Molecular interaction fields vs. quantum-mechanical-based descriptors in the modelling of lipophilicity of platinum(iv) complexes

Article abstract of DOI:10.1039/c2dt32360e

We report QSAR calculations using VolSurf descriptors to model the lipophilicity of 53 Pt(iv) complexes with a diverse range of axial and equatorial ligands. Lipophilicity is measured using an efficient HPLC method. Previous models based on a subset of these data are shown to be inadequate, due to incompatibility of whole molecule descriptors between carboxylato and hydroxido ligands. Instead, the interaction surfaces of complexes with various probes are used as independent descriptors. Partial least squares modelling using three latent variables results in an accurate (R² = 0.92) and robust model (Q² = 0.87) of lipophilicity, that moreover highlights the importance of size and hydrophobicity terms and the modest relevance of hydrogen bonding.

Full text of DOI:10.1039/c2dt32360e

Dalton
Transactions
View Article Online
View Journal | View Issue
PAPER
Molecular interaction fields vs. quantum-mechanical-
based descriptors in the modelling of lipophilicity of
platinum(IV) complexes†
Cite this: Dalton Trans., 2013, 42, 3482
Giuseppe Ermondi,a Giulia Caron,a Mauro Ravera,b Elisabetta Gabano,b
Sabrina Bianco,b James A. Plattsc and Domenico Osella*b
We report QSAR calculations using VolSurf descriptors to model the lipophilicity of 53 Pt(IV) complexes
with a diverse range of axial and equatorial ligands. Lipophilicity is measured using an efficient HPLC
method. Previous models based on a subset of these data are shown to be inadequate, due to incompat-
ibility of whole molecule descriptors between carboxylato and hydroxido ligands. Instead, the interaction
surfaces of complexes with various probes are used as independent descriptors. Partial least squares
modelling using three latent variables results in an accurate (R2 = 0.92) and robust model (Q2 = 0.87) of
lipophilicity, that moreover highlights the importance of size and hydrophobicity terms and the modest
relevance of hydrogen bonding.
Received 5th October 2012,
Accepted 5th December 2012
DOI: 10.1039/c2dt32360e
www.rsc.org/dalton
calculations.1 In this context lipophilicity is a key feature
because it is directly related to the ability of a molecule to
Introduction
It is well known that similar molecules, with just minor struc- cross passively cell membranes.2,3 The partition coefficient log
tural variations, can exhibit dramatically different biological Po/w is widely used to represent molecular lipophilicity because
behaviour. Consequently, the development of the quantitative it measures the differential solubility of a compound between
structure–activity relationships (QSAR), i.e. mathematical n-octanol (a model of the lipid bilayer of a cell membrane) and
relationships linking chemical structure and pharmacological water (the solvent in and out of cells). The importance of this
activity, helps the chemists to design out negative properties parameter is confirmed by the fact that log Po/w is one of the
and incorporate positive attributes to the molecules under properties identified by Lipinski in the “Rule of 5” for drug-
investigation. The final goal is to design potentially active com- like molecules.4 Lipophilicity of a molecule is generally
pounds prior to their synthesis, and to discharge drug candi- thought to arise from its size and polarity. Thus, a large,
dates with unfavourable pharmacokinetic/pharmacodynamic apolar molecule will have a tendency to partition into the
profiles, thus limiting the amount of in vitro and in vivo experi- organic phase, where formation of cavities is easier, while a
ments required in drug discovery and optimisation.
more polar molecule will tend to partition into water, the more
Considering the complexity of the ADME (Absorption, Dis- polar and hydrogen bonding solvent.5
tribution, Metabolism and Excretion) prediction of drug candi-
Although a lipophilicity experiment, by using the tra-
dates, today the original dependent variable “A” in QSAR, ditional shake-flask method, seems at first glance simple and
representing the desired therapeutic effect, may be replaced by easy, it is not a trivial matter because data can vary signifi-
other, generic properties “P” (QSPR, i.e. quantitative structure– cantly with experimental conditions. The main difficulty in the
property relationship) in order to limit the extension of shake-flask method comes from the significant errors affecting
the final log Po/w in the case of extreme values (i.e. very polar
or very lipophilic compounds), that comes from an unbalanced
aCASMedChem Research Group at the Centre for Innovation, Dipartimento di
partition in favour of the aqueous or the organic phase,
Biotecnologie Molecolari e Scienza della Salute, Università di Torino,
respectively. This is the case of the well-known Pt(II)-based
Via Quarello 11, 10135 Torino, Italy
bDipartimento di Scienze e Innovazione Tecnologica (DiSIT), Università del Piemonte
Orientale “Amedeo Avogadro”, Viale Michel 11, 15121 Alessandria, Italy.
E-mail: domenico.osella@unipmn.it
antitumor drug cisplatin and its analogues, characterized
by quite negative log Po/w values. Moreover the shake-flask
method is generally slow, expensive and, sometimes, poorly
reproducible.
For these reasons, alternative experimental methods have
been developed to measure the lipophilicity. Among them, the
cSchool of Chemistry, Cardiff University, Park Place, Cardiff CF10 3AT, Wales, UK
†Electronic supplementary information (ESI) available: Details of the syntheses
and characterization of complexes 1–22; list of MOPAC2009 descriptors; MIFs of
complex 15. See DOI: 10.1039/c2dt32360e
3482 | Dalton Trans., 2013, 42, 3482–3489
This journal is © The Royal Society of Chemistry 2013

View Article Online
Dalton Transactions
Paper
retention parameters in RP-HPLC (where n-octanol is ideally platinum drugs, allowing potential candidates to be proposed
replaced by the C18 chains functionalizing silica as the station- before lengthy synthesis and testing. In the literature a limited
ary phase, while the mobile phase consists of various mixtures number of publications report on models to predict log
of water and organic co-solvents) are often used, even if criti- Po/w values of Pt(II)19–24 and Pt(IV)20,22,24 complexes by using
cized by some authors as not truly replacing the shake-flask different descriptors and, more importantly, different mathe-
method.6–8 Since RP-HPLC retention is due to partitioning matical approaches. In this framework, a statistically accurate
between (polar) mobile and (apolar) stationary phases, there is model for the prediction of log Po/w of a series of Pt(IV) com-
a straightforward correlation between the partition coefficient plexes containing carboxylates or chlorides as axial ligands has
and the HPLC capacity factor k′ (k′ = (tR − t0)/t0, where t0 is the been recently published by us. A combination of surface area
retention time for an unretained compound and tR is the and atomic charges was necessary to build the model.25 This
retention time of the analyte). The log k′0 (k0′ is the HPLC study also demonstrated the ability of the PM6 semi-empirical
capacity factor extrapolated to 100% water) values of com- method26 to accurately reproduce the X-ray geometry of Pt(IV)
pounds with known log Po/w can be used to create a calibration complexes such as those discussed here.
curve (log Po/w = a log k′0 + b) to evaluate partition coefficients
from chromatographic data.
In order to evaluate the applicability of the above-reported
model to other Pt(IV) complexes, a new set of compounds has
Historically, log Po/w values are preferred because one been studied. In the present work 22 dihydroxido Pt(IV) com-
would like to have a comparison with standard experimental plexes (1–22 in Fig. 1) have been newly synthesised or, in some
lipophilicity values or with calculated log P values. However, cases, re-synthesised according to standard methods.27–30
log k′0 is a good lipophilicity index per se and is generally used Their extrapolated capacity factors log k′0 were measured and
because of its direct correlation with log Po/w values. Moreover, in silico properties evaluated by QSRR (Quantitative Structure–
being obtained only in an experimental way log k′0 does not Retention Relationship) methods together with the previously
require any conversion, in such a way that no further experi- studied Pt(IV) complexes (23–53 in Fig. 1).
mental error is introduced in the final lipophilicity value.
In recent years octahedral Pt(IV) complexes have emerged as
an alternative to traditional cisplatin-like compounds. They are
generally considered to be antitumor pro-drugs that can be
Results and discussion
Synthesis and characterization of Pt(IV) complexes
reduced in vivo to the active Pt(II) metabolite losing the axial
ligands (L) in the hypoxic tumor milieu.9 These compounds
The dihydroxido Pt(IV) complexes 1–22 were obtained by oxi-
dation30 of the parent Pt(II) compound27–29 with hydrogen per-
are more inert to ligand substitution reactions than Pt(II)
counterparts, leading to lower systemic toxicity from unwanted
side-reactions and increasing the likelihood of the drug reach-
several times with cold water. All the dihydroxido Pt(IV) com-
ing its cellular target.
oxide (see ESI† for the details of the syntheses). The samples
contained residual H2O2, removed by washing the compounds
plexes, except compounds 5–8, are poorly soluble in water.
Fig. 1 shows the entire set of the studied complexes.
Interestingly, satraplatin (trans,cis,cis-bis(acetato)ammine-
(cyclohexylamine)dichloridoplatinum(IV), JM216) is actually on
clinical trials as an orally active antitumor drug.10
Determination of extrapolated capacity factors, log k0′
Dihydroxido Pt(IV) complexes represent an interesting sub-
class of Pt(IV) compounds.11 In clinical experience iproplatin
All the 22 dihydroxido complexes under study were injected
into RP-HPLC and their log k′0 values, i.e. the capacity factor
(cis,trans,cis-dichloridodihydroxidobis(isopropylamine)plati-
extrapolated to 0% of organic co-solvent, namely MeOH, were
measured (Table 1). To obtain a more general QSRR relation-
ship, the log k′0 values of 31 other Pt(IV) complexes, determined
under the same experimental conditions, were added in this
study.25 The log k′0 of the entire series of Pt(IV) complexes cover
a wide range of values (ca. 8 log units). The increase of the
lipophilicity is expected to enhance cellular uptake by passive
diffusion.20,31,32
num(IV), also nicknamed CHIP or JM9) demonstrated excellent
activity in several phase II trials.12–14 However, this compound
was abandoned due to the lack of any superior performance
with respect to cisplatin.15
In a previous study, a relationship between cytotoxicity and
both reduction potential Ep and partition coefficient log Po/w
of Pt(IV) complexes has been established: the easier the
reduction and the higher the lipophilicity, the higher the cyto-
toxicity.16 These two chemico-physical properties, in turn
related to cellular uptake and activation by reduction in
QSRR analysis
cytosol, can be tuned through choice of the six ligands around The log k′0 of some compounds of the series (23–53) were suc-
the Pt(IV) centre, especially the axial ones.10,17 These ligands cessfully modelled with a limited set of QM descriptors.25 The
can also act as carriers or adjuvant agents, offering interesting same approach failed when applied to the whole series of 1–53
applications in controlled release and in drug targeting and complexes. This is probably due to the increased heterogeneity
delivery.18
of the training set, in turn due to the different axial ligands
The ability to predict relevant physicochemical properties present in 1–22 with respect to 23–53 compounds (OH instead
of platinum complexes directly from their structures would of carboxylates) that enormously enlarge the chemical domain
represent an important step in rational design of new of the molecules. This is shown by the analysis of the score
This journal is © The Royal Society of Chemistry 2013
Dalton Trans., 2013, 42, 3482–3489 | 3483

View Article Online
Paper
Dalton Transactions
Fig. 1 Sketch of the complexes under study: the dihydroxido complexes 1–22 are reported here for the first time, while complexes 23–53 were previously
studied.17,25
plot of the Principal Component Analysis (PCA) based on QM of large data tables, in which the number of the original vari-
descriptors (Fig. 2). PCA is a general tool for the interpretation ables is reduced by a projection of the objects (i.e. the
3484 | Dalton Trans., 2013, 42, 3482–3489
This journal is © The Royal Society of Chemistry 2013

View Article Online
Dalton Transactions
Paper
molecules) onto a smaller number of new variables termed molecular interaction fields (MIFs)35 obtained employing the
principal components (PC).
GRID force field.36–38 Interaction fields are obtained with
The PCs are orientated so that the first PC describes as different probes and the surfaces of the regions that encom-
much as possible of the original variation between the objects. pass interaction energy values under certain cutoff limits are
The second PC is orientated in an orthogonal manner to the calculated. In particular, water (OH2), hydrophobic (DRY),
first PC and is directed to describe as much as possible of the hydrogen bond donor (HBD, amide N1) and hydrogen bond
remaining variation and so on. The projection of objects onto acceptor (HBA, carbonyl O) probes have been considered in
a PC is called score. By plotting the scores for two PCs it is the present work.
possible to graphically find similarities and differences
Since VolSurf descriptors represent polarity and hydropho-
between objects. Fig. 2A shows that the two main PCs (PC1 bicity (as well as size and shape) of molecules, they are gener-
and PC2) based on the QM descriptors split the whole series of ally well suited for the modelling of lipophilicity indexes.39
compounds in the two groups (1–22 and 23–53) characterised This was shown in previous studies where robust QSRR
by different kinds of axial ligands. This result confirms that models based on VolSurf descriptors were used to model
the previously selected series of QM descriptors (Table S1†)25 the variation of different chromatographic lipophilicity
are not suited to exhaustively model the chromatographic be- indexes.40,41 Fig. 2B confirms the suitability of the application
haviour of the whole series of platinum complexes.
of VolSurf descriptors to the whole series of investigated com-
A totally different approach has been attempted, replacing plexes, with no discrimination of carboxylato, chlorido or
QM descriptors with descriptors based on 3D molecular fields, hydroxido species. As an example, Fig. 3 shows the visual rep-
i.e. VolSurf descriptors.33,34 Briefly, VolSurf is a computational resentation of two of the 92 VolSurf descriptors obtained with
procedure to find out molecular descriptors from 3D OH2 and DRY probes. They represent the volume of the mol-
ecular envelope which is accessible to, and interacts attrac-
tively with the OH2 probe at −5.0 kcal mol−1 and with the DRY
Table 1 log k’0 values for the complete 1–53 series of Pt(IV) complexes
probe at −0.2 kcal mol−1 for compound 15 (see Experimental
a
a
a
Cmpd
log k′0
Cmpd
log k′0
Cmpd
log k′0
for details and ESI† for more envelopes).
A relationship between log k′0 and the VolSurf descriptors
was obtained by the Partial Least Squares (PLS) regression
technique. A three latent variables (3LVs) model was found:
the three main components explained about 90% of the total
variance (R2 = 0.92) while the root mean square of the errors
(RMSE) was 0.3. To validate these models internal validation
was firstly used. Whereas some researchers in the QSAR field
support internal validation, others consider that internal vali-
dation is not a sufficient test to check the robustness of
models, and external validation is necessary.42 In this case,
however, since the sample size is relatively small, and thus
holding a portion of it back for testing would be wasteful, it
was preferred to use cross-validation, with multiple rounds
using different partitions.
1
2
3
4
5
6
7
8
−0.97
−0.50
0.64
−0.90
−0.75
−0.55
−0.24
−0.08
1.20
2.16
2.84
−0.96
−0.75
−0.89
−0.50
0.19
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.23
−0.56
0.98
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
2.29
3.53
−0.46
−0.16
0.31
1.98
5.11
0.58
−0.78
−0.66
0.37
−0.25
0.68
1.72
2.94
4.22
−0.60
−0.70
−0.06
−0.32
0.40
3.83
9
−0.83
−0.51
1.34
1.60
4.56
6.98
1.60
3.70
4.26
10
11
12
13
14
15
16
17
18
−0.58
−0.63
1.29
In particular the 53 compounds were assigned in a random
way to N groups, each one containing an equal (or nearly equal)
a log k′0 for complexes 23–53 are from ref. 25.
Fig. 2 The score plot of the first two main PCs (in parentheses the variance %): (A) QM descriptors, (B) VolSurf descriptors.
This journal is © The Royal Society of Chemistry 2013
Dalton Trans., 2013, 42, 3482–3489 | 3485

View Article Online
Paper
Dalton Transactions
Fig. 3 MIF generated by the OH2 probe at −5.0 kcal mol−1 (left) and by the DRY probe at −0.2 kcal mol−1 (right) for compound 15.
Table 2 BR values for the VolSurf PLS model of log k0’
Block
No. descriptors
BR
Size
Water
DRY
HBA
HBD
Others
7
23
16
15
15
16
3.24
0.74
1.46
0.35
0.58
0.94
Importance in Projection (VIP) plots are often used to over-
come this limit.43 Briefly, the VIPs show which descriptors are
the most important for the model, whereas the sign of the PLS
coefficients indicates the positive or negative contribution of
the descriptor to the investigated variable. However the use of
VIPs for model interpretation is somewhat subjective and
therefore remains difficult to understand.
Recently, some of us described a new method to obtain a
mechanistic interpretation of PLS models by introducing the
concept of block of descriptors.41 This approach can be
applied to any set of molecular descriptors but is particularly
suited for VolSurf+ descriptors. Since technical reasons
prevent the use of VolSurf+ software with platinum complexes,
here we slightly modified the method described above to adapt
it to VolSurf descriptors. The relevance of any block of descrip-
tors to the PLS model is calculated by the Block Relevance (BR)
parameter (see the Experimental part). Table 2 lists the BR
values for the model discussed above, and demonstrates that
log k′0 is mainly characterized by the block of size descriptors
(BR = 3.24), which is at least twice as important as any other
block.41
Fig. 4 Correlation between experimental (Exp) and calculated (Calc) log k’0.
number of compounds. Then models were built keeping one
of these groups out of the analysis until all of the compounds
were kept out once. The formation of the groups and the
validation was repeated 50 times. Results were similar
with different partitions: Q2 (LOO) = 0.87; Q2 (N = 4) = 0.85;
Q2 (N = 3) = 0.85.
Finally, to further validate the PLS model the order of
Y values which produced unacceptable R2 and Q2 values was
randomized (data not shown).
The relationship between experimental and calculated
values is shown in Fig. 4. A slope of approximately 1 and an
intercept of about 0 indicate very good correspondence
between experimental and calculated values.
The main drawback of PLS models is the difficulty of their
interpretation from the chemical standpoint. The Variable
To take into account the sign of the PLS coefficient, BR is
split into BR(+) relating to retention in the stationary phase,
and into BR(−) relating to retention in the mobile phase (see
the Experimental part). Fig. 5 shows the trend of BR(+) and
3486 | Dalton Trans., 2013, 42, 3482–3489
This journal is © The Royal Society of Chemistry 2013

View Article Online
Dalton Transactions
Paper
Purity of compounds was assessed by analytical RP-HPLC
(see below), elemental analysis and determination of Pt
content by inductively coupled plasma-optical emission spec-
troscopy (ICP-OES). Elemental analyses were carried out with
an EA3000 CHN Elemental Analyzer (EuroVector, Milano,
Italy). Platinum was quantified by means of a Spectro Genesis
ICP-OES spectrometer (Spectro Analytical Instruments, Kleve,
Germany) equipped with a crossflow nebulizer. In order to
quantify the platinum concentration the Pt 299.797 nm line
was selected. A platinum standard stock solution of 1000 mg L−1
was diluted in 1.0% v/v nitric acid to prepare calibration
standards.
Fig. 5 BR(+) and BR(−) for the PLS model of log k’0.
Synthesis of platinum complexes
BR(−) for the investigated system. A positive coefficient means
that an increase of the block of the considered descriptors
causes an increase in log k′0 (and thus the retention in station-
ary phase); the reverse is true for negative coefficients. The rel-
evance of the block of size and hydrophobic descriptors
(Fig. 5) indicates that the lipophilicity of these complexes is
mainly driven by the ligands rather than by platinum(IV) core
properties. The importance of this block of descriptors, and
their positive sign, is consistent with established ideas of the
molecular properties that affect lipophilicity.5 It is noteworthy
that this conclusion can be drawn only using MIFs based
descriptors because of the limits of QM descriptors in the mod-
elling of chromatography behaviour of platinum complexes.
The dihydroxido Pt(IV) complexes 1–22 (Fig. 1) were obtained
by oxidation with hydrogen peroxide30 of the parent Pt(II) com-
pound obtained from Dhara’s method and its adaptations.27–29
Briefly, K2[PtI4] was produced in solution by the reaction of
K2[PtCl4] (1 mmol) and KI (6 mmol), and reacted with an
amine ligand (3.3 mmol for monodentate amines, A, and
1.7 mmol for bidentate ones, A2) to give the cis-[PtA2I2] precipi-
tate. This was isolated by centrifugation and washed with
water, ethanol and diethyl ether. Upon reaction with AgNO3
(1.96 mmol) or Ag2SO4 (0.98 mmol), the corresponding diaqua
intermediate was formed. After removal of AgI, the diaqua
species reacted with chlorides or carboxylates to yield the final
Pt(II) complex. The yields were from 70 to 85%. This complex
(0.5 mmol) reacted with a tenfold excess of 35% hydrogen per-
oxide (440 μL) in water at RT for 24 h. The solvent was then
removed under reduced pressure and the dihydroxido product
was collected as a pale yellow solid, which was washed with
cold water, methanol and diethyl ether and dried in vacuo. The
yields were from 50 to 70% (see ESI† for details).
Experimental
General
K2[PtCl4] (Johnson Matthey and Co.) and all other chemicals
(Aldrich) were used without further purification.
The multinuclear NMR spectra were measured on a JEOL
Eclipse Plus operating at 400 MHz (1H), 100.5 MHz (13C) and
RP-HPLC method
1
85.9 MHz, respectively. H and 13C NMR chemical shifts were
Chromatographic analysis was used to evaluate the purity and
the capacity factors of the compounds. The chromatographic
conditions were:19 silica-based C18 stationary phase (5 μm
Phenomenex Gemini® C18 column 25 × 3 mm ID); mobile
phase containing 15 mM HCOOH aqueous solution with
different percentage of MeOH (flow rate = 0.75 mL min−1; iso-
cratic elution, UV-visible detector set at 210 nm). KCl was the
internal reference to determine the column dead-time (t0).45
Platinum complex solutions were 0.25 mM. A chromatogram
for each complex with every different eluant composition was
performed (the methanol fraction, φ, ranging from 20 to 70%)
and the corresponding retention time tR was used to calculate
log k′ (k′ = (tR − t0)/t0).
reported in parts per million (ppm) referenced to solvent reso-
nances; for measurements in D2O 1% methanol was added as
an internal reference. 195Pt NMR spectra were recorded using a
solution of K2[PtCl4] in saturated aqueous KCl as the external
reference. The shift for K2[PtCl4] was adjusted to −1628 ppm
from Na2[PtCl6] (δ = 0 ppm).
RP-HPLC and mass analysis were performed using a Waters
HPLC-MS instrument equipped with an Alliance 2695 separ-
ations module, a 2487 dual lambda absorbance detector, and
a 3100 mass detector. Electrospray ionization mass spectra
(ESI-MS) were obtained delivering a diluted solution of the
compound in methanol directly into the spectrometer source
at 0.01 mL min−1 44
The source and desolvation temperatures
.
From these data, the extrapolation of the log k′ to 0%
MeOH (log k′0), corresponding to the capacity factor in pure
water, for all compounds has been performed (eqn (1)).17,25,46
were set to 150 and 250 °C, respectively, with nitrogen used
both as a drying and a nebulizing gas. The cone and the capil-
lary voltages were usually 30 V and 2.70 kV, respectively. Quasi-
molecular ion peaks [M + H]+ or sodiated [M + Na]+ peaks were
log k′ ¼ log k′0 ꢀ Sφ
ð1Þ
assigned on the basis of the m/z values and of the simulated where S is a solute-dependent solvent strength specific to the
isotope distribution patterns.
organic modifier on the stationary phase under consideration.
This journal is © The Royal Society of Chemistry 2013
Dalton Trans., 2013, 42, 3482–3489 | 3487

View Article Online
Paper
Dalton Transactions
Quantum-mechanical (QM) descriptors
where i is the number of blocks (6 here, Table 2), N is the
number of descriptors for any block, VIP is the value of each
predictor fitting the PLS model. The higher the value of BR,
the more important is that block.
Depending on the sign of the PLS coefficient, BR was
broken down into BR(+) relating to retention in the stationary
phase, and BR(−) relating to retention in the mobile phase
(eqn (3)).
Surface area, volume, PSA (surface sum over oxygen and nitro-
gen atoms), frontier orbital energies, dipole moment and
atomic partial charges were extracted from the PM626
(MOPAC2009, http://openmopac.net/ and Spartan 08, Wave-
function, Irvine, CA, USA) fully optimised geometries of com-
plexes 1–22 (see Table S1, ESI†). The same set of data for 23–53
was taken from the literature.25
BRi ¼ BRiðþÞ þ BRiðꢀÞ
ð3Þ
Principal Component Analysis (PCA)
PCA was performed using Simca v.13 (Umetrics Sweden).
Conclusions
VolSurf model
We report the HPLC measurement and QSRR modelling of 53
Pt(IV) parameterization in the GRID force field implemented in Pt(IV) complexes as models of potential anti-cancer pro-drugs.
VolSurf v.4.1.2 software (http://www.moldiscovery.com/) was New values of log k′0 for 22 dihydroxido complexes have been
performed according to a procedure similar to that described measured, and combined with 31 previously reported values to
elsewhere for Pt(II).39 The geometries of fully optimized com- result in a large and diverse set of complexes whose lipophili-
plexes were saved in mol2 format and submitted to GRID to city spans 8 orders of magnitude, and hence act as a stringent
obtain the molecular interaction fields (MIFs). The binary.kont test of possible statistical models. We show that conventional
files of all structures were then submitted to VolSurf for the models based on whole molecule properties, extracted from
calculation of the 92 VolSurf descriptors. These were extracted semi-empirical PM6 calculations, are incapable of modelling
from MIFs obtained with the water (OH2), hydrophobic (DRY), this larger dataset, in contrast to an earlier study. On the con-
hydrogen bond donor (HBD, amide N1) and hydrogen bond trary, sampling to the interaction of each complex with a series
acceptor (HBA, carbonyl O) probes. Partial Least Squares (PLS) of probes and using these interactions as independent vari-
analysis was used as implemented in the VolSurf software.
ables results in an accurate and robust model. This VolSurf
Chemical interpretation of the PLS model was performed as model, using three latent variables, not only correlates and
described in detail elsewhere using VolSurf+ descriptors.39,40 predicts lipophilicity with good accuracy, but also highlights
Since for technical reasons VolSurf+ descriptors for platinum the molecular properties that determine lipophilicity, showing
complexes cannot be calculated, here we used VolSurf descrip- that size and hydrophobic interactions play major roles. This
tors. These latter were grouped in six blocks: (a) descriptors study confirms the superiority of VolSurf over QM descriptors
that characterize the size and shape of the solute (7 descriptors to model lipophilicity of platinum(IV) complexes, and suggests
in the text, briefly called size, color-code green), (b) 23 descrip- that they have excellent potential in modelling biologically
tors that express the solute’s interaction with water molecules active transition metal complexes in general.
(in the text indicated as water, colour-code light blue), (c) 15
descriptors that describe the solute’s ability to form hydrogen
bond interactions with the donor group of the probe (that
mimics the chromatographic system, colour-code blue, see
Acknowledgements
This research was carried out within the framework of the
European Union CMST COST Action CM1105 “Functional
metal complexes that bind to biomolecules” and Inter-Univer-
sity Consortium for Research on the Chemistry of Metal Ions
in Biological Systems (CIRCMSB, Bari, Italy).
below), (d) 15 descriptors expressing the solute’s ability to
form hydrogen bond interactions with the acceptor group of
the probe (that mimics the chromatographic system, colour-
code red, see below), (e) 16 descriptors describing the solute’s
propensity of the solute to participate in hydrophobic inter-
actions (in the text called DRY for short, colour-code yellow),
(f) 16 descriptors mainly describing the imbalance between
hydrophilic and hydrophobic regions (in the text called others,
colour-code grey). More details about the significance of the
descriptors can be found in the original paper.40
Notes and references
1 H. Kubinyi, in QSAR: Hansch Analysis and Related
Approaches (Methods and Principles in Medicinal Chem-
istry, vol. 1), ed. R. Mannhold, P. Krogsgaard-Larse and
H. Timmermann, VCH, Weinheim, 1993, ch. 1.
2 G. Caron and G. Ermondi, in Molecular Drug Properties
Measurement and Prediction (Methods and Principles in
Medicinal Chemistry, vol. 37), ed. R. Mannhold, Wiley-
VCH, Weinheim, 2008, ch. 12.
Block Relevance (BR) was defined as the ratio of the sum of
the squared Variable Importance Projection (VIP) values of a
given block of descriptors and the number of those descriptors
(eqn (2)).
P
Ni
j¼1 VIP2j
Ni
BRi ¼
ð2Þ
3488 | Dalton Trans., 2013, 42, 3482–3489
This journal is © The Royal Society of Chemistry 2013

View Article Online
Dalton Transactions
Paper
3 X. Liu, B. Testa and A. Fahr, Pharm. Res., 2011, 28, 962.
4 C. A. Lipinski, F. Lombardo, B. W. Dominy and P. J. Feeney,
Adv. Drug Delivery Rev., 2001, 46, 3.
5 M. H. Abraham, Chem. Soc. Rev., 1993, 22, 73.
6 W. J. Lambert, J. Chromatogr., A, 1993, 656, 469.
7 T. Braumann, J. Chromatogr., 1986, 373, 191.
21 I. V. Tetko, I. Jaroszewicz, J. A. Platts and J. Kuduk-
Jaworska, J. Inorg. Biochem., 2008, 102, 1424.
22 S. P. Oldfield, M. D. Hall and J. A. Platts, J. Med. Chem.,
2007, 50, 5227.
23 A. A. Toropov, A. P. Toropova and E. Benfenati, J. Math.
Chem., 2009, 46, 1060.
8 S. Martel, D. Guillarme, Y. Henchoz, A. Galland, 24 P. Sarmah and R. C. Deka, J. Comput. Aided Mol. Des., 2009,
J. L. Veuthey, S. Rudaz and P. A. Carrupt, in Molecular Drug 23, 343.
Properties Measurement and Prediction (Methods and Princi- 25 J. A. Platts, G. Ermondi, G. Caron, M. Ravera, E. Gabano,
ples in Medicinal Chemistry, vol. 37), ed. R. Mannhold,
Wiley-VCH, Weinheim, 2008, ch. 13.
L. Gaviglio, G. Pelosi and D. Osella, J. Biol. Inorg. Chem.,
2011, 16, 361.
9 M. D. Hall, H. R. Mellor, R. Callaghan and T. W. Hambley, 26 J. J. P. Stewart, J. Mol. Model, 2007, 13, 1173.
J. Med. Chem., 2007, 50, 3403. 27 S. C. Dhara, Indian J. Chem., 1970, 8, 193.
10 G. Doshi, G. Sonpavde and C. N. Sternberg, Expert Opin. 28 R. C. Harrison and C. A. McAuliffe, Inorg. Chim. Acta, 1980,
Drug Metab. Toxicol., 2012, 8, 103. 46, L15.
11 M. D. Hall, R. D. Dolman and T. W. Hambley, in Metal 29 F. D. Rochon and L. M. Gruia, Inorg. Chim. Acta, 2000, 306,
Complexes in Tumor Diagnosis and as Anticancer Agents 193.
(Metal Ions in Biological Systems, vol. 42), ed. A. Sigel and 30 C. M. Giandomenico, M. J. Abrams, B. A. Murrer,
H. Sigel, Dekker, New York, 2004, pp. 298–322.
J. F. Vollano, M. I. Rheinheimer, S. B. Wyer, G. E. Bossard
12 V. H. C. Bramwell, D. Crowther, S. O’Malley, R. Swindell,
and J. D. Higgins III, Inorg. Chem., 1995, 34, 1015.
R. Johnson, E. H. Cooper, N. Thatcher and A. Howell, 31 W. H. Ang, S. Pilet, R. Scopelliti, F. Bussy, L. Juillerat-Jean-
Cancer Treat. Rep., 1985, 69, 409. neret and P. J. Dyson, J. Med. Chem., 2005, 48, 8060.
13 N. J. Wheate, S. Walker, G. E. Craig and R. Oun, Dalton 32 A. R. Ghezzi, M. Aceto, C. Cassino, E. Gabano and
Trans., 2010, 39, 8113. D. Osella, J. Inorg. Biochem., 2004, 98, 73.
14 L. Pendyala, J. Wayne Cowens, G. B. Chheda, S. P. Dutta, 33 G. Cruciani, P. Crivori, P. A. Carrupt and B. Testa, J. Mol.
J. Patrick and J. Creaven, Cancer Res., 1988, 48, 3533. Struct. (THEOCHEM), 2000, 503, 17.
15 H. Anderson, J. Wagstaff, D. Crowther, R. Swindell, 34 P. Crivori, G. Cruciani, P. A. Carrupt and B. Testa, J. Med.
M. J. Lind, J. McGregor, M. S. Timms, D. Brown and
P. Palmer, Eur. J. Cancer Clin. Oncol., 1988, 24, 1471.
16 P. Gramatica, E. Papa, M. Luini, E. Monti, M. B. Gariboldi,
M. Ravera, E. Gabano, L. Gaviglio and D. Osella, J. Biol.
Inorg. Chem., 2010, 15, 1157.
Chem., 2000, 43, 2204.
35 G. Cruciani, R. Mannhold, H. Kubinyi and G. Folkers, Mol-
ecular Interaction Fields. Applications in Drug Discovery and
ADME Prediction, Wiley-VCH, Zurich, 2006.
36 D. N. Boobbyer, P. J. Goodford, P. M. Mcwhinnie and
R. C. Wade, J. Med. Chem., 1989, 32, 1083.
17 A. M. Montaña and C. Batalla, Curr. Med. Chem., 2009, 16,
2235.
37 P. J. Goodford, J. Med. Chem., 1985, 28, 849.
18 For general reviews see: (a) N. Graf and S. J. Lippard, Adv. 38 R. C. Wade and P. J. Goodford, J. Med. Chem., 1993, 36,
Drug Delivery Rev., 2012, 64, 993; (b) K. S. Lovejoy and 148.
S. J. Lippard, Dalton Trans., 2009, 10651; (c) E. Gabano, 39 G. Caron and G. Ermondi, in Comprehensive Medicinal
M. Ravera and D. Osella, Curr. Med. Chem., 2009, 16, 4544.
For recent, relevant publications on this topic see:
Chemistry, ed. B. Testa and H. van de Waterbeemd, Else-
vier, Oxford, 2nd edn, 2006, vol. 5, pp. 425–452.
(d) S. Mukhopadhyay, C. M. Barnés, A. Haskel, S. M. Short, 40 G. Caron, M. Ravera and G. Ermondi, Pharm. Res., 2011, 28,
K. R. Barnes and S. J. Lippard, Bioconjugate Chem., 2008, 640.
19, 39; (e) L. J. Parker, L. C. Italiano, C. J. Morton, 41 G. Ermondi and G. Caron, J. Chromatogr., A, 2012, 1252, 84.
N. C. Hancock, D. B. Ascher, J. B. Aitken, H. H. Harris, 42 (a) S. Wold, M. Sjostrom and L. Eriksson, Chemom. Intell.
P. Campomanes, U. Rothlisberger, A. De Luca, M. Lo Bello,
W. H. Ang, P. J. Dyson and M. W. Parker, Chem.–Eur. J.,
2011, 17, 7806; (f) L. Gaviglio, A. Gross, N. Metzler-Nolte
Lab. Syst., 2001, 58, 109; (b) T. Scior, J. L. Medina-Franco,
Q.-T. Do, K. Martínez-Mayorga, J. A. Yunes Rojas and
P. Bernard, Curr. Med. Chem., 2009, 16, 4297.
and M. Ravera, Metallomics, 2012, 4, 260; (g) J. Yang, 43 G. Caron and G. Ermondi, J. Med. Chem., 2005, 48, 3269.
X. Sun, W. Mao, M. Sui, J. Tang and Y. Shen, Mol. Pharma- 44 R. Bertani, R. Seraglia, D. Favretto, R. A. Michelin,
ceutics, 2012, 9, 2793.
19 J. A. Platts, S. P. Oldfield, M. M. Reif, A. Palmucci,
M. Mozzon, S. Mazzega Sbovata and S. Sassi, Inorg. Chim.
Acta, 2003, 356, 357.
E. Gabano and D. Osella, J. Inorg. Biochem., 2006, 100, 45 O. Heudi, S. Mercier-Jobard, A. Cailleux and P. Allain, Bio-
1199. pharm. Drug Dispos., 1999, 20, 107.
20 J. A. Platts, D. E. Hibbs, T. W. Hambley and M. D. Hall, 46 K. Valkó, L. R. Snyder and J. L. Glajch, J. Chromatogr., A,
J. Med. Chem., 2001, 44, 472.
1993, 656, 501.
This journal is © The Royal Society of Chemistry 2013
Dalton Trans., 2013, 42, 3482–3489 | 3489