Peptide Libraries

A peptide library is a large collection of peptides with systematically varied sequences, made so that all of them can be tested against a target at the same time: an antibody, a receptor, an enzyme or a microbe[1]. Libraries range from a few hundred peptides on a plastic plate to millions on resin beads and about a billion displayed on bacterial viruses[1][2].

The idea turns the usual approach around. Instead of designing one peptide and hoping it works, the chemist makes a huge number and lets the target pick the ones it binds. That creates two problems which run through the whole history of the method: how to make so many peptides quickly enough, and how to find out which one in the crowd did the binding. Every type of library is a different answer to these two questions.

How many peptides are possible

With 20 amino acids to choose from at each position, a chain of n residues can have 20ⁿ different sequences. The numbers grow very quickly:

Libraries can cover every possible sequence only for short peptides. A typical phage library of about a billion clones can hold most of the 64 million hexapeptides, but only a tiny fraction of the possible 15-mers[2]. That is less of a limitation than it seems. For many targets, antibodies among them, only three to five residues of a peptide make the contacts that matter, and the rest act as spacers[3].

Many peptides at once: pins, tea bags and spots

In the early 1980s, making even a hundred peptides was so slow and expensive that it was impractical in most laboratories, and the supply of peptides limited many studies[4]. The first answer was to run many syntheses side by side and keep track of where each peptide was.

In 1984 Mario Geysen and colleagues built peptides on the tips of polyethylene pins held in a block with the spacing of a 96-well plate, so that each pin could be dipped into its own well of reagent. They made all 208 overlapping hexapeptides of a 213-residue coat protein of foot-and-mouth disease virus and tested them with antibodies while the peptides were still on the pins. The antibodies raised against the whole virus recognised a stretch at residues 146–152. A second set of 120 hexapeptides, in which each of the six positions was filled in turn with each of the 20 amino acids, showed that two leucines, at positions 148 and 151, were essential for binding[5].

A year later Richard Houghten sealed portions of resin into numbered polypropylene mesh packets, like tea bags. The packets went through the common steps of each cycle together and were sorted into separate vessels only for the coupling step. In less than four weeks he made 248 different 13-residue peptides, 10–20 mg each, all variants of a segment of influenza haemagglutinin. Tested against a monoclonal antibody, they showed that a single aspartic acid, residue 101, was of unique importance[4].

SPOT synthesis, introduced by Ronald Frank in 1990, went further and dropped the separate vessels altogether. A droplet of reagent placed on a sheet of cellulose soaks in and forms a spot that acts as its own tiny reactor, while the steps common to all peptides are done by washing the whole sheet. A dye, bromophenol blue, turns each spot blue while it has free amino groups and yellow once they have reacted. Arrays from the 96 spots of a plate up to microscope-slide size can be tested directly on the membrane, and the method became a popular tool in immunology, especially for mapping the linear epitopes that antibodies bind, usually with peptides of 15–20 residues[6].

In all these formats every peptide has a known address, so a positive signal identifies the sequence directly. The cost is scale: such parallel libraries usually hold hundreds to a few thousand peptides[1].

Split and mix

To reach millions, the bookkeeping had to go. Árpád Furka and colleagues in Budapest, who first reported the idea in 1988, divided the resin into equal portions before each coupling, coupled each portion with a different amino acid and then mixed all portions together again. Repeating this split-couple-mix cycle gives a nearly equimolar mixture of every combination[7][3].

The method is remarkably economical. The number of different peptides is the product of the numbers of amino acids used at each position, while the number of couplings is only their sum. Furka's 180 pentapeptides needed 15 couplings instead of the 720 that separate syntheses would have taken[7].

One bead, one peptide

Furka analysed his products as mixtures. The decisive observation was made in 1991 by Kit Lam and colleagues in Arizona: in split-and-mix synthesis each bead meets only one amino acid in each cycle, and if every coupling is driven to completion, each bead ends up carrying a single sequence[8][3]. Simply coupling a mixture of activated amino acids would not work, because amino acids couple at different rates and every bead would carry a mixture of peptides[8]. A bead 100 µm across can carry up to 10¹³ copies of its one peptide[3].

This turned each bead into a separate, tiny test. Lam's group added the target, an antibody or streptavidin tagged with an enzyme or a fluorescent dye, to millions of beads at once. The few beads that bound it turned coloured and could be seen by eye against a background of colourless ones. They were picked out with fine forceps, stripped of the target, and their peptide was read by automated Edman sequencing: a single bead held 50–200 pmol of peptide, at least ten times more than the sequencer needed. A library of several million beads could be screened in 10–15 Petri dishes in an afternoon, and then washed and reused[8].

Leaving out cysteine, they made pentapeptides from 19 amino acids, up to 2,476,099 sequences. Against a monoclonal antibody raised to β-endorphin, whose epitope is YGGFL, about 2 million beads gave six positives; the best, YGGFQ, bound almost exactly as tightly as the natural epitope. Against streptavidin, 23 of 28 sequenced beads carried the motif His-Pro-Gln[8].

Because beads are distributed at random, a library needs more beads than possible sequences. Computer simulations cited by Lam and colleagues showed that about three times as many beads are needed to include 95% of all sequences (with 99% confidence), for example around 9.6 million beads for the 3.2 million pentapeptides of 20 amino acids. One person can screen about 10 million beads a day, and around 100 million is a practical upper limit for a library[3].

Libraries in solution

Bead libraries test peptides that are still attached to a solid. In the same issue of Nature, Houghten's group published the opposite approach: mixtures of free peptides in solution, which can be used in almost any existing assay[9]. Their first library consisted of hexapeptides of the form Ac-O₁O₂XXXX-NH₂. The first two positions were defined, giving 324 separate mixtures, while each X was an equimolar mixture of 18 amino acids. Each mixture contained 104,976 peptides, and the library as a whole about 34 million[9].

The active sequence was then found by stepwise narrowing. Against a monoclonal antibody, the mixture Ac-DVXXXX-NH₂ inhibited binding best. The group then made 20 new mixtures with the third position defined, picked the best, and repeated the process position by position, cutting the number of sequences 20-fold at each step. The end point, Ac-DVPDYA-NH₂, matched exactly the epitope already known for that antibody, and no information about the antigen or the antibody had been needed to find it. The four rounds took about four weeks[9].

The same 324 mixtures were also screened against bacteria and yeast. Starting from Ac-RRXXXX-NH₂, the process led to the hexapeptide Ac-RRWWCR-NH₂, which inhibited Staphylococcus aureus at 3.2–6.5 µg/mL while barely damaging red blood cells; 17 of the 20 final peptides were more active against staphylococci than magainin, a natural antimicrobial peptide[9].

Libraries made by bacteria: phage display

A different solution to both problems comes from biology. In phage display, introduced by George Smith in 1985, a short random DNA sequence is inserted into the gene for a coat protein of a filamentous bacteriophage, a virus that infects Escherichia coli. Each phage then carries the encoded peptide on its surface and the matching gene inside[1][2]. The phage particle itself does the bookkeeping: whatever peptide it displays, the instructions for it travel with it.

Screening, often called panning, uses ordinary affinity purification. The target is fixed to a surface, the phage library is passed over it, unbound phages are washed away and the bound ones are released. This enriches binding phages, often a million-fold or more, and because they are still infectious they can be multiplied in fresh bacteria and selected again. After two or three rounds, single clones are grown and the peptide sequence is read from their DNA[2]. A typical library holds about a billion clones; a single 1.5 mL tube can contain a few hundred trillion phage particles, and a library, once made, can be regrown almost for free[2].

The price of this convenience is chemistry. Early biological libraries could use only the 20 natural L-amino acids, with disulfide bonds as the only way to close a ring[1]. One workaround uses mirror-image targets: a protein domain was synthesised chemically from D-amino acids and used to select ordinary L-peptides, whose mirror images then bind the natural protein and resist degradation in the body[2]. Later display methods, above all mRNA display, accept unnatural and D-amino acids directly and are now used to screen libraries of cyclic peptides[1].

Finding the winner

Each format answers the second question, which peptide bound, in its own way[1][3]:

Standard Edman sequencing needs a free N-terminus and stops at the first building block that is not an α-amino acid[3]. For other building blocks, bead libraries use mass spectrometry or coding: a second, easily read set of molecules is attached to each bead alongside the test compound and records its synthetic history. One trick caps about 10% of the chains at every step, so that each bead carries a ladder of shortened sequences whose mass differences spell out the sequence. Because a positive bead may be unique, it can be cut in half so that one half is kept in reserve[3].

What libraries have found

The first applications were in immunology, mapping the epitopes recognised by antibodies down to single amino acids[5][4][9]. Libraries also turn up mimotopes, a term coined by Geysen's group: peptides that bind an antibody or receptor in the same place as its natural ligand without resembling the natural sequence[2]. In one bead library, a hexapeptide selected with an antibody against hepatitis A virus reacted specifically with the patients' antibodies even though it shared no sequence with the virus's coat proteins[3]. Other targets have included proteases, protein kinases, integrins and cell-surface receptors[3], and screening a patient's serum against a phage library can yield diagnostic peptides without studying the pathogen itself[2].

The early hopes for drug discovery were high, and by the account of Lam's own group they have not been fully met[1]. Pharmaceutical companies have moved towards smaller libraries designed around a particular target, but very large libraries remain the method of choice for targets whose structure is unknown[1]. The split-and-mix idea also travelled well beyond peptides: today the same procedure is used to put DNA barcodes on microbeads for reading the RNA of thousands of single cells[1].

References

1. Liu R, Li X, Lam KS (2017). Combinatorial chemistry in drug discovery. Current Opinion in Chemical Biology 38:117–126. DOI: 10.1016/j.cbpa.2017.03.017
2. Smith GP, Petrenko VA (1997). Phage display. Chemical Reviews 97:391–410. DOI: 10.1021/cr960065d
3. Lam KS, Lebl M, Krchňák V (1997). The "one-bead-one-compound" combinatorial library method. Chemical Reviews 97:411–448. DOI: 10.1021/cr9600114
4. Houghten RA (1985). General method for the rapid solid-phase synthesis of large numbers of peptides: specificity of antigen-antibody interaction at the level of individual amino acids. Proceedings of the National Academy of Sciences USA 82:5131–5135. DOI: 10.1073/pnas.82.15.5131
5. Geysen HM, Meloen RH, Barteling SJ (1984). Use of peptide synthesis to probe viral antigens for epitopes to a resolution of a single amino acid. Proceedings of the National Academy of Sciences USA 81:3998–4002. DOI: 10.1073/pnas.81.13.3998
6. Frank R (2002). The SPOT-synthesis technique: synthetic peptide arrays on membrane supports — principles and applications. Journal of Immunological Methods 267:13–26. DOI: 10.1016/S0022-1759(02)00137-0
7. Furka Á, Sebestyén F, Asgedom M, Dibó G (1991). General method for rapid synthesis of multicomponent peptide mixtures. International Journal of Peptide and Protein Research 37:487–493. DOI: 10.1111/j.1399-3011.1991.tb00765.x
8. Lam KS, Salmon SE, Hersh EM, Hruby VJ, Kazmierski WM, Knapp RJ (1991). A new type of synthetic peptide library for identifying ligand-binding activity. Nature 354:82–84. DOI: 10.1038/354082a0
9. Houghten RA, Pinilla C, Blondelle SE, Appel JR, Dooley CT, Cuervo JH (1991). Generation and use of synthetic peptide combinatorial libraries for basic research and drug discovery. Nature 354:84–86. DOI: 10.1038/354084a0