Computational predictions of binding affinities to dihydrofolate reductase: Synthesis and biological evaluation of methotrexate analogues

Article abstract of DOI:10.1021/jm0009639

The relative binding affinities to human dihydrofolate reductase of four new potential antifolates, containing ester linkages between the two aromatic systems, were estimated by free energy perturbation simulations. The ester analogue, predicted to exhibi

Full text of DOI:10.1021/jm0009639

3852
J . Med. Chem. 2000, 43, 3852-3861
Com p u ta tion a l P r ed iction s of Bin d in g Affin ities to Dih yd r ofola te Red u cta se:
Syn th esis a n d Biologica l Eva lu a tion of Meth otr exa te An a logu es
Malin Graffner-Nordberg,† J ohn Marelius,‡ Sofie Ohlsson,§ Åsa Persson,§ Go¨te Swedberg,# Paul Andersson,§
Sven E. Andersson, J ohan Åqvist,‡ and Anders Hallberg*,†
Department of Organic Pharmaceutical Chemistry, Uppsala Biomedical Center, Uppsala University, Box 574,
SE-751 23 Uppsala, Sweden; Department of Cell and Molecular Biology, Uppsala Biomedical Center, Uppsala University,
Box 596, SE-751 24 Uppsala, Sweden; Discovery, AstraZeneca R&D Lund, SE-221 87 Lund, Sweden; Department of
Pharmaceutical Microbiology, Uppsala Biomedical Center, Uppsala University, Box 581, SE-751 23 Uppsala, Sweden; and
Department of Internal Medicine, Lund University Hospital, SE-221 86 Lund, Sweden
Received April 3, 2000
The relative binding affinities to human dihydrofolate reductase of four new potential
antifolates, containing ester linkages between the two aromatic systems, were estimated by
free energy perturbation simulations. The ester analogue, predicted to exhibit the highest
binding affinity to human dihydrofolate reductase, and a reference ester (more structurally
related to methotrexate) were synthesized. As deduced from the measured IC50 values, the
calculated ranking of the ligands was correct although a greater difference in affinity was
indicated by the experimental measurements. Among the new antifolates the most potent
inhibitor exhibited a similar pharmacokinetic profile to methotrexate but lacked activity in a
complex antiarthritic model in rat in vivo.
In tr od u ction
an ester-containing bridge to a second aromatic ring. It
was revealed that compounds containing 2,4-diamino-
quinazoline as the bicyclic element only sluggishly
undergo enzymatic hydrolyses in human and rat liver
S9 fractions as well as in human duodenal mucosa
homogenate and, in particular, in human blood leuko-
cytes, the latter being target cells for immunosuppres-
sive drugs. These findings encouraged us to synthesize
and evaluate an analogue to MTX, where the methyl-
eneamino bridge was substituted with a suitable ester
unit, as a potential metabolically stable inhibitor of
human DHFR possessing a new linker type.
Dihydrofolate reductase (DHFR) is a well-established
target for drug intervention. Several inhibitors are
approved as therapeutic agents against a variety of
pathological disorders. Thus, the classical antifolate,
methotrexate (MTX), is used in cancer therapy and
frequently for the treatment of rheumatoid arthritis1-3
and psoriasis.4-6 MTX is also reported to be effective
against steroid-resistant asthma7,8 and inflammatory
bowel diseases.9,10 Research programs aimed at identi-
fying new classical antifolates have been pursued, and
numerous new derivatives with potential use against,
for example, rheumatoid arthritis have been reported.11
The nonclassical lipophilic inhibitors have been used for
a long time against bacterial infections (e.g., trimetho-
prim) and in the treatment of malaria (e.g., py-
rimethamine). Considerable efforts are at present de-
voted to identifying selective DHFR inhibitors for
treatment of infections caused by Pneumocystis carinii
and Toxoplasma gondii, two major opportunistic patho-
gens often found in patients with acute AIDS and other
immunodeficiency disorders.12 Trimetrexate13 and piri-
trexim14 are two potent, lipophilic, nonclassical inhibi-
tors from the new generation of antifolates used in the
clinic. In a previous metabolism study, a series of model
compounds was investigated with the objective of defin-
ing the molecular structural factors which affect the rate
of enzymatic ester cleavage and the tissue selectivity
of hydrolytic metabolism.15 The model compounds ex-
amined comprised bicyclic aromatic units connected by
We herein report the calculations of relative binding
affinities from free energy perturbation simulations that
guided the selection of the target compound to be
synthesized, the subsequent synthesis, and the biologi-
cal evaluation in a human DHFR assay. Furthermore,
to assess the metabolic stability of the ester bond of the
selected compound, a pharmacokinetic study was per-
formed and the compound was thereafter evaluated in
an antigen-induced arthritis (AIA) model.
Resu lts
Com p u ta tion a l P r ed iction of Bin d in g Affin ities.
Organic synthesis is time-consuming, and methods
allowing reliable predictions of affinity and specificity
of ligands prior to synthesis would be of paramount
value in the drug discovery process. Even with a three-
dimensional structure of the target available, compu-
tational predictions of binding affinity of a new ligand
are difficult when the exact structure of the ligand-
enzyme complex is unknown. A range of different
methods to approach this problem has been developed,16
including methods for scoring putative inhibitors from
single conformations of ligand-enzyme complexes17 and
estimations of absolute binding free energies using
* To whom correspondence should be addressed. Tel: +46 18
4714284. Fax: +46 18 4714976. E-mail: Anders.Hallberg@bmc.uu.se.
† Department of Organic Pharmaceutical Chemistry, Uppsala Uni-
versity.
‡ Department of Cell and Molecular Biology, Uppsala University.
§ AstraZeneca R&D Lund.
#
Department of Pharmaceutical Microbiology, Uppsala University.
Lund University Hospital.
10.1021/jm0009639 CCC: $19.00 © 2000 American Chemical Society
Published on Web 10/04/2000

Computational Prediction of Binding Affinity to DHFR
J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21 3853
F igu r e 1. MTX and four potential inhibitors of DHFR used in the calculations of binding free energies.
F igu r e 3. Thermodynamic cycle for relative binding affinity.
tively, were performed. From a pair of simulations, one
of the inhibitors free in water and the other comprising
the inhibitor bound to the solvated enzyme, the relative
free energy of binding can be calculated using the
thermodynamic cycle shown in Figure 3. The pKa values
of the bicyclic element of the 2,4-diaminopteridines and
the 2,4-diaminoquinazolines, respectively, differ by
about 2 units.20 Energy is required to protonate the
pteridines upon binding, whereas the quinazolines are
protonated in solution at pH 6.9 (used in IC50 determi-
nations), as illustrated in Figure 2. To compare the
energetics of the bound and free states of the ligand
using the thermodynamic cycle, the same protonation
state must be used in both environments. All simula-
tions were thus carried out using the protonation state
of the bound ligands, and the free energy for protonation
in water was added as a correction. The resulting free
energies, which are averages over independent simula-
tions of the forward and reverse transformation simula-
tions, are summarized in Table 1. The simulations
identified the quinazoline ester 1a as about 4 times
more potent as an inhibitor of human DHFR than the
pteridine ester 1b but less potent than MTX by 1 order
of magnitude (see Table 2). A methylene extension was
found to reduce the enzyme affinity as deduced from
calculations. Thus, the quinazoline compound 8a was
predicted to be approximately equipotent to 1b, while
the pteridine 8b was predicted to have the lowest
affinity among the compounds studied. We finally
selected 1a as the first candidate for synthesis and
biological evaluation. Furthermore, the lower affinity
of 1b that was predicted by the FEP calculations was
not obvious from modeling, and we therefore felt
prompted to also synthesize 1b as a reference compound
F igu r e 2. Binding processes of 2,4-diaminopteridines and 2,4-
diaminoquinazolines. The former is not protonated at N1 in
water at neutral pH, and protonation upon binding is associ-
ated with a free energy cost.
linear interaction energy methods based on conforma-
tional sampling by molecular dynamics (MD) simula-
tions.18
We used relative binding affinities calculated from
free energy perturbation (FEP) simulations19 for the
selection of one of the four esters depicted in Figure 1
as the compound likely to inhibit human DHFR most
efficiently. Binding free energies of the esters 1a ,b
relative to the reference compound MTX as well as of
8a ,b relative to 1a ,b were calculated. In this way the
largest perturbations (MTX f 8a ,b) are divided into two
steps with the two compounds 1a ,b as intermediates
along the FEP paths. Compound 1b, which is structur-
ally most similar to MTX, and the 2,4-diaminoquinazo-
line 1a both contain the same bridge (methyleneoxy-
carbonyl) as in the compounds previously examined in
the metabolism study.15 The other two candidates for
synthesis, the quinazoline 8a and the pteridine deriva-
tive 8b, possessed four-atom bridges (Figure 1). Energy
is required to protonate the pteridines upon binding,
whereas the quinazolines are protonated also in solution
at neutral pH, as illustrated in Figure 2.
MD simulations, during which MTX was gradually
transformed into 1a or 1b and 1a ,b into 8a ,b, respec-

3854 J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21
Graffner-Nordberg et al.
Sch em e 1a
a
Reagents and conditions: (a) tert-butyl alcohol, pyridine; (b) KOH, tert-butyl alcohol; (c) cesium carbonate, DMF; (d) TFA.
Sch em e 2a
a
Reagents and conditions: (a) DCI, 4-pyrrolidinopyridine, CH2Cl2; (b) 25% piperidine in DMF; (c) 5a or 5b, PyBOP, DIPEA, DMF; (d)
50% TFA/CH2Cl2.
Ta ble 1. Relative Binding Free Energies
Ta ble 2. Experimental and Calculated Inhibitory
Concentrations against Recombinant Human DHFR
pKa
bind
(kcal/mol)b
∆∆GFbiEnPd(AfB) ∆∆G (AfB)
compd compd
∆∆Gbind(AfB)
a
A
B
(kcal/mol)a
(kcal/mol)c
compd exptl IC50 (µM) exptl relative IC50 calcd relative Ki
MTX
1a
1b
8a
8b
0.002b
0.125b
3.1b
1
63
1
9
34
56
MTX
MTX
1a
1a
1b
1a
1b
1b
8a
8b
2.9 ( 0.3
2.1 ( 0.3
1.4 ( 0.6
1.1 ( 0.5
2.4 ( 1.2
-1.6
1.3 ( 0.3
2.1 ( 0.3
3.0 ( 0.6
1.1 ( 0.5
2.4 ( 1.2
0
1.6
0
0
1.5 × 103
nac
nac
nac
nac
1.9 × 103
a
a
The calculated free energy is the average of forward and
backward FEP summations from independent simulations of the
A f B and B f A transformations. The error given is the standard
Based on addition of relative free energies of binding from
b
Table 1 along the shortest path from MTX. The corresponding
IC50 values obtained with rat liver DHFR assay according to ref
46 were determined as 0.0025 µM (MTX), 0.0193 µM (1a ), and
0.12 µM (1b), respectively. c na, not available (compound not
synthesized).
b
error of the mean. Change in free energy of protonation of the
inhibitors in solution at pH 6.9, given by eq 1. c Total predicted
relative binding affinity ) ∆∆GFbiEnPd(A f B) + ∆∆GpKa (A f B).
bind
to allow for comparison of the theoretical and experi-
mental data.
was introduced with the objective of simplifying puri-
fications. Esterification of the commercially available
Fmoc-Glu-OtBu was conducted on ArgoGel with the acid
labile Wang linker as solid support (Scheme 2). Fmoc
deprotection by 20% piperidine in DMF yielded the
polymer-supported glutamic acid derivative 6, which
was further reacted with compound 5a , using PyBOP
as coupling reagent, eventually providing the amide 7a .
According to 13C NMR the amide bond formation oc-
curred smoothly. Hydrolysis of 7a in 50% TFA/dichlo-
romethane for 1.5 h cleaved both the tert-butyl ester and
the Wang linker in one step providing the target
compound 1a in 92% yield (four steps, total yield
starting from the Wang resin). The photolabile linker,
P-linker-2,22 was initially employed as an acid/base
stable linker. However, since the central ester bond
Ch em ical Syn th eses. Standard solution-phase chem-
istry was combined with solid-phase synthesis for the
preparation of compounds 1a ,b. A literature method21
starting from terephthaloyl chloride with pyridine in
tert-butyl alcohol and subsequent monodeprotection,
using potassium hydroxide in tert-butyl alcohol, yielded
the monoester 2 (Scheme 1). A coupling reaction of 2,4-
diamino-6-(bromomethyl)quinazoline (3a ) with cesium
carbonate as base in N,N-dimethylformamide (DMF)
afforded the ester 4a in a satisfactory yield. Selective
hydrolysis of the tert-butoxyl group of 4a in trifluoro-
acetic acid (TFA) delivered the acid 5a .
Since previously we had experienced difficulties in the
isolation of related compounds, solid-phase chemistry

Computational Prediction of Binding Affinity to DHFR
J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21 3855
Ta ble 3. Plasma Pharmacokinetic Parametersa in Male SD Rats (n ) 3)
a. After Intravenous Administration of MTX and Compound 1a
AUCinf (nmol/L × h) Cmax (nmol/L) Vss (L/kg) T1/2 (h)
8257 ( 3684 0.36 ( 0.16
9171 ( 2392 0.42 ( 0.076
b. After Oral Administration of MTX and Compound 1a
treatment
CL (mL/min/kg)
MTX iv
1a iv
1559 ( 528
1231 ( 489
0.9 ( 0.06
1.0 ( 0.3
11 ( 3.3
15 ( 5.4
treatment
AUC (nmol/L × h)
C
max (nmol/L)
Tmax (h)
F (%)
MTX po
1a po
838 ( 299
384 ( 207
266 ( 61.6
144 ( 30.1
0.67 ( 0.29
0.67 ( 0.29
5.8 ( 2.3
3.4 ( 2.0
a
AUC (area under the curve), Cmax (mean peak plasma concentration), Vss (volume of distribution at steady state), T1/2 (plasma half-
life), CL (clearance), Tmax (time required to reach Cmax), F (bioavailability).
Dih yd r ofola te Red u cta se Assa y. Compounds 1a ,b
and MTX were tested for their ability to inhibit recom-
binant human DHFR. The results are presented in
Table 2 as IC50 values, and also as IC50 values relative
to MTX. It was revealed that compound 1a was ap-
proximately 60 times less potent than MTX as an
inhibitor of human DHFR in the assay used. The
pteridine ester 1b exhibited a very poor inhibitory effect.
P h a r m a cok in etics in Ra ts. The pharmacokinetics
of the quinazoline ester 1a and of MTX were evaluated
in rats after intravenous and oral administration (aver-
age plasma concentrations, corrected for differences in
dose plotted versus time, are shown in Figure 4).
Compound 1a and MTX exhibited similar kinetics after
intravenous administration (CL: 15 and 11 mL/min/
kg, respectively). The pharmacokinetic parameters are
given in Table 3. The alcohol metabolite formed after
enzymatic hydrolysis and also the corresponding car-
boxylic acid (formed after subsequent oxidation of this
alcohol) were both observed in plasma.
An tigen -In d u ced Ar th r itis in Ra ts. Two compari-
sons were made; first the effect of MTX administered
orally was compared to that given ip. The data indicates
that MTX is effective irrespective of the mode of
administration although the antiarthritic effect was
somewhat slower in onset in the ip group (Figure 5a).
Second, the effect of compound 1a was compared to that
of MTX, where both were given ip daily. The ester
analogue 1a exerted no detectable antiarthritic effect
despite that it was given in a 25 times higher dose than
MTX (Figure 5b). Note that MTX is only 6 times more
potent than 1a as an inhibitor in the rat DHFR assay
(Table 2, see footnotes).
F igu r e 4. (a) Plasma concentration-time curves of MTX and
compound 1a after iv administration to male SD rats. The
values are the means ( SD for three rats. The doses of MTX
(9) and 1a (b) are normalized to 1 µmol/kg. (b) Plasma
concentration-time curves of MTX and compound 1a after po
administration to male SD rats. The values are the means (
SD for three rats. The doses of MTX (0) and 1a (O) are
normalized to 10 µmol/kg.
Discu ssion
Compound 1a is predicted to bind to human DHFR
with only 1.3 kcal/mol lower affinity than MTX; the less
favorable interactions of 1a indicated by the positive
remained intact after treatment with TFA with the
Wang linker, P-linker-2 offered no advantages in this
case. The main disadvantage using the latter linker was
that deprotection of the tert-butyl ester was required
prior to the photolysis. Furthermore, byproducts and
liberated poly(ethylene glycol) (PEG) were observed
after photolysis. Neither couplings of 3a or 2,4-diami-
noquinazoline-6-methanol15 with a polymer-supported
benzoic acid derivative nor early efforts to attach the
amino groups of the heterocycle to different linkers were
successful. The pteridine derivative 1b was prepared
analogosly to 1a , from commercially available 2,4-
diaminopteridine-6-methanol, by the synthetic route
outlined in Scheme 2.
FEP
∆∆G
are compensated in part by the difference in
bind
free energy of protonation (Table 1). From comparison
of the experimental structure of the DHFR-MTX
complex to the structure of 1a at the end of the
transformation of MTX shown in Figure 6, it is clear
that the quinazoline ring remains tightly held in place.
However, even with a well-defined binding mode of the
bicyclic ring system there may be conformational alter-
natives regarding the positioning of the rest of the
ligand. The nitrogens at positions 5 and 8 that distin-
guish the pteridine from the quinazoline moiety do not
appear to affect the binding mode of the ring system.
In fact, the only hydrogen bond from these nitrogens in

3856 J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21
Graffner-Nordberg et al.
F igu r e 6. Compound 1a in the binding site of human DHFR
at the end-point of the transformation from MTX (yellow) and
MTX in the crystal structure (green). Hydrogen bonds are
indicated in gray. The quinazoline superimposes well on the
pteridine moiety from MTX. The ester carbonyl is oriented
opposite to N-CH3. Interactions between the carboxylate
groups and several basic side chains and water molecules are
not shown.
the inhibitors is very flexible, can adopt different
conformations, and accepts hydrogen bonds or forms ion
pairs in different ways. This can be rationalized by the
presence of several positively charged side chains at
similar distances around the opening to the binding site,
which can make similar and interchangeable contacts
with the carboxylate groups.20
The difference in binding free energy between the
quinazoline and pteridine inhibitors is dominated by the
difference in free energy of protonation, which favors
the quinazolines by 1.6 kcal/mol at pH 6.9. In both the
quinazoline and pteridine frameworks, extending the
linker chain by one methylene group between the ester
and the phenyl groups (compounds 8a ,b) decreases the
affinity by 1.1-2.4 kcal/mol. The reason for this seems
to be that the longer linker chain cannot attain a relaxed
extended conformation in the complex, even though the
phenyl group is pushed somewhat outward. Further-
more, the displacement of the phenyl group disturbs the
interactions with Phe31 of the enzyme.
Adding relative free energies in Table 1 along the path
MTX f 1a f 1b f MTX gives 2.2 kcal/mol. This error
in closing the cycle of inhibitors is probably not due to
convergence problems in the individual transformations,
where results from independent forward and backward
simulations agree very well. Uncertainty about the
details of the binding conformation of 1a ,b reflected in
the initial structures for the 1a f 1b and 1b f 1a
simulations, which are the end-points of MTX f 1a and
MTX f 1b simulations, respectively, may contribute to
this error. A corresponding simulation starting from the
docked structures of 1a was performed for comparison
and gave a ∆∆Gbind of 3.9 kcal/mol, only marginally
different from the 3.0 kcal/mol reported in Table 1.
Generally, the amino acid side chains lining the
binding site undergo slight conformational changes
during the MD simulations, whereas backbone displace-
ments are negligible. In a typical simulation, the root-
mean-square (rms) coordinate deviations from the initial
F igu r e 5. (a) J oint swelling in AIA rats (n ) 6/time point
except vehicle po (n ) 5)) treated with MTX po (0, 4 mg/kg/
week starting 1 week before challenge), vehicle po (saline, 4),
MTX ip (9, 0.15 mg/kg/day starting on the day of challenge),
or vehicle ip (saline, 2). All data are from a separate experi-
ment and are given as mean ( SEM; *p < 0.05, **p < 0.01,
***p < 0.001 compared to the vehicle group (Fisher’s PLSD).
(b) J oint swelling in AIA rats (n ) 6/time point) after ip doses
of MTX (9, 0.15 mg/kg/day), compound 1a (b, 3.75 mg/kg/day),
or vehicle (saline, 2) respectively. All data are from a separate
experiment and are given as mean ( SEM; *p < 0.05, **p <
0.01 compared to the vehicle group (Fisher’s PLSD).
the MTX complex structure is to a water molecule (not
shown). A feature observed in nearly all the complex
simulations is that the orientation of the ester carbonyl
group is opposite to that of the N-CH3 group in MTX
as shown in Figure 6. This arrangement, while sterically
different, maintains the same direction of the dipole
moment of the ester linker chain as the methyleneamino
linker in MTX. In the water simulations, there is a
rotation about the bonds of the linker chain and no clear
preference for any particular orientation of the ester
group versus the bicyclic element is observed. It appears
that the ester group, irrespective of the orientation of
the carbonyl, is poorly ‘solvated’ by the protein. Trans-
ferring the more polar ester group to a rather hydro-
phobic pocket where it lacks specific interactions with
the protein is probably unfavorable compared to the less
polar N-CH3 group. As expected, transformation from
two- to three-atom linkers shifts the phenyl and
glutamate moieties outward somewhat. We find it
reassuring to see that the conformations of MTX at the
end-point of the 1a f MTX and 1b f MTX simulations
in the complex are very similar to that in the crystal
structure, except for the glutamate residue which is
partly exposed to the solvent. The glutamate residue of

Computational Prediction of Binding Affinity to DHFR
J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21 3857
structure for 15 active site residues was limited to about
1 Å. Hydrogen bonds and ion pairs with the pteridine/
quinazoline also remain intact during all the simula-
tions.
Con clu sion
It has been demonstrated that the estimation of
relative binding free energies by FEP simulations can
be useful for the selection of target compounds to be
synthesized for biological evaluation. Both of the ester
analogues, 1a,b, were predicted to inhibit human DHFR,
although to a lesser extent than MTX itself. Compound
1a was predicted to be more potent than 1b. Although
a larger difference in affinity was encountered in the
human DHFR assay, the calculated ranking of the
ligands was correct. The analogue 1a has a high affinity
for DHFR, and despite its ester link, 1a exhibits a
similar pharmacokinetic profile to MTX. Thus, it seems
that the ester link could provide a new suitable substi-
tute for the methyleneamino bridge and other fragments
connected to the ring system that are commonly found
in analogues of MTX. However, in a more complex in
vivo system, the antigen-induced arthritis model, the
ester 1a was not effective. A more extensive study is
necessary to elucidate this particular issue.
Long simulations and a large number of FEP steps
(λ points) were required to achieve convergence between
the independent simulations of the MTX f 1a and 1a
f MTX transformation processes. These transforma-
tions involve both the change of charge distribution
between the pteridine and quinazoline ring systems and
the creation and annihilation of atoms in the linker
chain and are associated with large absolute values of
∆G1 and ∆G2 (ca. 100 kcal/mol). These values have no
physical interpretation since they include also contribu-
tion from intramolecular interactions, whose absolute
energy values are determined by the force-field logic.
Most of the other transformation processes yielded
nearly identical results with only one-fourth of the
simulation time (4.5-ps data collection at about 60 λ
points).
The in vivo effects of MTX and 1a were evaluated in
a rat model. The amino acid sequence of the rat enzyme
(R. norvegicus) is not available in any database. The
sequence from another rodent (M. auratus) is 90%
identical to the human DHFR enzyme. No major dif-
ferences between rat and human DHFR in binding
affinity of the inhibitors were expected as deduced from
a homology model of the sequence of M. auratus showing
that the residues involved in binding are conserved. This
was confirmed by the experimental enzyme binding data
shown in Table 2.
Exp er im en ta l Section
Com p u ta tion a l Meth od s. Relative binding affinities were
calculated as relative free energies of binding using the free
energy perturbation (FEP) method.19 For another application
of this method to DHFR inhibitors, see Cummins and Gready.27
This procedure is based on a thermodynamic cycle as shown
in Figure 3. From the thermodynamic cycle it follows that the
relative binding affinity ∆∆Gbind(A f B) can be calculated
using:
∆∆Gbind(A f B) ) ∆Gbind(B) - ∆Gbind(A) ) ∆G2 - ∆G1 (1)
When MTX was given orally, the antiarthritic effect
was similar to that seen after ip administration, this
despite that the oral bioavailability was only about 5%
(Table 3b). Part of this discrepancy could be explained
by the effect of polyglutamation leading to intracellular
pharmacokinetics, which most likely differ substantially
from the extracellular pharmacokinetics which were
measured. The total dose given was also higher in the
oral group (3 × 4 ) 12 mg/kg) compared to the ip group
(10 × 0.15 ) 1.5 mg/kg). It seems less likely that these
factors entirely could explain the high effectiveness of
the oral administration. Our reason for comparing these
two groups was that we hypothesized that a large part
of the effect of MTX could take place in the gut mucosa.
Our results still suggest that this is possible. After oral
dosing the drug concentration is likely to be higher in
the gut wall than in plasma. Furthermore, one possible
mechanism of action for MTX is that it could induce
apoptosis in activated T-lymphocytes.23 Since a large
portion of the body’s T-cells are present in the gut
lymphatic system, it could simply be that a high drug
concentration here is important for a strong effect.
The free energies ∆G1 and ∆G2 pertain to the “alchemical”
processes where inhibitor A is transformed into inhibitor B in
solution and bound to the solvated enzyme, respectively. These
processes can be simulated with moderate computational
effort, as opposed to the actual binding processes. The enzyme
does not affect the calculation of ∆G1, which thus only involves
FEP transformation between the solvated inhibitors. The free
energy associated with transforming inhibitor A into inhibitor
B, whose interactions are described by the potentials VA and
VB, respectively, can be calculated using the perturbation
formula:
∆Gi(A f B) ) -RT ln exp(-(Vm′ - Vm)/RT)
(2)
m
∑
m
where i ) 1 or 2 as in eq 1. Vm ) (1 - λ)VA + λVB are
intermediate potentials used during the transformation and
λ∈ [0,1] is a mapping parameter that controls the transforma-
tion between the initial and final states. The free energy is
thus calculated as a sum over small steps in λ in which Vm
changes to Vm′. The ensemble averages ‚ are calculated from
m
molecular dynamics (MD) simulations on the potential Vm.
In a previous work20 that addressed the calculation of
absolute binding free energies the approximate linear interac-
tion energy (LIE) method18 was used. However, for the set of
rather similar inhibitors studied here one can expect that the
more rigorous FEP method should provide accurate relative
binding estimates with manageable computational expense.
Furthermore, the fluctuations in the inhibitors’ electrostatic
interaction energies with the surroundings due to the ionic
groups in the glutamate moiety tend to overshadow the
changes due to substitutions in the ring system and linker
chain.20 This situation makes the LIE method less suited for
eliciting small changes in affinity. The difficulties to achieve
convergent results with ionic groups at the ends of flexible
arms remain challenging also with FEP calculations of ∆∆G’s,
although less pronounced since the relative free energies are
The ester analogue 1a had no antiarthritic effect. This
result was clearly unexpected in light of the pharma-
cokinetic data and the effect on DHFR in vitro. A
possible explanation for this discrepancy would be if the
primary target for MTX in the antigen-induced arthritic
model was another folate-dependent enzyme other than
DHFR, e.g., AICARtf (5-aminoimidazole-4-carboxamide
ribonucleotide transformylase). Despite thorough re-
search in this area, the mechanism of action of MTX in
rheumatoid arthritis is still under dispute.23-26

3858 J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21
Graffner-Nordberg et al.
calculated directly rather than as the difference between two
absolute values. FEP calculations on this system also poses
new challenges: (1) Extending the linker chain of the inhibi-
tors by introducing groups in the interior of the chain requires
careful handling of both bonded and nonbonded interactions
within the molecule to avoid excessive strain. (2) Docking of
the ester-linked inhibitors is not trivial, despite the strong
similarities to MTX, since an extra element in the linker chain
is introduced and the dipole of the ester carbonyl group is
opposite that of the N-CH3 of MTX. (3) The preferred
conformation of the ester group, particularly the orientation
of the carbonyl, is not obvious: e.g., different orientations
occurred among the highest ranked conformations generated
by the docking program FLEXX.28
The ∆∆G’s obtained from eq 1 can only be calculated
assuming the same protonation state of the inhibitor in
solution and in the complex. This is valid for the 2,4-
diaminoquinazolines (pKa about 7.7)20 but not true for the 2,4-
diaminopteridines, which have a pKa of about 5.729 and are
known to become protonated at N1 upon binding.29 In such
cases ∆∆G must be corrected by the free energy associated
with protonation of the inhibitor in solution. This correction
takes the form:
enzyme-cofactor-inhibitor complexes, the center of the 18 Å
radius sphere was fixed at the binding site and protein atoms
outside this sphere (ca. 750) were not included in the simula-
tions. Partial charges for the inhibitors were assigned as in
earlier work20 based on semiempirical AM1-SM2 calculations.33
The complexes were equilibrated by a stepwise heating scheme
during 58 ps. All simulations were carried out at 300 K with
a time step of 1.5 fs. At each of the 116 λ points the system
was equilibrated for 3 ps followed by data collection during 9
ps, giving a total simulation time of 1450 ps. Independent
simulations of the forward and reverse transformation pro-
cesses were conducted to estimate the error of the FEP
calculations.
Ch em istr y. Gen er a l In for m a tion . Melting points (uncor-
rected) were determined in open glass capillaries on an
Electrothermal apparatus. Infrared (IR) spectra were recorded
on a Perkin-Elmer 1605 FT-IR spectrophotometer and are
recorded in vmax (cm-1). 1H and 13C NMR spectra were recorded
on a J EOL J NM-EX270 spectrometer at 270 and 67.8 MHz,
respectively and a J EOL J NM-EX400 spectrometer at 400 and
100 MHz, respectively. Thin-layer chromatography (TLC) was
performed by using aluminum sheets precoated with silica gel
60 F254 (0.2 mm) type E. Merck. Chromatographic spots were
visualized by UV light. Column chromatography was con-
ducted on silica gel S (0.032-0.063 mm; Riedel-deHae¨n) and
silica gel 60 (0.040-0.063 mm; E. Merck), unless otherwise
noted. Preparative TLC was performed on glass sheets pre-
coated with silica gel 60 F254 (2.0 mm; E. Merck). Centrifugal
chromatography was carried out on a Harrison Research
Chromatotron (model 7924T) with silica gel PF254 containing
gypsum (E. Merck) as solid phase. The elemental analyses
were performed by Mikro Kemi AB, Uppsala, Sweden, or
Analytische Laboratorien, Gummersbach, Germany, and were
within (0.4% of the calculated values. All commercial chemi-
cals were used without further purification. In the cases where
solid-phase syntheses were adopted, washings after all reac-
tions were performed with different combinations of solvents
using DMF, methanol, and dichloromethane. All reactions on
solid phase were performed in polypropylene syringe columns
fitted with a polypropylene frit. The latter was also used for
all the washings. 13C NMR on solid phase were conducted in
DMSO-d6 to achieve adequate spectra. The loading value for
the polymersupported glutamic acid derivative 6 was obtained
by amino acid analysis performed at the Department of
Biochemistry, Uppsala Biomedical Center, Uppsala, Sweden,
on 24-h hydrolysates with a LKB 4151 alpha plus analyzer
using ninhydrin detection.
Ter ep h th a lic Acid Mon o-ter t-bu tyl Ester (2).21 Pyridine
(4.0 g, 50 mmol) and tert-butyl alcohol (3.7 g, 49 mmol) were
added to terephthaloyl chloride (5.0 g, 25 mmol). Heat was
evolved and a solid cake was formed. The reaction mixture
was allowed to stir overnight before it was extracted with
diethyl ether and aqueous sodium bicarbonate. Drying (Mg-
SO4) and filtration yielded after concentration under reduced
pressure a fine white powder, which was recrystallized from
aqueous ethanol affording 4.3 g (62%) of pure diester: mp
116-117 °C (lit.21 mp 118 °C). The diester (2.5 mg, 90 mmol)
was dissolved in warm tert-butyl alcohol (9.3 g) and added to
a mixture of potassium hydroxide (0.50 g, 9.0 mmol) in tert-
butyl alcohol (6.8 g). The reaction mixture was warmed to 50
°C and stirred for 3 h. After cooling the solid was extracted
with large volumes of chloroform and 1 M HCl (0 °C) was
added and a final extraction into the organic layer was
performed. Drying (MgSO4) and evaporation of the organic
layer under reduced pressure yielded a mixture of both
starting material and product. The solid was dissolved in
methanol and a small portion of silica gel was added. The silica
plug was loaded on a silica column and the mixture was
purified by flash chromatography [chloroform:methanol (39:1
with a gradient to 9:1)] yielding 1.35 g (85%) of 2 based on
consumed starting material: 1H NMR (CDCl3) δ 10.50 (br s
1H, COOH), 8.17-8.06 (4H, ArH), 1.61 (s, 9H); 13C NMR
(CDCl3) δ 171.32, 164.80, 136.62, 132.54, 130.01 (2C), 129.45
(2C), 81.89, 28.12.
pKa
bind
∆∆G (A f B) ) RT ln 10(pKa(A) - pKa(B)) (3a)
when both inhibitors A and B have pKa lower than the pH at
which the affinities are to be calculated. If inhibitor B has pKa
> pH then the correction becomes:
∆∆GpbiKnd(A f B) ) RT ln 10(pKa(A) - pH)
(3b)
a
The free energy contribution for protonation of a 2,4-
diaminopteridine to the binding free energy relative to a 2,4-
diaminoquinazoline is thus calculated using eq 3b. At pH 6.9
the correction is 1.6 kcal/mol (see Table 1). The processes of
transforming one inhibitor into another were carried out using
hybrid molecules including all atoms particular to each of the
two species, in addition to the atoms common to both. By
turning appropriate atoms into ‘dummy atoms’ (with no
interactions), in combination with altering partial charges,
Lennard-J ones parameters for atoms changing type, and
altering the parameters for bond angles, torsion, and improper
torsion angles, the two inhibitors are modeled in the frame-
work of the hybrid molecule. In this way the transformation
between the two end-points (states) representing the two
inhibitors can be carried out along a smooth pathway via
intermediate linear combinations defined by a series of values
for the coupling parameter λ. Very closely spaced such λ points
turned out to be necessary to achieve reasonable convergence;
116 λ values were used to connect the initial and final states.
Bonds to atoms transformed into dummy atoms were retracted
to shorter lengths as required when atoms from the middle of
a chain were removed. These bond lengths were constrained
to the average lengths in the two end-points, weighted by λ.
Interactions between atoms in 1-4 positions (connected via 2
atoms) across or with the atoms being transformed to dummy
atoms were eliminated since they would otherwise lead to large
artificial repulsion in the intermediate states.
The model of DHFR was taken from the crystal structure
of the ternary complex of the human L22Y mutant with
NADPH and MTX by Lewis et al. (PDB entry 1DLS)30 where
prior to simulations the wild-type enzyme complex was mod-
eled from this structure. Ionic groups on the protein within
about 10 Å from the inhibitor were included, whereas those
far from the inhibitor were modeled as dipolar.20 The net
charge of the protein + cofactor was zero. Ligands other than
MTX were docked using the FLEXX docking program.28
The program Q31 was used for the MD simulations employ-
ing the GROMOS87 force field.32 Simulations of the solvated
inhibitors were carried out in 18 Å radius spheres of rigid
SPC32 water molecules with one central atom of the inhibitor
restrained to the center of the sphere. The systems were
equilibrated for 50 ps at 300 K. In the simulations of solvated

Computational Prediction of Binding Affinity to DHFR
J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21 3859
2,4-Dia m in o-6-(br om om eth yl)qu in a zolin e (3a ). Com-
pound 3a was prepared according to a method described
elsewhere in the preparation of 3b34 adding 30% hydrobromic
acid in acetic acid (18 mL) to a solution consisting of 2,4-
diaminoquinazoline-6-methanol15 (0.100 g, 0.53 mmol) in acetic
acid (14 mL). After stirring at room temperature for 24 h in
the absence of light, the mixture was poured down to cold
diethyl ether with stirring and allowed to stand in the
refrigerator for a couple of hours before the precipitate was
collected under nitrogen and washed with cold diethyl ether.
The formed hydrobromic salt was dried in vacuo at 40 °C
before it was used in the next step without further purification.
(2C, Wang), 129.21 (2C), 127.82 (2C), 127.04 (Wang), 124.31
(2C), 114.25 (2C, Wang), 109.92, 80.68 (C(CH3)3), 69.76 (PEG),
68.23 (PEG), 67.09 (PEG), 66.86, 65.34 (Wang), 52.68 (CH),
30.21, 27.56 (C(CH3)3), 25.71.
N-[4-[(2,4-Dia m in oqu in a zolin -6-yl)m eth oxyca r bon yl]-
ben zoyl]-L-glu ta m ic Acid (1a ). The resin 7a (0.200 g, 0.42
mmol) was treated with 50% TFA/dichloromethane (6 mL) for
1.5 h. The product was filtered off and washed with dichlo-
romethane, DMF, methanol, and dichloromethane, respec-
tively. The first washing fraction from the washing with
dichloromethane was evaporated under reduced pressure. The
following workup procedure was the same as for compound
5a , yielding 36 mg of the final compound (92% over 4 steps):
ter t-Bu tyl 4-[(2,4-Dia m in oqu in a zolin -6-yl)m eth oxyca r -
bon yl]ben zoa te (4a ). Freshly prepared 3a (0.53 mmol) was
added portionwise to a mixture of compound 2 (0.351 g, 1.58
mmol) and cesium carbonate (0.514 g, 1.58 mmol) in anhydrous
DMF (15 mL). The reaction mixture was allowed to stir at
room temperature for 5 days before it was filtered. A small
portion of silica gel was added and the solvent was evaporated
under reduced pressure. The silica plug was added on the top
of a silica column and the product was purified by flash
chromatography [chloroform:methanol (39:1) + aqueous NH3]
providing 65 mg (31% over 2 steps) after a further purification
by circular chromatography [chloroform:methanol (39:1) +
1
IR (KBr) 1724 (ester), 1652 (amide) cm-1; H NMR (DMSO-
d6) δ 8.73 (d, J ) 7.75 Hz, 1H, NH), 8.20 (app d, 1H, H-5),
8.10-7.98 (m, 4 + 2H, ArH + NH2), 7.83 (br s, 2H, NH2), 7.74
(dd, J ) 8.58, 1.65 Hz, 1H, H-7), 7.32 (d, J ) 8.58 Hz, 1H,
H-8), 7.09 (br s, 2H, NH2), 5.37 (s, 2H), 4.42-4.34 (m, 1H),
2.35 (app t, 2H), 2.16-1.88 (m, 4H); 13C NMR (DMSO-d6) δ
174.06, 173.84, 165.51, 165.12, 162.65, 158.54, 138.42, 133.98,
131.77, 129.28 (2C), 128.93, 127.80 (2C), 124.29, 121.27,
109.46, 66.52, 52.44, 30.72, 26.31 (one aromatic carbon miss-
ing). Anal. (C22H21N5O7‚1H2O) C, H, N.
4-[(2,4-Dia m in op ter id in -6-yl)m eth oxyca r bon yl]ben zo-
ic Acid (5b). Compound 5b was prepared as described above
in the synthesis of 5a starting from 1.56 mmol of 2,4-diamino-
6-(bromomethyl)pteridine (3b)35 and compound 2 (643 mg, 2.89
mmol) in the presence of cesium carbonate (2.5 g, 7.67 mmol).
The reaction mixture was allowed to stir at room temperature
in dry DMF (15 mL) for 2 days. After filtration and purification
132 mg of 4b together with the byproduct in the reaction (2,4-
diaminopteridine 6-methylacetate) was collected and allowed
to react further with trifluoroacetic acid (7 mL). Workup and
purification was performed following the same procedure as
for 5a yielding 77 mg of 5b as the only product (10% over 2
steps): 1H NMR (DMSO-d6) δ 8.90 (s, 1H, H-7), 8.13-8.04 (m,
4H, ArH), 7.84 (br s, 2H, NH2), 6.86 (br s, 2H, NH2), 5.49 (s,
2H); 13C NMR (DMSO-d6) δ 167.13, 165.54, 163.37, 163.11,
155.52, 150.88, 143.44, 135.60, 133.39, 130.21 (4C), 122.25,
66.24. Anal. (C15H12N6O4‚0.25CF3CO2H‚0.5H2O) C, H, N.
1
aqueous NH3]: IR (KBr) 1715 (ester) cm-1; H NMR (DMF-
d6) δ 8.24 (d, J ) 1.81 Hz, 1H, H-5), 8.18-8.08 (m, 4H, ArH),
7.68 (dd, J ) 8.58, 1.81 Hz, 1H, H-7), 7.44 (br s, 2H, NH2),
7.29 (d, J ) 8.58 Hz, 1H, H-8), 6.10 (br s, 2H, NH2), 5.42 (s,
2H), 1.59 (s, 9H); 13C NMR (DMF-d6) δ 171.84, 165.61, 164.77,
163.43, 135.58, 136.11, 134.05, 133.22, 129.88 (2C), 129.73
(3C), 127.83, 125.43, 124.26, 110.64, 67.48, 27.71 (3C). Anal.
(C21H22N4O4) C, H, N.
4-[(2,4-Dia m in oqu in a zolin -6-yl)m eth oxyca r bon yl]ben -
zoic Acid (5a ). Compound 4a (39 mg, 0.10 mmol) in trifluo-
roacetic acid (1 mL) was allowed to stir at room temperature
for 2 h before the solvent was evaporated under reduced
pressure. The crude product was triturated with diethyl ether
and the formed precipitate was filtered off. Water was added
and the pH was adjusted to 3.8. The flask was stored in the
refrigerator overnight, filtered and washed with water, ace-
tone, and diethyl ether, respectively, providing 22 mg (67%)
of a white solid: IR (KBr) 1713 (ester) cm-1; 1H NMR (DMSO-
d6 + 1 drop of DCO2D) δ 8.32 (app s, 1H, H-5), 8.11-8.05 (m,
4H, ArH), 7.90-7.88 (m, 1H), 7.48-7.45 (m, 1H), 5.41 (s, 2H);
13C NMR (DMSO-d6 + 1 drop of DCO2D) δ 166.94, 165.50,
163.24, 154.88, 139.29, 135.75, 135.25, 133.33, 132.46, 129.91
(4C), 124.81, 117.53, 109.53, 66.32. Anal. (C17H14N4O4‚0.5CF3-
CO2H‚1H2O) C, H, N.
P olym er -Su p p or ted 2,4-Dia m in op ter id in e Der iva tive
7b. Compound 7b was prepared as described for 7a starting
from 6 (150 mg, 63 µmol) and 5b (77 mg, 2.26 mmol): 13C NMR
(DMSO-d6) δ 172.42, 170.88, 166.17, 165.16, 163.40, 163.11,
158.44 (Wang), 155.86, 150.61, 142.60, 138.25, 134.91, 131.71
(Wang), 129.95 (Wang + 2C), 129.52 (Wang + 2C), 122.11,
114.35 (Wang), 81.11 (C(CH3)3), 69.90, 67.20, 65.73, 62.71,
52.90, 30.41, 27.68 (C(CH3)3), 25.88.
P olym er -Su p p or ted Glu ta m ic Acid Der iva tive 6. Argo-
Gel-Wang-OH (0.5 g, 0.21 mmol) was allowed to swell in
dichloromethane (8 mL). Fmoc-Glu-OtBu (0.45 g, 1.05 mmol),
N,N′-diisopropylcarbodiimide (163 µL, 1.05 mmol) and 4-pyr-
rolidinopyridine (15 mg, 0.10 mmol) were added and the vessel
was capped and placed in an overhead mixer for 4 h. The
reaction liquors were removed by filtering and the resin was
washed with dichloromethane, methanol, DMF and dichlo-
romethane, respectively. A mixture of 25% piperidine in DMF
was added to cleave off the Fmoc protecting group and after
15 min the solid phase was washed again with DMF, metha-
nol, and dichloromethane, respectively and dried in vacuo
overnight: 13C NMR (DMSO-d6) δ 174.83, 173.09, 158.81
(Wang), 129.97 (2C, Wang), 128.25 (Wang), 114.57 (2C, Wang),
81.11 (C(CH3)3), 70.56 (PEG), 67.43 (PEG), 66.04 (Wang), 54.26
(CH), 30.73, 29.85, 28.04 (C(CH3)3). Amino acid analysis: Glu,
0.42 mmol/g.
P olym er -Su p p or ted 2,4-Dia m in oqu in a zolin e Der iva -
tive 7a . The resin 6 (0.200 g, 84 µmol) was swelled for about
10 min in DMF (7 mL). PyBOP (131 mg, 0.25 mmol), N,N-
diisopropylethylamine (88 µL, 0.50 mmol) and compound 5a
(85 mg, 0.25 mmol) were added. The reaction proceeded for 4
h before the suspension was filtered off. The resin was washed
with DMF, methanol, and dichloromethane, respectively: 13C
NMR (DMSO-d6) δ 172.07, 170.67, 165.84, 165.13, 162.49,
160.94 (Wang), 158.34, 152.31, 138.01, 133.00, 131.99, 129.82
N-[4-[(2,4-Diam in opter idin -6-yl)m eth oxycar bon yl]ben -
zoyl]-L-glu ta m ic Acid (1b). The resin 7b was treated with
50% TFA/dichloromethane as in the case for 7a , yielding 14
mg of the final compound: 1H NMR (DMSO-d6) δ 8.90 (s, 1H,
H-8), 8.79 (d, J ) 7.92 Hz, 1H, NH), 8.12-7.98 (m, 4H, ArH),
7.68 (br s, 2H, NH2), 6.74 (br s, 2H, NH2), 5.48 (s, 2H), 4.45-
4.36 (m, 1H), 2.38-2.33 (m, 2H), 2.12-1.91 (m, 2H); 13C NMR
(DMSO-d6) δ 173.82, 173.16, 165.59, 165.05, 163.24, 162.80,
155.74, 150.33, 142.35, 138.22, 131.58, 129.35 (2C), 127.85
(2C), 121.60, 65.64, 52.11, 30.53, 25.98. Anal. (C20H19N7O7‚
0.25CF3CO2H‚0.75H2O) C, H; N: calcd, 19.2; found, 18.5.
Dih yd r ofola te Red u cta se Assa y. A plasmid with a cloned
gene encoding hDHFR was a generous gift from Raymond
Blakley (St. J ude Children’s Research Hospital). Plasmid DNA
was introduced into the E. coli strain DH5R36 by transforma-
tion. A 400-mL culture of transformed bacteria was grown in
LB medium37 to 2 × 108 cells, and expression of hDHFR was
induced by addition of isopropyl â-D-thiogalactopyranoside to
1 mM. Four hours after induction the cells were harvested by
centrifugation, washed with 100 mL of buffer A (0.05 M Tris-
hydrochloride (pH 7.2), 0.05 M potassium chloride, 1 mM
dithiothreitol, 1 mM disodium EDTA), recentrifuged and
finally resuspended in 3 mL of buffer A. The cells were
disrupted by sonication and the DHFR was purified further
as described by Tennhammar-Ekman and Sko¨ld.38 Reduced
nicotinamide adenine dinucleotide phosphate (NADPH) was

3860 J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21
Graffner-Nordberg et al.
obtained from Sigma. Dihydrofolic acid was prepared according
to Blakley39 and concentrations was determined by UV absorp-
tion using ∆ꢀ285 of 20 400 M-1 cm-1. The buffer used through-
out the studies consisted of 0.02 M phosphate (potassium), pH
6.9, 1 mM dithiothreitol, 1 mM ethylenediaminetetraacetic
acid (EDTA). Final concentrations (in 2 mL of buffer) were 60
µM NADPH, 36 µM dihydrofolate, and different concentrations
of the inhibitor. DHFR activity was observed by following the
decline in absorbance at 340 nm (NADPH maximum) using a
spectrophotometric assay monitoring the decrease in absor-
bance of the assay mixture at 30 °C with a Beckman DU 62
spectrophotometer. The cuvettes had a light path of 1 cm. Each
compound was prepared as a stock solution in buffer A. In a
few cases DMSO was needed, but the final concentrations of
DMSO in the enzymatic assay were in these cases less than
1%. Assays were performed, by incubating the enzyme,
NADPH, and the inhibitor in the buffer for 3 min before
dihydrofolate was added. The initial reaction rates were
determined and monitored continuously for 5 min. The activity
of recombinant human DHFR was assayed without inhibitor,
together with a series of concentrations suitable for IC50
determinations. The specific activity of the enzyme was 7.8
units/mg of protein, where 1 unit is defined as ∆340 nm/min.
Triplicate determinations were done at each concentration, and
the results were averaged. The concentrations of inhibitor
required to reduce DHFR activity by 50% (IC50) were deter-
mined by semilogarithmic plots of the data yielding normal
sigmoidal curves for most inhibitors using GraphPad Prism
(version 2.01 for Windows; GraphPad Software Inc., San Diego,
CA).
folate-dependent, possibly at the mucosal level. In the present
experiments we attempted to compare ip administration of
compound 1a and MTX to oral administration, which could
give hints to the site of action. We chose an oral dosing that
previously had been proven to be effective (4 mg/kg/week,
starting 1 week before challenge). The ip dose was chosen
based on results in preliminary experiments (0.15 mg/kg/day,
starting on the day of challenge). Since we did not know the
rate of polyglutamation for compound 1a and the bioavail-
ability seemed to be lower than that for MTX (see Table 3),
we administered 1a daily, given by ip injections. Compound
1a was given in a high dose (3.75 mg/kg/day) in order to
maximize the chance to detect an antiarthritic effect.
AIA was induced in female Dark Agouti rats with an
approximate body weight of 165 g. The rats were obtained from
Mo¨llegaards Breeding Center (Denmark) and kept in sawdust-
covered cages with food and water ad libitum. The temperature
was thermostatically maintained at 22 °C, the relative humid-
ity was 50%, and the light was on in 12-h periods from 6 a.m.
to 6 p.m. The rats were conditioned at least one week before
immunization. The experiments were approved by the Animal
Ethics Committee of Lund, Sweden. The animals were killed
after the last day of measurement.
For induction of AIA the rats were sensitized intradermally
at the tail root with 1 mg of methylated bovine serum albumin
(mBSA) dissolved in 50 µL of saline and emulsified in 50 µL
of Freund’s complete adjuvant. Ten days later the rats were
challenged with a unilateral intraarticular injection of 50 µg
of mBSA (1 µg/µL, dissolved in saline). The intensity of the
arthritis was quantified by measurement of joint swelling. This
measure mainly reflects the pannus mass.41 The same measure
is also important for the assessment of clinical RA develop-
ment. In the AIA model this parameter has a low variance
and it is the only nonradiological method to follow the joint
process longitudinally in single animals. For each day of
measurement the increase in joint diameter for each animal
was calculated. The individual values were then used in the
statistical analysis. Unless stated otherwise, the area under
the curve (AUC) was calculated for each animal and these
values were used in the statistical analysis, which was
performed by analysis of variance followed by the Fisher
protected least-significant difference (PLSD). Calculations
were performed on an Apple computer using StatView 4.0
(Berkeley, CA) as software. In all experiments each group
contained six animals and each experiment contained a
vehicle-treated control group. All presented data in Figure 5
are from a separate experiment and are given as mean ( SEM.
P h a r m a cok in etics in Ra ts. The pharmacokinetics of MTX
and 1a were evaluated in male SD rats, weighing 233-264 g,
after intravenous (1 µmol/kg) and oral (10 µmol/kg) adminis-
tration, three animals for each route of administration. The
test compounds were dissolved in a DMA:PEG400:citrate
buffer (40:40:20) vehicle to the concentration of 2 µmol/mL.
The intravenous dose (100 µL) was given as a bolus into a tail
vein. The oral dose (1 mL) was given by gavage. Blood was
collected by repeated sampling from the tail vein opposite to
the intravenous administration site at 2, 10, 30, 60, 120, 240,
and 360 min. No anaesthetics were used during the experi-
ment. Plasma, prepared by centrifugation of the blood for 5
min at 3000g in 4 °C, was precipitated with ethanol and the
supernatant was analyzed for content of compound using LC-
MS/MS technique. In the case of compound 1a , the metabolites
were also quantified. As standards for the bioanalysis, the test
formulations were added to blank plasma and diluted stepwise
giving rise to standard curves ranging from 2000 to 6000 nM.
The concentration of compound in the test formulations was
determined before and after each experiment by LC-MS. For
calculation of the pharmacokinetic parameters, the noncom-
partmental analysis model in WinNonlin Professional (version
2.1; Scientific Consulting) was used.
An tigen -In du ced Ar th r itis (AIA): In du ction an d Evalu -
a tion . MTX is presently the drug of choice for the treatment
of rheumatoid arthritis (RA). MTX is one of the few antirheu-
matoid drugs that works both in animal models and patients.
The models thus offer suitable in vivo systems for evaluation
of MTX and related analogues. One such model is the AIA in
rats. The acute joint inflammation in this model resembles in
several respects the corresponding reaction in RA patients.40-43
AIA is ameliorated by MTX when administrated in a similar
way as in the clinic (weekly oral dosing). The exact mechanism
of action is not established, but since folate supplementation
in excess can override the antiarthritic effect of MTX, it is
concluded that folate antagonism is essential.26 The effect of
weekly dosing which propably indicates that it is trapped
intracellularly due to polyglutamation.44
Ackn owledgm en t. We thank Prof. Sherry F. Queen-
er for skillful in vitro binding experiments on rat liver
DHFR. We also thank the Swedish Research Council
for Engineering Sciences (TFR), the Swedish Natural
Science Research Council (NFR), and the Swedish
Foundation for Strategic Research (SSF) for financial
support.
Refer en ces
(1) Andersen, P. A.; West, S. G.; O’Dell, J . R.; Via, C. S.; Claypool,
R. G.; Kotzin, B. L. Weekly Pulse Methotrexate in Rheumatoid
Arthritis. Clinical and Immunologic Effects in a Randomized,
Double-Blind Study. Ann. Intern. Med. 1985, 103, 489-496.
(2) Williams, H. J .; Willkens, R. F.; Samuelson, C. O. J .; Alarco´n,
G. S.; Guttadauria, M.; Yarboro, C.; Polisson, R. P.; Weiner, S.
R.; Luggen, M. E.; Billingsley, L. M.; Dahl, S. L.; Egger, M. J .;
Reading, J . C.; Ward, J . R. Comparison of Low-Dose Oral Pulse
Methotrexate and Placebo in the Treatment of Rheumatoid
Arthritis. A Controlled Clinical Trial. Arthritis Rheum. 1985,
28, 721-730.
(3) Weinblatt, M. E.; Coblyn, J . S.; Fox, D. A.; Fraser, P. A.;
Holdsworth, D. E.; Glass, D. N.; Trentham, D. E. Efficacy of Low-
Dose Methotrexate in Rheumatoid Arthritis. N. Engl. J . Med.
1985, 312, 818-822.
(4) Van Scott, E. J .; Auerbach, R.; Weinstein, G. D. Parenteral
Methotrexate in Psoriasis. Arch. Dermatol. 1964, 89, 550-556.
(5) Rees, R. B.; Bennett, J . H.; Maibach, H. I.; Arnold, H. L.
Methotrexate for Psoriasis. Arch. Dermatol. 1967, 95, 2-11.
The AIA model is also sensitive to oral tolerance: that is,
inhibition of arthritis by oral feeding of the arthritogenic
antigen prior to immunization, indicating a possible connection
between lymphoid organs in the gut and the antigen-present-
ing cells in the joint. Oral tolerance has been shown to be
abrogated by MTX45 which suggests that this connection is

Computational Prediction of Binding Affinity to DHFR
J ournal of Medicinal Chemistry, 2000, Vol. 43, No. 21 3861
(6) Weinstein, G. D.; Frost, P. Methotrexate for Psoriasis. A New
Therapeutic Schedule. Arch. Dermatol. 1971, 103, 33-38.
(7) Mullarkey, M. F.; Blumenstein, B. A.; Andrade, W. P.; Bailey,
G. A.; Olason, I.; Wetzel, C. E. Methotrexate in the Treatment
of Corticosteroid-Dependent Asthma. A Double-Blind Crossover
Study. N. Engl. J . Med. 1988, 318, 603-607.
(28) Rarey, M.; Kramer, B.; Lengauer, T.; Klebe, G. A Fast Flexible
Docking Method using an Incremental Construction Algorithm.
J . Mol. Biol. 1996, 261, 470-489.
(29) Cocco, L.; Roth, B.; Temple, C. J .; Montgomery, J . A.; London,
R. E.; Blakley, R. L. Protonated State of Methotrexate, Trimetho-
prim, and Pyrimethamine Bound to Dihydrofolate Reductase.
Arch. Biochem. Biophys. 1983, 226, 567-577.
(30) Lewis, W. S.; Cody, V.; Galitsky, N.; Luft, J . R.; Pangborn, W.;
Chunduru, S. K.; Spencer, H. T.; Appleman, J . R.; Blakley, R.
L. Methotrexate-resistant Variants of Human Dihydrofolate
Reductase with Substitutions of Leucine 22: Kinetics, Crystal-
lography, and Potential as Selectable Markers. J . Biol. Chem.
1995, 270, 5057-5064.
(31) Marelius, J .; Kolmodin, K.; Feierberg, I.; Åqvist, J . Q: An
Molecular Dynamics Program for Free Energy Calculations and
Empirical Valence Bond Simulations in Biomolecular Systems.
J . Mol. Graph. Model. 1998, 16, 213-225.
(32) van Gunsteren, W. F.; Berendsen, H. J . C. Groningen Molecular
Simulation (GROMOS) Library Manual; Biomos B.V.: Gronin-
gen, 1987.
(33) Cramer, C. J .; Truhlar, D. G. An SCF Solvation Model for the
Hydrophobic Effect and Absolute Free Energies of Aqueous
Solvation. Science 1992, 256, 213-217.
(34) Piper, J . R.; J ohnson, C. A.; Maddry, J . A.; Malik, N. D.; McGuire,
J . J .; Otter, G. M.; Sirotnak, F. M. Studies on Analogues of
Classical Antifolates Bearing the Naphthoyl Group in Place of
Benzoyl in the Side Chain. J . Med. Chem. 1993, 36, 4161-4171.
(35) Piper, J . R.; Montgomery, J . A. Preparation of 6-(Bromomethyl)-
2,4-pteridinediamine Hydrochloride and Its Use in Improved
Synthesis of Methotrexate and Related Compounds. J . Org.
Chem. 1977, 42, 208-211.
(36) Hanahan, D. Studies on Transformation of Escherichia coli with
Plasmids. J . Mol. Biol. 1983, 166, 557-580.
(37) Bertani, G. Studies on Lysogenesis. I. The Mode of Phage
Liberation by Lysogenic Escherichia coli. J . Bacteriol. 1951, 62,
293-300.
(38) Tennhammar-Ekman, B.; Sko¨ld, O. Trimethoprim Resistance
Plasmids of Different Origin Encode Different Drug-Resistant
Dihydrofolate Reductases. Plasmid 1979, 2, 334-346.
(39) Blakley, R. L. Crystalline Dihydropteroylglutamic Acid. Nature
1960, 188, 231-232.
(40) Andersson, S. E.; Lexmu¨ller, K.; Ekstro¨m, G. M. Physiological
Characterisation of mBSA Antigen-Induced Arthritis in the Rat.
I. Vascular Leakiness and Pannus Growth. J . Rheumatol. 1998,
25, 1772-1777.
(41) Andersson, S. E.; J ohansson, A.; Lexmu¨ller, K.; Ekstro¨m, G. M.
Physiological Characterization of mBSA Antigen Induced Ar-
thritis in the Rat. II. J oint Blood Flow, Glucose Metabolism, and
Cell Proliferation. J . Rheumatol. 1998, 25, 1778-1784.
(42) Andersson, S. E.; Lexmu¨ller, K.; J ohansson, A.; Ekstro¨m, G. M.
Tissue and Intracellular pH in Normal Periarticular Soft Tissue
and During Different Phases of Antigen-Induced Arthritis in the
Rat. J . Rheumatol. 1999, 26, 2018-2024.
(8) Mullarkey, M. F.; Webb, D. R.; Pardee, N. E. Methotrexate in
the Treatment of Steroid-Dependent Asthma. Ann. Allergy 1986,
56, 347-350.
(9) Kozarek, R. A.; Patterson, D. J .; Gelfand, M. D.; Botoman, V.
A.; Ball, T. J .; Wilske, K. R. Methotrexate Induces Clinical and
Histologic Remission in Patients with Refractory Inflammatory
Bowel Disease. Ann. Intern. Med. 1989, 110, 353-356.
(10) Peppercorn, M. A. Advances in Drug Therapy for Inflammatory
Bowel Disease. Ann. Intern. Med. 1990, 112, 50-60.
(11) DeGraw, J . I.; Colwell, W. T.; Piper, J . R.; Sirotnak, F. M.; Smith,
R. L. New Analogs of Methotrexate in Cancer and Arthritis.
Curr. Med. Chem. 1995, 2, 630-653.
(12) Klepser, M. E.; Klepser, T. B. Drug Treatment of HIV-Related
Opportunistic Infections. Drugs 1997, 53, 40-73.
(13) Queener, S. F.; Bartlett, M. S.; J ay, M. A.; Durkin, M. M.; Smith,
J . W. Activity of Lipid-Soluble Inhibitors of Dihydrofolate
Reductase Against Pneumocystis carinii in Culture and in a Rat
Model of Infection. Antimicrob. Agents Chemother. 1987, 31,
1323-1327.
(14) Kovacs, J . A.; Allegra, C. J .; Swan, J . C.; Drake, J . C.; Parrillo,
J . E.; Chabner, B. A.; Masur, H. Potent Antipneumocystis and
Antitoxoplasma Activites of Piritrexim, a Lipid Soluble Anti-
folate. Antimicrob. Agents Chemother. 1988, 32, 430-433.
(15) Graffner-Nordberg, M.; Sjo¨din, K.; Tunek, A.; Hallberg, A.
Synthesis and Enzymatic Hydrolysis of Esters, Constituting
Simple Models of Soft Drugs. Chem. Pharm. Bull. (Tokyo) 1998,
46, 591-601.
(16) Lamb, M. L.; J orgensen, W. L. Computational Approaches to
Molecular Recognition. Curr. Opin. Chem. Biol. 1997, 1, 449-
457.
(17) Oprea, T. I.; Marshall, G. R. Receptor-Based Prediction of
Binding Affinities. Perspect. Drug Discovery Des. 1998, 9, 35-
61.
(18) Hansson, T.; Marelius, J .; Åqvist, J . Ligand Binding Affinity
Prediction by Linear Interaction Energy Methods. J . Comput.-
Aided Mol. Des. 1998, 12, 27-35.
(19) Kollman, P. Free Energy Calculations: Applications to Chemical
and Biochemical Phenomena. Chem. Rev. 1993, 93, 2395-2417.
(20) Marelius, J .; Graffner-Nordberg, M.; Hansson, T.; Hallberg, A.;
Åqvist, J . Computation of Affinity and Selectivity: Binding of
2,4-Diaminopteridine and 2,4-Diaminoquinazoline Inhibitors to
Dihydrofolate Reductases. J . Comput.-Aided Mol. Des. 1998, 12,
119-131.
(21) Buckle, D. R.; Smith, H. An Improved Synthesis of Substituted
Benzoyl Acetates. J . Chem. Soc. 1971, 2821-2823.
(22) Åkerblom, E. B.; Nygren, A. S.; Agback, K. H. Six New Photo-
labile Linkers for Solid-Phase Synthesis. 1. Methods of Prepara-
tion. Mol. Divers. 1998, 3, 137-148.
(23) Genestier, L.; Paillot, R.; Fournel, S.; Ferraro, C.; Miossec, P.;
Revillard, J .-P. Immunosuppressive Properties of Methotrex-
ate: Apoptosis and Clonal Deletion of Activated Peripheral T
cells. J . Clin. Invest. 1998, 102, 322-328.
(24) Bannwarth, B.; Labat, L.; Moride, Y.; Schaeverbeke, T. Meth-
otrexate in Rheumatoid Arthritis. An Update. Drugs 1994, 47,
25-50.
(43) Andersson, S. E.; Lexmu¨ller, K.; Alving, K.; Ekstro¨m, G. M.
Periarticular Tissue Levels of Cytokine- and Endothelin-1-like
Immunoreactivity During the Course of Antigen-induced Ar-
thritis in the Rat. Inflamm. Res. 1999, 48, 491-496.
(44) Cichowicz, D. J .; Shane, B. Mammalian Folylpoly-γ-glutamate
Synthetase. 1. Purification and General Properties of the Hog
Liver Enzyme. Biochemistry 1987, 26, 504-512.
(25) Cronstein, B. N. Molecular Therapeutics. Methotrexate and its
Mechanism of Action. Arthritis Rheum. 1996, 39, 1951-1960.
(26) Andersson, S. E.; J ohansson, L.-H.; Lexmu¨ller, K.; Ekstro¨m, G.
M. Anti-Arthritic Effect of Methotrexate: Is It Really Mediated
by Adenosine? Eur. J . Pharm. Sci. 2000, 9, 333-343.
(27) Cummins, P. L.; Gready, J . E. Computer-Aided Drug Design:
A Free Energy Perturbation Study on the Binding of Methyl-
substituted Pterins and N5-Deazapterins to Dihydrofolate Re-
ductase. J . Comput-Aided Mol. Des. 1993, 7, 535-555.
(45) Ekstro¨m, G. M.; Szajewski, K.; Persson, Å. Commonly Used RA
Drugs Abrogate Oral Tolerance in Antigen-Induces Arthritis
(AIA). Scand. J . Immunol. 1997, 45, 566.
(46) Broughton, M. C.; Queener, S. F. Pneumocystis carinii Dihydro-
folate Reductase Used to Scren Potential Antipneumocystis
Drugs. Antimicrob. Agents Chemother. 1991, 35, 1348-1355.
J M0009639