From cell to clinic: where the extrapolation stops being safe

Everything in this department is written on the assumption that mechanism is worth understanding. It is also worth being explicit about what mechanism cannot do, because the gap between what receptor pharmacology explains and what a reader wants explained is wide and rarely acknowledged. Receptor data predicts average effects reasonably well. It predicts almost nothing about which individual will lose twenty-five per cent of their body weight and which will lose four.

Biased agonism, stated carefully

When the GLP-1 receptor is activated it can couple to Gα_s, raising cyclic AMP, and it can recruit beta-arrestin, which contributes to receptor internalisation and desensitisation. An agonist that favours the first over the second is described as G-protein-biased. The therapeutic argument for bias is that sustained cAMP signalling without proportionate internalisation should produce a more durable effect at the same occupancy.

The evidence for that argument is real but narrower than its popularity suggests. Bias is measured in transfected cell systems at receptor densities that bear no relationship to a beta cell or a vagal afferent, and the translation from a bias factor in vitro to a clinical difference in vivo has been demonstrated convincingly for very few ligands.¹ The Journal’s position is that bias is a legitimate and probably important variable, that it is one of several plausible explanations for the differences observed between molecules, and that anybody presenting it as the explanation is ahead of the data.

The accumulation arithmetic, worked

For a drug given at a fixed interval, the accumulation ratio at steady state is approximately 1 / (1 − e^−kτ), where τ is the dosing interval and k is the elimination rate constant, itself 0.693 divided by the half-life. For a once-weekly drug with a seven-day half-life, τ and the half-life are equal, e^−kτ is 0.5, and the accumulation ratio is 2. In plain terms: at an unchanged weekly dose, average concentration at steady state is roughly twice what it is after the first injection.

Time to steady state depends only on the half-life, not on the dose or the interval: about 94% of steady state after four half-lives, 97% after five. For a seven-day half-life that is four to five weeks. This is why a fixed four-week escalation step exists at all — it is approximately the time required for the previous dose to stop increasing — and why escalating faster than that means escalating onto a still-rising exposure curve.

These are approximations that assume linear kinetics and complete absorption. Both assumptions are reasonable for this class and neither is exact.

The oral non-peptide agonists

An orally bioavailable small molecule that activates a class B GPCR was, for a long time, considered close to impossible. The current crop of non-peptide GLP-1 receptor agonists achieves it by binding a site that overlaps only partially with the peptide binding pocket, stabilising an active conformation without the two-domain capture mechanism.

Pharmacologically this matters for three reasons. Absorption does not depend on a permeation enhancer, so bioavailability is far less variable and far less dependent on fasting state than oral semaglutide’s. Elimination is hepatic rather than largely renal and proteolytic, which changes the interaction profile. And potency at the receptor is achieved without a fatty-acid albumin depot, so the concentration-time profile looks like a conventional small molecule rather than a peptide. None of this predicts efficacy; all of it predicts a different practical drug.

The expression map, tissue by tissue

Pancreatic beta cells: receptor activation potentiates glucose-dependent insulin secretion, which is why the class does not cause hypoglycaemia in the way sulfonylureas do — the effect requires elevated glucose. Alpha cells: suppression of glucagon secretion, also glucose-dependent. Gastric smooth muscle and enteric neurons: reduced antral motility and delayed emptying. Vagal afferents: signalling to the brainstem that contributes to satiety and to nausea.

Brainstem — area postrema and nucleus tractus solitarius: integration of peripheral satiety signals, and the site most plausibly responsible for nausea and vomiting. Hypothalamic arcuate nucleus: modulation of POMC and AgRP neuron activity, the classical appetite circuit. Cardiac atria: heart-rate increase of a few beats per minute, consistently observed and of uncertain clinical significance. Renal vasculature and tubule: effects on natriuresis and glomerular haemodynamics that are the most plausible mechanism for the renal outcome findings.²

The heart-rate signal

A resting heart-rate increase of roughly two to four beats per minute is one of the most reproducible findings in the class, observed across molecules, doses and populations. The mechanism is probably direct: GLP-1 receptors are expressed in the sinoatrial node region, and receptor activation has chronotropic effects in isolated preparations.

What it means clinically is unresolved. The cardiovascular outcome trials that reported the heart-rate increase also reported reductions in major adverse cardiovascular events, so whatever the chronotropic effect represents it is not overwhelming the benefit in the populations studied. That is a statement about trial populations and event rates, not a mechanistic reassurance, and the Journal reports it as such.

The response distribution nobody can explain

In the large obesity trials, mean weight reduction is reproducible to within a percentage point or two across programmes. The distribution around that mean is wide and consistent: a substantial minority of participants lose more than a quarter of their body weight, and a smaller but non-trivial group lose almost nothing. Reported non-response rates — usually defined as failing to reach 5% reduction — run to roughly one participant in seven to one in ten depending on the molecule and dose.

Nothing measurable at baseline has been shown to predict which group an individual falls into with useful accuracy. Receptor polymorphisms have been examined and explain little. Baseline BMI, sex, diabetes status and age shift the mean modestly and the variance barely at all. The honest summary is that this is the largest unexplained quantity in the field, and that any source claiming to predict individual response is claiming something the literature does not support.³

Desensitisation, and what it does and does not explain

Receptor internalisation following agonist binding is well established in vitro, and the popular inference is that "the receptors get used to it", explaining plateaus. The inference outruns the evidence in two ways. First, plateaus in the trials occur at around sixty to seventy weeks and coincide closely with the point at which reduced body mass lowers energy requirement enough to re-establish balance, which is a sufficient explanation without invoking receptor changes. Second, weight regain on withdrawal is rapid and near-complete, which is difficult to reconcile with a model in which the receptor has become unresponsive.

The tolerability tachyphylaxis discussed above — the attenuation of nausea and gastric delay over weeks at a fixed dose — is separately well supported. Two different phenomena share a name, and conflating them produces confident conclusions about plateaus that the data does not license.

Why any of this belongs in a general publication

An argument could be made that receptor pharmacology is a specialist concern and that readers need practical guidance instead. The Journal’s position is the opposite, for a specific reason: almost every piece of bad advice circulating about this drug class is a mechanistic error with a practical conclusion attached.

Escalating on a fixed calendar regardless of symptoms is an error about accumulation kinetics. Splitting a weekly dose into daily fractions to reduce side effects is an error about half-life and steady state. Assuming a molecule with GIP activity is simply a stronger version of one without is an error about selectivity. Expecting weight to keep falling indefinitely is an error about energy balance. In each case the practical advice is wrong because the mechanism was misunderstood, and in each case understanding the mechanism is not much harder than memorising the rule.

A note on sources

Everything above is drawn from the peer-reviewed pharmacology and clinical literature and from regulatory assessment reports, which are more informative than the papers on questions of dose selection and exposure. Where a claim rests on in-vitro work in transfected cells, this piece says so, because the translation of such work to human physiology has failed often enough in this field to deserve a standing caveat.

Where the Journal reports a trial number it states the estimand behind it, because the treatment-policy and trial-product estimands differ by two to three percentage points in the obesity programmes and the difference is routinely lost in secondary coverage. Nothing here is a recommendation, and none of the compounds discussed as research chemicals are approved for human use.

A short glossary, because the words are used loosely

Agonist: a ligand that binds a receptor and produces a response. Full agonist: one producing the maximal response the system permits. Partial agonist: one producing less than maximal response even at full occupancy. Analogue: a molecule structurally derived from a natural ligand. Mimetic: a molecule reproducing a natural ligand’s effect without structural derivation.

Orthosteric site: the binding site the natural ligand occupies. Allosteric site: a distinct site whose occupancy modulates activity at the orthosteric one. Biased agonism: preferential activation of one downstream pathway over another. Tachyphylaxis: diminishing response to repeated administration. Steady state: the condition in which the rate of drug entering the body equals the rate leaving it.

Precision here is not pedantry. Several of the arguments this publication receives by post turn out, on inspection, to be disagreements about which of these words the writer meant.

What has actually changed in the last three years

Three things, on the Journal’s assessment. First, the demonstration that a dual agonist could produce weight reduction approaching bariatric-surgical magnitude moved the field’s expectations, and with them the design of every subsequent programme. Second, the cardiovascular and renal outcome results reframed the class from metabolic-cosmetic to cardiometabolic, which changed reimbursement arguments far more than it changed prescribing.

Third, and least remarked, the pharmacology of oral administration became tractable. That is a manufacturing and access story as much as a scientific one: an oral small molecule has a completely different cost structure, cold-chain requirement and supply profile from an injectable peptide, and if it holds up in phase 3 it will do more to change who can get treated than any of the receptor science described above.

Two things follow practically from the pharmacology above, and only two. Consistency of dosing interval matters more than consistency of hour. And an interruption long enough to clear the drug is an interruption long enough to reset tolerability, which means resumption is a fresh escalation and not a continuation. Everything else in this piece is background.