Where does an "EU/mg" endotoxin limit come from, and is a spec of 10 EU/mg tight or loose?

Two COAs for the same nominal product give me endotoxin specifications of "less than 10 EU/mg" and "less than 0.5 EU/mg" respectively. Both pass. The second vendor charges more and points at their tighter spec as a selling point.

What I cannot judge is whether 10 EU/mg is a real limit that means something, or a number somebody invented because it is a round figure and easy to pass. I gather there is a formula involving a threshold pyrogenic dose and a maximum dose, but every explanation I have found either states the formula without doing the arithmetic or does the arithmetic for a compound dosed in grams, where the answer comes out tiny.

Can someone work the derivation through for a peptide dosed in single-digit milligrams weekly? I would like to be able to look at a spec and say "that is a factor of X below the pharmacologically derived limit" instead of comparing two numbers I have no scale for. And separately — if the derived limit turns out to be loose, does the tighter spec actually buy anything, or is the second vendor selling me a number?

The formula is ELC = K / M and for a milligram-dosed weekly peptide it produces a surprisingly loose limit, which is why 10 EU/mg passes comfortably and why the tighter spec is worth less than it looks — though not worthless. Here is the derivation.

The two terms

K is the threshold pyrogenic dose: the amount of endotoxin per kilogram of body mass per hour that is taken as the threshold for a febrile response. For parenteral routes other than intrathecal, K = 5 EU/kg/hour. (Intrathecal is 0.2 EU/kg/hour, a 25-fold tighter figure, because the central nervous system has no tolerance for pyrogen.)

M is the maximum dose of product per kilogram of body mass administered in one hour. For a bolus given in a few seconds, M is simply the whole dose divided by body mass. Pharmacopoeial convention uses a 70 kg adult.

Worked for semaglutide at the 2.4 mg maintenance dose

The 2.4 mg weekly dose is the arm with the outcome data [1], so it is the right number to use for a worst-case limit.

• M = 2.4 mg / 70 kg = 0.0343 mg/kg
• ELC = 5 EU/kg / 0.0343 mg/kg = 145.8 EU/mg

Worked for tirzepatide at 15 mg

The highest studied maintenance dose [2]:

• M = 15 mg / 70 kg = 0.2143 mg/kg
• ELC = 5 / 0.2143 = 23.3 EU/mg

The full picture

That last row is why generic advice is useless. A drug dosed in hundreds of milligrams needs a sub-EU/mg limit; a peptide dosed in single milligrams does not. The limit scales inversely with dose, and these compounds are extraordinarily potent.

Does the 0.5 EU/mg spec buy anything?

Three things, none of which is "protection from pyrogenic reaction at label dose":

1. It is a process hygiene indicator. Endotoxin is a proxy for Gram-negative bioburden somewhere in the process — water system, glassware, an operator, a raw material. A batch at 8 EU/mg is technically within a 10 EU/mg spec and is also telling you that something in that facility is growing Pseudomonas. A batch at 0.05 EU/mg is telling you the water system is clean. That is genuinely useful information about the manufacturer, independent of the dose arithmetic.
2. Dose margin at low body mass and stacked administration. The 70 kg assumption is a convention. At 50 kg the semaglutide limit drops to 5 / (2.4/50) = 104 EU/mg. Still loose, but the margin is not what you calculated.
3. Multi-dose vials give repeated exposure. K is a per-hour figure and does not accumulate across a week, so this is a weak argument, but it is not nothing if a vial is being drawn from over a month.

So: 10 EU/mg is a real limit with a 2 to 15-fold margin depending on the compound and dose, and 0.5 EU/mg is a better hygiene signal rather than a meaningfully safer product at these doses.

The thing that actually matters more than either spec

Whether a test was run at all, on which batch, by what method, and with a positive product control that recovered. A specification with no accompanying result is a sentence, not data. Ask for:

• The method: gel-clot (limit test only), kinetic turbidimetric, kinetic chromogenic, or recombinant Factor C. Kinetic chromogenic and rFC give you a number; gel-clot gives you pass or fail against one sensitivity.
• The lysate sensitivity, lambda: typically 0.03 or 0.125 EU/mL for gel-clot, down to 0.005 EU/mL for chromogenic.
• The maximum valid dilution and the dilution actually used. MVD = ELC x sample concentration / lambda. For tirzepatide at a 23.3 EU/mg limit, tested at 1.0 mg/mL with lambda = 0.005: 23.3 x 1.0 / 0.005 = 4660-fold. A lab has enormous dilution headroom here, which is convenient because peptides frequently interfere with the assay and dilution is how you get rid of interference.
• The positive product control recovery, which must fall between 50% and 200%. Without it, a "less than lambda" result may just mean the product inhibited the reaction.

A report with the method, lambda, dilution factor, PPC recovery and a numeric result is worth far more than a tighter spec with none of that. Ask which of the two vendors can produce one.

edited 16 Sept 2025 by swab_and_wait — fixed an arithmetic slip in the third paragraph

Adding the assay mechanics, because the choice of method changes what a number means and the terminology on COAs is used loosely.

All the classical methods exploit the same biology: a clotting cascade in the amoebocyte lysate of the horseshoe crab. Lipopolysaccharide activates Factor C, which activates Factor B, which activates the proclotting enzyme, which in nature cleaves coagulogen into a gel.

• Gel-clot. Incubate sample with lysate for 60 minutes at 37 degrees, invert the tube. A firm gel that does not slump is positive. It is a binary result against the lysate's labelled sensitivity lambda. Cheap, robust, no instrument, and it cannot give you a number — only "less than lambda times the dilution factor" or "more than". If a COA reports "less than 0.125 EU/mL" with no curve, this is what was run.
• Kinetic turbidimetric. Read the onset time of turbidity as the gel forms; onset time is log-linear in endotoxin concentration. Gives a number over about three orders of magnitude. Sensitive to sample turbidity and colour.
• Kinetic chromogenic. Replace coagulogen with a synthetic peptide substrate carrying para-nitroaniline. The activated enzyme cleaves it and you read the yellow product at 405 nm; onset time again gives concentration. Standard curves typically span 0.005 to 5 EU/mL. This is the workhorse and it is what you want on a peptide COA.
• Recombinant Factor C. Recombinant Factor C plus a fluorogenic substrate, single-step, no crab. Compendial in the European Pharmacopoeia and available as an alternative method in the USP. It reads only the Factor C pathway, so it is blind to beta-glucans — an advantage, because beta-glucan contamination from cellulose depth filters and some cotton-plugged glassware causes false positives in lysate-based methods via the Factor G pathway.

Interference is the practical problem with peptides. Highly charged and amphiphilic molecules inhibit or enhance the cascade at concentrations well above a few tenths of a mg/mL. The lab handles it by diluting toward the MVD until the positive product control recovers within 50 to 200%. That works, but it means the actual limit of detection on your sample is lambda x dilution factor, not lambda. A result reported as "less than 0.5 EU/mg" from a chromogenic assay run at 100-fold dilution with lambda 0.005 has an effective detection limit of 0.5 EU/mL in the diluted sample, which is exactly the reported number — fine, but it means the assay had no headroom to see anything lower and the "less than" is a detection limit, not a measurement.

None of this is exotic and any of the independent labs will tell you the numbers if asked. Janoshik and Medutest both offer endotoxin as an add-on rather than as part of a standard purity panel, and it is a different price because it is a different assay on different equipment.

Counterpoint on the practical relevance, which I think the thread is underweighting in the other direction.

The dose arithmetic above is correct and gives a loose limit, but K = 5 EU/kg is a fever threshold derived from rabbit pyrogen studies and human volunteer work with purified LPS. It is not a no-effect level for everything endotoxin does. Sub-pyrogenic LPS exposure produces measurable cytokine responses, local injection-site inflammation and malaise in some people at doses well below the febrile threshold, and the between-individual variability in sensitivity is large.

Localised effects also do not follow the systemic per-kilogram arithmetic. Endotoxin deposited into subcutaneous tissue at a single site is a local bolus, and a persistent hard, red, itchy nodule at every site is one of the more common complaints about material with poor process hygiene. The K/M model says nothing about that because it was never designed to.

So my reading of the same numbers: the derived limit tells you when you are risking a systemic pyrogenic reaction, and 10 EU/mg keeps you clear of that at these doses. It does not tell you that 8 EU/mg material is as good as 0.05 EU/mg material, and the difference between those two batches is a real difference in what came out of a real facility.

Given that these are research-use-only compounds not approved for administration to humans, the framing I would actually use is: endotoxin level is the best single proxy available for whether a manufacturer controls their water, their glassware and their fill environment, and that proxy correlates with everything else you cannot measure. Buy the low number for what it implies, not for the pyrogen arithmetic.