Accepted answer
Neither lab is wrong. Chromatographic purity is not a property of a vial; it is a property of a vial as measured by a specified method, and your two labs specified different methods. The whole 1.3 points is accountable, and once you can do the accounting you can predict the direction of a disagreement from the method tables alone.
The two methods side by side
| Parameter | Lab A (reported 98.1%) | Lab B (reported 99.4%) |
|---|
| Column | C18, 150 x 4.6 mm, 3.5 um, 100 A pore | C18, 250 x 4.6 mm, 5 um, 300 A pore |
| Temperature | 40 C | 25 C |
| Mobile phase A | 0.1% TFA in water | 0.1% formic acid in water |
| Mobile phase B | 0.085% TFA in acetonitrile | 0.1% formic acid in acetonitrile |
| Gradient | 25 to 55% B over 60 min | 30 to 60% B over 30 min |
| Gradient slope | 0.50 %B/min | 1.00 %B/min |
| Flow rate | 1.0 mL/min | 1.0 mL/min |
| Detection | 214 nm | 220 nm |
| Mass on column | 20 ug | 60 ug |
| Reporting (disregard) threshold | 0.05% | 0.10% |
| Integration of the front shoulder | perpendicular drop | tangent skim |
| Main peak tailing factor | 1.2 | 1.6 |
| Resolution, main vs nearest impurity | 1.9 | 1.0 |
| Reported purity | 98.1% | 99.4% |
Accounting for the 1.3 points
Estimated contribution of each difference, in points of reported purity:
| Cause | Points | Direction and mechanism |
|---|
| Gradient slope, 0.50 vs 1.00 %B/min | 0.55 | Lab A resolves shoulders that merge into the main peak for Lab B |
| Reporting threshold, 0.05% vs 0.10% | 0.42 | Roughly six small peaks between 0.05 and 0.10% that Lab B is entitled to discard |
| Integration, perpendicular drop vs tangent skim | 0.18 | Tangent skim assigns part of the shoulder to the main peak |
| Detection, 214 vs 220 nm | 0.12 | Lower relative response for less-conjugated impurities at 220 nm |
| Mass on column, 20 vs 60 ug | 0.05 | Overload broadening and fronting hides small neighbours |
| Total | 1.32 | matches the observed 1.3 |
Gradient slope is the big one, and it works in a direction people find counterintuitive
A shallower gradient gives higher resolving power. Higher resolving power finds more impurities. Finding more impurities gives a lower purity number. So the better method reports the worse result, and a vendor who wants a high number runs a fast gradient.
Quantitatively, with the half-height resolution formula R = 1.18 (t2 - t1) / (W1 + W2): on Lab B's 30-minute gradient the main peak elutes at 18.42 min with a half-height width of 0.29 min, and the nearest impurity at 18.97 min with 0.31 min.
R = 1.18 x 0.55 / (0.29 + 0.31) = 0.649 / 0.60 = 1.08
At R = 1.08 there is a shared valley, and the shoulder is partly inside the main peak whatever integration you choose. On Lab A's 60-minute gradient the same pair separates to a gap of about 0.94 min with widths of 0.32 and 0.34:
R = 1.18 x 0.94 / (0.32 + 0.34) = 1.109 / 0.66 = 1.68
Baseline resolved. The shoulder integrates as its own peak and appears in the impurity table. Nothing about the vial changed; a peak moved from inside the main peak to outside it.
Reporting threshold is the second biggest and is pure convention
Every method sets a disregard limit below which peaks are not counted. 0.05% and 0.10% are both entirely conventional. In a peptide with a busy baseline there are frequently five to ten peaks in that window, and whether they are summed into the impurity total is a decision made once, in a document, years before your sample arrived. Lab B is not cheating; Lab B has a 0.10% threshold and is applying it.
Tailing and integration
Lab B's tailing factor of 1.6 against Lab A's 1.2 does quiet damage. A tailing main peak spreads into the region where late-eluting impurities live, so any small peak riding on that tail must be skimmed rather than baseline-separated, and tangent skim systematically assigns tail area to the main peak. Lab B's tailing is probably the 25 degree column temperature plus formic acid rather than TFA, so several of these differences share one root cause.
So is the cynical reading correct?
Largely, yes. A purity figure with no method attached is uninterpretable, and the space of defensible methods is wide enough to move a peptide's apparent purity by well over a point without anyone doing anything improper. A vendor selecting a 20-minute gradient, 0.10% threshold, 220 nm detection and generous integration will publish a genuinely higher number than a vendor running a pharmacopoeial-style method, on identical material.
What follows practically:
- Compare within a lab, never across labs. If you are ranking vendors, all the samples must go to the same service on the same method. Janoshik, Medutest and PeptideMeter each run their own conditions consistently, so their internal comparisons are valid and cross-service comparisons at the tenth-of-a-point level are not.
- Ask for the method, or at minimum the gradient time and the reporting threshold. Those two account for most of any gap you will see.
- Treat 98 and 99 as the same answer. The difference between 98.1 and 99.4 is method choice. The difference between 99 and 92 is the material. Calibrate your attention accordingly.
- Prefer the lower number when a lab reports one. A lab reporting 98.1 on a vial where another reports 99.4 is more likely to be the one resolving more, and the one you learn more from.
edited 30 Nov 2025 by Dr_Marek_Zielinski — added a caveat about sampling
6The better method reporting the worse number is the single most important sentence for anyone reading vendor COAs. – helena_vidmar 8 months ago 7Would add that Lab B loading 60 ug on a 4.6 mm column is on the edge of mass overload for a gradient peptide method. – bac_or_bust 10 months ago 4The disregard-limit contribution surprised me. Six peaks at 0.07% each really does add up to nearly half a point. – Dr_Colm_Fitzhenry 2 months ago add a comment