Translation of DNA into a library of 13 000 synthetic small-molecule macrocycles suitable for in vitro selection

Article abstract of DOI:10.1021/ja805649f

DNA-templated organic synthesis enables the translation, selection, and amplification of DNA sequences encoding synthetic small-molecule libraries. Previously we described the DNA-templated multistep synthesis and model in vitro selection of a pilot library of 65 macrocycles. In this work, we report several key developments that enable the DNA-templated synthesis of much larger (>10 000-membered) small-molecule libraries. We developed and validated a capping-based approach to DNA-templated library synthesis that increases final product yields, simplifies the structure and preparation of reagents, and reduces the number of required manipulations. To expand the size and structural diversity of the macrocycle library, we augmented the number of building blocks in each DNA-templated step from 4 to 12, selected 8 different starting scaffolds which result in 4 macrocycle ring sizes and 2 building-block orientations, and confirmed the ability of the 36 building blocks and 8 scaffolds to generate DNA-templated macrocycle products. We computationally generated and experimentally validated an expanded set of codons sufficient to support 1728 combinations of step 1, step 2, and step 3 building blocks. Finally, we developed new high-resolution LC/MS analysis methods to assess the quality of large DNA-templated small-molecule libraries. Integrating these four developments, we executed the translation of 13 824 DNA templates into their corresponding small-molecule macrocycles. Analysis of the resulting libraries is consistent with excellent (>90%) representation of desired macrocycle products and a stringent test of sequence specificity suggests a high degree of sequence fidelity during translation. The quality and structural diversity of this expanded DNA-templated library provides a rich starting point for the discovery of functional synthetic small-molecule macrocycles.

Full text of DOI:10.1021/ja805649f

Published on Web 10/29/2008
Translation of DNA into a Library of 13 000 Synthetic
Small-Molecule Macrocycles Suitable for in Vitro Selection
Brian N. Tse, Thomas M. Snyder, Yinghua Shen, and David R. Liu*
Howard Hughes Medical Institute and the Department of Chemistry and Chemical Biology,
HarVard UniVersity, 12 Oxford Street, Cambridge, Massachusetts 02138
Received July 20, 2008; E-mail: drliu@fas.harvard.edu
Abstract: DNA-templated organic synthesis enables the translation, selection, and amplification of DNA
sequences encoding synthetic small-molecule libraries. Previously we described the DNA-templated
multistep synthesis and model in vitro selection of a pilot library of 65 macrocycles. In this work, we report
several key developments that enable the DNA-templated synthesis of much larger (>10 000-membered)
small-molecule libraries. We developed and validated a capping-based approach to DNA-templated library
synthesis that increases final product yields, simplifies the structure and preparation of reagents, and reduces
the number of required manipulations. To expand the size and structural diversity of the macrocycle library,
we augmented the number of building blocks in each DNA-templated step from 4 to 12, selected 8 different
starting scaffolds which result in 4 macrocycle ring sizes and 2 building-block orientations, and confirmed
the ability of the 36 building blocks and 8 scaffolds to generate DNA-templated macrocycle products. We
computationally generated and experimentally validated an expanded set of codons sufficient to support
1728 combinations of step 1, step 2, and step 3 building blocks. Finally, we developed new high-resolution
LC/MS analysis methods to assess the quality of large DNA-templated small-molecule libraries. Integrating
these four developments, we executed the translation of 13 824 DNA templates into their corresponding
small-molecule macrocycles. Analysis of the resulting libraries is consistent with excellent (>90%)
representation of desired macrocycle products and a stringent test of sequence specificity suggests a high
degree of sequence fidelity during translation. The quality and structural diversity of this expanded DNA-
templated library provides a rich starting point for the discovery of functional synthetic small-molecule
macrocycles.
Introduction
templated synthesis approach to translating DNA sequences into
synthetic small molecules and synthetic polymers in which
Functional molecules emerge in living systems through cycles
of translation, selection, and amplification with mutation.
Scientists have successfully adapted features of biological
evolution to create DNA, RNA, and protein molecules with
tailor-made binding or catalytic properties. The scope of this
approach has traditionally been restricted to biopolymers that
can be accessed through the use of DNA or RNA polymerases1-4
or the ribosome.5-8 The potential benefits of applying translation
and selection-based methods to the discovery of functional
synthetic molecules has inspired new approaches to addressing
the problem of translating DNA sequences into structures not
necessarily compatible with polymerase enzymes or ribosomal
machinery. For example, our laboratory has developed a DNA-
Watson-Crick base pairing to a DNA template recruits DNA-
linked reagents to perform each synthetic step in a sequence-
programmed manner.9-15 Harbury and co-workers have devel-
oped an elegant “DNA display” method in which resin-bound
DNA hybridization directs split-and-pool combinatorial syn-
thesis, and have used DNA display to generate libraries of linear
peptides and peptoids.16-19 Neri and co-workers have used DNA
base pairing to bring together noncovalent combinations of small
pharmacophores that can suggest the synthesis of covalent small
molecules with desired binding properties.20-23
(9) Gartner, Z. J.; Liu, D. R. J. Am. Chem. Soc. 2001, 123, 6961–3.
(10) Gartner, Z. J.; Kanan, M. W.; Liu, D. R. J. Am. Chem. Soc. 2002,
124, 10304–6.
(11) Li, X.; Gartner, Z. J.; Tse, B. N.; Liu, D. R. J. Am. Chem. Soc. 2004,
126, 5090–2.
(1) Ellington, A. D.; Szostak, J. W. Nature 1990, 346, 818–22.
(2) Tuerk, C.; Gold, L. Science 1990, 249, 505–10.
(3) Singer, B. S.; Shtatland, T.; Brown, D.; Gold, L. Nucleic Acids Res.
1997, 25, 781–6.
(12) Calderone, C. T.; Liu, D. R. Angew. Chem., Int. Ed. 2005, 44, 7383–6.
(13) Sakurai, K.; Snyder, T. M.; Liu, D. R. J. Am. Chem. Soc. 2004, 127,
1660–1661.
(4) Shamah, S. M.; Healy, J. M.; Cload, S. T. Acc. Chem. Res. 2008, 41,
130–8.
(14) Rosenbaum, D. M.; Liu, D. R. J. Am. Chem. Soc. 2003, 125, 13924–5.
(15) Gartner, Z. J.; Tse, B. N.; Grubina, R.; Doyon, J. B.; Snyder, T. M.;
Liu, D. R. Science 2004, 305, 1601–5.
(5) Roberts, R. W.; Szostak, J. W. Proc. Natl. Acad. Sci. U.S.A. 1997,
94, 12297–302.
(16) Wrenn, S. J.; Weisinger, R. M.; Halpin, D. R.; Harbury, P. B. J. Am.
Chem. Soc. 2007, 129, 13137–43.
(6) Takahashi, T. T.; Austin, R. J.; Roberts, R. W. Trends Biochem. Sci.
2003, 28, 159–65.
(17) Halpin, D. R.; Harbury, P. B. PLoS Biol. 2004, 2, E173.
(18) Halpin, D. R.; Harbury, P. B. PLoS Biol. 2004, 2, E174.
(19) Halpin, D. R.; Lee, J. A.; Wrenn, S. J.; Harbury, P. B. PLoS Biol.
2004, 2, E175.
(7) Hanes, J.; Pluckthun, A. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 4937–
42.
(8) Wilson, D. R.; Finlay, B. B. Can. J. Microbiol. 1998, 44, 313–29.
9
10.1021/ja805649f CCC: $40.75
2008 American Chemical Society
J. AM. CHEM. SOC. 2008, 130, 15611–15626 15611

A R T I C L E S
Tse et al.
These approaches to the central problem of translating nucleic
acids into corresponding synthetic compounds are relatively
recent developments, and their ability to provide molecules with
desired functional properties will depend on their applicability
to appropriately sized libraries of synthetic structures. Past
efforts to synthesize and evaluate libraries of synthetic small
molecules have resulted in the discovery of chemical genetic
probes that perturb specific cellular functions in vivo in a
temporally controlled, dose-dependent, and reversible manner,24-30
as well as the discovery of lead compounds for the development
of new therapeutic agents.31-33
Despite these attractive features, the availability and use of
small- and medium-sized macrocycles have been limited by the
challenges associated with their synthesis.31,34,59-62 Macrocy-
clization reactions can be very sensitive to small structural
variations within linear precursors,60,63 can be especially difficult
for small peptide substrates that form medium-sized macro-
cycles,61,64-66 and can be prone to multimeric side-products
arising from intermolecular bond formation.61,62 These chal-
lenges are magnified in a solid-phase library synthesis format
in which the removal of truncated, multimeric, and acyclic
byproducts from desired macrocycles (all of which are linked
to the same bead) can be difficult.
Macrocycles are of special interest for the development of
biologically active small molecules.34-46 The increased rigidity
of macrocyclic compounds can greatly decrease the entropic
cost of their binding to biological targets, resulting in higher
potential binding affinities and specificities than those of
corresponding linear compounds.34,36,41,47-53 In addition, mac-
rocyclic peptide-like structures can possess higher bioavailabil-
ity, membrane permeability, and resistance to in ViVo degrada-
tion than their linear counterparts.34,36,40,41,54-58 These advantages
likely account for the significant representation of macrocyclic
compounds among biologically active secondary metabolites.38,39
We hypothesized that several features of DNA-templated
synthesis could facilitate access to macrocycles. DNA-templated
synthesis is compatible with aqueous solvent,67 extremely low
(nM) reactant concentrations,9 and selection-based purification
methods for bond formation or bond cleavage that are not
available to solid-phase synthesis10s all factors that could
promote macrocycle formation or facilitate macrocycle isolation.
We also anticipated that the ability of base pairing to hold
together relevant reactive groups at elevated effective molarities
during the macrocyclization step would further assist the ring
closure reaction. Indeed, these features enabled the successful
DNA-templated synthesis and model in Vitro selection of a pilot
library of 65 macrocycles.15
Here, we report the DNA-templated synthesis of a large
library of synthetic macrocycles suitable for in Vitro selection.
Achieving this goal required several significant methodological
advances. We developed a more robust and efficient library
synthesis route to these compounds and conducted a thorough
study of the compatibility of 36 building blocks and eight
scaffolds in DNA-templated macrocycle synthesis. Guided by
our previous studies on template secondary structure,68 we
designed a new set of DNA codons to support large-scale library
synthesis. We confirmed the efficacy, quality, and sequence
fidelity of our library synthesis method by examining a series
of small-molecule sublibraries with PAGE and high resolution
(20) Melkko, S.; Scheuermann, J.; Dumelin, C. E.; Neri, D. Nat. Biotechnol.
2004, 22, 568–74.
(21) Dumelin, C. E.; Scheuermann, J.; Melkko, S.; Neri, D. Bioconjug.
Chem. 2006, 17, 366–70.
(22) Melkko, S.; Zhang, Y.; Dumelin, C. E.; Scheuermann, J.; Neri, D.
Angew. Chem., Int. Ed. 2007, 46, 4671–4.
(23) Scheuermann, J.; Dumelin, C. E.; Melkko, S.; Zhang, Y.; Mannocci,
L.; Jaggi, M.; Sobek, J.; Neri, D. Bioconjug. Chem. 2008, 19, 778–
85.
(24) Schreiber, S. L. Bioorg. Med. Chem. 1998, 6, 1127–52.
(25) Stockwell, B. R. Nat. ReV. Genet. 2000, 1, 116–25.
(26) Dobson, C. M. Nature 2004, 432, 824–8.
(27) Stockwell, B. R. Nature 2004, 432, 846–54.
(28) Spring, D. R. Chem. Soc. Re.V 2005, 34, 472–82.
(29) Eggert, U. S.; Mitchison, T. J. Curr. Opin. Chem. Biol. 2006, 10, 232–7.
(30) Walsh, D. P.; Chang, Y. T. Chem. ReV. 2006, 106, 2476–530.
(31) Schreiber, S. L. Science 2000, 287, 1964–9.
(32) Ding, S.; Schultz, P. G. Curr. Top. Med. Chem. 2005, 5, 383–95.
(33) Marcaurelle, L. A.; Johannes, C. W. Prog. Drug Res. 2008, 66 (187),
189–216.
(51) Dinsmore, C. J.; Bogusky, M. J.; Culberson, J. C.; Bergman, J. M.;
Homnick, C. F.; Zartman, C. B.; Mosser, S. D.; Schaber, M. D.;
Robinson, R. G.; Koblan, K. S.; Huber, H. E.; Graham, S. L.; Hartman,
G. D.; Huff, J. R.; Williams, T. M. J. Am. Chem. Soc. 2001, 123,
2107–8.
(34) Driggers, E. M.; Hale, S. P.; Lee, J.; Terrett, N. K. Nat. ReV. Drug
DiscoV. 2008, 7, 608–24.
(35) Deber, C. M.; Madison, V.; Blout, E. R. Acc. Chem. Res. 1976, 9,
106–113.
(52) Dathe, M.; Nikolenko, H.; Klose, J.; Bienert, M. Biochemistry 2004,
43, 9140–50.
(36) Fairlie, D. P.; Abbenante, G.; March, D. R. Curr. Med. Chem. 1995,
2, 654–686.
(53) Udugamasooriya, G.; Saro, D.; Spaller, M. R. Org. Lett. 2005, 7, 1203–6.
(54) Charpentier, B.; Dor, A.; Roy, P.; England, P.; Pham, H.; Durieux,
C.; Roques, B. P. J. Med. Chem. 1989, 32, 1184–90.
(55) Rizo, J.; Gierasch, L. M. Annu. ReV. Biochem. 1992, 61, 387–418.
(56) Craik, D. J.; Simonsen, S.; Daly, N. L. Curr. Opin. Drug DiscoV.
DeVel. 2002, 5, 251–60.
(37) Tyndall, J. D.; Fairlie, D. P. Curr. Med. Chem. 2001, 8, 893–907.
(38) Clardy, J.; Walsh, C. Nature 2004, 432, 829–37.
(39) Wessjohann, L. A.; Ruijter, E.; Garcia-Rivera, D.; Brandt, W. Mol.
DiVers. 2005, 9, 171–86.
(40) Patton, G. C.; van der Donk, W. A. Curr. Opin. Microbiol. 2005, 8,
543–51.
(57) Shibata, K.; Suzawa, T.; Soga, S.; Mizukami, T.; Yamada, K.; Hanai,
N.; Yamasaki, M. Bioorg. Med. Chem. Lett. 2003, 13, 2583–6.
(58) Tugyi, R.; Mezo, G.; Fellinger, E.; Andreu, D.; Hudecz, F. J. Pept.
Sci. 2005, 11, 642–9.
(41) Fung, S.; Hruby, V. J. Curr. Opin. Chem. Biol. 2005, 9, 352–8.
(42) Craik, D. J.; Cemazar, M.; Daly, N. L. Curr. Opin. Drug DiscoV. DeVel.
2006, 9, 251–60.
(43) Kopp, F.; Marahiel, M. A. Nat. Prod. Rep. 2007, 24, 735–49.
(44) Turner, R. A.; Oliver, A. G.; Lokey, R. S. Org. Lett. 2007, 9, 5011–4.
(45) Scott, C. P.; Abel-Santos, E.; Wall, M.; Wahnon, D. C.; Benkovic,
S. J. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 13638–43.
(46) Tavassoli, A.; Benkovic, S. J. Nat. Protoc. 2007, 2, 1126–33.
(47) Gilon, C.; Halle, D.; Chorev, M.; Selinger, Z.; Byk, G. Biopolymers
1991, 31, 745–50.
(59) Illuminati, G.; Mandolini, L. Acc. Chem. Res. 1981, 14, 95–102.
(60) Woodward, R. B.; et al. J. Am. Chem. Soc. 1981, 103, 3213–3215.
(61) Davies, J. S. J. Pept. Sci. 2003, 9, 471–501.
(62) Blankenstein, J.; Zhu, J. Eur. J. Org. Chem. 2005, 2005, 16.
(63) Liau, B. B.; Gnanadesikan, V.; Corey, E. J. Org. Lett. 2008, 10, 1055–7.
(64) Besser, D.; Olender, R.; Rosenfeld, R.; Arad, O.; Reissmann, S. J.
Pept. Res. 2000, 56, 337–45.
(48) Cherney, R. J.; Wang, L.; Meyer, D. T.; Xue, C. B.; Wasserman, Z. R.;
Hardman, K. D.; Welch, P. K.; Covington, M. B.; Copeland, R. A.;
Arner, E. C.; DeGrado, W. F.; Decicco, C. P. J. Med. Chem. 1998,
41, 1749–51.
(65) Haddadi, M. E.; Cavelier, F.; Vives, E.; Azmani, A.; Verducci, J.;
Martinez, J. J. Pept. Sci. 2000, 6, 560–70.
(66) Taunton, J.; Collins, J. L.; Schreiber, S. L. J. Am. Chem. Soc. 1996,
118, 10412–10422.
(49) Gudmundsson, O. S.; Vander Velde, D. G.; Jois, S. D.; Bak, A.;
Siahaan, T. J.; Borchardt, R. T. J. Pept. Res. 1999, 53, 403–13.
(50) Gudmundsson, O. S.; Jois, S. D.; Vander Velde, D. G.; Siahaan, T. J.;
Wang, B.; Borchardt, R. T. J. Pept. Res. 1999, 53, 383–92.
(67) Li, C.-J.; Chan, T.-H. Organic reactions in aqueous media; Wiley
and Sons: New York, 1997.
(68) Snyder, T. M.; Tse, B. N.; Liu, D. R. J. Am. Chem. Soc. 2008, 130,
1392–401.
9
15612 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
The 12-membered DNA template libraries that produced mac-
rocycle libraries 21, 22, and 23 were assembled by ligating the
appropriate oligonucleotide 5′-halves and 3′-halves in a one-pot
mixture using the corresponding splint oligonucleotides, as de-
scribed above. Templates were then purified by PAGE, as described
above. The sequences for 21 (12 × 1 × 1), 22 (1 × 12 × 1), and
23 (1 × 1 × 12) are as follows:
LC/MS analysis. Integrating these findings, we completed the
DNA-templated synthesis of the full library of 13,824 macro-
cycles and generated sufficient material for hundreds of in Vitro
selections against biological targets of interest.
Materials and Methods
21: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTTGCTGAAAT-
GAT[codon 1]CTACCCACTC
All chemicals, unless otherwise specified, were purchased from
Sigma-Aldrich. All reagents for DNA synthesis, including modified
phosphoramidites and CPG resins, were purchased from Glen
Research. All reactions were performed at 25 °C unless otherwise
noted. All buffers were prepared at 25 °C to match reaction
conditions.
DNA Synthesis. DNA oligonucleotides were synthesized with
a PerSeptive Biosystems Expedite 8090 DNA synthesizer or with
a Bioautomation Corporation MerMade 12 DNA synthesizer, using
standard phosphoramidite protocols. Oligonucleotides were purified
by reverse-phase HPLC (Agilent) using a gradient of acetonitrile
(8-80%) in 100 mM triethylammonium acetate (TEAA), pH 7.0.
DNA was synthesized on standard CPG (1000 Å) beads, unless
otherwise noted. The oligonucleotides were quantitated by UV
spectroscopy (except where otherwise noted) using a NanoDrop
ND-1000 Spectrophotometer.
5′-segment for 21: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTT
3′-segments for 21 (12 in total): 5′-(P)-GCTGAAATGAT[codon
1]CTACCCACTC
Splint for 21: 5′-ATCATTTCAGCAACTTGAGGAA
22: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTT[codon 2]AT-
GATGGCTTTCTACCCACTC
5′-segment for 22: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTT
3′-segments for 22 (12 in total): 5′-(P)-[codon 2]ATGATG-
GCTTTCTACCCACTC
Splints for 22 (12 in total): 5′-ATCAT[anticodon 2]AACT-
TGAGGAA
23: 5′-(scaffold)-CCCTGTACAC[codon 3]AAGTTGCTGAAAT-
GATGGCTTTCTACCCACTC
5′-segments for 23 (12 in total): 5′-(scaffold)-CCCTGTACAC-
[codon 3]AAGTT
Secondary Structure Prediction for Codon Screening (see
also ref 68). Secondary structure prediction was performed using
the Oligonucleotide Modeling Platform (OMP, DNA Software, Inc.)
with the following parameters: 25 °C in 1.0 M NaCl with 100 nM
template. Reagent-template hybridization was simulated at 150 nM
reagent and 100 nM template under the same conditions.
DNA Templates. DNA templates were made by the enzymatic
ligation of two oligonucleotides mediated by a complementary splint
oligonucleotide. The 3′ halves of the templates (24, 27, or 33 bases
each) were synthesized with a 5′-phosphate group using the
Chemical Phosphorylation Reagent II (CPRII) phosphoramidite. The
5′ halves of the templates (21 or 24 bases each) were prepared
using the 5′-Amino-5 phosphoramidite; solid-phase synthesis
protocols were used (see Supporting Information) to append the
scaffold to the 5′ template halves following DNA synthesis. Ligation
reactions contained a 1:1:1 mixture of 5′-half:3′-half:splint, T4 DNA
ligase and T4 DNA Ligase Buffer (New England Biolabs). Ligation
reactions were maintained at 16 °C for 3 h. Ligation products were
precipitated with ethanol, 0.1 volumes of 3 M NaOAc (pH 5.0),
and 50 µg of glycogen/mL (Roche) and purified by PAGE. DNA
was recovered by excising the desired acrylamide gel band, and
crushing, freezing, and thawing the excised gel bands in 400 µL
10 mM Tris (pH 7.5) with 1 mM EDTA. Tubes containing the
resulting material were shaken for 3 h at 37 °C, and the desired
DNA was recovered by removal of the polyacrylamide using an
Ultrafree-MC centrifuge filter (Millipore) and precipitation with
ethanol as described above.
3′-segment for 23: 5′-(P)-GCTGAAATGATGGCTTTCTAC-
CCACTC
Splints for 23 (12 in total): 5′-ATCATTTCAGCAACTT[anti-
codon 3]
The general template structure for template libraries 24, 26, and
28 is as follows; each building-block codon is six nucleotides, while
the scaffold codon is three nucleotides (scaffold codons are listed
in the Supporting Information, Figure S2):
24/26/28: 5′-(scaffold)-CCCTGTACAC[codon 3]AAGTT[codon
2]ATGAT[codon 1]CTA[scaffold codon]-CATCCCACTC
Template libraries 24, 26, and 28 were assembled by ligation of
the appropriate oligonucleotide 5′-halves and 3′-halves in a one-
pot mixture using the corresponding splint oligonucleotides, as
described above. Templates were then purified by PAGE, as
described above. The sequences for 24, 26, and 28 are as follows:
5′-segments of 24, 26, and 28: 5′-(scaffold)-CCCTGTACAC-
[codon 3]AAGTT
3′-segments of 24, 26, and 28: 5′-(P)-[codon 2]ATGAT[codon
1]CTA[scaffold codon]-CATCCCACTC
Splints for 24, 26, and 28: 5′-ATCAT[anticodon 2]AACTT
[anticodon 3]
12-membered template library 24 was synthesized by performing
12 separate splint ligations (3A + [2A-1A], 3B + [2B-1B],... 3
L + [2L-1L]); each ligation used one 5′-half, one 3′-half, and the
corresponding splint. The resulting 12 templates were purified by
PAGE, as described above. The templates were pooled together to
form a final 12-membered template library 24.
1a: 5′-(scaffold)-GCATGTCTACCACGTTCTGAGCACACT-
GACTCCACTGTACACCTCGAG
Template libraries 26 and 28 were produced by one-pot mixed
ligations. The 3′-segments were assembled by split/pool oligo-
nucleotide synthesis: the first round of synthesis (5′-ATGAT[codon
1]CTA[scaffold codon]CATCCCACTC) was performed in 12
separate columns; the CPG was then removed from the columns
and pooled into a single container using acetonitrile. After agitation
by vortexing, the beads were redistributed to the columns evenly,
and subjected to a second round of oligonucleotide synthesis (5′-
[codon 2]). The CPG was pooled again, shaken, and transferred to
a column to add the 5′-CPRII phosphoramidite. The oligonucle-
otides were cleaved from CPG, prepared, and purified following
standard protocols for use in the ligation reaction with their
corresponding 5′-segments. This produced a mixture containing 144
different 3′-segments (12 × 12) per split/pool synthesis.
The splints for template libraries 26 and 28 were also generated
by split/pool oligonucleotide synthesis: the first round of synthesis
(5′- AACTT[anticodon 3]) was performed in 12 separate columns;
the CPG was then removed from the columns and pooled into a
5′-segment for 1a: 5′-(scaffold)-GCATGTCTACCACGTTCT-
GAGCAC
3′-segment for 1a: 5′-(P)-ACTGACTCCACTGTACACCTC-
GAG
Splint for 1a: 5′-GTGGAGTCAGTGTGCTCAGAAC
1b: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTTGCTGAAAT-
GATGGCTTTCTACCCACTC
5′-segment for 1b: 5′-(scaffold)-CCCTGTACACTTCCTCAAGTT
3′-segment for 1b: 5′-(P)-GCTGAAATGATGGCTTTCTAC-
CCACTC
Splint for 1b: 5′-ATCATTTCAGCAACTTGAGGAA
The general template structure for libraries 21 (12 × 1 × 1), 22
(1 × 12 × 1), and 23 (1 × 1 × 12) is shown below; each codon
is six bases in length:
5′-(scaffold)-CCCTGTACAC[codon 3]AAGTT[codon 2]AT-
GAT[codon 1]CTACCCACTC
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15613

A R T I C L E S
Tse et al.
single container using acetonitrile. After agitation by vortexing, the
beads were redistributed to the columns evenly, and subjected to a
second round of oligonucleotide synthesis (5′-ATCAT[anticodon
2]). The CPG was pooled, and the oligonucleotides were cleaved
from CPG and purified as above. This produced a mixture
containing 144 different splint oligonucleotides (12 × 12) per split/
pool synthesis.
µL of 50 mg/mL SIA (N-succinimidyl-iodoacetamide, Pierce) in
DMF and agitated for 5 min. The 3′-4-(diphenylphosphino)-
benzoic acid amide-linked oligonucleotide solution (∼100 µL of 1
mM DNA in 10 mM sodium phosphate buffer, pH 7.0) was then
added and the resulting mixture was agitated for 2 h. The mixture
was then diluted to 300 µL total, desalted by gel filtration (Nap-5),
and purified by reverse-phase HPLC using a gradient of acetonitrile
(8-80%) in 100 mM TEAA (pH 7.0).
For Hydrophobic Amino Acids: To improve the yield of
reactions with sparingly soluble amino acids (e.g., cyclohexylsta-
tine), the following protocol was used: 50 µL of 500 mM amino
acid in 500 mM NaOH was combined with 50 µL of 50 mg/mL
SIA in DMF and agitated for 5 min. To this solution, 50 µL of 1
M sodium phosphate buffer (pH 7.0) was added, followed by the
3′-4-(diphenylphosphino)-benzoic acid-amide linked oligonucleotide
solution (100 µL of 1 mM DNA in 10 mM sodium phosphate
buffer, pH 7.0). The resulting mixture was agitated for 2 h. Desalting
and purification were performed as described above.
The masses of all reagents were confirmed by MALDI-TOF mass
spectrometry (see Supporting Information, Figures S5, S6, and S7).
DNA-Templated Reactions and Capping Reactions. DNA-
templated reactions were performed at concentrations of 120 nM
template, or in the case of combinatorial syntheses, 120 nM per
template codon (1.44 µM template total). A concentration of 144
nM of each reagent was used (1.2 equivalents). Small-scale reactions
were performed in a nonstick 1.7 mL microcentrifuge tube (VWR).
Large-scale reactions (volume >1.5 mL) were performed in a 30
mL Oak Ridge polypropylene centrifuge tube (VWR).
Steps 1, 2, and 3 Amine Acylation. Templates and reagents
were combined in a solution containing 1 M NaCl, 100 mM pH
6.0 sodium 2-(N-morpholino)ethanesulfonate buffer (MES), 20 mM
N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride
(EDC), and 15 mM sulfo-N-hydroxysuccinimide (sNHS). 10% v/v
acetonitrile was present during library reactions (templates 24, 26,
and 28) to minimize binding to plasticware. The solution was briefly
heated to 55 °C and cooled to 25 °C prior to the addition of EDC
to facilitate DNA hybridization. After the addition of EDC, the
reaction was briefly agitated and left to react for 3 h at 25 °C.
Steps 1 and 2 Capping. To the previous reaction mixture (after
DNA-templated reaction), 1 µL of acetic anhydride was added per
200 µL of reaction solution (53 mM final concentration). This
reaction was briefly agitated and left to react for 2 h at 25 °C.
Steps 1 and 2 Cleavage. To the previous reaction mixture (after
capping), 25% v/v of 1 M NaOH was added to raise the pH and
trigger sulfone linker cleavage.10 The solution was briefly agitated
and left to react for 30 min at 25 °C. The pH was adjusted with
0.2 vol of 3 M NaOAc (pH 5.0) with 50 µg glycogen/mL and the
material was recovered by precipitation with isopropanol.
Macrocyclization and Purification of Macrocyclic Products.
The biotinylated step 3 products were captured from the reaction
mixture using PhyTip 1000+ columns loaded with streptavidin-
linked resin (Phynexus). The resin was washed once with 100 mM
pH 12.0 sodium N-cyclohexyl-3-aminopropanesulfonate buffer
(CAPS) at 45 °C, and three times with water (45 °C). The resin
was then exposed (∼3 min) to a solution of 50 mM NaIO4, 500
mM NaOAc pH 3.5, washed with 500 mM NaOAc pH 3.5, washed
with 25 mM NaOAc pH 3.5, and exposed to 100 mM pH 8.5 4-(2-
hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) to induce
macrocyclization and elution (1 h). Remaining product was eluted
from the resin by washing with 10% acetonitrile in water (v/v).
The HEPES solution and the 10% acetonitrile solution were
combined and precipitated with ethanol as described above to
recover macrocyclic products.
Characterization of DNA-Linked Species by Restriction
Digestion and MALDI-TOF for 1a-8a. The 5′ tripeptide scaffold
product (0.5-10 pmol) was hybridized with 50 pmol of a 5′- and
3′-biotinylated digestion oligonucleotide (10-mer, 5′-biotin-GTA-
GACATGC-biotinTEG) in 10-30 µL of NEB 4 buffer (New
England Biolabs). The digestion reaction with restriction endonu-
Template library 26 was produced by using the Lys-s scaffold
codon (5′-AAC) for the 3′-segments; the 5′-Lys-s-scaffold (see
Supporting Information) was attached to the 5′-segments. The
ensuing ligation, using 144 different 3′-segments, 12 different 5′-
segments, and 144 splints, produced library 26 with 1728 different
templates (12 × 144).
Template library 28 was produced by performing the synthesis
of library 26 for each corresponding 5′-scaffold/3′-scaffold codon
pair. The eight different libraries (1728 different templates each)
were then pooled together to form the complete template library
28 consisting of 13 824 different templates. The identities of all
appended 5′ scaffolds were confirmed by S1 nuclease digestion of
the templates (see below), followed by high-resolution LC/MS
analysis (Supporting Information, Figure S3).
Preparation of Step 1 and Step 2 DNA-Linked Reagents. The
commercial sources for the amino acids are listed in the Supporting
Information, Figures S5, S6, and S7.
For Hydrophobic Amino Acids: The corresponding reagent
oligonucletides were synthesized using 3′-amino-C7-CPG beads
(500 Å), and prepared using standard protocols. To attach the amino
acid, 50 µL of a ∼2 mM solution of the appropriate 3′-amine-
terminated DNA oligonucleotide was mixed with 50 µL of 100
mM amino acid in 1 M sodium phosphate buffer (pH 7.0). To the
resulting solution 25 µL of 100 mg/mL bis-[2-(succinimidooxy-
carbonyloxy)-ethyl]sulfone (BSOCOES, Pierce) in DMF was added.
The resulting mixture was sonicated to homogeneity and agitated
for 1 h at 25 °C. The product was diluted to 200 µL total, desalted
by gel filtration (Nap-5, Amersham Biosciences), and purified by
reverse-phase HPLC using a gradient of acetonitrile (8-80%) in
100 mM TEAA buffer (pH 7.0).
For Hydrophilic Amino Acids: When coupled to DNA, certain
amino acids (e.g., glycine) generate desired oligonucleotide con-
jugates that coelute on the HPLC with undesired BSOCOES-linked
oligonucleotide byproducts lacking the amino acid. To avoid this
problem, the concentration of amino acid was increased from 100
mM to 250 mM. The remainder of the synthesis and purification
was performed as described above.
Preparation of Step 3 DNA-Linked Reagents. Preparation
of 3′-Phosphine Reagents. The corresponding oligonucleotides
were synthesized on 3′-Amino-C7-CPG beads (500 Å). The 3′
FMOC group on the beads was removed using three consecutive
washes with 20% piperidine in DMF (5 min agitation at 25 °C for
each wash). The beads were then washed with DMF and acetoni-
trile. In a separate container, 4-(diphenylphosphino)-benzoic acid
was activated by reaction with N,N′-dicyclohexyl carbodiimide
(DCC) (100 µmol), 1-hydroxybenzotriazole (HOBt) (100 µmol),
and N,N-diisopropylethylamine (DIPEA) (100 µmol) in 600 µL dry
DMF for 1 h. The crude product was filtered through a glass frit to
remove the N,N′-dicyclohexylurea (DCU) and added to the oligo-
nucleotide beads with agitation for 4 h. The beads were washed
with DMF and acetonitrile. Treatment for 10 min with a solution
of 50:50 40% aqueous ammonium hydroxide:methylamine (v:v,
AMA) and 1 mg/mL tris-(2-carboxyethyl)phosphine hydrochloride
(TCEP) at 65 °C deprotected the modified oligonucleotide and
cleaved it from the resin. Products were dried in Vacuo and purified
by reverse-phase HPLC using a gradient of acentonitrile (8-80%)
in 100 mM TEAA (pH 7.0). After lyophilization, the samples were
resuspended in 10 mM sodium phosphate buffer (pH 7.0), divided
into 100 µL aliquots (∼100 nmol of DNA each), and immediately
frozen for later use.
For Hydrophilic Amino Acids: 50 µL of 500 mM amino acid
in 1 M sodium phosphate buffer (pH 7.0) was combined with 50
9
15614 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 1. Previously reported15 scheme for the synthesis of a DNA-templated macrocycle library, where R is -NHCH3 or tryptamine, and Ar is -(p-C6H4)-.
In contrast with the capping-based method described in the present work, each DNA-templated reaction in this scheme requires a bond formation step, a
streptavidin bead capture step, a washing step, and a product elution step.
clease Hpy8I (5 U, Fermentas), which cleaves the sequence 5′-
GTNNAC-3′, was added to effect template cleavage after the
seventh nucleotide. The digestion reaction was incubated for 1 h
at 37 °C. The digestion oligonucleotide byproducts were removed
by binding to the streptavidin-linked magnetic beads (80 µL,
Roche). The small-molecule products, linked to seven template
nucleotides, were collected from the supernatant and desalted using
Zip-Tips (Millipore). The sample was eluted directly onto a MALDI
analysis plate using 1 µL of 8:1 (50 mg/mL THAP in 1:1 water:
acetonitrile):(50 mg/mL ammonium citrate in water). Data was
collected on a Voyager DE MALDI-TOF mass spectrometer
(PerSeptive Biosystems) operated in reflector mode.
Expected mass (positive ion) for 1a ) 2612.6 Da, observed mass
) 2615.3 ( 6 Da.
Expected mass (positive ion) for 5a )2759.7 Da, observed mass
) 2762.4 ( 6 Da.
Expected mass (positive ion) for 6a ) 2896.7 Da, observed mass
) 2901.2 ( 6 Da.
S1 Nuclease Digestion for High-Resolution LC/MS Analysis.
Product was added to 50 µL pH 4.5 NH4OAc buffer in a glass
LC/MS vial. S1 nuclease (1 µL, 60 U, Promega) was then added.
Digestion was performed at 37 °C for 2 h. The mixture was frozen,
lyophilized, and resuspended in 0.1% (v:v) aqueous formic acid
for LC/MS analysis.
High Resolution LC/MS Characterization. Water was obtained
from a Milli-Q water purification system (Millipore). All solvents
and reagents used were mass spectrometry grade, including
methanol and acetonitrile (Fisher Scientific). For the LC/MS
analysis of 12-membered library 25 ultraperformance liquid chro-
matography (UPLC) was carried out with an ACQUITY UPLC
system (Waters) using a Waters BEH C18 column (2.1 mm i.d. ×
100 mm, 1.7-µm particles) and a linear gradient of 0.1% (v:v)
formic acid in acetonitrile (0%-100%) in 0.1% (v:v) aqueous
formic acid. The eluent was directly injected into a Q-TOF Premier
mass spectrometer (Waters) fitted with an electrospray interface.
Data was acquired and processed with MassLynx 4.1 software
(Waters). Mass spectra were recorded in the negative mode within
a m/z range of 100-1200. The masses for macrocycle libraries 21,
22, and 23 are shown in the Supporting Information, Figures S8,
S9, and S10.
NanoLC/MS Analysis of 1728-Membered Library 27. LC/
MS analysis was performed on a Q-TOF Premier mass spectrometer
equipped with a nanoelectrospray interface and the nanoACQUITY
UPLC system (Waters). The nano-LC used a flow of a rate of 0.2
µL per min through a BEH C18 nanoLC column (Waters, 75 µm
i.d. × 360 µm o.d. Ten cm, 1.7-µm particles) and a gradient of
methanol (1% to 99%) in 6 mM aqueous triethylammonium
bicarbonate (TEAB). The nanoLC column was connected to a
PicoTip Emitter (Waters, tubing i.d. 360 µm, o.d. 20 µm, tip i.d.
10 µm) by means of a ZVD microunion that allowed for a dead-
volume-free butt connection. The column was optimized for
maximum sensitivity of a standard raffinose (100 ng/nL) signal in
an identical mobile phase. Data was acquired and processed with
MassLynx 4.1 software. Mass spectra were recorded in negative
ion mode within the m/z range of 750-1300. Masses were identified
with a minimum mass accuracy of ( 0.05 Da.
Results and Discussion
Development of a Capping-Based DNA-Templated Macro-
cycle Library Synthesis. Previously we reported a DNA-
templated route for the preparation of a library of DNA-linked
synthetic macrocycles from a corresponding library of lysine-
linked DNA oligonucleotides (1, Figure 1).15 Our original
synthesis began with two successive DNA-templated amine
acylation reactions, each consisting of carbodiimide-mediated
amide bond formation using biotinylated DNA-linked amino
acid building blocks (2, 3), capture of the desired reaction
intermediate using streptavidin-linked beads, and base-induced
linker cleavage to release the coupled reaction product (5, 6).
A third DNA-templated amine acylation step was then per-
formed using a biotinylated DNA-linked building block contain-
ing a novel Wittig ylide linker (4). After capture of the step 3
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15615

A R T I C L E S
Tse et al.
Figure 2. Capping-based strategy for DNA-templated macrocycle synthesis. The use of a capping reagent (here, acetic anhydride) prevents unreacted or
improperly reacted material from participating in further reactions. The final streptavidin bead capture and macrocyclization steps achieve effective purification
of the final product, obviating the need to purify after each DNA-templated reaction as in Figure 1 and reducing the total number of required manipulations.
reaction product using streptavidin-linked beads, the im-
mobilized product was exposed to aqueous NaIO4 to unmask a
side-chain-linked aldehyde (7); elevating the pH of the buffer
to 8.5 then induced Wittig olefination and macrocyclization with
simultaneous cleavage of the biotinylated linker. The cleavage
of this linker caused the desired macrocycle product (but not
unreacted macrocycle precursors or step 3 reagents) to self-
elute from the beads (8). Using this strategy, we generated a
pilot library of 65 unique macrocyclic small molecules in a DNA
sequence-programmed manner.15 While effective, this approach
required that the products of each DNA-templated step undergo
capture with streptavidin-linked beads, washing, elution, and
buffer exchange. The removal of these requirements in a manner
that does not compromise library quality would significantly
enhance the ease and efficiency of DNA-templated library
synthesis.
In order to extend this DNA-templated synthesis to >100-
fold larger libraries, we developed a simpler, more robust route
to DNA-templated macrocycles. We hypothesized that the final
biotin pulldown, Wittig macrocyclization, and self-elution steps
achieve a sufficient purity of final macrocycle product such that
the material entering step 3 does not require purification after
steps 1 and 2. We further anticipated that hybridized reagent
oligonucleotides from previous steps would not impair subse-
quent reactions, so long as subsequent steps used template bases
closer to the site of small-molecule attachment. To avoid
truncated macrocycles arising from a failed step 1 or step 2
reaction, we envisioned using acetic anhydride to cap unreacted
templates after steps 1 and 2, thereby preventing any unreacted
or improperly reacted starting material from participating in step
3 and the macrocyclization reaction (Figure 2). This strategy is
reminiscent of the capping procedures used in solid-phase
peptide and oligonucleotide synthesis to prevent truncated
species from contaminating full-length products.69,70
viability. After each step, we removed an aliquot of the mixture
in order to monitor the progress of the synthesis using denaturing
polyacrylamide gel electrophoresis (PAGE); the resulting gel
reveals efficient product formation during each DNA-templated
reaction (Figure 3, lanes b, e, and h). Similarly, each cleavage
step (lanes d and g) shows quantitative disappearance of the
product band from the preceding lane.
To gauge the efficiency of the capping step, we removed
aliquots from each of the first two rounds of templated synthesis,
both before (lanes b and e) and after (lanes c and f) the addition
of acetic anhydride; intermediates were then cleaved at high
pH, enzymatically digested with Hpy8I to remove all but seven
nucleotides of the DNA template, and analyzed by MALDI-
TOF mass spectrometry. The resulting spectra show quantitative
conversion of unreacted step 1 starting material (1a) to acety-
lated product (9a) upon addition of acetic anhydride (Figure
3), with no adulteration of the desired product (5a). Similarly,
unreacted starting material from the second reaction (5a) was
converted to acetylated material (10a) after step 2, with no
visible modification to the desired product (6a). The persistence
of 9a in the second step confirms that acetylated products are
unreactive in subsequent steps (Figure 3).
The product of the third templated reaction (Figure 3, lane
h) was purified with streptavidin-linked beads, washed as
previously described, and exposed to aqueous sodium periodate
to enable cyclization. The final macrocyclic product was eluted
from the resin by exposure to pH 8.5 buffer, as confirmed by
PAGE analysis (see Supporting Information, Figure S1).
Restriction digestion followed by MALDI-TOF analysis (see
Supporting Information, Figure S1) revealed a product mass
consistent with the expected macrocycle 8a (expected: 2842.7
Da, found 2846.6 ( 6 Da). The overall yield was calculated by
PAGE and densitometry at 1% for all three templated reactions,
the streptavidin purification, and macrocyclization. Notably, no
significant peaks corresponding to any of the acetylated products
(oxidized versions of 9a and 10a) were seen, confirming the
effectiveness of the purification step.
To test the viability of this approach, we synthesized the
single macrocycle 8a from DNA template 1a and the corre-
sponding reagents 2a, 3a, and 4a (Figure 3) using the capping-
based method shown in Figure 2. The codons were selected
from our previous work15 to ensure template and reactant
This new capping-based strategy offers significant advantages
over the previous library-synthesis route. First, it prevents
material that has not reacted or that has reacted improperly from
continuing to participate in the synthesis, thereby eliminating
byproducts including contracted rings. Second, it reduces the
(69) Merrifield, B. Science 1986, 232, 341–7.
(70) Caruthers, M. H.; Beaton, G.; Wu, J. V.; Wiesler, W. Methods
Enzymol. 1992, 211, 3–20.
9
15616 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 3. Denaturing PAGE and MALDI-TOF analysis of the capping method applied to DNA-templated macrocycle precursor synthesis.
Development of an Optimized Codon Set for Library
Synthesis. The codons (variable regions) within a library of DNA
templates play crucial roles during and after library translation.
They mediate changes in effective molarity that promote the
efficient reactivity of otherwise highly dilute reactants, deter-
mine the sequence specificity of DNA hybridization (and
therefore the specificity of building-block reactivity), and reveal
the structures of active library members after in Vitro selection
and PCR amplification. We recently characterized the relation-
ship between template secondary structure and DNA-templated
reactivity, and discovered that optimal reactivity arises from
templates that have an intermediate degree of secondary
structure that is sufficient to promote template compaction but
insufficient to block reagent hybridization.68 We found that the
window for maximum reactivity is in the range of -7 to -3
kcal/mol of predicted template folding energy.
total number of manipulations required to generate the final
product by eliminating streptavidin capture steps, washing steps,
and exposure to plasticware. Indeed, all three DNA-templated
reactions are performed within the same vessel using the new
strategy. The removal of two sets of streptavidin capture,
washing, elution, and buffer exchange steps enabled us to
complete the macrocycle synthesis in less than three days
starting from reagents and templates, and in less than three
weeks starting entirely from commercially available starting
materials. Third, the fact that step 1 and step 2 reagents in this
new method are no longer biotinylated simplifies their prepara-
tion and reduces cost. Finally, we speculate that the reagent
DNA strands from completed steps assist in rigidifying template
DNA strands by hybridizing to the templates, eliminating
template secondary structural motifs that can potentially impede
reactivity by blocking proper reagent annealing.68 In summary,
this new methodology enhances the ease and efficiency of
multistep DNA-templated syntheses, and produces final product
in a more robust fashion.
Using the codon-generating algorithm shown in Figure 4, we
produced three new sets of 40 codons each such that a minimum
of three mismatches are present between any mismatched
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15617

A R T I C L E S
Tse et al.
Figure 4. Method for the computational generation, modeling, and screening of a codon set suitable for DNA-templated library synthesis.
(noncognate) template and reagent pair. We screened in silico
templates containing these codons using the Oligonucleotide
Modeling Platform (OMP),71,72 and eliminated those codons that
frequently resulted in templates with insufficient or excessive
folding energies.68 This strategy generated codon sets that result
in templates closely fitting the ideal energetic profile.
To encode 1728 building-block combinations in a 12 × 12
× 12 arrangement, we used 36 codons (3 sets of 12 each) from
the refined codon set. We synthesized 12 representative
templates (1A-2A-3A, 1B-2B-3B, etc. through 1L-2L-3L)
such that all 36 codons were represented within the 12-template
library. The corresponding 36 complementary reagent oligo-
nucleotides (step 1 reagents complementary to codons 1A to
1L, step 2 reagents complementary to codons 2A to 2L, and
step 3 reagents complementary to codons 3A to 3L) were also
synthesized. Each reagent contained an 11-base coding region
and a 4-base Ω constant region, as previously described.73 The
templates were synthesized with a 5′ amine to enable amine
acylation; reagents were prepared with a 3′ D-Phe group ending
in a carboxylic acid.68
To rigorously test the sequence fidelity of templates contain-
ing the new codon set, we exposed each of the 12 representative
templates (and by extension, each of the 36 codons) to each of
the 36 DNA-linked reagents under DNA-templated amine
acylation conditions, and measured the amount of resulting
product by PAGE. Product yields arising from sequence-
matched reactivity are shown along the diagonals of the matrices
in Figure 5; mismatched reactivity is represented by all off-
diagonal yields. The matched reagents regularly resulted in
reaction yields in the 60-70% range, with only five matched
combinations dropping below the 50% yield mark. In contrast,
none of the 396 mismatched combinations exhibited reactivity
resulting in more than 8% yield, and only five combinations
exceeded 5% yield. Notably, this study tested codon sequence
fidelity under more stringent conditions than would be present
Figure 5. DNA-templated reaction yields for all possible matched and
mismatched combinations of templates and reagents used in this work.
Product yields arising from sequence-matched reactivity are shown on the
diagonals of the matrices; mismatched reactivity yields are shown in off-
diagonal cells. Cells are colored by percent yield from highest (red) to lowest
(green) in spectral order.
(71) SantaLucia, J., Jr.; Hicks, D. Annu. ReV. Biophys. Biomol. Struct. 2004,
33, 415–40.
during normal library synthesis. In the above experiments, one
reagent was mixed with one template (matched or mismatched)
and allowed to react; in an actual library synthesis, the matched
reagent is always present in solution and its hybridization to
(72) SantaLucia, J., Jr. Methods Mol. Biol. 2007, 402, 3–34.
(73) Gartner, Z. J.; Grubina, R.; Calderone, C. T.; Liu, D. R. Angew. Chem.,
Int. Ed. 2003, 42, 1370–5.
9
15618 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 6. High-resolution LC/MS analysis of macrocycle 8b after S1 nuclease digestion. Macrocycle 8b was synthesized using the capping-based approach
from template 1b using the new codons 1A, 2A, and 3A.
the template precludes the hybridization of mismatched reagents,
reducing undesirable reactivity.
Based on our previous in Vitro selection studies,75 we estimated
that maintaining final product abundances within a range of
∼100-fold in the final library would facilitate in Vitro selection
by minimizing problems associated with more extreme stoichi-
ometry differences between products, such as the need for more
library material entering selection to enable active but under-
represented members to survive selection and PCR amplification.
To test the codon set for use in multistep DNA-templated
macrocycle synthesis, we repeated the synthesis of macrocycle
8a using a representative template (1b, Figure 6) from the new
codon set side-by-side with a template from the original, smaller
codon set.15 The template using the original codons produced
8a in 1% final overall yield of pure macrocycle. The new codon
set provided identical macrocycle 8b in 3% overall yield of pure
product. (The improvement in yield arises largely from the use
of the Ω architecture,73 as the end-of-helix reagents using the
same template 1b resulted in a 0.8% final overall yield.) To
further characterize the structure of the macrocycle 8b, we
digested the product with S1 nuclease to liberate the macrocycle
from the DNA template. S1 nuclease digestion leaves a single
nucleotide 5′-monophosphate at the 3′ end of its digestion
products,74 producing chemical species that are sufficiently small
to be analyzed by high-resolution electrospray LC/MS. The
crude digested sample yielded a mass spectrum consistent with
the expected structure (expected: 972.35 Da, observed: 972.34
( 0.05 Da), as shown in Figure 6. Collectively, the above results
indicate that the new codon set exhibits an excellent combination
of efficient reactivity toward matched reagents and low reactivity
toward mismatched reagents, while maintaining compatibility
with multistep DNA-templated macrocycle synthesis.
These results also validate the template design principles we
recently described68 that relate DNA-templated reaction ef-
ficiency with predicted template folding energy. High-quality
codon sets are critical to the success of any nucleic acid-
templated synthesis. The capability developed here to accurately
and rapidly design and test the reactivities of many library
codons and templates is essential to current and future multi-
plexed DNA-templated syntheses.
The Effect of Building-Block and Scaffold Variability on
DNA-Templated Macrocycle Synthesis. Next we performed a
series of experiments to determine the range of building blocks
compatible with an expanded macrocycle library synthesis.
We chose six different amino acids to represent a diverse
range of chemical functionalities (Figure 7): glycine (small and
flexible), 1-amino-1-cyclohexyl-carboxylic acid (R,R-disubsti-
tuted), proline (N-substituted and constrained), statine (extended
backbone), N′,N′,N′-trimethyllysine (charged), and 4-benzoyl-
D-phenylalanine (D-stereochemistry and hydrophobic). Using
macrocycle 8a as the centerpiece of our study, we synthesized
three separate series of macrocycles (11, 12, and 13), each
containing amino acid substitutions at one of the three variable
macrocycle positions (macrocycle series 11 for substitutions at
R1, series 12 for R2, and series 13 for R3). Each synthesis was
started from the last common template starting material (1a,
5a, and 6a, for 11, 12, and 13, respectively) and the codons
used throughout the study were identical between templates to
enable the direct comparison of reaction yields. The analogs of
reagents 2a, 3a, and 4a containing each of the six building
blocks in Figure 7 were also prepared. Macrocycle 8a was
synthesized side-by-side with each sublibrary to provide a
benchmark of reaction efficiency.
The relative yields for all three series of macrocycles are
shown in Figure 7, as calculated using densitometry with internal
standards. With the exception of 2- to 3-fold lower yields from
the use of the R,R-disubstituted building block in step 1 and
step 2, reactivity was quite uniform (Figure 7). In all 18 cases,
the products containing these six building blocks were viable
substrates for the remaining steps of the library synthesis,
including macrocyclization. These findings suggest that a wide
variety of building-block parameters including size, flexibility,
charge, hydrophobicity, and stereochemistry can be accom-
(75) Doyon, J. B.; Snyder, T. M.; Liu, D. R. J. Am. Chem. Soc. 2003, 125,
(74) Vogt, V. M. Methods Enzymol. 1980, 65, 248–55.
12372–3.
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15619

A R T I C L E S
Tse et al.
Figure 7. Systematic study of building-block compatibility with DNA-templated macrocycle synthesis. Using macrocycle 8a as a benchmark of “normal” reactivity,
we selected a set of six distinct amino acids for substitution, one at a time, into the macrocycle at each of the three building block positions (R1, R2, and R3).
Macrocycle series 11 (substitution at R1), 12 (substitution at R2), and 13 (substitution at R3) were synthesized using the corresponding reagent sets and starting
materials 1a, 5a, and 6a, respectively. Percent yields of each macrocycle as analyzed by PAGE using internal standards are shown in the table.
modated by our DNA-templated macrocycle synthesis. Given
the range of building-block reactivities observed over the three
sites of substitution, we estimate the ratio between the lowest
potential macrocycle yield and the highest, assuming indepen-
dent reactivity, to be ∼1:50; this ratio is within our target range
of 100-fold. To further decrease this range, we avoided R,R-
disubstituted amino acids in the first two steps in subsequent
macrocycle library syntheses.
To further expand the size and diversity of the final library,
the starting scaffold of template 1b was varied among lysine,
ornithine (14), diamino-butyric acid (15), and diamino-propionic
acid (16, Figure 8). Each of these diamino acids incrementally
contracts the macrocycle ring by reducing the number of side-
chain methylenes and enables access to significantly different
preferred macrocycle conformations. In addition, the tartaramide
group that gives rise to the aldehyde during macrocyclization,
originally located on the side-chain ε-amine of lysine in 1b,
can alternatively be placed on either the R-amine or side-chain
amine (“s-amine”) of the diamino acid in question (scaffolds
17, 18, 19, and 20, for the four respective diamino acids). These
two choices generate macrocycles of opposite building-block
orientation (Figure 8). Between the two different orientations
and four different amino acids, a total of eight different scaffolds
are possible. The effect of changing scaffolds on macrocycle
shape and functional-group presentation is demonstrated in
Figure 9, which shows that energy-minimized macrocycles with
the same three building blocks but using two different scaffolds
can have dramatically different predicted conformations.
To test the viability of each of the eight scaffolds, we
synthesized a series of templates using codons identical to those
of 5′-lysine(tartaramide) template 1b (codons 1A, 2A, and 3A),
but bearing the different scaffolds on their respective 5′ ends
(14-20). Using conditions and reagents identical to those used
to translate template 1b into macrocycle 8b, we translated
templates 14-20 into their corresponding macrocycles. Final
yields of macrocycles, ranging from 6.2% (for 19) to 2.6% (for
17), are shown in Figure 8 and indicate efficient synthesis with
all eight scaffolds. These results indicate the viability of using
variable macrocycle scaffolds as a source of library diversity.
The use of these eight scaffolds significantly augments the
variety of macrocyclic structures that can be accessed within
the framework of the library synthesis scheme.
Validation of 36 Building Blocks By Macrocycle Sublibrary
Synthesis. Guided by the above findings, we selected three sets
of 12 amino acid building blocks for step 1, step 2, and step 3
of the macrocycle library synthesis (Figure 10). The building
blocks were chosen on the basis of their diverse chemical
functionalities, compatibility with the synthetic route (no side-
chain free amines or carboxylic acids), and variable backbone
lengths. To assess their suitability for simultaneous reaction in
a single pot during each DNA-templated library synthesis step
and to establish their viability as macrocycle building blocks,
9
15620 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 8. Eight scaffolds for DNA-templated macrocycle syntheses. The tartaramide group (aldehyde precursor) can be placed on either the R-amine or
side-chain (“s”) amine, resulting in two different orientations of building blocks within the resulting macrocycles. The eight scaffolds (denoted by the circled
S) were transformed into their corresponding macrocycles using the same reagents used in the synthesis of macrocycle 8b from scaffold 1b; the respective
overall percent yields for final purified macrocycles are shown in the table above.
Figure 9. Three-dimensional models for two sets of MM2 energy-minimized macrocycles. The macrocycles within each set incorporate identical building
blocks, but use a different scaffold (either the Lys-s or Dpr-R scaffold). The different scaffolds can result in very different preferred conformations.
we synthesized three 12-membered DNA-templated sublibraries
by installing all 12 library building blocks for a given step into
the macrocycle while keeping the other two amino acids of the
macrocycle constant (libraries 21, 22, and 23, for variation at
R1, R2, and R3, respectively). These three test libraries col-
lectively tested the ability of all 36 library building blocks to
react in their designated positions and to generate corresponding
macrocyclic products.
We prepared all 36 of the necessary DNA-linked library
reagents using the new codon set shown in Figure 11 by standard
DNA synthesis, followed by small-molecule conjugation and
HPLC purification. Through a convergent, modular ligation-
based strategy (described below), we also generated three
corresponding 12-membered template libraries containing the
L-lysine s-amine scaffold linked to each of 12 oligonucleotides,
representing a 12 × 1 × 1 library (testing all 12 step 1 reagents,
for 21), a 1 × 12 × 1 library (testing all 12 step 2 reagents, for
22), and a 1 × 1 × 12 library (testing all 12 step 3 reagents,
for 23).
Each template library was then translated using the new
capping-based macrocycle synthesis method (Figure 2) to
generate its corresponding macrocycle library. Macrocycle
libraries 21, 22, and 23 were synthesized in overall yields of
1.9%, 2.4%, and 1.1%, respectively. While some of our earlier
macrocycle pilot library syntheses were sufficiently simple to
be analyzed by low-sensitivity, low-resolution MALDI mass
spectrometry of heptanucleotide-macrocycle conjugates,15 the
characterization of these 12-membered test libraries (and larger
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15621

A R T I C L E S
Tse et al.
Figure 10. Amino acid building blocks for macrocycle libraries 21, 22, and 23.
Figure 11. Template and reagent sequences that mediate the synthesis of macrocycle libraries 21, 22, and 23. The optimized template codons and corresponding
reagent anticodons resulting from the method shown in Figure 4 are shown.
subsequent libraries) requires an improved analysis method. As
we previously did with macrocycle 8b, we therefore digested
libraries 21, 22, and 23 with S1 nuclease and analyzed the
resulting products by high-resolution LC/MS analysis. The
removal of all but a single nucleotide 5′-monophosphate by S1
nuclease digestion enables the mass characterization of these
DNA-templated products to an accuracy of ( 0.05 Da or better.
The resulting spectra (Figure 12) confirm the presence of 34/
36 (94%) of the total possible macrocyclic products in these
three test libraries (explicit masses are listed in the Supporting
Information, Figures S8, S9, and S10). In addition, the number
of unexpected masses observed in significant abundance rep-
resented only a small fraction of the total observed masses,
suggesting the effectiveness of the macrocycle selection step
as a means of simultaneously effecting the macrocyclization
and purification of final products. Taken together, these results
validate the codons, starting scaffolds, and building blocks that
make up a diverse, densely functionalized 8 × 12 × 12 × 12
(13 824-membered) DNA-templated macrocycle library.
Evaluating Sequence Fidelity During Library Translation.
While the three 12-membered sublibraries assess the ability of
the 36 building blocks to participate in DNA-templated mac-
rocycle synthesis, they do not confirm the sequence fidelity of
each DNA-templated library synthesis reaction in which 12
reagents and thousands of templates are all present in one
solution. The codon sequence fidelity tests (Figure 5) indicate
that each DNA-templated coupling reaction can proceed se-
quence specifically, but they were not conducted in the context
of a multistep macrocycle synthesis. To rigorously evaluate the
sequence specificity of macrocycle translation, we synthesized
a sublibrary of exactly 12 templates using codon combinations
1A-2A-3A through 1L-2L-3L, such that each codon was
represented within the template set once; all other features (5′
scaffold, primer binding sites, and constant regions) of the
9
15622 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 12. LC/MS analysis of 12-membered macrocycle libraries 21, 22, and 23 following S1 nuclease digestion. In total, 34 of the 36 expected macrocyclic
species (94%) are visible.
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15623

A R T I C L E S
Tse et al.
Figure 13. High-resolution LC/MS analysis of the 12-membered (25) and 1728-membered (27) DNA-templated macrocycle sublibraries. From library 25,
a stringent test of sequence specificity, 10 of 12 expected macrocycle masses were detected, and only 8.6% of masses corresponding to sequence-mismatched
products were observed. For library 27, 94% of expected masses were observed. The macrocycles eluted over a total of ∼2400 s; the representative mass
spectrum shown here is of a 35-s elution window. Expected masses are in red and observed masses are in black.
templates were identical. This template library (24, Figure 13)
therefore encodes the synthesis of 12 distinct macrocycles, each
comprising three unique building blocks, that collectively use
all of the 36 DNA-linked reagents.
corresponding to the step 1 reagent set, 2A-2L for step 2, and
3A-3L for step 3) as did template library 24, but in a
combinatorial 12 × 12 × 12 format (Figure 13). Whereas library
24 encoded 12 distinct macrocycles, library 26 therefore encodes
all 1728 possible macrocycles.
If translation occurs with perfect sequence dependence, only
the 12 expected macrocyclic products would result from DNA-
templated library synthesis starting with 24. Cross-reactivity
between mismatched templates and reagents would appear as
unintended macrocycle products, of which 1716 ([12 × 12 ×
12]-12) are possible. For the translation of this library, we used
three complete sets of reagents linked to anticodons comple-
mentary to 1A-1L, 2A-2L, and 3A-3L. Compared with the
reagent sets used for the 1728-membered sublibrary, the reagent
sets for this 12-membered sequence-specificity test contain two
amino acid substitutions to facilitate analysis by replacing the
lowest yielding building blocks, as indicated by the spectra of
sublibraries 21 and 22 (Figure 12).
Execution of the library synthesis on template library 24 with
the step 1, 2, and 3 reagents resulted in the formation of
macrocycle library 25 (Figure 13) in 1.4% overall yield. High-
resolution LC/MS analysis of this library after S1 nuclease
digestion produced a spectrum in which 10 of the 12 expected
macrocycle masses were present with a signal above a threshold
of 3000 ion counts. Of the 1716 possible macrocycles corre-
sponding to unanticipated cross reactivity, masses consistent
with only 148 (8.6%) were seen at or above the same signal
threshold. The low fraction of nonsequence-programmed prod-
ucts formed during this stringent test of sequence specificity,
consistent with the explicit tests of mismatched codon-mediated
reactivity described above (Figure 5), suggests that nonsequence-
programmed macrocycle formation should represent only a small
fraction of total macrocycles in a full 1728-membered macro-
cycle sublibrary.
Template library 26 was then reacted as described above with
12-membered step 1, step 2, and step 3 reagent sets, followed
by streptavidin capture and macrocyclization, to yield final
macrocycle library 27 (Figure 13). The final yield, across all
three amide-bond formations, purification, and cyclization was
1.0%. In total, 48 picomoles of purified product were generated.
This library of up to 1728 macrocycles was then digested
with S1 nuclease and subjected to high-resolution nano-LC/
MS analysis (Figure 13). The resulting LC/MS spectrum of 27
was computationally compared to the expected macrocycle
masses of all 1,728 products by searching for the presence of
all expected product masses based on a minimum distinguishable
signal threshold empirically determined immediately prior to
injection of the library sample. Of the 1394 discrete expected
masses, 1317 (94%) were detected within the spectrum at a
signal level exceeding the threshold of 1000 ion counts. In
addition, an examination of representative portions of the LC/
MS spectrum revealed relatively few unanticipated masses (see
Figure 13 for an example). These results are consistent with
the presence of up to 1633 (95%) of the 1728 theoretical
macrocyclic species in the library. Taken together, these findings
suggest the robustness of the DNA-templated macrocycle library
synthesis scheme and the viability of synthesizing a large scale
DNA-templated library suitable for in Vitro selection.
DNA-Templated Synthesis of a 13 824-Membered Macro-
cycle Library. Integrating the above developments, we synthe-
sized the complete library of 13 824 DNA-templated macro-
cycles using all eight validated scaffolds (Figure 8) and all 36
building blocks (Figure 10). The starting template pool (28) was
generated by split-pool DNA synthesis and enzymatic ligation
as described above, but using eight different scaffolds (14-20)
instead of one scaffold. Since the structural elucidation of any
library member surviving selection requires the identification
of the starting scaffold as well as the three building blocks, it
is necessary to encode the identity of the scaffold in each
template sequence. As the scaffold-encoding sequence does not
DNA-Templated Synthesis and Characterization of
a
1728-Membered Macrocycle Sublibrary. Next we synthesized
and characterized the complete 1728-membered sublibrary
representing one scaffold and all possible combinations of the
36 macrocycle building blocks. Template library 26 was
generated by a combination of split-pool solid-phase oligo-
nucleotide synthesis and enzymatic ligation; this template library
contains the codons for all 36 unique reagents (1A-1L,
9
15624 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008

Library of 13 000 Synthetic Small-Molecule Macrocycles
A R T I C L E S
Figure 14. Synthesis of a 13 824-membered macrocycle library (32) from DNA template library 28. Eight template library-linked scaffolds were processed
in a single solution through three DNA-templated reactions, each using 12 building blocks each, to produce a final macrocycle library with 13 824 members.
The synthesis was performed in triplicate, with an average overall purified yield of 1.6% as determined by PAGE (one example is shown in the gel above).
In total, >375 pmol of library was produced, sufficient material for hundreds of in Vitro selections.
direct any chemical reactions through DNA hybridization, it
can be of minimal length, and placed anywhere within the
template between the two PCR primer binding sites. We chose
to encode the scaffold by using a three-base sequence near the
3′ (nonreacting) end of the templates (Figure 14; scaffold codons
are shown in Supporting Information, Figure S2), which were
otherwise prepared as described above.
The synthesis scheme for the 13 824-membered library is
shown in Figure 14. Using the capping strategy, the template
library (28) was reacted with the step 1 reagent set using EDC/
sNHS to effect amine acylation; products were capped with
acetic anhydride, and cleaved by pH elevation with hydroxide
to generate intermediate library 29. After recovery by precipita-
tion, 29 was reacted with the step 2 reagent set under standard
amine acylation conditions; products were cleaved, capped, and
precipitated as before to produce intermediate 30. This mixture
was reacted with the step 3 reagent set under amine acylation
conditions. The crude product was captured with streptavidin-
linked beads, washed extensively, and treated with aqueous
periodate to unmask the scaffold aldehyde group (intermediate
31). Upon exposure to pH 8.5 solution, the macrocyclized
material eluted from the streptavidin-linked beads to afford the
final macrocycle library product (32).
14). In total, sufficient material (>375 pmols) was synthesized
and isolated for hundreds of in Vitro selections.15,75 The
sublibrary characterization experiments described above (Figures
12 and 13) suggest that this library contains more than 13 000
correctly translated macrocycle members; as such, the library
represents one of the largest collections of synthetic macrocycles
of this size (17- to 25-membered rings) reported to date.76-82
This DNA-templated macrocycle library provides a rich
starting point for the discovery of functional synthetic macro-
cycles through in Vitro selections against biological targets. We
previously described in Vitro selections of small molecule-DNA
(76) Spatola, A. F.; Crozet, Y. J. Med. Chem. 1996, 39, 3842–6.
(77) McBride, J. D.; Freeman, H. N.; Leatherbarrow, R. J. Eur. J. Biochem.
1999, 266, 403–12.
(78) Jefferson, E. A.; Arakawa, S.; Blyn, L. B.; Miyaji, A.; Osgood, S. A.;
Ranken, R.; Risen, L. M.; Swayze, E. E. J. Med. Chem. 2002, 45,
3430–9.
(79) Kohli, R. M.; Walsh, C. T.; Burkart, M. D. Nature 2002, 418, 658–
61.
(80) Sedrani, R.; Kallen, J.; Martin Cabrejas, L. M.; Papageorgiou, C. D.;
Senia, F.; Rohrbach, S.; Wagner, D.; Thai, B.; Jutzi Eme, A. M.;
France, J.; Oberer, L.; Rihs, G.; Zenke, G.; Wagner, J. J. Am. Chem.
Soc. 2003, 125, 3849–59.
(81) Schmidt, D. R.; Kwon, O.; Schreiber, S. L. J. Comb. Chem. 2004, 6,
286–292.
The synthesis was repeated three times, with an average yield
of 1.6% as indicated by PAGE and densitometry analysis (Figure
(82) Goto, Y.; Ohta, A.; Sako, Y.; Yamagishi, Y.; Murakami, H.; Suga,
H. ACS Chem. Biol. 2008, 3, 120–9.
9
J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008 15625

A R T I C L E S
Tse et al.
conjugates that exhibited enrichment factors of ∼100- to 1000-
fold per round.75 When such selections are applied to the 13 824-
membered macrocycle library described above, two rounds of
selection in principle could result in a single active library
member predominating the molecules surviving selection. As
these in Vitro selections are quite general, require only very
simple equipment, and can be performed in less than one day
regardless of library size, efforts are currently underway to select
this unique library of DNA-templated macrocycles against a
wide variety of protein and RNA targets of biological and
biomedical interest.
examined the range of building blocks that are compatible with
macrocycle synthesis by incorporating a representative set of
different types of amino acids into corresponding macrocycles.
We expanded the structural diversity of the library through the
incorporation of a diverse set of 36 building blocks and eight
different scaffolds that alter the size and orientation of the
macrocyclic ring. Finally, we evaluated the efficacy and
sequence fidelity of our library syntheses by generating a series
of relevant macrocycle sublibraries and developing high-
resolution LC/MS-based methods for characterizing each pos-
sible product.
Integrating these developments, we synthesized a final library
of >13 000 DNA-linked synthetic macrocycles. To our knowl-
edge this library represents one of the largest collections of
medium-sized synthetic macrocycles reported to date. Sufficient
material of the completed macrocycle library was generated for
many in Vitro selections against biological targets of interest.
In addition, the principles and strategies described here facilitate
future DNA-templated synthesis efforts.
Conclusion
Biology-inspired approaches to the discovery of functional
synthetic molecules bring the efficiency of in Vitro selection,
the sensitivity of DNA amplification, and the ease of DNA
sequence analysis to bear on structures that can only be accessed
through synthetic organic chemistry. Previously, we used
multistep DNA-templated synthesis to translate a modest library
of 65 DNA templates into a corresponding library of synthetic
macrocycles.15
In the present work, we expanded the scope of DNA-
templated macrocycle library synthesis considerably through a
series of studies that each addressed a specific limitation of our
original approach. We increased the robustness and efficiency
of the synthetic methodology by adopting a streamlined capping-
based synthesis strategy that reduces the total number of
manipulations, eliminates impurities, and reduces cost. In
addition, we designed and computationally screened a new set
of library codons in accordance with the principles revealed in
our recent study on the relationship between template sequence
and DNA-templated reactivity.68 We experimentally validated
the ability of this new codon set to mediate DNA-templated
reactions efficiently and in a sequence-specific manner. We also
Acknowledgment. This research was supported by the NIH/
NIGMS (R01GM065865) and the Howard Hughes Medical Insti-
tute. TMS and BNT gratefully acknowledge NSF Graduate
Research Fellowships. TMS also acknowledges the support of an
ACS Division of Organic Chemistry Fellowship sponsored by
Organic Reactions, Inc. We thank Zev J. Gartner, Matthew W.
Kanan, and Christopher T. Calderone for many helpful discussions.
Supporting Information Available: Additional DNA se-
quences, procedures, experimental results, and the complete
author list for reference 60. This material is available free of
charge via the Internet at http://pubs.acs.org.
JA805649F
9
15626 J. AM. CHEM. SOC. VOL. 130, NO. 46, 2008