Computational structure-activity study directs synthesis of novel antitumor enkephalin analogs

Article abstract of DOI:10.1007/s00726-009-0329-5

The capability of a Support Vector Machines QSAR model to predict the antiproliferative ability of small peptides was evaluated by screening a virtual library of enkephalin-like analogs modified by incorporation of the (R,S)-(1-adamantyl)glycine (Aaa) residue. From an initial set of 390 compounds, the peptides, Tyr-Aaa-Gly-Phe-Met (2), Tyr-Aaa-Gly-Phe-Phe (3), Phe-Aaa-Gly-Phe-Phe (4) and Phe-Aaa-Gly-Phe-Met (5) were selected, synthesized and their antitumor activity was tested and compared to that of Met-enkephalin (1). The antiproliferative activity correlated with the computational prediction and with the foldamer-forming ability of the studied peptides. The most active compounds were the hydrophobic peptides, Phe-Aaa-Gly-Phe-Phe (4) and Phe-Aaa-Gly-Phe-Met (5), having a greater propensity to adopt folded structures than the other peptides.

Full text of DOI:10.1007/s00726-009-0329-5

Amino Acids (2010) 38:1185–1191
DOI 10.1007/s00726-009-0329-5
ORIGINAL ARTICLE
Computational structure–activity study directs synthesis
of novel antitumor enkephalin analogs
ˇ
M. Gredicak Æ F. Supek Æ M. Kralj Æ Z. Majer Æ
ˇ
´
´
M. Hollosi Æ T. Smuc Æ K. Mlinaric-Majerski Æ
ˇ
S. Horvat
Received: 20 May 2009 / Accepted: 14 July 2009 / Published online: 29 July 2009
Ó Springer-Verlag 2009
Abstract The capability of a Support Vector Machines
QSAR model to predict the antiproliferative ability of
small peptides was evaluated by screening a virtual library
of enkephalin-like analogs modified by incorporation of the
(R,S)-(1-adamantyl)glycine (Aaa) residue. From an initial
set of 390 compounds, the peptides, Tyr-Aaa-Gly-Phe-Met
(2), Tyr-Aaa-Gly-Phe-Phe (3), Phe-Aaa-Gly-Phe-Phe (4)
and Phe-Aaa-Gly-Phe-Met (5) were selected, synthesized
and their antitumor activity was tested and compared to
that of Met-enkephalin (1). The antiproliferative activity
correlated with the computational prediction and with the
foldamer-forming ability of the studied peptides. The most
active compounds were the hydrophobic peptides, Phe-
Aaa-Gly-Phe-Phe (4) and Phe-Aaa-Gly-Phe-Met (5), hav-
ing a greater propensity to adopt folded structures than the
other peptides.
Keywords Enkephalin analogs Á (1-Adamantyl)glycine Á
QSAR study Á CD spectroscopy Á In vitro
antiproliferative activity Á Peptide synthesis
Introduction
The native opioid growth factor (OGF), Met-enkephalin
(Tyr-Gly-Gly-Phe-Met, 1), not only plays a role in cell
proliferation and tissue organization during development,
cellular renewal, wound healing, and angiogenesis, but also
in cancer (Zagon et al. 2002; Smith et al. 2004). To obtain
peptides with improved or novel activity profiles, modifi-
cations using low molecular weight lipophilic adamantane
building blocks are reported (Miyazaki et al. 2008; Karle
´
et al. 2008; Basaric et al. 2007; Wanka et al. 2007). The
adamantyl groups provide the desired membrane perme-
ability and conformational constraint for efficient transport
in lipid membranes. The susceptibility of the modified pep-
tides to enzymatic degradation may also decrease. Recently,
we designed and synthesized new analogs of Met-enkephalin
(1) by substituting either Gly2 or Gly2-Gly3 in 1 and/or in its
shorter N-terminal fragments by unnatural adamantane-
derived amino acids (Horvat et al. 2006). Exclusively, the
peptide analogs containing Caa-dialkylated glycine or
Ca-alkylated glycine residues showed in vitro antitumor
activity in the cell lines HEp-2, HBL, SW-620 and Caco-2. In
particular, the pentapeptide Tyr-(R,S)-Aaa-Gly-Phe-Met (2)
[Aaa = (R,S)-(1-adamantyl)glycine] (Horvat et al. 2006)
(Fig. 1) showed the most potent tumoricidal activity by
inducing apoptosis in the cell lines HEp-2 and SW-620.
Since highly enriched diastereomers [(S)- and (R)-Aaa]-2
showed lower antiproliferative effect than the racemic one,
the obtained data suggested synergism of [(S)- and (R)-Aaa]-
2 when investigated together as a racemate.
Electronic supplementary material The online version of this
article (doi:10.1007/s00726-009-0329-5) contains supplementary
material, which is available to authorized users.
ˇ
ˇ
´
M. Gredicak Á K. Mlinaric-Majerski Á S. Horvat (&)
Division of Organic Chemistry and Biochemistry,
ˇ
´
ˇ
Ruder Boskovic Institute, Bijenicka c. 54, POB 180,
¯
Zagreb 10002, Croatia
e-mail: shorvat@irb.hr
ˇ
F. Supek Á T. Smuc
ˇ
´
Division of Electronics, Ruder Boskovic Institute,
¯
Zagreb 10002, Croatia
M. Kralj
Division of Molecular Medicine, Ruder Boskovic Institute,
¯
ˇ
´
Zagreb 10002, Croatia
´
Z. Majer Á M. Hollosi
Department of Organic Chemistry, Institute of Chemistry,
´
¨
Eotvos Lorand University, 1518 Budapest, Hungary
¨
123

ˇ
M. Gredicak et al.
1186
In this paper, we have undertaken a computational
QSAR study using human tumor cell line cytostatic activity
data from 22 previously synthesized and published peptide
compounds structurally related to Met-enkephalin con-
taining unnatural amino acids (Horvat et al. 2006). The
Support Vector Machines (SVM) algorithm was employed
to derive a model for regression of the compound’s cyto-
static activity against a set of 1,330 molecular descriptors.
We used this model to screen a virtual library of 390
enkephalin-like tri-, tetra-, and pentapeptides modified by
incorporation of the unnatural amino acid residue, (R,S)-(1-
adamantyl)glycine (Aaa). Three highly scoring novel
compounds were selected for synthesis and testing of
cytostatic activity on three tumor cell lines.
OH
S
O
O
H
H
N
N
OH
H2N
N
N
H
H
O
O
O
Tyr-Gly-Gly-Phe-Met (1)
OH
S
O
O
H
N
H
N
OH
H2N
N
N
H
H
O
O
O
Methods and materials
Tyr-Aaa-Gly-Phe-Met (2)
Construction of dataset
OH
Human tumor cell line cytostatic activity data from 22
previously synthesized and published compounds (Horvat
et al. 2006) related to Met-enkephalin derivatives were
collected. This included measurements of growth inhibition
at concentrations of 10-6 to 10-3 M on the HEp-2, HeLa,
HBL, HT-29, SW-620, Caco-2, and MCF-7 cell lines (28
measurements in total, per compound). To obtain an
overall estimate of the cytostatic activity of each com-
pound, we used principal component analysis, as imple-
mented in the Weka 3.5.6 package (Witten and Frank
2005). The first principal component (PC1) captured the
main source of variation in the cytostatic activity mea-
surements, and using PC1 alone described 44.7% of the
variance in the original 28 measurements. PC1 was com-
puted as a sum of weighted standardized cytostatic activity
measurements at different concentrations, making it a
dimensionless quantity. Here, the HBL and MCF-7 cell
lines contributed less to PC1 (data not shown), which was
probably not due to a genuine difference in biological
response of these two cell lines, but was a consequence of
missing data; 8 out of 22 compounds did not have cyto-
static activity measurements on HBL and MCF-7. The full
dataset, including the compounds’ structures, individual
cytostatic activity measurements and PC1, and molecular
descriptors (see below) is available from the authors’ web
site at http://anticancer.irb.hr and in the Supplementary
material (only structures and activities).
O
O
H
N
H
N
OH
H2N
N
N
H
H
O
O
O
Tyr-Aaa-Gly-Phe-Phe (3)
O
O
H
H
N
N
OH
H2N
N
N
H
H
O
O
O
Phe-Aaa-Gly-Phe-Phe (4)
S
O
O
H
N
H
N
OH
H2N
N
N
H
H
O
O
O
Phe-Aaa-Gly-Phe-Met (5)
Fig. 1 Structures of pentapeptides 1–5
Modeling cytostatic activity from structure
It is of relevance to understand how modifications in the
Met-enkephalin peptide sequence may lead to better per-
forming drugs and such understanding is possible through
mathematical approaches such as QSAR.
The 22 enkephalin-like compounds were submitted to the
E-Dragon web service (Tetko et al. 2005), that predicts a
probable three-dimensional structure of the molecules
123

Computational structure–activity study directs synthesis of novel antitumor enkephalin analogs
1187
using CORINA (Sadowski et al. 1994) and then computes
1,666 molecular descriptors per compound. CORINA
employs a rule-based system shown to be quite accurate in
prediction of 3D structures (Sadowski et al. 1994). After
removal of descriptors invariant within our data, 1,330
descriptors remained. The SVM algorithm (Burges 1998)
for regression (e-SVR variety) was applied to this data, as
implemented in the LibSVM software (Chang and Lin
2001) adapted for the Weka 3.5.6 data mining environment
(El-Manzalawy and Honavar 2005). As per recommenda-
tion by the LibSVM authors (Hsu et al. 2008), we utilized a
radial basis function kernel, and optimized the SVM’s
metaparameters ‘‘c’’ and ‘‘gamma’’ in an exhaustive search
for a combination yielding the best root-mean-square error
(RMSE) estimated on crossvalidation. Gamma was varied
exponentially from 2-15 to 23 in steps of 21, and c was
varied from 2-5 to 215 in steps of 21. Five runs of fourfold
crossvalidation were used to estimate the RMSE. The
optimal values of the parameters were found to be
gamma = 2-8 and c = 26. The final estimate of the trained
model’s performance on unseen data was performed using
leave-one-out crossvalidation and by computing the cor-
relation coefficient between the predicted and actual val-
ues, which was q = 0.734. This indicates that the model
has reasonable utility for screening a virtual library of
compounds.
optimization of the SVM metaparameters, we obtain a leave-
one-out crossvalidated Pearson’s correlation coefficient of
q = 0.741, comparable to q = 0.734 with CORINA. The
predictions for the virtual library were generally similar to
the original ones with CORINA, Pearson’s r = 0.532 (Fig. 1
in Supporting material). The differences that do exist are
likely attributable to the high conformational flexibility of
peptides which may prove challenging to the 3D structure
prediction approaches such as CORINA and FROG that
were primarily designed for smaller drug-like organic
molecules.
CD measurements
Circular dichroism (CD) spectra were obtained on a Jasco
J-810 spectropolarimeter in the 280–185 nm range in a
0.2 mm path length quartz cell at room temperature. The
sample concentrations ranged between 0.2 and 0.4 mg/ml.
CD spectra were taken in 2,2,2-trifluoroethanol (TFE;
Aldrich, NMR grade) and in a mixture of water and TFE in
a ratio of 1:1. The following parameters were used: spectral
band width 1 nm, response time 0.5 s, data pitch 0.1 nm,
scanning speed 50 nm/min, accumulation 5. The CD
spectral parameters were expressed as molar CD (De).
Proliferation assay
Construction and screening of the virtual library
The colon carcinoma (SW-620), breast carcinoma (MCF-
7), and cervical carcinoma (HeLa) cells were cultured as
monolayers and maintained in Dulbecco’s modified
Eagle’s medium (DMEM), supplemented with 10% fetal
bovine serum (FBS), 2 mM L-glutamine, 100 U/ml peni-
cillin, and 100 lg/ml streptomycin in a humidified atmo-
sphere with 5% CO2 at 37°C. The growth inhibition
activity was assessed as described previously, according to
the slightly modified procedure of the National Cancer
Institute, Developmental Therapeutics Program (Horvat
et al. 2009). Briefly, the cells were inoculated onto standard
96-well microtiter plates on day 0. Test agents were then
added in five consecutive tenfold dilutions (10-8 to
10-4 mol/l) and incubated for a further 72 h. Working
dilutions were freshly prepared on the day of testing. After
72 h of incubation, the cell growth rate was evaluated by
performing the MTT assay (Horvat et al. 2009) that detects
dehydrogenase activity in viable cells. The absorbance
(OD, optical density) was measured on a microplate reader
at 570 nm. Each test point was performed in quadruplicate
in three individual experiments. The results are expressed
as IC50, which is the concentration necessary for 50% of
inhibition. The IC50 values for each compound were cal-
culated from dose–response curves using linear regression
analysis by fitting the test concentrations that give PG
(percentage of growth) values above and below the
A virtual library of 390 compounds was constructed,
encompassing enkephalin-like tri-, tetra-, and pentapeptides
modified with the (R,S)-(1-adamantyl)glycine (Aaa) residue.
Our previous results clearly showed that all of the peptides
structurally related to Met-enkephalin containing the Aaa
residue exhibited increased cytostatic activity in vitro
(Horvat et al. 2006). The E-Dragon web service (Tetko et al.
2005) was used to compute the 1,330 relevant descriptors,
and applied the trained SVM model to the virtual library to
screen for novel potentially active compounds. We used
Marvin 5.1.1 (ChemAxon, http://www.chemaxon.com) to
visualize, browse, and otherwise handle the molecules in the
virtual library. The structures of the peptides in the virtual
library, their molecular descriptors, and the predicted cyto-
static activities are available from the authors’ web site at
http://anticancer.irb.hr and in the Supplementary material
(only structures and activities). To verify that our activity
predictions are not biased by choice of a specific algorithm
for generation of the peptides’ 3D conformations (CORI-
NA), we have re-trained the SVM model and screened the
virtual library with the 3D structures generated by the FROG
server (Bohme Leite et al. 2007) that couples a Monte Carlo
sampling procedure of the conformational space to a MMFF
force field for scoring the energy of the conformations. After
123

ˇ
M. Gredicak et al.
1188
reference value (i.e., 50%). If, however, for a given cell
line all of the tested concentrations produced PGs
exceeding the respective reference level of effect (e.g., PG
value of 50), then the highest tested concentration was
assigned as the default measurement value. In the screening
data report, the default values are denoted by a ‘‘[’’ sign
preceding the number (for details see Supplementary
material). Alternatively, if the PGs only slightly exceeded
the reference level of effect, the result was extrapolated.
Each result is a mean value from three separate
experiments.
as assessed by analytical RP HPLC. The spectroscopic data
of compounds 2–5 are provided in the Supplementary
material.
Results and discussion
We gathered human tumor cell line cytostatic activity data
from 22 previously synthesized and published compounds
(Horvat et al. 2006) related to Met-enkephalin derivatives,
and used the data to train a regression model that would
predict antitumor activity from a large number of molec-
ular descriptors.
Synthesis and purification of peptides
The method employed for this QSAR effort was the
SVM regression. SVMs have desirable properties of
robustness to noise and high generalization power in dif-
ficult situations with few training examples, and have
consequently become widely popular in other sciences,
such as computational biology (Noble 2004). Examples of
previous applications of SVMs in QSAR studies include a
classification problem involving dihydrofolate reductase
inhibition by pyrimidines (Burbidge et al. 2001) and large-
scale virtual screening of molecular databases for inhibitors
of multiple enzymes (Jorissen and Gilson 2005); for a
review, see Ivanciuc (2007).
Met-enkephalin (1) was purchased from Bachem AG
(Bubendorf, Switzerland). (R,S)-(1-adamantyl)glycine
(Aaa), Boc-Tyr(Boc)-Aaa-OH, and Boc-Phe-Aaa-OH were
prepared according to the procedure described in the
literature (Horvat et al. 2006).
The peptides 2–5 were synthesized manually from C- to
N-terminal by the solid-phase Fmoc method on a com-
mercially available preloaded Fmoc-Met or Fmoc-Phe
Wang resin (Bachem, p-alkoxybenzyl alcohol resin, 200–
400 mesh, 0.8 mmol/g loading) on a 0.05 mmol scale. The
consecutive steps in the solid-phase peptide synthesis
performed in each cycle were: (a) deprotection of the Fmoc
group by two treatments (1 and 30 min) with 20% piperi-
dine in DMF (v/v), (b) coupling by applying HBTU/HOBt/
DIPEA activation and a threefold excess of the appropriate
Fmoc-amino acid for 30 min or, in the last step, by
applying the dipeptide building blocks Boc-Tyr(Boc)-Aaa-
OH or Boc-Phe-Aaa-OH (prepared by solution methods)
for 3 h, (c) removal of the peptides from the resin by
treatment with a mixture of TFA:TIS:H2O in the ratio
9.5:0.25:0.25 (v/v/v) for 1 h. Successive deprotection and
coupling steps were monitored by positive and negative
Kaiser (ninhydrin) test, respectively. The peptides were
obtained as a filtrate in TFA and precipitated with cold dry
diisopropyl ether. Analysis and purification of crude pep-
tides were achieved by RP HPLC performed on a Varian
Pro Star 230 HPLC system using a Eurospher 100
reversed-phase C-18 semipreparative (250 9 8 mm ID,
5 lm) (flow rate 1.0 ml/min) or analytical (150 9 4.5 mm
ID, 5 lm) (flow rate 0.5 ml/min) column under isocratic
conditions using 57% MeOH/0.1% TFA for peptides 2 and
3 or 61% MeOH/0.1% TFA for peptides 4 and 5. UV
detection was performed at 254 and 215 nm using a Varian
Pro Star 335 photodiode-array detector. The trifluoroace-
tate ion present after preparative HPLC was removed using
an SPE cartridge. The cartridge was first eluted with water
and then with MeOH to recover the peptide compounds.
The effluent was evaporated and the residue dissolved in
water and lyophilized. Peptides 2–5 were at least 95% pure
The SVM regression model trained on our 22-compound
dataset was estimated to predict activity of unknown
molecules with reasonable accuracy; Pearson’s correlation
coefficient on leave-one-out crossvalidation was q =
0.734. Finally, we applied the regression model to a virtual
library of 390 enkephalin-like tri-, tetra-, and penta-
peptides modified by incorporation of (R,S)-(1-adaman-
tyl)glycine (Aaa) residues. The top ten peptides ranked by
their predicted activity (dimensionless, see ‘‘Methods and
materials’’ for details), and their predicted log P (hydro-
phobicity) and log S (aqueous solubility) values are given
in Table 1.
Peptides 2–5 (listed in Table 1) were synthesized using
a commercial Wang resin already anchored with either the
Met or the Phe residues, by the Fmoc technique using
piperidine deprotection and HBTU/HOBt/DIPEA as cou-
pling reagents. After removal from the resin with the usual
TFA/TIS/water cleavage procedure, the products were
purified by semipreparative RP HPLC. The prepared pep-
tides were very sparingly soluble in water.
All the newly synthesized compounds afforded appro-
priate mass spectrometry data. The NMR spectra were
consistent with the assigned structures and are provided in
the Supplementary material.
The biological activities of peptides 2–5 (as hydro-
chloride salts), and Met-enkephalin (OGF, 1) as a reference
compound, were evaluated by determining their capacity to
inhibit the proliferation of three tumor cell lines. In our
123

Computational structure–activity study directs synthesis of novel antitumor enkephalin analogs
1189
average, indicating somewhat better activity. This can be
seen from the concentration–response graphs, showing that
up to 20% of inhibition was obtained even at 10 lM
concentration of 5. Contrary to this, compound 3 was less
active, pointing to the influence of a change in hydropho-
bicity of the amino acid at position 1. Interestingly, the
SW-620 cell line was the most sensitive cell line, as it was
the case in our previous study (Horvat et al. 2006).
It was already established that replacement of Tyr res-
idue at position 1 with a Phe residue, in an attempt to
increase the hydrophobicity of a series of pentapeptides,
resulted in a significant increase in the Caco-2 cell per-
meability of these peptides (Knipp et al. 1997). Moreover,
the same authors found that other amino acid alterations
that induce a b-turn spatial conformation have an effect on
the peptides’ physico-chemical characteristics, primarily
hydrophobicity, and consequently increase the membrane
permeability of the peptides.
Table 1 Top-rated peptides as obtained by computational cytostatic
activity prediction, chosen from a virtual library of 390 enkephalin-
like peptides containing the (R,S)-(1-adamantyl)glycine (Aaa) residue
Peptide
Predicted activitya
log Pb
log Sb
Tyr-Aaa-Gly-Phe-Met (2)c
Phe-Aaa-Gly-Phe-Met (5)
Phe-Aaa-Gly-Phe-Phe (4)
Tyr-Aaa-Gly-Phe-Phe (3)
Trp-Aaa-Gly-Phe-Met
Tyr-Aaa-Gly-Phe-Gly
Trp-Aaa-Gly-Phe-Phe
Phe-Aaa-Gly
-5.77
-5.37
-5.07
-4.20
-4.18
-3.53
-3.50
-3.28
-3.27
-3.25
0.65
0.91
1.32
1.10
1.14
0.40
1.58
0.29
0.01
-0.12
-5.36
-5.62
-5.83
-5.65
-5.79
-5.02
-5.90
-4.68
-5.23
-4.47
Phe-Gly-Gly-Phe-Phe
Phe-Gly-Aaa-Gly
a
The predicted activity is dimensionless; more negative values sig-
nify stronger cytostatic activity
b
log P (hydrophobicity) and log S (aqueous solubility) values are
given as predicted by the E-Dragon web service (Tetko et al. 2005)
Therefore, to gain further understanding of structure–
activity relationship for peptides 1–5, their CD curves were
measured. The CD spectra of peptides 1–5 in TFE and
TFE-water (1:1) mixtures are presented in Fig. 3 (for CD
spectral parameters see Table 3 in the Supplementary
material). The CD spectra in TFE (Fig. 3a) of Met-
enkephalin (1) and its derivatives containing (R,S)-(1-ad-
amantyl)glycine (Aaa) in position two are determined by
the backbone conformation and the aromatic chromophores
of the amino acids, Tyr and/or Phe. The CD spectrum of
Met-enkephalin (1) in TFE indicated the presence of an
ensemble of conformers rather than one or two prevailing
ones, similarly to the structurally related Leu-enkephalin
(Tyr-Gly-Gly-Phe-Leu) discussed in one of our earlier
papers (Vass et al. 2008). The characteristic features of the
spectra of peptides 1–3 with Tyr in position one were a
positive 1La band between 227 and 223 nm, and a positive
1Bb band below 195 nm. Peptides 4 and 5 did not contain
Tyr, in their spectra the 1La band was blue-shifted to
*218 nm, while the 1Bb band was red-shifted and
appeared with decreased intensity.
c
Horvat et al. (2006)
previous paper (Horvat et al. 2006), we demonstrated that
the pentapeptide, Tyr-Aaa-Gly-Phe-Met (2), showed the
most potent antiproliferative activity, therefore we checked
its activity in three cell lines (SW-620, HeLa, and MCF-7),
along with the activity of three novel analogs of peptide 2,
having different amino acids at position 1 (Phe-Aaa-Gly-
Phe-Met, 5), 1 and 5 (Phe-Aaa-Gly-Phe-Phe, 4), or at
position 5 (Tyr-Aaa-Gly-Phe-Phe, 3) (Table 2; Fig. 2). The
reference compound, OGF (Met-enkephalin, 1) did not
exert any inhibitory activity at all in the tested concentra-
tion range, which is in accordance with the previously
published results (Horvat et al. 2006; Horvat et al. 2009). In
this study, peptide 2 showed slightly more pronounced
activity (probably due to its better solubility) than previ-
ously reported. However, compounds 4 and 5 inhibited the
growth of all cell lines at slightly lower concentrations on
The negative band below 205 nm in the spectra of
peptides 2, 4, and 5 reflects the presence of more ordered
structures and the spectral effect of the interaction between
the aromatic side chains and the steric constrain of (R,S)-
Aaa in position two. However, the type of folded structure
is difficult to predict because of the dominance of the CD
contribution of the aromatic chromophores.
Table 2 In vitro inhibition of the growth of tumor cells treated with
Met-enkephalin (OGF, 1) and Aaa-containing peptides 2-5
Peptide
IC50 (lM)a
SW-620
MCF-7
HeLa
Tyr-Gly-Gly-Phe-Met (1)
Tyr-Aaa-Gly-Phe-Met (2)
Tyr-Aaa-Gly-Phe-Phe (3)
Phe-Aaa-Gly-Phe-Phe (4)
Phe-Aaa-Gly-Phe-Met (5)
a
[100
[100
[100
C100
270 90b
73 25
47 39
60 32
55 33
[100
The CD spectra were also measured in TFE-water 1:1
mixture (Fig. 3b). The spectra of 2, 4, and 5 did not show
the negative band below 205 nm as a sign of decreased
population of folded conformers.
[100
69 33
64 36
59 38
94 69
By monitoring the spectral behavior of the studied
compounds, it can be suggested that 2, 4, 5 have similar
populations of conformers which are more fixed through
IC50, the concentration that causes a 50% reduction of cell
proliferation
b
The result is extrapolated
123

ˇ
M. Gredicak et al.
1190
Fig. 2 Dose–response profiles
for peptides 2–5. PG percentage
of growth
2
3
B
A 150
150
100
50
100
50
HeLa
HeLa
SW 620
MCF-7
SW 620
MCF-7
0
0
-50
-100
-50
-100
-9
-8
-7
-6
-5
-4
-3
-9
-8
-7
-6
-5
-4
-3
log concentration (M)
5
log concentration (M)
4
C
D
150
100
50
150
100
HeLa
HeLa
SW 620
MCF-7
50
0
SW 620
MCF-7
0
-50
-100
-50
-100
-9
-8
-7
-6
-5
-4
-3
-9
-8
-7
-6
-5
-4
-3
log concentration (M)
log concentration (M)
It should be recognized that adamantyl amino acids are
c-turn inducers for peptides (Kuroda et al. 1997). In addi-
tion, it was also demonstrated by theoretical (Bisetty et al.
2006) and experimental studies (Albericio et al. 2008), that
some other non-natural cage amino acids, containing rigid
trishomocubane moiety, were strong b-turn inducers in
medium-sized peptides.
(A) 10
5
0
Apparently, the results of antiproliferative activity are in
accordance with our conformation study using CD,
whereby the spectrum of Met-enkephalin (1) in TFE dif-
fered significantly from those of its adamantylated deriv-
ative 2 indicating a higher population of ordered (likely
folded) conformers than of the extended structure, while 1
showed a CD spectrum similar in shape to that of Phe5-
analog 3. On the other hand, the CD spectra of compounds
4 and 5 were similar, also showing spectral features com-
patible with the presence of folded structures. Accordingly,
the peptides with higher population of folded conformers
exerted better biological activity.
-5
1
2
3
4
5
-10
-15
200
220
240
λ / nm
260
280
(B) 20
10
0
Conclusion
To conclude, a computational QSAR model was used to
predict the antiproliferative ability of small peptides by
screening a virtual library of enkephalin-like analogs
modified by incorporation of the (R,S)-(1-adamantyl)gly-
cine (Aaa) residue. From an initial set of 390 compounds,
peptides 2–5 were selected, synthesized and screened for
cytostatic activity. The results clearly revealed antiprolif-
erative activity of compounds 2–5 on three tumor cell lines.
The most active compounds were the hydrophobic pep-
tides, Phe-Aaa-Gly-Phe-Phe (4) and Phe-Aaa-Gly-Phe-Met
(5). The antiproliferative activity can be correlated with the
lipophilicity and foldamer-forming ability of the studied
peptides, as determined by CD spectroscopy. The results
suggest that the applied SVM model can be used to screen
for novel potentially active compounds with reasonable
1
2
3
4
5
-10
-20
200
220
240
λ / nm
260
280
Fig. 3 The far-UV region circular dichroism spectra of peptides 1–5
in TFE (a) and in a TFE-water (1:1) mixture (b). The spectra were
obtained at room temperature and expressed as molar CD (De/
dm3 mol-1 cm-1) from 185 to 280 nm
the possible interactions (salt bridge and/or aromatic side
chain interaction modified by the steric hindrance of the
adamantyl group) than the conformations of peptides 1, 3.
123

Computational structure–activity study directs synthesis of novel antitumor enkephalin analogs
1191
synthesis and in vitro growth inhibition of human tumor cells.
Chem Biol Drug Design 73:253–257
Hsu CW, Chang CC, Lin CJ (2008) A practical guide to support
vector classification. URL: http://www.csie.ntu.edu.tw/*cjlin/
papers/guide/guide.pdf
Ivanciuc O (2007) Applications of support vector machines in
chemistry. In: Lipkowitz KB, Cundari TR (eds) Reviews in
computational chemistry, vol 23. Wiley-VCH, Weinheim, pp
291–400. URL: http://www.ivanciuc.org/Files/Reprint/Ivanciuc_
SVM_CCR_2007_23_291.pdf
confidence. In addition, the results confirm the previous
observation (Knipp et al. 1997) that the solution confor-
mation has an effect on the physico-chemical characteris-
tics and consequently, the biological activity of the
peptides. We could speculate that the membrane perme-
ability of the compounds is relevant for the magnitude of
their cytostatic effect on tumor cell lines.
Acknowledgments This work was supported by grants from
the Croatian Ministry of Science, Education and Sports (grants No.
098-0982933-2936, 098-0982464-2514, 098-0982933-2911, 098-
0000000-3168) and by a grant from the Hungarian Scientific Research
Fund (OTKA NI168466) and the Bilateral Hungarian/Croatian project
Jorissen RM, Gilson MK (2005) Virtual screening of molecular
databases using a support vector machine. J Chem Inf Model
45:549–561
Karle IL, Ranganathan D, Kumar MG, Nagaraj R (2008) Design,
synthesis, conformational and membrane ion transport studies of
proline–adamantane hybrid cyclic depsipeptides. Biopolymers
89:471–478
ˇ
No. CRO-5/2006. The authors thank Milica Perc and Enidija Luksa
for excellent technical assistance.
Knipp GT, Vander Velde DG, Siahaan TJ, Borchardt RT (1997) The
effect of b-turn structure on the passive diffusion of peptides
across Caco-2 cell monolayers. Pharm Res 14:1332–1340
Kuroda Y, Ueda H, Nozawa H, Ogoshi H (1997) Adamantyl amino
acid as c-turn inducer for peptide. Tetrahedron Lett 38:7901–
7904
Miyazaki A, Tsuda Y, Fukushima S, Yokoi T, Vantus T, Bokonyi G,
Szabo E, Horvath A, Keri G, Okada Y (2008) Synthesis of
somatostatin analogues containing C-terminal adamantane and
their antiproliferative properties. J Med Chem 51:5121–5124
Noble WS (2004) Support vector machine applications in computa-
tional biology. In: Schoelkopf B, Tsuda K, Vert J-P (eds) Kernel
methods in computational biology. MIT Press, Cambridge,
pp 71–92. URL: http://noble.gs.washington.edu/papers/noble_
support.html
Sadowski J, Gasteiger J, Klebe G (1994) Comparison of automatic
three-dimensional model builders using 639 X-ray structures.
J Chem Inf Comput Sci 34:1000–1008
Smith JP, Conter RL, Bingaman SI, Harvey HA, Mauger DT, Ahmad
M, Demers LM, Stanley WB, McLaughlin PJ, Zagon IS (2004)
Treatment of advanced pancreatic cancer with opioid growth
factor: phase I. Anti-Cancer Drugs 15:203–209
References
Albericio F, Arvidson PI, Bisetty K, Giralt E, Govender T, Jali S,
Kongsaeree P, Kruger HG, Prabpai S (2008) Trishomocubane
amino acid as a b-turn scaffold. Chem Biol Drug Des 71:125–
130
´
ˇ
´
´
´
´
Basaric N, Molcanov K, Matkovic M, Kojic-Prodic B, Mlinaric-
Majerski K (2007) Adamantane-retropeptides, new building
blocks for molecular channels. Tetrahedron 63:7985–7996
Bisetty K, Corcho FJ, Canto J, Kruger HG, Perez JJ (2006) A
theoretical study of pentacyclo-undecane cage peptides of the
type (Ac-X-Y-NHMe). J Peptide Sci 12:92–105
Bohme Leite T, Gomes D, Miteva MA, Chomilier J, Villoutreix JO,
´
Tuffery P (2007) Frog: a FRee Online drug 3D conformation
generator. Nucleic Acids Res 35(Web Server issue):W568–
W572
Burbidge R, Trotter M, Buxton B, Holden S (2001) Drug design by
machine learning: support vector machines for pharmaceutical
data analysis. Comput Chem 26:5–14
Tetko IV, Gasteiger J, Todeschini R, Mauri A, Livingstone D, Ertl P,
Palyulin VA, Radchenko EV, Zefirov NS, Makarenko AS,
Tanchuk VY, Prokopenko VV (2005) Virtual computational
chemistry laboratory—design and description. J Comput Aid
Mol Des 19:453–463
Burges CJC (1998) A tutorial on support vector machines for pattern
recognition. Data Min Knowl Discov 2:121–167
Chang CC, Lin CJ (2001) LIBSVM: a library for support vector
machines. URL: http://cramer.cs.nmt.edu/*kdd/libsvm.pdf
El-Manzalawy Y, Honavar V. WLSVM (2005) Integrating LibSVM
into Weka environment. URL: http://www.cs.iastate.edu/
*yasser/wlsvm
ˇ
ˇˇ ´
Vass E, Hollosi M, Horvat S, Roscic M (2008) CD and FTIR
spectroscopic studies of carbohydrate-modified opioid peptides.
Croat Chem Acta 81:647–656
ˇ
´
ˇ
Horvat S, Mlinaric-Majerski K, Glavas-Obrovac Lj, Jakas A,
Wanka L, Cabrele C, Vanejews M, Schreiner PR (2007) Gamma-
aminoadamantanecarboxylic acids through direct bond amida-
tions. Eur J Org Chem 147:4–1490
´
Veljkovic J, Marczi S, Kragol G, Roscic M, Matkovic M,
Milostic-Srb A (2006) Tumor-cell-targeted methionine-enkeph-
ˇˇ ´
´
´
alin analogues containing unnatural amino acids: design,
synthesis, and in vitro antitumor activity.
49:3136–3142
Witten IH, Frank E (2005) Data mining: practical machine learning
tools and techniques, 2nd edn. Morgan Kaufmann, San Francisco
Zagon IS, Verderame MF, McLaughlin PJ (2002) The biology of the
opioid growth factor receptor (OGFr). Brain Res Rev 38:351–
376
J Med Chem
ˇ
Horvat S, Kralj M, Perc M, Jeric I, Varga-Defterdarovic L, Jakas A,
´
Roscic M, Suman L, Gredicak M (2009) Novel side-chain
´
ˇ
ˇˇ ´
ˇ
glucosylated and adamantylated [Asp2/Glu2]enkephalin analogs:
123