Design, synthesis, and evaluation of antineoplastic activity of novel carbocyclic nucleosides

Article abstract of DOI:10.1002/minf.200900033

Cancer is the leading cause of death among men and women under age 85. Every year, millions of individuals are diagnosed with cancer. But finding new drugs is a complex, expensive, and very time-consuming task. Over the past decade, the cancer research co

Full text of DOI:10.1002/minf.200900033

DOI: 10.1002/minf.200900033
Design, Synthesis, and Evaluation of Antineoplastic
Activity of Novel Carbocyclic Nucleosides
Aliuska M. Helguera ,[a, c, d] J. E. Rodrꢀguez-Borges,[b] Olga CaamaÇo,[e] Xerardo Garcꢀa-Mera,*[e]
Maykel Pꢁrez Gonzꢂlez,[c] and M. Natꢂlia D. S. Cordeiro*[a]
This paper is dedicated to the memory of our dear colleague, Maykel Pꢁrez Gonzꢂlez, who unfortunately has passed away
very suddenly last year.
Abstract: Cancer is the leading cause of death among men
and women under age 85. Every year, millions of individu-
als are diagnosed with cancer. But finding new drugs is a
complex, expensive, and very time-consuming task. Over
the past decade, the cancer research community has
begun to address the in silico modeling approaches, such
as Quantitative Structure-Activity Relationships (QSAR), as
an important alternative tool for targeting potential anti-
cancer drugs. With the compilation of a large dataset of nu-
cleosides synthesized in our laboratories, or elsewhere, and
tested in a single cytotoxic assay under the same experi-
mental conditions, we recognized a unique opportunity to
attempt to build predictive QSAR models. Early efforts with
2D classification models built from part of this dataset were
very encouraging. Here we report a further detailed evalua-
tion of classification models to flag potential anticancer ac-
tivities derived from a variety of 3D molecular representa-
tions. A quantitative 3D-model model that discriminates an-
ticancer compounds from the inactive ones was attained,
which allowed the correct classification of 82% of com-
pounds in such a large and diverse dataset, with only 5%
of false inactives and 11% of false actives. The model devel-
oped here was then used to select and design a new series
of nucleosides, by classifying beforehand them as active/in-
active anticancer compounds. From the compounds so de-
signed, 22 were synthesized and evaluated for their inhibi-
tory effects on the proliferation of murine leukemia cells
(L1210/0), of which 86% were well-classified as active or in-
active, and only two were false actives, corroborating the
good predictive ability of the present discriminant model.
The results of this study thus provide a valuable tool for
the design of novel potent anticancer nucleoside ana-
logues.
Keywords: Antitumor agents · Nucleosides · QSAR · 3D-DRAGON descriptors · Leukemia
1 Introduction
[a] A. M. Helguera , M. N. D. S. Cordeiro
Cancer is one of the major causes of disease and death all
over the world.[1] For instance, in Europe, an estimated 3.2
million people were diagnosed with cancer in 2006, there
were 1.7 million cancer deaths, and the annual number of
new cases has increased dramatically.[1] Oncology has thus
become a leading therapeutic area for pharmaceutical re-
search with ca. 650 drugs undergoing clinical trials
(www.pharma.org). However, finding new drugs is a com-
plex, expensive, and very time-consuming task. Indeed,
bringing a new drug to the market can take up to 15 years
and cost around E620 million.[1] Lately, rational drug-design
strategies, such as Quantitative Structure–Activity Relation-
ships (QSAR) modeling methods, have emerged as a prom-
ising alternative or complementary tool towards the effec-
tive screening of potential drugs, being thus increasingly
attracting the attention of medicinal chemists and of the
pharmaceutical industry. QSAR modeling may be better re-
garded as an exercise to filter drug candidates, before they
are subjected to more intensive calculations such as dock-
REQUIMTE, Chemistry Department, University of Porto
Rua do Campo Alegre 687, 4169-007 Porto, Portugal
fax: +351 220402659
*e-mail: ncordeir@fc.up.pt
[b] J. E. Rodrꢀguez-Borges
CIQ, Chemistry Department, University of Porto
Rua do Campo Alegre 687, 4169-007 Porto, Portugal
[c] A. M. Helguera , M. P. Gonzꢁlez
CBQ, Department of Chemistry, Central University of Las Villas
Santa Clara, 54830, Villa Clara, Cuba
[d] A. M. Helguera
Department of Chemistry, Central University of Las Villas
Santa Clara, 54830, Villa Clara, Cuba
[e] O. CaamaÇo, X. Garcꢀa-Mera
Department of Organic Chemistry, Faculty of Pharmacy,
University of Santiago de Compostela
15706 Santiago de Compostela, Spain
fax: +34 981594912
*e-mail: xerardo.garcia@usc.es
Supporting Information for this article is available on the WWW
under http://dx.doi.org/10.1002/minf.200900033
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
213

Full Paper
A. M. Helguera et al.
ing or an experimental measurement of activity (in vitro)
and under real conditions (in vivo) lastly.
The 2D representation of molecules can encode impor-
tant information on adjacency, branching and relative dis-
tance among different functionalities in a numerical form
that determines a wide range of physicochemical and bio-
logical properties, but it does not take into account infor-
mation concerning conformational aspects, i.e. bond
lengths, bond angles and torsion angles.[34] Besides, recog-
nition of the importance of the three-dimensional (3D)
structure and stereochemistry of molecules to their biologi-
cal activity, increasing knowledge of the 3D structure of
biological macromolecules such as proteins, and awareness
of the limitations of classical 2D approaches, led to many
attempts to generate 3D descriptors either for comple-
menting 2D-QSAR models or for standalone 3D-QSAR
models. In this work, we systematically examined the use
of 3D descriptors along with linear discriminant analysis for
probing the anticancer activity of NAs. This strategy was
found to produce a final discriminant QSAR model that ex-
hibits very good cross-validation statistics and additionally,
perform well on an external test set comprising other NA
chemicals designed by us and with unknown activity. The
design, synthesis, and biological evaluation of this new
series of carbonucleosides are reported.
The nucleoside analogues (NAs) were among the first
chemotherapeutic agents introduced for the medical treat-
ment of cancer.[2] One of kind of NAs that have been syn-
thesized for fighting cancer are carbocyclic nucleosides,[3] in
which the methylene group replaces the oxygen atom of a
furanose ring. As such, the glycosidic bond is resistant to
nucleoside phosphorylases and hydrolases, making the car-
bocyclic nucleosides more stable towards metabolic degra-
dation.[4] Interestingly, in some cases, the substitution of
the sugar ring by a carbocyclic ring does not affect the
enzyme recognition (especially kinases as target en-
zymes).[5] Due to these features, carbocyclic nucleosides
have received much attention as potential chemotherapeu-
tic agents,[2] such as abacavir and entecavir (Figure 1).
Figure 1. Important carbocycle nucleosides.
2 Materials and Methods
2.1 QSAR Modeling
For years our research group has been engaged in the
design, synthesis and evaluation of nucleoside analogues in
an effort to develop potential anticancer or antiviral
agents[4,5]. We recognized an excellent opportunity when
we could assemble a set of over 200 NA compounds, previ-
ously synthesized in one of our laboratories[3,6–9] or else-
where[10–31], all measured in a single, consistent cytotoxic
assay against murine leukemia L1210/0 cells[32]. With the
desire to build statistical sound and predictive models from
such data, we have recently presented a detailed QSAR
analysis covering the most important two-dimensional (2D)
structural features that rule the anticancer activity within
NAs.[33] The 2D-structural information then gathered and
the QSAR model per se can well aid to discriminate active/
inactive NAs. That was confirmed by the classification re-
sults obtained with an external test set comprising 20 di-
verse carbonucleoside analogues. Indeed, for that external
set, the percentage of overall discrimination attained was
quite impressive (75%), given the diversity of the test set
and the complexity of the biologic response being mod-
eled. Finally, a clustering search analysis, to identify the sim-
ilarities to natural nucleosides of the well-predicted active
NAs, and a structural interpretation of the results, taking
into account the mechanism of action responsible for their
cytotoxic activity, has been carried out. The information
provided by this analysis showed us that, in particular, the
indan derivatives are closely related to adenosine and gua-
nosine.[33]
2.1.1 Data Set
All the compounds used here are primarily nucleoside ana-
logues, derived from purinic and pyrimidinic bases, and
were experimentally assayed for their inhibitory effects
(IC50) in the proliferation of L1210/0 cancer cells. These ex-
periments have been conducted at the Rega Institute for
Medical Research of the Katholieke Universiteit Leuven in
Leuven, Belgium, following the same in vitro assay proto-
col.[26] One can rely on the quality of such biological data,
which has been measured by a single protocol, at the same
laboratory, by even the very same staff.
The compounds were firstly clustered into two groups
according to their IC50 values. The first group – actives – in-
cludes all chemicals with IC50 <200 mM, while the second
one – inactives – includes those with IC50 ꢀ200 mM. This
classification criterion was adopted not only because over
that concentration chemicals can be too toxic and there-
fore lack biological value but also to get a reasonable ratio
of active/inactive chemicals in the dataset. We have also
discarded all chemicals with disconnected structures like
salts and polymers. Finally we managed to assemble a
large, balanced dataset of 300 chemicals comprising 107
actives and 193 inactives.
A necessary but delicate task in any QSAR modeling is
predictive validation, i.e. to assess model adequacy for new
compounds. In general, the most reliably way to predictive-
ly validate a model is by external validation, which consists
214
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
of making predictions for an independent set of data not
used in the model setup. Here we select a small subset (22
compounds) of the NAs after a design process to act as an
external test set. These compounds have been synthesized
in one of our laboratories following the procedures de-
scribed above or reported previously by us,[35–37] and four
of them have already been assayed experimentally in
L1210 cells and their activity reported,[36] while the others
were evaluated here for the first time. A complete list of
the training set (278 compounds) along with the reported
experimental cytotoxicity (IC50 values expressed in mM) is
given as Supporting Information.
derived by joining the MIM information with the geometric
interatomic distances in the molecule.
3D-molecular descriptors were computed after fully opti-
mizing the geometry of each molecule by the semi-empiri-
cal quantum-mechanics method, Austin Model 1 (AM1), im-
plemented in MOPAC program[38b]. By disregarding descrip-
tors with constant or near constant values inside each
class, a final subset of 671 descriptors was then used for
building the QSAR models.
2.1.3 Statistical Methods
Linear discriminant analysis (LDA), as implemented in the
STATISTICA software (version 7.0),[38c] was applied here to
find classification models (Eqs. 1–2) that best describe the
cytotoxic activity P, as a linear combination of the predictor
X-variables (3D descriptors).
In developing the models, P values of +1 and ꢁ1 were
assigned to active and inactive compounds, respectively,
but a posteriori probabilities are used instead to assert the
models’ classification of compounds. In particular, when the
probability of being active did not differ more than 5%
from that of being inactive, the case was considered as un-
classified (U) by the model.
2.1.2 Molecular Descriptors
Our models are based on six different blocks of 3D-descrip-
tors that are available in the DRAGON software package[38a]
,
with a successful history in structure–activity and structure–
property correlation. They include:
´
(i) Randic molecular profiles
A set of 41 descriptors derived from the distance distri-
bution moments of the geometry matrix, including both
molecular as shape profiles.
(ii) Geometrical descriptors
The forward stepwise (FS) technique was applied to
select the molecular descriptors (X-variables) with the high-
est influence on the anticancer activity. This technique
begins by including the variable which yields the best
linear fit in terms of explaining the response. The next vari-
able is included as that variable which most significantly
improves the existing model. Once this new model is deter-
mined, the variables included are tested to see if the model
can be improved by dropping them from the model. If the
model can be improved, the variable is removed and the
stepwise procedure is repeated until no further variables
are either included or removed.
As the variables included in the first developed model
(Eq.1) were strongly interrelated to each other, this may
well lead to a multicollinearity problem, and can cause
problems in interpreting the individual estimated coeffi-
cients. One very useful and informative approach of avoid-
A set of 74 descriptors consisting of different kinds of
conformation-dependent descriptors based on molecular
geometry.
(iii) Radial Distribution Function (RDF) descriptors
This set consists of 135 descriptors calculated from the
radial distribution function, that is to say, the probability
distribution to find an atom in a spherical volume centered
in each atom.
(iv) Molecule Representation of Structures based on Elec-
tron diffraction (3D-MoRSE) descriptors
The 3D-MoRSE approach considers the molecular infor-
mation derived from an equation used in electron diffrac-
tion studies. The 160 3D-MoRSE descriptors are calculated
by summing atomic weights viewed by different angular
scattering functions.
(v) Weighted Holistic Invariant Molecular (WHIM) descrip-
tors
ing multicollinearity is the orthogonalization procedure
[39]
´
These indices are calculated by a principal component
analysis on the centered Cartesian coordinates of the
atoms by using a weighted covariance matrix. Two kinds of
WHIM descriptors have been defined: 66 directional and
33 nondirectional descriptors.
suggested by Randic some year’s ago. In such a proce-
dure, after choosing a starting descriptor, subsequent de-
scriptors are added only as their orthogonal complements
to the descriptors already present. This has the advantages
that the equation coefficients are stable (i.e., they do not
change as new descriptors are added), and the new infor-
mation supplied by each additional descriptor is clearly dis-
tinguished in the final equation statistics. Here, to deal with
(vi) GEometry, Topology, and Atoms-Weighted AssemblY
(GETAWAY) descriptors
The GETAWAY descriptors are based on a leverage
matrix, similar to the one defined in statistics and em-
ployed for regression diagnostics, called molecular influ-
ence matrix (MIM). These descriptors are divided into two
different sets: the first set consists of 104 H-GETAWAY de-
scriptors, derived by using only the information provided
by the MIM, and the remaining 93 R-GETAWAY descriptors,
´
the multicollinearity problem, we resorted to Randic’s pro-
cedure and orthogonalized the variables following the
order selected by the FS scheme (Mor12m, ITH, ADDD,
MEcc, Mor04e, E3s, Mor19u, J3D, Au, and RDF120v). The re-
sulting orthogonal-descriptor model was standardized af-
terwards.
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
215

Full Paper
3 Experimental
A. M. Helguera et al.
J=7.86, 10.55 Hz, 4-HH), 1.87 (ddt, 1H, J=7.18, 8.10, 10.85 Hz, 1-H),
2.36–2.45 (m, 1H, 3-H), 3.28 (dd, 1H, J=5.89, 10.85 Hz, CHHOH),
3.39 (ddd, 1H, J=5.38, 8.21, 10.85 Hz, CHHOH), 4.12 (dd, 1H, J=
8.75, 13.81 Hz, CHHN), 4.23 (t, 1H, J=5.02 Hz, D2O exch., OH), 4.31
(dd, 1H, J=7.03, 13.81 Hz, CHHN), 8.69(s, 1H, 2-Hpurine), 8.75 (s, 1H,
8-Hpurine). 13C NMR (DMSO-d6): d=16.50 (CH3), 24.61 (CH3), 30.76
(CH2), 39.22 (C), 41.16 (CH2), 43.85 (CH), 44.85 (CH), 61.64 (CH2),
131.19 (CH), 147.72 (C), 149.32 (CH), 151.72 (C), 152.18 (C). Anal.
calcd. for C13H17ClN4O: C, 55.62; H, 6.10; N, 19.96. Found: C, 55.31;
H, 6.25; N, 19.83.
3.1 Chemistry
3.1.1 General
Silica gel was purchased from Merck. All other chemicals used
were of reagent grade and were obtained from Aldrich Chemical
Co. The reactions were monitored by thin layer chromatography
conducted on E. Merck TLC plates (silica gel 60 F-254, aluminum
back) and visualized with UV light or iodine or anisaldehyde solu-
tion. The products were purified by column chromatography on
silica gel (Merck 60, 230–240 mesh). Melting points were deter-
mined on a Reichert Kofler Thermopan or in capillary tubes on a
Bꢄchi 510 apparatus and are uncorrected. Infrared spectra were re-
corded on a Perkin-Elmer 1640-FT spectrophotometer and the
(+)-(1S,cis)-6,9-Dihydro-9-[3-(hydroxymethyl)-2,2-dimethylcyclo-
butylmethyl]-1H-purin-6-ona (4a). A mixture of 3a (0.40 g ) and
0.25N NaOH (17 mL) was refluxed for 6 h., the solvents under re-
duced pressure left a solid residue that was chromatographed
(silica gel AcOEt/MeOH 7:3) afforded 4a as a solid (0.34 g, 89%).
main bands are given in cmꢁ1 1H NMR (300 MHz) and 13C NMR
.
25
Mp 198–2008C (MeOH). [a]D +98.33 (c 0.3, MeOH). IR (KBr): v=
1
3445, 2958, 1648, 1586, 1548, 1521, 1457, 1364, 1021 cmꢁ1. H NMR
spectra (75.47 MHz) were recorded on a Bruker WM AMX spec-
trometer using TMS (tetramethylsilane) as an internal standard
(chemical shifts (d) in parts per million, J in hertz). Mass spectra
were performed on a Hewlett–Packard HP5988A mass spectrome-
ter by electron impact (EI), or on a Finnigan Trace-MS mass spec-
trometer by chemical ionization (CI). Optical rotations at the
sodium D-line were determined using a Perkin-Elmer 241 thermo-
stated polarimeter. Elemental analyses were obtained on a Perkin-
Elmer 240B microanalyser by the Microanalysis Service of the Uni-
versity of Santiago de Compostela. GLC analyses were carried out
on a Hewlett-Packard 5890 II apparatus provided with a flame ioni-
zation detector, using a semicapillary column (5 mꢅ0.53 mm i.d.,
film thickness 2.65 mm) and helium as carrier gas. The purity of the
compounds used on the biological tests was at least 95% and was
determined by combustion analysis. (1S,2R)-(cis)-1-Amino-2-inda-
nol, 99% was purchased on Aldrich. Inc.
(DMSO-d6): d=0.94 (s, 3H, CH3), 1.01 (s, 3H, CH3), 1.34 (c, 1H, J=
10.33 Hz, 4-HH), 1.72 (dt, 1H, J=7.86, 10.55 Hz, 4-HH), 1.89 (ddt,
1H, J=6.22, 8.03, 10.14 Hz, 3-H), 2.44 (tt, 1H, J=7.85, 8.98 Hz, 1-H),
3.25 (dd, 1H, J=6.45, 10.97 Hz, CHHOH), 3.37 (dd, 1H, J=8.10,
10.97 Hz, CHHOH), 3.96 (dd, 1H, J=8.60, 13.73 Hz, CHHN), 4.15 (dd,
1H, J=7.19, 13.73 Hz, CHHN), 4.23 (s, 1H, D2O exch., OH), 8.01 (s,
1H, 2-Hpurine), 8.05 (1s, s, 1H, 8-Hpurine), 12.29 (s, 1H, D2O exch., NH).
13C NMR (DMSO-d6): d=16.50 (CH3), 24.62 (CH3), 30.83 (CH2), 39.19
(C), 41.57 (CH2), 42.86 (CH), 43.31 (CH), 61.68 (CH2), 124.25 (CH),
140.55 (C), 145.73 (CH), 148.64 (C), 157.07 (C). Anal. calcd. for
C13H18N4O2: C, 59.53; H, 6.92; N, 21.36. Found: C, 59.31; H, 6.65; N,
21.33.
(+)-(1S,cis)-6,7-Dihydro-3-[3-(hydroxymethyl)-2,2-dimethylcyclo-
butyl]-3H-1,2,3-triazo-lo[4,5-d]pyrimidin-7-one (5a). A solution of
NaNO2 (0.12 g, 1.61 mmol) in water (3 mL) was added dropwise to
a mixture of 2a (0.35 g,1.30 mmol), and 1N HCl (7 mL) at approx.
ꢁ58C, the mixture was stirred at room temperature for 12 h, and
removal of the solvents left a solid (0.32 g) that was chromato-
graphed (silica gel AcOEt) afforded 5a as a white solid (0.31 mg,
(+)-(1R,cis)-3-[(5-Amino-6-chloropyrimidin-4-ylaminomethyl)-2,2-
dimethylcyclobutyl]methanol (2a). A mixture of freshly prepared
amino alcohol 1a[42] (3.0 g, 20.98 mmol), and 5-amino-4,6-dichloro-
pyrimidine (4.0 g, 24.4 mmol) in Et3N (13 mL) and n-BuOH (83 mL)
was refluxed under argon for 72 h, the solvents were removed
under reduced pressure, and the resulting oil was chromatograph-
ed (silica gel CH2Cl2/MeOH, 25 :1). Removal of the solvent left 2a
25
93%). Mp 184–1868C (Cyclohexane/AcOEt/MeOH). [a]D +7.76 (c
0.27, MeOH). IR (KBr): v=3308, 3054, 1717, 1653, 1636, 1591, 1457,
1384, 1010 cmꢁ1. H NMR (DMSO-d6): d=0.90 (s, 3H, CH3), 1.05 (s,
1
25
(5.11 g, 90%) as a white solid. Mp 181–1838C (Et2O/EtOH). [a]D
3H, CH3), 1.39 (c, 1H, J=10.39 Hz, 4-HH), 1.80 (dt, 1H, J=7.84,
10.44 Hz, 4-HH), 1.93 (ddt, 1H, J=6.20, 8.18, 9.92 Hz, 3-H), 2.42 (tt,
1H, J=7.76, 9.29 Hz, 1-H), 3.16 (dd, 1H, J=6.20, 11.17 Hz, CHHOH),
3.38 (dd, 1H, J=8.18, 11.17 Hz, CHHOH), 4.03 (t, 1H, J=5.02 Hz,
D2O exch., OH), 4.23 (dd, 1H, J=7.50, 13.99 Hz, CHHN), 4.52 (dd,
1H, J=8.02, 13.99 Hz, CHHN), 7.88 (s, 1H, 2Hpyrimidine), 12.57 (s, 1H,
D2O exch., NH). 13C NMR (DMSO d6): d=16.42(CH3), 24.64 (CH3),
30.68 (CH2), 41.41 (C), 42.30 (CH2), 43.93 (CH), 47.53 (CH), 61.64
(CH2), 129.78 (CH), 148.69 (C), 149.85 (C), 155.75 (C). Anal. calcd. for
C12H17N5O4: C, 54.74; H, 6.51; N, 26.60. Found: C, 54.71; H, 6.45; N,
26.44.
+13.02 (c 0.5, MeOH). IR (KBr): v=3196, 2951, 1654, 1577, 1458,
1
1423, 1225, 920 cmꢁ1. H NMR (DMSO-d6): d=0.96 (3H, s, CH3), 1.04
(s, 3H, CH3), 1.18 (c, 1H, J=10.03 Hz, 4-HH), 1.88–1.92 (m, 2H, 4-
HH+3-H), 2.09 (dt, 1H, J=7.52, 9.93 Hz, 1-H), 3.22 (dd, 1H, J=5.86,
7.47 Hz, CHHN), 3.24–3.30 (m, 1H, CHHN), 3.33 (t, 1H, J=7.62 Hz,
CHHOH), 3.37 (dd, 1H, J=5.29, 7.62 Hz, CHHOH), 4.74 (t, 1H, J=
5.00 Hz, D2O exch., NH), 7.70 (s, 1H, 2Hpyrimidin). 13C NMR (DMSO-d6)
d=16.46 (CH3), 25.10 (CH3), 31.43 (CH2), 39.06 (C), 40.85 (CH2),
42.26 (CH), 44.15 (CH), 61.95 (CH2), 123.77 (CH), 136.85 (C), 145.97
(C), 152.26 (C). Anal. calcd. for C12H19ClN4O: C, 53.23; H, 7.07; N,
20.69. Found: C, 53.31; H, 7.05; N, 20.58.
(+)-(1R,cis)-3-(7-Amino-3H-1,2,3-triazolo[4,5-d]pyrimidin-3-ylme-
(+)-(1R,cis)-3-(6-Chloro-9H-purin-9-yl)metil-2,2-dimethylcyclobu-
tylmethanol (3a). A mixture of 2a (1.50 g, 5.54 mmol), triethyl or-
thoformate (30 mL), 12N HCl (1.4 mL) was stirred at room tempera-
ture for 24 h and then concentrated under vacuum obtain a resi-
due that was treated with 0.5N HCl (30 mL) and THF (15 mL) for
2 h at room temperature. The resulting solution was brought to
pH 7 with 1N NaOH, and evaporation of the solvents under re-
duced pressure left a residue that was chromatographed (silica gel
CH2Cl2/MeOH, 20 :1) afforded 3a as an yellow oil (0.72 g, 47%).
til)-2,2-dimethylcyclobutylmetanol (6a). To a solution of 2a
(0.35 g, 1.30 mmol) in 1N HCl (3.5 mL in a salted ice bath (approx
ꢁ58C), a solution of NaNO2 (0.12 g, 1.61 mmol) in water (13 mL)
was added slowly enough to prevent the temperature from rising
above 08C. The mixture was stirred in the salted ice bath for
30 min, treated with 14N NH4OH (7 mL), and refluxed for 1 h. After
removal of water by azeotropic distillation with toluene and EtOH,
the resulting solid residue (0.36 g) was chromatographed (silica gel
CH2Cl2/MeOH, 8 :1 and 7:1). Removal of the solvents from the
latter eluate under reduced pressure left a solid residue (0.30 g)
that upon recrystallization from Hex/AcOEt/MeOH afforded 6a
25
[a]D +22.24 (c 0.24, MeOH). IR (film): v=3343, 2959, 1596, 1558,
1506, 1400, 1211, 648 cmꢁ1 1H NMR (DMSO-d6): d=0.94 (s, 3H,
.
25
CH3), 1.03 (s, 3H, CH3), 1.38 (c, 1H, J=10.32 Hz, 4-HH), 1.73 (dt, 1H,
(0.22 g, 65%) as a white solid. Mp 194–1968C. [a]D +6.20 (c 0.25,
216
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
MeOH). IR (KBr): v=3745, 3150, 2950, 1665, 1607, 1576, 1510, 1457,
content 630 mg, 4.46 mmol) was added slowly enough to cause
no rise in temperature, following which the reaction mixture was
stirred overnight at r.t.. Toluene and EtOH were then added to
form a low-boiling ternary azeotrope that was evaporated under
reduced pressure while the temperature was maintained below
408C, affording 11c (524 mg, 56%) as a white solid. Mp 170–28C.
1384, 730 cmꢁ1 1H NMR (DMSO-d6): d=0.87 (s, 3H, CH3), 1.06 (s,
.
3H, CH3), 1.39 (c, 1H, J=10.33 Hz, 4-HH), 1.79 (dt, 1H, J=7.83,
10.50 Hz, 4-HH), 1.92 (ddt, 1H, J=6.27, 7.96, 11.06 Hz, 1-H), 2.45 (tt,
1H, J=7.81, 10.24 Hz, 3-H), 3.27 (dd, 1H, J=6.27, 10.86 Hz,
CHHOH), 3.38 (dd, 1H, J=8.10, 10.86 Hz, CHHOH), 4.22 (t, 1H, J=
5.05 Hz, D2O, exch., OH), 4.33 (dd, 1H, J=7.41, 13.98 Hz, CHHN),
4.53 (dd, 1H, J=8.14, 13.98 Hz, CHHN), 8.03 (s, 1H, D2O exch.,
NHH), 8.28 (s, 1H, 2Hpyrimidine), 8.33 (s, 1H, D2O exch., NHH). 13C NMR
(DMSO-d6): d=16.42 (CH3), 24.65 (CH3), 30.67 (CH2), 41.30 (C), 42.30
(CH2), 43.94 (CH), 47.12(CH), 61.67 (CH2), 124.06 (CH), 148.93 (C),
156.55 (C), 156.92 (C). Anal. calcd. for C12H18N6O: C,54.95; H, 6.92;
N, 32.04. Found: C, 54.83; H, 6.75; N, 32.23.
IR (KBr): n=3293, 1701, 1666, 1626, 1542, 1474, 1335, 1256, 1205,
1
1187, , 965, 794, 761 cmꢁ1. H NMR (CDCl3): d=1.21–1.27 (t, 3H, =
t
7.05 Hz, CH3), 1.46–1.57 (dt, 1H, J=7.76, 12.39 Hz, 2b-H), 2.40–2.51
(dt, 1H, J=7.14, 12.39 Hz, 2a-H), 3.12–3.18 (t, 1H, J=6.32 Hz, 1b-H),
3.35–3.42 (part A of an ABM system, JAB =12.64 Hz, JAM =6.91 Hz,
CHHNH), 3.62–3.69 (part B of an ABM system, JBA =12.64 Hz, JBM
=
6.99 Hz, CHHNH), 3.90–3.98 (q, 2H, J=7.01 Hz, CH3CH2O), 4.94–4.99
(m, 1H, 3b-H), 5.32–5.35 (d, 1H, J=5.75 Hz, D2O exch., OH), 5.48–
5.53 (d, 1H, J=12.30 Hz, COCH), 7.21–7.35 (m, 4H, ArH), 7.53–7.58
(d, 1H, J=12.30 Hz, CHOEt), 8.65 (s, 1H, D2O exch., CH2NHCO),
10.07 (s, 1H, D2O exch., CONHCO). 13C NMR (CDCl3): d=16.2 (CH3),
42.1 (CH2), 43.1 (CH), 44.4 (CH2), 69.1 (CH2), 74.5 (CH), 100.1 (CH),
125.1 (CH), 125.9 (CH), 128.6 (CH), 129.2 (CH), 144.7 (C), 148.7 (C),
155.7 (C), 163.9 (CH), 169.5 (C). Anal. calcd. for C16H20N2O4: C, 63.14;
H, 6.62; N, 9.20. Found: C, 63.31; H, 6.45; N, 9.33.
(1S,cis)-N-[3-(Hydroxymethyl)-2,2-dimethylcyclobutyl]-N’’-(3-
ethoxypropanoyl)urea (8b). Dry benzene (50 mL) was added in
the dark, under argon, to silver cyanate (7.5 g) that had previously
been dried in vacuo over P2O5 at 1008C. The resulting suspension
was refluxed with vigorous stirring for 30 min, after which a solu-
tion of 3-ethoxypropenoyl chloride (3.0 g, 25 mmol) in dry benzene
(10 mL) was added dropwise. The resulting mixture was refluxed
with vigorous stirring for a further 30 min at room temperature for
3 h, and then left to settle. A sample of the supernatant (2.5 mL,
theoretically containing 1.03 mmol of 3-ethoxypropenoyl isocya-
nate) was transferred to a dropping funnel and added dropwise
under argon to a solution of 7b[44] (0.10 g, 0.77 mmol) in dry DMF
(4 mL) at ꢁ158C. The resulting mixture was left for 1 h to reach
room temperature and was then stirred overnight and concentrat-
ed under reduced pressure (oil pump) at a temperature below
408C. Removal of the solvents by repeated co-evaporation with
ethanol left a solid that was chromatographed (silica gel CHCl3/
EtOH, 95 :5) afforded compound 8b (0.12 g, 61%) a white solid. An
analytic sample was obtained by crystallization from cyclohexane/
EtOAc. Mp 139–1408C. IR (KBr): v=3281, 2985, 1673, 1547, 1234,
1-(1-Inden-3-ylmethyl)-1,2,3,4-tetrahydropyrimidine-2,4-dione
(12c). A solution of 11c (75 mg, 0.246 mmol) and 2N H2SO4 (5 mL)
in 1,4-dioxane (2 mL) was refluxed 30 min. The mixture was al-
lowed to cool and the solvent was evaporated under reduced
pressure, and the resulting solid residue was chromatographed
(silica gel, CH2Cl2/MeOH, 95 :5) affording 12c (47 mg, 79%) as a
white solid. Mp 175–1808C. IR (KBr): n=3019, 1718, 1559, 1458,
1
1423, 1380, 1340, 1242, 1202, 830, 745, 716 cmꢁ1. H NMR (CDCl3):
d=3.37 (d, 2H, J=1.30 Hz, 1’-H), 4.86 (d, 2H, J=1.40 Hz, CH2N),
5.62 (d, 1H, J=7.93 Hz, 5-H), 6.41 (s, 1H, 2‘-H), 7.14 (d, 1H, J=
7.93 Hz, 6-H), 7.16–7.32 (m, 3H, ArH), 7.43 (d, 1H, J=6.66 Hz, ArH),
9.05 (bs, 1H, D2O exch., NH). 13C NMR (CDCl3): d=37.8 (CH2),
44.5(CH2), 101.7 (CH), 119.4 (CH), 124.3 (CH), 125.4 (CH), 126.5 (CH),
131.8 (CH), 139.8 (C), 142.9 (C), 144.4 (C), 145.5 (CH), 151.3 (C),
164.0 (C). Anal. calcd. for C14H12N2O2: C, 69.99; H, 5.03; N, 11.66.
Found: C, 69.82; H, 5.17; N, 11.80.
1
1176, 1120 cmꢁ1. H NMR (CDCl3): d=1.02. (s, 3H, CH3), 1.21 (s, 3H,
CH3), 1.35 (t, 3H, J=7.05 Hz, CH3CH2), 1.48–1.57 (m, 1H, 4-HH),
1.94–2.01 (m, 1H, 4-HH), 2.33 (dt, 1H, J=7.82, 10.88 Hz, 3-H), 3.47–
3.63 (m, 3H, CH2OH+OH), 3.90–3.99 (m, 3H, CH2CH3 +1-H), 5.31
(d, 1H, J=12.22 Hz, CH-OEt), 7.65 (d, 1H, J=12.82 Hz, CHCO), 8.72
(d, 1H, J=7.75 Hz, NHCH), 9.17 (s, 1H, CONHCO). 13C NMR (CDCl3):
d=15.15 (CH3), 16.30 (CH3), 16.82 (CH3), 36.61 (CH2), 43.24 (CH),
49.30 (C), 64.72 (CH2), 76.73 (CH), 98.75 (CH), 153.82 (CH), 162.33
(C), 170.18 (C). Anal. calcd. for C13H22N2O4: C, 57.76; H, 8.20; N,
10.36. Found: C, 57.65; H, 8.35; N, 10.63.
(1S,2R)-cis-N-(2-Hydroxy-1-indanyl)-N’-(3-ethoxypropenoyl)urea
(14d). To a solution of aminoalcohol 13d (500 mg, 3.35 mmol) in
DMF (22 mL) under Ar at ꢁ258C, a solution of 3-ethoxy-2-propeno-
yl isocyanate 1.15 M in benzene (3.91 mL, theoretical isocyanate
content 630 mg, 4.46 mmol) was added slowly enough to cause
no rise in temperature, following which the reaction mixture was
stirred overnight at r.t.. Toluene and EtOH were then added to
form a low-boiling ternary azeotrope that was evaporated under
reduced pressure while the temperature was maintained below
(1S,cis)-1[3-(Hydroxymethyl)-2,2-dimethylcyclobuthyl]-1,2,3,4-
tetrahydropyrimidine-2,4-diona (9b). To a solution of 8b (0.18 g,
0.7 mmol) in dioxane (11 mL) was added 2N H2SO4 (14 mL), and
this mixture was refluxed for 30 min, allowed to cool, brought to
pH 7 with 2N NaOH, and concentrated to dryness. The residue was
extracted with ethanol (3ꢅ30 mL), and concentration of the ex-
tracts left a residue that was chromatographed (silica gel EtOAc/
MeOH, 5 :1) afforded 9b (0.15 g, 83%). Mp 135–1388C.1H NMR
(CDCl3): d=0.75. (s, 3H, CH3), 1.16 (s, 3H, CH3), 1.60–1.99 (m, 1H, 4-
HH), 2.03–2.09 (m, 2H, 4-HH +3-H), 3.34–3.74 (m, 2H, CH2OH), 4.28
(t, 1H, J=9.20 Hz, 1-H), 5.45 (d, 1H, J=7.90 Hz, 4-Hpyrimidine), 7.64 (d,
1H, J=7.95 Hz, 5-Hpyrimidine). 13C NMR (CDCl3): d =15.58(CH3), 24.20
(CH3), 30.09 (CH2), 41.05 (CH), 44.25 (C), 56.78 (CH), 61.51 (CH2),
100.26 (CH), 143.19 (CH), 152.41 (C), 164.42 (C). Anal. calcd. for
C11H16N2O3: C, 58.91; H, 7.19; N, 12.49. Found: C, 58.99; H, 7.17; N,
12.35.
24
408C, affording 14d (576 mg, 59%) as a white solid. [a]D +24.18
(c 0.5, DMSO). Mp 180–28C. IR (KBr): n=3648, 3421, 1697, 1675,
1611, 1559, 1540, 1458, 1177, 1128, 816, 744 cmꢁ1 1H NMR
.
(DMSOd6): d=1.21–1.27 (t, 3H, J=7.05 Hz, CH3), 1.54–1.65 (dt, 1H,
J=7.58, 12.52 Hz, 2b-H), 2.78–2.89 (dt, 1H, J=7.18, 12.52 Hz, 2a-H),
3.87–3.95 (q, 2H, J=7.05 Hz, CH3CH2O), 5.41–5.44 (d, 1H, J=
6.35 Hz, D2O exch., OH), 5.48–5.53 (d, 1H, J=12.13 Hz, COCH),
4.89–4.97 (q, 1H, J=6.45 Hz, 2b-H), 5.03–5.13 (q, 1H, J=8.12 Hz,
1b-H), 7.16–7.36 (m, 4H, ArH), 7.47–7.52 (d, 1H, J=12.13 Hz,
CHOEt), 8.84–8.87 (d, 1H, J=8.12 Hz, D2O exch., CHNHCO), 10.15 (s,
1H, D2O exch., CONHCO). 13C NMR (DMSOd6): d=14.7 (CH3), 40.1
(CH2), 57.6 (CH), 67.6 (CH2), 72.1 (CH), 98.8 (CH), 124.2 (CH), 125.0
(CH), 126.7 (CH), 127.6(CH), 142.52 (C), 145.9 (C), 153.7 (C), 162.3
(CH), 167.9 (C). Anal. calcd. for C15H18N2O4: C, 62.06; H, 6.25; N,
9.65. Found: C, 61.97; H, 6.39; N, 9.81.
(ꢂ)-cis-N-(2-Hydroxy-1-indanyl)-N’-(3-ethoxypropenoyl)urea
(11c). To a solution of aminoalcohol 10c[45] (500 mg, 3.06 mmol) in
DMF (22 mL) under Ar at ꢁ258C, a solution of 3-Ethoxy-2-propeno-
yl Isocyanate 1.15 M in benzene (3.91 mL, theoretical isocyanate
(1S,2R)-cis-1-(2-Hydroxy-1-indanyl)-1,2,3,4-tetrahydropyrimidine-
2,4-dione (15d). A solution of 14d (36 mg; 0.124 mmoles) and 2N
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
217

Full Paper
H2SO4 (8 mL) in 1,4-dioxane (4 mL) was refluxed 30 min. The mix-
ture was allowed to cool, neutralized with NaOH 2N and the sol-
vent was evaporated under reduced pressure. The resulting solid
residue was chromatographed (silica gel, CH2Cl2/MeOH, 30 :1) af-
A. M. Helguera et al.
(d, 2H, J=7.33 Hz), 7.83–7.85 (m, 2H). 13C NMR (CDCl3): d=29.8
(CH2), 41.7 (CH), 45.4 (CH), 46.7 (CH2), 61.2 (CH2), 123.2 (C), 128.0
(CH), 128.1 (CH), 128.2 (CH), 128.6 (CH), 129.0 (CH), 136.0 (C), 136.9
(C), 142.5 (CH), 142.9 (C), 143.5 (C), 149.1 (C), 153.7 (C), 157.0 (C),
157.3 (C), 159.4 (C). MS (EI, 70 eV): m/z (%)=486 (M+2, 21), 485 (M
+1, 18), 484 (M, 69), 413 (6), 316 (35), 315 (100), 289 (6), 288 (31),
285 (13), 283 (10). Anal. calcd. for C26H22ClN7O: C, 64.53; H, 4.58; N,
20.26. Found: C, 64.92; H, 6.36; N, 16.87.
24
fording 15d (27 mg, 89%) as a white solid. [a]D ꢁ100.26 (c 1.15,
CHCl3). Mp 99–1028C. IR (KBr): n=3402, 1688, 1463, 1387, 1261,
1055, 813, 747, 563 cmꢁ1. H NMR (CDCl3): d=2.87–2.94 (d, 1H, J=
1
16.95 Hz, 3‘-H), 3.08–3.17 (dd, 1H, J=5.62, 16.95 Hz, 3‘-H), 4.65 (s,
1H, D2O exch., OH), 4.93 (s, 1H, 2’-H), 5.22–5.25 (d, 1H, J=7.99 Hz,
5-H), 5.95–5.97 (d, 1H, J=5.02 Hz, 1’-H), 6.90–6.93 (d, 1H, J=
7.99 Hz, 6-H), 7.08–7.19 (m, 4H, ArH), 10.77 (bs, 1H, D2O exch., NH).
13C NMR (CDCl3): d=39.2 (CH2), 63.6 (CH), 71.7 (CH), 100.2 (CH),
124.3 (CH), 125.8 (CH), 127.3 (CH), 128.9 (CH), 136.4 (C), 141.9 (C),
144.8 (CH), 152.1 (C), 165.4 (C). Anal. calcd. for C13H12N2O3: C, 63.93;
H, 4.95; N, 11.47. Found: C, 64.11; H, 4.81; N, 11.39.
(ꢂ)-cis-7-(tert-Butyldimethylsilyloxymethyl)-5-(2-amino-6-cyclo-
pentylamine-9H-purin-9-ylmethyl)-1,4-diphenyl-6,7-dihydro-5H-
cyclopenta[d]pyridazine (20e).
A solution of 18e (103 mg,
0.17 mmol) and cyclopentylamine (0.10 mL, 1.04 mmol) in dry
EtOH (10 mL) was refluxed under Ar for 30 h. Removal of the sol-
vent under reduced pressure left a solid residue that was chroma-
tographed (silica gel, hexane/EtOAc, 1:3). Concentration of the
non-void fractions to dryness afforded 20e (95 mg, 86%) as a
white solid. Mp 113–1158C. IR (KBr): n=3325, 2952, 2858, 1596,
(ꢂ)-cis-[7-(tert-Butyldimethylsilyloxymethyl)-5-(2-amino-6-chloro-
9H-purin-9-ylmethyl)-1,4-diphenyl-6,7-dihydro-5H-cyclopenta[d]-
pyridazine (18e) and (ꢂ)-5-(tert-butyldimethyl-silyloxymethyl)-7-
methylidene-1,4-diphenyl-6,7-dihydro-5H-cyclopenta[d]pyrida-
1484, 1444, 1394, 1252, 1101, 1033, 922, 836, 772, 730, 698 cmꢁ1
.
1H NMR (CDCl3): d=ꢁ0.29 (s, 3H, Si(CH3)2), ꢁ0.21 (s, 3H, Si(CH3)2),
0.67 (s, 9H, C(CH3)3), 1.48–1.52 (m, 2H), 1.62–1.65 (m, 2H), 1.71–1.73
(m, 2H), 2.03–2.07 (m, 3H), 2.50 (dt, 1H, J=14.11, 9.98 Hz, 6-HH),
3.23 (dd, 1H, J=10.06, 5.33 Hz, 7-H), 3.47 (dd, 1H, J=10.06,
2.90 Hz, 5-H), 3.88–3.95 (m, 3H, HOCH2 +NCHH), 4.45–4.48 (m, 1H,
NCHH), 4.57 (s, 2H, D2O exch., NH2), 5.44 (d, 1H, J=7.69 Hz, D2O
exch, NH), 7.00 (s, 1H, 8-Hpurine), 7.43–7.53 (m, 6H), 7.79–7.82 (m,
2H), 7.90 (d, 2H, J=7.12 Hz). 13C NMR (CDCl3): d=ꢁ5.7 (CH3), ꢁ5.6
(CH3), 18.2 (C(CH3)3), 23.7 (CH2), 25.8 (C(CH3)3), 30.7 (CH2), 33.5 (CH2),
43.7 (CH), 46.1 (CH), 46.2 (CH2), 63.6 (CH2), 128.5 (CH), 128.6 (CH),
128.7 (CH), 129.3 (CH), 136.6 (C), 136.7 (CH), 136.9 (C), 142.8 (C),
142.9 (C), 154.8 (C), 157.6 (C), 157.8 (C), 159.8 (C). MS (EI, 70 eV):
m/z (%)=648 (M+2, 47), 647 (M+1, 100), 646 (M, 39), 430 (33),
429 (91), 339 (21), 319 (11), 309 (17), 308 (11), 307 (21), 297 ( 12),
289 (11), 287 (16), 283 (22), 279 (12), 219 (17), 213 (16), 199 (31),
197 (50). Anal. calcd. for C37H46N8OSi: C, 68.70; H, 7.17; N, 17.32.
Found: C, 69.06; H, 7.33; N, 17.49.
zine (17e).
A
solution of 2-amino-6-chloropurine (0.33 g,
1.97 mmol), 60% NaH (78.74 mg, 1.97 mmol) and 18-crown-6 ether
(0.30 g, 1.14 mmol) in dry DMF (30 mL) was stirred under argon at
558C for 1 h. A solution of 16e[43] (0.60 g, 1.14 mmol) in dry DMF
(25 mL) was added, and stirring at 558C was continued for a fur-
ther 24 h, after which the reaction mixture was diluted with CH2Cl2
(50 mL) and washed with H2O (4ꢅ60 mL). The organic phase was
dried over Na2SO4, and removal of the solvent left a solid residue
(0.40 g) that was chromatographed (silica gel, hexane/EtOAc, 1:1
and 1:2). The early fractions afforded 17e[47] (0.29 g, 59%) as a
white solid; the middle fractions gave unreacted 16e (20 mg); and
the late fractions provided 18e (66 mg, 10%). Mp 132–1338C. IR
(KBr): n=3323, 2927, 2854, 1613, 1561, 1517, 1462, 1382, 1254,
1102, 911, 836, 775, 699 cmꢁ1
.
1H NMR (CDCl3): d=ꢁ0.30 (s, 3H,
Si(CH3)2), ꢁ0.23 (s, 3H, Si(CH3)2), 0.65 (s, 9H, C(CH3)3), 1.96–2.05 (m,
1H, 6-HH), 2.53 (dt, 1H, J=14.0, 10.0 Hz, 6-HH), 3.28 (dd, J=10.13,
4.78 Hz, 1H, 7-H), 3.48 (dd, 1H, J=10.13, 2.60 Hz, 5-H), 3.89–4.04
(m, 3H, OCH2 +NCHH), 4.38–4.43 (m, 1H, NCHH), 5.12 (s, 2H, D2O
exch., NH2), 7.39–7.51 (m, 6H), 7.80–7.77 (m, 4H), 7.97 (s, 1H, 8-
(ꢂ)-[cis-7-(2-Amino-6-cyclopentylamine-9H-purin-9-ylmethyl)-
1,4-diphenyl-6,7-dihydro-5H-cyclopenta[d]pyridazin-5-yl]metha-
nol (21e): A 1 M solution of TBAF in THF (0.27 mL, 0.27 mmol) was
added dropwise to a solution of 20e (88 mg, 0.15 mmol) in the
same solvent (4 mL) that was stirring under Ar in an ice bath. The
solution was allowed to reach r.t., and stirring was continued for a
further 6 h. The solvent was removed under reduced pressure and
the residue was chromatographed (silica gel, CH2Cl2/MeOH, 97:3
and 95 :5). Concentration of the nonvoid fractions to dryness af-
forded 21e (68 mg, 94%) as a white solid. Mp 145–1468C. IR (KBr):
n=3331, 2948, 2865, 1599, 1486, 1446, 1396, 1347, 1244, 1077,
H
purine). 13C NMR (CDCl3): d=ꢁ5.7 (SiCH3), - 5.6 (SiCH3), 18.1
[C(CH3)3], 25.7 [C(CH3)3], 30.9 (CH2), 43.3 (CH), 46.1 (CH), 46.9 (CH2),
63.4 (CH2), 125.1 (CH), 128.4 (CH), 128.5 (CH), 128.6 (CH), 128.7
(CH), 129.3 (CH), 129.4 (CH), 136.4 (C), 136.6 (C), 141.9 (CH), 142.4
(C), 142.7 (C), 151.1 (C), 153.7 (C), 157.4 (C), 157.8 (C), 158.9 (C),
162.44 (C). MS (EI, 70 eV): m/z (%)=600 (M+2, 29), 599 (M +1, 23),
598 (M, 62), 430 (32), 429 (100), 413 (5), 315 (6). Anal. calcd. for
C32H36ClN7OSi: C, 64.25; H, 6.07; N, 16.39. Found: C, 64.56; H, 6.18;
N, 16.65.
1045, 908, 766, 699 cmꢁ1 1H NMR (CDCl3): d=1.48–1.52 (m, 2H),
.
(ꢂ)-[cis-7-(2-Amino-6-chloro-9H-purin-9-ylmethyl)-1,4-diphenyl-
1.58–1.65 (m, 2H), 1.69–1.73 (m, 2H), 1.91–1.95 (m, 1H), 2.02–2.05
(m, 2H, one of them D2O exch., OH+6-HH), 2.37 (dt, 1H, J=14.48,
9.96 Hz, 6-HH), 3.36 (dd, 1H, J=12.21, 3.70 Hz, 5-H), 3.54–3.60 (m,
2H, 7-H+HOCHH), 3.89 (d, 1H, J=9.38 Hz, HOCHH), 3.97–4.03 (m,
1H), 4.08–4.17 (m, 1H, NCHH), 4.40–4.51 (m 1H, NCHH), 4.62 (s, 2H,
D2O exch., NH2), 5.74 (s, 1H, D2O exchange, NH), 7.89–7.87 (m, 2H),
7.10 (s, 1H, 8-Hpurine), 7.47–7.61 (m, 6H), 7.77–7.79 (m, 2H). 13C NMR
(CDCl3): d=23.7 (CH2), 28.8 (CH2), 44.0 (CH2), 45.9 (CH), 46.8 (CH),
61.9 (CH2), 113.9 (C), 128.3 (CH), 128.4 (CH), 128.8 (CH), 129.0 (CH),
129.2 (CH), 129.6 (CH), 136.3 (CH), 136.9 (C), 142.4 (C), 143.5 (C),
155.1 (C), 157.4 (C), 158.6 (C), 159.7 (C). MS (EI, 70 eV): m/z (%)=
534 (M+2, 24), 533 (64), 532 (M, 26), 391 (26), 369 (10), 316 (10),
315 (40), 313 (10), 311 (10), 306 (19), 289 (14), 285 (12), 283 (13),
271 (7), 219 (18), 199 (11), 167 (14), 165 (11). Anal. calcd. for
C31H32N8O: C, 69.90; H, 6.06; N, 21.04. Found: C, 70.25; H, 6.19; N;
20.93.
6,7-dihydro-5H-cyclopenta[d]pyridazin-5-yl]methanol (19e):
A
1 M solution of TBAF in THF (0.29 mL, 0.29 mmol) was added drop-
wise to a solution of 18e (88 mg, 0.15 mmol) in the same solvent
(4 mL) that was stirring under Ar in an ice bath. The solution was
allowed to reach r.t., and stirring was continued for a further
30 min. The solvent was removed under reduced pressure and the
residue was chromatographed (silica gel, hexane/EtOAc, 1:1). Con-
centration of the nonvoid fractions to dryness afforded 19e
(70 mg, 98%) as a white solid. Mp 177–1788C. IR (KBr): n=3313,
3197, 2930, 1616, 1566, 1524, 1468, 1411, 1380, 1279, 1170, 1052,
1
912, 773, 698 cmꢁ1. H NMR (DMSOd6): d=1.98–2.02 (m, 1H, 6-HH),
2.52–2.59 (m, 1H, 6-HH), 3.09–3.13 (m, 1H), 3.29–3.31 (m, 1H), 3.36–
3.49 (m, 1H), 3.86–4.04 (m, 2H), 4.53–4.55 (m, 1H), 4.78 (t, 1H, J=
4.93 Hz, D2O exch., OH), 6.77 (s, 2H, D2O exch., NH2), 7.30–7.36 (m,
1H), 7.39–7.43 (m, 2H), 7.52–7.63 (m, 3H), 7.67 (s, 1H, 8-Hpurine), 7.72
218
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
(1R,cis)-3-[2-Amino-6-(cyclopropylamino)-9H-purin-9-ylmethyl]-
1,2,2-trimethylcyclopentylmethanol (23f). A solution of 22f[47]
(100 mg, 0.32 mmol) and cyclopropylamine (183 mg, 3.20 mmol) in
EtOH (12 mL) was refluxed for 20 h. Removal of the solvent under
reduced pressure left a solid from which was chromatography
(silica gel, EtOAc/hexane, 8 :2). Concentration of the non-void frac-
tions to dryness afforded 23f (85 mg, yield 80%) as a white solid.
counted in a coulter counter (Coulter Electronics Ltd, Har-
penden Herts, England) and the number of dead cells was
evaluated by staining with trypan blue. The IC50 (50% inhib-
itory concentration) was defined as the compound concen-
tration that inhibit cell proliferation by 50%, as compared
to untreated control[26]
.
25
M.p. 202–2048C. [a]D +28 (c 0.1, MeOH). IR (KBr): n=3346, 3222,
2964, 1652, 1596, 1487, 1399, 1356, 1027 cmꢁ1 1H NMR (CDCl3):
.
d=0.58–0.63 (m, 2H, cyclopropyl), 0.82–0.88 (m, 2H, cyclopropyl),
0.92 (s, 3H, CH3), 0.99 (s, 3H, CH3), 1.0 (s, 3H, CH3), 1.43–1.52 (m,
2H), 1.56–1.72 (m, 2H), 1.89 (b s, 1H, D2O exch., OH), 2.38–2.43 (m,
1H), 2.96–3.01 (m, 1H, 1-Hcyclopropyl), 3.47 and 3.61 (AB system, 2H,
J=10.7 Hz, CH2OH), 4.09 and 3.83 (AB part of a ABX system, 2H,
4 Results and Discussion
4.1 Model Setup
Our training set contains a diverse set of nucleosides ana-
logues, derived from purinic and pyrimidinic bases with di-
verse values of inhibitory effects in the proliferation of
L1210/0 cancer cells. Our first goal was to establish a dis-
criminant function based on the most relevant 3D molecu-
lar descriptors, after ascertaining them with an adequate
selection method. This model (from now on denoted as
Model-1) is given below together with the statistical param-
eters of the linear discriminant analysis (LDA).
J
AB =13.5 Hz, JAX =10.3 Hz, JBX =4.5 Hz, CH2N), 5.81 (b s, 1H, D2O
exch., NH), 4.78 (b s, 2H, D2O exch., NH2), 7.46 (s, 1H, 8-Hpurine). 13C
NMR (CDCl3): d=160.28 (C), 156,55 (C), 152.46 (C), 137.77 (CH),
115.02 (C), 69.56 (CH2), 49.05 (C), 48.75 (3CH3), 45.35 (CH2), 44.72
(C), 33.85 (CH2), 30.11 (CH2), 26.90 (CH2), 24.07 (CH), 23.81 (CH),
21.22 (CH). 18.71 (CH), 7.82 (CH2). MS (EI, 70 eV): m/z (%)=345
(M+1, 21), 344 (96), 329 (86), 299 (18), 204 (29), 203 (46), 191 (37),
190 (31), 189 (44) 175 (100), 174 (22), 173 (46), 163 (22), 162 (25).
Anal. Calcd. for: C18H28N6O: C, 62.76; H, 8.19; N, 24.40. Found: C,
63.02; H, 8.33; N, 24.52.
Model-1
3.2 Antitumor Activity
P ¼ 1:417 ꢃ J3Dꢁ0:225 ꢃ ADDD þ 78:975 ꢃ MEcc
All assays were carried out in flat-bottomed 96-well micro-
titer plates. To each well were added 5ꢅ104 murine leuke-
mia cells (L1210/0) and a given amount of the test com-
pound. The cells were allowed to proliferate for 48 hours at
378C on a humidified, CO2-controlled atmosphere. The
growth of the cells was linear during this 48 h incubation
period. At the end of the incubation period, the cells were
þ0:841 ꢃ RDF120vꢁ1:529 ꢃ Mor19u ꢃ 0:599 ꢃ Mor12m
þ0:250 ꢃ Mor04e þ 5:598 ꢃ E3sꢁ0:051 ꢃ Au
þ0:136 ꢃ ITHꢁ83:187
ð1Þ
N ¼ 278 1 ¼ 25:27 F ð10,267Þ ¼ 17:610
p < 10ꢁ5 l ¼ 0:602 D2 ¼ 2:819
Figure 2. Distribution of the standardized residuals for all cases studied.
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
219

Full Paper
A. M. Helguera et al.
P ¼ 1:357 ꢃ W1Mor12m þ 0:806 ꢃ W2ITHꢁ0:958
The 1 value, large sample size (N), large F index, and
small p value, are indicative of the model’s statistical signifi-
cance. In addition, the values of the Wilks’ l statistic (l can
take values from zero – perfect discrimination – to one – no
discrimination –) and of the Mahalanobis distance (a mea-
sure of the separation between the active and inactive
groups) show that the model displays an adequate discrim-
inatory power for differentiating both groups. The later is
also confirmed by the classification results; the model cor-
rectly classified 81.4% (83 out of 102) of the 102 cytotoxic
compounds and 81.8% (144 out of 176) of the 176 noncy-
totoxic compounds, giving rise to an overall 81.7% (227
out of 278) effective discrimination of the 278 training set
compounds. Furthermore, the percentage of unclassified
compounds in the training set is only 2.87% (8 out of 278),
while the percentages of false inactives and false actives
are 5.39% (15 out of 278) and 11.5% (32 out of 278), re-
spectively.
Further analysis of this discriminant model should only
be pursued after checking the reliability of the pre-adopted
assumptions. Firstly, LDA establishes a linear additive rela-
tion between the molecular descriptors and the underlying
bioactivity. In fact, this is the simplest mathematical form
that might be envisaged for the model in the absence of
any a priori information. Nevertheless, by looking at the dis-
tribution of the standardized residuals (observed minus
predicted divided by the square root of the residual mean
square) for all cases (Figure 2), no specific pattern can be
seen, therefore suggesting that the model does not exhibit
a non-linear dependence.
ꢃ W3ADDD þ 0:498 ꢃ W4MEcc þ 0:455 ꢃ W5Mor04e
þ 0:452 ꢃ W6E3sꢁ0:400 ꢃ W7Mor19uꢁ0:334 ꢃ W8Au
þ0:521 ꢃ W9RDF120v þ 0:345 ꢃ W10J3Dꢁ0:303
N ¼ 278 1 ¼ 25:27 F ð10,267Þ ¼ 17:244
p < 10ꢁ5 l ¼ 0:607 D2 ¼ 2:760
ð2Þ
where the symbol Wi X means the orthogonal complement
of variable X, while the subscript refers to the order select-
ed for orthogonalizing the variables. Table 2 summarizes
and defines the descriptors used in this model.
As can be seen, all descriptors were found to be statisti-
cally significant and the overall fitness of the model pre-
served as the statistics are as robust as before (see Equa-
tion 1), though the classification of active compounds
slightly improved (82.3%) with respect to that of the non-
orthogonal model (81.4%). By comparing Equation 2 with
Equation 1, one can see that there are no changes in either
the sign of the regression coefficients or of the constant.
Nevertheless the relative contributions of the variables in
Equation 2 are significantly different from those in Equa-
tion 1. Therefore, for purposes of QSAR interpretability, we
shall use the orthogonal-standardized descriptor model de-
fined in Equation 2.
Another important parametric assumption of LDA is mul-
tivariate normality.[40] Figure 3 shows the plots for the histo-
grams of frequency distribution of the discriminant func-
tion, divided by active and non-active groups, respectively,
which takes into account the interaction among variables.
Attached to each plot are also the results derived from the
Kolmogorov–Smirnov statistical test (d). Accordingly, a visual
inspection of the normality plots and frequency distribu-
tions for the discriminant function (Figure 3), as well as the
calculated d values for both groups (actives: d=0.099; in-
actives: d=0.041; p>0.200 in both cases), lead us to
accept the hypothesis of multivariate normality.
The assumption of non-multicollinearity between the de-
scriptors can be readily confirmed by analyzing the correla-
tion matrix. The results in Table 1 show that the pairs of
3D-descriptors (Au; ADDD), (Au; RDF10v), (ITH; ADDD),
(ITH; RDF120v), and (ITH; Au) are highly correlated with
each other, denoting thereby redundancy in the informa-
tion displayed by such pairs. Therefore, we have examined
the performance of orthogonal complements in modeling
the anticancer activity.
[39]
´
Following Randic’s procedure, we determined the or-
Let us now check the hypothesis of homocesdasticity. A
possible problem regarding the homogeneity of the (co)va-
riances is suggested by the Box’s M statistical test (p<0.01),
though this test can be overly sensitive to large data files[40]
thogonal complements for all variables in Model-1 that,
after standardization, allowed us to derive the following
best ten-variable equation (Model-2).
Table 1. Intercorrelation between the ten descriptors used in QSAR model-1 (Eq. 1). Significant correlations are marked in bold.
J3D
ADDD
MEcc
RDF120v
Mor19u
Mor12m
Mor04e
E3s
0.11
0.04
ꢁ0.49
0.07
Au
ITH
J3D
1.00
ꢁ0.34
ꢁ0.26
0.05
1.00
ꢁ0.38
0.70
0.08
0.16
0.74
0.00
ꢁ0.07
ꢁ0.19
0.04
0.10
0.33
0.13
0.14
0.41
ꢁ0.39
0.84
0.01
0.83
0.52
0.16
0.09
0.16
1.00
ꢁ0.38
0.88
ꢁ0.18
0.71
0.56
0.24
0.13
0.27
0.88
1.00
ADDD
MEcc
RDF120v
Mor19u
Mor12m
Mor04e
E3s
1.00
ꢁ0.04
1.00
0.45
1.00
ꢁ0.18
0.06
0.32
1.00
ꢁ0.33
1.00
ꢁ0.16
1.00
Au
ITH
220
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
Table 2. Symbols for the descriptors used in the QSAR models and their definition.
Symbol
Descriptor type
Descriptor definition
Mor12m
Mor04e
Mor19u
RDF120v
ITH
ADDD
MEcc
J3D
3D-MoRSE
3D-MoRSE
3D-MoRSE
RDF
GETAWAY
Geometrical
Geometrical
Geometrical
WHIM
3D-MoRSE – signal 12/weighted by atomic masses
3D-MoRSE – signal 04/weighted by atomic Sanderson electronegativities
3D-MoRSE – signal 19/unweighted
Radial Distribution Function – 12.0/weighted by atomic van der Waals volumes
Total information content on the leverage equality
Average distance/distance degree
Molecular eccentricity
3D-Balaban index
E3s
Au
3rd component accessibility directional WHIM index/weighted by atomic electrotopological states
A total size index/unweighted
WHIM
Figure 3. Histograms for the frequency distributions of the discriminant function (Eq. 1), considering active groups (a) and non-active
groups (b).
which is likely what happened here. This nevertheless in-
creases the likelihood that a case belongs to the higher dis-
persion group and, in this sense, adjusting the a priori
probabilities can greatly improve the overall classification
rate of the discriminant model.
A distinct, better threshold for the a priory classification
probability can be estimated by means of the Receiver Op-
erating Characteristics (ROC) curve.[41] This is a useful tech-
nique not only for obtaining the best thresholds but also
for organizing classifiers. As seen in Figure 4, the optimal
threshold for predicting the active chemicals with the pres-
ent QSAR model is 0.53. Further, one can see that the
model is not a random, but a truly statistically significant,
classifier, since the area under the ROC curve (=0.88) is sig-
nificantly higher than the area under the random classifier
curve (diagonal line).
Having checked the pre-adopted assumptions, it is now
important to access the model quality regarding its robust-
ness and how well it might be expected to generalize, i.e.
how well it will correctly predict the activity of new com-
pounds. This was first accomplished here by means of in-
ternal cross-validation (CV) of the model, using the leave-
group-out procedure. The statistics and classification results
reported in Table 3 correspond to five independent leave-
20%-out CV runs, each involving a different, randomly
chosen partition into a training and a test set.
As can be seen, the model is robust and shows little de-
pendence on the composition of the training and test sets,
since it displays good statistics (l, D2 and F) as previously
(see Eq. 2). It also appears to show a good predictive
power, judging from the averages computed for the overall
classification results in the training and predictive sets gen-
erated in each CV run (81.4% and 79.8%, respectively).
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
221

Full Paper
A. M. Helguera et al.
Figure 4. Receiver operating characteristic (ROC) curve for the final classification model (Eq. 2).
Table 3. Results from the cross-validation leave-group-out procedure. Results obtained with Model-2 (Eq. 2) after removing ca.20% of com-
pounds from the training set. %ACG(T) and %ACG(P) are the percentage of good overall classifications in the training and predicting sets,
respectively.
CV-run
l
D2
F
%ACG(T)
%ACG(P)
1
2
3
4
0.615
0.624
0.593
0.596
0.585
0.603
2.675
2.581
2.927
2.886
3.021
2.818
13.164
12.733
14.557
14.352
15.026
13.966
81.0
80.2
81.6
82.1
82.1
81.4
82.7
85.5
80.0
78.2
72.7
79.8
5
average
4.2 Discovery of Novel Carbonucleoside Anticancer
Compounds
The first group of compounds (4a, 5a and 6a) was syn-
thesized according to the strategy given in Scheme 1. As
shown in this Scheme, compound 4a was prepared using
standard chemistry for purine carbocyclic nucleosides ana-
logues: amino alcohol 1a[42] was condensed with 5-amino-
4,6-dichloropyrimidine, the resulting diamine 2a was cy-
clized with ethyl orthoformate to obtain the intermediate
3a (in 90% yield) and this 9-substituted 6-chloropurine was
further transformed into the final inosine 4a through hy-
drolysis with NaOH. The diazotation of compound 2a under
the usual conditions and with the usual subsequent proce-
dures[43] gave the 8-azapurine derivatives 5a and 6a (see
Scheme 1).
In this work, a design approach for the rational selection
(discovery) of novel compounds with anticancer activity is
applied. The design is conceived here as “a preliminary out-
line showing the main features of something to be execut-
ed”. In this way, we choose the family of carbonucleoside
compounds to be studied with the present approach.
The design process started by drawing the chemical
structures of some of the carbonucleosides obtained in one
of our laboratories and others not even synthesized. The
compounds so designed were evaluated by our discrimi-
nant model and when needed, synthesized afterwards (9
compounds). Notice that, even though some of these com-
pounds were recognized by the model as inactives, they
were nevertheless synthesized for the purpose of QSAR-
model validation.
On the other hand, the second group of compounds
(the uridine analogues 9b, 12c and 15d) were constructed
on the aminoalcohol 7b,[44] 10c[45] and 13d by condensation
with 3-ethoxy-2-propenoyl isocyanate[46] and cyclization of
222
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
Scheme 1. General procedure for the preparation of the test set nucleoside analogues 3a, 4a and 5a.
Scheme 2. General procedure for the preparation of the test set nucleoside analogue 9b.
Scheme 3. General procedure for the preparation of the test set nucleoside analogue 12c.
the urea 8b with conventional acidic conditions (see
Schemes 2, 3 and 4).
pounds were prepared from known meslylate 16e[47]: heat-
ing crude 16e for 24 h at 558C with 2-amino-6-chloropur-
ine, NaH and 18-crown-6 ether (see Scheme 5) afforded a
10% yield of 18e (lower temperatures and shorter reaction
times gave even smaller yields) together with unreacted
16e (10%) and a 59% yield of 17e,[47] the product of a
competing elimination reaction that was shown by TLC
monitoring to produce detectable amounts of 17e within
In a previous work, our research group synthesized 1(N)-
homocarbanucleosides in which the double bond of the cy-
clopentene ring of carbovir and abacavir was replaced with
a pyridazine.[10] Following that work, we have then synthe-
sized the third group of compounds (19e, 21e) that bear
the 2-amino group of carbovir and abacavir. Such com-
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
223

Full Paper
A. M. Helguera et al.
Scheme 4. General procedure for the preparation of the test set nucleoside analogue 15d.
Scheme 5. General procedure for the preparation of the test set nucleoside analogues 19e and 21e.
just a few minutes of the start of the reaction. Refluxing
18e with cyclopentylamine in ethanol gave 20e. Finally, de-
protection of 18e and 20e with TBAF afforded the new pu-
rinyl carbanucleosides 19e and 21e.
Finally, the reaction of (1R,cis)-3-(6-chloro-9H-purin-9-yl-
methyl)-1,2,2-trimethylcyclopentylmethanol (22f)[47] with cy-
clopentylamine afforded the new purinyl carbonucleoside
23f.
The carbonucleoside analogues now synthesized and the
others previously synthesized comprise a new set (22 com-
pounds) for external validation, which in turn is an absolute
requirement towards a final, truly assessment of the predic-
tive power of our QSAR model. So, following on, an experi-
mental evaluation of the inhibitory effects of these com-
pounds on the proliferation of L1210/0 cancer cells was car-
ried out. Four of them have already been evaluated experi-
mentally,[47] but the remaining have not. The results of the
classification and experimental pharmacological evaluations
are given in Table 4.
As can be seen in Table 4, Model-2 (Eq. 2) correctly classi-
fies 80% (4 out of 5) of actives NAs and 88% (15 out of 17)
of inactives. At the same time, the percentage of overall
classification is 86% (19 out of 22), corroborating the good
predictive ability of the present discriminant model. It can
224
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
Scheme 6. General procedure for the preparation of the test set nucleoside analogue 23f.
Table 4. The 22 carbonucleosides used in the external prediction set along with the observed cytotoxicity against the cellular line L1210/0
and the classification (a posteriori probabilities) according to QSAR model-2 (Eq. 2).
No.
IC50(mM)[a]
prob (%)[b]
Predicted class[c]
Reference
Active chemicals
24h
25g
21e
26g
7.89
42.46
65.71
176.76
13.94
99.236
95.449
94.772
90.270
5.019
+1
+1
+1
+1
ꢁ1
[10]
[12]
this work
[12]
this work
23f
Non-active chemicals
27j
28j
29j
5a
30j
4a
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
> 200
97.451
96.722
96.671
96.545
96.457
95.572
94.772
92.981
91.788
88.022
86.458
78.445
74.102
73.074
59.836
43.843
18.807
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
ꢁ1
+1
+1
[21]
[21]
[21]
this work
[21]
this work
this work
this work
this work
[68]
[68]
[11]
[68]
this work
[11]
6a
9b
12c
31k
32k
33i
34k
15d
35i
36g
19e
[12]
this work
[a] 50% inhibitory concentration or compound concentration required to reduce the proliferation of tumors cells by 50%. [b] A posteriori
probability of classifying a chemical as active or inactive, according to Equation 2. [c] Values of +1 and ꢁ1 stand for compounds with and
without cytotoxic activity, respectively.
be also seen that, of the six compounds that were theoreti-
cally classified by our model as active ones, four turn out
to be really inhibitors, and were thus considered as well-
classified. The other two compounds resulted inactive in
the experiment evaluation and should then be considered
as false actives. Furthermore, compound 26g showed some
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
225

Full Paper
A. M. Helguera et al.
activity against the cellular line; however, the values of IC50
for this compound are very high (the highest of all active
compounds) revealing that this compound has a poor anti-
tumor activity.
atomic electrotopological states) covariance matrix of
atomic coordinates. A plot of the molecular descriptor
values for each of these compounds is shown in Figure 5.
Based on this interpretation of Model-2, we thus tried to
modify the structural features of the noncytotoxic carbonu-
cleosides of the training set in order to turn them into cyto-
toxic ones. For instance, Figure 6 shows a 3D-structural
sketch of the inactive chemical 231 and of the new active
chemicals 26g (for simplicity, chemical 25g is not shown as
it has a similar pattern to chemical 26g.). As can be seen,
both NAs have the same carbocycles, but the substituents
on the C6-position of the purine base differ. For improving
the anticancer activity, we try to increase the atomic van
der Waals volume of the substituent at a distance of ca.
12 ꢆ, which is around the C6-position of purine (see
Figure 5). In so doing, by replacing its target NH2 substitu-
ent by a p-chlorophenyl (chemical 25g) or by a p-tolyl
(chemical 26g), that really enhanced the corresponding
RDF120v values of the new designed chemicals (see Fig-
ures 5 and 6). In addition, the descriptors Mor04e, E3s and
ITH also rose for both chemicals (see Figure 6). These de-
scriptors codify electronic factors (Mor04e, E3s) and molec-
ular complexity (ITH). In the particular case of the ITH de-
scriptor, its discriminant power is associated to its strong
dependence on molecular size.[48]
The two false active compounds constitute 9% of bad
classification of the entire data set of carbonucleosides
evaluated. Of these, compounds 36g and 19e are similar to
compounds 26g and 21e, respectively. In the first case, the
only difference between them is the change of the NH-cy-
clopropyl group to a C6H5CH3 group (36g .vs. 26g), while in
the second case the difference is the change of the Cl
atom to a NH-cyclopentyl group (19e .vs. 21e). From the
experimental values of the antitumor activity, it appears
that these differences are enough for suppressing that ac-
tivity. But nevertheless these structural differences are not
well-captured by the present approach.
Overall, five carbonucleoside compounds (13f, 21e, 24h,
25g, 26g) of the external set resulted to be active (IC50 <
200 mM) in the biological assays,[26] of which four were well
predicted by eq 2. The latter were formerly designed by a
careful interpretation of QSAR Model-2. In this model, ten
descriptors are correlated with the anticancer activity (see
Eq. 2 and Table 2), but the most important variables for
these four compounds are the descriptors[34]: RDF12v – ob-
tained by the Radial Distribution Function approach using
a spherical distance of 12 ꢆ weighted by atomic van der
Waals volumes; Mor04e and Mor12m – calculated by sum-
ming atomic weights (atomic electronegativities and
masses, respectively) viewed by a different electron diffrac-
tion angular scattering function; ITH – calculated as the
total information content provided by the leverage matrix;
and E3s – calculated by the projection onto the principal
components obtained from the weighted (in this case, by
A similar approach was followed on for the inactive
chemical 252, which allowed us to design the new active
chemical 21e (see Figure 7). Both are derivatives of the bi-
phenylcyclopentylpyridazine, but their substituents on the
purine base differ. By increasing the bulkiness of the sub-
stituent at the C6-purine position, e.g.: from cyclopropyl to
cyclopentyl (252 vs. 21e), as well as adding an amino
group on the C4-purine position (compound 21e), we could
Figure 5. Values of molecular descriptors Mor12m, ITH, ADD, MEcc, Mor04e, E3s, Mor19u, Au, RDF120v and J3D (eq 2) for the inactive
chemicals (dashed line) and active chemicals (solid line).
226
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
Figure 6. 3D structures of the inactive-chemical 231 (top) and active-chemical 26g (bottom). The maximum interatomic distances to the
substituent on the purine C6-position are shown. Substructures represented by sticks correspond to distances greater than 12 ꢆ.
then increase the value of the RDF120v descriptor (see
Figure 7), and so the biological activity. As can be seen,
Mor12m is also a discriminant descriptor between both
compounds but unlike descriptor RDF120v, its value de-
creased after the structural modification. As according to
the model (Eq. 2) it makes a positive contribution to the
biological activity, it seems that the descriptor Mor12m has
a less effect on the anticancer activity of these structures
than the RDF120v descriptor.
ring condensate to pyridazine is favorable to improve the
anticancer activity. This was also noticed in the previous
cases (see Figures 6 and 7), but in other positions of the
molecular structure. The remaining descriptors did not
show a significant variation.
3.3 2D and 3D Modeling in Balance
Finally, by examining the structural details of chemicals
268 (inactive) and 24h (active), it can be seen that the re-
placement of the hydroxymethyl group located on the C2-
position of the cyclopentyl moiety by the tert-butyldime-
thylsilyl-O-methoxyl group did improve the biological activ-
ity. By inspection of Figure 5, one can notice a significant
increase of the RDF120v value when going from the inac-
tive chemical (268) to the new active one (24h). It can be
easy understand by looking at Figure 8, which shows that
the degree of bulkiness of the substituent located on the
C2-position of the cyclopentyl ring, at an interatomic dis-
tance greater than 12 ꢆ, is greater for the active carbonu-
cleoside than for the inactive one. In other words, a sterical-
ly bulky substituent at the C2-position of the cyclopentyl
In a previous work, we developed a discriminant function
based on 2D-DRAGON descriptors.[33] The following 2D-
QSAR model was derived from a training set of 241 NA
compounds, by combining LDA along with the same varia-
ble selection technique.
2
3
P ¼ 1:828 ꢃ WD=Dr06 þ 3:646 ꢃ WpiPC10
5
6
þ 0:455 ꢃ WMATS8e ꢁ 3:708 ꢃ WpiPC09 ꢁ 0:549
7
ð3Þ
ꢃ WMPC10 þ 0:161
N ¼ 241 1 ¼ 48:2 Fð7,233Þ ¼ 35:74 p < 10ꢁ5
l ¼ 0:57 D2 ¼ 3:21
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
227

Full Paper
A. M. Helguera et al.
Figure 7. 3D structures of the inactive-chemical 252 (top) and active-chemical 21e (bottom). The maximum interatomic distances to the
substituent on the purine C6-position are shown. Substructures represented by sticks correspond to distances greater than 12 ꢆ.
Though 2D-DRAGON molecular descriptors can encode
information about adjacency, branching, etc., they cannot
take into account information regarding conformational as-
pects. With the desire to build a reliable QSAR model
based on 3D-DRAGON descriptors, which could be used to
flag even better potential anticancer activity, we recognized
a unique opportunity when we could assemble a set of
over 270 NAs tested in a single, consistent cytotoxic assay.
For a comparison with the 3D-QSAR model built here,
our former 2D-QSAR model[33] (see Eq. 3) was also used for
predicting the anticancer activity of the present external
test set. In so doing, we found that the model correctly
classified 80% (4 out of 5) of the active compounds, like
the 3D-QSAR model, and 94% (16 out of 17) of the inac-
tives ones, thereby revealing a slightly superior percentage
of external overall classification. Yet, the predicted data can
only be considered reliable for those compounds that fall
within the applicability domain on which the model was
derived.[49] Figure 9 shows a Williams plot, i.e. the plot of
the standardized residuals (y-axis) versus the computed lev-
erage values (x-axis) for each compound of the training set.
From this plot, one can establish the applicability domain
as the squared area within ꢂ2 standard deviations and the
leverage threshold h* (h* is generally fixed at 3p’/n, where
n is the number of training compounds and p’ the number
of model parameters). The leverage thresholds for the 2D
and 3D models are h*=0.074 and h*=0.119, respectively.
As can be noticed, for the 2D-QSAR model, 5 out of 22
compounds of the external test set are outside of this area,
and 3 of them have large standardized residual values
(greater than 2 standard deviation units). In contrast, for
the 3D-QSAR model, only 2 out of the 22 carbocyclic nu-
cleosides have a leverage greater than h* and they show
standard deviation values within the limits. Therefore the
228
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
Figure 8. 3D structures of the inactive-chemical 268 (top) and active-chemical 24h (bottom). The maximum interatomic distances to the
oxygen atom of the hydroxymethyl group and the tert-butyldimethylsilyl-O-methoxyl group located on the C2-position of the cyclopentyl
moiety are shown. Substructures represented by sticks correspond to distances greater than 12 ꢆ.
3D-QSAR model derived here has a greater applicability
domain than the 2D-QSAR model previously developed by
us,[33] most likely due to the fact that it includes 3D-struc-
tural information and was derived using a larger training
set.
Here it is important to remark that, if a chemical is out-
side the applicability domain according to a given, correctly
applied method, this is not a final argument for rejecting
the prediction; rather, it is an indication of the increase un-
certainty of the prediction. We can say that this is, in a stat-
istical sense, an incorrect application of a model, but it is
nevertheless possible that the model will generate a correct
result.
5 Conclusions
In this work, we have examined the ability of a large, di-
verse and consistently tested training set to provide predic-
tive QSAR models for probing potential anticancer activity,
specifically antileukemia activity. The training set included
278 NAs derived from purinic and pyrimidinic bases, and
was assembled from literature compounds with published
cytotoxic activity against L1210/0 cancer cells. For that pur-
pose, we have thoroughly evaluated LDA models in con-
junction with a variety of 3D structure representations and
feature selection algorithms. For this training set, the best
found 3D model showed good accuracy, robustness and
predictivity, as judged by several statistics and extensive in-
ternal cross-validation. The final 3D-model showed also
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
229

Full Paper
A. M. Helguera et al.
Figure 9. Comparison between the applicability domains of our final 3D-QSAR model (Eq. 2) and of the previous 2D-QSAR model (Eq. 3).
good predictive power with an external test set of 22 NA
compounds, which were previously designed by us. From
this external set, nine compounds have never been synthe-
sized, and the cytotoxic activity of eighteen of them is
being reported here for the first time. Therefore, the model
allowed the design, synthesis and biological evaluation of
novel NAs, which is of great relevance as new compounds
enrich the structural diversity of related databases and
could be used in forthcoming QSARs. Although previous ef-
forts with a 2D-classification model were successful, we rec-
ognized that 3D-QSAR models could give more useful infor-
mation, primarily because they highlight the 3D structural
features and stereochemistry of the compounds that are
relevant to their biological activity. In addition, the 3D-
QSAR model derived here was shown to have a greater ap-
plicability domain than that previously developed 2D-
model. Therefore, the structural information gathered and
the 3D-QSAR model per se will certainly help in the future
design of novel potent anticancer drugs.
References
[1] a) H. Frankish, Lancet 2003, 361, 1278; b) A. Jemal, R. Siegel, E.
Ward, Y. Hao, J. Xu, T. Murray, M. J. Thun, CA Cancer J. Clin.
2008, 58, 71.
[2] C. Simons, Carbocyclic Nucleosides; Gordon and Breach, UK
2001.
[3] a) J. M. Blanco, O. Caamano, F. Fernandez, X. Garcꢀa-Mera, A. R.
Hergueta, C. Lopez, J. E. Rodriguez-Borges, J. Balzarini, E. De
Clercq, Chem. Pharm. Bull. (Tokyo) 1999, 47, 1314; b) M. J. Fig-
ueira, J. M. Blanco, O. CaamaÇo, F. Fernꢂndez, X. Garcꢀa-Mera,
C. Lꢇpez, Arch. Pharm. Pharm. Med. Chem. 1999, 332.
[4] M. T. Crimmins, Tetrahedron 1998, 54, 9229.
[5] M. I. Nieto, O. CaamaÇo, F. Fernꢂndez, G. Gꢇmez, J. Balzarini, E.
De Clercq, Nucleosides, Nucleotides Nucleic Acids 2002, 21, 243.
[6] M. J. Figueira, O. CaamaÇo, F. Fernꢂndez, J. E. Rodrꢀguez-
Borges, J. Balzarini, E. De Clercq, Synthesis 2004, 12, 1991.
[7] A. R. Hergueta, F. Fernandez, C. Lopez, J. Balzarini, E. De
Clercq, Chem. Pharm. Bull. (Tokyo) 2001, 49, 1174.
[8] C. Balo, F. Fernꢂndez, E. Lens, C. Lꢇpez, Arch. Pharm. Pharm.
Med. Chem. 1997, 265.
[9] J. M. Blanco, O. Caamano, F. Fernandez, J. E. Rodriguez-Borges,
J. Balzarini, E. de Clercq, Chem. Pharm. Bull. (Tokyo) 2003, 51,
1060.
[10] S. F. Wnuk, C. S. Yuan, R. T. Borchardt, J. Balzarini, E. De Clercq,
M. J. Robins, J. Med. Chem. 1997, 40, 1608.
Acknowledgment
[11] E. Estrada, E. Uriarte, A. Montero, M. Teijeira, L. Santana, E. De
Clercq, J. Med. Chem. 2000, 43, 1975.
[12] H. Gonzalez-Diaz, D. ViÇa, L. Santana, E. de Clercq, E. Uriarte,
Bioorg. Med. Chem. 2006, 14, 1095.
[13] S. Prekupec, D. Svedruzic, T. Gazivoda, D. Mrvos-Sermek, A.
Nagl, M. Grdisa, K. Pavelic, J. Balzarini, E. De Clercq, G. Folkers,
The authors acknowledge the Portuguese Fundażo para a
CiÞncia e a Tecnologia (FCT) (SFRH/BD/22692/2005) and the
Xunta de Galicia (Project 07CSA008203PR) for financial sup-
port, and appreciate the assistance of Yunierkis Pꢂrez-Castil-
lo for help in performing some of the calculations.
230
www.molinf.com
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2010, 29, 213 – 231

Antineoplastic Activity of Novel Carbocyclic Nucleosides
L. Scapozza, M. Mintas, S. Raic-Malic, J. Med. Chem. 2003, 46,
5763.
[34] R. Todeschini, V. Consonni, Handbook of Molecular Descriptors,
Wiley-VCH, Weinheim, Germany, 2000.
[14] A. A. Moosavi-Movahedi, S. Hakimelahi, J. Chamani, G. A. Kho-
darahmi, F. Hassanzadeh, F. T. Luo, T. W. Ly, K. S. Shia, C. F. Yen,
M. L. Jain, R. Kulatheeswaran, C. Xue, M. Pasdar, G. H. Hakime-
lahi, Bioorg. Med. Chem. 2003, 11, 4303.
[15] S. Raic-Malic, D. Svedruzic, T. Gazivoda, A. Marunovic, A. Her-
gold-Brundic, A. Nagl, J. Balzarini, E. De Clercq, M. Mintas, J.
Med. Chem. 2000, 43, 4806.
[16] L. Santana, M. Teijeira, E. Uriarte, J. Balzarini, E. De Clercq, Eur.
J. Med. Chem. 2002, 37, 755.
[17] G. H. Hakimelahi, N. W. Mei, A. A. Moosavi-Movahedi, H.
Davari, S. Hakimelahi, K. Y. King, J. R. Hwu, Y. S. Wen, J. Med.
Chem. 2001, 44, 1749.
[35] O. CaamaÇo, G. Gꢇmez, F. Fernꢂndez, M. D. Garcꢀa, X. Garcꢀa-
Mera, E. De Clercq, Synthesis 2004, 17, 2855.
[36] J. M. Blanco, O. CaamaÇo, F. Fernꢂndez, G. Gꢇmez, M. I. Nieto,
J. Balzarini, E. Padalko, E. De Clercq, Nucleosides Nucleotides
1997, 16, 159.
[37] a) F. Fernꢂndez, X. Garcꢀa-Mera, M. Morales, L. VilariÇo, O. Caa-
maÇo, E. De Clercq, Tetrahedron 2004, 60, 9245; b) F. Fernꢂn-
dez, X. Garcꢀa-Mera, C. Lꢇpez, M. Morales, J. E. Rodrꢀguez-
Borges, Synthesis 2005, 20, 3549; c) C. Balo, J. M. Blanco, F. Fer-
nandez, E. Lens, C. Lꢇpez, Tetrahedron 1998, 54, 2833.
[38] a) R. Todeschini, V. Consonni, M. Pavan, Dragon Software, Ver-
sion 2.1. 2002; b) J. Frank, MOPAC, Version 6.0, S. R. Laborato-
ries, US Air Force Academy, Colorado Springs, CO, 1993;
c) STATISTICA (data analysis software system), Version 7.0, Stat-
soft Inc, 2002.
[18] G. H. Hakimelahi, A. A. Moosavi-Movahedi, T. Sambaiah, J. L.
Zhu, K. S. Ethiraj, M. Pasdar, S. Hakimelahi, Eur. J. Med. Chem.
2002, 37, 207.
[19] S. Hatse, L. Naesens, E. De Clercq, J. Balzarini, Biochem. Phar-
macol. 1999, 58, 311.
[20] C. Teran, L. Santana, M. Teijeira, E. Uriarte, E. De Clercq, Chem.
Pharm. Bull. (Tokyo) 2000, 48, 293.
´
[39] M. Randic, New. J. Chem. 1991, 15, 517.
[40] M. H. Pestana, J. N. Gageiro, Anꢁlise de dados para ciÞncias so-
ciais. A complementaridade do SPSS; Lisboa: EdiÅꢈes Sꢀlabo,
2008.
[21] Z. X. Wang, L. I. Wiebe, J. Balzarini, E. De Clercq, E. E. Knaus, J.
Org. Chem. 2000, 65, 9214.
[22] S. F. Wnuk, C. S. Yuan, R. T. Borchardt, J. Balzarini, E. De Clercq,
M. J. Robins, J. Med. Chem. 1994, 37, 3579.
[23] N. G. Kundu, P. Das, J. Balzarini, E. De Clercq, Bioorg. Med.
Chem. 1997, 5, 2011.
[24] I. A. Mikhailopulo, N. E. Poopeiko, T. I. Pricota, G. G. Sivets, E. I.
Kvasyuk, J. Balzarini, E. De Clercq, J. Med. Chem. 1991, 34,
2195.
[25] S. Raic-Malic, A. Hergold-Brundic, A. Nagl, M. Grdisa, K. Pavelic,
E. De Clercq, M. Mintas, J. Med. Chem. 1999, 42, 2673.
[26] E. De Clercq, J. Balzarini, P. F. Torrence, M. P. Mertes, C. L.
Schmidt, D. Shugar, P. J. Barr, A. S. Jones, G. Verhelst, R. T.
Walker, Mol. Pharmacol. 1981, 19, 321.
[27] E. De Clercq, J. Descamps, J. Balzarini, J. Giziewicz, P. J. Barr,
M. J. Robins, J. Med. Chem. 1983, 26, 661.
[28] M. J. Robins, P. W. Hatfield, J. Balzarini, E. De Clercq, J. Med.
Chem. 1984, 27, 1486.
[41] F. Provost, T. Fawcett, Analysis and Visualization of Classifier
Performance Comparison Under Class and Cost Distributions,
American Association for Artificial Intelligence, 1997.
[42] F. Fernꢂndez, C. Lꢇpez, A. R. Hergueta, Tetrahedron 1995, 51,
10317.
[43] R. Vince, S. Daluge, J. Org. Chem. 1980, 45, 531.
[44] C. Balo, O. CaamaÇo, F. Fernꢂndez, C. Lꢇpez, Tetrahedron:
Asymmetry 2005, 16, 2593.
[45] M. Morales, F. Fernꢂndez, X. Garcꢀa-Mera, J. E. Rodrꢀguez-
Borges, in XI National Congress of the Spanish Society of Thera-
peutic Chemistry: Valencia (Spain), 1999.
[46] F. Fernꢂndez, X. Garcꢀa-Mera, M. Morales, J. E. Rodrꢀguez-
Borges, Synthesis 2001, 2, 239.
[47] M. I. Nieto, J. M. Blanco, O. CaamaÇo, F. Fernꢂndez, X. Garcꢀa-
Mera, J. Balzarini, E. Padalko, J. Neytsb, E. De Clercq, Nucleo-
sides Nucleotides 1998, 17, 1255.
[48] V. Consonni, R. Todeschini, M. Pavan, P. Gramatica, J. Chem. Inf.
Comput. Sci. 2002, 42, 693.
[49] T. I. Netzeva, A. P. Worth, T. Aldenberg, R. Benigni, M. T. D.
Cronin, P. Gramatica, J. S. Jaworska, S. Kahn, G. Klopman, C. A.
Marchant, G. Myatt, N. Nikolova-Jeliazkova, G. Y. Patlewicz, R.
Perkins, D. W. Roberts, T. W. Schultz, D. T. Stanton, J. J. M. van
de Sandt, W. Tong, G. Veith, C. Yang, ATLA 2005, 33, 155.
[29] R. N. Hunston, A. S. Jones, C. McGuigan, R. T. Walker, J. Balzari-
ni, E. De Clercq, J. Med. Chem. 1984, 27, 440.
[30] L. A. Al-Razzak, D. Schwepler, C. J. Decedue, J. Balzarini, E. De
Clercq, M. P. Mertes, J. Med. Chem. 1987, 30, 409.
[31] M. Bobek, S. H. An, D. Skrincosky, E. De Clercq, R. J. Bernacki, J.
Med. Chem. 1989, 32, 799.
[32] E. De Clercq, J. Balzarini, J. Descamps, C. F. Bigge, C. T. Chang,
P. Kalaritis, M. P. Mertes, Biochem. Pharmacol. 1981, 30, 495.
[33] A. M. Helguera, J. E. Rodriguez-Borges, X. Garcꢀa-Mera, F. Fer-
nandez, M. N. Cordeiro, J. Med. Chem. 2007, 50, 1537.
Received: September 10, 2009
Accepted: January 25, 2010
Published online: March 5, 2010
Mol. Inf. 2010, 29, 213 – 231
ꢃ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
231