A lab ran a gut bacterium through an autoclave at 121 degrees, the same cycle used to sterilize surgical instruments, and human cells responded to the powder that came out. For decades we were told probiotics work because they are alive: little armies of good bacteria, marching in to take up residence. The more interesting question turns out not to be whether dead bacteria work. It is how you killed them, and whether anyone checked.
Lit Review Friday · Learn Something with Thaena · Episode 2 · Published July 2026 · Reading time: ~25 minutes
- Do probiotics have to be alive to work? For a large class of effects, no. Researchers autoclaved the gut bacterium Parabacteroides goldsteinii RV-01 at 121 °C for fifteen minutes, a protocol that leaves nothing viable, and the resulting powder still blunted the inflammatory response that E. coli provoked in human colon cells (Lin et al., 2024, Int J Mol Sci). The activity appears to ride on the surface chemistry rather than the living cell, though the authors are careful to call that an inference.
- Does the temperature used to kill the bacteria matter? It can, and hardly anyone has checked. Most published heat-killed preparations sit between 70 and 100 °C (Piqué et al., 2019, Int J Mol Sci). Very few studies have ever run more than one temperature on the same organism, and the two that did reached opposite conclusions on different readouts.
- What is a postbiotic? The field definition is "a preparation of inanimate microorganisms and/or their components that confers a health benefit on the host" (Salminen et al., 2021, Nat Rev Gastroenterol Hepatol). A probiotic is the live organism. A postbiotic is the inactivated organism and what it is made of.
- Has any of this been tested in people? For a different organism, yes. A randomized placebo-controlled trial gave participants pasteurized Akkermansia muciniphila, and the heat-treated arm improved insulin sensitivity and cholesterol against placebo (Depommier et al., 2019, Nat Med). The autoclaved P. goldsteinii work has not been through a human trial.
🎧 Learn Something Weekly Podcast Listen — Episode 2
Also on Spotify.
The Question Almost Nobody Asks
An autoclave is not a gentle machine. It is a pressure vessel that holds saturated steam at 121 degrees Celsius for fifteen minutes, and it is the standard by which a hospital decides a scalpel is safe to put inside a person. Nothing that concerns us here survives it. That is the entire point of owning one.
In 2024, a group in Taiwan put a gut bacterium through that cycle on purpose. Not to destroy it. To keep it.
The organism was Parabacteroides goldsteinii RV-01, a strain isolated from the stool of a healthy adult. They grew it, spun the cells down out of the broth, autoclaved the pellet, freeze-dried it, and ground it into a powder. Then they put that powder on human colon cells and watched what happened. The cells responded. Not to a living organism, because there was no longer one. To the shape of what was left.
That result is the part that gets quoted. It is not the reason the paper stuck with me.
The reason is that before they settled on the autoclave, they did something the field almost never does. They ran the same bacterium through three different heat treatments and asked which one left the most behind.
If you go looking, as I did, for other studies that tested more than one inactivation temperature on the same organism and compared what survived, you find almost nothing. The field picks a temperature, uses it, publishes, and moves on. Most published preparations cluster between 70 and 100 degrees, which a 2019 review describes as the usual working range, with heat the default method in most cases. Autoclaving appears in that literature too, as a minority choice. What is genuinely rare is not the harsh temperature. It is anyone running the experiment that would tell you whether the temperature they picked was the right one.
What this organism is
Parabacteroides is a genus of gut commensals, and across a dozen human populations it averages about 1.27 percent of the gut microbiota. That is the genus, which also includes P. distasonis, P. merdae, and P. johnsonii. P. goldsteinii alone is a fraction of that fraction. It is not sold in consumer probiotic products today, though at least one company is working to change that. It is a normal resident of a normal gut that has spent the last several years attracting research attention, mostly favorable. In mice, research groups have reported it countering obesity, easing colitis through valine-isobutyrate metabolism, and producing an unusual lipopolysaccharide that calms airway inflammation. A 2025 review in Frontiers in Microbiology reads it as a serious next-generation probiotic candidate.
What They Did: 121 Degrees, 15 Minutes, Nothing Left Alive
Cells were grown for 24 hours at 37 degrees. Then they were separated from the growth medium by centrifugation. Only then were they autoclaved, at 121 °C for 15 minutes, mixed with maltodextrin at one part cells to six parts carrier, freeze-dried, and milled.
Hold onto that middle step. The broth gets thrown away. Everything the bacterium secreted while it was alive is centrifuged off and discarded, and what goes into the autoclave is washed cell biomass. It will matter twice later.
Non-viability was confirmed by flow cytometry with a live/dead stain, and the resulting count is reported in total fluorescence units, which the paper defines plainly as the sum of dead cells. There is no ambiguity in the language and none in the protocol. This is not "heat-treated." It is not "inactivated." Those words are doing marketing work in a lot of places. Autoclaved is a number and a duration.
All five studies were run to OECD guidelines under GLP conditions in a certified facility.
- Bacterial reverse mutation (Ames), OECD 471. Five Salmonella Typhimurium strains, 5 mg/plate, with and without metabolic activation. Negative.
- In vitro chromosomal aberration, OECD 473. CHO-K1 cells, 5 mg/mL. Negative.
- In vivo micronucleus, OECD 474. ICR mice, 2000 mg/kg. Negative.
- 28-day repeated-dose oral, OECD 407. Sprague-Dawley rats. No deaths, no clinical signs. NOAEL 1500 mg/kg/day.
- 90-day repeated-dose oral, OECD 408. Sprague-Dawley rats, ten per sex per group. NOAEL at the highest dose tested.
That is the complete standard genotoxicity triad plus both a subacute and a subchronic oral study. The authors derive an acceptable daily intake of 900 mg for a 60 kg adult, with a hundredfold safety margin.
One result in that file deserves more attention than it gets. The organism carries tetQ, a tetracycline-resistance gene. In the finished ingredient, tetQ was undetectable by PCR, which the authors say excludes horizontal spread of that gene. A live bacterial product cannot make that claim, because a live cell keeps its genome. This is a concrete, measured benefit of the harsher process, and it has nothing to do with whether the preparation works. It has to do with what it cannot do to you. Oddly, the live organism is susceptible to tetracycline anyway, and the paper says the reason is not yet understood.
Autoclaved, Pasteurized, and Tyndallized Are Not the Same Thing
This is the most confused thing in the consumer postbiotic aisle. "Heat-killed" is a category, not a protocol, and the protocols are not equivalent.
- Pasteurized: moderate heat, typically 70 °C for 30 minutes. This is what was used for Akkermansia muciniphila in both the mouse and human studies below.
- Tyndallized: repeated cycles of heating and cooling, designed to catch spore-formers between germinations. Generally in the same gentle band.
- Autoclaved: saturated steam at 121 °C, 15 minutes, under pressure.
In one 2025 review's survey of published inactivation conditions, most entries sat at or below 90 °C and only a couple used 121 °C. The gentle band is where the field lives. The harsh end is a minority choice rather than an unexplored one.
The distinction comes down to a single molecule.
The Belgian group working on Akkermansia muciniphila identified a specific protein on that organism's outer membrane, Amuc_1100, that engages a receptor on host cells and reproduces part of the whole bacterium's effect on its own. They also reported that it is stable at the temperatures used for pasteurization. So they had a protein they wanted to keep, and they chose a temperature that keeps proteins. That is not caution. That is a design decision that follows directly from knowing what your active ingredient is.
Lin's paper states that autoclaving denatures the proteins, and uses that fact affirmatively to argue the allergenic risk of the ingredient is low. Amuc_1100 would not be expected to survive an autoclave.
Two teams, two different actives, two different correct answers. "Heat-killed" on a label tells you the cells are dead. It tells you almost nothing about what is left.
What Survived the Autoclave
Heat is not a single force that destroys a single thing. It is selective, and which molecules it ruins depends on what those molecules are made of.
Proteins are the fragile part. They work by being folded into a specific three-dimensional shape, and heat unfolds them.
Carbohydrates and lipids are the durable part. The outer surface of a Gram-negative bacterium is mostly sugar and fat: lipopolysaccharide anchored in the membrane, capsular polysaccharide wrapped around the outside. These structures were never folded, so there is nothing to unfold. Heat does chemistry on them, but far more of them comes through intact than protein does. The clearest evidence is negative: autoclaving does not destroy endotoxin. Stripping that activity out of pharmaceutical glassware takes dry heat at 250 degrees, not 121.
So what comes out of that autoclave is: the proteins are gone, the surface architecture is largely intact, and everything the cell had secreted was washed away before the cycle even started.
- The functional readout is measured. The autoclaved powder demonstrably suppressed an inflammatory signal in human colon cells. That part is data.
- The attribution of that effect to specific surviving molecules is inference, and the authors say so in their own word. Their discussion states that the anti-inflammatory lipopolysaccharide and capsular polysaccharide that "may survive" autoclaving were "speculated to constitute active components."
- The capsular polysaccharide is the weakest link. It was never isolated, purified, or measured. It was identified by searching the genome for genes resembling those of a known polysaccharide in Bacteroides fragilis, and the two matches sit at 69.3 and 38.6 percent protein identity. That is a thin basis for a molecular claim, and the paper does not pretend otherwise.
A note on a number you may see quoted: the paper reports the finished ingredient as roughly 95 to 96 percent carbohydrate. That is overwhelmingly the maltodextrin carrier, not the bacterium.
How Does Your Body Sense a Dead Bacterium?
Your innate immune receptors do not detect life. They detect molecular shapes.
Toll-like receptor 4 reads the acylation pattern of lipid A, the greasy anchor at the base of bacterial lipopolysaccharide. Toll-like receptor 2 reads lipoproteins and lipoteichoic acids, and the fine chemistry of those wall molecules tunes whether the response leans inflammatory or calming. NOD2, sitting inside the cell rather than on its surface, reads muramyl dipeptide, the minimal fragment of the peptidoglycan wall common to essentially all bacteria. Not one of these receptors has any mechanism for asking whether the cell that presented the shape is metabolically active. There is no pulse check. There is a shape, and a receptor that fits it.
Since the shapes in question are made of carbohydrate and lipid, and carbohydrate and lipid are what best survives 121 degrees, a heat-killed bacterium still presents much of the same face to your immune system.
The specific mechanism in this paper
In HCT116 human colon epithelial cells, the team ran three conditions. Autoclaved E. coli alone, which drove a strong inflammatory response, as expected. Autoclaved RV-01 alone, which produced no significant activation of its own. And RV-01 first, then E. coli, which significantly blunted the E. coli response. So the powder is quiet on its own and it makes the cell quieter when something noisy shows up.
The proposed reason is structural. Searching the RV-01 genome, the team found it lacks LpxM, the enzyme that attaches the sixth acyl chain to lipid A. The classic inflammatory lipid A, the E. coli version that TLR4 reads as danger, carries six fatty-acid tails. RV-01 is predicted to build only five. A penta-acylated lipid A still fits the receptor, but it does not trigger it. It occupies the site and blocks it.
If that prediction is right, the bacterium is not sending a calming signal. It is sitting in the alarm's keyhole so the alarm cannot be turned. Worth saying plainly: Lin's team never isolated RV-01's lipid A and never ran a TLR4 experiment on it. This is a genome-based prediction that fits the functional result, not a measured structure.
Two things follow that are easy to conflate. This mechanism belongs to the organism, not to the autoclave. Penta-acylation is written into P. goldsteinii's genome; heating did not create it and does not confer it on anything else. It also makes RV-01 typical of its phylum rather than exceptional, since Bacteroidota characteristically build hypo-acylated lipid A with low inflammatory activity. What belongs to the autoclave is narrower, and it is the subject of the next section.
What Happens When Someone Actually Tests the Temperature
Before Lin's team settled on 121 degrees, they ran the comparison.
Same strain, same batch, four conditions: live, pasteurized at 70 °C for 30 minutes, heat-treated at 100 °C for 15 minutes, autoclaved at 121 °C for 15 minutes. One readout, measuring whether host receptors could still see the cells at all.
The autoclaved cells came out on top. They signaled as strongly as the live bacterium, and more strongly than either gentler preparation.
Not as much as the sentence above makes it sound. This is one assay, reported in a supplementary figure, using a single engineered reporter cell line built to light up when one particular receptor is engaged. That line is murine, not human, and it is not a gut cell. There was one dose and one timepoint. No health outcome was measured. What it demonstrates is that sterilization did not render the cells invisible, and that the harshest treatment left the most of that particular signal intact. It does not demonstrate that autoclaving is better, and nobody should read it that way.
Even bounded that tightly, it is a rare piece of work, because it is a real within-study comparison. Same organism, same batch, same assay, only the heat differs. That design is what lets you say anything at all about temperature, and almost nothing else in this field has it.
The other group that asked, and got a different answer
A Chinese group ran the same kind of comparison in 2023 on two commercial postbiotic preparations, one a Lacticaseibacillus, one a Bifidobacterium, across five temperatures from 70 up to 121 °C.
On a test-tube radical-scavenging measure, both held steady through the gentle range and dropped at 121 °C. For the first, scavenging fell from about 93 percent in the live cells to roughly 65 percent after autoclaving. For the second, from about 36 percent to 23. On an inflammatory signal measured in macrophage cells, the first preparation's 80 °C and 100 °C versions performed better than its autoclaved version. The second showed no temperature effect at all.
They also profiled the chemistry and found around a dozen compounds abundant at 80 and 100 degrees had almost disappeared by 121, phenyllactic acid among them. A handful of new compounds appeared that had not been there before.
So two groups have asked whether the temperature matters, and they got opposite answers on different readouts in different organisms. That is not a contradiction to resolve. It is the actual state of the knowledge.
One detail suggests they may not be measuring the same thing at all, and it is that middle step from earlier. The 2023 group measured their chemistry in the cell-free liquid the bacteria had been growing in. Lin's process centrifuges that liquid away before the autoclave ever runs. The two studies are largely measuring different fractions of the same organism.
Do Probiotics Have to Be Alive to Work?
For a large class of effects, no. The activity that depends on molecular shape survives the death of the cell, because the shape does. The leading review of heat-killed probiotics puts it about as directly as a review ever does: bacterial viability, or even cell-wall integrity, is not an essential condition for the intestinal effects of probiotics.
What does not survive is everything the organism would have gone on to do: the metabolites it would have kept producing, the substrates it would have kept fermenting, the way it would have competed for space against its neighbors. A live probiotic is an ongoing process. A postbiotic is a finished object. Those are different products, and you are trading one set of properties for another rather than getting a strictly better version.
The field has a formal definition. The International Scientific Association for Probiotics and Prebiotics defines a postbiotic as "a preparation of inanimate microorganisms and/or their components that confers a health benefit on the host." Inactivated whole cells qualify. Purified metabolites with no cellular biomass do not. The definition was contested on publication and the argument has not fully settled, which is a reasonable thing to know about a word that appears on a great many labels.
The Akkermansia Precedent: Pasteurized, and Tested in People
The RV-01 work stops at cell lines and rat toxicology. Nobody has given it to a person. A closely parallel story has already been through a human trial with a different organism, and it is the strongest evidence in this piece.
In 2017, that Belgian group reported something they described as unexpected: pasteurizing Akkermansia muciniphila enhanced its ability to reduce fat mass development, insulin resistance, and dyslipidemia in obese and diabetic mice. Those are the authors' own words, and in mice the comparative is theirs to make.
Two years later they took it into people. A randomized, double-blind, placebo-controlled exploratory trial: 40 overweight or obese, insulin-resistant volunteers enrolled, 32 completed three months of daily supplementation with either live A. muciniphila, pasteurized A. muciniphila, or placebo.
- Insulin sensitivity +28.6% (p=0.002)
- Insulinemia −34.1% (p=0.006)
- Insulin resistance score −32.6% (p=0.004)
- Plasma total cholesterol −8.7% (p=0.02)
- Body weight and fat mass: not significant (p≈0.09 for both)
Pasteurized Akkermansia, human volunteers, randomized against placebo, measurable metabolic change. It is the proof of principle for the category.
Now the part that is usually reported badly, and I want to get it right because the sloppy version is more flattering and less true. It is often written up as "pasteurized outperformed live." The trial does not support that.
The pasteurized arm was the one that separated from placebo. The live arm did not. But look underneath: the live arm's point estimate for insulin sensitivity was larger, at +42.4 percent against the pasteurized arm's +28.6. It failed to reach significance because only nine people completed in that group, and the noise around that estimate was roughly three times wider. That is a power problem, not an efficacy result. No statistical comparison of pasteurized against live was reported, because the trial was never built to make one.
And two results that cut the other way
The first is the sharpest argument against treating gentle heat as the safe default. In 2022, a team at UCLouvain characterized a newly isolated human gut commensal, Dysosmobacter welbionis, that prevented diet-induced obesity in mice. Then they pasteurized it at 70 degrees for 30 minutes, the same protocol that improved Akkermansia. It completely abolished the benefit. Same gentle temperature, same lab lineage, opposite outcome.
The second is about whether any of this survives contact with people. Earlier this year, a tyndallized Clostridium butyricum postbiotic that had performed well in animal models went into a randomized, double-blind, placebo-controlled trial in 96 adults with mild-to-moderate knee osteoarthritis, dosed for twelve weeks. Both groups improved. Neither beat the other. There were no significant between-group differences on any clinical endpoint, and no consistent movement in any of the biomarkers measured. The trial was industry-authored, which makes a null result more credible rather than less.
The one that failed was the gentle protocol, the conventional choice, in the largest human trial in this piece. Heat-killing a bacterium makes it stable and removes the risk of infection. Whether it does anything useful is a separate question, and the temperature you used is only one of the things that decides it.
Does That Mean Live Bacteria Are Worse?
No. And the reason it is not that simple is the most interesting thing I found while putting this together.
In 2026, a paper came out of the Levy and Thaiss labs describing this same species doing something considerably less flattering. Charting the mouse microbiome across a lifespan, they found P. goldsteinii rising with age, and transplanting an aged microbiota into young animals impaired their memory. Mono-colonizing germ-free mice with it induced cognitive impairment and blunted the hippocampal response to novelty. The controls were good: Alistipes, Lachnospiraceae, and Lactobacillus also shift with age and none of them did this.
Then they isolated the cause. They took the culture broth, filtered out everything larger than 3 kilodaltons so that no cells and no large molecules remained, and gave mice that filtrate alone. It reproduced the entire effect. The agent they identified was a medium-chain fatty acid, 3-hydroxyoctanoic acid, acting on a receptor on myeloid immune cells and driving inflammation that degrades the vagus nerve's ability to carry signal from gut to brain.
The damage was not done by the bacterium. It was done by something the bacterium secreted into its surroundings. And Lin's process, as you now know, spins the cells down and throws that liquid away.
So these two papers are not describing a live-versus-dead contrast at all. They are describing two different molecular libraries produced by one organism. The cell surface is one library. The secreted metabolites are another, read by an entirely different class of receptor. One paper studied the surface. The other studied the broth. They point in opposite directions because they are looking at different things.
- These are different strains. Lin used RV-01, a fresh isolate from a healthy human donor. The 2026 work used the type strain. Genome comparison puts RV-01 at roughly 98 percent identity to other P. goldsteinii strains, which sounds close and is not. Lin's own discussion warns that strains of one species can behave as double-edged swords.
- These are different models with no shared endpoint. Cell lines and rat toxicology on one side; mouse cognition, hippocampal imaging, and vagal recording on the other.
- There is no head-to-head, and the obvious experiment has not been run. Nobody has given autoclaved P. goldsteinii to an aging mouse and tested its memory.
- "Live is harmful" would be flatly wrong as a general claim. In 2025 alone, separate groups reported live P. goldsteinii ameliorating colitis, mediating the anti-obesity effect of a lysine-restricted diet, and warranting development as a probiotic candidate.
- Our reading that discarding the broth would remove the harmful fatty acids is an inference, not a finding. No heat-killed preparation was tested in the 2026 work. It is mechanistically reasonable. It is not demonstrated.
So this is an open question, not a verdict. What the two papers together suggest is that asking whether an organism is good or bad is the wrong question, and asking which of its chemistries you are delivering is the right one.
There Is No Right Temperature for Postbiotics
I think the field gets this wrong when it treats gentle heat as the obviously safer choice.
Keeping the cell structurally whole is not the universal requirement that the gentle-heat default assumes. It is a real requirement when the thing doing the work is a protein, and proteins do not survive an autoclave. One group found exactly that, and chose 70 degrees because the molecule they cared about holds up there. Another group used the same 70 degrees on a different organism and erased its effect completely. A third ran the same bacterium through three treatments and the harshest one held up best in their assay. A fourth found the harshest one cost them a measurable amount of chemistry.
Which means there is no correct temperature for postbiotics. There is only a correct temperature for a particular preparation with a particular active in it, and the only way to find it is to run the comparison and look at what came out.
Most of the field has never run that comparison. That is the actual gap here, and it is a bigger one than the argument about which number is right.
And once you accept that what you deliver matters more than whether it is alive, a second question follows: does it matter what time of day you deliver it? We went into that one separately, in Your Gut Is On a Clock.
From One Strain to the Whole Community
This is where the science meets what we are doing at Thaena, and it is where I have to be precise about what transfers and what does not.
Lin's team took one characterized strain, sterilized it, and kept the chemistry. ThaenaBiotic® applies the same logic at a different scope: instead of a single species, the collective metabolic output of a screened, healthy human donor's gut community. Hundreds of species. 10,000+ molecular signals: short-chain fatty acids, indoles, dipeptides, amino acid derivatives, and a great deal we have not finished cataloging. Donor screening first, then autoclave sterilization of the whole stool matrix, pH stabilization, and freeze-drying to hold those compounds intact.
The starting material is the same one fecal microbiota transplantation uses, and so is the underlying observation: that a healthy donor's whole gut community does something no isolated organism reproduces. Where we depart from FMT is on the mechanism. FMT moves a living ecosystem and depends on it taking hold, and donor strains measurably do take hold, though how much of the benefit that establishment actually accounts for is still an open argument. We move the chemistry that ecosystem had already made, and ask nothing of it afterward. Nothing living is transferred, and nothing has to establish for it to work.
There is a trade-off in that, and it runs against us in one important respect. A defined single strain is characterizable in a way a donor-derived community is not. Lin could sequence their organism's genome, name the missing enzyme, and point at a specific molecule. We cannot do that for every constituent of a whole-community preparation, and any claim that we can is overstated. The formal field definition of a postbiotic requires molecular characterization of the source organisms, and a preparation derived from an undefined microbial community sits uneasily against the letter of that requirement. We would rather state that than be asked it.
What we believe we gain in exchange is the emergent chemistry that arises only in a functioning community. A single strain makes what a single strain makes. A community makes things that arise from the interactions between its members: cross-feeding, sequential metabolism, molecules that no isolated organism produces alone. If what the host is responding to is the signaling environment of a healthy microbiome, then delivering that environment, rather than one contributor to it, is a reasonable thing to attempt.
We believe this is a plausible mechanism. The literature is consistent with it. It is a thesis, not a proven clinical outcome, and we have not run the randomized trial that would test it. We would like to.
What we have run
In C. elegans, the millimeter-long roundworm that most of what we know about the genetics of aging came from, we added ThaenaBiotic to the worms' environment and then challenged them with an oxidant. The treated worms stayed vigorous longer and lived longer under that oxidative stress, comparably to rapamycin and resveratrol. When we sequenced their RNA, the genes that moved were the oxidative-stress machinery: glutathione S-transferases, glutathione peroxidases, methionine sulfoxide reductase.
That is a worm, and worms are not people. What it tells us is that the protection was dose-dependent, and that at the level of gene expression the mixture moves something specific rather than something general. It does not tell us it will do the same thing in you.
ThaenaBiotic is one tool in your toolbox. Not the answer, not the missing piece. One input into an ecosystem that has a great many other things going on.
The Honest Limitations
- The temperature comparison is one assay in one supplementary figure. A murine reporter line, one receptor, one dose, one timepoint, no health outcome. It organizes a story. It does not settle one.
- No human data on autoclaved P. goldsteinii. Human cell lines, yes. Humans, no.
- No efficacy study of the autoclaved preparation in any animal. The animal work is toxicology in rats, with no benefit endpoints, plus germ-free colonization studies that used the live organism.
- Safety testing is not efficacy testing. That five-study dossier establishes that the ingredient does not cause harm at high doses. It says nothing about whether it helps.
- A cell line is not a gut. HCT116 is a colorectal carcinoma line. No immune compartment, no mucus layer, no resident microbiota, no nervous system.
- The study is industry-funded. It was paid for by the company commercializing the ingredient; the senior author co-founded that company. The work is GLP, OECD-guideline, and independently testable, so this does not invalidate it. It does mean this is a commercial safety dossier supporting a food-ingredient filing, not a disinterested academic characterization.
- The active-component attribution is explicitly speculative, in the authors' own word, and the capsular polysaccharide behind it was never isolated.
None of this makes the finding uninteresting. It constrains the language. What we have is a well-characterized single strain with a real safety file, a plausible mechanism, and one of the very few direct tests of whether the killing method matters. That is what the beginning of a real field looks like. It is not the end of one.
Frequently Asked Questions
Do probiotics have to be alive to be effective?
For a large class of effects, no. Researchers autoclaved Parabacteroides goldsteinii RV-01 at 121 °C for fifteen minutes, a protocol that leaves nothing viable, and the resulting powder still blunted the inflammatory response that E. coli provoked in human colon cells (Lin et al., 2024). A 2019 review of the field concluded that bacterial viability is not an essential condition for the intestinal effects of probiotics (Piqué et al., 2019). What does not survive is the ongoing chemistry a living organism would keep producing.
Are heat-killed probiotics the same as pasteurized probiotics?
No, and the difference matters more than the labels suggest. Pasteurization is moderate heat, around 70 °C for 30 minutes. Tyndallization is repeated heating and cooling cycles in a similar range. Autoclaving is saturated steam at 121 °C under pressure. Each preserves a different fraction of the cell: gentle heat can keep certain proteins intact, while autoclaving denatures them and leaves the carbohydrate and lipid surface. "Heat-killed" on a label tells you the cells are dead. It does not tell you what is left.
Does the temperature used to kill the bacteria change the result?
It can, and surprisingly few studies have tested it. Two that did reached different conclusions: one found its autoclaved preparation retained more signaling activity than gentler versions of the same organism (Lin et al., 2024), while another found 121 °C reduced activity and degraded specific compounds relative to 80 and 100 °C in two different preparations (Sun et al., 2023). A third group found that gentle pasteurization completely abolished the benefit of a promising gut commensal (Le Roy et al., 2022). The right temperature depends on the preparation and has to be determined rather than assumed.
Can dead bacteria still affect your body?
Yes. The sensors your cells use to notice bacteria read molecular shapes, not whether a cell is alive. Those shapes are mostly built from sugars and fats on the bacterial surface, which are far more heat-stable than proteins, so a killed cell still presents much of the same face to the tissue around it. How much of that face survives depends on how it was killed.
Are postbiotics safe?
That depends entirely on the preparation, because non-viable and safe are two different claims. It is why the RV-01 team ran a full safety program rather than assuming: three genotoxicity assays, a 28-day and a 90-day repeated-dose oral study, with no observed adverse effect at the highest dose tested (Lin et al., 2024). Sterilization removes the risk of infection, and in this case the antibiotic-resistance gene the organism carries was undetectable in the finished ingredient. That is a result someone had to measure, not something sterilization guarantees. It also does not by itself establish that a preparation does anything beneficial, and safety data should never be read as efficacy data.
What is Parabacteroides goldsteinii?
It is a common gut commensal. Its genus, Parabacteroides, averages roughly 1.27 percent of gut bacteria across a dozen human populations, and this species is one member of that genus. It has drawn research interest for producing an unusual surface lipid that dampens rather than triggers inflammatory signaling, and it counters obesity and colitis in animal models. It is not sold in consumer probiotic products today.
The Bottom Line
The story we were told about probiotics was that the benefit rides on the organism being alive. The evidence keeps suggesting that a good deal of it rides on the organism's shape, and shape survives things life does not.
But the question that turned out to be more interesting is the one almost nobody asks. Every heat-killed product on a shelf was made at some temperature. Somebody picked that number. And in most cases nobody ever went back and checked whether a different number would have left more behind.
- Dead does not mean inert. Autoclaved P. goldsteinii RV-01 was still recognized by host receptors and still suppressed inflammatory signaling in human colon cells.
- Most heat-killed preparations are made between 70 and 100 °C. Autoclaving is a minority choice, and the comparison that would justify either is almost never run.
- The two groups who did compare temperatures disagreed, on different readouts in different organisms. That is the real state of the evidence.
- Gentle is not automatically safe. Pasteurizing one promising commensal at 70 °C abolished its benefit entirely, and the one gentle-protocol product to reach a large human trial this year beat placebo on nothing.
- Autoclaving has one benefit nothing gentler can claim: it destroyed a detectable antibiotic-resistance gene that the live organism carries.
- Alive is a variable, not a virtue. The same species reads as beneficial or harmful depending on which of its chemistries reaches you.
- Thaena's thesis is an extension of this logic, not an equivalent of it. One characterized strain versus a whole donor community: more emergent chemistry, less characterizability. We believe it is a plausible mechanism. It is a thesis, not a proven clinical outcome.
We still do not know which molecules, in any of these preparations, are doing the work. Nobody does yet. The field is at the stage of having established that the effect is real and not yet having established what carries it.
That is an honest place to be.
References
- Lin T-L, Chen W-J, Hung C-M, Wong Y-L, Lu C-C, Lai H-C. Characterization and safety evaluation of autoclaved gut commensal Parabacteroides goldsteinii RV-01. Int J Mol Sci. 2024;25(23):12660. https://doi.org/10.3390/ijms252312660 · FREE FULL TEXT
- Piqué N, Berlanga M, Miñana-Galbis D. Health benefits of heat-killed (tyndallized) probiotics: an overview. Int J Mol Sci. 2019;20(10):2534. https://doi.org/10.3390/ijms20102534 · FREE FULL TEXT
- Sun Z, Zhao Z, Fang B, et al. Effect of thermal inactivation on antioxidant, anti-inflammatory activities and chemical profile of postbiotics. Foods. 2023;12(19):3579. https://doi.org/10.3390/foods12193579 · FREE FULL TEXT
- Le Roy T, Moens de Hase E, Van Hul M, et al. Dysosmobacter welbionis is a newly isolated human commensal bacterium preventing diet-induced obesity and metabolic disorders in mice. Gut. 2022;71(3):534–543. https://doi.org/10.1136/gutjnl-2020-323778 · FREE FULL TEXT
- Zhu Y, et al. The role of inactivation methods in shaping postbiotic composition and modulating bioactivity: a review. Foods. 2025;14(13):2358. https://doi.org/10.3390/foods14132358 · FREE FULL TEXT
- Depommier C, Everard A, Druart C, et al. Supplementation with Akkermansia muciniphila in overweight and obese human volunteers: a proof-of-concept exploratory study. Nat Med. 2019;25(7):1096–1103. NCT02637115. https://doi.org/10.1038/s41591-019-0495-2 · FREE FULL TEXT
- Plovier H, Everard A, Druart C, et al. A purified membrane protein from Akkermansia muciniphila or the pasteurized bacterium improves metabolism in obese and diabetic mice. Nat Med. 2017;23(1):107–113. https://doi.org/10.1038/nm.4236
- Lee HB, Lee J, Jeong K, Yang J, Jung YH, Moon JS. Randomized, double-blind, placebo-controlled trial of ID-CBT5101, a tyndallized Clostridium butyricum postbiotic, in adults with mild-to-moderate knee osteoarthritis. J Microbiol Biotechnol. 2026;36:e2512024. https://doi.org/10.4014/jmb.2512.12024 · FREE FULL TEXT
- Cox TO, Devason AS, de Araujo A, et al. Intestinal interoceptive dysfunction drives age-associated cognitive decline. Nature. 2026;652(8109):442–450. https://doi.org/10.1038/s41586-026-10191-6 · FREE FULL TEXT
- Salminen S, Collado MC, Endo A, et al. The International Scientific Association of Probiotics and Prebiotics (ISAPP) consensus statement on the definition and scope of postbiotics. Nat Rev Gastroenterol Hepatol. 2021;18(9):649–667. https://doi.org/10.1038/s41575-021-00440-6 · FREE FULL TEXT
- Li Z, et al. Therapeutic potential of the gut commensal bacterium Parabacteroides goldsteinii in human health and disease treatment. Front Microbiol. 2025;16:1618892. https://doi.org/10.3389/fmicb.2025.1618892 · FREE FULL TEXT
- Grangette C, Nutten S, Palumbo E, et al. Enhanced antiinflammatory capacity of a Lactobacillus plantarum mutant synthesizing modified teichoic acids. Proc Natl Acad Sci USA. 2005;102(29):10321–10326. https://doi.org/10.1073/pnas.0504084102 · FREE FULL TEXT
- Girardin SE, Boneca IG, Viala J, et al. Nod2 is a general sensor of peptidoglycan through muramyl dipeptide (MDP) detection. J Biol Chem. 2003;278(11):8869–8872. https://doi.org/10.1074/jbc.C200651200
This post accompanies the Lit Review Friday episode of Learn Something with Thaena.