Synthesis and phospholipidosis effect of a series of cationic amphiphilic compounds: a case study to evaluate in silico and in vitro assays

Article abstract of DOI:10.1007/s00044-017-2093-5

In recent years, a large number of in silico and in vitro assays have been developed for safety assessment in early drug discovery. These methods are usually validated using datasets of known drugs with large chemical diversity, while application to homologous series has been rarely explored. Here we report a case study about phospholipidosis (PLD) risk evaluation for a dataset of nine compounds, designed and synthesized to modulate the physico-chemical properties typical of cationic amphiphilic compounds (CADs), representing the main class of PLD inducers. Our aim was to investigate the effect of structure modification on PLD induction according to a number of standard in silico and in vitro methods. As a result, we found that different in silico methods lead to conflicting results when applied to our series of weak PLD inducers, thus the apparently easy-to-use definition of CADs requires special attention. Moreover, when weak inducers are tested in vitro, the revealed PLD effect may vary based on the purity grade of the tested compound and the features of the selected assay. Finally, we have shown that slight modifications on a chemical scaffold can have an impact on the PLD effect. This study also exemplifies that current in silico methods possibly overestimate the PLD induction effect of cationic amphiphilic compounds compared to the in vitro, with the risk of discarding promising compounds based on incorrect safety liabilities.

Full text of DOI:10.1007/s00044-017-2093-5

MEDICINAL
CHEMISTRY
RESEARCH
Med Chem Res (2018) 27:679–692
DOI 10.1007/s00044-017-2093-5
ORIGINAL RESEARCH
Synthesis and phospholipidosis effect of a series of cationic
amphiphilic compounds: a case study to evaluate in silico and
in vitro assays
Susan Lepri1 Aurora Valeri2 Sandra Buratta3 Martina Ceccarelli2
●
●
●
●
Desirée Bartolini4 Renzo Ruzziconi1 Laura Goracci
2
●
●
Received: 19 April 2017 / Accepted: 3 October 2017 / Published online: 25 October 2017
© Springer Science+Business Media, LLC 2017
Abstract In recent years, a large number of in silico and
in vitro assays have been developed for safety assessment in
early drug discovery. These methods are usually validated
using datasets of known drugs with large chemical diver-
sity, while application to homologous series has been rarely
explored. Here we report a case study about phospholipi-
dosis (PLD) risk evaluation for a dataset of nine com-
pounds, designed and synthesized to modulate the physico-
chemical properties typical of cationic amphiphilic com-
pounds (CADs), representing the main class of PLD indu-
cers. Our aim was to investigate the effect of structure
modification on PLD induction according to a number of
standard in silico and in vitro methods. As a result, we
found that different in silico methods lead to conflicting
results when applied to our series of weak PLD inducers,
thus the apparently easy-to-use definition of CADs requires
special attention. Moreover, when weak inducers are tested
in vitro, the revealed PLD effect may vary based on the
purity grade of the tested compound and the features of the
selected assay. Finally, we have shown that slight mod-
ifications on a chemical scaffold can have an impact on the
PLD effect. This study also exemplifies that current in silico
methods possibly overestimate the PLD induction effect of
cationic amphiphilic compounds compared to the in vitro,
with the risk of discarding promising compounds based on
incorrect safety liabilities.
●
●
Keywords Phospholipidosis Cationic amphiphilic drugs
●
●
●
Toxicophore HepG2 cells Fluorescence assays Organic
synthesis
Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s00044-017-2093-5) contains supplementary
material, which is available to authorized users.
Introduction
Susan Lepri and Aurora Valeri contributed equally to this work.
Drug failure due to safety issues represents a major factor in
the productivity gap for pharmaceutical companies
(Blomme and Will 2016; Waring et al. 2015). Thus, several
strategies have been recently developed to move the safety
assessment of new chemical entities early in drug discovery,
and new technologies are being applied for this purpose
(Waring et al. 2015; DiMasi et al. 2010). An exhaustive
review about toxicology strategies in drug discovery has
been recently published (Blomme and Will 2016), including
in silico and in vitro assays.
* Laura Goracci
laura.goracci@unipg.it
1
Department of Chemistry, Organic Chemistry Section, Biology
and Biotechnology, University of Perugia, via Elce di Sotto 8,
06123 Perugia, Italy
2
Department of Chemistry, Laboratory for Chemoinformatics and
Molecular Modelling, Biology and Biotechnology, University of
Perugia, via Elce di Sotto 8, 06123 Perugia, Italy
3
Department of Chemistry, Laboratory of Biochemistry and
Molecular Biology, Biology and Biotechnology, University of
Perugia, via del Giochetto, 06123 Perugia, Italy
Among the toxicity risk assessments moved to an early
discovery phase, the evaluation of phospholipidosis (PLD)
associated to drugs apparently represents a simple task
(Chatman et al. 2009). PLD is a lipid storage disorder
4
Department of Pharmaceutical Sciences, Nutrition and Clinical
Biochemistry Laboratory, University of Perugia, via del Giochetto,
06123 Perugia, Italy

680
Med Chem Res (2018) 27:679–692
characterized by the accumulation of phospholipids in the
lysosomes of the affected tissues upon drug treatment,
forming microscopic multilamellar inclusion bodies (Chat-
man et al. 2009; Ikeda et al. 2008; Reasor et al. 2006;
Halliwell 1997). In recent years, much effort has been
devoted to predict the PLD effect of drugs (Ploemen et al.
2004; Tomizawa et al. 2006; Pelletier et al. 2007; Ivanciuc
2008; Hanumegowda et al. 2010; Lowe et al. 2012; Kruhlak
et al. 2008; Goracci et al. 2013), as well as to develop
efficient in vitro assays (Kasahara et al. 2006; Bhandari
et al. 2008; van de Water et al. 2011) for fast and cost-
effective PLD evaluation. Although the mechanism is not
yet fully understood, knowledge about this adverse event is
rather advanced, and the chemical features of PLD inducers
seem quite simple to be described. Indeed, cationic
amphiphilic drugs (CADs) are considered the most relevant
class responsible for this disorder (Reasor et al. 2006;
Kodavanti and Mehendale 1990; Lullmann et al. 1975),
although a number of aminoglycoside or macrolide anti-
biotics are also PLD inducers (PLD+) (Munic et al. 2011;
Mingeot-Leclercq and Tulkens 1999). Despite this side-
effect, CADs are, for their amphiphilic nature, elected drugs
to penetrate the blood brain barrier, and thus are often
developed to treat neurological diseases (Muehlbacher et al.
2012). In addition, they are also used in a wide range of
therapeutic areas, including the treatment of arrhythmia,
allergy, and fungal infections (Kodavanti and Mehendale
1990). The definition of CADs is apparently rather trivial.
CADs are usually described as drugs that share two struc-
tural features: a rigid hydrophobic moiety (primarily aro-
matic rings) and a polar “head group” including an amine
group, which is mainly in its protonated state under phy-
siological conditions (Reasor et al. 2006). Due to this
simple definition and the general PLD/CADs qualitative
correlation, a number of rule-based methods for PLD pre-
dictions mainly based on CADs chemical features (pKa,
CLogP, volume of distribution, and amphiphilic moment)
have been developed in the last 15 years (Ploemen et al.
2004; Tomizawa et al. 2006; Pelletier et al. 2007; Hanu-
megowda et al. 2010; Fischer et al. 2012). Interestingly,
depending on the PLD prediction methods, physico-
chemical properties defining CADs are quantitatively dif-
ferent. For example, Ploemen et al. (2004) set as pre-
liminary reference values pKa > 8 and ClogP > 1, while
Muehlbacher et al. (2012) used pKa > 7.4 and logP > 3. In
addition, Katzmi et al. (2013) set a pKa > 6 as reference
value, while Tomizawa et al. (2006) introduced the net
charge (NC) calculated at pH 4.0 for CADs definition
(CLogP > 1 and 1 ≤ NC ≤ 2). Finally, for Nadanaciva et al.
(2011) CADs having a basic center with a pKa between 6.5
and 11 are likely to cause lysosomotropism. Thus,
depending on the threshold used, new chemical entities
(NCE) may or may not be classified as CADs. In this
context, the need to improve the definition for CADs
inducing PLD has been recently raised (Slavov et al. 2014;
Sun et al. 2012; Choi et al. 2013), considering that, as noted
by Choi (2013), not all CADs induce PLD. Consequently,
the importance of describing the spatial distribution of the
chemical groups leading to the amphiphilic nature has been
also addressed to refine the CADs definition. To this aim,
the peculiar chemical nature of PLD-inducing CADs has
been recently summarized by two toxicophore models by
Slavov et al. (2014) and by us (Goracci et al. 2015). A
deeper understanding of the PLD effect associated with
CADs should avoid being too general thereby limiting the
exclusion of potentially good drugs in an early phase.
However, if from the chemistry perspective a better
definition of CADs inducing PLD is in progress, on the
biological side, eventual anomalous effects of CADs
towards PLD induction could be related also to the choice
of the in vitro assay used for toxicological evaluation.
In the present study, we report a case study to evaluate
the effect of chemical features of CADs and the biological
assays used on the determination of the PLD induction.
Indeed, we realized that focused investigations into this
issue are extremely scarce (Quaglino et al. 2004). Our
starting point was the N-(2-chlorobenzyl)quinuclidin-3-
amine (1) (Fig. 1), which possesses all the chemical features
generally attributed to a PLD+ compound. Indeed, com-
pound 1 has a calculated LogP value of 2.5, with the most
basic center having a pKa of 9.45, and it is predicted as a
PLD+ according to various in silico models for PLD risk
assessment (see Table 1). In addition, compound 1 nicely
fits the Slavov two-center toxicophore (Slavov et al. 2014),
with a distance between the tertiary amino group (the most
basic center) and the aromatic ring of 5.3 Å, and our
recently published toxicophore model (Goracci et al. 2015).
Despite these chemical features, compound 1 recently
resulted to be a false positive in vitro, using two different
biological assays (Goracci et al. 2015).
Therefore, we decided to further investigate this case, re-
synthesizing compound 1 with higher purity (>99%), and
synthesizing eight new analogs (compounds 2–9, Table 1).
The PLD effect of compounds 1–9 was predicted according
to four in silico methods, and then determined by a well-
established cell-based in vitro assay using two different
fluorescent phospholipids as probes.
Fig. 1 Structure of compound 1

Med Chem Res (2018) 27:679–692
681
Table 1 Calculated pKa and LogP for compounds 1–9
LogPa PLD (P)b PLD (T)c PLD (PLS-DA)d PLD (PLD-phore)e
a
Comp Structure
pKa
1
9.45 (b1), 5.29 (b2)
2.5
+
+
+
+
2
3
4
5
5.89 (b1)
2.9
−
−
−
−
9.42 (b1), 5.42 (b2)
2.8
+
+
+
+
+
+
+
+
+
+
+
+
10.17 (b1), 5.98 (b2) 2.0
9.45 (b1), 5.29 (b2)
3.0
6
7
15.08 (a1), 8.18 (b1) 1.2
−
+
+
−
−
9.42 (b1), 5.42 (b2)
3.3
1.9
+
+
+
8
8.64 (b1), 4.48 (b2)
−
+
+
+

682
Med Chem Res (2018) 27:679–692
Table 1 continued
a
Comp Structure
pKa
8.82 (b1)
LogPa PLD (P)b PLD (T)c PLD (PLS-DA)d PLD (PLD-phore)e
9
2.3
−
+
+
+
a
The pKa and Log P values were predicted using MoKa (v.2.6.5) (Cruciani et al. 2009; Milletti et al. 2010)
b,c
d
PLD effect calculated using Ploemen and Tomizawa predictive rules, respectively (Ploemen et al. 2004; Tomizawa et al. 2006)
PLD prediction based on the PLS-DA model (Goracci et al. 2013)
e
PLD risk assessment based on PLD-phore alignment (Goracci et al. 2015)
Materials and methods
Computational study
PLD effect prediction by PLD-phore
The FLAP software (Molecular Discovery, UK) was used to
study the similarity of the tested compounds with the pre-
viously defined PLD-phore (Goracci et al. 2015). In parti-
cular, the FLAP algorithm (Baroni et al. 2007) was used to
align the structures of compounds 1, 3–9 to the PLD-phore
previously generated using the FLAPpharm module (Cross
et al. 2012a, b). Briefly, the PLD-phore was generated by
the automatic alignment of five CADs reported to be PLD
inducers, in their protonated state. Subsequently the com-
mon interacting features in terms of their GRID molecular
interaction fields, pseudo-fields, and atomic points are
extracted. An extensive description of the FLAP and
FLAPpharm algorithms as well as the applications in
defining the pharmacophores have been reported elsewhere
(Baroni et al. 2007; Cross et al. 2012a, b; Sirci et al. 2012;
Lepri et al. 2014; Tondi et al. 2016; Chapy et al. 2015;
Artese et al. 2013). Compounds 1 and 3–9, each one con-
sidered in its protonated form and in a maximum of 50
possible conformations to mimic flexibility, were thus
aligned on the PLD-phore. During this alignment process a
constraint was used for the PLD-phore, defining that its H-
bond donor region should be a charged H-bond donor atom.
Then, the best alignment in term of the FLAP Glob-Prod
descriptor (covering shape, donor, acceptor, and hydro-
phobic similarity) was selected.
PLD effect predicted according to Ploemen (P) (Ploemen
et al. 2004) and Tomizawa (T) (Tomizawa et al. 2006)
models The rule-based methods P and T are based on Eqs.
(1) and (2), respectively:
2
2
Predicted PLD þ : if ðCLogPÞ þ ðpKaMB
Þ
ꢀ 90 with CLogP ꢀ 1 and pKaMB ꢀ 8
ð1Þ
2
2
Predicted PLDÀ : if ðCLogPÞ þðpKaMB
<90 with CLogP<1 and pKaMB<8
Þ
Predicted PLDþ : if CLogP>1 and 1 ꢁ NC ꢁ 2
Predicted PLDÀ : if CLogP<1 and NC<1
ð2Þ
where pKaMB refers to the pKa of the most basic center in
the molecule and NC correspond to the net charge at pH
4.0. Values calculated for compound 1–9 are reported in
Table 1. The pKaMB, CLogP, and NC values were
calculated using MoKa (Molecular Discovery Ltd, UK)
(Cruciani et al. 2009; Milletti et al. 2010).
PLD effect prediction by PLS-DA
Chemistry
An already published PLS-DA model of 331 PLD inducing
and non-inducing compounds (accuracy = 0.78, sensitivity
= 0.71), was used to predict the PLD risk for compounds in
Table 1. Using the VolSurf+ software (Cruciani et al.
2000), compounds 1–9 were imported in their most abun-
dant protonation state at pH 7.4 and all the 128 physico-
chemical and ADME related descriptors were used as
variables. A complete list of the VolSurf+ descriptors has
been reported elsewhere (Mannhold et al. 2006).
All commercial products were acquired form Sigma Aldrich
1
and used without further purification (≥97% pure). H and
13C nuclear magnetic resonance (NMR) spectra were
recorded at 400 and 100.6 MHz, respectively, on Bruker
Avance II 400 MHz spectrometer at room temperature.
Chemical shifts (δ) are given in parts per million (ppm)
relative to the internal standard tetramethylsilane. Peak
multiplicities are reported as s (singlet), d (doublet),

Med Chem Res (2018) 27:679–692
683
dd (double doublet), t (triplet), dt (double triplet), q (quar-
tet), m (multiplet), or br s (broad singlet). Coupling con-
stants (J) are given in Hz. High resolution mass
spectrometry (HRMS) spectra were registered on Agilent
Technologies 6540 UHD Accurate Mass Q-TOF liquid
chromatography/mass spectrometry (LC/MS) system.
C12H1135ClN2 219.0689 (M + H+), found 219.0689
(M + H+).
N1-(2-Chlorobenzyl)-N2,N2-diethylethane-1,2-diamine
dihydrochloride (3) Yield 87%, white solid, mp 115–116 °C;
1H NMR (400 MHz, MeOD): δ = 7.78 (d, J = 6.3 Hz, 1H, H-
6), 7.58 (d, J = 7.3 Hz, 1H, H-3), 7.54–7.37 (m, 2H, H-5, H-
4), 4.52 (s, 2H, ArCH2N), 3.87–3.67 (m, 2H, NCH2CH2NEt),
3.67–3.53 (m, 2H, NCH2CH2NEt), 3.49–3.33 (m, 4H,
NCH2CH3), 1.55–1.26 (m, 6H, NCH2CH3); 13C NMR (101
MHz, MeOD): δ = 134.3 (C, C-1), 132.0 (CH, C-6), 131.4
(CH, C-4), 129.9 (CH, C-3), 128.7 (C, C-2), 127.7 (CH, C-5),
48.8 (CH2, ArCH2N), 48.0 (CH2, NCH2CH3), 47.5 (CH2,
NCH2CH2NEt), 42.2 (CH2, NCH2CH2NEt), 7.9 (CH3,
NCH2CH3); HRMS-ESI: m/z calcd. for C13H2135ClN2
241.1472 (M + H+), found 241.1469 (M + H+).
Synthesis of N-substituted 2-chlorobenzylamines (1–7)
A solution of 2-chlorobenzaldehdye (11 mmol) and suitable
amine (10 mmol) in MeOH (35 mL) was stirred at r.t. for 24
h. NaBH4 (40 mmol) was carefully added in small portion at
0 °C and the reaction was kept at room temperature for
further 4 h. The reaction was carefully quenched with
saturated aqueous NaHCO3 solution and evaporated to
dryness. The residue was purified by flash chromatography
on SiO2 (eluent DCM/MeOH/Et3N 95:5:1, if not stated
otherwise) to afford 2-chlorobenzylamines 1–7 (Table 1).
The resulting amino compound was dissolved in methanol
and HCl 4.0 M in dioxane (40 mmol) was added, the solu-
tion was stirred for 30 min and then concentrated again to
give the hydrochloric salt in quantitative yield.
N-(2-Chlorobenzyl)piperidin-3-amine dihydrochloride (4)
1
Yield 79%, white solid, mp (dec.) 270 °C; H NMR (400
MHz, MeOD): δ = 7.75 (dd, J = 1.5, 7.3 Hz, 1H, H-6),
7.65–7.57 (m, 1H, H-3), 7.56–7.41 (m, 2H, H-4, H-5),
4.63–4.39 (m, 2H, ArCH2N), 3.87 (d, J = 12.1 Hz, 1H, H-
2′), 3.79 (t, J = 10.9 Hz, 1H, H-1′), 3.46 (d, J = 12.8 Hz,
1H, H-4′), 3.24 (t, J = 11.7 Hz, 1H, H-2′), 3.17–3.00 (m,
1H, H-4′), 2.48 (d, J = 10.6 Hz, 1H, H-6′), 2.18 (d, J =
13.9 Hz, 1H, H-5′), 2.03–1.73 (m, 2H, H-5′, H-6′); 13C
NMR (101 MHz, MeOD): δ = 134.5 (C, C-1), 132.1 (CH,
C-6), 131.5 (CH, C-4), 129.9 (CH, C-3), 128.9 (C, C-2),
127.7 (CH, C-5), 52.1 (CH, C-1′), 46.0 (CH2, ArCH2N),
43.6 (CH2, C-2′), 43.3 (CH2, C-4′), 24.9 (CH2, C-6′), 20.2
(CH2, C-5′); HRMS-ESI: m/z calcd. for C12H1735ClN2
225.1159 (M + H+), found 225.1159 (M + H+).
N-(2-Chlorobenzyl)quinuclidin-3-amine
dihydrochloride
(1-syn) Yield 70%, white solid, mp 141–142 °C; 1H
NMR (400 MHz, dimethyl sulphoxide (DMSO)-d6): δ =
10.97 (s, 1H, NH), 10.57 (s, 1H, NH), 9.87 (s, 1H, NH),
7.92 (d, J = 6.1 Hz, 1H, H-6), 7.69–7.54 (m, 1H, H-3),
7.54–7.40 (m, 2H, H-4, H-5), 4.57–4.14 (m, 2H, ArCH2N),
3.94–3.74 (m, 1H, H-1′), 3.68 (t, J = 11.4 Hz, 1H, H-2′),
3.53–3.40 (m, 1H, H-7′), 3.39–3.33 (m, 1H, H-2′),
3.27–2.96 (m, 3H, H-4′, H-7′), 2.66–2.58 (m, 1H, H-6′),
2.41–2.20 (m, 1H, H-5′), 2.10–1.76 (m, 1H, H-8′),
1.90–1.76 (m, 2H, H-5′, H-8′); 13C NMR (101 MHz,
DMSO-d6): δ = 134.3 (C, C-1), 133.1 (CH, C-6), 131.5
(CH, C-4), 130.0 (CH, C-3), 129.9 (C, C-2), 127.9 (CH, C-
5), 52.7 (CH, C-1′), 48.6 (CH2, C-2′), 46.2 (CH2, ArCH2N),
45.8 (CH2, C-7′), 45.3 (CH2, C-4′), 22.0 (CH2, C-8′), 21.9
(CH, C-6′), 17.1 (CH2, C-5′); HRMS-ESI: m/z calcd.
for C14H1935ClN2 251.1315 (M + H+), found 251.1311
(M + H+).
N-(2-Chlorobenzyl)-1-ethylpiperidin-3-amine dihydrochlor-
ide (5) Yield 80%, white solid, mp 110–111 °C; 1H NMR
(400 MHz, MeOD): δ = 7.72 (d, J = 6.1 Hz, 1H, H-6), 7.58
(d, J = 7.7 Hz, 1H, H-3), 7.54–7.38 (m, 2H, H-4, H-5),
4.58–4.44 (m, 2H, ArCH2N), 4.03–3.78 (m, 1H, H-2′),
3.91–3.76 (m, 1H, H-1′), 3.76–3.54 (m, 1H, H-4′), 3.34 (q,
J = 6.5 Hz, 2H, NCH2CH3), 3.24–2.89 (m, 1H, H-2′),
3.07–2.92 (m, 1H, H-4′), 2.56–2.35 (m, 1H, H-6′),
2.25–2.08 (m, 1H, H-5′), 2.07–1.89 (m, 1H, H-6′),
1.90–1.74 (m, 1H, H-5′), 1.43 (t, J = 6.7 Hz, 3H,
NCH2CH3); 13C NMR (101 MHz, MeOD): δ = 134.5 (C,
C-1), 132.1 (CH, C-6), 131.6 (CH, C-4), 129.9 (CH, C-3),
128.8 (C, C-2), 127.8 (CH, C-5), 53.0 (CH2, NCH2CH3),
52.5 (CH, C-1′), 51.3 (CH2, C-2′), 51.0 (CH2, C-4′), 46.1
(CH2, ArCH2N), 24.7 (CH2, C-6′), 20.8 (CH2, C-5′), 8.3
(CH3, NCH2CH3). HRMS-ESI: m/z calcd. for
C14H2135ClN2 253.1472 (M + H+), found 253.1472
(M + H+).
N-(2-Chlorobenzyl)pyridin-3-amine dihydrochloride (2)
Yield 75%, white solid, mp 130–131 °C; purified by col-
umn chromatography (eluent DCM/MeOH, 98:2); 1H NMR
(400 MHz, MeOD): δ = 8.04 (s, 1H, H-2′), 8.00 (t, J = 3.0
Hz, 1H, H-6′), 7.80–7.69 (m, 2H, H-4′, H-5′), 7.53–7.42 (m,
2H, H-3, H-6), 7.37–7.28 (m, 2H, H-4, H-5), 4.58 (s, 2H,
NCH2Ar); 13C NMR (101 MHz, CD3OD): δ = 147.7 (C, C-
3′), 134.2 (C, C-1), 133.3 (C, C-2), 129.5 (CH, C-3), 129.0
(CH, C-4), 128.8 (CH, C-6), 127.9 (CH, C-6′), 127.4 (CH,
C-5′), 127.2 (CH, C-4′), 127.1 (CH, C-5), 123.9 (CH, C-2′),
44.0 (CH2, NCH2Ar); HRMS-ESI: m/z calcd. for

684
Med Chem Res (2018) 27:679–692
2-((2-Chlorobenzyl)amino)ethanol hydrochloride (6)
Yield 81%, white solid, mp 134–135 °C; H NMR (400
cooled to 0 °C, 0.77 mL of water was added, followed by
0.77 mL of 15% NaOH (aq) and 2.31 mL of water. The
resulting suspension was stirred for 1 h, the solid was filtered
and rinsed with EA. The organic phase was separated and
evaporated under vacuum; the residue was treated with HCl in
dioxane (2 mL, 4.0 M) and evaporated again to give N,N-
Diethyl-2-(isoindolin-2-yl)ethanamine dihydrochloride; 1H
NMR (400 MHz, MeOD): δ = 7.45 (s, 4H, H-3, H-4, H-5, H-
6), 4.87 (s, 4H, H-7, H-8), 4.27–3.92 (m, 2H, NCH2CH2NEt),
3.85–3.59 (m, 2H, NCH2CH2NEt), 3.40 (q, J = 7.3 Hz, 4H,
NCH2CH3), 1.44 (t, J = 7.3 Hz, 6H, NCH2CH3); 13C NMR
(101 MHz, MeOD): δ = 133.1 (C, C-1, C-2), 128.9 (CH, C-4,
C-5), 122.6 (CH, C-3, C-6), 58.8 (CH2, C-7, C-8), 48.1 (CH2,
NCH2CH2NEt), 47.7 (CH2, NCH2CH3), 46.8 (CH2,
NCH2CH2NEt), 7.7 (CH3, NCH2CH3); HRMS-ESI: m/z
calcd. for C14H22N2 219.1861 (M + H+), found 219.1855
(M + H+).
1
MHz, MeOD): δ = 7.68 (dd, J = 1.9, 7.5 Hz, 1H, H-6), 7.58
(dd, J = 1.4, 7.5 Hz, 1H, H-3), 7.53–7.34 (m, 2H, H-4, H-
5), 4.46 (s, 2H, ArCH2N), 4.02–3.78 (m, 2H,
NCH2CH2OH), 3.29–3.18 (m, 2H, NCH2CH2OH); 13C
NMR (101 MHz, MeOD): δ = 134.5 (C, C-1), 131.9 (CH,
C-6), 131.2 (CH, C-4), 129.8 (CH, C-3), 129.1 (C, C-2),
127.6 (CH, C-5), 56.2 (CH2, NCH2CH2OH), 49.2 (CH2,
NCH2CH2OH), 47.7 (CH2, CH2, ArCH2N); HRMS-ESI: m/
z calcd. for C9H1235ClNO 186.0686 (M + H+), found
186.0677 (M + H+).
N1,N1-Diethyl-N2-(naphthalen-1-ylmethyl)ethane-1,2-dia-
mine dihydrochloride (7) Yield 77%, white solid, mp
1
126–127 °C; H NMR (400 MHz, MeOD): δ = 8.30 (dd, J
= 0.8, 8.6 Hz, 1H, H-10), 8.04 (d, J = 8.3 Hz, 1H, H-4),
8.02–7.98 (m, 1H, H-7), 7.84 (dd, J = 1.0, 7.1 Hz, 1H, H-
2), 7.72 (ddd, J = 1.4, 6.9, 8.4 Hz, 1H, H-9), 7.63 (ddd, J =
1.5, 6.9, 8.0 Hz, 1H, H-8), 7.61–7.57 (m, 1H, H-3), 4.87 (s,
2H, ArCH2N), 3.84–3.73 (m, 2H, NCH2CH2NEt),
3.69–3.57 (m, 2H, NCH2CH2NEt), 3.36 (q, J = 7.0 Hz, 4H,
NCH2CH3), 1.41 (t, J = 7.0 Hz, 6H, NCH2CH3); 13C NMR
(101 MHz, MeOD): δ = 134.0 (C, C-5), 131.2 (C, C-6),
130.4 (CH, C-4), 129.2 (CH, C-2), 128.7 (CH, C-7), 127.2
(CH, C-9), 126.7 (C, C-1), 126.3 (CH, C-8), 125.1 (CH, C-
3), 122.7 (CH, C-10), 48.4 (CH2, ArCH2N), 47.8 (CH2,
NCH2CH3), 47.4 (CH2, NCH2CH2NEt), 41.9 (CH2,
NCH2CH2NEt), 7.8 (CH3, NCH2CH3); HRMS-ESI: m/z
calcd. for C7H24N2 257.2018 (M + H+), found 257.2018
(M + H+).
2-((2-Chlorobenzyl)oxy)-N,N-dimethylethanamine
hydrochloride (9)
A solution of 2-dimethylaminoethanol (0.66 g, 7.4 mmol) in
anhydrous DMF (2 mL) was carefully added to a stirred
suspension of NaH (60% dispersion in mineral oil, 0.29 g,
7.4 mmol) in anhydrous DMF (50 mL). After 30 min, 2-
chlorobenzylbromide (1.00 g, 4.9 mmol) in anhydrous DMF
(5 mL) was added dropwise, and stirring was continued
overnight. The mixture was dropped in water and extracted
with DCM. The organic phase was washed with water
(three times) and evaporated to give the crude oil, purified
by flash chromatography on silica gel (eluent DCM/MeOH/
Et3N, 94:5:1). The resulting amino compound was dis-
solved in methanol and HCl 4.0 M in dioxane (40 mmol)
was added, the solution was stirred for 30 min and then
concentrated again to give the hydrochloric salt as pale
N,N-diethyl-2-(isoindolin-2-yl)ethan-1-amine (8)
1
Yield 50%, white solid, mp (dec.) 238 °C; N,N-diethy-
lethylenediamine (1.88 g, 2.27 mL, 16.2 mmol) was added
to a stirred solution of phthalic anhydride (2.00 g, 14.5
mmol) in toluene (100 mL). Catalytic p-TsOH was added
and the solution refluxed in a Dean Stark apparatus for 2 h.
The solution was concentrated under reduced pressure and
the residue was taken up in DCM, washed with NaOH and
brine. The organic phase was concentrated to afford 2-(2-
(diethylamino)ethyl)isoindoline-1,3-dione (10) as a yellow
oil (99% yield), used in the following step without any
further purification. 1H NMR (400 MHz, CD3Cl): δ =
7.97–7.78 (m, 2H, Ar–H), 7.78–7.64 (m, 2H, Ar–H), 3.78
(t, J = 7.0 Hz, 2H, NCH2CH2NEt), 2.72 (t, J = 7.0 Hz, 2H,
NCH2CH2NEt), 2.59 (q, J = 7.1 Hz, 4H, NCH2CH3), 1.01
(t, J = 7.1 Hz, 6H, NCH2CH3). Phthalimide 10 (1.00 g,
4.06 mmol) was slowly added to a stirred suspension of
LiAlH4 (0.77 g, 20.3 mmol) in THF (10 mL) at 0 °C; after
the addition, the mixture was refluxed for 24 h. After being
yellow solid (1.2 g, yield 99%); mp 108–109 °C; H NMR
(400 MHz, MeOD): δ = 7.61–7.51 (m, 1H, H-6), 7.47–7.40
(m, 1H, H-3), 7.40–7.29 (m, 2H, H-4, H-5), 4.74 (s, 2H,
ArCH2O), 3.89 (t, J = 4.7 Hz, 2H, OCH2CH2N), 3.43 (t, J
= 4.7 Hz, 2H, OCH2CH2N), 2.94 (s, 6H, NCH3); 13C NMR
(101 MHz, MeOD): δ = 134.8 (C, C-1), 133.3 (C, C-2),
130.0 (CH, C-6), 129.3 (CH, C-4), 129.1 (CH, C-3), 126.8
(CH, C-5), 70.2 (CH2, ArCH2O), 63.7 (CH2, OCH2CH2N),
56.7 (CH2, OCH2CH2N), 42.3 (CH3, NCH3); HRMS-ESI:
m/z calcd. for C11H16ClNO 214.0993 (M + H+), found
214.0999 (M + H).
Purity analysis
Purities of the final compounds were determined by LC-MS
on Agilent Technologies 6540 UHD Accurate Mass Q-TOF
LC/MS. The mobile phase was a mixture of water (A) and
acetonitrile (B), both containing formic acid at 0.1%.

Med Chem Res (2018) 27:679–692
685
Chromatographic separation was carried out with Acquity
UPLC BEH C18 1.7 µm (C18, 150 × 2.1 mm) column, with
a flow rate of 0.650 mL/min in a 10 min gradient elution.
Gradient elution was as follows: 99.5:0.5 (A/B) to 5:95 (A/
B) over 8 min, 5:95 (A/B) for 2 min, and then reversion
back to 99.5:0.5 (A/B) over 0.1 min. New synthesized
compounds were ≥98% (>99% for compound 1-syn), while
purity of 1-com was 96%, according to previous result
(Goracci et al. 2015).
CH3COOH 50:49:1, v/v/v). Thus, gentle shaking for 10 min
complete the dissolution. Optical density (OD) was mea-
sured at 550 nm using a microplate reader (Beckman
Coulter DTX880) and blanks, which contain no cells as a
reference. The relative uptake of NR (%) by viable cells was
calculated by comparing absorbance of compound-treated
cells and absorbance of control cells exposed to the vehicle
(0.05% DMSO) (OD (treated cells)/OD (control vehicle) ×
100). Each experiment was performed in triplicate.
NBD-PC and NBD-PE uptake: Cells were plated in 96-
well black plates and allowed to attach for 24 h as described
above. Cells were then treated with each of the compound
tested together with the fluorescent phospholipid analog
NBD-PC, according to Kasahara et al. (2006) or NBD-PE,
according to Bhandari et al. (2008). Briefly, the medium in
each well was removed, and 50 µL of medium containing
80 µM NBD-PC or 100 µM NBD-PE plus 50 µL of medium
containing the tested drugs were added. After 24 h, the cells
were washed twice with PBS, and 50 µL of PBS was added
to each well. Phospholipid fluorescence was measured with
a microplate fluorescence reader (excitation and emission
wavelengths of 485/538 nm and 462/536 nm for NBD-PC
and NBD-PE, respectively). After the phospholipid fluor-
escence was measured, cells were incubated for further 20
min at 37 °C with Hoechst 33342 solution (20 µg/mL in
PBS). The Hoechst 33342 fluorescence was then measured
(excitation and emission wavelengths of 355 and 460 nm,
respectively). Values for NBD-fluorescence were normal-
ized to those of Hoechst 33342 fluorescence.
Biological assay
Cell culture and treatments
HepG2 cells (human hepatoma cell line) were purchased
from the American Type Culture Collection (ATCC® HB-
8065™, Manassas, VA, USA) and all culture reagents were
purchased from Lonza (Breda, Netherlands). 3-Amino-7-
dimethylamino-2-methylphenazine hydrochloride (Neutral
Red) and Hoechst 33342 were purchased from Sigma-
Aldrich (St Louis, MO, USA). 1-acyl-2-[12-(7-nitro-2,1,3-
benzoxadiazol-4-yl)amino]dodecanoyl]-glycero-3-phos-
phocholine (NBD-PC) was purchased from Avanti Polar
Lipids, Inc. (Alabaster, AL) and N-(7-nitrobenz-2-oxa-1,3-
diazol-4-yl)-1,2-dihexadecanoyl-sn-glycero-3-phosphoetha-
nolamine triethylammonium salt (NBD-PE), was purchased
from Molecular Probes (Eugene, OR, USA). HepG2 cells
were maintained in modified Eagle’s Medium supplemented
with 10% fetal bovine serum, 50 U/mL penicillin, 50 µg/mL
streptomycin, 2 mM L-glutamine and 1% non-essential
amino acids solution in a humidified atmosphere contain-
ing 5% CO2 at 37 °C. Cells were passaged as needed using
0.5% trypsin-EDTA solution and experiments were per-
formed between passages 10 and 25. For experimental
purposes, cells collected and suspended at a density of 1 ×
105/mL in fresh medium, were seeded on culture plates.
Briefly, 2 × 104 cells were plated in 96-well plates for NR
uptake and in 96-well black plates for PLD assay with
fluorescent phospholipids. After 24 h, culture medium was
replaced with fresh medium containing vehicle (0.05%
DMSO) or test compounds and then treated cells were
incubated for 24 h.
Statistical analysis: Data are expressed as mean
SD
from three independent experiments done at least in tripli-
cate. Statistical comparisons were performed using Stu-
dent’s t-test. Differences were considered statistically
significant when p < 0.05.
Results and discussion
Design and synthesis of analogs of compound 1
As mentioned in the introduction, physico-chemical prop-
erties have proved to be useful to evaluate a possible PLD
risk effect associated with CAD-like NCE (Ploemen et al.
2004; Tomizawa et al. 2006; Hanumegowda et al. 2010;
Fischer et al. 2012). In our previous work, the false-positive
result of compound 1 was attributed, among other causes, to
the steric hindrance at the most basic center (i.e., the qui-
nuclidine ring). Indeed, the quinuclidine moiety remains
barely explored in PLD induction studies, with only three
examples reported in the literature database (i.e. quinine,
quinidine, and palonosetron) (Ploemen et al. 2004; Tomi-
zawa et al. 2006; Hanumegowda et al. 2010; Orogo et al.
2012). Thus, eight analogs of 1 were designed and
Cytotoxicity assay
Neutral red (NR) uptake: NR uptake into lysosomes of
viable cells was performed according to Repetto et al.
(2008) with slight modification. Briefly, following exposure
to the selected compounds the medium was removed and
50 µL of NR dye (100 µg/mL) was added to each well. After
incubation at 37 °C for 90 min, cells were washed with
phosphate buffered saline (PBS) and dye was extracted by
adding 100 μL of acidified ethanol solution (ethanol: H2O:

686
Med Chem Res (2018) 27:679–692
synthesized to tune the acid-base properties, the lipophili-
city and the steric hindrance at the head-group moiety.
Calculated pKa and LogP values are reported in Table 1.
As a first step, the quinuclidine moiety was replaced with
pyridine, triethylamino group, piperidine, ethyl piperidine,
and hydroxyethyl moiety, to give compounds 2–6, respec-
tively. In compound 2, although the substitution with the
pyridine has a minor impact on the LogP, the pKa is sig-
nificantly reduced by four units. In addition, the second
basic center (the benzyl amino group) becomes also irrele-
vant, being now linked to an aromatic ring. Based on these
physico-chemical features, compound 2 was expected to
lose the CAD-like features and thus to be a PLD-.
Designing compounds 3–5, the effect of the steric hin-
drance associated with the quinuclidine ring was investi-
gated. The greatest flexibility was provided in compound 3,
maintaining a tertiary amine as the most basic center and
keeping the distance between the two basic centers constant.
The cyclohexyl ring was retained in compounds 4 and 5,
which differ for the absence or the presence of a further
alkyl group at amine group, respectively. For these three
structures the possible PLD effect was more uncertain,
because such fine modulations of a scaffold for PLD studies
have been scarcely studied so far (Quaglino et al. 2004).
In compound 6, the most basic center was totally
removed and replaced with an alcoholic functional group.
Despite this substitution, compound 6 preserves its basic
nature while its LogP is significantly reduced. Compounds
7 and 8 are analogs of compound 3, in which the hydro-
phobic moiety was modified in order to increase or decrease
the lipophilicity of the scaffold, respectively. The alkyl-
diethylamine moiety of 3 was selected in this study because
this group is frequent among CADs (e.g., amiodarone,
amiodaquine, chloroquine, and coralgil). Finally, compound
9 was synthesized replacing the benzyl amino basic centre
with a benzyl ether; in addition, the tertiary amine head was
replaced with a dimethylamino moiety because the reagent
was readily available and the dimethylamine is also com-
mon in PLD+ drugs (e.g., imipramine, chlorpromazine, and
iprindole).
PLD effect prediction are reported in the Materials and
methods section.
As shown in Table 1, compound 2 is the only compound
univocally predicted as a PLD non-inducer, in agreement
with the lack of a strong basic center. Compounds 8 and 9
are generally predicted as PLD+ compounds, excepting for
the P model. On the contrary, compound 6 is generally
predicted as PLD- excepting for T model. Finally, com-
pounds 1, 3–5, and 7 are univocally reported as PLD+.
Differently from the P and the T methods, the PLS-DA
and the PLD-phore approaches allow a visual inspection of
the projection on the model or the compound/toxicophore
matching, respectively (Fig. 2). Concerning the projection
on the PLS-DA model, compounds 2 and 6 are clearly
projected on PLD- region (Fig. 2a, b, red circles), while the
remaining compounds are located at edge between PLD+
(blue) and PLD- (red) chemical space. This suggests that,
although compounds 1, 3–5, and 7–9 are predicted as PLD+
drugs, their PLD effect is expected rather weak, with
compound 7 resulting the one with the highest associated
risk of PLD. Concerning the toxicophore similarity, the best
alignment for each structure is reported in Fig. 2c, with the
exception of compound 2. Indeed, it has already been
proven (Slavov et al. 2014; Goracci et al. 2015) that the
toxicophore models are generally valid only when applied
to strong bases. Figure 2c shows that the alignments of 3–5,
8, and 9 are very similar to that of compound 1, following
the Slavov two-center model (Slavov et al. 2014). Com-
pound 7, bearing a naphthalene moiety, shows the best
fitting of the hydrophobic region of the toxicophore, also
fitting the Slavov three-center model (Slavov et al. 2014).
On the contrary, for the hydroxyl compound 6, the align-
ment to the hydrophobic region of the toxicophore is totally
lost, due to the orientation of the remaining basic center to
the polar region.
Synthesis of compounds 1–9
All tested compounds were synthesized as described in
Schemes 1–3. In particular, the compounds 1–7 were
obtained by reductive amination of 2-chlorobenzaldehyde
and 1-naphthaldeyde with the suitable amine. The reaction
was carried out in methanol using sodium borohydride as
reducing agent (Schemes 1–2).
Compound 8 was obtained by condensation of phthalic
anhydride with N,N-diethyl-1,2-ethylendiamine (10), fol-
lowed by reduction (Scheme 2). Finally, benzyl ether
derivative 9 was obtained in high yield by reaction of 2-
chlorobenzyl bromide with N,N-dimethylethanolamine
(Scheme 3).
Predicted PLD effect of compounds 1–9
The PLD effect of compound 1 and of its designed analogs
2–9 was predicted using four in silico methods (Table 1):
the Ploemen’s method (P) (Ploemen et al. 2004), the
Tomizawa’s method (T) (Tomizawa et al. 2006), the pro-
jection on a recently published partial least squares dis-
criminant analysis (PLS-DA) model (Goracci et al. 2013),
and the similarity towards the toxicophore (PLD-phore)
(Cruciani et al. 2009). Additional information about the

Med Chem Res (2018) 27:679–692
687
Fig. 2 Compound 1 and its analogous external prediction on our
already published model (Goracci et al. 2013). PLD+ and PLD-
compounds are reported in blue and red, respectively. Yellow pro-
jections correspond to compounds in Table 1. In particular, com-
pounds 2 and 6 are highlighted in (a); in b, a detailed inspection of the
compounds 1, 3–5, and 7–9 projections, is shown with an enlargement
of (a). In c, the alignment of compounds 1, 3–9 on the PLD-phore
model generated using FLAP is reported (Goracci et al. 2015; Baroni
et al. 2007). Thus, based on physico-chemical properties and fitting to
the PLD-phore, compounds 3–5, 7–9 are predicted as likely to induce
phospholipid accumulation (PLD+ compounds), while compounds 2
and 6 are predicted as PLD- (color figure online)
see Materials and methods section) with respect to the
commercial sample (AN-465/42886418 from SPECS,
named here 1-com, 96% pure by LC-MS) were compared.
Viability of HepG2 cells, following 24 h exposure to var-
ious concentrations of 1-com or 1-syn, was measured by
neutral red (NR) uptake. This assay is based on the ability
of viable cells to take up the supravital dye NR by active
transport, following its incorporation and binding to lyso-
somes. Alteration in the cell surface lead to decreased
uptake and binding of NR (see Materials and methods
section for more details). As shown in Fig. 3, the 1-syn
caused a decrease of NR uptake of about 70% at the highest
concentration tested, while the 1-com had no effects. The
IC50 for the synthesized compound is 124.2 μM.
Scheme 1 Synthesis of compounds 1–6. Reagents and conditions: (i)
1. MeOH, r.t. 24 h; 2. NaBH4, 4 h; (ii) HCl in dioxane 4.0 M
Biological evaluation
Comparison of cytotoxicity and NBD-phospholipid uptake
induction for the commercial and the newly synthesized
compound 1
The biological properties of the newly synthesized com-
Afterwards, the possible PLD induction by 1-com and 1-
pound 1 (1-syn, purity >99% by LC-MS; for more details
syn was evaluated by fluorescence assays, measuring the

688
Med Chem Res (2018) 27:679–692
Scheme 2 Synthesis of compounds 7–8. Reagents and conditions: (i) 1. MeOH, r.t. 24 h; 2. NaBH4, 4 h; (ii) HCl in dioxane 4.0 M; (iii) p-TsOH,
toluene, reflux; (iv) 1. LiAlH4, THF, reflux, 24 h; 2. HCl in dioxane 4.0 M
com did not exhibit phospholipid accumulation at the tested
compound concentration. However, when the same assay
was performed with 1-syn at 100 µM, an increased uptake
of NBD-PC was observed; in fact, the fluorescent ratio
between NBD-PC and Hoechst was 1.53 0.24, compar-
able to the one of amiodarone at 12 µM concentration (1.56
0.27). Thus, the lower purity of 1-com might have
negatively influenced the assignment in our previous work
(Goracci et al. 2015). When NBD-PE was used instead of
NBD-PC, a significant increased accumulation of the
fluorescent phospholipid was observed for both 1-syn and
1-com at the highest concentrations tested (Fig. 4f). These
results suggest that compound 1 can be classified as a
potential weak PLD+, although previously reported as a
PLD- one (Goracci et al. 2015) confirming the quality of the
two toxicophore models for PLD risk prediction (Slavov
et al. 2014; Goracci et al. 2015). It is noteworthy that also
the reference compound amiodarone increased the uptake of
NBD-PE at a greater extent respect to NBD-PC, suggesting
a greater sensitivity of NBD-PE with respect to NBD-PC. In
addition, the purity grade of the tested compounds resulted
as an important aspect to consider in the case of high-
throughput screening campaigns.
Scheme 3 Synthesis of compound 9. Reagents and conditions: (i)
NaH, DMF, r.t., 16 h
Fig. 3 Effect of 1-com (yellow) and 1-syn (green) compound 1 on
HepG2 cell viability. HepG2 were treated for 24 h with selected
compounds at the indicated concentrations and viability was assessed
by NR uptake assay. Results are expressed as percentage viability of
cells exposed to selected compounds with respect to vehicle control
(Cveh). Statistical analysis was performed using Student’s t-test.
Values are the mean SD of three independent experiments done in
triplicate (***p < 0.001) (color figure online)
Cytotoxicity and NBD-PE uptake induction for the analogs
of compound 1
accumulation of the two fluorescent lipids NBD-PC and
NBD-PE (Fig. S1, Supporting Information). As mentioned
in the Introduction section, the phosphatidylethanolamine
derivative was reported to be a more sensitive probe for
PLD risk detection than the phosphatidylcholine one
(Bhandari et al. 2008). To take into account cell number, the
NBD-PC and NBD-PE fluorescence intensities (Fig. 4a, b)
were normalized using Hoechst 33342 (Fig. 4c, d), as
described elsewhere (Kasahara et al. 2006; Goracci et al.
2015). Using NBD-PC (Fig. 4e) we could confirm that 1-
Viability of HepG2 cells for the eight synthesized analogs
2–9 were assayed by NR uptake, as previously described for
compound 1. Figure 5 shows the lysosomal uptake of NR
by HepG2 cells incubated at various concentrations of
compounds 2–9; AMIO and 1-syn are reported as refer-
ences at 12 and 100 μM, respectively. Among the analogs,
only compound 2, bearing the pyridine group, decreased
NR uptake at 100 μM (Fig. 5), indicating an impairment of
lysosomal function. The other analogs did not significantly
affect NR uptake at all the concentration tested. Afterwards,

Med Chem Res (2018) 27:679–692
689
Fig. 4 Measurement of NBD-PC (a) and NBD-PE (b) uptake in
HepG2 cells treated with the commercial (yellow, 1-com) and the
newly synthesized (green, 1-syn) compound 1. HepG2 cells were
incubated with increasing concentrations of compound 1 in the pre-
sence of 40 μM NBD-PC and 50 μM NBD-PE. Amiodarone (AMIO)
at 12 µM concentration was used as positive control. After measure-
ment of NBD-PC and NBD-PE fluorescence, nuclei were stained with
Hoechst 33342 (c–d). Fluorescence intensities of NBD-PC, NBD-PE,
and Hoechst 33342 were measured at 485/538, 462/536, and 355/460
nm (Ex/Em) respectively, and results were expressed as percentage of
vehicle controls (Cveh). Normalized values were calculated by
dividing the NBD-PC or NBD-PE value by the Hoechst 33342 value
(e–f). Statistical analysis was performed using Student’s t-test. Results
represent the mean SD of three independent experiments done in
triplicate (*p < 0.05; ***p < 0.001) (color figure online)
the PLD effect of the analogs of compound 1 was evaluated
by the NBD-PE fluorescence assay. Figure 6 shows the
level of accumulation of the fluorescent lipid at increasing
concentrations of tested compounds, before and after nor-
malization with Hoechst 33342. AMIO and 1-syn were
used as positive controls, at 12 and 100 μM, respectively.
Among compounds 2–9, only compound 7 displayed a
degree of NBD-PE accumulation comparable to that of
100 μM 1-syn with a significant lipid accumulation even at
50 μM. A weak effect was also observed for compound 4,

690
Med Chem Res (2018) 27:679–692
Fig. 5 Effect of compounds 1-syn, and 2–9 on HepG2 cell viability.
HepG2 were treated for 24 h with the selected compounds at the
indicated concentrations and viability was assessed by NR uptake
assay. Amiodarone (AMIO) at 12 µM concentration was used as
positive controls. Results are expressed as percentage viability of cells
exposed to selected compounds with respect to vehicle control (Cveh).
Statistical analysis was performed using Student’s t-test. Values are
the mean SD of three independent experiments done in triplicate
(**p < 0.01; ***p < 0.001)
but at 100 μM only. It is noteworthy that compound 7
represents the only tested analog almost fitting the three-
centered toxicophore model by Slavov (Slavov et al. 2014).
Indeed, the distances connecting the tertiary amino group to
the centroids of the two aromatic rings are 6.3 and 6.6 Å
(i.e., in the ranges 5.5–7 and 3.5–7 Å, proposed for this
model), while the distance between the two centroids of the
aromatic rings is slightly lower (3.2 Å) than the proposed
range (4–5 Å) (Fig. S2, in the Supporting Information).
However, compounds 1, 4, and 7 can be generally classified
as weak PLD inducers, compared to the effect exerted by
AMIO. Indeed, the fluorescent NBD-PE/Hoechst ratio for
compound 1, 4, and 7 at 100 μM was 1.53 0.24, 1.39
0.1, and 1.71 0.12, respectively, while for AMIO a ratio
of 2.10 0.09 at 12 μM was observed. Concerning com-
pound 2, although cell viability studies suggested an
impairment of lysosomal function, this effect seems to be
not associated to lipid accumulation induction.
Fig. 6 Measurement of NBD-PE uptake in HepG2 treated with
compounds 2–9. Amiodarone (AMIO) and compound 1-syn at 12 μM
and 100 μM concentrations, respectively, were used as positive
controls. NBD-PC fluorescence (a) and Hoechst 33342 fluorescence
(b) were measured and results are expressed as percentage of
vehicle treated cells (Cveh). Normalized values were calculated
as ratio between the NBD-PE and the Hoechst 33342 values (c).
Statistical analysis was performed using Student's t-test. Results
represent the mean S.D. of three independent experiments done in
triplicate (*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 drug-treated HepG2
cells vs. vehicle controls)
A comparison between biological findings for NBD-PE
uptake induction and in silico prediction in Table 1 suggests
that for this series of compounds, false positive assignments
rather than false negative ones are likely to occur, with
predictions based on the Ploemen’s rules being the less
affected ones. However, Ploemen’s rules were still not
found to be a general solution for PLD prediction, per-
forming worse than other methods when applied to a
broader chemical space (van de Water et al. 2011; Goracci
et al. 2015). Therefore, although the three new PLD indu-
cers 1, 4, 7 were successfully predicted by all the tested in
silico methods despite the weakness of their effect, the high
number of false positives using in silico tools could lead to
discard promising compounds at an early discovery phase.
Conclusions
In summary, we believe that this case study leads to three
major considerations. First of all, even though the
assumption that CADs are PLD inducers is commonly
accepted, CAD-like compounds can be PLD- or weak PLD
+ compounds. Therefore, considering the relevant role of
CADs in many therapeutic areas, the application of in silico
tools tout court without considering uncertainty levels can
lead to the discarding of promising compounds. Secondary,
the selection of the in vitro assay is determinant when weak
CAD-like PLD inducers are tested. Thus, homogeneous
data from the same assay should be used for model gen-
eration. Finally, when weak PLD inducers or CAD-like

Med Chem Res (2018) 27:679–692
691
Cruciani G, Crivori P, Carrupt PA, Testa B (2000) Molecular fields in
quantitative structure–permeation relationships: the VolSurf
approach. J Mol Struct 503:17–30
Cruciani G, Milletti F, Storchi L, Sforna G, Goracci L (2009) In silico
pKa prediction and ADME profiling. Chem Biodivers
6:1812–1821
DiMasi JA, Feldman L, Seckler A, Wilson A (2010) Pharmaceutical
innovation in the 21st century: new drug approvals in the first
decade, 2000–2009. Clin Pharmacol 87:272–277
Fischer H, Atzpodien EA, Csato M, Doessegger L, Lenz B, Schmitt G,
Singer T (2012) In silico assay for assessing phospholipidosis
potential of small druglike molecules: training, validation, and
refinement using several data sets. J Med Chem 55:126–139
Goracci L, Ceccarelli M, Bonelli D, Cruciani (2013) Modeling
phospholipidosis induction: reliability and warnings. J Chem Inf
Model 53:1436–1446
Goracci L, Buratta S, Urbanelli L, Ferrara G, Di Guida R, Emiliani C,
Cross S (2015) Evaluating the risk of phospholipidosis using a
new multidisciplinary pipeline approach. Eur J Med Chem
92:49–63
Halliwell WH (1997) Cationic amphiphilic drug-induced phospholi-
pidosis. Toxicol Pathol 25:53–60
Hanumegowda UM, Wenke G, Regueiro-Ren A, Yordanova, Corradi
JP, Adams SP (2010) Phospholipidosis as a function of basicity,
lipophilicity, and volume of distribution of compounds. Chem
Res Toxicol 23:749–755
non-inducers are included in the dataset used for in silico
model generation, they may introduce a serious risk of
noise. Indeed, due to the lack of quantitative data on PLD
induction, PLD datasets available in the literature are based
only on two classes and weak inducers could be cause of
discordant PLD effect, as their definition as PLD+ or PLD-
might be method dependent. Thus, in our opinion, avoiding
the addition of so weak PLD inducers in PLD datasets
might be a valuable strategy to work with homogeneous and
reliable data, and the limit of prediction for borderline
compounds should be taken into account. In this context,
the use of the PLS-DA approach could be a valuable
strategy to visually inspect borderline compounds.
Acknowledgements The authors would like to thank Prof. Fran-
cesco Galli for generously sharing his laboratory facilities for biolo-
gical analysis, and Dr. Simon Cross for helpful comments and for
English revision. Financial support from the Italian MIUR within the
“FIRB-Futuro in Ricerca 2010” Program—Project RBFR10×500 is
gratefully acknowledged.
Compliance with ethical standards
Ikeda K, Hirayama M, Hirota Y, Asa E, Seki J, Tanaka Y (2008)
Drug-induced phospholipidosis is caused by blockade of man-
nose 6-phosphate receptor-mediated targeting of lysosomal
enzymes. Biochem Biophys Res Commun 377:268–274
Ivanciuc O (2008) Weka machine learning for predicting the phos-
pholipidosis inducing potential. Curr Top Med Chem
8:1691–1709
Conflict of interest The authors declare that they have no competing
interests.
References
Kasahara T, Tomita K, Murano H, Harada, Tsubakimoto K, Ogihara
T, Ohnishi S, Kakinuma C (2006) Establishment of an in vitro
high-throughput screening assay for detecting phospholipidosis-
inducing potential. Toxicol Sci 90:133–141
Artese A, Cross S, Costa G, Distinto S, Parrotta L, Alcaro S, Ortuso F,
Cruciani G (2013) Molecular interaction fields in drug discovery:
recent advances and future perspectives. Wires Comput Mol Sci
3:594–613
Kazmi F, Hensley T, Pope C, Funk RS, Loewen GJ, Buckley DB,
Parkinson A (2013) Lysosomal sequestration (trapping) of lipo-
philic amine (cationic amphiphilic) drugs in immortalized human
hepatocytes (Fa2N-4 cells). Drug Metab Dispos 41:897–905
Kodavanti UP, Mehendale HM (1990) Cationic amphiphilic drugs and
phospholipid storage disorder. Pharmacol Rev 42:327–354
Kruhlak NL, Choi SS, Contrera JF, Weaver JL, Willard JM, Hastings
KL, Sancilio LF (2008) Development of a phospholipidosis
database and predictive quantitative structure-activity relationship
(QSAR) models. Toxicol Mech Methods 18:217–227
Baroni M, Cruciani G, Sciabola S, Perruccio F, Mason JS (2007) A
common reference framework for analyzing/comparing proteins
and ligands. Fingerprints for ligands and Proteins (FLAP): theory
and application. J Chem Inf Model 47:279–294
Bhandari N, Figueroa DJ, Lawrence JW, Gerhold DL (2008) Phos-
pholipidosis assay in HepG2 cells and rat or rhesus hepatocytes
using phospholipid probe NBD-PE. Assay Drug Dev Technol
6:407–419
Blomme EA, Will Y (2016) Toxicology strategies for drug discovery:
present and future. Chem Res Toxicol 29:473–504
Lepri S, Nannetti G, Muratore G, Cruciani G, Ruzziconi R, Mercorelli
B, Palu G, Loregian A, Goracci L (2014) Optimization of small-
molecule inhibitors of influenza virus polymerase: from
thiophene-3-carboxamide to polyamido scaffolds. J Med Chem
57:4337–4350
Chapy H, Goracci L, Vayer P, Parmentier Y, Carrupt PA, Decleves X,
Scherrmann JM, Cisternino S, Cruciani
G
(2015)
Pharmacophore-based discovery of inhibitors of a novel drug/
proton antiporter in human brain endothelial hCMEC/D3 cell
line. Br J Pharmacol 172:4888–4904
Lowe R, Mussa HY, Nigsch F, Glen RC, Mitchell JB (2012) Pre-
dicting the mechanism of phospholipidosis. J Cheminform 4:2
Lullmann H, Lullmann-Rauch R, Wassermann O (1975) Drug-induced
phospholipidoses. II. Tissue distribution of the amphiphilic drug
chlorphentermine. Crit Rev Toxicol 4:185–218
Mannhold R, Berellini G, Carosati E, Benedetti P (2006) AOME
properties and their prediction major causes for failure in drug
development are unsuitable pharmacokinetic properties of drug
candidates including absorption, distribution, metabolism, and
excretion (ADME), which were traditionally measured at rather
late stages of drug development. Nowadays, the testing of ADME
properties is done much earlieri. In: Cruciani G (ed) Molecular
interaction fields: applications in drug discovery and ADME
prediction. Wiley-VCH, New York, 2006, vol. 27 173–196
Chatman LA, Morton D, Johnson TO, Anway SD (2009) A strategy
for risk management of drug-induced phospholipidosis. Toxicol
Pathol 37:997–1005
Choi SS, Kim JS, Valerio LG, Sadrieh N (2013) In silico modeling to
predict drug-induced phospholipidosis. Toxicol Appl Pharmacol
269:195–204
Cross S, Baroni M, Goracci L, Cruciani G (2012a) GRID-based three-
dimensional pharmacophores I: FLAPpharm, a novel approach
for pharmacophore elucidation. J Chem Inf Model 52:2587–2598
Cross S, Ortuso F, Baroni M, Costa G, Distinto S, Moraca F, Alcaro S,
Cruciani G (2012b) GRID-based three-dimensional pharmaco-
phores II: PharmBench, a benchmark data set for evaluating
pharmacophore elucidation methods.
J Chem Inf Model
52:2599–2608

692
Med Chem Res (2018) 27:679–692
Milletti F, Storchi L, Goracci L, Bendels S, Wagner B, Kansy M,
Cruciani G (2010) Extending pKa prediction accuracy: high-
throughput pKa measurements to understand pKa modulation of
new chemical series. Eur J Med Chem 45:4270–4279
Mingeot-Leclercq MP, Tulkens PM (1999) Aminoglycosides:
nephrotoxicity. Antimicrob Agents Chemother 43:1003–1012
Molecular Discovery Ltd. http://www.moldiscovery.com/. Accessed
11 Feb 2016
structure-activity relationship. Am J Physiol Lung Cell Mol
Physiol 287:L438–L447
Reasor MJ, Hastings KL, Ulrich RG (2006) Drug-induced phospho-
lipidosis: issues and future directions. Expert Opin Drug Saf
5:567–583
Repetto G, del Peso A, Zurita JL (2008) Neutral red uptake assay for the
estimation of cell viability/cytotoxicity. Nat Protoc 3:1125–1131
Sirci F, Goracci L, Rodriguez D, an Muijlwijk-Koezen J, Gutierrez-de-
Muehlbacher M, Tripal P, Roas F, Kornhuber J (2012) Identification
of drugs inducing phospholipidosis by novel in vitro data.
ChemMedChem 7:1925–1934
Munic V, Banjanac M, Kostrun S, Nujic K, Bosnar M, Marjanovic N,
Ralic J, Matijasic M, Hlevnjak M, Erakovic Haber V (2011)
Intensity of macrolide anti-inflammatory activity in J774A.1 cells
positively correlates with cellular accumulation and phospholi-
pidosis. Pharmacol Res 64:298–307
Nadanaciva S, Lu S, Gebhard DF, Jessen BA, Pennie WD, Will Y
(2011) A high content screening assay for identifying lysoso-
motropic compounds. Toxicol In Vitro 25:715–723
Nonoyama T, Fukuda R (2008) Drug-induced phospholipidosis:
Pathological aspects and its prediction. J Toxicol Pathol 21:9–24
Orogo AM, Choi SS, Minnier BL, Kruhlak NL (2012) Construction
and consensus performance of (Q)SAR models for predicting
phospholipidosis using a dataset of 743 compounds. Mol Inf
31:725–739
Pelletier DJ, Gehlhaar D, Tilloy-Ellul, Johnson TO, Greene (2007)
Evaluation of a published in silico model and construction of a
novel Bayesian model for predicting phospholipidosis inducing
potential. J Chem Inf Model 47:1196–1205
Ploemen JP, Kelder J, Hafmans T, van de Sandt H, van Burgsteden
JA, Salemink PJ, van Esch E (2004) Use of physicochemical
calculation of pKa and CLogP to predict phospholipidosis-
inducing potential: a case study with structurally related piper-
azines. Exp Toxicol Pathol 55:347–355
Teran H, Mannhold R (2012) Ligand-, structure- and
pharmacophore-based molecular fingerprints: a case study on
adenosine A(1), A (2A), A (2B), and A (3) receptor antagonists. J
Comput Aided Mol Des 26:1247–1266
Slavov SH, Wilkes JG, Buzatu DA, Kruhlak NL, Willard JM, Hanig
JP, Beger RD (2014) Computational identification of a phos-
pholipidosis toxicophore using (13)C and (15)N NMR-distance
based fingerprints. Bioorg Med Chem 22:6706–6714
SPECS. http://www.specs.net/. Accessed 18 Dec 2015
Sun H, Shahane S, Xia M, Austin CP, Huang R (2012) Structure based
model for the prediction of phospholipidosis induction potential
of small molecules. J Chem Inf Model 52:1798–1805
Tomizawa K, Sugano K, Yamada H, Horii I (2006) Physicochemical
and cell-based approach for early screening of phospholipidosis-
inducing potential. J Toxicol Sci 31:315–324
Tondi D, Cross S, Venturelli A, Costi M, Cruciani G, Spyrakis F
(2016) Decoding the structural basis for carbapenem hydrolysis
by class A β-lactamases: fishing for a pharmacophore. Curr Drug
Targets 17:983–1005
van de Water FM, Havinga, Ravesloot WT, Horbach GJ, Schoonen
WG (2011) High content screening analysis of phospholipidosis:
validation of a 96-well assay with CHO-K1 and HepG2 cells for
the prediction of in vivo based phospholipidosis. Toxicol In Vitro
25:1870–1882
Waring MJ, Arrowsmith J, Leach AR, Leeson PD, Mandrell S, Owen
RM, Pairaudeau G, Pennie WD, Pickett SD, Wang J, Wallace O,
Weir A (2015) An analysis of the attrition of drug candidates
from four major pharmaceutical companies. Nat Rev Drug Dis-
cov 14:475–486
Quaglino D, Ha HR, Duner E, Bruttomesso D, Bigler L, Follath F,
Realdi G, Pettenazzo A, Baritussio A (2004) Effects of metabo-
lites and analogs of amiodarone on alveolar macrophages: