Dynamic Association between the Catalytic and Lectin Domains of Human UDP-GalNAc:Polypeptide α-N-Acetylgalactosaminyltransferase-2*

The family of UDP-GalNAc:polypeptide α-N-acetylgalactosaminyltransferases (ppGalNAcTs) is unique among glycosyltransferases, containing both catalytic and lectin domains that we have previously shown to be closely associated. Here we describe the x-ray crystal structures of human ppGalNAcT-2 (hT2) bound to the product UDP at 2.75 Å resolution and to UDP and an acceptor peptide substrate EA2 (PTTDSTTPAPTTK) at 1.64 Å resolution. The conformations of both UDP and residues Arg362–Ser372 vary greatly between the two structures. In the hT2-UDP-EA2 complex, residues Arg362–Ser373 comprise a loop that forms a lid over UDP, sealing it in the active site, whereas in the hT2-UDP complex this loop is folded back, exposing UDP to bulk solvent. EA2 binds in a shallow groove with threonine 7 positioned consistent with in vitro data showing it to be the preferred site of glycosylation. The relative orientations of the hT2 catalytic and lectin domains differ dramatically from that of murine ppGalNAcT-1 and also vary considerably between the two hT2 complexes. Indeed, in the hT2-UDP-EA2 complex essentially no contact is made between the catalytic and lectin domains except for the peptide bridge between them. Thus, the hT2 structures reveal an unexpected flexibility between the catalytic and lectin domains and suggest a new mechanism used by hT2 to capture glycosylated substrates. Kinetic analysis of hT2 lacking the lectin domain confirmed the importance of this domain in acting on glycopeptide but not peptide substrates. The structure of the hT2-UDP-EA2 complex also resolves long standing questions regarding ppGalNAcT acceptor substrate specificity.

The first committed step of carbohydrate addition to mucin-type glycoproteins is catalyzed by a family of UDP-GalNAc:polypeptide ␣-Nacetylgalactosaminyltransferases (ppGalNAcTs), 2 yielding the Tn antigen (GalNac-␣-1-O-Ser/Thr). This family is large (with Ϸ24 mammalian isoforms) and phylogenetically conserved with Drosophila expressing 14 isoforms, at least one of which is essential for development (1,2), and Caenorhabditis elegans expressing 9 isoforms (3). Subsequent elongation of the Tn structure yields an array of eight distinct "core" glycans that can be further modified by many of the glycosyltransferases resident in the Golgi. The embryonic lethality resulting from the knock-out of one of these core glycosyltransferases (the core 1 ␤1,3-galactosyltransferase) in mice underscores the biological importance of mucin-type glycans (4). The repertoire of O-glycans has been implicated in diverse biological processes including host defense (5), lymphocyte homing (6), and tumor metastasis (7), and the first example of a human disease (familial tumoral calcinosis) caused by the loss of function of a ppGalNAcT-T (ppGalNAcT-3) was recently reported (8). However, there appears to be functional redundancy among ppGalNAcT members because mice in which isoforms 4, 5, or 13 are ablated do not present with any obvious phenotype (9 -11), whereas mice in which ppGalNAcT-1 has been ablated are viable but show lymph node B-cell retention deficits (12).
The primary structure of ppGalNAcTs is similar to other type II Golgi membrane glycosyltransferases, but the ppGalNAcTs are unique among glycosyltransferases in possessing a C-terminal, ricin-type lectin domain of Ϸ130 residues containing three putative carbohydrate-binding sites (13). Biochemical analyses suggest that this domain functions in the transfer of GalNAc to glycopeptide but not peptide substrates (14,15). We recently reported the first x-ray crystal structure of a ppGalN-AcT, murine ppGalNAcT-1 (mT1). The structure revealed that the catalytic and lectin domains are closely associated, sharing Ϸ645 Å 2 /domain of interaction surface area. The structure also provided a molecular understanding for the conservation of many of the residues of the ppGalNAcTs. The mT1 crystal structure contained a bound manganese ion essential for activity but did not contain either the donor UDP-GalNAc or an acceptor substrate. To determine the molecular details of substrate binding, we have now solved the x-ray crystal structures of human ppGalNAcT-2 (hT2) bound to both UDP and to UDP and an acceptor peptide substrate EA2 (PTTDSTTPAPTTK). These structures suggest that the association of ppGalNAcTs catalytic and lectin domains can be dynamic and also reveal the molecular basis of substrate recognition by the ppGalNAcTs. cloned between the MluI/AgeI sites of pKN55-N6His-TEV. The catalytic domain of hT2 (residues 75-440) was PCR-amplified using the primers 5Ј-ACCACGGCTTGAAAGTACGGTGGCCAGACTTT and 5Ј-ACCACCGGTCTATGGAACCCTTAACTCTGGATAGAC and cloned between the MluI/AgeI sites of pKN55-N6His-TEV. The plasmids were linearized and electroporated into Pichia pastoris strain SMD1168 to create stable transformants as previously described (16).
Pichia transformants were grown at 30°C in rich medium (2% peptone, 1% yeast extract, 1% casamino acids, 1% yeast nitrogen base, 1% glycerol) to an A 600 ϭ 2-4. The cells were centrifuged, resuspended in 1 ⁄ 10 the volume of the same medium in which 2% methanol was substituted for glycerol, and induced for 16 h at 20°C. The cells were removed by centrifugation, and the supernatant was adjusted to 10 mM ␤-mercaptoethanol (␤-ME) and 5 mM EDTA. The supernatant was concentrated and diafiltered Ϸ4000-fold against 20 mM NaPO 4 (pH 7.5-8) and 0.1-0.2 M NaCl (diafiltration buffer) using a Millipore tangential flow membrane with a 10-kDa molecular mass cut-off. The sample was concentrated and applied to a 5-ml HiTrap chelate column (GE Biosciences) and eluted using a 5-column volume gradient of 0 -500 mM imidazole in diafiltration buffer. For some purifications, the column was washed with diafiltration buffer containing 25 mM imidazole and eluted with a linear gradient of 25-500 mM imidazole in diafiltration buffer. The product fractions were pooled and incubated with an equimolar amount of TEV protease at 4°C overnight in 50 mM NaPO 4 (pH 8), 25 mM imidazole, 0.2 M NaCl, and 10 mM ␤-ME (cleavage buffer). The sample was centrifuged and passed over a nickel-nitrilotriacetic acid resin (New England Biolabs) in cleavage buffer to remove the six-histidine peptide and TEV protease, and hT2 was dialyzed against 2 mM Tris (pH 8), 0.5 mM EDTA, and 10 mM ␤-ME at 4°C. Crystals were grown by hanging drop vapor diffusion at room temperature. Ternary complex (hT2-UDP-EA2-Mn 2ϩ ) crystal growth was initiated by mixing 0.5-1 l of protein solution containing 5.8 mg/ml hT2, 2 mM Tris (pH 8.0), 0.5 mM EDTA, 10 mM ␤-ME, 10 mM UDP, 10 mM MnCl 2 , and 5 mM EA2 with an equal volume of precipitant solution containing 23-25% polyethylene glycol 1000, 100 mM Hepes (pH 7.0). Binary complex (hT2-UDP-Mn 2ϩ ) crystal growth was initiated by mixing 0.5-1 l of protein solution containing 5.8 mg/ml hT2, 2 mM Tris (pH 8.0), 0.5 mM EDTA, 10 mM ␤-ME, 10 mM UDP, 10 mM MnCl 2 , and 5 mM EA2 with an equal volume of precipitant solution containing 7-10% polyethylene glycol 6000, 100 mM Hepes (pH 7.0). Although EA2 was included in the crystallization solution, no electron density for the peptide was observed in the crystal structure. The crystals were grown over 0.3 ml of precipitant solution in 48-well plates, appeared in 3-4 days, and were transferred briefly (30 -60 s) to a mother liqueur solution lacking protein but containing 10% glycerol before flash cooling in a 95-100 K N 2 stream or liquid propane.
Diffraction intensities from single binary complex crystals were collected using 1.0°oscillations on an in-house Raxis-IV detector and a rotating anode generator (Rigaku/MSC) or at SER-CAT beamline 22ID at the Advanced Photon Source. Diffraction intensities from single ternary complex crystals were collected using 0.5°oscillations on the in-house Raxis-IV detector. Intensities from 560 (ternary complex inhouse) or 110 (binary complex Advanced Photon Source) or 90 (binary complex in-house) frames were integrated and scaled using the programs DENZO/SCALEPACK (17). The hT2 binary complex crystal structure was solved by molecular replacement using the program Phaser (18) and a search model prepared from separate catalytic (residues 95-427) and lectin (residues 428 -548) domains of the mT1 crystal structure (Protein Data Bank code 1XHB) in which nonconserved residues were changed to alanine. Model building was done using XtalView (19). A partial model (74% complete) of the hT2 binary complex was built and refined against a 3.2 Å data set using several rounds of torsional simulated annealing in CNS (20) before changing to a higher resolution (2.75 Å) data set. The two noncrystallographic symmetry-related monomers of the hT2 binary complex were kept identical until the final rounds of energy minimization and B-factor refinement. The hT2 ternary complex structure was solved by molecular replacement using the program Phaser and a search model of separate catalytic and lectin domains of the hT2 binary complex structure without UDP. Domain contact areas were calculated, and Figs. 1-3 were created using the program CCP4MG (21). Protein sequence alignments were created using ClustalX (22) and edited using Seaview (23). The structures were aligned using LSQMAN (24) and optimized using the "improve" option. Fig. 4 was created using PyMol.
Glycopeptides were synthesized by Anaspec, and enzyme activity was measured as previously described (25). The reactions were initiated by adding 0.05 pmol of enzyme, and incubation times were such that not more than 10% of the limiting substrate was converted to product. EA2 and Muc5Ac-3,13 were varied from 46.

RESULTS
Overall Protein Fold-Binary complex crystals (hT2-UDP-Mn 2ϩ ) contained two molecules in the asymmetric unit, and electron density was observed for all residues except Thr 90 -Asn 102 /Lys 103 , Ala 476 , Gly 477 , and Gln 571 . Because the monomers are structurally similar (root mean square deviation ϭ 0.39 Å) and each contains UDP and Mn 2ϩ , only details for the A monomer are described. Ternary complex crystals (hT2-UDP-EA2-Mn 2ϩ ) contained a single monomer in the asymmetric unit, and electron density was observed for all residues except Leu 569 -Gln 571 . Phi/psi angles of three residues in each complex (Lys 192A , Lys 323A and Lys 323B for the binary complex and Lys 323 , Val 330 , and Met 493 for the ternary complex) were in disallowed regions of the Ramachandran plot, but because electron density for each residue was well defined, these angles were unchanged. The crystallographic data are shown in Table 1.
The catalytic domains of mT1 and hT2 are structurally similar (Fig.  1). The average root mean square deviation between corresponding C ␣ carbons of the catalytic domains varies from 0.94 (hT2 binary/ternary complexes) to 1.13 Å (hT2 binary complex/mT1). Electron density for an additional 39 (ternary complex) or 42 (binary complex) residues compared with the mT1 structure was observed at the N termini of the hT2 structures. These amino acids form two short helices connected by a random coil (Fig. 1). Electron density for all residues of the random coil (Gly 88 -Asn 105 ) is observed in the ternary complex but is absent for residues Thr 90 -Asn 102 /Lys 103 of the binary complex. In the hT2-UDP-EA2 complex, this random coil is stabilized by a hydrogen bond from the main chain oxygen atom of Asn 102 to the side chain of Arg 362 . The association of these helices and the random coil with the remainder of the catalytic domain is further stabilized primarily through association with an adjacent ␣-helix (Arg 149 -Lys 162 ). These interactions include hydrogen bonds between the side chain of highly conserved Ser 109 and the side chains of residues Ser 150 (highly conserved) and invariant Arg 154 . A stretch of amino acids (Arg 347 -Thr 358 ) could not be built for mT1 because of a lack of electron density (16). However, electron density for all of the corresponding hT2 residues (Arg 362 -Ser 373 ) was seen in both binary and ternary hT2 complexes but differs greatly (Fig. 1). Several amino acids within this flexible loop mediate UDP binding as discussed below.
As expected from the mT1 structure, the lectin domain of each of the hT2 structures forms a ␤-trefoil fold, but the orientation of this domain relative to the catalytic domain in the two hT2 structures differs from that of mT1 and from each other (Fig. 1). The catalytic and lectin domains of mT1 form a close association in which Ϸ645 Å 2 of each domain is buried. This interaction is substantially reduced to Ϸ325 Å 2 /domain in the hT2 binary complex and the two domains of the hT2 ternary complex do not associate except for the amino acids connecting them. In fact, residues Gln 443 -Ala 446 , which form the first strand of a ␤-sheet in the lectin domain of the hT2-UDP complex, unfold from this  sheet in the hT2-UDP-EA2 structure and extend the peptide tether linking the catalytic and lectin domains.
UDP Binding-The binding of UDP differs dramatically between the binary and ternary complexes (Fig. 2). Compared with the ternary complex, UDP is inverted in the binary complex with the ribose group shifted out of its ternary complex pocket to face bulk solvent. Indeed, UDP is a product of the reaction catalyzed by hT2, and its observed conformation in the binary complex is consistent with UDP leaving the active site following catalysis. Residues Arg 362 -Ser 373 of a flexible loop fold out of the way to accommodate this orientation, and the C terminus of the loop is lengthened beyond Ser 373 by several amino acids (Gly 374 -Ala 378 ) that unwind from an adjacent ␣-helix ( Fig. 2A). Residues within the loop move by as much as 25.7 Å in the ternary complex to position Arg 362 , His 365 , and Tyr 367 to interaction with UDP. Two residues (Val 330 and Trp 331 ) of a shorter mobile loop also move in to complete the seal over UDP. Similar conformational changes occurring in corresponding loops of other glycosyltransferases have been described (26). The ribose 2Ј hydroxyl hydrogen bonds to Arg 362 and invariant Glu 147 via a water molecule, and the ribose 3Ј hydroxyl forms a direct hydrogen bond to Ser 225 (the X of the DXH motif). The 3Ј hydroxyl is also positioned within hydrogen bonding distance of the peptide oxygen of Thr 143 and the amide nitrogen of Ser 225 , which could assume hydrogen bonding duties in ppGalNAcT isoforms in which this serine is replaced by alanine (Table 2). In contrast, there is no obvious substitute for the loss of the hydrogen bond between Thr 143 and the O 2 of the uridine ring in isoforms in which this threonine is replaced by the hydrophobic residues valine, isoleucine, alanine, or proline ( Table 2).
Despite the large differences in the loop residue positions and UDP orientation, the conformation of several residues that hydrogen bond to the uridine ring and the phosphate moieties of UDP is substantially unaltered between the binary and ternary complexes (Fig. 2). These residues include Thr 143 , Asp 176 , and Arg 201 , which bind the uridine ring, and Asp 224 and His 226 of the DXH motif and His 359 , which bind the phosphate groups via coordination with the Mn 2ϩ ion. Several residues of mT1 corresponding to those of hT2 mediating UDP binding have been mutated, and the activity of enzymes carrying these mutations have been described (27). Adjacent to UDP is a cavity occupied by five water molecules (see Fig. 4) presumed to be the GalNAc-binding pocket based upon similarly located pockets shown to be the sites of sugar binding for other retaining glycosyltransferases (28 -30). This pocket is lined by invariant (Arg 208 and Glu 334 ) and highly conserved (Trp 331 and Asn 335 ) residues.
EA2 Binding-A schematic diagram of EA2 binding indicating the hydrogen bonds and hydrophobic interactions it forms with hT2 is shown in Fig. 3. Electron density for the first 4 residues of EA2 was absent so only residues Ser 5 -Lys 13 are shown. EA2 binds in an extended conformation with each amino acid except Lys 13 , assuming phi/psi angles favored by ␤-strands. The binding of acceptor substrates in an extended conformation was previously hypothesized based upon secondary structure predictions of residues flanking potential glycosylation sites (31). The side chain hydroxyl of Thr 7 , shown to be the preferred residue of initial glycosylation of EA2 by hT2 and several other isoforms (32), forms a strong hydrogen bond with a ␤-phosphate oxygen of UDP and is ideally located to be the GalNAc acceptor. Analysis of EA2 binding shows that the majority of hydrogen bonds between hT2 and EA2 occur between EA2 residues Ser 5 -Pro 8 , whereas hydrophobic interactions dominate the binding of residues Ala 9 -Lys 13 .

TABLE 2
The UDP-binding residues of active ppGalNAcT isoforms are highly conserved Residues (hT2 numbering) were chosen based on the hT2-UDP-EA2 complex. Those with a gray background show a greater than 50% identity for that position. Isoforms include human, mouse, and/or rat (designated with a capital T followed by a single number), Drosophila (pGANT), C. elegans (Gly), and Toxoplasma gondii (Tg).
EA2 binds in a shallow cleft on the surface of hT2 that broadens toward the C-terminal end of EA2 and narrows toward the N terminus of EA2 (Fig. 4). Residues Pro 8 -Lys 13 of EA2 bind against one side of the cleft with the cyclic side chain of Pro 10 inserted into a pocket formed by Val 255 , Leu 270 , Trp 282 , and Phe 361 and stacked against the side chain of Trp 282 . The opposite side of the cleft contains additional, shallower pockets, two of which are occupied by water molecules. Threonine 7 points into the UDP-binding pocket with the hydroxyl group hydrogen bonded to a ␤-phosphate oxygen atom and the methyl group directed into a hydrophobic cavity lined by the side chains of Phe 280 , Ala 307 , and Phe 361 . Threonine 6 packs against a ridge in the surface of the enzyme formed by residues Arg 362 -His 365 of the flexible loop with the hydroxyl group of T6 hydrogen bonding to the main chain carbonyl of Arg 362 . Approximately 540 Å 2 of surface is covered by EA2, and only 22% of this area is contributed by UDP and residues within the mobile loop of hT2. Thus, 78% of the EA2-binding site is preformed.
Catalytic Domain Activity-The lack of interaction between the hT2 catalytic and lectin domains observed in the ternary complex suggests that the catalytic domain may not require the lectin domain for activity. Based on the crystal structure we designed and expressed the hT2 catalytic domain (residues 74 -440) lacking the entire lectin domain and compared its activity to the full-length enzyme against peptide and glycopeptide acceptors. As shown in Table 3, both k cat and K m values (and thus the k cat /K m ratio) for peptides EA2 and Muc5Ac are similar for full-length hT2 and the hT2 catalytic domain. However, removal of the hT2 lectin domain reduced glycopeptide k cat /K m ratios of the catalytic domain 4 -18-fold compared with full-length hT2 (Table 3). Thus, the absence of the hT2 lectin domain affected the transfer of GalNAc to the glycopeptides but not to the peptide substrates tested. For the Muc5Ac-3 glycopeptide, the smaller k cat /K m value was dominated by a reduced k cat , whereas for the Muc5Ac-13 glycopeptide it was dominated by an increase in K m . For the Muc5Ac-3/13 glycopeptide, the diminished k cat /K m value was caused by both a smaller k cat and larger K m . Because both the k cat and K m values represent a collection of individual rate constants (and thus are apparent catalytic constants) that have not been determined, the specific step(s) of the catalytic mechanism most affected by the absence of the lectin domain remains unknown. K m values for UDP-GalNAc were similar for full-length hT2 (11.5 Ϯ 2.4 M) and the hT2 catalytic domain (7.9 Ϯ 2.3 M).

DISCUSSION
The properties of ppGalNAcT acceptor substrates have been studied for nearly 15 years through data base analyses of known O-glycosylation sites (33,34), in vitro studies using defined peptide acceptors (31,35), sequencing of tissue-extracted mucins (36,37), and more recently systematic variation of the amino acids flanking acceptor threonine residues (38). Common findings among these investigations are a strong bias for proline 3 residues C-terminal to the site of glycosylation (the "ϩ3 site") and a preference of threonine over serine for glycosylation. The molecular basis for the proline preference is explained by the hT2-UDP-EA2 structure, which shows that the ϩ3 proline of EA2 (Pro 10 ) inserts into a cavity surrounded by hydrophobic residues (Val 255 , Leu 270 , Trp 282 , and Phe 361 ) and stacks against the side chain of Trp 282 (Fig. 4). Each of these "proline pocket" residues except Leu 270 is conserved in the T1 isoform (Table 4), which also prefers proline at the ϩ3 site, although not to the same degree as hT2 (39). There is also considerable variability in the residues lining this pocket, and the extent to which they and others control acceptor specificity will be revealed as peptide preferences for individual, purified isoforms are determined. In turn, this information should lead to more reliable predictions of isoform-specific substrates and/or sites of mucin-type glycosylation than are currently available (40). The preference for glycosylating threonine versus serine residues can be explained by the fact that loss of the methyl group would lead to a loss of interaction energy provided by the hT2 residues that bind the methyl group.
The previously determined structure of mT1 revealed that the catalytic and lectin domains form a close association in which each of the three carbohydrate-binding sites of the lectin domain face the same side of the enzyme as the catalytic site (Fig. 1). This suggested a model in which these carbohydrate-binding sites permit the capture of an array of previously glycosylated substrates for subsequent GalNAc transfer and that was consistent with the mT1-catalyzed, in vitro pattern of glycosylation of a Muc1-based peptide (16). The structures of hT2 presented in this paper indicate that, at least for this isoform, a new variable should be considered in a model of substrate capture that incorporates conformational changes within the residues linking the two domains. This "flexible tether" model suggests that lectin domain mobility endows certain ppGalNAcTs with an even greater capacity to adapt to and capture glycosylated substrates, thus ensuring the high density of glycosylation characteristic of mucin domains. Whether this model applies to other isoforms (including T1) will require solving the binary and/or ternary complex structures of additional ppGalNAcTs. Determining whether substrate binding triggers lectin domain mobility will require solving the corresponding apo enzyme structures. It should be noted that the relative orientations of the catalytic and lectin domains seen in the hT2 binary and ternary complexes may have been influenced in part by crystal packing forces and thus may not accurately reflect in vivo conformations. However, our conclusion that the two conformations observed imply an inherent flexibility between the two domains is not dependent on either of the specific orientations.
Our result showing that the lectin domain plays a role in the transfer of GalNAc to glycopeptide but not peptide substrates is consistent with studies showing that point mutations in the putative carbohydratebinding sites of the lectin domain diminish activity toward glycopeptide but not peptide substrates (14,15). However, it was noted that a lectindependent model of substrate capture is not the only mechanism used by ppGalNAcTs, at least with some glycopeptides used to monitor transferase activity (16). This conclusion was based on results showing that ppGalNAcT-7 and T-10 transfer GalNAc to threonine 6 when presented with EA2 containing GalNAc on threonine 7 (41). The structure of EA2 bound to hT2 shows how GalNAc could be attached to the residue C-terminal to the current site of glycosylation because the side chain of this residue points away from the enzyme surface (Fig. 4).
Our finding that the hT2 lectin domain is dispensable for catalytic activity contrasts with a prior study examining the effects of lectin domain truncations on the function of rat ppGalNAcT-1 (rT1). It was shown that the removal of 12 or more residues from the C terminus of the rT1 lectin domain eliminated enzymatic activity for both peptide (PPDAATAAPL) and apomucin acceptors, even though enzyme expression was only moderately reduced (14). Our results show that ppGalNAcTs lacking a lectin domain may still be active. One such transferase, Gly 8 , has been identified in C. elegans, but no activity has been demonstrated for this isoform, perhaps because a suitable acceptor substrate has not been found (3).
The pattern of substrate binding for several glycosyltransferases is ordered sequential with the sugar nucleotide donor binding first followed by the acceptor substrate (28 -30, 42). In contrast, a kinetic inves-tigation of the mechanism of bovine ppGalNAcT-1 using erythropoetin-derived peptide EPO-T (PPDAATAAPLR) indicated that substrate binding follows a random sequential pattern (43). The current structure of EA2 bound to hT2 helps rationalize this finding because the EA2binding site and, by extension, the EPO-T binding site is largely independent of interactions with both UDP and residues Arg 362 -Ser 373 of the flexible loop. However, with full-length protein substrates that make more extensive interactions with the loop residues, binding may well become ordered sequential with UDP-GalNAc binding first followed by acceptor protein.
The catalytic mechanism of retaining glycosyltransferases remains undetermined, but recently an aspartic acid has been identified as a potential nucleophile for the lipopolysaccharyltransferase LgtC (44). Because this aspartate is 8.9 Å away from the donor substrate, the authors noted that a conformational change would be required during catalysis for it to function as the nucleophile. Inspection of hT2 for potential nucleophiles surrounding the putative GalNAc pocket identified residues Arg 208 , Glu 334 , and Asn 335 (Fig. 4). However, the closest approach these residues make to the ␤-phosphate oxygen to which GalNAc would be attached is 7 Å (Asn 335 ), similar to the situation for LgtC. Structures of hT2 with bound UDP-Gal-NAc and the products UDP and the glycopeptide should help to define the

TABLE 4 Conservation of ppGalNAcT residues mediating EA2 binding
The details are as in Table 2. ppGalNAcT catalytic mechanism and perhaps that of other retaining glycosyltransferases.
A key challenge for investigating ppGalNAcT function is the identification of unknown, isoform-specific ppGalNAcT protein substrates. The large size of this transferase family coupled with the fact that each shares a common donor precludes using radiolabeled UDP-GalNAc for identifying such protein substrates in cells or tissues expressing multiple ppGalNAcTs because it would not be possible to know which isoform was responsible for labeling a given substrate. A similar situation applies to the Src family of protein kinases, each of which uses ATP as the donor substrate. A structure-based approach has been used to create ATP analogs that serve as substrates for mutant but not wild-type Src kinases (45,46). Labeling of cell lysates with the ATP analog and mutant kinases has been successful in identifying direct substrates of a given Src kinase (45). The hT2 structures determined in this study will aid the design of UDP-GalNAc analog/mutant ppGalNAcT pairs that can be used to identify isoform-specific ppGalNAcT acceptor substrates and thus will help to define the biological functions of this transferase family.