The Decades When Peptides Fell Out of Fashion

Put yourself in the chair of a medicinal chemist in the early 1990s, handed a disease target and asked where to start. Behind you sits one of the most consequential drug classes medicine has produced — insulin, in clinical use since the 1920s, turning a fatal diagnosis into a managed condition, and a run of hormone therapies built on the same principle. In front of you sits a robot that can test tens of thousands of small synthetic compounds a week against a purified protein. You do not choose a peptide. Almost nobody did, for about twenty years. The industry turned away from peptide drugs because they failed on four practical counts: they could not be swallowed, they were destroyed in the bloodstream within minutes, they were costly to manufacture at scale, and they violated nearly every rule of thumb the era had adopted for what a drug ought to look like 14. Whether that reasoning was sound at the time is not the interesting question. It was. The question is what it took to undo.

It is tempting to file the episode under fashion — peptides in, peptides out, peptides back in. But fashion is the wrong metaphor. What changed was a list of specific technical objections, each answered separately, mostly by people not trying to rehabilitate peptides at all.

The drugs that came first

A peptide is a short chain of amino acids — the same building blocks proteins are made from, run to perhaps five or fifty rather than several hundred. That length happens to be the body's own vocabulary for signalling: a great many hormones, and much of what one tissue says to another, is written in peptides.

Which is why they arrived in medicine so early. Insulin established the principle that a molecule the body already makes can be extracted, purified and given as a drug, decades before anyone could describe how it worked. A series of pituitary and hypothalamic peptides followed it into the clinic — hormones governing labour, water balance, growth and reproduction, each a naturally occurring signal repurposed as a therapy 1.

The appeal was not subtle. Evolution had already done the medicinal chemistry: these molecules were potent at very low concentrations, unusually selective, and broke down into amino acids rather than into anything needing a toxicology programme of its own. Set against that inheritance, walking away looks eccentric. It was not.

The rules a peptide breaks

Start with the plainest problem. A peptide cannot ordinarily be swallowed, because the digestive tract is purpose-built to dismantle exactly this kind of molecule. Proteases — enzymes that cut amino acid chains at recognisable points — begin the work in the stomach and finish it in the small intestine, and whatever survives still has to cross the gut wall, which a large, water-loving molecule does very poorly. So the route is injection, which narrows the range of conditions a drug can plausibly treat.

Getting past the gut would have bought little time anyway. Proteases circulate in blood too, and the kidney filters small molecules out of the bloodstream efficiently. Between them, an unmodified peptide can be cleared in minutes. A drug with a half-life of minutes is not a tablet or a convenience — it is an infusion line, or nothing 4.

Third, cost. Assembling a chain of thirty residues one residue at a time is a different industrial proposition from a six-step route to a compound of forty atoms, and the gap in cost per kilogram was then wide enough to shape portfolio decisions by itself.

Then the decisive objection, which was not really about peptides at all. The era belonged to small-molecule discovery. Combinatorial chemistry promised enormous libraries generated in parallel rather than one compound at a time; high-throughput screening promised to test them against a target by the tens of thousands, automatically. Together they implied that finding a drug was becoming a numbers problem best answered by making the numbers very large. Peptides could not be fed into that machine — slow to make in variety, and filtered out at the far end against criteria they could not satisfy.

A modest peptide fails that filter comprehensively, and the scale of the mismatch is the point 4.

• Molecular weight under roughly 500 — a thirty-residue peptide runs to around 3,000, six times over.
• No more than five hydrogen-bond donors — the backbone contributes one at every amide link, before any side chain is counted.
• No more than ten hydrogen-bond acceptors — the backbone alone passes that mark before a short chain is half built.
• A lipophilicity suited to crossing a membrane unaided — peptides are polar by construction.

Here is the honest version, and it requires nobody to have behaved badly. Peptides were excluded by criteria written to describe a different kind of molecule — an accurate summary of what small orally absorbed drugs looked like, generalised, quietly, into a definition of what a drug was.

Where the expertise survived

Peptides never went to zero, and that matters more than it sounds. They persisted where the main objection had already been settled by circumstance: injection was routine, and no alternative molecule existed to be preferred. Endocrinology above all — diabetes care, fertility medicine, disorders of growth, the branches of oncology that work by suppressing a hormonal axis. In each the therapy was a peptide because the natural signal was a peptide, and nothing else would speak to the receptor. Arguing that patients would not accept an injection carried no weight in a clinic where they already did 1.

So the know-how survived there: methods for proving a chain is the one you meant to make, formulation tricks that keep it stable in a vial, regulatory precedent, the plant and the people who knew how to run it. When the field turned back, it did not start from a standing position. It picked up equipment kept warm for two decades by a specialty that never had the option of abandoning it 3.

Turning minutes into weeks

The clearance problem looked, for a long time, like a fact of nature. It turned out to be an engineering problem with several independent solutions, catalogued by the 2000s well enough to be taught rather than rediscovered 5.

The first family of fixes attacks the enzyme's recognition step. A protease does not chew indiscriminately; it binds a particular short stretch of chain and cuts at a particular bond. Substitute an unnatural amino acid there — a mirror-image residue, or one carrying an extra methyl group on the backbone nitrogen — and the active site no longer fits what it is holding. The scissors are still there; the handle has changed 5.

The second is cyclisation: joining the ends of the chain, or bridging two side chains partway along, so the molecule becomes a loop rather than a length of string. Proteases generally need their substrate stretched into an extended conformation before they can cut it, and a constrained ring cannot adopt that shape. A rigid molecule also pays a smaller entropic penalty on binding, so cyclisation often improves affinity as well as survival 4.

The third changed the arithmetic most dramatically. Attach a fatty acid chain to the peptide — lipidation — and it binds reversibly to albumin, the most abundant protein in blood. Albumin is cleared very slowly and is far too large for the kidney to filter, so while the peptide is riding it, it is effectively invisible to both mechanisms that were removing it. It detaches, acts, reattaches. Minutes become days 53.

Meanwhile the cost objection quietly eroded: better resins, more reliable couplings, routes that build two fragments separately and join them at the end, purification at a scale nobody had needed in 1975. None of it was a breakthrough. It was twenty years of the kind of process chemistry that produces no headlines and eventually moves a cost line by an order of magnitude 32.

The targets that would not cooperate

The third strand turns the story around, because it has nothing to do with peptides improving. It is about which targets turned out to matter, and how badly they suited the tools of the screening era.

A small molecule does its best work in a pocket. An enzyme's active site is a deep concave cleft — a hole with a shape — and a rigid compound of a few dozen atoms dropped into it makes many contacts at once, held on all sides. That geometry is what the screening machine was implicitly optimised for, and against such targets it worked extremely well.

But an enormous amount of biology is decided elsewhere, at protein-protein interactions: two large proteins meeting across a broad, shallow, comparatively featureless surface. Binding strength comes overwhelmingly from contact area, and here a small molecule runs out of options. It can touch only a patch of a wide flat face, and there is no cleft to sit in, so nothing holds it — a few weak contacts, and it drifts off. Interrupting an interaction like that needs a molecule that can spread along the surface, matching it over a distance. Which is a fairly exact description of a peptide 42.

The same logic applies to a receptor family that became commercially central. G-protein-coupled receptors are the cell's largest set of signalling switches, and the class B subgroup — receptors for the body's larger hormone peptides, including those governing appetite and insulin release — are addressed by their natural ligands in a distinctive way. The hormone threads into a long groove on a sizeable extracellular domain, then reaches down into the receptor core, engaging along its whole length. There is no compact pocket to fill. Years of small-molecule effort here produced comparatively little, and in hindsight that is not mysterious: these receptors evolved to be spoken to by something long 2.

The vindication nobody set out to win

The metabolic medicines that have dominated the last decade — the incretin drugs acting on class B receptors, and the multi-receptor agonists that followed — are peptides, and peptides carrying precisely the modifications described above: substitutions where a protease would otherwise cut, a fatty acid chain to keep them bound to albumin, sequences tuned for selectivity. Their half-lives are measured in days rather than minutes because somebody solved that problem deliberately.

They are also, by any reasonable measure, among the most commercially significant medicines in the world. The class set aside in the 1980s for being unruly, expensive and un-drug-like became the class that worked on the targets nothing else could reach. More than eighty peptide drugs have reached the market worldwide, and the pipeline behind them has grown rather than shrunk — the opposite trajectory to the one the field was on when the screening robots arrived 2.

What actually changed

It would be easy, and cheap, to end on the industry having got it wrong. It did not, in any sense that matters. Judge the 1985 decision on the information available in 1985 and it holds up: a molecule that must be injected, survives minutes, costs a great deal per gram and cannot reach anything inside a cell is a poor bet against a library of a hundred thousand orally available compounds — provided you believe the library will deliver. Plenty of thoughtful people believed exactly that, on good evidence.

What changed afterwards was not anyone's opinion. There was no moment of reappraisal, no argument won, no paper that persuaded the field peptides had been undervalued all along. The objections were dismantled one at a time, over about twenty years, by people working on much narrower problems: a chemist keeping one molecule intact for one study, a process group taking a cost line down by a third, a structural biologist mapping a receptor for reasons of their own 3.

That is the shape progress takes here, and it predicts what comes next. The field's current objections — oral delivery still unsolved for most peptides, targets inside the cell still largely out of reach, very long chains still expensive — are simply the next list 2. They will not be crossed off by anyone deciding peptides deserve more credit. They will go the same undramatic way the last list did: one obstacle at a time, by people who were mostly looking at something else.