Chapter 35 — Key Takeaways

The frame

There are four routes to a peptide drug, and all four are in active use.

1. Find it in nature — isolate a molecule some organism already makes.
2. Start from an endogenous ligand and modify it — the Chapter 33 route.
3. Screen an enormous library and select what binds — display technologies.
4. Design it computationally — structure-based and de novo design.

They are roughly chronological. The newest has not retired the oldest, and most interesting drugs involve more than one.

Every one of them answers the same question — how do I obtain a molecule that binds this target? — and none of them answers will binding this target help a sick person? That distinction is the chapter.

Venom

• Venoms are pre-optimized pharmacological libraries. Evolution has spent millions of years selecting peptides that act fast, potently, and specifically on vertebrate nervous, cardiovascular, and muscular targets. That is precisely the design brief of a drug.
• Venom peptides are selective because indiscriminate venom is metabolically wasteful. Venom is expensive and slow to regenerate; specificity is cheap and effective.
• Many are disulfide-stapled and protease-resistant, arriving pre-solved for the stability problem Chapter 33 attacks by design. Stability is not drug-likeness — they are still undeliverable orally, still renally cleared, still shut out of the brain.

The Gila monster, and the correction to it

• Exendin-4 was identified in the venom of the Gila monster (Heloderma suspectum), work associated with John Eng in the early 1990s.
• It shares roughly half its residues with human GLP-1 — enough to activate the human GLP-1 receptor, different enough at the critical position that human DPP-4 does not cleave it.
• Synthetic exendin-4 became exenatide, approved in 2005 as the first GLP-1 receptor agonist.
• The irony: Chapter 33's entire lipidation-and-substitution program exists to solve a problem a lizard had already solved. The first drug in the century's most consequential peptide class was not designed. It was found.
• The counterweight: exenatide's duration was still inadequate for many patients, and the drugs that displaced it — liraglutide, semaglutide — were engineered. Nature supplied the lead; chemistry supplied the drug. Both clauses.

Other natural sources

• Captopril descends from bradykinin-potentiating peptides in the venom of the Brazilian pit viper (Bothrops jararaca), via the injectable peptide teprotide. Ondetti, Cushman, and colleagues designed a small molecule reproducing the key interactions; approved in the early 1980s. Captopril is not a peptide — it is medicine's earliest and most commercially important peptidomimetic. Same arc as orforglipron in Chapter 33: peptide lead → small-molecule drug.
• Ziconotide derives from ω-conotoxin MVIIA (Conus magus), blocks N-type calcium channels, and is approved for severe chronic pain — only intrathecally, because it does not cross the blood-brain barrier. Chapter 4's delivery problem at its most extreme: the molecule works, so the route had to become surgical.
• Magainins, from Xenopus laevis skin, identified by Michael Zasloff in the late 1980s, opened the antimicrobial peptide field (Chapter 25).

Rational design and display

• Route 2 is the origin of most peptide drugs in this book: insulin, GLP-1, somatostatin, GnRH, vasopressin, and PTH analogs. The target, receptor, and physiology are known from the start.
• The cost: you inherit the endogenous ligand's selectivity profile, including its off-target activity — which is why so many peptide drugs share their parent hormone's side effects.
• Phage display (George Smith, mid-1980s; developed for antibodies by Greg Winter; a share of the 2018 Nobel Prize in Chemistry, with Frances Arnold for directed evolution) rests on a physical link between a displayed peptide and the DNA encoding it. Because each phage carries its own instructions, the winners can be sequenced.
• Ribosome and mRNA display extend this to cell-free systems, permitting libraries far larger than can be transformed into bacteria.
• Display technologies do not design anything. They are selection, industrialized — evolution run on a bench in a week. Their output is a sequence that binds, with no explanation attached.

Structure, prediction, and design

• Class B GPCRs — the GLP-1 receptor's family — were historically very hard to crystallize. Cryo-EM changed that, producing structures of activated receptor–ligand–G-protein complexes, which is what made §33.9's multi-agonist ratio problem a tractable engineering question.
• AlphaFold2 produced a decisive result at CASP14 in 2020; the work led to a share of the 2024 Nobel Prize in Chemistry for Demis Hassabis and John Jumper, shared with David Baker for computational protein design.
• What changed: sequence-to-structure went from career-length to minutes, and hundreds of millions of predicted structures were released publicly.
• What did not change: (1) structure is not function; (2) short peptides frequently have no single stable structure to predict — exactly the molecules this book is about; (3) predicting a structure is not predicting a drug; (4) the predictions are predictions, least confident precisely where disorder — and much interesting biology — lives.
• De novo design (David Baker and others; ProteinMPNN, RFdiffusion) has produced binders with no natural counterpart, some hitting targets at high affinity on the first pass. For the first time, the space of possible peptides is being searched by design rather than by selection or by luck.

The point of the chapter

Binding was never the bottleneck.

• Every technique above accelerates finding a molecule that binds a target.
• The dominant causes of clinical failure are lack of efficacy in humans and unacceptable toxicity — and neither is fixed by generating candidates faster. Chapter 9 (Phase 2 optimism), Chapter 10 (attrition), Chapter 16 (surrogates), and Chapter 22 (substance P antagonists that bound beautifully and did not treat pain) each supply a piece of this.
• A pipeline whose narrow point is a three-year outcomes trial does not run faster because the first step got quicker.

And be fair to the technology. Cheaper, faster discovery means more shots on goal, and it makes economically marginal targets — rare diseases, neglected infections — viable to pursue at all. That is a real and significant gain. It is simply not the same claim as "AI will cure disease faster."

The general rule, which will outlive every technology named in this chapter: A technology that improves one stage of a pipeline improves the whole pipeline only if that stage was the constraint.

Ratings issued in this chapter

Dossier

Field 12, sharpened. For any ⚠️ entry, write the specification of the study that would resolve it: population, endpoint, comparator, duration, size, blinding — plus both thresholds, the one that would upgrade the rating and the one that would downgrade it. A specification with only one direction is a wish.

A reader who cannot describe the study that would settle their question does not yet know what they are uncertain about.

Coda: record each compound's origin — found, derived, selected, or designed — and then record that origin carries no evidentiary weight whatsoever. Exenatide came from a lizard and is ✅; several ❌-rated compounds are exact fragments of human proteins. An origin story is the most common substitute for evidence in peptide marketing and one of the least informative facts about a molecule (Chapter 31 makes the identical argument about veterinary use).