Thermodynamic analysis of mRNA cap binding by the human initiation factor eIF4E via free energy perturbations

Article abstract of DOI:10.1021/ja9064359

Eukaryotic mRNAs are appended at the 5′ end, with the 7-methylguanosine cap linked by a 5′-5′-triphosphate bridge to the first transcribed nucleoside (m7GpppX). Initiation of cap-dependent translation of mRNA requires direct interaction between the cap st

Full text of DOI:10.1021/ja9064359

Published on Web 11/19/2009
Thermodynamic Analysis of mRNA Cap Binding by the
Human Initiation Factor eIF4E via Free Energy Perturbations
†
‡
‡
§
Cristiano R. W. Guimar a˜ es, David J. Kopecky, Jeff Mihalic, Shanling Shen,
§
†
‡
†
Shawn Jeffries, Stephen T. Thibault, Xiaoqi Chen, Nigel Walker, and
,
†
Mario Cardozo*
Departments of Molecular Structure, Medicinal Chemistry, and Oncology, Amgen, Inc.,
1
120 Veterans BouleVard, South San Francisco, California 94080
Received July 30, 2009; E-mail: mcardozo@amgen.com
Abstract: Eukaryotic mRNAs are appended at the 5′ end, with the 7-methylguanosine cap linked by a
′-5′-triphosphate bridge to the first transcribed nucleoside (m7GpppX). Initiation of cap-dependent
5
translation of mRNA requires direct interaction between the cap structure and the eukaryotic translation
initiation factor eIF4E. Biophysical studies of the association between eIF4E and various cap analogs have
7
demonstrated that m GTP binds to the protein ca. -5.0 kcal/mol more favorably than unmethylated GTP.
In this work, a thermodynamic analysis of the binding process between eIF4E and several cap analogs
has been conducted using Monte Carlo (MC) simulations in conjunction with free energy perturbation (FEP)
calculations. To address the role of the 7-methyl group in the eIF4E/m7GpppX cap interaction, binding
7
free energies have been computed for m GTP, GTP, protonated GTP at N(7), the 7-methyldeazaguanosine
7
5
′-triphosphate (m DTP), and 7-deazaguanosine 5′-triphosphate (DTP) cap analogs. The MC/FEP
7
simulations for the GTPfm DTP transformation demonstrate that half of the binding free energy gain of
m GTP with respect to GTP can be attributed to favorable van der Waals interactions with Trp166 and
7
reduced desolvation penalty due to the N(7) methyl group. The methyl group both eliminates the desolvation
penalty of the N(7) atom upon binding and creates a larger cavity within the solvent that further facilitates
7
7
the desolvation step. Analysis of the pair m GTP-m DTP suggests that the remaining gain in affinity is
related to the positive charge created on the guanine moiety due to the N(7) methylation. The charge
provides favorable cation-π interactions with Trp56 and Trp102 and decreases the negative molecular
charge, which helps the transfer from the solvent, a more polar environment, to the protein.
Introduction
complex contains the scaffold protein, eIF4G, and an RNA
helicase, eIF4A. Upon cap binding, eIF4F and another initiation
6
Eukaryotic mRNAs are appended at the 5′ end, with the
-methylguanosine cap linked by a 5′-5′-triphosphate bridge
factor, eIF4B, unwind secondary structure within the 5′ un-
7
translated region of mRNA and enable recruitment of the
1
to the first transcribed nucleoside (m7GpppX). The cap
structure plays a pivotal role in several stages of gene expression,
such as promoting mRNA splicing, facilitating mRNA export
to the cytoplasm, and protecting mRNA against nucleolytic
5
ribosome to the translation start codon.
eIF4E is the least abundant initiation factor; its recruitment
to mRNA is suspected to be the rate-limiting step in new protein
7
synthesis. The accessibility of eIF4E in mammals is regulated
2
degradation. The cap structure is also important for mRNA
by interactions with the small eIF4E-binding proteins, 4E-BP1,
3
stabilization and ribosome recruitment for translation. In cap-
dependent translation, a direct interaction is required between
the cap structure and the eukaryotic initiation factor 4E (eIF4E)
for efficient translation initiation. eIF4E, which is phyloge-
netically highly conserved, is the smallest subunit (25 kDa) of
the eIF4F heterotrimeric complex. In addition to eIF4E, this
4
E-BP2, and 4E-BP3, which prevent productive interactions
8,9
between eIF4E and eIF4G. Phosphorylation of 4E-BPs by
active mTOR, the mammalian target of rapamycin, leads to a
significant loss of their binding ability to eIF4E, which facilitates
4
,5
10
association of eIF4E and eIF4G, and new protein synthesis.
Upregulation of eIF4E cap dependent protein translation is
commonly observed in multiple human tumor types. Activation
of eIF4E lies downstream of growth factor signaling through
†
Department of Molecular Structure.
Department of Medicinal Chemistry.
Department of Oncology.
‡
§
(
1) Niedzwiecka, A.; Marcotrigiano, J.; Stepinski, J.; Jankowska-Anyszka,
M.; Wyslouch-Cieszynska, A.; Dadlez, M.; Gingras, A.-C.; Mak, P.;
Darzynkiewicz, E.; Sonenberg, N.; Burley, S. K.; Stolarski, R. J. Mol.
Biol. 2002, 319, 615–635.
(6) Niedzwiecka, A.; Darzynkiewicz, E.; Stolarski, R. Biochemistry 2004,
43, 13305–13317.
(7) Raught, B.; Gingras, A.-C. Int. J. Biochem. Cell Biol. 1999, 31, 43–
57.
(
(
(
(
2) Furuichi, Y.; Shatkin, A. J. AdV. Virus Res. 2000, 55, 135–184.
3) Miura, K. AdVan. Biophys. 1981, 14, 205–238.
(8) Haghighat, A.; Mader, S.; Pause, A.; Sonenberg, N. Eur. Mol. Biol.
Org. J. 1995, 14, 5701–5709.
4) Pain, V. M. Eur. J. Biochem. 1996, 236, 747–771.
(9) Mader, S.; Lee, H.; Pause, A.; Sonenberg, N. Mol. Cell. Biol. 1995,
15, 4990–4997.
5) Gingras, A.-C.; Raught, B.; Sonenberg, N. Annu. ReV. Biochem. 1999,
6
8, 913–963.
(10) Reiling, J. H.; Sabatini, D. M. Oncogene 2006, 25, 6373–6383.
10.1021/ja9064359  2009 American Chemical Society
J. AM. CHEM. SOC. 2009, 131, 18139–18146 9 18139

A R T I C L E S
Guimar a˜ es et al.
7
Figure 1. Interactions between m GpppA and human eIF4E observed in the crystal structure (Brookhaven Protein Data Bank code 1WKW).
the PI3K-Akt-mTor pathway; signaling via this network is
one with the backbone NH of Trp102, cation-π interactions
with Trp102 and Trp56, and van der Waals interactions between
the N(7)-methyl group and Trp166 (Figure 1). The triphosphate
moiety forms an extensive hydrogen-bonding network with
Arg112, Arg157, Lys162, and water molecules. Finally, the
adenosine moiety interacts with the flexible C-terminal loop
11
frequently upregulated during human tumorigenesis. Overex-
pression of eIF4E causes accelerated cell division, malignant
5
transformation, and inhibition of apoptosis. eIF4E overexpres-
sion occurs up to 30-fold in many cancers, including head and
6
neck, breast, and lung cancers. Levels of eIF4E are also
7
elevated in some colon carcinomas in comparison to normal
region, unobserved for the complex with m GTP.
1
2,13
colon cells.
In certain types of cancer, the relationship
Biophysical studies at 20 °C of the association between
14
1
between 4E-BP1 and eIF4E might be a predictor of metastasis.
murine eIF4E and various cap analogs have been reported.
Both eIF4E and 4E-BP1 are frequently overexpressed in colon
carcinoma, but the 4E-BP1 levels are the most elevated in
There are clearly profound effects on the binding free energies
obtained following modifications of the cap structure. One of
the most striking differences is a 5000-fold drop in the
association constant or a binding free energy ca. 5.0 kcal/mol
1
4
patients that have little or no metastatic disease. This over-
expression of eIF4E observed in several tumor types suggests
that the development of an inhibitor of eIF4E could be a suitable
anticancer therapeutic approach, since it has the potential to
interfere in cap-dependent translation of mRNA and in turn
decrease cell proliferation.
7
less favorable for GTP as compared to m GTP. The only direct
interaction involving the 7-methyl group is a nonspecific van
der Waals contact with Trp166 (Figure 1), suggesting the eIF4E
7
binding enhancement for m GTP might derive from favorable
X-ray crystal structures of human eIF4E and the cap analogs
m GTP and m GpppA as well as the ternary complex of eIF4E,
cation-π interactions between the resultant positive charge at
7
7
7
the m G moiety and the Trp102 and Trp56 residues. Another
7
15,16
m GpppA, and a 4E-BP1 fragment have been determined.
The 4E-BP1 fragment does not significantly affect the overall
contribution for the lower eIF4E affinity of GTP is potentially
associated with its greater desolvation penalty upon binding;
GTP has an additional negative charge when compared to
1
6
tertiary structure and cap-binding scaffold of eIF4E. eIF4E
forms a temple-bell-shaped surface of eight antiparallel ꢀ-struc-
tures, three R-helices, and ten loop structures, where the
N-terminal region corresponds to the handle of the bell. The
m G moiety of m GpppA displays multiple interactions with
eIF4E; two hydrogen bonds with the Glu103 side chain and
7
m GTP. Also, the N(7) atom of the former is available to form
hydrogen-bond interactions with water molecules in the unbound
state.
1
5
7
7
In this work, a thermodynamic analysis of the eIF4E binding
17
process has been conducted using Monte Carlo (MC) simula-
tions in conjunction with free energy perturbation (FEP)
1
8-21
calculations.
More specifically, the relative eIF4E binding
(
(
11) Easton, J. B.; Houghton, P. J. Oncogene 2006, 25, 6436–6446.
12) Rosenwald, I. B.; Chen, J. J.; Wang, S.; Savas, L.; London, I. M.;
Pullman, J. Oncogene 1999, 18, 2507–2517.
7
free energy between m GTP and GTP was computed. To
understand the role of the 7-methyl group in the greater affinity
of m GTP, relative binding free energies were also computed
(
(
(
13) Berkel, H. J.; Turbat-Herrera, E. A.; Shi, R.; de Benedetti, A. Cancer
Epidemiol. Biomarkers PreV. 2001, 10, 663–666.
7
14) Martin, M. E.; Perez, M. I.; Redondo, C.; Alvarez, M. I.; Salinas, M.;
Fando, J. L. Int. J. Biochem. Cell Biol. 2000, 32, 633–642.
15) Tomoo, K.; Shen, X.; Okabe, K.; Nozoe, Y.; Fukuhara, S.; Morino,
S.; Sasaki, M.; Taniguchi, T.; Miyagawa, H.; Kitamura, K.; Miura,
K.-I.; Ishida, T. J. Mol. Biol. 2003, 328, 365–383.
(17) Allen, M. P.; Tildesley, D. J. Computer Simulations of Liquids;
Clarendon Press: Oxford, U.K., 1987.
(18) Zwanzig, R. J. Chem. Phys. 1954, 22, 1420–1426.
(19) Jorgensen, W. L.; Ravimohan, C. J. Chem. Phys. 1985, 83, 3050–
3056.
(
16) Tomoo, K.; Matsushitaa, Y.; Fujisakia, H.; Abiko, F.; Shena, X.;
Taniguchib, T.; Miyagawac, H.; Kitamurac, K.; Miurad, K.-I.; Ishidaa,
T. BBA 2005, 1753, 191–208.
(20) Jorgensen, W. L. Acc. Chem. Res. 1989, 22, 184–189.
(21) Kollman, P. A. Chem. ReV. 1993, 93, 2395–2417.
1
8140 J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009

FEP Simulations of mRNA Cap Binding by eIF4E
A R T I C L E S
Figure 2. (a) Thermodynamic cycles used for the calculation of relative free energies. ∆Ghyd, ∆Gtransf, and ∆Gbind are the absolute free energies of hydration,
are the free energy changes for the transformation of ligand A into B
g w p
transfer from the gas phase to the solvated protein, and binding. ∆G , ∆G , and ∆G
in the gas phase, water, and protein. (b) Transformation sequence used in the FEP simulations.
+
for protonated GTP at N(7) (GTPH ) and the 7-methyldeaza-
and transfer from the gas phase to the protein, thus facilitating
7
guanosine 5′-triphosphate (m DTP) and 7-deazaguanosine 5′-
the rationalization of the affinity order. The required ∆G(AfB)
terms are computed by transforming A into B in the different
environments through the FEP methodology using the sequence
triphosphate (DTP) cap analogs. In addition, in order to verify
the influence of hydration on the binding process, the relative
free energies of hydration for the analogs were calculated. The
experimental result for nucleotide analogs with a positive charge
and neutral core validates our hypothesis of the various binding
contributions. This in turn provides insight into ligand design
aimed to develop an eIF4E inhibitor with potential application
in anticancer therapy.
1
9
shown in Figure 2b.
∆∆Gbind ) ∆Gbind(B) - ∆Gbind(A) ) ∆G (A f B) -
p
∆
G (A f B) ) ∆∆G
- ∆∆Ghyd (3)
transf
w
Protein-Ligand Complexes. The crystal structure for the 2.1
7
Computational Details
Å ternary complex between human eIF4E, m GpppA, and the
4
E-BP1 peptide (only the Pro47-Pro66 sequence of human 4E-
Free Energy Changes. Figure 2 illustrates the thermodynamic
2
2
BP1 was determined) was employed (Brookhaven Protein Data
Bank code 1WKW). All cocrystallized water molecules as
well as 4E-BP1 were removed from the complex. m GpppA
was converted into m GTP by simple deletion of adenosine.
The segment from N-terminal to Gln26 was not determined
experimentally because of low electron density. As this region
is very far from the binding site, it was not considered in the
cycles used to calculate the free energy changes. Since free
energy is a thermodynamic state function, the cycle on the left
side of Figure 2a gives the relative free energy of hydration
1
6
7
7
(
eq 1), where A and B are any two analogs, ∆Ghyd is the absolute
free energy of hydration, and ∆G (AfB) and ∆G (AfB) are
w
g
16
the free energies associated with the transformation of ligand
A into B in water and in the gas phase, respectively.
7
model, which only contains residues 27 to 217 plus m GTP. A
restrained conjugate-gradient energy minimization for the
complex between eIF4E and m GTP was performed to alleviate
∆
∆Ghyd ) ∆Ghyd(B) - ∆Ghyd(A) ) ∆G (A f B) -
w
7
∆
G (A f B) (1)
g
bad contacts in the crystal structure. The energy minimization
step was stopped when a backbone deviation of 0.3 Å with
respect to the starting structure was reached. Degrees of freedom
for the protein backbone atoms were not sampled in the MC
simulations. Only the bond angles and dihedral angles for the
The cycle on the right-hand side of Figure 2a gives the relative
free energy of transfer from the gas phase to the solvated protein
between A and B. In eq 2, ∆Gtransf is the absolute free energy
of transfer and ∆G
the transformation of A into B in the solvated eIF4E.
p
(AfB) is the free energy associated with
7
side chains of residues with any atom within 10 Å from m GTP
were varied. The ligands, however, are fully flexible in the MC
∆
∆Gtransf ) ∆Gtransf(B) - ∆Gtransf(A) ) ∆G (A f B) -
2
simulations except as specified below. The pK value of the
p
7
1
m GTP γ-phosphate group is close to 7. As the association
∆
G (A f B) (2)
g
7
between m GTP and eIF4E is practically the same for pH values
1
Finally, relative binding free energies, ∆∆Gbind, can be
ranging from 6.0 to 8.5, the γ-phosphate group was considered
rewritten as the difference between ∆∆Gtransf and ∆∆Ghyd as
shown in eq 3, where ∆Gbind is the absolute binding free energy.
In this manner, the binding process is separated into dehydration
here in its monoanionic state. Thus, partial atomic charges
7
+
totaling -2 e for m GTP and GTPH and -3 e for GTP, DTP,
7
23
and m DTP were computed using the CM1A procedure.
(
22) Wong, C. F.; McCammon, J. A. J. Am. Chem. Soc. 1986, 108, 3830–
832.
(23) Storer, J. W.; Giesen, D. J.; Cramer, C. J.; Truhlar, D. G. J. Comput-
Aided Mol. Design 1995, 9, 87–110.
3
J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009 18141

A R T I C L E S
Guimar a˜ es et al.
9
The preparation of the enzymatic system was much facilitated
∆G
p
(AfB) values and 0.9 × 10 configurations per ∆G
g
(AfB)
2
4
+
by use of the Chop delegate program. Charge neutrality for
the protein was imposed by assigning normal protonation states
at physiological pH to basic and acidic residues near the active
site and making the adjustments for neutrality to the most distant
residues. As for the protonation and tautomeric states for
histidine residues, the defaults in the Chop software were
employed. The entire system was solvated with a 22 Å radius
water cap consisting of 852 molecules. A half-harmonic potential
value. Since no cutoffs were used for the GTPH fGTP
6
transformation, 5 × 10 configurations of full equilibration and
6
10 × 10 configurations of data collection were applied for each
window in water or protein to reduce the computational cost
for these simulations. This is possible because only the partial
charges (the hydrogen atom connected to the N(7) atom has no
Lennard-Jones parameters) are perturbed in this transformation,
providing a very fast convergence for the free energy values
obtained. The independent FEP simulations for each pair of
ligands were all performed in the same direction, according to
the convention shown in Figure 2b. The energy-minimized
2
with a force constant of 1.5 kcal/mol ·Å was applied to water
molecules at distances greater than 22 Å from the center of the
2
5
system to discourage evaporation. As discussed previously,
7
the use of a spherical cap of water rather than periodic boundary
conditions affects the calculated free energies of hydration in
simple systems. Therefore, to cancel any potential errors, a 22
Å cap with 1474 water molecules was also used for the AfB
crystal structure for the complex between eIF4E and m GTP
provided the initial configuration for the FEP simulations in all
complex environments. The other five starting configurations
per transformation were randomly selected from the equilibration
phase of each respective first FEP run at a λ value of 0.05. A
similar scheme was applied to the simulations in the gas phase
and water.
w
unbound perturbations, yielding ∆G (AfB).
MC Simulations. The AfB transformations in all environ-
ments were performed using the single topology approach by
melding the force field parameters for bond lengths, bond angles,
3
0
MCPRO 2.0 was used to perform all MC calculations.
1
9
torsions, and nonbonded interactions. In order to keep the
number of atoms constant, dummy (DM) atoms were introduced
for hydrogens that exist in one state and have no counterpart in
the other. Bond lengths for perturbations requiring an atom
mutation or an annihilation or creation of a hydrogen were not
sampled and were treated as geometrical parameters. Bonds
containing a DM atom have an equilibrium value of 0.5 Å.
Shrinking the bond length for DM atoms is a common practice
Established procedures including Metropolis and preferential
sampling were employed, and statistical uncertainties were
obtained from the batch means procedure with batch sizes of 1
× 10 configurations. Attempted moves of the cap analogs in
water occurred every 10 configurations, while, in the complex,
attempted moves of the protein and cap analogs occurred every
10 and 80 configurations, respectively. The TIP4P model was
used for water, and the complexes were represented with the
OPLS-AA force field, with the exception of the CM1A atomic
6
17
31
1
9,26-29
32
to improve convergence in FEP simulations.
The bond
angles involving dummy atoms have the same parameters as
their counterparts in the other state. Associated unphysical
contributions to the free-energy differences cancel in a ther-
charges for the ligands.
Experimental Details
7
2
6,28
In Vitro IC50 Determination of eIF4E m GTP Binding.
modynamic cycle.
The MC/FEP calculations for the AfB
Human eIF4E protein (NP_001959) was produced as an N terminal
FLAG-HIS fusion protein in BL21 bacterial cells and purified by
transformations were executed at 25 °C using double-wide
1
9
sampling. Residue-based cutoffs of 10 Å were employed in
7
m GTP-Sepharose affinity chromatography (GE Healthcare).
+
all transformations except the one from GTPH to GTP. As
7
Protein was eluted with 100 µM m GDP and dialyzed extensively
this transformation involves a charge mutation from -2 to -3
e, no cutoffs were applied. In this manner, since the water-
phase and solvated eIF4E models have equal radii and the same
dielectric constant outside the spherical system, the missing
long-range electrostatic interactions between the solvent mol-
ecules not included in the finite models and the ligands with
into storage buffer. For binding analysis, eIF4E protein (50nM final)
was biotinylated and bound to streptaviden-coated scintillation
3
proximity assay beads (GE Healthcare) in the presence of H-
7
m GTP (1mCi/mL, Moravek Biochemicals) in binding buffer (20
mM HEPES, pH 7.4, 50 mM KCL, 1 mM DTT and 0.5 mM
EDTA). To determine competitive inhibition of eIF4E binding,
compounds were added in an 11-point, 3-fold dilution dose
series, and loss of radioactive signal was monitored by scintil-
lation counting. IC50 values were calculated by nonlinear
regression curve fitting using GraphPad Prism analysis software
different molecular charges would cancel out when computing
+
∆
∆Gbind for the GTPH fGTP transformation.
The initial and final states were coupled using 10 windows
with values for the coupling parameter (λ) evenly distributed
(
GraphPad Software).
Chemical Synthesis. Synthetic route and procedure for the
between 0 and 1 (0.05, 0.15, ..., 0.85, 0.95). Initial relaxation
6
of the solvent was performed for 5 × 10 configurations,
preparation of compounds 1 and 2 are provided in the Supporting
Information.
6
followed by 20 × 10 configurations of full equilibration and
6
3
0 × 10 configurations of data collection for each window in
Results and Discussion
water or protein, whereas for the gas-phase simulations, 5 ×
6
6
Long MC/FEP simulations were performed generating smooth
free-energy curves for the AfB transformations in all environ-
ments (Figure 3). As noted above (eqs 1-3), the binding process
may be decomposed into dehydration and transfer from the gas
1
0 configurations of equilibration and 10 × 10 configurations
of data collection were employed. The ∆G values obtained for
each window are averages of six independent MC runs, giving
9
a total of 3.3 × 10 configurations per ∆G
w
(AfB) and
phase to the protein. The relative free energies, ∆∆Gtransf
∆Ghyd, and ∆∆Gbind, between all cap analogs are reported in
Table 1.
,
∆
(
(
(
(
24) Tirado-Rives, J. Chop; Yale University: New Haven, CT, 2002.
25) Essex, J.; Jorgensen, W. L. J. Comput. Chem. 1995, 16, 9510–972.
26) Boresch, S.; Karplus, M. J. Phys. Chem. A 1999, 103, 103–118.
27) Pearlman, D. A.; Kollman, P. A. J. Chem. Phys. 1991, 94, 4532–
(30) Jorgensen, W. L. MCPRO, Version 2.0; Yale University: New Haven,
CT, 2005.
4
545.
(
(
28) Price, D. J.; Jorgensen, W. L. J. Comput-Aided Mol. Design 2001,
(31) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, W.; Klein,
M. L. J. Chem. Phys. 1983, 79, 926–935.
1
5, 681–695.
29) Guimar a˜ es, C. R. W.; Boger, D. L.; Jorgensen, W. L. J. Am. Chem.
Soc. 2005, 127, 17377–17384.
(32) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. J. Am. Chem. Soc.
1996, 118, 11225–11236.
1
8142 J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009

FEP Simulations of mRNA Cap Binding by eIF4E
A R T I C L E S
Figure 3. Free-energy curves obtained in the gas phase, water, and the protein for all transformations. Error bars representing two standard deviations from
+
the six independent MC runs per window are also plotted. Error bars cannot be visualized for the GTPH fGTP transformation due to the plot scale.
Table 1. Calculated Relative Free Energies of Transfer from the
Gas Phase to eIF4E (∆∆Gtransf), Hydration Free Energies
transformation, cutoffs cannot be used because a charge mutation
from -2 to -3 e is required. However, shorter simulations can
be performed as only the partial charges are mutated; the free
energy values converge very fast in this type of transformation.
(
∆∆Ghyd)a, and Binding Free Energies (∆∆Gbind) between All Cap
Analogs
b
7
+
7
A f B m GTP
GTPH
GTP
DTP
m DTP
+
In addition, GTPH is a valuable molecule to understand the
7
m GTP 0.0 9.0 ( 0.3 -117.9 ( 0.6 -117.5 ( 0.7 -118.9 ( 0.7
7
greater affinity of m GTP, as it has the same net charge and
0
0
.0 0.4 ( 0.4 -124.3 ( 0.6 -123.2 ( 0.7 -122.6 ( 0.7
the ability to establish cation-π interactions with Trp56 and
.0 8.6 ( 0.3
6.4 ( 0.7
5.7 ( 0.8
3.7 ( 0.8
GTPH+
GTP
0.0
-126.9 ( 0.5 -126.5 ( 0.6 -127.9 ( 0.6
-124.7 ( 0.5 -123.6 ( 0.6 -123.0 ( 0.6
Trp102. The charge distributions for m GTP and GTPH are
very similar, with the positive charge resulting from methylation
or protonation delocalized throughout the guanine ring.
7
+
0
.0
.0
0
-2.2 ( 0.6
-2.9 ( 0.7
0.4 ( 0.4
1.1 ( 0.4
-0.7 ( 0.4
0.0
-4.9 ( 0.7
-1.0 ( 0.4
1.7 ( 0.4
-2.7 ( 0.4
-1.4 ( 0.2
0.6 ( 0.2
-2.0 ( 0.2
0.0
0.0
+
0
0
.0
.0
The MC configurations for GTPH in water show that the
protonated N(7) atom forms hydrogen bonds with the solvent.
In the gas phase, the protonated N(7) atom is hydrogen bonded
to the γ-phosphate group. The ∆∆Ghyd value of only 0.4 kcal/
DTP
0
0
.0
.0
7
7
+
m DTP
mol for the m GTPfGTPH transformation (Table 1) indicates
that the more favorable interactions with the solvent for the latter
are offset by the loss of the intramolecular hydrogen bond upon
solvation, resulting in similar hydration free energies for both
analogs. When transferred from the gas phase to eIF4E,
however, the loss of the intramolecular hydrogen bond with the
γ-phosphate group is not counterbalanced, as the protonated
N(7) atom does not form any hydrogen bonds with the protein
0
0
.0
.0
a
Values are in kcal/mol. ∆∆Gtransf, ∆∆Ghyd, and ∆∆Gbind, which are
the first, second, and third values for each comparison in the table, were
obtained according to eqs 1-3. ∆∆Gbind are highlighted in bold for
clarification. The errors in ∆∆G were calculated by propagating the
statistical uncertainties obtained from the batch means procedure with
6
batch sizes of 1 × 10 configurations for the individual ∆G
i
values of
values. For
b
+
each window used to compute the ∆G
each AfB transformation and corresponding relative free-energy values,
p
, ∆G
w
, and ∆G
g
(Figure 4); this reduces ∆Gtransf for GTPH by 9.0 kcal/mol with
7
respect to m GTP. As a consequence, the eIF4E binding for
ligand A is in the table column and ligand B in the table row.
the first is 8.6 kcal/mol less favorable. Thus, even though
+
GTPH is able to establish cation-π interactions with Trp56
7
7
m GTP versus GTP. The transformation from m GTP to GTP
and Trp102, burying the cationic hydrogen-bond donor greatly
+
was performed in two steps, with GTPH being a hypothetical
7
+
penalizes its eIF4E binding with respect to m GTP.
intermediate state between the two analogs (Figure 2). GTPH
The MC simulations show that the N(7) atom of GTP is
hydrogen bonded to the solvent in the water phase. No
intramolecular hydrogen bonds between the N(7) atom and the
hydroxyl of the γ-phosphate group were observed in the gas
was introduced to reduce the computational cost for the
m GTPfGTP transformation, as it avoids the need to run long
7
MC simulations with no cutoffs for nonbonded interactions.
7
+
More specifically, since the m GTPfGTPH transformation
involves the perturbation of bonded terms and Lennard-Jones
parameters, the free energy values converge very slowly and
demand long MC simulations. As the molecular charge is the
same in both states, the computational cost can be reduced by
+
phase. In this manner, and differently from GTPH , GTP does
not have a penalty associated with the loss of intramolecular
hydrogen bonds upon solvation, thus facilitating its hydration
process. In addition, as the electrostatic component of the
solvation free energy is proportional to the square of the
+
employing residue-based cutoffs of 10 Å. In the GTPH fGTP
J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009 18143

A R T I C L E S
Guimar a˜ es et al.
7
+
7
Figure 4. Monte Carlo snapshots illustrating the complexes between eIF4E and m GTP, GTPH , GTP, DTP, and m DTP.
3
3
molecular charge, GTP, with a net charge of -3 e, should be
correction would be obtained for ∆∆Gtransf as the solvated eIF4E
has the same radius and dielectric constant outside the finite
system; this correction then cancels out when computing
+
significantly better hydrated than GTPH , which has a net
charge of -2 e. The ∆∆Ghyd value in Table 1 indicates that the
+
hydration of GTP is -124.7 kcal/mol more favorable than
∆∆Gbind between GTPH and GTP. Application of the Born
+
the one for GTPH . The correction of this value to account for
equation to calculate the relative hydration between the 22 Å
radius supermolecules with charges of -2 and -3 e, consisting
of the solutes and all 1474 water molecules, results in a
correction of -37.3 kcal/mol and provides a ∆∆Ghyd of -162.0
kcal/mol. The quality of this value can be attested by simple
application of the Born equation to calculate the relative
solvation free energy between ions with charges of -2 and -3
the missing long-range electrostatic interactions between the
solvent molecules not included in the finite models and the
ligands with different molecular charges can be obtained via
3
3
the Born equation. It should be noted that exactly the same
(
33) Born, M. Z. Phys. 1920, 1, 45–48.
8144 J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009
1

FEP Simulations of mRNA Cap Binding by eIF4E
A R T I C L E S
e and that have radii of ca. 5 Å. The agreement between the
of GTP by 1.1 kcal/mol. The final ∆∆Gbind value for the
GTPfDTP transformation is then -0.7 kcal/mol.
∆
∆Ghyd value calculated in this way and the one obtained from
the FEP simulations and corrected as described above is very
good.
7
For the DTPfm DTP transformation, the addition of a
7
methyl group decreases the solvation of m DTP by 0.6 kcal/
The ∆∆Gtransf value in Table 1 indicates that the transfer of
GTP from the gas phase to the protein is -126.9 kcal/mol
mol, which is likely related to the more unfavorable free energy
required to form a larger cavity within the solvent; this
component of the solvation free energy comprises the entropic
penalty associated with the reorganization of the solvent
molecules around the solute and the work done against the
(
-164.2 kcal/mol if corrected by the Born equation) more
+
favorable than the one for GTPH . The eIF4E binding site is
quite polar and exposed to the solvent. Thus, the additional
negative charge of GTP provides a ∆Gtransf value that is
significantly more favorable than the one obtained for its
protonated form. Also, as mentioned above, the transfer of
7
solvent pressure in creating the cavity. The transfer of m DTP
from the gas phase to eIF4E is more favorable than that for
DTP by -1.4 kcal/mol, indicating that the van der Waals
interactions with Trp166 are indeed beneficial (Figure 4). Thus,
+
GTPH from the gas phase to the protein is greatly penalized
7
due to the loss of intramolecular ionic hydrogen bonds in the
gas phase and the burying of the cationic hydrogen-bond donor
in a hydrophobic pocket of the protein. Hence, despite its
significant desolvation penalty, GTP has greater affinity (-2.2
the ∆∆Gbind value of -2.0 kcal/mol for the DTPfm DTP
transformation consists of both more favorable desolvation and
interactions with the protein due to the additional methyl group.
The combination of the results for the GTPfDTP and
+
kcal/mol) for eIF4E than GTPH .
7
DTPfm DTP transformations provide ∆∆Ghyd, ∆∆Gtransf, and
7
+
7
The combination of the results for the m GTPfGTPH and
∆∆Gbind between GTP and m DTP (Table 1). With less solvation
+
GTPH fGTP transformations provides ∆∆Gtransf of -117.9
and more favorable transfer from the gas phase to eIF4E of 1.7
7
kcal/mol, ∆∆G h7 yd of -124.3 kcal/mol, and ∆∆Gbind of 6.4 kcal/
mol between m GTP and GTP (Table 1). The calculated value
is in very good agreement with the experimental value of 5.0
and -1.0 kcal/mol, respectively, m DTP binds more favorably
than GTP by -2.7 kcal/mol. This corresponds to almost 55%
7
of the experimental binding free energy gain of m GTP with
1
kcal/mol, even without the application of a polarizable force
respect to GTP (-5.0 kcal/mol) or 42% of the calculated value
(-6.4 kcal/mol). In summary, the comparison of GTP with the
deazaguanosine analogs indicates that a considerable portion
field, believed to be important for the complete description of
cation-π interactions. Contributions from polarization effects
might not be as significant as for typical cation-π interactions,
since the positive charge is highly delocalized throughout the
guanine ring. Like before, the large negative values for ∆∆Gtransf
and ∆∆Ghyd are caused by the additional negative charge of
GTP. In summary, the reduced affinity for GTP is derived from
a large desolvation penalty that is not compensated for in the
protein. The N(7) methylation (i) provides van der Waals
interactions with Trp166, (ii) generates a positive charge that
establishes cation-π interactions with Trp56 and Trp102, and
7
of the relative binding free energy between m GTP and GTP
can be attributed to favorable van der Waals interactions with
Trp166 and reduced desolvation penalty due to the methyl
group; the methyl group not only eliminates the penalty
associated with the desolvation of the N(7) atom upon binding
but also creates a larger cavity within the solvent that further
facilitates the desolvation step. Analysis of the pair
7
7
m GTP-m DTP suggests that the remaining boost in affinity
is related to the positive charge created on the guanine moiety
due to the N(7) methylation. It provides favorable cation-π
interactions with Trp56 and Trp102 and decreases the negative
molecular charge, which helps the transfer from the solvent, a
more polar environment, to the protein.
(
iii) facilitates the desolvation step because it reduces the
molecular charge and eliminates the interactions between the
N(7) atom of GTP and water molecules in the unbound state.
Role of the Methyl Group in the eIF4E Binding. Even though
the binding process was separated into dehydration and transfer
from the gas phase to the protein to facilitate the rationalization
of the affinity order, it is still very difficult to pinpoint the
Experimental Validation for the Cation-π Contribution. A
medicinal chemistry program aimed at identifying a small
molecule inhibitor of eIF4E has been conducted. The lead
7
specific contributions of the methyl group to ∆∆Ghyd, ∆∆Gtransf
,
optimization process started from the cap analog m GTP. The
7
and ∆∆Gbind between m GTP and GTP. The net charge
difference between these analogs further complicates the analysis
due to the long-range electrostatic contributions to solvation and
transfer from the gas phase to eIF4E. Therefore, in order to
investigate what makes GTP have a poorer affinity for eIF4E,
initial goal of this effort was to obtain a tool compound showing
eIF4E inhibitory activity in a relevant cellular assay. To this
end, an inhibitor with high in vitro potency and suitable
physicochemical properties for cellular permeability was re-
quired. During the course of the lead optimization effort, a large
number of compounds were synthesized and the corresponding
SAR results will be reported independently. Efforts focused upon
reducing the total net charge of the molecule in order to improve
cellular permeability while maintaining in vitro potency com-
7
the GTPfDTP and DTPfm DTP transformations were also
performed. DTP is an interesting model system to understand
the penalty associated with the desolvation of the GTP N(7)
7
atom upon binding, while m DTP, which cannot interact with
7
Trp56 and Trp102 through cation-π interactions, addresses the
binding contributions derived from the van der Waals interac-
tions between the methyl group and Trp166.
parable to that of m GTP. Indeed, a suitable inhibitor design
approach was to remove the positive charge from the nucleotide
core. However, as mentioned previously, there is a 5000-fold
7
Table 1 shows that the transfer from the gas phase to eIF4E
is only 0.4 kcal/mol less favorable for DTP than GTP,
suggesting that both analogs interact with similar strength with
the protein through salt bridges, hydrogen bonds, and π-π
stacking interactions (Figure 4). However, the interactions
between the solvent and the N(7) atom in the unbound state
provide a more favorable hydration free energy for GTP. Thus,
the greater desolvation penalty reduces the binding free energy
loss in the binding affinity from m GTP to GTP (∆∆Gbind
∼
1
7
+5 kcal/mol). Nevertheless, our FEP studies of the m GTPf
m DTP transformation suggested that the contribution of the
7
cation-π interaction is worth about half of the total binding
7
free energy loss from m GTP to GTP (Table 1). This result
encouraged the synthesis of analogs with a neutral core which
maintains other key groups important for binding. Here we
illustrate the accuracy of our free energy estimations with two
J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009 18145

A R T I C L E S
Guimar a˜ es et al.
Table 2. Molecular Structures and the Corresponding Experimental
IC50 Values and Binding Free Energies for the eIF4E Inhibitors
Monte Carlo (MC) as the sampling technique. The MC/FEP
simulations provided informative free energy results and
computed structures. The rationalization of the affinity order
7
7
+
obtained (m GTP > m DTP > DTP > GTP > GTPH ) was
greatly facilitated by the separation of the relative binding free
energies into dehydration and transfer from the gas phase to
+
the protein. The hypothetical analog GTPH binds the least
favorably to eIF4E because of the loss of the intramolecular
hydrogen bond between the cationic hydrogen-bond donor and
the γ-phosphate group that greatly penalizes the transfer from
the gas phase to eIF4E. GTP has the second weakest affinity
mainly because of loss of interactions between the solvent and
the N(7) atom in the unbound state that are not compensated
for in the protein. Simple transformation of GTP into DTP
provides a gain in binding free energy of -0.7 kcal/mol, while
7
methylation of the latter generating m DTP gives another -2.0
kcal/mol; the methyl group provides both favorable van der
Waals interactions with Trp166 and reduced desolvation penalty
due to larger cavity within the solvent that facilitates the
7
7
desolvation step of m DTP. The more favorable m DTP-eIF4E
binding of -2.7 kcal/mol with respect to GTP corresponds to
almost 55% of the experimental relative binding free energy
7
a
gain of m GTP with respect to GTP (-5.0 kcal/mol) or 42% of
Calculated from ∆Gbind ) RT ln IC50 at 298 K.
the calculated value (-6.4 kcal/mol). This suggests that almost
half of the relative binding free energy between m GTP and
GTP can be attributed to both favorable van der Waals
interactions with Trp166 and reduced desolvation penalty due
to the methyl group. The greater affinity of m GTP for eIF4E
7
examples of analogs where the main structural modification is
the removal of the positive charge from the nucleotide core.
Compounds 1 and 2 (Table 2) have the same substituents at
the 7- and 9-positions of the nucleotide core. However, in 2
the core is neutral as a result of replacing the 9-position nitrogen
atom in 1 with a carbon atom. On the basis of the fact that all
of the substituents in 1 and 2 are identical, this pair of
compounds provides an ideal case to study the importance of
the cation-π interaction. The estimated ∆Gbind for compounds
and 2, based on the measured IC50 values, are -9.9 and -7.1
kcal/mol, respectively. Thus, the loss in binding affinity for 2
of 2.8 kcal/mol is in good agreement with the calculated
.7((0.8) kcal/mol value for the m GTPfm DTP transforma-
tion (Table 1) and can be mainly attributed to the lack of
cation-π interaction. GTP pays additional penalties upon
binding, since its nitrogen lone pair is buried toward the protein
surface, resulting in a large desolvation penalty without a
compensating intermolecular interaction and reduced vdW
interactions with the protein by not having the 7-position
substituted with a methyl group. From the calculated binding
7
7
when compared to m DTP is due to the positive charge created
on the guanine moiety because of the N(7) methylation. It
provides favorable cation-π interactions with Trp56 and Trp102
and decreases the negative molecular charge, which likely
reduces the desolvation penalty upon binding. Finally, the FEP
simulations indicate that the cation-π interactions with Trp56
and Trp102 are important, but not as critical for eIF4E binding.
1
7
When compared to m GTP, the affinity drop of GTP is 5000-
fold while that for m DTP is only 50-fold. In other words, this
7
7
3
7
work suggests that, when designing eIF4E inhibitors, neutral
guanine analogs containing appropriate features can safely
replace the positively charged 7-methylguanine core without
significant losses in affinity. This hypothesis is in full agreement
with the inhibitory potency observed for compound 2 and it
provides a valid rationale for designing neutral eIF4E inhibitors
with physicochemical properties suitable for cellular activity.
7
free energy difference between m DTP and GTP (Table 1), these
two effects combined account for about 2.7 kcal/mol of the total
loss of binding from m GTP to GTP.
Acknowledgment. Gratitude is expressed to John Eksterowicz
for helpful discussions.
7
Conclusions
Supporting Information Available: Synthetic route and
procedure for the preparation of compounds 1 and 2. This
material is available free of charge via the Internet at http://
pubs.acs.org.
Striking aspects of the association between eIF4E and the
cap analogs m GTP, GTP, protonated GTP at N(7) (GTPH ),
7
+
7
-deazaguanosine 5′-triphosphate (DTP), and 7-methyldeaza-
7
guanosine 5′-triphosphate (m DTP) have been addressed through
the present free energy perturbation (FEP) simulations using
JA9064359
1
8146 J. AM. CHEM. SOC. 9 VOL. 131, NO. 50, 2009