CoMFA analysis of tgDHFR and rlDHFR based on antifolates with 6-5 fused ring system using the all-orientation search (AOS) routine and a modified cross-validated r2-guided region selection (q2-GRS) routine and its initial application

Article abstract of DOI:10.1016/j.bmc.2009.12.066

We report the development of CoMFA analysis models that correlate the 3D chemical structures of 80 compounds with 6-5 fused ring system synthesized in our laboratory and their inhibitory potencies against tgDHFR and rlDHFR. In addition to conventional CoMFA analysis, we used two routines available in the literature aimed at the optimization of CoMFA: all-orientation search (AOS) and cross-validated r²-guided region selection (q²-GRS) to further optimize the models. During this process, we identified a problem associated with q²-GRS routine and modified using two strategies. Thus, for the inhibitory activity against each enzyme (tgDHFR and rlDHFR), five CoMFA models were developed using the conventional CoMFA, AOS optimized CoMFA, the original q²-GRS optimized CoMFA and the modified q²-GRS optimized CoMFA using the first and the second strategy. In this study, we demonstrate that the modified q²-GRS routines are superior to the original routine. On the basis of the steric contour maps of the models, we designed four new compounds in the 2,4-diamino-5-methyl-6-phenylsulfanyl-substituted pyrrolo[2,3-d]pyrimidine series. As predicted, the new compounds were potent and selective inhibitors of tgDHFR. One of them, 2,4-diamino-5-methyl-6-(2′,6′-dimethylphenylthio)pyrrolo[2,3-d]pyrimidine, is the first 6-5 fused ring system compound with nanomolar tgDHFR inhibitory activity. The HCl salt of this compound was also prepared to increase solubility. Both forms of the drug were tested in vivo in a Toxoplasma gondii infection mouse model. The results indicate that both forms were active with the HCl salt significantly more potent than the free base.

Full text of DOI:10.1016/j.bmc.2009.12.066

Bioorganic & Medicinal Chemistry 18 (2010) 1684–1701
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry
journal homepage: www.elsevier.com/locate/bmc
CoMFA analysis of tgDHFR and rlDHFR based on antifolates with 6–5 fused
ring system using the all-orientation search (AOS) routine and a modified
cross-validated r2-guided region selection (q2-GRS) routine
and its initial application q
a
a
b
Aleem Gangjee a, , Xin Lin , Lisa R. Biondo , Sherry F. Queener
*
a Division of Medicinal Chemistry, Graduate School of Pharmaceutical Sciences, Duquesne University, Pittsburgh, PA 15282, United States
b Department of Pharmacology and Toxicology, School of Medicine, Indiana University, Indianapolis, IN 46202, United States
a r t i c l e i n f o
a b s t r a c t
Article history:
We report the development of CoMFA analysis models that correlate the 3D chemical structures of 80
compounds with 6–5 fused ring system synthesized in our laboratory and their inhibitory potencies
against tgDHFR and rlDHFR. In addition to conventional CoMFA analysis, we used two routines available
in the literature aimed at the optimization of CoMFA: all-orientation search (AOS) and cross-validated r2-
guided region selection (q2-GRS) to further optimize the models. During this process, we identified a
problem associated with q2-GRS routine and modified using two strategies. Thus, for the inhibitory activ-
ity against each enzyme (tgDHFR and rlDHFR), five CoMFA models were developed using the conven-
tional CoMFA, AOS optimized CoMFA, the original q2-GRS optimized CoMFA and the modified q2-GRS
optimized CoMFA using the first and the second strategy. In this study, we demonstrate that the modified
q2-GRS routines are superior to the original routine. On the basis of the steric contour maps of the models,
we designed four new compounds in the 2,4-diamino-5-methyl-6-phenylsulfanyl-substituted pyrrol-
o[2,3-d]pyrimidine series. As predicted, the new compounds were potent and selective inhibitors of
tgDHFR. One of them, 2,4-diamino-5-methyl-6-(20,60-dimethylphenylthio)pyrrolo[2,3-d]pyrimidine, is
the first 6–5 fused ring system compound with nanomolar tgDHFR inhibitory activity. The HCl salt of this
compound was also prepared to increase solubility. Both forms of the drug were tested in vivo in a Tox-
oplasma gondii infection mouse model. The results indicate that both forms were active with the HCl salt
significantly more potent than the free base.
Received 27 August 2009
Revised 18 December 2009
Accepted 28 December 2009
Available online 6 January 2010
Keywords:
Antifolates
Dihydrofolate reductase
CoMFA analysis
Selective inhibitors
Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction
gle agents which not only display high potency but are also
selective against DHFR from P. carinii and/or T. gondii over mam-
Infections caused by opportunistic pathogens Pneumocystis cari-
nii (pc) and Toxoplasma gondii (tg) are the leading cause of morbid-
ity and mortality in immunocompromised patients such as those
with AIDS.2 Dihydrofolate reductase (DHFR) inhibitors are the cur-
rent drugs of choice for the treatment of these infections. Ideally,
these drugs should efficiently inhibit the growth of pathogenic
cells via DHFR inhibition without affecting the mammalian DHFR.
Unfortunately, due to their lack of potency and/or selectivity, com-
binations of current DHFR inhibitors with other agents such as sul-
fa drugs are often required for synergistic effects or to decrease
host toxicity, which leads to high costs. Discontinuation of therapy
is necessary in many cases as a result of severe side effects. There-
fore, efforts continue to be directed toward the development of sin-
malian DHFR, such as rat liver (rl) DHFR. In recent reports, DHFR
inhibitors with 6–5 fused ring system including furo[2,3-
a
d]pyrimidine, pyrro[2,3-d]pyrimidine and purine derivatives have
shown good activity and selectivity against these pathogenic
organisms, especially T. gondii.3–16 Computational techniques such
as QSAR models can be used to assist the rational design of more
potent and selective DHFR inhibitors with 6–5 fused ring system.
Most models for predicting DHFR inhibition in the literature17–47
published to-date use homologous data sets of DHFR inhibitors with
a specific heterocyclic core (e.g., quinazolines, pyrimidines). Matti-
oni and Jurs48 recently developed QSAR models that correlated
chemical structure and inhibition potency for three types of DHFR:
rlDHFR, pcDHFR, and tgDHFR. The results, however, did not give
structural information about the binding sites.
More recently, Sutherland and Weaverb49 generated compara-
tive molecular similarity indices analysis (CoMSIA) three-dimen-
sional quantitative structure–activity relationship (3D-QSAR)
q
See Ref. 1.
* Corresponding author. Tel.: +1 412 396 6070; fax: +1 412 396 5593.
E-mail address: gangjee@duq.edu (A. Gangjee).
0968-0896/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmc.2009.12.066

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1685
Table 1
modelsfortheinhibitoryactivitiesagainstpcDHFRandrlDHFRusing
Structures and IC50 values (lM) of the compounds used in developing the models
a data set of 406 structurally diverse DHFR inhibitors. Gangjee and
Lin50 also reported comparative molecular field analysis (CoMFA)
and CoMSIA analysis of pcDHFR, tgDHFR and rlDHFR based on the
biological data of 179 compounds synthesized in our laboratory.50
These are general models for pcDHFR, tgDHFRand rlDHFRinhibitors.
Even though some compounds with 6–5 fused ring system were in-
cluded in the training set and/or test set, the majority of the drugs
used in the development of these models were compounds with
6–6 fused ring system, such as quinazolines and pyridopyrimidines.
In addition, certain monocyclic, tricyclic and tetracyclic compounds
were also included in the data sets.
H
N
NH2
N
R
N
O
H2N
Compd
R
tgDHFR
rlDHFR
Ref.
1
2
3,4,5-Cl3
3,4-Cl2
>3.9
89.3
25.6
35.2
3
3
As a continuation of our previous work and the importance of
both potency and selectivity associated with the 6–5 fused ring sys-
tems, we report the development of CoMFA analysis models that
correlate the 3D chemical structures of 80 compounds3,4,6–9,12,14,15
with 6–5 fused ring system synthesized in our laboratory and their
inhibitory potencies for tgDHFR and rlDHFR. In addition to conven-
tional CoMFA analysis, we used two routines available in the litera-
ture aimed at the optimization of CoMFA: all-orientation search
(AOS)51 and cross-validated r2-guided region selection (q2-GRS)52
to further optimize the models. During this process, we identified
a problem associated with q2-GRS routine and modified it with
two different strategies. Thus, for the inhibitory activity against
each enzyme (tgDHFR and rlDHFR), five CoMFA models were
developed using the conventional CoMFA, AOS optimized CoMFA,
the original q2-GRS optimized CoMFA and the modified q2-GRS
optimized CoMFA with strategy one and two.
H
NH2
N
N
R
N
N
H
H2N
Compd
R
tgDHFR
rlDHFR
Ref.
3
4
5
6
7
8
9
10
3,4,5-(OCH3)3
3,4-(OCH3)2
4-OCH3
2,5-(OCH3)2
2,5-(OC2H5)2
3,4-Cl2
8.1
4.3
6.0
1.7
5.3
1.4
1.1
3.9
56.3
116.0
63.0
156.0
70.0
14.4
59.3
>252
4
4
4
4
4
4
4
4
2,3-(CH)4
H
NH2
N
X
R
2. Computational details
N
2.1. Data set and biology activity
N
H2N
H
The structures of the compounds used to develop the CoMFA
modelalongwiththeirDHFRinhibitoryactivitiesarelistedinTable1.
Compd
X
R
tgDHFR
rlDHFR
Ref.
11
12
13
14
15
16
17
N(CH3)
N(CH3)
N(CH3)
2,5-(OCH3)2
3,4-Cl2
2,3-(CH)4
3,4-(OCH3)2
3,4-Cl2
3.40
1.00
0.87
2.60
11.6
0.81
9.20
>12.0
3.00
8.20
16.7
5.3
6
6
6
6
6
6
6
2.2. Structure alignment
S
S
S
S
The pcDHFR bound structures of 81 (PDB 1daj)53 (Fig. 1) was
used as the template. For each class of compounds only the struc-
ture with the plain phenyl side chain was flexibly fit to the tem-
plate in MMFF94 force field using the Flexible Align module
implemented in MOE 2004.0354 with the default options followed
by the addition of the substituents on the phenyl in a consistent
fashion, with preference for the solvent-exposed edge of the phe-
nyl ring: the 20-position was occupied before the 60-position and
the 30-position before the 50-position, with the ortho position tak-
ing precedence over the meta position when both are substituted.
The newly added substituents were then relaxed in MOE with the
rest of the molecule being fixed using MMFF94 force field with de-
fault option. The resulting structures were exported as MOL2 files
and were subsequently imported into SYBYL 7.0,55 followed by the
calculation of Gasteiger–Hückel charges. The hydrogen atoms of
each molecule were further optimized, while the heavy atoms
were held still, with a gradient convergence of 0.05 kcal/Å and up
to 1000 iterations using the Powell method in the TRIPOS force
field. The final compound aggregate is shown in Figure 2.
2,3-(CH)4
3,4-(CH)4
3.0
82.9
NH2
N
Ar
n
N
N
H
H2N
N
Compd
n
Ar
tgDHFR
rlDHFR
Ref.
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
C6H5
28
59.9
13
3.7
1.88
7
3.5
32.4
52.3
0.9
252
15.3
>25
30.1
280
29.8
1.6
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
3,4,5-(OCH3)3C6H2
2,3,4-(OCH3)3C6H2
2,4,6-(OCH3)3C6H2
2,4,5-(OCH3)3C6H2
2,5-(OCH3)2C6H3
3,5-(OCH3)2C6H3
3,4-(OCH3)2C6H3
2,4-(OCH3)2C6H3
3,4-Cl2C6H3
2.2
1.8
0.84
1.5
1.7
3.0
14.2
2.4
6.7
1
16.8
32.9
27
30.5
0.5
49
3,5-(OCH3)2-4-OHC6H2
2-NO2-3,4-(OCH3)2C6H2
4-C6H5C6H4
2.3. Training set and test set
2-Naphthyl
9-Fluorenyl
C6H5
Among the 80 compounds listed in Table 1, and 76 compounds
were active (had detectable IC50 values) against tgDHFR and 71 com-
pounds were active against rlDHFR. A dataset containing only the
NH-2,5-OCH3C6H3
S-2-Naphtyl
>29
107
45
(continued on next page)

1686
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
Table 1 (continued)
Table 1 (continued)
R
NH2
N
XAr
NH2
N
N
N
O
H2N
N
H
H2N
Compd
R
tgDHFR
rlDHFR
Ref.
Compd
X
Ar
C6H5
tgDHFR
rlDHFR
Ref.
74
75
76
77
78
79
80
2,5-Cl2
1.24
4.05
1.03
1.6
1.3
1.01
2.7
4.8
11.8
4.8
0.66
1.5
1.09
8.6
15
15
15
15
15
15
15
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
S
S
S
NH
NH
O
NH
NH
N(CH3)
NH
N(CH3)
NH
NH
NH
NH
N(CH3)
N(CH3)
S
>26
19
11.6
37
252
23
12.3
12
8
8
8
8
8
8
8
8
8
8
8
9
9
9
9
9
9
9
9
9
9
9
9
2,5-(OCH3)2
3,4,5-(OCH3)3
4-Benzyloxy
2,3-(CH)4
3,4-(CH)4
H
1-Naphthyl
2-Naphthyl
1-Naphthyl
2-Naphthyl
2-Naphthyl
4-C6H5OC6H4
2-C6H5C6H4
2-Naphthyl
2,5-Cl2C6H3
3,4,5-Cl3C6H2
1-Anthracene
2-Fluorene
38
36.5
60.3
16.2
137
14.6
71.9
34.3
>51
>10
12.6
>13
241
24.7
49
>42
32.4
45.4
23.6
>47
21.5
15.4
27.7
4.5
NH2
N
N
3-(N-Ethylcarbazole)
O
2-(9-Hydroxyfluorene)
3-(2-Methoxydibenzofuran)
3-(N-Ethylcarbazole)
2-Biphenyl
63
O
N
H2N
392
22.6
105
23
79
38
HN
COOH
S
S
S
S
3-Biphenyl
4-Biphenyl
2-C6H5OC6H5
3-C6H5OC6H5
31
351
18
39
65
HOOC
19
259
S
4-C6H5OC6H5
81
NH2
N
S
N
R
H2N
N
H
Compd
R
tgDHFR
rlDHFR
Ref.
59
60
61
62
63
64
65
66
2,3-(CH)4
3-Cl
4-Cl
2-OCH3
3-OCH3
4-OCH3
2,4-Cl2
2,5-(OCH3)2
0.16
6.94
8.93
1.02
4.76
2.69
20
4.57
47.7
19
2.32
2.06
31.3
110
7.8
12
12
12
12
12
12
12
12
Figure 1. The structure and pcDHFR bound conformation of 81.
0.17
active compounds against a particular enzyme were used to develop
the CoMFA model. A diversity subset56 which contained approxi-
mately 20% of the compounds (14 for rlDHFR and 15 for tgDHFR)
with the farthest distance from the least active compound based
on the 2-D fingerprints (MACCS Structural Keys bit packed version)
using Tanimoto coefficient56 as the similarity metric were extracted
using MOE and was used as the test set. The remaining compounds
were used as the training set. This was carried out to make sure that
the test set is diverse enough to justify the models. Instead of the
least active compound, any compound within the dataset could have
be chosen to calculate the diversity subset, however, we used the
least active compound here in all cases just to be consistent.
R
NH2
N
S
N
N
H
H2N
Compd
R
tgDHFR
rlDHFR
Ref.
67
68
69
70
71
72
73
2,3-(CH)4
4-Cl
3,4-(CH)4
3-Cl
2,4-Cl2
2,5-(OCH3)2
H
18.8
31.8
31
76.8
28.2
87
40
37.4
31
14
14
14
14
14
14
14
2.4. Conventional CoMFA
64
CoMFA was performed using the QSAR module in SYBYL 7.0. For
each training set compound, the CoMFA descriptors, steric (Len-
nard–Jones 6–12 potential) and electrostatic (Coulombic potential)
34.6
>46
>33
81

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1687
sure of the predictivity, fluctuates with the orientation of the
aligned molecular aggregate on the computer screen by up to 0.5
q2 unit. The reason for this fluctuation in q2 values lies in the fact
that conventional CoMFA samples the continuous molecular field
at discrete lattice points and calculates the steric and electrostatic
field energies at each lattice point with distance-sensitive func-
tions, such as the Lennard–Jones 6–12 potential. When the molec-
ular aggregate rotates, so does the molecular field surrounding the
aggregate. The lattice box in CoMFA, however, is always axis-
aligned and does not rotate along with the field. Thus, different
points in the same molecular field are mapped onto the lattice
points resulting in different field energy values. These values, when
processed subsequently by partial least squares (PLS) to produce
the final model causes a variation in the q2 value and, conse-
quently, the predictivity of the model.
The AOS routine51 optimizes the field sampling by rotating the
molecular aggregate systematically and picking the orientation
that produces the highest q2 value. The details of the AOS routine
were described previously elsewhere.51 Briefly, the compounds
aggregate was rotated about the x, y, and z axes systematically
with an increment of 30° using the STATIC ROTATE command in
SYBYL. For each orientation, a conventional CoMFA was performed
as described above and the predictive value of the model was eval-
uated using leave-one-out (LOO) cross-validation with sample-dis-
tance partial least squares (SAMPLS). The orientation that gave the
highest q2 value was selected to produce the final model. A SYBYL
Programming Language (SPL) script was written to perform the
AOS routine as described51 automatically. For the inhibitory activ-
ity against each enzyme, a new model was obtained by optimizing
the model generated with the conventional CoMFA using the AOS
routine.
2.4.2. The original q2-GRS CoMFA
The q2-GRS routine was developed by Cho and Tropsha52 to
overcome the deficiency of CoMFA analysis mentioned above.
The q2-GRS routine first subdivides the rectangular box obtained
initially with conventional CoMFA, which, as mentioned above, ex-
tends beyond the van der Waals envelopes of all molecules by 4.0 Å
along the principal axes of the Cartesian coordinate system, into
125 small boxes and perform 125 independent analyses using
20
15
10
5
Figure 2. The compound aggregate. (A) Front view; (B) top view.
field energies were calculatedusing the SYBYL default parameters. The
CoMFAregionwasdefinedtoextendbeyondthevander Waalsenve-
lopes of all molecules by 4.0 Å along the principal axes of the Carte-
sian coordinate system. A distance dependent dielectric constant
wasused. Ansp3 carbon atomwith+1.0 chargewasusedastheprobe
atom to calculate steric and electrostatic fields. The steric and elec-
trostatic contributions were truncated at 30 kcal/mol, and electro-
static contributions were dropped at lattice intersections with
maximum steric interactions. The CoMFA steric and electrostatic
fields generated were scaled by the CoMFA standard option in SYBYL
.
0
In order to improve the cross-validated q2 to an acceptable level, a
0
5
10
15
20
small number of outliers were dropped from the training set.
X
Figure 3. Front view of the q2-GRS grids (black), apparently, these grids do not
coincide with that of the conventional CoMFA, and the distances between the grid
points at the opposite side of the border (purple) of two adjacent subregions can be
much less then 1.0 Å.
2.4.1. All-orientation search (AOS) CoMFA
As first reported by Cho and Tropsha52 the cross-validated r2
(q2) value of CoMFA analysis, which serves as a quantitative mea-

1688
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
probe atoms placed within each box with the step size of 1.0 Å. Fol-
lowed by the selection of only those small boxes for which a q2 is
higher than a specified optimal cutoff value (default 0.1). Finally,
CoMFA analysis is repeated with the combined region of small
boxes selected at the previous step to generate the final model.
Cho and Tropsha52 argued that it is contradictory for the con-
ventional CoMFA routine to emphasize the limited areas of
three-dimensional space as important for biological activity in
the final result even though it assumes equal importance of all lat-
tice points during the PLS analysis, which happens to be the calcu-
lation that leads to the final result. Thus q2-GRS was designed to
eliminate those areas of three-dimensional space where changes
in steric and electronic fields do not correlate with changes in bio-
logical activity from the analyses. For the inhibitory activity against
each enzyme, a new model was obtained by optimizing the model
generated with the conventional CoMFA using the original q2-GRS
routine.
20
15
10
5
0
2.4.3. The modified q2-GRS CoMFA
0
5
10
15
20
A close inspection of the final combined region file generated by
the original q2-GRS routine revealed a potential problem. For the
training set of tgDHFR, the region box generated by conventional
CoMFA has a dimension of 19.95 Å ꢀ 20.49 Å ꢀ 18.66 Å. If this re-
gion were to be defined using 1 Å as step size, there would be
20 ꢀ 21 ꢀ 19 = 7980 lattice points. In the final combined region
generated by q2-GRS using the default parameters, 74 boxes or
X
Figure 5. Front view of the modified q2-GRS grids using the first strategy (blue),
which coincide with that of the conventional CoMFA (green).
subregions (59.2%) of the total 125 were selected and the rest 51
boxes (40.8%) were eliminated. Each of the subregions had the
Figure 4. A comparison of the contour maps of the tgDHFR training set generated with the conventional CoMFA (left) and with the original q2-GRS with the cutoff set at
ꢁ1000.0 (so that all subregions are selected) (right). The areas of the q2-GRS contour maps where the grid points are very close to each other (these areas are not present in
the conventional CoMFA maps) are highlighted with the white symbols. (A) The steric contour maps, (B) the electrostatic contour maps.

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1689
Figure 6. A comparison of the contour maps of the tgDHFR training set generated with the conventional CoMFA (left) and with the first strategy modified original q2-GRS
with the cutoff set at ꢁ1000.0 (so that all subregions are selected) (right). Superimposed on their respective region box(es) (cyan). (A) The steric contour maps, (B) the
electrostatic contour maps.
dimension of one fifth of the region box generated by conventional
CoMFA. In this case, it is 3.99 Å ꢀ 4.10 Å ꢀ 3.73 Å. Thus, each box
20
contained 5 ꢀ 5 ꢀ 4 = 100 grid points. Thus the total number of
grid points in the combined region is 74 ꢀ 100 = 7400 grid points,
which is 96.5% of the total 7980 lattice points found in the conven-
tional CoMFA region box. Apparently, the original q2-GRS routine
15
forced 92.7% of the lattice points into 59.2% of the space. In the
conventional CoMFA region, the distance between any given pair
of immediately adjacent grid points is 1 Å. In the combined region
generated by the original routine of q2-GRS, in order to fit more
10
grid points into the space, the distance between some of the lattice
points must be less than 1 Å. Since within each subregion, the dis-
tance of any given pair of immediately adjacent to each other is
also 1 Å, the shorter distances could only occur between grid
points from adjacent boxes.
5
0
To understand this behavior of the original q2-GRS routine, let
us consider a putative cubic region box, whose lowest corner is
at (0, 0, 0) in the Cartesian space, with the dimension of
21.5 Å ꢀ 21.5 Å ꢀ 21.5 Å. From our observation, CoMFA normally
rounds the dimensions down to the nearest integers. If the decimal
part of a dimension is equal to or greater than 0.99, however, CoM-
FA treats this dimension as the rounded up integer, for example,
0
5
10
15
20
X
Figure 7. Front view of the modified q2-GRS subregions (purple) and grids (green)
using the second strategy.

1690
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
21.99 will be rounded up to 22 whereas 21.989 will still be
rounded down to 21. Thus, the dimension of the conventional
CoMFA grid in our example would be 21 Å ꢀ 21 Å ꢀ 21 Å and the
number of grid points would be 223 = 10648.
dimensional space near the borders of the adjacent subregions by
putting more grid points at these sites.
To further illustrate the shortcoming of the original q2-GRS rou-
tine, we first conducted a conventional CoMFA analysis of the
tgDHFR training set. For this analysis, the step size of the region file
was set to 1.0 Å instead of the default 2.0 Å. The q2 of this analysis
was 0.627. Next, we did the analysis using the original q2-GRS rou-
tine, with the cutoff set to ꢁ1000.0 instead of the default 0.1, so
that all of the 125 regions were selected in the final region. At only
6.06, the q2 of this analysis was significantly lower than that of the
conventional CoMFA analysis. The steric and electrostatic
stdev * coefficient contour maps derived from these two analyses
were compared in Figure 4. There were areas in the q2-GRS contour
maps where the grid points were in close vicinity to each other.
To overcome this problem, we modified the original q2-GRS
routine with two different strategies.
The q2-GRS routine divides the region box into 125 subregions.
As in our example, the size of each subregion box is 4.3 Å ꢀ
4.3 Å ꢀ 4.3 Å. CoMFA treats each individual subregion the same
way as the conventional region box. In our example, the size of
the grid within each subregion is rounded down to 4 Å ꢀ 4 Å ꢀ
4 Å. The number of grid points within each subregion is 53 = 125.
Assuming that all the subregions are selected in the final combined
region, there would be a total of 1252 = 15625 grid points. That is
47% more than the 10648 grid points in the conventional CoMFA
grid. As shown in Figure 3, the closest distance occurs between
the grid points at the opposite side of the border of two adjacent
subregions. This distance is the decimal part of the dimension of
the subregion, as in our example, it is 0.3 Å. Apparently, in the
worst case scenario, this distance could be 0 Å, and in this extreme
scenario some of the grid points would be counted twice, thrice or
more (up to eight times).
In the first strategy, the lowest corner of each modified subre-
gion is the lowest grid point of the conventional CoMFA grid en-
closed by the original q2-GRS subregion, and the highest corner
of each modified subregion is the highest grid point of the conven-
tional CoMFA grid enclosed within the subregion of the original q2-
GRS routine (Fig. 5). The advantage of this strategy is that only the
grid points of the conventional CoMFA grid would be included in
the final combined region.
Thus, in the final combined region, this uneven distribution of
grid points will always be present, unless in the unlikely scenario
where none of the selected subregions are adjacent to each other.
The original q2-GRS introduces unnecessary weight on the three-
Figure 8. Comparison of the contour maps of the tgDHFR training set generated with the conventional CoMFA (left) and with the second strategy modified original q2-GRS
with the cutoff set at ꢁ1000.0 (so that all subregions are selected) (right). (A) the steric contour maps, (B) the electrostatic contour maps.

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1691
Where Y0pred is the predicted value of the target property (pIC50
using the final model. The statistical F-test was also performed
and the results indicated that all models were statistically
significant.
CoMFA coefficient maps were generated by interpolation of the
pairwise products between the PLS coefficients and the standard
deviations of the corresponding CoMFA descriptor values.
)
We tested this modified routine on the tgDHFR training set by
lowering the q2 cutoff to ꢁ1000.0 so that all the 125 subregions
were included in the final combined region, which gave exactly
the same q2 (0.627) as that of the conventional CoMFA (with
1.0 Å as the step size instead of the default 2.0 Å). As shown in Fig-
ure 6, except for those gaps between subregions, the stdev * coeffi-
cient contour maps derived from this analysis were exactly the
same as that of the original CoMFA analysis.
The drawback of this strategy is that the size of the subregions
varies. In our putative example illustrated in Figure 5, the number
of grid points in each subregion varies from 64 to 125, making the
subregions less comparable to each other. Therefore, a modification
using a second strategy was devised. In this second strategy, the re-
gion box is divided into 125 equal-sized subregions and the distance
between the adjacent subregions is 1 Å (Fig. 7), whereas in the origi-
nal routine, the adjacent subregions touch each other. In this case,
eventhough thegrid pointsdo notmatchthatof theconventionalre-
gion, the size of the subregions and the number of grid points within
these subregions are the same, much like the original routine.
This second modified routine was also tested on the tgDHFR
training set by lowering the q2 cutoff to ꢁ1000.0, which gave a
q2 of 0.631; this was slightly higher than that of the conventional
CoMFA using step size 1.0 Å (0.627). The stdev * coefficient contour
maps from this analysis was compared with the conventional
CoMFA in Figure 8.
3. Results and validation
The tgDHFR training set has a total of 61 compounds, which in-
clude 2–12, 14–27, 30, 31, 33, 37–39, 42–44, 46–50, 53–55, 57–
64, 67–73, 76, 78–80 (Table 1). The tgDHFR test set contains 15
compounds, which are 13, 28, 29, 32, 34, 35, 40, 51, 52, 56, 65,
66, 74, 75 and 77 (Table 1). A total of five models were generated
using the conventional CoMFA, AOS optimized CoMFA, the original
q2-GRS optimized CoMFA and the modified q2-GRS optimized
CoMFA using the first and the second strategy. The key statistical
parameters associated with these models are shown in Table 2.
The graph the actual pIC50 versus the predicted pIC50 values for
the training set and the test set by the five CoMFA models are
shown in Figure 9A–E, respectively. The predicted pIC50 values of
the training set compounds using the five models as well as the
residual values are given in Appendix Tables 1, 3, 5, 7 and 9,
respectively. The predicted pIC50 values of the test set compounds
and the residual values using the five models are given in Appendix
Tables 2, 4, 6, 8 and 10, respectively.
For the inhibitory activity against each enzyme, models were
obtained by using both modified q2-GRS routine.
Though, AOS CoMFA gave the best internal cross-validated q2
value 0.699 (ONC = 7) among the five models, this q2 value was
only obtained with the maximum allowed seven components. Fur-
thermore, the AOS CoMFA model only gave the lowest external
predictive r2 (0.365), which was worse than that of the conven-
tional CoMFA (0.422). In contrast, the original q2-GRS CoMFA mod-
el showed the lowest internal cross-validated q2 value 0.622
(ONC = 3), lower than that of the conventional CoMFA model
(0.639, ONC = 5), it however had the second best external predic-
tive r2 (0.465), whereas the modified q2-GRS model using the first
strategy had a q2 value of 0.639 (ONC = 3), which was the same as
that of the conventional CoMFA, and its external predictive r2 value
(0.447) was only slightly better than that of the conventional
model.
2.5. PLS analysis
The CoMFA, descriptors derived above were used as explanatory
variables, andpIC50 (ꢁlog IC50)valueswereusedasthetargetvariable
in PLS regression analyses to derive 3D QSAR models using the imple-
mentation in the SYBYL package. The predictive value of the models
was evaluated byleave-one-out (LOO) cross-validation with SAMPLS.
The cross-validated coefficient, q2, was calculated using Eq. 1,
P
2
ðYpred ꢁ Yactual
Þ
q2 ¼
ð1Þ
P
2
ðYactual ꢁ Ymean
Þ
Where Ypred, Yactual, and Ymean are predicted, actual, and mean val-
P
2
ues of the target property (pIC50), respectively.
ðYpred ꢁ YactualÞ
Finally, the q2-GRS model modified with the second strategy
afforded the best external predictive r2 (0.499) and also gave a sat-
isfactory internal cross-validated q2 at 0.647 (ONC = 3), which was
the second best among the five models only after the AOS CoMFA.
The stdev * coefficient contour maps generated using this model is
further discussed in the next section.
The rlDHFR training set has a total of 57 compounds, which in-
clude 1–6, 8, 13–27, 30, 31, 33, 36–41, 43, 44, 52–54, 56–63, 66–
71, 74–76, 78–80 (Table 1). The rlDHFR test set contains 14 com-
pounds, which are 7, 9, 12, 28, 32, 35, 42, 45, 49, 51, 55, 64, 65
and 77 (Table 1). A total of five models were generated using the
conventional CoMFA, AOS optimized CoMFA, the original q2-GRS
optimized CoMFA and the modified q2-GRS optimized CoMFA using
the first and the second strategy. The key statistical parameters
is the predictive sum of squares (PRESS). The number of compo-
nents giving the lowest PRESS value or the optimal number of com-
ponents (ONC) was used to generate the final PLS regression
models. To prevent overfitting, no more than 7 components were al-
lowed. The conventional correlation coefficient r2 and its standard
error, s, were subsequently computed for the final PLS models using
the following formulae, respectively.
P
2
ðY0pred ꢁ Yactual
ðYactual ꢁ Ymean
Þ
r2
¼
ð2Þ
P
2
Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
ðY0pred ꢁ Yactual
S ¼
ð3Þ
N
Table 2
Statistical data for tgDHFR CoMFA results
Conventional
AOS
q2-GRS (original)
q2-GRS (modified 1)
q2-GRS (modified 2)
CV-r2 (q2)
Opt. no. of comp.
Std. error of estimate
Non-CV-r2
0.639
5
0.236
0.897
95.613
0.422
0.699
7
0.141
0.964
204.900
0.365
0.622
3
0.322
0.800
76.108
0.465
0.639
3
0.291
0.837
97.300
0.447
0.647
3
0.311
0.814
83.198
0.499
F value
Predictive r2

1692
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
(A) Conventional CoMFA for tgDHFR
inhibition
(C) Original q2-GRS CoMFA for tgDHFR
inhibition
7
6.5
6
7
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(D) q2-GRS CoMFA modified using the first
strategy for tgDHFR inhibition
(B) AOS CoMFA for rlDHFR inhibition
7
6.5
6
7
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(E) q2-GRS CoMFA modified using the
second strategy for tgDHFR inhibition
7
6.5
6
5.5
5
4.5
4
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Figure 9. Predictions for the training ( ) and test ( ) sets for tgDHFR inhibitory activities. The solid line is the regression line for the training set predictions whereas the
dotted lines indicate the 1.0 log margins.
Table 3
Statistical data for rlDHFR CoMFA (value from original training set/value from the training set after dropping three outliers)
Conventional
AOS
q2-GRS (original)
q2-GRS (modified 1)
q2-GRS (modified 2)
CV-r2 (q2)
Opt. no. of comp.
Std. error of estimate
Non-CV-r2
0.367/0.465
3/3
0.348/0.317
0.703/0.715
41.822/41.812
ꢁ0.025/0.267
0.478/0.536
3/3
0.338/0.294
0.720/0.755
45.837/51.373
0.163/0.281
0.375/0.452
3/3
0.380/0.334
0.646/0.682
32.246/35.810
0.171/0.296
0.415/0.468
3/3
0.366/0.329
0.672/0.692
36.150/37.454
0.233/0.326
0.419/0.431
3/3
0.367/0.339
0.671/0.673
36.003/34.252
0.188/0.320
F value
Predictive r2

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1693
associated with these models are shown in Table 3. The graph of
the actual pIC50 versus the predicted pIC50 versus the predicted
pIC50 for the training set and test set by the five CoMFA models
are shown in Figure 10A–E, respectively. The predicted pIC50 values
of the training set compounds using the five models as well as the
residual values are given in Tables 11, 13, 15, 17 and 19 in Appen-
dix, respectively. The predicted pIC50 values of the test set com-
pounds and the residual values using the five models are given
in Tables 12, 14, 16, 18 and 20 in Appendix, respectively.
both modified q2-GRS routines gave better q2 values than that of
the original routine. AOS CoMFA continued to give the highest q2
values among the five models and the first strategy modified q2-
GRS model afforded the best predictive r2.
In order to obtain a reasonable CoMFA model for rlDHFR, three
outliers 21, 26 and 33, which had an absolute residual value great-
er than 1.0 during the cross-validation of the conventional CoMFA
model, were dropped from the previous rlDHFR training set, while
the test set compounds remained the same. A total of five models
were again generated using the conventional CoMFA, AOS opti-
mized CoMFA, the original q2-GRS optimized CoMFA and the mod-
ified q2-GRS optimized CoMFA using the first and the second
A q2 P 0.5 is generally considered as an indication that the
model is internally predictive. In this case, however, none of the
models afforded a q2 greater than 0.5. However, it was evident that
(A) Conventional CoMFA for rlDHFR
(C) Original q2-GRS CoMFA for rlDHFR
inhibition
inhibition
7
7
6.5
6
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(B) AOS CoMFA for rlDHFR inhibition
(D) q2-GRS CoMFA modified using the first
strategy for rlDHFR inhibition
7
6.5
6
7
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(E) q2-GRS CoMFA modified using the
second strategy for rlDHFR inhibition
7
6.5
6
5.5
5
4.5
4
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Figure 10. Predictions for the training ( ) and test ( ) sets for rlDHFR inhibitory activities. The solid line is the regression line for the training set predictions whereas the
dotted lines indicate the 1.0 log margins.

1694
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
strategy. The key statistical parameters associated with these mod-
els are shown in Table 3. The graph of the actual pIC50 versus the
predicted pIC50 values for the training set and test set by the five
CoMFA models are shown in Figure 11A–E, respectively. The pre-
dicted pIC50 values of the training set compounds using the five
models as well as the residual values are given in Tables 21, 23,
25, 27 and 29 in Appendix, respectively. The predicted pIC50 values
of the test set compounds and the residual values using the five
models are given in Tables 22, 24, 26, 28 and 30 in Appendix,
respectively.
the first strategy gave the second best q2 value (0.468,
ONC = 3) and the best predictive r2 (0.326). The stdev * coeffi-
cient contour maps of the AOS CoMFA is further discussed in
the following section.
4. The stdev * coefficient contour maps
4.1. tgDHFR
The second strategy modified q2-GRS CoMFA model was used to
construct the stdev * coefficient contour maps for the tgDHFR
(Fig. 12).
In this case, only AOS CoMFA model gave a q2 value greater
than 0.5 (0.536, ONC = 3). The modified q2-GRS model using
(C) Original q2-GRS CoMFA for rlDHFR
(A) Conventional CoMFA for rlDHFR
inhibition after dropping three outliers from
inhibition after dropping three outliers from
the training set
the training set
7
7
6.5
6
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(B) AOS CoMFA for rlDHFR inhibition after
dropping three outliers from the training set
(D) q2-GRS CoMFA modified using the first
strategy for rlDHFR inhibition after dropping
three outliers from the training set
7
6.5
6
7
6.5
6
5.5
5
5.5
5
4.5
4
4.5
4
3.5
3
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Actual
(E) q2-GRS CoMFA modified using the
second strategy for rlDHFR inhibition after
dropping three outliers from the training set
7
6.5
6
5.5
5
4.5
4
3.5
3
3
3.5
4
4.5
5
5.5
6
6.5
7
Actual
Figure 11. Predictions for the training ( ) and test ( ) sets for rlDHFR inhibitory activities (three outliers were dropped from the training set). The solid line is the regression
line for the training set predictions whereas the dotted lines indicate the 1.0 log margins.

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1695
In the CoMFA steric field (Fig. 12A), the green (sterically favor-
able) and yellow (sterically unfavorable) contours represent 80%
and 20% level contributions, respectively.
and 30-substituents are surrounded by a number of small green
regions. The 6-S bridge is also close to a yellow region. There
are two green regions also near the 50 and 60-substituent posi-
tion. Some small green regions are found near the 5-methyl
group.
For the purine analogs (compounds 18–35, Table 1), green re-
gions are found near the 20-, 30- and 50-substituents and yellow re-
gions near 40- and 60-substituents.
Some yellow regions are found near the 7-position of the ring
systems indicating that the furo[2,3-d]pyrimidine is more steri-
cally conducive to the activity than pyrrolo[2,3-d]pyrimidines,
whose 7-NH is slightly bulkier than the oxygen. The ring of the
purine compounds is slightly tilted as compared to the other ana-
logs, positioning it further away from these yellow regions, indicat-
ing that the purine ring is probably better than the other two ring
systems in terms of steric effect for the inhibitory activity against
tgDHFR.
The CoMFA electrostatic contour map for tgDHFR inhibitory
activity is depicted in Figure 12B. The red (negative charge favor-
able) and blue (negative charge unfavorable) contours in the CoM-
FA electrostatic field represent 80% and 20% level contributions,
respectively. The molecules are enveloped by blue regions. Some
patches of red regions are also present.
The 20-substituent (on the phenyl ring) of 5-substituted pyrrol-
o[2,3-d]pyrimidines and furo[2,3-d]pyrimidines such as com-
pounds 6, 11, 45, 72 and 74 (Table 1) lies between a yellow
region and a green region. A cluster of several small green regions
which includes the aforementioned green region is found near the
30-substituent of 5-substituted pyrrolo[2,3-d]pyrimidines and
furo[2,3-d]pyrimidines such as compounds 2, 4, 14, 46, 70 and
76 (Table 1). Yellow regions are found near 40-substituent and 50-
substituent of the same classes of compounds. The 1-naphthyl
moiety, attached to the 5-position of a pyrro[2,3-d]pyrimidine or
furo[2,3-d]pyrimidine including compounds 9, 13, 16, 37, 39, 67
and 78 has more contact with green regions than yellow regions.
However, the reversed can be said for the 2-naphthyl moiety (com-
pounds 17, 31, 38, 40, 44 and 79). This is evident because com-
pounds 16, 37, 39, 78 that contain the 1-naphthyl moiety all
have greater predicted activities than their 2-naphthyl analogs
17, 38, 40 and 79, respectively. A small green region is also found
near the bridge atom of all 5-substituted compounds that is di-
rectly attached to the 5-position of pyrrolo[2,3-d]pyrimidines or
furo[2,3-d]pyrimidines. A green region is also located near the 9-
methyl group of compounds 11, 12, 13, 44, 46, 51 and 52 (Table
1). The 6-methyl group of compounds 67–73 (Table 1) lies within
a yellow region.
Large blue regions are found on top of the bridge atoms. Blue re-
gions are also found near the amino protons. Red regions are present
in the vicinity of the 7-NH and 7-O of the pyrrolo[2,3-d]pyrimidine
and furo[2,3-d]pyrimidines, respectively. The 6-methyl group of
compounds 67–73 (Table 1) lies near a red region.
For the 5-methyl-6-phenylsulfanyl-substituted pyrro[2,3-
d]pyrimidines including compounds 59–66 (Table 1), both 20-
Figure 12. (A) Orthogonal view of the steric fields generated with the second strategy modified q2-GRS CoMFA model of tgDHFR: yellow indicates regions where bulky
groups decrease activity, whereas green indicates regions where bulky groups increase activity. (B) Orthogonal view of the electrostatic fields generated with the second
strategy modified q2-GRS CoMFA model of tgDHFR: red indicates regions where more negatively charged groups increase activity, whereas blue indicates regions where more
positively charged groups increase activity.

1696
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
For the purine analogs (compounds 18–35, Table 1), two red
gions, but 2-naphthyl seems to have more contact with the green
region. A green region is found near the 9-methyl group of com-
pounds 11, 12, 13, 44, 46, 51 and 52 (Table 1).
regions and one blue region are found near the 20-methoxy oxygen
(e.g., compound 21, Table 1) and one red region near the 20-methoxy
methyl. Afew blue regionsare close tothe 30-methoxymethylgroup.
Some blue regions are found near the 40-methoxy methyl of the pur-
ine analogs. One blue region is found near 50-methoxy methyl of the
purine analogs.
The 20-substituents of the purines lie in the green region, part of
30-substituent lies inside the green region, but it is also in contact
with the yellow region. The 40-OCH3 is found inside the yellow re-
gion and the 50-OCH3 hardly touches that yellow region. The 2-
biphenyl contacts some small yellow regions, however, 3-biphenyl
is in the previously mentioned green area. The 2-naphthyl of the
purine analogs lies in a yellow region.
For the 5-methyl-6-phenylsulfanyl-substituted pyrro[2,3-d]-
pyrimidines including compounds 59–66 (Table 1), a red region is
near the 20-methoxy oxygen. The 30-, 40-, 50- and 60-substituents
are surrounded by blue regions. A blue region is located near the
5-methylgroup. Some red regionsarefound nearthe6-sulfurbridge.
For the 5-methyl-6-substituted pyrrolo[2,3-d]pyrimidines, the
1-naphthyl is in contact with the green region. 20-OCH3 and 30-
OCH3 lie within the green region.
4.2. rlDHFR
The CoMFA electrostatic contour map for rlDHFR inhibitory
activity is depicted in Figure 13B. The red (negative charge favor-
able) and blue (negative charge unfavorable) contours in the CoM-
FA electrostatic field represent 80% and 20% level contributions,
respectively.
For the 5-substituted pyrrolo[2,3-d]pyrimidines and furo[2,3-
d]pyrimidines, one relatively large blue region is found near the
20- and 30-substituents along with two small red regions located
near the 20-substituent alone. The 40-substituent overlaps with a
red region. A moderately sized blue region is found between the
50- and 60-substituents. A small blue region is near the 60-substitu-
ent alone. This small blue region is also close to the sulfur atom of
the –CH2–S– bridge in compounds 14–17, 36–38 and 53–58 (Table
1) and the nitrogen atom of the –CH2–NH– or –CH2–N(CH3)–
bridge of compounds 1–13, 39, 40 and 42–52 (Table 1). One blue
region is found near the bridge carbon atom which is directly
attached to the 5-position of the furo[2,3-d]pyrimidines or
The AOS CoMFA model was used to construct the stdev * coeffi-
cient contour maps for the rlDHFR (Fig. 13). In the CoMFA steric
field (Fig. 13A), the green (sterically favorable) and yellow (steri-
cally unfavorable) contours represent 80% and 20% level contribu-
tions, respectively.
The 1-naphthyl moiety of 5-substituted pyrrolo[2,3-d]pyrimi-
dines and furo[2,3-d]pyrimidines with a –CH2–NH– bridge is in
contact only with the yellow region whereas the 2-naphthyl con-
tacts both the green and the yellow regions. In contrast, the 2-
naphthyl moiety of 5-substituted pyrrolo[2,3-d]pyrimidines and
furo[2,3-d]pyrimidines with a –CH2–S– bridge is in contact only
with the yellow region whereas the 1-naphthyl contacts both the
green and the yellow regions. For the 3-carbon bridge at the 5-po-
sition of pyrrolo[2,3-d]pyrimidines and furo[2,3-d]pyrimidines,
both the 1- and 2-naphthyl contact both the green and yellow re-
Figure 13. (A) Orthogonal view of the steric fields generated with AOS CoMFA model of rlDHFR (after dropping three outliers from the training set): yellow indicates regions
where bulky groups decrease activity, whereas green indicates regions where bulky groups increase activity. (B) Orthogonal view of the electrostatic fields generated with the
AOS CoMFA model of rlDHFR (after dropping three outliers from the training set): red indicates regions where more negatively charged groups increase activity, whereas blue
indicates regions where more positively charged groups increase activity.

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1697
pyrrolo[2,3-d]pyrimidines. Three large red regions are located near
the 7-position –NH– or –O– for the pyrrolo[2,3-d]pyrimidine and
furo[2,3-d]pyrimidine, respectively.
For the purine analogs (compounds 18–35, Table 1), one rela-
tively large blue region is found near the 20- and 30-substituents
with two small red regions overlapping with the 2-substituents
and a red region near the 30-substituent. One small blue region is
found near the oxygen of the 50-methoxy group.
over rlDHFR activity may be possible for the 2,4-diamino-5-
methyl-6-phenylsulfanyl-substituted pyrrolo[2,3-d]pyrimidines, if
both sides of the phenyl ring are substituted (i.e., any combination
of 20, 30 and 50, 60). Indeed, among the eight previously reported 5-
methyl-6-phenylsulfanyl-substituted
pyrrolo[2,3-d]pyrimidines
59–66,12 the only compound that fits this criteria was 66 (20,50-
diOMePhenyl), which was the most potent tgDHFR inhibitor
among the series (59–66) with an IC50 value of 0.17
lM similar
In the case of 5-methyl-6-substituted pyrro[2,3-d]pyrimidines
including compounds 59–66 (Table 1), the 20- and 30-substituent
each overlaps with a red region and a small blue region is located
near the 50-region. The sulfur bridge is sandwiched between a large
red region and a relatively small blue region.
to 59 (1-naphthyl, IC50 = 0.16 M) and was also the most selective
tgDHFR inhibitor as compared to rlDHFR with a selectivity ratio of
45.9 in contrast to 28.6 for 59.
Thus it was of interest to synthesize additional analogs with
disubstitution on either side of the phenyl ring as suggested by the
steric contour maps. We synthesized and evaluated compounds
90–94 (Scheme 1). Four of these compounds (91–94) had substitu-
ents on both sides of the phenyl ring. The 2-naphthyl compound
90 was designed on the basis of the excellent activity and selectivity
of the closely related 1-naphtyl compound 59 described above.
l
4.2.1. Application of the model
As an initial application we elected to use the steric contour
maps of the tgDHFR and rlDHFR CoMFA (Fig. 14A and B, respec-
tively), which indicated that selective tgDHFR inhibitory activity
Figure 14. (A) Stereo view of the steric fields of the tgDHFR CoMFA model with the structure of compound 62, there are green (steric favorable) regions near both sides of the
phenyl moiety. (B) Stereo view of the steric fields of the rlDHFR CoMFA model with the structure of compound 60, only one side of the phenyl moiety is close to a green (steric
favorable) region.

1698
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
Table 4
These compounds were obtained via a previously reported an
oxidative thiolation by Gangjee et al.12 of the known compound
8457 (Scheme 1) with appropriately substituted thiols 85–89 in
the presence of iodine as shown in Scheme 1.12
Inhibition concentrations (IC50, in
lM) dihydrofolate reductases from T. gondii and rat
liver and selectivity ratiosa
Compd
tgDHFR
1.6
0.00511
0.572
0.212
0.686
rlDHFR
rl/tg
The thiol 89 was the only one that was not commercially avail-
able and was synthesized from the corresponding aniline.58 Com-
pounds 90–94 were evaluated as inhibitors of tgDHFR and
rlDHFR, and the results (IC50) are reported in Table 4.
90
91
92
93
94
8.1
5.1
0.388
7.6
75.9
13.3
14.2
ND
3
24.6 (12%)b
All four compounds designed on the basis of the CoMFA models
(91–94) were sub-micromolar inhibitors of tgDHFR and had selec-
tivity ratios of more than 10-fold against tgDHFR as compared to
the mammalian standard rlDHFR. Compound 91 was remarkable
in terms of both its IC50 value and selectivity ratio (5.11 nM and
75.9-fold, respectively) and is the first nanomolar inhibitor of
tgDHFR with a 6–5 fused ring system. In contrast, the analog 90
a
These assays were carried out at 37 °C under conditions of substrate (90
dihydrofolic acid) and cofactor (119 M NADPH) in the presence of 150 mM KCl.
Numbers in parentheses are percent inhibition at the given concentration.
lM
l
b
On the basis of the results in isolated tgDHFR and the cell cul-
ture inhibitory data compound 91 was evaluated in a T. gondii
infection mouse model. The HCl salt 95 was also evaluated in a
side-by-side evaluation with 91 and the data are presented in
Table 6.
was a micromolar inhibitor (IC50 = 1.6
tivity ratio of only fivefold.
lM) of tgDHFR with a selec-
The promising result for compound 91 against isolated tgDHFR
prompted us to further investigate this compound in cell culture
and in vivo. However, 91 had poor solubility properties. To im-
prove the solubility characteristics, 91 was converted to its more
soluble hydrochloride salt 95 (Scheme 1) by carefully adding 1 N
hydrochloric acid to a suspension of 91 in methanol under sonica-
tion until all the compound went into solution. Filtration and evap-
oration of the filtrate to dryness afforded the HCl salt as a white
powder.
Compounds 91 and 95 were initially evaluated in vitro as inhib-
itors of the growth of T. gondii cells in culture along with the stan-
dard compound pyrimethamine and the IC50 values are tabulated
in Table 5. These results suggest that the free base 91 and the
HCl salt 95 do not differ in the inhibition of T. gondii cells in culture,
and both are potent inhibitors of T. gondii cell growth in vitro.
Cytotoxicity of compound 91 was then evaluated against hu-
man embryonic lung fibroblast. Compound 91 inhibited these
Compound 91 has been studied in two acute infection mouse
models of RH strain of T. gondii. The intraperitoneal route was cho-
sen, based on past experience showing activity by this route with
other antifolates and on the knowledge that 91 lacked adequate
solubility to administer appropriate doses intravenously. The dose
of 50 mg/kg/day reflects the amount of compound injected but
upon harvest of the mice in the first study, we observed drug
deposits in the peritoneal cavity for many of the animals; there-
fore, the actual available dose was below 50 mg/kg/day for many
mice. In the first study, folic acid was given to one treated group
to offset any potential toxicity due to antifolate activity in mam-
malian cells; in fact, we saw no gross evidence of toxicity and folic
acid had little effect on the activity of the compound. Hence, folic
acid was not used in the second study.
The rationale for the second study was to determine if the HCl
salt 95 was more effective than the parent compound. Control lev-
els of infection and the results with 91 alone were comparable to
the first study, in that 91 reduced the median counts per 1000ꢀ
fibroblast cells only 9% at 260 lM indicating that compound 91
is relatively nontoxic against human fibroblast cells compared to
T. gondii cells in culture.
Scheme 1. Reagents and conditions: (a) CH2(CN)2, CH3OH, NEt3, room temperature; (b) EtOH, NH2)2C@NHꢂHCl, NaOH, NaOCH3, reflux; (c) EtOH–H2O 2:1, I2, reflux; (d)
hydrochloric acid, MeOH, sonication.

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1699
Table 5
IC50 of cell growth of 91, 95 and pyrimethamine in culture from T. gondiia
5. Experimental section
Compd
IC50
(
l
M)
All evaporations were carried out in vacuo with a rotary evapo-
rator. Analytical samples were dried in vacuo (0.2 mmHg) in an
Abderhalden drying apparatus over P2O5 at 70 °C. Thin-layer chro-
matography (TLC) was performed on silica gel plates with fluores-
cent indicator. Spots were visualized by UV light (254 and 365 nm).
All analytical samples were homogeneous on TLC in at least two
different solvent systems. Purification by column and flash chro-
matography was carried out using Merck Silica Gel 60 (200–
400 mesh). The amount (weight) of silica gel for column chroma-
tography was in the range of 50–100 times the amount (weight)
of the crude compounds being separated. Columns were dry-
packed unless specified otherwise. Solvent systems are reported
as volume percent of mixture. Melting points were determined
on a Mel-Temp II melting point apparatus and are uncorrected.
Proton nuclear magnetic resonance (1H NMR) spectra were re-
corded on a Bruker WH-300 (300 MHz) spectrometer. The chemi-
cal shift (d) values are reported as parts per million (ppm)
relative to tetramethylsilane as internal standard; s = singlet,
d = doublet, t = triplet, q = quartet, m = multiplet, br s = broad sin-
glet. Elemental analyses were performed by Atlantic Microlab,
Inc., Norcross, GA. Elemental compositions were within 0.4% of
the calculated values. Fractional moles of water or organic solvents
frequently found in some analytical samples of antifolates could
not be removed despite 24 h of drying in vacuo and were con-
firmed, where possible, by their presence in the 1H NMR spectrum.
All solvents and chemicals were purchased from Aldrich Chemical
Co. and Fisher Scientific and were used as received.
91
95
1.92
2.15
0.7
Pyrimethamine
a
T. gondii cell culture inhibition was assessed by measuring the incorporation of
[3H]uracil by T. gondii cells.
field by more than twofold; the drop was statistically significant (P
<0.035). The group treated with 95 showed even more effective
lowering of the counts of surviving organisms; the difference in
counts for this group was statistically different from both the con-
trol and 91 alone, with P values of <0.0001. Furthermore, there was
100% survival of the mice for 91 and 95 compared to 25% for con-
trol on the 5th day. The mice were harvested on the 6th day. Clin-
damycin was used as the positive control and also provided a 100%
survival of mice.
In summary, CoMFA analysis models that correlate the 3D
chemical structures of 80 compounds with 6–5 fused ring system
synthesized in our laboratory and their inhibitory potencies for
tgDHFR and rlDHFR were developed. In addition to conventional
CoMFA analysis, two routines available in the literature aimed at
the optimization of CoMFA: all-orientation search (AOS) and
cross-validated r2-guided region selection (q2-GRS) were used to
further optimize the models. During this process, a potential prob-
lem associated with q2-GRS routine was identified and corrected
by modifications using two strategies. Thus for the inhibitory activ-
ity against each enzyme (tgDHFR and rlDHFR), five CoMFA models
were developed using the conventional CoMFA, AOS optimized
CoMFA, the original q2-GRS optimized CoMFA and the modified
q2-GRS optimized CoMFA using the first and the second strategy.
This work demonstrated that the modified q2-GRS routines are
superior to the original routine.
On the basis of the steric contour maps, four new compounds
(91–94) were designed, synthesized and biologically evaluated. All
of the compounds had IC50 values that were at least sub-micromolar
against tgDHFR inhibitors with 10-fold or better selectivity ratios as
compared to rlDHFR. In particular, compound 91 was identified as
the first nanomolar 6–5 fused ring system inhibitor of tgDHFR.
Compound 91 and its HCl salt 95 were potent inhibitors of
T. gondii cell growth in culture with similar activities. On the basis
of the cell culture inhibitory data, both compounds were evaluated
in a T. gondii infection mouse model. The in vivo results confirmed
that compound 91 has demonstrable activity against T. gondii in an
acute in vivo model, and the HCl salt 95 is significantly more active
than the parent compound 91, a result likely due to increased sol-
ubility and better uptake from the peritoneal cavity.
5.1. 2,4-Diamino-5-methyl-6-(2-naphthylthio)pyrrolo[2,3-
d]pyrimidine (90)
To a solution of 84 (0.10 g, 0.61 mmol) in a mixture of ethanol/
water (2:1, 50 mL) was added 2-naphthylthiol 85 (0.20 g,
1.22 mmol) and the reaction mixture was heated to reflux for
4 h. The mixture was cooled to room temperature and concen-
trated under reduced pressure. After evaporation of the solvent un-
der reduced pressure, the residue was washed with ethyl acetate
followed by the addition of 200 mL of methanol and the pH ad-
justed to 10 with concentrated NH4OH. The suspension was left
at room temperature for 30 min and filtered. The residue was
washed well with methanol and air dried to give compound 90
(0.077 g, 39%): mp 276.2–282.2 °C (decomp.); TLC Rf 0.50 (CHCl3/
MeOH, 5:1, silica gel); 1H NMR (Me2SO-d6) d 2.35 (s, 3H, 5-CH3),
5.68 (s, 2H, 2-NH2), 6.33 (s, 2H, 4-NH2), 7.15–7.84 (m, 7H, C10H7),
11.06 (s, 1H, 7-H). HRMS (EI) m/e calcd for C17H15N5S
321.104817, found (M+) 321.105293.
5.2. 2,4-Diamino-5-methyl-6-(20,60-
dimethylphenylthio)pyrrolo[2,3-d]pyrimidine (91)
Table 6
Statistical analysis of two studies in T. gondii mouse model
Group
Route
Dose (mg/kg/day)
Median
Mean
SEM
P value*
Compound 91 (0.94 g, 51%) was obtained from 84 (1.0 g,
6.1 mmol), 2,6-dimethyl-benzenethiol 86 (1.7 g, 12.3 mmol), and
iodine (3.1 g, 12.3 mmol): mp 286.8–288.4 °C (decomp.); TLC Rf
0.70 (CHCl3/MeOH, 5:1, silica gel); 1H NMR (Me2SO-d6) d 2.20 (s,
3H, 5-CH3), 2.33 (s, 6H, 20,60-diMe), 5.44 (s, 2H, 2-NH2), 6.07 (s,
2H, 4-NH2) 7.07 (s, 3H, aromatic), 10.62 (s, 1H, 7-H). Anal.
(C15H17N5Sꢂ0.20H2O) C, H, N, S.
Control
91
91 + FA
N/A
ip
ip
N/A
50
50
14.0
7.0
6.0
17.9
12.5
9.8
3.3
2.5
2.0
N/A
0.055
0.019
Control
91
95
N/A
ip
ip
N/A
50
50
13.5
5.0
1.0
12.6
8.9
2.4
2.0
1.5
0.4
N/A
0.035
<0.0001
FA = folic acid, which was administered in drinking water at 10 mg/mL.
5.3. 2,4-Diamino-5-methyl-6-(30,50-
dimethylphenylthio)pyrrolo[2,3-d]pyrimidine (92)
The data analyzed are counts of T. gondii per 1000x field in total harvested peri-
toneal, the 10 microliters sampled onto a 1 cm2 area of a microscope slide for
analysis.
*
Relative to the control within each experiment. The non-parametric Mann–
Compound 92 (0.065 g, 7.0%) was obtained from 84 (0.50 g,
3.06 mmol), 30,50-dimethyl-benzenethiol 87 (0.85 g, 6.13 mmol)
Whitney test was used to compare two groups and the non-parametric Kruskal–
Wallis ANOVA was used to compare several groups within one experiment.

1700
A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
and iodine (1.56 g, 6.13 mmol): mp 285.5–288.9 °C (decomp.); TLC
Rf 0.67 (CHCl3/MeOH, 5:1, silica gel); 1H NMR (Me2SO-d6) d 2,17 (s,
6H, 3,5-(CH3)2), 2.32 (s, 3H, 5-CH3), 5.63 (s, 2H, 2-NH2), 6.29 (s, 2H,
4-NH2), 6.62 (s, 2H, 20,60-H2), 6.75 (s, 2H, 40-H), 10.93(s, 1H, 7-H).
Anal. (C15H17N5S) C, H, N, S.
5.8. Determination of IC50 values
DHFR was assayed without inhibitor and with a series of con-
centrations of inhibitors to produce 10–90% inhibition. At least
three concentrations were required for calculation. Semilogarith-
mic plots of the data yielded normal sigmoidal curves for most
inhibitors. The 50% inhibitory concentration (IC50) was calculated
from these curves using Prism 3.0.
5.4. 2,4-Diamino-5-methyl-6-(20,40,50-
trichlorophenylthio)pyrrolo[2,3-d]pyrimidine (93)
5.9. Source of T. gondii
Compound 93 (0.036 g, 8.4%) was obtained from 84 (0.20 g,
1.23 mmol), 20,40,50-trichloro-benzenethiol 88 (0.52 g, 2.45 mmol)
and iodine (0.62 g, 2.45 mmol): mp >300 °C (decomp.); TLC Rf
0.65 (CHCl3/MeOH, 5:1, silica gel); 1H NMR (Me2SO-d6) d 2.30 (s,
3H, 5-CH3), 5.72 (s, 2H, 2-NH2), 6.36 (s, 2H, 4-NH2), 6.53 (d, 1H, aro-
matic), 7.90 (s, 1H, aromatic), 11.10 (s, 1H, 7-H). Anal.
(C13H10Cl3N5S) C, H, N, Cl, S.
A clinical isolate of T. gondii was obtained from the Department of
Pathology, Indiana University School of Medicine, after a single pas-
sage in a female BALB/c mouse (Harlan Industries, Indianapolis). The
organisms were passaged in mice twice more, increasing the num-
ber of mice at each passage. After the final passage, the peritoneal
exudate was pooled and centrifuged, and the organisms were resus-
pended in RPMI medium containing 10% fetal calf serum. Frozen
stocks were prepared by adding 5% DMSO to the medium and freez-
ing slowly over 8–15 h. Stocks were stored in liquid nitrogen.
5.5. 2,4-Diamino-5-methyl-6-(30,40,50-
trimethoxyphenylthio)pyrrolo[2,3-d]pyrimidine (94)
Compound 94 (0.023 g, 5.4%) was obtained from 84 (0.20 g,
1.23 mmol), 30,40,50-trimethoxy-benzenethiol 89 (0.49 g, 2.45 mmol)
and iodine (0.62 g, 2.45 mmol). 230.5–233.1 °C (decomp.); TLC Rf
0.55 (CHCl3/MeOH, 5:1, silica gel); 1H NMR (Me2SO-d6) d 2.34
(s, 3H, 5-CH3), 3.59 (s, 3H, 40-OCH3), 3.66 (s, 6H, 30,50-(OCH3)2),
5.60 (s, 2H, 2-NH2), 6.25 (s, 2H, 4-NH2), 6.38 (s, 2H, 20,60-H2), 7.04
(d, 2H, 20, 60-2H), 10.94 (s, 1H, 7-H). HRMS (EI) m/e calcd for
C16H19N5O3S 361.120862, found (M+) 361.120657.
5.10. Culture of T. gondii for enzyme production
By using a Chinese hamster ovary cell line that lacks DHFR
(American Type Culture Collection, 3952 CL, CHO/dhfr-), T. gondii
cells were grown and maintained in Iscove’s modified Eagle’s med-
ium with 10% fetal calf serum, 1% penicillin/streptomycin, 100 mM
hypoxanthine, and 10 mM thymidine. To each 75 cm2 tissue cul-
ture flask containing the monolayer of cells was added an inocu-
lum of approximately 107 organisms. Organisms (4 ꢀ 108) were
harvested from each flask, within 6–8 days.
5.6. 2,4-Diamino-5-methyl-6-(20,60-
dimethylphenylthio)pyrrolo[2,3-d]pyrimidine hydrochloride
salt (95)
5.11. Preparation of enzymes from T. gondii
T. gondii organisms are minimally contaminated with mamma-
lian host cells when harvested from tissue culture, without detect-
able mammalian DHFR activity.58 When prepared as noted above,
DHFR from cultured T. gondii has been shown to yield IC50 values
similar to those reported in the literature.58 The kinetics for cofactor
and substrate are also similar to reported values in the literature.58
Compound 91 (0.8 g) suspended in 100 mL of methanol was
submerged in water inside of a sonicator. 1 N Hydrochloric acid
was added drop-wise to the suspension until a clear solution was
obtained, which was filtered and the filtrate evaporated to dryness
under reduced pressure to afford the salt as a white powder: 1H
NMR (Me2SO-d6) d 2.26 (s, 3H, 5-CH3), 2.34 (s, 6H, 20,60-diMe),
7.12 (s, 3H, aromatic), 7.28 (s, 2H, 2-NH2), 8.02 (s, 2H, 4-NH2),
11.76 (s, 1H, 7-H), 12.01 (s, 1H, 1-H). Anal. (C15H17N5Sꢂ1.0HClꢂ
0.50H2O) C, H, N, S, Cl.
5.12. Uracil incorporation by cultured T. gondii
For T. gondii grown in culture, uracil is incorporated into nucleic
acid, but mammalian cells do not. Thus incorporation of uracil is
used as an index of growth of T. gondii on host cells.58 T. gondii is
grown on HEL (human embryonic lung) cells with Minimum
Essential Medium (MEM) supplemented with glutamine (2 mM),
5.7. Dihydrofolate reductase (DHFR) assay
The spectrophotometric assay for DHFR was modified to opti-
mize for temperature, substrate concentration, and cofactor con-
centration for each enzyme form assayed. The standard assay
contained sodium phosphate buffer pH 7.4 (40.7 mM), 2-mercap-
toethanol (8.9 mM), NADPH (0.117 mM), 1–3.7 IU of enzyme activ-
ity (1 IU = 0.005 OD units/min), and dihydrofolic acid (0.092 mM).
KCl (150 mM) was included in the assay for T. gondii and rat liver
DHFR, because it stimulated the enzymes 1.4- and 2.63-fold,
respectively. The first three reagents were combined in a dispos-
able cuvette and brought to 37 °C. Drug dilutions were added at
this stage. The enzyme was added 30 s before the reaction was ini-
tiated with dihydrofolic acid. The reaction was followed for 5 min
with continuous recording. Activity under these conditions of as-
say was linear with enzyme concentration over at least a fourfold
range. Background activity measured with no added dihydrofolic
acid was zero with the enzyme obtained from cultured T. gondii
and near zero for other forms of DHFR. All DHFR inhibitors were
tested against rat liver DHFR as well as against pathogen DHFR
to allow assessment of selectivity.
penicillin/streptomycin (100 units/mL and 100 lg/mL, respec-
tively), and fetal bovine serum (10%). The experiment was carried
out as described previously.58
5.13. In vivo testing of drugs against T. gondii
Female BALB/c mice (18–20 g) were injected intraperitoneally
with 5 ꢀ 103 trophozoites of T. gondii from culture. Drug treatment
started immediately if the drugs were given in drinking water of
food; if the drugs were injected, treatment started 4 h after inocu-
lation. Survival or counts of T. gondii present in peritoneal exudates
or liver were monitored as an index of drug efficacy.
Acknowledgments
This work was supported in part by the National Institutes of
Health, National Institute of Allergy and Infectious Diseases
AI047759 (A.G.) and NCI CA98850 (A.G.).

A. Gangjee et al. / Bioorg. Med. Chem. 18 (2010) 1684–1701
1701
26. Hirst, J. D.; King, R. D.; Sternberg, M. J. E. J. Comput. Aided Mol. Des. 1994, 8, 405.
27. Stanton, D. T.; Murray, W. J.; Jurs, P. C. Quant. Struct. Act. Relat. 1993, 12, 239.
28. So, S.-S.; Richards, W. G. J. Med. Chem. 1992, 35, 3201.
29. Selassie, C. D.; Li, R.-L.; Poe, M.; Hansch, C. J. Med. Chem. 1991, 34, 46.
30. Debnath, A. K.; Lopez de Compadre, R. L.; Debnath, G.; Shusterman, A. J.;
Hansch, C. J. Med. Chem. 1991, 34, 786.
31. Andrea, T. A.; Kalayeh, H. J. Med. Chem. 1991, 34, 2824.
32. Selassie, C. D.; Fang, Z.-X.; Li, R.-L.; Hansch, C.; Debnath, G.; Klein, T. E.;
Langridge, R.; Kaufman, B. T. J. Med. Chem. 1989, 32, 1895.
33. Li, R.-L.; Poe, M. J. Med. Chem. 1988, 31, 366.
Supplementary data
Supplementary data (Elemental analysis and Appendix Tables
1–30.) associated with this article can be found, in the online ver-
sion, at doi:10.1016/j.bmc.2009.12.066.
References and notes
34. Booth, R. G.; Selassie, C. D.; Hansch, C.; Santi, D. V. J. Med. Chem. 1987,
30, 1218.
35. Selassie, C. D.; Fang, Z.-X.; Li, R.-L.; Hansch, C.; Klein, T. E.; Langridge, R.;
Kaufman, B. T. J. Med. Chem. 1986, 29, 621.
36. Ghose, A. K.; Crippen, G. M. J. Med. Chem. 1985, 28, 333.
37. Hansch, C.; Hathaway, B. A.; Guo, Z.-R.; Selassie, C. D.; Dietrich, S. W.; Blaney, J.
M.; Langridge, R.; Volz, K. W.; Kaufman, B. T. J. Med. Chem. 1984, 27, 129.
38. Hathaway, B. A.; Guo, Z.-R.; Hansch, C.; Delcamp, T. J.; Susten, S. S.; Freisheim, J.
H. J. Med. Chem. 1984, 27, 144.
39. Hopfinger, A. J. J. Med. Chem. 1983, 26, 990.
40. Khwaja, T. A.; Pentecost, S.; Selassie, C. D.; Guo, Z.-R.; Hansch, C. J. Med. Chem.
1982, 25, 153.
41. Hansch, C.; Li, R.-L.; Blaney, J. M.; Langridge, R. J. Med. Chem. 1982, 25, 777.
42. Ghose, A. K.; Crippen, G. M. J. Med. Chem. 1982, 25, 892.
43. Li, R.-L.; Dietrich, S. W.; Hansch, C. J. Med. Chem. 1981, 24, 538.
44. Coats, E. A.; Genther, C. S.; Dietrich, S. W.; Guo, Z.-R.; Hansch, C. J. Med. Chem.
1981, 24, 1422.
45. Crippen, G. M. J. Med. Chem. 1980, 23, 599.
46. Dietrich, S. W.; Blaney, J. M.; Reynolds, M. A.; Jow, P. Y. C.; Hansch, C. J. Med.
Chem. 1980, 23, 1205.
47. Blaney, J. M.; Hansch, C.; Silipo, C.; Vittoria, A. Chem. Rev. 1984, 84, 333.
48. Mattioni, B. E.; Jurs, P. C. J. Mol. Graph. Modell. 2003, 21, 391.
49. Sutherland, J. J.; Weaverb, D. F. J. Comput. Aided Mol. Des. 2004, 18, 309.
50. Gangjee, A.; Lin, X. J. Med. Chem. 2005, 48, 1448.
51. Wang, R.; Gao, Y.; Liu, L.; Lai, L. J. Mol. Model. 1998, 4, 276.
52. Cho, S. J.; Tropsha, A. J. Med. Chem. 1995, 38, 1060.
53. Cody, V.; Galitsky, N.; Luft, J. R.; Pangborn, W.; Gangjee, A.; Devraj, R.; Queener,
S. F.; Blakley, R. L. Acta Crystallogr., Sect. D 1997, 53, 638.
54. Molecular Operating Environment (MOE 2004.03), C. C. G., Inc, 1255 University
St., Suite 1600, Montreal, Quebec, Canada, H3B 3X3.
1. Presented in part at the 231st ACS National Meeting, Atlanta, GA, March 26–30,
2006; MEDI-349.
2. Klepser, M. E.; Klepser, T. B. Drugs 1997, 53, 40.
3. Gangjee, A.; Devraj, R.; McGuire, J. J.; Kisliuk, R. L.; Queener, S. F.; Barrows, L. R.
J. Med. Chem. 1994, 34, 1169.
4. Gangjee, A.; Mavandadi, F.; Queener, S. F.; McGuire, J. J. J. Med. Chem. 1995, 38,
2158.
5. Gangjee, A.; Mavandadi, F.; Kisliuk, R. L.; McGuire, J. J.; Queener, S. F. J. Med.
Chem. 1996, 39, 4563.
6. Gangjee, A.; Mavandadi, F.; Queener, S. F. J. Med. Chem. 1997, 40, 1173.
7. Gangjee, A.; Vasudevan, A.; Queener, S. F. J. Med. Chem. 1997, 40, 3032.
8. Gangjee, A.; Guo, X.; Queener, S. F.; Cody, V.; Galitsky, N.; Luft, J. R.; Pangborn,
W. J. Med. Chem. 1998, 41, 1263.
9. Gangjee, A.; Dubash, N. P.; Queener, S. F. J. Heterocycl. Chem. 2000, 37, 935.
10. Gangjee, A.; Vidwans, A.; Elzein, E.; McGuire, J. J.; Queener, S. F.; Kisliuk, R. L. J.
Med. Chem. 2001, 44, 1993.
11. Gangjee, A.; Yu, J.; Kisliuk, R. L. J. Heterocycl. Chem. 2002, 39, 833.
12. Gangjee, A.; Lin, X.; Queener, S. F. J. Med. Chem. 2004, 47, 3689.
13. Gangjee, A.; Zeng, Y.; McGuire, J. J.; Mehraein, F.; Kisliuk, R. L. J. Med. Chem.
2004, 47, 6893.
14. Gangjee, A.; Jain, H. D.; Queener, S. F. J. Heterocycl. Chem. 2005, 42, 589.
15. Gangjee, A.; Ye, Z.; Queener, S. F. J. Heterocycl. Chem. 2005, 42, 1127.
16. Gangjee, A.; Zeng, Y.; Ihnat, M.; Warnke, L. A.; Green, D. W.; Kisliuk, R. L.; Lin,
F.-T. Bioorg. Med. Chem. 2005, 13, 5475.
17. Ivanciuc, O.; Ivanciuc, T.; Cabrol-Bass, D. Theochem. J. Mol. Struct. 2002, 582, 39.
18. Burbidge, R.; Trotter, M.; Buxton, B.; Holden, S. Comput. Chem. 2001, 26, 5.
19. Garg, S.; Achenie, L. E. K. Biotechnol. Prog. 2001, 17, 412.
20. Zheng, W.; Tropsha, A. J. Chem. Inf. Comput. Sci. 2000, 40, 185.
21. Selassie, C. D.; Gan, W.-X.; Kallander, L. S.; Klein, T. E. J. Med. Chem. 1998, 41,
4261.
55. SYBYL Version 7.0; Tripos Associates: St. Louis, M.
22. Burden,F.R.;Rosewarne, B.S.;Winkler,D. A. Chemom.Intell. Lab. Syst. 1997, 38, 127.
23. Ivanciuc, O. Rev. Roum. Chim. 1996, 41, 645.
24. Marlowe, C. K.; Selassie, C. K.; Santi, D. V. J. Med. Chem. 1995, 38, 967.
25. Hirst, J. D.; King, R. D.; Sternberg, M. J. E. J. Comput. Aided Mol. Des. 1994, 8, 421.
56. Godden, J. W.; Xue, L.; Bajorath, J. J. Chem. Inf. Comput. Sci. 2000, 40, 163.
57. Taylor, E. C.; Patel, H. H.; Jun, J.-G. J. Org. Chem. 1995, 60, 6684.
58. Offer, J.; Boddy, C. N. C.; Dawson, P. E. J. Am. Chem. Soc. 2002, 124, 4642.