A three-dimensional quantitative structure-activity relationship study of heparin-binding epidermal growth factor shedding inhibitors using comparative molecular field analysis

Article abstract of DOI:10.1021/jm0110385

Despite the lack of structural information on the heparin-binding (HB) epidermal growth factor (EGF) shedding putative target enzyme, the design of potent HB-EGF shedding inhibitors has been attempted by means of comparative molecular field analysis (CoMFA), a well-established 3D-QSAR technique. Two different binding modes, obtained by docking a flexible representative into the MMP-3 and TACE target enzymes, were considered as alignment rules for an in-house data set of 50 HB-EGF shedding inhibitors. CoMFA models were derived with the standard steric, electrostatic, and Bohacek and McMartin's H-bond molecular fields. These fields were used individually or in combination. For both alignments, the H-bond field alone yielded the best statistical models. From the analysis of the CoMFA contours, ideas for testing the size of the S1′ pocket and suggestions for the design of new inhibitors came forward, resulting in the synthesis and testing of four new inhibitors. Three of four compounds turned out to possess from good (IC₅₀= 0.56 and 0.60 μM) to excellent (IC₅₀ = 0.13 μM) inhibitory activity. The hypothesis that, upon binding, the S1′ pocket in the vicinity of the R¹ benzene ring must be narrow in size was confirmed by the weak activity (IC₅₀ = 1.1 μM) of the fourth compound. The experimental profile of these new inhibitors does suggest the MMP-3 alignment as the most plausible one for HB-EGF shedding inhibitors.

Full text of DOI:10.1021/jm0110385

J . Med. Chem. 2002, 45, 781-788
781
A Th r ee-Dim en sion a l Qu a n tita tive Str u ctu r e-Activity Rela tion sh ip Stu d y of
Hep a r in -Bin d in g Ep id er m a l Gr ow th F a ctor Sh ed d in g In h ibitor s Usin g
Com p a r a tive Molecu la r F ield An a lysis
Roberta Bursi,*,† Masaaki Sawa,‡,§ Yasuyuki Hiramatsu,‡,# and Hirosato Kondo‡,
Department of Molecular Design & Informatics, N.V.Organon, P.O. Box 20, 5340 BH Oss, The Netherlands, and Department of
Chemistry, R&D Laboratories, Nippon Organon K.K., 1-5-90, Tomobuchi-cho, Miyakojima-ku, Osaka 534-0016, J apan
Received September 7, 2001
Despite the lack of structural information on the heparin-binding (HB) epidermal growth factor
(EGF) shedding putative target enzyme, the design of potent HB-EGF shedding inhibitors has
been attempted by means of comparative molecular field analysis (CoMFA), a well-established
3D-QSAR technique. Two different binding modes, obtained by docking a flexible representative
into the MMP-3 and TACE target enzymes, were considered as alignment rules for an in-
house data set of 50 HB-EGF shedding inhibitors. CoMFA models were derived with the
standard steric, electrostatic, and Bohacek and McMartin’s H-bond molecular fields. These
fields were used individually or in combination. For both alignments, the H-bond field alone
yielded the best statistical models. From the analysis of the CoMFA contours, ideas for testing
the size of the S1′ pocket and suggestions for the design of new inhibitors came forward,
resulting in the synthesis and testing of four new inhibitors. Three of four compounds turned
out to possess from good (IC50 ) 0.56 and 0.60 µM) to excellent (IC50 ) 0.13 µM) inhibitory
activity. The hypothesis that, upon binding, the S1′ pocket in the vicinity of the R1 benzene
ring must be narrow in size was confirmed by the weak activity (IC50 ) 1.1 µM) of the fourth
compound. The experimental profile of these new inhibitors does suggest the MMP-3 alignment
as the most plausible one for HB-EGF shedding inhibitors.
In tr od u ction
putative enzyme seems to be very similar to the active
site of other zinc-dependent metalloproteases, such as
the matrix metalloproteinases (MMPs), since several
MMP inhibitors blocked the phorbol ester 12-O-tetrade-
canoylphorbol-13-acetate (TPA) induced HB-EGF secre-
tion.
Heparin-binding (HB) epidermal growth factor (EGF)-
like growth factor is a member of the EGF family that
stimulates growth and differentiation.1,2 HB-EGF is
synthesized as a membrane-anchored precursor (proHB-
EGF), and then mature HB-EGF is released from the
cell surface by regulated proteolytic processing as other
transmembrane proteins.3 The importance of such
transmembrane protein shedding for cytokine biology
has been increasingly recognized.4 HB-EGF has been
implicated as a participant in a variety of normal and
aberrant processes such as wound healing, blastocyst
implantation, SMC hyperplasia, atherosclerosis, and
tumor growth.5 Thus, agents, which inhibit HB-EGF
production, would be effective in the treatment of such
diseases.
To our knowledge, the responsible enzyme for HB-
EGF shedding has not been identified as yet, and no
structure-activity relationship (SAR) studies on HB-
EGF shedding inhibitors have been reported in the
literature. Several studies, however, have suggested
that zinc-dependent metalloproteases are involved in
the processing of proHB-EGF.6,7 The active site of the
The rational design of new HB-EGF shedding inhibi-
tors is strongly disturbed by a lack of structural
information on the responsible target enzyme. For this
reason, we have reported the SAR of a series of non-
peptidic hydroxamate-based bicyclic derivatives as in-
hibitors of HB-EGF shedding to obtain a better under-
standing of the SAR of the responsible enzyme.8-10
Nowadays, a 3D quantitative structure-activity re-
lationship (3D-QSAR)11-13 technique such as Compara-
tive Molecular Field Analysis (CoMFA)14 is routinely
used in structure-based drug design to obtain topological
information on the active site of the investigated target
in terms of molecular fields (mostly steric and electro-
static) complementarity. It has been already shown in
the literature that this approach can strongly support
and help the design of novel and potent inhibitors.15-19
In this study, a CoMFA analysis of a in-house series of
HB-EGF shedding inhibitors was performed, and the
active site’s topographical features of the putative
sheddase were discussed. Moreover, we designed and
synthesized new inhibitors based on the CoMFA results
and tested their in vitro efficacy.
* To whom correspondence should be addressed. Tel: (+31) (0)412-
661468. Fax: (+31) (0)412-662539. E-mail: roberta.buma@organon.com.
† N.V.Organon.
‡ Nippon Organon K.K.
§ New address: Chemistry Research Laboratories, Drug Research
Division, Dainippon Pharmaceutical Co., Ltd., Osaka 564-0053, J apan.
#
New address: Medicinal Chemistry, Discovery Research Labora-
Meth od s
tories, Shionogi & Co., Ltd., Osaka 553-0002, J apan.
New address: Manufacturing Technology R & D Laboratories,
Shionogi & Co., Ltd., Hyogo 660-0813, J apan.
Ch em istr y. The structures of the 50 compounds used in
this study are listed in Tables 1 and 2. These compounds have
10.1021/jm0110385 CCC: $22.00 © 2002 American Chemical Society
Published on Web 01/18/2002

782 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4
Bursi et al.
Ta ble 1. Chemical Structures and HB-EGF Shedding Inhibition Data of 1,2,3,4-Tetrahydroisoquinoline Derivatives
been synthesized from sulfonyl chlorides and 1,2,3,4-tetrahy-
droisoquinoline derivatives or 5,6,7,8-tetrahydropyrido[3,4-b]-
pyrazine derivatives as described in the patent and elsewhere
(Scheme 1).8,9 Thus, 1,2,3,4-tetrahydroisoquinoline derivatives
or 5,6,7,8-tetrahydropyrido[3,4-b]pyrazine derivatives were
treated with appropriate sulfonyl chlorides (commercially
available or prepared from corresponding sulfonic acids with
thionyl chloride or corresponding aromatic compounds with

QSAR Study of HB-EGF Shedding Inhibitors
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4 783
Ta ble 2. Chemical Structures and HB-EGF Shedding Inhibition Data of 5,6,7,8-Tetrahydropyridopyrazine Derivatives
characteristic data (1H NMR, MS spectra, and combustion
analyses) that fulfilled Organon criteria (purity g 95%) for
Sch em e 1
biological testing.
HB-EGF Sh ed d in g In h ibition Assa y.7 Fibrosarcoma HT-
1080 transfectants expressing human placental alkaline phos-
phatase (AP)-tagged HB-EGF in MEM (containing 10% FCS)
as culture medium were seeded in 96-well plates at a density
of 2 × 105 cells/well and incubated for 24 h. The cells were
washed with PBS and preincubated with the test compounds
in MEM (containing 1% DMSO) for 30 min. TPA (60 nM) was
added to stimulate inducible processing, and the plate was
incubated for 60 min. A 0.1 mL aliquot of the supernatant was
transferred to a 96-well plate and heated for 10 min at 65 °C
in order to inactivate endogenous alkaline phosphatases. A 0.1
mL of substrate solution (1 M diethanolamine, 0.01% MgCl2,
1 mg/mL p-nitrophenyl phosphate, pH 9.8) was added to each
well, and the plate was incubated for 2 h. AP activity was then
determined by the measurement of absorbance at 405 nm with
a microplate reader. The IC50 values were determined with
different inhibitor concentrations by using GraphPad Prism
Version 3.0 (GraphPad Software, Inc.).
Com p u ta tion a l P r oced u r es. Within CoMFA, structural
superposition is a condicio sine qua non. Structures must be
aligned with respect to a reference compound or template,
which mimics the binding mode of the investigated compounds
with respect to their corresponding target. When experimental
information is available, CoMFA potential is maximized. In
absence of such information, structural alignment can become
chlorosulfonic acid) in dioxane-water in the presence of
organic bases. In the case of the 6,7-dihydroxyl analogue, the
reaction was carried out in the presence of sodium tetraborate
and sodium hydroxide to protect hydroxyl groups. The result-
ing sulfonamides were converted to the corresponding hydrox-
amic acids by commonly used procedures. All compounds gave

784 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4
Bursi et al.
F igu r e 1. Two-dimensional structure of compound 37.
F igu r e 3. Conformation of 37-TACE.
F igu r e 2. Conformation of 37-MMP.
a problem, particularly when flexible compounds need to be
superimposed.
As can be seen in Tables 1 and 2, in most of the HB-EGF
inhibitors present in our data set the R1 substituents on the
sulfonamide moiety are p-substituted phenyl rings and these
para substituents are quite flexible. In few inhibitors, o- or
m-substitutions come forward as well, and their orientation
with respect to the sulfonamide-bridging moiety is clearly a
critical issue. Further, R1 substituents are not only (un)-
substituted phenyl moieties but also include heterocycles, like
(un)substituted thiophens or pyridines. The position of the
heteroatom with respect to the sulfonamide bridge is also
critical.
F igu r e 4. Compounds chosen as representatives of o-, m-
substituted phenyl and thiophen/pyridin R1 substituents of
the sulfonamide.
Since meaningful CoMFA models could only be produced
when all these issues were properly dealt with, modeling work
was performed in order to explore all alignment possibilities
and to arrive at (a) highly probable binding mode(s) of these
inhibitors toward their putative enzyme. The computational
strategy, which was followed in this study, is explained below.
Ca lcu la tion of th e Exp ected Bioa ctive Con for m a tion -
(s) of F lexible In h ibitor s. Compound 37 (Figure 1) was
chosen as a typical flexible representative of this data set. A
docking study was then performed on both the MMP-3 (bis-
sulfonamide inhibitor complex, PDB code: 1BQO)20 and TACE
(peptide-based inhibitor complex, PDB code: 1BKC).21 These
enzymes were chosen as target enzymes since the inhibition
profiles of our compounds on MMP-3, TACE, and HB-EGF
shedding look similar.10
Compound 37 was manually docked into the active sites of
these enzymes. First, compound 37 was placed in the active
site of the MMP-3 to occupy the S1′ pocket with the R1 group.
Then, the sulfonamide moiety of compound 37 was superim-
posed on one of the sulfonamide moieties of the bis-sulfonamide
inhibitor. Then, the protein-ligand complexes were energy
minimized, treating all ligand atoms and retaining all protein
atoms. Next, the MMP-3/compound 37 complex model was
superimposed on TACE/peptide-based inhibitor complex to
overlap the C-R chains of the enzymes. Then, the peptide-based
inhibitor was replaced by compound 37, and the obtained
protein-ligand complexes were subsequently energy mini-
mized, treating all ligand atoms and retaining all protein
atoms. This led to two possible conformations of compound 37,
the 37-MMP and the 37-TACE conformation, respectively.
These are shown in Figures 2 and 3. Docking and energy
minimizations were performed with the TRIPOS force field22
within the SYBYL program.23 All computations were per-
formed on an Octane Silicon Graphics workstation.
F igu r e 5. Orientation of sides A and B, respectively.
Ta ble 3. Comparison of the Energy Differences of the
Protein-Ligand Complexes
MMP-3
TACE
energy difference
energy difference
compd
(A - B) kcal/mol
(A - B) kcal/mol
38
51
36
33
1.0
-4.0
-2.5
-1.5
-1.6
5.8
-3.6
-2.3
thiophen/pyridine R1 substituents of the sulfonamide. These
four compounds were docked in their two possible correspond-
ing conformations, which we will call here for simplicity “side
A” and “side B”, into the active sites of MMP-3 and TACE (see
Figure 5). The eight complexes were energy minimized (by
always keeping the structure of the protein rigid), and the
energy differences between side A and side B were obtained.
The results are shown in Table 3. Since three of four complexes
possess the lowest energy when the ligand is present in the
conformation pointing to side A in both MMP-3 and TACE
active sites, we have chosen side A as alignment orientation
of this subgroup of inhibitors. Docking and energy minimiza-
tions were performed with the SYBYL program.
CoMF A Mod els Develop m en t. The remaining data set
was energy minimized, and Coulson partial charges were
calculated at the AM1 semiempirical level24 for all compounds.
The data set was then aligned on each of the two templates,
37-MMP and 37-TACE, in two steps: first, a rigid alignment
was performed on the tetrahydropyridopyrazine and the
sulfonamide moieties; second, the flexible substituents of the
phenyl ring in R1 moieties were aligned by flexible fitting to
Ca lcu la tion of th e Exp ected Dir ection s of o-, m -
Su bstitu en ts a t th e P h en yl Rin g or th e Su lfu r /Nitr ogen
At om in t h e Th iop h en /P yr id in e R in g At t a ch ed t o t h e
Su lfon a m id e. Compounds 36, 51, 38, and 33 (see Figure 4)
were chosen as representatives of o-, m-substituted phenyl and

QSAR Study of HB-EGF Shedding Inhibitors
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4 785
Ta ble 4. CoMFA Models Based on the 37-MMP Alignmenta
match the orientation of the phenylpropoxy substituent of 37-
MMP and 37-TACE, respectively. CoMFA models were derived
on the two resulting data sets.
grid
#
field(s)
(Å) q2
s
LV r2
s
F
field %
SYBYL standard steric (van der Waals radii), electrostatic
(Coulson partial charges), and H-bond25,26 fields were applied.
The H-bond field is based on the Lee and Richards accessible
surface (L&R surface), which consists of a van der Waals
radius plus a probe radius away from the ligand. In Bohacek’s
study, it was shown that this surface can display regions for
hydrophobic, hydrogen-bond acceptor, and hydrogen-bond
donor properties around the ligand, which are accurate and
complementary to the target protein. Regions on the protein
surface where hydrogen bonds can be formed can be expected
to be hydrophilic, while regions which lack hydrogen-bonding
groups are likely to be hydrophobic. Further, the H-bond field
has an advantage on the electrostatic calculation, since it is
not force field dependent while atomic partial charges are.
Within CoMFA, the H-bond field has been implemented in
an acceptor (“steric” field type) and a donor (“electrostatic” field
type) component, as given by Bohacek and McMartin. The
green and yellow contours will therefore display the favorable
and unfavorable acceptor field-type regions, while the red and
blue contours will display the favorable and unfavorable donor
field-type regions.
ster
ster
elec
elec
H-bond
H-bond
2
1
2
1
2
1
0.15 0.71
0.35 0.64
0.30 0.66
0.21 0.68
0.16 0.72
0.41 0.61
1
5
4
1
3
4
0.38 0.60 29.4
0.87 0.29 59.8 ster (1.00)
0.83 0.33 65.6
0.42 0.58 35.2 elec (1.00)
0.70 0.43 36.3
0.90 0.25 98.4 acc (0.56)/
don (0.44)
0.91 0.25 83.5
0.89 0.26 89.9 ster (0.46)/
elec (0.54)
0.69 0.44 51.2
0.95 0.18 162.3 ster (0.22)/
acc (0.42)/
ster + elec
ster + elec
2
1
0.37 0.63
0.37 0.63
5
4
ster + H-bond
ster + H-bond
2
1
0.23 0.68
0.45 0.59
2
5
don (0.36)
elec + H-bond
elec + H-bond
2
1
0.21 0.70
0.34 0.63
3
2
0.75 0.39 46.6
0.77 0.37 79.5 elec (0.24)/
acc (0.40)/
don (0.36)
ster + elec +
H-bond
ster + elec +
H-bond
2
1
0.22 0.68
0.43 0.61
2
5
0.69 0.43 51.8
0.95 0.18 158.6 ster (0.18)/
elec (0.18)/
acc (0.34)/
Field values were not smoothed, and the cutoff values of
steric and electrostatic interactions were kept to the default
values of 30.0 kcal/mol. A smooth transition was chosen
between the cutoff plateaus for the steric and the electrostatic
calculations, and every Coulombic electrostatic energy calcula-
tion was performed using a distance-dependent dieletric ꢀ )
R. The CoMFA standard scaling procedure was used through-
out the analyses.
Performing the CoMFA analyses at 1 Å grid distance
ensured rotational and translational invariance of the statisti-
cal results (q2 and r2). Cross-validated correlation coefficients,
q2, were obtained with SAMPLS,27 and subsequent final
correlation coefficients, r2, were calculated with conventional
PLS.28 In every analysis, the standard error of estimate, s,
associated with both q2 and r2 is provided as well as the F-ratio,
which is the ratio between r2 and 1- r2 (explained to
unexplained). The larger is the F-ratio, the better is the model.
Correlation chances were checked repeatedly by performing
PLS analyses on randomized activity (-log IC50 (µM)) sets. No
“randomized” models turned out to have cross-validated cor-
relation coefficient values, q2, close to the correct one.
don (0.30)
a
Molecular fields were calculated at 2 (default) and 1 Å grid
distances, respectively. AM1 semiempirical charges were employed
to account for electrostatic interactions. The H-bond field is
described in terms of H-bond acceptor and donor components.
This might well be due to an intrinsic correlation
between these fields. The combination of the steric and
the H-bond fields, on the contrary, resulted in some,
although limited, improvement of the corresponding
individual models. To a certain extent, this result was
disappointing since more was expected from the com-
bination of two fields, which, if taken alone, well
describe this data set. Plausible conclusion of this result
is that, as for the electrostatics counterpart, the H-bond
field based on the L&R accessible surface is correlated
to a certain extent to the steric field. When all three
molecular fields were employed, no significant statistical
improvements were obtained.
In the last column on the right-hand side of Table 4,
field contributions to each analysis are displayed. From
these values, it can be clearly seen that the H-bond field
accounts for the largest model contributions when it is
combined either with the steric or the electrostatic
fields. In the three-fields combination, the H-bond field
is still the major contribuent (64%) to the model.
In conclusion, two statistically significant models were
obtained for the 37-MMP alignment: the H-bond field
alone and in combination with the steric field. Since the
latter provides slightly better statistics (q2 and s values)
than the former, but with the requirement of an extra
component (5 LV against 4 LV), the H-bond field alone
was preferred to the combination and was called model
1.
The results of the CoMFA models based on the 37-
TACE alignment are shown in Table 5. In this case, no
significant differences were found between the models
obtained at 2 or 1 Å grid distances. For consistency,
however, with the former analysis, we will focus our
attention again on the highest resolution models. While
looking at the results, one can see that these models,
on average, required a smaller number of components
with respect to their corresponding models shown in
Table 4. This generally caused smaller final correlation
Resu lts a n d Discu ssion
CoMF A Mod els. Several models were derived by
applying steric, electrostatic, and H-bond fields indi-
vidually and in combination. In Table 4, the results for
the 37-MMP alignment are shown. Clear differences
were found between cross-validated correlation coef-
ficients (q2) and the number of latent variable (LV)
values obtained at 2 (default) and 1 Å grid distances
for each analysis, suggesting some grid dependence of
the statistical results. We will focus therefore our
attention only on the results obtained at the highest
resolution (1 Å). When molecular fields were used
individually, the H-bond field resulted to be the best
(q2 ) 0.41, s ) 0.61, 4 LV, r2 ) 0.90, s ) 0.25), followed
by the steric field (q2 ) 0.35, s ) 0.64, 5 LV, r2 ) 0.87,
s ) 0.29). The electric field alone did not provide an
acceptable model. This result was quite surprising since
in most compounds, although quite hydrophobic in
nature, large polar substituents are present. When
molecular fields were combined, the addition of the
electrostatic field to the steric or the H-bond field did
not improve the statistics of the corresponding indi-
vidual models. In the particular combination of the
electrostatic and the H-bond field, the result is poor.

786 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4
Ta ble 5. CoMFA Models Based on the 37-TACE Alignmenta
Bursi et al.
grid
#
field(s)
(Å) q2
s
LV r2
s
F
field %
ster
ster
elec
elec
H-bond
H-bond
2
1
2
1
2
1
0.27 0.66
0.35 0.63
0.18 0.69
0.20 0.69
0.42 0.59
0.42 0.60
2
3
1
1
3
3
0.55 0.52 28.5
0.74 0.40 44.5 ster (1.00)
0.36 0.61 27.2
0.38 0.60 29.8 elec (1.00)
0.79 0.36 59.2
0.82 0.34 67.6 acc (0.59)/
don (0.41)
ster + elec
ster + elec
2
1
0.25 0.66
0.28 0.66
1
2
0.44 0.57 38.0
0.70 0.43 53.8 ster (0.49)/
elec (0.51)
ster + H-bond
ster + H-bond
2
1
0.40 0.60
0.38 0.61
2
2
0.73 0.40 64.7
0.73 0.40 65.9 ster (0.28)/
acc (0.39)/
don (0.33)
elec + H-bond
elec + Hbond
2
1
0.38 0.62
0.37 0.62
3
3
0.82 0.34 67.4
0.83 0.32 74.9 elec (0.24)/
acc (0.43)/
F igu r e 6. CoMFA contours of model 1, 37-MMP. The green
and yellow contours display favorable and unfavorable H-bond
acceptor field-type regions, while the red and blue contours
display the favorable and unfavorable H-bond donor field-type
regions, respectively.
don (0.33)
ster + elec +
H-bond
ster + elec +
H-bond
2
1
0.37 0.62
0.35 0.63
2
2
0.73 0.40 63.1
0.73 0.41 62.0 ster (0.23)/
elec (0.17)/
acc (0.32)/
don (0.28)
the most important one. Clearly, the presence of H-bond
donating and accepting groups at this site of the
inhibitors is very important for the desired activity
profile. At the same time, these contours suggest a
highly hydrophilic region at the complementary site of
the putative enzyme-binding pocket.
In the second case, only the acceptor component is
present, and the green and yellow regions are present
almost 360° all around the flexible chains, the green
regions being the closest ones to the substituent. The
analysis shows that the presence of acceptor group(s)
in this region of the S1′ pocket will indeed increase the
affinity of the inhibitors, but that these groups may not
be large in size. This result suggests that a hydrophilic
region (H-bond donors) must be present in the corre-
sponding site of the putative enzyme, but that the
available space in the cavity upon binding must be quite
narrow.
Further, minor H-bond field contributions are present
at the very end of the S1/S2 pocket alongside the
tetrahydropyridopyrazine moiety and below the tet-
rahydropyridopyrazine plane at several a˚ngstro¨ms of
distance. Few small field contributions are also present
around the sulfonic bridge. No H-bond field contribu-
tions were found above the tetrahydropyridopyrazine
plane, suggesting that this region must be hydrophobic
in nature. This result agrees with the description of the
S1/S2 sites in both MMP-3 and TACE,15 where a
π-stacking interaction is envisaged between ligand and
enzyme.
a
Molecular fields were calculated at 2 (default) and 1 Å grid
distances, respectively. AM1 semiempirical charges were employed
to account for electrostatic interactions. The H-bond field is
described in terms of H-bond acceptor and donor components.
coefficients values (r2). Despite these differences, it can
be concluded that the general statistical trend that was
identified in the models based on the 37-MMP align-
ment comes forward in these models as well.
The best model was obtained with the H-bond field
alone (q2 ) 0.42, s ) 0.60, 3 LV, r2 ) 0.82, s ) 0.34),
while the poorest one was again the electrostatic field.
The H-bond field was called model 2. Within the frame-
work of this alignment, no field combination yield statis-
tically improved models, and the individual field con-
tributions are comparable to the ones given in Table 4.
CoMF A Con tou r s. In absence of information on the
experimental binding mode of these compounds toward
their putative target enzyme, one could be tempted of
identifying such binding mode in the alignment, which
has produced the best statistical results. Unfortunately,
this may not be the case. We have previously discussed29
that better statistical results do not necessarily point
to the actual binding mechanism and that in some cases
experimental alignments can lead to worse statistics
than a priori, theoretically derived, and nonphysical
structural superpositions. For this reason, we regarded
models 1 and 2 as equally plausible, and we proceeded
to analyze the molecular field contours of both.
The H-bond contours of model 1 are displayed in
Figure 6. While looking at these contours, it must be
borne in mind that the acceptor component contribu-
tions of the H-bond field are depicted in green (favor-
able) and yellow (unfavorable), while the donor compo-
nent contributions are given in red (favorable) and blue
(unfavorable). The model provides major information
mostly around the hydroxamic acid substituent on the
tetrahydropyridopyrazine and around the R1 flexible
substituent (S1′ pocket) covering approximately a 6 Å
distance from the benzene ring.
Finally, no H-bond contributions are present at the
very end of the flexible R1 substituents, suggesting the
presence of a region in the ligands (and therefore in the
enzyme), where hydrophobic interactions prevail.
The H-bond contours of model 2 are displayed in
Figure 7. Bearing in mind the results that were just
discussed for model 1, it was interesting to see that
similar information could be extracted from the contours
of model 2 despite the completely different orientation
of the flexible chains of the R1 substituents.
The only differences are some larger donor contours
alongside the tetrahydropyridopyrazine moiety at the
very end of the S1/S2 pocket and the presence of a small
donor contour in the S1′ pocket.
In the first case, the region around the hydroxamic
acid group is described by both donor and acceptor
H-bond field components, the donor component being

QSAR Study of HB-EGF Shedding Inhibitors
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4 787
Ta ble 6. Chemical Structures and HB-EGF Shedding Inhibition Data of 5,6,7,8-Tetrahydropyridopyrazine Derivatives, CoMFA
Predictions, and Percentage Error (YExp - YPred/YExp
)
group. In compound 55, the ethylene group was ad hoc
chosen to test the actual available space of this narrow
site, and it was expected to be potentially less active
because of its bending shape.
The 5,6,7,8-tetrahydropyrido[3,4-b]pyrazine scaffold
was chosen as mother skeleton in all four compounds
in order to compensate for potential hydrophobicity of
the new substituents.8
In silico predictions, synthesis, and testing were
performed on these four compounds. Computational and
experimental results are shown in Table 6. Compounds
52 and 53 exhibited excellent inhibitory activity against
HB-EGF shedding with sub-micromolar IC50 values,
although their activity was slightly less than the cor-
responding parent compound 24. Compound 54 exhib-
ited excellent inhibitory activity against HB-EGF shed-
ding (IC50 ) 0.13 µM), which places it among the most
potent inhibitors of the in-house data set considered
here, while the activity of compound 55 showed moder-
ate activity as expected (IC50 ) 1.1 µM). Both CoMFA
models 1 and 2 turned out to have well predicted three
of the four compounds, as the error percentages show.
Interestingly, both models had underestimated com-
pound 54. An important difference, however, between
the two models needs to be outlined: while model 1 did
predict semiquantitatively the experimental trend and,
in particular, the relative activities of compounds 54 and
55, model 2 did not. Although limited to a small set of
compounds, these results seem to point at model 1 as
the model, which better describes the most probable
binding mode of these HB-EGF shedding inhibitors.
F igu r e 7. CoMFA contours of model 2, 37-TACE. The green
and yellow contours display favorable and unfavorable H-bond
acceptor field-type regions, while the red and blue contours
display the favorable and unfavorable H-bond donor field-type
regions, respectively.
By the analyses of the contours of models 1 and 2, it
could be concluded that for this data set of inhibitors
both alignments tend to provide similar information on
the binding cavity of the HB-EGF putative target
enzyme.
Eva lu a tion of New In h ibitor s. The CoMFA models
1 and 2 were applied to the design of new HB-EGF
shedding inhibitors.
If we look at the S1′ pocket, according to these models
the addition of groups to the end of the alkoxyphenyl
moiety may increase the activity as long as these groups
are narrow and polar (H-bond acceptors) in the vicinity
of such moiety and bulky and hydrophobic at the very
end. By making use of this information and checking
upon synthetic feasibility, four new inhibitors (see Table
6) were designed with the following strategy: two
compounds (52 and 53) were designed with a narrow
and polar (H-bond acceptors) long tail. In one of the two,
compound 52, a bulky and hydrophobic substituent was
added at the very end of the tail. Two compounds (54
and 55) were designed with short and polar (high
electron density) tails. In compound 54, the acetylene
moiety was expected to well fit the narrow space present
in the cavity upon binding in the vicinity of the phenyl
Con clu sion s
At present, the design of potent HB-EGF shedding
inhibitors as well as the identification of the HB-EGF
shedding putative target enzyme are great challenges
for many scientists at universities and pharmaceutical
companies.
In this study, we have attempted the design of new
and potent compounds by means of performing com-
parative molecular field analysis on an in-house con-
generic data set of HB-EGF shedding inhibitors.
Plausible alignment rules were derived from struc-
tural information provided by the MMP-3 and TACE

788 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 4
Bursi et al.
(10) Sawa, M.; Kiyoi, T.; Kurokawa, K.; Kumihara, H.; Yamamoto,
M.; Miyasaka, T.; Ito, Y.; Hirayama, R.; Inoue, T.; Kirii, Y.;
Nishiwaki, E.; Ohmoto, H.; Maeda, Y.; Ishibishi, E.; Inouhe, Y.;
Yoshino, K.; Kondo, H. New Type of Metalloproteinase Inhibi-
tor: Design and Synthesis of New Phosphonamide-Based Hy-
droxamic Acid. Submitted for publication.
(11) 3D-QSAR in Drug Design. Theory, Methods and Applications;
Kubinyi, H., Ed.; ESCOM: Leiden (NL), 1993.
(12) 3D-QSAR in Drug Design. Volume 2. Ligand-Protein Interactions
and Molecular Similarity; Kubinyi, H., Folkers, G., Martin, Y.
C., Eds.; Kluwer/ESCOM: Dordrecht (NL), 1997.
(13) 3D-QSAR in Drug Design. Volume 3. Recent Advances; Kubinyi,
H., Folkers, G., Martin, Y. C., Eds.; Kluwer/ESCOM: Dordrecht
(NL), 1998.
(14) Cramer, R. D., III; Patterson, D. E.; Bunce, J . D. Comparative
Molecular Field Analysis (CoMFA). 1. Effect of Shape on Binding
of Steroids to Carrier Proteins. J . Am. Chem. Soc. 1988, 110,
5959-5967.
(15) Matter, H.; Schwab, W.; Barbier, D.; Billen, G.; Haase, B.; Neises,
B.; Schudok, M.; Thorwart, W.; Schreuder, H.; Brachvogel, V.;
Lo¨nze, P.; Weithmann, K. U. Quantitative Structure-Activity
Relationship of Human Neutrophil Collagenase (MMP-8) Inhibi-
tors Using Comparative Molecular Field Analysis and X-ray
Structure Analysis. J . Med. Chem. 1999, 42, 1908-1920.
(16) Matter, H.; Schwab, W. Affinity and Selectivity of Matrix
Metalloproteinase Inhibitors: A Chemometrical Study from the
Perspective of Ligands and Proteins. J . Med. Chem. 1999, 42,
4506-4523.
(17) Fichera, M.; Cruciani, G.; Bianchi, A.; Musumarra, G. A 3D-
QSAR Study on the Structural Requirements for Binding to CB1
and CB2 Cannabinoid Receptors. J . Med. Chem. 2000, 43, 2300-
2309.
(18) Martinez, A.; Gil, C.; Abasolo, M. I.; Castro, A.; Bruno, A. M.;
Perez, C.; Prieto, C.; Otero, J . Benzothiadiazine Dioxide Dibenzyl
Derivatives as Potent Human Cytomegalovirus Inhibitors: Syn-
thesis and Comparative Molecular Field Analysis. J . Med. Chem.
2000, 43, 3218-3225.
(19) Gnerre, C.; Catto, M.; Leonetti, F.; Weber, P.; Carrupt, P.-A.;
Altomare, C.; Carotti, A.; Testa, B. Inhibition of Monoamine
Oxidases by Functionalized Coumarin Derivatives: Biological
Activities, QSARs, and 3D-QSARs. J . Med. Chem. 2000, 43,
4747-4758.
(20) Pikul, S.; Dunham, K. L. M.; Almstead, N. G.; De, B.; Natchus,
M. G.; Anastasio, M. V.; McPhail, S. J .; Snider, C. E.; Taiwo, Y.
O.; Rydel, T.; Dunaway, C. M.; Gu, F.; Mieling, G. E. Discovery
of Potent, Achiral Matrix Metalloproteinase Inhibitors. J . Med.
Chem. 1998, 41, 3568-3571.
(21) Maskos, K.; Fernandez-Catalan, C.; Huber, R.; Bourenkov, G.
P.; Bartunik, H.; Ellestad, G. A.; Reddy, P.; Wolfson, M. F.;
Rauch, C. T.; Castner, B. J .; Davis, R.; Clarke, H. R. G.; Petersen,
M.; Fitzner, J . N.; Cerretti, D. P.; March, C. J .; Paxton, R. J .;
Black, R. A.; Bode, W. Crystal Structure of the Catalytic Domain
of Human Tumor Necrosis Factor-R-converting Enzyme. Proc.
Natl. Acad. Sci. U.S.A. 1998, 95, 3408-3412.
(22) Clark, M.; Cramer, R. D., III.; Van Opdenbosch, N. Validation
of the general purpose Tripos 5.2 Force Field. J . Comput. Chem.
1989, 10, 982-1012, and SYBYL manual (SYBYL 6.6) including
corrections to the Tripos 5.2 force field version.
complexes, leading to two possible hypothetical binding
modes of our flexible inhibitors in the putative binding
cavity of the HB-EGF shedding target enzyme.
CoMFA models were subsequently obtained with
different molecular fields. Surprisingly, most models
showed modest cross-validated correlation coefficients
(q2 < 0.4), despite the high degree of relation of the
inhibitors considered. The H-bond molecular field did
not only yield the best statistical models for both
alignments, but it also provided a simple and intuitive
description of the structure-activity relationship of
these compounds in terms of H-bond acceptor/donor
contributions and complementary hydrophobicity.
From considerations based on synthetic feasibility and
CoMFA analyses and contours, novel inhibitors were
designed, synthesized, and tested for model validation.
The experimental results did confirm both models as
predictive, did unravel three new compounds with
excellent HB-EGF shedding inhibitory activity, and did
suggest for these inhibitors the MMP-3-based alignment
as the most probable one.
Despite the complexity of the objective before us, the
necessary assumptions, and the intrinsic limitations of
the techniques to our disposal, this study confirmed the
utility of CoMFA in particular and rational drug design
in general to set steps forward in this specific area of
research.
Ack n ow led gm en t. We greatly acknowledge Dr. S.
Higashiyama (School of Medicine, Osaka University,
Osaka, J apan) for the gift of the expression vector of
HB-EGF fused with human placental alkaline phos-
phatase (AP) and Dr. J . de Vlieg (Molecular Design &
Informatics) for support during the preparation of this
manuscript.
Refer en ces
(1) Higashiyama, S.; Abraham, J . A.; Miller, J .; Fiddes, J . C.;
Klagsbrun, M. A Heparin-Binding Growth Factor Secreted by
Macrophage-Like Cells that is Related to EGF. Science 1991,
251, 936-939.
(2) Hashimoto, K.; Higashiyama, S.; Asada, H.; Hashimura, E.;
Kobayashi, T.; Sudo, K.; Nakagawa, T.; Damm, D.; Yoshikawa,
K.; Taniguchi, N. Heparin-binding Epidermal Growth Factor-
like Growth Factor is an Autocrine Growth Factor for Human
Keratinocytes. J . Biol. Chem. 1994, 269, 20060-20066.
(3) Piepkorn, M.; Pittelkow, M. R.; Cook, P. W. Autocrine Regulation
of Keratinocytes: The Emerging Role of Heparin-binding,
Epidermal Growth Factor-related Growth Factors. J . Invest.
Dermatol. 1998, 111, 715-721.
(4) Mu¨llberg, J .; Althoff, K.; J ostock, T.; Rose-J ohn, S. The Impor-
tance of Shedding of Membrane Proteins for Cytokine Biology.
Eur. Cytokine Network 2000, 11, 27-37.
(5) Raab, G.; Klagsbrun, M. Heparin-binding EGF-like Growth
Factor. Biochim. Biophys. Acta 1997, 1333, F179-F199. A
review.
(6) Prenzel, N.; Zwick, E.; Daub, H.; Leserer, M.; Abraham, R.;
Wallasch, C.; Ullrich A. EGF Receptor Transactivation by
G-protein-coupled Receptors Requires Metalloproteinases Cleav-
age of proHB-EGF. Nature 1999, 402, 884-888.
(7) Tokumaru, S.; Higashiyama, S.; Endo, T.; Nakagawa, T.; Miya-
gawa, J .; Yamamori, K.; Hanakawa, Y.; Ohmoto, H.; Yoshino,
K.; Shirakata, Y.; Matsuzawa, Y.; Hashimoto, K.; Taniguchi, N.
Ectodomain Shedding of Epidermal Growth Factor Receptor
Ligands is Required for Keratinocyte Migration in Cutaneous
Wound Healing. J . Cell. Biol. 2000, 151, 209-219.
(23) SYBYL 6.7. Tripos Assoc., 1699 S. Hanley Rd., St. Louis, MO
63144.
(24) Stewart, J . J . P. MOPAC 6.0, Quantum Chemical Program
Exchange 455, 1990.
(25) Bohacek, R. S.; McMartin C. Definition and Display of Steric,
Hydrophobic, and Hydrogen-Bonding Properties of Ligand Bind-
ing Sites in Proteins Using Lee and Richards Accessible Sur-
face: Validation of a High-Resolution Graphical Tool for Drug
Design. J . Med. Chem. 1992, 35, 1671-1684.
(26) Kellogg, G. E. Finding Optimum Field Models for 3D QSAR.
Med. Chem. Res. 1997, 7: 6/7, 417-427.
(27) Bush, B. L.; Nachbar, R. B., J r. Sample-distance Partial Least
Squares: PLS optimized for many variables, with application
to CoMFA. J . Comput.-Aided Mol. Des. 1993, 7, 587-619.
(28) Wold, S.; Albano, C.; Dunn, W. J .; Edlund, U.; Esbensen, K.;
Geladi, P.; Hellberg, S.; J ohansson, E.; Lindberg, W.; Sjostrom,
M. In Chemometrics: Mathematics and Statistics in Chemistry;
Kowalski, B., Ed.; Dordrecht, The Netherlands, 1984.
(29) Bursi, R.; Grootenhuis, P. D. J . Comparative molecular field
analysis and energy interaction studies of thrombin-inhibitor
complexes. J . Comput.-Aided Mol. Des. 1999, 13, 221-232.
(8) Yoshiizumi, K.; Yamamoto, M.; Miyasaka, T.; Ito, Y.; Kumihara,
H.; Sawa, M.; Kiyoi, T.; Yamamoto, T.; et al. Synthesis, HB-EGF
Shedding Inhibitory Activities and MMP Inhibitory Activities
of Hydroxamic Acids Having Pyrido[3,4-b]pyrazine Skeleton.
Submitted for publication.
(9) Hashimoto, J .; Higashiyama, S.; Yoshino, K.; Yoshiizumi, K.;
Yamamoto, M.; Kiyoi, T.; Kurokawa, K.; Kondo, H.; Sawa, M.;
Kumihara, H. Keratinocyte Growth Inhibitors and Hydroxamic
Acid Derivatives. PCT Patent Application. WO 01/70269.
J M0110385