QSAR on antiproliferative naphthoquinones based on a conformation- independent approach

Article abstract of DOI:10.1016/j.ejmech.2014.02.057

The antiproliferative activities of a series of 36 naphthoquinone derivatives were subjected to a Quantitative Structure-Activity Relationships (QSAR) study. For this purpose a panel of four human cancer cell lines was used, namely HBL-100 (breast), HeLa (cervix), SW-1573 (non-small cell lung) and WiDr (colon). A conformation-independent representation of the chemical structure was established in order to avoid leading with the scarce experimental information on X-ray crystal structure of the drug interaction. The 1179 theoretical descriptors derived with E-Dragon and Recon software were simultaneously analyzed through linear regression models based on the Replacement Method variable subset selection technique. The established models were validated and tested through the use of external test sets of compounds, the Leave-One-Out Cross Validation method, Y-Randomization and Applicability Domain analysis.

Full text of DOI:10.1016/j.ejmech.2014.02.057

European Journal of Medicinal Chemistry 77 (2014) 176e184
Contents lists available at ScienceDirect
European Journal of Medicinal Chemistry
journal homepage: http://www.elsevier.com/locate/ejmech
Original article
QSAR on antiproliferative naphthoquinones based on a
conformation-independent approach
,
b
c
d
Pablo R. Duchowicz a , Daniel O. Bennardi , Daniel E. Bacelo , Evelyn L. Bonifazi ,
*
Carla Rios-Luci e, José M. Padrón e, Gerardo Burton d, Rosana I. Misico d
a Instituto de Investigaciones Fisicoquímicas Teóricas y Aplicadas INIFTA (UNLP, CCT La Plata-CONICET), Diag. 113 y 64, C.C. 16, Sucursal 4, 1900 La Plata,
,
*
Argentina
b Catedra de Química Organica, Facultad de Ciencias Agrarias y Forestales, Universidad Nacional de La Plata (UNLP), 60 y 119, B1904AAN La Plata,
Argentina
c Departamento de Química, Facultad de Ciencias Exactas y Naturales, Universidad de Belgrano, Villanueva 1324, C1426BMJ Ciudad de Buenos Aires,
Argentina
d Departamento de Química Orgánica y UMYMFOR (CONICET-UBA), Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Pabellón 2,
Ciudad Universitaria, C1428EGA Ciudad de Buenos Aires, Argentina
e BioLab, Instituto Universitario de Bio-Orgánica “Antonio González” (IUBO-AG), Centro de Investigaciones Biomédicas de Canarias (CIBICAN),
Universidad de La Laguna, C/Astrofísico Francisco Sánchez 2, 38206 La Laguna, Spain
a r t i c l e i n f o
a b s t r a c t
Article history:
The antiproliferative activities of a series of 36 naphthoquinone derivatives were subjected to a Quan-
titative StructureeActivity Relationships (QSAR) study. For this purpose a panel of four human cancer cell
lines was used, namely HBL-100 (breast), HeLa (cervix), SW-1573 (non-small cell lung) and WiDr (colon).
A conformation-independent representation of the chemical structure was established in order to avoid
leading with the scarce experimental information on X-ray crystal structure of the drug interaction. The
1179 theoretical descriptors derived with E-Dragon and Recon software were simultaneously analyzed
through linear regression models based on the Replacement Method variable subset selection technique.
The established models were validated and tested through the use of external test sets of compounds, the
Leave-One-Out Cross Validation method, Y-Randomization and Applicability Domain analysis.
Ó 2014 Elsevier Masson SAS. All rights reserved.
Received 16 December 2013
Received in revised form
5 February 2014
Accepted 25 February 2014
Available online 26 February 2014
Keywords:
QSAR theory
Multivariable linear regression
Molecular descriptors
́
́
Cancer
Naphthoquinone derivative
1. Introduction
with other structural factors. Within the quinone group, the
naphthoquinones (NQs) represented an important class of com-
Quinones widely occurred in animals, plants and microorgan-
isms, and often carried out indispensable roles in the biochemistry
of energy production by providing vital links in the respiratory
chain of living cells. These compounds acted as inhibitors of elec-
tron transport, uncouplers of oxidative phosphorylation, and gave
rise to a wide range of cytostatic and antiproliferative activities [1].
Hence, quinones such as doxorubicin, salvicine, saframycin, sain-
topin, actinomycin D, etc. were widely used in cancer chemo-
therapy [2]. The National Cancer Institute (NCI) had identified the
quinone unit as an important pharmacologic element for the
cytotoxic activity, and it had been speculated that the latter was due
to an intrinsic chemical property of the quinone core associated
pounds with featured biological activities, where the 1,2- and 1,4-
naphthoquinones had been considered privileged structures in
Medicinal Chemistry [3e5].
Lapachol (1), a natural naphthoquinone, and many related
heterocyclic derivatives, had been investigated during the past
years mainly due to their antimicrobial [6,7], antifungal [8,9], try-
panocidal [10], and anticancer activities [11]. b-Lapachone (2), a
cyclization product of lapachol, had been demonstrated cytotoxic
activity against a variety of cancer cells both in vitro and in vivo, and
it was currently under phase II clinical trials as a new anticancer
drug [12]. Although the cytotoxic action of b-lapachone had been
known for more than 20 years [13], the detailed mechanism of
action remained largely unknown.
It had been shown that the introduction of a hydroxyl group peri
at the carbonyls of the 1,4-naphthoquinone nucleus increased its
reduction potential by means of the formation of hydrogen bonds
* Corresponding authors.
E-mail addresses: pabloducho@gmail.com (P.R. Duchowicz), misicori@qo.fcen.
uba.ar (R.I. Misico).
http://dx.doi.org/10.1016/j.ejmech.2014.02.057
0223-5234/Ó 2014 Elsevier Masson SAS. All rights reserved.

P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
177
with the semiquinone radical and quinone dianion [14]. This effect
might have been responsible for the enhancement in the biological
activity of these NQs. There were few examples in the literature of
natural or synthetic NQs with the characteristics above mentioned
and all of them showed an enhanced antitumor activity as a direct
consequence of the presence of such group [15e18].
In recent years, we had reported the synthesis and anti-
proliferative activities in human solid tumor cell lines of NQs
related to lapachol and b-lapachone, with the goal of identifying
and 2D QSAR methods had been better suited for the development
of relationships between the molecular structure and the biological
activities of compounds [33].
In this work, we reported on the synthesis of a set of ortho-
furanonaphthoquinones (4e9) with a hydroxyl group at position 6
or 9, and the evaluation of their antiproliferative activities against a
panel of representative human solid tumor cell lines comprising
HBL-100 (breast), HeLa (cervix), SW1573 (non-small cell lung), and
WiDr (colon). We examined the experimental values of the anti-
proliferative activities observed for these compounds and the NQs
previously synthesized by us (1e3, 10e36), and developed a QSAR
study to describe the structure-activity function. Herein, we
described an approach that did not consider the conformational
representation of the chemical structure by only relying on the
constitutional and topological aspects of the molecules as previ-
ously argued [34] (Fig. 1).
new antitumor agents. Our main strategy was the introduction of
hydroxy or methoxy groups on the benzene ring, resulting in 7-
hydroxy-b-lapachone (3) as a lead with enhanced activity over
the parent drug
b
-lapachone [19,20].
The well-known basis of the Quantitative StructureeActivity
Relationships (QSAR) theory had been the hypothesis that the
biological activity exhibited by a chemical compound was mainly
determined by its molecular structure [21e23]. This fact did not
offer specific details on the usually complex mechanism/path of the
process that resulted in the final biological effect. However, it was
possible to get some insight into the underlying mechanism of
action by means of the QSAR-based predicted activities. Although
naphthoquinones had been widely studied in the past in QSAR
studies of their antiproliferative activities, many of such studies
involved very few molecules [24e27], and others did not consider
the simultaneous exploration of a great number of molecular de-
scriptors for obtaining the best ones to be used in the QSAR model
[28,29].
Among the main serious drawbacks that had suffered different
widely used 3D-QSAR methods, such as Molecular Docking based
QSAR [30], Comparative Molecular Field Analysis (CoMFA) [31] or
Comparative Molecular Similarity Indices Analysis (CoMSIA) [32],
the following ones were worth mentioning: i) the lack of accurate
X-ray crystal structures showing the drug interaction, and ii) 3D-
QSAR methods were not applicable to structurally diverse com-
pound datasets [23]. In contrast, the conformation-independent 1D
2. Results and discussion
2.1. Chemistry
Compounds 1e3 and 10e36 were prepared according to pre-
viously reported in-house methods [19,20]. The hydroxydunniones
(4e7) were obtained from 5-hydroxydunniol (32) and 8-
hydroxydunniol (33) under different acidic conditions as shown
in Scheme 1. Thus, treatment of 32 with methanesulfonic acid gave
selectively compound 4 while reaction of the latter with 20% hy-
drochloric acid under reflux gave 37. Compound 6 was obtained in
15% yield upon reaction of 37 with concentrated sulfuric acid
overnight (compound 37 was recovered in 74% yield).
In a similar way, treatment of 8-hydroxydunniol (33) with
methanesulfonic acid gave quantitatively compound 5. Isomeriza-
tion of the latter compound required harsher conditions than those
used above, treatment with concentrated hydrochloric acid at 85 ꢀC
overnight gave 38 in 50% yield (39% of unreacted 5 was recovered).
Fig. 1. Structures of naphthoquinones analyzed.

178
P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
Table 1
GI50 values of naphthoquinones against human solid tumor cells.a
Compound HBL-100 (breast) HeLa (cervix) SW1573 (lung) WiDr (colon)
1b
7.8 (ꢂ4.6)
0.38 (ꢂ0.14)
0.036 (ꢂ0.009)
2.2 (ꢂ0.1)
0.54 (ꢂ0.11)
2.0 (ꢂ0.1)
0.34 (ꢂ0.01)
3.0 (ꢂ0.2)
0.46 (ꢂ0.05)
1.8 (ꢂ0.1)
0.46 (ꢂ0.02)
1.9 (ꢂ0.3)
0.41 (ꢂ0.05)
0.60 (ꢂ0.12)
21 (ꢂ4.1)
2.3 (ꢂ0.8)
1.0 (ꢂ0.4)
34 (ꢂ6.5)
36 (ꢂ9.8)
2.0 (ꢂ0.2)
2b
0.84 (ꢂ0.32)
3c
4
5
6
7
8
9
10
0.21 (ꢂ0.04) 0.029 (ꢂ0.007) 2.0 (ꢂ0.2)
1.9 (ꢂ0.1)
1.9 (ꢂ0.1)
0.90 (ꢂ0.03)
1.80 (ꢂ0.07)
3.0 (ꢂ0.2)
0.21 (ꢂ0.01)
1.6 (ꢂ0.3)
1.5 (ꢂ0.3)
1.7 (ꢂ0.4)
1.3 (ꢂ0.3)
0.42 (ꢂ0.11)
2.6 (ꢂ0.4)
30 (ꢂ13)
2.4 (ꢂ0.2)
0.99 (ꢂ0.05)
1.3 (ꢂ0.1)
0.48 (ꢂ0.03)
2.70 (ꢂ0.01)
1.00 (ꢂ0.02)
1.6 (ꢂ0.2)
0.80 (ꢂ0.29)
1.3 (ꢂ0.3)
0.40 (ꢂ0.13)
0.70 (ꢂ0.23)
18 (ꢂ2.7)
19 (ꢂ1.1)
5.2 (ꢂ1.3)
2.1 (ꢂ0.7)
16 (ꢂ5.3)
26 (ꢂ2)
Scheme 1. Synthesis of hydroxydunniones 4e7. Reagents and conditions: a) MsOH; b)
HCl; c) H2SO4 (c).
2.2 (ꢂ0.3)
1.8 (ꢂ0.1)
1.9 (ꢂ0.1)
2.0 (ꢂ0.03)
2.0 (ꢂ0.2)
6.3 (ꢂ1.9)
19 (ꢂ2.4)
>100
Compound 7 was obtained in 32% yield upon reaction of 38 with
concentrated sulfuric acid (55% of unreacted 38 was recovered).
Attempts to carry out the isomerization of compounds 4 and 5 to 6
and 7, respectively with concentrated sulfuric acid were
unsuccessful.
11
12b
13b
14b
15b
16c
17b
18b
19c
20b
21b
22c
23c
24c
25c
26c
27c
28c
29c
30c
31b
32b
33b
34b
35b
36b
22 (ꢂ2.9)
26 (ꢂ4.5)
Due to the interesting antiproliferative activity exhibited by 5-
hydroxy-dehydro-iso-a-lapachone (36) [17] we decided to carry
20 (ꢂ6.9)
3.5 (ꢂ0.6)
18 (ꢂ1)
4.8 (ꢂ0.2)
5.8 (ꢂ1.7)
17 (ꢂ2.0)
20 (ꢂ2.7)
28 (ꢂ5.5)
68 (ꢂ16)
26 (ꢂ0.6)
18 (ꢂ1.6)
>100
14 (ꢂ2.2)
82 (ꢂ13)
1.9 (ꢂ0.1)
2.1 (ꢂ0.2)
21 (ꢂ1.6)
27 (ꢂ0.8)
86 (ꢂ25)
2.1 (ꢂ0.03)
1.9 (ꢂ0.1)
31 (ꢂ1.6)
23 (ꢂ1.8)
18 (ꢂ2.1)
2.0 (ꢂ0.1)
1.9 (ꢂ0.2)
17 (ꢂ2.3)
out the synthesis and antiproliferative evaluation of the isomeric
orthoquinone 8. Thus, 5-hydroxylapachol (14) was reduced with
zinc and acetylated in situ to give the intermediate peracetylated
hydroquinone (not isolated) that was immediately oxidized with
selenium dioxide (Scheme 2). Subsequent deacetylation gave 5-
hydroxylomatiol (39). Compound 8 was obtained in 50% overall
yield by cyclization of 39 with aqueous sulfuric acid at low tem-
perature and short reaction time. Using the same reaction
sequence, compound 9 was obtained in 54% overall yield from 8-
hydroxylapachol (15) (Scheme 2).
19 (ꢂ0.6)
18 (ꢂ0.8)
15 (ꢂ2.6)
12 (ꢂ1.2)
>100
14 (ꢂ4.5)
3.1 (ꢂ0.1)
2.7 (ꢂ0.7)
>100
2.6 (ꢂ0.8)
28 (ꢂ3.1)
3.4 (ꢂ0.2)
28 (ꢂ4.1)
1.8 (ꢂ0.2)
40 (ꢂ8.6)
1.1 (ꢂ0.2)
1.8 (ꢂ0.4)
2.5 (ꢂ0.3)
1.7 (ꢂ0.2)
5.1 (ꢂ0.4)
1.6 (ꢂ0.03)
1.3 (ꢂ0.4)
28 (ꢂ1.6)
24 (ꢂ5.3)
2.2 (ꢂ0.34)
0.8 (ꢂ0.41)
2.4 (ꢂ0.7)
19 (ꢂ0.9)
27 (ꢂ3.0)
0.62 (ꢂ0.16)
0.69 (ꢂ0.03)
2.3 (ꢂ0.1)
0.41 (ꢂ0.16)
2.7 (ꢂ0.4)
0.18 (ꢂ0.05)
1.1 (ꢂ0.1)
31 (ꢂ3)
0.36 (ꢂ0.01)
0.46 (ꢂ0.01)
2.3 (ꢂ0.1)
0.45 (ꢂ0.15)
1.4 (ꢂ0.5)
0.23 (ꢂ0.01)
0.56 (ꢂ0.01)
36 (ꢂ6.3)
2.2. Biology
16 (ꢂ1.8)
3.8 (ꢂ1.3)
2.8 (ꢂ1.3)
0.83 (ꢂ0.35)
2.3 (ꢂ0.5)
2.8 (ꢂ0.5)
0.94 (ꢂ0.36)
2.0 (ꢂ0.5)
2.2.1. Antiproliferative activity of the synthesized compounds
The naphthoquinones 1e36 were screened for their anti-
proliferative effects on a panel of human solid tumor cells: HBL-100
(breast), HeLa (cervix), SW1573 (non-small cell lung), and WiDr
(colon) and the results, expressed as GI50, were obtained using the
SRB assay [35]. Table 1 shows the antiproliferative activities of all
the naphthoquinones assayed in the QSAR study. The experimental
antiproliferative potencies were converted into logarithm
pGI50 ¼ ꢁlog10GI50, for the QSAR study. The variation range of the
activities in each human solid tumor cell line was: HBL-100 (breast)
[ꢁ1.4914, 1.4437], HeLa (cervix) [ꢁ2.0000, 0.6778], SW-1573 (non-
small cell lung) [ꢁ2.0000, 1.5376], and WiDr (colon) [ꢁ2.0000,
ꢁ0.2553].
a
Values are given in
m
M and are means of three to six experiments (ꢂstandard
deviation).
b
Data from Bioorg. Med. Chem. 18 (2010), 2621e2630.
Data from Eur. J. Med. Chem. 53, 2012, 264e274.
c
2.3. Model development
2.3.1. Molecular descriptors selection in MLR
In recent years, theoretical and experimental researchers had
focused an increasing attention on finding the most efficient tools
for selecting molecular descriptors in QSAR studies. There were
available a great number of feature selection methods to search for
the best descriptors from a pool of variables, and the Replacement
Method (RM) [36,37], employed here, had been successfully
applied elsewhere [38e42]. In brief, the RM was an efficient opti-
mization tool which generated Multivariable Linear Regression
(MLR) based models on the training set by searching in a set having
D descriptors for an optimal subset having d << D ones with
smallest training set standard deviation (Strain) or smallest root
mean square deviation (RMSDtrain). Table 1S included a list of
mathematical equations used in the present study, with definitions
for S and RMSD. The quality of the results achieved with this
technique approached that obtained by performing an exact
(combinatorial) full search of molecular descriptors although, of
course, required much less computational work. However, in some
cases, the RM could get trapped in a local minimum of S. Although
such local minima provided acceptable models, an improvement of
the method had been developed in the Enhanced Replacement
Method (ERM) [43,44]. The Matlab 7.12 software was used in all our
calculations [45].
Direct comparison between para and ortho tricyclic isomers
showed comparable activity for furanonaphthoquinones, while in
the case of pyranonaphthoquinones generally the
b isomer was
more active than the . Comparison between o-furanonaph-
a
thoquinones with a hydroxyl group at position 6 or 9 of the aro-
matic ring showed that the presence of a peri hydroxyl increased
antiproliferative activity in some cell lines, while for other lines the
activity was comparable. Compound 3 gave the best results of all
the evaluated NQs in three of the tested cell lines, being 29 times
more active than b-lapachone on SW1573 cells.
Scheme 2. Synthesis of compounds 8 and 9. Reagents and conditions: a) Zn, Ac2O, Et3N,
rt, b) i. SeO2, Ac2O, HAcO, 150 ꢀC, ii. MeOH, KOH 6 M, reflux; c) H2SO4:H2O (5:3), 0 ꢀC.

P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
179
2.3.2. Model validation
HBL-100:
The most efficient validation strategy had always consisted on
using an external test set of molecules, never seen by the model
during the calibration of its parameter values. For this purpose, the
training set included 30 naphthoquinone derivatives and the test
set the remaining 6 compounds. This partition was chosen for each
human solid tumor cell line in such a way that both sets shared
similar qualitative structure-activity characteristics, as the training
set should be representative of the molecular diversity of all the
compounds under study and uniformly span over the whole ac-
tivity range.
The Cross-Validation technique of Leave-One-Out (loo) was
practiced [46]. The parameters R2loo and Sloo (square of the corre-
lation coefficient and standard deviation of Leave-One-Out)
measured the stability of the model upon exclusion of molecules.
According to the literature, R2loo should be greater than 0.5 for a
validated model.
The Y-Randomization procedure [47] was also applied in order
to verify that the model was robust. This technique consisted on
scrambling the experimental property values in such a way that
they did not correspond to the respective compounds. After
analyzing 50,000 cases of Y-Randomization, the standard deviation
obtained (Srand) had to be a poorer value than the one found by
considering the true calibration (S).
pGI50 ¼ ꢁ1:502ICRꢁ3:860GATS5eꢁ2:742BELe7þ8:640
R2train ¼ 0:68; Strain ¼ 0:44; Ri2j max ¼ 0:19; oð2:5SÞ ¼ 0
(1)
R2loo ¼ 0:55; Sloo ¼ 0:52; Srand ¼ 0:51; Rt2est ¼ 0:80; Stest ¼ 0:58
HeLa:
pGI50 ¼ ꢁ4:684GATS5eꢁ57:240JGI6ꢁ0:840Oꢁ060þ6:057
(2)
R2train ¼ 0:67; Strain ¼ 0:38; Ri2j max ¼ 0:50; oð2:5SÞ ¼ 0
R2loo ¼ 0:56; Sloo ¼ 0:44; Srand ¼ 0:41; Rt2est ¼ 0:78; Stest ¼ 0:46
SW-1573:
pGI50 ¼ ꢁ7:216EEig11xþ10:046EEig11rꢁ7:883BELp7þ8:666
(3)
R2train ¼ 0:73; Strain ¼ 0:42; Ri2j max ¼ 0:91; oð2:5SÞ ¼ 0
2.3.3. Applicability domain analysis
The applicability domain (AD) for the QSAR models was also
explored, as not even a predictive model was expected to reliably
predict the modeled activity for the whole universe of mole-
cules. The AD was a theoretically defined area that depended on
the descriptors and the experimental activity [48]. Only the
molecules falling within this AD were not considered model
extrapolations. One possible way to characterize the AD was
based on the leverage approach [49], which allowed to verify
whether a new compound could be considered as interpolated
(with reduced uncertainty, reliable prediction) or extrapolated
outside the domain (unreliable prediction). Each compound i
had a calculated leverage value (hi) and there existed a warning
leverage value (h*); Table 1S included the definitions for hi and
h*. When hi > h* for a test set compound, then a warning should
be given: it meant that the prediction was the result of sub-
stantial extrapolation of the model and could not be treated as
reliable.
R2loo ¼ 0:65; Sloo ¼ 0:48; Srand ¼ 0:48; Rt2est ¼ 0:86; Stest ¼ 0:45
WiDr:
pGI50 ¼ ꢁ1:939ICRþ2:336 ATS6vꢁ4:715 BELe7ꢁ0:158
R2train ¼ 0:76; Strain ¼ 0:32; Ri2j max ¼ 0:31; oð2SÞ ¼ 0
(4)
R2loo ¼ 0:68; Sloo ¼ 0:37; Srand ¼ 0:40; Rt2est ¼ 0:96; Stest ¼ 0:23
here, Rijmax denoted the maximum correlation coefficient between
descriptor pairs, o(2S) indicated the number of outlier compounds
having a residual (difference between experimental and calcu-
lated activity) greater than 2 times Strain and lower than three
times Strain
.
Equations (1)e(4) had an acceptable predictive power on the
external test sets including each one 6 naphthoquinone com-
pounds, according to R2test and Stest parameters. Such models
approved the internal validation process of Cross-Validation
through the exclusion of one molecule at a time. The Y-Randomi-
zation technique demonstrated that all these equations had
Strain < Srand and thus valid structureeactivity relationships were
achieved. We checked that these equations accomplished with the
external validation criteria recommended in Ref. [46] to assure
predictive capability, that was to say:
2.3.4. Degree of contribution for a descriptor
In order to find out the relative importance of the j-th descriptor
in the linear model, the regression coefficients (bsj , see Table 1S)
were standardized. The larger was the absolute value of bsj , the
greater was the importance of such descriptor [50].
2.3.5. Discussion of QSAR results
The most representative structural features of the training set
of 30 naphthoquinones were identified in the panel of four hu-
man solid tumor cell lines, by applying the RM technique for
searching the best 1e4 descriptor linear regressions. We followed
the common practice of keeping the model’s size as small as
possible, in order to avoid any possible fortuitous correlation.
According to this, we chose structureeactivity relationships
having 3-descriptors with an acceptable predictive power on the
test set. The best MLR found were listed in Table 2S, while a brief
description of the involved descriptors was provided in Table 3S.
It was appreciated from Table 2S that the RMSDtrain parameter
continued improving beyond such number of 3 variables, but
RMSDtest did not improve. The following models were
established:
ꢃ 1 ꢁ R02=R2test < 0:1 or 1 ꢁ R002=Rt2est < 0:1
ꢃ 0.85 ꢄ k ꢄ 1.15 or 0.85 ꢄ k’ ꢄ 1.15
ꢃ R2m > 0:5
The R20, R002, k, k’ and R2m parameters appeared defined in Table 1S
and were presented in Table 4S for each model.
Fig. 2 plotted the predicted antiproliferative pGI50 activities as
function of the experimental values for the training and test sets
(numerical data were provided in Table 5S), showing that there
existed a tendency for the points to have a straight line trend in

180
P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
Fig. 2. Predicted antiproliferative activities as a function of experimental values for A. HBL-100; B. HeLa; C. SW-1573; D. WiDr.
each cell line. The dispersion plots of residuals (i.e. residuals as a
function of predicted activities) in Fig. 1S revealed that residuals
tended to obey a random pattern around the zero line, suggesting
that the assumption of the MLR technique was fulfilled. The cor-
relation matrices of Eqs. (1)e(4) were given in Table 6S, and showed
the absence of very high correlations between descriptors pairs.
The numerical values taken by the molecular descriptors were
included in Table 7S.
The conformation-independent descriptors appearing in Eqs.
(1)e(4) belonged to five different families [51]: i) Topological: ICR,
radial centric information index; ii) 2D-Autocorrelation: GATS5e,
Geary autocorrelation of lag 5 weighted by Sanderson electroneg-
ativity; ATS6v, Broto-Moreau autocorrelation of lag 6 (log function)
weighted by van der Waals volume; JGI6, mean topological charge
index of order 6; iii) BCUT: BELe7, lowest eigenvalue 7 of Burden
matrix weighted by Sanderson electronegativity; BELp7, lowest
eigenvalue n. 7 of Burden matrix weighted by atomic polarizability;
iv) Edge Adjacency Index: EEig11x, eigenvalue 11 from Edge Adja-
cency matrix weighted by edge degree; EEig11r, eigenvalue 11 from
Edge Adjacency matrix weighted by resonance integral; v) Atom
Centred Fragment: O-060, number of AleOeAr/AreOeAr/R.O.R/
ReOeC ¼ X groups. In addition, none of these descriptors were of
the TAE type.
GATS5e (0.69) when compared to ICR (0.36) and BELe7 (0.28). This
meant that GATS5e contributed more than the rest of the variables
to the numerical variation of the activity in the training set, and this
was the main reason why this specific descriptor was selected
during the application of the RM approach.
As every molecular descriptor in Eqs. (1)e(4) took positive nu-
merical values (Table 7S), it was concluded that positive regression
coefficients made the associated descriptors to have a positive
contribution on the predicted potencies, while negative coefficients
made the descriptors to have a negative contribution on the pre-
dictions. Therefore, the higher were the positive contributions in
the equations (also considering the sign of the constant), then the
greater was the tendency on the predicted pGI50 activity to have a
high value in a naphthoquinone derivative. Of course, as the
established models were statistical in nature, a highly negative
predicted activity meant for instance that it was quite probable for
the compound to display a poor potency.
Table 2
Relative importance of the molecular descriptors of Eqs. (1)e(4). The absolute
standardized regression coefficients are given in parentheses.
Eq. (1)
Eq. (2)
Eq. (3)
Eq. (4)
ICR (0.36)
GATS5e (0.69)
JGI6 (0.46)
EEig11r (2.09)
ATS6v (0.80)
BELe7 (0.28)
O-060 (0.58)
BELp7 (0.88)
BELe7 (0.63)
GATS5e (0.96)
EEig11x (1.27)
ICR (0.52)
The relative importance of the molecular descriptors involved in
Eqs. (1)e(4) was presented in Table 2. For instance, in Eq. (1) the
absolute standardized regression coefficient had a greater value for

P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
181
The analysis of the applicability domains of Eqs. (1)e(4)
revealed that all the compounds included in the test sets of each
cell line belonged to the AD of the models (hi < h*). Thus, the
predicted anticancer activities for such naphthoquinones were
considered as reliable. The leverage values were provided in
Table 8S.
Finally, an alternative validation of the established QSAR was
performed. It was corroborated what happened when these
structureeactivity relationships were exposed to independent
data, e.g. put the antiproliferative activity of HBL-100 into the SW-
1573 model and vice versa. The results found were provided in
Table 3. According to the Strain and Stest parameters, it was seen that
Eqs. (1)e(4) were specific to the cell line for which they were
calibrated, as the statistical parameters deteriorated the models
when applied on other data. Therefore, the herein established QSAR
were able to discriminate in the tumor cell panel.
on a Nicolet Magna 550 FT-IR spectrophotometer, values were
given in cmꢁ1. 1H and 13C NMR spectra were recorded on a Bruker
Avance II 500 spectrometer at 500.13 and 125.77 MHz respectively,
in deuterochloroform unless indicated otherwise. Chemical shifts
were given in ppm downfield from TMS as internal standard, J
values are given in Hz. Multiplicity determinations and 2D spectra
(COSY, HSQC and HMBC) were obtained using standard Bruker
software. High-resolution mass spectra (HRMS) were measured on
an Agilent LCTOF, high resolution TOF analyzer or a Bruker
micrOTOF-Q II spectrometer with ESI ionization in positive mode.
Reactions were monitored using thin-layer chromatography (TLC)
on aluminum backed precoated Silica Gel 60 F254 plates (E Merck).
In general, naphthoquinones were highly colored and were visible
on a TLC plate; colorless compounds were detected using UV light.
Flash chromatography was carried out using silica gel 60 (230e400
mesh) with the solvent system indicated in the individual pro-
cedures. All solvent ratios were quoted as v/v. Solvents were
evaporated at reduced pressure and ca. 45 ꢀC.
3. Conclusions
4.2. Chemistry
The search of new antitumor agents had been considered of
main interest in the Medicinal Chemistry field. In this work, the
relationships between the chemical structures of new hydrox-
ynaphthoquinone derivatives to their antiproliferative potencies
were investigated. The established linear models were specific to
the each cancer cell line and had an acceptable predictive perfor-
mance on the test sets, and were able to fulfill other necessary
mathematical conditions, such as Cross-Validation, Y-Randomiza-
tion and Applicability Domain analysis. Finally, the contributions of
molecular descriptors were discussed, in order to elucidate the
Lapachol (1) was extracted from the powdered wood of Tabebuia
impetiginosa (Bignoniaceae) and was purified by a series of re-
crystallizations [52]. From this quinone,
a
-lapachone (3,4-dihydro-
2,2-dimethyl-2H-benzo-[g]chromene-5,10-dione, 20) and
b-lapa-
chone (3,4-dihydro-2,2-dimethyl-2H-benzo[h]chromene-5,6-dione,
2) were obtained by intramolecular cyclization using concentrated
hydrochloric acid and concentrated sulfuric acid, respectively [53].
Compounds 14 and 32 were obtained from 2-hydroxy-1,4-
naphthoquinone. In a similar way, 15 and 33 were synthesized from
3-hydroxy-1,4-naphthoquinone. Treatment of 5-hydroxylapachol
(14) with dimethyl sulfate lead to compounds 17 and 18. The cycli-
zation of 14 with CAN gave naphthoquinones 34 and 35. The addition
of iodine to compounds 14, 32, and 33 gave the iodinated naph-
thoquinones 10e13, 31 and 36 as described previously.
5-Methoxylapachol (16) and 7-methoxylapachol (19) were
prepared from commercially available 5-methoxytetralone and
7-methoxytetralone (SigmaeAldrich), respectively. The pyr-
anonaphthoquinones 21/25, 3/23, 22/26 and 28/24 were obtained
by cyclization under acidic conditions of lapachol derivatives 14, 15,
16 and 19, respectively. Quinones 27, 29 and 30 were obtained from
compounds 3 and 28 as previously described by us.
structural characteristics for
a naphthoquinone derivative to
exhibit a favorable potency. For instance, when the molecular
structures leaded to high positive contribution of descriptors in the
QSAR equations, then the greater was the tendency on the pre-
dicted activity to have a high value in a naphthoquinone derivative.
The use of the conformational representation of the chemical
structure was avoided, and thus no-experimental information on X-
ray crystal structure of the drug interaction was required. From the
computational point of view, the exclusion of 3D structural aspects
also avoided problems associated with ambiguities that resulted
from an incorrect geometry optimization due to the existence of
compounds in various conformational states. Such kind of prob-
lems would also have leaded to the loose of predictive capability of
the QSAR. Some of us had been working in the use of new methods
based on constitutional or topological approximations to QSAR
studies and new results would be published shortly elsewhere.
4.2.1. 9-Hydroxy-b-dunnione (4)
To a stirred solution of 5-hydroxydunniol (32, 25.5 mg,
0.1 mmol) in CH2Cl2 at ꢁ15 ꢀC was added methanesulfonic acid
(0.4 mL). After 3 min the reaction was quenched with water and
extracted with CH2Cl2. The organic layer was dried over anhydrous
Na2SO4, filtered, and evaporated under vacuum. Compound 4 was
obtained as red crystals (25.5 mg, 100%); mp 144e146 ꢀC; IR (KBr)
4. Experimental section
4.1. General
3450, 2956, 2929, 2870, 1693, 1644, 1602, 1466, 1303, 1259 cmꢁ1
;
1H NMR (500.13 MHz, CDCl3): 7.87 (1H, s, OH), 7.62 (1H, dd, J ¼ 7.5,
1.1 Hz, H-6), 7.37 (1H, dd, J ¼ 8.3, 7.6 Hz, H-7), 7.11 (1H, dd, J ¼ 8.4,
1.1 Hz, H-8), 4.72 (1H, q, J ¼ 6.7 Hz, H-2), 1.46 (3H, d, J ¼ 6.7 Hz,
CH3-10), 1.39 (3H, s, CH3-11), 1.21 (3H, s, CH3-12); 13C NMR
(125.77 MHz, CDCl3): 181.0 (C-5), 175.4 (C-4), 167.7 (C-9b), 155.0
(C-9), 133.3 (C-7), 131.4 (C-5a), 124.6 (C-8), 123.3 (C-6), 122.6
(C-3a), 111.2 (C-9a), 94.6 (C-2), 43.2 (H-3), 25.5 (CH3-11), 20.3
(CH3-12), 14.7 (CH3-10); HRMS found: 259.0975 (M þ H)þ calcd for
Melting points were taken on a Fisher-Johns apparatus and are
uncorrected. IR spectra were recorded in thin films using KBr disks
Table 3
Comparison of specificity of developed QSAR models on different human cancer cell
lines. The selected results are shown in bold.
Eq. 1
Eq. 2
Eq. 3
Eq. 4
HBL-100
HeLa
Strain ¼ 0.44
Stest ¼ 0.58
Strain ¼ 0.42
Stest ¼ 0.49
Strain ¼ 0.49
Stest ¼ 0.88
Strain ¼ 0.48
Stest ¼ 0.53
Strain ¼ 0.53
Stest ¼ 0.83
Strain ¼ 0.38
Stest ¼ 0.46
Strain ¼ 0.54
Stest ¼ 1.30
Strain ¼ 0.53
Stest ¼ 0.74
Strain ¼ 0.45
Stest ¼ 0.87
Strain ¼ 0.44
Stest ¼ 1.27
Strain ¼ 0.42
Stest ¼ 0.45
Strain ¼ 0.42
Stest ¼ 0.38
Strain ¼ 0.54
Stest ¼ 1.21
Strain ¼ 0.50
Stest ¼ 0.65
Strain ¼ 0.61
Stest ¼ 0.57
Strain ¼ 0.32
Stest ¼ 0.23
C
15H15O4: 259.0965.
4.2.2. Hydroxy- -dunnione (37)
a
SW-1573
WiDr
A suspension of 4 (43 mg, 0.17 mmol) in 20% HCl (2.7 mL) was
heated to 85 ꢀC overnight. The reaction was quenched with water
and extracted with CH2Cl2. The organic layer was dried

182
P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
over anhydrous Na2SO4, filtered, and evaporated under vacuum.
Compound 37 was obtained as orange crystals (43 mg, 100%); mp
105e106 ꢀC; IR (KBr) 2963, 2923, 2850, 1680, 1633, 1602 cmꢁ1; 1H
NMR (500.13 MHz, CDCl3): 12.40 (1H, d, J ¼ 0.4 Hz, OH), 7.61 (1H,
dd, J ¼ 7.4, 1.2 Hz, H-8), 7.51 (1H, ddd, J ¼ 7.9, 7.5, 0.5 Hz, H-7), 7.23
(1H, dd, J ¼ 8.4, 1.2 Hz, H-6), 4.61 (1H, q, J ¼ 6.7 Hz, H-2), 1.49 (3H, s,
H-11), 1.45 (3H, d, J ¼ 6.7 Hz, H-10), 1.30 (3H, s, H-12); 13C
NMR(125.77 MHz, CDCl3): 188.7 (C-4), 177.9 (C-9), 161.3 (C-5), 159.4
(C-9a), 134.9 (C-7), 131.6 (C-8a), 130.5 (C-3a), 125.8 (C-6), 119.2 (C-
8), 115.1 (C-4a), 92.1 (C-2), 45.0 (C-3), 25.8 (CH3-11), 20.6 (CH3-12),
14.2 (CH3-10); HRMS found: 259.0972 (M þ H)þ calcd for C15H15O4:
259.0965.
4.2.6. 6-Hydroxy-b-isodunnione (7)
A solution of 38 (6.0 mg, 0.02 mmol) in concentrated H2SO4
(0.6 mL) was stirred at room temperature overnight. The reaction
was cooled to 0 ꢀC, quenched with water and extracted with CH2Cl2.
The organic layer was dried over anhydrous Na2SO4, filtered, and
evaporated under vacuum. The residue was purified by column
chromatography on silica gel (hexane/EtOAc, gradient) to give 7 as
red crystals (1.9 mg, 32%) and 38 (3.3 mg, 55%) was recovered
unreacted; mp 117e118 ꢀC (lit. [55] 116e117 ꢀC); IR (KBr): 2918,
2845, 1735, 1648, 1615, 1592 cmꢁ1 1H NMR (500.13 MHz, CDCl3):
;
11.94 (1H, s, OH), 7.52 (1H, dd, J ¼ 8.5, 7.4 Hz, H-8), 7.18 (1H, dd,
J ¼ 7.3, 1.0 Hz, H-9), 7.11 (1H, dd, J ¼ 8.6, 1.0 Hz, H-7), 3.20 (1H, q,
J ¼ 7.1 Hz, H-3), 1.51 (3H, s, CH3-10), 1.49 (3H, s, CH3-11), 1.27 (3H, d,
J ¼ 7.1 Hz, CH3-12); 13C NMR (125.77 MHz, CDCl3): 185.7 (C-5),175.2
(C-4), 167.3 (C-9b), 164.3 (C-6), 137.5 (C-8), 127.9 (C-9a), 123.0 (C-7),
119.9 (C-3a), 117.5 (C-9), 113.6 (C-5a), 95.4 (C-2), 43.5 (C-3), 28.9
(CH3-10), 22.4 (CH3-11), 14.1 (CH3-12); HRMS found: 281.0794
(M þ Na)þ calcd for C15H14NaO4: 281.0784.
4.2.3. 9-Hydroxy-b-isodunnione (6)
To a stirred solution of 37 (43 mg, 0.17 mmol) in CH2Cl2
at ꢁ15 ꢀC was added concentrated H2SO4 (0.8 mL) and stirring
continued at room temperature overnight. The reaction mixture
was cooled to 0 ꢀC, diluted with water and extracted with CH2Cl2.
The organic layer was dried over anhydrous Na2SO4, filtered, and
evaporated under vacuum. The residue was purified by column
chromatography on silica gel (hexane/EtOAc, gradient) to give 6 as
dark red crystals (6.5 mg, 15%) and recovered 37 (32 mg, 74%); mp
4.2.7. 5-Hydroxylomatiol (39)
A suspension of 5-hydroxylapachol (14, 50 mg, 0.2 mmol) and
zinc dust (100 mg) in acetic anhydride (1 mL) and triethylamine
75e76 ꢀC; IR (KBr) 3440, 2973, 1642, 1604, 1257 cmꢁ1 1H NMR
;
(10 mL) was stirred at room temperature for 1.5 h. The solution was
(500.13 MHz, CDCl3): 7.87 (1H, d, J ¼ 0.5 Hz, OH), 7.64 (1H, dd,
J ¼ 7.5, 1.1 Hz, H-6), 7.38 (1H, ddd, J ¼ 8.4, 7.5, 0.3 Hz, H-7), 7.11 (1H,
dd, J ¼ 8.5, 1.1 Hz, H-8), 3.12 (1H, q, J ¼ 7.1 Hz, H-3), 1.53 (3H, s, CH3-
10), 1.50 (3H, s, CH3-11), 1.22 (3H, d, J ¼ 7.1 Hz, CH3-12);13C NMR
(125.77 MHz, CDCl3): 181.0 (C-5), 175.7 (C-4), 167.2 (C-9b), 155.1 (C-
9), 133.4 (C-7), 131.4 (C-5a), 124.6 (C-8), 123.3 (C-6), 119.4 (C-3a),
111.3 (C-9a), 98.4 (C-2), 42.5 (C-3), 29.0 (C-10), 22.6 (C-11), 14.1 (C-
12); HRMS found: 281.0786 (M þ Na)þ calcd for C15H14NaO4:
281.0784.
filtered and the filtrate was diluted with water and extracted with
CH2Cl2. The organic layer was dried over anhydrous Na2SO4,
filtered, and evaporated under vacuum to give a pale yellow oil
(47.6 mg). This oil was dissolved in a mixture of acetic anhydride
(2.9 mL) and glacial acetic acid (1.8 mL) and selenium dioxide
(9 mg) added. The mixture was heated under reflux for 1 h, cooled
and diluted with water. The resulting oil was extracted with
diethyl ether, washed successively with water and 2% sodium
hydroxide, dried over anhydrous Na2SO4, filtered and concen-
trated. The residual viscous oil (82.3 mg) was dissolved in meth-
anol (3.2 mL) and treated with 6 M KOH (3.2 mL), and the solution
was heated under reflux for 15 min. After cooling the mixture was
carefully acidified with HCl 5% and extracted with diethyl ether.
The organic layer was washed with water, dried over anhydrous
Na2SO4, filtered, and evaporated under vacuum. The residue was
purified by column chromatography on silica gel (hexane/EtOAc,
gradient) to give 39 as yellow crystals (10.1 mg, 19%); mp 166e
167 ꢀC; IR (KBr): 3339, 3224, 2956, 2926, 2850, 1735, 1645, 1620,
4.2.4. 6-Hydroxy-b-dunnione (5)
Compound 5 (14.6 mg, 100%) was obtained as red crystals from
compound 33 (14.6 mg, 0.06 mmol) and methanesulfonic acid
(0.8 mL) following the procedure described for compound 4; mp
151e152 ꢀC (lit [54]. 151e152 ꢀC); IR (KBr): 2958, 2916, 2848, 1733,
1640, 1615, 1592 cmꢁ1 1H NMR (500.13 MHz, CDCl3): 11.93 (1H, s,
;
OH), 7.53 (1H, dd, J ¼ 8.6, 7.3 Hz, H-8), 7.19 (1H, dd, J ¼ 7.3, 1.0 Hz, H-
9), 7.10 (1H, dd, J ¼ 8.6, 1.0 Hz, H-7), 4.65 (1H, q, J ¼ 6.7 Hz, H-2), 1.46
(3H, d, J ¼ 6.7 Hz, CH3-10), 1.44 (3H, s, CH3-11), 1.26 (3H, s, CH3-12);
13C NMR (125.77 MHz, CDCl3): 185.5 (C-5), 174.9 (C-4), 167.7 (C-9b),
164.3 (C-6), 137.5 (C-8), 127.7 (C-9a), 123.3 (C-3a), 122.9 (C-7), 117.4
(C-9), 113.5 (C-5a), 92.8 (C-2), 44.1 (C-3), 25.7 (CH3-11), 20.3 (CH3-
12), 14.5 (CH3-10); HRMS found: 259.0972 (M þ H)þ calcd for
1472, 1467 cmꢁ1 1H NMR (500.13 MHz, CDCl3): 12.48 (1H, s, OH-
;
5), 7.61 (1H, dd, J ¼ 7.4, 1.0 Hz, H-8), 7.54 (1H, t, J ¼ 7.9 Hz, H-7),
7.27 (1H, dd, J ¼ 8.3, 0.9 Hz, H-6), 5.48 (1H, dt, J ¼ 7.3, 1.2 Hz, H-20),
3.97 (2H, s, H-40), 3.34 (2H, d, J ¼ 7.3 Hz, H-10), 1.83 (3H, s, CH3); 13C
NMR(125.77 MHz, CDCl3): 190.6 (C-4), 180.9 (C-1), 160.8 (C-5),
154.4 (C-2), 136.7 (C-30), 134.8 (C-7), 129.7 (C-8a), 125.7 (C-6), 122.6
(C-3), 120.7 (C-20), 119.0 (C-8), 114.3 (C-4a), 68.1 (C-40), 21.4 (C-10),
C
15H15O4: 259.0965.
13.6 (C-50); HRMS found: 297.0742 (M
15H14NaO5: 297.0733.
þ
Na)þ calcd for
4.2.5. 8-Hydroxy- -dunnione (38)
a
C
A solution of 5 (17.1 mg, 0.07 mmol) in concentrated HCl
(3.2 mL) was heated at 85 ꢀC overnight. The reaction was quenched
with water and extracted with CH2Cl2. The organic layer was dried
over anhydrous Na2SO4, filtered, and evaporated under vacuum.
The residue was purified by column chromatography on silica gel
(hexane/EtOAc, gradient) to give 38 as yellow crystals (8.5 mg, 50%)
and 5 (6.7 mg, 39%) was recovered unreacted; mp 117e118 ꢀC; IR
(KBr): 2964, 2926, 1645, 1609 cmꢁ1; 1H NMR (500.13 MHz, CDCl3):
11.66 (1H, s, OH), 7.58 (2H, m, H-5 y 6), 7.19 (1H, m, H-7), 4.60 (1H, q,
J ¼ 6.6 Hz, H-2), 1.47 (3H, s, CH3-11), 1.45 (3H, d, J ¼ 6.7 Hz, CH3-10),
1.28 (3H, s, CH3-12); 13C NMR(125.77 MHz, CDCl3): 183.3 (C-9),
181.5 (C-4), 161.8 (C-8), 158.4 (C-9a), 136.9 (C-6), 133.7 (C-4a), 131.8
(C-3a), 123.8 (C-7), 118.9 (C-5), 114.6 (C-8a), 91.9 (C-2), 45.2 (C-3),
25.7 (CH3-11), 20.5 (CH3-12), 14.2 (CH3-10); HRMS found: 259.0972
(M þ H)þ calcd for C15H15O4: 259.0965.
4.2.8. 8-Hydroxylomatiol (40)
Compound 40 (21.4 mg, 49%) was obtained as yellow crystals
from 8-hydroxylapachol (15, 42 mg, 0.16 mmol) following the
procedure described for compound 39; mp 104e105 ꢀC; IR (KBr):
3386, 2920, 2848, 1714, 1626, 1452 cmꢁ1 1H NMR (500.13 MHz,
;
CDCl3): 11.09 (1H, s, OH-8), 7.64 (2H, m, H-5 and H-6), 7.20 (1H, dd,
J ¼ 7.7, 1.9 Hz, H-7), 5.48 (1H, tsep, J ¼ 7.2, 1.5 Hz, H-20), 4.00 (2H, s,
H-40), 3.36 (2H, dd, J ¼ 7.0, 0.4 Hz, H-10), 1.84 (3H, s, CH3); 13C
NMR(125.77 MHz, CDCl3): 184.8 (C-1),183.6 (C-4),161.2 (C-8),152.6
(C-2), 137.7 (C-5 o 6), 137.0 (C-30), 132.6 (C-4a), 123.9 (C-3), 123.2 (C-
7), 120.7 (C-20), 119.7 (C-5 o 6), 112.9 (C-8a), 68.6 (C-40), 22.2 (C-10),
13.8 (C-50); HRMS found: 297.0738 (M þ Na)þ calcd for C15H14NaO5:
297.0733.

P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
183
4.2.9. 9-Hydroxy-dehydro-iso-
b-lapachone (8)
4.4. Biology
5-Hydroxylomatiol (39, 6.2 mg, 0.02 mmol) was treated at 0 ꢀ
C
with a mixture of H2SO4eH2O (5:3) (1 mL). The mixture was stirred
until the 5-hydroxylomatiol was completely dissolved and the so-
lution was then immediately poured into water and extracted with
CH2Cl2. The organic layer was dried over anhydrous Na2SO4,
filtered, and evaporated under vacuum. The residue was purified by
column chromatography on silica gel (hexane/EtOAc, gradient) to
give 8 as violet crystals (2.6 mg, 50%); mp 120e121 ꢀC; IR (KBr)
All starting materials were commercially available research-
grade chemicals and used without further purification. RPMI
1640 medium was purchased from Flow Laboratories (Irvine, UK),
fetal calf serum (FCS) was from Gibco (Grand Island, NY), tri-
chloroacetic acid (TCA) and glutamine were from Merck (Darm-
stadt, Germany), and penicillin G, streptomycin, DMSO and
sulforhodamine B (SRB) were from Sigma (St Louis, MO).
3157, 2962, 2921, 1628, 1594, 1543, 1260 cmꢁ1
;
1H NMR
(500.13 MHz, CDCl3): 7.88 (1H, s, OH), 7.74 (1H, dd, J ¼ 7.5, 1.1 Hz, H-
6), 7.49 (1H, dd, J ¼ 8.3, 7.5 Hz, H-7), 7.21 (1H, dd, J ¼ 8.5, 1.1 Hz, H-
8), 5.62 (1H, dd, J ¼ 10.2, 7.8 Hz, H-2), 5.17 (1H, d, J ¼ 0.9 Hz, H-11a),
5.09 (1H, dd, J ¼ 1.3, 1.1 Hz, H-11b), 3.28 (1H, dd, J ¼ 15.4, 10.4 Hz, H-
3a), 2.95 (1H, dd, J ¼ 15.4, 7.8 Hz, H-3b), 1.83 (3H, dd,J ¼ 1.3, 0.9 Hz,
CH3); 13C NMR(125.77 MHz, CDCl3): 180.6 (C-5), 175.3 (C-4), 169.0
(C-9b), 154.9 (C-9), 141.0 (C-10), 133.7 (C-7), 131.3 (C-5a), 124.8 (C-
8), 123.7 (C-6), 115.1 (C-11), 114.5 (C-3a), 110.9 (C-9a), 92.0 (C-2),
30.0 (C-3), 16.7 (CH3); HRMS found: 257.0800 (M þ H)þ calcd for
4.4.1. Cells, culture and plating
The human solid tumor cell lines HBL-100, HeLa, SW1573, and
WiDr were used in this study. These cell lines were a kind gift from
Prof. G.J. Peters (VU Medical Center, Amsterdam, The Netherlands).
Cells were maintained in 25 cm2 culture flasks in RPMI 1640 sup-
plemented with 5% heat inactivated fetal calf serum and 2 mM L-
glutamine in a 37 ꢀC, 5% CO2, 95% humidified air incubator. Expo-
nentially growing cells were trypsinized and resuspended in anti-
biotic containing medium (100 units penicillin G and 0.1 mg of
streptomycin per mL). Single cell suspensions displaying >97%
viability by trypan blue dye exclusion were subsequently counted.
After counting, dilutions were made to give the appropriate cell
densities for inoculation onto 96-well microtiter plates. Cells were
C
15H13O4: 257.0808.
4.2.10. 6-Hydroxy-dehydro-iso-b-lapachone (9)
Compound 9 (11 mg, 54%) was obtained as red crystals from 8-
hydroxylomatiol (40, 21.4 mg, 0.08 mmol) following the procedure
described for compound 8; mp 150 ꢀC; IR (KBr) 2930, 1645, 1615,
1581 cmꢁ1; 1H NMR (500.13 MHz, CDCl3): 11.94 (1H,d, J ¼ 0.3, OH),
7.55 (1H, dd, J ¼ 8.4, 7.3 Hz, H-8), 7.24 (1H, dd, J ¼ 7.3, 1.0 Hz, H-9),
7.14 (1H, dd, J ¼ 8.7, 1.0 Hz, H-7), 5.47 (1H, dd, J ¼ 10.5, 7.8 Hz, H-2),
5.13 (1H, d, J ¼ 0.9 Hz, H-11a), 5.03 (1H, t, J ¼ 1.3 Hz, H-11b), 3.29
(1H, dd, J ¼ 15.6, 10.5 Hz, H-3a), 2.95 (1H, dd, J ¼ 15.6, 7.8 Hz, H-3b),
1.80 (3H, t, J ¼ 1.0 Hz, CH3);13C NMR(125.77 MHz, CDCl3): 185.2 (C-
5), 174.9 (C-4), 169.2 (C-9b), 164.5 (C-6), 141.8 (C-10), 137.7 (C-8),
127.1 (C-9a), 123.4 (C-7), 117.5 (C-9), 115.2 (C-3a), 113.9 (C-11), 113.4
(C-5a), 89.8 (C-2), 31.1 (C-3), 16.8 (CH3); HRMS found: 257.0809
(M þ H)þ calcd. for C15H13O4: 257.0808.
inoculated in a volume of 100
(SW1573 and HBL-100) of 15,000 (HeLa), and 20,000 (WiDr) cells
per well, based on their doubling times.
mL per well at densities 10,000
Acknowledgments
We thank the financial support provided by the National
Research Council of Argentina (CONICET) PIP11220100100151 and
PIP11220100100447 projects and to Ministerio de Ciencia, Tecno-
logía e Innovación Productiva for the electronic library facilities. Co-
financed by the EU Research Potential (FP7-REGPOT-2012-CT2012-
31637-IMBRAIN), the European Regional Development Fund
(FEDER), and the Spanish Instituto de Salud Carlos III (PI11/00840).
4.3. Molecular descriptors calculation
Appendix A. Supplementary data
The compounds were first drawn with HyperChem for Windows
[56], and then the .hin files were converted into .sdf by using the
Open Babel 2.3.1 software [57]. A set of 931 conformation-
independent molecular descriptors were computed using E-
Dragon [58]. This useful and well-known descriptors database
included thirteen descriptor families: Constitutional, Topological,
Walk and Path Count, Connectivity Index, Information Index, Edge
Adjacency Index, Topological Charge Index, Burden Eigenvalues,
Eigenvalue-Based Index, 2D-Autocorrelation, Functional Group
Count, Atom-Centred Fragment, and Molecular Property [51].
In addition, atomic charge density-based descriptors were ob-
tained, encoding electronic and structural information relevant to
the chemistry of intermolecular interactions, by means of Recon 5.5
[59]. This sort of descriptors were not provided by E-Dragon, while
the robustness of Recon had previously been demonstrated else-
where [60,61]. Recon was an algorithm for the reconstruction of
molecular charge densities and charge density-based electronic
properties of molecules, using atomic charge density fragments
precomputed from ab initio wavefunctions. The method was based
on the Quantum Theory of Atoms in Molecules [62]. A library of
atomic charge density fragments had been built in a form that al-
lows for the rapid retrieval of the fragments and molecular as-
sembly. The .sdf molecular format was employed as input for the
generation of 248 Transferable Atom Equivalent (TAE) descriptors
developed by Breneman et al. [63]. In this way, the total number of
calculated molecular descriptors was of 1179 variables.
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.ejmech.2014.02.057.
References
[1] G. Powis, Free radical formation by antitumor quinones, Free Radical Biology &
Medicine 6 (1989) 63e101.
[2] C. Asche, Antitumour quinones, Mini-Reviews in Medicinal Chemistry
5
(2005) 449e467.
[3] L. Constantino, D. Barlocco, Privileged structures as leads in medicinal
chemistry, Current Medicinal Chemistry 13 (2006) 65e85.
[4] J. Li, K.K.-C. Liu, S. Sakya, Synthetic approaches to the 2004 new drugs, Mini-
Reviews in Medicinal Chemistry 5 (2005) 1133e1144.
[5] R.W. De Simone, K.S. Currie, S.A. Mitchell, J.W. Darrow, D.A. Pippin, A, Privi-
leged structures: applications in drug discovery, Combinatorial Chemistry &
High Throughput Screening 7 (2004) 473e494.
[6] K.O. Eyong, G.N. Folefoc, V. Kuete, V.P. Beng, K. Krohn, H. Hussain,
A.E. Nkengfack, M. Saeftel, S.R. Sarite, A. Hoerauf, Newbouldiaquinone A: a
naphthoquinone-anthraquinone ether coupled pigment, as a potential anti-
microbial and antimalarial agent from Newbouldia laevis, Phytochemistry 67
(2006) 605e609.
[7] V. Kuete, K.O. Eyong, G.N. Folefoc, V.P. Beng, H. Hussain, K. Krohn,
A.E. Nkengfack, Antimicrobial activity of the methanolic extract and of the
chemical constituents isolated from Newbouldia laevis, Die Pharmazie 62
(2007) 552e556.
[8] J. Breger, B.B. Fuchs, G. Aperis, T.I. Moy, F.M. Ausubel, E. Mylonakis, Antifungal
chemical compounds identified using a C. elegans pathogenicity assay, PLoS
Pathogens 3 (2007) 168e178.
[9] A. Portillo, R. Vila, B. Freixa, T. Adzet, S. Cañigueral, Antifungal activity of
Paraguayan plants used in traditional medicine, Journal of Ethno-
pharmacology 76 (2001) 93e98.

184
P.R. Duchowicz et al. / European Journal of Medicinal Chemistry 77 (2014) 176e184
[10] A.V. Pinto, S.L. De Castro, The trypanocidal activity of naphthoquinones: a
review, Molecules 14 (2009) 4570e4590.
[11] H. Hussain, K. Krohn, V.U. Ahmad, G.A. Miana, I.R. Green, Lapachol: an over-
view, Arkivoc 2 (2007) 145e171.
[36] P.R. Duchowicz, E.A. Castro, F.M. Fernández, Alternative algorithm for the
search of an optimal set of descriptors in QSAR-QSPR studies, MATCH: Com-
munications in Mathematical and in Computer Chemistry 55 (2006) 179e192.
[37] P.R. Duchowicz, E.A. Castro, F.M. Fernández, M.P. González, A new search al-
gorithm of QSPR/QSAR theories: normal boiling points of some organic mol-
ecules, Chemical Physics Letters 412 (2005) 376e380.
[38] P.R. Duchowicz, A. Talevi, L.E. Bruno-Blanch, E.A. Castro, New QSPR study for
the prediction of aqueous solubility of drug-like compounds, Bioorganic &
Medicinal Chemistry Letters 16 (2008) 7944e7955.
[39] M. Goodarzi, P.R. Duchowicz, C.H. Wu, F.M. Fernández, E.A. Castro, New hybrid
genetic based support VECTOR regression as QSAR approach for analyzing
flavonoids-GABA(A) complexes, Journal of Chemical Information and
Modeling 49 (2009) 1475e1485.
[12] X. Cheng, F. Liu, T. Yan, X. Zhou, L. Wu, K. Liao, G. Wang, Metabolic profile,
enzyme kinetics, and reaction phenotyping of
man liver and intestine in vitro, Molecular Pharmaceutics 9 (2012) 3476e3485.
[13] R.J. Boorstein, A.B. Pardee, -Lapachone greatly enhances MMS lethality to
b-lapachone metabolism in hu-
b
human fibroblasts, Biochemical and Biophysical Research Communications
118 (1984) 828e834.
[14] P.S. Guin, S. Das, P.C. Mandal, Electrochemical reduction of quinones in
different media: a review, International Journal of Electrochemical Science
2011 (2011) 1e22.
[15] H.-X. Chang, T.-C. Chou, N. Savaraj, L.F. Liu, C. Yu, C.C. Cheng, Design of anti-
neoplastic agents based on the “2-phenylnaphthalene-type” structural
pattern. 4. Synthesis and biological activity of 2-chloro-3-(substituted phe-
noxy)-1,4-naphthoquinones and related 5,8-dihydroxy-1,4-naphthoquinones,
Journal of Medicinal Chemistry 42 (1999) 405e408.
[16] K.-I. Hayashi, F.-R. Chang, Y. Nakanishi, K.F. Bastow, G. Cragg, A.T. Mc Phail,
H. Nozaki, K.-H. Lee, Antitumor Agents. 233 lantalucratins A-F, new cytotoxic
naphthoquinones from Lantana involucrata, Journal of Natural Products 67
(2004) 990e993.
[40] A.B. Pomilio, M.A. Giraudo, P.R. Duchowicz, E.A. Castro, QSPR analyses for
aminograms in food: citrus juices and concentrates, Food Chemistry 123
(2010) 917e927.
[41] A. Talevi, M. Goodarzi, E.V. Ortiz, P.R. Duchowicz, C.L. Bellera, G. Pesce,
E.A. Castro, L.E. Bruno-Blanch, Prediction of drug intestinal absorption by new
linear and non-linear QSPR, European Journal of Medicinal Chemistry 46
(2011) 218e228.
[42] G. Pasquale, G.P. Romanelli, J.C. Autino, J. García, E.V. Ortiz, P.R. Duchowicz,
Quantitative structure-activity relationships on chalcone derivatives: mos-
quito larvicidal studies, Journal of Agricultural and Food Chemistry 60 (2012)
692e697.
[17] R.M. Khan, S.M. Mlungwana, 5-Hydroxylapachol:
a cytotoxic agent from
Tectona grandis, Phythochemistry 50 (1999) 439e442.
[18] M. Yamashita, M. Kaneko, H. Tokuda, K. Nishimura, Y. Kumeda, A. Iida, Syn-
thesis and evaluation of bioactive naphthoquinones from the Brazilian me-
dicinal plant, Tabebuia avellanedae, Bioorganic & Medicinal Chemistry 17
(2009) 6286e6291.
[19] E.L. Bonifazi, C. Rios-Luci, L.G. León, G. Burton, J.M. Padrón, R.I. Misico, Anti-
proliferative activity of synthetic naphthoquinones related to lapachol. First
synthesis of 5-hydroxylapachol, Bioorganic & Medicinal Chemistry 18 (2010)
2621e2630.
[43] A.G. Mercader, P.R. Duchowicz, F.M. Fernández, E.A. Castro, Replacement
method and enhanced replacement method versus the genetic algorithm
approach for the selection of molecular descriptors in QSPR/QSAR theories,
Journal of Chemical Information and Modeling 50 (2010) 1542e1548.
[44] A.G. Mercader, P.R. Duchowicz, F.M. Fernández, E.A. Castro, Advances in the
replacement and enhanced replacement method in QSAR and QSPR theories,
Journal of Chemical Information and Modeling 51 (2011) 1575e1581.
[45] Matlab, Matlab 7.0, The MathWorks, Inc., Masachussetts, USA, http://www.
mathworks.com.
[20] C. Rios-Luci, E.L. Bonifazi, L.G. León, J.C. Montero, G. Burton, A. Pandiella,
R.I. Misico, J.M. Padrón,
b
-Lapachone analogs with enhanced antiproliferative
[46] A. Golbraikh, A. Tropsha, Beware of q2!, Journal of Molecular Graphics and
Modelling 20 (2002) 269e276.
[47] S. Wold, L. Eriksson, Statistical validation of QSAR results, in: H. van de
Waterbeemd (Ed.), Chemometrics Methods in Molecular Design, VCH,
Weinheim, 1995, pp. 309e318.
[48] P. Gramatica, Principles of QSAR models validation: internal and external,
QSAR & Combinatorial Science 26 (2007) 694e701.
[49] L. Eriksson, J. Jaworska, A.P. Worth, M.T. Cronin, R.M. McDowell, P. Gramatica,
Methods for reliability and uncertainty assessment and for applicability
evaluations of classification- and regression-based QSARs, Environmental
Health Perspectives 111 (2003) 1361e1375.
activity, European Journal of Medicinal Chemistry 53 (2012) 264e274.
[21] C. Hansch, A. Leo, Exploring QSAR, in: Fundamentals and Applications in
Chemistry and Biology, American Chemical Society, Washington, D. C., 1995.
[22] H. Kubinyi, QSAR: Hansch Analysis and Related Approaches, Wiley-Inter-
science, New York, 2008.
[23] T. Puzyn, J. Leszczynski, M.T.D. Cronin (Eds.), Recent Advances in QSAR
Studies: Methods and Applications, Springer Science&Business Media B.V.,
Netherlands, 2010.
[24] S.G. Renou, S.E. Asís, M.I. Abasolo, D.G. Bekerman, A.M. Bruno, Mono-
arylhydrazones of a-lapachone: synthesis, chemical properties and antineo-
plastic activity, Die Pharmazie 58 (2003) 690e695.
[50] N.R. Draper, H. Smith, Applied Regression Analysis, 1981. New York.
[51] R. Todeschini, V. Consonni, Molecular Descriptors for Chemoinformatics
(Methods and Principles in Medicinal Chemistry), Wiley-VCH, Weinheim,
2009.
[25] R.P. Verma, C. Hansch, Elucidation of structure-activity relationships for 2- or
6-substituted-5,8-dimethoxy-1,4-naphthoquinones, Bioorganic & Medicinal
Chemistry 12 (2004) 5997e6009.
[26] R.P. Verma, Understanding topoisomerase I and II in terms of QSAR, Bio-
organic & Medicinal Chemistry 13 (2005) 1059e1067.
[52] A.G. Ravelo, A. Estévez-Braun, E. Pérez-Sacau, The chemistry and biology of
lapachol and related natural products:
a- and b-lapachones, Studies in Natural
[27] B. Kumar, Quantitative structure activity relationship (QSAR) modeling of 2-X-
5,8-dimethoxy-1,4-naphthoquinones against L1210 cells, International Jour-
nal of Pharmacy and Pharmaceutical Sciences 4 (2012) 445e448.
Products Chemistry 29 (2003) 719e760.
[53] S.C. Hooker, The constitution of lapachol and its derivatives. Part V. The
structure of Paternò’s “isolapachone, Journal of the American Chemical Soci-
ety 58 (1936) 1190e1197.
[54] K. Inoue, S. Ueda, H. Nayeshiro, H. Inouye, Structures of unusually prenylated
naphthoquinones of Streptocarpus dunii and its cell cultures, Chemical and
Pharmaceutical Bulletin 30 (1982) 2265e2268.
[28] L. Saíz-Urra, M.P. González, M. Teijeira, 2D-autocorrelation descriptors for pre-
dicting cytotoxicity of naphthoquinone ester derivatives against oral human
epidermoid carcinoma, Bioorganic & Medicinal Chemistry 15 (2007) 3565e3571.
[29] A. Takano, K. Hashimoto, M. Ogawa, J. Koyanagi, T. Kurihara, H. Wakabayashi,
H. Kikuchi, Y. Nakamura, N. Motohashi, H. Sakagami, K. Yamamoto, A. Tanaka,
Tumor-specific cytotoxicity and type of cell death induced by naphtho[2,3-b]
furan-4,9-diones and related compounds in human tumor cell lines: rela-
tionship to electronic structure, Anticancer Research 29 (2009) 455e464.
[30] A. Bargiotti, L. Musso, S. Dallavalle, L. Merlini, G. Gallo, A. Ciacci, G. Giannini,
W. Cabri, S. Penco, L. Vesci, M. Castorina, F.M. Milazzo, M.L. Cervoni,
M. Barbarino, C. Pisano, C. Giommarelli, V. Zuco, M. De Cesare, F. Zunino,
Isoxazolo(aza)naphthoquinones: a new class of cytotoxic Hsp90 inhibitors,
European Journal of Medicinal Chemistry 53 (2012) 64e75.
[55] V. Guay, P. Brassard, Synthesis of (ꢂ)-7- and 8-Hydroxydunnione, Journal of
Natural Products 49 (1986) 122e125.
[56] HyperChem 7, Hypercube, Inc., http://www.hyper.com.
[57] OpenBabel 2.3.1, OpenBabel 2.3.1, http://openbabel.org/wiki/Windows_GUI.
[58] E-Dragon, Milano Chemometrics and QSAR Research Group, VCCLAB, Virtual
Computational Chemistry Laboratory, 2005. http://michem.disat.unimib.it/
chm.
[59] Recon, Recon, Rensselaer Polytechnic Institute, Troy, New York, USA, 2002.
http://www.drugmining.com.
[31] Y. Ma, J.-G. Wang, B. Wang, Z.-M. Li, Integrating molecular docking, DFT and
[60] B.K. Lavine, C.E. Davidson, C. Breneman, W. Katt, Electronic Van der Waals
SURFACE property descriptors and genetic algorithms for developing
structure-activity correlations in olfactory databases, Journal of Chemical In-
formation and Modeling 43 (2003) 1890e1905.
CoMFA/CoMSIA approaches for a series of naphthoquinone fused cyclic
a-
aminophosphonates that act as novel topoisomerase II inhibitors, Journal of
Molecular Modeling 17 (2011) 1899e1909.
[32] E. Pérez-Sacau, R.G. Díaz-Peñate, A. Estévez-Braun, A.G. Ravelo, J.M. García-
Castellano, L. Pardo, M. Campillo, Synthesis and pharmacophore modeling of
naphthoquinone derivatives with cytotoxic activity in human promyelocytic
leukemia HL-60 cell line, Journal of Medicinal Chemistry 50 (2007) 696e706.
[33] J. Verma, V.M. Khedkar, E.C. Coutinho, 3D-QSAR in drug design e a review,
Current Topics in Medicinal Chemistry 10 (2010) 95e115.
[34] P.R. Duchowicz, N.C. Comelli, E.V. Ortiz, E.A. Castro, QSAR study for carcinoge-
nicity in a large set of organic compounds, Current Drug Safety 7 (2012) 282e288.
[35] P.O. Miranda, J.M. Padrón, J.I. Padrón, J. Villar, V.S. Martín, Prins-type synthesis
and SAR study of cytotoxic alkyl chloro dihydropyrans, ChemMedChem 1
(2006) 323e329.
[61] A. Worachartcheewan, C. Nantasenamat, T. Naenna, C. Isarankura-Na-Ayud-
hya, V. Prachayasittikul, Modeling the activity of furin inhibitors using arti-
ficial neural network, European Journal of Medicinal Chemistry 44 (2009)
1664e1673.
[62] R.F.W. Bader, Atoms in Molecules-a Quantum Theory, 1990. U.K.
[63] C.M. Breneman, L.W. Weber, Transferable atom equivalents. Assembling ac-
curate electrostatic potential fields for large molecules from ab initio and
PROAIMS results on model systems, in: G.A. Jeffrey, J.F. Piniella (Eds.), The
Application of Charge Density Research to Chemistry and Drug Design, 1991.
New York.