Characterization and crystal structure of a high-affinity pentavalent receptor-binding inhibitor for cholera toxin and E. coli heat-labile enterotoxin

Article abstract of DOI:10.1021/ja0202560

Multivalent ligand design constitutes an attractive avenue to the inhibition of receptor recognition and other biological events mediated by oligomeric proteins with multiple binding sites. One example is the design of multivalent receptor blockers target

Full text of DOI:10.1021/ja0202560

Published on Web 07/04/2002
Characterization and Crystal Structure of a High-Affinity
Pentavalent Receptor-Binding Inhibitor for Cholera Toxin and
E. coli Heat-Labile Enterotoxin
Ethan A. Merritt,† Zhongsheng Zhang,‡ Jason C. Pickens,§ Misol Ahn,†
Wim G. J. Hol,*,†,| and Erkang Fan‡
Contribution from Department of Biochemistry, Department of Biological Structure,
Biomolecular Structure Center, Department of Chemistry, and Howard Hughes Medical
Institute, UniVersity of Washington, Box 357742, Seattle, Washington 98195
Received February 19, 2002
Abstract: Multivalent ligand design constitutes an attractive avenue to the inhibition of receptor recognition
and other biological events mediated by oligomeric proteins with multiple binding sites. One example is
the design of multivalent receptor blockers targeting members of the AB5 bacterial toxin family. We report
here the synthesis and characterization of a pentavalent inhibitor for cholera toxin and Escherichia coli
heat-labile enterotoxin. This inhibitor is an advance over the symmetric pentacyclen-derived inhibitor
described in our earlier work in that it presents five copies of m-nitrophenyl-R-D-galactoside (MNPG) rather
than five copies of â-D-galactose. The approximately 100-fold higher single-site affinity of MNPG for the
toxin receptor binding site relative to galactose is found to yield a proportionate increase in the affinity and
IC50 measured for the respective pentavalent constructs. We show by dynamic light scattering that inhibition
of receptor binding by the pentavalent inhibitor is due to 1:1 inhibitor:toxin association rather than to inhibitor-
mediated aggregation. This 1:1 association is in complete agreement with a 1.46 Å resolution crystal structure
of the pentavalent inhibitor:toxin complex, which shows that the favorable single-site binding interactions
of MNPG are retained by the five arms of the 5256 Da pentavalent MNPG-based inhibitor and that the
initial segment of the linking groups interacts with the surface of the toxin B pentamer.
Introduction
the ligands being presented is variable, and the affinity gain
from multivalent presentation is due primarily to an increase in
Cholera toxin is a secreted toxin with overall AB5 architecture.
Five identical receptor-binding sites on the cholera toxin B
pentamer mediate binding to the epithelial cell surface of the
human host through specific interaction with ganglioside GM1.
Receptor binding by cholera toxin is both an attractive model
system for exploring the design of multivalent antagonists and
a compelling target for the design of prophylactic drugs against
the initial, acute, result of infection by Vibrio cholerae. The
same mode of binding to GM1 is exhibited by the closely related
Escherichia coli heat-labile enterotoxin, the causative agent of
traveler’s diarrhea. We have undertaken the design of multi-
valent receptor-binding antagonists against cholera toxin and
heat-labile enterotoxin, with a particular focus on exploiting the
5-fold symmetry of the binding sites on the toxin B pentamer.
Previous designs for multivalent ligands have usually em-
ployed a generic scaffold, to which are attached many copies
of a chemical moiety with relatively poor individual affinity
for the target site.1-15 In such cases the resulting geometry of
the effective local concentration of the monovalent ligand.
Furthermore, for highly multimeric ligands it is difficult to
distinguish ligand-mediated aggregation of the protein from an
actual gain in effective affinity. In contrast, if the multivalent
presentation of the liganding group is geometrically restrained
to match the specific arrangement of binding sites on the target
protein, then a substantial additional increase in affinity is to
(2) Gordon, E. J.; Sanders, W. J.; Kiessling, L. L. Nature 1998, 392, 30-31.
(3) Reuter, J. D.; Myc, A.; Hayes, M. M.; Gan, Z.; Roy, R.; Qin, D.; Yin, R.;
Piehler, L. T.; Esfand, R.; Tomalia, D. A.; Baker, J. R. J. Bioconjug. Chem.
1999, 10, 271-278.
(4) Thoma, G.; Duthaler, R. O.; Magnani, J. L.; Patton, J. T. J. Am. Chem.
Soc. 2001, 123, 10113-10114.
(5) Kramer, R. H.; Karpen, J. W. Nature 1998, 395, 710-713.
(6) Profit, A. A.; Lee, T. R.; Lawrence, D. S. J. Am. Chem. Soc. 1999, 121,
280-283.
(7) Glick, G. D.; Toogood, P. L.; Wiley, D. C.; Skehel, J. J.; Knowles, J. R.
J. Biol. Chem. 1991, 266, 23660-23669.
(8) Kanai, M.; Mortell, K. H.; Kiessling, L. L. J. Am. Chem. Soc. 1997, 119,
9931-9932.
(9) Mammen, M.; Dahmann, G.; Whitesides, G. M. J. Med. Chem. 1995, 38,
4179-4190.
(10) Connolly, D. T.; Townsend, R. R.; Kawaguchi, K.; Bell, W. R.; Lee, Y.
C. J. Biol. Chem. 1982, 257, 939-945.
* Corresponding author. E-mail: hol@gouda.bmsc.washington.edu.
† Department of Biochemistry and Biomolecular Structure Center.
‡ Department of Biological Structure and Biomolecular Structure Center.
§ Department of Chemistry.
(11) Lee, Y. C. FASEB J. 1992, 6, 3193-3200.
(12) Kingery-Wood, J. E.; Williams, K. W.; Sigal, G. B.; Whitesides, G. M. J.
Am. Chem. Soc. 1992, 114, 7303-7305.
(13) Spevak, W.; Nagy, J. O.; Charych, D. H.; Schaefer, M. E.; Gilbert, J. H.
J. Am. Chem. Soc. 1993, 115, 1146-1147.
| Howard Hughes Medical Institute.
(1) Mammen, M.; Choi, S. K.; Whitesides, G. M. Angew. Chem., Int. Ed. 1998,
37, 2755-2794.
(14) Thompson, J. P.; Schengrund, C.-L. Glycoconj. J. 1997, 14, 837-845.
(15) Roy, R. Curr. Opin. Struct. Biol. 1996, 6, 692-702.
9
8818
J. AM. CHEM. SOC. 2002, 124, 8818-8824
10.1021/ja0202560 CCC: $22.00 © 2002 American Chemical Society

Multivalent Inhibitor for Cholera Toxin
A R T I C L E S
In this report, we describe solution studies addressing two
important issues in structure-based multivalent ligand design.
The first question is whether proportional affinity gains can be
realized in a multivalent ligand if it is built from better
monovalent ligands. The second question is whether discrete
complexes, rather than random aggregates, are formed in
solution between the designed multivalent ligand and its target
protein. The solution studies point to the absence of multipen-
tamer aggregates in the presence of the pentavalent inhibitor.
We also report a high-resolution crystal structure of the complex
between our designed pentavalent ligand and the B pentamer
of cholera toxin. The crystal structure both supports the solution
study results and highlights new opportunities for the design of
improved pentavalent ligands by revealing interactions between
the first linkage group and the toxin’s surface.
Figure 1. Conceptual design for the pentavalent ligand, showing modular
design based on symmetric core, variable numbers of linker units, and
monovalent “fingers” that block the toxin’s receptor binding site. The toxin
B pentamer target structure is shown with galactose (black) bound at each
of the five receptor binding sites.
be expected. In such cases the binding of a ligand to one site
partially orients and positions additional copies of the ligand to
bind at other sites. A portion of the net entropic cost due to the
loss of rotational and translational freedom of the ligand in
solution is thus paid once for the entire set of multiple binding
events, rather than being paid in full for binding at each site.
The expected decrease in the entropic cost of binding for a rigid
multivalent ligand should lead to a dramatic increase in affinity
relative to both free, monovalent, ligands and to randomly
presented multivalent ligands such as those based on dendrimers
or linear polymers.
Results
Synthesis. The synthesis of the monovalent and pentavalent
ligands is shown in Scheme 1. The acid-bearing MNPG
analogue 1 was synthesized according to previously reported
procedures20 using galactose pentaacetate and 3-hydroxy-5-
nitrobenzoic acid. One mono-Boc protected linker fragment 2
was attached to 1 through amide bond formation.19 Subsequent
acetate removal and enzymatic digestion produced 3 in the pure
R-anomeric form.20 TFA deprotection produced the soluble
monovalent ligand 4. The synthesis of the pentavalent ligand 7
was again straightforward through activation of 4 by dimethyl
squarate and subsequent reaction with the core-linker module
6, following our established synthetic route.19 Although ideally
one would like to obtain a pentavalent ligand with the optimal
linker length as revealed in our previous study,19 the poor
solubility associated with longer MNPG-based finger + linker
units during synthesis prevented us from obtaining the final
product for longer linkers. As a consequence, only ligand 7 with
two basic linker units was prepared.
Inhibition. The affinity of the ligands toward toxin B
pentamer was measured in terms of ability to inhibit receptor
binding (IC50) using a cholera toxin B pentamer (CTB) enzyme-
linked adhesion assay.21 As listed in Table 1, the pentavalent
ligand 7 showed about 260-fold affinity enhancement over the
monovalent ligand 4. This enhancement is very close to the
equivalent gain when a galactose-based finger is used in the
equivalent pentavalent ligand.19
Light Scattering. To further study the state of pentavalent
ligand-toxin B pentamer complexes in solution, we performed
dynamic light scattering (DLS) experiments with varying
concentration ratios of ligand 7 and the E. coli heat-labile
enterotoxin B pentamer (LTB), at concentrations well above
the measured IC50 levels. The high concentration ensures that
ligand-protein complex formation is maximized. DLS measures
the distribution of the hydrodynamic radius (Rh) of solution
species, making it a useful tool for determining aggregation
behavior in solution. The DLS measurements can distinguish
whether the primary mode of interaction in solution is aggrega-
tion or the formation of a 1:1 complex. In the latter case one
would observe little change in Rh as the ligand is added to the
free toxin B pentamer,19 while in the case of aggregation one
Multivalent inhibition of AB5 toxins has previously been
investigated by several groups. Presentation of four to eight
copies of the full GM1 receptor oligosaccharide via poly-
(propylene imine) dendrimers yielded up to a 1000× gain in
IC50 for cell binding by cholera toxin.14,16 Kitov et al.17 designed
a decavalent inhibitor for Shiga toxin, with over a millionfold
gain in affinity, based on an asymmetric scaffold. Shiga toxin
presents three receptor binding sites per B monomer, of varying
affinity, and the intent was to block up to 10 of these sites upon
binding a single decavalent inhibitor. In this case, however, the
toxin:inhibitor complex was found to consist of two toxin
pentamers sandwiching a single inhibitor, with the five highest
affinity binding sites occupied on each of the two toxin
pentamers.17 The expected gain in affinity due to multivalent
presentation of a ligand to a pentavalent target has been modeled
by Gargano et al.18 and quantified experimentally for linear
polymers derived from the Shiga toxin receptor trisaccharide.
The observed gain of >5 × 103 compared well with the
predicted gain of >104.18 A gain of >104 was also found by
Fan et al.,19 who explored the effect of optimizing the effective
radius of a symmetric pentavalent inhibitor for cholera toxin
and E. coli heat-labile enterotoxin using a modular design with
multiple flexible linker units. This same design has also been
chosen for the work presented here and is used to link five
single-site antagonists to a central pentacyclen core. The linkages
in the current design are not rigid, leading to a star-shaped
molecule with five flexible arms, each arm terminating in a
single “finger” which binds to a GM1 binding site on the toxin
B pentamer (Figure 1).
(16) Thompson, J. P.; Schengrund, C. L. Biochem. Pharmacol. 1998, 56, 591-
597.
(17) Kitov, P. I.; Sadowska, J. M.; Mulvey, G.; Armstrong, G. D.; Ling, H.;
Pannu, N. S.; Read, R. J.; Bundle, D. R. Nature 2000, 403, 669-672.
(18) Gargano, J. M.; Ngo, T.; Kim, J. Y.; Acheson, D. W. K.; Lees, W. J. J.
Am. Chem. Soc. 2001, 123, 12909-12910.
(20) 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
(21) Minke, W. E.; Roach, C.; Hol, W. G. J.; Verlinde, C. L. M. J. Biochemistry
1999, 38, 5684-5692.
(19) Fan, E.; Zhang, Z.; Minke, W. E.; Hou, Z.; Verlinde, C. L. M. J.; Hol, W.
G. J. J. Am. Chem. Soc. 2000, 122, 2663-2664.
9
J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002 8819

A R T I C L E S
Merritt et al.
Scheme 1. Pentavalent Ligand Synthesis
Table 1. Inhibition of Receptor Binding
Table 3. Crystallographic Data
gain in inhibition
compared to
single finger
data collection
IC50a (esd)
(µM)
resolution (highest shell)
unique data measured
completeness
50-1.46 Å (1.51-1.46Å)
84 902 (7 083)
97% (82%)
compound
galactose-based finger19
5000 (200)
16 (8)
195 (96)
0.9 (0.2)
R
merge on intensities
0.081 (0.475)
galactose-based pentavalent ligand19
MNPG-based finger 4
312
263
model
R
Rfree
0.149
0.191
MNPG-based pentavalent ligand 7
4098, Biso ) 14 Å2, A ) 0.50
120, Biso ) 21 Å2, A ) 0.59
105, Biso ) 34 Å2
567, Biso ) 32 Å2, A ) 0.51
a IC50 values given are a weighted average, with the corresponding
estimated standard deviation, derived from either three or four replicate
ELISA experiments.
protein atoms
ligand atoms (finger)
ligand atoms (linker)
water molecules
Table 2. Dynamic Light Scattering
data used in refinement
reflections (working set)
reflections (Rfree set)
cutoffs
80 506
4 244
27-1.46 Å
solution components
R (δ) (nm)
h
MW (kDa)
LTB (11.06 µM)
3.191 (0.251)-3.442 (0.454)
3.186 (0.254)-3.338 (0.393)
3.135 (0.276)-3.226 (0.227)
3.186 (0.254)-3.311 (0.114)
12.15 (7.27)
54
54
51
54
1160
7:LTB (11.06:22.12 µM)
7:LTB (11.06:11.06 µM)
7:LTB (22.12:11.06 µM)
CPRG:LTB19
stereochemistry
rms nonideality bond lengths
rms nonideality 1-3 lengths
overall coordinate ESU
(Cruickshank DPI)
0.019 Å
0.039 Å
0.08 Å
overall coordinate ESU
(maximum likelihood)
0.04 Å
would observe scattering from much larger particles. The DLS
experiments were carefully performed by separately prefiltering
ligand and protein stock solutions prior to mixing. Then
appropriate volumes of ligand and protein solutions were mixed
to reach the desired concentration and ratio of ligand to protein
in solution. These samples were then used directly for measure-
ment without further treatment or filtration. This eliminated the
potential for removal or breakup of aggregates if they were ever
formed in substantial quantities. The results from DLS are listed
in Table 2. As it clearly shows, the Rh measured for ligand-
protein mixtures is very close to that of a single toxin B
pentamer, regardless of the ratio of ligand to protein in solution.
Therefore, one can safely conclude that, indeed, formation of a
1:1 complex between the pentavalent ligand 7 and the toxin B
pentamer is the major mode of association in solution.
Table 2 shows the hydrodynamic radius (Rh) as measured by
DLS for varying concentration ratios of the E. coli heat-labile
enterotoxin B pentamer (LTB) and the pentavalent ligand 7.
The mean Rh along with the associated distribution width δ
(polydispersity) is shown as the total range of values obtained
from four replicate experiments. The corresponding calculated
molecular weight is an approximation based on the assumption
of scattering by spherical particles, MW ) (1.68Rh)2.3398. The
final row in Table 2 is a positive control consisting of a mixture
of LTB with a monovalent ligand (CPRG) known to cause
aggregation (Wendy Minke, personal communication). Unlike
the preparation of the 7:LTB mixtures, the CPRG:LTB prepara-
tion was filtered after mixing to remove visible particulate
precipitation; thus the measured Rh is an underestimate of the
extent of aggregation.
Crystal Structure. Crystals of the toxin:inhibitor 7 complex
are isomorphous to those of the toxin:receptor complex, and
similarly diffract to near-atomic resolution.22 Refinement of the
present structure against 1.46 Å data yielded an excellent
crystallographic model (Table 3), and allowed modeling of a
large portion of the inhibitor. The asymmetric unit of the crystal
contains one toxin pentamer and one pentavalent inhibitor. The
five crystallographically independent receptor binding sites are
each to be occupied by one finger of the pentavalent inhibitor.
The terminal phenyl-R-D-galactoside moiety adopts in each case
(22) Merritt, E. A.; Kuhn, P.; Sarfaty, S.; Erbe, J. L.; Holmes, R. K.; Hol, W.
G. J. J. Mol. Biol. 1998, 282, 1043-1059.
9
8820 J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002

Multivalent Inhibitor for Cholera Toxin
A R T I C L E S
toxin pentamer is paired with the receptor-binding surface of a
crystallographically related pentamer. The gap between such
pentamer pairs in the crystal is large enough to accommodate
two pentacyclen cores, as required by the observed fully
occupied receptor-binding sites. The central pores of these
crystallographically related pentamer pairs are not perfectly
aligned, but they are close enough that we cannot rule out a
binding mode in which one or more monovalent arms from each
toxin:ligand complex swap reciprocally with arms from the
paired complex. Such swapping would be stochastic, in that it
would vary from one unit cell to the next in the crystal, and
could contribute to the lack of clear electron density for the
first set of linking groups attached to the pentacyclen core.
Nevertheless, the simplest interpretation of the well-defined
portions of the toxin:ligand complex in the crystal is that there
is 1:1 association of pentavalent inhibitor to toxin pentamer.
This is borne out by the absence of higher order assemblies of
toxin:ligand multimers seen in solution by dynamic light
scattering.
Discussion
The binding interaction between cholera toxin and the
branched oligosaccharide of ganglioside GM1 has been studied
crystallographically at high resolution,22,27 revealing that the
receptor binding site on the toxin is relatively broad and shallow.
This makes it a difficult target for inhibitor design (see Figure
4 in Pickens et al.).20 The pentasaccharide of the natural receptor,
ganglioside GM1, is a fairly rigid molecule whose interaction
with the toxin buries 400 Å2. The terminal galactose residue of
GM1 is the most deeply buried portion of the receptor after
binding, and all of the more than 25 receptor fragments and
galactose-based inhibitors investigated crystallographically are
observed to bind with the galactose moiety in the same position
and conformation. Thus there is good reason to expect that, as
long as the terminal galactose is retained, chemical variation
of the linker and core moieties of a multivalent inhibitor will
not perturb the fundamental binding mode of the fingers.
The starting point for our inhibitor design was therefore the
identification and characterization of small galactose derivatives
that bind to the toxin as single-site inhibitors.21 These are poor
competitive inhibitors for the natural receptor, but interpretation
of their relative affinities as correlates of the detailed molecular
interactions observed crystallographically allowed us to identify
key structural features. Among these is an extensive hydrogen
bonding network involving side chains from the toxin, particu-
larly from residues in the 50-60 loop, and a well-defined set
of tightly associated water molecules at the toxin surface.23 In
particular we found improved affinity from R-anomeric sugar
substituents and from displacement of one or more of the tightly
associated water molecules (Figure 2). Both of these features
are exhibited by the compound m-nitrophenyl-R-D-galactoside
(MNPG),24,26 and the present crystal structure of the 7:CTB
complex shows that these favorable binding interactions are
retained by a pentavalent receptor antagonist presenting five
MNPG moieties linked to a central 5-fold symmetric core.
Structure-guided inhibitor design based on derivatizing the
anomeric oxygen of the galactose has yielded in the best instance
to date an affinity of Kd = 10 µM.20 This is >103 better than
Figure 2. Binding mode of monovalent receptor inhibitor MNPG (yellow),
showing R-anomer of galactose and canonical tightly associated water
molecules including the water at site 2 that is displaced by the nitrophenyl
group of the ligand.23,24 Superposition (green) of the crystallographically
observed conformation of one finger of the pentavalent inhibitor 7.
the favorable bound conformation that was our design target,
analogous to that seen for MNPG. The water molecule tightly
associated with other CTB:sugar complexes through hydrogen
bonding to the amide hydrogen of residue Gly 33 and to the
hydroxyl O6 of galactose is displaced by one oxygen of the
nitrophenyl (Figure 2).23,24 This was by no means guaranteed
to be the case, as earlier attempts to vary or extend the core
phenyl-R-D-galactoside of MNPG resulted in other, less favor-
able, binding modes.25,26 Difference electron density for the first
of the two linker units attached to each finger is present at each
site and can be modeled up to the first squaric acid moiety
(Figure 3). Density for this portion of the inhibitor at the five
crystallographically independent sites is consistent with a single
favored binding mode. The root-mean-square deviation of
refined positions for the inhibitor atoms, based on least-squares
superposition of CR atoms in the five corresponding toxin
monomers, was 0.25 Å. The second linker group and the
pentacyclen core of the pentavalent ligand are not visible in
the crystal structure, but the dimensions of a simple model are
appropriate to fit both the CTB pentamer and the surrounding
crystal lattice (Figure 4).
The crystal structure is indicative of 1:1 binding of the
pentavalent ligand to the 5-fold symmetric CTB pentamer. The
crystal packing is such that the receptor-binding surface of each
(23) Merritt, E. A.; Sixma, T. K.; Kalk, K. H.; van Zanten, B. A. M.; Hol, W.
G. J. Mol. Microbiol. 1994, 13, 745-753.
(24) Merritt, E. A.; Sarfaty, S.; Feil, I. K.; Hol, W. G. J. Structure 1997, 5,
1485-1499.
(25) Minke, W. E.; Pickens, J.; Merritt, E. A.; Fan, E.; Verlinde, C. L. M. J.;
Hol, W. G. J. Acta Crystallogr., Sect. D 2000, D56, 795-804.
(26) Fan, E.; Merritt, E. A.; Zhang, Z.; Pickens, J.; Roach, C.; Ahn, M.; Hol,
W. G. J. Acta Crystallogr., Sect. D 2001, D57, 201-212.
(27) Merritt, E. A.; Sarfaty, S.; van den Akker, F.; L’hoir, C.; Martial, J. A.;
Hol, W. G. J. Protein Sci. 1994, 3, 166-175.
9
J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002 8821

A R T I C L E S
Merritt et al.
Figure 3. Stereopair of the finger and first linker unit of 7 fit into electron density at one of the five identical binding sites in the crystal structure of the
CTB:7 complex. Experimental electron density is contoured at 1.5σ (gray) and 3.0σ (green) in a σA-weighted (mFo - Fc) difference map calculated before
addition of any ligand atoms to the crystallographic model.
Figure 4. Molecular surface of the CTB pentamer shown with a composite
model of the modular components of the pentavalent inhibitor 7. The finger
+ linker substructure observed in the crystal structure is shown in green;
an energy-minimized model for the pentacyclen core is shown in red; a
Figure 5. Extrapolation of the affinity gain that may be achieved by
second linker unit is shown in purple in a fully extended conformation to
further optimization of the pentavalent MNPG-based inhibitor. The
indicate the maximum distance spanned by a single linker unit. Note that
improvement in IC50 gained by increasing the number of linker modules
the location of the squaric acid moiety belonging to the first linker unit,
in a galactose-based inhibitor is shown in the upper line of the figure.19
although shown here in green, is poorly indicated by experimental elec-
The best affinity was found for a construct containing four linker modules,
tron density as compared to the rest of the finger + linker substructure
corresponding to an expected mean ligand dimension matching that of the
(Figure 3).
target toxin B pentamer. The MNPG-based pentavalent inhibitor reported
here contains two linker modules, corresponding to a mean ligand dimension
the affinity of galactose itself, but still many orders of magnitude
worse than that of the natural receptor. Thus there is at present
only a remote prospect of finding a simple galactose derivative
with sufficient single-site affinity to be medically relevant. These
compounds are nonetheless of interest as candidates for mul-
tivalent presentation. The present structure bears this out,
showing that our five-armed core + linker + finger construct
retains the desired favorable binding mode of the single-site
inhibitor MNPG (Figure 2). Furthermore, we have now shown
that improved affinity in a single-site ligand confers a parallel
improvement in the affinity of a pentavalent construct presenting
five copies of this ligand. Pentavalent presentation of â-galactose
in a symmetric construct with two linker units resulted in a 310-
fold increase in affinity over a single â-galactose based finger-
linker unit.19 Pentavalent presentation of MNPG in a symmetric
construct with two linker units results in a 260-fold increase
compared to a single MNPG-finger-linker unit (4) (Table 1).
In an earlier paper we showed that the affinity of a pentavalent
receptor antagonist for the E. coli heat-labile enterotoxin B
pentamer (LTB) was strongly dependent on the effective length
in solution that is smaller than the predicted optimum. If a similar
dependence of IC50 on total linker length holds for the MNPG-based
inhibitor, then further improvement should be possible, as indicated by the
second line shown running parallel to the line fit to previously observed
data.19
in solution of the five arms.19 In that pentavalent design, based
on a much weaker single-site ligand, the optimal number of
linker units was shown to be four, corresponding to an effective
solution radius of the pentavalent inhibitor that matches the
distance between nonadjacent receptor binding sites on the toxin
pentamer (Figure 5). By comparison the pentavalent construct
characterized here has only two linker units per arm, and is
thus too short for optimal complementarity to the target set of
binding sites on the protein. The present two-linker construct
with fingers based on MNPG already exhibits 100-fold greater
inhibition of toxin:receptor binding than does the analogous two-
linker pentavalent ligand based on galactose fingers (Table 1).
We therefore predict that optimization of the arm length, using
modified linker units with greater solubility, would produce a
further 100-fold gain in affinity, parallel to the gain previously
9
8822 J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002

Multivalent Inhibitor for Cholera Toxin
A R T I C L E S
seen for the galactose-based construct, bringing the net IC50
into the subnanomolar range (Figure 5).
designed that would optimize these interactions with the protein.
Ideally this would both yield enthalpic gains from the specific
interactions and yield entropic gains due to a smaller loss of
entropy arising from the decreased number of rotatable bonds
in the unbound ligand. It is also possible that a modified core
moiety, replacing 6 in the current scheme, can be designed to
interact favorably with the toxin surface in the region of the
central pore (Figure 4). The molecular weight and physical
dimensions of such a fully optimized pentavalent ligand will
be quite large, but the site of biological action, epithelial cell
surfaces in the intestinal lumen, means that ligand size is not
by itself a problem. A receptor-binding antagonist for these
toxins should be orally delivered but need not, and ideally would
not, be transported into the bloodstream. This is in distinction
to the more typical requirement that a drug enter the bloodstream
and be delivered to intracellular targets elsewhere in the body.
Thus multivalent approaches to toxin inhibition such as de-
scribed here may one day lead to prophylaxis of cholera and E.
coli induced traveler’s diarrhea.
Characterization of the specific mode of interaction between
a multivalent ligand and its target protein is of great theoretical
interest, yet can pose a great challenge. For example, when a
multivalent ligand exhibits increased activity compared to a
monovalent ligand, it can be difficult to prove whether this gain
is due to an increase in intrinsic affinity or due to other effects
such as aggregation, precipitation, or the combination of several
factors. This is especially true when the gain is characterized
through inhibition assays rather than by direct measurement of
affinities at equilibrium. Therefore, it is not surprising that there
is still active debate about the true gain in intrinsic affinity from
multivalency.28,29 Any heterogeneity of the multivalent ligand
itself adds to the difficulty of interpreting assay results. Thus
conventional multivalent approaches based on polymers,4,8,9,30
dendrimers,3,14-16 or other nonhomogeneous backbones1,12,13
suffer from a lack of precise control over the number and the
geometry of monovalent ligands attached to each backbone
molecule. In contrast, a modular multivalent approach such as
we report here enables the synthesis and isolation of a single
species multivalent ligand. We are therefore able to use a variety
of biophysical tools to study the solution behavior of the ligand-
protein complex. We chose to use DLS to study the interaction
between 7 and toxin B pentamer in solution because DLS is
very sensitive to aggregate formation. The DLS experiments
provide crucial evidence on the nature of complexes formed in
solution between the pentavalent ligand 7 and toxin B pentamer.
As listed in Table 2, there is no sign of significant formation of
large aggregates or precipitation in this pentavalent ligand-
toxin B pentamer system under a variety of conditions. Over a
range of ligand:toxin concentration ratios from 1:2 to 2:1, the
observed effective hydrodynamic radius of the complex does
not significantly vary (Table 2). This is strong evidence that
1:1 complex formation between ligand 7 and the toxin B
pentamer is the major event in solution at micromolar concen-
tration or lower. Therefore, one can safely conclude that the
gain in inhibition potency measured in our competitive inhibition
assay is due not to aggregation, but rather to a gain in the
intrinsic affinity of the pentavalent ligand for the toxin B
pentamer.
Experimental Section
Synthesis. (A) General. Common solvents (reagent grade or
HPLC grade) were used as purchased from commercial sources with-
out further purification. HPLC purification was performed on an HP
1100 quaternary pump system with a variable wavelength detector. The
C18 preparative column was purchased from Vydac (21 × 250 mm,
10-15 µm). 1H NMR spectra were obtained at 300 MHz on a Bruker
AC-300 instrument, while mass spectra were obtained from a Bruker
Esquire 3000 electrospray ion trap mass spectrometer.
(B) {3-Nitro-5-[3-(2-{2-[3-(t-Boc-amino)propoxy]ethoxy}ethoxy)-
propylaminocarbonyl]phenyl}-r-D-galactopyranoside (3). A general
synthetic procedure is as follows: To a solution containing crude 1
(300 mg, 0.58 mmol)20 in a total of 15 mL of EtOAc, 36 mg (0.19
mmol) of cyanuric chloride was added. After the cyanuric chloride was
completely dissolved, 60 µL (0.58 mmol) of N-methylmorpholine was
added while stirring vigorously. After 2 h, the monoprotected diamine
2 (189 mg, 0.58 mmol) was added dropwise. The mixture was allowed
to stir for 12 h, and then was filtered through a 0.4 µm nylon syringe
filter to remove insoluble byproducts. The solvent was then removed
by rotary evaporation and the crude sugar-protected product 3 brought
up in 10 mL of MeOH. Approximately 10 mg of sodium was added
and the solution was allowed to stir for 2 h, and then passed through
10 mL of Dowex cation-exchange resin (8% cross-linking, 200 mesh)
in the acidic form. HPLC purification (solvent A, 0.1% TFA; solvent
B, CH3CN gradient 0 to 15% B over 5 min, then to 60% B over 25
min) gave 68 mg of a 1:1 anomeric mixture of 3. Yield: 18%. The
anomeric mixture was brought up in 10 mL of 0.1 M phosphate buffer
(pH 7, 1 mM MgCl2) containing 500 units of â-galactosidase. The
enzymatic digest was continued overnight. After removing the enzyme
by filtration, the solution was subjected to HPLC repurification using
the same conditions as above giving 10.5 mg of 3. Recovery: 30%.
An ideal receptor antagonist would exhibit perfect comple-
mentarity to the full set of binding sites on the target; in the
case of cholera toxin and heat-labile enterotoxin, it would
contain five rigidly linked GM1 antagonists, each of which
would bind to one site on the toxin and block toxin:receptor
interactions at that site. Our current work falls short of this ideal,
as the core and linker units that join the binding site ligands
are flexible. Nevertheless, a substantial portion of one linker
unit of each arm is well-ordered in the current 7:CTB complex
(Figures 3 and 4), making van der Waals contact with the side
chains of residues Glu 11, Tyr 12, His 13, Lys 34, and Arg 35.
This may partially explain the increase in IC50 observed for
the MNPG-based finger 4 relative to MNPG itself (Table 1),
and suggests that a rigidified variant of the linker can be
1
ESI-MS: 648.4 [M + H]+. H NMR: d 8.33 (s, 1H), 8.16 (s, 1H),
7.97 (s, 1H), 5.70 (s, 1H), 3.98 (m, 3H), 3.87 (t, 1H), 3.70-3.45 (m,
16H), 3.10 (t, 2H), 1.90 (m, 2H), 1.69 (m, 2H), 1.42 (s, 9H).
(C) {3-Nitro-5-[3-(2-{2-[3-aminopropoxy]ethoxy}ethoxy)prop-
ylaminocarbonyl]phenyl}-r-D-galactopyranoside (4). Compound 3
(21 mg, 32.3 µmol) was treated with 2.0 mL of 1:1 TFA:CH2Cl2 for 5
min at room temperature. After rotary evaporation of solvent, the resi-
due was dissolved in 3 mL of water and neutralized with aqueous
NaOH to pH 2. HPLC purification (solvent A, 0.1% aqueous TFA; B,
CH3CN using same gradient as for compound 3) produced 15 mg of 4.
Yield: 84%. ESI-MS: 548.4 [M + H]+.
(28) Dimick, S. M.; Powell, S. C.; McMahon, S. A.; Moothoo, D. N.; Naismaith,
J. H.; Toone, E. J. J. Am. Chem. Soc. 1999, 121, 10286-10296.
(29) Lundquist, J. J.; Debenham, S. D.; Toone, E. J. J. Org. Chem. 2000, 65,
8245-8250.
(30) Mourez, M.; Kane, R. S.; Mogridge, J.; Metallo, S.; Deschatelets, P.;
Sellman, B. R.; Whitesides, G. M.; Collier, J. R. Nat. Biotechnol. 2001,
19, 958-961.
(D) Pentavalent Ligand (7). Compound 3 (10 mg, 15.4 µmol) was
treated with 1.0 mL of 1:1 TFA:CH2Cl2 for 5 min at room temperature.
9
J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002 8823

A R T I C L E S
Merritt et al.
After rotary evaporation of solvent, the residue was dissolved in ∼1
mL of water and neutralized with aqueous NaOH to pH ∼7. Then a
solution of dimethyl squarate (5.0 mg, 35 µmol) in 1.0 mL of MeOH
was added. This mixture was adjusted to pH 8 by 0.1 M NaHCO3, and
was stirred at room temperature overnight. After neutralization with
0.1% aqueous TFA to pH 7, the mixture was subjected to HPLC
purification (solvent A, H2O; B, CH3CN gradient 0 to 15% B over 15
min, then to 55% B over additional 30 min) to produce 7.2 µmol of
the activated compound 5. Yield: 47%. ESI-MS: 658.5 [M + H]+.
This 7.2 µmol of compound 5 was mixed with 0.66 µmol of
compound 619 in a 4.0 mL solution of 1:1 H2O:MeOH. The pH of the
solution was adjusted to pH 9.0 by 0.1 M NaHCO3 (∼0.3 mL). After
stirring at room temperature for 36 h, the mixture was acidified to pH
2.0 by TFA and subjected to HPLC purification (solvent A, 0.1%
aqueous TFA; B, CH3CN gradient 20 to 50% B over 30 min). A total
of 0.59 µmol of final pentavalent ligand 7 in 3.0 mL of aqueous solution
was obtained. The yield was determined by photometric measurement
of the absorbance at 290 nm (ꢀ290 ) 1.568 × 105 M-1 cm-1). The final
pure compound was kept in aqueous solution at pH 7 and used directly
for assays and cocrystallization experiments. Yield: 89%. ESI-MS:
at 5.0 mg/mL in 100 mM Tris HCl at pH 7.5, 1 µL of 0.17 mM 7 in
the same buffer, and 1 µL well buffer containing 50 mM NaCl, 100
mM Tris HCl at pH 7.5, and 42% PEG 300. Crystals were isomorphous
to the previously determined complex of CTB with the GM1 pentasac-
charide.27
X-ray intensities from a single flash-frozen crystal were measured
using a wavelength of 0.9793 Å at the APS Structural Biology Center
beamline 19ID. The data were integrated and scaled using programs
HKL2000 and TRUNCATE.32,33 Isotropic refinement of the protein and
well-ordered water molecules not in the region of the receptor-binding
site was carried out in REFMAC version 434 using data to 1.5 Å and
a riding hydrogen model. This yielded residuals R ) 0.191 and Rfree
)
0.221. Clear electron density was present in σA weighted difference
maps (mFo - Fc) for the “finger” portion of the inhibitor at all five
binding sites, and fragmentary density for the start of the attached linker
moieties (Figure 3). Refinement was continued with the addition of
individual atomic displacement parameters Uij to the model and
inclusion of data to 1.46 Å, bringing the residuals to R ) 0.165 and
Rfree ) 0.210. At this point the inhibitor fingers (from the sugar to the
amide nitrogen) were built into difference density and added to the
refinement. This yielded residuals R ) 0.148 and Rfree ) 0.191. Portions
of the dilinker groups attached to the amide were modeled into the
remaining difference density at each of the binding sites (Figure 3)
and refined with isotropic thermal parameters, yielding the final model
as described in Table 3.
1052.8 [M + 5H]5+, 1315.7 [M + 4H]4+, and 1754.0 [M + 3H]3+
.
IC50 Measurements. The CTB:GD1b enzyme-linked adhesion
assay was performed as previously reported.21 Test samples consisted
of 6 ng/mL CTB horseradish peroxidase conjugate (Sigma-Aldrich,
Milwaukee, WI) preincubated with desired ligands for 2 h at room
temperature. Independent experiments for each inhibitor were carried
out in triplicates and validated against a concentration gradient of 0,
1.5, 3, 6, and 12 ng/mL toxin peroxidase conjugate. IC50 values were
calculated from triple data sets of at least seven different concentrations
of competitive ligands by nonlinear regression,31 and reported as a
weighted average of three or four independent determinations.
Dynamic Light Scattering (DLS). Protein or ligand in phosphate
buffered saline (150 mM NaCl/10 mM phosphate, pH 7.2) was
individually filtered through inorganic membrane filters (Whatman,
Anodisc13, 0.02 µm) into separate vials. Then various portions of
protein and ligand were mixed and, without further filtration, transferred
into a sample cell for measurement. DLS measurement was done on a
DynaPro99 instrument (Protein Solutions Inc.) illuminated by a 25 mW,
832.8-nm-wavelength, solid-state laser at 25 °C. Data analysis was
performed using the dynamics Version 5.25.44 software provided with
the instrument. Inverse Laplace transform (“regularization fit”) analysis
was used to find the mean and standard deviation (polydispersity) of
the hydrodynamic radius (Rh) distribution for the molecule/complex
species in solution.
E. coli heat-labile enterotoxin B pentamer (LTB) was used for DLS
experiments in the presence or absence of the pentavalent ligand 7.
The DLS measurements were carried out with the following solutions:
(1) LTB (11.06 µM); (2) LTB:7 (11.06:22.12 µM); (3) LTB:7 (11.06:
11.06 µM); (4) LTB:7 (22.12:11.06 µM). Independent experiments were
performed three to four times for each set of solution.
Protein Production. Cholera toxin B pentamer and E. coli heat-
labile enterotoxin B pentamer were separately expressed in E. coli,
using constructs kindly provided by Claudia Roach, and purified by
galactose affinity chromatography essentially as described in Minke
et al.25
Model fitting, placement, and real-space refinement of ligand and
water molecules was carried out using XFIT.35 The choice of restraint
weights for Uij parameters was guided by analysis of the overall
distribution of anisotropy using PARVATI,36 and the anisotropic
treatment of water molecules was additionally restrained toward isotropy
using a local modification to REFMAC.37 The resulting mean value of
atomic anisotropy, A , is given in Table 3 for protein, ligand, and
solvent atoms. No noncrystallographic symmetry restraints were used
during refinement. Figures were generated using RASTER3D,38
MSMS,39 and XFIT.35 Structure factors and the refined model have
been deposited with the Protein Databank (accession code 1LLR).
Acknowledgment. This work was supported by the NIH
(AI34501, AI44954). We thank Claudia Roach for generating
the expression system for CTB, and Wendy Minke and
Christophe Verlinde for stimulating discussions. Use of the
Argonne National Laboratory Structural Biology Center beam-
line 19ID at the Advanced Photon Source was supported by
the U. S. Department of Energy, Office of Biological and
Environmental Research, under Contract No. W-31-109-ENG-38.
JA0202560
(32) Otwinowski, Z.; Minor, W. Methods Enzymol. 1997, 276, 307-326.
(33) Collaborative Computational Project Number 4. Acta Crystallogr., Sect. D
1994, D50, 760-763.
(34) Murshudov, G. N.; Vagin, A. A.; Dodson, E. J. Acta Crystallogr., Sect. D
1997, D53, 240-255.
(35) McRee, D. Practical Protein Crystallography; Academic Press: San Diego,
1993.
(36) Merritt, E. A. Acta Crystallogr., Sect. D 1999, D55, 1109-1117.
(37) Murshudov, G.; Lebedev, A.; Vagin, A. A.; Wilson, K. S.; Dodson, E. J.
Acta Crystallogr., Sect. D 1999, D55, 247-255.
Crystallization and Structure Determination. Crystals of CTB
complexed with 7 grew from sitting drops consisting of 1 µL of protein
(38) Merritt, E. A.; Bacon, D. J. Methods Enzymol. 1997, 277, 505-524.
(39) Sanner, M. F.; Spehner, J.-C.; Olson, A. J. Biopolymers 1996, 38, 305-
320.
(31) Bowen, W. P.; Jerman, J. C. Trends Pharmacol. Sci. 1995, 16, 413-417.
9
8824 J. AM. CHEM. SOC. VOL. 124, NO. 30, 2002