80
C. Sehon et al. / Bioorg. Med. Chem. Lett. 16 (2006) 77–80
puts for QSAR regression analysis. These descriptors of-
fer the advantage of requiring less computational time
to generate than whole molecule descriptors. However,
Cammareta6 has eloquently outlined the dangers of
using such descriptors when a system is not additive.
Thus, any linear regression analysis that uses exclusively
fragment-based descriptors cannot by definition be any
more than anecdotally relevant if the system in question
is found to be non-additive. In fact, for libraries such as
those we consider here, it can be shown that the additive
model described in the previous paper represents an
upper limit to the accuracy obtainable through frag-
ment-based QSAR. To adequately describe non-additive
effects in this case, for instance, descriptors that simulta-
neously treat both rings A and B are required.
0.6
0.4
0.2
0
∆ pKI
-0.2
-0.4
-0.6
-0.8
A
B
C
Ar2
D
F
E
E
D
Ar1
C
B
A
Another advantage of determining the inherent additiv-
ity in a system is a more practical one. It is often neces-
sary for the medicinal chemist to optimize multiple
properties in a molecular series simultaneously (i.e.,
activity and bioavailability). In these instances, knowl-
edge about a systemꢁs additivity would serve to provide
confidence that when one portion of a series is altered to
optimize a second property, the SAR trends for the pri-
mary target will not be altered. Conversely if a system
has a low degree of additivity then it might be advanta-
geous to adopt a strategy of combinatorial analoging to
avoid missing key parts of the SAR.
Figure 5. Difference between experimental pKI and that predicted by
the best least-squares fit to the additive model for each compound in
the matrix.
Ar1
N
N
O
Ar2
8.3
OH
7.8
Me
7.3
pKI
6.8
6.3
5.8
Acknowledgments
We thank Dale Boger for helpful discussions on the syn-
thesis of compound 4, and Jiejun Wu and Heather
McAllister for analytical support.
Me
O
MeSO
2
N
O
Ar1
Ar2
Me N
2
Figure 6. CCK1 binding affinity for library 2.
Supplementary data
An additional library was made which more clearly illus-
trates the extent of the non-additive relationships in this
series of CCK1 receptor antagonists (Fig. 6). In this
case, when Ar2 is naphthyl or methylenedioxyphenyl
the SAR resulting from changes in Ar1 is roughly flat.
However, when Ar2 is dimethylaminophenyl the SAR
is quite pronounced, resulting in compounds having
>300-fold differences in binding affinity. These results
demonstrate the potential magnitude of non-additive
SAR. It is clear from these results that non-additive ef-
fects can have a significant impact on medicinal chemis-
try programs that are managed with the expectation of
additive behavior, and illustrates the usefulness of gener-
ating a combinatorial matrix of compounds to verify
additive behavior before embarking on linear analoging.
Supplementary data associated with this article can be
References and notes
1. For recent comprehensive reviews of CCK pharmacology
and medicinal chemistry see (a) Herranz, R. Med. Res. Rev.
2003, 23, 559; (b) Tullio, P.; Delarge, J.; Pirotte, B. Exp.
Opin. Investig. Drugs 2000, 9, 129.
2. Varga, G. Curr. Opin. Invest. Drugs 2002, 3, 621.
3. McClure, K.; Huang, L.; Sehon, C.; Hack, M.; Morton,
M.; Barrett, T.; Shankley, N.; Breitenbucher, J. G. Bioorg.
Med. Chem. Lett. 2005, accompanying paper.
4. Shen, D.-M.; Shu, M.; Chapman, K. T. Org. Lett. 2000, 2,
2789.
This case also demonstrates the importance of exercising
caution when using certain QSAR models to predict bio-
logical activity. For instance, fragment-based descrip-
tors such as TPSA, C log P, MW, H-bond donors, and
acceptors, etc., have become increasingly common as in-
5. Plucinska, K.; Kataoka, T.; Yodo, M.; Cody, W. L.; He, J.
X.; Humblot, C.; Lu, G. H.; Lunney, E.; Major, T. C.;
Panek, R. T.; Schelkum, P.; Skeean, R.; Marshall, G. R.
J. Med. Chem. 1993, 36, 1902.
6. Cammareta, A. J. Med. Chem. 1972, 15, 573.