A structure-based design of new C2- and C13-substituted taxanes: Tubulin binding affinities and extended quantitative structure-activity relationships using comparative binding energy (COMBINE) analysis

Article abstract of DOI:10.1039/c3ob40407b

Ten novel taxanes bearing modifications at the C2 and C13 positions of the baccatin core have been synthesized and their binding affinities for mammalian tubulin have been experimentally measured. The design strategy was guided by (i) calculation of interaction energy maps with carbon, nitrogen and oxygen probes within the taxane-binding site of β-tubulin, and (ii) the prospective use of a structure-based QSAR (COMBINE) model derived from an earlier series comprising 47 congeneric taxanes. The tubulin-binding affinity displayed by one of the new compounds (CTX63) proved to be higher than that of docetaxel, and an updated COMBINE model provided a good correlation between the experimental binding free energies and a set of weighted residue-based ligand-receptor interaction energies for 54 out of the 57 compounds studied. The remaining three outliers from the original training series have in common a large unfavourable entropic contribution to the binding free energy that we attribute to taxane preorganization in aqueous solution in a conformation different from that compatible with tubulin binding. Support for this proposal was obtained from solution NMR experiments and molecular dynamics simulations in explicit water. Our results shed additional light on the determinants of tubulin-binding affinity for this important class of antitumour agents and pave the way for further rational structural modifications. The Royal Society of Chemistry 2013.

Full text of DOI:10.1039/c3ob40407b

Organic &
Biomolecular Chemistry
View Article Online
View Journal
PAPER
A structure-based design of new C2- and
C13-substituted taxanes: tubulin binding affinities and
extended quantitative structure–activity relationships
using comparative binding energy (COMBINE) analysis†
Cite this: DOI: 10.1039/c3ob40407b
Claire Coderch,‡a Yong Tang,‡b Javier Klett,c Shu-En Zhang,b Yun-Tao Ma,b
Wang Shaorong,b Ruth Matesanz,d Benet Pera,d Angeles Canales,d
Jesús Jiménez-Barbero,d Antonio Morreale,c J. Fernando Díaz,d Wei-Shuo Fang*b
and Federico Gago*a
Ten novel taxanes bearing modifications at the C2 and C13 positions of the baccatin core have been syn-
thesized and their binding affinities for mammalian tubulin have been experimentally measured. The
design strategy was guided by (i) calculation of interaction energy maps with carbon, nitrogen and
oxygen probes within the taxane-binding site of β-tubulin, and (ii) the prospective use of a structure-
based QSAR (COMBINE) model derived from an earlier series comprising 47 congeneric taxanes. The
tubulin-binding affinity displayed by one of the new compounds (CTX63) proved to be higher than that
of docetaxel, and an updated COMBINE model provided a good correlation between the experimental
binding free energies and a set of weighted residue-based ligand–receptor interaction energies for 54
out of the 57 compounds studied. The remaining three outliers from the original training series have in
common a large unfavourable entropic contribution to the binding free energy that we attribute to
taxane preorganization in aqueous solution in a conformation different from that compatible with
tubulin binding. Support for this proposal was obtained from solution NMR experiments and molecular
dynamics simulations in explicit water. Our results shed additional light on the determinants of tubulin-
binding affinity for this important class of antitumour agents and pave the way for further rational struc-
tural modifications.
Received 26th February 2013,
Accepted 12th March 2013
DOI: 10.1039/c3ob40407b
www.rsc.org/obc
tubulin depolymerization in dividing mammalian cells.1 The
resulting increased stability of MT inhibits the normal
Introduction
Paclitaxel (PXL, a.k.a. Taxol®) was originally discovered in the dynamic reorganization of the microtubular network that is
bark of the Pacific yew tree (Taxus brevifolia) and was found to essential for vital interphase and mitotic cellular functions. In
promote the assembly of microtubules (MT) by preventing clinical use since 1992, PXL became a first-in-class antimitotic
drug which is currently indicated in the chemotherapy of a
variety of human tumours including ovarian, breast and non-
small-cell lung carcinomas. Notwithstanding the remarkable
potency of this antineoplastic agent, its limited availability
from its natural source, its poor aqueous solubility and the
emergence of resistance in cancer cells, partly due to over-
expression of multidrug efflux transporters,2 made it desirable
from the start to obtain new analogues (taxanes) more readily
and with an improved pharmacological profile.3–7 This goal
has been partially achieved with the clinical development and
marketing of the semisynthetic derivatives docetaxel (DXL,
Taxotere®) and, more recently, cabazitaxel (Jevtana®), both of
which partially overcome some of PXL’s limitations. Neverthe-
less, structure-based design efforts in the search for novel
taxanes9 were hampered for many years because of difficulties
aÁrea de Farmacología, Departamento de Ciencias Biomédicas, Universidad de
Alcalá, E-28871 Alcalá de Henares, Unidad Asociada al Instituto de Química Médica
del CSIC, Madrid, Spain. E-mail: federico.gago@uah.es
bState Key Laboratory of Bioactive Substances and Functions of Natural Medicines,
Institute of Materia Medica, Chinese Academy of Medical Sciences & Peking Union
Medical College, 2A Nan Wei Road, Beijing 100050, P. R. China.
E-mail: wfang@imm.ac.cn
cUnidad de Bioinformática, Centro de Biología Molecular Severo Ochoa (CSIC/UAM),
Cantoblanco, E-28049 Madrid, Spain
dCentro de Investigaciones Biológicas, Consejo Superior de Investigaciones
Científicas, E 28040 Madrid, Spain
†Electronic supplementary information (ESI) available: Details of synthetic pro-
cedures, compound characterizations, and additional computations. See DOI:
10.1039/c3ob40407b
‡Authors contributed equally to this paper.
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.

View Article Online
Paper
Organic & Biomolecular Chemistry
associated with tubulin crystallization, the low affinity of PXL was measured and an energy-refined binding mode for PXL,
for unpolymerized tubulin subunits and the poor water solubi- DXL and CTX40 to β-tubulin was proposed and then
lity of the drug.10
explored using molecular dynamics (MD) simulations in
The first reliable three-dimensional model of tubulin in explicit solvent. Support for the conformation of bound DXL
atomic detail was reported in 1998.11 This long sought-after was provided by the excellent agreement obtained with data
structure, deposited in the Protein Data Bank (PDB) as entry acquired from transfer nuclear Overhauser effect spectroscopy
1TUB, was obtained by electron crystallography of zinc- (TR-NOESY) measurements using MT-bound DXL.8 Impor-
induced sheets of bovine tubulin bound to PXL. At this point tantly, the thermodynamic characterization of the binding of
it became clear that PXL binds to the luminal side of the β these taxanes to MT revealed not only a range of ∼10 kcal
subunit in the α,β-tubulin heterodimer. Subsequent refine- mol−1 in binding free energies (ΔG) but also distinct enthalpic
ment improved the resolution to 3.5 Å (PDB entry 1JFF)12 but (ΔH) and entropic (−TΔS) contributions. The existence of
neither the acidic C-terminal tails nor the N-terminal loop rather large enthalpy/entropy compensations22–24 within this
including residues 35–60 in the α subunit could still be traced series (Fig. 2)8 attests to the difficulties that can be en-
accurately. Moreover, due to the limited certainty about side- countered when theoretical methods are used to try and
chain orientation for many residues at the taxane-binding site rationalize the quantitative SAR (QSAR) for these and other
(and elsewhere), the precise set of interactions between tubulin-binding ligands.25
β-tubulin and bound PXL13 (and related taxanes) cannot be
In the following we describe how we probed the taxane-
entirely clarified from this low-resolution structure.
binding site in β-tubulin with methyl carbon, amine nitrogen,
The initial pharmacophore that was established for taxanes and hydroxyl oxygen functional groups in our search for
on the basis of the so-called “T-shaped” or T-Taxol binding unexploited “regions of attraction”26 that could be filled with
conformation14–17 has been the foundation for several struc- suitable substituents in attempts to increase the binding
ture-guided modifications18,19 and structure–activity relation- affinity over that of existing congeneric ligands. Synthetic con-
ship (SAR) studies.6,16,20,21 Recent collaborative work from our siderations limited the number of novel C2- and C13-modified
laboratories resulted in the report of an extended series of 44 taxanes to ten derivatives (Table 1) whose binding affinities
semisynthetic taxanes (CTX), which have in common the bac- were experimentally measured using the same method as
catin core present in PXL, DXL and cephalomannine (CPM), reported earlier.8 This new series was then employed to update
but additionally incorporate a variety of changes at positions a previous exploratory QSAR model (Fig. 3) that was derived by
C2, C7, C10 and C13 (Scheme S1 in the ESI†).8 A corollary of applying a chemometric tool known as comparative binding
this endeavour was the finding of an optimal combination of energy (COMBINE) analysis27 on a whole set of modelled
substituents that yielded the high-affinity taxane CTX40 ligand–receptor complexes. In order to account for the grossly
(Fig. 1), the cytotoxicity of which is unaffected by P-glyco- outlying behaviour of CTX42, CTX43 and CTX44 (Fig. 4) from
protein (P-gp) overexpression. Furthermore, the binding of all the original training series, MD simulations in aqueous solu-
of these 47 compounds to glutaraldehyde-stabilized bovine MT tion were carried out for these three analogues and also for
CTX40 with a view to assessing possible conformational differ-
ences between the free and bound states of these rather
unique taxanes. Confirmatory data could be obtained from
solution NMR experiments only for the most water-soluble
analogue CTX42.
Materials and methods
Synthetic procedures
For C2 modification, the 2-ethyl baccatins were prepared from
10-deacetyl baccatin III (10-DAB) following a procedure similar
to that described in the literature42 consisting of first reacting
the 2-debenzoyl baccatin intermediates with mesyl chloride
and then transforming to the C2 ethers upon treatment with
sodium phenolate. Subsequent reduction of the 13-oxo group,
incorporation of the C13 side chain bearing the 3′N-Boc group,
and further desilylation led to the final products CTX55 and
CTX56.
As it is known that chloroformates derived from tertiary
alcohols bearing electron-donating groups (EDGs) are unstable
whereas those bearing electron-withdrawing groups (EWGs)
Fig. 1 Chemical structures of paclitaxel (PXL), docetaxel (DXL), cephaloman-
nine (CPM) and CTX40.
are relatively stable,28 p-nitrobenzoyl carbonates were used
Org. Biomol. Chem.
This journal is © The Royal Society of Chemistry 2013

View Article Online
Organic & Biomolecular Chemistry
Paper
Fig. 2 Thermodynamic parameters (kcal mol−1) for the binding of the taxanes studied to the PXL site in bovine β-tubulin, as reported in Table 1 of ref. 8. Note that
actual values correspond only to the length of the coloured block, i.e. are not cumulative.
Table 1 Novel taxanes incorporating modifications at positions C2 and C13
(either on C3’ or 3’N)
Fig. 3 Correlation between the binding free energies calculated in the cross-
validated COMBINE model and the experimental values for compounds in the
Compound
R2
R3′N
RC3′
Compound
R2
R3′N
RC3′
training set ( ) upon exclusion of the three outliers ( ). This model, with only
5 LV, accurately predicted the affinity of CTX40 ( ), which was not included in
model derivation.
CTX55
CTX56
CTX57
CTX58
CTX59
A
B
C
C
C
A
A
B
C
D
A
A
A
A
A
CTX60
CTX61
CTX62
CTX63
CTX64
C
C
C
C
C
E
F
A
A
A
A
A
B
C
D
analogs CTX57, CTX60 and CTX61, the 3′-debenzoyl PXL was
transformed into a carbamate by reacting with the correspond-
instead of chloroformates during the carbamoylation of the ing chloroformates, whereas for EDG-bearing 3′N-modified
3′-NH2 for some taxanes. For the syntheses of EWG-bearing taxanes CTX58 and CTX59, 3′-debenzoyl PXL was transformed
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.

View Article Online
Paper
Organic & Biomolecular Chemistry
samples were then incubated at 37 °C for 30 min before
measurement.
The increased solubility of CTX42 relative to the other
amide-containing taxanes allowed the preparation of a sample
of this free compound in the same buffer described above.
TR-NOESY information was acquired for CTX42 in the pres-
ence of MT with a mixing time of 200 ms at 310 K. 2D NOE in
the rotating frame (ROESY) data were acquired for CTX42 in
the free state because the NOESY cross peaks for the free mole-
cule were essentially zero at 310 K. All of the spectra were
obtained in a Bruker 500 MHz spectrometer.
Fig. 4 Chemical structures of the SAR outliers that have in common an amide
bond at the C2 position in place of an ester.
Theoretical methods
into a carbamate by reacting first with the corresponding
p-nitrobenzoyl carbonate,29 and then with the corresponding
alcohol. By subsequent desilylation the 3′N-modified DXL
analogues CTX57 to CTX61 were obtained.
The refined structure of the α,β-tubulin heterodimer at 3.5 Å
resolution (Protein Data Bank code 1JFF) was used for mole-
cular modelling and automated docking of DXL and CTX40, as
already described.8 In brief, addition of missing hydrogen
atoms and computation of the protonation state of ionisable
groups in β-tubulin at pH 6.5 were carried out using the H++
Web server,33 which relies on AMBER force-field para-
meters34,35 and finite difference solutions to the Poisson–
Boltzmann equation. The GDP molecule was conserved and
the protein side-chains in the nucleotide binding site were
slightly reoriented so as to reproduce the same interactions
that are observed in the 2.51 Å resolution X-ray crystal
structure of γ-tubulin bound to 5′-guanosine-diphosphate-
monothiophosphate36 (PDB entry 1Z5V). Following energy
minimization, MD simulations at 300 K were run for these two
ligand–protein complexes for 10 ns under periodic boundary
conditions in truncated octahedra each containing ∼10 000
TIP3P water molecules and 11 randomly placed sodium ions
to neutralize the total charge of the systems. After this time,
the temperature was gradually decreased from 300 K to 273 K
over 1200 ps. Energy minimization of the resulting “cooled”
structures employing 5000 cycles of steepest descent followed
by 4000 cycles of conjugate gradient provided the “frozen”
templates that were used as representative structures to build
the remaining complexes for all the CTX derivatives. The need
for two templates arose from the fact that the C2 substituent
can be placed slightly differently inside the hydrophobic
For C3′ modifications, the syntheses of the 3′-thiophenyl
and 3′-(m-hydroxymethyl)phenyl β-lactams were accomplished
as previously reported,30 with the addition of Schiff base to the
ketene (Staudinger reaction) as the key step. During the syn-
thesis of 3′-benzo[d][1,3]dioxol-5-yl β-lactam, di-(3,4-methylene-
dioxy-benzaldehyde) substituted Schiff base was applied to the
Staudinger reaction, and subsequent N-deprotection with
AcOH furnished the β-lactam. Further kinetic resolution of
racemic β-lactam and the incorporation of the Boc group
onto the N atom furnished optically active N-Boc-β-lactam,
which was conjugated to 7-triethylsilyl (TES) baccatin III to
afford the corresponding 2′,7-silyl-protected DXL analogues.
Subsequent desilylation furnished 3′-modified taxanes CTX62
to CTX64.
Measurement of the equilibrium binding constants
Stabilized, moderately crosslinked MT were prepared from pur-
ified calf brain tubulin as reported before.31 Binding constants
and thermodynamic parameters were determined by displace-
ment of the fluorescent taxane derivative Flutax-2 from the
PXL-binding site in these MT.32
NMR experiments
Samples of MT-bound CTX42 were prepared using a 300 μM binding pocket, depending on the absence or presence of sub-
concentration of CTX42 and 20 μM of tubulin in D2O, 10 mM stituents on the phenyl ring, in such a way that the stacking
KPi, 0.1 mM GMPCPP, 6 mM MgCl2 pH* 6.7. The tubulin interaction between the latter and the side chain of His229 is
samples were prepared by removing sucrose, Mg2+, and H2O also different. Therefore the bound conformation of DXL was
from the storage buffer of a 20 mg sample of frozen tubulin used as the template for building models of those taxanes that
using a two-step procedure by chromatography in a drained have no substituent on this ring whereas those bearing bulky
centrifuge column of Sephadex G-25 medium (6 × 1 cm) equili- substituents on this phenyl were modelled from CTX40. Of
brated in D2O, 10 mM KPi, 10 μM GTP pH* 7.0 in the cold, fol- note, during the MD simulations we observed the long resi-
lowed by
a
second chromatography using another dence times of two water molecules that bridged a hydrogen
Sephadex G-25 medium column (15 × 0.9 cm) equilibrated in bonding interaction between the carbonyl oxygen present in
D2O, 10 mM KPi, pH* 7.0. Tubulin was centrifuged, and its the C2 linker and the side-chains of Asp226 and His229. Con-
concentration was determined spectrophotometrically by sequently, we decided to keep these two water molecules
employing an extinction coefficient of 107 000 M−1 cm−1 in (WAT1 and WAT2) as part of the ligand–receptor complexes
10 mM phosphate buffer containing 1% SDS. Tubulin was (Fig. 5). These two extra residues were later shown to be instru-
diluted to 20 μM and 0.1 mM GMPCPP and 6 mM MgCl2 mental for an accurate prediction of taxanes CTX22–27, which
(final pH 6.7) were added prior to drug incorporation. The have in common the presence of non-ester C2 linkers.
Org. Biomol. Chem.
This journal is © The Royal Society of Chemistry 2013

View Article Online
Organic & Biomolecular Chemistry
Paper
variables).27 To this end, each complex was centred in a cubic
box with a grid spacing of 0.5 Å and a minimum separation of
11 Å was allowed between any solute atom and the box walls.
The solute dielectric constant was set to 4, while the solvent
dielectric medium was set to 80. The dielectric boundary was
calculated using a radius of 1.4 Å for the spherical water
probe. The potentials at the grid points delimiting the box
were calculated analytically by treating each charge atom as a
Debye–Hückel sphere.
The data matrix containing the computed energy com-
ponents (X variables) was then subjected to a multivariate stat-
istical analysis named ‘projection to latent structures’ (a.k.a.
partial least squares, PLS)41 in order to find a correlation with
the experimental binding free energies (Y variable). For cross-
validation purposes, a group consisting of 5 randomly chosen
complexes was excluded from the training set and used as a
test set, as described,39 and this procedure was repeated a total
of 10 times. By doing this the robustness of the model is chal-
lenged more rigorously than when the very conservative leave-
one-out method is employed. The dimensionality of the model
(i.e. the number of latent variables, LV, extracted) was chosen
as that corresponding to the peak cross-validated regression
coefficient (q2) value.
Fig. 5 Taxane-binding site in β-tubulin (green C atoms) as found in the refined
complex with DXL (grey C atoms) which includes two water molecules in the
vicinity of Asp226 (1JFF numbering). The orange and cyan mesh regions corre-
spond, respectively, to locations giving rise to a favourable interaction energy
with a methyl carbon atom probe (contoured at −1.5 kcal mol−1) or a hydroxyl
oxygen (contoured at −6.0 kcal mol−1), as calculated by our in-house program
cGRILL.
To study the conformational preferences of CTX40, CTX42,
CTX43 and CTX44 in aqueous solution, each of these taxanes
was extracted from its respective complex and immersed in
The remaining taxanes bearing modifications on the C2, a cubic box of approximately 1400 TIP3P water molecules.
C7, C10 and C13 substituents were built within the molecular Following energy minimization, a 100 ns MD simulation at
graphics program PyMOL using as templates the common bac- 300 K was carried out for each taxane. The resulting trajec-
catin cores of the refined CTX40 and DXL structures in their tories were processed with the ptraj module of AMBER to
respective complexes with β-tubulin. The geometries of the iso- monitor the rms deviations from the initial T-taxol geometry.
lated new analogues were fully optimized at the Hartree–Fock Desolvation energies of representative structures of each con-
level using the Gaussian 03 program37 and the 3-21G basis set. formation were obtained using the Generalized Born method
Charge distributions for all the ligands were then obtained by that is implemented in the AMBER 10 distribution. The
fitting the quantum mechanically calculated (HF/6-31G*//HF/ energies of each conformer were then used to calculate the
3-21G*) molecular electrostatic potential (MEP) to a RESP probabilities of the microstates (Pi), as defined by a Boltzmann
point charge model.38 Covalent and non-bonded parameters distribution:
X
for the inhibitors were derived, by analogy or through inter-
Pi ¼ neꢀE =RT
=
neꢀE =RT
i
i
polation, from those already present in the AMBER34,35 database
j
(ff03). The tubulin–taxane complexes were refined by energy
minimization using only the steepest descent algorithm in the where n is the number of structures belonging to each confor-
sander module until the root mean square (rms) value of the mational cluster i, Ei is the energy of the average structure
potential energy gradient was below 0.1 kcal mol−1 Å−1
.
extracted from each cluster, T is the temperature in kelvin,
The set of 47 refined ligand–receptor complexes (including R is the gas constant expressed in kcal mol−1 and j is the total
WAT1 and WAT2) was then used as input to the gCOMBINE number of clusters in the ensemble.
program,39 which calculates all the van der Waals (AMBER
The molecular graphics program PyMOL (v. 0.99rc6,
force field) and electrostatic interaction energies between each DeLano Scientific, LLC, Palo Alto, CA)42 was employed for
ligand and every protein residue (plus selected water mole- molecular editing, visualization and representation. The ener-
cules). The electrostatic contributions were calculated using getic characterization of the taxane-binding site was performed
Goodford’s implementation of the images method26 and a using the cGRILL program, an improved version of our CGRID
uniform dielectric constant of 4.0 Debye. For each complex, code43 that relies on interaction energy calculations for
the electrostatic contributions to the desolvation energies of selected probes on 3D grids, as pioneered by Goodford in his
both the protein and the ligand, calculated by numerically well-known GRID program.26 The cubic grid for cGRILL calcu-
solving the linear Poisson–Boltzmann equation using the lations was defined as the space delimited by the axis-parallel
finite difference method as implemented in the program box containing the bound ligand, augmented by 5 Å in each
DelPhi,40 were used as additional descriptors (a.k.a. external axis direction.
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.

View Article Online
Paper
Organic & Biomolecular Chemistry
positions of the common baccatin scaffold. When the whole
set of complexes was subjected to a COMBINE analysis, the
clearly outlying behaviour of CTX42, CTX43 and CTX44 (Fig. 4)
Results and discussion
Modification rationale and binding affinities
The structures of the ten novel taxanes (CTX55–CTX64) that was made apparent.
were designed, synthesized and tested are shown in Table 1,
In relation to these compounds, which are the only ones
and their thermodynamic binding parameters are listed in bearing an amide linker at the C2 substituent, it has been
Table 2. The rationale for synthesizing CTX55 and CTX56 was known for many years that both the nature and stereo-
to explore the effect of replacing the O-benzoyl moiety typically chemistry of the 2-aroyl group in PXL are of great importance
present at the C2 position with an extended N-containing aro- for activity. In particular, loss of the 2-aroyl ester, or change of
matic ring system (i.e. a quinoline or an isoquinoline, respecti- its orientation from α to β, were shown to be detrimental. In
vely) that might improve the stacking interactions with the line with these findings, when CTX42, the 2α-benzoylamido
imidazole of His229, in accordance with the calculated affinity analogue of DXL, was first synthesized back in 2001 in one of
maps for the C probe (Fig. 5) and the weight assigned to this our laboratories,23 it was found that its cellular activity was
residue in the retrospective COMBINE analysis (Table S1 in the less than that of DXL although it was not totally lost as was the
ESI†) of the previously reported series.8 In taxanes CTX57 to case with the C-13 amido analogues of PXL. When the
CTX61, one of the methyl groups in the tert-butoxycarbonyl affinities of CTX42–44 for stabilized MT were measured several
attached to the NH bonded to C3′ on the C13 substituent (3′N years later,8 it was found that they were indeed dramatically
position) was replaced with a small substituent in an attempt decreased relative to those of their ester-containing closest
to fill in a protein cavity in the vicinity of the side chains of counterparts CTX21, CTX11 and CTX6. Furthermore, it was
Lys19 and Glu22. Finally, either a thiophene (CTX62) or a sub- learnt that their favourable enthalpic binding to tubulin was
stituted phenyl (CTX63–CTX64) was attached to C3′ in place of strongly opposed by a large positive entropic contribution
the common unsubstituted phenyl ring of DXL to try and (Fig. 2). Obviously, these compounds were included in our
improve the interactions within this subsite.
initial QSAR analysis for completeness but anybody interested
Among these novel taxanes, only CTX63 and CTX64 were in designing new higher-affinity taxanes would be strongly
shown to display binding affinities higher than or similar to, biased against incorporating amide-bearing substituents at
respectively, those measured for DXL under identical con- position 2.
ditions (Table 2). The better binding affinity reported for
Nonetheless, from a modelling perspective, the fact that
CTX63 presumably arises from the favourable hydrogen this sole modification is translated into a largely decreased
bonding interaction that the benzylic hydroxyl group at C3′ in binding free energy is striking because (i) the proposed
this ligand establishes, in the vicinity of Glu27, with the side binding modes for all taxanes are very similar (Fig. 6), and (ii)
chains of Ser374 and Thr376, in accordance with the cGRILL no protein contacts are apparent in this region that could
maps.
account for these differences, in contradistinction to what has
been described for C2 S-linked taxanes.44 Therefore we
decided to deal with these three molecules separately and
exclude them from the initial QSAR model because entropy
cannot be expected to be adequately captured within the
energy descriptors that are normally used in COMBINE
analysis.24
Structure–affinity relationship
At this point we thought it would be of interest to test the pre-
dictive ability of the original COMBINE model and assess
whether this chemometric tool could provide clues for the lack
of improved affinity in CTX55–CTX62. In our original training
set8 the 47 taxanes studied spanned a range of binding free
energies of ∼10 kcal mol−1 and differed from one another in
the nature of the groups attached to the C2, C7, C10 and C13
When CTX42–44 were removed from the training set,
a robust cross-validated COMBINE model was obtained (r2
=
0.94 and q2 = 0.89 for the best model; average r2 = 0.94 0.03
Table 2 Apparent binding affinities of the novel taxanes for the paclitaxel-binding site in β-tubulin (× 107 M−1), measured at different temperatures, and calculated
thermodynamic binding parameters
30 °C
32 °C
35 °C
ΔG (35 °C) (kcal mol−1
)
ΔH (kcal mol−1
)
ΔS (cal mol−1 K−1
)
CTX55
CTX56
CTX57
CTX58
CTX59
CTX60
CTX61
CTX62
CTX63
CTX64
0.019 0.003
0.036 0.005
0.630 0.135
1.84 0.61
0.021 0.001
0.90 0.12
0.018 0.005
0.035 0.012
0.674 0.131
1.67 0.47
0.019 0.002
0.98 0.10
0.015 0.003
0.029 0.005
0.595 0.042
1.87 0.09
0.017 0.002
0.85 0.15
1.52 0.28
2.10 0.29
90 14
−7.29 0.1
−7.69 0.1
−9.54 0.05
−10.25 0.02
−7.26 0.07
−9.75 0.1
−9.87 0.1
−10.30 0.1
−12.38 0.1
−11.69 0.02
n.d.
n.d.
n.d.
n.d.
−2.87 0.7
−2.15 0.7
n.d.
−2.87 0.96
−7.65 1.4
−6.93 3.1
−2.39 1.4
−7.41 1.2
0.021 0.002
0.026 0.003
n.d.
0.022 0.003
0.008 0.004
0.011 0.015
0.033 0.005
0.013 0.004
1.87 0.29
2.49 0.57
1.77 0.24
2.59 0.64
96
4
97
22
2
9
25 20
20 0.1
n.d., not determined.
Org. Biomol. Chem.
This journal is © The Royal Society of Chemistry 2013

View Article Online
Organic & Biomolecular Chemistry
Paper
confer resistance to PXL.46,47 On the other hand, the negative
PLS coefficient that modulates the electrostatic contribution
to ligand desolvation strongly suggests that this factor contrib-
utes toward explaining the binding affinity differences, in con-
trast to the small positive coefficient that is assigned to the
desolvation of the binding pocket.
The binding affinities of the new analogues were overpre-
dicted to different extents using this model (Fig. S2 in the
ESI†). Nonetheless, the calculated ΔG values for the taxanes
bearing modifications at C2 and C3′, including the most
potent CTX63 and CTX64, deviated from the experimental
measurements much less than those modified at 3′N. Of
these, CTX59 clearly behaved as an outlier. This result is likely
a consequence of the lack of accurate information in the initial
PLS model regarding the protein residues that interact with
the 3′N substituents because this region was totally unexplored
in the original training series. As discussed before, lead optim-
ization is a cyclic process involving repeated rounds of com-
pound design, synthesis and testing, and QSAR models
derived at one stage need to be continually updated incorpor-
ating as much new information as possible.27 To prove this
point we added the 3′N-substituted analogues CTX57–CTX60
to the original training set and tried to predict the affinity of
Fig. 6 Superposition of all 57 taxanes studied inside the PXL-binding site of
β-tubulin (1JFF numbering). For simplicity, only the cartoon representation of
one of the protein molecules (that from the DXL–tubulin complex) enveloped
by a semitransparent grey van der Waals surface is shown. Notice the two water the remaining novel taxanes. This updated model (r2 = 0.93,
q2 = 0.77 for the best run and average r2 = 0.94 0.01, q2 = 0.72
molecules in the vicinity of Asp226 that were incorporated into the comparative
binding energy analysis.
0.04 over 10 runs using 6
1 latent variables) was able to
predict the binding free energies of CTX55, CTX56 and CTX61
to CTX64 remarkably well (Fig. S3 in the ESI†). Finally, when
and q2 = 0.86 0.02 for the 10 runs, using 5 1 latent vari- the whole set comprising 54 taxanes was subjected to
ables; Fig. 3). An alternative set of complexes in which His229 COMBINE analysis using random groups of five complexes for
was modelled as neutral yielded much worse figures of merit cross-validation, similarly good figures of merit (r2 = 0.94, q2 =
(Fig. S1 in the ESI†). This result appears to support our choice 0.79 for the best model and average r2 = 0.94 0.01, q2 = 0.77
of protonated state for this amino acid in the taxane-binding 0.2 over 10 runs using 7 1 latent variables) were obtained
site.
(Fig. 7) and updated PLS coefficients were assigned (Table 3)
The relative importance of the β-tubulin residues that best that will be used in future work.
help explain the differences in binding free energies, as
judged from the assigned partial least squares (PLS) pseudo-
Entropic contribution to binding
coefficients (Table S1 in the ESI†), depends on the actual mag- Despite the reservations expressed above about considering
nitude of the calculated interaction energies because these taxanes bearing an amide linker instead of an ester between
signed coefficients are used to properly weigh these terms.27,45 the C2-substituent and the baccatin core, we thought that it
Since the energy values are usually negative (i.e. attractive), a could be worthwhile to try to understand the origin of the
positive coefficient will make the calculated binding free unfavourable entropic contribution to the Gibbs free energy of
energy more negative, and vice versa. Thus, improved van der binding (Fig. 2). In this regard, it is usually agreed that sol-
Waals interactions between the C13 moiety and both Val23 vation entropy changes originating from the release of water
and Asp26 are predicted to result in gains in binding molecules brought about by the burial of a significant hydro-
affinity. The same can be said for those established between phobic surface are favourable for binding48 whereas the loss of
the baccatin core and the C2 substituent and residues Asp226, conformational degrees of freedom by the inhibitor and/or
His229, Ala233, Phe272 and Leu275. In contrast, the stronger some residues in the protein is unfavourable.49 Thus, a
the interaction of the ligands with other residues, particularly common goal in molecular design is to minimize the ligand
Cys213, Leu217 and Leu219, the lower the affinity, as this most entropy loss by restraining conformational flexibility and pro-
likely reflects the existence of steric clash.45
moting a configuration in the unbound state that optimally
Regarding electrostatics, the importance of the interactions matches the binding site,50 even though doubts have been cast
with His229, Arg278 and WAT2 is reflected in the positive on this assertion.51 This “preorganization” may lead to ligands
coefficients assigned to the terms involving these two protein with affinities in the nanomolar range even in the presence
residues and one of the water molecules. In this respect, it is of unfavourable binding enthalpies.52 Conversely, favourable
noteworthy that the mutation of Arg278 to Ser is known to binding enthalpies can be largely offset by unfavourable
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.

View Article Online
Paper
Organic & Biomolecular Chemistry
and conformational differences between several taxanes in
solution and those bound to tubulin (“T-shaped”) have been
amply reviewed,16 we speculated that CTX42, CTX43, and
CTX44 might be preferably stabilized in aqueous solution in a
conformation that is not compatible with binding to the site.
To test this hypothesis, NMR experiments were carried out to
characterize the conformation of CTX42, the most water-
soluble derivative, in both the free and bound states. In
addition, we performed several MD simulations in explicit
water to sample the conformational space of the three C2
amide-containing taxanes and also of CTX40 as a representa-
tive example of an ester-containing analogue whose binding is
entropically favoured.
NMR experiments and MD simulations
The conformation of MT-bound CTX42 was deduced from
analysis of TR-NOESY cross peaks (Fig. 8). The strong negative
cross peaks observed for CTX42 in the presence of MT indicate
binding of the taxane to the tubulin preparation (NOESY cross
peaks were essentially zero for the free ligand). The fact that
the TR-NOESY spectrum of CTX42 in the presence of MT
Fig. 7 Updated COMBINE model (7 latent variables) derived from 54 taxane–
tubulin complexes including all the taxanes reported in the original training set
8
(
)
except for CTX42, CTX43 and CTX44, plus the novel taxanes reported
herein ( ).
Table 3 Selected projection to latent structure (PLS) pseudo-coefficientsa for
the amino acids and water molecules that contribute the most to explaining the
predicted binding free energy differences in the final updated COMBINE model
derived from 54 tubulin–taxane complexes
Residue
vdW
Ele
Residue
vdW
Ele
Lys19
Glu22
Val23
Asp26
Glu27
Cys213
Leu217
Leu219
Asp226
His229
Leu230
Ala233
Phe272
−0.72
−1.02
0.45
−0.67
Pro274
Leu275
Thr276
Ser277
Arg278
Gln282
Pro360
Arg369
Gly370
WAT1
0.20
0.46
−0.1
0.21
−0.22
−0.18
0.34
0.25
−0.10
0.24
0.18
0.35
0.22
−0.1
0.26
0.25
−0.15
−0.48
−0.33
0.12
0.42
0.16
−0.20
−0.48
0.46
−0.23
0.17
0.46
0.12
−0.48
0.54
WAT2
−0.32
0.64
0.89
−0.21
DesolvRb
DesolvLb
−0.27
a Only PLS pseudo-coefficients with absolute values ≥|0.1| for the van
der Waals (vdW) and electrostatic (Ele) interactions are shown.
b DesolvR and DesolvL denote receptor and ligand desolvation,
respectively.
entropic contributions, as clearly seen in the series presented
here. In this case, although we tried to take into account, at
least in part, the cost of ligand and binding-site desolvation
for all the complexes studied, we did not find an easy way to
incorporate the configurational entropies22,51 of the ligands,
the water molecules and the protein into the COMBINE model
even though we did include some apparently crucial water-
mediated interactions. Since early conformational studies on
PXL and DXL reported the existence of hydrophobically col-
Fig. 8 (a) ROESY and (b) TR-NOESY spectra acquired for free and microtubule-
lapsed structures in mixtures of water and organic solvents53 bound CTX42.
Org. Biomol. Chem.
This journal is © The Royal Society of Chemistry 2013

View Article Online
Organic & Biomolecular Chemistry
Paper
shows an NOE pattern similar to that described for DXL8 change in the ligand. In contrast, for taxanes that are preorga-
points to a similar conformation for both molecules in the nized for binding to the site, e.g. CTX40, the ligand-dependent
bound state, which is in accordance with our modelled com- entropic component is likely to be favourable.
plexes (Fig. 6). Clear NOEs are observed between the C13 tert-
butoxy and C2 aromatic ring protons. The 4-OAc group also
gives NOEs with both aromatic rings, and they are stronger
with the protons of the phenyl ring on the C13 substituent,
Conclusions
which also gives an NOE contact with the methyl group at C12.
In contrast, no ROE contacts were detected between the C3′
tert-butoxy protons and those of the C2 aromatic ring in the
ROESY of free CTX42. Moreover, weak ROEs were observed
between the 4-OAc group and the C13 phenyl ring protons
whereas no ROEs were observed between this latter moiety and
the C12 methyl group. All these data support the hypothesis of
different conformations for CTX42 in the free and MT-bound
states.
The MD trajectories showed that the probability that the
microstate corresponds to the bound conformation was 0.78
for CTX40 but only 0.0005, 0.0001, and 0.03, respectively, for
CTX42, CTX43 and CTX44. Therefore these amide derivatives
in aqueous solution appear to be locked in an alternative con-
formation (Fig. 9) that is fully compatible with the ROESY
spectrum of free CTX42 and is more energetically favourable
in aqueous solution than the tubulin-bound conformation by
∼6.0 kcal mol−1 for CTX44 and ∼4.0 kcal mol−1 for CTX42 and
CTX43. Collectively, these results support the view that the
entropic penalty associated with tubulin binding (Fig. 2) may
arise from an energetically unfavourable conformational
We have synthesized ten novel taxanes that incorporate new
substituents at positions C2 and C13 of the baccatin core and
have also measured their tubulin binding affinities. The
design principle was guided by the calculation of affinity maps
inside the protein binding site as well as by the results pro-
vided by a comparative binding energy (COMBINE) QSAR
analysis on a previously reported series. Compound CTX63 is
endowed with higher affinity than docetaxel and makes use of
a previously unexplored region in the taxane-binding site.
A serious limitation of the QSAR study was found when it
was realized that the affinities of a few analogues whose
binding is characterized by a large unfavourable entropic com-
ponent could not be correlated with the calculated interaction
and desolvation energies. To account for this anomalous
behaviour we hypothesized, on the basis of the results of MD
simulations in aqueous solution, that the most populated con-
formation of these outliers in the free state is unsuitable for
binding and is stabilized by hydrophobic collapse. NMR
data obtained for the only representative taxane that could
be studied tubulin-free in a water–DMSO mixture (CTX42)
provided qualitative support to this hypothesis.
Fig. 9 Structural comparison of the tubulin-bound conformations (pale pink sticks) and the four main conformations found in the MD simulations of free CTX40
(top) and CTX42 (bottom). All hydrogens have been removed for clarity. For each taxane the lowest-energy and most populated conformer is boxed and that corre-
sponding to the tubulin-bound conformation has the C atoms coloured in green. Er is the energy of the average structure extracted from each cluster relative to the
lowest-energy conformation whereas Pi is the probability of the microstate. Note that the bound conformation is unlikely for CTX42 free in solution (and also for
CTX41 and CTX43, see Fig. S4 in the ESI†) but it corresponds to the major conformer of CTX40 free in solution.
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.

View Article Online
Paper
A robust QSAR model was derived from a total of 57 taxane–
Organic & Biomolecular Chemistry
6 C. G. Yang, I. Barasoain, X. Li, R. Matesanz, R. Liu,
F. J. Sharom, D. L. Yin, J. F. Diaz and W. S. Fang,
ChemMedChem, 2007, 2, 691–701.
7 I. Ojima, J. Chen, L. Sun, C. P. Borella, T. Wang,
M. L. Miller, S. Lin, X. Geng, L. Kuznetsova, C. Qu,
D. Gallager, X. Zhao, I. Zanardi, S. Xia, S. B. Horwitz,
J. Mallen-St Clair, J. L. Guerriero, D. Bar-Sagi, J. M. Veith,
P. Pera and R. J. Bernacki, J. Med. Chem., 2008, 51, 3203–
3221.
tubulin complexes using the COMBINE methodology, which
estimates the relative contributions to the binding affinities of
a number of protein residues and also (optionally) the electro-
static contribution to ligand and receptor desolvation. Our
results favour the view that His229, which is an important
residue for binding, is doubly protonated on the imidazole
ring and also that some water molecules get trapped in the
binding process.
Given the known additivity of contributions from optimal
substituents to the binding affinity of taxanes,8 attempts are
currently being made to use this information to further
expand the series and synthesize more analogues that incor-
porate, for example, the bridging water’s hydrogen-bonding
functionality into their own structure or a benzylic alcohol in
8 R. Matesanz, I. Barasoain, C. G. Yang, L. Wang, X. Li, C. de
Ines, C. Coderch, F. Gago, J. J. Barbero, J. M. Andreu,
W. S. Fang and J. F. Diaz, Chem. Biol., 2008, 15, 573–585.
9 R. Geney, L. Sun, P. Pera, R. J. Bernacki, S. Xia,
S. B. Horwitz, C. L. Simmerling and I. Ojima, Chem. Biol.,
2005, 12, 339–348.
place of a phenyl on C3′ in CTX40. The results of this ongoing 10 M. Suffness, Taxol: science and applications, CRC Press,
collaborative effort will be reported in due course.
Boca Raton, Florida, 1995.
11 E. Nogales, S. G. Wolf and K. H. Downing, Nature, 1998,
391, 199–203.
12 J. Lowe, H. Li, K. H. Downing and E. Nogales, J. Mol. Biol.,
2001, 313, 1045–1057.
Acknowledgements
13 L. Sun, C. Simmerling and I. Ojima, ChemMedChem, 2009,
4, 719–731.
14 S. Forli, F. Manetti, K. H. Altmann and M. Botta,
ChemMedChem, 2010, 5, 35–40.
15 J. P. Snyder, J. H. Nettles, B. Cornett, K. H. Downing and
E. Nogales, Proc. Natl. Acad. Sci. U. S. A., 2001, 98, 5312–
5316.
16 D. G. Kingston, S. Bane and J. P. Snyder, Cell Cycle, 2005, 4,
279–289.
17 A. A. Alcaraz, A. K. Mehta, S. A. Johnson and J. P. Snyder,
J. Med. Chem., 2006, 49, 2478–2488.
18 W. P. Walters, J. Green, J. R. Weiss and M. A. Murcko,
J. Med. Chem., 2011, 54, 6405–6416.
19 I. Ojima, S. Chakravarty, T. Inoue, S. Lin, L. He,
S. B. Horwitz, S. D. Kuduk and S. J. Danishefsky, Proc. Natl.
Acad. Sci. U. S. A., 1999, 96, 4256–4261.
20 I. Ojima and M. Das, J. Nat. Prod., 2009, 72, 554–565.
21 T. Ganesh, C. Yang, A. Norris, T. Glass, S. Bane,
R. Ravindra, A. Banerjee, B. Metaferia, S. L. Thomas,
P. Giannakakou, A. A. Alcaraz, A. S. Lakdawala, J. P. Snyder
and D. G. Kingston, J. Med. Chem., 2007, 50, 713–725.
22 N. Singh and A. Warshel, Proteins, 2010, 78, 1724–1735.
The authors are very grateful to Prof. José Manuel Andreu (CIB,
CSIC) for sustained advice, encouragement and useful discus-
sions. C.C. enjoys an FPU grant awarded by the Spanish Minis-
try of Science and Innovation (AP2007-01225). A.M. and J.K.
acknowledge financial support from Comunidad de Madrid
(AMAROUTO program to the Fundación Severo Ochoa) and
Ministerio de Economía y Competitividad (BFU2011-24595),
respectively. This work was financed by the National Natural
Science Foundation of China (NSFC grant 30930108 to W.S.F.),
the Spanish Ministerio de Ciencia e Innovación (BIO2008-
04384 and BIO-2010-16351 to F.D.), Comisión Interministerial
de Ciencia y Tecnología (SAF2009-13914-C02-02 to F.G.), and
Comunidad de Madrid (S-BIO/0214/2006 and S2010-BMD-2457
to A.M., F.D. and F.G.). The computer resources, technical
expertise and assistance provided by the Barcelona Super-
computing Center (Centro Nacional de Supercomputación) are
gratefully acknowledged.
Notes and references
1 P. B. Schiff, J. Fant and S. B. Horwitz, Nature, 1979, 277, 23 W.-S. Fang, Q.-C. Fang and X.-T. Liang, Tetrahedron Lett.,
665–667. 2001, 42, 1331–1333.
2 P. Breedveld, J. H. Beijnen and J. H. Schellens, Trends 24 A. Morreale and F. Gago, in Physico-Chemical and Compu-
Pharmacol. Sci., 2006, 27, 17–24.
3 I. Ojima, J. C. Slater, E. Michaud, S. D. Kuduk,
P. Y. Bounaud, P. Vrignaud, M. C. Bissery, J. M. Veith,
tational Approaches to Drug Discovery, ed. F. Luque and
X. Barril, Royal Society of Chemistry, 1st edn, 2012,
pp. 244–272.
P. Pera and R. J. Bernacki, J. Med. Chem., 1996, 39, 3889– 25 C. K. Coderch, J. Klett, A. Morreale, J.-F. Díaz and F. Gago,
3896. ChemMedChem, 2012, 7, 836–843.
4 Y. F. Wang, Q. W. Shi, M. Dong, H. Kiyota, Y. C. Gu and 26 P. J. Goodford, J. Med. Chem., 1985, 28, 849–857.
B. Cong, Chem. Rev., 2011, 111, 7652–7709.
5 I. Ojima, J. C. Slater, S. D. Kuduk, C. S. Takeuchi,
27 C. Pérez, M. Pastor, A. R. Ortiz and F. Gago, J. Med. Chem.,
1998, 41, 836–852.
R. H. Gimi, C. M. Sun, Y. H. Park, P. Pera, J. M. Veith and 28 S. T. M. Simila and S. F. Martin, J. Org. Chem., 2007, 72,
R. J. Bernacki, J. Med. Chem., 1997, 40, 267–278.
5342–5349.
Org. Biomol. Chem.
This journal is © The Royal Society of Chemistry 2013

View Article Online
Organic & Biomolecular Chemistry
Paper
29 P. Bertrand and J. P. Gesson, J. Org. Chem., 2007, 72, 3596–
J. A. Pople, Gaussian 03 (Revision B.04), Gaussian, Inc.,
3599.
Pittsburgh, PA, 2003.
30 R. Brieva, J. Z. Crich and C. J. Sih, J. Org. Chem., 1993, 58, 38 C. I. Bayly, P. Cieplak, W. Cornell and P. A. Kollman,
1068–1075. J. Phys. Chem., 1993, 97, 10269–10280.
31 J. F. Diaz, I. Barasoain and J. M. Andreu, J. Biol. Chem., 39 R. Gil-Redondo, J. Klett, F. Gago and A. Morreale, Proteins,
2003, 278, 8407–8419. 2010, 78, 162–172.
32 R. M. Buey, I. Barasoain, E. Jackson, A. Meyer, 40 A. Nicholls and B. Honig, J. Comput. Chem., 1991, 12, 435–
P. Giannakakou, I. Paterson, S. Mooberry, J. M. Andreu and
J. F. Diaz, Chem. Biol., 2005, 12, 1269–1279.
33 J. C. Gordon, J. B. Myers, T. Folta, V. Shoja, L. S. Heath and
A. Onufriev, Nucleic Acids Res., 2005, 33, W368–W371.
34 W. D. Cornell, P. Cieplak, C. I. Bayly, I. R. Gould,
445.
41 H. Abdi, Wiley Interdisciplinary Reviews: Computational Stat-
istics, 2010, vol. 2, pp. 97–106.
42 W. L. DeLano, The PyMOL molecular graphics system,
DeLano Scientific, LLC, Palo Alto, CA, 2006.
K. M. Merz, D. M. Ferguson, D. C. Spellmeyer, T. Fox, 43 C. Pérez and A. R. Ortiz, J. Med. Chem., 2001, 44, 3768–
J. W. Caldwell and P. A. Kollman, J. Am. Chem. Soc., 1995,
117, 5179–5197.
35 W. D. Cornell, P. Cieplak, C. I. Bayly, I. R. Gould,
K. M. Merz, D. M. Ferguson, D. C. Spellmeyer, T. Fox,
3785.
44 L. Wang, A. A. Alcaraz, R. Matesanz, C. G. Yang,
I. Barasoain, J. F. Diaz, Y. Z. Li, J. P. Snyder and W. S. Fang,
Bioorg. Med. Chem. Lett., 2007, 17, 3191–3194.
J. W. Caldwell and P. A. Kollman, J. Am. Chem. Soc., 1996, 45 F. Rodriguez-Barrios and F. Gago, J. Am. Chem. Soc., 2004,
118, 2309–2309. 126, 2718–2719.
36 H. Aldaz, L. M. Rice, T. Stearns and D. A. Agard, Nature, 46 H. Freedman, J. T. Huzil, T. Luchko, R. F. Luduena and
2005, 435, 523–527. J. A. Tuszynski, J. Chem. Inf. Model., 2009, 49, 424–436.
37 M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, 47 C. Ferlini, G. Raspaglio, S. Mozzetti, L. Cicchillitti,
M. A. Robb, J. R. Cheeseman, J. A. Montgomery, T. Vreven,
K. N. Kudin, J. C. Burant, J. M. Millam, S. S. Iyengar,
F. Filippetti, D. Gallo, C. Fattorusso, G. Campiani and
G. Scambia, Cancer Res., 2005, 65, 2397–2405.
J. Tomasi, V. Barone, B. Mennucci, M. Cossi, G. Scalmani, 48 P. W. Snyder, J. Mecinovic, D. T. Moustakas, S. W. Thomas
N. Rega, G. A. Petersson, H. Nakatsuji, M. Hada, M. Ehara,
K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima,
Y. Honda, O. Kitao, H. Nakai, M. Klene, X. Li, J. E. Knox,
III, M. Harder, E. T. Mack, M. R. Lockett, A. Heroux,
W. Sherman and G. M. Whitesides, Proc. Natl. Acad.
Sci. U. S. A., 2011, 108, 17889–17894.
H. P. Hratchian, J. B. Cross, V. Bakken, C. Adamo, 49 D. H. Williams, J. P. L. Cox, A. J. Doig, M. Gardner,
J. Jaramillo, R. Gomperts, R. E. Stratmann, O. Yazyev,
A. J. Austin, R. Cammi, C. Pomelli, J. W. Ochterski,
P. Y. Ayala, K. Morokuma, G. A. Voth, P. Salvador,
U. Gerhard, P. T. Kaye, A. R. Lal, I. A. Nicholls, C. J. Salter
and R. C. Mitchell, J. Am. Chem. Soc., 1991, 113, 7020–
7030.
J. J. Dannenberg, V. G. Zakrzewski, S. Dapprich, 50 M. J. Todd, I. Luque, A. Velazquez-Campoy and E. Freire,
A. D. Daniels, M. C. Strain, O. Farkas, D. K. Malick, Biochemistry, 2000, 39, 11876–11883.
A. D. Rabuck, K. Raghavachari, J. B. Foresman, J. V. Ortiz, 51 G. R. Marshall, J. Comput. Aided Mol. Des., 2012, 26, 3–8.
Q. Cui, A. G. Baboul, S. Clifford, J. Cioslowski, 52 I. Luque, M. J. Todd, J. Gomez, N. Semo and E. Freire,
B. B. Stefanov, G. Liu, A. Liashenko, P. Piskorz,
Biochemistry, 1998, 37, 5791–5797.
I. Komaromi, R. L. Martin, D. J. Fox, T. Keith, A. Laham, 53 D. G. Vander Velde, G. I. Georg, G. L. Grunewald,
C. Y. Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill,
B. Johnson, W. Chen, M. W. Wong, C. Gonzalez and
C. W. Gunn and L. A. Mitscher, J. Am. Chem. Soc., 1993,
115, 11650–11651.
This journal is © The Royal Society of Chemistry 2013
Org. Biomol. Chem.