge (Millipore) with a 1 square foot membrane surface area. Further concentration to 40 ml was carried out using an Amicon stirred cell with a Millipore YM-10 ultrafiltration membrane (10,000 molecular weight cut-off). To separate contaminating BSA, this preparation was supplemented with 10 ml of 5ϫ binding buffer (100 mM Tris⅐HCl, pH 7.4, 2.5 M NaCl, 5.0 mM MgCl 2 , 5 mM CaCl 2 ) and passed through a 10-ml concanavalin A-Sepharose affinity column (Amersham Biosciences). Unretained material was removed with 200 ml of binding buffer (1ϫ) prior to elution of the C2GnT-I protein with 20 mM Tris⅐HCl buffer, pH 7.4, containing 0.5 M NaCl and 0.5 M methyl-␣-D-glucopyranoside. Six 50-ml fractions were collected and concentrated to 10 ml by using Centricon® filter units (Millipore), followed by dialysis in 10 mM MES buffer, pH 6.7, at 4°C. The sample was then concentrated to 2.5 ml, incubated with 2 mM DTT and 2 mM EDTA for 1 h at room temperature to remove contaminating metals, and finally subjected to dialysis against 5 mM Tris⅐HCl, pH 7.5, at 4°C. C2GnT-I protein concentration was adjusted to 1 mg/ml, in a volume of ϳ5 ml. A sample containing the same volume of Tris⅐HCl buffer was used to determine the background signal. Samples were submitted for metal analysis to the ANALEST Laboratory at the University of Toronto and tested by inductively coupled plasma atomic emission spectrometry by using the model Optima 3000 DV (Dual View) from PerkinElmer Life Sciences.

Effect of Reductants on Murine ProtA-C2GnT-I Activity-
During the course of experiments aimed at producing relatively large amounts of partially purified ProtA-C2GnT-I protein for use in high-throughput screening efforts (35), we observed a significant loss of enzymatic activity upon storage of the protein construct at 4°C. The decrease in activity appeared to be time-and concentration-dependent and was not caused by protein precipitation or proteolytic degradation, as evidenced by Coomassie Blue staining after SDS-PAGE (not shown).
The presence of highly conserved Cys residues in the enzyme (Fig. 1) prompted us to investigate the implication of catalytically essential thiols in the loss of activity. Fig. 2A shows that a 15-min preincubation of the enzyme with 1 mM concentrations of DTT, 2-mercaptoethanol, tris-(carboxyethyl)phosphine, GSH, and sodium sulfite caused a 10 -15-fold increase in C2GnT-I activity. The extent of activation correlated well with the degree of protein inactivation, and a Ͼ100-fold increase in activity was measured when 1 mM DTT was added to an enzyme preparation with negligible residual activity after storage for 2 months at 4°C. Activation of C2GnT-I by DTT treatment was observed over a relatively wide concentration range (i.e. 0.2-10 mM) (Fig. 2B). Kinetic analysis of untreated and DTTtreated enzyme indicated that changes in activity were associated with V max , without affecting the K m values for either substrate (Table I). Taken together, these observations suggested that ProtA-C2GnT-I inactivation during storage may be caused by air oxidation and pointed to a requirement for unmodified thiol(s) in the core protein.
Inactivation of ProtA-C2GnT-I by Thiol-reactive Agents-To confirm the requirement for unmodified thiol(s), ProtA-C2GnT-I was preincubated with either IA, DTNB, or NEM for various times prior to measuring enzyme activity. Treatment with IA and DTNB resulted in a time-dependent inactivation pattern, whereas NEM exhibited a very rapid inhibitory effect (Fig. 3A). The inhibitory effect caused by DTNB could be substantially reversed upon addition of excess DTT (10 mM) (Fig.  3B). In contrast, DTT failed to restore enzyme activity inhibited by NEM (not shown), consistent with the formation of a thioether.
Substrate-mediated Protection from IA and DTNB Inactivation-As shown in Fig. 4, the donor substrate UDP-GlcNAc protects ProtA-C2GnT-I against inactivation by IA (A) and DTNB (C) in a concentration-dependent fashion. Conversely, preincubation with acceptor substrate Gal␤1-3GalNAc␣-pNp at concentrations of up to 20 times the K m value failed to protect the enzyme against inactivation by both alkylating agents (Fig. 4, B and D). These results suggest that inactivation should be ascribed to a reaction with the sulfhydryl group(s) of the enzyme that are positioned in close proximity to the UDP-GlcNAc binding site.
Confirmation of Amino Acid Sequence and Utilization of N-Linked Glycosylation Sites-By using LC/ESI-MS/MS analysis in combination with protein data base searching, we confirmed more than 95% of the amino acid sequence predicted for the truncated form (aa 37-428) of murine C2GnT-I. The enzyme has two potential N-linked glycosylation sites, corresponding to Asn 58  Analysis of Cysteines in "Non-reduced" and DTT-treated Murine ProtA-C2GnT-I-To investigate the functional and structural significance of cysteine residues in C2GnT-I, the pattern of free thiols and disulfide bridges was investigated by using two enzyme aliquots, each containing 28 g of protein. One preparation was exposed to 1 mM DTT, whereas the other was analyzed as a non-reduced protein. In the presence of reductant, the specific activity of the enzyme was 15-fold higher than that of the untreated sample. Nevertheless, both proteins showed the same pattern of free and cross-linked Cys residues and an identical network of four intramolecular disulfide bonds. However, in the absence of DTT, one of the two free Cys residues was found to be engaged in an intermolecular bridge causing dimerization of the protein.
Identification of Free Cys Residues-Mouse C2GnT-I contains 10 Cys residues, 9 of which are conserved throughout the ␤1,6-Core2/I-GnT family (Fig. 1). We detected two tryptic peptides (aa 212-224 and aa 235-246), each containing a Cys residue (i.e. Cys 217 and Cys 235 ) that was alkylated by M-biotin. Peptide 212-224 was detected as a doubly charged ion at m/z   (Fig. 5A). MS/MS analysis of the precursor ion at m/z 1035.5 confirmed the peptide sequence (Fig. 5B), and demonstrated that Cys 217 was conjugated to the biotin reagent. Nevertheless, alkylation of Cys 217 was incomplete. In fact, peptide 212-224 could also be detected as a doubly charged ion at m/z 763.8, and MS/MS analysis of the precursor ion at m/z 763.8 confirmed the peptide sequence with an unmodified Cys residue (not shown). Fig. 6A shows the mass spectrum and the corresponding inset zoom scan of the doubly charged ion (M ϩ 2H) 2ϩ at m/z ϭ 934.4 for peptide 235-246. The ion intensity for this peptide was weak, likely due to the presence of three negatively charged Glu residues that may be responsible for decreasing the ionization efficiency in the positive ion mode. The observed mass matched the calculated (M ϩ 2H) 2ϩ ion mass of 934.4 Da for the sequence containing a biotinylated derivative. MS/MS analysis of the precursor ion at m/z ϭ 934.4 (Fig. 6B) confirmed the peptide sequence and demonstrated that Cys 235 was modified by M biotin.
In conclusion, the results demonstrate that Cys 217 and Cys 235 are both free residues in murine ProtA-C2GnT-I. Since no other peptides with free thiol groups could be detected, we concluded that the other highly conserved Cys residues (i.e. Cys 59 , Cys 100 , Cys 151 , Cys 172 , Cys 199 , Cys 372 , Cys 381 , and Cys 413 ) had to be cross-linked to form disulfide bonds.
Identification of Intramolecular Disulfide Bond Pairs-To demonstrate the participation of eight conserved Cys in disulfide bonds and elucidate their bonding pattern, we initially computed all possible disulfide combinations. Subsequently, the data generated from the LC/ESI-MS/MS analyses of untreated and DTT-treated ProtA-C2GnT-I, after tryptic or chymotryptic digestion, were screened for possible matches to the hypothetical combinations. The results revealed a set of ions with five, four, and three charges at m/z 809, 1011.3, and 1347.2, respectively, that were consistent with the presence of a disulfide-bonded pair of peptides linked by Cys 59 and Cys 413 (Fig. 7A). The inset graphic of Fig. 7A shows that the monoisotopic peak of the triply charged ion has a mass of 1346.6 Da, which matches the calculated mass (M ϩ H) of 4037.8 Da for a disulfide-bonded pair of peptides containing amino acid residues 45-61 and 402-419. In Fig. 7B, the MS/MS spectrum of the triply charged precursor ion at m/z 1346.6 contains dominant fragment ions at m/z 1594.5 for (Y 18 y 9 ) 2ϩ and 1659.1 for (Y 12 y 17 ) 2ϩ . These fragments were generated from preferential cleavage of the amide bonds of Pro 52 and Pro 408 , respectively, and correspond to those predicted to occur in Pro-containing peptides, as indicated in studies reported previously (41). Frag-  Fig. 9A shows the mass spectrum obtained from a chymotryptic digest of ProtA-C2GnT-I that demonstrates the presence of a disulfide linkage between Cys 100 and Cys 172 . The triply charged ion at m/z 974.7 corresponds to the predicted m/z for a disulfide-bonded peptide pair containing amino acids 88 -103 and 164 -173, and matches the calculated mass of the peptide pair with conversion of Asn 95 to Asp after PNGase F treatment. MS/MS analysis of the precursor ion at m/z 974.4 (Fig. 9B) is dominated by fragments generated from the C termini of the peptides, Y n y m (n ϭ 2-8 and 10; m ϭ 9 and 13-16), confirming the sequences and the presence of a disulfide bond.
From the chymotryptic digest, we also detected triply and quadruply charged ions corresponding to another pair of peptides, composed of amino acids 151-163 and 191-204, linked by a disulfide bridge between Cys 151 and Cys 199 (Fig. 10A). The inset zoom scan mass spectrum shows that the triply charged ion at m/z 1071.8 is the dominant ion, matching the calculated mass of 1071.9. MS/MS analysis of the precursor ion at m/z 1072.3 (Fig. 10B) shows that Y 14  were initially tested for catalytic efficiency under routine assay conditions. The specific activity of the mutant enzyme was ϳ90% (8.2 Ϯ 0.4 versus 9.2 Ϯ 0.2 mol/min/mg) of that exhibited by the DTT-treated wild-type protein, and K m values of the two enzymes for both substrates were virtually identical. Additionally, both reactions were linear with time for at least 12 h. UDP was found to inhibit both enzymes in a competitive mode with donor substrate, with apparent K i values of 0.14 (wild type) and 0.15 mM (mutant), suggesting that Cys 217 does not affect the UDP binding site of the protein. A summary of these results is presented in Table II.
Unlike the non-reduced, wild-type enzyme, the (Cys 217 3 Ser) C2GnT-I mutant could be stored at 4°C for at least 3 months without any loss of activity (Fig. 12A). Furthermore, we observed that the mutant protein was insensitive to sulfhydrylmodifying reagents such as DTNB, iodoacetamide, and N- ethylmaleimide, as well as the heavy metals HgCl 2 and ZnCl 2 (Fig. 12B). These data, taken together, unequivocally show that Cys 217 is the only thiol that, although not required per se for activity, causes inactivation of the enzyme if modified by sulfhydryl reagents.
Fold Recognition/Threading and Homology Modeling-Crystal structures of glycosyltransferases elucidated to date suggest a limited number of protein folds (42), which have been grouped within two distinct structural superfamilies, SpsA/GnT I and T4BGT. On a topological comparison, the T4BGT superfamily consists of two similar domains, each with a fold resembling a Rossmann nucleotide-binding motif (43), that are separated by a deep central cleft where UDP-glucose binds (44). Conversely, members of the SpsA/GnT I superfamily (45) share a single, structurally similar catalytic domain, where a DXD motif serves to coordinate an essential divalent metal and stabilize binding to the nucleotide donor. In the T4BGT family, where the DXD motif is absent, the dynamics of pyrophosphate stabilization differs among enzymes and is likely dependent upon interactions with basic residues (44).
Due to the lack of primary sequence identity with C2GnT-I, we initially examined both T4BGT and SpsA/GnT I folds to identify the most suitable template for threading and homology modeling. To this end, the distribution of disulfide bridges in murine C2GnT-I was exploited to determine distance and structural restraints. Fig. 13 shows the final sequence alignments between the putative catalytic domain of murine C2GnT-I (aa 117-428) and T4BGT, whereas the corresponding threading model is shown in Fig. 14A. The cysteine network fits well into the C2GnT-I/T4BGT model; for example, the C terminus forms an extended ␣-helix packed against the N-terminal domain that, in fact, brings the termini in close proximity and accommodates the bridge between Cys 59 and Cys 413 without affecting other structural segments. In this model, the free, conserved Cys 217 residue lies within the active site, with partial accessibility, whereas the non-conserved Cys 235 appears on a small flat surface, made of extended ␤-sheets, that is positioned outside the active site.
The threading model for C2GnT-I (aa 117-428), based on the SpsA/GnT I fold, is shown in Fig. 14B. Here, the position of Cys 217 is confirmed nearby the active site, although buried underneath the pocket. A number of observations argue against the use of this template as follows: (i) absence of a consensus DXD motif; (ii) unlikely rationalization of the observed bridge between Cys 59 and Cys 413 ; and (iii) lack of metals in the core protein, as shown by inductively coupled plasma atomic emission spectrometry (not shown). In conclusion, these results strongly suggest that the likely fold for the murine C2GnT-I is that of the T4BGT structural superfamily. DISCUSSION The structural and functional relevance of highly conserved Cys residues has been recently substantiated for a number of glycosyltransferases (reviewed in Ref. 46), but not for the enzymes of the ␤1,6-Core2/I-GnT family.
In the present study, the arrangement of disulfide bonds and free thiols in the C2GnT-I protein was elucidated by using a murine truncated form of the enzyme, fused to protein A and expressed in Chinese hamster ovary cells. The analysis was performed via protease digestion followed by capillary LC/ESI-MS/MS according to a previously validated procedure (47), which was designed to prevent the occurrence of thiol/disulfide FIG. 12. Stability of (Cys 217 3 Ser) mutant C2GnT-I enzyme during storage at 4°C and resistance to thiol-reactive agents. A, freshly prepared samples of wild-type (OE) and (Cys 217 3 Ser) mutant (q) C2GnT-I (2 g in 10 ml of MES buffer, pH 6.7) were split into 0.5-ml aliquots prior to storage at 4°C in tightly capped Eppendorf tubes. Aliquots were tested at different time points, as shown. Enzyme assays, run in triplicate, were performed as described under "Experimental Procedures", using 2 ng of protein. B, the activity of the (Cys 217 3 Ser) mutant C2GnT-I was measured in the presence of thiol-reactive agents that were preincubated with the enzyme (2 ng) for 30 min at 37°C prior to starting the reaction. Final concentrations of agents in the reaction mixture were as follows: DTNB, 50 M; IA, 500 M; NEM, 10 M; HgCl 2 , 10 M; ZnCl 2 , 500 M. C is the control. The same concentrations of HgCl 2 and ZnCl 2 inhibited the wild-type enzyme, in the absence of reductants, Ͼ95 and 70%, respectively. exchange reactions, thereby ensuring correct assignments for cysteine residues.
Our results confirm more than 95% of the predicted amino acid sequence for murine C2GnT-I and indicate that the two possible N-glycosylation sites, corresponding to Asn 58 and Asn 95 residues, are both utilized. This observation is consistent with a study of a human truncated C2GnT-I form expressed in the baculovirus/Sf9 insect cell system, which verified the presence of N-glycans in both predicted sites (48). However, it must be noted that native proteins may exhibit a different pattern of glycosylation than recombinant proteins, in view of the apparent lack of N-glycosylation on the C2GnT-I enzyme from HL60 cells and, more generally, the existence of tissue-specific enzyme forms with marked differences in size (49).
To date, murine C2GnT-I is the only glycosyltransferase that has been shown to contain four intramolecular disulfide bonds, with a pattern of Cys pairing between highly conserved residues that is identical in both the presence and absence of 1 mM DTT. Thus, notwithstanding the dramatic effect of DTT upon C2GnT-I enzyme activity, these findings suggest that the reductant does not promote any thiol/disulfide exchange or disulfide shuffling and that the covalent interactions between Cys residues are most likely required to maintain a stable structural protein geometry rather than directly affecting the catalytic properties of the enzyme. That the four disulfide bonds define important conformational arrangements is clearly evidenced by the recent observations of Yang et al. (50), which showed that in human recombinant C2GnT-I, substitution of Cys into Ser for seven of the eight residues corresponding to those engaged in disulfide bonds in the murine protein caused a total loss of catalytic activity. 3 Interestingly, the disulfide linkage formed between Cys 59 and Cys 413 brings the amino acid residues located near the N and C termini of the catalytic domain within close proximity. This distinctive arrangement is similar to that occurring in ␣(1,3/1,4) fucosyltransferase III (51) and polysialyltransferase ST8Sia IV (52), where site-directed mutagenesis demonstrated that such steric conformation may be associated with acceptor substrate specificity (53,54) and catalytic efficiency (52), respectively. Similarly, the catalytic activity of GM2 synthase depends on the formation of disulfide-bonded homodimers, whose antiparallel orientation also brings the N and C termini in close proximity, most likely to establish a catalytic domain (55).
Based upon the limited number of protein folds that have been proposed for glycosyltransferases (42), one could speculate that all members of the C2GnT family share a similar structural conformation. For example, it is possible that the pattern of disulfide bonds observed in murine C2GnT-I is also conserved in other enzymes of the ␤1,6-Core2/I-GnT family. However, this is not the case for the ␣(1,3/1,4) fucosyltransferase family, whose members display four conserved Cys residues but different disulfide patterns, for example in ␣(1,3/1,4) fucosyltransferase III and VII (51,56). We have proposed previously that these structural differences may be important in the modulation of acceptor substrate specificities, a hypothesis that could also be valid for the ␤1,6-Core2/I-GnT family in light of the differential functionalities of C2GnT-I and -III versus C2GnT-II and IGnT.
Two free Cys residues have been identified in the murine C2GnT-I protein, one of which, Cys 217 , was identified as a free thiol in both the presence and absence of 1 mM DTT; the second, Cys 235 , is in a reduced form only in the presence of reductant, but engages in an intermolecular bridge in a non-reduced milieu. Protein dimerization is indicative of the presence of Cys 235 on the outer surface of the protein, and in this respect, our proposed three-dimensional model positions this residue on a flat area outside the active site.
The biological explanation for the formation of C2GnT-I homodimers has yet to be elucidated. There are studies suggesting that disulfide-mediated oligomerization of Golgi enzymes may contribute to a differential localization/retention in the organelle (57,58) or be responsible for other important functional roles. For example, the catalytically active forms of ␣-mannosidase II (59), UDP-GlcNAc:dolichol-P GlcNAc-1-Ptransferase (60), ␤1,4-galactosyltransferase (61), and GM2 synthase (55) are represented by dimeric proteins. Conversely, the ␣2,6-sialyltransferase homodimer, which accounts for ϳ30% of the total protein in the Golgi, is essentially inactive, due to a weak affinity for the sugar nucleotide donor. However, its unaltered ability to bind galactose and galactose-terminated substrates suggests that it may act as a lectin (62).
The fact that Cys 235 is not conserved within the ␤1,6-Core2/ I-GnT family suggests that this residue does not play a role in catalysis. Furthermore, our finding that a Cys 217 3 Ser mutant of the murine recombinant C2GnT-I, in which Cys 235 is the only free thiol, retains activity during storage, rules out the hypothesis that dimerization is associated with loss of enzyme activity. It is possible that dimerization of the recombinant protein may simply represent a mechanical event, resulting from suboptimal storage conditions. In this regard, we note that the native enzyme from mouse kidney is expressed as a 50-kDa monomeric protein (63).
Classic biochemical data using thiol-reactive agents unequivocally show that the activity of the mouse recombinant C2GnT-I is thiol-dependent. Because of air oxidation, the enzyme is unstable when stored at 4°C. The loss of activity is gradual and correlates with both time and enzyme concentration. However, the addition of reducing agents was found to restore full enzyme activity, regardless of the rate of inactivation. The effect of DTT on C2GnT-I activity could be observed within a relatively wide range of concentrations, with detrimental consequences only at Ն20 mM, most likely due to the disruption of the disulfide bond network. The enzyme reacti-vation following treatment with DTT or other less potent reducing agents, such as sodium sulfite, suggests the occurrence of sulfenic acid (R-SOH) in the enzyme solution, which can be recycled back to thiol (64), rather than sulfinic and sulfonic acid.
The high level of conservation of Cys 217 and its position in the active site, evidenced by the three-dimensional model shown in Fig. 14A, pinpoint Cys 217 as the free thiol that may influence enzyme activity. In fact, the Cys 217 3 Ser mutant of C2GnT-I was insensitive to air oxidation and thiol-reactive agents, thereby demonstrating that Cys 235 is not implicated in enzyme inactivation. Interestingly, the mutant enzyme displayed a specific activity and kinetic properties comparable to those of the DTT-treated wild-type protein (Table II). These findings, taken together, establish that Cys 217 is not required for enzyme activity per se but represents the only thiol that causes enzyme inactivation when modified.
The unaltered K m values of the "oxidized" enzyme and the mutant enzyme for both substrates, as compared with those of the "reduced" form, lead to the conclusion that Cys 217 is not essential for UDP-GlcNAc binding. This is in apparent contrast with the observation that UDP-GlcNAc protects C2GnT-I from IA-and DTNB-mediated inactivation, which suggests that the thiol is in very close proximity to the UDP binding site, similar to the corresponding residue in the human enzyme (50). In this regard, recent crystallographic studies on glycosyltransferases describing the nature of the interaction with UDP-donor substrate (44,45,65) indicate that the binding sites contain a predominance of positively charged side chains that may offset the high negative charges of the phosphate groups. For example, the free Cys residues of N-acetylgalactosaminyltransferase I (66) and glucuronosyltransferase (67) are essential for substrate binding, as evidenced by the fact that their serine mutants display much weaker affinities for the corresponding UDP-sugar donor.
Our observations raise the possibility that Cys 217 plays a structural role in the enzyme. For example, this residue may be part of a loop that folds down over UDP-GlcNAc, similar to the thiol found in the active site of the galactosyltransferase LgtC from Neisseria meningitidis (68). Oxidation or other modifications of Cys 217 could then prevent the proper folding of the loop by causing steric hindrance and/or affecting charge status and energetics. This, in turn, might result in inactivation of the enzyme. This hypothesis would explain the unchanged K m values measured for the residual, non-reduced C2GnT-I for both substrates. Additionally, our proposed three-dimensional model rationalizes sensitivity to steric hindrance, given the location of Cys 217 within a region with partial accessibility.
In conclusion, the results shown in this study provide structural information on murine C2GnT-I that can represent, in the absence of crystals, an excellent starting point for virtual screenings and the rational design of enzyme inhibitors. Additionally, the biochemical analyses leading to the identification of a free thiol that affects enzyme activity, without being involved in catalysis, strongly enhance the value and potential of a (Cys 217 3 Ser) C2GnT-I mutant for practical use in highthroughput screening platforms geared toward the selective identification of specific enzyme inhibitors.