Compound Monograph  ·  No. 20  ·  Research Use Only

Cagrilintide A hormone found as a disease, and rebuilt as a drug

In 1901 a pathologist in Baltimore opened the pancreas of a man who had died of diabetes and found the islets replaced by a waxy, structureless material he could describe but not identify. It took eighty-five years to find out what it was, and the answer was not what anyone expected: not a residue of insulin, but a second hormone the beta cell had been making all along, whose defining physical property was that it clumped. Turning that hormone into a medicine meant engineering away the exact property by which it had been discovered. Two drug generations did it. Cagrilintide is the second, and this document is about what it changed, what it demonstrably does, and how much smaller its own evidence base is than the attention around it suggests.

Compiled by South Beach Longevity · 2 August 2026
Copyright 2026
Corpus 1,607 scientific full texts · ~32,253 printed-page equivalents
Metadata layer 5,959 screened PubMed records of 6,288 returned · 42 references
Source project 05 · Therapeutic Peptide Research Library
Constraint No human use, dose, route or schedule is recommended anywhere in this document

01Eighty-five years of hyaline

Eugene Opie was twenty-eight and working at Johns Hopkins when he published, in 1901, a description of what happens to the pancreas in diabetes. Among the changes he catalogued was one he called hyaline degeneration of the islands of Langerhans: the insulin-producing tissue displaced by a translucent, homogeneous substance that took up stains in a characteristic way and could be described in no more useful terms than that (Opie, 1901). He had no way of knowing what it was made of. Nobody did, for a very long time.

The delay is worth pausing on, because it is not a story about slow science. The material sat in plain sight in thousands of autopsy specimens across four continents for most of a century. It was recognised as amyloid — the generic name for a class of insoluble protein deposits — and the reasonable assumption, held for decades, was that it derived from insulin or one of insulin's precursors, since insulin was what those cells made. The problem was mechanical. Amyloid is defined by being insoluble. Every technique available for purifying and sequencing a protein begins by getting it into solution, and this material's whole identity was its refusal to go. The substance defeated identification for the same reason it accumulated in the pancreas.

The problem was solved twice, independently, within about eighteen months of each other, by two groups approaching from opposite ends.

02Two laboratories, and two names

Per Westermark, in Sweden, came at it from the deposit. He and his colleagues extracted the fibril protein from the endocrine pancreas and found that it was not insulin-derived at all but a novel peptide belonging to the calcitonin gene–related peptide family (Westermark et al., 1986). That family assignment mattered more than it can have seemed at the time: it is the first hint of the receptor biology on which every drug in this document depends, arriving nearly forty years before anyone would use it.

The following year the same group published the observation that settles the question of what the peptide is. Amyloid fibrils from human insulinoma and from the islets of diabetic cats were derived from a protein also present in normal islet cells (Westermark et al., 1987). The word carrying the weight is normal. A substance found only in diseased tissue is a product of the disease. A substance found in healthy tissue, which sometimes aggregates, is a normal constituent behaving badly — which is to say, a hormone. They proposed the name islet amyloid polypeptide.

Garth Cooper, working at Oxford with Robert Turner and Kenneth Reid, came at it from the chemistry. His group purified and sequenced a thirty-seven-residue peptide from amyloid-rich pancreases taken from people who had died with type 2 diabetes (Cooper et al., 1987a). They initially called it diabetes-associated peptide (Cooper et al., 1987b; Clark et al., 1987), then renamed it amylin when they proposed that it was a hormone in its own right, one regulating glycogen metabolism in skeletal muscle (Cooper et al., 1988; Cooper et al., 1989).

One correction is due to the original framing, and it matters because the early Cooper papers are still cited loosely. The hypothesis those papers advanced was that amylin causes insulin resistance in skeletal muscle (Leighton and Cooper, 1988) — that it was, in effect, a diabetogenic hormone. That is not the basis of the modern drug class. The therapeutic rationale that produced pramlintide and cagrilintide rests on an entirely different set of actions, described below, and the muscle hypothesis has not carried forward. The discovery papers are landmarks; they are not a description of why anyone now gives the molecule to anybody.

03What the hormone actually does

Amylin is made by the pancreatic beta cell, stored in the same secretory granules as insulin, and released with it when a meal arrives. That co-secretion is the structural fact from which its function follows: it is not an independent signal but a companion to insulin, released on the same schedule and reporting the same event.

Three actions are consistently attributed to it. It reduces meal size and brings forward the point at which eating stops. It slows the rate at which the stomach empties into the small intestine. And it suppresses the secretion of glucagon after a meal, opposing the hormone that raises blood glucose at precisely the moment when insulin is trying to lower it. Taken together these describe a hormone whose job is to regulate the rate at which nutrients arrive — to keep the post-meal delivery of glucose within a range that insulin can manage.

There is a species difference here that constrains the reading of every animal experiment in this document, and it is not a technicality. Human amylin aggregates; rat amylin does not. The difference traces to a short stretch of the sequence around residues 20 to 29, where the rodent peptide carries prolines that the human peptide lacks. Rats and mice do not develop islet amyloid. The chemistry that made the human hormone undruggable is absent in the animals used to study it — which turned out, as Part Two describes, to be the source of the solution rather than merely a nuisance.

The aggregation itself proceeds by a route now well characterised for amyloid-forming peptides generally. Soluble monomer associates into small oligomers; those template the formation of protofilaments in which the peptide backbones stack in register, hydrogen-bonded to one another along the length of the growing fibril in the arrangement called cross-β; protofilaments wind together into the mature fibrils that Opie saw. The process is self-templating, which is why it runs to completion once started, and why it happens in a manufacturing vessel as readily as in an islet.

That is the state of knowledge the drug programmes inherited. A thirty-seven-residue beta-cell hormone with a coherent physiological job, a clear therapeutic logic in diabetes and obesity, and one disqualifying property: it turns itself into an insoluble solid. No formulation chemistry can manufacture, store or inject a peptide that does that. The molecule had to be changed before it could be a medicine, and the question was what to change.

04Pramlintide: the first problem solved, the second not

The solution to the aggregation problem was sitting in the rat. If rodent amylin does not form fibrils, and the sequence difference between rodent and human amylin is concentrated in a short stretch, then the obvious experiment is to move the rodent residues into the human peptide and see whether the aggregation goes away with them. It did. Substituting proline at positions 25, 28 and 29 — the three positions where the rat sequence differs in the aggregation-prone region — produces a peptide that retains amylin's actions and does not turn into a solid.

The chemistry is unusually clean, and worth understanding because it is inherited unchanged by cagrilintide. A β-sheet is held together by a ladder of hydrogen bonds running between backbone amide groups on adjacent strands. Proline is the one amino acid whose side chain loops back and bonds to its own backbone nitrogen, which means that nitrogen has no hydrogen left to donate. A proline in the middle of an amyloidogenic stretch is a missing rung. The ring also stiffens the backbone against adopting the extended conformation the sheet requires. One residue, two independent reasons the sheet cannot propagate.

The resulting molecule, pramlintide, was approved in the United States in 2005 as an adjunct to mealtime insulin. It was the first amylin analogue to reach a patient, and it demonstrated that amylin-receptor agonism does what the physiology predicted: reduced post-meal glucose excursions, suppressed glucagon, slower gastric emptying, modest weight loss.

And it was a difficult drug to take. Its elimination half-life is approximately forty-eight minutes. A signal that short has to be re-created before every meal, which means an injection before every meal, on top of the insulin injections the patient is already giving themselves. Pramlintide never became a widely used medicine, and the reason was not that it failed to work. It was the schedule.

05The analogues that did not arrive

Between pramlintide and cagrilintide there is a gap of roughly two decades in which amylin analogues were developed and did not reach the market. The most advanced was davalintide, a chimeric peptide combining elements of amylin and calcitonin, developed by the same company that produced pramlintide and discontinued during clinical development.

This is recorded because a monograph that lists only the survivors misrepresents how hard the problem was. The amylin receptor is not an easy target: it is not a single protein but an obligate complex, its pharmacology depends on which accessory subunit a given cell happens to express, and the therapeutic window is bounded by nausea in a way that has ended more than one programme. The twenty-year gap is the honest measure of the difficulty, and it is also the reason the current wave of amylin analogues — there are now several — arrived together rather than in sequence. They are all downstream of the same enabling technology.

06The technology that was borrowed

That technology did not come from the amylin field. It came from the incretin field, where the problem of turning a minutes-long peptide signal into a weekly one had already been solved, twice, for glucagon-like peptide 1.

The method is to attach a fatty acid to the peptide. Not for any effect on the receptor — the fatty acid is not part of the pharmacophore — but because a fatty acid binds reversibly to human serum albumin, the most abundant protein in plasma, which circulates for weeks. A peptide tethered to a fatty acid spends most of its time attached to albumin, and while attached it is invisible to the two processes that would otherwise destroy it. The albumin–peptide complex is far too large to pass the glomerular filter, so the kidney cannot clear it. And the bound peptide is sterically shielded from the plasma proteases that would cut it up.

The consequence is a depot made out of the patient's own blood. Only the small unbound fraction can reach a receptor, and as that fraction is consumed more peptide dissociates from albumin to replace it. The circulating pool behaves as a slow-release reservoir. A second depot forms at the injection site, where the acylated peptide self-associates and is absorbed over days rather than minutes.

The technique had produced liraglutide and then semaglutide. Applying it to a different hormone family was not conceptually novel; the interesting part was where to attach the chain, and what that would cost.

07Cagrilintide: the modification ledger

Cagrilintide was developed at Novo Nordisk and described in the medicinal chemistry literature in 2021, where it appears as compound 23 of a structure–activity series (Kruse et al., 2021). It carries two development codes, AM833 and NNC0174-0833, and received its International Nonproprietary Name in 2022 under the -lintide stem that designates amylin analogues.

Its identity, as registered, is a thirty-seven-residue peptide with the molecular formula C₁₉₄H₃₁₂N₅₄O₅₉S₂, a molecular weight of approximately 4,409, and CAS registry number 1415456-99-3 (PubChem CID 171397054; ChEMBL4802169). The two sulphur atoms are the cysteines of the disulphide bridge that closes the N-terminal loop, retained from native amylin.

The ledger of what was changed separates cleanly into inherited and new.

Inherited from pramlintide: the three prolines. Cagrilintide is not built on the human hormone. Its backbone is pramlintide's, which means the aggregation problem arrives already solved and none of the molecule's own innovations are addressed to it.

New, in the middle: glutamate at position 14 in place of asparagine, and arginine at position 17 in place of valine, positioned so that the two side chains form an intramolecular salt bridge. This is a helix-stabilising pair, and its stated purpose is the same as the prolines' — to further counteract the fibril formation that afflicts the human hormone — by locking the N-terminal helix into a defined conformation. Section 09 examines what it does at the receptor, which is less than it is usually credited with.

New, at the far end: proline at position 37, where the parent peptides carry tyrosine. This substitution is small, easy to overlook, and routinely stated backwards in secondary descriptions of the molecule. It is also, on the structural evidence, the change that gives cagrilintide its distinctive receptor profile. Section 09 sets out why.

New, and the reason the molecule exists: acylation at the N-terminal α-amino group of Lys1 — the peptide's own N-terminus, not the lysine side chain, which is left free — through a single γ-glutamyl linker to a twenty-carbon fatty diacid. There is no polyethylene-glycol spacer: the discovery programme tried them and abandoned them. The attachment point was not arbitrary either. The structure–activity work reports that acylation at other positions — among them 2, 10, 14 to 17 and 29 — gave reduced potency, and the N-terminal region was identified as the most promising.

The pharmacokinetic result is the largest single change in the molecule's behaviour, and it is worth stating as a ratio rather than as a pair of numbers. Native human amylin has an elimination half-life of about twelve minutes in people. Pramlintide, about forty-eight. Cagrilintide's reported half-life is 159 to 195 hours — between six and a half and eight days. That is roughly two hundred to two hundred and forty-five times pramlintide's, and seven to nine hundred times the native hormone's. The fatty acid does not change what the molecule does at the receptor; it changes how long it keeps doing it.

Two qualifications belong with those figures. The cagrilintide range is not a monotherapy measurement: it comes from a phase 1b trial in which the molecule was given together with semaglutide, and no monotherapy half-life has been published. And the widely repeated thirteen-minute figure for native amylin is a rat value, relayed into the human literature by a review; the human measurement is 11.8 ± 0.9 minutes by intravenous infusion.

With weekly administration the arithmetic consequence is accumulation to a steady state over roughly the first month — four to five weeks, from the half-life, though no primary study of this compound reports a measured time to steady state — after which concentrations oscillate within a relatively narrow band rather than falling to baseline between doses. The receptor is engaged continuously.

Which brings the design back to the word in Section 03. Pramlintide re-creates a mealtime signal at mealtimes; it is a phasic drug imitating a phasic hormone. Cagrilintide does something categorically different. It holds the amylin receptor occupied all week, including at three in the morning and between meals, when the hormone it is imitating would be absent. That is not a more convenient version of pramlintide. It is a different pharmacological proposition, and Section 11 returns to what is and is not known about whether the difference matters.

08One peptide, four receptors

The amylin receptor is not a protein. It is a partnership, and understanding that is the key to everything else in this Part.

The signalling component is the calcitonin receptor, a class B1 G protein-coupled receptor that works perfectly well on its own and responds to calcitonin. What converts it into an amylin receptor is the addition of a second, much smaller protein: a receptor activity-modifying protein, or RAMP, of which there are three. The calcitonin receptor partnered with RAMP1 is the amylin receptor AMY1R; with RAMP2, AMY2R; with RAMP3, AMY3R.

The RAMP does not signal. It has a single transmembrane pass and no machinery of its own. What it does is sit against the receptor and remodel the surface where the ligand binds, changing which peptides the complex recognises. The consequence is worth stating slowly, because it inverts the usual way of thinking about hormone specificity:

Cagrilintide is non-selective across all four, and the way that fact is usually stated is wrong in an instructive way.

Section 09 explains where that came from, because the answer is not the one the molecule's summary descriptions usually give.

One observation from the pharmacology constrains how that non-selectivity should be read. Both selective amylin-receptor agonists and dual calcitonin/amylin agonists reduce appetite and body weight; a calcitonin-receptor-selective agonist does not. Activity at the calcitonin receptor is therefore not, on its own, the mechanism of weight loss. It is an additional property of a molecule whose weight effect runs through the amylin receptors.

09What the structures actually show

In 2025 a group determined cryo-electron microscopy structures of cagrilintide bound to all four receptors in their active, Gs-coupled states (Cao et al., 2025). The consensus maps reached 2.2 Å for the calcitonin receptor, 2.2 Å for AMY1R, 2.7 Å for AMY2R and 3.0 Å for AMY3R. This is the highest-quality mechanistic evidence that exists for the molecule, and it is worth reading carefully because two of the claims commonly made about cagrilintide's design do not survive it.

The binding follows the two-domain mechanism characteristic of class B1 receptors. The peptide's C-terminal region is captured by the extracellular domains of the receptor and, where present, the RAMP; its N-terminal region inserts into the transmembrane bundle and drives activation. Residues 19 to 25 form what the field calls the bypass motif, a conserved arrangement that is a principal determinant of the high potency of amylin analogues at amylin receptors. Cagrilintide forms it.

The engineered salt bridge is real, and does something other than advertised

The paired substitutions at positions 14 and 17 — glutamate for asparagine, arginine for valine — do form the intramolecular salt bridge they were designed to form. The structures show glutamate 14 adopting a downward rotamer and making hydrogen-bonded and electrostatic contacts with arginine 17, on the same face of the peptide helix, in the active-state complexes. The authors call it an ionic lock. Rat amylin, which has asparagine at position 14, puts that residue in an upward rotamer instead, exposed to solvent and hydrogen-bonded to the backbone of the second extracellular loop.

What the pair was incorporated for, however, is helix stabilisation to counteract the fibril formation of human amylin, not receptor potency — and the discovery paper's own structure–activity table shows it cost potency at the calcitonin receptor. In matched compounds differing only at position 17, introducing the arginine that completes the salt bridge moved the half-maximal concentration at the human calcitonin receptor from 107 to 447 picomolar: four times worse. The structures show why the effect is not neutral either — the bulk of the two larger side chains pushes the peptide helix away from the second extracellular loop and slightly widens the binding pocket. The ionic lock is a real feature of the molecule and a real contributor to its conformation. It is not the reason the molecule engages the calcitonin receptor.

The reason is a proline at the far end

The C-terminus is where cagrilintide's receptor behaviour is decided, and the relevant substitution is one that secondary descriptions of the molecule routinely state backwards.

Rat amylin carries tyrosine at position 37. Cagrilintide carries proline. The change is tyrosine to proline, and it is not cosmetic.

In the amylin receptors, tyrosine 37 of rat amylin sits above a conserved tryptophan in the receptor's extracellular domain, forming a perpendicular π-stacking interaction with it — and, critically, additional contacts with the RAMP. Those RAMP contacts are a principal reason amylin prefers amylin receptors over the bare calcitonin receptor. Cagrilintide's proline occupies the same pocket, buries itself completely among the calcitonin receptor's own extracellular-domain residues, makes extensive hydrophobic contact with the same tryptophan, and forms no direct interaction with RAMP1 or RAMP3 at all. The authors state the consequence plainly: the absence of those RAMP contacts is consistent with cagrilintide's ability to bind the calcitonin receptor protomer on its own with high affinity.

The same table supplies the quantitative half of the argument. Adding the C-terminal proline to the salt-bridged compound took the human calcitonin receptor from 447 back to 62 picomolar — a sevenfold gain — and collapsed the AMY3R-over-CTR selectivity ratio from four to one. The discovery paper says as much in its own words: the molecule was designed with "clear inspiration from calcitonin, for example, a C-terminal proline and a salt bridge in the 8–18 region."

What the structures do not show

Two limits deserve stating, because published depictions of this molecule routinely draw past them.

The fatty acid is barely resolved at all. Density for the conjugated lipid appeared only in receptor-focused maps of two of the four complexes, and even there modelling stopped at the γ-glutamyl linker and a short adjacent segment. Counted from the deposited coordinates, at most nine of the twenty fatty-acid carbons were ever modelled in any structure, and in the AMY3R complex neither the linker nor the lipid was modelled at all. Where the linker could be traced, it exits the receptor in different places depending on which complex is being looked at — between the second extracellular loop and the top of transmembrane helix six at AMY1R, but along transmembrane helix five at the calcitonin receptor. And it merges into a detergent micelle, because these are detergent-solubilised complexes; there is no membrane. Diagrams that draw the chain taking one defined route out of the receptor and into a lipid bilayer are drawing an inference, and drawing it in the wrong medium.

And the acylation matters less to the receptor than the duration. The same study solved structures of a non-lipidated version of the peptide and found the lipid conjugation does not alter the conformation of either the peptide or the receptor once the peptide is engaged. In cell-based reporter assays the non-lipidated backbone is equipotent at AMY3R. It is not equipotent everywhere, though: at the bare calcitonin receptor it is about sixfold weaker, a difference the authors report as significant. So the fatty acid is overwhelmingly a pharmacokinetic device, as designed — but it makes a modest contribution at the one receptor where cagrilintide's profile differs most from its parent's.

What the structures do add, beyond the static picture, is a difference in motion. Cagrilintide binds in an amylin-like way but induces conformational dynamics at these receptors that differ from those induced by rat amylin or salmon calcitonin — in the AMY3R complex, for instance, the RAMP3 extracellular domain is markedly less well ordered, indicating substantial loss of the interface between it and the receptor. The authors suggest those dynamic differences may contribute to the molecule's clinical efficacy. That is a hypothesis offered by the structural work, not a demonstrated mechanism, and it should be read as one.

10Where the drug acts, and a mechanism that was revised

An intact blood-brain barrier does not admit a 4,400-dalton peptide, still less one spending most of its time bound to albumin. Cagrilintide nonetheless acts on neurons. The reason is anatomical rather than pharmaceutical.

The area postrema sits on the floor of the fourth ventricle in the dorsal medulla and is a circumventricular organ: its capillaries are fenestrated and its blood-brain barrier is incomplete rather than absent. Large circulating molecules — specifically, the free fraction not bound to albumin — reach these neurons directly, with no transport mechanism required. It is the established primary site of action for amylin's satiating effect, and signals from it ascend to the adjacent nucleus of the solitary tract, to the lateral parabrachial nucleus, and onward to hypothalamic feeding circuitry.

For cagrilintide specifically, two studies published within a year of each other tell a more interesting story than either alone, and the second revises the first.

The first used genetically modified mice (Carvas et al., 2025). Animals lacking RAMP1 and RAMP3 — and therefore lacking AMY1R and AMY3R while retaining the calcitonin receptor — were compared with wild-type littermates after twenty-three weeks on a high-fat diet, over a three-week treatment period with cagrilintide at 3 nmol/kg subcutaneously once daily.

Three things follow. Cagrilintide's weight effect depends on the amylin receptors rather than on the calcitonin receptor alone — salmon calcitonin, acting on the same family, moved weight in the opposite direction in the same experiment. Neuronal activation in the area postrema is largely RAMP-dependent, though the fifty-seven per cent reduction is a halving and not an abolition. And the study was funded in part by an investigator-led industry consortium grant, which is stated here because it should be.

The second study went after the area postrema hypothesis directly, and it did not survive contact (Ludwig et al., 2026). Chemogenetically activating the relevant area postrema neurons in rats failed to affect long-term food intake or body weight. Knocking down Prlh — the gene for prolactin-releasing peptide — in the dorsal vagal complex abolished cagrilintide's effects. And it did so selectively: semaglutide's effects in the same animals were unaffected.

Everything in this section is rodent work. Cagrilintide reduces body weight in people, and the anatomy of the area postrema is human anatomy, but the receptor and circuit attributions rest on knockout, knockdown and chemogenetic experiments that can only be done in animals. No cagrilintide brain-distribution study, autoradiography or human central-nervous-system data exists.

11A phasic hormone given as a tonic drug

This is the section the rest of the document has been building toward, and it is the one where the honest answer is that nobody knows.

Amylin, as Section 03 described, is a meal-associated signal. It is released with insulin when food arrives, acts over the following hour or two, and clears. Its half-life of roughly thirteen minutes is not an inefficiency to be engineered around; it is a specification. The hormone is a report that a meal is in progress, and a report that never stops is not the same report.

Pramlintide honours that design. It is given before meals, acts for about an hour, and disappears. Whatever else can be said about a drug requiring three injections a day, it is pharmacologically faithful to the hormone it imitates.

Cagrilintide is not. A weekly injection holding plasma concentrations within a narrow band means the amylin receptors of the area postrema are occupied continuously — during meals, between meals, overnight, and on days when the person eats nothing at all. The transformation is from a signal to a set-point.

Is that better, worse, or simply different? The arguments run both ways and neither is settled.

For: continuous occupancy is what produces continuous appetite suppression, and the clinical results are the results of the tonic drug, not of a phasic one. Whatever the theoretical objection, the weight comes off.

Against, or at least unresolved: receptor systems exposed continuously to an agonist frequently adapt, and the mouse data above contain a signal consistent with adaptation — food intake fell for days, not for weeks. Whether the human food-intake effect similarly attenuates while weight continues to fall, and what happens over years rather than months, has not been established for this molecule. No monotherapy trial has run long enough to find out.

The question is not academic. It is the difference between a drug that works by restoring a physiological signal and one that works by overriding a regulatory system, and the two would be expected to behave differently on withdrawal, over long exposure, and in combination with other appetite-acting agents. The evidence to distinguish them does not yet exist.

12The size of the evidence base

Before any individual result, the shape of the whole — because the shape changed sharply in 2025 and much of the commentary has not caught up.

A systematic review published in 2024 searched for every randomised trial in which people received cagrilintide alone or in fixed combination with semaglutide. From 678 screened articles it found three randomised trials enrolling 430 people in total (Dutta et al., 2024). One studied the molecule on its own, at phase 2; the other two studied it in combination, and one of those was a phase 1b trial with twenty-four people in its comparator arm. That was the entire randomised literature at the point the combination entered phase 3.

Then REDEFINE 1 reported (Garvey et al., 2025), and it turns out to matter for the single agent in a way its coverage did not convey. The trial randomised 3,417 adults in a 21:3:3:7 ratio to the combination, to semaglutide alone, to cagrilintide alone, or to placebo. Three hundred and two people received cagrilintide 2.4 mg as monotherapy for sixty-eight weeks, inside a phase 3 protocol, against a placebo arm of 705 and an active arm of 302 in the same trial.

13The phase 2 dose-finding trial

The phase 2 trial established the dose–response (Lau et al., 2021; NCT03856047). It ran at fifty-seven sites in ten countries and enrolled adults without diabetes who had a body-mass index of at least 30, or at least 27 with hypertension or dyslipidaemia. Participants were randomised 6:1 to weekly subcutaneous cagrilintide at one of five doses, to daily liraglutide 3.0 mg, or to volume-matched placebo, for twenty-six weeks including up to six weeks of dose escalation, followed by six weeks off treatment.

On the trial-product estimand, weight reductions were greater at every dose than on placebo: 6.0 per cent at 0.3 mg rising to 10.8 per cent at 4.5 mg (6.4 to 11.5 kg), against 3.0 per cent (3.3 kg) on placebo, with treatment differences of 3.0 to 7.8 percentage points, all P < 0.001. The active comparator, liraglutide 3.0 mg daily, produced 9.0 per cent (9.6 kg), and cagrilintide at its top dose beat it by 1.8 percentage points (P = 0.03).

Gastrointestinal adverse events occurred in 41 to 63 per cent of cagrilintide recipients against 32 per cent on placebo, nausea in 20 to 47 per cent against 18 per cent. Ten per cent of participants discontinued treatment permanently, four per cent for adverse events, at similar rates across groups. The trial was funded by Novo Nordisk.

What it was not designed to answer matters as much as what it found. It was powered for weight change and tolerability over twenty-six weeks. It had no cardiovascular endpoint, nothing capable of showing whether the weight returns, and no body-composition measurement adequate to establish what tissue was lost. Those are not failings of a dose-finding trial; they are a description of what it leaves open.

14The phase 3 monotherapy arm, and what it settles

REDEFINE 1 ran all four arms under one protocol for sixty-eight weeks, which makes its internal comparisons the most informative human data that exist on this molecule. On the treatment-policy estimand — the trial's own coprimary analysis, which counts everyone regardless of adherence:

All figures treatment-policy estimand, P < 0.001 for each comparison against placebo. The same trial's trial-product analysis gives −22.7 per cent for the combination against −2.3 per cent on placebo. Note that 20.4 appears twice in the primary report as two different quantities — the treatment-policy weight change, and the trial-product between-group difference — so any sentence quoting it has to name the estimand.

Two things are settled by that table which nothing before it could settle.

Cagrilintide alone is less effective than semaglutide alone. The pooled phase 2 comparison had been null with 98 per cent heterogeneity across two small trials — neither equivalence nor inferiority (Dutta et al., 2024). Under one protocol, at scale, over sixty-eight weeks, the amylin analogue produced 11.5 per cent against the GLP‑1 agonist's 14.9. It is a real and substantial drug; it is not the stronger of the two.

The combination is sub-additive. Against placebo, cagrilintide alone is worth 8.5 percentage points and semaglutide alone 11.9. Simple addition predicts 20.4 points over placebo; the combination delivers 17.4. The amylin mechanism therefore adds real and substantial effect — five and a half percentage points on top of semaglutide — while the pair together retain about 85 per cent of the sum of their separate effects. The two overlap somewhere, which is exactly what Section 10's circuit evidence would predict for two drugs converging on the same behaviour by different routes.

And one result that goes the other way

In people with type 2 diabetes the ranking reverses. A 32-week phase 2 trial randomised ninety-two participants on metformin, with or without an SGLT2 inhibitor, to CagriSema, semaglutide or cagrilintide, all escalated to 2.4 mg (Frias et al., 2023; NCT04982575). Mean weight change was −8.1 per cent with cagrilintide against −5.1 per cent with semaglutide and −15.6 per cent with the combination.

That is the opposite ordering to REDEFINE 1, and the honest reading is that the two trials are not measuring the same thing: different populations (diabetes against no diabetes), very different sizes (thirty per arm against three hundred), and different durations. The large, long, single-protocol trial in people without diabetes is the stronger evidence, and it should be given precedence — but the diabetes result is not explained away by that, and it leaves open the possibility that the relative standing of the two mechanisms depends on the population.

The same trial is unambiguous about glycaemia, and there the amylin analogue is clearly the weaker agent: HbA1c fell 0.9 percentage points with cagrilintide against 1.8 with semaglutide and 2.2 with the combination. Whatever cagrilintide is doing to body weight, it is not primarily a glucose-lowering drug.

15CagriSema is a different product

The combination pairs cagrilintide 2.4 mg with semaglutide 2.4 mg in one weekly injection. The mechanistic rationale is that the two act at different receptors on partly separate neuronal populations and should therefore add rather than overlap. Section 14 shows they mostly do, and not entirely.

REDEFINE 1's coprimary result was −20.4 per cent against −3.0 per cent on placebo at week 68 on the treatment-policy estimand (estimated difference 17.3 percentage points, 95% CI 16.6 to 18.1, P < 0.001), from 3,417 randomised participants. Gastrointestinal adverse events affected 79.6 per cent of the combination group against 39.9 per cent on placebo, mainly transient and mild to moderate; 6 per cent discontinued for adverse events against 3.7 per cent on placebo.

The programme also produced a result that has been less widely repeated. In an open-label head-to-head against tirzepatide 15 mg over eighty-four weeks, with 809 people randomised and a mean baseline weight of 114.2 kg, the combination reached 23.0 per cent against tirzepatide's 25.5 per cent on the efficacy estimand, and 20.2 against 23.6 per cent on the treatment-regimen estimand. The primary endpoint, which was non-inferiority to tirzepatide, was not met.

Both statements about that trial are true and should be held together. Twenty-three per cent weight loss is a very large effect. And the trial the sponsor designed to show the combination was not worse than the leading competitor did not show it. For a document about the amylin analogue the consequence is specific: a two-drug combination that does not beat a single molecule from another class raises a question about how much the second mechanism is worth in practice, and Section 14's arithmetic — sub-additive, roughly seven-tenths of the standalone effect retained — is the closest thing to an answer that currently exists.

16Tolerability, including two findings in the drug's favour

The adverse-effect profile of amylin-receptor agonism is gastrointestinal and dose-related. Nausea is the characteristic effect, appears during dose escalation, and diminishes with continued exposure. It is also what bounds the dose, and it has ended earlier programmes in this class.

In the pooled phase 2 data, treatment-emergent adverse events, serious adverse events, injection-site reactions, nervous-system disorders, nausea and diarrhoea did not differ significantly between cagrilintide 2.4 mg and the GLP‑1 comparators over 26 to 32 weeks, nor between the combination and semaglutide alone over 20 to 32 weeks (Dutta et al., 2024). Gastrointestinal events and vomiting were significantly more frequent with the combination than with semaglutide alone.

Vomiting was significantly less frequent with cagrilintide monotherapy than with semaglutide or liraglutide. If that holds at scale it is clinically meaningful, because gastrointestinal intolerance is a leading reason people stop taking drugs in this class. It rests on small numbers and should be read as a signal rather than an established difference. Against it, the 32-week diabetes trial reported adverse events in 80 per cent of the cagrilintide arm against 71 per cent on semaglutide — small numbers again, and in the other direction.

The second favourable finding is firmer, because it comes from the phase 3 arm. Cagrilintide alone lowered systolic blood pressure by 5.2 mmHg and diastolic by 2.9 mmHg at week 68, against 2.8 and 1.7 on placebo (Verma et al., 2026). Whether that is an independent effect or a consequence of the weight loss is not established by those data.

Two safety questions that often attach to peptides in this class have been answered for cagrilintide and answered negatively, which is worth recording because negative safety findings are rarely repeated. A dedicated thorough QT study found no clinically relevant prolongation of the corrected QT interval (Gabe et al., 2024). And a clinical pharmacokinetic study found that renal or hepatic impairment does not meaningfully affect the compound's pharmacokinetics, safety or tolerability (Nielsen et al., 2026) — which is what the albumin-bound disposition described in Section 06 would predict, and a useful confirmation that the prediction holds.

Injection-site reactions deserve caution rather than reassurance. The pooled relative risks were 1.90 (95% CI 1.00 to 3.62) and 3.51 (95% CI 0.85 to 14.46) across the two comparisons — intervals wide enough to be compatible with no effect and with a threefold increase. The point estimates should not be quoted without them.

Against placebo in phase 2, cagrilintide lowered very-low-density lipoprotein cholesterol by 0.12 mmol/L (95% CI 0.03 to 0.21, P = 0.01) and triglycerides by 0.31 mmol/L (95% CI 0.06 to 0.56, P = 0.01).

17What is still not known

A reader encountering this compound through the research-use market will find it described in terms borrowed from the combination's phase 3 headline. Those terms overstate it by nearly a factor of two, and the correct number — 11.5 per cent over sixty-eight weeks, in 302 people, against 3.0 per cent on placebo — is a good deal more interesting for being real.

18Where the molecule actually stands

Two statements, and the difference between them is the whole of this section.

Cagrilintide, as a single agent, is investigational and is approved nowhere in the world. There is no marketing authorisation for it in any jurisdiction and no regulatory submission for it as a monotherapy — despite sixty-eight weeks of phase 3 exposure in three hundred people, and two dedicated phase 3 monotherapy trials now running.

CagriSema, the fixed combination with semaglutide, is under regulatory review. Novo Nordisk filed a New Drug Application with the United States Food and Drug Administration on 18 December 2025 on the strength of the REDEFINE programme, and a decision is expected in the latter part of 2026. As of the compilation date of this document the combination is not approved in the United States or the European Union.

Material sold outside clinical supply — in the research-use market, by the milligram, under this compound's name or the combination's — is not a regulated pharmaceutical product. It has not been assessed by any medicines regulator for identity, purity, potency, sterility or endotoxin content, and nothing in the published record described in this document attaches to it.

19The field the molecule created

A reliable indicator of a compound's standing is whether its competitors measure themselves against it. By that test cagrilintide has been consequential out of proportion to its own development programme: it is the comparator in the published development papers of the analogues now following it.

The names are not incidental. All five carry the -lintide stem, which the international naming convention reserves for amylin derivatives and mimics. The shared stem is why the literature on any one of them contaminates the literature on the others, and it is a problem this document had to solve mechanically before it could be written — see the Apparatus.

The strategic question the field is testing is not whether amylin agonism reduces body weight. It does, by about nine and a half percentage points over placebo across sixty-eight weeks. The question is whether that is enough to carry a product, or whether amylin agonism earns its place as the second component of a combination. Cagrilintide's own development is now hedged across both answers: the large phase 3 programme is of the combination, while RENEW 1 and RENEW 2 test the single agent against placebo. Petrelintide is being developed as a standalone from the outset, which is what makes the two programmes worth watching against each other.

20What would change the picture

The open questions have specific experimental shapes, and naming them is more useful than listing hopes.

Body composition, measured properly, in people. The claim that amylin agonism spares lean mass better than incretin agonism is mechanistically plausible and, in mice, supported. In humans it is unestablished for this molecule. Given that muscle loss is the central live controversy about this whole drug class, it is the single most conspicuous missing measurement, and REDEFINE 1's four-arm design was the obvious opportunity to make it.

Whether the diabetes reversal is real. Section 14 records a genuine conflict: cagrilintide outperformed semaglutide on weight in a thirty-person diabetes trial and underperformed it in a three-hundred-person trial without diabetes. Those are compatible — populations differ, and so do the endpoints that matter in them — but they are not reconciled, and a properly sized trial in type 2 diabetes would reconcile them.

Whether the tonic signal behaves over years like the phasic one. Section 11 sets out why continuous amylin-receptor occupancy is not the same intervention as restoring a meal-associated hormone. The relevant evidence is long-term: receptor regulation, whether the food-intake effect that disappears within days in mice attenuates similarly in people, and whether weight loss is maintained on continued exposure. Sixty-eight weeks is the longest exposure reported, and no off-treatment follow-up of the monotherapy arm has been published.

The vomiting signal, in numbers large enough to trust. Cagrilintide monotherapy produced significantly less vomiting than semaglutide or liraglutide in pooled small trials, and a higher overall adverse-event rate than semaglutide in another. Both rest on tens of people. The REDEFINE 1 monotherapy arm and the two RENEW trials have the numbers to settle it.

An outcome trial of the single agent. None exists. A cardiovascular outcome trial is running — REDEFINE 3, 7,101 participants with established cardiovascular disease, primary completion estimated September 2027 — but it tests the combination, so a positive result will not separate the contribution of the amylin analogue from that of a GLP‑1 agonist already known to reduce cardiovascular events. Every claim in this document concerns weight, glycaemia, blood pressure and laboratory measures.

21References

Generated from verified NCBI records rather than from recall. Author lists, journal names, volumes, pages and identifiers were re-fetched from PubMed for every entry below, and the returned title of each was read back and confirmed to be the paper intended. This series has twice shipped reference lists drafted from memory in which identifiers pointed at real but unrelated papers, and the build refuses to run if any identifier fails to resolve.

Sources with no PubMed record — the trial-registry entries, the chemical and pharmacological database records, the naming-council statement and the company announcements — are listed separately after the indexed references and are cited to the instrument itself rather than to any summary of it.

Sources without a PubMed record

22How this document was assembled

The evidence base was built against project 05, the Therapeutic Peptide Research Library, and against NCBI directly. Every file carrying a document extension in the project’s two full-text stores was opened and its extracted text searched: 33,200 files, returning 51 assets that named the compound at all.

The reading corpus is 1,607 unique documents, roughly 32,253 printed-page equivalents, keyed by identifier across the local stores and a PubMed Central harvest and counted once. Of those, 121 name cagrilintide substantively, 5 name only the combination, and 1,456 are the amylin, receptor-pharmacology and neuroanatomy literature the compound sits inside.

Why the corpus is so much larger than the compound’s own literature

Because the compound’s own literature is small. The PubMed arm that searches for cagrilintide and its development codes returns 150 records — 114 for the molecule and 62 for the combination. That is the smallest core surface of any compound in this series. Harvesting only those would have produced a document with nothing to say about the forty years of amylin biology that produced the molecule, so two further arms were harvested and are reported separately: a lineage arm of 3,720 records covering amylin, pramlintide and the other analogues, and a pharmacology arm of 2,801 covering the calcitonin receptor, the RAMPs and the area postrema. The union is 6,288 records.

A separate PubMed Central body-text sweep for the compound returned 589 articles, 484 of them invisible to the indexed search because they name cagrilintide only inside their Results or Methods. Retrieval returned 2,589 full texts; a substantive-use screen on their own text retained 1,582 and discarded the rest as passing mentions, reference-list-only occurrences, or documents with no retrievable body.

Two identity problems, and what each one cost

The development code belongs to another drug. The internal designation for this molecule, AM833, was used decades earlier by a different company for fleroxacin, a fluoroquinolone antibacterial — and PubMed’s automatic term mapping expands the string to that drug. In this harvest, 27 records matched the bare code and nothing else. All 27 were rejected, and 26 of them were fluoroquinolone papers. A case-sensitive, word-boundaried match — which is everything the identity rule normally requires — still admits every one of them, because the string genuinely is present and genuinely is bounded. Only corroboration by subject matter separates them.

The naming stem belongs to four other real drugs. The suffix -lintide is reserved by international convention for amylin analogues, and it is shared with pramlintide, davalintide, petrelintide and eloralintide. Pramlintide is cagrilintide’s own backbone, so every subject-matter term that corroborates one corroborates the others and corroboration cannot separate them; only the discriminating token can. In the project’s library index a plain %lintide% prefilter proposes 298 passages and the boundaried matcher admits 91 — 69.5% of what the stem proposes is a different molecule.

A merged corpus figure must be keyed by identifier and never summed. 20 documents were present in both stores; adding the two page totals would have reported 32,854 printed-page equivalents against a true 32,253.

The figures, and why none of the supplied artwork is in this document

Five commissioned plates were supplied against a caption list describing six. Every printed value on them was checked against the primary literature, the deposited structural coordinates and the trial registry, as the house standard requires. All five were withheld, and every figure in this monograph is authored from verified values instead.

Three of the five failures were detectable without opening a single source, by internal arithmetic alone. The structure plate drew 36 residue positions for a molecule it labelled as having 37, omitting position 36, and named the C-terminal substitution backwards — a claim its own printed molecular formula refutes, since C₁₉₄H₃₁₂N₅₄O₅₉S₂ is achievable only with a proline at that position. The pharmacokinetic plate’s axis was labelled logarithmic and had decade spacings of 167, 148 and 143 pixels, placing every data point on a different scale from its own ticks. The trial plate’s bars were plotted on a scale sixteen per cent larger than the axis they shared an origin with, so that the highest-dose bar ran past the axis maximum.

The other two required the literature. The receptor plate printed four potency values that appear in no primary source, sit one to two and a half log units below every published measurement, and rank the calcitonin receptor last — which no dataset does, and which contradicts the explanatory note printed beneath them on the same plate. The anatomy plate drew the hypothalamic arcuate nucleus inside the cerebellum. Separately, four of the six trial values on the evidence plate matched neither of the source trial’s two estimands, and its placebo arm was missing entirely.

The standard distinguishes a discrepancy, which is captioned in place, from a defect, which is withheld — the test being whether the wrong thing is the figure’s payload. A sequence diagram whose residues are wrong cannot be repaired by a note, because a reader reads the residues. The full verification record, value by value, is held with the project artwork.

23Evidence handling

Findings are labelled by the kind of study that produced them, in the sentence that reports them. Randomised human trials, single-arm human studies, knockout and chemogenetic animal experiments, cultured-cell measurements, structural determinations and narrative reviews are different kinds of claim, and the difference is stated rather than left to the reader to infer. Animal and in-vitro results are never phrased so as to imply a human outcome, and the species is named every time. Every mechanistic attribution in Part Three is rodent work, and says so.

Two products are kept strictly separate, and this is the most important editorial rule in the document. Cagrilintide is a single molecule. CagriSema is a fixed combination of that molecule with semaglutide 2.4 mg, which produces roughly fifteen per cent weight loss on its own. Almost all the recent literature and effectively all the public attention concern the combination, and its headline figure is close to twice the amylin analogue’s. No combination result is quoted anywhere in this document as evidence about the single agent. Where the two appear in the same sentence, the sentence says which is which.

Several pivotal papers are cited but were not read in full. The discovery paper, the phase 2 monotherapy trial, the phase 1b combination trial, the phase 2 diabetes trial and both phase 3 reports appear in journals that deposit no open-access full text, so none of them has a PubMed Central identifier and none is in the reading corpus. Their figures here come from the structured abstract of the primary report, from the trial registry record, or — for the individual arms of the phase 3 trial — from an open-access secondary analysis of the same trial. This is stated rather than allowed to hide behind the corpus figure, because a corpus number that implied the pivotal papers had been read in full would be a claim about coverage this document does not have.

Recency is weighted but not blindly. The 2025 and 2026 literature is treated as current on trial results, on regulatory status and on receptor pharmacology, where it genuinely supersedes what came before — a 2026 circuit study overturned the standard account of where this drug acts, and this document follows it. It is not treated as current where a recent review merely restates an old measurement: the thirteen-minute half-life for native amylin is repeated across the modern literature as a human value, and it is a rodent one.

Where evidence conflicts, both sides are given with the reason one does or does not supersede the other. Two conflicts here are live and are presented as live: whether cagrilintide outperforms or underperforms a GLP‑1 agonist as monotherapy, where a small trial in type 2 diabetes and a large one without diabetes point in opposite directions; and whether the area postrema or the nucleus of the solitary tract carries the sustained effect. Where a widely repeated claim is not supported by the primary record — that the C-terminal substitution runs from proline to tyrosine, that pramlintide is weak at the calcitonin receptor, that the molecule is stalled at phase 2 — it is named as unsupported rather than quietly omitted.