The number at the bottom of the certificate

In February we bought six vials of the same lot of a research-grade semaglutide from a single supplier, opened none of them, and posted one each to four independent testing services. The vials came from the same box, carried the same batch number, and had been stored together since arrival. If purity is a property of the material, four laboratories analysing four vials from one lot should return four closely similar numbers.

They returned 99.1%, 98.4%, 97.2% and 96.6%.

The spread is 2.5 percentage points, which in a market where suppliers advertise 99% and buyers reject 97% is the difference between a sale and a refund. None of the four laboratories made an error. Each answered the question it was asked, using a method it disclosed, and each answer is defensible. What the exercise demonstrates is that "purity" is not a property of a vial in the way that mass is. It is the output of a method, and the method is a set of choices.

What the instrument actually measures

Reversed-phase high-performance liquid chromatography separates a mixture by passing it through a column packed with a hydrophobic stationary phase while the composition of the mobile phase changes over time. Compounds that interact more strongly with the stationary phase elute later. A detector at the column outlet — almost always an ultraviolet absorbance detector in this application — produces a trace of signal against time, and the software integrates the area under each peak.

The purity figure is then, in almost every case, the area of the main peak expressed as a percentage of the total integrated area. That definition contains four hidden decisions, and every one of them was made differently by at least one of our four laboratories.¹

Decision one: the gradient

How quickly the mobile phase composition changes determines how well adjacent species separate. A shallow gradient — say 25% to 45% acetonitrile over forty minutes — will resolve impurities that a steep gradient runs together with the parent peak. Two of our four laboratories used gradients of roughly that shallowness. One used a twenty-minute run. The fourth used a twelve-minute generic peptide method.

The twelve-minute method returned the highest purity figure. This is not a coincidence and it is not dishonesty: a fast method genuinely cannot see what a slow one can, and a laboratory running high sample volumes at low cost has a legitimate commercial reason to use one. The buyer, however, is entitled to know which they paid for.

Decision two: the detection wavelength

Peptide bonds absorb strongly at around 214 nm; aromatic side chains absorb at 280 nm. A method reading at 214 nm sees essentially every peptide-bonded species in the sample, including small fragments. A method reading at 280 nm sees only species containing tyrosine, tryptophan or phenylalanine, and will be blind to fragments that lack them.

Three of our laboratories reported at 214 nm; one at 220 nm. The difference between those two is modest. The difference between either and 280 nm would not have been, and 280 nm methods do circulate in this trade.

Decision three: the integration threshold

Chromatographic software does not integrate every wobble in the baseline. It applies a threshold, below which a feature is treated as noise. Set the threshold at 0.1% of the main peak area and a great many small impurities disappear from the calculation; set it at 0.02% and they appear, each subtracting from the reported purity.

Only two of the four reports we received stated a threshold. They were 0.05% and 0.1%. The laboratory using 0.1% reported a purity 0.7 percentage points higher than the one using 0.05% on samples from the same lot, which is roughly the magnitude one would predict from the number of small peaks visible in the two chromatograms.

Decision four: what counts as "total"

The denominator is the total integrated peak area — but which peaks are included? Solvent-front features, injection artefacts and mobile-phase impurities are conventionally excluded, and reasonably so. The convention differs on where the front ends. One of our reports excluded everything eluting before 2.5 minutes; another excluded everything before 1.2 minutes. A truncated peptide fragment eluting at 1.8 minutes is an impurity in the second method and does not exist in the first.

Read the table in the other direction and it becomes a price list. The 99.1% came back in four days for a low three-figure sum. The 96.6% took three weeks, cost roughly four times as much, and came with two chromatograms and a written interpretation. The trade quotes the first number and pays for the first service.

What the four numbers agree on

It would be a mistake to conclude from this that purity testing is arbitrary. The four laboratories agreed on a great deal. All four found a single dominant peak at the expected retention time relative to their own reference. All four found the same two largest impurities, in the same order of magnitude. None found evidence of a substituted or unrelated compound. None found a total impurity burden above 3.5%.

In other words, the material was what it claimed to be, and it was reasonably pure by any of the four methods. The disagreement was entirely about the second decimal place of a number that the trade treats as though it were the first significant figure.

Why the trade settled on one number

Purity became the industry’s single figure of merit for a straightforward reason: it is cheap to obtain, easy to compare, and it sounds like a quality statement. Peptide content — the fraction of the vial’s mass that is actually the peptide, as opposed to counter-ions, residual solvent and water — is a more useful number for anyone doing arithmetic with a vial, and it costs substantially more to determine.² Sterility is a different discipline entirely. Identity confirmation by mass spectrometry is routine in a regulated laboratory and optional here.

The result is a market that competes on the one measurement that is easiest to flatter. A supplier wanting a higher number does not have to make a better product; it can run a faster gradient.

What we think should change

Three things, none of which require a regulator. Certificates should state the gradient, the wavelength and the integration threshold, because all three are known to whoever produced the document and none is commercially sensitive. Purity figures should be reported to one decimal place at most, because a single injection on a generic method does not support two. And any supplier quoting a figure it did not generate should say which laboratory did, and on what lot.

Two of the twenty companies in the Journal’s dossier programme already do all three. One of them, having been shown a chromatogram that disagreed with its own, wrote back describing its integration convention in enough detail that the disagreement resolved arithmetically. That correspondence took a working day and settled a question that would otherwise have become a dispute about honesty.

The wider point is the one this department keeps returning to. A measurement is a claim about a procedure. Publish the procedure and the claim can be checked; withhold it and the number is a decoration, however many decimal places it carries.