Synthesis and preliminary evaluation of a library of polycyclic small molecules for use in chemical genetic assays

Article abstract of DOI:10.1021/ja992144n

(-)-Shikimic acid, 3, was converted into both enantiomers of epoxycyclohexenol carboxylic acid, 7, which were attached to a solid support via a photocleavable linker. Tandem acylation-1,3-dipolar cycloaddition with nitrones 11a-g yielded tetracyclic templates 12a-g. After development of several efficient coupling reactions of iodobenzyl tetracycles 12b-d and completion of extensive validation protocols, a splitpool synthesis yielded a binary encoded library calculated to contain 2.18 million polycyclic compounds. These compounds are compatible with miniaturized cell-based 'forward' chemical genetic assays designed to explore biological pathways and 'reverse' chemical genetic assays designed to explore protein function. As a simple illustration of the potential of these compounds, several were shown to activate a TGF-β-responsive reporter gene in mammalian cells.

Full text of DOI:10.1021/ja992144n

J. Am. Chem. Soc. 1999, 121, 9073-9087
9073
Synthesis and Preliminary Evaluation of a Library of Polycyclic
Small Molecules for Use in Chemical Genetic Assays
Derek S. Tan, Michael A. Foley, Brent R. Stockwell, Matthew D. Shair, and
Stuart L. Schreiber*
Contribution from the Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology,
and HarVard Institute of Chemistry and Cell Biology, HarVard UniVersity, 12 Oxford St.,
Cambridge, Massachusetts 02138
ReceiVed June 23, 1999
Abstract: (-)-Shikimic acid, 3, was converted into both enantiomers of epoxycyclohexenol carboxylic acid,
7, which were attached to a solid support via a photocleavable linker. Tandem acylation-1,3-dipolar
cycloaddition with nitrones 11a-g yielded tetracyclic templates 12a-g. After development of several efficient
coupling reactions of iodobenzyl tetracycles 12b-d and completion of extensive validation protocols, a split-
pool synthesis yielded a binary encoded library calculated to contain 2.18 million polycyclic compounds.
These compounds are compatible with miniaturized cell-based “forward” chemical genetic assays designed to
explore biological pathways and “reverse” chemical genetic assays designed to explore protein function. As
a simple illustration of the potential of these compounds, several were shown to activate a TGF-â-responsive
reporter gene in mammalian cells.
Introduction
Biological systems can be explored using genetic methods,
where mutations in genes are introduced and the resulting
biological effects are examined. In the classical or “forward
genetic” approach, a large collection of randomly generated
mutations is introduced and subsequently screened in search of
a specific biological effect. The mutated gene is then identified
and the roles of its cognate wild-type gene and encoded protein
product in the biological pathway are investigated further.
Progress in whole genome sequencing1 has facilitated a
related method of exploring biological systems. Such sequencing
efforts are uncovering myriad novel genes with unknown
functions. In the “reverse genetic” approach, a deletion or
“knockout” mutation is targeted to a known gene of unknown
function. This is followed by a broad search for all resulting
biological effects, allowing the function of the gene to be
inferred.
The chemical genetic approach2,3 is complementary to the
genetic approach (Figure 1). Instead of mutations, small
molecules are used to activate or inactivate proteins by direct
interactions. The “forward chemical genetic” approach involves
the search for small molecules, beginning with a large random
collection, that modulate a specific biological pathway or
process. The active small molecule and its protein partner are
then identified, leading to an understanding of the protein’s role
in the pathway. In the “reverse chemical genetic” approach, a
small molecule having a known and specific protein target is
used to alter the function of its target. By determining the
pathways and processes altered by the small molecule, the
functions of its target can be inferred.
Figure 1. Genetic and chemical genetic approaches.
Chemical genetic approaches allow protein function to be
altered rapidly and conditionally. They are also dependent upon
the identification of small molecules that bind to proteins with
high specificity. Natural products and natural product-derived
compounds provide many of the most striking examples,
particularly in terms of their specificity (Figure 2). Examples
of interest to our laboratory include rapamycin, which led to
the discovery of the immunophilin FKBP12 and the amino acid
sensor FRAP,4,5 and trapoxin, which led to the discovery of
the chromatin-remodeling histone deacetylases (HDACs)6,7 and
nucleosome remodeling and deacetylating (NRD) complex.8 The
natural products Taxol and discodermolide bind and activate,
rather than inactivate, the protein tubulin,9 which itself was
discovered as the protein target of colchicine.10-12 Inactivating
small molecules have also been converted into activating ones
(4) Brown, E. J.; Schreiber, S. L. Cell 1996, 86, 517-520.
(5) Brown, E. J.; Albers, M. W.; Shin, T. B.; Ichikawa, K.; Keith, C. T.;
Lane, W. S.; Schreiber, S. L. Nature 1994, 369, 756-758.
(6) Hassig, C. A.; Schreiber, S. L. Curr. Opin. Chem. Biol. 1997, 1,
300-308.
(7) Taunton, J.; Hassig, C. A.; Schreiber, S. L. Science 1996, 272, 408-
411.
(8) Tong, J. K.; Hassig, C. A.; Schnitzler, G. R.; Kingston, R. E.;
Schreiber, S. L. Nature 1998, 395, 917-921.
(9) Hung, D. T.; Chen, J.; Schreiber, S. L. Chem. Biol. 1996, 3, 287-
293.
(1) Collins, F. S.; Patrinos, A.; Jordan, E.; Chakravarti, A.; Gesteland,
R.; Walters, L.; and the members of the DOE and NIH planning groups
Science 1998, 282, 682-689.
(2) Schreiber Group Website. http://www-schreiber.chem.harvard.edu
(accessed Jun 1999).
(10) Borisy, G. G.; Taylor, E. W. J. Cell Biol. 1967, 34, 525-533.
(11) Borisy, G. G.; Taylor, E. W. J. Cell Biol. 1967, 34, 535-548.
(12) Shelanski, M. L.; Taylor, E. W. J. Cell Biol. 1967, 34, 549-554.
(3) Schreiber, S. L. Bioorg. Med. Chem. 1998, 6, 1127-1152.
10.1021/ja992144n CCC: $18.00 © 1999 American Chemical Society
Published on Web 09/21/1999

9074 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
Figure 3. Abbreviated spacer, photolabile linker, and support.
of effective forward and reverse chemical genetic screens. Two
that we are finding to be of great value are the cytoblot technique
(for forward chemical genetics)16,17 and small molecule printing
(for reverse chemical genetics).18
We are developing an integrated program aimed at advancing
chemical genetics. At the front end of this effort is a series of
binary encoded,19 split-pool syntheses of complex small mol-
ecules having structural features reminiscent of natural products.
A preliminary account of a first such synthesis was recently
reported.20 That study addressed numerous issues that we feel
will be of relevance to future studies, especially the development
of library validation protocols. Herein, we provide a full account
of these investigations.
Template Synthesis. Decreasing the degrees of conforma-
tional freedom in small molecules can facilitate protein binding
and specificity by reducing the entropic cost of binding. In
natural products, this conformational restriction results from
macrocyclic, polycyclic, or highly substituted linear structures
(Figure 2). An appropriate constellation of hydrogen bonding
and hydrophobic functional groups provides the enthalpic
driving force for protein binding. We are synthesizing similarly
constructed molecules using efficient multistep syntheses that
include several coupling steps. The use of the split-pool strategy
allows the efficient syntheses of collections of 104-107 spatially
segregated compounds, which can then be used in phenotype-
based cytoblot assays or binding-based small molecule printing
assays. Since purification is not practical in large split-pool
syntheses, reactions generating products in g90% purity are of
considerable value. Ideally, every reaction should increase the
diversity of the library; thus, batch transformations and protect-
ing group manipulations are undesirable.21 Selection of a solid
support and linker that allow release of the synthetic compound
from a single bead into miniaturized wells is a critical aspect
of the experimental design. Linkers and beads can also present
significant synthetic challenges since their structural elements
may be incompatible with numerous reagents used in conven-
tional synthetic chemistry.
Figure 2. Natural products used in chemical genetic studies.
by combined genetic modification of target proteins and
chemical modification of natural products to generate chemical
inducers of dimerization (CIDs).13
It seems unlikely that natural products alone will provide the
hypothetical “complete” set of small molecules that would allow
the functions of all proteins, as well as their individual domains,
to be determined. For chemistry to have its maximal effect on
biology, efficient methods for discovering this set of small
molecules are in great demand. Organic synthesis will play a
central role, yet syntheses must be devised that will allow the
products to be used effectively in both forward and reverse
chemical genetic studies. The strategy of split-pool (also known
as mix-and-split) synthesis, involving a simple mechanical
intervention at strategic steps in a conventional synthesis,
provided the first practical means of synthesizing vast numbers
of spatially segregated small molecules.14,15 This strategy ideally
requires that an early building block of the synthetic compounds
be immobilized onto small polymeric supports. Its most effective
implementation will require the development of syntheses of
small molecules having structural features optimal for protein
binding and specificity. There are many challenges involved in
the identification of these structural features and in the develop-
ment of effective syntheses of molecules equipped with them.
There are also many challenges associated with the development
In our initial studies, we selected TentaGel S NH2,22 a poly-
(ethyleneglycol)-polystyrene copolymer, as a solid support with
compounds attached via a photolabile linker developed by
Geysen and co-workers (Figure 3).23 This support-linker com-
(17) Mayer, T. U.; Kapoor, T. M.; Haggarty, S. J.; King, R. W.; Schreiber,
S. L.; Mitchison, T. J. Science. In press.
(18) MacBeath, G.; Koehler, A. N.; Schreiber, S. L. J. Am. Chem. Soc.
1999, 121, 7967-7968.
(19) Ohlmeyer, M. H. J.; Swanson, R. N.; Dillard, L. W.; Reader, J. C.;
Asouline, G.; Kobayashi, R.; Wigler, M.; Still, W. C. Proc. Natl. Acad.
Sci. U.S.A. 1993, 90, 10922-10926.
(20) Tan, D. S.; Foley, M. A.; Shair, M. D.; Schreiber, S. L. J. Am.
Chem. Soc. 1998, 120, 8565-8566.
(13) Crabtree, G. R.; Schreiber, S. L. Trends Biochem. Sci. 1996, 21,
418-422.
(14) Furka, A.; Sebestye´n, F.; Asgedom, M.; Dibo´, G. Int. J. Pept. Protein
Res. 1991, 37, 487-493.
(15) Lam, K. S.; Salmon, S. E.; Hersh, E. M.; Hruby, V. J.; Kazmierski,
W. M.; Knapp, R. J. Nature 1991, 354, 82-84.
(16) Stockwell, B. R.; Haggarty, S. J.; Schreiber, S. L. Chem. Biol. 1999,
6, 71-83.
(21) Bartlett and co-workers have proposed related criteria for the
synthetic design of split-pool libraries: Marx, M. A.; Grillot, A.-L.; Louer,
C. T.; Beaver, K. A.; Bartlett, P. A. J. Am. Chem. Soc. 1997, 119, 6153-
6167.
(22) Rapp Polymere Home Page. http://www.rapp-polymere.com (ac-
cessed Jun 1999).
(23) Brown, B. B.; Wagner, D. S.; Geysen, H. M. Mol. DiV. 1995, 1,
4-12.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9075
Scheme 1
Scheme 2
bination allows compounds to be released from the beads in
the presence of living cells in aqueous media by exposure to
long-wave (365 nm) ultraviolet light.24-26 An ω-aminocaproic
acid (Aca) spacer was coupled after the linker to minimize steric
interactions between compounds and the linker.
Tamura and co-workers have described the synthesis of a
tricyclic compound, 2, by a tandem transesterification-cycload-
dition reaction of nitrone methyl ester 1 and cyclohexen-3-ol
(Scheme 1).27,28 Titanium tetrachloride catalyzes the initial
transesterification to a nitrone cyclohexenyl ester intermediate.
Presumably, isomerizations to an s-cis-like conformation of the
ester C-O bond and to the E configuration of the nitrone CdN
bond are required to allow the appropriate transition state
geometry for the 1,3-dipolar cycloaddition reaction to form 2.29
The tricyclic structure, synthesized by a modified sequence, was
selected as a core template for a library of small molecules.
Both enantiomers of epoxycyclohexenol 7 can be synthesized
from the natural product (-)-shikimic acid, 3 (Scheme 2).30-32
Introduction of an epoxide functionality in the allylic alcohol
component of the Tamura tandem reaction was expected to
provide a handle for diversification of the cycloadduct or,
alternatively, a potentially reactive functional group for interac-
tion with cellular proteins. Thus, (+)-7 was attached to the solid
support using standard coupling conditions to yield 10 (Scheme
3).33 Because of the epoxide functionality, titanium tetrachloride
was initially replaced with a milder organotin transesterification
catalyst.34 Tamura had demonstrated that this catalyst was also
effective in mediating the tandem reaction. Exposure of 10 to
nitrone ester 1 and the organotin catalyst generated tetracycle
12a with complete stereo- and regioselectivity. Unfortunately,
the solid-phase tandem reaction proceeded sluggishly upon
scale-up. Replacement of the transesterification reaction by a
direct acylation of 10 with nitrone carboxylic acid 11a35 in the
presence of PyBroP, DIPEA, and DMAP36 again generated the
Scheme 3
desired tetracycle 12a. This reaction proceeded satisfactorily
upon scale-up although multiple couplings were generally
required, presumably due to the sensitivity of the activated
nitrone carboxylic acid species.
Reaction Survey. The tetracyclic template, 12, is rigid and
densely functionalized. A variety of building blocks37 can be
coupled to the central octahydrobenzisoxazole structure, poten-
tially without the use of protecting groups (Figure 4). The
isoxazoline nitrogen can be substituted with various functional
groups, installed at either the nitrone or the tetracycle stage.
The electrophilic lactone and epoxide can react with nucleo-
philes while simultaneously unmasking the C-6 and C-5 alcohols
for subsequent reactions. Furthermore, reductive N-O bond
(24) Huang, J.; Schreiber, S. L. Proc. Natl. Acad. Sci. U.S.A. 1997, 94,
13396-13401.
(25) You, A. J.; Jackman, R. J.; Whitesides, G. M.; Schreiber, S. L. Chem.
Biol. 1997, 4, 969-975.
(26) Borchardt, A.; Liberles, S. D.; Biggar, S. R.; Crabtree, G. R.;
Schreiber, S. L. Chem. Biol. 1997, 4, 961-968.
(27) Tamura, O.; Okabe, T.; Yamaguchi, T.; Gotanda, K.; Noe, K.;
Sakamoto, M. Tetrahedron 1995, 51, 107-118.
(28) Tamura, O.; Okabe, T.; Yamaguchi, T.; Kotani, J.; Gotanda, K.;
Sakamoto, M. Tetrahedron 1995, 51, 119-128.
(29) Tamura et al. propose reaction of the Z-nitrone via an exo transition
state. However, orbital overlap appears to be more favorable in the reaction
of the E-nitrone via an endo transition state as indicated in Scheme 1.
(30) McGowan, D. A.; Berchtold, G. A. J. Org. Chem. 1981, 46, 2381-
2383.
(31) Wood, H. B.; Ganem, B. J. Am. Chem. Soc. 1990, 112, 8907-
8909.
(32) Mitsunobu, O. Synthesis 1981, 1-28.
(33) All resin-bound products were photocleaved and analyzed by
standard techniques (TLC, HPLC, NMR, MS).
(36) Abbreviations not cited in text: DMF ) N,N-dimethylformamide,
DMA ) N,N-dimethylacetamide, NMP ) N-methylpyrrolidinone, PyBroP
) bromotripyrrolidinophosphonium hexafluorophosphate, HATU ) N-[(di-
methylamino)-1H-1,2,3-triazolo[4,5-b]pyridin-1-ylmethylene]-N-methylmethan-
aminium hexafluorophosphate, PyBOP ) benzotriazol-1-yloxytripyrroli-
dinophosphonium hexafluorophosphate, TFFH ) N,N,N′,N′-tetramethylfluo-
roformamidinium hexafluorophosphate, DIPC ) 1,3-diisopropylcarbodi-
imide, CDI ) 1,1′-carbonyldiimidazole, DIPEA ) N,N-diisopropylethylamine,
DMAP ) 4-(dimethylamino)pyridine, cod ) 1,5-cyclooctadiene, DPPF )
1,1′-bis(diphenylphosphino)ferrocene, DBN ) 1,5-diazabicyclo[4.3.0]non-
5-ene, LC-MS ) tandem high-pressure liquid chromatography-mass
spectrometry, HR-FAB-MS ) high-resolution fast atom bombardment mass
spectrometry, HR-TOF-ESI-MS ) high-resolution time-of-flight electro-
spray induction mass spectrometry, EC-GC ) electron capture detection
gas chromatography.
(34) Otera, J.; Dan-oh, N.; Nozaki, H. J. Org. Chem. 1991, 56, 5307-
5311.
(37) The term “building block” is preferred over the commonly used
term “monomer” since the products are non-oligomeric.
(35) Keirs, D.; Overton, K. Heterocycles 1989, 28, 841-848.

9076 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
Scheme 5
Figure 4. Potential building block coupling sites on 12.
Scheme 4
We next considered strategies for introducing substituents at
the isoxazolidine nitrogen. The most obvious approach was to
perform solution-phase syntheses of a large number of the
nitrone carboxylic acids, 11, then apply these compounds to
the solid-phase tandem reaction. However, it is more efficient
to synthesize a limited number of nitrone acids that contain a
reactive functional group. After tetracycle formation, this group
can undergo further solid-phase reactions with readily available
building blocks, increasing library diversity without requiring
multiple solution-phase syntheses. Aryl iodides were selected
because they can be functionalized using palladium-catalyzed
reactions. Three iodobenzyl nitrone acids, 11b-d, were readily
synthesized from the iodobenzyl alcohols with trituration at the
final step sufficient to provide pure compounds (Scheme
5).35,42,43 Treatment of 10 with 11b-d and PyBroP/DIPEA/
DMAP yielded the desired iodobenzyl tetracycles 12b-d
(Scheme 3).
The benzyl iodide substituents are amenable to a number of
different palladium-catalyzed reactions (e.g., cross couplings,
aminations, etherifications, carbonylations). This creates the
potential for increased diversity through the application of both
“building block-based” and “reaction-based” diversity. The
former arises by introducing building blocks with a common
functional group at a given site, the latter by using reactions
that couple building blocks with different functional groups to
the same site. With this approach in mind, Stille,44,45 Suzuki,46-49
and Sonogashira/Castro-Stephens50-54 reactions were all suc-
cessfully performed on 13d at room temperature to form 18-
20 (Scheme 6). Unfortunately, attempts at Heck55,56 and catalytic
cleavage can provide two additional orthogonal handles for
functionalization.
(42) Borch, R. F.; Bernstein, M. D.; Durst, H. D. J. Am. Chem. Soc.
1971, 93, 2897-2904.
(43) Acheson, R. M.; Lee, G. C. M. J. Chem. Soc., Perkin Trans. 1 1987,
2321-2332.
(44) Stille, J. K. Angew. Chem., Int. Ed. Engl. 1986, 25, 508-524.
(45) Despande, M. S. Tetrahedron Lett. 1994, 35, 5613-5614.
(46) Miyaura, N.; Suzuki, A. Chem. ReV. 1995, 95, 2457-2483.
(47) Backes, B. J.; Ellman, J. A. J. Am. Chem. Soc. 1994, 116, 11171-
11172.
Preliminary experiments showed that lactone 12a reacted
readily with primary amines in THF to form γ-hydroxyamides
such as 13a (Scheme 4). The liberated C-6 alcohol could then
be acylated with phenyl isocyanate to form 15. Alternatively,
the epoxide underwent Yb(OTf)3-mediated aminolysis38 to form
aminodiols such as 14, although this product was recovered in
low yield and purity. Nitriles were used to open the epoxide of
12a in another Yb(OTf)3-mediated reaction to form 16 and
17,39,40 however these compounds were also recovered in low
yield. Further investigations indicated that the Geysen linker
was prone to â-elimination under these conditions, highlighting
the need for exceedingly mild reaction conditions when using
this linker-support combination.41
(48) Frenette, R.; Friesen, R. W. Tetrahedron Lett. 1994, 35, 9177-
9180.
(49) Guiles, J. W.; Johnson, S. G.; Murray, W. V. J. Org. Chem. 1996,
61, 5169-5171.
(50) Sonogashira, K.; Tohda, Y.; Hagihara, N. Tetrahedron Lett. 1975,
4467-4470.
(51) Stephens, R. D.; Castro, C. E. J. Org. Chem. 1963, 28, 3313-3315.
(52) Young, J. K.; Nelson, J. C.; Moore, J. S. J. Am. Chem. Soc. 1994,
116, 10841-10842.
(38) Meguro, M.; Asao, N.; Yamamoto, Y. J. Chem. Soc., Perkin Trans.
1 1994, 2597-2601.
(53) Collini, M. D.; Ellingboe, J. W. Tetrahedron Lett. 1997, 38, 7963-
7966.
(39) Ryan, R. J.; Julia, S. Tetrahedron 1973, 29, 3649-3654.
(40) Lindsay Smith, J. R.; Norman, R. O. C.; Stillings, M. R. J. Chem.
Soc., Perkin Trans. 1 1975, 1200-1202.
(54) Odingo, J.; Sharpe, B. A.; Oare, D. Presented at the 213th National
Meeting of the American Chemical Society, San Francisco, CA, April 1997;
ORGN 574.
(55) Heck, R. F. In ComprehensiVe Organic Synthesis; Trost, B. M.,
Fleming, I., Eds.; Permangon Press: Oxford, 1991; Vol. 4, pp 833-863.
(41) Sternson, S. M.; Schreiber, S. L. Tetrahedron Lett. 1998, 39, 7451-
7454.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9077
Scheme 6
Scheme 7
an acceptable level for split-pool synthesis (17% for 13a and
21% for 13d).
We investigated other isoxazolidine nitrogen substituents that
might reduce or eliminate the aminolysis side reaction problem.
Chloroformates have been used to dealkylate tertiary amines,
yielding secondary carbamates.64 Application to the tertiary
isoxazoline nitrogen would provide various carbamate products
that might not undergo the aminolysis side reaction. Further-
more, use of the appropriate chloroformate would allow
subsequent carbamate cleavage to form the free isoxazoline that
could be functionalized by alkylation or acylation after lactone
aminolysis.65-68 Treatment of 12a and 12d with a large excess
of phenylchloroformate resulted in formation of the desired
carbamates 21a and 21d (Scheme 7). A smaller excess of
DIPEA was added to quench adventitious HCl and prevent
halohydrin formation at the epoxide. Unfortunately, we were
unable to drive this reaction beyond 50% conversion and
therefore did not pursue it further.
Three iodophenyl nitrones, 11e-g69,70 (Scheme 5), and the
corresponding iodophenyl tetracycles, 12e-g (Scheme 3), were
also synthesized in an effort to overcome the aminolysis side
reaction problem.71 We were encouraged to find that tetracycle
12g could be aminolyzed with 4-methoxybenzylamine to
γ-hydroxyamide 22 (Scheme 8) with minimal side reaction
(<10% yield of 10). Several other pilot reactions with these
iodophenyl tetracycles appeared promising, including the facile
epoxide (thio)acidolyisis of 12e to form 26 and 27 (Scheme 9).
Alcohol 27 was readily acylated with propionyl chloride to yield
28. However, presumably due to the electron-rich nature of the
aryl iodides, these iodophenyl compounds proved poor substrates
amination57-59 reactions appeared to result in decomposition
of the starting material or â-elimination of the linker.
The palladium chemistry seemed well-suited for a late step
of the synthesis because of the mild, site-selective nature of
these reactions. However, when 4-iodobenzyltetracycle 12d was
aminolyzed to γ-hydroxyamide 13d (Scheme 4), a significant
amount (59%) of epoxycyclohexenol 10 was formed along with
the desired product. Reexamination of the aminolysis of the
parent desiodotetracycle 12a to γ-hydroxyamide 13a revealed
24% formation of 10. Whether this side reaction occurred by a
true cycloreversion or some other mechanism,60 it presented a
serious challenge because of the stringent requirements for purity
and yield (g90%) in split-pool synthesis. After screening a wide
range of acylation catalysts,61 we found that addition of the
tautomeric catalyst 2-hydroxypyridine62,63 substantially reduced
the amount of 10 formed during the reaction, however not to
(56) Hiroshige, M.; Hauske, J. R.; Zhou, P. Tetrahedron Lett. 1995, 36,
4567-4570.
(57) Wolfe, J. P.; Wagaw, S.; Marcoux, J.-F.; Buchwald, S. L. Acc. Chem.
Res. 1998, 31, 805-818.
(62) Rony, P. R. J. Am. Chem. Soc. 1969, 91, 6090-6096.
(63) Openshaw, H. T.; Whittaker, N. J. Chem. Soc. C 1969, 89-91.
(64) Hobson, J. D.; McCluskey, J. G. J. Chem. Soc. C 1967, 2015-
2017.
(58) Hartwig, J. F. Acc. Chem. Res. 1998, 31, 852-860.
(59) Ward, Y. D.; Farina, V. Tetrahedron Lett. 1996, 37, 6993-6996.
(60) Analysis of the supernatant from the 12d f 13d aminolysis reaction
seems to indicate that the side reaction is not a true cycloreversion.
Additional experiments indicate that electron-withdrawing substituents on
the aromatic ring potentiate this reaction. (For 12 f 13: R ) 4-CF3-Bn,
69% side reaction; R ) 4-I-Bn, 59%; R ) 4-Cl-Bn, 32%; R ) Bn, 24%;
R ) 4-tBu-Bn, 13%; R ) 3-I-Bn, 46%; R ) 2-I-Bn, 31%.) Replacement
of the ω-aminocaproic acid spacer with glycine or removal of the spacer
altogether actually decreases the amount of 10 formed. (For 12, R ) 4-I-
Bn: Aca spacer, 59% side reaction; Gly spacer, 36%; no spacer 29%.)
Interestingly, substitution at the benzylic carbon eliminates the side reaction
altogether (for 12, R ) 4-I-R-Me-Bn, 0% side reaction). Moreover, solution-
phase aminolysis of the tetracycle below results in no side reaction. Control
experiments verified that the side reaction is not caused by photocleavage.
(65) Montzka, T. A.; Matiskella, J. D.; Partyka, R. A. Tetrahedron Lett.
1974, 1325-1327.
(66) Olofson, R. A.; Schnur, R. C.; Bunes, L.; Pepe, J. P. Tetrahedron
Lett. 1977, 1567-1570.
(67) Olofson, R. A.; Martz, J. T.; Senet, J.-P.; Piteau, M.; Malfroot, T.
J. Org. Chem. 1984, 49, 2081-2082.
(68) Jung, M. E.; Lyster, M. A. J. Chem. Soc., Chem. Commun. 1978,
315-316.
(69) McClure, K. F.; Danishefsky, S. J. J. Am. Chem. Soc. 1993, 115,
6094-6100.
(70) Kamm, O. Organic Syntheses; Wiley & Sons: New York, 1941;
Collect. Vol. I, pp 445-447.
(71) Use of the appropriate coupling reagent proved critical to the success
of these tandem reactions. After extensive screening of coupling conditions,
HATU/DMAP was found effective in forming 12e and 12g, while only
PyBOP/DMAP successfully mediated conversion to 12f. The following
reaction conditions were unsuccessful: (1) in situ activation of acid with
PyBroP, DIPEA, (DMAP, (4 Å MS, in NMP, DMF, dioxane, or CH2-
Cl2; (2) preactivation of acid with oxalyl chloride in CH2Cl2; (3) preacti-
vation of acid with cyanuric fluoride in CH2Cl2; (4) in situ activation of
acid with DIPC, DMAP, CH2Cl2; (5) in situ activation of acid with CDI,
catalytic pyridine, CH2Cl2; (6) in situ activation of acid with 2,4,6-
trichlorobenzoyl chloride, CH2Cl2; and (7) esterification with the Otera
organotin catalyst.
(61) Aminolysis in the presence of sodium cyanide in MeOH, sodium
hydride in DMA, sodium methoxide in benzene, ferric chloride in MeOH,
lithium chloride in MeOH, and lithium hydroxide in MeOH all appeared
to result in degradation of the linker. Aminolysis in the presence of the
Otera organotin catalyst in toluene did not reduce the amount of epoxy-
cyclohexenol recovered.

9078 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
Scheme 8
Scheme 10
Scheme 11
Scheme 9
variety of acylation catalysts, bases, and solvents in efforts to
cap this hydroxyl group.73 Conditions for efficient reactions with
anhydrides (31a) and aryl isocyanates (31b) were developed
(Scheme 11). However, in the latter case, even after extensive
resin washing, the photolysis product of carbamate 31b con-
tained substantial amounts of di(4-methoxyphenyl)urea. Al-
though isocyanate reactions have been reported on polystyrene
solid supports,74-79 it is possible that the poly(ethyleneglycol)
portion of the TentaGel polymer prevented effective removal
of the urea byproduct.
The alkynes introduced by the Castro-Stephens reaction
provide a potential handle for coupling of a fourth diversity
element. Considering the mild reaction conditions required by
our linker-support combination, we investigated a recently
reported rhodium-catalyzed hydroacylation reaction.80 Since an
ample number of salicyl aldehydes are commercially available,
this reaction could generate substantial additional diversity.81
Exposure of propionate 31a to salicyl aldehyde in the presence
of [Rh(cod)Cl]2 and DPPF yielded the expected regioisomers
for the palladium-catalyzed reactions and yielded impure
products (23-25).
Ultimately, we overcame the aminolysis side reaction problem
by changing our reaction sequence. Iodobenzyl tetracycle 12d
was first converted to alkynylbenzyl tetracycles 29a and 29b
using the Sonogashira/Castro-Stephens reaction (Scheme 10).
2-Hydroxypyridine-mediated lactone aminolysis to γ-hydroxya-
mides 30a and 30b proceeded effectively without any detectable
formation of epoxycyclohexenol 10.72
Lactone aminolysis not only couples an amine diversity
element but also simultaneously unmasks the C-6 hydroxyl
group that can undergo further reactions. Acid chlorides,
anhydrides, sulfonyl chlorides, chloroformates, carbamoyl chlo-
rides, and aryl and alkyl isocyanates were combined with a
(74) Hobbs DeWitt, S.; Kiely, J. S.; Stankovic, C. J.; Schroeder, M. C.;
Reynolds Cody, D. M.; Pavia, M. R. Proc. Natl. Acad. Sci. U.S.A. 1993,
90, 6909-6913.
(75) Kick, E. K.; Ellman, J. A. J. Med. Chem. 1995, 38, 1427-1430.
(76) Kurth, M. J.; Ahlberg Randall, L. A.; Takenouchi, K. J. Org. Chem.
1996, 61, 8755-8761.
(77) Beebe, X.; Schore, N. E.; Kurth, M. J. J. Org. Chem. 1995, 60,
4196-4203.
(72) Despite considerable efforts to understand this vexing side reaction,
both its exact mechanism and a rationalization of the ameliorating effects
of alkynylbenzyl substituents remain mysterious to us.
(73) Where appropriate, DMAP, PBu3 (Vedejs, E.; Diver, S. T. J. Am.
Chem. Soc. 1993, 115, 3358-3359), and Cu(I)Cl (Duggan, M. E.; Imagire,
J. S. Synthesis 1989, 131-132) were assayed as acylation catalysts; DIPEA,
pyridine, and 2,6-lutidine were assayed as bases; CH2Cl2, THF, and DMF
were assayed as solvents.
(78) Beebe, X.; Schore, N. E.; Kurth, M. J. J. Am. Chem. Soc. 1992,
114, 10061-10062.
(79) Meyers, H. V.; Dilley, G. J.; Durgin, T. L.; Powers, T. S.;
Winssinger, N. A.; Zhu, H.; Pavia, M. R. Mol. DiV. 1995, 1, 13-20.
(80) Kokubo, K.; Matsumasa, K.; Miura, M.; Nomura, M. J. Org. Chem.
1997, 62, 4564-4565.
(81) The 1996-1997 Aldrich Structure Index lists 24 applicable salicyl
aldehydes.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9079
Scheme 12
Scheme 13
32a and 32b in approximately a 1:1 ratio (Scheme 12). However,
this reaction proved quite sensitive to sterics as compounds 33-
35 all failed to react under these conditions.
byproduct to an acceptable 5-10% while still resulting in
complete conversion of 39 to 40. Additionally, terminal diynes
were coupled efficiently by using (PPh3)4Pd as the palladium
source and a larger excess of diyne (50 equiv instead of 20
equiv for monoynes) to avoid intrabead cross-linking. Attempts
to perform this reaction with catalytic amounts of palladium or
copper iodide resulted in slow reaction rates.
Preliminary studies had indicated that inclusion of 2-hy-
droxypyridine in the lactone aminolysis reaction (40 f 41)
yielded the γ-hydroxyamide products in higher purity. However,
this modification led to epoxide aminolysis as a competing side
reaction. Application of 25 equiv of amine, 5 equiv of
2-hydroxypyridine, and shorter reaction time (14-16 h instead
of 20-24 h) resulted in complete conversion with little or no
evidence of epoxide aminolysis. R-Branched amines could also
open the lactone efficiently when applied in larger excess (50
Reaction Optimization. The three highly efficient Sono-
gashira/Castro-Stephens, lactone aminolysis, and esterification
reactions offered the opportunity to generate a library of
substantial size and diversity. Additional diversity could be
generated by the use of the three iodobenzyl nitrone isomers
11b-d, both enantiomers of epoxycyclohexenol 7, and several
different spacer elements between the photolinker and the library
(Scheme 13). These reactions were studied in detail to define
the scope, limitations, and optimal conditions for each.
Initial efforts at the Sonogashira/Castro-Stephens reaction (39
f 40) yielded approximately 20% of an unidentified low
molecular weight byproduct. However, shorter reaction times
(15-120 min instead of 16 h) reduced the amount of this

9080 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
equiv of amine, 10 equiv of 2-hydroxypyridine). Unfortunately,
we were unable to develop conditions for lactone opening
reaction with secondary amines, anilines, N,N-dialkylhydrazines,
O-alkylhydroxylamines, hydrazides (R-CO-NHNH2), carbazates
(RO-CO-NHNH2), or (thio)carbazides (RHN-C(O/S)-NHNH2).
Exposure of the tetracycle to hydrazine in MeOH appeared to
result in degradation of the photolinker.
Scheme 14
A limited number of symmetrical anhydrides are com-
mercially available and useful for the esterification reaction (41
f 42).82 Thus, a general method of coupling carboxylic acids
to the C-6 hydroxyl group was desired. After surveying several
conditions,83 we found that the reaction proceeded efficiently
when DIPC and DIPEA were used to preactivate the carboxylic
acids as symmetrical anhydrides that were subsequently added
to the resin with DMAP.84 Straight-chain, R-branched, conju-
gated, and aromatic carboxylic acids were all coupled success-
fully under these conditions. However, we were unable to find
conditions that would allow coupling of N-Boc- or N-Fmoc-
protected amino acids. Completion of these reaction optimization
experiments set the stage for the detailed library validation
studies required prior to library synthesis.
Library Validation Protocols. Split-pool synthesis provides
the theoretical means to synthesize the full matrix of every
combination of building blocks in a multistep synthesis. Such
large numbers of molecules will likely be required for successful
outcomes in chemical genetic screens. However, these syntheses
present enormous analytical challenges. We have developed a
four-stage validation protocol to provide maximum confidence
that a complex, split-pool synthesis of encoded molecules yields
the anticipated products in high purity and efficiency. The first
three protocols are concerned with the synthetic molecules and
the fourth with the encoding step.
First, the suitability of the reaction sequence for library
synthesis was demonstrated by execution of the entire reaction
sequence six times, each time using different building blocks.
The fully elaborated final products, 42a-f, were recovered in
80-90% purity following photolysis. These products, as well
as the 20 intermediates preceding them (38a, 38d, 39a-f, 40a-
alkynes that were unsuitable for the Sonogashira/Castro-
Stephens reaction; however, in the aminolysis and esterification
reactions, electron-poor amines and electron-rich or enolizable
acids generally did not react with suitable efficiency. In addition,
several of the acids were insoluble under the reaction conditions.
Of the building blocks tested, 23 alkynes, 54 amines, and 44
acids reacted with g90% conversion and purity. These building
blocks, along with a limited number of less optimal candidates
(generally reacting with g70% conversion and purity), were
selected for inclusion in library synthesis.
Third, a small test library was generated from iodobenzyltet-
racycle 39b (Figure 9) to investigate whether any unforeseen
complications, such as interactions between building blocks
coupled at different sites, might arise during synthesis in a split-
pool format. The building blocks were carefully selected such
that every product within each final acylated pool would have
a unique mass (Figure 10), allowing analysis by LC-MS. Thus,
the tetracycle-containing resin was divided into eight portions
and the seven alkynes were coupled to the first seven portions.
The eighth portion was left as the parent aryl iodide, representing
a “skip codon”.86 After pooling and splitting, the seven amines
were coupled and the eighth portion of resin was left as the
lactone-closed skip codon. Finally, after a third round of pooling
and splitting, the seven acids were coupled and the eighth portion
was left as the free C-6 hydroxyl skip codon. Because all eight
final pools, designated 43{X,X,1} through 43{X,X,8},87 con-
1
f, 41a-f), were fully characterized by multidimensional H
NMR, HR-FAB-MS, TLC, and HPLC. This experiment showed
that the reaction sequence could be used to synthesize library
members in satisfactory purity.
Second, potential building blocks were tested by reaction with
a selected substrate at each step (Scheme 14). While it is
impossible to test the complete matrix of building block
combinations, this experiment indicated which building blocks
are compatible with the coupling reactions. Thus, 50 alkynes
(Figure 5) were tested in reactions with iodobenzyltetracycle
39d, 87 amines (Figure 6) in reactions with alkynylbenzyltet-
racycle 40d, and 98 acids (Figure 7) in reactions with γ-hy-
droxyamide 41d. Nearly every commercially available terminal
alkyne was tested, along with a variety of amines and acids
representing different steric and electronic functional groups.
Photocleavage products were analyzed by HPLC and LC-MS85
and their purities and percent conversions were estimated from
these data (Figure 8). There were no obvious trends among the
(82) The 1996-1997 Aldrich Structure Index lists 54 applicable sym-
metrical anhydrides compared to over 1000 carboxylic acids.
(83) DIPC (0.5 and 1.0 equiv relative to acid), HATU, and TFFH
(Carpino, L. A.; El-Faham, A. J. Am. Chem. Soc. 1995, 117, 5401-5402)
were assayed as coupling reagents with DIPEA in CH2Cl2, both with and
without DMAP.
(84) Atuegbu, A.; Maclean, D.; Nguyen, C.; Gordon, E. M.; Jacobs, J.
W. Bioorg. Med. Chem. 1996, 4, 1097-1106.
(85) Certain acid-sensitive products were also analyzed by TLC and FAB-
MS.
(86) Combs, A. P.; Kapoor, T. M.; Feng, S.; Chen, J. K.; Daude´-Snow,
L. F.; Schreiber, S. L. J. Am. Chem. Soc. 1996, 118, 287-288.
(87) Test library compounds are designated as 43{Ra,Rb,Rc} where Ra
signifies the alkyne building block, Rb signifies the amine building block,
and Rc signifies the acid building block. Pools of compounds are signified
by Rn ) X, where X represents all eight building blocks at a given position.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9081
Figure 5. Alkyne building blocks tested in Sonogashira/Castro-Stephens reaction. Building blocks reacting with g80% conversion and purity are
denoted by x, 50-80% by ), <50% by ×. Building blocks included in the full-scale library are denoted by *.
tained the same eight γ-butyrolactone compounds corresponding
to the aminolysis skip codon, a total of 456 compounds was
generated.
purpurea lectin,94 matrilysin,95 and aspartyl proteases.96 Still’s
polyhaloaromatic EC-GC tags, 44, were selected since they are
relatively unreactive and can be coupled directly to the
polystyrene backbone of beads using mild carbene insertion
chemistry (Scheme 15).19,97
Each pool was photocleaved to yield a mixture of 64
compounds that were analyzed by LC-MS (Figure 11). Of the
456 expected masses, all 456 (100%) were detected at some
level, 418 (92%) were detected at g10% of the average intensity
for the given pool, and 400 (88%) were detected at g20% of
the average intensity for the given pool. All of the weak signals
resulted from compounds having one of two building blocks at
the amine position. 1-(3-Aminopropyl)-2-pyrrolidinone (Amine
6) is known to cyclize to DBN with loss of H2O. Since strong
bases had been found to be incompatible with our linker-
support combination, this building block was excluded from
full-scale library synthesis. The skip codon (Amine 1) left
lactone-closed tetracycles that were partially hydrolyzed during
the final acylation step. As a result, during full-scale library
synthesis, the aminolysis skip codon pool was set aside before
the final pooling, splitting, and acylation steps.
Binary Encoding. Assuming an ideal, efficient split-pool
synthesis, each support carries a single compound. Several
solutions to the problem of compound identification have been
developed, falling into two general categories: recursive de-
convolution and encoding.88 Since recursive deconvolution
requires several rounds of resynthesis, we chose the particularly
powerful binary encoding strategy. This method has been used
to discover small molecules that bind to carbonic anhydrase,89,90
SH3 domains, 86,91,92 angiotensin-converting enzyme,93 Bauhinia
Unfortunately, the published procedures gave inconsistent and
unsatisfactory results in our hands. Substitution of the reported
rhodium bis(trifluoroacetate) catalyst with a bulkier rhodium
bis(triphenylacetate) catalyst98 suppressed solution-phase dia-
zoketone dimerization and substantially improved the efficiency
of tag-bead coupling to form cycloheptatrienes 45. We also
found that, after reaction with an initial set of tags, attachment
of subsequent tags to the same beads required multiple
couplings. It is possible that the initial reactions occurred at
the most accessible sites in the polymer, making subsequent
reactions more difficult. Finally, the reaction conditions for
oxidative cleavage of the tags from 45 with ceric ammonium
nitrate (CAN) were optimized, reducing the required cleavage
time to 10 min from the reported 4 h. This improved the yields
of the polyhaloaromatic alcohol products, 46, and allowed rapid
and consistent analysis by EC-GC.
Full-Scale Library Planning and Synthesis. Completion of
the validation protocols above set the stage for full-scale encoded
library synthesis. First, building blocks were selected for each
step of the synthesis (Scheme 13). ω-Aminocaproic acid and
glycine were selected as spacer elements with the “no spacer”
skip codon providing a third structure for 37. Use of both
enantiomers of epoxycyclohexenol 7 resulted in six structures
(88) Czarnik, A. W. Curr. Opin. Chem. Biol. 1997, 1, 60-66.
(89) Burbaum, J. J.; Ohlemeyer, M. H. J.; Reader, J. C.; Henderson, I.;
Dillard, L. W.; Li, G.; Randle, T. L.; Sigal, N. H.; Chelsky, D.; Baldwin,
J. J. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 6027-6031.
(90) Baldwin, J. J.; Burbaum, J. J.; Henderson, I.; Ohlemeyer, M. H. J.
J. Am. Chem. Soc. 1995, 117, 5588-5589.
(94) Liang, R.; Yan, L.; Loebach, J.; Ge, M.; Uozumi, Y.; Sekanina,
K.; Horan, N.; Gildersleeve, J.; Thompson, C.; Smith, A.; Biswas, K.; Still,
W. C.; Kahne, D. Science 1996, 274, 1520-1522.
(95) Schullek, J. R.; Butler, J. H.; Ni, Z.-J.; Chen, D.; Yuan, Z. Anal.
Biochem. 1997, 246, 20-29.
(91) Kapoor, T. M.; Hamilton Andreotti, A.; Schreiber, S. L. J. Am.
Chem. Soc. 1998, 120, 23-29.
(96) DiIanni Carroll, C.; Johnson, T. O.; Tao, S.; Lauri, G.; Orlowski,
M.; Gluzman, I. Y.; Goldberg, D. E.; Dolle, R. E. Bioorg. Med. Chem.
Lett. 1998, 8, 3203-3206.
(92) Morken, J. P.; Kapoor, T. M.; Feng, S.; Shirai, F.; Schreiber, S. L.
J. Am. Chem. Soc. 1998, 120, 30-36.
(93) Maclean, D.; Schullek, J. R.; Murphy, M. M.; Ni, Z.-J.; Gordon, E.
M.; Gallop, M. A. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 2805-2810.
(97) Nestler, H. P.; Bartlett, P. A.; Still, W. C. J. Org. Chem. 1994, 59,
4723-4724.
(98) Callot, H. J.; Metz, F. Tetrahedron 1985, 41, 4495-4501.

9082 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
Figure 6. Amine building blocks tested in lactone aminolysis reaction. Salts indicated in parentheses were neutralized in situ with DIPEA. Building
blocks reacting with g80% conversion and purity are denoted by x, 50-80% by ), <50% by ×. Building blocks included in the full-scale library
are denoted by *.
for 38. Inclusion of all three iodobenzyl nitrone carboxylic acids
11b-d led to 18 iodobenzyltetracycle structures for 39.
The three remaining diversity positions, corresponding to the
Sonogashira/Castro-Stephens, lactone aminolysis, and C-6
acylation reactions, allowed the use of substantially larger
numbers of building blocks. Optimal use of the binary encoding
tags dictates that 2n - 2 building blocks should be used at a
given position. This accounts for one skip codon (the “all one”
code) and allows for exclusion of the undesirable “all null” code
that cannot be differentiated from a failed tagging reaction. As
a practical matter, coupling of up to 26 - 2 ) 62 building blocks
at a given step was deemed feasible.
were selected, representing a range of sizes and functional
groups (Figures 6 and 7). Coupling of 62 different amines with
exclusion of a 63rd aminolysis skip codon portion would result
in 35154 structures for 41. As discussed above, the 558 lactone-
closed compounds corresponding to the aminolysis skip codon
would not react at the final acylation step. Therefore, the total
number of final library compounds, 42, resulting from acylation
with 62 acids and exclusion of a 63rd skip codon portion is
calculated as follows: [(62 × 558 ) 34596) × 63 ) 2179548]
+ 558 ) 2180106 compounds.
Synthesis of three copies of the library was planned, based
upon a calculation that indicates that three copies should be
screened to ensure 95% confidence that every compound has
been sampled.99 Although this calculation does not address the
number of copies required to ensure that every possible
compound has been synthesized, we recognized that, if neces-
sary, the library could be resynthesized on larger scale in the
future.
Only 25 - 2 ) 30 building blocks were selected for the
Sonogashira/Castro-Stephens reaction because of the relatively
small number of available terminal alkynes. Most of the alkynes
that reacted efficiently during building block screening (Figure
5) were selected. Several racemic alkynes were also included,
although diastereomeric products would likely result. Further-
more, several alkynols were included, despite their potential
reactivity in the final acylation reaction. Control experiments
indicated that these hydroxyl groups were efficiently acylated
by a variety of alkyl and aromatic acids under the DIPC-
mediated coupling conditions. Coupling of 30 different alkynes
with exclusion of a 31st skip codon portion would result in 558
structures for 40.
Library synthesis began with coupling of Fmoc-protected
Geysen linker to 90 µm TentaGel S NH2. After deprotection,
the resin, 36, was split into three portions by weight and labeled
with the tags corresponding to the spacer position. For the fourth
validation protocol, after each tagging step in the synthesis,
several beads were removed from every portion of the resin
and the tags were cleaved and analyzed to ensure adequate
Wider selections of building blocks were available at the
amine and acid positions. For each reaction, 62 building blocks
(99) Nolan, G. P., FACS Screening Web Page. http://www.stanford.edu/
group/nolan/FACSscrn.html (accessed Jun 1999).

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9083
Figure 7. Carboxylic acid building blocks tested in acylation reaction. Building blocks reacting with g80% conversion and purity are denoted by
x, 50-80% by ), <50% by ×. Underlining indicates starting material 41d was of questionable purity. Building blocks included in the full-scale
library are denoted by *.
incorporation levels. Fmoc-Aca-OH and Fmoc-Gly-OH were
then coupled to two of the portions and deprotected under
standard conditions. The resin, 37, was pooled, mixed, and split
into two equal portions. After tagging, one enantiomer of
epoxycyclohexenol 7 was coupled to each portion. Resin 38
was then pooled, mixed, and split into three equal portions for
tagging and reactions with iodobenzyl nitrones 11b-d to yield
resin 39. These tetracycle-containing resins were pooled, mixed,
and split into 31 equal portions. Each was tagged and the
appropriate 30 terminal alkynes were coupled using the Sono-
gashira/Castro-Stephens reaction. The 31st portion of resin was
set aside as the skip codon. Resin 40 was pooled, mixed, and
split into 63 portions. In this case, the 63rd portion of resin,
corresponding to the aminolysis skip codon, was 1/63rd the size
of the other 62 portions. This modification was required to avoid
overrepresentation of lactone-closed compounds in the com-
pleted library. The 62 large portions of resin were tagged and
reacted with the appropriate amines to yield resin 41. The 63rd

9084 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
Figure 8. Representative LC-MS data for testing of building blocks.
Top: Starting material and product structures. Bottom: Mass spectrum
of product peak. Inset: 214 nm UV trace, Total Ion Count (TIC) trace,
product mass trace, and starting material mass trace. While the starting
material is difficult to detect in the UV and TIC traces, a small amount
is clearly seen in the single mass trace at 4% relative intensity compared
to the product. Note the slight (0.09 min) delay between the UV detector
and mass detector retention times.
Figure 9. Tetracycle and building blocks used in the test library.
skip codon portion was set aside for the remainder of the
synthesis to avoid hydrolysis of the lactone-closed compounds
during the final acylation step. The remaining 62 portions of
resin were pooled, mixed, and split into 63 equal portions. After
tagging, the appropriate 62 acids were coupled and the 63rd
portion of resin was left as the unreacted C-6 hydroxy
compounds. Finally, the 63 acylation portions and the lactone
aminolysis skip codon portion were pooled and mixed to yield
the final library 42, calculated to contain three copies of 2180106
compounds. The entire process was completed by two of us
(D.S.T. and M.A.F.) working over a period of 3 weeks. The
bulk of this time was spent verifying that the encoding tags
had successfully coupled to every portion of the resin during
each step of the synthesis.
Cell Permeation and Pathway Modulation. It seemed
worthwhile to begin an analysis of these compounds by
screening the 456-compound test library (Figure 9) in cellular
assays even before the full-scale library was completed.
Although our compounds were designed to contain structural
features common to natural products, we had no general sense
of their ability to either permeate cells or alter cellular pathways.
These initial studies relied upon traditional nonminiaturized
assays, requiring more material than was contained on a single
synthesis bead. Thus, the test library was screened as mixtures
of compounds cleaved in bulk. The eight final pools, 43{X,X,1}
through 43{X,X,8}, each contained 64 compounds. The com-
pounds in each pool were photolyzed from the resin, recovered,
and dissolved in DMSO at an estimated concentration of 1 mM
per compound (64 mM overall).100 The eight pools, assayed at
Figure 10. Alkyne and amine building block masses and the resulting
64 unique γ-hydroxyamide product masses. Acylation of the C-6 alcohol
with a carboxylic acid shifts all of the product masses for that pool by
the same value (mass of the acid minus water).
concentrations up to 10 µM per compound, showed no sup-
pression of rapamycin-based growth inhibition in S. cereVi-
siae.101 In addition, none of the pools, assayed at concentrations
up to 12.5 µM per compound, showed inhibitory activity in a
Xenopus laeVis oocyte extract assay that indicates modulation
of the cyclin B degradation pathway.102
However, all eight pools showed a significant inhibitory effect
on mink lung cell proliferation when assayed at a concentration
of 1 µM per compound (Figure 12). Moreover, when the library
was assayed at a concentration of 250 nM per compound, pool
43{X,X,8} (Figure 13), was found to activate a TGF-â-
(100) Based upon previous results, a 50% yield was assumed.
(101) The rapamycin concentration was 100 nM. J. Huang and S. L. S.,
unpublished results.
(102) R. W. King, unpublished results.
(103) Carcamo, J.; Zentella, A.; Massague, J. Mol. Cell. Biol. 1995, 15,
1573-1581.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9085
Figure 12. Mink lung cell proliferation assay. Lane 1: No treatment.
Lane 2: 0.1% DMSO control. Lanes 3-10: Pools 43{X,X,1} through
43{X,X,8} at 1 µM concentration per compound. Data represent the
average of three experiments with error bars indicating one SD.
Scheme 15
The 64 compounds in 43{X,X,8} were resynthesized as eight
pools, designated 43{X,1,8} through 43{X,8,8}, each containing
eight different compounds. Each pool was aminolyzed with a
different amine and all pools were acylated with Acid 8. As a
negative control, pool 43{X,X,3} was also resynthesized as eight
pools, 43{X,1,3} through 43{X,8,3}. Surprisingly, pool 43-
{X,8,3}, assayed at a concentration of 1 µM per compound,
was a stronger activator of the TGF-â-responsive reporter gene
than any of the other 15 eight-compound pools or either of the
parent 64-compound pools (data not shown).
Figure 11. Representative LC-MS data for test library pool 43{X,X,4}
acylated with Acid 4. Top: Mass spectrum averaged over the entire
chromatogram. Middle: TIC trace and single mass traces. Bottom:
Corresponding structures. The 550 and 540 single mass traces represent
typical “clean” product signals. Smaller peaks (6.02 min, 6.58 min) in
the 578 single mass trace arise from isotopic compositions of lower
mass products (FW ) 575, 576). The 381 single mass trace is
representative of a “weak” product signal.
A final round of resynthesis deconvoluted pool 43{X,8,3},
yielding single compounds 43{1,8,3} through 43{8,8,3}. These
eight compounds and all of the 16 intermediates preceding them
1
were recovered in high purity as determined by H NMR and
HR-TOF-ESI-MS analysis. All 24 compounds were screened
and compounds 43{5,8,3} and 43{6,8,3} were found to activate
the TGF-â-responsive reporter gene (Figure 15). However, the
alkynylbenzyltetracycle (43{5,1,1} and 43{6,1,1}) and γ-hy-
droxyamide (43{5,8,1} and 43{6,8,1}) precursors to these
compounds were even more active. The six active compounds
were purified by silica gel chromatography and reassayed,
verifying that the desired major product is responsible for the
responsive reporter gene103 in a stably transfected mink lung
cell line (Figure 14).104 Since this library is not encoded with
chemical tags, a recursive deconvolution strategy was used to
investigate this activity further.
(104) Stockwell, B. R.; Schreiber, S. L. Curr. Biol. 1998, 8, 761-770.

9086 J. Am. Chem. Soc., Vol. 121, No. 39, 1999
Tan et al.
pathway modulating properties of members of the library. These
results also highlight the shortcoming of screening mixtures of
compounds. Active compounds may be masked by competing
cytotoxic, cytostatic, or antagonistic compounds in the same
pool. Alternatively, activity may arise from synergistic effects
between two or more compounds, making deconvolution to a
single structure impossible. For these reasons, we generally
avoid screening mixtures in high-throughput chemical genetic
screens.
Chemical Genetic Screens Using Compounds Released
from Single Beads. Solid-phase synthesis was originally
developed to facilitate the purification of peptides from reaction
mixtures. Much of the more recent solid-phase organic synthesis
has been used for a similar purpose, but involves non-oligomeric
small molecules. Split-pool synthesis provides a new and
arguably more powerful incentive for performing solid-phase
organic synthesis: It yields large numbers of spatially segregated
small molecules. To take full advantage of this feature, assay
formats must be developed that use compounds derived from
single beads. In our early studies of assay formats using single
beads, two such systems were used. Binding-based assays were
performed by detecting soluble proteins that had been recruited
to individual beads via their interactions with a tethered small
molecule.86,91,92 Phenotype-based assays were performed using
cells and synthesis beads contained in small volumes of cell
culture. These “nanodroplets” were generated either stochasti-
cally26 or on a molded, poly(dimethylsiloxane) (PDMS) grid.25
Although these assays have yielded useful information, they
suffer from several shortcomings, including the fact that the
synthesis beads are not easily used more than once. Moreover,
the synthesis described above was performed on 90 µm
TentaGel beads, having a loading capacity of only approximately
80-100 picomolar equivalents per bead. In mink lung cell
cytoblot assays searching for small molecule suppressors of
trapoxin’s and rapamycin’s actions and using a subset of
approximately 77000 beads,105 these shortcomings manifested
themselves in several ways, including the need for resynthesis
of compounds corresponding to apparent positives. On the other
hand, our early experience in screening these synthesis beads
has revealed the ability to recover positive beads, cleave their
EC-GC tags, and decode them successfully.
Figure 13. Activators of the TGF-â-responsive reporter gene.
To address the problem noted above, colleagues at the
Harvard Institute of Chemistry and Cell Biology have developed
instrumentation and robotics that provide efficient arraying of
synthesis beads into high-density wells, releasing and transfer-
ring of compounds into high-density stock solutions, and
transferring of these solutions into high-density PDMS assay
plates.106 These assay plates are the preferred format for cytoblot
assays,16 and the high-density stock solutions are also well-
suited for small molecule printing.18 To allow efficient release
of compounds into storage wells without the need for removal
of toxic byproducts, we have explored various linkers. To
provide sufficient amounts of released compounds to allow their
use in large numbers of assays, we have explored beads with
higher loading capacities.
Figure 14. TGF-â-responsive reporter gene assay. Lane 1: No
treatment. Lane 2: 0.1% DMSO control. Lane 3: 200 pM TGF-â1.
Lanes 4-11: Pools 43{X,X,1} through 43{X,X,8} assayed at a
concentration of 250 nM per compound. Data represent a single
experiment with fold activation calculated relative to untreated cells.
Conclusion
We have completed the design, development, validation, and
split-pool synthesis of a binary encoded library of complex small
molecules having structural features reminiscent of natural
products. Highly efficient coupling reactions were used to attach
activity in each case. The EC50 for the strongest activator, 43-
{6,1,1}, is approximately 50 µM.
Discovery of these compounds, while somewhat fortuitous,
gave us a measure of confidence in the cell permeating and
(105) B. R. S. S. J. Haggarty, A. J. You, B. K. Wagner, and S. L. S.,
unpublished results.
(106) L. Walling and R. W. King, unpublished results.

Synthesis of a Library of Polycyclic Small Molecules
J. Am. Chem. Soc., Vol. 121, No. 39, 1999 9087
Figure 15. TGF-â-responsive reporter gene assay. Lane 1: 1 nM TGF-â1. Lanes 2 and 3: Pool 43{X,8,3} and a mixture of individually synthesized
compounds 43{1,8,3} through 43{8,8,3}, assayed at 2.5 (black bars), 1.25 (striped bars), and 0.625 µM (white bars) per compound. Lanes 4-27:
Alkynylbenzyltetracycle precursors 43{1,1,1} through 43{8,1,1}, γ-hydroxyamide precursors 43{1,8,1} through 43{8,8,1}, and acylated final
compounds 43{1,8,3} through 43{8,8,3}, assayed at 25, 12.5, and 6.25 µM per compound. Data represent the average of two experiments with fold
activation calculated relative to untreated cells (data not shown). Note that the background signal from the instrument has not been subtracted.
Experimental Section
a variety of building blocks to a rigid, densely functionalized
tetracyclic template. We have described a four-stage library
validation protocol that provides maximum confidence that
library members are synthesized in high purity and efficiency.
Preliminary biological assays have identified molecules that
activate a TGF-â-responsive reporter gene in mammalian cells.
This work also highlights the need for further development
of mild, chemoselective coupling reactions that are compatible
with both the wide range of functional groups found in split-
pool libraries and the solid support-linker combinations useful
in chemical genetic assays. Furthermore, our preliminary
biological results indicate the dangers involved in screening
mixtures of compounds. The development of assay formats that
instead use individual compounds should eliminate these
concerns.
See Supporting Information.
Acknowledgment. D.S.T. was supported by NDSEG and
Roche Fellowships; M.A.F. was supported by Glaxo-Wellcome;
B.R.S. is a Howard Hughes Medical Institute Predoctoral
Fellow; M.D.S. was an NIH postdoctoral fellow. We are grateful
to Dr. Andrew Tyler, Ms. Jennifer Lynch, and Mr. John
Athanasopoulos for expert mass spectrometry support. We thank
Mr. Kuenley Chiu and Ms. Lygia Snow for assistance with EC-
GC tag analysis. The Harvard University Mass Spectrometry
and NMR Facilities were funded by grants from the NSF (CHE-
9020043, CHE88-14019) and NIH (S10-RR06716, 1-S10-
RR04870-01). Financial support for this work was provided by
the National Institute of General Medical Sciences. The ICCB
is supported by the National Cancer Institute, and Merck & Co.
S.L.S. is an Investigator at the Howard Hughes Medical Institute.
Our current research efforts are aimed at adapting the
synthetic chemistry described in detail herein to new linkers
and beads. The results are highly encouraging, and we can now
anticipate adapting the underlying chemistry, as well as other
synthetic pathways in development, in numerous ways. Such
syntheses should allow many thousands of experiments where
synthetic small molecules are used to modulate the circuitry of
cellular systems.
Supporting Information Available: Complete experimental
procedures and analytical data for demonstration compounds;
building block testing; test library synthesis, screening, and
deconvolution; binary encoding; and full-scale library synthesis
(PDF). This material is available free of charge via the Internet
at http://pubs.acs.org.
JA992144N