Proc. Natl. Acad. Sci. USA
`Vol. 89. PP. 5381-5383, June 1992
`Chemistry
`
`Encoded combinatorial chemistry
`(chemical repertoire/encoded libraries/commaless code)
`
`SYDNEY BRENNER AND RICHARD A. LERNER
`
`Departments of Chemistry and Molecular Biology, The Scripps Research Institute, 10666 North Torrey Pines. La Jolla, CA 92037
`
`Contributed by Sydney Brenner, March 3, 1992
`
`The diversity of chemical synthesis and the
`ABSTRACT
`power of genetics are linked to provide a powerful, versatile
`method for drug screening. A process of alternating parallel
`combinatorial synthesis is used to encode individual members
`of a large library of chemicals with unique nucleotide se-
`quences. After the chemical entity is bound to a target, the
`genetic tag can be amplified by replication and utilized for
`enrichment of the bound molecules by serial hybridization to a
`subset of the library. The nature of the chemical structure
`bound to the receptor is decoded by sequencing the nucleotide
`tag.
`
`There is an increasing need to find new molecules that can
`efiectively modulate a wide range of biological processes, for
`applications in medicine and agriculture. A standard way to
`search for novel chemicals is to screen collections of natural
`materials, such as fermentation broths, plant extracts, or
`libraries of synthesized molecules. Assays can range in
`complexity from simple binding reactions to elaborate phys-
`iological preparations. The screens often only provide leads,
`which then require further improvement either by empirical
`methods or by chemical design. The process is time-
`consuming and costly but is unlikely to be replaced totally by
`rational methods even when they are based on detailed
`knowledge of the three-dimensional structure of the target
`molecules. Thus, what we might call “irrational drug de-
`sign”—the process of selecting the correct molecules from
`large ensembles or repertoires—requires continual improve-
`ment both in the generation of repertoires and in the methods
`of selection.
`Recently there have been several developments in using
`peptides or nucleotides to provide libraries of compounds for
`discovery of leads. The methods were originally developed to
`speed up the determination of epitopes recognized by mono-
`clonal antibodies. For example, the standard serial process of
`stepwise search of synthetic peptides has been replaced by a
`variety of highly sophisticated methods in which large arrays
`of peptides are synthesized in parallel and screened with
`acceptor molecules labeled with fluorescent or other reporter
`groups (1, 2). The sequence of any elfective peptide can be
`decoded from its address in the array. In another approach,
`combinatorial libraries of peptides are synthesized on resin
`beads such that each resin bead contains about 20 pmol of the
`same peptide (3). The beads are exposed to labeled acceptor
`molecules. Those with bound acceptor are identified by
`visual inspection and physically removed, and the peptide is
`sequenced directly. In principle, this method could be used
`with other chemical entities, provided one has a sensitive
`method for sequence determination.
`A difierent method of solving the problem of identification
`in a combinatorial peptide library is used by Houghten er al.
`(4). For hexapeptides of the 20 natural amino acids, separate
`libraries are synthesized, each with the first two amino acids
`
`The publication costs of this article were defrayed in part by page charge
`payment. This article must therefore be hereby marked “adverti.rement"
`in accordance with 18 U.S.C. §1734 solely to indicate this fact.
`
`fixed and the remaining four positions occupied by all pos-
`sible combinations. An assay, based on competition for
`binding or some other activity, is then used to find the library
`with an active peptide. On the basis of this result, 20 new
`libraries are synthesized and assayed to determine the effec-
`tive amino acid in the third position. The process is reiterated
`in this fashion until the active hexapeptide is defined. This is
`analogous to the method used in searching a dictionary: the
`peptide is decoded by using a series of sieves, and this makes
`the search logarithmic. A powerful biological method has
`recently been described in which the library of peptides is
`presented on the surface of a bacteriophage such that each
`phage displays a particular peptide and contains within its
`genome the corresponding DNA sequence (5, 6). The library
`is prepared by synthesizing a repertoire of random oligonu-
`cleotides to generate all combinations, followed by their
`insertion into a phage vector. Each of the sequences is cloned
`in one phage and the relevant peptide can be selected by
`finding those that bind to the particular target. The phages
`recovered in this way can be amplified and the selection
`repeated. The sequence of the peptide is decoded by se-
`quencing the DNA. Another “genetic” method has been
`applied by Tuerk and Gold (7) and Ellington and Szostak (8),
`using libraries of synthetic oligonucleotides that themselves
`are selected for binding to an acceptor and then amplified by
`the polymerase chain reaction (PCR). In this case, however,
`the repertoire is limited to nucleotides or nucleotide ana-
`logues that preserve specific Watson-Crick pairing and can
`be copied by a polymerase.
`The main advantages of the genetic methods reside in the
`capacity for cloning and amplification of DNA sequences,
`which allows enrichment by serial selection and provides a
`facile method for decoding the structure of active molecules.
`However, the genetic repertoires are restricted to nucleotides
`and peptides composed of natural amino acids, whereas a
`more extensive chemical repertoire is required to populate
`the entire universe of binding sites. In contrast, chemical
`methods can provide limitless repertoires, but they lack the
`capacity for serial enrichment and there are difficulties in
`discovering the structures of selected active molecules. We
`have now devised a way of combining the virtues of both
`methods through the construction of encoded combinatorial
`chemical
`libraries,
`in which each chemical sequence is
`labeled by an appended “genetic" tag, itself constructed by
`chemical synthesis. In effect, we implement a “retrogenetic”
`way of specifying each chemical structure.
`In outline, we perform two alternating parallel combina-
`torial syntheses so that the genetic tag is chemically linked to
`the chemical structure being synthesized. In each case,
`addition of a monomeric chemical unit to a polymeric struc-
`ture is followed by addition of an oligonucleotide sequence
`which is defined as “encoding” that chemical unit. The
`library is built up by the repetition of this process after
`pooling and division. Active molecules are selected by bind-
`ing to a receptor, and amplified copies of their retrogenetic
`tags are obtained by the PCR. DNA strands with the appro-
`priate polarity can then be used to enrich for a subset of the
`
`5381
`
`

`
`5382
`
`Chemistry: Brenner and Lerner
`
`Proc. Natl. Acad. Sci. USA 89 (1992)
`
`library by hybridization with the matching tags, and the
`process can then be repeated on this subset. Thus serial
`enrichment is achieved by a process of purification, exploit-
`ing linkage to a nucleotide sequence that can be amplified.
`Finally, the structures of the chemical entities are decoded by
`cloning and sequencing the products of PCR.
`
`DesignoftheCodeandtIIeGeneticTag
`
`It is essential to choose a coding representation in such a way
`that no significant part of the sequence can occur by chance
`in some other unrelated combination. Suppose we allocate a
`triplet to each of the chemical units used. Then, because the
`method allows us to "cover all combinations and permutations
`of an alphabet of chemical units, unless we are careful, we
`could find that two difierent combinations have closely
`related sequences which difier only by a frame shift and
`which could not be easily distinguished by hybridization.
`This, potentially the greatest source of errors, can be elim-
`inated by choosing a commaless code (9). The particular
`commaless triplet code that we have chosen allows 20 unique
`representations, as shown in Table 1.
`The sequences for the PCR primers must be chosen so that
`they do not occur within any coding segment and so that they
`can be readily removed from the final PCR product because
`we do not want them to dominate the selective hybridization.
`This can be achieved by building in sites for restriction
`enzymes with the appropriate polarity of cutting. One of the
`restriction enzymes should cut at a site that permits the
`incorporation of a biotinylated nucleotide, such as biotinyl-
`dU'I'P, into the strand complementary to the coding strand.
`All of the above conditions have been met in the following
`design:
`
`S’-AGCTACTTCCCIIGG [coding sequence] GGGCCCTATTCTTAG-3'
`3‘-TCGA'.l'GARGGG'.|!!§§[anticoding smm ATRAGAATC-5’
`Sty I
`Apa I
`
`After cleavage with both restriction enzymes we have
`5'-AGCTACTTCC
`CIIG6 [coding sequence] GGGCC
`CTITTCTTAG-3'
`3'-TCGITGIAGGGIIC
`Qllnticoding strandlg
`CCGGGATAIGIBTC-5’
`
`The internal fragment can be cloned in an appropriate vector
`to sequence the individuals. The temiinal overhang of the Sty
`
`Table 1. Commaless code used in this study
`ttt
`tct
`tat
`TTC
`tcc
`tac
`TTA
`tea
`taa
`TTG
`tcg
`tag
`
`ctt
`CTC
`CTA
`CTG
`
`att
`ATC
`ATA
`ATG
`
`cct
`ccc
`cca
`ccg
`
`act
`ACC
`ACA
`ACG
`
`cat
`cac
`caa
`cag
`
`at
`aac
`aaa
`aag
`
`tgt
`tgc
`tga
`tgg
`
`cgt
`cgc
`cga
`egg
`
`agt
`agc
`aga
`agg
`
`ggt
`gat
`gct
`gtt
`ggc
`gac
`GCC
`GTC
`gga
`GAA
`GCA
`GTA
`ggg
`GAG
`GCG
`GTG
`“Sense triplets” a.re XYZ; nonsense triplets are xyz.
`
`I site can be filled in with dCTP and biotinyl-dUTP (BTP)
`which, because an asymmetric site was chosen, will ap-
`pend the biotinylated nucleotides to only one of the cleavage
`products.
`CIlGG[codin8 SGCIIIGIIGGIGGOCC CTITTCTTAG-3'
`5’-AGCTACTTCCC
`3'-TCGATGAAGGGITC 3BC§[amicoding strandlg
`CCGQGATAAGAATC-5'
`
`The biotinylated fragment can be bound to avidin and, alter
`denaturation, provides the strand suitable for hybridization
`and selection of the appropriate coding strands:
`
`Avidin-BB CQ[anticoding strand]Q
`
`The two PCR primers are the two sequences 5’-AGCTACT-
`TCCCAAGG (Sty I primer) and 5'-CTAAGAATAGGGCCC
`(Apa I primer). Adding a biotin to the 5’ end of the Apa I
`primer would allow the isolation of the whole strand con-
`taining the anticoding sequence.
`We should have at least 15 nucleotides in the coding region
`for effective hybridization. Thus, in a library of degree d 2
`5, that is, composed of five or more successive chemical
`units, we could code each unit by a triplet. That would allow
`an alphabet (A) of up to 20 difl‘erent units, each corresponding
`to one of the triplets defined above. The complexity of the
`combinatorial library is A‘. Libraries with a smaller degree,
`say d = 3, should be coded by sextuplets, which, in the
`simplest case, could be a repeated triplet (this size is chosen
`because any combination of triplets still obeys the commaless
`condition). In the same way, the size of the alphabet can be
`extended by using combinations of triplets to code for the
`chemical units.
`
`AFormalExarnple
`
`As an illustration we discuss how a library of degree d = 3 is
`made with an alphabet of two amino acids, glycine and
`methionine. In this case, we use sextuplets to give us a
`reasonable length of coding sequence. To make the se-
`quences as difierent as possible we code each amino acid by
`a combination of two different triplets as follows:
`
`Gly = CACATG, Met = ACGGTA.
`
`Step I. We begin with some appropriate linker, LINK,
`attached to some solid-state surface and synthesize the first
`PCR oligonucleotide sequence on one end,
`in the usual
`3’-to-5' direction, to give
`GGGCCCTITTCTTIG-LINK
`
`Step 2. This product is divided into two aliquots for parallel
`synthesis. In each synthesis, one amino acid is added to
`LINK and the oligonucleotide sequence is extended by the
`corresponding code to give the following products:
`
`CACATGGGGCCCTITTCTTIG-LI NK—G1y
`ACGGTllGGGCCC'.l!l'.l!'.l!C'.l!!l!AG—LINK—Het
`
`Step 3. The elongated products are pooled and again split
`into two parts for parallel synthesis, yielding
`
`CllCATGCACATGGGGCCCIITTCTTlG—LIllK—Gly—G1y
`CACATGllCGGTllGGGCCC!l!l!l!C!|!'l!lG—LINK-net-G1y
`ACGGTACACATGGGGCCC'£lTTCT'l!lG—LIIIK-G1y-—llet
`AcccrnAcccneeccccnncr-no-L1m<-net-net
`
`Steps 4 and 5. Once more the products are pooled and
`divided into two aliquots for parallel synthesis. This results
`in an ensemble of eight tripeptide sequences, each encoded
`by a unique sequence of 18 nucleotides. The second PCR
`oligonucleotide is added to the ensemble of products to give
`
`

`
`Chemistry: Brenner and Lerner
`
`Proc. Natl. Acad. Sci. USA 89 (1992)
`
`5383
`
`IGCTACTTCCCLLGGCACATGCACATGCACATGGGGCCCTITICTTAG-LINK-Gly—Gly—G1y
`
`AGCTACTTCCCAAGGCACATGCACATGACGGTAGGGCCCTATTCTTAG-LINK-Met-Gly—G1y
`
`IGCTACTTCCCAAGGCACATGACGGTACACATGGGGCCCTITTCTTlG—LINK—G1y-Met—Gly
`
`IGCTACTTCCCAAGGCACATGACGGTAACGGTAGGGCCCTATTCTTAG—LINK-Met-Met-G1y
`
`IGCTICTTCCCAAGGACGGTACACATGCACATGGGGCCCTATTCTTAG-LINK—G1y—Gly-Met
`
`IGCTACTTCCCAAGGACGGTACACATGACGGTAGGGCCCTATTCTTAG-LINK-Met-Gly—Het
`
`AGCTICTTCCCAAGGACGGTAACGGTACACATGGGGCCCTATTCTTAG—LINK—Gly—Met-Met
`AGCTACTTCCCALGGACGGTAACGGTAACGGTAGGGCCCTATTCTTAG-LINK—Met—Met—Het
`
`Implementation
`
`Although natural amino acids are used in the example dis-
`cussed above, the system is not limited to these, nor, for that
`matter,
`to peptides. The chemistry required for making
`encoded libraries is constrained only by the compatibility of
`the two alternating syntheses. Partly this involves the choice
`of the protecting groups, and the methods used to deprotect
`one chain while the other remains blocked. And, of course,
`each product needs to survive through the synthesis of the
`other. One can imagine many different ways of joining the
`chemical entities together. and one could even use mixed
`syntheses, provided that the rules of mutual compatibility are
`obeyed.
`We have recently, in principle, solved the synthetic pro-
`cedures for peptides (K. Janda, S. Ramcharitar, S.B., and
`R.A.L., unpublished results). Even within this field there is
`a choice of alphabets that extends well beyond the 20 natural
`a-amino acids. The only requirement is that we be able to
`make an amide bond. Thus, the amino and carboxylic groups
`can be located on a wide variety of compounds so that we can
`make libraries with many different backbone structures. We
`can also combine different backbones, if we define alphabets
`where, for example, both the number of carbon atoms and
`their configurations in the backbone are varied. New amino
`acids can be easily invented with unusual heterocyclic rings,
`such as thiazole-alanine or purine-alanine. These rings are
`components of natural effector molecules and often provide
`core chemical functions for important drugs. Libraries made
`with such alphabets will allow us to explore the combinatorial
`association of known effector chemical functions.
`It is also useful to consider how large the combinatorial
`library should be. The PCR provides a very sensitive detec-
`tion method, allowing even a few molecules to be seen.
`However, we need to have some reasonable concentration of
`each of the species present to cross the binding threshold of
`the acceptor molecule being assayed. If, for example, we set
`this as 1 p.M and want 1 ml of the library, then we need to
`make at least 1 nmol of each of the species. Libraries with
`complexities of up to 10‘, giving us a total amount of 10 pmol
`of product, would seem reasonable. Because of this recip-
`rocal relationship, more complex libraries could be made if
`the binding threshold is lowered.
`
`Discussion
`
`Traditional chemical synthesis proceeds by careful design,
`sequentially linking atoms or groups of atoms to a growing
`core structure. The process has the advantage that the
`product of each step can be analyzed, thereby allowing
`continuous evaluation ofthe effectiveness ofa given strategy.
`Indeed, the analyzed results of these individual steps ulti-
`mately become the corpus of synthetic organic chemistry. A
`major technical revolution occurred with the advent of solid-
`state methods for the synthesis of polymeric molecules (10).
`Here, since a limited number of suitably protected oligomeric
`
`units are added via a common covalent bond, the results of
`the individual transformations can be predicted, and, to first
`approximation,
`it is necessary to analyze only the final
`product. In addition, the relationship of the monomeric units
`to each other and the extent of conformational space that is
`occupied can be estimated. Our method permits the study of
`the efficacy of combinatorial associations of diverse chemical
`units without the necessity of either synthesizing them one at
`a time or knowing their interactions in advance. It also allows
`easy identification of the most effective molecules through a
`common method of nucleic acid sequencing. Once the chem-
`ical polymers are decoded, more precise questions about
`critical
`interactions and conformations can be asked by
`reversion to classical chemical methods. Further, we expect
`that many receptors will interact with sets of related but not
`identical chemical entities such that major clues as to critical
`interactions can be deduced from the shared features of the
`sets.
`
`Our method also provides a method of amplification, again
`by exploiting a common procedure of nucleic acid hybrid-
`ization. In any screening procedure where large libraries of
`compounds or effector molecules are being studied,
`the
`absolute number of different nonspecific interactions may be
`large, but the specific ligand or efiector is represented many
`more times than any individual background molecule. In such
`a situation the signal-to-noise ratio rapidly increases after
`repeated cycles of amplification and selection, and the spe-
`cific molecule becomes highly enriched after only a few
`iterations. For both identification and selection, our method
`exploits the power of genetic systems. By coupling genetics
`and the versatility of organic chemical synthesis we have
`extended the range of analysis to chemicals that are not
`themselves part of biological systems.
`
`We thank Kim Janda, Bernie Gilula, and Jerry Joyce for helpful
`comments on the manuscript.
`
`4.
`
`1. Fodor, S. P. A., Read, J. L., Pirrung, M. C.. Stryer, L., Tsar’,
`Lu, A. & Solas, D. (1991) Science 251, 767-773.
`2. Geysen, H. M., Meloen, R. H. & Barteling, S. J. (1984) Proc.
`Natl. Acad. Sci. USA 81, 3998-4002.
`3. Lam, K. S., Salmon, S. E., Hersh, E. M., Hruby, V. J., Kaz-
`mierski, W. M. & Knapp, R. J. (1991) Nature (London) 354,
`82-84.
`I-loughten, R. A., Pinilla, C., Blondelle, S. E., Appel, J. R.,
`Dooley, C. T. & Cuervo, J. H. (1991) Nature (London) 354,
`84-86.
`Scott, J. K. & Smith, G. P. (1990) Science 249, 386-390.
`Cwirla, S. E., Peters, E. A., Barrett, -R. W. & Dower, W. J.
`(1990) Proc. Natl. Acad. Sci. USA 87, 6378-6382.
`Tuerk, C. & Gold, L. (1990) Science 24, 505-510.
`Ellington, A. D. & Szostak, J. W. (1990) Nature (London) 346,
`818-822.
`9. Crick, F. H. C., Griffith, J. S. & Orgel, L. E. (1957) Proc.
`Natl. Acad. Sci. USA 43, 416-421.
`10. Merrifield, B. (1984) Les Prix Nobel (Almqvist & Wiksell
`lntemational, Stockholm), pp. 127-153.
`
`
`
`9°.".°‘§"