Proc. Natl. Acad. Sci. USA
`Vol. 90, pp. 10700-10704, November 1993
`Chemistry
`
`Generation and screening of an oligonucleotide-encoded synthetic
`peptide library
`(encoded synthetic libraries/combinatorial chemistry/peptide diversity)
`
`MICHAEL C. NEEDELS, DAVID G. JONES, EMILY H. TATE, GREGORY L. HEINKEL, LYNN M. KOCHERSPERGER,
`WILLIAM J. Dowan, RONALD W. BARRETT, AND MARK A. GALLop*
`Alfymax Research Institute, 4001 Miranda Avenue, Palo Alto, CA 94304
`
`Communicated by Peter G. Schultz, August 2, 1993
`
`We have prepared a library of =10‘ difierent
`ABSTRACT
`peptide sequences on small, spherical (10-um diameter) beads
`by the combinatorial chemical coupling of both L- and D-amino
`acid building blocks. To each bead is covalently attached many
`copies of a single peptide sequence and, additionally, copies of
`a unique single-stranded oligonucleotide that codes for that
`peptide sequence. The oligonucleotide tags are synthesized
`through a parallel combinatorial procedure that eflectively
`records the process by which the encoded peptide sequence is
`assembled. The collection of beads was screened for binding to
`a fluoreseently labeled anti-peptide antibody using a fluores-
`cence-activated cell sorting instrument. Those beads to which
`the antibody bound tightly were isolated by fluorescence-
`activated sorting, and the oligonucleotide identifiers attached
`to individual sorted beads were amplified by the PCR. Se-
`quences of the amplified DNAs were determined to reveal the
`identity of peptide sequences that bound to the antibody with
`high affinity. By combining the capacity for information stor-
`age in an oligonucleotide code with the tremendous level of
`amplification possible through the PCR, we have devised a
`means for specifying the identity of each member of a vast
`library of molecules synthesized from both natural and unnat-
`ural chemical building blocks. In addition, we have shown that
`the use of flow cytometry instrumentation permits facile iso-
`lation of individual beads that bear high-afi'mity ligands for
`biological receptors.
`
`Ligands for macromolecular receptors can be identified by
`screening diverse collections of peptides produced through
`either molecular biological or synthetic chemical techniques.
`Recombinant peptide libraries have been generated by in-
`serting degenerate oligonucleotides into genes encoding cap-
`sid proteins of filamentous bacteriophage (1-3) or the DNA-
`binding protein Lac I (4). These random libraries may contain
`>109 different peptides, each fused to a larger protein se-
`quence that
`is physically linked to the genetic material
`encoding it. Chemical approaches to generating peptide li-
`braries are not limited to combinatorial syntheses using just
`the 20 genetically coded amino acids. By expanding the
`building block set to include unnatural amino acids,
`the
`accessible sequence diversity is dramatically increased. In
`several of the strategies described for creating synthetic
`peptide libraries, the dilferent peptides are tethered to a solid
`support in a spatially segregated manner (5, 6). Large librar-
`ies of soluble peptides have been prepared as peptide pools
`using the “tea-bag" method of multiple synthesis. Active
`peptides within these degenerate mixtures are identified
`through an iterative process of screening and sub-library
`resynthesis (7, 8).
`
`Using the split-synthesis protocol pioneered by Furka et al.
`(9), Lam et al. (10) have prepared libraries containing ~10‘
`peptides attached to 100- to 200-um-diameter resin beads.
`The bead library is screened by incubation with a labeled
`receptor: beads binding to the receptor are identified by
`visual inspection and are selected with the aid of a micro-
`manipulator. Each bead contains 50-200 pmol of a single
`peptide sequence that may be determined directly either by
`Edman degradation or mass spectrometry analysis. In prin-
`ciple, one could create libraries of greater diversity using this
`approach by reducing the bead dimensions. The sensitivity of
`peptide-sequencing techniques is limited to =1 pmol, how-
`ever, clearly limiting the scope of direct peptide-sequencing
`analysis. Moreover, neither analytical method provides for
`straightforward and unambiguous sequence analysis when
`the library building block set is expanded to include D- or
`other nonnatural amino acids.
`We recognized that the products of a combinatorial peptide
`synthesis on resin beads could be explicitly specified if it
`were possible to attach an identifier tag to the beads coinci-
`dent with each amino acid-coupling step in the synthesis.
`Each tag would then convey which amino acid monomer was
`coupled in a particular step of the synthesis, and the overall
`sequence of a peptide on any bead could be deduced by
`reading the set of tags on that bead. We now describe the use
`of single-stranded oligonucleotides to encode a combinatorial
`synthesis on 10-um-diameter polystyrene beads. Peptides
`and nucleotides are assembled in parallel, alternating syn-
`theses so that each head bears many copies of both a single
`peptide sequence and a unique oligonucleotide identifier tag.
`We have generated an encoded synthetic library of some 8.2
`x 105 heptapeptides and screened it for binding to an anti-
`dynorphin B monoclonal antibody (mAb) D32.39 (4), using a
`fluorescence-activated cell sorting (FACS) instrument to
`select individual beads that strongly bind the antibody.
`
`MATERIALS AND METHODS
`
`Reagents and General Methods. The monodisperse 10-p.m-
`diameter bead material used in this work was a custom-
`synthesized macroporous styrene/divinylbenzene copoly-
`mer functionalized with a 1,12-diaminododecane linker. All
`protected amino acids were obtained from Bachem.
`Parallel Synthesis of a 69-Base Oligonucleotide and the
`Peptide I-I-Arg-Gln-Phe-Lys-Val-Val-Thr-NH2 (RQFKVVT).
`The C-terminal seven amino acid fragment of the opioid
`peptide dynorphin B was synthesized in parallel with a
`69-mer oligodeoxyribonucleotide (ST08) on 10-um-diameter
`beads. The sequence of ST08 was 5’-AI‘_C CAA I‘_C_'I QLC
`CAC ATC TCTATA CTA TCA TCA CC [TA TC CT AT TT
`
`The publication costs of this a.rticle were defrayed in part by page charge
`payment. This article must therefore be hereby marked “advertisement"
`in accordance with 18 U.S.C. §l734 solely to indicate this fact.
`
`Abbreviations: Fmoc, 9-tluorenylmethoxycarbonyl; DMT, di-
`methoxytrityl; mAb, monoclonal antibody; FACS, fluorescence-
`activated cell sorting.
`‘To whom reprint requests should be addressed.
`
`10700
`
`

`
`Chemistry: Needels et al.
`
`Proc. Natl. Acad. Sci. USA 90 (1993)
`
`10701
`
`TT AC] CTC ACI CAC ILTQ CAI TEE AC-3'. Underlined
`portions of this sequence correspond to PCR-priming sites,
`while the region in italics is homologous to the primer used
`for sequencing this template. The 14-base sequence enclosed
`in brackets represents the coding region of the template.
`The beads were first treated with a mixture of succinimidyl
`4-0-DMT-oxybutyrate (where DMT is dimethoxytrityl; Mo-
`lecular Probes) and the 1-oxybenzotriazole ester of either
`N-Fmoc-2,4-dimethoxy-4’-(carboxymethyloxy)benzhydry-
`lamine (i.e., the acid-cleavable Knorr carboxamide linker,
`where Fmoc is 9-fluorenylmethoxycarbonyl) or N-Fmoc-
`Thr(tert-butyl)-OH (for noncleavable experiments). The ratio
`of Fmoc-protected amino groups to DMT-protected hydroxyl
`residues on the beads was determined spectrophotometri-
`cally to be =~—20:1. The beads were subjected to 20 cycles of
`oligonucleotide synthesis on an automated synthesizer using
`3’-0-methyl-N,N-diisopropyl phosphoramidites of the fol-
`lowing nucleosides: N‘-benzoyl-5’-0-DMT-(7-deaza)-2’-
`deoxyadenosine (Berry and Associates, Ann Arbor, MI),
`N‘-benzoyl-5'-0-DMT-2'-deoxycytidine, and 5’-0-DMT-
`thymidine (Glen Research, Sterling, VA). The beads were
`then removed from the instrument and treated for 5 min with
`10% (vol/vol) piperidine in dimethylformamide to remove the
`Fmoc protecting group. After coupling the first amino acid
`residue [N-Fmoc-Thr(tert-butyl)-OH],
`the beads were
`treated with a tetrahydrofuran solution of acetic anhydride
`and l-methylirnidazole to cap any unreacted amines. All
`peptide coupling reactions were run for 20 min and contained
`0.11 M Fmoc-amino acid, 0.1 M 0-(benzotriazol-1-yl)-
`1,1,3,3-tetramethyluronium hexafluorophosphate, 0.1 M
`1-hydroxybenzotriazole, and 0.3 M diisopropylethylamine in
`dimethylformamide/CH2Clz, 1:1. The beads were then sub-
`jected to two cycles of nucleotide addition on the synthesizer
`(detritylation with trichloroacetic acid; tetrazole-catalyzed
`phosphitylation; capping with acetic anhydride; oxidation
`with iodine in acetonitrile/water). Sequential steps of amino
`acid coupling and dinucleotide addition were repeated until
`synthesis of the peptide sequence RQFKVVT and construc-
`tion of the oligonucleotide coding region had been completed.
`After an additional 35 cycles of oligonucleotide synthesis, the
`beads were treated sequentially with piperidine/dimethylfor-
`mamide, 1:9 for 8 min; thiophenol/triethylamine/dioxane,
`1:2:2, for 4 hr; ethylenediarnine/ethanol, 1:1, for 5 hr at 55°C;
`and trifluoroacetic acid/water, 20:1, for 1 hr to fully depro-
`tect both the peptide and oligonucleotide chains. In experi-
`ments using the acid-cleavable linker, the supernatant from
`the trifluoroacetic acid deprotection reaction was concen-
`trated in vacuo, and the isolated crude peptide was then
`analyzed by HPLC using a Rainin reverse-phase C13 column.
`Construction of an Encoded Library. The parallel synthesis
`chemistry outlined above was used in the construction of the
`library. The sites of peptide synthesis were differentiated
`from DNA-synthesis sites in this experiment by coupling to
`all beads a mixture of N-Fmoc-Thr(tert-butyl)-oxybenzotri-
`azole and succinimidyl 4-0-DMT-oxybutyrate, as has been
`described. Sequences of oligonucleotide tags in the library
`deviated from ST08 only within the coding region. The
`3'-conserved region of the oligonucleotide ST08 was first
`synthesized on a total bead mass of 35 mg (=1.75 X 103
`beads). The Fmoc protecting group was removed, and the
`bead mass was divided into seven equal parts. To each
`aliquot was coupled one of seven difierent a-N-Fmoc-
`protected amino acids (side-chain protecting groups are
`shown in parenthesis): Arg(N°-2,2,5,7,8-pentamethylchro-
`man-6-sulfonyl), Gln(tn'tyl), Phe, Lys(tert-butoxycarbonyl),
`Val, D-Val, and Thr(tert-butyl). Each part was then subjected
`to two rounds of automated oligonucleotide synthesis. The
`respective sequences of the appended dinucleotides that
`specified uniquely each difi‘erent amino acid residue were
`TA, TC, CT, AT, TT, CA, and AC. The beads were then
`
`pooled and mixed thoroughly; the entire bead mass was then
`subjected to Fmoc deprotection. This cycle of bead parti-
`tioning, peptide coupling, oligonucleotide-dimer synthesis,
`bead recombination, and Fmoc removal was repeated for a
`total of seven times. The final Fmoc protecting group was not
`removed. Rather, the pooled bead mass was subjected to 35
`cycles of oligonucleotide synthesis. The library was then fully
`deprotected as described above.
`FACS Analysis of Antibody Binding to Beads. A portion of
`the library (typically 0.5-2 mg of beads) was suspended in
`blocking buffer [phosphate—buffered saline (PBS)/ 1% bovine
`serum albumin/0.05% Tween-20] and incubated at room
`temperature for 1 hr. The beads were pelleted by centrifu-
`gation and resuspended in a solution of mAb D32.39 (10
`;:.g/ml in blocking buffer) (4). The suspension was incubated
`on ice for 30 min, pelleted by centrifugation, and washed with
`blocking buffer. Beads were then suspended in a solution of
`phycoerythrin-conjugated goat anti-mouse antibody (Molec-
`ular Probes) for 20 min on ice. The beads were washed in
`blocking bufier and diluted in PBS for delivery into the FACS
`instrument (Becton Dickinson FACStar"'“‘). Beads that had
`bound mAb D32.39 were identified by their acquired fluo-
`rescence. Individual beads from both the most brightly
`stained 0.17% of the library and from the region having the
`lowest fluorescence (~98%) were sorted into PCR microcen-
`trifuge vials.
`_
`PCR of Bead-Bound Template and Sequencing of PCR
`Product. PCRs consisting of 45 amplification cycles were
`done with Taq polymerase (Perkin—Elmer) according to the
`manufacturer’s instructions. The reactions contained dUTP
`and uracil DNA glycosidase (GIBCO/BRL) to prevent car-
`ryover contamination with soluble product from previous
`amplifications (11). Biotinylated PCR product from individ-
`ual reactions was isolated with streptavidin-coated magnetic
`beads (Dynal, Great Neck, NY). After alkaline elution of the
`nonbiotinylated strand and washing, each bead sample was
`treated with sequencing mixture. Dideoxynucleotide se-
`quencing was done by using the primer 5 '-ATC TCT ATA
`CTA TCA-3’ (SP15) and B31 polymerase (Bio-Rad), accord-
`ing to the manufacturer’s instructions, except that a 1:100
`ratio of deoxy- to dideoxynucleotide triphosphates (Pharma-
`cia) was used.
`Determination of Peptide-Binding Affmities. Binding afl'in-
`ities of various peptides for mAb D32.39 were measured in a
`competition binding experiment. A tracer peptide (LR-
`RASLGGGRRQFKVVT; 50 pM) containing the known epi-
`tope for mAb D32.39 fused to a consensus substrate sequence
`for cAMP-dependent protein kinase was radiolabeled to high
`specific activity with [7-33P]ATP (12) and mixed with various
`concentrations of the peptide of interest (10 ;.cM to 1 pM). The
`peptide mixtures were added to polystyrene wells coated
`with mAb D32.39 (0.1 pg/ml). Samples were incubated 2 hr
`at 4°C, the wells were washed with PBS, and the radioactivity
`associated with each well was counted and used to generate
`a competitive binding curve. Under the conditions of the
`assay the IC5o should be close to the dissociation constant
`(Kd) for the peptide.
`
`RESULTS
`
`Establishing a practical bead-based oligonucleotide-encoded
`peptide library methodology demands that several key tech-
`nical criteria be met. These criteria include (i) the develop-
`ment of mutually compatible chemistries for parallel assem-
`bly of peptides and oligonucleotides, (ii) the selection of bead
`material with appropriate physical characteristics, (iii) the
`facile isolation of small beads bearing ligands that bind a
`target receptor, and (iv) successful PCR amplification and
`sequencing of template DNA from single beads. We have
`found that the properties of 10-um-diameter beads fashioned
`
`

`
`10702
`
`Chemistry: Needels et al.
`
`Proc. Natl. Acad. Sci. USA 90 (1993)
`
`olic ethylenediamine (debenzoylation of protected cytidine
`and 7-deaza-adenine residues). These mild, anhydrous ami-
`nolysis conditions did not adversely afiect protected peptide
`sequences (19), which were deblocked using trifluoroacetic
`acid under standard conditions.
`The carboxyl~terminal region of opioid peptide dynorphin
`B (YGGFLRRQFKVVT) has been previously shown to
`represent the epitope of anti-dynorphin B mAb D32.39 (4):
`the soluble heptapeptide RQFKVVT binds mAb D32.39 with
`high affinity (K; = 0.5 nM). To test the efficacy of our
`chemical methods, a parallel synthesis of this peptide and a
`69-base oligodeoxyribonucleotide was performed on orthog-
`onally difierentiated beads bearing an acid-cleavable Fmoc-
`protected carboxarnide (Knorr) linker. The beads were ex-
`posed to full oligonucleotide and then peptide-deprotection
`conditions, and the trifluoroacetic acid supernatant contain-
`ing the cleaved peptide was analyzed by reverse-phase
`HPLC. Fig. 1a shows that the crude peptide from the parallel
`synthesis consists of a single major component (coeluting
`with authentic RQFKVVT; data not shown) and that this
`crude product is not significantly difi'erent from that gener-
`ated in a control peptide synthesis in which no oligonucleo-
`tide chemistry occurred (Fig. lb). Fig. 2 demonstrates that
`the integrity of the DNA template containing c7 dA was also
`maintained through the course of the parallel synthesis chem-
`istry.
`.
`Construction of a Large Encoded Combinatorial Library.
`An encoded library designed to contain 823,543 (77) different
`heptapeptides attached to 10-p.m beads was constructed by a
`combinatorial synthesis using the seven amino acids arginine,
`glutamine, phenylalanine, lysine, valine, D-valine, and thre-
`onine. a-N-Fmoc-protected threonine and 0—DMT-protected
`y-oxybutyrate residues were first coupled to all the beads to
`provide the orthogonally differentiated amino and hydroxyl
`groups for this synthesis. Starting with a total bead mass of
`35 mg (1.75 X 103 beads) ensured that each peptide sequence
`appeared on ~200 different beads in the library. Peptide
`microsequencing analysis of an aliquot of the library con-
`firmed that the seven amino acids were stochastically dis-
`tributed among every position of the degenerate heptapeptide
`mixture (note that L-valine and D-valine are indistinguished in
`the Edman degradation procedure).
`The binding of mAb D32.39 to control beads and to the
`bead library was analyzed by flow cytometry. Fig. 3a shows
`that beads carrying the positive control sequence RQFKVVT
`4
`
`8.
`
`30
`
`b
`
`30
`
`’
`
`min.
`
`0
`
`O&<
`
`0
`
`FIG. 1. Reversed-phase HPLC chromatograms of crude peptide
`RQFKVVT. Asterisks mark peak corresponding to authentic mate-
`rial. (a) Peptide synthesized in parallel with 69-mer oligonucleotide
`ST08. (b) Peptide from control synthesis (no nucleotide chemistry).
`
`from a macroporous styrene/divinylbenzene copolymer and
`derivatized with a dodecylamine linker are generally satis-
`factory for this work. The amino group loading of these beads
`was estimated to be ~100 p.mol/g by exhaustive acylation
`with Fmoc-glycine, followed by piperidine cleavage of the
`Fmoc group and spectrophotometric quantitation of the
`released piperidine-dibenzofulvene adduct
`(£302 = 7800
`liter-mol'1‘cm‘1). With 5 x 109 beads per g, this corresponds
`to a maximum peptide loading of ~20 fmol per bead. Acy-
`lation of the beads with a mixture of an appropriately
`protected amino acid and an w-hydroxy acid provided or-
`thogonally differentiated amino and hydroxyl groups from
`which the peptide and nucleotide chains, respectively, could
`be extended. The average stoichiometry of peptide to oligo-
`nucleotide per bead is controlled by varying the ratio of
`amino and hydroxy acids coupled to the initial bead mass.
`Test peptide syntheses (5-mers to 12-mers) on these beads
`equipped with a trifluoroacetic acid-cleavable Knorr linker
`(13) using standard Fmoc chemistry were found to proceed
`with high fidelity that was indistinguishable from syntheses
`performed on conventional peptide synthesis resin, as deter-
`mined by HPLC analysis of the crude cleaved peptide
`carboxarnides (data not shown).
`Parallel Synthesis of Peptides and Oligonucleotides. Parallel
`synthesis strategies require (i) the use of a set of protecting
`groups on the amino acids and nucleotide building blocks that
`are mutually orthogonal and (ii) that each of the polymer
`chains be stable to the reagents used in the synthesis and
`deprotection of the second chain. Although, in principle, a
`variety of protection/deprotection schemes could be envis-
`aged, we preferred to use Fmoc/tert-butyl protection on the
`peptide building blocks because of the extensive commercial
`availability of natural and unnatural amino acids protected in
`this manner. However, the tert-butyl-based peptide side-
`chain protecting groups require treatment with strong acid
`(typically trifluoroacetic acid) for removal, conditions that
`lead to rapid depurination of oligonucleotides containing
`either 2’-deoxyadenosine (dA) or 2’-deoxyguanosine (dG)
`(14). This problem has been circumvented by using 7-deaza-
`2’-deoxyadenosine (c7dA) for dA in the template oligonucle-
`otide tag. The glycosidic bonds of deazapurine nucleosides
`are resistant to acid-catalyzed hydrolysis (15), and oligonu-
`cleotides incorporating these monomers are faithfully copied
`by thermostable polymerases used in the PCR (16, 17).
`Although not used in this work, acid-resistant guanosine
`analogs could also be incorporated into the template DNA.
`5’-0-dimethoxytrityl 2'-deoxynucleoside 3’-(0-methyl-
`N,N-diisopropyl)phosphoramidites were used in all parallel
`syntheses. The reagent (I2/collidine/H20/acetonitrile) used
`to convert the nucleotide phosphite intermediates to phos-
`photriesters in the DNA-synthesis protocol was not found to
`adversely affect either the readily oxidized residues tryp-
`tophan and methionine or any of the other protected amino
`acids used in this work (data not shown). Complete removal
`of the 5'-0-DMT group from the growing oligonucleotide
`chain was achieved in ~40 sec using 1% trichloroacetic acid
`in dichloromethane, whereas all of the acid-labile side-chain
`protecting groups used conventionally in Fmoc/tert-butyl
`chemistry were inert to treatment with 1% trichloroacetic
`acid for 1 hr. Quantitative deprotection of the as-amino
`residues required 5- to 10-min treatment with piperidine/
`dimethylformamide (10% vol/vol) and also resulted in partial
`demethylation of the protected polynucleotide phosphotri-
`esters (tr/2 = 45 min). Control experiments indicated that any
`aberrant phosphitylation of the resulting phosphodiester spe-
`cies during subsequent nucleotide chain elongation was re-
`versed by the final oligonucleotide deprotection steps, as
`noted by other workers (18). At the completion of the parallel
`synthesis, the DNA was fully deprotected by treatment with
`thiophenolate (phosphate 0-demethylation) and then ethan-
`
`

`
`Chemistry: Ncedels et al.
`
`Proc. Natl. Acad. Sci. USA 90 (1993)
`
`10703
`
`-1-l-<l—(—i>O -70 bp
`/—l>-iOO'4>
`
`,)
`
`FIG. 2. Amplification and sequence analysis of oligonucleotide
`ST08 synthesized in parallel with peptide RQFKVVT. (a) Ethidium
`bromide-stained agarose gel electrophoresis of products from PCR
`amplifications of ST08 template attached to single sorted beads.
`Lanes: 1 and 6, DNA markers; 2, c7dA-containing template from
`parallel synthesis; 3, dA-containing template after 1-hr treatment
`with 95% trifluoroacetic acid/5% H20; 4, untreated dA-containing
`template; 5, zero bead control. (b) Sequencing gel of the PCR
`amplification product from an individual bead; DNA sequence of the
`template coding region is shown at right.
`
`and a 69-mer oligonucleotide tag are strongly stained by the
`antibody, whereas blank beads are unstained. By contrast,
`only a small fraction of the encoded library bound mAb
`D32.39 (see Fig. 3b). Analysis of 105 events indicated that
`~2% of the library stained above background levels. Signif-
`icantly, this binding to mAb D32.39 was specific for the
`combining site, as it could be completely blocked by prelu-
`cubating the mAb with soluble RQFKVVT peptide (Fig. 3c).
`Individual beads from the library having fluorescence inten-
`sities comparable with the positive control beads were sorted
`into microcentrifuge tubes for tag amplification by PCR
`(beads with fluorescence in the top 0.17% of the population
`were collected). Nucleotide sequences were obtained from 12
`sorted beads, and the deduced peptide sequences are given in
`Table 1. Representative peptide sequences obtained from
`single beads having fluorescence that was not significantly
`above background are also tabulated for comparison.
`These data are consistent with an earlier study showing
`that the preferred recognition sequence of mAb D32.39 is
`localized to the six-amino acid fragment RQFKVV of dynor-
`phin B (4). Interestingly, D-valine appears best tolerated at
`positions outside the consensus motif. The range of affinities
`of peptides that were selected (Kd z 0.3-1400 nM) was not
`unexpected, given the design of the binding assay—i.e.,
`bivalent primary antibody with labeled second antibody
`detection. We anticipate that manipulation of the binding
`valency (for example, directly labeled monovalent receptor)
`
`
`
`Particle Scatter
`
`Particle Scatter
`
`
`
`LogFluorescence
`
`Particle Scatter '
`
`FIG. 3. Flow cytometric analysis of binding of mAb D32.39 to
`10-um beads bearing peptide and oligonucleotide. Approximately 105
`events are recorded in each experiment. Fluorescence intensity is
`shown on vertical axis. (a) A 1:1 mixture of underivatized (lower
`population) and RQFKVVT (upper population) beads as negative
`and positive controls. (b) Binding of mAb to library. (c) Specific
`binding to library blocked by preincubation of mAb with 10 pM
`RQFKVVT.
`
`and the stringency of wash conditions will improve the
`capacity to isolate only the highest-affinity ligands.
`
`DISCUSSION
`
`We have developed chemistry to prepare a highly diverse
`oligonucleotide-encoded synthetic peptide library on micro-
`scopic beads by combinatorial synthesis. While this work
`was in progress, the concept of an oligonucleotide-encoded
`chemical synthesis was proposed independently by Brenner
`and Lerner (20). More recently, two other groups have shown
`that an L-amino acid peptide strand may be used to encode
`the combinatorial assembly of molecular structures that are
`not amenable to direct sequence analysis (21, 22). It seems
`likely that constraints on the sensitivity and throughput of the
`Edman procedure will ultimately restrict the scope of this
`peptide-coding approach to analyzing libraries of limited
`diversity.
`Encoding a combinatorial synthetic procedure with oligo-
`nucleotides provides a mechanism for addressing the major
`limitations of ambiguity and sensitivity encountered in the
`direct structural analysis of minute quantities of ligands
`isolated from large libraries. The high capacity of DNA for
`information storage can be exploited to archive the precise
`details of a library’s construction. In the example above, we
`used a “codon” structure of two contiguous nucleotides
`comprising three bases (c7 dA, dC, and T), capable of
`encoding a synthesis incorporating up to 32 = 9 amino acid
`building blocks (only seven were used in this library). Ifc7 dG
`were also included in the coding template, then a combina-
`torial synthesis using 1000 different monomers could be
`accommodated by using a “codon” size of just 5 nt (45 =
`1024).
`A second outstanding advantage inherent in using an
`oligonucleotide-based coding scheme is the ability to achieve
`
`Table 1.
`
`Amino acid sequences of peptides on beads that bind mAb D32.39
`
`High fluorescence intensity
`
`Low fluorescence intensity
`
`Sequence
`Tl-‘RQFKV ( T)
`TTRRFRV ( T)
`TVRQFKT ( T)
`QVRQFKT ( T)
`RQFRTVQ ( T )
`KQFKVTK( T)
`
`Kd, nM
`0.29
`4.3
`8.8
`16
`76
`340
`
`Sequence
`QQFKVVQ ( T)
`KQFKVTQ ( T)
`TQFKVTK ( T )
`TF'RvFRV( T )
`FRRQFRV ( T )
`RQF'KQVQ(T)
`
`RQFKVVT
`
`0.51
`
`(positive control)
`
`Kg, nM
`370 ‘
`410
`560
`1400
`ND
`ND
`
`Sequence
`QTVTVKK ( T)
`QQVQRQT ( T )
`KTQVVQF‘ ( T )
`QVTQVRV ( T )
`FVVTVRV ( T )
`
`Kd, mM
`>1
`>0.4
`ND
`ND
`ND
`
`A library of peptide-beating beads was screened for antibody ligands by using an indirect fluores-
`cence assay and FACS instrumentation. Sequences from the highly fluorescent beads are aligned to
`show consensus with the D32.39 epitope. Aflinities of selected soluble peptides for mAb D32.39 were
`determined by RIA. ND, not determined; ('1'), threonine linker residue.
`
`

`
`10704
`
`Chemistry: Needels et al.
`
`Proc. Natl. Acad. Sci. USA 90 (I993)
`
`tremendous levels of target amplification through the PCR.
`We are therefore able to work with tiny quantities of DNA
`template and, hence, to use solid supports of microscopic
`dimensions in our syntheses. This will facilitate the construc-
`tion and screening of libraries that far exceed the diversity
`accessible through other tethered synthetic library tech-
`niques. Moreover, these libraries will employ manageable
`quantities of bead material that can therefore be assayed for
`receptor binding using practical volumes of biological re-
`agents.
`Standard FACS instrumentation permits bead (cell) fluo-
`rescence analysis rates of =10‘ events per sec or, when
`operated in single bead cloning mode, sort rates that are 5- to
`10-fold slower. In assaying very large libraries (e.g., >>107
`beads) some form of affinity-selective prescreen could be
`used before individual bead isolation with the cell sorter. For
`example, receptor-coated sub-micron-sized superparamag-
`netic particles are frequently used to affinity-purify specific
`cells from large, mixed populations by magnetic activated
`sorting (23). It should be noted that to have a high probability
`of detecting very rare binding events, it is essential that each
`compound be present on many beads in the library. A
`practical upper limit for the size of an encoded library
`constructed from 10-pm particles, assuming a 100-fold re-
`dundancy, is probably 103-109 compounds synthesized on
`=101°—10“ beads. Even larger libraries might be prepared
`using smaller beads, but conventional cytometers are un-
`likely to detect or manipulate particles much less than -1 pm
`in diameter.
`High-throughput screening of collections of chemically
`synthesized molecules and of natural products (such as
`microbial fermentation broths) has traditionally played a
`central role in the search for lead compounds for the devel-
`opment of new pharmacological agents. The remarkable
`surge of interest in combinatorial chemistry and the associ-
`ated technologies for generating and evaluating molecular
`diversity represent significant milestones in the evolution of
`this paradigm of drug discovery (24). To date, peptide chem-
`istry has been the principle vehicle for exploring the utility of
`combinatorial methods in ligand identification. This may be
`ascribed to the availability of a large and structurally diverse
`range of amino acid monomers, a relatively generic, high-
`yielding solid-phase coupling chemistry, and the synergy
`with biological approaches for generating recombinant pep-
`tide libraries. Moreover, the potent and specific biological
`activities of many low-molecular-weight peptides make these
`molecules attractive starting points for therapeutic drug
`discovery. Unfavorable pharmacodynamic properties, such
`as poor oral bioavailability and rapid clearance in vivo, have
`limited the more widespread development of peptidic com-
`pounds as drugs, however. This realization has recently
`inspired workers to extend the concepts of combinatorial
`organic synthesis beyond peptide chemistry to create librar-
`ies of known pharmacophores like benzodiazepines (25), as
`well as unnatural polymeric molecules, such as oligomeric
`N-substituted glycines (“peptoids") (26) and oligocarba-
`mates (27). We anticipate that the coding strategy illustrated
`here could greatly facilitate the screening of large collections
`of compounds that cannot be directly sequenced and are
`produced by multistep synthesis.
`We have shown here that oligonucleotide encryption pro-
`vides a powerful mechanism for recording the structural
`identity of every member of a vast library of tethered peptides
`generated through a combinatorial synthesis. This technique
`should also be broadly applicable to encoding the combina-
`torial assembly of other nonpeptidic structures, providing the
`parallel synthetic schemes remain orthogonal and compati-
`ble. The net outcome of a combinatorial synthesis is unam-
`biguously defined only for a sequence of reactions that each
`
`proceed in very high yield to afford single products. This
`situation is approximated by peptide and DNA synthesis
`chemistries, and the resulting product structures are explic-
`itly specified by the order of the building blocks and/or
`coupling reactions used in the synthesis. However, most
`synthetic organic reactions are more idiosyncratic, giving
`variable yields and frequently multiple products (such as
`regio- and stereoisomeric structures). Using such chemistry
`to synthesize combinatorial libraries on solid supports would
`yield a mixture of products on each bead in the library. In the
`most general case, the encryption of a synthesis may not
`uniquely specify the chemical structure of an associated
`entity. Rather, it may encode the exact synthetic protocol
`(e.g., reagents, reaction conditions", etc.) by which a member
`of the library was constructed. The library would be screened
`to identify “active recipes,” which then could be reproduced
`on a preparative scale and fractionated (if necessary) to
`isolate the bioactive component(s). Encoded library technol-
`ogies have considerable potential to expand the scope of
`combinatorial chemistry and its applications to drug discov-
`cry.
`
`7.
`
`99°
`
`1. Cwirla, S. E., Peters, E. A., Barrett, R. W. & Dower, W. J. (1990)
`Proc. Natl. Acad. Sci. USA 87, 6378-6382.
`2. Scott, J. K. & Smith, G. P. (1990) Science 249, 386-390.
`3. Devlin, J. J., Panganiban, L. C. &._Devlin, P. E. (1990) Science 249,
`404-406.
`4. Cull, M. G., Miller, J. F. & Schatz, P. J. (1992) Proc. Natl. Acad.
`Sci. USA 89, 1865-1869.
`5. Geysen, I-I. M.,Meloen, R. H. &Barteling, S. J. (1984) Proc. Natl.
`Acad. Sci. USA 81, 3998-4002.
`6. Fodor, S. P. A., Read, J. L., Pirrung, M. C., Stryer, L., Lu, A. T.
`& Solas, D. (1991) Science 251, 767-773.
`1-loughten, R. A., Pinilla, C., Blondelle