Lot sampling and retains: what a CoA result is representative of

A purity figure on a certificate of analysis is a measurement of something. The question that determines what the figure is worth is what that something was: how many vials were drawn from the lot, at what points in the filling sequence, whether they were pooled before injection, and whether the material tested was the finished vial at all or the bulk powder it was filled from. None of those are reported on a typical research-grade CoA. The number appears with a method, a date, and a lot code, and the reader is left to supply the connection between the result and the vial in front of them. That connection is a sampling plan, and it is an inference the document does not make. This post sets out where sampling sits in a release workflow, what homogeneity assumptions it rests on, how retain samples change what can be reconstructed later, and where the boundary falls between a lot-level result and a vial-level claim.

The sample is the unit of measurement, not the lot

Analytical chemistry measures what is placed in the instrument. A reversed-phase run reports the composition of the solution loaded onto the column; a Karl Fischer titration reports the water content of the aliquot weighed into the vessel. Everything beyond that — the statement that the lot is 99.1% pure, that the lot contains 2.1% residual moisture — is an extrapolation from the aliquot to the population it was drawn from, and the strength of that extrapolation is entirely a property of how the aliquot was selected.

This is uncontroversial in principle and routinely elided in practice. A CoA that reads “Purity (RP-HPLC): 99.2%” reports a chromatogram. Whether that chromatogram describes the lot depends on whether the injected solution was representative of the lot, which depends on the sampling plan. A single vial pulled from the front of a filling run and a composite drawn across the run can both produce a number that looks identical on the page and means something materially different.

The distinction matters most where the analyte is not uniformly distributed. Chromatographic purity of a well-mixed bulk powder is, in most cases, genuinely close to uniform — the impurity profile was established during synthesis and purification, before the material was ever divided. Attributes established during or after filling behave differently. Fill weight, residual moisture, subvisible particulate counts, and container closure integrity are all properties of individual units, and a result from one unit constrains the others only through a statistical argument that has to be made explicitly.

Bulk testing and finished-unit testing answer different questions

Much of what appears on a research-grade CoA is bulk data. The synthesis house characterizes the purified peptide — purity, identity by mass spectrometry, peptide content by amino acid analysis or nitrogen determination, counterion, residual solvents — and issues a document describing that bulk. A downstream party that fills the bulk into vials may reissue or pass along that document, sometimes with the filler’s own lot code applied.

The bulk figures remain accurate descriptions of the bulk. What they cannot describe is anything the filling and lyophilization steps introduced or changed. Residual moisture in a finished vial is a function of the lyophilization cycle and the closure’s seating, not of the bulk powder’s water content. Particulate counts reflect the fill line, the stopper, and the vial. Oxidation-sensitive residues can pick up measurable change during a long cycle under an imperfect vacuum. A CoA carrying bulk purity alongside a finished-lot code describes the material at an earlier point in its history than the code implies.

Reading a CoA usefully therefore begins with a question the document often does not answer directly: at what stage was this tested? Some documents make it plain, giving a bulk lot and a fill lot separately, or labelling sections “bulk release” and “finished product.” Many do not, and the only available signal is which attributes appear. A panel consisting of purity, identity, content, counterion, and residual solvents is a bulk panel. Finished-unit testing adds moisture, appearance after reconstitution, fill weight or content uniformity, and — in a sterile context — sterility and endotoxin.

Homogeneity is an assumption that has to earn its place

Sampling plans in a pharmaceutical setting are built around a homogeneity argument. If the material is demonstrably uniform, a small number of units characterizes the whole; if it is not, the sample count and the positions sampled have to rise until the plan captures the variation. The argument is normally made once, during process validation, and then relied on for routine release.

For lyophilized peptides, the sources of non-uniformity are reasonably well understood. Shelf position in the lyophilizer produces a heat-transfer gradient: vials at the edge of a shelf receive radiant heat from the chamber wall that centre vials do not, which shortens primary drying and can leave edge and centre vials at different residual moisture. Fill-volume drift across a run produces content variation. Stopper seating varies unit to unit. Each of these is a reason that a defensible plan samples across the run — beginning, middle, end — and across shelf positions, rather than taking a convenience sample.

Where a CoA reports a single value for a unit-level attribute with no indication of how many units contributed, the reader cannot distinguish between a plan that sampled nine vials across three shelf positions and one that sampled the first vial off the line. Both produce one number. Only one of them supports a statement about the lot.

Composite sampling sits between these. Pooling aliquots from multiple units and running the pool once gives a genuine lot average at a fraction of the analytical cost, and for attributes where the mean is the quantity of interest it is a reasonable design. It also destroys the variance. A composite of ten vials, one of which is substantially out of family, returns a mean that looks acceptable and reports nothing about the outlier. Where uniformity is the attribute in question, compositing is the wrong instrument regardless of how many units feed it.

What retains make reconstructable

A retain — a reserved sample from a released lot, stored under the labelled conditions for a defined period — is the mechanism by which a question raised after release can be answered against the material as released rather than against memory. In a regulated finished-drug context, retention requirements are specified: US 21 CFR 211.170 sets the reserve sample at twice the quantity needed for all specification tests other than sterility and pyrogens, held one year past the labelled expiry. The equivalent practice for research-grade material is a matter of supplier policy, and policies vary from a formal retain programme to nothing at all.

The practical consequence shows up when a result is disputed. If a buyer’s independent testing disagrees with the CoA, there are several candidate explanations: the two laboratories used different methods, the material changed in transit or storage, the wrong material was tested, or the original result was wrong. Distinguishing among these requires material of known provenance from the original lot, tested alongside the disputed sample. Without a retain, the comparison is between one laboratory’s current result and another laboratory’s historical report on material no longer available, and the disagreement is generally unresolvable.

Retains also carry the stability record. A retain stored under labelled conditions and pulled at intervals is the observation that supports a retest date. A retest date assigned without any such observation is an assumption carried forward from a related product or from general expectations about the compound class. Both appear on documents as a date.

When evaluating a supplier’s analytical documentation, the presence of a retain programme is therefore informative beyond its stated purpose. It indicates that lots are tracked as discrete populations, that material is stored under controlled conditions past release, and that the organization has a mechanism for investigating a result rather than only for producing one.

From lot result to vial: what the inference requires

Connecting a CoA figure to a specific vial requires several links, each of which can be absent without the document showing a gap.

The lot code on the vial has to match the lot the document describes, and the document has to describe the finished lot rather than an upstream bulk. Where a filler applies a new code to material from a supplier’s bulk, the CoA may reference either, and the relationship between the two codes is internal to the filler.

The attribute has to be one that transfers from the lot to the unit. Chromatographic purity generally does, because it was established before division. Moisture, fill content, and particulate burden are unit properties, and transfer only under the homogeneity argument the sampling plan was built to support.

The storage history between release testing and the vial’s arrival has to be short enough or controlled enough that the released figure still describes the material. Release data is a measurement at a point in time. A lot released at 99.2% purity and held eighteen months at ambient temperature is not a 99.2% lot at the point of use, and no document reissues itself to say so.

None of this makes a CoA uninformative. It makes it a description of a defined thing — a measurement, by a stated method, on a sample of stated or unstated provenance, at a stated date. The figure is a fact about that measurement. Everything the reader wants to conclude about the vial is an inference from it, and the sampling plan is the part of the argument the document usually leaves out. Knowing which questions the plan would have to answer is what separates reading a certificate from reading a number on one.

The sample is the unit of measurement, not the lot

Bulk testing and finished-unit testing answer different questions

Homogeneity is an assumption that has to earn its place

What retains make reconstructable

From lot result to vial: what the inference requires

Further reading