Mimics of ganglioside GM1 as cholera toxin ligands: replacement of the GalNAc residue.

Article abstract of DOI:10.1039/b210503a

Two new cholera toxin (CT) ligands (4 and 5) are described. The new ligands were designed starting from the known GM1 mimics 2 and 3 by replacement of their GalNAc residue with the C4 isomer GlcNAc. As predicted by molecular modelling, the conformational

Full text of DOI:10.1039/b210503a

View Article Online / Journal Homepage / Table of Contents for this issue
Mimics of ganglioside GM1 as cholera toxin ligands:
replacement of the GalNAc residue†
Anna Bernardi,*a Daniela Arosio,a Leonardo Manzoni,b Diego Monti,b Helena Posteri,a
Donatella Potenza,a Silvia Maria and Jesús Jiménez-Barberoc
a Universita’ di Milano – Dipartimento di Chimica Organica e Industriale e Centro di Eccellenza
CISI, via Venezian 21, 20133 Milano, Italy
b CNR-Istituto di Scienze e Tecnologie Molecolari (ISTM), Via Golgi 19, 20133 Milano, Italy
c Dept. de Estructura y Función de Proteínas, Centro de Investigaciones Biológicas CSIC,
Velázquez 144, 28006 Madrid, Spain
Received 28th October 2002, Accepted 13th December 2002
First published as an Advance Article on the web 5th February 2003
Two new cholera toxin (CT) ligands (4 and 5) are described. The new ligands were designed starting from the
known GM1 mimics 2 and 3 by replacement of their GalNAc residue with the C4 isomer GlcNAc. As predicted
by molecular modelling, the conformational properties of the equivalent pairs 2–4 and 3–5 are very similar and
their affinity for CT is of the same order of magnitude. NMR experiments have also proved that 5 occupies the
GM1-binding site of the toxin and have revealed its bound conformation.
Introduction
Carbohydrate–protein complexes are formed in the initial steps
of a large number of physiological and pathological processes,
which range from cell–pathogen interaction, to cell–cell recog-
nition, to tumor metastasis, etc.1 Interference with these
recognition events could be used to modulate or alter signal
transmission, or to prevent the onset of diseases. Therefore the
synthesis of functional sugar mimics capable of antagonising
oligosaccharides at the protein receptor level has attracted a
great deal of attention as a way to develop drugs with good
stability and synthetic availability.2
Our group is currently involved in an effort to devise a
rational approach to the design and synthesis of glycomimetics
by scaffold replacement.3 In brief, we have proposed the use of
conformationally stable cyclic diols4 to replace non-pharmaco-
phoric parts of bioactive oligosaccharides, while preserving
the correct pharmacophore orientation. Computational tools
can be used to predict the three-dimensional structures of the
mimics and compare them with the structure of the natural
ligand. When supported by adequate experimental work,
molecular modelling also makes it possible to obtain at least
qualitative predictions on the binding mode of new substrates,
and to design further simplifications of the glycomimetic struc-
tures aimed at reducing the carbohydrate-likeness of the mimics
and increasing their drug-like properties.
This approach has been validated5 using as a model system
the recognition pair composed of the head-group of ganglio-
side GM1 1 (Chart 1) and the two bacterial enterotoxins
(cholera toxin (CT) and heat-labile toxin of E. coli (LT)) that
use it as their target on cell membranes. GM1 interacts with CT
and LT using the Gal and NeuAc residues at the oligosacchar-
ide non-reducing end, as has been shown by biochemical6 and
structural7 studies. Not long ago, we described the rational
design and the synthesis of the pseudo-oligosaccharide 25,8
(Chart 1), a functional and structural mimic of ganglioside
GM1 based on the use of a conformationally restricted cyclo-
hexanediol (CHD)3 to replace the reducing end of the ganglio-
side head-group, which is not involved in toxin binding. Mimic
† Electronic supplementary information (ESI) available: synthetic
details, product characterisations and full NOE contact list. See http://
www.rsc.org/suppdata/ob/b2/b210503a/
Chart 1
T h i s j o u r n a l i s © T h e R o y a l S o c i e t y o f C h e m i s t r y 2 0 0 3
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
785

View Article Online
2 was found to be as active as GM1 in binding to CT.5 More
recently, we reported three second-generation mimics9 that
were conceived by replacing the sialic acid (NeuAc) moiety of 2
with simple α-hydroxyacids. The most active compound of this
series was the (R)-lactic acid-containing ligand 3 (Chart 1).
In an effort to further simplify the synthetic complexity of
the structure we also examined the pseudo-tetrasaccharide 4
(Chart 1), which is derived from 2 by replacing the GalNAc
residue with an N-acetylglucosamine (GlcNAc). Modification
of the hexosamine was considered because it would simplify
the synthesis of the artificial receptors (GalNAc is actually
made from GlcNAc), while not negatively interfering with the
formation of the toxin complex, since the GalNAc of GM1
interacts with the protein only via the N-acetyl group. Further-
more, the 4-hydroxy group of GalNAc in the experimental
CT–GM1 complex7 and in the computational LT–GM110 and
LT–25 models is located outside the protein binding pocket
and fully exposed to the solvent. Assuming that 4 will bind to
CT and/or LT with the same general mode of GM1 and 2,
it appeared likely that inversion of configuration at C4 of
the hexosamine could allow new H-bond contacts to be
formed between the toxin and the substrate. In principle, these
may or may not compensate for the loss in complex solvation
due to the burying of the hydroxy group in the binding cavity.
To evaluate the potential of 4 as a GM1 antagonist, computer
simulations were used to model the free ligand 4 and its LT
complex, followed by comparison with the corresponding
structures of 2.11
The full synthetic sequence and product characterisations are
reported in the Supplementary Information.† Briefly, the known
acceptors 68 and 79 (Scheme 1) were glycosylated at the axial
hydroxy group using the trichloroacetimidate 8 as a Galβ-1,3-
GlcNAc donor, and TfOH or TMSOTf as the promoter. The
adducts were routinely deprotected to yield the free ligands.
NMR studies of the free ligands 4 and 5
Two-dimensional 400-MHz NOESY and ROESY spectra of 4
and 5 were obtained in D2O. The relevant inter-residual con-
tacts are collected in Table 1, and compared with the data
reported in the literature for 1,12 25 and 313 in D2O. Complete
spectral assignments and full contact lists are collected in the
Supplementary Information.†
The NMR data obtained for 4 (Table 1, column 4) show the
characteristic set of NOE cross-peaks (Fig. 1b) which is also
observed for 112 and 25 (Table 1, columns 1 and 2; Fig. 1a). The
data are consistent with one major conformation of the oligo-
saccharide framework featuring the Galβ-1,3-GlcNAcβ-1,4-
The following predictions were made:11
1. Changing the hexosamine in the pseudo GM1 structure
should not modify the overall molecular conformation, and, in
particular, should not alter the relative position of the Gal and
NeuAc binding determinants.
2. Compared to 2, the new compound 4 will indeed insert one
more hydroxy group within the protein binding site. Molecular
dynamics simulations suggested that, in turn, this may trigger a
series of rearrangements and reorientations of side chains and
crystallographic water molecules in the toxin, leading to new H-
bond contacts which may result in enhanced affinity of the new
inhibitor.
3. Free Energy Perturbation (FEP) calculations performed by
carrying out the 2
4 mutation in solution and in the protein
complex suggested that the GlcNAc mimic 4 should be a
stronger binder than its parent compound 2.
We now report the experimental data relative to the GlcNAc-
containing mimics 4 and 5, the latter being the GlcNAc-
containing version of the second generation binder 3. The
ligands have been synthesised and their solution structure has
been studied by NMR spectroscopy, confirming the compu-
tational predictions.11 The complex formed by 5 with the recog-
nition element of CT, CTB5, has been studied by TR-NOE
NMR, and the bound conformation of the ligand has been
determined. Competition experiments, carried out by adding
oGM1 to a solution of the CTB5–5 complex, could also be
monitored by NMR, and have confirmed that the GlcNAc-
containing ligand and the ganglioside are indeed competing
for the same binding site. Affinity constants for CTB5 have been
obtained for 4 and 5 by fluorescence titrations, and found to be
similar to those of the parent compounds 2 and 3, respectiv-
ely. Thus, the computational predictions on the GlcNAc
binders were revealed to be qualitatively correct, but the FEP
calculations were quantitatively incorrect.
Results
Fig. 1 a. Observed NOE contacts in the NMR spectra of 2. b.
Observed NOE contacts in the NMR spectra of 4. c. Overlap of the
calculated structures of 2 (light blue) and 4 (white). For the Galβ-1,3-
GalNAcβ-1,4-CHD pseudo-trisaccharide the glycosidic angles ꢀ and ψ
are defined as H1–C1–O–Cn and C1–O1–Cn–Hn, respectively. For the
Synthesis of the ligands
The GlcNAc-containing ligands 4 and 5 were prepared follow-
ing the synthetic scheme adopted for the parent compounds 28
and 39 and reported in Scheme 1.
NeuAcα-2,3-CHD fragment ꢀ and ψ are defined as (O᎐)C1–C2–O–Cn
᎐
and C2–O–Cn–Hn, respectively.
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
786

View Article Online
Table 1 Relevant NOE contacts in ligands 1–5a
Residue
1b
2c
3d
4e
5e free state
5e CTB5 complex
NeuAc [N]
N-3ax
N8
GII-3 (s)
CHD3 (s/m)
—
CHD3 (s/m)
—
—
g
g
GN1 (m)f
CHD4 (w)
(R)-lactic acid
H
—
—
CHD3 (s)
CHD4 (s)
CHD2eq (m)
GN1 (w)
—
CHD3 (s)
CHD4 (s)
CHD2eq (m/w)
GN1 (w)
CHD3 (m)
—
CHD2eq (s)
—
Me
CHD3 (s)
GN1 (m)
CHD3 (s)
GN1 (m)
CHD3 (m)
—
GalNAc [GN]
GlcNAc [GN]
Gal [G]
GN1
GN1
G1
GII-4 (s)
CHD4 (s)
CHD4 (s)
—
—
—
—
—
—
CHD4 (s)
CHD4 (s)
CHD4 (s)
GN3 (s)
GN3 (s)f
GN2 (w)
GN4 (w)f
GN3 (s)
GN2 (w)
GN4 (w)
GN3 (s)
GN2 (w)
GN4 (w)
GN3 (s)
GN3 (s)
GN2 (w)
h
a ROESY cross-peaks in D2O solution. In parentheses the observed intensities; s: strong, m: medium, w: weak. b From ref. 12. c From ref. 5. d From
ref. 13. e This work. f Data from DMSO-d6 solution (ref. 31); not seen in D2O, due to signal overlap. g N8–GN1 not measurable, due to signal overlap.
h GN2 overlaps with GN3.
Scheme 1 Synthesis of 4 and 5.
CHD pseudo-trisaccharide and the NeuAcα-2,3-CHD frag-
ment exclusively in the syn, syn (ꢀ, ψ 55Њ, 0Њ) and the anti, syn
(ꢀ, ψ Ϫ170Њ, Ϫ30Њ) conformations, respectively (for the defin-
ition of the ꢀ, ψ inter-glycosidic torsion angles, see the caption
of Fig. 1). A marked conformational restriction is a distinctive
feature of the oGM1 pentasaccharide and its mimics. In gener-
al, all experiments find that the ganglioside head-group can be
broken down into two areas: a so-called “core trisaccharide”,
the GalNAcβ1-4(Neu5Acα2-3)Gal trisaccharide which is often
described as “rigid”, and more mobile regions corresponding to
the external sugars at both ends of the head-group. However,
also the mobility of the Galβ1-3GalNAc fragment at the non
reducing end is actually limited to ample oscillations around
well-defined average glycosidic torsion angles. Dynamic simu-
lations of the oligosaccharide suggest that the NMR data show-
ing two weak and equivalent NOE contacts between Gal-H1
and GalNAc-H2 and H4 (Table 1, column 1; see Fig. 1 for the
equivalent contacts in 2) are accounted for by a model
whereby the Galβ1–3GalNAc bond populates a single, broad
energy minimum, rather than two individual conformations.14
The energy well is centered at ꢀ, ψ 55Њ, Ϫ5Њ, which is also the
conformation observed in the CTB5–oGM1 X-ray structure
(ꢀ, ψ 55Њ 10Њ, 0Њ 10Њ).7 Therefore it can safely be asserted
that oGM1 exists largely as one main conformer. The same
behaviour has been observed for 2, which assumes the con-
formation depicted in Fig. 1c (light blue structure).5 In 4, the
flexibility of the Galβ-1,3-GlcNAcβ-1,4-CHD fragment is
further reduced by the H-bond interaction of GlcNAc-OH4
with the Gal oxygen, which constrains the amplitude of the
oscillations closer to the observed bound conformation. In fact,
the conformational analysis of 411 led to 18 conformers within
10 kJ molϪ1 from the global minimum that all shared a common
conformation of the core fragment. This conformation is com-
pletely superimposable with the NMR solution structure of 2,
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
787

View Article Online
Fig. 2 Experimentally observed side-chain conformations in the (R)-lactic acid-containing ligand 5. The intensity of the observed NOE cross-peaks
is indicated: w:weak; s:strong; m/w:medium/weak. a. Free state. The mutually exclusive NOE contacts H/H4 and H/H2eq reveal the existence of
two different conformations in water solution. The conformation depicted is the prevalent one. b. Bound state. The single conformation selected by
the toxin is not the most abundant one in the free state.
Fig. 3 NMR spectra of 5. a) NOESY spectrum of 5 in the free state. b) TR-NOESY spectrum of 5 in the CTB5 complex. c)TR-NOESY spectrum
of the CTB5–5 complex after adding oGM1. Blue cross-peaks are due to the free state form, red cross-peaks are TR-NOE signals arising from the
bound state of 5.
which also corresponds to the bound structure observed in the
CT–GM1 complex (Fig. 1c). The experimental data that are
now available for 4 confirm the computational prediction by
showing exclusively the expected pattern of NOE contacts
(Fig. 1b).
For the (R)-lactic acid-containing ligand 5 the NMR data
(Table 1, column 5, free state) are once more consistent with a
syn, syn disposition of the Galβ1-3GlcNAcβ-1,4-CHD pseudo-
trisaccharide, but show a substantial flexibility of the (R)-lactic
acid side-chain. At least two conformations are represented
with χ 60Њ (major, Fig. 2a) and χ 180Њ (minor), where χ is the
NOE contacts from H to CHD4 and CHD2eq (Fig. 3a), which
should arise from the χ 60Њ and χ 180Њ conformations (Fig. 2),
respectively. The same pattern of NOE contacts was also
observed for the GalNAc derivative 313 (Table 1, column 3).
Thus, the NMR experiments for the unbound ligands suggest
that the GalNAc-containing and the GlcNAc-containing mole-
cules are indeed adopting identical conformations, regardless
of the nature of the hexosamine.
NMR studies of the CT–5 complex
improper dihedral angle H3–C3–C –C(᎐O) (see Fig. 2) defining
the position of the carboxylate relative to the cyclohexanediol
ring. This is inferred by the presence of two mutually exclusive
Information about the conformation of complexed ligands
can be derived from transferred nuclear Overhauser effect
(TR-NOE) studies,15 provided that the exchange between the
᎐

O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
788

View Article Online
Fig. 4 TR-NOESY spectrum of the CTB5–5 complex after addition of the GM1 oligosaccharide 1 (oGM1). In-phase (red) TR-NOESY cross-
peaks belonging to the H3ax of the NeuAc residue in oGM1 (N3ax) show that oGM1 is bound to CT.
complexed and uncomplexed states is sufficiently fast.16 When a
ligand interacts with a macromolecule, the cross-relaxation
rates of its bound states become opposite in sign to those of the
free ligand and generate negative NOEs. So, in the TR-NOESY
experiment the cross-peaks which are due to bound species are
in phase with the diagonal signals, whereas unbound species
give rise to cross-peaks which are out of phase. The bound form
in-phase signals can be exploited to assess the conformational
properties of the bound state. The TR-NOE technique allows
the study in solution of how a ligand fits into a protein binding
site. The phase behaviour of NOESY cross-peaks can be used
to analyze the affinity of individual ligands within mixtures of
compounds,17 and can also be exploited for competition
experiments.18 Our group has recently used it to study the com-
plexes formed by CTB5 and a group of GM1 mimics, including
3.13
H/CHD2eq cross-peaks typical of the free state of 5 (CHD4 s,
CHD2eq m/w) is also restored. Thus, it clearly appears that 5 is
no longer bound to the toxin. Furthermore, the NeuAc-H3eq/
Gal-H3 NOE contact typical of GM1 (see Table 1) is observed
(Fig. 4) as an in-phase (red) cross-peak in the same spectrum,
proving that GM1 has indeed displaced 5 from the toxin bind-
ing site. This experiment clearly establishes that oGM1 and 5
are competing for the same binding site in CTB5, as expected
on the basis of the computational results.
Binding studies
The interaction of cholera toxin B5 pentamer (CTB5) with Gal-
containing ligands can be studied using the intrinsic fluor-
escence of the Trp88 residue in the toxin binding site.21 Binding
of 1–3 to CTB5 induces bathochromic shifts in the emission
spectrum whose extent depends on the ligand.9 oGM1 binding
is also accompanied by an increase in the intensity of fluor-
escence emission, whereas a variable decrease in fluorescence is
seen upon binding of 2 (small decrease) and 3 (larger decrease).
Surprisingly, when 0.5 µM CTB5 was titrated with the GlcNAc-
containing ligands 4 and 5 no bathochromic shift occurred.
However, a small decrease in fluorescence was observed with 4
and a larger one with 5 (see Supplementary Information,
Figs. S1 and S2).† The binding isotherms of GalNAc-containing
and GlcNAc-containing ligands were therefore compared by
analysing the intensity data. This requires a relatively high
CTB5 concentration (0.5 µM). However, wavelength titrations
performed for 1 and 2 at lower toxin concentrations (0.1 µM)
gave comparable results (See Supplementary Information,
Fig. S3).†
The affinities of 1,19 25 and 4 (see below) for CTB5 are on the
upper limit of what can be revealed by the TR-NOE tech-
nique,20 but the study of the CTB5–5 complex yielded much
important information. No difference is observed between the
free-state and bound-state conformations of the Gal-GlcNAc
and GlcNAc-CHD fragments of 5, which are always found to
be syn (ꢀ, ψ 55Њ, 0Њ; see Table 1, column 5, CTB5 complex). In
contrast, a clearly different set of cross-peaks was observed for
the NOEs involving the hydroxyacid protons HL and CH3 (see
Table 1) upon passing from the free to the bound state. In par-
ticular, the cross-peaks observed for HL in both free and bound
states are shown in Fig. 3a and 3b, respectively. The out-of-
phase NOESY cross-peaks relative to the free state of 5 are
depicted in blue, whereas the in-phase TR-NOESY peaks aris-
ing from its bound state are shown in red, like the diagonal
signals. The free state spectrum (Fig. 3a) shows the set of NOE
contacts (blue cross-peaks) discussed above, and sketched in
Fig. 2a (CHD3 and CHD4 strong, CHD2eq m/w). The TR-
NOESY spectrum obtained upon addition of CTB5 to the
solution is shown in Fig. 3b (see also Table 1). The strong HL/
CHD-H4 cross-peak found in the free state NOESY of 5
(Fig. 3a) almost disappears in the TR-NOESY spectrum of the
CTB5–5 complex, and the small residual signal is still out-of-
phase (blue), therefore it still belongs to the free state of the
ligand. The in-phase (red) cross-peaks belonging to the bound
state of 5 show a close H/CHD-H2eq contact, much stronger
than in the free state. This implies a conformational transi-
tion of the hydroxyacid side-chain, with the χ 180Њ conformer
(Fig. 2b) being selected in the bound state preferentially with
respect to the most abundant χ 60Њ conformer (Fig. 2a). The
behaviour of 5 described here closely parallels the observations
already reported for 3,13 and is strongly suggestive of a common
binding mode for the two ligands in the toxin.
Binding of 1 to CTB is known to occur cooperatively.19 The
observed concentration of the ligand at 50% saturation (IC50) in
our titration (0.5 µM CTB5, see Fig. 5) is 1.5 µM. Under the
same conditions, the intrinsic association constant, determined
by calorimetric titration, is 1.05 × 106 MϪ1.19 The first gener-
ation mimic 2 was also found to bind cooperatively and with
comparable affinity (IC50 1.2 µM, see Fig. 5).9 Binding of the
Finally, a competition experiment was performed by adding
oGM1 to the CTB5–5 complex (Fig. 3c). oGM1 has a much
higher affinity for CT than 5 (see below), therefore in a competi-
tion experiment it is expected to bind selectively to the toxin
and displace 5 from its binding site. Indeed, the cross-peaks of 5
which are in phase (red) in the TR-NOESY spectrum of its CT
complex (e.g. H/CHD3 and H/CHD2eq in Fig. 3b) become
out-of-phase (blue) upon addition of oGM1 to the mixture
(Fig. 3c), and the relative intensity of the H/CHD4 and
Fig. 5 Fluorescence intensity titrations of CTB5 (0.5 µM) with
o-GM1 1 (black square, dotted line); ps-GM1 2 (empty circles, solid
line); the GlcNAc-containing analogue 4 (black circles, broken line).
The absolute values of the variation of fluorescence intensity emission
at 350 nm (|δI|) have been normalised to 10, and plotted against the
total concentration of the ligands (nM).
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
789

View Article Online
GlcNAc analogue 4 fits a simple 1 : 1 isotherm and doesn’t
show any cooperativity (Fig. 5), the IC50 is 1.3 µM and the
dissociation constant, determined by nonlinear regression
analysis, is 1.8 µM. Although quantitative comparison is com-
plicated by the different cooperativity behaviour, the relative
binding affinities of 1, 2, and 4 toward CTB5 can be appreci-
ated by comparing the corresponding binding isotherms,
collected in Fig. 5 and showing that the three ligands clearly
appear to be of comparable potency.
The two lactic acid-containing compounds 3 and 5, both
binding CTB5 without any cooperativity effect, allow an easier
estimation of relative affinity. The binding isotherms obtained
in a titration of 0.5 µM CTB5 are reported in Fig. 6. The dis-
sociation constants calculated by non-linear regression analysis
are 190 and 450 µM for 3 and 5, respectively.
with the nature of the hydroxyacid side-chain,23 but in all the
analogues we have studied so far cooperativity is lost.9,23
Much less is known about the role played by the hexosamine.
In the X-ray structure of the CTB5–oGM1 complex the Gal-
NAc residue does not exhibit any directed or mediated H-bond
interaction with the protein, but the acetamide methyl group
makes a van der Waals contact with the Cβ of His13.7 There is
no natural glycolipid which contains a Galβ1,3-GlcNAcβ-1,4-
(NeuAcα-2,3-)Gal fragment, and, to the best of our knowledge,
no binding data are available for this oligosaccharide or for
larger entities that contain it. In a previous paper,11 we evalu-
ated the potential of 4, a GlcNAc analogue of psGM1 2, to
behave as a GM1 antagonist using computer simulations to
model the free ligand 4 and its LT complex. On the basis of this
computational work it was expected that exchanging GalNAc
with GlcNAc would yield a new molecule with the same over-
all molecular conformation of 2, capable of fitting the GM1-
binding site of CT and displaying a higher affinity than
psGM1.
The experimental results described above concerning 4 and
its lactic acid analogue 5 have now shown that the prediction
was qualitatively correct, in particular:
1. NMR experiments have shown that the conformation of
GlcNAc-containing ligands 4 and 5 in the free state and in the
CTB5–5 complex do not differ from those observed for
the GalNAc substrates 25 and 313 (free state) and CTB5–313
complex.
2. Competition experiments using the TR-NOESY technique
have unequivocally shown that 5 binds into the GM1-binding
site of CTB5, from which it can be displaced by the higher-
affinity natural ligand oGM1.
3. Fluorescence emission titrations of CTB5 have shown that
the affinities of the GlcNAc-containing ligands 4 and 5 are of
the same order of magnitude as those measured for the corre-
sponding GalNAc-containing compounds 2 and 3. However, in
contrast with the behaviour of 2, binding of 4 to CTB5 is not
cooperative. Furthermore, the increase of affinity predicted by
FEP calculations11 in the 2 to 4 mutation is not borne out by
experiment.
Fig. 6 Titrations of CTB5 0.5 µM with 3 (empty triangles, solid line)
and 5 (black triangles, broken line). The absolute values of the variation
of fluorescence intensity emission at 350 nm have been normalised to
10, and plotted against the total concentration of the ligands (µM). The
dissociation constants calculated by non-linear regression analysis are
190 and 450 µM, respectively.
Thus, it appears that, compared to the parent ganglioside
and pseudo-gangliosides, the GlcNAc compounds 4 and 5 are
indeed showing a comparable affinity for CT, as expected based
on the conformational analysis and molecular dynamic simu-
lation of the toxin–4 complex.11 However, contrary to the FEP
predictions,11 4 is not more active than 2, nor does 5 appear to
be any more active than 3.
The structural reasons leading to the cooperative behaviour
of GM1 binding to CT are unclear, thus it is very difficult to
understand what leads to the loss of cooperativity in some of
the psGM1 analogues. Computational analysis (let alone pre-
diction) of cooperativity in systems of this size is currently
unfeasible, so at this time we can only try a speculative inter-
pretation of the available experimental data. Based on X-ray
structures of bound and unbound CT and LT, the main struc-
tural effect that sugar binding has on these toxins is a tightening
of the 51–60 loop of the B subunits, a loop that connects beta
strand β4 to helix α2 of the B monomer.24 The position of the
α2 helix controls the size of the pore in the B5 doughnut,25 and
thus it may regulate translocation of the A fragment through
the pore and ultimately through the host cell membrane.
Through the α2 helix, carbohydrate binding to one site could
in principle be signalled to the adjacent B monomer. However,
this same tightening is elicited by sugars that are cooperative
binders, such as GM1, and sugars that are non-cooperative
binders, such as galactose.24 When we first determined the
affinity of 3 and other analogues obtained by sialic acid
replacement,9 we found that none of them displayed cooper-
ative binding and speculated that the communications between
CTB monomers elicited by 1 and 2 may be mediated by the
NeuAc residue. The NeuAc side-chain, in fact, is in contact
with the Gly-33 residue of the Bϩ1 monomer through one
water molecule (W2) located at crystallographic site two in the
CT–GM1 complex.7 This molecule is not seen in the known
X-ray structures of isolated enterotoxins,24 but it is present in
many24 (but not all)26 of the known toxin–ligand structures.
Neither of the new GlcNAc-containing mimics 4 and 5
appears to exhibit cooperative behaviour in binding to CTB5,
Discussion
The complex formed by ganglioside GM1 [Galβ1-3GalNAcβ1-
4(NeuAcα2-3)Galβ1-4Glcβ1-1Cer] and the two AB5 entero-
toxins CT and LT is one of the best characterised protein–
carbohydrate pairs. Both toxins recognise the pentasaccharide
head-group of ganglioside GM1 (o-GM1, 1) on the host cell
epithelial surface using the B5 pentamer. The pentamer has a
characteristic, doughnut-like shape, with a central pore. Once
the toxin is attached to the membrane, a fragment of the A
monomer moves through the pore and is finally translocated
through the host cell membrane into the cytosol. Biochemical6
and structural17 data have shown that there are five binding
sites in the toxin, and that the ganglioside binds to them using
the two terminal sugars at its non-reducing end, a galactose
(Gal) and a sialic acid (NeuAc). Binding of oGM1 to CT dis-
plays positive cooperativity, as determined by calorimetric titra-
tions,19 and in biological settings the interaction between the B5
pentamer and several membrane-bound GM1 molecules is a
text-book case of high-affinity multivalent interaction.22
We have shown that it is possible to replace the reducing end
lactose of oGM1 with an appropriate diol as in psGM1 2 with
no loss of affinity or cooperativity effect.5 The sialic acid can
also be replaced with various hydroxyacids, as in the lactic acid
derivative 3.9 In this case, the amount of affinity retained varies
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
790

View Article Online
independent of the presence of the sialic acid moiety (see Fig.
6), apparently suggesting that NeuAc is not the cooperativity
determinant. However, it should be noted that during the
dynamics simulations of the LT–4 complex11 a displacement of
the water molecule W2 was observed from the position it occu-
pies in CT–17 and LT–211 and toward the 4-hydroxy group of
the GlcNAc residue of 4. Thus, in the LT–4 complex W2 loses
its interaction with Gly-33(Bϩ1) and participates in a different
H-bond network.11 On the basis of this dynamics simulation
and of the above speculation on inter-monomer communic-
ation, it is still possible that the cooperativity behaviour observed
for 1 and 2 is indeed determined by the sialic acid side-chain
via the W2 water molecule. The latter, however, is displaced
from its normal position by the GlcNAc-containing ligands
and cannot transmit its signal to the adjacent (Bϩ1) monomer
through the Gly-33 residue. If this is not the case, more subtle
effects must be operating in defining the cooperativity
behaviour of ganglioside binding to CT.
Free Energy Perturbation (FEP) calculations have recently
been systematically applied to a series of antibody–sugar com-
plexes using GLYCAM, TIP3P water, RESP charges and step-
wise perturbations.27 The simulations were shown to reproduce
reasonably the known geometries of ligand–protein complexes,
while the calculated values of ∆∆G of binding were found to be
qualitatively reproduced and to depend heavily on the choices
made about the protonation state of an His located in the vicin-
ity of the binding site. Our estimate of the ∆∆G of binding
between 2 and 4 was obtained11 with a similar set of parameters
(GLYCAM, TIP3P water, ESP-derived charges on the carbo-
hydrate atoms) and the mutation was performed in the 4 to 2
sense using the slow-growth algorithm rather than stepwise
perturbations of the ligand. The energetic gain calculated for 4
(3.8 1.9 kcal molϪ1) resulted from both a better solvation of
the GalNAc ligand 2 in the free state, and a more favourable
interaction of 4 with the protein. Since, as we have discussed
above, the pseudo-ganglioside oligosaccharides experimentally
display a limited internal flexibility,5,13,14 incomplete sampling
of the free ligands doesn’t appear to be a likely explanation of
the rather large error in the relative free energies computed.
However this might not hold in the case of the protein-com-
plex, where sampling of the protein’s degrees of freedom is also
important (vide infra). The use of LT rather than CT in the
calculation was also taken into account as a possible source of
error in the calculation. However, preliminary fluorescence
titrations28 obtained with a sample of the entire LT toxin29
showed the same trends observed with CT, i.e. equivalent pairs
of ligands (3 and 5, and 2 and 4) have similar affinity for LT as
well as for CT.
samine C4 in 2 would yield a new molecule with overall shape
and conformational properties very similar to psGM1. It also
allowed a prediction that the new molecule 4 would be able to
interact with the cholera toxin in a way similar to its GalNAc
analogue. The question remains open of how to reliably achieve
quantitative prediction of relative binding affinities in this and
similar systems.
In conclusion, two new artificial ligands of the cholera toxin,
the pseudo sugars 4 and 5 are described. The new ligands were
designed starting from the known GM1 mimics 2 and 3 by
replacement of their GalNAc residue with the C4 isomer Glc-
NAc. Such substitution had been suggested by inspection of the
CT–GM1 complex, and supported by computational predic-
tions, which suggested that the three-dimensional shape of the
new ligands and their mode of interaction with CT would be
similar to those of the starting structures. These predictions are
now confirmed by the experimental results showing that the
conformational properties of the equivalent pairs 2–4 and 3–5
are indeed very similar and that their affinity for CT is of the
same order of magnitude. NMR experiments have also allowed
the gathering of information on the structure of the CT–5
complex showing that 5 occupies the GM1-binding site of the
toxin and that it binds with a conformation similar to the one
adopted by 3 in the CT–3 complex. Since GalNAc is normally
synthesised by C4 inversion of GlcNAc, the GlcNAc ligands
constitute a new class of GM1 mimics with improved synthetic
accessibility.
Experimental section
Synthesis
The synthesis of compound 6 was described in ref. 8. The full
synthetic sequence leading to 4 and 5 and the product charac-
terisations are reported in the Supplementary Information.
NMR
NMR spectra of 4 and 5 were recorded at 25–30 ЊC in D2O, on
Varian Unity 500 MHz or Bruker AVANCE 400 MHz spectro-
meters. For the experiments with the free ligands, the corre-
sponding compound (1–1.5 mg) was dissolved in D2O and the
solution was degassed by passing N2. COSY, TOCSY and
HSQC experiments were performed using standard sequences
at temperatures between 298 and 310 K. NOESY experiments
were performed with mixing times of 500, 700, 1000 ms (4) and
400, 600, 700, 800 ms (5). ROESY experiments30 were per-
formed with mixing times of 50, 100, 150, 200, 300 ms (4) and
100, 200, 250, 300 ms (5).
The most likely explanation for the error in the computed
relative free energies lies in the computational treatment of the
pseudo-ganglioside–toxin complex part of the FEP. First,
incomplete sampling of the complex simulation cannot be ruled
out, particularly considering that a substantial amount of side-
chain rearrangements occur in the protein binding site on pass-
ing from the LT–4 to the LT–2 complex. Furthermore, the state
of protonation of His-57, one of the amino acids most involved
in the rearrangement, was not addressed at all during the FEP
simulation.
Although achieving a quantitative prediction of relative
binding energies for oligosaccharide–protein complexes may
well require a lot more effort, this work shows that compu-
tational tools can be used with success to design new inhibitors
of carbohydrate–protein interaction, and that they yield quali-
tatively useful results. Indeed, at the start of this project, noth-
ing could suggest that inversion of the hexosamine C4 in
psGM1 2 would yield a ligand with good CT affinity. We have
already noted that, to the best of our knowledge, the natural
counterpart of 4 has never been described, and certainly has
never been tested as a ligand for bacterial enterotoxins. The
calculations correctly predicted that inversion of the hexo-
The cholera toxin CTB pentamer (CTB5) was purchased
from List Biological Laboratories Inc. The commercial sample
was ultrafiltered to remove EDTA and tris salt, redissolved in
phosphate buffer and subjected to two cycles of freeze-drying
with D2O to remove traces of H2O. The sample was then dis-
solved with D2O, and the solution transferred to the NMR tube
to give a final concentration ca. 0.1 mM. TR-NOESY experi-
ments were performed with mixing times of 100, 200 and 300
ms, for a ca. 50 : 1 molar ratio of 5 : lectin. TR-ROESY experi-
ments were also performed with mixing times of 100, 200, 250,
300 ms. In the competition experiment, 1.5 mg of oGM1 1 (gift
from Professor Sandro Sonnino, Dipartimento di Chimica e
Biochimica Medica, Universita’ di Milano) were added to the
same solution and the TR-NOESY spectra were recorded as
described above.
Titrations
Fluorescence titrations were performed with an LS50 Perkin
Elmer fluorimeter, using a pH 7.5 tris buffer to dissolve CTB5
(0.5 µM) and ligands. Fluorescence from TRp-88 was measured
with an excitation of 280 nm and an emission from 300 to 450
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
791

View Article Online
9 A. Bernardi, L. Carrettoni, A. Grosso Ciponte, D. Monti and
S. Sonnino, Bioorg. Med. Chem. Lett., 2000, 10, 2197–2200.
10 A. Bernardi, L. Raimondi and F. Zuccotto, J. Med. Chem., 1997, 40,
1855–1862.
11 A. Bernardi, M. Galgano, L. Belvisi and G. Colombo, J. Comput.-
Aided Mol. Des., 2001, 15, 117–128.
nm. The emission spectra are reported as Supplementary
Information. The Kd reported in the text were determined by
non-linear regression analysis, using Sigma Plot 2.0 (Jandel
Corporation).
12 P. Brocca, P. Berthault and S. Sonnino, Biophys. J., 1998, 74, 309–
318.
13 A. Bernardi, D. Potenza, A. M. Capelli, A. García-Herrero,
F. J. Cañada and J. Jiménez-Barbero, Chem.-Eur. J., 2002, 8, 4597–
4612.
14 P. Brocca, A. Bernardi, L. Raimondi and S. Sonnino, Glycoconjugate
J., 2000, 17, 283–299.
15 A. A. Bothner-By and R. Gassend, Ann. N. Y. Acad. Sci., 1973, 222,
668–676.
16 P. L. Jackson, H. N. Moseley and N. R. Krishna, J. Magn. Reson.,
Ser. B, 1995, 107, 289–292; V. L. Bevilacqua, D. S. Thomson and
J. H. Prestegard, Biochemistry, 1990, 29, 5529–5537; V. L.
Bevilacqua, Y. Kim and J. H. Prestegard, Biochemistry, 1992, 31,
9339–9349.
17 M. Vogtherr and T. Peters, J. Am. Chem. Soc., 2000, 122, 6093–6099.
18 M. Mayer and B. Meyer, J. Med. Chem., 2000, 43, 2093–2099.
19 A. Schoen and E. Freire, Biochemistry, 1989, 28, 5019–5024.
20 An accurate NMR study of these complexes is in progress, and will
be reported in due course.
Acknowledgements
We thank Professor Sandro Sonnino (Dipartimento di Chimica
e Biochimica Medica, Universita’ di Milano) for a generous gift
of oGM1, and for helpful discussions throughout this project.
We thank Giorgio Colombo (IBMR, Milano) for carefully
reading this manuscript and providing his comments. Fund-
ing was provided by MIUR, CNR, COST D-13/012, Azioni
Integrate.
References
1 (a) H. Lis and N. Sharon, Chem. Rev., 1998, 98, 637–674; (b) R. A.
Dwek, Chem. Rev., 1996, 96, 683–720; (c) A. Varki, Glycobiology,
1993, 3, 97–130.
2 P. Sears and C. H. Wong, Angew. Chem., Int. Ed., 1999, 111, 2446–
2471; P. Sears and C. H. Wong, Angew. Chem., Int. Ed., 1999, 38,
2300–2324 and references therein.
21 J. A. Mertz, J. A. McCann and W. D. Picking, Biochem. Biophys.
Res. Commun., 1996, 226, 140–144.
3 A. Bernardi, D. Arosio, L. Manzoni, F. Micheli, S. Pasquarello and
P. Seneci, J. Org. Chem., 2001, 66, 6209–6216.
4 trans-Cyclohexane-1,2-diol has been used by many authors as an
effective replacement for 3,4-disubstituted N-acetylglucosamine in
the synthesis of sialyl Lewis X mimics. For a review, see: E. E.
Simanek, G. J. McGarvey, J. A. Jablonowski and C.-H. Wong,
Chem. Rev., 1998, 98, 833–862.
5 A. Bernardi, A. Checchia, P. Brocca, S. Sonnino and F. Zuccotto,
J. Am. Chem. Soc., 1999, 121, 2032–2036.
6 C.-L. Schengrund and N. J. Ringler, J. Biol. Chem., 1989, 264,
13233–13237 and references therein.
7 (a) E. A. Merritt, S. Sarfaty, F. van den Akker, C. L’Hoir,
J. A. Martial and W. G. J. Hol, Protein Sci., 1994, 3, 166–175;
(b) E. A. Merritt, P. Kuhn, S. Sarfaty, J. L. Erbe, R. K. Holmes and
W. G. J. Hol, J. Mol. Biol., 1998, 282, 1043–1059.
8 A. Bernardi, G. Boschin, A. Checchia, M. Lattanzio, L. Manzoni,
D. Potenza and C. Scolastico, Eur. J. Org. Chem., 1999, 6, 1311–
1317.
22 M. Mammen, S.-K. Chio and G. M. Whitesides, Angew. Chem.,
Int. Ed., 1998, 37, 2755–2794.
23 S. Cattaldo Tesi di Laurea, Universita’ di Milano a.a. 2000–2001.
24 E. A. Merritt, T. K. Sixma, K. H. Kalk, B. A. M. van Zanten and
W. G. J. Hol, Mol. Microbiol., 1994, 13, 745–753.
25 A. T. Amn, S. Fraser, E. A. Merritt, C. Rodrigherio, M. Kenny,
M. Ahn, W. G. J. Hol, N. A. Williams, W. I. Lencer and T. R. Hirst,
Proc. Natl. Acad. Sci. USA, 2001, 98, 8536–8541.
26 W. E. Minke, C. Roach, W. G. J. Hol and C. L. M. J. Verlinde,
Biochemistry, 1999, 38, 5684–5692.
27 A. Pathiaseril and R. J. Woods, J. Am. Chem. Soc., 2000, 122, 331–
338.
28 D. Arosio, unpublished.
29 Generous gift of Mariagrazia Pizza at Chiron Vaccines, Siena.
30 J. Dabrowski, T. Kozár, H. Grosskurth and N. E. Nifant’ev, J. Am.
Chem. Soc., 1995, 117, 5534–5539.
31 D. Acquotti, L. Poppe, J. Dabrowski, C.-W. V.d. Lieth, S. Sonnino
and G. Tettamanti, J. Am. Chem. Soc., 1990, 112, 7772–7778.
O r g . B i o m o l . C h e m . , 2 0 0 3 , 1, 7 8 5 – 7 9 2
792