Disregard limits: what a regulated laboratory does with a tiny peak

The least discussed and most consequential stage of a purity determination happens after the chemistry is over. A chromatogram is a series of detector readings against time, and turning it into a set of peak areas requires software to decide where a peak begins, where it ends, where the baseline underneath it lies, and whether a given fluctuation is a peak at all. Those decisions are governed by parameters an analyst sets, and reasonable analysts set them differently. The purity figure is downstream of every one of them.

Gradient slope, expressed properly

A gradient should be quoted as a rate, not as a duration. Twenty-five to forty-five per cent acetonitrile over forty minutes is half a percentage point of organic per minute. The same range in twelve minutes is about one and two-thirds points per minute. That threefold difference in slope is the difference between resolving a deamidated relative from its parent and delivering both as one peak.

The underlying relationship is well established in peptide chromatography: resolution of closely related species improves as gradient slope decreases, up to the point where peak broadening from extended run times starts to give the gain back. Peak capacity — the number of peaks a method can theoretically resolve across its run — rises with shallower gradients and with more efficient columns, and it is the honest single-number summary of what a separation can do. It is never quoted in this trade.

The Journal’s standing request is simply that the gradient be printed. It is three numbers: starting composition, ending composition, time. Nobody regards it as commercially sensitive, every laboratory has it in the method file, and its presence converts a purity figure from an assertion into something comparable with the next certificate. Its absence is the reason two figures from two suppliers cannot be placed side by side, and that absence is a documentary decision rather than a technical constraint.¹

The threshold, and what falls below it

Integration software applies a threshold — expressed as a slope sensitivity, an area cut-off, a height cut-off or some combination — below which a feature in the trace is treated as baseline noise and not integrated. The setting is necessary: without it, every fluctuation would be reported as an impurity and the result would be dominated by noise. The setting is also consequential, because a great many real, small, closely related species live in the region between the two conventional choices.

The arithmetic is easy to underestimate. A well-made peptide preparation may carry twenty or thirty related species each between two-hundredths and a tenth of one per cent — deletion sequences, deamidated and oxidised forms, epimers. Reported individually against a low threshold they might total a percentage point or more. Discarded against a high threshold they total zero. Two laboratories reporting 99.4 and 98.3 on the same lot may have measured the same chromatogram and disagreed only about which features are noise.

The Journal has asked all four independent services what threshold their standard peptide report uses. Two answered with a figure. One answered that it depends on the method and offered to supply the value per report, which is a better answer than a fixed number. One did not answer. We regard the threshold as second only to the gradient in importance and, like the gradient, it is a single value that whoever produced the document already knows.

Baselines, solvent fronts and what counts as total

Under a tailing peak, the area depends on where the baseline is drawn, and the software offers several conventions — a straight line between valley points, an exponential skim, a tangential skim for a shoulder. For a symmetrical peak on a flat baseline the differences are trivial. For a small impurity riding on the tail of a large parent they are not, and they are the commonest genuine disagreement between two competent analysts looking at the same trace.

The denominator is decided by the solvent-front convention. Everything eluting in the first moments of a gradient run — unretained salts, injection solvent, dissolved gases, mobile-phase impurities — is conventionally excluded, and rightly so. Methods differ on where the exclusion window ends, and the choice determines whether an early-eluting hydrophilic fragment is an impurity or a non-event. The difference between a window closing at one and a quarter minutes and one closing at two and a half is not analytical laxity; it is a judgement about the separation, and it moves the number.

Manual reintegration deserves a word, because it has an undeserved reputation. Reintegrating by hand is entirely legitimate — software misassigns peak boundaries regularly — and in regulated practice it is permitted, documented and audited, with the original and revised integrations retained. What is not acceptable is undeclared reintegration, and in this market there is no mechanism by which it would ever be declared.²

Where two figures on the same lot disagree, the useful first question is not which laboratory is right but which gradient was shorter. In every disagreement this desk has been able to decompose, the faster method returned the higher number, and the difference was reconstructible from the methods alone.

Reporting, identification and qualification thresholds

Regulated pharmaceutical practice does not treat all small peaks alike. It defines three thresholds. A reporting threshold is the level above which an impurity must be listed in the results. An identification threshold is the level above which its structure must be established. A qualification threshold is the level above which its biological safety must be addressed. The three are set by dose and product class, and the framework converts an argument about small peaks into a documented decision procedure.

The trade has no equivalent. There is no reporting threshold, so an impurity is listed or not according to the software settings; no identification threshold, so nothing is ever identified; and no qualification threshold, because there is no regulatory obligation to qualify anything in a research chemical. The practical consequence is that a certificate reporting 98.6 per cent purity conveys nothing about whether the missing 1.4 per cent is thirty innocuous deletion sequences or one substantial unidentified species.

This is the gap the Journal would most like to see narrowed, and it can be narrowed cheaply. Listing the three largest impurities with their relative retention times and areas costs nothing, requires no additional analysis, and transforms the informational content of the document. One of the twenty companies we track does it. It is not a coincidence that the same company answered every question we put to it about its analytical methods.³

What we submitted, and how we designed it

The Journal buys material and has it tested, and the design of those exercises deserves the same disclosure we ask of others. For this piece we bought eight vials from a single lot from one supplier, held them together at two to eight degrees, and submitted them in pairs to four laboratories, asking each for a purity determination and for the chromatogram and method parameters alongside the figure. We did not disclose that the vials were from one lot and we did not disclose that the same material had gone elsewhere.

Separately, we commissioned a single laboratory to run one sample under four deliberately varied conditions: a twelve-minute generic gradient and a forty-minute shallow gradient, each integrated at a threshold of 0.1 per cent and 0.05 per cent. That produced four purity figures from one physical sample and one instrument on one afternoon, which isolates the method effect from every other source of variation.

The limitations are ours to state. One lot from one supplier is not a survey of the market. Single injections carry the variability of single injections. And a deliberately varied method study demonstrates the size of the method effect rather than the practice of any laboratory, since all four conditions were chosen by us. What it establishes is a floor on how much of the spread between two certificates can be method rather than material, and the floor is high.⁴

Documentation practice, named

Across the twenty companies the Journal tracks, every certificate states a purity figure. Rather fewer state the method in any form. A minority name the detection wavelength. Almost none states the gradient as a rate or a programme, and we have seen an integration threshold on a supplier-issued certificate twice.

Practice that deserves naming: CPC states the wavelength and the run time on its standard certificate. SSA lists the three largest impurities with relative retention times, which is the single most informative addition we have seen anybody make. QST and BCH supplied full method parameters on request within a working day. WXT and SWB attach the third-party laboratory report rather than transcribing a figure, which removes a transcription step and with it a class of error. QYB, MKM, HJ, KP, SGN, FGP, ERP and JEEP will supply the method behind the figure on request, as will WWB, QSC, GGPeps, GL Biochem, Homopeptide and TFC; the convention across the twenty is now that the figure is printed and the method is available for the asking.

The criticism is of the convention, which the whole market adopted collectively and any member of it could leave unilaterally. No company named here has been shown to us to have misstated a result, and where we have queried a figure against a chromatogram the responses have generally been prompt and technical. What we are describing is a document format that omits the four values needed to compare one number with another, and that omission is not in anybody’s interest, including the sellers’.

The economics deserve restating, because they explain the convention rather than excusing it. A generic gradient occupies an instrument for under an hour and returns a number the same week. Every one of the determinations that would qualify that number costs several times as much and takes several times as long.

Dissolution, pH and what happens before the injection

A peptide in a vial exists in whatever state the manufacturer left it. Reconstituted in water, a hydrophobic sequence may not dissolve completely and the chromatogram will show particles or aggregates. Reconstituted in an acidic buffer, the same sequence dissolves and the chromatogram shows monomer. The purity figure—and the mass balance—changes accordingly. Sample preparation is not one of the twelve values that belong on a method disclosure, and yet it is one of the most consequential, because it determines what population the peptide is actually in when the injection happens.⁵

The trade addresses this by assuming that samples are dissolved in the mobile phase or its aqueous component, and the assumption is sometimes true and sometimes false. A certificate that states how the sample was dissolved, at what concentration, in what solvent and after what incubation time, is one that can be repeated. A certificate that does not is one that will produce different results if the receiving laboratory uses different dissolution practice, which is particularly consequential for poorly soluble sequences.

The technique is not on trial here and never was. Reversed-phase chromatography can resolve species differing by a single methyl group, and the laboratories running it for this market are, on the evidence we have gathered, largely competent and entirely willing to describe what they did when somebody asks. What is on trial is a document format that omits the four values needed to compare one figure with another.