Why does autoclaving not destroy endotoxin, and how many vials would you need to test to actually catch contamination?

Two things I want to understand properly rather than by rule of thumb.

First: I keep reading that endotoxin survives autoclaving. Autoclaving is 121 degrees under pressure for 15 minutes and it reliably kills bacterial spores, which are the most heat-resistant biological structures anyone routinely deals with. Why would a molecule from a bacterial cell wall be harder to destroy than a spore? Is there a number attached to this, or is it a qualitative claim?

Second: I have seen it asserted that the pharmacopoeial sterility test is statistically almost useless, which seems like a strong thing to say about a compendial method. If a lot of 4,000 vials has some contamination in it, and a lab tests the compendial number of containers, what is the actual probability the test catches it? I would like the arithmetic, because I am trying to decide how much weight to put on a "sterility: pass" line if I ever see one, and "almost useless" and "compendial requirement" are hard to hold in the same head.

Related, if anyone has seen one: what does a sterility failure actually look like in practice, and how do labs distinguish a genuine failure from their own analyst contaminating the tube?

On the third part of the question, since nobody has covered it: what a sterility failure looks like and how the false positives get sorted out.

Physically it is unremarkable. A clear tube of Fluid Thioglycollate Medium or Soybean-Casein Digest Medium goes turbid, usually somewhere between day 2 and day 7, occasionally as late as day 13 for a slow-growing organism. Sometimes there is a surface pellicle, sometimes discrete colonies on the membrane, sometimes only a faint haze that requires comparison against the negative control tube to call at all. Anaerobic growth in thioglycollate often appears as a band at a specific depth rather than uniform turbidity.

Then the investigation, which is where the interesting judgement happens. The isolate is subcultured and identified — MALDI-TOF or 16S rRNA sequencing. The identity is the primary evidence about origin:

• Staphylococcus epidermidis, Micrococcus, Corynebacterium, Cutibacterium: skin flora. Strongly suggests the analyst or the testing environment, not the product. The investigation then looks for the same organism in that day's environmental monitoring plates, personnel monitoring, and the media lot's own controls.
• Pseudomonas, Burkholderia, Ralstonia: water-system organisms. Points at the product or its process water.
• Bacillus and moulds: ambiguous. Environmental, could be either.

USP 71 permits the test to be invalidated and repeated only if there is documented evidence that the test itself was defective — a demonstrated breach in aseptic technique, a failed media growth-promotion control, a monitored environmental excursion during the test. Absent that, the batch fails and cannot be released. What is emphatically not permitted is "retest until pass", and the reason the rules are written so tightly is that the temptation is enormous: a failure destroys a batch, and the base rate of genuine laboratory false positives is real but low, single-digit percent of positives in a competent lab.

The relevance to a buyer: if a vendor ever tells you a batch "failed sterility initially but passed on retest", the correct question is what the documented invalidation cause was and what the isolate was identified as. If the answer is vague, the honest interpretation is that the batch failed and was released anyway.

In practice you will never see this conversation, because research-grade peptide vials are not sterility tested and no failure is available to discuss. Which is itself the answer to how much weight to put on the absence of the line.

One addition on the endotoxin side that follows from the depyrogenation arithmetic and matters for anyone reconstituting.

Since endotoxin cannot be removed by filtration and is not destroyed by any heat you can apply to a peptide, every route by which LPS could enter a solution is a one-way door. That includes:

• The diluent. Bacteriostatic water and sterile water for injection from a reputable pharmaceutical source are manufactured to a compendial endotoxin limit — typically 0.25 EU/mL for water for injection. Laboratory-grade or distilled water is not, and a bottle of deionised water from a bench system can carry hundreds of EU/mL if the polishing loop has a biofilm. Deionised water is famously a worse endotoxin source than tap water, because the ion-exchange resin bed is an excellent bacterial habitat.
• The syringe and needle. Sterile is not the same as non-pyrogenic, although single-use medical devices are generally both.
• Anything reused. A rinsed and dried container is sterile-ish and pyrogenic, because rinsing removes organisms and leaves their cell walls behind.

The practical rule that falls out: endotoxin in a finished solution is the sum of what was in the powder and what was in everything it touched, and nothing downstream reduces it. That is a stronger constraint than most handling advice acknowledges, and it is why the diluent's own compendial specification is not a triviality.

Again, for the record: research-use-only compounds, not approved for human administration, and anything touching a decision about a person belongs with a clinician.