Psoralen derivatives as inhibitors of NF-κB/DNA interaction: Synthesis, molecular modeling, 3D-QSAR, and biological evaluation

Article abstract of DOI:10.1021/jm3009647

Some new psoralen derivatives were synthesized and evaluated as inhibitors of NF-κB/DNA interaction, with the aim to investigate the structural determinants required to inhibit this interaction. Starting from molecular docking studies, several possible protein binding sites were proposed and several three-dimensional quantitative structure-activity relationship (3D-QSAR) models were built using the docked poses of 29 (the most active psoralen in the series) as templates for alignment of the inhibitors. The selected best model was validated through the prediction of the activity of 17 novel compounds. All the experimental data agreed with the computational experiments, supporting the reliability of the computational approach. The hypothesis about the interaction with NF-κB was also supported by surface plasmon resonance based assays using compound 29. All the collected data allowed the identification of compound 29 as a potential candidate for the development of pharmaceutical strategies against the inflammatory phenotype of cystic fibrosis.

Full text of DOI:10.1021/jm3009647

Article
pubs.acs.org/jmc
Psoralen Derivatives as Inhibitors of NF-κB/DNA Interaction:
Synthesis, Molecular Modeling, 3D-QSAR, and Biological Evaluation
Giovanni Marzaro,† Adriano Guiotto,† Monica Borgatti,‡ Alessia Finotti,‡ Roberto Gambari,‡
Giulia Breveglieri,‡ and Adriana Chilin*,†
†Dipartimento di Scienze del Farmaco, Universita
̀
degli Studi di Padova, Via Marzolo 5, 35131 Padova, Italy
‡Dipartimento di Scienze della Vita e Biotecnologie, Universita
̀
di Ferrara, Ferrara, Italy
S
* Supporting Information
ABSTRACT: Some new psoralen derivatives were synthesized and evaluated as
inhibitors of NF-κB/DNA interaction, with the aim to investigate the structural
determinants required to inhibit this interaction. Starting from molecular docking
studies, several possible protein binding sites were proposed and several three-
dimensional quantitative structure−activity relationship (3D-QSAR) models were built
using the docked poses of 29 (the most active psoralen in the series) as templates for
alignment of the inhibitors. The selected best model was validated through the
prediction of the activity of 17 novel compounds. All the experimental data agreed with
the computational experiments, supporting the reliability of the computational
approach. The hypothesis about the interaction with NF-κB was also supported by
surface plasmon resonance based assays using compound 29. All the collected data allowed the identification of compound 29 as
a potential candidate for the development of pharmaceutical strategies against the inflammatory phenotype of cystic fibrosis.
approaches, including virtual screening, molecular docking,
electrophoretic mobility shift assay, and expression of NF-κB
regulated genes studied by RT-PCR and Bio-plex analyses,17 we
found a good correlation between molecular interaction with
the NF-κB factor and inhibition of NF-κB activity studied in
cell-free conditions (EMSA) and in cellular systems. In
particular, psoralens such as those reported in Figure 1 were
INTRODUCTION
■
NF-κB is a transcription factor involved in the control of a large
number of normal cellular and organism processes, such as
immune and inflammatory responses, cellular growth and
apoptosis.1−3 The deregulation of NF-κB is associated with
many diseases such as AIDS and other viral infections, arthritis
and other inflammatory diseases, cancer, and genetic
disorders.4−7 In this way, targeting NF-κB could be of great
interest in order to find new therapeutic agents, mainly
anticancer and/or anti-inflammatory compounds. In the case of
cystic fibrosis (CF), one of the most important NF-κB
regulated targets is the gene coding the proinflammatory
cytokine IL-8.8,9 In fact, IL-8 is the most important
inflammatory protein induced by Pseudomonas aeruginosa
infection.10,11 With respect to this issue, we have recently
characterized the transcriptional control of the human IL-8
gene, demonstrating the role of NF-κB using the transcription
factor decoy approach with decoy molecules mimicking NF-κB
binding sites.12 In conclusion, targeting IL-8 gene expression
appears to be therapeutically relevant and NF-κB should be
considered among the most important molecules to inhibit, in
order to achieve anti-inflammatory and IL-8 inhibitory effects
on CF cells.
Figure 1. Structure of reference compounds.
considered as interesting leads for the development of
pharmaceutical strategies against the inflammatory phenotype
of CF, since they exhibited high efficiency in inhibiting NF-κB/
DNA interactions and also IL-8 gene expression in TNF-α
treated IB3-1 CF cells. Unfortunately, this interesting feature
should be considered just as a proof-of-concept based on a
limited number of molecules; in agreement, no structure−
activity relationship was attempted.
With the aim to establish a structure−activity relationship
and to define the structural determinants required to inhibit the
NF-κB/DNA interaction through NF-κB binding, we synthe-
sized and evaluated several new psoralen derivatives. The
Among NF-κB inhibitors, we found several psoralen
derivatives. For instance, 4,6,4′-trimethylangelicin (TMA) was
proven to inhibit the recruitment of NF-κB to the IL-8 gene
promoter and to exert inhibitory effects on the accumulation of
IL-8 mRNA and on the secretion of IL-8 protein.13 Moreover,
other psoralen derivatives able to inhibit NF-κB/DNA
interactions and the related IL-8 gene expression were
identified.14−19 On the basis of a combination of different
Received: July 5, 2012
Published: February 17, 2013
© 2013 American Chemical Society
1830
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
binding site of the compounds was proposed by combining
molecular docking and 3D-QSAR studies. In fact, although the
crystallographic structure of NF-κB was reported,20 no
structure with bound inhibitor was resolved to date. Another
major issue with NF-κB is its molecular weight (higher than 37
kDa), which impairs the specific characterization of the whole
protein through NMR experiments.21 Consequently, no
information regarding the binding modes of known NF-κB/
DNA interaction inhibitors is reported in literature. In the past
years, however, the increase in efficiency of computational tools
has highlighted the following: (i) molecular docking studies, in
particular “blind docking” strategies,22,23 are useful tools in the
identification of putative binding site(s); (ii) robust and
effective 3D-QSAR models can also be generated when the
tridimensional structure of the binding site is unknown;24 (iii)
validated 3D-QSAR models can be used for the so-called
“receptor mapping techniques”,25 which allow the generation of
hypothetical model of the binding site (receptor site model).26
In this respect, a crucial aspect of the 3D-QSAR modeling is the
choice of the ligands conformations. Indeed, a major limitation
of the “receptor mapping techniques” is that the computed
model site is usually derived from the minimized ligands
structures that could not correspond to the bioactive
conformations.24 As a consequence, a so computed model
site could not perfectly overlap with the real protein binding
site.
synthesized according to previously reported methods,27,28
slightly modified to take advantage of MAOS (microwave
assisted organic synthesis), as described for compound 2.19
Acid derivatives were activated as acyl chlorides and then
reacted with the opportune amines or amino acids to give the
desired amides, paying attention to reaction temperature to
avoid decarboxylation of psoralen acids (Scheme 1).
Screening of Inhibitors of NF-κB/DNA Interaction by
EMSA Assays. In order to obtain a first indication on
biological activity, electrophoretic mobility shift assays (EMSA)
were performed using purified p50 NF-κB protein. This
approach represents a standard technology very useful in
determining the potency of inhibitors of NF-κB/DNA
interactions.10,15,17 Figure 3 reports a representative EMSA
study where compounds were first incubated with NF-κB
protein and then the 32P-labeled target NF-κB oligonucleotide
was added.
This analysis was performed for all 37 synthesized
compounds and used for comparison with chrome azurol
(CAS), aurin (AU), pyrocatechol violet (PV), and pyrogallol
red (PR), considered as reference compounds. These
compounds were reported to bind to the NF-κB DNA-binding
domains, leading to inhibition of NF-κB/DNA interactions.29
The EC50 values, reported in Table 2, showed that the most
efficient inhibitors of NF-κB/DNA binding were compounds 4,
29, and 32. Their activity is higher than that found for other
known NF-κB inhibitors, such as CAS, AU, PV, and PR,
exhibiting inhibition of NF-κB/DNA interactions at EC50
ranging between 0.03 and 0.1 mM, as detailed in Table 2.29
Since it is well-known that psoralens intercalate into DNA,30
linear dichroism experiments were performed to determine the
intercalative properties of the new derivatives. However, no
differences in the dichroic spectra were found between free
DNA and DNA incubated with our compounds (see Figure S1
in Supporting Information), thus suggesting that the nucleic
acid was not the major target of our psoralen derivatives.
Computational Studies. Since the preliminary biological
experiments suggested NF-κB function as the molecular target,
docking studies were combined with 3D-QSAR modeling to
make a reliable hypothesis on the possible binding mode of the
psoralen derivatives to NF-kB protein, following the strategy
depicted in Figure 4.
Thus, in this work we tried to join the useful aspects of
molecular docking and 3D-QSAR modeling to derive a binding
site for psoralens.
RESULTS AND DISCUSSION
Chemistry. Several structural changes were introduced in
the points of diversity of the psoralen structure (Figure 2) to
■
Figure 2. Points of diversity in the psoralen structure.
The idea to merge molecular docking with 3D-QSAR
modeling to hypothesize a binding site for psoralens arose
from the attempt to join the different aspects of the two
computational techniques previously mentioned, i.e., (i) to
identify putative binding sites through docking studies, (ii) to
develop 3D-QSAR models also in the absence of information
regarding the binding site, (iii) to extract information on the
electrostatic and steric features of the binding site starting from
3D-QSAR models.
Moreover, in the 3D-QSAR modeling, if the compounds
alignment is made starting from the crystal (or the docked)
pose of a template ligand, the molecular fields can be visualized
in context with the protein target, thereby verifying the
coherence between the computed model and the protein
structure.31
explore the role of specific positions with respect to biological
activity, providing a small library of 37 compounds. The two
positions of the furan ring were substituted with one or two
methyl groups, a tert-butyl group, or a further condensed
cyclohexane ring, with the aim to investigate the significance of
steric hindrance, as well as for the benzene portion, which was
unsubstituted or substituted with a methyl function. The chain
length in the spacer portion was modulated with one to two
methylene spacers. The amide substituent on the side chain was
modified, changing the nature and the position of the carboxylic
group and also replacing it with a chlorine atom. Thus, the
newly synthesized compounds, as reported in Table 1, were
grouped in seven series according to furan ring substituents and
to the absence or presence of the methyl group on the central
benzene ring.
Thus, if a good coherence between the steric and the
electronic fields of a robust 3D-QSAR model and an NF-κB
region were obtained, we could find a plausible binding site for
our psoralens. However, instead of generating a receptor site
model (which requires other computational studies in addition
to 3D-QSAR modeling) and to “move” it on the protein
All the compounds were prepared with the same synthetic
strategy that consisted of (i) condensation of the furan ring on
the desired alkyl benzopyranylacetate or propionate, (ii)
hydrolysis of the ester function, and (iii) final condensation
with the opportune amine. In detail, starting (7-oxo-7H-
furo[3,2-g][1]benzopyran-6-yl)acetic or propionic acids were
1831
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
Table 1. Synthesized Compounds
structure until a full overlap was found, we restricted the
protein search space by detecting some putative binding sites
through a “blind docking” on compound 29. In this way, we
were able to keep down the computational costs and to
propose different reference poses that were in turn used for the
psoralens tridimensional alignment.
Docking Studies. The crystal structure of the NF-κB p50−
p50 homodimer in complex with DNA (PDB code 1NFK) was
1832
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
a
Scheme 1
a
Reagents and conditions: (a) 1-chloropinacolone or chloroacetone or
3-chloro-2-butanone or 2-chlorocyclohexanone, TEA/H2O, micro-
wave, 130 °C, 20 min;19 (b) 1 M NaOH, propan-2-ol, microwave, 120
°C, 5 min;19 (c) (i) SOCl2, rt, 3 h; (ii) amine, THF, TEA/H2O, rt. See
Table 1 for R specification.
Figure 3. Representative EMSA assays for compounds 25 and 29 with
different ability to inhibit NF-κB p50/DNA interaction. 10 ng of
human NF-κB p50 protein and different concentrations of compounds
were preincubated for 20 min at room temperature. Then 0.25 ng of
32P-labeled DNA/DNA target molecules was added to the samples for
a further 20 min at room temperature. Protein/DNA complexes were
separated by polyacrylamide gel electrophoresis, and autoradiography
was performed. The arrow indicates protein/DNA complexes. The
asterisk (∗) indicates the free 32P-labeled NF-κB oligonucleotides.
Figure 4. Computational strategy used for the identification of
psoralen binding site.
docking box) and AutoDock that performs the conformational
analysis and the scoring of the ligand based on the
precomputed maps. Initially the docking box was allowed to
contain the whole DNA-binding region of the enzyme, thus
performing the “blind docking”.22,23 As previously mentioned
and in accordance with the EMSA data, at this stage, compound
29 was chosen as “probe”, allowing the identification of four
putative binding sites. During the “blind docking” procedure,
the amide bond of the ligand was left free to rotate. In fact,
quantum mechanics (QM) calculations showed that the energy
difference between the trans and the cis amide conformations
was quite low (2.0 kcal/mol), with the trans conformation
characterizing the global minimum. The QM results were also
supported by NOESY experiments that showed, among the
others, a spatial correlation between the methylene spacer and
the amide aromatic ring (possible only in the case of the cis
amide conformation; see Figure S2 in Supporting Information).
The NOESY spectrum was recorded in DMSO-d6 instead of
D2O or D2O/NaOD, although the solvent could affect the
conformational equilibrium. In fact, in the presence of a large
amount of water, the methylene signal and the water peak fell
together, thus impairing the determination of the spatial
correlation. However, the QM and the NMR data suggested
that we cannot discard the existence of the cis-conformation a
priori. Since the blind docking procedure yielded some poses
with the amide dihedral quite far from planarity, the docking
studies were then performed again to obtain more exhaustive
poses. One docking box of appropriate dimensions was
centered on each previously determined site. At this stage,
the amide rotation was not allowed: the appropriate amide
Table 2. EC50 Values (mM) To Obtain 50% Inhibition of
NF-κB/DNA Interaction
compd
EC50
compd
EC50
compd
EC50
1
0.80
0.60
0.08
0.01
0.60
0.53
0.10
0.06
0.06
0.06
0.67
0.08
0.80
0.06
15
16
17
18
19
20
21
22
23
24
25
26
27
28
5.00
0.60
0.06
0.04
4.00
0.80
0.04
6.00
0.60
0.10
1.00
0.06
0.08
0.10
29
0.01
0.10
0.10
0.01
1.00
0.07
0.10
0.10
0.06
0.05
0.05
0.03
0.10
2
30
3
31
4
32
5
33
6
34
7
35
8
36
9
37
10
11
12
13
14
PV
AU
PR
CAS
chosen for the docking studies, employing the AutoDock 4.2
software (see the Experimental Section for a detailed
description of the docking experiments). AutoDock consists
of two engines, AutoGrid that computes several atom-specific
affinity maps along with one electrostatic map and one
desolvation map inside the binding site (collected into the
1833
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
conformations were chosen in accordance with the dihedral
values obtained in the blind docking (see Table 3). The results
of the docking studies are summarized in Figure 5.
Table 3. Conformations of Compound 29 Used in the
Docking Refinement Procedure
H−N−O−C dihedral
H−N−O−C dihedral
energy penalty
a
b
c
site
(blind docking)
(docking refinement)
(kcal/mol)
1
2
3
4
30° (cis-like)
61° (cis-like)
43° (cis-like)
180° (trans)
0° (cis)
0° (cis)
0° (cis)
180° (trans)
29.7
5.2
21.3
1.1
a
Dihedral values measured in the poses obtained in the blind docking
b
studies. Dihedral values forced in the docking refinement procedure
(cis conformation was chosen when the dihedral of the blind docking
obtained pose was closer to 0° than to 180°; trans conformation was
c
chosen when the dihedral was closer to 180° than to 0°). QM-based
energy difference between the docked pose (obtained after the
docking refinement protocol) and the global minimum energy
(obtained through QM-based structure minimization). All the QM
calculations were performed “in water”.
In all the cases, the carboxylic function of the compound was
found to establish H-bonds with positive charged amino acid
residues (Lys144(A) and Lys145(A) in site 1, Arg51(A) in site
2, Lys144(A) in site 3, Lys272(A) and Lys275(B) in site 4),
abundant in the DNA-binding region of NF-κB. Moreover, the
oxygen atom of the furan ring was also found to interact with
amide protons of the backbone. The computed binding scores
(ADT scores) were, in all the cases, almost equal, ranging from
−6.36 to −6.42, while the QM-based energy differences
between the docked poses and the minimized conformation
indicated that poses 1 and 3 were less reliable than poses 2 and
4 (see Table 3).
3D-QSAR Modeling. As previously mentioned, the
selection of the most probable binding site was then carried
out by means of 3D-QSAR studies, through the computation of
one model for each of the four putative binding sites. Thus, no
binding site was discarded a priori on the basis of the energy
penalties derived from the QM calculations. The reliabilities of
the sites were assessed on the basis of the statistical parameters
of the 3D-QSAR models (r2, q2lmo, SDEC, and SDEP) and of
the coherence between each model features and the related
binding site characteristics. Briefly, all the synthesized
compounds were superimposed onto each of the four docked
poses of compound 29, employing the Discovery Studio 3.1
suite.32 The structures of all the derivatives (except for
compound 29) were considered flexible, and the better
superimpositions were, in all the cases, obtained through the
“field fit” methodology with 80% steric and 20% electrostatic
contributions. As the docking studies had highlighted the
importance of the interaction of the carboxylate anion with
positively charged residues, the protonation state of the
compounds was also considered. Before the superimposition
process, all the molecular structures were hydrogenated at pH
7.4 and all the carboxylic functions were negatively charged.
The 3D-QSAR models were then built with the Open3-
DQSAR33 software as described in the Experimental Section.
The characteristics of the models are summarized in Table 4.
The model built on binding site 1 showed the lowest
correlation (r2 = 0.58), whereas the model built on binding site
3 showed good correlation in training but poor correlation in
internal validation (q2 = 0.25). On the basis of these results, the
Figure 5. Putative binding sites for psoralen derivatives. (A) Depiction
of the four binding sites with respect to the DNA-binding region
(white dotted circle) of NF-κB. (B−E) Interactions predicted for
compound 29 with protein residues at each site. The hydrogen bonds
are highlighted as yellow dotted lines. The poses obtained in the “blind
docking” are depicted as lines, while the refined poses are depicted as
sticks.
two binding sites were discharged. As reported in Table 4, the
number of active variables used for the PC analysis on sites 1
and 3 was considerably higher than the number for sites 2 and
4, reflecting not optimal molecular superimpositions. Indeed,
the models 1 and 3 were built on the poses that also showed
the highest energy penalties in QM calculations (Table 3).
Finally, the number of PCs used was higher than for the models
built on sites 2 and 4. Indeed, it is well-known that using a large
number of independent variables leads to models with good
correlations in training but poor performances in cross-
validation or in prediction. Also, binding site 2 was not further
considered: although the QSAR model built on this site showed
good performances in both training and validation, the
molecular fields did not agree with the protein structure. For
example, a sterically favorable area clashed against Val209, while
a sterically unfavorable area was placed in a protein cavity. On
the contrary, the model built on binding site 4, besides showing
the best performances, fully agreed with the protein structure
(vide post), thus indicating that this could reasonably be the
binding site for psoralen derivatives. All the calculated EC50
1834
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
Table 4. Statistics of 3D-QSAR Models
a
b
c
d
e
binding site
1
protein residues
r2
SDEC
q2
SDEP
active variables
4234
PC
lmo
Lys144(A)
Lys145(A)
Phe148(A)
Arg51(A)
0.579
0.447
0.152
1.050
5
2
3
4
0.842
0.819
0.947
0.222
0.273
0.163
0.654
0.252
0.731
0.411
0.599
0.374
1584
3998
1122
3
4
3
Lys144(A)
Lys272(A)
Lys275(B)
Asp276(B)
a
b
c
Amino acid residues of NF-κB involved in interactions with 29. Standard deviation error of calculations. Standard deviation error of predictions.
d
Number of independent variables (i.e., sum of the grid points for the steric and for the electrostatic fields retained after the variables reduction
e
procedures described in the Experimental Section) used for the PC analysis. Number of principal components (independent variables) in the
model. The number of PCs reported allowed the highest value of q2.
values, the standard errors, and the leverage values are reported
in Table S1 in Supporting Information.
13, 22; see Figure S4 in Supporting Information). The length
of the spacer portion (one or two methylene units) was not
fundamental (compare, for example, 29 with 32), since the
terminal carboxylic functions were always located in the same
manner (see Figure S5 in Supporting Information). However,
in almost all the cases, when an additional methylene function
was added to the amide portion (e.g., 5, 6, 11), the activity
remarkably decreased (see Figure S6 in Supporting Informa-
tion). The steric unfavorable areas clashed against Lys272A and
Lys275B (yellow surfaces in Figure 6B), whereas the steric
favorable areas were well accommodated in the protein cavities
(green surfaces in Figure 6C), justifying the slightly lower
activities of less hindered compounds (compare, for example,
21 with 29). On these bases, the selected model also explained
why in almost all the cases the presence of a methyl group in
the central benzene ring negatively affected the activities (see,
for example, 21 in comparison with 24 or 29 in comparison
with 35). We thus supposed that the compounds belonging to
classes B, C, E, and G could be accommodated in the binding
site only by inducing a change in the position of Lys275(B). To
confirm this hypothesis, compound 35 was docked in binding
site 4 allowing the Lys275(B) side chain flexibility. Satisfac-
torily, the obtained pose (Figure 7) fully agreed with that
Binding Mode and SAR Analysis. The pseudo
coefficients for the steric and the electrostatic fields of the
selected model were extracted and visualized to evaluate the
agreement of the model with the receptor structure (see an
example in Figure 6, and see also Figure S3 in Supporting
Information).
Figure 7. Comparison of the binding modes predicted for compound
29 (white sticks) and compound 35 (yellow sticks). NF-κB is depicted
as green and blue colored backbone. The Lys275 residue is depicted in
sticks colored accordingly to the related ligand. Note that in the case of
compound 35, the movement of the Lys275 side chain (induced by
the presence of the methyl group in the ligand) causes a loss of the
hydrogen bond with the pyran ring.
Figure 6. Pseudo coefficients of the selected 3D-QSAR model in
context with compound 29 (atom-colored sticks) and the target (cyan
and green ribbon backbone in parts A and B; gray and lemon surfaces
in part C). Red surfaces indicate negatively charged atoms favorable
volumes (level = −0.0008). Yellow surfaces indicate sterically
unfavorable volumes (level = −0.0006). Green surfaces indicate
sterically favorable volumes (level = +0.0006).
The negatively charged atoms favorable volumes (red
surfaces in Figure 6A) were surrounded by positively charged
residues (Lys272A, Arg305A, and Lys272B). This finding
highlighted the importance of an anionic function in the tested
molecules and justified the low activity of compounds not
bearing a carboxylic function (15, 19, 20) or of compounds
with a carboxylic function not properly accommodated (e.g., 1,
observed for compound 29. However, the loss of the hydrogen
bond between Lys275(B) and the oxygen atom of the pyran
ring (due to the induced movement of the amino acid side
chain) probably caused the reduction in the binding efficiency
(ADT-Scorecompd29 = −6.42; ADT-Scorecompd35 = −5.68).
Validation of the 3D-QSAR Model. In all the previously
discarded models, the compounds were characterized by the cis
1835
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
Table 5. Compounds in the Test Set and Measured EC50 (mM)
Figure 8. Selected 3D-QSAR model. (A) Williams’ plot indicating the applicability domain of the model. As all the compounds fall within
3
standard deviation units (horizontal lines) and have a leverage lower than the critical value (0.324 for the present model, vertical line), no outliers
were present. (B) Actual versus predicted pEC50 for psoralen derivatives in training (blue triangles), cross-validated (red triangles), and test (green
triangles) sets.
amide conformations. In order to exclude that the selected
model outperformed the others only because of trans amide
conformations, an additional model was also built using the
fully minimized conformation of compound 29 as a template
for the molecular alignment. However, in this case also, lower
statistical parameters were found (r2 = 0.74; q2 = 0.42),
suggesting that the active conformations could be different
from the lowest energy ones, as observed in many other cases.34
Also in this case, it cannot be stated that this is a confirmation
that site 4 is the real binding site for psoralens, but the
computational approach allowed us to propose at least
reasonable bioactive conformations.
The selected model was then used to predict the EC50 values
of a number of properly synthesized psoralen derivatives (Table
5). Following a previously reported approach,31 in order to test
the model, compounds structurally very close to those used in
the training set were included in the test set (38−41) as well as
molecules whose properties vary significantly (42−54), mainly
in terms of pKa properties and H-bond donor/acceptor
characteristics.
All the 17 molecules were synthesized with the same
synthetic strategy depicted in Scheme 1 and were evaluated by
means of EMSA experiments. The QSAR model was found to
predict in a very satisfactory manner the activities of the novel
compounds (r2
= 0.73, SDEP = 0.26; see Table S1 in
prediction
Supporting Information for all the calculated EC50 values, the
standard errors, and the leverage values). Interestingly, when
the activities were predicted with the discarded 3D-QSAR
models, very low correlations were found, with r2 of prediction
ranging from 0.21 to 0.32.
None of the newly synthesized derivatives were more active
than parent compound 29. These data confirmed the
importance of a bulky substituted furan (compare compounds
38−41 with 29) and of a terminal residue (i.e., the carboxylate
1836
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
Figure 9. BIA analysis of the binding of compound 29 to NF-κB. (A) Immobilization of NF-κB p50 on the sensor chip CM5 (research grade) by the
amine coupling procedure, as described in the Experimental Section. (B) Representative sensorgrams obtained after injection of compound 29 and
compound 50. Stable interactions are represented by RUres (residual RUs). (C) Summary of the bindings of 29 and 50 obtained in three
independent experiments. The presented results are the RU
SD bound to the NF-κB sensor chip. (D) Bindings of the double stranded
oligonucleotide 5′-CGCTGGGGACTTTCCACGG-3′ (sense strand) to the NF-κB sensor chip in the absence (solid line) or in the presence
(dotted line) of preloaded compound 29.
anion) able to form strong interactions with the positively
charged amino acids present in the DNA binding region of the
transcription factor.
Finally, the domain of applicability was also determined
(Figure 8).35 As no outliers were identified, all 54 compounds
fit into the 3D-QSAR model.
consistent with the EMSA experiments. In fact, when a double
stranded oligonucleotide on the NF-κB sensor chip preloaded
with compound 29 was injected, a full inhibition of DNA/NF-
κB interaction was observed (Figure 9C). While the possibility
that interaction of compound 29 with NF-kB might lead to
additional effects altering NF-kB/DNA interactions (including,
but not limited to, unfolding, even partial, of NF-kB) cannot be
ruled out, the BIA data supported the EMSA results and
suggested that compound 29 deserves further studies to
determine its possible inhibitory activity on NF-kB dependent
genes, including IL-8.
Effects of Compound 29 on the Expression of IL-8
Genes in TNF-α Treated IB3-1 Cells. Since the EMSA
experiments agreed in indicating psoralen 29 as the most
interesting NF-κB inhibitor, the study on the effect on cystic
fibrosis cell line was performed with this compound.
In order to determine the effects on IL-8 mRNA
accumulation, the expression of IL-8 gene was studied by
RT-PCR analysis of RNA extracted from TNF-α treated CF
IB3-1 cells cultured in the presence of increasing concen-
trations of compound 29 for 24 h. As already pointed out, the
accumulation of IL-8 mRNA was chosen to be analyzed
because IL-8 is one of the most expressed interleukins involved
in inflammatory processes and is strongly induced by TNF-α
treatment.36
Figure 10A shows the results obtained by quantitative RT-
PCR concerning the effects of compound 29 on TNF-α
induced IL-8 mRNA accumulation in cell. A clear induction of
IL-8 transcripts following TNF-α treatment was evident
(CTRL+). Significant inhibitory effects on IL-8 mRNA
accumulation were detected following exposure of TNF-α
induced IB3-1 cells to compound 29.
Surface Plasmon Resonance (SPR) Based Biospecific
Interaction Analysis (BIA) of Interactions between
Compound 29 and NF-κB. Since the docking experiments
suggested a direct binding of compound 29 to NF-κB, this
possibility was also investigated through biospecific interaction
analysis using the SPR-based biosensor BIAcore X-100
(Biacore, GE Healthcare). We first immobilized NF-κB p50
on a CM5 sensor chip using the amine coupling technology
(see the representative binding sensorgram shown in Figure
9A). Then we injected 29 (which was proposed to display
highly efficient binding to NF-κB) and 50 (which was proposed
to display low efficient binding to NF-κB). The results obtained
(depicted in Figure 9B and Figure 9C) suggested that
compound 29 was able to bind to NF-κB immobilized on the
sensor chip with higher efficiency than compound 50. In three
independent interactions experiments 765.5
150 RU
representing stable complexes between compound 29 and
7139.5 250 RU of immobilized NF-κB were found (Figure
9C). As a matter of fact, by this approach the binding site was
not precisely definable, as well as the stoichiometry of the
binding. However, taken together, the obtained results are
compatible with binding of compound 29 to the NF-κB factor.
In addition, it is important to emphasize that the same
experiment done with compound 50 showed a significant lower
binding activity (Figure 9C), in agreement with the computa-
tional studies. The SPR-based BIA analysis not only supported
the conclusions driven by the docking approach but was fully
1837
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
(iii) the length of the spacer between the psoralen nucleus and
the amide function; (iv) the nature of the amide moiety. The
evaluation of the inhibition of the NF-κB/DNA interaction
through EMSA analysis allowed the identification of 29 as the
most promising compound in the series. By use of 29 as
reference compound in molecular docking studies, four
hypothetical binding sites have been proposed. Then the
selection of the most plausible binding site was obtained by
comparison of several 3D-QSAR models. The final selected
model was validated through the prediction of the activities of
the other 17 psoralen derivatives, whose structural features
were quite different from those of the molecules in the training
set. The computational methodologies allowed also the
definition of a structure−activity relationship. In particular,
with respect to the four points of diversity, (i) the presence of a
steric hindrance on the furan ring positively modulated the
ability to inhibit the protein−DNA interaction, (ii) the
presence of a methyl group in the central benzene ring caused
a reduction in activity, (iii) the length of the spacer portion
(one or two methylene units) did not significantly influenced
the activities, and (iv) the amide moiety required a terminal
carboxylic function that probably interacted with positively
charged amino acids (as Lys and Arg).
To the best of our knowledge, this work constitutes the first
example of the combination of molecular docking and 3D-
QSAR modeling to propose a hypothetical binding mode.
Although it was not possible to demonstrate the accuracy of the
proposed binding site with structural evidence, all the
experimental data fully agreed with the computational experi-
ments, thus supporting, at least in part, the potential of the
proposed computational strategy. The selected model will be
used in the future for the rational design of novel and more
potent inhibitors.
Figure 10. Effects of compound 29 on TNF-α induced IL-8 gene
expression in IB3-1 cells: (A) quantitative RT-PCR analysis of IL-8
mRNA fold difference relative to untreated cells (CTRL −); (B) Bio-
plex analysis of IL-8 protein present in supernatants; (CTRL −)
untreated IB3-1 cells; (CTRL +) TNF-α treated IB3-1 cells. For each
experiment duplicate RT-PCR determinations were performed.
Results represent the average SD of three independent experiments.
The hypothesis of the binding of compound 29 to NF-κB
was further supported by the results obtained through
biospecific interaction analysis. The results were compatible
with the binding of compound 29 to NF-κB, leading to stable
formation of a complex that prevented the binding of NF-κB to
oligonucleotides, in full agreement with the EMSA experiments.
These data suggested that the interaction of compound 29 with
NF-κB played a key role in the inhibition of NF-κB/DNA
interactions. Furthermore, the inhibition of the NF-κB/DNA
interaction by preloaded compound 29 on immobilized NF-κB
suggested a binding to the DNA-binding site of NF-κB, in
agreement with molecular docking and 3D-QSAR studies.
Clearly, further efforts will be necessary to determine
whether compound 29 has a unique binding site and, more
importantly, the accuracy of the proposed binding site. In this
respect BIA experiments might be repeated using mutant NF-
kB proteins and/or recombinant NF-kB contructs mimicking
specific domains of NF-kB. In addition to the SPR-based BIA
approach, other technologies might be highly recommended,
including NMR of specific 13C−15N labeled domains of the
protein.
These results were also confirmed using Bio-plex technology
to analyze the release of IL-8 protein (pg/mL) in supernatants
collected from the same IB3-1 cells induced with TNF-α and
treated with compound 29. The results, shown in Figure 10B,
confirmed the inhibitory activity of compound 29 on
expression of IL-8 protein in IB3-1 cells at 25 μM. These
data suggest that compound 29 might specifically inhibit the
expression of the IL-8 proinflammatory gene.
To ensure that the reduction in IL-8 expression did not arise
from the impairment of cell viability, the IC50 of compound 29
was determined and cell viability evaluated by trypan blue
exclusion test. The measured IC50 was 355 9.2 μM; thus, it
can be excluded that the measured biological properties of the
compound could be due to cytotoxic effects.
CONCLUSION
■
NF-κB is a very important transcription factor involved in
several human pathologies, including osteoporosis,37 rheuma-
toid arthritis,38 and cancer.39,40 In addition, NF-κB is one of the
master transcription factors responsible for inflammation in CF
cells infected with Pseudomonas aeruginosa.11 Accordingly,
compounds that bind to NF-κB, thus interfering with the
NF-κB/DNA interaction and inhibiting IL-8 expression in CF
cells, are of great importance.10
Despite the fact that the data do not fully explain the
inhibition of NF-kB/DNA interaction by compound 29 and do
not allow exclusion of interactions of compound 29 with other
proteins/molecules present in tissue culture conditions or
within the cellular system employed, the collected biological
data suggest possible biomedical applications of compound 29.
In this respect, it is relevant that NF-κB inhibitors, such as
parthenolide41 and azithromycin,42 are employed in clinical
Starting from the structure of the previously reported
psoralen derivatives depicted in Figure 1, several novel
analogues have been synthesized. Four points of structural
diversity have been considered: (i) the furan substituents; (ii)
the presence/absence of a methyl in the central benzene ring;
1838
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
hydrogen atoms were merged with ADT, thus preparing the
appropriate PDBQT files for the molecular docking studies. The
superimpositions of molecules have been done using the Discovery
Studio 3.1 suite.32 All the 3D-QSAR studies have been performed
using Open3DQSAR 2.02 software.33
trials for CF; thus, compound 29 could also be an interesting
candidate for CF treatment.
Taken together, these data indicate that psoralens belonging
to this set might find an important place in the set of molecules
of interest for the development of pharmaceutical strategies
against the inflammatory phenotype of CF.
Blind Molecular Docking. The structure of NF-κB in complex
with the DNA has been imported in Chimera, and the water molecules
and the nucleic acid were removed. The so handled protein structure
was imported in ADT. The hydrogen atoms were added, and the
Gastaiger charges were computed. After the addition of the AutoDock
atom types the nonpolar hydrogen atoms were merged and the protein
was saved in PDBQT file format required for AutoDock and AutoGrid
tools. The entire protein structure has been contained in a box
(coordinate of the center: x = −2.37, y = 13.23, z = 19.60) with 0.55 Å
grid spacing and 80 × 80 × 80 points in order to allow the
computation of the atom-affinity maps through AutoGrid 4.2
application. The molecular docking simulations were then performed
with AD, using compound 29 as flexible ligand and with the searching
parameters previously reported for the blind docking strategy.23 The
amide bond of compound 29 was left free to rotate. Cluster analysis
was performed on docked results, with a root-mean-square tolerance
of 2.0 Å. All the poses with binding energy ranging from the minimum
ADT-Score to ADT-Score + 1 were then extracted, leading to the
identification of four different plausible binding sites for psoralen
derivatives.
Binding Poses Refinement. A docking refinement study has been
performed at each of the previously determined binding sites. In all the
cases, the boxes for AutoGrid were characterized by 0.375 Å spacing
and by 54 × 54 × 54 points. The boxes were centered at x = 12.43, y =
2.50, z = 21.80 (site 1); x = 11.92, y = 18.50, z = 20.68 (site 2); x =
6.27, y = 12.09, z = 20.78 (site 3); x = −11.11, y = 18.23, z = 15.33
(site 4). For each site, a total of 10 runs with a maximum of 2 500 000
energy evaluations were carried out, using the Lamarkian genetic
algorithm (LGA) searching engine. At each site, the amide bond of
compound 29 was constrained to the cis conformation or the trans
conformation, depending on the dihedral angle determined in the
“blind docking” procedure (see Table 3). Cluster analysis was
performed on docked results, with a root-mean-square tolerance of
2.0 Å, finally selecting the lowest energy pose.
Quantum Mechanical Studies. The fully minimized structure of
compound 29 (trans amide conformation) has been determined with
the Firefly QC package48 (which is partially based on the GAMESS-
US source code),49 setting the parameters as follows: $basis gbasis =
N31 ngasuss = 6 ndfunc = 1; npfunc = 1; dffsp = .true.; $pcm solvnt =
water; $contrl scftyp = rohf; runtyp = energy; dfttyp = b3lyp; icharge =
−1; mult = 1. The minimized structure of compound 29 with the cis
amide conformation has been determined using the same parameters,
with the amide dihedral angle constrained at 0°. All the QM-based
energies have been computed using the same parameters as above,
with the only exception of runtyp, which was set to “energy”.
Flexible Target Molecular Docking. The PDBQT file of NF-κB
was modified, allowing the Lys275(B) side chain flexibility. The box
for AutoGrid was characterized by 0.375 Å spacing and by 54 × 54 ×
54 points. The grid was centered at x = −11.11, y = 18.23, z = 15.33.
The molecular docking simulation was then performed with AD using
compound 35 as flexible ligand and with the searching parameters
described in the section “Binding Poses Refinement”.
3D-QSAR Modeling. The alignments of the compounds were
performed with the Discovery Studio 3.1 suite, using the “field fit”
algorithm. This methodology is based on the computation of a steric
and an electrostatic field for each molecule. Then the molecules are
superimposed in a manner such that the differences among the
computed fields are lower than possible. In the present case, the
contributions to the alignment were 80% for the steric field and 20%
for the electrostatic one. Compound 29 was considered as the rigid
template, using the coordinates previously determined through the
molecular docking studies (for the models built on the four putative
binding sites) or the QM-based fully minimized structure. All the other
molecular structures were considered flexible. The torsional incre-
ments were set to 10° for all the rotatable bonds except for the amide
EXPERIMENTAL SECTION
■
Chemistry. All commercial chemicals and solvents used were
analytical grade and were used without further purification. Analytical
thin layer chromatography (TLC) was performed on precoated silica
gel plates (Merck 60-F-254, 0.25 mm). Melting points were
determined on a Gallenkamp MFB-595-010M melting point apparatus
and are uncorrected. The 1H NMR spectra were recorded on a Bruker
300-AMX spectrometer with TMS as an internal standard. Coupling
constants are given in Hz, and the relative peaks area were in
agreement with all assignments. Elemental analyses were performed on
a Perkin-Elmer 2400 analyzer. Mass spectra were obtained using a
Applied Biosystem Mariner system 5220 with direct injection of the
sample. Starting (3′-tert-butyl-5′-methyl-7′-oxo-7H-furo[3,2-g][1]-
benzopyran-6′-yl)acetic and propionic acids (type A), (3′-tert-butyl-
5′,9′-dimethyl-7′-oxo-7H-furo[3,2-g][1]benzopyran-6′-yl)acetic and
propionic acids (type B), (3′,5′,9′-trimethyl-7′-oxo-7H-furo[3,2-g][1]-
benzopyran-6′-yl)acetic acid (type C), (2′,3′,5′-trimethyl-7′-oxo-7H-
furo[3,2-g][1]benzopyran-6′-yl)acetic acid (type D), (2′,3′,5′,9′-
tetramethyl-7′-oxo-7H-furo[3,2-g][1]benzopyran-6′-yl)acetic and pro-
pionic acids (type E), (4′-methyl-2′-oxo-6′,7′,8′,9′-
tetrahydrobenzofuro[3,2-g]-1-benzopyran-3′-yl)acetic and propionic
acids (type F), (4′,11′-dimethyl-2′-oxo-6′,7′,8′,9′-
tetrahydrobenzofuro[3,2-g]-1-benzopyran-3′-yl)acetic and propionic
acids (type G) were prepared according to literature methods,27,28,43
modified as described for the already reported compound 2.19 All the
amine reagents (2-aminobenzoic acid, 3-aminobenzoic acid, 4-
aminobenzoic acid, 4-aminomethylbenzoic acid, 4-aminophenylacetic
acid, DL-phenylalanine, 2-chlorobenzylamine, 4-aminobutyric acid, 3-
aminophenol, 4-aminophenol, 3-trifluoromethylaniline, 4-amino-N-
acetylaniline, 4-amino-N-methanesulfonylaniline, 4-aminobenzamide,
p-phenylendiamine) were used as purchased. Purity for all the tested
compounds was determined by elemental analyses and was found to
be equal to or more than 95%.
General Procedures for Amides 3−12, 14−21, and 23−54. A
solution of the above-mentioned starting acid (1.0 mmol) in thionyl
chloride (25 mL) was stirred at room temperature until a clear
solution was formed. The solvent was evaporated under a nitrogen
stream at room temperature, and the residue was dissolved in dry
tetrahydrofuran (10 mL). The solution was added dropwise to a
mixture of the amine (2.0 mmol) in triethylamine (0.3 mL, 2.1 mmol)
and water (5 mL). After reaction completion, the mixture was diluted
with water (40 mL) and acidified with 1 M HCl to pH 3. The resulting
precipitate was filtered off and crystallized from propan-2-ol to give 3−
12, 14−21, 23−54.
See Supporting Information for the data of all the synthesized
compounds.
Computational Methodologies. All the computational studies
were carried out on a 4 CPU (Intel Core2 Quad CPU Q9550, 2.83
GHz) ACPI x64 Linux workstation with Ubuntu 12.04 operating
system. The NF-κB/DNA complex structure was downloaded from
Protein Data Bank (PDB code 1NFK) and modified with Chimera
5.3.1 software.44 All the QM studies have been performed with Firefly
8.0 software at the B3LYP level with 6-31+G(d,p) basis set. All the
input files for Firefly have been prepared with the graphical interface of
Avogadro 1.1.0 software. For the molecular docking and the 3D-QSAR
studies, the psoralen structures have been prepared with MarvinSketch
5.5 software.45 The lowest energy conformations and the degree of
protonation at pH 7.4 were determined with OpenBabel 2.2.3
software,46 using the MMFF94s force field. All the docking studies
have been performed with AutoDock (AD) 4.2 software with
AutoDock Tools 1.5.4 (ADT) graphical interface.47 For compound
29 and 35, the Gasteiger charges were computed and the nonpolar
1839
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
bonds whose torsion values were set to 180° or to 0°, depending on
the amide conformations determined in the docking studies. The
aligned data sets were used for the computation of five 3D-QSAR
models. The steric field (calculated using an sp3-hybridized carbon
atom as probe) and a molecular mechanics based electrostatic field
were computed in a grid box that exceeded the largest molecule of 1 Å
in all directions. For both fields, a cutoff of 10 kcal/mol was used. The
variables with values lower than 0.05 were zeroed, while the variables
with standard deviations lower than 0.1 were removed. The active
variables retained after these variables reduction procedures were
grouped following the smart region definition approach described by
Pastor et al.50 Then the subset of variables that allowed the
computation of the model with the higher q2 value was selected by
means of fractional factorial design.51 In all the cases, the models were
cross-validated by means of 20% leave-multiple-out methodology. The
q2 values were used to determine the adequate number of principal
components. The full list of commands inputted in Open3DQSAR is
reported in Supporting Information. Compounds 1−37 were used for
the generation and the cross-validation of the 3D-QSAR models.
Compounds 38−54 were used for the external validation.
Biology. Electrophoretic Mobility Shift Assay (EMSA). EMSA
was perfomed as previously described.52 Briefly, double-stranded
synthetic oligodeoxynucleotides mimicking the NF-κB binding (NF-
κB, sense: 5′-CGC TGG GGA CTT TCC ACG G-3′) were employed.
Oligodeoxynucleotides were labeled with γ32P-ATP using 10 units of
T4-polynucleotide-kinase (MBI Fermentas) in 500 mM Tris-HCl, pH
7.6, 100 mM MgCl2, 50 mM DTT, 1 mM spermidine, 1 mM EDTA in
the presence of 50 mCi γ32P-ATP in a volume of 20 μL for 45 min at
37 °C. The mixture were brought to 150 mM NaCl, and 150 ng of
complementary oligodeoxynucleotide was added. Reaction temper-
ature was increased to 100 °C for 5 min and left diminishing to room
temperature overnight. Binding reactions were set up in a total volume
of 20 μL containing buffer TF plus 5% glycerol, 1 mM dithiothreitol,
10 ng of human NF-κB p50 protein with or without 10 ng of NF-κB
p65 protein (Promega) and different concentrations of compounds.
After an incubation of 20 min at room temperature, 0.25 ng of 32P-
labeled oligonucleotides was added to the samples for further 20 min
at room temperature, and then they were electrophoresed at constant
voltage (200 V) under low ionic strength conditions (0.25× TBE
buffer of 22 mM Tris borate and 0.4 mM EDTA) on 6%
polyacrylamide gels. Gels were dried and subjected to standard
autoradiographic procedures.
obtain IC50 concentration and cell viability evaluated by trypan blue
exclusion test.
Quantitation of IL-8 Transcripts. Total RNA was isolated
(TRIzol Reagent, Sigma-Aldrich, WI, U.S.) and retrotranscribed
(Promega Corporation, Madison, WI, U.S.). The resulting cDNA
was quantified by relative quantitative RT-PCR. The sequences of the
oligonucleotides used for amplification of IL-8 mRNA were 5′-GTG
CAG TTTT GCG GAC AAG T-3′ (forward) and 5′-TTA TGA ATT
CTC AGC CCT CTT CAA AAA CTT CTC-3′ (reverse). For
GAPDH mRNA, they were 5′-AAG GTC GGA GTC AAC GGA TTT
-3′ (forward) and 5′-ACTG TGG TCA TGA GTC CTT CCA-3′
(reverse). Each 25 μL of total reaction volume contained 1 μL of
cDNA, 10 pmol of primers, 1 × iQ SYBR Green Supermix (Bio-Rad
Laboratories, Hercules, CA). RT-PCR reactions were performed for a
total of 40 cycles (95 °C for 10 s, 68 °C for 30 s, and 72 °C for 60 s)
using an iCycler IQ (Bio-Rad Laboratories, Hercules, CA). The
relative proportions of each template amplified were determined based
on the threshold cycle (Tc) value for each PCR reaction. The ΔΔCt
method was used to compare gene expression data. RT-PCR was
usually performed in duplicate using two cDNA preparations for each
independent experiment. Changes in mRNA expression level were
calculated following normalization with the GAPDH calibrator gene
and expressed as fold change over untreated samples. Mean
SD
values were determined for each fold difference for at least three
independent experiments.
IL-8 Release. IL-8 in tissue culture supernatants released from the
cells under analysis was measured by Bio-Plex cytokine assay (Bio-Rad
Laboratories, Hercules, CA)56 as described by the manufacturer. The
premixed multiplex beads of the IL-8 human cytokine were used. An
amount of 50 μL of IL-8 standards or samples (supernatants recovered
from treated cells) was incubated with 50 μL of anti-IL-8 conjugated
beads in 96-well filter plates for 30 min at room temperature with
shaking. Plates were then washed by vacuum filtration three times with
100 μL of Bio-Plex wash buffer. An amount of 25 μL of diluted
detection antibodies was added, and plates were incubated for 30 min
at room temperature with shaking. After three filter washes, 50 μL of
streptavidin−phycoerythrin was added, and the plates were incubated
for 10 min at room temperature with shaking. Finally, plates were
washed by vacuum filtration three times, beads were suspended in Bio-
Plex assay buffer, and samples were analyzed on a Bio-Rad 96-well
plate reader using the Bio-Plex suspension array system and Bio-Plex
Manager software (Bio-Rad Laboratories, Hercules, CA).
BIA Technology. Interactions of selected compounds and NF-κB
were studied by surface plasmon resonance (SPR) analysis on BIAcore
X-100 (Biacore, GE Healthcare).53,54 The sensor chip CM5 (research
grade), with a carboxymethylated dextran matrix, was used. CM5
sensor chips containing NF-κB were obtained by the amine coupling
procedure: Briefly, after activation of carboxyl groups on the sensor
chip surface by a 70 μL pulse of a solution, 0.2 M 1-ethyl-3-(3-
dimethylaminopropyl)carbodiimide (EDC), 0.05 M N-hydroxysucci-
nimide (NHS) (10 μL/min flow rate), 50 μL of NF-κB (12.8 μg/mL)
were injected with a flow rate of 5 μL/min. A 70 μL pulse of 1 M
ethanolamine (10 μL/min flow rate) was finally performed to
deactivate excess reactive groups. Binding to selected compounds (1
mM) or double stranded DNA carrying NF-κB binding sequence (28
μg/mL) was performed in HBS buffer (10 mM Hepes, pH 7.4, 150
mM NaCl, 3.4 mM EDTA, and 0.05% surfactant P2). All procedures
were performed at 25 °C and at a 5 μL/min flow rate.
Cell Cultures. IB3-1 cells (LGC Promochem), derived from a CF
patient with a ΔF508/W1282X mutant genotype and immortalized
with adeno12/SV40,55 were grown in LHC-8 basal medium
(Invitrogen, Carlsbad, CA), supplemented with 5% FBS in the
absence of gentamycin, at 37 °C/5% CO2. For the proliferation assay,
IB3-1 cells were seeded at a density of 60 000 cells in 24-well plates in
LHC-8 medium in the presence of 5% FBS and drug was added at
serial dilutions and incubated for a further 3 days. After this period
cells were detached with trypsin/EDTA and resuspended in LHC-8
medium and 5% FBS. The cell number per milliliter was determined
by a ZF Coulter counter (Coulter Electronics, Hialeah, FL, U.S.) to
ASSOCIATED CONTENT
* Supporting Information
■
S
Results of LD measurements; NOESY spectra of compound
29; results of the compounds superimposition and pseudo-
coefficients for the steric and electrostatic fields of the selected
model; superimposition of compounds 29 and 1, 29 and 32, 29
and 5; actual and predicted EC50 values, standardized residuals,
and leverage values of psoralen derivatives; detailed data of all
the synthesized compounds; list of commands used to compute
the 3D-QSAR models. This material is available free of charge
via the Internet at http://pubs.acs.org.
AUTHOR INFORMATION
■
Corresponding Author
*Phone: 00390498275349. E-mail: adriana.chilin@unipd.it.
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
The project was supported by Telethon (Grant GGP10124),
CIB (Consorzio Interuniversitario di Biotecnologie), and FFC
(Italian Cystic Fibrosis Research Foundation, Contract FFC#2/
2010). The BIAcore X-100 was obtained with “Fondo Grandi
1840
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
(14) Nicolis, E.; Lampronti, I.; Dechecchi, M. C.; Borgatti, M.;
Tamanini, A.; Bianchi, N.; Bezzerri, V.; Mancini, I.; Giri, M. G.;
Rizzotti, P.; Gambari, R.; Cabrini, G. Pyrogallol, an active compound
from the medicinal plant Emblica officinalis, regulates expression of
pro-inflammatory genes in bronchial epithelial cells. Int. Immunophar-
macol. 2008, 8, 1672−1680.
(15) Piccagli, L.; Fabbri, E.; Borgatti, M.; Bezzerri, V.; Mancini, I.;
Nicolis, E.; Dechecchi, M. C.; Lampronti, I.; Cabrini, G.; Gambari, R.
Docking of molecules identified in bioactive medicinal plants extracts
into the p50 NF-kappaB transcription factor: correlation with
inhibition of NF-kappaB/DNA interactions and inhibitory effects on
IL-8 gene expression. BMC Struct. Biol. 2008, 8, 38.
(16) Nicolis, E.; Lampronti, I.; Dechecchi, M. C.; Borgatti, M.;
Tamanini, A.; Bezzerri, V.; Bianchi, N.; Mazzon, M.; Mancini, I.; Giri,
M. G.; Rizzotti, P.; Gambari, R.; Cabrini, G. Modulation of expression
of IL-8 gene in bronchial epithelial cells by 5-methoxypsoralen. Int.
Immunopharmacol. 2009, 9, 1411−1422.
(17) Piccagli, L.; Fabbri, E.; Borgatti, M.; Bianchi, N.; Bezzerri, V.;
Mancini, I.; Nicolis, E.; Dechecchi, C. M.; Lampronti, I.; Cabrini, G.;
Gambari, R. Virtual screening against p50 NF-kappaB transcription
factor for the identification of inhibitors of the NF-kappaB-DNA
interaction and expression of NF-kappaB upregulated genes.
ChemMedChem 2009, 4, 2024−2033.
(18) Piccagli, L.; Borgatti, M.; Nicolis, E.; Bianchi, N.; Mancini, I.;
Lampronti, I.; Vevaldi, D.; Dall’Acqua, F.; Cabrini, G.; Gambari, R.
Virtual screening against nuclear factor kappaB (NF-kappaB) of a
focus library: identification of bioactive furocoumarin derivatives
inhibiting NF-kappaB dependent biological functions involved in
cystic fibrosis. Bioorg. Med. Chem. 2010, 18, 8341−8349.
(19) Borgatti, M.; Chilin, A.; Piccagli, L.; Lampronti, I.; Bianchi, N.;
Mancini, I.; Marzaro, G.; dall’Acqua, F.; Guiotto, A.; Gambari, R.
Development of a novel furocoumarin derivative inhibiting NF-kappaB
dependent biological functions: design, synthesis and biological effects.
Eur. J. Med. Chem. 2011, 46, 4870−4877.
(20) Ghosh, G.; van Duyne, G.; Ghosh, S.; Sigler, P. B. Structure of
NF-kappa B p50 homodimer bound to a kappa B site. Nature 1995,
373, 303−310.
(21) Iwai, H.; Zuger, S. Protein ligation: applications in NMR studies
of proteins. Biotechnol. Genet. Eng. Rev. 2007, 24, 129−145.
(22) Hetenyi, C.; van der Spoel, D. Efficient docking of peptides to
proteins without prior knowledge of the binding site. Protein Sci. 2002,
11, 1729−1737.
(23) Hetenyi, C.; van der Spoel, D. Blind docking of drug-sized
compounds to proteins with up to a thousand residues. FEBS Lett.
2006, 580, 1447−1450.
(24) Verma, J.; Khedkar, V. M.; Coutinho, E. C. 3D-QSAR in drug
designa review. Curr. Top. Med. Chem. 2010, 10, 95−115.
(25) Talele, T. T.; Kulkarni, V. M. Three-dimensional quantitative
structure−activity relationship (QSAR) and receptor mapping of
cytochrome P-450(14 alpha DM) inhibiting azole antifungal agents. J.
Chem. Inf. Comput. Sci. 1999, 39, 204−210.
(26) Walter, D. E.; Muhammad, T. D. Genetically Evolved Receptor
Models (GERM): A Procedure for Construction of Atomic-Level
Receptor Site Models in the Absence of a Receptor Crystal Structure.
In Genetic Algorithms in Molecular Modeling; Devillers, J., Ed.;
Academic Press: London, 1996; pp 193−210.
(27) Veselovskaya, M. V.; Shilin, S. V.; Garazd, M. M.; Khilya, V. P.
Modified coumarins. 9. Synthesis of amino-acid derivatives of 3-(2,3,5-
trimethyl-7-oxofuro[3,2-g]chromen-6-yl)propanoic acid. Chem. Nat.
Compd. 2003, 39, 177−181.
(28) Nagorichna, I. V.; Dubovik, I. P.; Garazd, M. M.; Khilya, V. P.
Modified coumarins. 10. Synthesis of substituted 2-(7-oxofuro[3,2-
g]chromen-6-yl)acetic acids. Chem. Nat. Compd. 2003, 39, 253−261.
(29) Sharma, R. K.; Chopra, S.; Sharma, S. D.; Pande, V.; Ramos, M.
J.; Meguro, K.; Inoue, J.; Otsuka, M. Biological evaluation, chelation,
and molecular modeling studies of novel metal-chelating inhibitors of
NF-kappaB-DNA binding: structure activity relationships. J. Med.
Chem. 2006, 49, 3595−3601.
Attrezzature of Ferrara University”. M.B. is supported by Chiesi
Foundation-2011 and FFC (Italian Cystic Fibrosis Research
Foundation, Grant FFC#1/2012).
ABBREVIATIONS USED
■
AD, AutoDock; ADT, AutoDockTools; CF, cystic fibrosis; BIA,
biospecific interaction analysis; DMSO, dimethylsulfoxide;
DTT, dithiothreitol; EDTA, ethylediaminetetraacetic acid;
EMSA, electrophoretic mobility shift assay; ESI-TOF, electro-
spray ionization time of flight; FBS, fetal bovine serum; HRMS,
high resolution mass spectrometry; IL-8, interleukine 8; MAOS,
microwave assisted organic synthesis; NMR, nuclear magnetic
resonace; NOESY, nuclear Overhauser effect spectroscopy;
PDB, Protein Data Bank; QSAR, quantitative structure−activity
relationship; RT-PCR, real time polymerase chain reaction;
SDEC, standard deviation error of calculations; SDEP, standard
deviation error of predictions; SPR, surface plasmon resonance;
TBE, Tris/borate/ethylediaminetetraacetic acid; TEA, triethyl-
amine; TLC, thin layer chromatography; TMA, 4,6,4′-
trimethylangelicin; TNF, tumor necrosis factor
REFERENCES
■
(1) Hayden, M. S.; Ghosh, S. NF-kappaB, the first quarter-century:
remarkable progress and outstanding questions. Genes Dev. 2012, 26,
203−234.
(2) Brown, K. D.; Claudio, E.; Siebenlist, U. The roles of the classical
and alternative nuclear factor-kappaB pathways: potential implications
for autoimmunity and rheumatoid arthritis. Arthritis Res. Ther. 2008,
10, 212.
(3) Siebenlist, U.; Franzoso, G.; Brown, K. Structure, regulation and
function of NF-kappa B. Annu. Rev. Cell Biol. 1994, 10, 405−455.
(4) Perkins, N. D. The diverse and complex roles of NF-kappaB
subunits in cancer. Nat. Rev. Cancer 2012, 12, 121−132.
(5) Victoriano, A. F.; Okamoto, T. Transcriptional control of HIV
replication by multiple modulators and their implication for a novel
antiviral therapy. AIDS Res. Hum. Retroviruses 2012, 28, 125−138.
(6) Batra, S.; Balamayooran, G.; Sahoo, M. K. Nuclear factor-kappaB:
a key regulator in health and disease of lungs. Arch. Immunol. Ther.
Exp. 2011, 59, 335−351.
(7) Ben-Neriah, Y.; Karin, M. Inflammation meets cancer, with NF-
kappaB as the matchmaker. Nat. Immunol. 2011, 12, 715−723.
(8) Black, H. R.; Yankaskas, J. R.; Johnson, L. G.; Noah, T. L.
Interleukin-8 production by cystic fibrosis nasal epithelial cells after
tumor necrosis factor-alpha and respiratory syncytial virus stimulation.
Am. J. Respir. Cell Mol. Biol. 1998, 19, 210−215.
(9) Jones, A. M.; Martin, L.; Bright-Thomas, R. J.; Dodd, M. E.;
McDowell, A.; Moffitt, K. L.; Elborn, J. S.; Webb, A. K. Inflammatory
markers in cystic fibrosis patients with transmissible Pseudomonas
aeruginosa. Eur. Respir. J. 2003, 22, 503−506.
(10) Bezzerri, V.; Borgatti, M.; Nicolis, E.; Lampronti, I.; Dechecchi,
M. C.; Mancini, I.; Rizzotti, P.; Gambari, R.; Cabrini, G. Transcription
factor oligodeoxynucleotides to NF-kappaB inhibit transcription of IL-
8 in bronchial cells. Am. J. Respir. Cell Mol. Biol. 2008, 39, 86−96.
(11) Sadikot, R. T.; Zeng, H.; Joo, M.; Everhart, M. B.; Sherrill, T. P.;
Li, B.; Cheng, D. S.; Yull, F. E.; Christman, J. W.; Blackwell, T. S.
Targeted immunomodulation of the NF-kappaB pathway in airway
epithelium impacts host defense against Pseudomonas aeruginosa. J.
Immunol. 2006, 176, 4923−4930.
(12) Bezzerri, V.; Borgatti, M.; Finotti, A.; Tamanini, A.; Gambari, R.;
Cabrini, G. Mapping the transcriptional machinery of the IL-8 gene in
human bronchial epithelial cells. J. Immunol. 2011, 187, 6069−6081.
(13) Tamanini, A.; Borgatti, M.; Finotti, A.; Piccagli, L.; Bezzerri, V.;
Favia, M.; Guerra, L.; Lampronti, I.; Bianchi, N.; Dall’Acqua, F.;
Vedaldi, D.; Salvador, A.; Fabbri, E.; Mancini, I.; Nicolis, E.; Casavola,
V.; Cabrini, G.; Gambari, R. Trimethylangelicin reduces IL-8
transcription and potentiates CFTR function. Am. J. Physiol.: Lung
Cell. Mol. Physiol. 2011, 300, L380−L390.
1841
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842

Journal of Medicinal Chemistry
Article
(30) Hearst, J. E. Psoralen photochemistry and nucleic acid structure.
J. Invest. Dermatol. 1981, 77, 39−44.
Selection of the best predictive PLS model. J. Chemom. 1992, 6, 347−
356.
(52) Borgatti, M.; Bezzerri, V.; Mancini, I.; Nicolis, E.; Dechecchi, M.
C.; Lampronti, I.; Rizzotti, P.; Cabrini, G.; Gambari, R. Induction of
IL-6 gene expression in a CF bronchial epithelial cell line by
Pseudomonas aeruginosa is dependent on transcription factors
belonging to the Sp1 superfamily. Biochem. Biophys. Res. Commun.
2007, 357, 977−983.
(53) Gambari, R.; Feriotto, G.; Rutigliano, C.; Bianchi, N.; Mischiati,
C. Biospecific interaction analysis (BIA) of low-molecular weight
DNA-binding drugs. J. Pharmacol. Exp. Ther. 2000, 294, 370−377.
(54) Feriotto, G.; Corradini, R.; Sforza, S.; Bianchi, N.; Mischiati, C.;
Marchelli, R.; Gambari, R. Peptide nucleic acids and biosensor
technology for real-time detection of the cystic fibrosis W1282X
mutation by surface plasmon resonance. Lab. Invest. 2001, 81, 1415−
1427.
(55) Dechecchi, M. C.; Nicolis, E.; Bezzerri, V.; Vella, A.; Colombatti,
M.; Assael, B. M.; Mettey, Y.; Borgatti, M.; Mancini, I.; Gambari, R.;
Becq, F.; Cabrini, G. MPB-07 reduces the inflammatory response to
Pseudomonas aeruginosa in cystic fibrosis bronchial cells. Am. J. Respir.
Cell Mol. Biol. 2007, 36, 615−624.
(56) Borgatti, M.; Mazzitelli, S.; Breveglieri, G.; Gambari, R.;
Nastruzzi, C. Induction by TNF-alpha of IL-6 and IL-8 in cystic
fibrosis bronchial IB3-1 epithelial cells encapsulated in alginate
microbeads. J. Biomed. Biotechnol. 2010, 2010.
(31) Nurminen, E. M.; Pihlavisto, M.; Lazar, L.; Szakonyi, Z.;
Pentikainen, U.; Fulop, F.; Pentikainen, O. T. Synthesis, in vitro
activity, and three-dimensional quantitative structure−activity relation-
ship of novel hydrazine inhibitors of human vascular adhesion protein-
1. J. Med. Chem. 2010, 53, 6301−6315.
(32) Discovery Studio Modeling Environment, release 3.1; Accelrys
Software Inc.: San Diego, CA, 2005−2011.
(33) Tosco, P.; Balle, T. Open3DQSAR: a new open-source software
aimed at high-throughput chemometric analysis of molecular
interaction fields. J. Mol. Model. 2011, 17, 201−208.
(34) Lipkowitz, K. B.; Boyd, D. B. Reviews in Computational
Chemistry; John Wiley & Sons, Hoboken, NJ, 1997.
(35) Gramatica, P. Principles of QSAR models validation: internal
and external. QSAR Comb. Sci. 2007, 26, 794−801.
(36) Borgatti, M.; Mancini, I.; Bianchi, N.; Guerrini, A.; Lampronti,
I.; Rossi, D.; Sacchetti, G.; Gambari, R. Bergamot (Citrus bergamia
Risso) fruit extracts and identified components alter expression of
interleukin 8 gene in cystic fibrosis bronchial epithelial cell lines. BMC
Biochem. 2011, 12, 15.
(37) Boyce, B. F.; Yao, Z.; Xing, L. Functions of nuclear factor
kappaB in bone. Ann. N.Y. Acad. Sci. 2010, 1192, 367−375.
(38) Piva, R.; Penolazzi, L.; Borgatti, M.; Lampronti, I.; Lambertini,
E.; Torreggiani, E.; Gambari, R. Apoptosis of human primary
osteoclasts treated with molecules targeting nuclear factor-kappaB.
Ann. N.Y. Acad. Sci. 2009, 1171, 448−456.
(39) Atkinson, G. P.; Nozell, S. E.; Benveniste, E. T. NF-kappaB and
STAT3 signaling in glioma: targets for future therapies. Expert Rev.
Neurother. 2010, 10, 575−586.
(40) Wong, K. K.; Jacks, T.; Dranoff, G. NF-kappaB fans the flames
of lung carcinogenesis. Cancer Prev. Res. 2010, 3, 403−405.
(41) Saadane, A.; Masters, S.; DiDonato, J.; Li, J.; Berger, M.
Parthenolide inhibits IkappaB kinase, NF-kappaB activation, and
inflammatory response in cystic fibrosis cells and mice. Am. J. Respir.
Cell Mol. Biol. 2007, 36, 728−736.
(42) Cigana, C.; Nicolis, E.; Pasetto, M.; Assael, B. M.; Melotti, P.
Anti-inflammatory effects of azithromycin in cystic fibrosis airway
epithelial cells. Biochem. Biophys. Res. Commun. 2006, 350, 977−982.
(43) Garazd, M. M.; Garazd, Ia, L.; Ogorodniichuk, A. S.; Khilia, V. P.
Synthesis of modified psoralen analogues. Bioorg. Khim. 2004, 30,
324−333.
(44) Pettersen, E. F.; Goddard, T. D.; Huang, C. C.; Couch, G. S.;
Greenblatt, D. M.; Meng, E. C.; Ferrin, T. E. UCSF Chimeraa
visualization system for exploratory research and analysis. J. Comput.
Chem. 2004, 25, 1605−1612.
(45) Marvin, version 5.5.0.1, Program B; ChemAxon: Budapest,
Hungary; www.chemaxon.com/products.
(46) O’Boyle, N. M.; Banck, M.; James, C. A.; Morley, C.;
Vandermeersch, T.; Hutchison, G. R. Open Babel: an open chemical
toolbox. J. Cheminf. 2011, 3, 33.
(47) Morris, G. M.; Huey, R.; Lindstrom, W.; Sanner, M. F.; Belew,
R. K.; Goodsell, D. S.; Olson, A. J. AutoDock4 and AutoDockTools4:
automated docking with selective receptor flexibility. J. Comput. Chem.
2009, 30, 2785−2791.
(48) Granovsky, A. A. Firefly, version 7.1.G; http://classic.chem.msu.
su/gran/firefly/index.html.
(49) Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.; Elbert, S. T.;
Gordon, M. S.; Jensen, J. H.; Koseki, S.; Matsunaga, N.; Nguyen, K. A.;
Su, S.; Windus, T. L.; Dupuis, M.; Montgomery, J. A. General atomic
and molecular electronic structure system. J. Comput. Chem. 1993, 14,
1347−1363.
(50) Pastor, M.; Cruciani, G.; Clementi, S. Smart region definition: a
new way to improve the predictive ability and interpretability of three-
dimensional quantitative structure−activity relationships. J. Med. Chem.
1997, 40, 1455−1464.
(51) Baroni, M.; Clementi, S.; Cruciani, G.; Costantino, G.; Riganelli,
D.; Oberrauch, E. Predictive ability of regression models. Part II:
1842
dx.doi.org/10.1021/jm3009647 | J. Med. Chem. 2013, 56, 1830−1842