What does the mechanism of semaglutide predict that STEP 5 did not test?

Conditions: semaglutide · STEP 5.

This is one of those things that everyone repeats and nobody derives.

This matters practically, not just academically, because it changes what I would do next.

Is the standard explanation correct, and if so, what is the evidence for it?

Start from the receptor and the rest follows: which receptors, in what ratio, with what signalling bias, reached at what concentration.

Receptor desensitisation as a plateau mechanism is plausible and poorly evidenced. GLP-1R internalises on agonist binding and recycles, and biased agonists that internalise less have been argued to sustain signalling better. Whether any of that operates at the timescale of a four-month clinical plateau — against the much simpler explanation that energy expenditure fell with mass — is not established.

More usefully, orforglipron is not a peptide at all, which changes everything downstream: it is orally bioavailable without an absorption enhancer, it has no fasting or water requirement of the same kind, it does not require cold chain, and it cannot be assayed by any of the peptide methods discussed elsewhere on this site.

I would separate what is established structurally from what is inferred clinically. The chemistry is settled; the attribution of clinical effect to specific receptor arms mostly is not.

The mechanism is settled enough to be useful and unsettled enough to be interesting, which is a reasonable place for a field to be.

Dose equivalence across agents with different receptor profiles is not a defensible concept, and the attempt to construct it is the most common analytical error in this area.

Comparing a 2.4 mg dose of one agonist to a 15 mg dose of another tells you nothing, because the molar potencies at their respective receptors differ, the receptor profiles differ, and the exposure per milligram differs. The only defensible comparison is between clinical outcomes in trials with comparable populations and durations, which is why SURMOUNT-5 exists and why indirect comparisons should be read sceptically.

Concretely, oral bioavailability of a 4 kDa peptide is essentially zero without help. Oral semaglutide is co-formulated with sodium N-(8-[2-hydroxybenzoyl]amino)caprylate, which raises local gastric pH and transiently increases transcellular permeability in a small area of gastric mucosa. It works, and it delivers roughly one per cent of the dose, which is why the oral tablet strengths are an order of magnitude above the injectable and why fasting and water volume are not optional details.

The structural basis of semaglutide’s pharmacokinetics — Aib-8, the Arg34Lys substitution and the C18 diacid–AEEA linker at Lys26 — is described in the original medicinal chemistry publication, and it is worth reading once because it makes the design logic explicit[1].

The limitation here is that almost all of the human mechanistic work is in the licensed agents, so mechanistic claims about the investigational tri-agonists rest on animal and early-phase data.

The chemistry is the interesting part and it is also the well-documented part. Read the medicinal chemistry papers; they are short and they explain the design.

Put another way, GLP-1R is a class B G-protein-coupled receptor signalling predominantly through Gs and cyclic AMP, and most of the interesting pharmacology in this class is about where that signalling happens rather than how hard it is driven.

Amylin co-agonism adds to a GLP-1 effect rather than duplicating it because the two act through different circuits: amylin signals through the area postrema via calcitonin receptor complexes, GLP-1 through both the area postrema and the arcuate nucleus. Two non-redundant satiety signals summate, which is the design rationale for a co-formulation rather than a higher dose of either.

The caveat is that mechanism explains and does not predict. A clean mechanistic story has repeatedly failed to survive a Phase 3 in metabolic medicine.

If you want to reason about a new agent, start from its receptor profile and its half-life. Almost everything else follows.

On the detail: the half-life is a formulation achievement rather than an intrinsic property. Native GLP-1 has a plasma half-life of a couple of minutes; everything in this class is a set of modifications engineered to defeat that.

The Aib substitution at position 8 replaces alanine with α-aminoisobutyric acid, which is sterically hindered enough that dipeptidyl peptidase-4 cannot cleave the N-terminal dipeptide. That single change takes the half-life from minutes to hours. The C18 diacid on a linker at Lys26 then binds albumin reversibly, which both shields the molecule from renal filtration and creates a depot that releases slowly — taking hours to about a week.

Tirzepatide’s imbalanced receptor pharmacology, with greater potency at GIPR than at GLP-1R, is characterised in its pharmacology publication and is the starting point for any mechanistic discussion of the agent[1].

Worth noting that receptor pharmacology measured in a transfected cell line is a starting point, not a physiological measurement, and potency ratios do not transfer cleanly in vivo.

Do not convert doses between agents. There is no exchange rate, and constructing one is how people arrive at an order-of-magnitude error.

edited 10 Dec 2024 by RP_C18 — added a caveat about sampling