Anti-enteroviral activity of new MDL-860 analogues: Synthesis, in vitro/in vivo studies and QSAR analysis

Article abstract of DOI:10.1016/j.bioorg.2019.02.020

A series of 60 nitrobenzonitrile analogues of the anti-viral agent MDL-860 were synthesized (50 of which are new) and evaluated for their activity against three types of enteroviruses (coxsackievirus B1, coxsackievirus B3 and poliovirus 1). Among them, six diaryl ethers (20e, 27e, 28e, 29e, 33e and 35e) demonstrated high in vitro activity (SI > 50) towards at least one of the tested viruses and very low cytotoxicity against human cells. Compound 27e possesses the broadest spectrum of activity towards all tested viruses in the same way as MDL-860 does. The most active derivatives (27e, 29e and 35e) against coxsackievirus B1 were tested in vivo in newborn mice experimentally infected with 20 MLD₅₀ of coxsackievirus B1. Compound 29e showed promising in vivo activity (protection index 26% and 4 days lengthening of mean survival time). QSAR analysis of the substituent effects on the in vitro cytotoxicity (CC₅₀) and anti-viral activity of the nitrobenzonitrile derivatives was carried out and adequate QSAR models for the anti-viral activity of the compounds against poliovirus 1 and coxsackievirus B1 were constructed.

Full text of DOI:10.1016/j.bioorg.2019.02.020

BioorganicChemistry85(2019)487–497
Contents lists available at ScienceDirect
Bioorganic Chemistry
journal homepage: www.elsevier.com/locate/bioorg
Anti-enteroviral activity of new MDL-860 analogues: Synthesis, in vitro/in
vivo studies and QSAR analysis
Ivanka Nikolovab, Ivaylo Slavcheva, Martin Ravutsova, Miroslav Dangalova, Yana Nikolovaa,
Irena Zagranyarskaa, Adelina Stoyanovab, Nadya Nikolovab, Lucia Mukovab, Petar Grozdanovb,
Rosica Nikolovac, Boris Shivachevc, Victor E. Kuz'mind,e, Liudmila N. Ognichenkod,e
,
⁎
⁎
Angel S. Galabovb, , Georgi M. Dobrikova,
a Institute of Organic Chemistry with Centre of Phytochemistry, Bulgarian Academy of Sciences, bl. 9, Acad. G. Bonchev Str., Sofia 1113, Bulgaria
b Stephan Angeloff Institute of Microbiology, Bulgarian Academy of Sciences, bl. 26, Acad. G. Bonchev Str., Sofia 1113, Bulgaria
c Institute of Mineralogy and Crystallography, Bulgarian Academy of Sciences, bl. 107, Acad. G. Bonchev Str., Sofia 1113, Bulgaria
d A.V. Bogatsky Physical-Chemical Institute NAS of Ukraine, Department of Molecular Structure and Chemoinformatics, Odessa, Ukraine
e Odessa National Polytechnic University, Department of Theoretical Foundation of Chemistry, Odessa, Ukraine
A R T I C L E I N F O
A B S T R A C T
Keywords:
A series of 60 nitrobenzonitrile analogues of the anti-viral agent MDL-860 were synthesized (50 of which are
new) and evaluated for their activity against three types of enteroviruses (coxsackievirus B1, coxsackievirus B3
and poliovirus 1). Among them, six diaryl ethers (20e, 27e, 28e, 29e, 33e and 35e) demonstrated high in vitro
activity (SI > 50) towards at least one of the tested viruses and very low cytotoxicity against human cells.
Compound 27e possesses the broadest spectrum of activity towards all tested viruses in the same way as MDL-
860 does. The most active derivatives (27e, 29e and 35e) against coxsackievirus B1 were tested in vivo in
newborn mice experimentally infected with 20 MLD50 of coxsackievirus B1. Compound 29e showed promising in
vivo activity (protection index 26% and 4 days lengthening of mean survival time). QSAR analysis of the sub-
stituent effects on the in vitro cytotoxicity (CC50) and anti-viral activity of the nitrobenzonitrile derivatives was
carried out and adequate QSAR models for the anti-viral activity of the compounds against poliovirus 1 and
coxsackievirus B1 were constructed.
Coxsackieviruses
QSAR
Synthesis
Nitrobenzonitriles
MDL-860
Anti-enteroviral
1. Introduction
initially linked to human disease). New strains of coxsackievirus B1
(CVB1), enterovirus-A71 (EV-A71), and enterovirus-D68 (EV-D68) have
Enteroviruses are members of the Picornaviridae family, comprising
small non-enveloped viruses with single stranded positive sense RNA
genome. They are usually agents of mild infections but also cause en-
cephalitis, myocarditis, poliomyelitis, acute heart failure, and diabetes
mellitus. Enteroviruses are subject to significant changes over time
because of errors introduced during genome replication. Intraspecies
recombination between enteroviruses is also common, further pro-
moting genetic diversity. This genetic plasticity allows for widespread
epidemics and sporadic outbreaks to occur. Enteroviruses are now
classified into 15 distinct species. Among them are polioviruses (causal
agents of poliomyelitis in humans and nonhuman primates), coxsackie
A viruses (associated with herpangina, human central nervous system
disease, and flaccid paralysis in suckling mice), coxsackie B viruses
(human central nervous system and cardiac disease, diabetes, spastic
paralysis in mice), and the echoviruses (nonpathogenic in mice, and not
emerged as causes of recent outbreaks in the United States, South-
Eastern Asia, and other countries, including more severe disease man-
ifestations than previously described. A recent outbreak of CVB1 has
once again demonstrated the epidemic potential of enteroviruses. In
mid-2007, cases of severe neonatal disease due to CVB1 were re-
cognized nearly simultaneously in several USA cities. In general, this
virus was recognized as one of the most commonly circulating en-
teroviruses in USA between 2009 and 2013 [1].
Significant progress has been made in the global effort to interrupt
poliovirus transmission and eradicate polio. However, attempts to
eliminate poliovirus 1 (PV1) circulation are still running in countries
like Afghanistan, Pakistan and Tadjikistan, but progress has been de-
layed by factors that have made vaccination unavailable for approxi-
mately 5–25% of children in the region [1]. In addition, to the best of
our knowledge, an efficient and approved chemotherapy against
⁎ Corresponding authors.
E-mail addresses: galabov@microbio.bas.bg (A.S. Galabov), gmdob@orgchm.bas.bg (G.M. Dobrikov).
https://doi.org/10.1016/j.bioorg.2019.02.020
Received 22 November 2018; Received in revised form 3 February 2019; Accepted 6 February 2019
Availableonline12February2019
0045-2068/©2019ElsevierInc.Allrightsreserved.

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
The aim of the present study is the synthesis of new MDL-860
analogues possessing unaltered 2-cyano-5-nitro substituted benzene
ring as a common fragment, in order to prove limits of possible varia-
tions in the other ring of MDL-860, leading to improved antiviral ac-
tivity.
2. Results and discussion
2.1. Chemistry
Four series of MDL-860 analogues were synthesized and evaluated
for anti-viral activity. Detailed synthetic procedures and analytical data
are presented in Supplementary data. The synthesis of non-commercial
starting compounds was also described in Supplementary data. All
target compounds were obtained through simple one-step nucleophilic
aromatic substitution reactions of series of phenols (1a-36a), thiols (1b-
9b), amines (1c-11c) and N-heterocycles (1d-3d) with 2-chloro-5-ni-
trobenzonitrile (1) in presence of a base. Due to the electron-deficient
nature of 1, chlorine substitution was performed in relatively mild
conditions with no catalyst needed [12]. All compounds were purified
by column chromatography and/or recrystallization.
Fig. 1. Structure of MDL-860.
polioviruses (as well as against coxsackieviruses) does not exist [2].
Such therapy could be complementary to vaccination, especially in
problematic regions worldwide.
The evolution of drug resistance is a challenge for successful de-
velopment of new anti-enteroviral agents with promising activity and
potential for further development [3]. Nevertheless, few compounds
have been in clinical trials so far – disoxaril, pleconaryl, pirodavir and
its analogues, some isoxazoles, imidazolidinones, chalcones and fla-
vanes [4]. Apart from drug-resistance, some additional issues were
observed – drug-drug interactions, low in vivo activity, side effects etc.
[1,5]. Recent studies have shown reborn interest in the synthesis and in
vitro evaluation of new small molecules active against enteroviruses
(PV1, coxsackievirus B3 (CVB3) and coxsackievirus B5) [6,7]. Diary-
lether MDL-860 (2-(3,4-dichlorophenoxy)-5-nitrobenzonitrile, also
known as DNB) (Fig. 1) was firstly reported in the 1980’s. Later on, it
was recognized that MDL-860 and its analogues possess a broad-spec-
trum of in vitro activity against picornaviruses along with very low
cytotoxicity toward human cells [8,9].
The synthesis of aryl ethers 1e-36e (Table 1) and aryl thioethers 1f-
10f (Table 2) from 1 and the corresponding phenols (1a-36a) and
thiophenols (1b-9b) was performed at 80 °C in dry DMSO generally
using powdered KOH as a base. In some cases NaOH or K2CO3 were
used instead (15e, 20e, 21e, 36e, 5f and 8f). Reaction progress was
monitored by TLC. Compound 11e was obtained from 1 and in situ
generated CF3CH2ONa (from dry CF3CH2OH and NaH) in refluxing
CF3CH2OH. Similar procedure was applied for the preparation of benzyl
ether 12e. Compound 10f was obtained as a single product through
spontaneous cyclisation during the reaction between 1 and mercapto-
benzimidazole (9b) [13]. The aryl thioether 9f was therefore not iso-
lated. It should be mentioned that the preparation of compound 36e (2-
(2,5-diiodophenoxy)-5-nitrobenzonitrile) was initially not aimed. In-
stead, we had planned to obtain a 3,4-diiodo-analogue of MDL-860.
Unfortunately, the synthesis of this derivative from 3-iodophenol
turned out to be quite challenging. All attempts led to isolation of 2,5-
diiodophenol (36a) with ca. 95% purity. It was not possible to further
purify compound 36a, thus it was used for the synthesis of 36e as it
was. The purification of compound 36e was a serious challenge as well.
Column chromatography purification followed by several successive
recrystallizations led to isolation of 36e with ca 95% purity. According
Our recent studies were focused on the mechanism of action of
MDL-860 [10] and further development of new MDL-860 analogues
(Fig. 2) active against CVB1, CVB3 and PV1 [11]. In the light of current
research [8–11], it is clear that most promising is the development of
new MDL-860 analogues bearing unaltered 2-cyano-4-nitro ring. In our
previous work, initial screening of twelve MDL-860 analogues resulted
in three compounds (A1, A7 and A8) with high activity against PV1 and
CVB1 and only two compounds (A2 and A3) with moderate activity
towards CVB3. In addition, compounds A1, A7 and A8 exhibited ac-
tivity towards CVB1 experimental neuroinfection in newborn mice
[11].
Fig. 2. Recently published analogues of MDL-860.
488

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Table 1
Synthesis of diaryl ethers 1e-36e.
ROH
ROe
Ether
ROH
ROe
Ether
ROH
ROe
Ether
1a
1e
13a
13e
25a
25e
2a
3a
2e
3e
14a
15a
14e
15e
26a
27a
26e
27e
4a
5a
4e
5e
16a
17a
16e
17e
28a
29a
28e
29e
6a
7a
6e
7e
18a
19a
18e
19e
30a
31a
30e
31e
8a
8e
20a
20e
32a
32e
9a
9e
21a
22a
21e
22e
33a
34a
33e
34e
10a
10e
11a
12a
11e
12e
23a
24a
23e
24e
35a
36a
35e
36ea
a
Compound 36e contains ca. 5% of the 3,4-diiodo isomer.
to X-ray data obtained (see Section 2.2) on crystal grown form that
mixture compound 36e was found to contain 5% of its 3,4-diiodo
substituted analogue, thus indicating the presence of 3,4-diiodophenol
in 36a.
Table 2
Synthesis of thioethers 1f-10f.
A series of arylamines 1g-11g (Table 3) were synthesized through
reaction of 1 with amines 1c-11c at 100 °C in a mixture of dry N,N-
diisopropylethylamine (DIPEA) and N-methylmorpholine (NMM).
Amine 1g was synthesized from 1 and dry DMF (as convenient in situ
source of dimethylamine) at 130 °C according to a described procedure
[14]. In some cases (6g, 7g and 8g) decomposition products (black tar)
and unreacted 1 were observed. N-arylation of heterocycles 1d-3d ap-
plying common conditions (NaH/dry DMSO) afforded the corre-
sponding compounds 1h-3h in good to excellent yields (Scheme 1). A
trifluoromethyl substituted analogue (4) of MDL-860 was prepared
from 1-fluoro-4-nitro-2-(trifluoromethyl)benzene (2) and 3,4-di-
chlorophenol (3) in DMSO (Scheme 2).
RSH RS─
Thio- RSH RS─
Thio- RSH RS─
Thio-
ether
ether
ether
1b
2b
1f
4b
5b
4f
7b
8b
7f
2f
3f
5f
8f
3b
6b
6f
9b
9fa
10f
It should be pointed out that most of the compounds obtained in this
study were synthesized for the first time. The synthesis of compounds
4e [15], 11e [16–20], 1g [12,14,21–25], 8g [12,26], 10g [27] and 1h
[25] was described elsewhere in different context studies, not being
related to the present biological investigations. The anti-viral activity of
MDL-860 [28], 16e [28], 17e [28], 23e [8,28] and 1f [28] against
other enteroviruses is discussed in Section 2.3.
a
Not isolated.
489

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Table 3
Synthesis of arylamines 1g-11g.
Amine
R1R2Ne
Aryl-amine
Amine
R1R2Ne
Aryl-amine
Amine
R1R2Ne
Aryl-amine
1c
1g
5c
5g
9c
9g
2c
2g
6c
6g
10c
11c
10g
11g
3c
4c
3g
4g
7c
8c
7g
8g
2.2. X-ray
Previously published data for compounds A1-A12 [11] are presented
for comparison. It was demonstrated that compound 27e (by analogy
with MDL-860) possesses the broadest spectrum of activity in this study
(against PV1, CVB1 and CVB3). Compound 35e was active against
CVB1 and CVB3. Significant activity towards PV1 and CVB1 demon-
strated by 28e. Compounds 20e and 29e were effective only against
CVB1 and 22e, 30e and 33e – only against PV1. Moderate activity
against PV1 and CVB1 was demonstrated by compound 4e; against PV1
– only by 17e, 18e, 24e and 31e; compound 13e was active against
CVB1 and 14e – against CVB3. It should be mentioned that the anti-
viral activity of 16e, 17e, 23e and 1f was investigated in the early
1980‘s against rhinoviruses and Coxsackie A21 virus [8,28]. Their ac-
tivities were generally higher compared to those observed against PV1,
CVB1 and CVB3 in this study (Table 5). Only MDL-860 could be con-
sidered to possess wide spectrum of activity – it demonstrated high in
vitro activity against all tested rhinoviruses and coxsackie viruses.
Among the thioethers, 3f and 8f demonstrated weak activity against
CVB1. The main disadvantage of 8f appears to be its high cytotoxicity
(CC50 = 18.7 μM). Since thioethers are able to oxidize easily in biolo-
gical media, it is not clear whether 3f and 8f are the active compounds
or just prodrugs. Thus, further investigation of sulfone analogues of 3f
and 8f is necessary.
Application of single crystal X-ray diffraction was used in this study.
This method was necessary in order to confirm the structure of 36e, on
the other hand the results obtained were useful to elucidate the struc-
ture of the impurity in 36e. Thus, the crystal structure of compound
36e was elucidated by single crystal X-ray diffraction (Fig. 3). Single
crystals were obtained by slow evaporation of a concentrated solution
of 36e in isopropanol. The most important crystallographic data and
refinement parameters for 36e are shown in Table 4, while bond dis-
tances, angles, torsion angles and other details about structure solution
and refinement are listed in Supplementary data (Tables S1 and S2,
Figs. S1 and S2). The crystal structure revealed the presence of impurity
(ca. 5%) of 2-(3,4-diiodophenoxy)-5-nitrobenzonitrile. It is interesting
to note that I4 shifts from its “original” I2 position and the distance
between iodines I3 and I4 from the minor component is 3.918 Å The
hypothetical I2…I3 distance being 3.232 Å. The tow ring systems
(diiodophenoxy and nitrobenzonitrile) are essentially planar (rmsd of
0.01 Å for both) though the angle between their mean planes is 81.5°
e.g. the bridging O1 allows rotation of the ring systems along CeO1
bond.
2.3. Virology
The data presented in Table 5 unambiguously show that even small
changes in the MDL-860 molecule dramatically influence the in vitro
activity. Interestingly, new active compounds could be found ex-
clusively among the diarylethers. Exploring in more detail the group of
2-cyano-5-nitro substituted ethers (compounds A1-A5, A7, A8, A10, A-
The newly synthesized MDL-860 derivatives (1-36e, 1-10f, 1-11g,
1-3h and 4) were subjected to in vitro screening study for activity to-
wards PV1, CVB1 and CVB3 (Table 5). The CPE inhibition test was
used, following the procedure of Borenfreund and Puerner [29].
Scheme 1. Synthesis of N-arylated heterocycles 1h-3h.
490

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Scheme 2. Synthesis of –CF3 derivative 4.
Fig. 3. ORTEP drawing of compound 36e showing the atomic numbering system and the observed disorder of the iodine’s (minor component of 5.58% is shown as
dashed lines).
11 and 1e-36e), it is clear that the presence of two and three halogen
substituents at different positions in the secondary benzene ring is op-
timal for high activity. On the other hand, poly-halogenated ethers (A2,
15e-17e) are inactive. Most of the diarylethers (with a few exceptions:
12e, 15e, 18e, 24e, 32e) are showing very low toxicity. In the light of
the recently discovered mechanism of action of MDL-860 [10], one
could argue that all active diarylethers have a similar mechanism of
action, namely, an irreversible covalent modification of phosphatidy-
linositol-4 kinase III beta (PI4KB). PI4KB is one of the most important
enzymes in mammals, responsible for replication of enteroviruses in the
host cells [30]. It could be assumed that alteration of the halogen
substituents in diarylethers is important. This may cause small changes
in the shape and geometry of the molecules but may impact sig-
nificantly the PI4KB modification and in vitro activity, respectively.
Other type of substituents (e.g. A4, A5, 1e-3e, 5e, 6e, 13e, 14e), or
the presence of heterocyclic moieties (A10, A11, 4e, 7e-10e) instead of
benzene ring, generally led to lack of activity. Other series of com-
pounds (1f-10f, 1g-11g and 1h-3h, containing different bridge het-
eroatoms, i.e. S, N) were completely inactive (except for 8f). It is no-
teworthy that even very close isosteric analogues of MDL-860 (like
thioether 1f or ethers 4 and 12e) are also inactive. Probably these
compounds are not able to modify PI4KB or they undergo biochemical
transformations before reaching the enzyme.
Table 4
Most important crystallographic and data refinement parameters for compound
36e.
Empirical formula
C13H6I2N2O3
Formula weight
Temperature/K
Crystal system
Space group
492.00
290
Monoclinic
P21/c
a/Å
21.1900(7)
b/Å
5.60500(10)
c/Å
13.2171(4)
α/°
90
β/°
106.026(4)
γ/°
90
Volume/Å3
1508.79(8)
Z
4
ρcalc g/cm3
2.166
μ/mm−1
32.829
F(0 0 0)
912.0
Crystal size/mm3
Radiation
0.3 × 0.25 × 0.12
Cu Kα (λ = 1.54184)
2Θ range for data collection/°
Index ranges
8.684–148.524
−22 ≤h ≤ 26, −6 ≤ k ≤ 6, −15 ≤ l ≤ 16
Reflections collected
Independent reflections
Data/restraints/parameters
Goodness-of-fit on F2
Final R indexes [I > = 2σ (I)]
Final R indexes [all data]
Largest diff. peak/hole/e Å−3
CCDC number
9013
3015 [Rint = 0.0544, Rsigma = 0.0431]
3015/0/202
1.026
Compound 4 is the only MDL-860 analogue in this study, possessing
a different substituent in the primary aromatic ring (eCF3 instead of
eCN). The role of the substituents in this ring is still unclear and further
studies of such series of compounds is necessary. For example, some
published results [8] show that replacement of the cyano group with
carboxyl group in MDL-860 leads to carboxylic acid with good antiviral
activity. Moreover, this replacement automatically allows improvement
R
1 = 0.0511, wR2 = 0.1304
R1 = 0.0706, wR2 = 0.1456
1.11/−1.32
1,876,618
491

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Table 5
In vitro screening data for anti-enteroviral activity.
Compounda
Cytotoxicity CC50 (μM)
PV1
CVB1
CVB3
IC50 (μM)
SI
IC50 (μM)
SI
IC50 (μM)
SI
MDL-860
A1
493.0
320.0
119.1
570.3
22.0
6.8
2.7
NA
NA
NA
NA
NA
6.8
2.7
NA
NA
NA
234.0
NA
NA
NA
30.6
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
79.0
22.8
1.0
11.0
32
72.5
118.0
−
0.8
0.8
10.9
256.1
NA
16.2
NA
0.7
0.75
NA
NA
190.1
152.2
NA
NA
NA
22.8
NA
NA
NA
NA
NA
142.0
NA
NA
12.7
NA
NA
NA
NA
NA
NA
2.1
NA
NA
NA
NA
NA
NA
6.4
6.1
2.9
NA
NA
NA
NA
NA
3.7
NA
NA
NA
24.9
NA
NA
NA
NA
3.1
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
586.9
405.0
10.9
2.2
−
2.7
NA
5.8
29.6
NA
NA
426.0
NA
NA
9.9
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
54.0
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
6.8
NA
NA
NA
NA
NA
NA
NA
1.0
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
182.0
−
A2
20.5
19.6
−
A3
−
A4
−
A5
94.3
−
5.8
−
−
A6
675.0
355.0
517.5
53.8
−
1.5
−
A7
52.2
191.6
−
507.1
690.0
−
A8
−
A9
5.4
−
A10
A11
A12
1e
718.0
570.8
492.2
123.0
175.0
123.0
332.0
346.6
280.0
450.0
576.0
423.0
367.0
165.0
10.1
−
−
−
3.0
3.2
−
−
2.1
−
−
−
2e
−
−
−
3e
−
−
−
4e
11.5
−
14.5
−
−
5e
−
6e
−
−
−
7e
−
−
−
8e
−
−
−
9e
−
−
−
10e
11e
12e
13e
14e
15e
16e
17e
18e
19e
20e
21e
22e
23e
24e
25e
26e
27e
28e
29e
30e
31e
32e
33e
34e
35e
36e
1f
−
2.5
−
−
−
−
−
−
−
291.0
680.0
47.1
−
22.9
−
−
−
12.5
−
−
−
195.0
287.0
18.7
2.4
12.6
18.7
8.6
6.2
−
−
−
−
−
−
−
95.0
−
−
199.0
572.0
219.0
272.0
30.7
95
−
NA
6.8
NA
1.8
NA
NA
11.0
2.7
NA
4.3
10.0
5.2
1.0
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
255.0
NA
NA
NA
NA
NA
−
−
32.2
−
−
−
−
−
17.0
−
−
−
92.0
−
−
547.0
785.0
234.0
342.0
200.0
155.0
30.5
−
−
−
71.3
86.6
−
122.6
38.3
117.9
−
115.4
−
−
46.0
15.5
5.8
107
−
−
−
−
−
−
107.0
215.0
493.0
273.0
132.0
211.0
187.5
161.3
13.6
−
−
−
−
−
133.2
−
493.0
−
−
−
−
−
2f
−
−
−
3f
−
7.5
−
−
4f
−
−
5f
−
−
−
6f
14.6
−
−
−
7f
16.5
−
−
−
8f
18.7
−
6.0
−
−
10f
1g
349.2
651.0
332.6
336.7
495.4
617.0
12.6
−
−
−
−
−
2g
−
−
−
3g
−
−
−
4g
−
−
−
5g
2.4
−
−
−
6g
−
−
7g
346.6
199.0
55.4
−
−
−
8g
−
−
−
9g
−
−
−
10g
336.4
−
−
−
(continued on next page)
492

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Table 5 (continued)
Compounda
Cytotoxicity CC50 (μM)
PV1
CVB1
CVB3
IC50 (μM)
SI
IC50 (μM)
SI
IC50 (μM)
SI
11g
1h
2h
3h
4
454.0
332.8
94.2
NA
NA
NA
NA
NA
−
−
−
−
−
NA
NA
NA
NA
NA
−
−
−
−
−
NA
NA
NA
NA
NA
−
−
−
−
−
339.9
22.8
NA – not active.
a
MDL860 was used as a reference compound; compounds A1-A12 were already published in respect of their synthesis and activity against PV1, CVB1 and CVB3
[11].
Table 6
ether structures imply extremely poor solubility in water. The nature of
the substituents does not allow chemical modification of the active
compounds in order to improve solubility and/or membrane transport
(e.g. conversion to prodrugs – salts, esters, etc.). Further formulation of
the in vitro active compounds through preparation of nanoparticles or
complexes with water soluble polymers could increase significantly the
in vivo effects.
Study of the in vivo activity of compounds 27e, 29e, and 35e against CVB1
experimental neuroinfection in newborn mice. Data are from three independent
experiments (average).
Compound/Dose Survivors/
MST
SD,
Δ, days Mortality, % PI, %
Total
daysa
27e/25 mg/kg
27e/50 mg/kg
29e/25 mg/kg
29e/50 mg/kg
35e/25 mg/kg
35e/50 mg/kg
MDL-860
0/16
3/21
6/23
0/26
0/17
1/23
0/27
0/16
3.2
5.4
7.0
3.1
6.1
4.6
6.1
3.0
0.5ns
0.5ns
1.0**
0.4 ns
0.8*
0.2
100
86
0
+2.4
+4.0
+0.1
+3.1
+1.6
+3.1
−
14.2
26
0
74
2.4. QSAR analysis
100
100
96
0
0.6ns
0.9*
4.3
0
QSAR analysis of substituent effects on the in vitro cytotoxicity
(CC50) and anti-viral activity (against PV1, CVB1, CVB3) of the 5-ni-
trobenzonitrile derivatives was carried out. A dataset consisting of 72 5-
nitrobenzonitrile derivatives and one 3-trifluoromethylnitrobenzene
derivative was used in this study. Structural descriptors for all in-
vestigated compounds were calculated using the simplex representation
of molecular structure (SiRMS) approach [31,32]. The calculation of
descriptors was carried out at the 2D level of molecular structure re-
presentation using the Dragon program. In this case only molecular
topology is taken into account, i.e. all information is extracted from the
structural formula. It should be noted that 2D-QSAR models are the
most popular in structure-property studies [31,32]. The efficiency of
such models is due to the fact that the topological model of the mole-
cular structure implicitly contains information about the possible con-
formations of the molecule.
100
100
Placebo
0.3
0
a
One-way ANOVA (Bonferroni’s multiple comparison post-test); MST –
mean survival time; PI – protection index; SD – standard deviation.
** p < 0.01vs. placebo group.
* p < 0.05 vs. placebo group; ns – not significant.
of water solubility through possible formation of salts.
The most active derivatives (27e, 29e and 35e) against CVB1 were
tested for in vivo activity in newborn mice experimentally infected with
20 MLD50 CVB1. Compounds 27e, 29e and 35 were administered
subcutaneously as daily doses of 25 and 50 mg/kg following 12-days
course since the day of viral inoculation. The results obtained showed
moderate protective effects of 27e and 29e. A marked lengthening of
the mean survival time was observed for 29e (25 mg/kg) and 35e
(25 mg/kg) (Table 6 and Fig. 4). Taking into account this lack of ac-
tivity, along with the previously reported promising results for com-
pounds A1 (PI 50%), A7 (PI 33%) and A8 (PI 11%) [11], it could be
suggested that there is no correlation between in vitro and in vivo ac-
tivity. Since the pharmacological properties and especially the me-
chanisms of transport across the cell membranes for these diaryl ethers
are unknown, it is difficult to explain these results. Moreover, the diaryl
The simplex approach is based on isolating and counting the
number of molecular fragments (pairs, triples, quadruples of atoms) in
which a certain sequence of changes in some property is observed. That
is, in the framework of SiRMS, any molecule can be represented as a
system of different specific fragments (simplexes) of fixed composition
and topology. Various atomic characteristics can be used for the vertex
differentiation in the simplex, such as the uniqueness of the atom (atom
nature or
a more detailed type), partial charge, lipophilicity,
Fig. 4. Individual effects of compounds 27e, 29e, 35e and MDL-860 in experimental neurotropic infection with Coxsackievirus B1 in newborn mice.
493

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
electronegativity, refraction, van der Waals interactions, H-bond
donor/acceptor potential, etc. For atomic characteristics having real
values (refraction, electronegativity, etc.) at the preliminary stage, the
range of values is divided into a certain number of groups. The number
of groups (G) is a tuning parameter of models and can vary (as a rule
G = 3–7). In addition, electronegativity, refraction, molecular weight
and octanol-water partition coefficient (LogP) are calculated as integral
descriptors that describe the whole molecule.
Scrambling procedure were lower in indices than in the final model:
R2(scr) = 0.21
0.02, Q2(scr) = 0.10
0.02. Thus, the non-randomness
of the established relationship between the structure of the studied
compounds and their cytotoxicity (Fig. 5) can be stated. Nevertheless,
even the approximated QSAR model is not of sufficient quality to pre-
dict cytotoxicity. We used it as an auxiliary for a qualitative assessment
of the effect of substituents on cytotoxicity only (see Fig. 6).
The clear mechanistic interpretation is one of the advantages of
SiRMS approach [31]. On the basis of developed QSAR models the in-
fluence of each atom over a particular property can be calculated. The
contribution of each atom in the molecule can be defined as the ratio of
the sum of PLS regression coefficients for all simplexes containing this
atom to the number of atoms in the simplex. The atomic contribution
depends on the number of simplexes that include this atom. The
number of simplexes is not constant. It varies in different molecules and
depends on other constituents. Thus, this contribution is non-additive.
The analysis of such information allows selecting different fragments
which have negative or positive influence on a considered property.
The relative influence of different substituents in the 2-position of
the 5-nitrobenzonitrile moiety on cytotoxicity (CC50) is shown in Fig. 6.
It could be stated that the introduction of fluorine and chlorine atoms
into the aromatic ring promotes greater cytotoxicity.
Moreover, certain parameters calculated by Dragon program were
also used [33]. Most of the Dragon descriptors are integral structural
characteristics of the molecule, but many of them are difficult to in-
terpret. Thus, simplex descriptors and Dragon descriptors were used for
the development of models. A total of about 8000 structural descriptors
were calculated for the evaluated molecules. The relationships between
the calculated molecular descriptors and the investigated properties of
these molecules were established by methods of partial least squares
(PLS) [34] and random forest (RF) [35]. Primarily, mutually-correlated
and constant parameters were eliminated. Then the procedure of Trend
Vector [36] was used to form initial sets of molecular descriptors. The
procedure of Genetic Algorithm was used for development of PLS-
models [37].
In this study, QSAR model reflecting the structural influence of in-
vestigated compounds on their cytotoxicity was developed. In the
preliminary model development it was found that compound 27e is an
outlier, so this molecule was excluded from the dataset. Thus, 72 mo-
lecules were included in the training set for model development. QSAR
model was built using the PLS method with three latent variables
(based on 25 descriptors in the final) and with the following statistical
characteristics: determination coefficient for the training set R2 = 0.82,
coefficient of determination for cross-validation (leave one out)
Q2 = 0.70, standard error Sts = 84, Scv = 110 for training set and cross-
validation, respectively. A “randomization” procedure (Y-Scrambling)
was used to confirm the “non-randomness” of the developed QSAR
model [31]. The statistical characteristics obtained using the Y-
Further in this study, QSAR models reflecting the structural influ-
ence of the investigated compounds on the in vitro activity (IC50
)
against poliovirus 1 (PV1) were developed. It was found that among the
73 compounds, only 20 exhibit anti-PV1 activities. Thus, the original
activity values were coded as follows: 1-active and 0-inactive. In this
case, RF method was used for decision of classification task. RF models
were constructed according to the described original RF algorithm [35].
RF is an ensemble of single decision trees. Each tree has been grown as
follows: (i) A bootstrap sample, which will be a training set for the
current tree, is produced from the whole training set of N compounds.
Compounds which are not in the current tree training set are placed in
an out-of-bag (OOB) set (OOB set size is ∼N/3). (ii) The best split by
Fig. 5. Observed versus predicted diagram of cytotoxicity (CC50) values for 73 molecules; compound 27e is an outlier.
494

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Fig. 6. Relative influence of different substituents in the 2-position of 5-nitrobenzonitrile on cytotoxicity (CC50).
Table 7
Table 8
Statistical parameters for classification models PV1.
Statistical parameters for classification models CVB1.
No
OBB set
MCC
AC
SP
SE
No
OBB set
MCC
AC
SP
SE
1
2
3
20 + 53 = 73
0.58
0.90
0.90
0.84
0.95
0.95
0.91
0.91
0.89
0.65
1.0
4
5
6
19 + 54 = 73
0.24
0.90
0.71
0.75
0.95
0.93
0.93
0.91
0.85
0.26
1.0
20 + 20 + 53 = 93
20 + 20 + 20 + 53 = 113
19 + 19 + 54 = 92
19 + 19 + 19 + 54 = 111
1.0
1.0
Matthew’s
correlation
coefficient:
MCC = (TP × TN − FP × FN)/
((TP + FP) × (TP + FN) × (TN + FP) × (TN + FN))1/2
Accuracy: AC = (TP + TN)/(TP + TN + FP + FN).
Specify: SP = TN/(TN + FP).
.
the count of active ones was duplicated. In the first series, 20 active
molecules (all of active and inactive – 73 molecules) were used; in the
second series the count of active molecules was increased twofold, i.e.
40 active molecules (a total of 93 molecules); in the third series the
count of active molecules was increased threefold – 60 molecules (a
total of active and inactive – 113 molecules).
Sensitivity: SE = TP/(TP + FN).
TP = true positive; TN = true negative; FP = false positive; FN = false nega-
tive.
The resulting QSAR models for the training set showed an un-
mistakable classification. The predictive ability of the QSAR models
was evaluated using the “out-of-bag” (OOB) procedure [35]. The
quality of the classification models was assessed according to the fol-
lowing statistical characteristics (Table 7):
classification and regression tree (CART) algorithm [35] among the m
randomly selected descriptors from whole set of M ones in each node is
chosen. The value of m is just one tuning parameter for which RF
models are sensitive. (iii) Each tree is grown to the largest possible
extent (there is no pruning).
As it can be seen from Table 7, the balancing of models leads to a
significant quality improvement. Model 2 could be considered as the
most appropriate.
Since the dataset is unbalanced, i.e. the count of active and inactive
molecules is significantly different, a special procedure for balance was
used. The count of inactive molecules was constant (53 molecules) and
Fig. 7. Relative influence of different substituents in the 2-position of 5-nitrobenzonitrile on the activity against PV1.
495

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Fig. 8. Relative influence of different substituents in the 2-position of 5-nitrobenzonitrile on the activity against CVB1.
Fig. 9. General formula of all obtained compounds (left formula) and active 2-phenoxy-5-nitrobenzonitriles (right formula).
Interpretation analysis of the QSAR models allowed for estimation
860 were synthesized and evaluated for activity against different types
of enteroviruses. All compounds contain a 5-nitrobenzonitrile moiety
bridged through a heteroatom (O, N and S) to a second aromatic ring
bearing different substituents (Fig. 9, left formula). The compounds
were subjected to an in vitro screening study for activity towards PV1,
CVB1 and CVB3 viruses. The most active ones (6 compounds with
SI > 50) were found among the diarylether derivatives (O as a bridge
heteroatom). Moreover, the nature of the substituents in the secondary
aromatic ring was found to be crucial for the anti-viral activity – only
diarylethers containing two to three halogen substituents in the sec-
ondary benzene ring demonstrated promising anti-viral activity. Other
type of substituents or the presence of heterocyclic moieties instead of a
benzene ring generally led to lack of activity. The most active against
CVB1 derivatives (27e, 29e and 35e) were tested for in vivo activity in
newborn mice experimentally infected with 20 MLD50 of CVB1. Com-
pound 29e showed promising activity (protection index 26% and 4 days
lengthening of mean survival time). Based on the experimental data
obtained (including data from our previous study [11]), a general-
ization of the structure of the most active compounds against PV1,
CVB1 and CVB3 viruses is shown in Fig. 9 (right formula).
of the relative influence of the substituents in the 5-nitrobenzonitrile
moiety on activity (Fig. 7). It can be seen, that nitrogen-containing
substituents promote the given activity to a greater extent.
Further, classification QSAR models regarding the in vitro activity
(IC50) against coxsackieviruses B1 (CVB1) were developed. The model
developments were carried out in a similar manner, except that in this
case there were 19 active and 54 inactive molecules. This dataset was
also balanced. RF method was used for decision of classification task.
Like the previous task, Model 5 is the best model with twice the
number of active compounds (Table 8). The relative influence of sub-
stituents on antiviral activity is given in Fig. 8.
As it can be seen from this sequence, the character of the influence
of substituents on anti-viral activity against CVB1 differs significantly
from the similar influence on the anti-PV1 activity. It can be noted that
the presence of cytisine moiety promotes both types of activity.
Unfortunately, there weren’t adequate QSAR models for the anti-
viral activity of the investigated compounds against CVB3. Obviously,
this is due to the high imbalance in the training set which consists of
only 8 active compounds out of 73.
Thus, classification QSAR models with adequate statistical char-
acteristics were obtained to estimate the anti-viral activity of in-
vestigated compounds against poliovirus and coxsackiviruses. These
models will be used in further studies for virtual screening and mole-
cular design of new anti-viral agents corresponding to the “domain
applicability” (DA) of developed QSAR models.
QSAR analysis of substituent effects on the in vitro cytotoxicity
(CC50) and anti-viral activity (against PV1, CVB1, CVB3) of the ni-
trobenzonitrile derivatives was carried out, including data from our
previous study [11]. The results obtained allowed to perform qualita-
tive assessment of the effect of substituents on cytotoxicity of the
compounds and to construct adequate QSAR models for anti-viral ac-
tivity against PV1 and CVB1. These models could be useful for further
virtual screening and molecular design of new anti-viral agents in ac-
cordance with the “domain applicability” (DA) of developed QSAR
models.
3. Conclusions
In summary, a series of 60 analogues of the anti-viral agent MDL-
496

I. Nikolova, et al.
BioorganicChemistry85(2019)487–497
Our study has revealed that the presented nitrobenzonitrile deri-
Infect. Dis. 3 (2017) 585–594.
[11] G.M. Dobrikov, I. Slavchev, I. Nikolova, A. Stoyanova, N. Nikolova, L. Mukova,
R. Nikolova, B. Shivachev, A.S. Galabov, Synthesis and anti-enterovirus activity of
new analogues of MDL-860, Bioorg. Med. Chem. Lett. 27 (2017) 4540–4543.
[12] H.P. Baudet, The replaceability of the halogen atom in 1-chloro- and 1-bromo-2-
cyano-4-nitrobenzene, Recl. Trav. Chim. Pays-Bas 43 (1924) 707–726.
[13] T. George, R. Tahilramani, Condensed 1,3-benzothiazines. Facile rearrangement of
3-alkyl-8-nitro-s-triazolo[3,4-b](1,3,4)benzothiadiazepine, J. Org. Chem. 36 (1971)
2190–2192.
vatives are promising class of anti-viral agents possessing high in vitro
activity and selectivity toward PV1, CVB1 and CVB3 viruses accom-
panied with very low cytotoxicity. The poor water solubility of these
compounds is a possible explanation for the absence of correlation
between their in vitro and in vivo activity. Nevertheless, further in vivo
experiments could be undertaken after an appropriate formulation of
the active in vitro compounds. Thus, we have developed this class of
compounds, including previously unexplored variation of the sub-
stituents in the southern hemisphere (Fig. 9). It could be concluded that
within the current study, we exhausted all the possibilities for suc-
cessful variations of the substituents in the southern hemisphere of
MDL-860. It seems that changes in the northern hemisphere could be
much more perspective for further investigations.
[14] Z. Brzozowski, J. Sławiński, Reaction products of activated aromatic and hetero-
aromatic chlorides with N, N-disubstituted formamides, Synth. Commun. 40 (2010)
1639–1645.
[15] R.H. Bradbury, J.G. Kettle, J.S. Scott, B.C. Barlaam, Quinazoline derivatives as erbb
receptor tyrosine kinases, WO 2005118572 A1, in: Google Patents, 2005.
[16] P.B. Chappell, B.T. O'Neill, M.D. Saltarelli, Pharmaceutical uses for fluoroalk-
oxybenzylamino derivatives of nitrogen containing heterocycles,
US20080132538A1, in: Google Patents, 2000.
[17] H. Ushio, S. Ishibuchi, Y. Naito, N. Sugiyama, T. Kawaguchi, K. Chiba, M. Ohtsuki,
Y. Naka, Amide compounds and medicinal use thereof, WO 2000047558 A1, in:
Google Patents, 2000.
Acknowledgements
[18] J.A. Lowe, T.J. Rosen, Fluoroalkoxybenzylamino derivatives of nitrogen containing
heterocycles, US5773450 A, in: Google Patents, 1998.
[19] J.P. Idoux, J.T. Gupton, C.K. McCurry, A.D. Crews, C.D. Jurss, C. Colon, R.C. Rampi,
Aromatic fluoroalkoxylation via direct aromatic nucleophilic substitution, J. Org.
Chem. 48 (1983) 3771–3773.
The authors acknowledge the generous financial support from
Bulgarian Science Fund – project B02/11 12.12.2014 ‘‘Synthesis and
anti-enterovirus activity of novel diaryl ethers and their complexes with
cyclodextrins”. The financial support of the Bulgarian Science Fund for
the purchase of Bruker Avance II+ 600 NMR spectrometer in the fra-
mework of the Program ‘Promotion of the Research Potential through
Unique Scientific Equipment’ – project UNA-17/2005 is gratefully ac-
knowledged. Technical support from Research and Development and
Innovation Consortium, Sofia Tech Park (Sofia, Bulgaria) was gratefully
acknowledged.
[20] J.T. Gupton, J.P. Idoux, C. Colon, R. Rampi, Aromatic nucleophilic substitution: the
reaction of sodium 2,2,2-trifluoroethoxide with chlorobenzonitriles in HMPA,
Synth. Commun. 12 (1982) 695–700.
[21] J. Harnett, M. Auguet, D.L.P.E. Chabrier, Novel lipoic acid derivatives, their pre-
paration, and pharmaceutical compositions containing them, WO2000059899 A1,
in: Google Patents, 2000.
[22] M. Makosza, S. Voskresensky, M. Białecki, A. Kwast, Vicarious nucleophilic sub-
stitution of hydrogen in 2,4-dinitroanilines and cyanonitroanilines, Pol. J. Chem. 73
(1999) 1969–1977.
[23] J.T. Gupton, J.P. Idoux, G. Baker, C. Colon, A.D. Crews, C.D. Jurss, R.C. Rampi,
Reaction of activated aryl and heteroaryl halides with hexamethylphosphoramide,
J. Org. Chem. 48 (1983) 2933–2936.
Appendix A. Supplementary material
[24] V.L. Plakidin, V.N. Vostrova, Reaction of 2,4-disubstituted chlorobenzenes with 3-
alkylamino and 3-dialkylaminopropionitriles, Zh. Org. Khim. 18 (1982) 342–352.
[25] J.F.K. Wilshire, The preparation of 2-fluoro-5-nitrobenzonitrile and the proton
magnetic resonance spectra of some compounds containing the N-(2-Cyano-4-ni-
trophenyl) group, Aust. J. Chem. 20 (1967) 1663–1670.
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.bioorg.2019.02.020.
[26] B.B. Dey, Y.G. Doraiswami, Reactivity of the chlorine atom in the benzene nucleus,
J. Indian Chem. Soc. 10 (1933) 309–320.
References
[27] J.F.K. Wilshire, Intramolecular hydrogen bonding in some ortho-substituted N-
methyl, N-benzyl and N-aryl-4-nitroanilines: a proton magnetic resonance study,
Aust. J. Chem. 35 (1982) 2497–2504.
[1] D. Lugo, P. Krogstad, Enteroviruses in the early 21st century: new manifestations
and challenges, Curr. Opin. Pediatr. 28 (2016) 107–113.
[28] L.D. Markley, Substituted benzonitriles having antiviral activity, US4332820A, in:
Google Patents, 1982.
[2] E. De Clercq, G. Li, Approved antiviral drugs over the past 50 years, Clin. Microbiol.
Rev. 29 (2016) 695–747.
[29] E. Borenfreund, J.A. Puerner, A simple quantitative procedure using monolayer
cultures for cytotoxicity assays (HTD/NR-90), J. Tissue Cult. Methods 9 (1985) 7–9.
[30] N.-Y. Hsu, O. Ilnytska, G. Belov, M. Santiana, Y.-H. Chen, P.M. Takvorian, C. Pau,
H. van der Schaar, N. Kaushik-Basu, T. Balla, C.E. Cameron, E. Ehrenfeld, F.J.M. van
Kuppeveld, N. Altan-Bonnet, Viral reorganization of the secretory pathway gen-
erates distinct organelles for RNA replication, Cell 141 (2010) 799–811.
[31] V.E. Kuz’min, A.G. Artemenko, E.N. Muratov, P.G. Polischuk, L.N. Ognichenko,
A.V. Liahovsky, A.I. Hromov, E.V. Varlamova, Virtual screening and molecular
design based on hierarchical QSAR technology, in: T. Puzyn, J. Leszczynski,
M.T. Cronin (Eds.), Recent Advances in QSAR Studies: Methods and Applications,
Springer, Netherlands, Dordrecht, 2010, pp. 127–176.
[3] A.S. Galabov, I. Nikolova, R. Vassileva-Pencheva, A. Stoyanova, Antiviral combi-
nation approach as a perspective to combat enterovirus infections, PRILOZI 36
(2015) 91.
[4] A.M. De Palma, I. Vliegen, E. De Clercq, J. Neyts, Selective inhibitors of picorna-
virus replication, Med. Res. Rev. 28 (2008) 823–884.
[5] F.G. Hayden, D.T. Herrington, T.L. Coats, K. Kim, E.C. Cooper, S.A. Villano, S. Liu,
S. Hudson, D.C. Pevear, M. Collett, M. McKinlay, Efficacy and safety of oral ple-
conaril for treatment of colds due to picornaviruses in adults: results of 2 double-
blind, randomized, placebo-controlled trials, Clin. Infect. Dis. 36 (2003)
1523–1532.
[6] Y. Xing, J. Zuo, P. Krogstad, M.E. Jung, Synthesis and structure-activity relationship
(SAR) studies of novel pyrazolopyridine derivatives as inhibitors of enterovirus
replication, J. Med. Chem. 61 (2018) 1688–1703.
[32] V.E. Kuz’min, A.G. Artemenko, E.N. Muratov, Hierarchical QSAR technology based
on the simplex representation of molecular structure, J. Comput. Aided Mol. Des. 22
(2008) 403–421.
[7] A. Carta, G. Sanna, I. Briguglio, S. Madeddu, G. Vitale, S. Piras, P. Corona,
A.T. Peana, E. Laurini, M. Fermeglia, S. Pricl, A. Serra, E. Carta, R. Loddo,
G. Giliberti, Quinoxaline derivatives as new inhibitors of coxsackievirus B5, Eur. J.
Med. Chem. 145 (2018) 559–569.
[33] Dragon Professional version 5.5, 2007. http://www.talete.mi.it.
[34] S. Rännar, F. Lindgren, P. Geladi, S. Wold, A PLS kernel algorithm for data sets with
many variables and fewer objects. Part 1: theory and algorithm, J. Chemom. 8
(1994) 111–125.
[8] L.D. Markley, Y.C. Tong, J.K. Dulworth, D.L. Steward, C.T. Goralski, H. Johnston,
S.G. Wood, A.P. Vinogradoff, T.M. Bargar, Antipicornavirus activity of substituted
phenoxybenzenes and phenoxypyridines, J. Med. Chem. 29 (1986) 427–433.
[9] G. Pürstinger, A.M. De Palma, G. Zimmerhofer, S. Huber, S. Ladurner, J. Neyts,
Synthesis and anti-CVB 3 evaluation of substituted 5-nitro-2-phenoxybenzonitriles,
Bioorg. Med. Chem. Lett. 18 (2008) 5123–5125.
[35] L. Breiman, Random forests, Mach. Learn. 45 (2001) 5–32.
[36] R.E. Carhart, D.H. Smith, R. Venkataraghavan, Atom pairs as molecular features in
structure-activity studies: definition and applications, J. Chem. Inf. Comput. Sci. 25
(1985) 64–73.
[37] K. Hasegawa, Y. Miyashita, K. Funatsu, GA strategy for variable selection in QSAR
studies: GA-based PLS analysis of calcium channel antagonists, J. Chem. Inf.
Comput. Sci. 37 (1997) 306–310.
[10] M. Arita, G. Dobrikov, G. Pürstinger, A.S. Galabov, Allosteric regulation of phos-
phatidylinositol 4-Kinase III Beta by an antipicornavirus compound MDL-860, ACS
497