The note without the incretin

Every large weight-loss result of the last decade has had a GLP-1 receptor somewhere in it. Semaglutide is GLP-1 alone; tirzepatide adds GIP; retatrutide adds glucagon on top. Even the amylin combinations that followed kept a GLP-1 agonist in the regimen. In December 2025 The Lancet published a 48-week trial in which the top dose of a once-weekly injection took off a mean 20% of body weight, and the GLP-1 receptor was nowhere in the molecule (Lancet 2025). Clinical

The drug is eloralintide, an analog of amylin, the hormone the beta cell releases in the same granule as insulin and that almost nobody names. This piece is about how the molecule is built, because the engineering is the argument: each edit to the 37-residue chain answers a specific failure of the native hormone, and one of those edits, receptor selectivity, is the reason the result is read as a verdict on amylin rather than on a blend. The trials come after, with their caveats intact.

Every claim is tagged by where it comes from: a public registry or regulatory record, human trial data, or animal and cell work. The tags grade provenance, not confidence.

Reference

A public registry record, regulatory fact, or the molecule's own chemistry.

Clinical

Data from human participants — randomized and placebo-controlled unless stated.

Preclinical

Animal or cell-culture data; no human equivalent published.

Community

Forum-reported and anecdotal — signal, not evidence.

Amylin has no receptor gene of its own. It signals through the calcitonin receptor, CTR, after that receptor has been clamped to one of three accessory proteins, the RAMPs. CTR with RAMP1 is the amylin-1 receptor, AMY1; with RAMP2, AMY2; with RAMP3, AMY3. The same core protein, dressed three ways, listens to amylin; undressed, it listens to calcitonin, the thyroid hormone that tells osteoclasts to stop resorbing bone. Reference

That is the design problem in one sentence. A peptide shaped like amylin will fit the dressed receptor, but the undressed one is built from the same protein, so it will usually fit that too. The first long-acting amylin analog, cagrilintide, does: it is routinely described as a dual amylin and calcitonin receptor agonist. Eloralintide was engineered to tell the two apart. In cell lines expressing one receptor at a time, it was about 12-fold more potent at human AMY1 than at human CTR, and about 11-fold more potent at AMY1 than at AMY3 (Briere et al., Mol Metab 2025). Albumin binding, which the fatty-acid tail below exists to provide, did not erase that preference. Preclinical

Why spend a discovery programme on selectivity? Two reasons, and the honest one is less tidy. The tidy reason is pharmacology: CTR is expressed in bone, kidney and parts of the brain where an amylin drug has no business, and a molecule that leaves it alone has fewer places to cause trouble. The evidence reason is an aversion assay. Rats learn to avoid a flavour paired with a drug that makes them feel unwell, and eloralintide produced significantly less of that conditioned taste avoidance than cagrilintide at matched exposure. That is the entire experimental basis, so far, for the hope that a selective amylin agonist will be a gentler one. It is a rat result. Preclinical

Lilly has not published a labelled structure of eloralintide in the clinical papers. The public chemistry registry has one. PubChem’s record gives the formula C₂₀₁H₃₁₉N₄₉O₆₅S₂, a molecular weight of about 4,526 daltons, and a systematic name long enough to read the whole peptide from end to end (PubChem CID 175663130). Four things in that name are not in native amylin, and together they are the drug. Reference

• A ring that cannot be opened. Native amylin begins with a disulfide bond between cysteines 2 and 7, a six-residue loop that the receptor needs to see. Disulfides are fragile: they reduce, scramble and swap partners in a formulation and in plasma. Eloralintide keeps the loop but replaces the sulfur–sulfur bond with a methylene-thioacetal, S–CH₂–S, one carbon inserted between the two sulfurs. The registry name shows it as a 1,3-dithia ring. The loop keeps its shape and loses its chemistry.
• Three residues the ribosome cannot make. The name contains an α-methyl-phenylalanine, a phenylalanine with an extra methyl on the backbone carbon, and an N-methylated amide in the chain. Both are standard moves against the two things that kill a peptide: proteases, which cannot cut a backbone they cannot bind, and aggregation, which native human amylin is famous for. An extra methyl at the right position stiffens the local conformation and blocks the hydrogen bond a fibril would need.
• A lysine with a tail. A lysine side chain carries, through two γ-glutamate spacers, an icosanedioic acid: a C20 chain with a carboxylic acid at each end. The Lilly papers place the acylation at position 26, which is an isoleucine in the native sequence, so the lysine was introduced to hold the tail. The free acid on the far end is what albumin binds.
• The rest left alone. The ring, the methyls and the tail are edits to a scaffold that is still recognisably amylin: 37 residues, the same amidated tyrosine at the C-terminus the receptor requires, and the sequence motifs that give the AMY1 preference. Selectivity was not bolted on; it was found in the sequence and then protected by the chemistry around it.

Compare the two earlier amylin drugs. Pramlintide made exactly three substitutions, prolines borrowed from rat amylin, to stop the fibrils, and changed nothing else: it lasts about 48 minutes and is injected at every meal. Cagrilintide added a C20 diacid for albumin binding and reached a week. Eloralintide is the first to rebuild the disulfide, methylate the backbone and acylate the chain all at once, and the pharmacokinetics are where that shows. Reference

The 12-week phase 1 study is the one with the numbers. Dosed once a week without escalation, eloralintide reached peak plasma concentration three to five and a half days after injection and had a terminal half-life of 310 to 366 hours, which is 13 to 15 days. Exposure rose in proportion to dose, with dose-normalised ratios of 1.1 for area under the curve and 1.0 for peak concentration, which is what you want from a drug that will be titrated (Bhattachar et al., Diabetes Obes Metab 2026). Clinical

Two consequences follow from a half-life longer than the dosing interval. First, the drug accumulates: a weekly dose of something that lasts two weeks reaches steady state only after about five half-lives, roughly ten weeks, which is why every trial escalates slowly and why a “fast start” arm is a real pharmacological experiment rather than a convenience. Second, the signal never switches off between doses. Native amylin is a pulse released with each meal; eloralintide is a plateau. The hormone the body uses to say enough after eating is being delivered as a constant, and the trials below are, in part, a test of what a constant amylin signal does to a system built for pulses. Reference

There are three human studies of eloralintide alone and one of the combination. Taken in order, they read as a dose-finding programme that got the answer it was looking for, with two side effects it may not have expected. Clinical

• Single ascending dose, 48 healthy volunteers. Doses from 0.04 to 12 mg, once. Four weeks after a single 4 mg or 12 mg injection, mean weight was down 2.5% and 4.4%, against +0.6% on placebo. Sixteen adverse events in nine participants, fifteen of them mild. Two participants reported four gastrointestinal events between them. A single dose of a two-week drug producing measurable weight loss a month later is the half-life made visible.
• Twelve weeks of weekly dosing, 100 adults with obesity or overweight. Five dose cohorts, no escalation. Weight loss ranged from 2.6% to 11.3% at week 12. The most common events were decreased appetite (19%), headache (12%) and fatigue (11%); diarrhoea 10%, nausea 8%, vomiting 4%. One serious adverse event, judged unrelated. Only 29% of participants were women.
• Forty-eight weeks, 263 adults, phase 2. Seven arms: placebo, four fixed doses from 1 to 9 mg, and two escalation schedules. Mean weight change −9%, −12%, −18% and −20% at 1, 3, 6 and 9 mg; −20% and −16% on the 6→9 and 3→9 escalations; −0.4% on placebo. Baseline weight 109 kg, mean BMI 39, 78% female, 46 US centres, no diabetes. The paper's stated conclusion is 'clinically meaningful, dose-dependent reductions in bodyweight' and 'generally well tolerated'.

Now the side-effect table, which is where the tolerability story gets more interesting than the press release. Nausea in the phase 2 trial was 11% at 1 mg and 13% at 3 mg, close to placebo’s 14%, but 64% in the arm that started at 6 mg with no escalation, 33% at 9 mg, and 54% on the 6→9 schedule. Fatigue, barely mentioned in the GLP-1 literature, was 29% at 6 mg, 43% at 9 mg and 46% on 6→9, against 12% on placebo (Lancet 2025). Clinical

One more comparison is worth making carefully, because it is the one everyone makes carelessly. Cagrilintide’s own phase 2, in 2021, produced 10.8% weight loss at its top dose after 26 weeks (Lau et al., Lancet 2021). Eloralintide’s 20% came at 48 weeks, in a heavier population, with a different escalation. The two numbers are not a ranking of two drugs. They are evidence that the amylin axis, played alone, keeps delivering past the 26-week mark, which until 2025 was an open question. Clinical

If a selective amylin agonist loses 20% on its own, the obvious experiment is to add it to the best incretin. Lilly ran it in the harder population, adults with obesity and type 2 diabetes, where every class loses less weight. The phase 2b trial randomised 367 people across ten arms for 48 weeks and reported at EASD in Milan on 30 September 2026. It is a press release and a conference presentation; the paper has not been published (Lilly, September 2026, NCT06603571). Clinical

• The headline. Eloralintide 9 mg plus tirzepatide 15 mg: −23.3% body weight and −2.9 A1C points from a baseline of 8.1%. Tirzepatide 15 mg alone: −14.8% and −2.4. Eloralintide alone: −8.2%, −12.3% and −11.1% at 3, 6 and 9 mg. Placebo: −3.0%.
• The arithmetic. The combination beat tirzepatide by 8.5 points of weight, and eloralintide alone contributed 11 to 12 points in this population. Less than fully additive, which is what two satiety signals converging on the same hindbrain circuits would predict, and still the largest number reported in people with type 2 diabetes to date.
• The cost. Discontinuation for adverse events ran from 10.8% to 27.0% across the combination arms, against 2.9% on tirzepatide alone and 0% to 10.8% on eloralintide alone. The release attributes the events to escalation and promises an optimised schedule for phase 3. A quarter of an arm leaving a trial is the number to watch when the paper appears.
• The odd placebo. Placebo discontinuation was 16.7%, higher than most active arms, and placebo weight loss was 3.0%. Both are unusual and neither is explained in the release.

The design language matters here. Cagrilintide with semaglutide, the combination called CagriSema, is two pens; amycretin is the attempt to write GLP-1 and amylin into one peptide. EloraTZP is a third answer: two separate molecules, co-formulated into one injection, each free to keep its own pharmacokinetics. A two-week amylin agonist and a five-day dual incretin in the same syringe is a mixture, not a molecule, and Lilly says phase 3 of that co-formulation begins by the end of 2026. Reference

It is worth separating the claims, because they are being sold as one.

• Proved: the amylin axis can carry GLP-1-class weight loss alone. A 20% mean reduction at 48 weeks from a molecule with no incretin activity is the cleanest demonstration yet that satiety has more than one door. That was the question the amylin renaissance was built on, and it is now answered in 263 people.
• Proved: the chemistry holds. A thioacetal ring, backbone methylation and a C20 diacid on a 37-residue amylin scaffold give a two-week half-life with dose-proportional exposure and no loss of receptor preference. The molecule does what the registry record says it should.
• Not proved: that selectivity is why it is tolerable. The only comparison with a non-selective amylin agonist is a rat taste-aversion assay. The phase 2 nausea rates at flat 6 mg are not low. Fatigue is a new, dose-dependent signal with no mechanism offered.
• Not proved: durability, safety, and the pulse question. Forty-eight weeks is the longest exposure published. Amylin's native receptor family runs bone and calcium through calcitonin, and a drug 12-fold selective is not a drug that never touches CTR. Nothing in the record yet says what a years-long constant amylin signal does to a system that evolved for meal-sized pulses.