Redox Polypharmacology as an Emerging Strategy to Combat Malarial Parasites

Article abstract of DOI:10.1002/cmdc.201600009

3-Benzylmenadiones are potent antimalarial agents that are thought to act through their 3-benzoylmenadione metabolites as redox cyclers of two essential targets: the NADPH-dependent glutathione reductases (GRs) of Plasmodium-parasitized erythrocytes and m

Full text of DOI:10.1002/cmdc.201600009

DOI: 10.1002/cmdc.201600009
Full Papers
Redox Polypharmacology as an Emerging Strategy to
Combat Malarial Parasites
Pavel Sidorov,[a, b] Israel Desta,[c, d] Matthieu ChessØ,[c] Dragos Horvath,[a] Gilles Marcou,[a]
Alexandre Varnek,[a] Elisabeth Davioud-Charvet,*[c] and Mourad Elhabiri*[c]
3-Benzylmenadiones are potent antimalarial agents that are
thought to act through their 3-benzoylmenadione metabolites
as redox cyclers of two essential targets: the NADPH-depen-
dent glutathione reductases (GRs) of Plasmodium-parasitized
erythrocytes and methemoglobin. Their physicochemical prop-
erties were characterized in a coupled assay using both targets
and modeled with QSPR predictive tools built in house. The
substitution pattern of the west/east aromatic parts that con-
trols the oxidant character of the electrophore was highlighted
and accurately predicted by QSPR models. The effects centered
on the benz(o)yl chain, induced by drug bioactivation, marked-
ly influenced the oxidant character of the reduced species
through a large anodic shift of the redox potentials that corre-
lated with the redox cycling of both targets in the coupled
assay. Our approach demonstrates that the antimalarial activity
of 3-benz(o)ylmenadiones results from a subtle interplay be-
tween bioactivation, fine-tuned redox properties, and interac-
tions with crucial targets of P. falciparum. Plasmodione and its
analogues give emphasis to redox polypharmacology, which
constitutes an innovative approach to antimalarial therapy.
Introduction
cy than that achieved with monotherapy, and 4) the expected
toxicity is decreased because of synergistic drug interactions
that allow low effective doses. To illustrate these different ad-
vantages, one can refer to reported cases for recently market-
ed antimalarial drug combinations. Atovaquone, used alone,
has rapidly (in a single year) induced resistance in Plasmodium
falciparum parasites following mutations of cytochrome b in
the region located at a putative drug binding site. Malarone
(the combination of atovaquone and proguanil) rescued the
marketed atovaquone development. Numerous artemisinin-
based combination therapies (ACT) have become available and
are widely used malaria drugs, owing to a significant drop
(45%) in mortality and morbidity in malaria cases since 2000.
These spectacular results are partly due to the massive use of
oral ACT, designed to have a high cure rate and few side ef-
fects and to decrease malaria transmission. For example, coar-
tem (the combination of artemether and lumefantrine), artekin
(the combination of dihydroartemisinin and piperaquine), and
artequin (the combination of artesunate and mefloquine), have
been demonstrated in clinical practice to be very effective
against drug resistance.
Drug combination for the treatment of infectious diseases
such as malaria, AIDS, and tuberculosis is becoming a major
strategy in the research and development of effective regi-
mens, because drug resistance of pathogens spreads in many
areas of the tropical world, making these diseases a major
public health disaster. Effective combination regimens offer nu-
merous benefits over single targeted monotherapy: 1) the risk
of developing rapid resistance is high for drugs targeting
a single step of a selected pathway, 2) the probability of induc-
ing multiple individual target-based gene mutations is low for
agents acting on multiple targets or whose targets are prod-
ucts of multiple genes, 3) the use of multitarget drugs or sev-
eral drugs can achieve a therapeutic effect with greater effica-
[a] P. Sidorov, Dr. D. Horvath, Dr. G. Marcou, Prof. A. Varnek
Laboratoire de Chemoinformatique
UMR 7140 CNRS–UniversitØ de Strasbourg
1 rue Blaise Pascal, Strasbourg 67000 (France)
[b] P. Sidorov
Butlerov Institute of Chemistry, Kazan Federal University
1/29 Lobachevskogo str., Kazan 420008 (Russia)
In addition to drug combinations, the identification of a mul-
titarget drug is a promising and appealing strategy that may
provide the desired drug discovery breakthrough required to
treat malaria. The parasites need to develop resistance against
one single agent that has numerous targets, exponentially de-
laying the onset of resistance, such as in a combination thera-
py. One example of the success of this research strategy is the
recent discovery of pyrrolopyrazines as potent multitarget anti-
malarial agents[1] from in-depth bioinformatics analysis and
target panel screening, suggesting IspD and multi-kinase inhib-
ition. Therefore, it is believed that composite computational
approaches may provide unique access to deciphering poly-
pharmacological effects of new bioactive chemical agents.
[c] I. Desta, M. ChessØ, Dr. E. Davioud-Charvet, Dr. M. Elhabiri
Laboratoire de Chimie Bioorganique et Medicinale
UMR 7509 CNRS–UniversitØ de Strasbourg
European School of Chemistry, Polymers and Materials (ECPM)
25 Rue Becquerel, 67087 Strasbourg (France)
E-mail: elisabeth.davioud@unistra.fr
elhabiri@unistra.fr
[d] I. Desta
New York University Abu Dhabi (NYUAD), Saadiyat Island, Abu Dhabi (UAE)
Supporting information and the ORCID identification number(s) for the
author(s) of this article can be found under http://dx.doi.org/10.1002/
cmdc.201600009.
This article is part of a Special Issue on Polypharmacology and
Multitarget Drugs. To view the complete issue, visit:
http://onlinelibrary.wiley.com/doi/10.1002/cmdc.v11.12/issuetoc.
ChemMedChem 2016, 11, 1339 – 1351
1339
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Scheme 1. Bioactivation of 3-benzylmenadiones (benzylMD) through a cascade of redox reactions, from their 3-benzoylmenadione (benzoylMD) metabolites,
catalyzing both the glutathione reductase (Target 1) and methemoglobin (Target 2) redox cycling in P. falciparum-infected RBCs. The continuous conversion of
metHb(FeIII) to oxyHb(FeII) was spectrophotometrically evidenced by monitoring hemoglobin absorbance in a coupled assay using the NADPH-GR system in
the presence of the redox cycler.
A unique class of multitarget agents is represented by
redox-active agents: in their oxidized states, they can be re-
duced by Target 1 and regenerated by Target 2 catalysis
through an oxidation step from their reduced state
(Scheme 1). Methylene blue[2,3] and 3-[4-(trifluoromethyl)ben-
zyl]menadione, henceforth called plasmodione,[3–5] are repre-
sentatives of such redox cyclers, acting both as catalysts and
NADPH-consuming turncoat inhibitors of multiple targets, that
is, NADPH-dependent disulfide reductases[6] and methemoglo-
bin(FeIII) (metHb)[2,7] with potent inhibition of P. falciparum-par-
asitized red blood cell (pRBC) growth in vitro,[3–5,8] moderate
in vivo activities in orally administered P. berghei-infected
mice,[4] and very low rates of spontaneous drug resistance.[8]
In their reduced state (i.e., the redox properties of plasmo-
dione or methylene blue are seemingly the key parameters of
their peculiar antimalarial activities), these redox cyclers play
a critical role in multiple and interrelated vital processes of
P. falciparum infecting RBCs. Above all, they can markedly alter
the hemoglobin digestion process of the parasite by reversing
the Hb(FeII)$metHb(FeIII) equilibrium toward the less digesti-
ble hemoglobin state. Alternatively, they can induce significant
oxidative stress to the parasites by reacting with oxygen to
to the parasite. Furthermore, we recently demonstrated[5] that
these redox cyclers (or their metabolites), in oxidized or re-
duced states, continuously generate ROS in P. falciparum
pRBCs, but can also alter hemoglobin catabolism. The generat-
ed ferrylhemoglobin(FeIV) ultimately leads to hemichrome pre-
cipitation and subsequent and early phagocytosis of the
pRBCs. Last but not the least, these redox cyclers can be con-
verted into other metabolites of interest, acting on other pro-
cesses vital to the parasite. Benzoxanthones, which have been
described as efficient hemozoin inhibitors, can be produced in
the pRBCs. Conferring fine-tuned redox properties to well-tail-
ored compounds can represent a smart and efficient polyphar-
macological strategy to combat malarial parasites, although
detailed mechanisms are not yet completely understood.
Therefore, building computational models to predict redox
properties of putative antimalarial leads would undoubtedly
speed up the development and delivery of multitarget redox-
active drug candidates. The chemical space of possible ana-
logues is virtually infinite, but resources for synthesis and prop-
erty measurement are not. Synthesis can be assisted by com-
puter-aided design of molecules with desired properties, based
on routinely used quantitative structure-property relationship
(QSPR) models,[9] to prioritize the molecules to be synthesized
and tested. Such models are learned by experimental data;[10]
structures with known desired properties are represented by
molecular descriptors,[11–13] quantitative characteristics are cal-
culated from the structure, then machine learning methods[14]
are applied in order to establish a mathematical relationship
between the experimental property (P) and descriptors (D).
CÀ
lead to O2 /H2O2 species that will generate a hostile milieu in
the presence of ferrous compound (see above, conditions for
production of reactive oxygen species [ROS]). As these versatile
compounds also act as efficient catalysts and effective sub-
strates of NADPH-dependent disulfide reductases, they exhaust
the P. falciparum antioxidant defense by overconsuming
NADPH and generating a pro-oxidant environment deleterious
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1340
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Once this function is established, it may be applied on untest-
ed or virtual compounds to guide chemists toward molecules
possessing desirable properties. The molecular descriptors may
represent various characteristics of compounds, from physico-
chemical properties to particular structural motives and fea-
tures.
conditions.[23,24] In the present work, the silver-catalyzed cou-
pling reaction represented an easy access to benzyl-substitut-
ed menadione derivatives 1–2, including new representatives
1h–1u (Scheme 2, route A). For the benzylic oxidation of 3-
benzylmenadione, we used CrO3 and periodic acid H5IO6 to
obtain the desired 3-benzoylmenadiones 3–4 in moderate to
good yields (33–67%); new representatives (3h–3t) are de-
scribed here (Table S1). The reaction progress was easily fol-
Chemoinformatics-based prediction models of redox poten-
tial (especially concerning the key family of quinone com-
pounds) are rare. Predictions of thermodynamic functions of
the formation of the radical anion intermediate, as the first
step of quinone reduction, using quantum chemical (QC)
methods, have been more frequently reported. Often, the
redox potential is predicted as a function of Gibbs free energy
of oxidized and reduced forms, calculated by either Hartree–
Fock[15–17] and DFT[18] calculations, or faster but less precise
semiempirical methods. The influence of solvent on redox po-
tentials of quinones was established as an MLR model,[19]
based on QC descriptors and solvent properties (Guttmann ac-
ceptor number, dipole moment). Another QC characteristic,
the electrophilicity index,[20] has been reported to robustly ex-
plain the redox potentials of a small dataset of quinones of dif-
ferent families (benzo-, naphtho-, and anthraquinones with var-
ious substituents).[21] However, the use of computationally
heavy QC calculations limits the applicability of these models
for virtual screening of a vast number of molecules. Therefore,
a model based on simpler molecular descriptors is desirable.
Recently, our group has reported the possibilities of QSPR
modeling for prediction of the redox potential of substituted
2-methyl-1,4-naphthoquinones (menadiones).[22] An online
evaluation model was built and made available to predict the
electrochemical properties of these quinones and redox-active
analogues to aid medicinal chemists in targeting their synthet-
ic efforts toward redox-active agents.
1
lowed by H NMR, as the singlet of the bridging CH2 group dis-
appeared over the course of the reaction until complete con-
version.[4] To mask the polarity of the carboxy group of acids
1 f and 3 f, the corresponding amides were prepared
(Scheme 2, route B).
Previous results indicated increased antimalarial activity for
b-cyanoamides, prepared from acids and 3-aminopropionitrile,
as intermediates in the synthesis of tetrazoles.[25] The acid was
first converted into its acyl chloride with SOCl2, and finally, 3-
aminopropionitrile was added to afford amides 1m and 3m.
Carboxylic acid 3 f was converted into its corresponding acid
chloride and then stirred in MeOH to form methyl ester 3n
using a standard protocol (Scheme 2, route C). Finally, starting
from the previously reported 2-bromo-1,4-dimethoxy-3-methyl-
naphthalene, accessible through bromination at the aromatic
ring with Br2 in CH2Cl2 at 08C (86%), the bromide was allowed
to react with acyl chlorides (Scheme 2, route D). The resulting
methanone intermediates 5h and 5o were oxidized by CAN to
give the desired benzoyl derivatives 3h and 3o.
Physicochemical investigations
We describe and discuss herein the physicochemical results
that were obtained by first measuring the electrochemical
properties (cyclic voltammetry [CV] and square-wave voltam-
metry [SWV]) of two homogenous series of 3-benzyl- and 3-
benzoylmenadiones (Table 1, Table S2). These redox-active
compounds differ by the substitution patterns of their benz(o)-
yl units (eastern region), which highlighted the key role of this
moiety (electronic and steric effects) on the antimalarial activity
of these derivatives.
In the present work, we extended our model to predict the
oxidant character of polysubstituted 3-benzylmenadione deriv-
atives and their putative metabolites for further optimization
of the early lead plasmodione. First, all redox potential values
of both sub-series of compounds were experimentally deter-
mined by cyclic and square-wave voltammetries. As 3-benzoyl-
menadiones act as catalytic inhibitors of both human and
P. falciparum glutathione reductase (GR), we then applied
a multitarget drug screening assay using the redox cycler in an
in vitro reconstitution system in the presence of both targets,
that is, Target 1: the human glutathione reductase (hGR), cou-
pled to Target 2: metHb(FeIII). To biochemically probe the com-
putationally predicted electron-transfer properties, we subject-
ed polysubstituted 3-benzylmenadione derivatives to metHb
reduction activity kinetics in the previously established cou-
pled assay, based on the NADPH-dependent hGR and metHb.
Also, the benzoyl C=O group, which results from endoge-
nous oxidation (bioactivation) of the benzylic related prodrugs,
has a significant impact on redox properties. Complementarily
to the electrochemical approach, we furthermore analyzed the
capacity of the menadione analogues, in their reduced states,
to efficiently reduce methemoglobin metHb(FeIII) to oxyhemo-
globin oxyHb(FeII) in
a reduction assay coupled to the
hGR/NADPH system under quasi-physiological conditions
(Scheme 1). The latter continuously regenerated the reduced
species of our substrates. This experiment reflected the capaci-
ty of the substrates to be efficiently reduced by hGR (Target 1)
under NADP flux and to transfer its electron(s) to ferric species
(Target 2). Together, these electrochemical, absorption spec-
troscopy, and kinetic data represent a valuable physicochemi-
cal dataset that highlights key aspects related to the mecha-
nism of action of this class of multitarget compounds toward
parasitic pathogens.
Results and Discussion
Chemistry
The silver-catalyzed decarboxylation reaction is an efficient
method to synthesize 2-methyl-1,4-naphthoquinone deriva-
tives starting from carboxylic acids, under previously described
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1341
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Scheme 2. Synthesis of benzyl- and benzoyl-substituted derivatives of menadione and plumbagin 1–4. Reagents and conditions: A: a) phenylacetic acid deriva-
tive (2.0 equiv), AgNO3 (0.1 equiv), (NH4)2S2O8 (1.3 equiv), CH3CN/H2O, 2 h, 858C; b) CrO3 (0.2 equiv), H5IO6 (7.0 equiv), CH3CN, 24 h, RT. B: a) SOCl2 (8 mL),
reflux, 2 h; b) 3-aminopropionitrile (1.0 equiv), CH2Cl2, RT, 1 h. C: a) SOCl2 (10 equiv), reflux, 3 h; b) MeOH (5 mL), RT, 3 h. D: a) 1. nBuLi (1.1 equiv), dry THF,
À788C, 15 min, 2. acyl chloride (1.2 equiv); b) CAN (3 equiv), CH3CN/H2O, 15 min, RT or 1. BBr3 (1 equiv), CH2Cl2, 2. oxidation in open air.
Plasmodium parasites use methemoglobin as a source of
amino acids for their own growth and digest it more quickly
than hemoglobin. Under the acidic conditions of the digestive
food vacuole of Plasmodium parasites, hemoglobin (HbFeII) or
oxyhemoglobin (oxyHb or oxyHb(FeII)) is quickly oxidized to
methemoglobin (metHb or metHb(FeIII)). Consequently, the re-
duction of metHb(FeIII) to oxyHb(FeII) can significantly slow
metHb digestion. The ability of our redox cyclers to inhibit
both human and plasmodial glutathione disulfide reductases
(hGR, PfGR) of the parasitized erythrocytes, combined with
their potency to reduce metHb(FeIII) to oxyHb(FeII), contribute
to a rise oxidative stress and interfere with hemozoin forma-
tion, with both phenomena leading to the death of the para-
sites. This pernicious and continuous futile NADPH flux under
the catalysis of these disulfide reductases in the presence of
redox-active compounds (which behave as substrates or elec-
tron acceptors) can dramatically increase the flux of toxic re-
duced species within the parasite and leads to its death by
shifting the equilibrium metHb(FeIII) to oxyHb(FeII) heme spe-
cies.
effects on the menadione core (i.e., the western region of the
3-benz(o)ylmenadione derivatives) and found that the two
one-electron transfers were sensitive to the nature of the sub-
stituent and the position of substitution.[22] The redox poten-
tials of the substituted 3-benz(o)ylmenadiones (Table 1 and
Figures S3–S42) were measured by CV (Table 2) and SWV
(Table S3) at 258C using a glassy carbon electrode in DMSO
solvent and tetra-n-butylammonium hexafluorophosphate
[N(nBu)4PF6] as the supporting electrolyte. These two electro-
chemical techniques are complementary, and the conditions
are exactly similar to those employed for investigation on the
substituted menadiones.[22] It is also noteworthy that the elec-
trochemical properties measured by SWV (Table S3) were
found to be in excellent agreement with those measured by
CV. SWV is a rapid, sensitive, and accurate technique that ena-
bles subtraction of background currents. In addition, it allows
direct and accurate measurement of E1/2 for (quasi)reversible
reactions, even with proximate potential values. The CV vol-
tammograms were recorded across a potential range from
+0.5 V to À2.2 V vs. KCl(3m)/Ag/AgCl reference electrode (+
0.210 V vs. NHE).[47]
In aprotic solvents, 1,4-naphthoquinone (1,4-NQ) reduction
occurs through two successive one-electron transfers. The
monoradical anion (1,4-NQ·À) is formed in the Epc1 reduction
step and is then reduced to its related dihydronaphthoquinone
dianion (1,4-NQ2À) in a second Epc2 step (Figure 1). The 1,4-NQs
Electrochemical properties of benz(o)ylmenadiones
In this section, we describe the electrochemical properties of
the 3-benz(o)ylmenadiones to evaluate the influence of the
benzyl substitution. We previously evaluated the substitution
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1342
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Table 1. Structures of 3-benz(o)ylmenadione (Benz(o)ylMD) derivatives, including 3-benzhydrolmenadione 6a.
BenzylMD[a]
R1
R2
BenzoylMD[b]
R1
R2
1a
1b
1c
1d
1e
1 f
1g
1h
1i
H
H
H
H
H
H
H
H
H
H
H
H
H
4’-Br
4’-F
3a
3b
3c
3d
3e
3 f
3g
3h
3i
3j
3k
3l
3m
3n
3o
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
4’-Br
4’-F
4’-CF3
4’-OH
4’-tBu
4’-COOH
4’-NO2
4’-H
4’-Me
4’-Cl
4’-CN
4’-OMe
4’-CF3
4’-OH
4’-tBu
4’-COOH
4’-NO2
4’-H
4’-Me
4’-Cl
4’-CN
4’-OMe
1j
1k
1l
1m
4’-CONH(CH2)2CN
4’-CONH(CH2)2CN
4’-COOCH3
2’-F
1p
1q
1r
H
H
H
H
2’-Br, 4’-OCH3
2’-OMe
3’,5’-(OCH3)2
2’,5’-(OCH3)2
3q
3r
3s
3t
H
H
H
H
2’-OMe
3’,5’-(OCH3)2
2’,5’-(OCH3)2
2’-F,5’-OMe
1s
1u
2a
H
5-OH
2’,3’,4’,5’,6’-(F)5
4’-Br
4a
6a[c]
5-OH
H
4’-Br
4’-Br
[a] Benzylmenadione scaffold: 1 and 2. [b] Benzoylmenadione scaffold: 3 and 4. [c] Benzhydrolmenadione scaffold: 6.
Table 2. Electrochemical data (first and second redox waves for the benz(o)yl series; a third redox wave has been evidenced for the benzoyl series) mea-
sured by CV for all benz(o)ylmenadione derivatives examined in this work.[a]
Compd
E11/2(DE) [V(mV)]
E21/2(DE) [V(mV)]
DE1/2 [V]
Compd
E11/2(DE) [V(mV)]
E21/2(DE) [V(mV)]
E31/2(DE) [V(mV)]
DE1/2 [V]
menadione
À0.61(93)
À0.62(71)
À0.63(88)
À0.59(76)
À0.65(94)
À0.64(80)
À0.65(80)
À0.60(70)
À0.63(80)
À0.64(86)
À0.62(84)
À1.30(96)
À1.29(85)
À1.30(86)
À1.27(78)
À1.40(290)
À1.38(80)
À1.35(109)
À1.38(80)
À1.33(128)
À1.35(93)
À1.30(84)
0.69
0.67
0.67
0.68
0.75
0.74
0.7
0.74
0.70
0.71
0.68
1a
1b
1c
1d
1e
1 f
1g
1h
1i
3a
3b
3c
3d
3e
3 f
3g
3h
3i
3j
3k
3l
3m
3n
3o
3q
3r
À0.46(78)
À0.46(86)
À0.43(80)
À0.53(80)
À0.47(78)
À0.49(72)
À0.42(90)
À0.47(92)
À0.48(90)
À0.45(82)
À0.43(80)
À0.49(86)
À0.45(82)
À0.44(88)
À0.45(74)
À0.54(88)
À0.47(77)
À0.54(96)
À0.45(76)
À1.12(80)
À1.14(82)
À1.12(84)
À1.17(80)
À1.18(66)
À1.29(160)
À1.17(88)
À1.19(72)
À1.14(41)
À1.13(60)
À1.09(70)
À1.17(54)
À1.09(70)
À1.10(92)
À0.99(140)
À1.17(88)
À1.22(77)
À1.16(88)
À1.11(74)
À1.63[b]
À1.65[b]
À1.62(60)
nd
À1.62[b]
À1.61[b]
À1.54[b]
À1.61[b]
À1.63[b]
À1.62[b]
À1.39(86)
nd
0.66
0.68
0.69
0.64
0.75
0.8
0.75
0.72
0.66
0.68
0.66
0.68
0.64
0.66
0.54
0.63
0.75
0.62
0.66
1j
br
À1.74(156)
À1.55[b]
br
1q
1r
À0.65(80)
À0.65(82)
À1.36(94)
À1.35(82)
0.71
0.7
À1.49[b]
3s
3t
br
À1.64(62)
1u
plumbagin
2a
À0.60(80)
À0.44(90)
À0.47(67)
À1.25(132)
À1.06(80)
À1.03(32)
0.65
0.62
0.56
4a
À0.30(82)
À0.93(76)
À1.49[b]
0.63
[a] CVs measured in DMSO with N(nBu)4PF6 (0.1m) electrolyte support at 258C; v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working electrode:
glassy carbon disk with 0.07 cm2 area. Compounds used for the validation test shown in blue. [b] SWV measurements; nd: not determined; br: broad
signal.
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1343
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
200 mVsÀ1), while those in the second redox process
are often close to 0.8. The cathodic ipc and anodic ipa
intensities also vary linearly with the square root of
the potential scan rate v1/2 (ip =(2.69”105) n3/2 A D1/2
[1,4-NQ] v1/2),[28] which confirms that these electron
transfers are diffusion-controlled processes.[29] The
voltage scan rate, however, has no influence on the
reduction and oxidation potentials, which further
substantiates this property. The current intensities
Figure 1. Oxidation/reduction processes of 1,4-NQs in aprotic solvents.
(except for the proton-donating systems, such as 3-benzylme-
nadiones 1d and 1 f, and 3-benzoylmenadiones 3d and 3 f,
Tables 1 and 2) are not subjected to any inter- or intramolecu-
lar proton transfers and therefore the electrochemical proper-
ties are strictly associated with successive formation of the two
anionic reduced species (Figure 1).
Substitution by either benzyl or benzoyl derivatives has no
effect on the global electrochemical behavior of the 1,4-NQ
core (i.e., two successive one-electron transfers). Table 2 shows
the voltammetric data (CV) of each of the substituted 3-ben-
z(o)ylmenadiones (including menadione[22] and plumbagin) in-
vestigated in this work. With the exception of hydroxylated
menadiones, two consecutive one-electron quasi-reversible
waves (E1pc-E1 ~67–96 mV and E2pc-E2 ~41–140 mV) were
pa
pa
constantly observed. In addition to these two quasi-reversible
Figure 3. CV profiles of 1i (1.05 mm, black) and 3i (0.99 mm, blue), mea-
sured in DMSO with N(nBu)4PF6 (0.1m) electrolyte support at 258C.
v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working electrode:
glassy carbon disk with 0.07 cm2 area. A third redox wave, attributed to the
benzoyl carbonyl group, can be observed at more negative values.
steps for the benzylmenadiones series, a third redox wave at
an intermediate potential value (E3 ~À1.07 V) was observed
1/2
for 4’-nitrobenzylmenadione 1g that results from reduction of
the nitro moiety (Figure 2).[26,27] Irrespective of the 3-benzoyl-
menadione derivative considered, a third (or fourth for 3g)
weak, broad, and ill-defined wave was observed at more nega-
tive values (E31/2 @À1.5 V) that can be related to the oxidation/
reduction of the benzoyl carbonyl unit (Table 2, Figure 3, and
Table S3).
(Table S3) of the second reduction step (ipc2) leading to 1,4-
NQ2À were systematically found to be weaker with respect to
the reduction step (ipc1) leading to the monoradical anion 1,4-
NQ·À. This electrochemical behavior may result from compro-
The degree of reversibility for the first electron transfer is in-
dicated by the ratio ipa/ipc, which is close to 1 (at v=
portionation reactions (E11/2-E2 ~0.65–0.74 V and lnKcomp rang-
1/2
ing from ~23 to 26)[30–33] and/or by fast and irreversible dimeri-
zation[33] between 1,4-NQ2À and 1,4-NQ to afford the electro-in-
2À
active dimer (1,4-NQ)2
.
With the exception of the nitro-containing compound,
a clear relationship (Figure 4) was observed between the half-
wave potentials of the first (E11/2) and second (E21/2) electro-
chemical processes. This feature strongly suggests that the
substituents on the benz(o)yl core induce the same electronic
effects on both the 1,4-NQ unit and its one-electron-reduced
semiquinone, 1,4-NQ·À. However, exceptions were observed for
benzoylmenadiones 3o (steric effect of the 2’-F substituent)
and 3q/3s (steric effect of the 2’-OCH3 substituent), nitro-con-
taining benz(o)ylmenadiones 1g and 3g (formation of a one-
electron-reduced NO2 intermediate), and benz(o)yl-plumbagin
derivatives 2a and 4a (strong hydrogen bond between the 5-
OH group and the 1,4-NQ carbonyl group).
Figure 2. CV profiles of 1g (0.91 mm, blue) and 3h (1.19 mm, black), mea-
sured in DMSO with N(nBu)4PF6 (0.1m) electrolyte support at 258C.
v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working electrode:
glassy carbon disk with 0.07 cm2 area. A third redox wave, attributed to the
nitro group, can be observed between the two one-electron transfers of 1,4-
NQ.
To gain insight into the influence of the benz(o)yl substitu-
ents on the electrochemical properties of the 1,4-NQ core, the
Hammett[34,35] approach (Figure 5) was used, as previously
done for simpler menadiones.[22] The 1,4-NQ reactivity can be
anodically or cathodically shifted by adding electron-releasing
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1344
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
series, Figure 5 illustrates the variation of DEn (n=1 or 2) and
1/2
demonstrates a linear dependence of DE11/2 and DE2 with the
1/2
electronic substituent constants (S(sm +sp)). This feature indi-
cates that substituents borne by the benz(o)yl moiety electron-
ically influence the 1,4-NQ core in a related manner.
Overall, for 3-benz(o)ylmenadiones, the reaction constant
(1R) for the first electron transfer (Figure 5) was calculated to
be 0.68(1). This value was significantly lower than those mea-
sured for the diversely substituted menadiones (11 =2.20),[22]
R
demonstrating that substitution on the western region (i.e.,
1,4-NQ core) of the benz(o)ylmenadiones has a much stronger
effect on the oxidant character than does substitution on the
eastern side (i.e., benz(o)yl unit). As electron delocalization
through conjugation can be excluded for the 3-benz(o)ylmena-
diones because the benzylic methylene or benzoyl carbonyl
are not good electronic relays, other interactions, such as CH–
p or dipole–dipole, can be proposed to explain the sensitivity
of the 1,4-NQ electrophore to the benz(o)yl substitution pat-
tern. In contrast with the 3-benz(o)ylmenadiones, their corre-
sponding one-electron-reduced species are much more sensi-
tive to substitution on the eastern side, as shown by the reac-
Figure 4. Variation in E2 (V, second redox step) as a function of E11/2 (V, first
1/2
redox step). v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working
electrode: glassy carbon disk with 0.07 cm2 area; auxiliary electrode: Pt wire.
The dashed line is only provided as a guide. & 3-Benzylmenadiones; * 3-
benzoylmenadiones; ~ compound with particular behavior (see text).
tion constant (1R) for the second electron transfer (12 =1.5(2)).
R
This might be rationalized by other interactions (e.g., anion–
p···) that take place within the negatively charged one-elec-
tron-reduced species. As a general rule, and regardless of the
nature of the substitution and its position (east vs. west), elec-
tron-withdrawing substituents decrease the electronic density
of the 1,4-NQ electroactive unit and facilitate the reduction of
these compounds, while electron-donating substituents (Me,
tBu) lessen the propensity of the corresponding redox com-
pounds to be reduced (Figure 5).
In the strict implementation of Hammett-Zuman equation,
its intercept should be null. This is almost the case, within ex-
perimental errors, for DE11/2 and DE2 (Figure 5). This suggests
1/2
that the two consecutive electron transfers are not complicat-
ed by coupled chemical reactions such as proton transfer.
Comparison of the 3-benzoylmenadiones (i.e., putative bio-
activated metabolite) and the benzyl analogues (i.e., putatively
acting as prodrug) shows that the benzylic oxidation believed
to occur within the pRBCs plays a crucial role in electrochemi-
cal characteristics and may thereby explain the observed anti-
Figure 5. Plots of DE11/2 (&) and DE2 (*) vs. S(sm +sp) for the substituted
benz(o)ylmenadiones series. Solvent: DMSO; I=0.1m N(nBu)4PF6; T=258C;
v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working electrode:
glassy carbon disk of 0.07 cm2 area; auxiliary electrode: Pt wire. DE11/
1/2
2 =0.040(5)”(S(sm +sp))À0.006(3) (R2 =0.820); DE2
1/
2 =0.09(1)”(S(sm +sp))À0.003(6) (R2 =0.815). For the 4’-substituted systems,
we used the sum (sm +sp) to describe the electronic effect of a substituent.
In the case of polysubstituted menadiones, the contribution of each sub-
stituent was considered.
malarial activity. Both one-electron redox waves E11/2 and E2
1/2
(Figure 5) were substantially shifted to more positive potential
values, meaning that the oxidized compounds at the benzylic
position became more oxidized and therefore were more
prone to transfer their electron to ferric targets such as
metHb(FeIII) (see below). With some exceptions that are dis-
cussed below, the redox potentials were anodically shifted by
~160 to 200 mV (see Table 4 below). This is clearly the result of
the new substitution by a keto functionality that stabilizes the
p-electrons of the menadione core (regardless of the oxidation
state) by delocalization and conjugation. Deviation from this
trend was observed for benzyl/benzoyl pairs displaying steric
interactions (1q/3q) and for the phenolic/carboxylic-substitut-
ed systems (1d/3d and 1 f/3 f), the nitro-substituted deriva-
tives (1g/3g) and the plumbagin analogues (2a/4a). Steric in-
teraction, intramolecular proton transfer within the reduced
or -withdrawing substituents either to the electrophore[22] or to
the benz(o)yl moiety (Table 2). The half-wave potentials (En
1/2(ref-
erence), n=1 or 2) of the references (1h for the benzyl series and
3h for the benzoyl series) were shifted to new values (En
1/
2(sample), n=1 or 2) by introducing one or several substituents
(Table 2 and Figure 5). With the exception of peculiar systems
(nitro derivatives 1g and 3g and plumbagin analogues 2a and
4a) and those inducing steric interactions (2’-substituted-3-
benz(o)ylmenadiones were not considered, Figure 5), the po-
tential shifts could be theoretically predicted by the Hammett-
Zuman relationship. Irrespective of the benzyl or benzoyl
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1345
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
species, and an additional redox pair centered on a nitro func-
tionality or strong hydrogen bond are among the processes
that can modulate the potential gap between the oxidized
and reduced species (Table 2).
Table 3. Statistical characteristics of consensus multilinear regression
(ISIDA MLR) and support vector machine (EED SVM) models for cross-vali-
dation (XV) and external test challenges.[a]
Descriptor
Training (10”3ÀXV)
Test
RMSE
R2
RMSE
R2
0.032 (0.027)[b] 0.86 (0.90)[b]
0.054 (0.030)[b] 0.60 (0.88)[b]
QSPR modeling studies of the redox properties
ISIDA MLR
EED SVM
0.080
0.082
0.85
0.84
Previous work[22] reported QSPR models for redox potential
that were built on a set of various quinones and indolone-N-
oxides, both novel and taken from literature. One of the ex-
plored modeling protocols, based on ISIDA molecular fragment
counts (the best compromise between accuracy and technical
web deployment costs) has been posted on our web server for
public use. Basically, it predicts the redox potential values by
adding fragment-specific increments for each of the key frag-
ments shown, in the training stage, to best explain experimen-
tal property values. The approach accepts a structure file for
organic compounds, then proceeds, for each molecule, to the
detection and counting of the mentioned key fragments. Each
occurrence of a key fragment triggers a fragment-specific in-
crement, expressed in volts (which may be positive or nega-
tive, as calibrated by hand using training compounds) to be
summed to the predicted redox potential value. Note that sev-
eral other theoretical models based on different molecular de-
scriptor schemes—notably Electronic Effect Descriptors (EED),
designed for the purpose of modeling reactivity-related prop-
erties—have also been explored, with very promising results.
Revisiting the technicalities of the previous modeling work is
not the scope of the present paper. The reader is advised to
refer to the previous article for details on the employed molec-
ular descriptor schemes, which are also adopted in the present
work.
[a] RSME: root-mean-squared error; R2: determination coefficient. [b] Ad-
justed statistical parameters when the outlier (2a) is excluded from the
test set.
the best descriptor space to host optimal SVM models, EED
terms clearly outperformed individual descriptor spaces. How-
ever, models based on different ISIDA molecular fragments
were combined into a consensus model (in which only models
with R2 of cross-validation >0.5 are accepted). This consensus
effect, over many different ISIDA descriptor spaces, interesting-
ly compensated for the advantage of EED over each individual
ISIDA fragmentation scheme and also for the alleged advant-
age of nonlinear modeling. As shown, the single descriptor-
space (EED) nonlinear model and the multi-fragmentation con-
sensus approach eventually performed equally, and very well,
in a threefold cross-validation challenge.
It is, nevertheless, of greater practical interest to evaluate
a model’s performance using the external test set. As Table 3
shows, consensus models based on ISIDA descriptors per-
formed well in both cases, with the R2 of the test being close
to the R2 of cross-validation. The model with EED descriptors
had a globally lower R2, but this was due to one molecule
being predicted more poorly, which, due to the modest size of
the test set, had a greater impact on the overall score
(Figure 6). The poorly predicted molecule (2a) is the derivative
with a hydroxy substituent, and the phenomenon has been re-
ported previously.[22] However, the updated training set now in-
cludes examples of analogues with such an intramolecular hy-
drogen bond and, accordingly, the structural pattern was rec-
ognized when using ISIDA fragment counts. The monitored
atom and bond sequences, with lengths ranging from two to
ten, include the corresponding O=CÀC:CÀO pattern (“:” repre-
sents an aromatic bond) of the fragment responsible for the
intramolecular hydrogen bond and only appear in the context
of the presence of a 5-hydroxy substituent. Therefore, the
model learned to associate this fragment with a negative incre-
ment for the predicted redox potential (the hydrogen bond
polarizes the quinone carbonyl even further, rendering the qui-
none system more electron-depleted). Of course, the presence
of such a fragment itself is not sufficient to trigger a decrease
in the redox potential; it must be determined whether the
fragment is in the right context, that is, involving the actual
quinone carbonyl and the adjacent phenolic hydroxy group in
a hydrogen bonding position. Should the mentioned sequence
appear in other moieties of a molecule, not connected to the
quinone system, the model would also add the associated
(negative) increment to the predicted redox potential and
would likely result in an error. This shows that even the signifi-
cantly extended compound set used in this work is still prone
Expansion of the chemical space of interest to include ben-
zoyl derivatives naturally raises the question of the compe-
tence of the previous model with respect to this new chemo-
type, which has not been previously employed for training.
Therefore, the first logical step was to challenge the old model
to make a prediction for the newly synthesized compounds.
The predictions by the model, are, on the absolute, quite inac-
curate (root-mean-squared error [RMSE]=0.176). This is not
surprising, as the tested compounds are all derivatives of new
families; the molecules contain benzyl and benzoyl substitu-
ents that were never present in the training set of the model,
which had no chance to learn their impact on the redox po-
tential value (Figure S43). The experimental, predicted (with
and without the applicability domain), and corrected (by cor-
rection coefficients) values are available as Supporting Informa-
tion (Table S4).
Table 3 reports the statistical parameters for new models
built on a combination of old and new data with different
methods. The training set for modeling consisted of 81 mole-
cules (all previously used data and examples of both new fami-
lies combined), and 14 benzyl and benzoyl derivatives were
manually selected—or tested only after model building—to
serve as an external test set.
The cross-validation R2 scores were quite high for all model-
ing methods. Note that, in the evolutionary competition for
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1346
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
strates as long as NADPH was present. This assay was recently
established as a relevant in vitro model in our laboratory.[4,7]
The UV/visible absorption spectrum of metHb(FeIII) between
300 and 700 nm was characterized by a Soret band of the FeIII
heme, whose maximum absorbance is centered at 405 nm and
was used as a valuable spectroscopic probe. Upon metHb(FeIII)
reduction by the enzymatically generated reduced species of
the redox-active compounds, the occurrence of the oxyHb(FeII)
species was associated with a bathochromic shift in the ab-
sorption maximum from 405 to ~410 nm (now corresponding
to a low spin hexacoordinated ferrous ion), as well as the for-
mation of two new less intense absorptions at ~536 and
~576 nm. As an example, Figure 7 depicts the spectral absorp-
tion variation recorded for 1c, 3b, and 6a under these experi-
mental conditions. These compounds represent some of the
putative metabolites series that can be generated within the
pRBCs, namely the benzyl-, benzhydrol-, and benzoylmena-
diones.
Kinetic data (pseudo-first-order rate constant, kred, sÀ1) pro-
cessed for a large series of compounds are shown in Table 4,
together with the E11/2 potential values previously determined
using CV. This important dataset allowed us to highlight inter-
esting properties related to the mechanism of action of the an-
timalarial benz(o)ylmenadiones. First, the benzylmenadiones
(Figure 7A), irrespective of their substitutions, were not reac-
tive with respect to metHb(FeIII). This feature can be clearly re-
lated to their redox potentials, which are ~160 to 200 mV cath-
odically shifted with respect to their benzoyl analogues
(Table 4).
Figure 6. Plot of E11/2 for test set compounds, experimental values vs. those
predicted by all models: ISIDA MLR (*), predicted by ISIDA/QSPR software,
and EED SVM (&), predicted by the web server application. The dotted line
represents perfect fit. The outlier (benzyl derivative with hydroxy substituent
in the naphthoquinone moiety) is highlighted.
to significant biases and sources of artifacts. The rule “presence
of a O=CÀC:CÀO ! presence of intramolecular hydrogen
bond ! lowered redox potential” was learned on the basis of
training examples and also happened to hold for the test set
compounds but was clearly not a general characteristic.
In contrast, EED descriptors do not explicitly consider con-
nectivity patterns but only focus on through-bond electronic
effects, ignoring all through-space non-bonded influences. As
such, they cannot capture the atypical effect in 5-OH ana-
logues, even if these were present in the training set. They fail
to learn any specific rule for the 5-OH analogues and therefore
commit an important error of +0.17 V when predicting the
redox potential of 2a. However, they are not prone to fail in
the above-discussed scenario of external predictions for com-
pounds with O=CÀC:CÀO moieties not responsible for redox
behavior. If 2a is discarded as an outlier, then RMSE=0.030 V
and R2 =0.88 (comparatively, for ISIDA descriptors RMSE=
0.027, R2 =0.90).
Bioactivation through benzylic oxidation leading to a benzhy-
drol (e.g., 6a, Figure 7C, Table 4) or a benzoyl (Figure 7B,
Table 4) analogue markedly favors metHb(FeIII) reduction. The
higher oxidant character of the benzhydrol- or benzoylmena-
diones therefore appears to be a key parameter. Notably, the
redox properties of the redox cycler must be subtly fine-tuned,
as their redox potentials should be in a narrow range of poten-
tial values, defined by the GR(flavine FAD/FADH2)/NADPH[36]
and the metHb(FeIII)/Hb(FeII).[37] Deviation from this prevents
either reduction of the electrophore by the GR/NADPH pair or
reduction of the metHb(FeIII) by the reduced electrophore. This
feature is exemplified in Figure 8 by benzylmenadiones 1c or
1u whose E11/2 values (~0.6 V) are close to that of menadione
(Table 2) and are not proper for efficient reduction by the GR/
NADPH pair. On the other hand, the system with the greatest
oxidant effects examined so far in this extensive work (i.e.,
plumbagin analogue 4a, E11/2 =À0.30 V) was far from being
the most efficient system to rapidly target metHb(FeIII) in the
reduction assay coupled to the hGR/NADPH system in vitro.
Even though a dependence of the rate constant of the
metHb(FeIII) reduction as a function of the redox potential E11/2
can be deduced as anticipated within a limited potential range
as defined above, several systems deviate from this linear var-
iation and were found to be very informative. As previously
noted, plumbagin derivative 4a deviates from this trend be-
cause of its particular structure and subsequent redox behavior
(i.e., a strong hydrogen bond between the 5-OH group and
one of the carbonyl groups that renders the compound too
The comprehensive table of predictions for test set com-
pounds by all methods is available as Supporting Information
(Tables S4 and S5). SVM models based on EED descriptors are
also available for users in a web application (http://infochim.u-
strasbg.fr/webserv/VSEngine.html). When an SDF file is upload-
ed in the application, it automatically detects the redox center
of the molecule. Evidently, the applicability domain of the
model accepts only molecules with a carboxyl redox center,
preferably a quinone or indolone moiety. Performance with
other scaffolds of redox-active compounds is not guaranteed.
Reduction of methemoglobin to hemoglobin
To evaluate the potency of our compounds to reduce metHb-
(FeIII) to oxyHb(FeII) and to assess their redox characteristics on
reduction kinetics, we employed a reduction assay coupled to
the hGR/NADPH system in vitro (under quasi-physiological con-
ditions), which regenerated the reduced species of our sub-
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1347
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Table 4. Electrochemical data (first redox wave), measured by cyclic vol-
tammetry (CV)[a] and kinetic data, related to the reduction of methemo-
globin in the hGR/NADPH coupled assay.[b]
Compd
E11/2
[V]
kred
]
Compd
E11/2
[V]
kred
[10À4 sÀ1
DE11/2
[mV]
[sÀ1
]
1a
1b
1c
1d
1e
1 f
1g
1h
1i
À0.62
À0.63
À0.59
À0.65
À0.64
À0.65
À0.60
À0.63
À0.64
À0.62
–
–
nr
–
–
–
–
–
–
–
3a
3b
3c4
3d
3e
3 f
3g
3h
3i
3j
3k
3l
3m
3n
3o
3q
3r
À0.46
À0.46
À0.43
À0.53
À0.47
À0.49
À0.42
À0.47
À0.48
À0.45
À0.43
À0.49
À0.45
À0.44
À0.45
À0.54
À0.47
À0.54
À0.45
5.0(3)
7.7(3)
13.1
101(2)
14.2(4)
36.2(8)
39.0(6)
–
160
170
160
120
170
160
180
160
160
170
–
1j
17(2)
54(2)
11.4(3)
82(3)
16.6(5)
–
15.1(4)
9.0(3)
15.0(3)
26(3)
1q
1r
À0.65
À0.65
–
–
110
180
3s
3t
1u
2a
À0.60
À0.47
nr
–
4a
6a
À0.30
23(1)
33(1)
170
–
[a] v=200 mVsÀ1; reference electrode: KCl(3m)/Ag/AgCl; working elec-
trode: glassy carbon disk of 0.07 cm2 area; auxiliary electrode: Pt wire.
Compounds used for the validation test shown in blue. [b] Solvent: H2O,
phosphate buffer (47 mm) pH 6.9, EDTA (1 mm), KCl (200 mm); T=25.08C;
NADPH (120 mm)+hGR (80 nm)+metHb (8 mm)+GSSG (20 mm)+sub-
strate (40 mm). nr: no reaction; –: not measured.
Figure 7. UV/visible absorption spectra recorded as a function of time in the
coupled assay in the presence of the hGR/NADPH system. A) Spectral varia-
tions showing metHb(FeIII) reduction in the presence of benzylmenadione
1c. B) Spectral variations showing metHb(FeIII) reduction in the presence of
benzoylmenadione 3b. C) Spectral variations showing metHb(FeIII) reduction
in the presence of benzhydrolmenadione 6a. Solvent: H2O, phosphate
buffer (47 mm) pH 6.9, EDTA (1 mm), KCl (200 mm); T=25.08C; NADPH
(120 mm)+hGR (88 nm)+metHb (8 mm)+GSSG (20 mm)+substrate (40 mm);
(1) t=0; (2) t=120 min.
Figure 8. Schematic representation of the influence of the redox potential
E11/2 (V) on the kinetics of metHb(FeIII) reduction. The dashed line is provided
only as a guide.
highly oxidizing, that is, the E11/2 redox potential is anodically
shifted to À0.30 V). Likewise, the benzylmenadiones, which are
significantly poorer oxidizing agents than the benzoyl ana-
logues, are not reactive, because their redox properties are
outside the potential range of efficacy. Compounds 3s and 3r
behave differently because of steric interactions of the benzylic
2’-methoxy substitution. This is confirmed by derivative 3t,
which bears a less bulky 2’-fluorine substitution and follows
the observed trend (Figure 8). Surprisingly, the two derivatives
bearing a cyano group on the benzoyl unit (3m and 3k) were
found to be very reactive in the reduction assay. This would
suggest an unexpected role for the cyano group that signifi-
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1348
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
cantly boosts the rate of metHb reduction. It should be men-
tioned that the electron transfer between the reduced ben-
zoylmenadione and the ferric target might occur either
through an inner (the involved redox sites interact) or outer
(the involved redox sites remain separate before, during, and
after the electron transfer event) sphere mechanism. This
would suggest that the cyano group might interact with the
ferric reactive center (or, to a lesser extent, with the globin)
and favor the reduction process. Compound 3d, bearing a 4’-
hydroxy substituent on the benzoyl unit, was found to be the
most reactive system (kred =1.01”10À2 sÀ1) among the series
examined and supports the idea that, in addition to the intrin-
sic redox character of the compound, its ability to interact with
one of the putative ferric targets is also of importance to the
global antimalarial activity. Indeed, we previously showed that
some of the putative metabolites of 1c (i.e., plasmodione) effi-
ciently prevent formation of b-hematin (i.e., synthetic hemo-
zoin), which is one of the main mechanism by which P. falcipa-
rum escapes heme toxicity after methemoglobin digestion.
redox potential of various quinones and indolone-N-oxides
with reasonable accuracy, and can therefore be used for the
computer-aided design of new antimalarial agents. We as-
sessed the potency of the drug candidates by evaluating the
electron transfer efficiencies and kinetics between the reduced
3-benz(o)ylmenadiones generated by reduction by hGR under
excess NADPH cofactor and methemoglobin metHb(FeIII). The
3-benzylmenadiones were not reactive, while the benzoyl ana-
logues efficiently reduced metHb(FeIII). Even though a relation-
ship can be drawn between the E11/2 potential values and the
apparent metHb(FeIII) reduction rate constants, the most
potent systems were those displaying a cyano or hydroxy
group on the benzoyl position that was thought to interact
with the ferric center through an inner-sphere electron-transfer
mechanism.
Experimental Section
Physicochemical characterization
Solvents and materials: Electrochemical and absorption spectro-
scopic properties of substituted 3-benz(o)ylmenadiones (Figures S1
and S2) and plumbagin (5-hydroxymenadione) were examined in
spectroscopic grade DMSO (>99.9%, Sigma–Aldrich). All stock sol-
utions were prepared by weighing solid products using a Mettler
Toledo XA105 Dual Range balance. The complete dissolution of the
ligands was obtained using an ultrasonic bath (Bandelin Sonorex
RK102).
Conclusions
The electrochemical properties of a series of polysubstituted 3-
benz(o)ylmenadiones acting as efficient redox cyclers toward
pRBCs were thoroughly investigated. This homogeneous set of
compounds supplemented a preliminary work, focused on the
evaluation of the steric/electronic effects of diverse substitu-
tion on the western part of the 3-benz(o)ylmenadiones, that is,
the 1,4-NQ moiety. All of these compounds were observed to
display a common pattern of two successive one-electron re-
versible transfers, separated by a potential gap, ranging from
0.6 to 0.8 V. The potential values of both electron transfers
were shown to be drastically altered by the nature and/or the
position of the substituents. We clearly identified the substitu-
ents and their subsequent electronic/steric/mesomeric proper-
ties that control the oxidant character of the electroactive me-
nadione derivatives. The most important effects were by far
those brought by the 1,4-NQ substitution, while those cen-
tered on the benz(o)yl unit were less critical but, however, of
significance in the context of their redox cycling capacities. Im-
portantly, this comprehensive study allowed us to evaluate the
impact of the substitution (position and nature) on the redox
properties of a promising class of multitargeted antimalarial
drugs and will undoubtedly pave the way toward better drug
design.
Electrochemistry: The redox potentials of the substituted 3-ben-
z(o)ylmenadiones and plumbagin were measured by cyclic voltam-
metry (CV) and square-wave voltammetry (SWV) in DMSO (Table 1).
DMSO can be classified as a dipolar aprotic protophilic solvent and
was assumed as an suitable solvent[38,39,40] to characterize the elec-
trochemical properties of our menadione analogues (measurable
potential limits of +0.9 V to À3.9 V with respect to Fc/Fc+, E= +
0.524 V/Ag/AgCl).[39] CV and SWV voltammograms were recorded at
room temperature (23Æ18C) in DMSO with 100 mm tetra-n-buty-
lammonium hexafluorophosphate [N(nBu)4PF6] as a supporting and
inert electrolyte.[41] CVs of the menadione or plumbagin derivatives
(~10À3 m) were first performed using a Voltalab 50 potentiostat/gal-
vanostat (Radiometer Analytical MDE15 polarographic stand,
PST050 analytical voltammetry, and CTV101 speed control unit),
controlled by the Voltamaster 4 electrochemical software. A con-
ventional three-electrode cell (10 mL) was employed in our experi-
ments with a glassy carbon disk (GC, s=0.07 cm2) set in a Teflon
rotating tube as a working electrode, a platinum (Pt) wire as
a counter electrode, and KCl(3m)/Ag/AgCl reference electrode (+
210 mV vs. NHE).[42] Prior to each measurement, the surface of the
GC electrode was carefully polished with a 0.3 mm aluminum oxide
suspension (ESCIL) on a silicon carbide abrasive sheet of 800/2400
grit. The GC electrode was then copiously washed with water and
dried with paper towels and argon. The electrode was installed
into the voltammetry cell along with a Pt wire counter electrode
and the reference. Solutions containing the menadione or plumba-
gin derivatives were vigorously stirred and purged with O2-free
argon (Sigma Oxiclear cartridge) for 15 min before the voltamme-
try experiments were initiated and maintained under an argon at-
mosphere during measurement. The voltage sweep rate was
varied from 50 to 300 mVsÀ1, and several cyclic voltammograms
were recorded from +0.5 V to À2.2 V. Peak potentials were mea-
sured at a scan rate of 200 mVsÀ1 unless otherwise indicated.
We also demonstrated that bioactivation of the drugs
through putative benzylic oxidation markedly influence the ox-
idant character of the one- or two-electron-reduced species
through a significant anodic shift in their redox potentials
(Table 4). This accounts for the new keto functionality that in-
creased delocalization and conjugation of the p-electrons of
the menadione core, regardless of its oxidation state. The anti-
malarial efficiency seemingly results from both the fine-tuned
redox properties of the drugs and the bioactivation (i.e., ben-
zylic oxidation among other possibilities) with respect to sever-
al endogenous crucial partners, such as disulfide reductase and
ferric targets. Developed QSPR models are able to predict the
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1349
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Redox potentials were determined from oxidation and reduction
potentials. Secondly, SWV voltammograms were recorded using
the same analytical apparatus. A series of optimization studies
were performed using the SWV parameters:[43] optimized values for
these parameters were: pulse height (DEp), 50 mV; pulse width
(tstep), 50 ms; step potential (Estep), 5 mV, step amplitude, 1 mV; scan
rate, 20 mVsÀ1; start, +0.5 V, end, À2.2 V.
Methemoglobin reduction hGR/NADPH coupled assay: Stock sol-
utions of 2 mm of the drugs in DMSO and of 0.4 mm methemoglo-
bin (from human hemoglobin, Sigma–Aldrich) were freshly pre-
pared in water. Stock solutions of ~4 mm NADPH and 1 mm GSSG
(oxidized l-glutathione, Serva) were prepared daily in hGR buffer
(KH2PO4 [2.79 g], K2HPO4·3H2O [6.04 g], EDTA [0.372 g], and KCl
[14.91 g] in 1 L H2O, pH 6.9). The pH was adjusted by dropwise ad-
dition of KOH (5m). The final NADPH concentration was calculated
groups are present, so the descriptors are calculated as an average
value between the two.
Second, ISIDA fragment count descriptors[55] that encode a molecu-
lar structure by the vector constituted occurrences of the frag-
ments of each type. They offer several advantages over other
types of descriptors: versatility, simplicity and rapidity of calcula-
tions, various applicability domain schemes[55,56] (fragment control
and bounding box strategies), and full support from our web
server Predictor application (http://infochim.u-strasbg.fr/cgi-bin/
predictor.cgi), which allows them to be easily accessible by general
public. With that in mind, 44 different fragmentation schemes, rep-
resenting atom and bond sequences in different lengths ranging
from 2 to 10, were used for training and building a consensus
model.
ISIDA/QSPR software[57] was used to build multilinear regression[58]
models on the training set consisting of 81 molecules. The external
test set contained 14 compounds that were either voluntarily kept
aside or actually measured after model fitting. The software offers
consensus model building with any number of fragmentation
schemes, as well as variable selection and applicability domain
monitoring during the model selection process.
from the absorbance recorded in the optical cell at 340 nm (e340
=
6.22 mmÀ1 cmÀ1). All the solutions were kept at 08C during the ex-
periments.
Recombinant human glutathione reductase (hGR) enzyme was pu-
rified and quantified (stock I, 0.124 mm), as described previous-
ly.[44,45] A hGR stock II solution was then prepared by dilution (1/10)
of the enzyme stock solution with hGR buffer (pH 6.9). In a Hellma
optical cell (l=1 cm), 902.9 mL of hGR buffer, 20 mL of methemoglo-
bin stock solution (final concentration of 8 mm), 20 mL of drug
stock solution (40 mm final concentration) and ~30 mL of NADPH
solution (120 mm final concentration) were pre-incubated in the
cell compartment of the absorption spectrophotometer at 258C
(Varian Cary Dual Cell Peltier thermostat). Then, 7.1 mL of the hGR
stock II solution (88 nm final concentration) was added to initiate
the reaction, which was monitored by UV/Vis absorption spectro-
photometry (200–800 nm) on an Agilent Cary 5000 spectropho-
tometer using the Scanning Kinetics method. A spectrum was
taken every 2 min between 300 and 650 nm over 2 h. For fast ki-
netics, the spectral window was limited to the Soret band of meth-
emoglobin (380–500 nm) and a spectrum was taken every 18 s
over 30 min. The kinetic data were processed with the Specfit pro-
gram.[46–51]
In addition, given the significantly larger training set size with re-
spect to the previous work, nonlinear modeling (SVM epsilon-re-
gression) was also created, using the evolutionary libsvm tuning
tool[59] for modeling. This ensures optimal selection of the descrip-
tor space (of both ISIDA and EED descriptor candidates) and of
SVM parameters yielding maximal robustness (i.e., best cross-vali-
dation performance) models. After optimization, EED descriptors
were selected as the most powerful and were used henceforth in
the SVM model.
The models were built by both methods following a three-fold
cross-validation[60] procedure repeated ten times and were addi-
tionally evaluated on the designated external set. For ISIDA de-
scriptors, a consensus model was applied with ISIDA/QSPR’s FMF
module. The top EED-based SVM model was deployed on the web
server application (http://infochim.u-strasbg.fr/webserv/VSEngi-
ne.html) and then used for external validation.
Datasets and modeling
Acknowledgements
This section provides an overview of the data preparation and
modeling workflow used in this work. The dataset described in our
previous work[22,52] was enriched by novel in-house compounds
and now consists of various benzo-, naphtho-, anthraquinones,[21]
and indolone-N-oxides[53] taken from literature, as well as previous-
ly synthesized in-house menadione derivatives[22] and novel com-
pounds described in this work. The equivalence of experimental
measurement conditions was established. In total, we selected
95 compounds measured under similar experimental conditions.
The structures were standardized (aromatization and charge-sepa-
rated representation of nitro groups) prior to modeling using the
ChemAxon Standardizer tool, version 14.10.20 (http://www.che-
maxon.org).
The authors thank the International Center for Frontier Research
in Chemistry (IC-FRC Innovation 2015 Program) in Strasbourg for
supporting our project entitled, “Computer-Aided Design of Novel
Antimalarial Naphthoquinones (CAD-NQ)”, and for creating
a proper framework for the scientific collaborations. This work
was supported in part by the Centre National de la Recherche
Scientifique (CNRS) and the University of Strasbourg (UMR 7509
CNRS-Unistra), Laboratoire d’Excellence (LabEx) ParaFrap (grant
LabEx ParaFrap ANR-11-LABX-0024 to E.D.-C.). P.S. and A.V. thank
the Program of Competitive Growth of the Kazan Federal Univer-
sity for support. E.D.-C. acknowledges former PhD students and
co-workers Holger Bauer (3h), Laure Johann (3o), and Kar›ne
Urgin (3r–3t) for compound preparation. The New York Universi-
ty Abu Dhabi Undergraduate Research Program is gratefully ac-
knowledged for providing summer research funding to I.D.
Two types of descriptors were used: Electronic Effect Descriptors
(EED)[22] and ISIDA fragment descriptors.[54] EED characterize the
global impact of the chemical environment on the reactivity of
a given key atom (K) in a molecule. They are computed as simple
additive topological distance-weighted contributions of each of
the nitrogen atoms of the compound. Here, the carbon atom in
the carboxyl group of the quinone or indolone moiety was chosen
as a key atom. In the case of the quinone ring, two carboxyl
Keywords: chemoinformatics
drugs · QSPR · redox potential
·
menadione
·
multitarget
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1350
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
[1] D. Reker, M. Seet, M. Pillong, C. P. Koch, P. Schneider, M. C. Witschel, M.
Rottmann, C. Freymond, R. Brun, B. Schweizer, B. Illarionov, A. Bacher,
M. Fischer, F. Diederich, G. Schneider, Angew. Chem. Int. Ed. 2014, 53,
7079–7084; Angew. Chem. 2014, 126, 7199–7204.
Moraes, V. F. Ferreira, F. C. de Abreu, A. V. Pinto, R. C. Montenegro, L. V.
Costa-Lotufo, M. O. F. Goulart, Toxicol. in Vitro 2012, 26, 585–594.
[32] P. S. Guin, S. Das, P. C. Mandal, Int. J. Electrochem. Sci. 2008, 3, 1016–
1028.
[2] O. Blank, E. Davioud-Charvet, M. Elhabiri, Antioxid. Redox Signaling
2012, 17, 544–554.
[33] D. O. Wipf, K. R. Wehmeyer, R. M. Wightman, J. Org. Chem. 1986, 51,
4760–4764.
[3] K. Ehrhardt, E. Davioud-Charvet, H. Ke, A. B. Vaidya, M. Lanzer, M. De-
ponte, Antimicrob. Agents Chemother. 2013, 57, 2114–2120.
[4] T. Müller, L. Johann, B. Jannack, M. Brückner, D. A. Lanfranchi, H. Bauer,
C. Sanchez, V. Yardley, C. Deregnaucourt, J. SchrØvel, M. Lanzer, R. H.
Schirmer, E. Davioud-Charvet, J. Am. Chem. Soc. 2011, 133, 11557–
11571.
[5] M. Bielitza, D. Belorgey, K. Ehrhardt, L. Johann, D. A. Lanfranchi, V. Gallo,
E. Schwarzer, F. Mohring, E. Jortzik, D. L. Williams, K. Becker, P. Arese, M.
Elhabiri, E. Davioud-Charvet, Antioxid. Redox Signaling 2015, 22, 1337–
1351.
[34] L. P. Hammett, J. Am. Chem. Soc. 1937, 59, 96–103.
[35] P. Zuman, Substituent Effects in Organic Polarography, Plenum Press,
New York, 1967.
[36] a) D. M. Veine, L. D. Arscott, C. H. Williams, Jr., Biochemistry 1998, 37,
15575–15582; b) “Flavoenzyme Structure and Function—Approaches
Using Flavin Analogues”: D. Edmondson, S. Ghisla, in Methods in Molec-
ular Biology, Vol. 131 (Eds.: S. K. Chapman, G. A. Reid), Humana, Totowa,
1999, pp. 157–179.
[37] F. W. Scheller, N. Bistolas, S. Liu, M. Jänchen, M. Katterle, U. Wollenberg-
er, Adv. Colloid Interface Sci. 2005, 116, 111 –120.
[6] D. Belorgey, D. A. Lanfranchi, E. Davioud-Charvet, Curr. Pharm. Des.
2013, 19, 2512–2528.
[38] I. M. Kolthoff, Treatise on Analytical Chemistry, Part 1, Theory and Practice,
Thermal Methods, Wiley, New York, 1993.
[7] L. Johann, D. A. Lanfranchi, E. Davioud-Charvet, M. Elhabiri, Curr. Pharm.
Des. 2012, 18, 3539–3566.
[39] A. Ashnagar, J. M. Bruce, P. L. Dutton, R. C. Prince, Biochim. Biophys. Acta
Gen. Subj. 1984, 801, 351–359.
[8] K. Ehrhardt, C. Deregnaucourt, A.-A. Goetz, T. Tzanova, B. Pradines, S. H.
Adjalley, S. Blandin, D. Bagrel, M. Lanzer, E. Davioud-Charvet, unpublish-
ed results.
[40] N. G. Tsierkezos, J. Solution Chem. 2007, 36, 289–302.
[41] K. Izutsu, Electrochemistry in Nonaqueous Solutions, Wiley-VCH, Wein-
heim, 2002.
[9] C. Hansch, A. Leo, Exploring QSAR: Fundamentals and Applications in
Chemistry and Biology, American Chemical Society, Washington DC,
1995.
[10] A. Tropsha, P. Gramatica, V. K. Gombar, QSAR Comb. Sci. 2003, 22, 69–
77.
[11] R. Todeschini, V. Consonni, R. Mannhold, H. Kubinyi, H. Timmermann,
Handbook of Molecular Descriptors, Wiley, Weinheim, 2008.
[12] F. Ruggiu, G. Marcou, A. Varnek, D. Horvath, Mol. Inf. 2010, 29, 855–868.
[13] A. Varnek, D. Fourches, D. Horvath, O. Klimchuk, C. Gaudin, P. Vayer, V.
Solov’ev, F. Hoonakker, I. V. Tetko, G. Marcou, Curr. Comput.-Aided Drug
Des. 2008, 4, 191–198.
[42] D. T. Sawyer, A. Sobkowiak, J. L. Roberts, Electrochemistry for Chemists,
Wiley, New York, 1995.
[43] V. Mirceski, S. Komorsky-Lovric, M. Lovric, Square-Wave Voltammetry:
Theory and Application, Springer, Berlin, 2007.
[44] A. Nordhoff, U. S. Bucheler, D. Werner, R. H. Schirmer, Biochemistry 1993,
32, 4060–4066.
[45] P. M. Färber, L. D. Arscott, C. H. Williams, K. Becker, R. H. Schirmer, FEBS
Lett. 1998, 422, 311 –314.
[46] H. Gampp, M. Maeder, C. J. Meyer, A. D. Zuberbühler, Talanta 1985, 32,
95–101.
[47] F. J. C. Rossotti, H. S. Rossoti, R. J. Whewell, J. Inorg. Nucl. Chem. 1971,
33, 2051–2065.
[48] H. Gampp, M. Maeder, C. J. Meyer, A. D. Zuberbühler, Talanta 1985, 32,
257–264.
[49] H. Gampp, M. Maeder, C. J. Meyer, A. D. Zuberbühler, Talanta 1986, 33,
943–951.
[50] D. W. Marquardt, J. Soc. Ind. Appl. Math. 1963, 11, 431–441.
[51] M. Maeder, A. D. Zuberbühler, Anal. Chem. 1990, 62, 2220–2224.
[52] a) E. Davioud-Charvet, T. Müller, H. Bauer, R. H. Schirmer (Centre Nation-
al de la Recherche Scientifique), Int. PCT Pub. No. WO 2009118327 A1
20091001, 2009; b) E. Davioud-Charvet, D. A. Lanfranchi, L. Johann, D. L.
Williams, E. C. Rodo (Centre National de la Recherche Scientifique), Int.
PCT Pub. No. WO 2012131010 A1 20121004, 2012.
[53] K. Reybier, T. H. Y. Nguyen, H. Ibrahim, P. Perio, A. Montrose, P. L. Fabre,
F. Nepveu, Biochemistry 2012, 88, 57–64.
[54] A. Varnek, D. Fourches, F. Hoonakker, V. P. Solov’ev, J. Comput.-Aided
Mol. Des. 2005, 19, 693–703.
[55] D. Horvath, G. Marcou, A. Varnek, J. Chem. Inf. Model. 2009, 49, 1762–
1776.
[56] I. V. Tetko, I. Sushko, A. K. Pandey, H. Zhu, A. Tropsha, E. Papa, T. Oberg,
R. Todeschini, D. Fourches, A. Varnek, J. Chem. Inf. Model. 2008, 48,
1733–1746.
[57] A. Varnek, D. Fourches, V. Solov’ev, O. Klimchuk, A. Ouadi, I. Billard, Sol-
vent Extr. Ion Exch. 2007, 25, 433–462.
[58] C. Agostinelli, J. Appl. Stat. 2002, 29, 825–840.
[59] D. Horvath, J. B. Brown, G. Marcou, A. Varnek, Challenges 2014, 5, 450–
472.
[14] E. Alpaydin, Introduction to Machine Learning, MIT Press, Cambridge,
2004.
[15] K. Alizadeh, M. Shamsipur, J. Mol. Struct. 2008, 862, 39–43.
[16] J. Cape, M. Bowman, D. Kramer, Phytochemistry 2006, 67, 1781–1788.
[17] M. Namazian, P. Norouzi, R. Ranjbar, J. Mol. Struct. 2003, 625, 235–241.
[18] M. Namazian, P. Norouzi, J. Electroanal. Chem. 2004, 573, 49–53.
[19] M. R. Hadjmohammadi, K. Kamel, P. Biparva, J. Solution Chem. 2011, 40,
224–230.
[20] P. Thanikaivelan, V. Subramanian, J. R. Rao, V. U. Nair, Chem. Phys. Lett.
2000, 323, 59–70.
[21] A. Behesti, P. Norouzi, M. R. Ganjali, Int. J. Electrochem. Sci. 2012, 7,
4811–4821.
[22] M. Elhabiri, P. Sidorov, E. Cesar-Rodo, G. Marcou, D. A. Lanfranchi, E. Da-
vioud-Charvet, D. Horvath, A. Varnek, Chem. Eur. J. 2015, 21, 3415–
3424.
[23] J. M. Anderson, J. K. Kochi, J. Am. Chem. Soc. 1970, 92, 1651–1659.
[24] a) N. Jacobsen, K. Torssell, Liebigs Ann. Chem. 1972, 763, 135–147; b) N.
Jacobsen, K. Torssell, Acta. Chem. Scand. 1973, 27, 3211–3216.
[25] C. Biot, H. Bauer, R. H. Schirmer, E. Davioud-Charvet, J. Med. Chem.
2004, 47, 5972–5983.
[26] K. Blumenstiel, R. Schçneck, V. Yardley, S. L. Croft, R. L. Krauth-Siegel,
Biochem. Pharmacol. 1999, 58, 1791–1799.
[27] a) O. Hammerich in Organic Electrochemistry Revised and Expanded,
5th ed. (Eds: O. Hammerich, B. Speiser), CRC, Boca Raton, Chap. 30,
pp. 1149–1201; b) W. H. Smith, A. J. Bard, J. Am. Chem. Soc. 1975, 97,
5203–5210.
[28] a) J. E. B. Randles, Faraday Soc. Trans. 1948, 44, 327–338; b) A. Sevcik,
Coll. Czechoslovak Chem. Commun. 1948, 13, 349–377.
[29] J. Mauzeroll, A. J. Bard, Proc. Natl. Acad. Sci. USA 2004, 101, 7862–7867.
[30] E. N. da Silva, Jr. , M. de Moura, A. V. Pinto, M. Pinto, M. de Souza, A. J.
Araujo, C. Pessoa, L. V. Costa-Lotufo, R. C. Montenegro, M. O. de Moraes,
V. F. Ferreira, M. O. F. Goulart, J. Braz. Chem. Soc. 2009, 20, 635–643.
[31] A. J. Araffljo, A. A. de Souza, E. N. da Silva, Jr., J. D. B. Marinho, M. de
Moura, D. D. Rocha, M. C. Vasconcellos, C. O. Costa, C. Pessoa, M. O. de
[60] D. M. Hawkins, S. C. Basak, D. Millsk, J. Chem. Inf. Comput. Sci. 2003, 43,
579–586.
Received: January 6, 2016
Published online on March 7, 2016
ChemMedChem 2016, 11, 1339 – 1351
www.chemmedchem.org
1351
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim