The reproducible spectrum: the single most useful thing a report can carry

What the pharmacopoeias require here is instructive, because the requirement is not onerous. A compendial identity test states the technique, the reference against which the article is compared, and the acceptance criterion. That is the whole architecture: technique, reference, criterion. A certificate reading “LC-MS: conforms” supplies a technique in the loosest possible sense, no reference, and no criterion. It is not a false statement. It is a statement with no content.

Which source a laboratory chooses, and why it should say so

The choice between the two techniques is not a matter of quality but of question. A synthesis chemist watching a coupling proceed wants a fast, salt-tolerant check that the chain has grown by the expected residue, and MALDI on a bench instrument answers that in minutes. An analytical laboratory asked whether a submitted vial contains the labelled compound and nothing closely related to it needs the resolving power and the accuracy that electrospray into a high-field analyser provides, coupled to a chromatographic separation so that species which co-elute can at least be assigned to retention times.

Both appear in this market, and reports rarely distinguish them. That matters because the two techniques have different blind spots. MALDI can induce loss of labile modifications during desorption, so a phosphorylated or otherwise fragile species may be under-represented. Electrospray suppresses ionisation of some analytes in the presence of others, so a minor component of a mixture may be absent from a spectrum in which it is genuinely present.

A certificate stating the source therefore tells a reader which class of error to consider. The Journal has stopped asking suppliers for more testing and started asking them for this line instead, on the grounds that it costs nothing and changes what the existing test can be said to support.

Mass accuracy, and what a tolerance ought to be

Mass accuracy is the difference between the measured mass and the true mass, expressed in parts per million of the measured value. It depends on calibration, on the stability of the instrument’s electronics and temperature, on the number of ions arriving at the detector, and on whether an internal calibrant was co-analysed with the sample. It is not a fixed property of an instrument; it is a property of a measurement made on an instrument on a particular day.

Certificates in this market seldom state a tolerance at all. Where they do, the figure is usually expressed in daltons rather than parts per million and is generous: ±0.5 or ±1.0 dalton is common, which at incretin molecular weights corresponds to 120 to 240 parts per million and is achievable on almost any instrument sold in the last thirty years. A tolerance that no plausible measurement could fail is not an acceptance criterion. It is a formality.

What a meaningful criterion looks like is not mysterious. State the theoretical mass and its convention, state the observed mass, state the deviation in parts per million, and state the limit above which the result would have been reported as non-conforming. Four numbers, all of them already known to the analyst. The compendial framework for validating an analytical procedure asks for exactly this kind of specificity about what a test can discriminate, and the framework predates this market by decades.¹

Sequence coverage as a reported number

Where a peptide map is performed, the headline output is a coverage figure: the percentage of residues in the expected sequence accounted for by identified fragments. Ninety-five per cent coverage sounds close to complete and is worth interrogating, because the five per cent that is missing is not randomly located. Very short fragments elute in the solvent front and are lost. Very hydrophobic fragments retain on the column. Regions between closely spaced cleavage sites produce peptides too small to identify unambiguously.

The consequence is that the uncovered fraction tends to sit in the same places for a given protease and a given sequence, which means a laboratory reporting ninety-five per cent coverage in run after run has ninety-five per cent coverage of a specific ninety-five per cent. A second digest with a different enzyme is the conventional remedy, and a report that used two orthogonal proteases is doing something a report using one cannot.

For a reader assessing a document, the useful questions are which enzyme, what coverage, and whether the uncovered residues are identified. A map that names the missing stretch has told you where the residual uncertainty lives. A map that reports a percentage alone has told you a number whose meaning depends on information it withheld — which is, in a different guise, the same complaint this department makes about purity figures reported without a gradient.

Six lines that would make an identity claim checkable

The Journal has settled on a short list, arrived at by writing to laboratories and asking what they could supply without additional work. Six lines. The ionisation source and mode. The analyser, named by class at minimum and by model preferably. The theoretical mass, with the convention stated as monoisotopic or average. The observed mass, with the charge state from which it was derived. The deviation, expressed in parts per million. And the acceptance criterion that was applied.

Every one of those is in front of the analyst at the moment the report is generated. None is commercially sensitive. Together they convert a verdict into a measurement, because they allow a reader to determine what the test could have detected and what it could not. A document carrying those six lines can be assessed by somebody who has never seen the sample; a document reading “MS: conforms” cannot be assessed at all, by anybody, including the person who wrote it.

The compendial approach to identity testing is built on the same three elements — a technique, a reference, and a criterion — and asks for them to be stated because a test whose discriminating power is undocumented has not been validated in any meaningful sense.² We are not asking this market to become a regulated one. We are asking it to print what it already knows.

How an identity result travels once it leaves the laboratory

Follow a mass spectrum through the market and its meaning changes at every step. A laboratory issues a report to whoever submitted the sample, stating what was observed on a named instrument on a named date. The submitter — a vendor, in most cases — extracts a figure and a verdict onto a certificate of analysis for the lot. A reseller reproduces the certificate, or a portion of it. A listing page distils the whole chain into a phrase: identity verified.

Nothing dishonest need happen at any step for the final phrase to support far more than the original report does. The instrument’s resolving power is lost at step two. The convention behind the theoretical mass is lost at step two or three. The date, the batch and the submitter’s identity survive unevenly. By the time the claim reaches a buyer it has become a property of the product rather than a record of a measurement on one vial from one lot on one day.

This is the structural reason the Journal reports identity claims by asking for the underlying laboratory report rather than the certificate. When a supplier supplies it, the claim usually holds up and often turns out to be stronger than the certificate suggested. When a supplier cannot locate it, that is itself information about how far back the documentary chain reaches, and we report that too, without inferring anything about the material.

The uncomfortable conclusion of all this is not that identity testing in this market is worthless. It is that identity testing here is doing considerably less work than the language attached to it suggests, and that the shortfall is documentary rather than analytical. The instruments are capable. The laboratories are competent. What is missing is six lines on a page, and the reason they are missing is that nobody has ever declined a purchase for want of them.