J. Leban et al. / Bioorg. Med. Chem. Lett. 14 (2004) 55–58
57
set, and a new lead compound, 80 times more active
than 1, emerged (Fig. 2).
The analysis of the QSAR model gave further hints for
important featuresof a good DHODH inhibitor:
Activity increased with rising total van der Waals sur-
face area aswell asnegatively charged and hydrophobic
van der Waal sus rface area. Negative contributions
were attributed to increased molecular weight and
volume.
To shed further light on the binding mode of these
compoundswe co-cr yt sa llized the compoundswith
truncated human DHODH and solved the structure.
The compoundshave been localized in the binding si te
according to the expectation, however subtle differences
in binding between compoundsmake os me bind more
like A771726 and otherslike brequinar (unpubl ihs ed
observation). We are also in the process to characterize
the kinetic of the enzyme inhibition. Analogueswith
heteroatom substitutions in the pentacyclic ring, side
chainson the pentacyclic ring and other hydrophobic
residues in place of the biphenylic ring system were also
prepared and tested. These findings and details of
biological data will be presented in future publications.
0
Figure 1. Reagentsand condition s: (a) Pd, MeOH, KF, reflux, 8 h; (a
Pd-tetrakis, DME/water (75/25), Cs CO , reflux, 8 h; (b) DCM, rt,
6 h.
)
2
3
1
In conclusion, we used a structure-based, combined in
silico and medicinal chemistry approach to discover
novel and potent inhibitorsof human DHODH. Such
compounds promise to become useful therapeutic
agents to treat autoimmune diseases such as rheumatoid
arthritis, multiple sclerosis, lupus erythematosus and
ulcerative colitis.
Acknowledgements
We thank Martin Kralik, Jan Mies, Tanja Wieber,
Heike Endreß and Larissa Bahrjanyj for technical
assistance, Matthias Dormeyer, Michael Gassen and
Kristina Wolf for helpful discussions, Gerhard Keil-
hauer and Daniel Vitt for support.
Figure 2. Calculated versus measured IC50 of enzyme inhibition.
Compoundsli st ed in Table 1 were numbered.
References and notes
with the core scaffold of 1 and bearing various us b-
stituents on the aromatic rings. First, a training set was
generated by synthesis and biological testing of 55
compounds was performed. These subset of compounds
was sufficiently diverse to allow the setup of a QSAR
model. Each molecule wascharacterized by se venty 2D
and internal coordinate dependent 3-D descriptors
1
2
3
¨ ¨
. Jockel, J.; Wendt, B.; Loffler, M. Biochem. Pharmacol.
998, 56, 1053.
. Fairbanks, L. D.; Bofill, M.; Ruckemann, K.; Simmonds,
H. A. J. Biol. Chem. 1995, 270, 29682.
. Alldred, A.; Emery, P. Exp. Opin. Pharmacother. 2001, 2, 125.
4. Sutton, A. E.; Clardy, J. Tetrahedron Lett. 2001, 42, 547.
Pospisil, P.; Kuoni, T.; Scapozza, L.; Folkers, G. J.
Recept. Signal Transduction Res. 2002, 22, 1, 141–154.
. Liu, S.; Neidhardt, E. A.; Grossman, T. H.; Ocain, T.;
Clardy, J. Struct. Fold Des. 2000, 8, 25.
. McLean, J. E.; Neidhardt, E. A.; Grossman, T. H.; Hed-
strom, L. Biochemistry 2001, 40, 2194.
. Knecht, W.; Loffler, M. FEBS Lett. 2000, 467, 27.
¨
. Davis, J. P.; Cain, G. A.; Pitts, W. J.; Magolda, R. L.;
Copeland, R. A. Biochemistry 1996, 35, 1270.
. deJulian-Ortiz, J. V.; Galvez, J.; Munoz-Collado, C.;
Garzia-Domenech, R.; Garcia-Domenech, R.; Gimeno-
Cardona, C. J. Med. Chem. 1999, 42, 3308.
1
1
3
delivered from the MOE package. Partial least square
analysis (PLS) of a subset of 44 molecules and their
log(IC ) lead to a cross-validated correlation coefficient
5
5
0
of 0.60 and a root mean square error of 0.36. The opti-
mal number of components used was seven. The result-
ing QSAR model allowed usthe prediction of log(IC 50)
valuesfor the complete virtual library and the se lection
of 30 high scoring compounds for subsequent synthesis.
Their average logarithmic activity was1–2-fold higher,
compared to the training set. Seven new molecules
showed activities even beyond (higher than) the training
6
7
8
9