Quality and COAs

Endotoxin and EU/mg: the contaminant no purity test can see

Endotoxin survives steam sterilisation, is not what a sterilising filter is rated to remove, and sits orders of magnitude below anything an HPLC purity figure can resolve. It is measured by a separate assay with its own distinctive failure modes, including one where the assay reports zero and the biology disagrees.

Written for Analytical chemists, cell-culture researchers, and quality reviewers · Updated August 7, 2026

Key points

Purity, sterility and endotoxin are three independent measurements
Steam sterilisation does not destroy endotoxin; depyrogenation is a dry-heat process
The endotoxin unit is a potency defined against a reference standard, not a mass
Masked endotoxin can be undetectable by Factor C assays and still activate cells

Research-use boundary

This article is educational and limited to non-clinical research context, documentation practices, and terminology. It does not provide preparation, administration, dosing, treatment, diagnostic, cosmetic, or consumer-use guidance, and nothing here indicates that any material is suitable for human or animal use.

Sterile does not mean endotoxin-free, and pure does not either

Bacterial endotoxin is not destroyed by steam sterilisation, it is not what a sterilising-grade filter is rated to remove, and it is present at concentrations several orders of magnitude below anything a purity chromatogram can resolve. All three statements are ordinary facts of the analytical literature, and together they produce a result that surprises people the first time they meet it: a material can be highly pure by HPLC, sterile by every test applied to it, and still carry enough endotoxin to drive a cell-based readout on its own. Endotoxin is measured by a separate assay because nothing else on a certificate is looking for it.

Where it comes from and why killing the organism does not help

Endotoxin is lipopolysaccharide, a structural component of the outer membrane of Gram-negative bacteria. It is not a secreted toxin that a live organism has to be present to produce; it is part of the wall, and it is released when cells lyse. That inverts the usual intuition about contamination control. A process step that kills bacteria liberates the lipopolysaccharide that was in their membranes, so the step that makes a solution sterile can leave the endotoxin load unchanged or higher. Sterility testing and endotoxin testing therefore answer genuinely different questions, and passing one says nothing about the other.

Why filtration does not solve it

A sterilising-grade filter is specified by its ability to retain microorganisms, not by molecular size, and endotoxin is very much smaller than the bacteria it came from. It is also not a single species: work published in 1977 examined lipopolysaccharide across the range of aggregation states it adopts in solution — vesicles, micelles, and detergent-solubilised forms — because the composition of the surrounding solution decides which state predominates, and the effective size follows from that. Removal in that study was achieved with membranes of 10,000 nominal molecular weight limit. That is ultrafiltration, not sterilising filtration, and the gap between the two ratings is the reason a filtered solution can be sterile and still carry endotoxin.

Why heat does not solve it either

Endotoxin is thermally stable in a way that vegetative organisms are not. Steam sterilisation cycles that reliably kill bacteria leave lipopolysaccharide largely intact, which is why removing it by heat is a separate process with its own name and its own conditions. Depyrogenation is a dry-heat operation, and the pharmacopoeial acceptance criterion is a three-log reduction of an endotoxin challenge rather than the six-log microbial reduction that defines sterilisation. A validated dry-heat cycle of 200 degrees Celsius for 60 minutes has been shown to deliver a four-log reduction, comfortably exceeding that requirement. Those are glassware and equipment conditions, not conditions any material survives, which is precisely the point: endotoxin is controlled by keeping it out, not by removing it afterwards.

Why no chromatogram will ever show it

The concentrations that matter are extraordinarily small. A study of commercially supplied recombinant proteins exposed immune cells to lipopolysaccharide across 0.002 to 2 nanograms per millilitre and found activation of human CD1c-positive dendritic cells at amounts equivalent to the contamination actually present in some of those preparations. Put that on the scale a purity method works at. In a solution at one milligram per millilitre, 2 nanograms per millilitre of lipopolysaccharide is 0.0002 percent of the dissolved material by weight; 0.002 nanograms per millilitre is 0.0000002 percent. A purity figure reported to a tenth of a percent is between two and five orders of magnitude too coarse to see either. Endotoxin is also not a peptide, does not carry the backbone chromophore a peptide method is detecting, and does not behave like one on a reverse-phase separation. It is invisible to that measurement in every sense that matters.

What an endotoxin unit actually is

EU/mg reads like a concentration, and it is one, but the numerator is not a mass. The endotoxin unit is a unit of biological potency in the Limulus amebocyte lysate cascade, defined by calibration against a reference standard endotoxin rather than by weighing anything. That distinction has a practical consequence: lipopolysaccharide structure varies between organisms and serotypes, and potency per unit mass varies with it, so two materials reporting the same EU/mg do not necessarily carry the same mass of endotoxin. The unit is deliberately defined this way because activity, not weight, is what the assay responds to and what the number is trying to capture.

The denominator is not free of assumptions either

Normalising to milligrams pushes the figure straight into the net-content problem. If EU/mg is calculated against gross powder weight, and the powder is 70 percent peptide with the balance counter-ion and water, then the endotoxin burden per milligram of actual peptide is higher than the certificate states by the reciprocal of that fraction. Whether a certificate means gross or net is rarely stated and rarely the same between suppliers, so the comparison between two certificates is often not the comparison it appears to be.

The assay has two arms, and only one of them is looking for endotoxin

Limulus amebocyte lysate contains two initiating proteases. Factor C is activated by bacterial endotoxin. Factor G is activated by beta-(1,3)-D-glucans, an entirely different class of molecule found in fungal cell walls and, more awkwardly, leaching from cellulose. Both pathways converge on the same clotting endpoint, so a glucan-driven reaction is indistinguishable from an endotoxin-driven one in the standard assay. This is not a theoretical concern: a cell-therapy product was reported in 2002 to have returned false positive endotoxin results traced to glucans acquired from a sterilising cellulose filter — the filtration step intended to control contamination was itself the source of the interference.

Which is what the recombinant methods change

USP General Chapter 86, Bacterial Endotoxins Test Using Recombinant Reagents, was published in November 2024 and became official in May 2025. It describes recombinant Factor C and recombinant cascade reagent methods as alternatives to the lysate-based test of General Chapter 85. Because these reagents reconstruct the Factor C arm without Factor G, they are not activated by beta-glucans, which removes an entire class of false positive rather than correcting for it. The same specificity can be obtained in the other direction: removing Factor C from lysate leaves a reagent that responds only to glucans, which is how a suspected glucan interference is confirmed rather than guessed at.

Interference runs both ways, which is why a spiked control is mandatory

A sample matrix can suppress the cascade as easily as it can trigger it, and a peptide solution is a demanding matrix: pH, buffer salts and residual acid from synthesis all reach the assay. The control that catches this is a spike of known endotoxin added into the sample matrix, run alongside the sample. Published NIH protocols set the acceptance range at 50 to 200 percent recovery of that spike: recovery below 50 percent indicates inhibition, and recovery above 200 percent indicates either enhancement or endotoxin already present in the sample. Outside that window the result is invalid, not merely uncertain. A reported endotoxin figure without an inhibition and enhancement check behind it is a number whose validity was never established.

The failure mode where the assay reports zero and the biology disagrees

Low endotoxin recovery is the most instructive result in this field. In certain solution compositions — characteristically a polysorbate detergent combined with a citrate or phosphate buffer — endotoxin that is demonstrably present becomes progressively undetectable by lysate and Factor C based assays over hold time. The obvious reading is that the endotoxin has been neutralised. It has not. Masked endotoxin that Factor C assays cannot detect still activates a TLR4-driven NF-kB reporter cell line, and in primary human monocytes it still induces pro-inflammatory cytokines and surface activation markers at low concentrations. The assay stopped seeing it; the cells did not.

Why this belongs in the confounder list, not the safety file

For anyone running a cell-based measurement, endotoxin is first and foremost an experimental confound. Lipopolysaccharide is among the most potent activators of innate immune signalling known, and a contaminant that drives cytokine expression on its own will produce a signal that gets attributed to whatever else was in the tube. The pattern is the same one the counter-ion creates: the confound scales with the amount of material added, so it mimics a dose-response, it does not present as noise, and replication does not reveal it because every replicate carries the same contamination. The 2014 recombinant-protein study makes the point concretely — the endotoxin levels were generally stated on the datasheets, and were sufficient to activate the cells anyway.

How to read an endotoxin figure on a certificate

Whether endotoxin was tested at all. An absent test is information, not a pass.
The method: gel-clot, kinetic turbidimetric, kinetic chromogenic, or a recombinant Factor C method, since they differ in what can interfere with them.
Whether the result is a measurement or a limit. A value written as less than a figure reports the sensitivity of the run, not the content of the vial.
Whether inhibition and enhancement were assessed, and whether spike recovery fell inside 50 to 200 percent.
Whether EU/mg is normalised to gross powder weight or to net peptide content.
For materials in detergent-containing or chelator-containing solutions, whether hold-time recovery was examined at all.

What the number does not establish

An endotoxin result characterises a lot. It is not a clearance, a qualification, or an approval of the material for any purpose, and it says nothing about the many properties it does not measure — identity, potency, sterility, or anything else. VSA BioScience materials are supplied for laboratory research use only and are not for human or animal use; an endotoxin figure on a research-use certificate documents a property of the batch and nothing beyond it.

The short version

Endotoxin is the contaminant that every other test on the certificate is structurally incapable of finding. It outlives the organism that carried it, passes the filter that made the solution sterile, survives the autoclave, and sits far below the resolution of any purity method. The assay that does look for it has two ways of being wrong — a second activation pathway that reads glucans as endotoxin, and a masking phenomenon that hides endotoxin the cells can still see. Which is why the useful question about an EU/mg figure is never just the value. It is which method produced it, and what was done to prove that method could be trusted on that sample.

Common questions

Does sterile filtration remove endotoxin?

No. A sterilising-grade filter is rated for retention of microorganisms, not for molecular size, and bacterial endotoxin is far smaller than the organisms it came from. Published filtration work removed endotoxin from contaminated solutions using membranes with a 10,000 nominal molecular weight limit, which is ultrafiltration rather than sterilising filtration. A filtered solution can therefore be sterile and still carry endotoxin.

Does autoclaving destroy endotoxin?

No. Bacterial endotoxin is heat stable and is not destroyed by ordinary steam sterilisation cycles. Removing it by heat requires dry-heat depyrogenation at much higher temperatures: the pharmacopoeial requirement is a three-log reduction, and a validated dry-heat cycle of 200 degrees Celsius for 60 minutes has been shown to achieve a four-log reduction. Sterilisation and depyrogenation are different processes with different conditions.

What does EU/mg mean on a certificate of analysis?

EU/mg is endotoxin units per milligram of material. The endotoxin unit is a measure of biological potency in the Limulus amebocyte lysate cascade, defined against a reference standard endotoxin, rather than a measure of mass. Because different lipopolysaccharide structures differ in potency per unit weight, two materials reporting the same EU/mg do not necessarily carry the same mass of endotoxin.

Why can an HPLC purity figure not detect endotoxin?

Endotoxin is present at concentrations far below the resolution of a purity method. Lipopolysaccharide has been shown to activate sensitive immune cells at 0.002 to 2 nanograms per millilitre; in a one milligram per millilitre solution that is between roughly 0.0002 percent and 0.0000002 percent of the material by weight, which is orders of magnitude below the smallest peak a method reporting to 0.1 percent would integrate. Endotoxin is also not a peptide and does not behave like one on a reverse-phase separation.

What is low endotoxin recovery?

Low endotoxin recovery, or LER, is the phenomenon in which endotoxin present in a solution becomes undetectable by Limulus amebocyte lysate and Factor C based assays, typically in formulations combining a polysorbate detergent with a citrate or phosphate buffer. It matters because masked endotoxin has been shown to remain biologically active: it still activates a TLR4 reporter cell line and induces pro-inflammatory cytokines in primary human monocytes at low concentrations.

What is the difference between the LAL assay and recombinant Factor C?

Limulus amebocyte lysate contains two proteases, Factor C and Factor G, activated by bacterial endotoxin and by beta-(1,3)-D-glucans respectively, and both pathways converge on the same clotting endpoint, so glucans can read as endotoxin. Recombinant assays described in USP General Chapter 86 use the Factor C arm and are not activated by glucans. USP 86 was published in November 2024 and became official in May 2025 as an alternative to the compendial test in USP 85.

Sources

Biological Activity of Masked Endotoxin Scientific Reports (PubMed 28317862) Primary source for masked endotoxin being undetectable in Factor C based assays while still activating a TLR4-NF-kB reporter line and inducing pro-inflammatory cytokines and surface activation markers in primary human monocytes.
Residual endotoxin contaminations in recombinant proteins are sufficient to activate human CD1c+ dendritic cells PLOS ONE (PubMed 25478795) Primary source for commercially supplied recombinant proteins carrying endotoxin at and sometimes above their stated levels, and for cell activation across the 0.002 to 2 nanograms per millilitre LPS range.
Filtration removal of endotoxin (pyrogens) in solution in different states of aggregation Applied and Environmental Microbiology (PubMed 335971) Primary source for endotoxin existing in vesicle, micelle and detergent-solubilised states depending on solution composition, and for removal requiring filters of 10,000 nominal molecular weight limit.
Validation of dry heat inactivation of bacterial endotoxins PDA Journal of Pharmaceutical Science and Technology (PubMed 7804819) Primary source for a validated dry-heat cycle of 200 degrees Celsius for 60 minutes achieving a four-log endotoxin reduction against a three-log requirement.
Detection and Quantification of Gram-Negative Bacterial Endotoxin Contamination in Nanoparticle Formulations by Kinetic Chromogenic LAL Assay NCI Nanotechnology Characterization Laboratory (NCBI Bookshelf NBK612349) Published NIH protocol giving the inhibition/enhancement control acceptance range of 50 to 200 percent spike recovery, and the interpretation of results below and above it.
Detection and Quantification of beta-(1,3)-D-Glucan Contamination in Nanoparticle Formulations by Factor C Depleted LAL (Glucatell) Assay NCI Nanotechnology Characterization Laboratory (NCBI Bookshelf NBK604968) Published NIH protocol for Factor C responding to endotoxin and Factor G to beta-(1,3)-D-glucans, and for glucans producing false positives in the standard endotoxin assay.
False positive endotoxin results in a DC product caused by (1-3)-beta-D-glucans acquired from a sterilizing cellulose filter Cytotherapy (PubMed 12568992) Documented case of glucans leached from a sterilising cellulose filter producing false positive endotoxin results in a cell product.
General Chapter 86, Bacterial Endotoxins Test Using Recombinant Reagents United States Pharmacopeia Compendial source for the recombinant Factor C and recombinant cascade reagent methods offered as alternatives to the Limulus amebocyte lysate test of USP 85.