Glycobiology vol. 13 no. 5 pp.

377±386, 2003
DOI: 10.1093/glycob/cwg042

Molecular modeling of glycosyltransferases involved in the biosynthesis of

blood group A, blood group B, Forssman, and iGb3 antigens and their
interaction with substrates

a2,3, Christelle Breton2, Jitka

Helena Heissigerov
Moravcova , and Anne Imberty1,2
3
2 surfaces (erythrocytes and/or vascular endothelium) of var-
Centre de Recherches sur les Macromolecules Vegetales, CNRS
(affiliated with Universite Joseph Fourier), 601 rue de la Chimie,
BP 53, 38041 Grenoble cedex 9, France; and 3Institute of Chemical
Technology, Technicka 5, 166 28 Prague, Czech Republic
Received on November 18, 2002; revised on December 9, 2002; accepted
on December 17, 2002
blood transfusion and organ allo- and xenotransplantation.
A and B blood group oligosaccharides are located on cell
ious mammals (Oriol, 1987) whereas the xeno-antigen, the
Forssman antigen, and the isogloboside 3 (iGb3) glycolipids
are not normally produced by human cells. The terminal
structures of these oligosaccharides are represented in
Scheme 1.

Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010

A terminal a1-3 linked Gal or GalNAc sugar residue is the
common structure found in several oligosaccharide antigens,
such as blood groups A and B, the xeno-antigen, the Forssman
antigen, and the isogloboside 3 (iGb3) glycolipid. The enzymes
involved in the addition of this residue display strong amino
acid sequence similarities, suggesting a common fold. From a
recently solved crystal structure of the bovine a3-galactosyl-
transferase complexed with UDP, homology modeling meth-
ods were used to build the four other enzymes of this family in
their locked conformation. Nucleotide-sugars, the Mn2 ‡ ion,
and oligosaccharide acceptors were docked in the models.
Nine different amino acid regions are involved in the substrate
binding sites. After geometry optimization of the complexes
and analysis of the predicted structures, the basis of the
specificities can be rationalized. In the nucleotide-sugar bind-
ing site, the specificity between Gal or GalNAc transferase
activity is due to the relative size of two clue amino acids. In
the acceptor site, the presence of up to three tryptophan
residues define the complexity of the oligosaccharide that
can be specifically recognized. The modeling study helps in
rationalizing the crystallographic data obtained in this family
and provides insights on the basis of substrate and donor
recognition.
The enzymes that are involved in the biosynthesis of
those Gal(NAc)a1-3Gal(NAc)-containing antigens display
strong amino acid sequence similarities, suggesting a com-
mon fold, and all belong to the same family, GT6, accord-
ing to the classification in the CAZY database (http://
afmb.cnrs-mrs.fr/~cazy/CAZY/index.html). The a3-galacto-
syltransferase (a3GalT), responsible for the biosynthesis of
a-Gal epitope (or xeno-antigen), is found in many mamma-
lian species, including most primates and New World mon-
keys, but not in humans and their closest relatives because
of the mutational inactivation of the gene (Galili and Swan-
son, 1991). Species lacking the a3GalT do not have the a-
Gal epitope on glycoconjugates, and about 1% of their
circulating antibodies are directed against this antigen,
causing the hyperacute rejection in xenotransplantation.
Because several crystal structures of a3GalT have been
solved recently (Boix et al., 2001, 2002; Gastinel et al.,
2001), structurally and mechanistically this enzyme is a
model for the several other related retaining glycosyltrans-
ferases of varying donor and acceptor substrate specificity,
such as Forssman glycolipid synthase (Forss-S) (Haslam
and Baenziger, 1996), isogloboside 3 synthase (iGb3-S)

Key words: blood group antigens/galactosyltransferase/

glycosyltransferase/molecular modeling

Introduction
Carbohydrates in the form of glycoproteins and glycolipids
play crucial roles in various signaling and molecular recog-
nition processes, affect the stability and structure of pro-
teins, and are epitopes recognized by the immune system
(Varki, 1993). A core a-galactosyl(1-3)galactose moiety is
the common structure of several oligosaccharide antigens
present on mammalian cell surfaces. These antigens have
been well characterized because of their importance for

1 Scheme 1. Terminal structures of the AB blood group/antigens, the

To whom correspondence should be addressed; e-mail:
imberty@cermav.cnrs.fr xeno-antigen, the Forssman antigen, and the iGb3 antigen.

Glycobiology vol. 13 no. 5 # Oxford University Press 2003; all rights reserved. 377
H. Heissigerova et al.
Table I. Specificity and reaction products of the glycosyltransferases related to blood group A and B synthases
Enzyme Donor
a3GalT UDP-Gal
GTB UDP-Gal
iGb3-S UDP-Gal
GTA UDP-GalNAc
Forss-S UDP-GalNAc
a
The monosaccharides that carry the acceptor hydroxyl are indicated in boldface type.
(Keusch et al., 2000), and the histo-blood group A and B
glycosyltransferases (GTA and GTB) (Yamamoto et al.,
1990).
Enzymes belonging to this a3-Gal(NAc)T family share
common features. They all use a UDP-nucleotide sugar as
donor, retain the configuration of the Gal (or GalNAc)
transferred, and their activity is strictly dependent upon
the presence of a divalent cation (generally Mn2 ‡ ). Never-
theless, they differ by their fine specificity: a3GalT, blood
group B transferase, and iGb3-S only use UDP-Gal as
donor, whereas blood group A transferase and Forss-S
use UDP-GalNAc. In addition, the mouse AB glycosyl-
transferase derived from a cis-AB gene can transfer sugar
from both donors (Yamamoto et al., 2001). The amino acid
basis for Gal versus GalNAc specificity has been well defined
for the blood group A and B transferases, which differ by
only four amino acids (Yamamoto and McNeill, 1996). On
the other hand, very little is known about the basis for the
acceptor specificity: N-acetyllactosamine and lactose are
the acceptor for a3GalT and iGb3-S, respectively. Blood
group enzymes require substitution by a L-fucose on posi-
tion 2 of the acceptor galactose, whereas Forss-S requires a
N-acetyl group at the same location. Details for donor and
acceptor specificities are listed in Table I.
Like the majority of glycosyltransferases (Paulson and
Colley, 1989), a3Gal(NAc)T enzymes are type II membrane
proteins with a short N-terminal cytoplasmic tail, single-
transmembrane domain, stem, and C-terminal catalytic
region. Despite their importance, structural information
about glycosyltransferases is rare, and only 12 different gly-
cosyltransferases have been crystallized until now (see
review; U È nligil and Rini, 2000; Breton et al., 2002). Although
they share little or no sequence similarity, these 12 glycosyl-
transferases appear to adopt only two different folds, which
have been named BGT fold and SpsA fold for reference to the
first structure solved in each case. Among the known struc-
tures, only five are retaining glycosyltransferases, namely,
bovine a3GalT (Boix et al., 2001, 2002; Gastinel et al., 2001),
human blood group A and B synthase (Patenaude et al.,
2002), a bacterial a4-galactosyltransferase (LgtC) (Persson
et al., 2001), and rabbit glycogenin that initiates the biosynth-
esis of glycogen (Gibbons et al., 2002). All of these retaining
glycosyltransferases adopt the SpsA fold.
Analysis and comparison of glycosyltransferase crystal
structures in the goal of elucidating the basis of substrate
recognition and catalysis has been complicated by the
occurrence of large movement of one or two loops involved
378
Acceptora Product
bGal1,4bGlcNAc-R aGal1,3bGal1,4bGlcNAc-R
aFuc1,2bGal1,4bGlcNAc-R aGal1,3[aFuc1,2]bGal1,4bGlcNAc-R
bGal1,4bGlcb-ceramide aGal1,3bGal1,4bGlc1-ceramide
aFuc1,2bGal1,4bGlcNAc-R aGalNAc1,3[aFuc1,2]bGal1,4bGlcNAc-R
bGalNAc1,3aGal-R aGalNAc1,3bGalNAc1,3aGal-R
in substrate binding. First evidenced in inverting glycosyl-
transferases for b2-GlcNAc transferase (U È nligil et al., 2000)
and then b4-Gal transferase (Ramakrishnan and Qasba,
2001), this conformational change has also been demon-
strated to occur in retaining glycosyltransferases (Boix et al.,
Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010
2001). In a3GalT, opening of the acceptor site is dependent
on a donor substrate-induced conformational change (Boix
et al., 2002). Among retaining glycosyltransferases, only
a3GalT (Boix et al., 2001, 2002) and LgtC (Persson et al.,
2001) have been crystallized in this substrate buried state,
also called Form II. From the several crystal structures that
have presently been obtained in Form I (open) and Form II
(closed), it can now be inferred that not only the presence of
UDP or UDP-sugar is necessary for obtaining the ``locked''
conformation but also that this substrate should be added
to the crystallization medium. Crystals obtained in the
absence of substrate and then soaked with UDP do bind
the substrate in the active site, but the ordering of loops
does not occur in the solid state. Owing to the difficulties in
obtaining cocrystals of glycosyltransferases with their sub-
strates, molecular modeling is an alternative to approaching
the role of flexible loops in substrate binding.
In the present article, we propose to use the recent high-
resolution structure of a3GalT (Boix et al., 2001) as a tem-
plate to model the other related enzymes with a3Gal(NAc)
transferase activity listed in Table I. Docking of nucleotide-
sugars on one hand and of oligosaccharide acceptor on the
other will lead to the comparison of the architecture of the
binding sites. Comparison with the recently solved blood
group A and B synthases (Patenaude et al., 2002) validates
our modeling method and allows for comparing the two
conformational forms of the enzymes. In addition, the dif-
ferent sequence motifs involved in substrate binding have
been defined, and their precise role has been investigated.
Results
Description of the overall structure of the models
Figure 1 displays the sequence alignments of the enzymes of
interest for the study. This alignment is the basis of the
homology building, and only the catalytic regions are
shown. This glycosyltransferase family displays a high level
of sequence identity. The percentage of identity varies from
44% to 55% (with the exception of GTA and GTB, which
display 98% identity). There are very few insertions or dele-
tions in the regions corresponding to secondary structures.

Fig. 1. Amino acid sequence alignment of bovine a3GalT and related enzymes of the a3Gal(NAc)T family. Secondary structure elements of a3GalT are
indicated above the sequences and numbered as in Boix et al. (2001). Conserved amino acids have a black background and preserved ones a gray
background. Regions involved in ligands binding are boxed. The stars indicate the four amino acids that are different between GTA and GTB sequences.
Loop regions display more variability, but their sizes are
comparable, thus facilitating the modeling study.
Because all glycosyltransferases belong to the SpsA fold
superfamily, the enzymes modeled here adopt a globular
shape. Figure 2A displays the overall structure of the GTA
model. One face is responsible for binding the nucleotide-
sugar and the acceptor. Because the crystal structure of
a3GalT complexed with UDP was taken as the reference
molecule (Boix et al., 2001), all models are in their locked
Form II conformation, with the nucleotide-sugar almost
completely buried under the C-terminal region. The accep-
tor binding site appears as a deep crevice adjacent to the
nucleotide-binding site. These two binding sites are made
up with residues belonging to nine peptide regions, identi-
fied as ligand binding regions (LBRs) and labeled from A
to I in Figures 1 and 2. Because there is no significant
differences in loop size, the overall structure of the four
proteins modeled here does not present large variation,
and only one has been displayed.
The nucleotide-sugar binding site: predicted interactions
between enzymes and UDP-Gal/GalNAc
The UDP-sugar binding domain is located between the
central b-sheet (b2±b5) and two long a-helices (a3 and a4)
and is capped by the C-terminal region. Figure 3A gives
insights into the binding site pockets of four models. The
model for binding UDP-Gal by iGb3-S is very similar to the
prediction for a3GalT and is not displayed. The Mn2 ‡
binding site is made up by the 225 DVD227 (numbering rela-
tive to a3GalT), the very conserved LBR-C region. The
uridine moiety is bound through the conserved
134
FA(IV)(GK)(KR)Y139 motif in LBR-A, whereas the
ribose ring interacts with amino acids of LBR-A, -B, and
-C. The C-terminal region makes direct contacts with the
Molecular modeling of a3Gal(NAc)transferases
Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010
phosphate atoms through Lys359 (LBR-H) and Arg365
(LBR-I), both positions absolutely conserved among the
enzymes studied here.
The Gal/GalNAc moiety interacts with different regions:
LBR-B, -C, -E, and -F. Details of the hydrogen bonds
involving the sugar moiety have been listed in Table II.
Depending on the model, five to seven hydrogen bonds
are established between the sugar moiety and the proteins
(see Figure 4 for GTA/GalNAc). Four of them are con-
served in the whole family: the first aspartate amino acid of
the DXD motif (LBR-C) receives a hydrogen bond from O-
3 hydroxyl group. An acidic amino acid from region LBR-F
(Asp316 in a3GalT) receives hydrogen bonds from both
O-4 and O-6, this latest hydroxyl receiving a hydrogen
bond from the conserved Ser amino acid of LBR-B. The
Gal/GalNAc specificity is ensured by LBR-E, which con-
tains the 277 FYYX(GA)(GA)282 motif. From our modeling,
it appears that the possibility to accept an N-acetyl group in
the binding site is controlled by a pair of amino acids
facing each other (His280 and Ala282 in a3GalT). The
steric hindrance created by Met266/Ala268 in GTB and
by His280/Ala282 in a3GalT (His253 in iGb3-S) is clearly
displayed in Figure 3A. The enzymes that can use GalNAc
have a glycine at one of these two positions (Leu266/Gly268
in GTA and Gly261/Ala263 in Forss-S), therefore allowing
for an opening of this subsite. These particular two residues
correspond to two of the four amino acids differentiating the
human blood group A and B glycosyltransferases.
The acceptor binding site: interactions between
enzyme and oligosaccharide acceptor
Acceptors of various size (disaccharides and trisaccharides)
have been docked into protein models. The molecules that
have been used in the modeling studies are those listed in
379
H. Heissigerova et al.

Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010

Fig. 2. (A) Graphical representation of the model of human GTA in interaction with Mn2 ‡ , UDP-GalNAc, and trisaccharide acceptor. The Connolly
protein surface is represented with different colors for the different peptide regions involved in substrates binding. These loops are labeled according to
Figure 1. (B) Same representation for the crystal structure of GTA in interaction with Mn2 ‡ , UDP, and H-disaccharide analog (Patenaude et al., 2002).
(C) Stereoscopic representation of the binding site of the GTA model. (D) Same representation for GTA crystal structure. (E) Ribbon representation of
the peptide chain in the GTA model, together with space fill representation of the amino acids involved in capping the nucleotide binding site.

Table I, although the noncarbohydrate part (ceramide)
has not been taken into account in the modeling. When
compared to the donor nucleotide sugar, it can be stated
that the acceptor oligosaccharides are more exposed to the
solvent and establish a limited number of contacts with the
protein surface. Nevertheless, four regions seem to play an
important role in binding the acceptor and defining the
specificity, namely, LBR-D, -F, -G, and -H. These regions
have been color coded in Figure 3B where the proteins are
represented by their accessible surfaces. Minor contacts are
also established with residues of LBR-E (Tyr278 in a3GT).
Details of hydrogen bonds and hydrophobic contacts are
listed in Tables III and IV.
In all enzymes of this family, the primary acceptor is a
Gal or GalNAc residue. This can be correlated to a con-
served structural feature: LBR-F is almost invariant and

380
 Molecular modeling of a3Gal(NAc)transferases

brings a Trp residue in contact with the galactose acceptor

A (Trp314 in a3GalT). As previously observed in several
S185 S185
carbohydrate-binding proteins (Vyas, 1991; Rini, 1995),
the aromatic rings stack against the CH of the galactose
D302 D302

rings and ensure specificity of binding for a galactose at this

E303
E303

G268
A268 position.
L266 M266
For two of the enzymes, a3GalT and iGb3-S, a terminal
GTA GTB
galactose is required for acceptor, and no substitution is
tolerated at the C-2 position of this residue. Examination of
E297
S180 D316
S199 the models indicated that region LBR-H, and parti-
E298
E317 cularly the presence of an additional bulky Trp (Trp356
G261
in a3GalT), is responsible for the fine specificity. The
A263 H280 A282 Trp356 in a3GalT makes a barrier for any substitution on
Forss-S α3GalT
the Gal ring, whereas in the other enzymes the smaller Ala
or Thr residue together with basic residues 323 RK (Forss-S)
or 328 QL (GTA and GTB) form a pocket accommodating
either Fuc or NHAc group on C2 of the Gal ring. The size
B of the first amino acid of region LBR-H (Trp356 in
a3GalT) seems to distinguish whether the first monosac-

Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010

W300 W287
charide of acceptor could be branched or not.
Tryptophan residues are also involved in defining the
H220
F236 H233 H223
K346 K332

acceptor specificity at longer range, that is, with the

W329
G235
A343
L329

reducing sugar of the di- or trisaccharide. Again, a3GalT,

W222 E315
Q328

I316

GTA iGb3-S

W295 W314

H228 Q247
Y231
K340 K359
W250
W356
T337
G230
K324
R323 W249
H342

I343
Forss-S α3GalT

Fig. 3. (A) Models of interaction between different glycosyltransferases

and the nucleotide-sugars. The protein is represented by its
Connolly surface that has been sliced to display the interior of the
binding site. Color coding represents the electrostatic potential (from
red for positively charged area to blue for negatively charged area).
The Mn2 ‡ cation is represented by a magenta sphere and the nucleotide
sugar by sticks. (B) Models of interaction between different
glycosyltransferases and the acceptor oligosaccharide. The Connolly Fig. 4. Stereoscopic representation of the hydrogen bond network between
surface of the protein that has been color coded according GalNAc and surrounding amino acids in the model of the interaction
to Figure 2A. between human GTA and UDP-GalNAc.

Table II. Analysis of the models: hydrogen bonds between the Gal or GalNAc of the nucleotide sugars and the amino acids of the different
glycosyltransferases

Ligand atom a3GalT GTB iGb3-S GTA Forss-S

O (N-acetyl) Gly267(N) Ala263(N)

O (N-acetyl) Gly268(N)
O-2 (Gal) His280(NE2) His253(NE2)
O-3 Asp225(OD1) Asp211(OD1) Asp198(OD1) Asp211(OD1)
O-3 Asp211(OD2) Asp211(OD2) Asp206(OD2)
O-4 Asp316(OD2) Asp302(OD2) Asp289(OD2) Asp302(OD2) Glu297(OE2)
O-4 O6(Acceptor) Gln296NE1)
O-6 Asp316(OD1) Asp302(OD1) Asp289(OD2) Asp302(OD1) Glu297(OE1)
O-6 Asp289(OD1) Trp295(O)
O-6 Ser199(OG) Ser185(OG) Ser172(OG) Ser185(OG) Ser180(OG)

381
H. Heissigerova et al.

which uses lactose or N-acetyllactosamine as acceptor,
displays a Trp residue stacking with the Glc (or GlcNAc)
ring at the reducing end, thus limiting the possibilities of
substitution or branching for this residue. This selection of
the flat, ribbonlike conformationÐwhich can be adopted
by equatorial±equatorial linked carbohydrate (such as cel-
lulose and chitin, but also lactose and N-acetlyllactosamine)
by pavement of Trp residuesÐhas previously been observed
in polysaccharide-binding protein module. Blood group A
and B transferases, which use either aFuc1-2bGal1-
3GlcNAc (type 1) or aFuc1-2bGal1-4GlcNAc (type 2)
acceptors, present a PG/S motif in the LBR-D region
instead of the AW motif characteristic of a3GalT.
Role of the nine ligand binding regions
The present dissection of the acceptor binding site
allows researchers to clarify the role of each binding
regions in cation, donor, and acceptor binding. The amino
acids that are predicted to directly interact with the substrate
LBR-E, which contains a variable sequence responsible
for the specificity toward Gal and GalNAc.
 The well-conserved region LBR-F mostly ensures specifi-
city for galacto configuration of the acceptor terminal
residue.
 The variable LBR-H region defines the substitution that
can be accepted on this terminal sugar of the acceptor.
This loop, together with the C-terminal peptide LBR-I,
undergoes the conformational change that locks the
nucleotide sugar in the binding site. Those two regions
also make contacts with the pyrophosphate group.
 The variable LBR-D region plays a role in the specificity
for the reducing part of the acceptor.
 Finally, LBR-G only interacts with Fuc or NHAc
moiety of the acceptor in GTA, GTB, and Forss-S.
Discussion

Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010

have been labeled in Figure 5. The specific role of each of
the nine sequence motifs can be summarized as follows.
 The most highly conserved regions, LBR-A and LBR-C,
interact with the uracil and ribose rings and ensure part
of the coordination of the Mn2 ‡ cation.
 Recognition of the sugar moiety of the donor involves
loops LBR-B, -C, -E, and -F. A special role is devoted to
Comparison with the crystal structure of bovine a3GalT
The modeling study started from the crystal structure of the
complex a3GalT/UDP (Boix et al., 2001). Further attempts
to crystallize a3GalT in the presence of UDP-Gal yielded to
the cleavage of the substrate in the active site (Boix et al.,
2002). In one structure the cleaved galactose residue is still
present in the site. After cleavage, the sugar undergoes

Table III. Analysis of the models: hydrogen bonds between the acceptor sugar and the amino acids of the different glycosyltransferases

Ligand atom a3GalT GTB iGb3-S GTA Forss-S

O40 Gln247(NE2) His233(NE2) His220(NE2) His233(NE2) His228(NE2)

0 00 00 00 00
O5 O3 (GlcNAc) O3 (GlcNAc) O3 (Glc) O3 (GlcNAc)
O60 Tyr278(OH) Tyr264(OH) Tyr251(OH) Tyr264(OH) Tyr259(OH)
O60 Glu317(OE1) Glu303(OE1) Glu290(OE1) Glu303(OE1) Glu298(OE1)
O200 Trp250(NE1) His223(NE2)

Table IV. Analysis of the models: amino acids of the different glycosyltransferases that are involved in van der Waals interaction with the acceptor sugar

a3GalT GTB iGb3-S GTA Forss-S

Gln247 His233 His220 His233 His228

Trp249 Ser235 Trp222
Trp250 Phe236 His223 Phe236 Tyr231
Tyr278 Tyr264 Tyr251 Tyr264 Tyr259
Trp314 Trp300 Trp287 Trp300 Trp295
Glu317 Glu303 Glu290 Glu303 Glu298
Arg323
Leu329 Leu329 Lys324
Trp356 Trp329
Lys359 Lys346 Lys332 Lys346 Lys340
Tyr361 His348 Tyr334 His348

382

Fig. 5. Alignments of the nine sequence motifs that are predicted to form the binding site. Residues labeling are as follows: plus sign, interacting with Mn2 ‡ ;
open squares, with uridine and ribose; filled diamonds, with pyrophosphate; open circles, with Gal or GalNAc of donor; and filled squares, with
oligosaccharidic acceptor. The amino acids that play a crucial role in specificity have been boxed.
rotation and translation movements of small amplitude
that allow it to maximize the number of hydrogen bonds
in the site.
Our docking of N-acetyllactosamine in the acceptor site
was also done from the a3GalT/UDP complex. Because the
crystal structure of the enzyme in complex with LacNAc has
been recently published (Boix et al., 2002), this gives a direct
comparison between the model and the crystal structure
and therefore a validation of our modeling approach.
Indeed, the location of the acceptor disaccharides is cor-
rectly predicted, especially all hydrophobic interaction, with
a particular role of the tryptophan residues. Small varia-
tions exist between the predicted and observed hydrogen
bond networks, mainly due to the orientation of the hydro-
xyl group at O-60 .
Comparison with the crystal structures of blood
group A and B transferases
Crystal structures of both blood group A and B transferases
have been very recently elucidated in the native state and as
complexes with UDP and acceptor analog (Patenaude et al.,
2002). Superimposition of the protein backbone (Asp83±
Arg176 and Cys196±Pro345) of the model structure on the
crystal structure yielded a rms of 0.94 A Ê and 0.87 AÊ for
GTA and GTB enzymes, respectively, thus confirming the
quality of the homology modeling. In the crystal structures
complexed with UDP and H-type substrate, two peptide
regions adjacent to the active site are disordered: one loop
(177±195) and the C-terminus region (346±354) therefore
resulting in the open conformation (Form I) of the binding
site cleft. On the opposite, the modeled structures corre-
spond to the Form II conformation, which has been pro-
posed to be induced by the binding of the nucleotide sugar
(Persson et al., 2001; Boix et al., 2002; Ramakrishnan et al.,
2002). The roles of regions LBR-B, LBR-H, and LBR-I in
binding the ligands could then be inferred from the model.
Figure 2 displays a comparison of Form I (crystal) and
Form II (model) of GTA.
The differences are maximum for the nucleotide-sugar
binding site (Figures 2C and 2D): this substrate is comple-
tely buried in Form II. Side chains of amino acids from the
C-terminal region (Lys346, His348, and Arg252) stack
together and join amino acids from the basis of the long
a4 helix (Trp181 and Gln182) to form a lid over the pyr-
ophosphate moiety of the nucleotide sugar (Figure 2E).
These five amino acids show a high degree of conservation
among the family of glycosyltransferases studied. Such clos-
ing of the lid by hydrophobic contacts is also observed in
LgtC structure (Persson et al., 2001) where Pro248 in the
Molecular modeling of a3Gal(NAc)transferases
C-terminal region interacts with the His78±Ile79 cluster.
When ordering these two regions in our model of GTA,
additional contacts are established between the protein and
the nucleotide-sugar: Lys346 (LBR-H) and Arg252 (LBR-I)
of the C-terminal domain make salt bridges to the phos-
phate groups, whereas Val184 (LBR-B) interacts with the
uracil ring. The GalNAc moiety of the nucleotide-sugar
also presents additional contact with the protein. The orien-
Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010
tation of Gal/GalNAc proposed from the present modeling
study varies slightly from the one that has been deduced (also
from modeling) but starting from the Form I crystals struc-
ture, where very few contacts were predicted to occur. In our
model, GalNAc is stabilized by a larger number of hydrogen
bonds (Figure 4), including Ser185, which was disordered in
the crystal. Also, the N-acetyl moiety interacted with
Gly268, one of the two crucial amino acids in term of AB
specificity (see later discussion), therefore explaining why
GalNAc is favored over Gal in the larger site of GTA.
In the acceptor site, the differences between Forms I and
II are less drastic (Figure 2C and D). Nevertheless, ordering
of bulky Lys346 reduces the space available for the fucose
residue. In the present model, this results not only in a
different orientation of the fucose but also for the Gal
moiety. This latter residue, presents a slightly different
orientation than in the GTA or GTB crystals, resulting in
a better stacking with Trp300, as previously observed in the
crystal structure of a3GalT/lactose complex (Boix et al.,
2002).
Comparison with biochemical data on GTA and GTB
The differences in amino acids between GTA and GTB are
limited to four residues: Arg176Gly, Gly235Ser, Leu266-
Met, and Gly268Ala (Yamamoto et al., 1990). Among these
four differences, three are located in regions that we identify
as LBRs. As discussed, Leu266Met and Gly268Ala substi-
tutions (region LBR-E) have a crucial effect on the shape of
the nucleotide binding pocket and directly affect the speci-
ficity for the nucleotide sugar. Indeed, mutagenesis
studies demonstrated that when only one of the two posi-
tions is substituted by its analog counterpart in the other
enzyme, the resulting enzyme displays both A and B activity
(Yamamoto et al., 1996; Seto et al., 1999). The mouse
glycosyltransferase that naturally displays this dual specifi-
city also has only one difference (Met to Gly) with human
BGT (Yamamoto et al., 2001). Our modeling study ratio-
nalizes the concerted action of the two amino acids
involved. As shown in Figure 3A, they form together a
bottleneck that size controls the possibility to accept an N-
acetyl group in a subsite.
383
H. Heissigerova et al.
The substitution Gly235Ser is located in the acceptor site
(region LBR-D) and may be involved in small differences in
acceptor recognition that have been defined by chemical
mapping of the acceptors (see later discussion). Arg176Gly
mutation does not affect the A-specificity but gives an 11-
fold increase in kcat (Seto et al., 1997). This amino acid is
not involved in the substrate binding but is located at the
surface of the enzyme. Analysis of the surface indicates that
it is located about 8 AÊ from LBR-A, which closes the site
above the nucleotide sugar. Depending on the orientation
of the Arg176 side chain, it can participate to a basic cluster
with K123 and K124 of LBR-A, two basic amino acids of A
and B transferases at the surface of the binding site. Mod-
ifying this basic cluster could affect the turnover of the
nucleotide-sugar and the catalysis products. Progress in
the understanding of the catalytic mechanism is needed to
explain fully the role of these residues.
Chemical mapping studies with modified synthetic accep-
tors demonstrated a special role for O-4 of the galactose
residue (Lowary and Hindsgaul, 1993, 1994). Indeed, in the
present models, this hydroxyl group is involved in a strong
hydrogen bond with His233 for both GTA and GTB
enzymes. Mapping studies on the fucose residue indicated
some differences in specificity because only the B enzyme
requires the methyl group at C6-Fuc (Mukherjee et al.,
2000). In our model, this methyl group interacts with
Leu329 in a region (LBR-G) where there is no difference
between A and B transferases. Nevertheless, in our predic-
tion, the methyl group of the fucose is also spatially close to
the region LBR-D, where GTA and GTB differs by a
Gly235Ser substitution.
Conclusion
The elucidation of the catalytic mechanism of glycosyltrans-
ferases remains one of the most challenging problems in
structural glycobiology (Ly et al., 2002). Particularly, the
catalytic event that allows transfer of a monosaccharide
with retention of configuration has not yet been elucidated:
the double displacement reaction via formation of a glycosyl-
enzyme intermediate that has been proposed by analogy
with glycosylhydrolases (Gastinel et al., 2001) has been
revised recently and is now almost abandoned (Boix et al.,
2002; Ly et al., 2002), although no clear alternative mechan-
ism could be proposed. In this context, the aim of the
present study was to clarify the role of conformational
changes as well as the basis of substrate and acceptor bind-
ing in one family of glycosyltransferases and to rationalize
their specificities. The sequence motifs playing a role in
ligand binding have been identified: the sugar donor speci-
ficity (UDP-Gal versus UDP-GalNAc) is due to the relative
size of two crucial amino acids, whereas in the acceptor
site, tryptophan residues play a key role in defining the
fine specificity. Because glycosyltransferases are widely
used for oligosaccharide synthesis in biotechnological
approaches, such a modeling study could provide some
structural basis for rational engineering of specificities and
design of new transferase activities.
The present article has been limited to enzymes belonging
to one family of homologous enzymes; it may nevertheless
be extended to other glycosyltransferases. Among the
384
crystal structures that have been determined recently, all
of the enzymes for which Mn2 ‡ is required for activity
share the same fold, despite a lack of sequence similarities.
In such cases, fold recognition methods are a powerful tool
for identifying family of enzymes that can share the same
fold (Breton et al., 2002). Combination of fold recognition
study and molecular modeling may help in predicting the
acceptor specificities of putative glycosyltransferase
sequences from newly determined genomes. This approach
may be helpful in the emerging area of glycogenomics.
Materials and methods
Multiple alignment of amino acid sequences was performed
with the ClustalX program (Thompson et al., 1994) using
the following sequences: bovine a3GalT (GenBank acces-
sion number J04989), canine Forss-S (U66140), human
GTA (J05175) and GTB (AF134414), and rat iGb3-S
Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010
(AF248543). The homology modeling COMPOSER
program (Blundell et al., 1988) of the Sybyl software
(SYBYL, St. Louis, MO) was used to build the different
enzyme models. Conserved a-helices and b-strands (struc-
turally conserved regions in COMPOSER) were built from
the highest-resolution structure of a3GalT (Boix et al.,
2001; code 1K4V in the Protein Data Bank; Berman et al.,
2000). Loops were then modeled by using the most similar
fragments in a library containing all crystal structures of
glycosyltransferases available. Resulting models were then
screened using the PROCHECK program (Laskowski et al.,
1993), and backbone conformations lying outside the
allowed regions of Ramachandran map were further opti-
mized. Hydrogen was added on all atoms, and partial
atomic charges were derived using the Pullman procedure.
UDP-Gal/UDP-GalNAc was docked into the binding site
in a conformation and location similar to what has been
observed for UDP-Gal complexed with a3GalT (Boix et al.,
2001) and for UDP-2F-Gal complexed with LgtC (Persson
et al., 2001). A Mn2 ‡ ion was located between the pyro-
phosphate group and aspartate groups of the DXD motif of
protein. Atom types and energy parameters available for
carbohydrates (Imberty et al., 1999) were used together with
parameters developed for the sugar±pyrophosphate linkage
(Petrova et al., 1999). For disaccharide acceptors, the con-
formation at the glycosidic linkage was selected according
to crystal structures, when available (PeÂrez et al., 2000), and
to previously calculated energy maps (Imberty et al., 1995).
In all cases, the monosaccharides on the nonreducing side
have been located in the binding site as observed for lactose
and LacNAc acceptor in complex with a3GalT (Boix
et al., 2002).
Four models were therefore generated: GTA/UDP-Gal-
NAc/aFuc1-2bGal1-4bGlcNAc/Mn2 ‡ , GTB/UDP-Gal/
aFuc1-2bGal1-4bGlcNAc/Mn2 ‡ , Forss-S/UDP-GalNAc/
bGalNAc1-3aGal/Mn2 ‡ , and iGb3-S/UDP-Gal/bGal1-
4bGlc/Mn2 ‡ . The complex a3GalT/UDP-Gal/bGal1-
4bGlcNAc/Mn2 ‡ was also generated for comparison. In
all complexes, several cycles of energy minimization were
performed to optimize the geometry of all ligands and also
of protein side chains in the binding site and its vicinity.
Energy calculations were performed using the TRIPOS

force field (Clark et al., 1989) in the Sybyl package with
addition of energy parameters developed for carbohydrates
(Imberty et al., 1999).
Validation of the models was performed by superimposi-
tion with the crystal structure of GTA and GTB enzymes
complexed with UDP and disaccharidic acceptor analog
(code 1LZI and 1LZJ; Patenaude et al., 2002).
Acknowledgments
H.H. is supported by a grant from the French Ministere des
Affaires Etrangeres. This research was partially supported
by the Assocation pour la Recherche contre le Cancer.
Abbreviations
GT, glycosyltransferase; GTA, blood group A transferase;
GTB, blood group B transferase; Forss-S, Forssman
synthase; iGb3, isogloboside 3; iGb3-S, isogloboside 3
synthase; LBR, ligand binding region; LgtC, a bacterial
a4-galactosyltransferase.
References
Berman, H.M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T.N.,
Weissig, H., Shindyalov, I.N., and Bourne, P.E. (2000) The protein
data bank. Nucleic Acids Res., 28, 235±242.
Blundell, T., Carney, D., Gardner, S., Hayes, F., Howlin, B., Hubbard, T.,
Overington, J., Singh, D.A., Sibanda, B.L., and Sutcliffe, M. (1988)
18th Sir Hans Krebs lecture. Knowledge-based protein modelling and
design. Eur. J. Biochem., 172, 513±520.
Boix, E., Swaminathan, G.J., Zhang, Y., Natesh, R., Brew, K., and
Acharya, K.R. (2001) Structure of UDP complex of UDP-galactose:
b-galactoside-a-1,3-galactosyltransferase at 1.53A resolution reveals a
conformational change in the catalytically important C-terminus.
J. Biol. Chem., 276, 48608±48614.
Boix, E., Zhang, Y., Swaminathan, G.J., Brew, K., and Acharya, K.R.
(2002) Structural basis of ordered binding of donor and acceptor
substrates to the retaining glycosyltransferase: alpha-1,3 galactosyl-
transferase. J. Biol. Chem., 277, 28310±28318.
Breton, C., Heissigerova, H., Jeanneau, C., Moravcova, J., and Imberty, A.
(2002) Comparative aspects of glycosyltransferases. In Drickamer, K.
and Dell, A. (eds.), Glycogenomics: the impact of genomics and
informatics in glycobiology. Portland Press, London, pp. 23±32.
Clark, M., Cramer, R.D.I., and van den Opdenbosch, N. (1989)
Validation of the general purpose Tripos 5.2 force field. J. Comput.
Chem., 10, 982±1012.
Galili, U. and Swanson, K. (1991) Gene sequences suggest inactivation of
alpha-1,3-galactosyltransferase in catarrhines after the divergence of
apes from monkeys. Proc. Natl. Acad. Sci. USA, 88, 7401±7404.
Gastinel, L.N., Bignon, C., Misra, A.K., Hindsgaul, O., Shaper, J.H., and
Joziasse, D.H. (2001) Bovine a1,3-galactosyltransferase catalytic
domain structure and its relationship with ABO histo-blood group
and glycosphingolipid glycosyltransferases. EMBO J., 20, 638±649.
Gibbons, B.J., Roach, P.J., and Hurley, T.D. (2002) Crystal structure of
the autocatalytic initiator of glycogen biosynthesis, glycogenin. J. Mol.
Biol., 319, 463±477.
Haslam, D.B. and Baenziger, J.U. (1996) Expression cloning of Forssman
glycolipid synthetase: a novel member of the histo-blood group ABO
gene family. Proc. Natl. Acad. Sci. USA, 93, 10697±106702.
Imberty, A., Mikros, E., Koca, J., Mollicone, R., Oriol, R., and Perez, S.
(1995) Computer simulation of histo-blood group oligosaccharides.
Energy maps of all constituting disaccharides and potential energy
surfaces of 14 ABH and Lewis carbohydrate antigens. Glycoconj. J.,
12, 331±349.
Molecular modeling of a3Gal(NAc)transferases
Imberty, A., Bettler, E., Karababa, M., Mazeau, K., Petrova, P., and
Perez, S. (1999) Building sugars: The sweet part of structural biology.
In Vijayan, M., Yathindra, N., and Kolaskar, A.S. (eds.), Perspectives
in structural biology. Indian Academy of Sciences and Universities
Press, Hyderabad, pp. 392±409.
Keusch, J.J., Manzella, S.M., Nyame, K.A., Cummings, R.D., and
Baenziger, J.U. (2000) Expression cloning of a new member of the
ABO blood group glycosyltransferases, iGb3 synthase, that directs
the synthesis of isoglobo-glycosphingolipids. J. Biol. Chem., 275,
25308±25314.
Laskowski, R.A., MacArthur, M.W., Moss, D.S., and Thornton, J.M.
(1993) PROCHECK: a program to check the stereochemical quality of
protein structures. J. Appl. Crystallogr., 26, 283±291.
Lowary, T.L. and Hindsgaul, O. (1993) Recognition of synthetic deoxy
and deoxyfluoro analogs of the acceptor alpha-L-Fuc p-(1 ! 2)-beta-
D-Gal p-OR by the blood-group A and B gene-specified glycosyl-
transferases. Carbohydr. Res., 249, 163±195.
Lowary, T.L. and Hindsgaul, O. (1994) Recognition of synthetic
O-methyl, epimeric, and amino analogues of the acceptor alpha-L-
Fuc p-(1 ! 2)-beta-D-Gal p-OR by the blood-group A and B gene-
specified glycosyltransferases. Carbohydr. Res., 251, 33±67.
Ly, H.D., Lougheed, B., Wakarchuk, W.W., and Withers, S.G. (2002)
Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010
Mechanistic studies of a retaining alpha-galactosyltransferase from
Neisseria meningitidis. Biochemistry, 41, 5075±5085.
Mukherjee, A., Palcic, M.M., and Hindsgaul, O. (2000) Synthesis and
enzymatic evaluation of modified acceptors of recombinant blood
group A and B. Carbohydr. Res., 326, 1±21.
Oriol, R. (1987) Tissular expression of ABH and Lewis antigens in
humans and animals: expected value of different animal models in the
study of ABO-incompatible organ transplants. Transplant. Proc., 19,
4416±4420.
Patenaude, S.I., Seto, N.O., Borisova, S.N., Szpacenko, A., Marcus, S.L.,
Palcic, M.M., and Evans, S.V. (2002) The structural basis for specificity
in human ABO(H) blood group biosynthesis. Nature Struct. Biol., 9,
685±690.
Paulson, J.C. and Colley, K.J. (1989) Glycosyltransferases. Structure,
localization, and control of cell type-specific glycosylation. J. Biol.
Chem., 264, 17615±17618.
Perez, S., Gautier, C., and Imberty, A. (2000) Oligosaccharide conforma-
tions by diffraction methods. In Ernst, B., Hart, G., and Sinay, P.
(eds.), Oligosaccharides in chemistry and biology: a comprehensive
handbook. Wiley/VCH, Weinheim, pp. 969±1001.
Persson, K., Ly, H.D., Dieckelmann, M., Wakarchuk, W.W.,
Withers, S.G., and Strynadka, N.C.J. (2001) Crystal structure of
the retaining galactosyltransferase LgtC from Neisseria meningitidis
in complex with donor and acceptor sugar analogs. Nature Struct.
Biol., 8, 166±175.
Petrova, P., Koca, J., and Imberty, A. (1999) Potential energy
hypersurfaces of nucleotide-sugars: Ab initio calculations, force-field
parametrization, and exploration of the flexibility. J. Am. Chem. Soc.,
121, 5535±5547.
Ramakrishnan, B. and Qasba, P.K. (2001) Crystal structure of lactose
synthase reveals a large conformational change in its catalytic
component, the beta1,4-galactosyltransferase-I. J. Mol. Biol., 310,
205±218.
Ramakrishnan, B., Balaji, P.V., and Qasba, P.K. (2002) Crystal structure
of beta-1,4-galactosyltransferase complex with UDP-Gal reveals an
oligosaccharide binding site. J. Mol. Biol., 318, 491±502.
Rini, J.M. (1995) Lectin structure. Annu. Rev. Biophys. Biomol. Struct., 24,
551±577.
Seto, N.O., Palcic, M.M., Compston, C.A., Li, H., Bundle, D.R., and
Narang, S.A. (1997) Sequential interchange of four amino acids from
blood group B to blood group A glycosyltransferase boosts catalytic
activity and progressively modifies substrate recognition in human
recombinant enzymes. J. Biol. Chem., 272, 14133±14138.
Seto, N.O., Compston, C.A., Evans, S.V., Bundle, D.R., Narang, S.A.,
and Palcic, M.M. (1999) Donor substrate specificity of recombinant
human blood group A, B and hybrid A/B glycosyltransferases
expressed in Escherichia coli. Eur. J. Biochem., 259, 770±775.
Thompson, J.D., Gibson, T.J., Plewniak, F., Jeanmougin, F., and
Higgins, D.G. (1994) The CLUSTAL_X windows interface: flexible
385
H. Heissigerova et al.

strategies for multiple sequence alignment aided by quality analysis Yamamoto, F. and McNeill, P.D. (1996) Amino acid residue at codon 268
tools. Nucleic Acids Res., 22, 4876±4882. determines both activity and nucleotide-sugar donor substrate
È nligil, U.M. and Rini, J.M. (2000) Glycosyltransferase structure and
U specificity of human histo-blood group A and B transferases. In vitro
mechanism. Curr. Opin. Struct. Biol., 10, 510±517. mutagenesis study. J. Biol. Chem., 271, 10515±10520.
È nligil, U.M., Zhou, S., Yuwaraj, S., Sarkar, M., Schachter, H., and
U Yamamoto, F., Clausen, H., White, T., Marken, J., and Hakomori, S.
Rini, J.M. (2000) X-ray crystal structure of rabbit N-acetylglucosami- (1990) Molecular genetic basis of the histo-blood group ABO system.
nyltransferase I: catalytic mechanism and a new protein superfamily. Nature, 345, 229±233.
EMBO J., 19, 5269±5280. Yamamoto, M., Lin, X.H., Kominato, Y., Hata, Y., Noda, R., Saitou, N.,
Varki, A. (1993) Biological roles of oligosaccharides: all of the theories are and Yamamoto, F. (2001) Murine equivalent of the human histo-
correct. Glycobiology, 3, 97±130. blood group ABO gene is a cis-AB gene and encodes a glycosyl-
Vyas, N. (1991) Atomic features of protein±carbohydrate interactions. transferase with both A and B transferase activity. J. Biol. Chem., 276,
Curr. Opin. Struct. Biol., 1, 732±740. 13701±13708.

Downloaded from glycob.oxfordjournals.org by guest on November 6, 2010

386