An endotoxin figure on a certificate of analysis is a single number carrying a unit most bench researchers rarely handle: endotoxin units per milligram. It sits next to a purity percentage and looks like the same kind of claim. It is not. Purity is a mass fraction resolved by separation. Endotoxin is a biological activity, measured by a clotting cascade borrowed from an arthropod, calibrated against a reference preparation, and reported in units with no fixed mass equivalent. The two numbers answer different questions, and a material can score well on one while telling you nothing about the other.

This guide covers the three classical lysate formats, what recombinant factor C changes, how to read and compute EU/mg, and the specific ways an endotoxin result can be wrong in either direction.

The cascade the assay borrows

Every lysate-based endotoxin assay reads out a proteolytic cascade from horseshoe crab amebocytes. Endotoxin activates factor C, which activates downstream factors, which ultimately convert a clottable protein into a gel. The important structural fact is that this cascade has a second entry point: factor G, activated by (1→3)-β-D-glucans rather than by endotoxin. Anything that starts the cascade produces signal, and the instrument cannot tell you which branch started it.

That architecture explains most of what follows. The three classical formats differ only in how they observe the cascade, not in what triggers it, so all three share the same specificity problem, the same interference profile, and the same dependence on a reference endotoxin for calibration.

Gel-clot: the endpoint method

Gel-clot is the oldest and simplest format. Lysate and sample are mixed, incubated, and the tube is inverted. A firm gel that holds when the tube is turned over is a positive; anything else is a negative. There is no instrument and no curve — the readout is binary, and quantitation comes from where the positives stop in a dilution series.

Each lysate lot carries a labeled sensitivity, conventionally written as lambda (λ), in EU/mL. Suppose a lot is labeled λ = 0.03 EU/mL and a sample is tested as a two-fold series. If the last dilution to gel is 1:8, the arithmetic is 0.03 EU/mL × 8 = 0.24 EU/mL in the undiluted sample. If the last positive were 1:16, the answer would be 0.48 EU/mL. Note what that implies: gel-clot resolution is the dilution factor itself. A two-fold series can only ever place a result within a factor of two, and the reported value is a bracket presented as a point.

Gel-clot remains a referee method in several pharmacopoeial frameworks, which is worth knowing when two laboratories disagree. It needs no reader, is difficult to misconfigure, and fails visibly. Its limitation is resolution.

Turbidimetric and chromogenic LAL

The two photometric formats watch the same cascade continuously instead of at a single endpoint. Turbidimetric assays track the increase in optical density as the coagulogen polymerizes. Chromogenic assays substitute a synthetic peptide substrate carrying a chromophore for the natural clottable protein; the activated enzyme cleaves it and the resulting color is read at a fixed wavelength. Both run in either endpoint or kinetic mode. Kinetic versions record the time taken to reach a defined threshold — onset time — and fit log(onset time) against log(standard concentration), which is why they cover several orders of magnitude on one plate rather than the single factor-of-two step a gel-clot series resolves.

Kinetic chromogenic LAL is the workhorse for routine release testing, and it is the comparator against which alternatives are usually judged. A quality-control laboratory testing roughly 15,000 water samples annually reported that its kinetic chromogenic assay and a recombinant factor C assay were the two most appropriate methods for rapid endotoxin detection in water, while a second LAL-based system performed less well on their criteria. [8]

Two practical points apply to both photometric formats. The standard curve must be constructed on the same plate, from the same reference endotoxin, on the same day — cross-plate or historical curves are not equivalent. And every sample must be accompanied by a positive product control: the same sample spiked with a known amount of endotoxin. Without the spike you cannot distinguish a genuinely clean sample from a sample that suppressed the assay.

Recombinant factor C

Recombinant factor C assays replace the crab-derived lysate with a single cloned protein — the cascade's first component — expressed in a recombinant system and paired with a fluorogenic substrate. Because factor G is simply absent from the reagent, the glucan entry point does not exist. This is the central technical claim for rFC, and it is a structural argument rather than an empirical one: the branch is not suppressed, it is not there.

The comparative literature is reasonably consistent. Across four complex vaccine matrices, rFC results were at least equivalent to LAL results, with two informative exceptions running in opposite directions: rFC was adequate but slightly less suitable for one product containing proteases, because the steps used to inactivate those proteases degraded assay performance, while LAL was adequate but less suitable for a product containing glucans. [7] In purified pharmaceutical water and cleaning-validation samples, an rFC assay was less sensitive to interference than the laboratory's kinetic chromogenic LAL method, with better accuracy, repeatability and a shorter time to result. [8] A six-year proficiency-testing program comparing one kinetic chromogenic LAL assay against two rFC assays found all results met compendial acceptance criteria and that the rFC methods produced better endotoxin recovery rates; the authors work for a commercial endotoxin testing service, which they disclose. [9]

The more interesting divergence is structural. Lipopolysaccharide is not one molecule. When LPS preparations from several Gram-negative species were compared across an rFC assay, two LAL assays and a human monocyte assay, potencies agreed for E. coli, B. cepacia and two Salmonella preparations but differed significantly for K. pneumoniae, S. marcescens, B. pertussis and P. aeruginosa. In that study only the rFC assay correlated significantly with IL-6 release from a monocytic cell line, and the authors concluded rFC is a good replacement for conventional LAL. [5] The scope matters: this was a panel of purified LPS preparations, not a survey of real contamination events.

An environmental comparison found the same broad picture with a different texture: in house dust, LAL gave higher endotoxin estimates than rFC, the two were highly correlated (r = 0.86), spike recovery was 113.6% for LAL and 99.7% for rFC, and — contrary to the obvious hypothesis — the LAL–rFC difference did not correlate with β-glucan levels in those samples. [3] Glucan is a real interference mechanism, but it is not the explanation for every discrepancy.

What EU/mg actually means

The endotoxin unit is defined by activity against a reference standard endotoxin, not by mass. This is the single most misread aspect of an endotoxin figure. The mass-to-activity ratio is a property of a particular endotoxin preparation, and it must be determined, not assumed. A collaborative study established that the E. coli O55:B5 lot then serving as a control standard endotoxin contained approximately 4.5 EU per ng. [1]

That number makes the scale concrete. Since 1 mg = 106 ng, 4.5 EU/ng corresponds to 4.5 × 106 EU/mg for that preparation. Inverting it: 1 EU is about 1 ÷ 4.5 = 0.22 ng of that particular LPS. So a hypothetical specification of 1 EU/mg of peptide corresponds to roughly 0.22 ng of LPS per mg of peptide, which as a mass fraction is 0.22 ÷ 106 = 2.2 × 10-7, or about 0.22 parts per million — 0.000022% of the material. Endotoxin specifications sit orders of magnitude below anything a chromatographic purity assay resolves, which is the arithmetic reason the two numbers cannot substitute for each other. The ratio is specific to that endotoxin lot and preparation; a different LPS would give a different conversion.

Converting a reported result to EU/mg is straightforward arithmetic. Take a 10 mg vial reconstituted in 2 mL, giving 5 mg/mL. If the assay reports 0.25 EU/mL on that solution, then 0.25 EU/mL ÷ 5 mg/mL = 0.05 EU/mg, and the vial contains 0.25 EU/mL × 2 mL = 0.5 EU in total. Run the conversion the other way before you accept a certificate: a figure quoted in EU/mL without the concentration it was measured at is uninterpretable.

The same arithmetic governs the maximum valid dilution. If a specification of 1 EU/mg has been set for a material tested at 5 mg/mL, the limit concentration is 1 EU/mg × 5 mg/mL = 5 EU/mL, and with λ = 0.03 EU/mL the MVD is 5 ÷ 0.03 = 166.7, which is used as 166-fold — the MVD is a ceiling, so it is rounded down rather than to nearest. Testing beyond the MVD dilutes real contamination below detection; testing below it may leave interfering matrix in the tube. The 1 EU/mg here is illustrative arithmetic, not a recommended figure.

Interference: inhibition, enhancement, and glucan

Interference is not an edge case; it is the default assumption until disproved for a given matrix, and it runs both ways. Enhancement makes a clean sample look contaminated; inhibition makes a contaminated sample look clean. Spike recovery of the positive product control tells you which you have, and the conventional acceptance window is 50–200%.

The glucan route deserves specific attention because it produces false positives that look entirely ordinary. β-glucans from fungal cell walls, cellulose filters and plant-derived excipients activate factor G. LAL is known to cross-react with plant-derived β-glucans, and this false-positive behavior is a recognized limitation of the method. [10] Where a glucan source is plausible, the options are a glucan-blocking buffer, a factor G-depleted lysate, or an rFC assay that has no factor G at all. What you cannot do is reason your way past a positive result on the grounds that glucan was probably responsible.

Low endotoxin recovery

Low endotoxin recovery, or masking, is the failure mode most likely to make an endotoxin certificate misleading. It is commonly defined as recovery of less than 50% of a known endotoxin spike, and the proposed mechanisms are physical: endotoxin aggregating, binding to protein, or organizing into micelles or vesicles such that it no longer presents to the cascade. [11] Formulations combining a chelating buffer with a surfactant are the classic context. The practical consequence is a false negative — the endotoxin is present and the assay does not see it.

Two findings should temper how confidently anyone talks about LER. First, susceptibility depends on which endotoxin you spike. Comparing highly purified endotoxin against crude extracts from different bacterial species grown under different conditions, masking varied with source, strain and growth conditions, and the authors concluded it is not possible to predict whether a given native contamination would be susceptible; detecting LER requires a spike that is itself susceptible to it. [6] A hold-time study that passes with purified endotoxin has therefore not fully cleared the matrix.

Second, the phenomenon is not universal. A study of mammalian cell-produced biologic drug products and drug substances in citrate, histidine, phosphate and sodium acetate formulations containing polysorbate, held up to six months and including a preserved peptide-containing formulation, reported acceptable recoveries in every case and emphasized the importance of appropriate study design in assessing endotoxin recovery. [4] And in products that do show LER, masking was observed not only in the bacterial endotoxins test but also in the monocyte activation test and the rabbit pyrogen test, which the authors read as suggesting the products biologically inactivate the spiked endotoxin over time rather than merely hiding it from one assay. [12] That is a meaningfully different interpretation from "the test is blind," and it is not settled.

Why endotoxin is separate from purity

A purity percentage from HPLC is a ratio of peak areas: how much of the material that eluted was the target compound. Peptide reference standards are characterized by a battery of orthogonal methods — HPLC and GC for content and impurities, mass spectrometry and NMR for identity — with a quantitative value assigned to bulk material by a mass-balance approach and transferred to the vialed material, and some features such as chiral or isobaric residues require additional techniques for full characterization. [13] Endotoxin appears in none of these. At the sub-parts-per-million mass fractions computed above it is invisible to a chromatogram, and it is chemically unrelated to the synthesis impurities a purity method is designed to resolve.

Endotoxin also enters at different points. Synthesis impurities come from the synthesis; endotoxin comes from water, containers, filters, buffers and handling, at any step. Removal behaves differently too: in one downstream process, anion exchange chromatography removed endotoxin, cation exchange removed it partially, and ultrafiltration/diafiltration did not remove it at all. [11] Nothing about a high purity figure constrains the endotoxin figure, in either direction.

Depyrogenation of contact surfaces is a separate control again, and it is thermally demanding. Dry heat is the common approach; one validation study noted that the two Ph. Eur. variants of 250 °C for 30 minutes and 200 °C for 60 minutes are not equivalent, that a 3-log reduction was the USP and FDA requirement, and recommended conditions delivering a 4-log reduction for glass containers without fillers. [2] Autoclaving sterilizes; it does not depyrogenate.

What is convention and what is measured

Being honest about the boundary is the point of this page. The cascade chemistry, the assay comparisons, the masking mechanisms and the mass-to-activity conversions cited above are measured. Several widely repeated conventions are not, or are not to the same standard.

  • The 50–200% spike recovery window. Widely applied and compendially entrenched, but it is a chosen acceptance band, not a measured breakpoint.
  • Two-fold dilution series. Convention. Nothing measures better at a factor of two than at a factor of 1.5.
  • What a supplier's endotoxin figure covers. A certificate reports a result on the lot as tested, at a stated concentration, by a stated method. It says nothing about the vial after it is opened, reconstituted with a solvent of unknown endotoxin content, and aliquoted. Endotoxin does not appear spontaneously, but it arrives readily with water and surfaces.
  • Non-endotoxin pyrogens. Neither LAL nor rFC detects them. A clean endotoxin result is a clean endotoxin result and not a general absence of pyrogenic material — which is part of why monocyte activation methods exist at all.

The reasonable posture toward an endotoxin line item is: check the method and the format, check the concentration the result was measured at, check whether a positive product control was run and recovered, and convert to EU/mg yourself. A number without those four things attached is not yet a measurement.