It Isn’t Low. It’s Unstable.

By George ForrestTechnology

Four Phase III trials, more than nine thousand patients, and one drug discontinued after it met its endpoint. Oral peptide delivery has a headline figure everyone repeats. The more useful fact is how far that figure moves — and how much of it a glass of water can take away.

A patient takes a tablet with a glass of water. Nothing about that sentence sounds like a variable.

For most medicines it isn’t one. For the small group of peptide drugs that have been made to work by mouth, it is one of the largest variables in the whole system — larger, on the published numbers, than the difference between some of the formulations the field has spent three decades arguing about.

Those drugs deliver roughly one per cent of their dose, for reasons that are well understood and set out in the companion to this article: a digestive tract built to destroy peptides, an epithelium sized to exclude them, and a permeation enhancer that opens the barrier for about twenty minutes. This piece is about something else — not the size of the number, but its stability.

The number everyone quotes is an average

Oral semaglutide, the most successful product in the category by a distance, is reported at around 0.8% absolute bioavailability under label conditions. That figure is accurate, widely cited, and nearly useless on its own.

The same population pharmacokinetic analysis that gives 0.8% also gives the within-subject variability: 137%. Not between people — within the same person, from one dose to the next. A patient taking the identical tablet under the identical instruction does not reliably absorb the same amount twice.

That is not a footnote to the mean. It is a different fact about the system, and it has a different set of causes.

What a swallow of water costs

The most direct evidence comes from a human pharmacoscintigraphy study of oral semaglutide, in which twenty-six volunteers took the tablet with either 50 mL or 240 mL of water while the tablet itself was imaged.

  • The tablet took 85 minutes to erode completely with 50 mL of water, and 57 minutes with 240 mL.
  • Exposure — both AUC and peak concentration — was roughly 70% higher with the smaller volume. Longer erosion and later gastric emptying correlated significantly with higher exposure.

An extra 190 mL of water — less than a small glass — costs the patient the majority of their dose.

Oral calcitonin, taken through three Phase III trials before being abandoned, produced the same shape of result. Its developers reported two- to three-fold higher exposure at 50 mL of water than at 200 mL, and that dosing within four hours after a meal reduced plasma exposure by up to 74%. Their published list of lessons learned is, read plainly, a list of ways the dose can be lost between the glass and the gut.

These are not tolerability instructions. They are dose multipliers, printed on the patient leaflet rather than in the strength on the box.

The pattern repeats. An antisense oligonucleotide delivered with an enhancer averaged 9.5% bioavailability across four formulations, with an intra-subject range of 1.96% to 27.5% — a fourteen-fold spread. An oral basal insulin reported coefficients of variation of 82–132% on AUC and 85–163% on peak concentration, and lost around 65% of its exposure if food followed too quickly.

It does not travel between species either

The instability also sits between the animal model and the human being, and the clearest demonstration of that was published by the company with more experience of these formulations than anyone.

In 2025 a team at Novo Nordisk combined two permeation enhancers in a single tablet and tested it in beagle dogs and then in 192 healthy human subjects. In dogs the combination worked, improving exposure by 22% for a GLP-1 analogue and 51% for a PCSK9 inhibitor over the single-enhancer reference. In humans the AUC ratio against the same reference was 0.94, with a 95% confidence interval of 0.75 to 1.17. No difference. Their conclusion was that there is “a lack of translatability of the permeation enhancement capability of C10 in the stomach between dogs and human”.

A second study, published this year, added a pH modifier to protect the enhancer from stomach acid. In dogs, exposure rose 57%. In humans it again landed level with the reference, and the authors concluded that differences in gastric pH “cannot solely account for the lack of translation”.

Two well-resourced clinical studies, both negative, both reported openly. Negative results at that scale are expensive to generate and easy to leave in a drawer.

The variable nobody varies

There is a reason to think the instability and the container are the same story, and it comes from the most basic expression in the field. How much drug crosses an epithelium depends on three things: how permeable the barrier is, how steep the concentration difference is, and how much area the drug is crossing. Permeation enhancers act on the first. Dose acts on the second. Almost nothing in thirty years of published work has acted on the third.

The area term is not a theoretical nicety. When Novo Nordisk’s researchers mapped where oral semaglutide is actually absorbed — pyloric ligation in dogs, immunofluorescence of the gastric mucosa — they reported that absorption “takes place in the stomach, is confined to an area in close proximity to the tablet surface”. Their own image shows the peptide detectable directly beneath the tablet, on a two-millimetre scale bar.

Set that against the anatomy. The gastric mucosa is roughly 0.05 square metres — about 500 square centimetres. The small intestine, counting villi and microvilli, is about 30 square metres. The face of a tablet is on the order of one square centimetre.

A drug designed to be absorbed across the gut is, in practice, being absorbed across roughly a five-hundredth of the stomach — and a stomach is less than a five-hundredth of the small intestine.

That reframes the dosing-condition data. Water volume, meal timing and gastric emptying do not obviously change how permeable the epithelium is. What they change is where the tablet sits, how long it stays there, how fast it erodes, and how much of it is in contact with tissue rather than with fluid. They are variables of area and residence rather than of permeability — and they are the variables that move exposure by 70%.

Two honest qualifications. Permeability and area are not independent, so this is not a matter of multiplying one by the other and reading off an answer. And no published study in this class reports the contact area at all — which is rather the point. The field has thirty years of careful measurement on one term of the expression and none on another.

What instability costs

The commercial consequence is easier to see in one failure than in any argument about mechanism.

An oral basal insulin met its Phase II endpoint. Fasting plasma glucose was 7.1 mmol/L against 6.8 for injected insulin glargine, a difference of 0.3 at p = 0.46 — non-inferior over eight weeks, no severe adverse events, hypoglycaemia no worse than the comparator. The drug worked. It was discontinued anyway, and the stated reason was that “I338 doses were high and, therefore, production of the required quantities of I338 for wide public use was deemed not commercially viable”.

The manufacturing arithmetic was survivable on its own. What is not survivable is having to size every dose for the worst case, because the average is not what any individual patient gets. Low mean and high variance compound: one sets the cost of goods, the other sets the margin you must build on top of it.

Behind that sit four Phase III trials — one in oral heparin, two in osteoarthritis, one in osteoporosis — enrolling more than 9,100 patients between them, all of which missed their primary endpoints. The osteoporosis trial, at 4,665 patients over thirty-six months, reported that pharmacokinetic analysis “confirmed exposure” but that “the drug levels were markedly lower than expected”.

What would change if the field measured it this way

None of this is hidden. Every figure above is published, most of it in the primary clinical reports, and the people who generated it have catalogued the reasons at length — Maher and Brayden’s 2021 review is titled directly for the question of why these formulations show low efficacy in humans, and it sets out luminal dilution, the narrow permeability window, loss of contact between drug and enhancer, proteolysis, and the mass of enhancer a swallowable tablet can carry. None of the five has a settled answer in the published literature.

What has not followed is a change in how results are reported.

  • Bioavailability is still quoted as a single number. A mean without a coefficient of variation, in this category, describes almost nothing a formulator or a prescriber can act on.
  • Dosing conditions are treated as adherence instructions rather than as dose. On the numbers above, water volume and meal timing belong in the pharmacokinetics section, not the patient leaflet.
  • Animal models are still reported as though they forecast human exposure for this class. Two published human studies now say that, for one widely used enhancer in the stomach, they do not.
  • Contact area is not reported at all. A study that measures permeability to four significant figures and never records how much tissue the formulation touched has measured one half of the problem.

A category whose defining problem is variance should be powered, analysed and reported on variance. That is a change in convention rather than in chemistry, and it costs nothing to make.

Until it happens, the honest version of the headline figure is not “about one per cent”. It is “about one per cent, most of the time, depending on where the tablet landed and how much water went down with it”.

George Forrest is CEO and co-founder of Ibumix, a UK pharmaceutical company working on oral bio-enhancers. This article surveys published work; it does not describe Ibumix’s own.

Sources