Phage Display: How Peptide Drugs Are Actually Discovered

Drug discovery

Phage display is the reason a surprising number of peptide drugs share a strange property: nobody designed them. They were selected, out of a pool of a billion candidates, by a process closer to fishing than to chemistry. The technique won a share of the 2018 Nobel Prize in Chemistry, and understanding it explains something that otherwise looks like a mistake — why a drug that activates a hormone receptor can have no chemical resemblance whatsoever to the hormone.

The problem it solves

If you want a molecule that binds a particular protein, the rational approach is to study the protein, work out what its natural partner looks like, and design something similar. That works when you know the answer already. It works badly when you do not.

The alternative is to make an enormous number of different molecules and let the target pick. The obstacle has always been bookkeeping: if you mix a billion peptides together and one of them sticks, you have a few molecules of an unknown substance and no way to find out what it was. Sequencing a picogram of peptide is not a realistic proposition.

George Smith's 1985 solution was to make each candidate molecule carry its own instruction manual. He inserted foreign DNA into gene III of a filamentous bacteriophage — a thread-like virus that infects E. coli — so that the encoded peptide appeared on the outside of the virus particle, fused to the pIII coat protein, while the DNA encoding it sat safely inside. Bind the particle, and you have captured the sequence too. Grow the particle in bacteria, and you have as much of it as you like. That link between what a molecule does and what it is — between phenotype and genotype — is the entire trick.

Smith described the constructs as “filamentous fusion phage.” In his Nobel lecture he put the scale plainly: a target immobilised on the surface of a small plastic Petri dish can be exposed to something like ten trillion virus particles representing a hundred billion clones.

How a phage display selection actually runs

The randomness is built in at the DNA level, using degenerate codons. The standard NNK codon — any base, any base, then G or T — is a mixture of 32 codons that between them encode all twenty amino acids. String six or seven of those together in the vector and you have a library.

What follows is affinity selection, universally called biopanning. The library is incubated with the immobilised target, unbound phage are washed away, whatever stuck is released with low-pH buffer, and the survivors are amplified in bacteria before the next round. The New England Biolabs protocol, which is the closest thing the field has to a standard recipe, runs three or four rounds and then sequences individual clones.

Five copies or two hundred: why valency matters

Where on the virus you display the peptide changes what you are able to find. The pIII coat protein is present at five copies per particle, and all five can carry a short peptide without destroying infectivity. The major coat protein pVIII is present at roughly 2,700 copies, of which about 10% can reliably be fused, giving around 200 copies of the peptide per particle.

That difference is not cosmetic. Two hundred copies of a weak binder can hold onto a target through avidity — many poor grips beating one good one — so pVIII display will happily recover ligands with affinities that would never survive a pIII selection. High valency is therefore useful when you have no starting point at all, and misleading when you want a molecule that will work as a monomer in a patient. Commercial peptide libraries are pIII-based for exactly this reason.

The uncomfortable part: most hits are wrong

A selection always returns something. Whether it returns something that binds your target is a separate question, and the literature on this is blunt. A widely cited 2011 review is titled “Phage Display: Selecting Straws Instead of a Needle from a Haystack.”

The false positives fall into two families. Selection-related artifacts are clones that genuinely bind something — just not your target. Candidates include the polystyrene plate itself, the streptavidin or protein A used for capture, the biotin on your biotinylated target, the milk or albumin used for blocking, and any contaminant in the target preparation. The classic example is the tripeptide motif HPQ, which binds streptavidin at the same site as biotin and turns up so reliably that its absence from a streptavidin control experiment is treated as a troubleshooting failure.

Propagation-related artifacts do not bind anything. They simply grow faster. Because every round ends with amplification in E. coli, a clone with even a small replication advantage — from the displayed peptide interfering less with assembly, or from a mutation elsewhere in the phage genome — compounds that advantage exponentially across rounds until it dominates the pool.

Deep sequencing has since shown that much of the damage is self-inflicted. Sequencing the naive library after a single amplification already reveals propagation bias, and in at least one published comparison a single round of panning followed by next-generation sequencing identified true binders that repeated rounds would have buried. The traditional three-or-four-round protocol is a workaround for cheap sequencing, not a law of nature.

What it actually produced

The most striking thing about display-derived drugs is how little they resemble the natural ligands they replace.

Romiplostim (Nplate, approved 2008) activates the thrombopoietin receptor to raise platelet counts in chronic immune thrombocytopenia. Its FDA label describes it as a “peptibody” — an IgG1 Fc domain carrying thrombopoietin-receptor-binding peptides — and then states, flatly, that romiplostim has no amino acid sequence homology to endogenous TPO. A molecule that switches on a hormone receptor while sharing nothing with the hormone is the cleanest possible illustration of what selection buys you that design does not. Worth a caveat, though: the Affymax programme behind it used phage display alongside a peptides-on-plasmids system and non-phage libraries for affinity maturation, so “a phage display drug” oversimplifies its ancestry.

Peginesatide (Omontys) is the unambiguous peptide-from-phage-display case, and also a cautionary one. Its parent peptide came from phage libraries screened against the erythropoietin receptor; again, the label notes it has no sequence homology to erythropoietin. It was approved in March 2012 for anaemia in dialysis patients, recalled in February 2013 after postmarketing reports of serious hypersensitivity reactions including anaphylaxis, and formally withdrawn in 2019.

Ecallantide (Kalbitor, 2009) came from a phage library built on a human protein scaffold rather than a random peptide, and is a 60-amino-acid plasma kallikrein inhibitor for hereditary angioedema attacks. Adalimumab (2002) is the antibody side of the same story — the Nobel Foundation calls it “the world’s first pharmaceutical based on a human antibody.”

Where the field went next

Phage display has a hard ceiling built into it: the peptides are made by a bacterial ribosome reading the ordinary genetic code, so you get the twenty natural amino acids and nothing else. Natural peptides are chewed up by proteases and do not cross the gut wall, which is why almost everything in the list above is injected.

Cell-free display methods removed that ceiling. In mRNA display, a puromycin linker covalently fuses each peptide to its own message during in vitro translation — no bacteria, no transformation step capping library size, and, critically, the translation machinery can be fed non-natural building blocks. Charging transfer RNAs with N-methylated and D-amino acids produces macrocyclic peptides that proteases struggle to recognise.

The payoff arrived recently. Zilucoplan, a 15-residue macrocyclic complement C5 inhibitor identified by mRNA display, was approved by the FDA in October 2023 for generalised myasthenia gravis. In July 2026 the FDA approved enlicitide, a macrocyclic peptide PCSK9 inhibitor from the same class of screening technology — and it is a tablet. A peptide taken by mouth was, for most of phage display's history, close to a contradiction in terms.

Frequently asked questions

Did phage display win the Nobel Prize?

It shared one. The 2018 Nobel Prize in Chemistry was divided, one half to Frances Arnold for the directed evolution of enzymes, the other half jointly to George Smith and Gregory Winter “for the phage display of peptides and antibodies” — a quarter share each.

Is a peptide from a phage display screen a drug?

No. It is a hit: a sequence that survived a selection. Between a hit and a medicine sit orthogonal binding confirmation, artifact screening, affinity maturation, stability engineering, pharmacokinetics and clinical trials. The great majority of published screen hits never leave the paper they appeared in.

Why do so many display-derived peptides have to be injected?

Because they are built from the same twenty amino acids as the proteins your digestive tract exists to break down. Protease susceptibility and poor membrane permeability are structural consequences of the chemistry, not formulation problems — which is covered in more detail in our guide to why peptides are not usually pills.

What replaced phage display?

Nothing replaced it; it is still in routine use and still the cheapest way to run a selection. But for peptide therapeutics specifically, mRNA display has taken the lead, because escaping the genetic code is what makes protease-resistant macrocycles possible.

Related reading

• Peptide reconstitution calculator — work out concentration and volume for a given vial.
• Peptide library — reference pages for individual compounds.
• Venom-derived peptide drugs: from snake bite to pharmacy

References

1. Smith GP. Filamentous fusion phage: novel expression vectors that display cloned antigens on the virion surface. Science 1985;228:1315–1317. pubmed.ncbi.nlm.nih.gov
2. The Royal Swedish Academy of Sciences. The Nobel Prize in Chemistry 2018 — popular science background. nobelprize.org
3. Smith GP. Phage display: simple evolution in a Petri dish. Nobel Lecture, 8 December 2018. nobelprize.org
4. Vodnik M, Zager U, Strukelj B, Lunder M. Phage display: selecting straws instead of a needle from a haystack. Molecules 2011;16:790–817. pmc.ncbi.nlm.nih.gov
5. US FDA. Nplate (romiplostim) prescribing information, BLA 125268 — peptibody description and the absence of TPO sequence homology. accessdata.fda.gov
6. New England Biolabs. Ph.D. phage display peptide library kits, instruction manual — library complexity and panning protocol. neb.com
7. US FDA. Approval letter for Zilbrysq (zilucoplan), NDA 216834, 17 October 2023. accessdata.fda.gov
8. Federal Register. Withdrawal of approval of the new drug application for Omontys (peginesatide), 13 February 2019. federalregister.gov

Informational only — not medical advice. VialHelp does not sell peptides and does not recommend any product, treatment or dose. Nothing here should be read as guidance on using any compound named above. Discuss any medical question with a qualified healthcare professional. Intended for readers 21+.