Dissecting the Cholera Toxin-Ganglioside GM1 Interaction by Isothermal Titration Calorimetry

Article abstract of DOI:10.1021/ja0378207

The complex of cholera toxin and ganglioside GM1 is one of the highest affinity protein-carbohydrate interactions known. Herein, the GM1 pentasaccharide is dissected into smaller fragments to determine the contribution of each of the key monosaccharide residues to the overall binding affinity. Displacement isothermal titration calorimetry (ITC) has allowed the measurement of all of the key thermodynamic parameters for even the lowest affinity fragment ligands. Analysis of the standard free energy changes using Jencks' concept of intrinsic free energies reveals that the terminal galactose and sialic acid residues contribute 54% and 44% of the intrinsic binding energy, respectively, despite the latter ligand having little appreciable affinity for the toxin. This analysis also provides an estimate of 25.8 kJ mol^-1 for the loss of independent translational and rotational degrees of freedom on complexation and presents evidence for an alternative binding mode for ganglioside GM2. The high affinity and selectivity of the GM1-cholera toxin interaction originates principally from the conformational preorganization of the branched pentasaccharide rather than through the effect of cooperativity, which is also reinvestigated by ITC.

Full text of DOI:10.1021/ja0378207

Published on Web 01/13/2004
Dissecting the Cholera Toxin-Ganglioside GM1 Interaction by
Isothermal Titration Calorimetry
†
‡
,†
W. Bruce Turnbull, Bernie L. Precious, and Steve W. Homans*
Contribution from the Astbury Centre for Structural Molecular Biology,
School of Biochemistry and Molecular Biology, UniVersity of Leeds, Leeds LS2 9JT, U.K., and
Centre for Biomolecular Sciences, UniVersity of St. Andrews, St. Andrews KY16 9ST, U.K.
Received August 8, 2003; E-mail: s.w.homans@leeds.ac.uk
Abstract: The complex of cholera toxin and ganglioside GM1 is one of the highest affinity protein-
carbohydrate interactions known. Herein, the GM1 pentasaccharide is dissected into smaller fragments to
determine the contribution of each of the key monosaccharide residues to the overall binding affinity.
Displacement isothermal titration calorimetry (ITC) has allowed the measurement of all of the key
thermodynamic parameters for even the lowest affinity fragment ligands. Analysis of the standard free
energy changes using Jencks’ concept of intrinsic free energies reveals that the terminal galactose and
sialic acid residues contribute 54% and 44% of the intrinsic binding energy, respectively, despite the latter
ligand having little appreciable affinity for the toxin. This analysis also provides an estimate of 25.8 kJ
-
1
mol for the loss of independent translational and rotational degrees of freedom on complexation and
presents evidence for an alternative binding mode for ganglioside GM2. The high affinity and selectivity of
the GM1-cholera toxin interaction originates principally from the conformational preorganization of the
branched pentasaccharide rather than through the effect of cooperativity, which is also reinvestigated by
ITC.
Introduction
nisms: (1) through positive cooperativity between the protein
5
-7
subunits
and (2) through the multivalent interaction of the
Cell surface protein-carbohydrate interactions mediate the
pentameric toxin with up to five copies of GM1 at a cell
invasion and colonization of a wide range of pathogenic bacteria
8
1
surface; the micromolar affinity is thus amplified to give a
and viruses. Furthermore, a number of bacteria, including
9
nanomolar avidity. As CTB/LTB-GM1 adhesion is a prereq-
Clostridium tetani, Escherichia coli, and Vibrio cholerae, also
release protein toxins that similarly exploit adhesion to cell
surface carbohydrates as a means of entering target cells,
uisite for the cell entry and thus subsequent activation of the
2
toxin, there is a continuing interest in the design of high-affinity
1
0-12
13
monovalent
and multivalent inhibitors based on the
usually of the gut wall. Such toxins are the primary virulence
factors of many bacteria and the causative agents of tetanus,
cholera, and haemolytic uremic syndrome among other diseases.
The archetypal bacterial toxins are the cholera and heat-labile
toxins derived from V. cholerae and E. coli, respectively. These
toxins share an AB5-type structuresa single A-subunit of 27
kDa which possesses a toxic ADP-ribosyltransferase activity
and five identical B-subunits which bind selectively to the cell
surface glycolipid ganglioside GM1 (Figure 1). The cholera
toxin B-subunit (CTB) and heat-labile toxin B-subunit (LTB)
share 80% sequence identity. CTB interacts with the soluble,
monovalent oligosaccharide portion of GM1 (GM1os, 1) with
oligosaccharide structure. Furthermore, as a consequence of its
high affinity, the CTB-GM1os complex is an ideal system for
(
4) Gangliosides GM1, GM2, GM3, GA1, GD1b, and FucGM1 are abbreviated
in the style of L. Svennerholm (J. Neurochem. 1963, 10, 613-623) and
are illustrated in Figure 1. Other abbreviations: AFM, atomic force
microscopy; CTB, cholera toxin B-subunit; Gal, galactose; GalNAc,
N-acetylgalactosamine; GM1os, GM1 oligosaccharide; ITC, isothermal
titration calorimetry; LTB, heat-labile toxin B-subunit; Neu5Ac, N-acetyl-
neuraminic acid; SAR-by-NMR, structure-activity relationships by NMR
spectroscopy; SPR, surface plasmon resonance; TLC, thin-layer chroma-
tography.
3
4
(
5) Sch o¨ n, A.; Freire, E. Biochemistry 1989, 28, 5019-5024.
(6) (a) Sattler, J.; Schwarzmann, G.; Staerk, J.; Ziegler, W.; Wiegandt, H.
Hoppe-Seyler’s Z. Physiol. Chem. 1977, 358, 159-163. (b) Sattler, J.;
Schwarzmann, G.; Knack, I.; Roehm, K. H.; Wiegandt, H. Hoppe-Seyler’s
Z. Physiol. Chem. 1978, 359, 719-723.
5
a micromolar dissociation constant at 37 °C, which places it
among the highest affinity protein-carbohydrate interactions
known. This binding is further enhanced through two mecha-
(
7) Schafer, D. E.; Thakur, A. K. Cell Biophys. 1982, 4, 25-40.
(
8) Mammen, M.; Choi, S. K.; Whitesides, G. M. Angew. Chem., Int. Ed. 1998,
37, 2755-2794.
(
9) Lauer, S.; Goldstein, B.; Nolan, R. L.; Nolan, J. P. Biochemistry 2002, 41,
1742-1751.
†
University of Leeds.
University of St. Andrews.
(
10) (a) Bernardi, A.; Checchia, A.; Brocca, P.; Sonnino, S.; Zuccotto, F. J.
Am. Chem. Soc. 1999, 121, 2032-2036. (b) Minke, W. E.; Hong, F.;
Verlinde, C. L. M. J.; Hol, W. G. J.; Fan, E. J. Biol. Chem. 1999, 274,
33469-33473. (c) Bernardi, A.; Arosio, D.; Sonnino, S. Neurochem. Res.
2002, 27, 539-545.
‡
(
1) (a) Karlsson, K. A. Annu. ReV. Biochem. 1989, 58, 309-350. (b) Karlsson,
K. A. Mol. Microbiol. 1998, 29, 1-11. (c) Menozzi, F. D.; Pethe, K.; Bifani,
P.; Soncin, F.; Brennan, M. J.; Locht, C. Mol. Microbiol. 2002, 43, 1379-
1
386.
(11) Pickens, J. C.; Merritt, E. A.; Ahn, M.; Verlinde, C. L. M. J.; Hol, W. G.
J.; Fan, E. Chem. Biol. 2002, 9, 215-224.
(12) Bernardi, A.; Arosio, D.; Manzoni, L.; Monti, D.; Posteri, H.; Potenza, D.;
Mari, S.; Jimenez-Barbero, J. Org. Biomol. Chem. 2003, 1, 785-792.
(
2) Merritt, E. A.; Hol, W. G. J. Curr. Opin. Struct. Biol. 1995, 5, 165-171.
3) Fukuta, S.; Magnani, J. L.; Twiddy, E. M.; Holmes, R. K.; Ginsburg, V.
Infect. Immun. 1988, 56, 1748-1753.
(
10.1021/ja0378207 CCC: $27.50 © 2004 American Chemical Society
J. AM. CHEM. SOC. 2004, 126, 1047-1054
9
1047

A R T I C L E S
Turnbull et al.
Figure 1. Composite structure of gangliosides used in binding studies with CTB.
structures of the bound and free proteins reveals that there is
only a small change to the backbone conformation (albeit mostly
18
around the galactose binding site) on complexation. Further-
1
9
more, solution structures of GM1os indicate that the ligand is
essentially preorganized for a near lock-and-key interaction with
CTB.
Considering the importance of this interaction to the progres-
sion of cholera (and traveler’s diarrhea in the case of the E.
coli heat-labile toxin), it is not surprising that there have been
several studies reported on the selectivity and structural basis
of complexation for GM1 and analogous ligands. Techniques
3
,20
as broad ranging as solid-phase and TLC overlay assays,
2
1,22
surface plasmon resonance (SPR) biosensing,
fluorescence
and flow cytometry (FACS), atomic force
microscopy (AFM), and isothermal titration calorimetry
ITC) have been applied to the problems of deconvoluting
the kinetics and thermodynamics of this multivalent interaction
and to rationalize the exquisite binding selectivity for GM1.
The general trend in binding affinities/avidities that emerges
2
3,24
9
spectroscopy
2
5
2
6
(
Figure 2. Schematic representation of the CTB-GM1os complex after
Merritt et al., ref 14.
from these studies is GM1 ≈ FucGM1 > GD1b . GM2 >
gaining further insight into the factors governing affinity and
selectivity in protein-carbohydrate interactions.
3,20,22,26,27
GA1 > GM3 (Figure 1).
Structure-activity relation-
ship studies of the sialic acid functional groups indicate that
loss of the negative charge has a greater influence on affinity
than removal/alteration of the acetamide group or glycerol side
The crystal structure of the CTB-GM1os complex has been
refined to 1.25 Å resolution.1 It shows a bivalent interaction
of the branched GM1os pentasaccharide, which has been likened
to the carbohydrate holding the protein in a “two-fingered grip”
comprising a sialic acid thumb and a Galâ(1f3)GalNAc
forefinger (Figure 2). There are extensive intermolecular hydro-
gen-bonding contacts, both directly between the ligand and
receptor and also via bridging water molecules. In terms of
buried surface area, the terminal Gal, GalNAc, and Neu5Ac
residues contribute 39%, 17%, and 43% of the intermolecular
4,15
2
8
chain.
However, it is more difficult to extract precise information
on the intrinsic contribution of each monosaccharide residue in
the GM1 ligand from experiments conducted under multivalent
conditions, as the mode of presentation of the ligand groups
2
9
can strongly affect experimental results, even giving quite
2
1,22
conflicting views of intrinsic selectivities.
Therefore, to try
1
4
(18) (a) Merritt, E. A.; Sixma, T. K.; Kalk, K. H.; van Zanten, B. A. M.; Hol,
W. G. J. Mol. Microbiol. 1994, 13, 745-753. (b) Zhang, R. G.; Westbrook,
M. L.; Westbrook, E. M.; Scott, D. L.; Otwinowski, Z.; Maulik, P. R.;
Reed, R. A.; Shipley, G. G. J. Mol. Biol. 1995, 251, 550-562.
contacts. A large number of crystal structures of complexes
between galactose derivatives and CTB or LTB have also been
reported,11,16
17
including that of the Tn antigen, which represents
(
19) (a) Acquotti, D.; Poppe, L.; Dabrowski, J.; Vonderlieth, C. W.; Sonnino,
S.; Tettamanti, G. J. Am. Chem. Soc. 1990, 112, 7772-7778. (b)
Richardson, J. M.; Milton, M. J.; Homans, S. W. J. Mol. Recog. 1995, 8,
the Galâ(1f3)GalNAc forefinger of GM1os. In all such
complexes, the orientation of galactose and the structure of the
binding site are essentially identical. Comparison of the
3
58-362. (c) Brocca, P.; Berthault, P.; Sonnino, S. Biophys. J. 1998, 74,
309-318. (d) Brocca, P.; Bernardi, A.; Raimondi, L.; Sonnino, S.
Glycoconjugate J. 2001, 17, 283-299.
(20) Ångstr o¨ m, J.; Teneberg, S.; Karlsson, K.-A. Proc. Natl. Acad. Sci. U.S.A.
1994, 91, 11859-11863.
(
13) (a) Thompson, J. P.; Schengrund, C.-L. Biochem. Pharm. 1998, 56, 591-
5
97. (b) Vrasidas, I.; de Mol, N. J.; Liskamp, R. M. J.; Pieters, R. J. Eur.
(21) Kuziemko, G. M.; Stroh, M.; Stevens, R. C. Biochemistry 1996, 35, 6375-
6384.
J. Org. Chem. 2001, 4685-4692. (c) Merritt, E. A.; Zhang, Z.; Pickens, J.
C.; Ahn, M.; Hol, W. G. J.; Fan, E. J. Am. Chem. Soc. 2002, 124, 8818-
(22) MacKenzie, C. R.; Hirama, T.; Lee, K. K.; Altman, E.; Young, N. M. J.
Biol. Chem. 1997, 272, 5533-5538.
8
824. (d) Zhang, Z.; Merritt, E. A.; Ahn, M.; Roach, C.; Hou, Z.; Verlinde,
C. L. M. J.; Hol, W. G. J.; Fan, E. J. Am. Chem. Soc. 2002, 124, 12991-
(23) Picking, W. L.; Moon, H.; Wu, H.; Picking, W. D. Biochim. Biophys. Acta
1995, 1247, 65-73.
1
2998. (e) Schengrund, C. L. Biochem. Pharm. 2003, 65, 699-707.
(
(
(
14) Merritt, E. A.; Sarfaty, S.; van den Akker, F.; L’Hoir, C. L.; Martial, J. A.;
(24) Mertz, J. A.; McCann, J. A.; Picking, W. D. Biochem. Biophys. Res.
Commun. 1996, 226, 140-144.
Hol, W. G. J. Protein Sci. 1994, 3, 166-175.
15) Merritt, E. A.; Kuhn, P.; Sarfaty, S.; Erbe, J. L.; Holmes, R. K.; Hol, W.
G. J. J. Mol. Biol. 1998, 282, 1043-1059.
16) (a) Merritt, E. A.; Sarfaty, S.; Feil, I. K.; Hol, W. G. J. Structure 1997, 5,
(25) Cai, X.-E.; Yang, J. Biochemistry 2003, 42, 4028-4034.
(26) Masserini, M.; Freire, E.; Palestini, P.; Calappi, E.; Tettamanti, G.
Biochemistry 1992, 31, 2422-2426.
1
485-1499. (b) Fan, E.; Merritt, E. A.; Zhang, Z.; Pickens, J. C.; Roach,
(27) Schengrund, C. L.; Ringler, N. J. J. Biol. Chem. 1989, 264, 13233-13237.
(28) (a) Lanne, B.; Schierbeck, B.; Karlsson, K.-A. J. Biochem. (Tokyo) 1994,
116, 1269-1274. (b) Lanne, B.; Schierbeck, B.; Angstrom, J. J. Biochem.
(Tokyo) 1999, 126, 226-234.
C.; Ahn, M.; Hol, W. G. J. Acta Crystallogr., D 2001, D57, 201-212.
17) van den Akker, F.; Steensma, E.; Hol, W. G. J. Protein Sci. 1996, 5, 1184-
(
1
188.
1048 J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004

Dissecting the CTB−GM1 Interaction by ITC
A R T I C L E S
Chart 1. Ligands Used in the ITC Titrations
to understand how this particular protein-carbohydrate interac-
tion works so efficiently, we have dissected the soluble GM1os
pentasaccharide (1) into smaller fragments (Chart 1) and eval-
uated their intrinsic binding affinities for CTB using ITC.
heat of dilution was accounted for during the fitting process according
to Sigurskjold. No significant difference in the thermodynamic param-
eters was observed whether heat of dilution was used as a fitting
parameter or measured in a blank titration of ligand into buffer. How-
ever, when using very high concentrations of the low-affinity ligand,
e.g., for ligand 7, it was preferable to add the same concentration of
ligand to both the syringe and the cell to minimize dilution effects and
modify Sigurskjold’s fitting model accordingly.
Materials and Methods
CTB (>95% purity by SDS-PAGE) was purchased from Sigma,
and the E. coli LTB was prepared according to published methods.30
R-Methyl sialoside (4) was obtained from Chess (Germany), methyl
â-galactopyranoside (5) was from Aldrich, and 3′-sialyl lactose (6) was
from Dextra (U.K.). Compound 7 was prepared according to literature
Fluorescence Spectroscopy. Fluorescence titrations were conducted
using a Perkin-Elmer LS50B luminescence spectrometer at 25 °C using
an excitation wavelength of 290 nm and monitoring emission between
310 and 450 nm at a scan speed of 50 nm/min. Solutions of ligand 7
3
1
methods. All other reagents were purchased from Aldrich and were
used without further purification. Full synthetic details for compounds
(
0-500 mM) and CTB (1 µM) were allowed to equilibrate for a few
minutes prior to recording each spectrum. Data were analyzed in Origin
.0 using the built-in binding model.
1
-3 are given in the Supporting Information.
5
Isothermal Titration Calorimetry. ITC experiments were under-
taken using either an MCS or VP-ITC calorimeter from Microcal Inc.,
with cell volumes of 1.309 and 1.409 mL, respectively. Unless stated
otherwise, all ITC titrations were conducted at 25 °C in 50 mM Tris‚
HCl buffer at pH 7.4 containing 200 mM NaCl, 3 mM EDTA, and 1
Results
Ligand Synthesis. To avoid any uncertainties associated with
using anomeric mixtures of the smaller oligosaccharide frag-
ments, the analogous methyl glycosides of Galâ(1f3)GalNAc
mM NaN . LTB-subunit concentrations of 145-265 µM were used
3
for direct titrations of GM1os fragments, and 10 µM CTB-subunit was
used for both direct and displacement titrations with GM1os. The protein
was typically dialyzed twice against 2 L of Tris‚HCl buffer. Ligand
(3), sialic acid (4), and galactose (5) were used for the ITC ex-
periments. Disaccharide 3 was synthesized as described in the
1
Supporting Information using an approach based on a literature
concentrations were determined by H NMR spectroscopy against a 5
3
4
synthesis of the analogous allyl glycoside. In the case of 1
and its agalacto analogue GM2os (2), it was more convenient
to synthesize the ligands by selective enzymatic degradation of
mM ethanol internal standard using a single-pulse experiment. Samples
were then freeze-dried and redissolved in an appropriate volume of
the same Tris‚HCl buffer that had been used for protein dialysis. For
displacement assays, CTB was preincubated with a high concentration
3
5
bovine brain gangliosides. Although the resulting oligosac-
charides were obtained as anomeric mixtures, the CTB-GM1os
crystal structure clearly indicates that the reducing terminal
glucose residue points away from the protein and into solution
and should therefore not have any effect on the binding
thermodynamics.
(2-200 mM) of the low-affinity ligand and 110 µM GM1os was titrated
into the mixture. Typically, one 2 µL injection and 25 8 µL injections
of ligand were added at 4 min intervals. The first data point was
routinely deleted before curve fitting to allow for diffusion of ligand
across the syringe tip during the pretitration equilibration period.
Nonlinear least-squares curve fitting was conducted in Microcal
Origin using Microcal’s one-site model, Sigurskjold’s displacement
ITC. In preliminary titration experiments using mono- to
trisaccharide fragments of GM1os and LTB, all exhibited greater
32
model, or a multiple-interacting-sites model based on those described
7
5
than millimolar K values. In the case of methyl galactoside 5,
by Schafer and Thakur and Sch o¨ n and Freire, but implemented using
d
3
3
Microcal Origin LabTalk as described previously. Heats of dilution
were subtracted before curve fitting with the one-site and multiple-
interacting-sites models, but in the case of the displacement model,
this gave a hyperbolic curve even at relatively high protein
concentration (145 µM) (Figure 3a), but in the case of sialyl
lactose 6, no heat release beyond the heat of dilution was
detected (Figure 3b). Although this observation could result from
(
(
(
29) Horan, N.; Yan, L.; Isobe, H.; Whitesides, G. M.; Kahne, D. Proc. Natl.
Acad. Sci. U.S.A. 1999, 96, 11782-11786.
30) O’Dowd, A. M.; Botting, C. H.; Precious, B.; Shawcross, R.; Randall, R.
E. Vaccine 1999, 17, 1442-1453.
(34) (a) Lay, L.; Nicotra, F.; Panza, L.; Russo, G.; Adobati, E. HelV. Chim.
Acta 1994, 77, 509-514. (b) Bernardi, A.; Boschin, G.; Checchia, A.;
Lattanzio, M.; Manzoni, L.; Potenza, D.; Scolastico, C. Eur. J. Org. Chem.
1999, 1311-1317.
(35) (a) Kolodny, E. H.; Brady, R. O.; Quirk, J. M.; Kanfer, J. N. J. Lipid Res.
1970, 11, 144-149. (b) Irie, R. F.; Tai, T.; Morton, D. L.; Cahan, L. D.;
Paulson, J. C. U.S. Patent 4,557,931. (c) Ito, M. Trends Glycosci.
Glycotechnol. 1990, 2, 399-402.
31) Tarasiejska, Z.; Jeanloz, R. W. J. Am. Chem. Soc. 1958, 80, 6325-6327.
(
a) Jacquinet, J. C.; Zurabyan, S. E.; Khorlin, A. Y. Carbohydr. Res. 1974,
3
2, 137-143.
(32) Sigurskjold, B. W. Anal. Biochem. 2000, 277, 260-266.
(
33) Yung, A.; Turnbull, W. B.; Kalverda, A. P.; Thompson, G. S.; Homans, S.
W.; Kitov, P.; Bundle, D. R. J. Am. Chem. Soc. 2003, 125, 13058-13062.
J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004 1049

A R T I C L E S
Turnbull et al.
Figure 4. The interaction of CTB and GM1os can be described by a
sequential binding model with 10 stepwise stability constants, 7 of which
(shown in red) are sufficient to describe a path among all 8 states. Enthalpy
changes relative to the unbound state are shown in blue.
6
Cooperativity was first identified from nonlinear Scatchard
plots of equilibrium dialysis data which were fitted to the Hill
equation to give a cooperativity coefficient of 1.25. This value
lies toward the lower limit for positive cooperativity in a
pentameric system which can exhibit values in the range of one
Figure 3. Titrations of LTB with (a) 5 and (b) sialyl lactose 6 with ligand
dilution experiments conducted in the absence of protein overlayed in red.
The corresponding integrated data are shown in panels c and d, respectively.
Residuals of the curve fitting for the GalâOMe data scaled up by a factor
of 2 (panel c, blue circles) demonstrate the efficiency of the baseline
subtraction of dilution processes.
(
for no cooperativity) to five. Reanalysis of these data by Schafer
7
and Thakur, using a model comprising seven independent
stepwise binding constants, suggested a 2-fold increase in
affinity for a second or subsequent molecules of GM1os binding
5
to a CTB5 pentamer. Sch o¨ n and Freire employed a similar
either a very low ∆H° or a very low affinity, the difficulty in
detecting binding of GM3 (Figure 1) to either LTB or CTB in
multivalent binding assays9 would support the latter reason.
Similarly, no interaction could be detected when sialyl lactose
was added to a mixture of LTB and disaccharide 3. Thus, the
possibility of an intrasubunit cooperative effect in which the
sialic acid binding affinity could be enhanced by the presence
of a galactosyl ligand (acting to preorganize bridging water
molecules in the binding site) was discounted.
model for analyzing ITC data for the interaction. However, here
the model was further constrained by the assumption that
cooperativity would manifest itself only through nearest neigh-
bor interactions and could thus be described by a ∆G for an
isolated CTB-GM1os interaction and an additional cooperative
enhancement of ∆g for each adjacent filled binding site. An
alternative description of this model (Figure 4) employs three
binding constants
,20
Protein concentrations that are at least 10 times higher than
the Kd are required to achieve titration curves with the sigmoidal
-∆G/RT
K ) e
1
36
shape that is preferred for curve fitting. However, for weak
interactions, such concentrations are often prohibitively high
on account of protein solubility and/or availability. Although
titrations employing protein concentrations below the Kd can
-
(∆G+∆g)/RT
K ) e
2
-
(∆G+2∆g)/RT
K ) e
3
3
7
still give reasonable estimates of the binding parameters, an
alternative approach that proved to be much more sensitive and
economical in terms of protein consumption, while providing
with corresponding enthalpy changes
H ° ) ∆H
∆
1
3
8
sigmoidal curves, was to use a displacement assay. In this
approach, a low concentration of the receptor is preincubated
with the low-affinity ligand and then the high-affinity ligand is
titrated into the mixture. A concise displacement model has been
described32 previously for determining binding constants in
systems with higher affinities than can be measured by
conventional direct titration, as long as a lower affinity ligand
with known thermodynamic parameters is available. The same
model is also equally applicable for dealing with low-affinity
interactions where a well-characterized higher affinity ligand
∆
H ° ) ∆H + ∆h
2
∆
H ° ) ∆H + 2∆h
3
However, the model can be simplified further to only two
stepwise equilibrium constants and two stepwise enthalpy
changes as K1K3 ) K2K2 and ∆H1° + ∆H3° ) 2∆H2°. A more
detailed account of the fitting equations for this model has been
33
described elsewhere.
3
9
is available. However, this analysis assumes that each of the
two competing ligands binds to the receptor following a simple
two-state model. Whereas the effects of multivalency can be
excluded from the experiments by using soluble monovalent
GM1os, any cooperative effects will remain.
Sch o¨ n and Freire concluded5 that where a ligand binds
adjacent to a site that is already filled, there is a 4-fold increase
in binding affinity accompanied by a change in the enthalpy of
-1
interaction from -22 to -33 kcal mol . Such a dramatic
difference in the stepwise enthalpy change should give rise to
a binding isotherm that deviates significantly from the sigmoidal
shape that is typical of the noninteracting sites model.40
However, we were unable to reproduce these previous observa-
(
36) Wiseman, T.; Williston, S.; Brandts, J. F.; Lin, L. N. Anal. Biochem. 1989,
1
79, 131-137.
(
37) Turnbull, W. B.; Daranas, A. H. J. Am. Chem. Soc. 2003, 125, 14859-14866.
(
(
38) Zhang, Y. L.; Zhang, Z. Y. Anal. Biochem. 1998, 261, 139-148.
39) Christensen, T.; Gooden, D. M.; Kung, J. E.; Toone, E. J. J. Am. Chem.
Soc. 2003, 125, 7357-7366.
(40) Fisher, H. F.; Tally, J. Biochemistry 1997, 36, 10807-10810.
1050 J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004

Dissecting the CTB−GM1 Interaction by ITC
A R T I C L E S
observations of a cooperative effect giving a 2-fold increase in
7
binding affinity.
However, the values of ∆G2° and ∆H2°, which represent the
average values for stepwise association constants and enthalpy
changes when all five binding sites are filled, differ from ∆G°
and ∆H° for the noninteracting sites model by only 3% and
0
.2%, respectively. Furthermore, the apparent values of ∆G°
and ∆H° resulting from the cooperative model vary by only
.5% and 1.5%, respectively, during the course of the titration.
Consequently, as cooperativity has not been observed for any
3
2
4
fragments of GM1os, the noninteracting sites approximation
was adopted for curve fitting in displacement experiments. We
would suggest that differences in our results from those reported
by Sch o¨ n and Freire may result from improvements in the
sensitivity of calorimeters or in curve fitting associated with
employing stepwise, rather than cumulative, enthalpy changes.36
For the displacement titrations 10 µM CTB was preequili-
brated with the low-affinity ligand prior to titration with GM1os.
In all cases, sigmoidal curves demonstrating saturation of the
receptor were still achieved on adding only 2 equiv of GM1os
with respect to CTB-subunits. Figure 5b,d shows how the
presence of 25 mM disaccharide 3 perturbs the GM1os-CTB
titration curve. Compared to the curve for GM1os-CTB alone
Figure 5. Titrations of 110 µM GM1os into 10 µM CTB in the absence
(a) and presence (b) of 25 mM Galâ(1f3)GalNAcâOMe. The best fitting
curves based on the cooperative model (red) and noninteracting sites model
green) are shown in (c), and the displacement model (red) is shown in (d).
(
Residuals of the curve fitting scaled up by a factor of 2 (blue circles in
panel d) show that dilution processes have been accommodated adequately
during analysis.
(Figure 5c), Galâ(1f3)GalNAcâOMe reduces the apparent
Table 1. Thermodynamic Parameters for GM1os Binding to CTB
at 25 °Ca
enthalpy of interaction and association constant for the curve,
while increasing the heat of dilution that is apparent toward
the end of the titration. This last phenomenon is a consequence
of having a high concentration of the low-affinity ligand that is
present in the cell, but absent from the injectant solution. In
the case of 2, initial curve fitting returned a binding stoichi-
ometry of n ) 0.85, whereas a second titration undertaken with
identical samples of CTB and GM1os, but in the absence of 2,
gave n ) 1.00. This observation would imply that 15% of the
binding sites were already filled with the high-affinity ligand
prior to starting the titration, which corresponds to 0.06%
GM1os contaminating the sample of GM2os. Therefore, the total
concentrations of GM1os at each step of the titration were
recalculated using a modification of the equation used37 in the
Origin program:
cooperative
model/
noninteracting
sites model/
kJ mol-1
cooperative
model/
kJ mol-1
noninteracting
sites model/
kJ mol-1
kJ mol-
1
∆
∆
∆
∆
G1° -38.8 ( 0.1b
Τ∆S1° -35.0 ( 0.2
G2° -40.3 ( 0.1 -41.6 ( 0.1c Τ∆S2° -32.2 ( 0.1 -30.8 ( 0.1
H1° -73.8 ( 0.2
H2° -72.5 ( 0.1 -72.4 ( 0.1
a
Binding stoichiometry was 1.004 ( 0.001 for the noninteracting sites
model and is fixed at five ligands per CTB pentamer for the cooperative
model. Errors are those reported by Origin on the curve fitting. c The two
models are equivalent if ∆G1° ) ∆G2°, etc.
b
tions in any of our experiments, regardless of whether they were
conducted at 15, 25, or 37 °C, in either Tris‚HCl or phos-
phate buffer and using several different batches of CTB. In all
cases, we found that the data fitted well to a simple noninter-
acting sites model, returning values for K and ∆H° that were
independent of buffer ionization enthalpy. It was therefore
decided to undertake a comparison of the standard one-site
model and the cooperative model for the GM1os-CTB interac-
tion.
As the binding constant at 25 °C is approaching the upper
limit for direct titrations by ITC (on account of levels of signal-
to-noise ratio),36 data averaging was used to minimize random
noise which can be better described by a model with more fitting
parameters. Thus, data from three titrations, all run under
identical conditions, were averaged and fitted using both the
standard noninteracting sites model and the cooperative model
∆
V(i)[X]
∆V(i)
syr
[
X]t(i) ) [X] +
1 -
(1)
(
0
)(
)
V0
2V0
where [X]t(i) is the GM1os concentration following the ith
injection, [X]0 is the initial concentration of GM1os in the cell
prior to starting the titration, ∆V(i) is the sum total volume of
ligand added following the ith injection, [X]syr is the concentra-
tion of ligand in the syringe, and V0 is the effective volume of
the cell. The second bracketed term accounts for the fact that
liquid is displaced from the sensitive part of the cell each time
that an addition of ligand is made from the syringe. Curve fitting
with these revised GM1os concentrations returned n ) 0.99.
Thermodynamic parameters obtained for all systems are listed
in Table 2.
(
Figure 5a,c, Table 1). Although both models described the data
4
1
well, an F-test demonstrated that the small improvement in
fit for the cooperative model was nevertheless statistically
significant (>95% probability) over and above the effect of
having an extra variable fitting parameter compared to the one-
site model. Therefore, our results are in accord with previous
The dissociation constants vary over 7 orders of magnitude
from the nanomolar range for the GM1os pentasaccharide 1 to
the high millimolar range for methyl sialoside 4. The errors
returned by the fitting program are all small compared to the
parameter values except for T∆S° for 2 (for which the parameter
value is very low) and for methyl sialoside 4. In the latter case,
(
41) Gans, P. Data fitting in the chemical sciences by the method of least squares;
Wiley: Chichester, U.K., 1992.
J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004 1051

A R T I C L E S
Turnbull et al.
Table 2. Thermodynamic Parameters for GM1os and Its Fragment Ligands Binding to CTB at 25 °C
∆G°/kJ mol-
1
∆H°/kJ mol-1
Τ∆S°/kJ mol-1
nb
ligand
d
K /mM
a
-5
1
2
3
4
(4.33 ( 0.14 ) × 10
-41.6 ( 0.1
-15.2 ( 0.2
-12.0 ( 0.3
-3.8 ( 1.2 d
-72.4 ( 0.1
-18.0 ( 2.0
-42.1 ( 1.8
-44.4 ( 35.7
-30.8 ( 0.1
-2.8 ( 2.0
-30.1 (1.9
-40.6 ( 34.6
1.00
0.99
1.06
1.06
c
2.0 ( 0.2
7.6 ( 0.8
210 ( 100
(
-3.8 ( 1.7)
5
7
14.8 ( 1.6
118 ( 12
-10.4 ( 0.3
-5.3 ( 0.2
-37.4 ( 2.0
-38.4 ( 1.9
-27.0 ( 2.0
-33.1 ( 1.9
0.94
1.04
a
Errors are those reported by Origin on the curve fitting. b All errors on binding stoichiometries were reported to be less than 1%. c Adjusted value taking
d
into account the presence of 1.5 µM GM1os present in the GM2os/CTB mixture at the start of the titration. Value and error used in subsequent calculations
see the text).
(
4
), falls a long way short of ∆G° for the bivalent ligand. This
4
2
is perhaps not surprising, as Jencks has noted that ∆G° for
fragments of a bivalent ligand should not be additive as a
consequence of the entropic penalty to be paid on complex
formation that results from a loss of independent rotational and
translational degrees of freedom for the ligand and receptor.
Whereas for a bivalent ligand this penalty is paid only once, in
the case of two complementary fragments binding to the receptor
to form a ternary complex, this penalty would be paid twice.
Consequently, Jencks recommended invoking the concept of
i
intrinsic free energy change ∆G , which is defined as the change
in free energy on complexation in the absence of the entropic
penalty described above. In other words, once half of a bivalent
ligand has bound to its receptor, the ligand has already paid
the entropic penalty for complex formation, so the second half
of the ligand now interacts with an enhanced intrinsic affinity,
Figure 6. (a) Quenching of the CTB-Trp88 fluorescence with increasing
concentrations of GalNAcâOMe is described well by (b) a simple two-
state binding model.
i
i
i
∆G ) -RT ln K . Therefore, ∆G for a fragment of a ligand is
a measure of the contribution that fragment makes to the overall
interaction. For systems in which there is no significant change
in the structure of either the ligand or the receptor on binding,
the errors originate from the broad minimum in the 3-D surface
defining the curve fitting process. Extensive simulations revealed
that reasonable combinations of Kd and ∆H° lay in a steep-
i
∆
G for a fragment of a bivalent ligand can be estimated, to a
-
1
walled valley running from Kd ) 100 mM/∆H° ) -21 kJ mol
first approximation, by determining the shortfall in ∆G° of the
complementary fragment with respect to ∆G° for the full
bivalent ligand:
-
1
to Kd ) 400 mM/∆H° ) -79 kJ mol . Consequently, a ∆G°
-
1
value of -3.8 ( 1.7 kJ mol for methyl sialoside was carried
through subsequent calculations. Duplicate displacement experi-
ments returned values of Ka that were within (20% of the values
listed above and (7% for direct titrations run in triplicate with
GM1os. As these errors were larger than those for the precision
of the fitting reported in Table 2, the corresponding errors in
i
A
∆G ) ∆GAB° - ∆GB°
(2)
i
Thus, ∆G for the galactose residue can be estimated by
subtracting ∆G° for GM2os from ∆G° for GM1os. Furthermore,
whereas ∆G° values for fragment ligands are not additive, their
intrinsic free energies are:
-
1
-1
∆
G° of (0.5 kJ mol (for 2, 3, and 5) and (0.2 kJ mol (for
1
) were used in subsequent calculations.
The interaction of GalNAcâOMe (7) was studied by both
i
i
A
i
∆
GAB ) ∆G + ∆G
(3)
displacement ITC and fluorescence titration (Figure 6). The sole
tryptophan residue in CTB is located at the bottom of the
galactose binding site where its indole ring stacks against the
hydrophobic R-face of the galactosyl residue. It is well
established that, on close interaction with this group, the
fluorescence spectrum of CTB exhibits a change in intensity
B
i
i
Thus, ∆G for GM1os can be calculated by adding ∆G for both
fragments 3 and 4. Finally, the difference between ∆G° and
G for GM1os is a measure of the largely entropic penalty to
be paid on bringing two particles together to form a complex.
Jencks referred to this as the Gibbs connection energy (∆G ),
and for an ideal system this parameter also accounts for the
shortfall in the sum of ∆G° for the fragment ligands relative to
G° for the bivalent ligand:
i
∆
s
12,24
and/or a shift in its wavelength for maximum emission.
fact that both assays return identical dissociation constants (ca.
10 mM) further validates the displacement ITC method for
The
1
∆
measuring low-affinity interactions and also indicates that, in
the absence of a galactose residue, GalNAcâOMe will instead
occupy the galactose binding site, albeit with a reduced affinity
compared to GalâOMe (5).
s
i
∆
G ) ∆GAB° - ∆GAB ) (∆G ° + ∆G °) - ∆GAB° (4)
A B
A summary of the standard, intrinsic, and connection free
energy changes are given in Figure 7. An initial analysis of the
data, however, provides inconsistent values for ∆G depending
Discussion
s
From the results in Table 2, it is clear that the sum of ∆G°
for complementary fragments of GM1os (i.e., 2 + 5 and 3 +
(42) Jencks, W. P. Proc. Natl. Acad. Sci. U.S.A. 1981, 78, 4046-4050.
1052 J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004

Dissecting the CTB−GM1 Interaction by ITC
A R T I C L E S
Table 3. Summary of Revised Intrinsic Free Energies for GM1os
and Its Fragment Ligands Binding to CTB at 25 °C
i a
∆
G /
% BSA
GM1os−CTBc
1
% ∆GGM1osi
fragment
GM1os
kJ mol-
-67.4 ( 1.9
-31.2 ( 0.5
-37.8 ( 1.7
-29.6 ( 0.5
-36.2 ( 1.8
-1.6 ( 0.7b
100
46
56
44
54
2
100
60
56
43
39
17
GM2os
Galâ(1f3)GalNAc
R-Neu5Ac
â-Gal
â-GalNAc
a
-1
Revised values assuming ∆G° ) -5.4 ( 1.8 kJ mol for GM2os
binding to CTB in the same orientation as GM1os. Based on ∆GGalNAci
b
Figure 7. Summary of the free energy changes for ligands 1-5. Standard
free energy changes are shown in black, intrinsic free energy changes in
gray, and Gibbs connection energies in white. Up and down arrows indicate
positive and negative changes, respectively.
c
)
∆GGalGalNAc° - ∆GGal°. Change in buried surface area for GM1os-
CTB interaction as reported in ref 14.
consideration that GalNAcâOMe binds in this site with a higher
affinity than Neu5AcROMe alone has for CTB, it is not
unreasonable to predict that, in the absence of the terminal
galactose residue, the GM2os fragment instead occupies the
galactose binding site while repositioning the sialic acid residue
to give alternative additional contacts with the protein. This
on which disconnection (i.e., GalâOMe/GM2os or Neu5AcROMe/
Galâ(1f3)GalNAcOMe) is chosen for the calculation. The most
likely explanation for this observation is that one (or more) of
the ligands binds to the receptor in a position that is distinct
from the site occupied by the corresponding fragment of GM1os.
3
,20,27
42
hypothesis would account for the observation
that, at a
Such “nonproductive binding” will only occur if the fragment
ligand finds a higher affinity site in the absence of its
complementary fragment, or in other words, binding of a
fragment to its preferred site is destabilized once part of the
bivalent ligand. Considering that crystallographic data are
available for complexes of a wide range of galactose derivatives,
all of which show the galactose residue bound in an identical
orientation, yet no structures have been published for either the
GM2os or sialic acid complexes, it is most probable that it is
one of these complexes that differs from that predicted using
the GM1os-CTB structure. This hypothesis is reinforced by
the observation that the 50-200-fold increase in the binding
affinity of GM2os vs Neu5AcROMe is far greater than can be
accounted for by the intrinsic free energy change for the
multivalent surface, CTB/LTB binds more strongly to GM2 than
to GA1, yet shows no appreciable affinity for GM3. Further-
more, the ability of the galactose binding site to accommodate
substitutions at C-2 of the ligand is consistent with the reported
similar affinities of GM1 and fucosyl-GM1 (fucGM1) for the
2
6
toxin. Although, in theory, this hypothesis could be tested by
fluorescence titration, the experiment was precluded in the
current study by the fact that the GM2os was contaminated by
ca. 0.06% GM1os. Although this contaminant was too low to
be detected by HPLC, NMR spectroscopy, or electrospray mass
spectrometry, a reduction in the binding stoichiometry was
detected in the ITC experiment corresponding to partial oc-
cupancy of the CTB binding site prior to starting the displace-
ment titration. Whereas the contaminating GM1os could be
accommodated explicitly in calculating the thermodynamic
parameters for this displacement titration, it would be impossible
to distinguish the effect on CTB fluorescence of GM2os from
that of the much higher affinity GM1os.
-
1
additional GalNAc residue of 1.4 kJ mol (calculated from
i
∆
GGalNAc ) ∆GGalGalNAc° - ∆GGal°).
It follows from eq 3 that
s
∆
G ° + ∆G ° ) ∆G ° + ∆G ° ) ∆G ° + ∆G (5)
A B A′ B′ AB
A revised breakdown of the intrinsic free energy contribu-
tions, employing the predicted value of ∆G° for GM2os, is given
in Table 3. The Gibbs connection energy is calculated to be
where A-B and A′-B′ are pairs of complementary fragments
resulting from different disconnections of bivalent ligand AB.
Therefore, by knowing ∆G° for three of the fragment ligands,
it is possible to estimate what ∆G° for the fourth ligand should
be. Such a calculation would predict that the sialoside fragment
-
1
2
5.8 ( 1.9 kJ mol at 298 K. There have been several estimates
i
-1
for ∆G published varying from 8.7 to over 80 kJ mol at 298
K. Bearing in mind that no real system exhibits ideal behavior,
43
the current analysis should give only lower limits to the values
4
should have a Kd of 4 mM, which is considerably higher
i
s
of ∆G and ∆G . It is therefore worth noting that following a
review of the published literature, Lundquist and Toone
concluded that “the interaction free energy is likely no greater
affinity than the observed value. Alternatively, if the Kd for
Neu5AcROMe is assumed to be 100-400 mM, then GM2os
should have a Kd of ca. 50-200 mM (∆G° ) -5.4 ( 1.8 kJ
-
1
-1 44
-
1
-1
than 6 kcal mol ” (ca. 25 kJ mol ).
mol , based on ∆G° ) -3.8 ( 1.7 kJ mol for 4)sa much
lower affinity than is measured experimentally. This would be
feasible if the GM2os tetrasaccharide does not bind to CTB in
the same orientation as the corresponding fragment of GM1os,
but rather it finds a higher affinity site elsewhere on the receptor.
Derivation of intrinsic free energy changes also allows a
means of dissecting the individual contributions of each of the
three key monosaccharide residues that participate in the
interaction with CTB. The results summarized in Table 3
correlate reasonably well with a similar division of the change
in buried surface area on binding that was previously calculated
from the crystal structure of the complex. It is interesting to
i
Similarly, on the basis of ∆G for GalNAc, it would be
expected that 7 should have no measurable affinity for the
protein. However, 7 does bind to CTB, albeit weakly, as is
evident from both calorimetric and fluorimetric experiments.
Moreover, in this case, the fluorescence titrations provide
compelling evidence for a shift in binding site to that occupied
by the terminal galactose residue in GM1os. Taking into
(
43) Burkhalter, N. F.; Dimick, S. M.; Toone, E. J. In Carbohydrates in
Chemistry and Biology. Part I: Chemistry of Saccharides; Ernst, B., Hart,
G. W., Pierre, S., Eds.; Wiley-VCH: Weinheim, Germany, 2000; Vol. 2,
pp 863-914.
(44) Lundquist, J. J.; Toone, E. J. Chem. ReV. 2002, 102, 555-578.
J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004 1053

A R T I C L E S
Turnbull et al.
note that the sialic acid residue contributes almost half of the
stability of the complex even though, in isolation, this monosac-
charide has little appreciable affinity for the receptor. Drug
discovery strategies such as SAR-by-NMR45 rely on identifying
small molecules that may be accommodated stimultaneously
in a receptor binding site and subsequently tethered together to
form high-affinity ligands. Considering the very poor 200 mM
Kd for Neu5AcROMe, it is unlikely that the GM1os ligand
would have been discovered by such an approach! Of course,
in this case it is presumably the CTB receptor that has evolved
to exploit ligands available in mammalian glycobiology, rather
than vice versa.
affinity for GM2os is very different from that of the ligands
that bear a terminal galactose residue. However, in the absence
of structural information for this complex, further discussion
would be inappropriate.
Conclusions
The relatively high monovalent affinity of GM1 for CTB
arises from the extensive intermolecular contacts between a
ligand and receptor that are essentially preorganized for binding.
On the other hand, the high selectivity of this interaction arises
from having the two key monosaccharide recognition sites
separated in the oligosaccharide by two other residues, thus
preventing any corresponding fragments of GM1os from
achieving interactions of comparable efficiency. That high-
affinity ligands can be achieved through a chelation strategy is
certainly not a new concept, but the extent to which gains in
∆G° can be achieved in cases for which the structure of the
linker is optimal is less well appreciated. Ligands with poor
affinities in the millimolar range are conventionally dismissed
as inactive, yet here, Nature makes efficient use of weakly
binding building blocks to construct ligands of great functional
importance.
Furthermore, the importance of linker design in such a
strategy is highlighted by a GM1os analogue synthesized by
Hindsgaul and co-workers in which the internal lactose unit is
replaced by a simple ethylene linker between the sialic acid
46
and GalNAc residues. This GM1os analogue binds with a
micromolar affinity corresponding to an unfavorable ∆∆G° of
-
1
ca. 8 kJ mol relative to GM1os. It is the relative rigidity of
the GM1os pentasaccharide, which is preorganized in a con-
formation suitable for binding, that confers its impressive
selectivity over related oligosaccharide ligands and allows a
s
Jencks’ concept of intrinsic free energy changes provides a
useful framework for dissecting the contributions of individual
fragments of a complicated ligand. Such an analysis has pro-
reasonable estimate of ∆G to be made using this system. Indeed,
synthetic linkers designed to mimic the rigid conformation of
the internal galactosyl residue provide GM1 analogues with
-1
1
0a,c,12
vided an estimate of 25.8 ( 1.9 kJ mol for the entropic penalty
to be paid on bringing two molecules together to form a
complex. Furthermore, this analysis has led us to consider that
the selectivity of the toxin for GM2 over GA1 may be a
consequence of GM2 binding in a manner that differs from that
of the corresponding fragment of GM1sa hypothesis that
remains to be tested. Determination of the intrinsic contributions
of each monosaccharide residue to the overall interaction can
also provide direction in the rational design of GM1os ana-
logues. This work supports the strategy that has been fol-
affinities for CTB similar to those for the natural ligand.
However, it must be expected that there will still be some loss
of residual conformational entropy upon binding. Therefore, as
this analysis has assumed that both ligand and receptor are rigid
s
bodies, it is possible that both ∆G and the contribution of the
GalNAc residue have been underestimated. Nevertheless, it is
unlikely that the contribution of the GalNAc residue would rise
above 10% of the total, so its role in the complex is probably
less supramolecular than structural, acting to hold the sialic acid
and galactose residues in the appropriate spatial arrangement
for complexation.
1
1,12
lowed
to date in using galactose as a source of selectivity
in ligand design, but it also emphasizes the potential gains that
may be had from further exploring the conformational space
occupied by the sialic acid residue in GM1os.
Analysis of the enthalpic and entropic contributions to the
free energy changes goes further to support this conclusion.
Addition of the GalNAc residue to the terminal galactose
provides a modest increase in ∆H° which is largely offset by a
similar increase in the unfavorable entropy term. However, from
here, there is virtually no further change in the entropic term
upon introducing the sialic acid residue into the ligand. The
net contribution of the sialic acid is almost entirely enthalpic,
Acknowledgment. This work was supported through the
award of a Wellcome Trust International Prize Travelling
Research Fellowship to W.B.T. The VP-ITC unit was funded
through a Joint Infrastructure Fund award. We thank Dr. B. W.
Sigurskjold for providing definition files for his displacement
model.
-
1
enabling a favorable increase of -26.4 kJ mol in the free
energy term. It is worth noting that the balance of enthalpic
and entropic contributions giving rise to the more favorable
Supporting Information Available: Experimental details for
the preparation of compounds 1-3. This material is available
free of charge via the Internet at http://pubs.acs.org.
(
45) Shuker, S. B.; Hajduk, P. J.; Meadows, R. P.; Fesik, S. W. Science 1996,
74, 1531-1534.
46) Hindsgaul, O.; Schriemer, D. C. U.S. Patent 6,054,047, 2000.
2
(
JA0378207
1054 J. AM. CHEM. SOC.
9
VOL. 126, NO. 4, 2004