Two documents, two disciplines: purity and sterility are not the same trade

We began looking at this seriously after a reader sent us a certificate with a query attached. The certificate reported 99.2 per cent purity for a lyophilised peptide, and the reader wanted to know whether that meant the remaining 0.8 per cent could be bacterial. It is a reasonable question and the answer is that the figure has nothing to do with bacteria in either direction. A vial containing a substantial microbial burden would very probably return a purity figure indistinguishable from a clean one, because the mass involved is far below the integration threshold of any method in commercial use.

Two questions that share a piece of paper

Consider what happens physically when a certificate is produced. A few milligrams of lyophilised powder are weighed, dissolved in an aqueous mobile phase with an organic modifier, injected onto a reversed-phase column and separated over a programmed gradient while an ultraviolet detector records absorbance. The output is a trace. Software integrates the areas beneath its features, and the main peak area as a proportion of the total becomes the purity figure.

Every step of that procedure is blind to microbial contamination. A bacterial cell contributes no ultraviolet-absorbing peak at any retention time a peptide method would record. Endotoxin, a lipopolysaccharide, is not usefully detected at the wavelengths used for peptide bond absorbance and would in any case be present at a mass fraction several orders of magnitude below any integration threshold in commercial use. A vial holding a hundred colony-forming units and a vial holding none produce chromatograms that no analyst could distinguish.

This is not a defect of the method. Reversed-phase chromatography is an excellent way of determining what proportion of the chromatographically visible material is the intended species, and that is what it is being asked. The defect is in the reading. A document answering one question is being filed as evidence about five.

Bioburden, and why the number before the filter matters

A sterilising filter is not an unlimited barrier. Its qualification is expressed as a retention capability under a defined challenge — conventionally a high titre of a small bacterial species per square centimetre of membrane — and its performance in use depends on the load presented to it. A bulk solution carrying a heavy microbial burden presents a filter with a harder problem than one carrying a light burden, and it presents a second problem the filter cannot address at all: the endotoxin released by organisms that die upstream passes through the membrane unimpeded.

Regulated manufacture therefore specifies a pre-filtration bioburden limit, tests against it on every batch, and treats an excursion as an investigation rather than a curiosity. The specification is usually expressed in colony-forming units per hundred millilitres, and a well-controlled process runs far below it.

Of the twenty companies the Journal wrote to, four stated that pre-filtration bioburden is determined on every batch, three stated that it is determined periodically, and the remainder did not answer the question. We regard that distribution as the single most informative result of the correspondence, because bioburden testing is inexpensive, is performed on the bulk rather than the finished container, and is the earliest point at which a problem is cheap to fix.

The filter, and the test that proves it survived

Sterilising-grade filtration through a membrane of 0.22 micron nominal rating is the operation on which aseptic processing rests. The membrane is qualified by challenge with a small bacterium at high concentration under the process conditions, and a filter that permits no passage under that challenge is accepted as sterilising for that fluid.

The critical practice is not the filtration but the integrity test that follows it. A membrane can be damaged during installation, during sterilisation, or by pressure excursions in use, and a damaged membrane looks exactly like an intact one. Bubble point, diffusive flow and pressure hold tests each detect a breach by measuring gas behaviour across a wetted membrane, and a post-use test is the only evidence that the filter was intact while the product was passing through it. A pre-use test alone establishes nothing about the state of the membrane at the end of the run.

This is the question the Journal has found most useful when assessing whether a fill operation is a real one. It is specific, it has a yes or no answer, the answer is recorded in the batch documentation as a matter of course, and it cannot be answered plausibly by anybody who is not actually running the process.

The statistics, worked in one paragraph

Suppose a batch of ten thousand vials in which one vial in a thousand is contaminated — a rate that would be a serious finding in any regulated operation and is invisible to any buyer. The probability that a single randomly chosen vial is clean is 0.999. The probability that all twenty sampled vials are clean is 0.999 raised to the twentieth power, which is approximately 0.980. The sterility test therefore passes this batch about ninety-eight times in a hundred.

Push the contamination rate up tenfold, to one vial in a hundred, and the test still passes the batch roughly eighty-two times in a hundred. To reach an even chance of detection at a one per cent contamination rate you would need to sample about seventy containers; to have a reasonable prospect of catching a one-in-a-thousand rate you would need to sample several hundred, which for most batches means testing a substantial fraction of the product.

This is not a criticism of the compendial test, which is designed as a final check against gross failure and performs that function. It is the reason no serious manufacturer treats a passed sterility test as the basis of the sterility claim, and the reason that a research supplier offering to have a vial sterility-tested on request is offering something considerably weaker than it sounds.

Three method families, and what each returns

The gel-clot method is the oldest and simplest: lysate is combined with the sample, incubated, and the tube inverted. A firm clot that does not slip is a positive. It is a limit test, and by testing serial dilutions it becomes semi-quantitative. Its virtues are robustness and independence from instrumentation; its limitation is resolution.

Turbidimetric methods read the increasing turbidity produced by clotting protein formation, either as an endpoint at fixed time or kinetically as the time to reach a defined turbidity. Chromogenic methods substitute a synthetic peptide substrate that releases a chromophore when cleaved by the activated enzyme, and read absorbance. The kinetic chromogenic variant — measuring the time to a defined absorbance change against a standard curve — is the method of record for most modern release testing, offering quantitation across several orders of magnitude from a small sample volume.

All three are compendial, all three require a demonstration that the sample matrix neither inhibits nor enhances the reaction, and all three are calibrated against an international reference endotoxin rather than against a mass. A result reported without the method and without the inhibition-enhancement result is, once again, a number without a procedure.¹

The endotoxin limit, calculated

The limit is derived, not looked up. For a parenteral product other than an intrathecal one, the threshold pyrogenic dose is taken as five endotoxin units per kilogram of body weight per hour. The endotoxin limit for the product is that figure divided by the maximum dose per kilogram administered within an hour. For an intrathecal route the threshold falls to 0.2 endotoxin units per kilogram, a twenty-five-fold reduction that reflects the absence of the systemic buffering the bloodstream provides.

Work it for a concrete case. A seventy-kilogram subject has an hourly allowance of three hundred and fifty endotoxin units. A vial nominally containing five milligrams of peptide, reconstituted to two millilitres, from which a fifth of a millilitre is drawn, delivers a tenth of the vial contents. If the whole vial carried three hundred and fifty endotoxin units, that draw would deliver thirty-five — a tenth of the allowance. The limit expressed per milligram of peptide is what a certificate should carry, because it is the only form of the number that survives a change in reconstitution volume.

None of the arithmetic is difficult. What is missing from this trade is not the calculation but the measured numerator.²

Low endotoxin recovery, and the clean result that is wrong

The most uncomfortable finding in endotoxin testing over the past fifteen years is that certain formulation matrices cause added endotoxin to become undetectable over time. Spike a known quantity of reference endotoxin into a solution containing a non-ionic surfactant and a chelating buffer, hold it, and the recoverable endotoxin declines — sometimes to a small fraction of what was added — while nothing has been removed. The lipopolysaccharide aggregate state has changed, and the assay cannot see what it cannot bind.

The phenomenon is called low endotoxin recovery, and it matters because the combination of polysorbate with citrate or phosphate is extremely common in peptide and protein formulations. A hold-time study — spiking the actual product matrix and measuring recovery across the intended sample storage interval — is the standard mitigation, and it is now expected as part of method suitability for products in that formulation space.

The implication for a reader is narrow but worth stating. A negative endotoxin result on a surfactant-containing formulation, reported without a hold-time recovery study, is weaker evidence than it appears. A negative result on a reconstituted lyophilised peptide in plain water for injection, tested promptly, is considerably stronger.³

What the cake tells you

A lyophilised plug is a physical record of the cycle that produced it. A well-formed cake occupies close to the volume of the solution that was frozen, has a matte, uniform appearance, retains a defined edge where it met the glass, and dissolves rapidly and completely on addition of diluent. That appearance indicates that the product was frozen below its critical formulation temperature and dried without the structure collapsing.

Departures are informative. A shrunken, translucent or glassy plug suggests collapse during primary drying — the ice sublimed faster than the amorphous matrix could hold its structure — which is associated with higher residual moisture and a shorter shelf life. A cake that has partly detached and moves freely suggests mechanical handling after drying, which is cosmetic. A cake showing melt-back at the base, or a visible meniscus line, suggests a shelf temperature excursion. Slow or incomplete reconstitution suggests either collapse or an unfavourable surface.

None of these observations is a purity finding, a sterility finding or an endotoxin finding, and it is important not to overclaim. What they are is the only direct evidence about process control that arrives in the box, and reading them costs nothing. The Journal photographs every cake it receives before opening the vial, for exactly that reason.⁴

Documentation practice, named and criticised

The Journal tracks the release documentation of twenty companies. On sterility and endotoxin the picture is close to uniform. Every one publishes a purity figure. A minority publish an identity confirmation. Two publish peptide content. On the microbiological attributes, the standard document is silent, and the silence is not annotated: there is no line reading that sterility has not been determined, which would at least be informative.

Some of the practices we would like to see are already in use somewhere in the group. SSA reports peptide content on its certificates, having begun after correspondence with this publication. CPC and SWB describe their fill environment in general terms on request. QST and BCH answered our five questions in full. WXT and FGP declined on the grounds that research-use products are not represented as sterile injectables, which is a legally sound answer that concedes the point of the exercise. Several others did not respond, and we record non-response as non-response rather than as evasion.

What we are criticising is a documentary convention, not the conduct of any company named here. None of the twenty has been shown to us to have misrepresented anything. The convention is that a chemistry certificate stands in for a release package, and it is a convention this trade adopted collectively and could abandon the same way.

What this article is not asserting

Precision about the scope of a criticism is part of the job, so it is worth spending a paragraph on what is not being said. This piece does not allege that any company named in it has sold contaminated material. It does not allege that any of them has concealed a result, falsified a document or misrepresented a test. It does not claim that the products discussed are dangerous, and it does not claim that they are safe, because neither claim is supportable from the evidence we have.

What the piece asserts is documentary. A certificate describing chemistry is being read as a release package covering microbiology. That mismatch is created by the format of the document rather than by anybody’s intent, and it is closed by adding lines rather than by changing behaviour.

There is also a legal point the Journal has no wish to elide. Research-use-only material is not approved for human use in any jurisdiction, is not required to meet parenteral standards, and is not represented by its sellers as meeting them. Everything in this article about endotoxin limits and sterility assurance describes the framework that would apply to a parenteral medicine. Applying that framework to a research chemical is a comparison, not a compliance requirement, and readers should hold both halves of that sentence at once.

A note on method and sourcing

The compendial material in this article is drawn from the current general chapters of the United States Pharmacopeia and the European Pharmacopoeia, read in the original rather than in summary, and from the international standards on aseptic processing and on laboratory competence. Where a chapter has changed status recently — as the recombinant reagent chapters have — we say so, because a reader consulting an older edition will find a different framing.

Where the Journal reports a number it obtained itself, it states the laboratory’s accreditation status, the method family, whether method suitability was established, and the number of determinations. Where we report what a company told us, we distinguish an answer from a refusal and a refusal from a non-response, because those three things are routinely collapsed in coverage of this trade and they are not the same.

Corrections to this department are handled by the standards desk, which reads every letter and records the outcome in the log. Readers who believe a paragraph here overstates its evidence are asked to write to standards@compoundjournal.com; readers with documents to send, including certificates they would like read, should write to letters@compoundjournal.com. We do not publish correspondents’ names without permission and we do not identify the source of a certificate.

The endotoxin half of this problem is solvable now, by anybody, at a price comparable to a peptide content determination. That is the single most practically useful sentence in this article. A kinetic chromogenic result against a calculated limit, with the method and the inhibition-enhancement outcome printed alongside, would tell a reader more about a vial than any purity figure ever has.