> he used an LLM to write his rebuttal post, and this was sort of a “hallucination”.
The more I think about the latter possibility, the more likely it seems. It fits the pattern of confidently wrong assertion-with-citation that you get from current models, Scott just mentioned dealing with o3 hallucinations in the writing of a recent piece, and he seems like the kind of guy who would have a custom GPT trained on his corpus.
----
I believe he's said previously that he does this. I can't search my old comments on Substack, but I had a conversation ~6 months ago with someone after he issued a correction like "o3 said this so I included it, but it's actually not true at all, my bad" for something that was obviously not true if you spent more than ~20 seconds thinking about it. I'm really worried that AI is slowly replacing the necessity of struggling with hard or controversial concepts, and taking the benefits of that struggle: understanding, along with it.
The interesting thing about your stuff is that it actually seems testable in-vivo. Unlike other crackpot theories, which are usually unfalsifiable or make no clear hypothesis at all, it seems relatively straightforward to me, a layman, that you can just act on someone's gut microbiome to introduce strains of bacteria that aren't currently there. Add the cholesterol-eating bacteria (or whatever the mechanism is, I remember it increases cholesterol in urine or something like that), and see if someone's cholesterol decreases. If it does, then that's pretty strong evidence that the theory is sound, and repeated intervention over a larger group would slowly turn it from hypothesis to fact. I'm sure it's actually much more complicated than this, but whatever the specifics, it seems like there's a broad-strokes path to falsifiability, so your stuff is therefore quite interesting.
Y'know a few weeks ago Freddie de Boer put out a live debate challenge to Matt Yglesias, and one of the comments was like "I wanna see this debate but is there a reason it has to be live, like what's wrong with written exchange" and one answer that I didn't see brought up is that it at least lets you know nobody's serving you LLM slop
> The more I think about the latter possibility, the more likely it seems. It fits the pattern of confidently wrong assertion-with-citation that you get from current models,
to be fair, Scott was like this long before modern LLMs were developed.
I'm no microbiologist, as a lowly purely clinical MD this is helpful in spelling out your positions... Seems to me this is absolutely worth further research, though if I were consenting a treatment-resistant schizophrenic for an FMT trial, it would sound something like "there is a pretty bizarre and unproven theory this might cure you, (or maybe possible make you less treatment resistant?), but we think it's worth a try, even if unlikely. Do you want to be the one to try?" Not sure how the IRB will like the proposal but I'm for it. If you are not already in touch with people doing psych clinical research I could *attempt to connect you but not at all sure if I'll get traction with that department - no close contacts!
As others have commented I think you probably oversell your case a bit which leaves it a bit vulnerable, though perhaps it is a useful strategy to draw in attention, too. I think there is probably more compatibility between your theories then acknowledged -- you are not claiming that our genetic immune factors do not play a role here, and the prevailing theory already includes hand waving about environmental triggers. We don't have much of a disease model to contextualize this within other "dysbioses" and so your theory might sound as if you are calling it an "infection" and I think it is fair to object this topic would seem to be more complex. I hope we DO start to have an understanding of that as a disease class to maybe overturn some of the "functional" disease classes which have been growing!
For a relatively healthy person, I'm starting to appreciate that I would Not want to monkey around on the cutting edge of biohacking around this topic. I might be way off base here but if this is part of our genetic inheritance as you say, and something more mutable then the rest, is it possible that globalization in addition to antibiotics is actually part of the origins of the modern maladies? Like, for me, as someone with a family history of both MS and AS, I really do not want to accidentally run into any ETX or any Dialister either, if I'm somehow lucky enough not to have those lurking in my system so far... and maybe my family members who did encounter it came from an inheritance where that was not the prevailing immune problem they were set up to handle... maybe thats way off but it's a question I have and suddenly makes me have some Chesterton fence timidness, despite being bullish on the field generally.
>you are not claiming that our genetic immune factors do not play a role here, and the prevailing theory already includes hand waving about environmental triggers
So that's the funny thing, right? Like, if your definition of "80% genetic" leaves 75% of real-world MZ twin concordance unexplained...fine, I guess we're saying the same thing, so I think I've found the other "20%" and can we please focus on that.
>For a relatively healthy person, I'm starting to appreciate that I would Not want to monkey around on the cutting edge of biohacking around this topic
It's very funny when people e.g. try a new vitamin regimen and call it "biohacking"—like I'm sorry dude but if there's no risk of corrupting the filesystem you're not really "hacking" anything. But to your point: yeah I am happy and healthy and while I'm willing to test stuff on myself because that's the only really ethical way to do a first-in-humans for most things, I wouldn't be doing it just to try and "optimize" anything. My goal is of course to make it so that it's less like hacking and more like the App Store, where if you want an upgrade, you can just install the thing and trust that it's going to work.
The Dialister thing is interesting, I wanted to ramble on in the post about it but the example had gone on long enough lol. Thing is: correlation is not causation! I said Dialister and Phascolarctobacterium are the only two that really eat succinate, and this is true, but some other taxa use succinate in auxiliary ways; it's not their main fuel but a metabolic accelerant for things like C. diff and Veillonella (jury is out on Veillonella but personally I'm sus of her).
So it may be that Dialister isn't really bad, just a poor man's Phascolarctobacterium, because it's not very good at its job; apparently it doesn't take up and consume succinate as quickly, which might make it easier for an opportunist like C. diff to get a foothold. I think part of ecosystem resilience is having every niche well-filled, so that everything you eat breaks down into stuff that either you or a symbiont can use.
I've had an idea in my mind for a while now, to make a deckbuilding game to learn/teach this stuff (things like Magic the Gathering and Balatro are the only way I've ever seen people learn the names and properties of 400 different lil guys by heart). The core mechanic would be about building combos—like, if lunch today features arabinose and glucose, your Bacteroides turns arabinose into acetate+succinate, and then a Faecalibacterium can use the acetate + glucose to make butyrate, and maybe that's points. But I think one of the core mechanics would have to be about making sure there's nothing left "on the table", so you'd wanna have a Phascolarcto in there too, to mop up the succinate. I doubt anyone would actually want to play such a game, and there's probably a way to have a computer do the metabolic network modeling, but where's the fun in that?
>is it possible that globalization in addition to antibiotics is actually part of the origins of the modern maladies?
The deck building game sounds... so fabulous but entirely unapproachable lol. An interesting way of trying to map this though, for the dedicated. Yes I've been following your blog for years, actually I got to meet you briefly back in '23 at vibecamp :) ... and to be fair I suppose the concept is really not novel - in med school we learned the model of autoimmunity that a susceptible host meets a bug and triggers and aberrant response. Amazing you've been identifying real targets now. I've just been in the mode of focusing on getting diverse plant fiber to support a diverse microbiome as the best current practice - and it's a good reminder that it is not really the more diverse the better. As well as getting an infectious side effect you could give yourself the ticking time bomb. I suppose I was just having a brief daydream of getting out of the sick care system and getting more involved with this stuff myself, but it's not quite ready for broad clinical use, best to wait for your app store to develop ;)
Also yeah, I know it'd be a tough sell to literally anyone lol. But yes please put me in touch with your clin psych research contacts if you don't mind. (shoot me an email at SDSkolnick@gmail.com)
I'm on your side, and this looks like a convincing rebuttal.
But if you're a lone outsider trying to single-handedly force a paradigm shift, you can't afford snark, put-downs, or unprovable accusations of bad faith. These are usually good excuses to dismiss someone as a credible actor, and a man in your position doesn't get any excuses. A single lapse in message discipline can tank your entire project.
> But if you're a lone outsider trying to single-handedly force a paradigm shift, you can't afford snark, put-downs, or unprovable accusations of bad faith.
I'm not sure that's true anymore. Before the post-Manhattan project bureaucratization of science snark and put-downs were rather common on scientific writing. Now between the rise of the internet and the current administration's attacks of the science bureaucracy we may be getting back to the old system.
I just wanted to note that, regarding concordance, there's an environmental term that I'm pretty sure Scott is implying, given his spreadsheet. It can be additive or multiplicative, depending on the mechanism. Critically: if the environment changes the prevalence of the disease changes, and so do the concordance rates.
So you can't calculate concordance rates from just knowing the percent of the variance that is explained by genes.
I'm not saying that any of this is material to your argument.
Also, I'm a huge fan of phage therapy which has always seemed on the cusp of fruition for the last 20 years. So given it's perpetual nascence maybe it's a bad thing to consider, but it seems like the perfect tool for launching a targeted strike against a particular strain of gut bacteria. (Several phages might be needed, since some strains of e.coli are immune to a given phage that others are susceptible to.)
My personal assumption is that other microbes, like T. gondii, also contribute to schizophrenia. And the effects of T. gondii are pretty well known. But if R. gnavus abundance contributes to a schizophrenia risk score then this should be one way to test the matter. Right?
yes! Subtracting is always gonna be more difficult than adding to a microbiome, but phage therapy is probably the way. The reason it's perpetually on the cusp is that, exactly as you point out, it doesn't scale easily. But as a species we're putting in a lot more squeeze for a lot less juice on plenty of fronts. Just about convincing the right people to allocate the resources effectively.
I am glad you wrote this, since the rebuttal + further comments exchange on ACX seemed like a lot of Bayes damage. And it was generally well-written and entertaining! But I notice that I am still not convinced - that these answers to most, but not all, of Scott's objections feel unsatisfying. There's a reciprocity between the two pieces, of both parties swatting away trivially inconvenient annoyances to their preferred models and then going on to further explicate them. I do take your point that the original Holy Grail post was just the tl;dr version, and it's to your credit that you include some of the missing pieces here. Strong claims require strong evidence! But the bar is really quite high, and while stronger, as a layman I am not sure even this 2.0 corpus clears it. (It is entirely possible that a combination SME and math deficit means I'm missing the meat of the argument, and from an expert point of view it's a slam-dunk...yet it's part of the Rationalist training to notice that gut feeling* of "this argument doesn't seem fully coherent", and not be intellectually satisfied until that itch is properly scratched.)
Probably gonna flesh this out later into a point-by-point. Preferably before Scott does a Contra Contra ("No, Rescuing The President Is Not Manliness-Dependent"), those never go anywhere and just lead to mutual retrenchment with no Yes, And.
*pune, or play on words, not intended, I am definitely stealing Branch-COVIDIAN though
Ended up procrastinating on a response due to a bad bacterial infection (yeah, ironic) and am ashamed at now having been scooped by Scott anyway. Lucky Hunter made a lot of great points too, in both directions. Since most of the technical minutiae have been addressed one way or another...
1) Concordance and missing heritability: I wish to note that Scott didn't write just the one article on schizo genetic inheritence that he linked in the rebuttal (which, yes, he did goof bigly on + got called + added to Official List Of Mistakes). It was a whole series of posts over a good length of time, going back to SSC, complete with other interlocutors that he also Contra'd, oblique references to Hey Guys Remember When I Demonstrated Schizophrenia Is Mostly Genetic, and recurring callbacks like the toxoplasma thing. It's "part of the brand", so to speak. Therefore I don't find it surprising at all that he responded quickly and strongly to your post - it's good grist material to springboard one of his own, you're a known reader, author and audience have overlapping interest in the subject, etc.
2) Accusations of LLM use: come on, this is silly. For starters, it's an obvious Don't Threaten Me With A Good Time. As an AI booster, the valence on "he sometimes uses LLMs!" against Scott is...probably not what you'd intend? It'd be like making the claim that Zvi Mowshowitz uses Twitter to find source material. Scott also explicitly claimed he didn't "[make] extensive use of AI" for any posts prior to Missing Heritability: MMTYWTK* , and that it was employed for research rather than writing. Now, Contra Skolnick did come out a few days later, that's true...but it seems really dumb to take flak for using o3 (both on that post and in the earlier Contra Cowen On USAID), write a good defense laying out that Actually, No, I Know What I'm Doing With These Tools, and then hypothetically have ChatGPT slop out a pro forma reply with such poor QC that he'd get called on it. Does Scott sometimes do dumb things? Sure! But since I'm not convinced on the speed argument (the man is not exactly known for the slowness of his writing), and you and I are both unwilling to assume malicious lying...
...that leaves us with basically a vibes argument, the penumbras and enumerations of LLM. Once upon a time, if a writer made a mistake citing his own corpus, we'd call that "making a mistake". In current_year, it's a harbinger of LLM slop, because...uh...humans are usually infallible at recollection of noncentral facts that only need to be directionally correct to support their arguments? (Again, valence issue, Scott fucks up way more often than just this one time!) If we're armchair quarterbacking, then I'd counterpropose that that was a particularly egregious and embarrassing error for Scott, so he'd be more rather than less liable to conveniently forget the correction. Likewise, I get the sense from some of your other comments that you're particularly skeptical of AI, at least when it comes to writing. Are you sure this isn't just reaching for a convenient Axe of Grinding? Are there any other smoking finger gunprints of LLM malfeasance in the post? If he's hypothetically got a custom-trained one, it sure does a damn good job of sounding just like Scott...(and as another commenter notes, it's not like "confidently wrong assertion" is a novel tone either)
3) Fuzzy gnavus: look, not to pull a tu quoque, but it's weird to do a bunch of throat-clearing about one's empiricist boner fides and then just go ahead and Roll To Disbelieve against the conclusion of the paper's actual authors + sundry others. They already did the experiment! And as others have pointed out, it's weird that we haven't had an antibiotics -> gnavus wipe -> miraculous scizo recovery case study to date, outside of like one maybe-sort-of-if-you-squint-and-play-reference-class-tennis case. You claim this would be very difficult and/or unethical to intentionally test with confidence, and thus probably won't happen...but that's the same Fully General Counterargument of disprovability you complain about wrt particle physicists' hedging! I do think you're on solid ground with your more general claim that Antibiotic Overuse Is Really Bad, Acktually, and it seems easy enough to directly test gnavus-cide if one bothers to try. But this really feels like a lamplight key search situation (though it's not like keys are never found near lampposts). Even if one accepts the Model 2 of many gnavus, now handle it -> severe symptoms -> clozapine, it still feels like one of those "necessary but not sufficient" constructions. Increased gnavus prevalence being found in all sorts of other diseases, few if any of which seem to have similar relation to tryptamine/phenethylamine levels, makes it seem kind of like the "vitD gets depleted while sick but we can't really prove D supplementation cures people faster" thing.
4) Adopt-A-Microbiome: Scott says no correlation between adopter and adoptee, instead it's between adoptee and bio parents. You clap back that There Is No Evidence that non-schizo adoptees acquire the schizo risk of their schizo adopters. Which...what? I don't understand if you're making the assertion that there *is* evidence, we just don't find it because sample size is too small to establish base rates (I'd imagine schizophrenia is a good way to get screened out of adopting *and* adoption? But surely it's happened a nonzero number of times in both directions?). Either way, I'm confused about how this is a particularly objectionable objection. Someone putting their bet on Genetic should be expected to bring out a strong line of argument like correlation with bio parents, no? Likewise, I'd have expected a counterargument from you more along the lines of "actually, the microbiomes of adopters and adoptees are more similar than adoptees and bio parents". Which would dovetail nicely with Scott's link that spouses have more similar microbiomes than siblings! But that shot wasn't taken, and instead it was...somewhere between a non-sequitur and a tone argument. Why?
Do wish to reiterate humility here: almost all the actual science parts of this go way over my head, I'm probably missing a lot, this is just me nitpicking weird structural deficiencies in the elephant rather than helping flesh it out. I am glad it is being fleshed out though! Even if you or Scott turn out to be dead-wrong about everything, it's important to test hypotheses (and existing consensus!) against Worthy Opponents. A productive and entertaining exchange of ideas, with even technically illiterate idiots like me leaving a bit more informed. Would that some of your other spicy takes got the same sort of trial-by-fire...you've got so many ideas that are just on the cusp of being actionable, and "sound right", but need more refinement.
Re: whether psychosis feels like a psychedelic or an amphetamine, I don't have schizophrenia, but have bipolar I and have gone through mania/psychosis 3 times.
My mania *absolutely* resembles a psychedelic / stimulant experience - to the extent that the first time I had a psychotic break, for part of the lead-up I was convinced I'd been dosed with LSD.
Imagine being on a kind of euphoric psychedelic + ritalin (or what I'd imagine meth to be like) for days/weeks/a month straight, and completely losing touch with reality as the trip gets deeper- transforming into an agitated, 'bad trip' over time.
It's unique, but it doesn't feel categorically different. The closest analogue I have to early mania was when I tried DMT and didn't "break through," plus a stimulant.
The things that do stand out are the hyper-confidence, the grandiosity, the feeling of the conscious experience radically speeding up, and the partial memory blackout.
It just so happens that I – a fiftysomething male who's never had symptoms of schizophrenia – had my gut biome profiled 3 times in 2024. All samplings showed a relative abundance of R gnavus significantly above the 0.1% to 0.3% loosely conjectured as 'usual': 4.000% in March, 1.820% in September, 7.735%(!) in October.
That last value is potentially near the 6+ "log fold change" differential abundance that's associated with tough schizophrenia cases in the paper [Vasileva, Yang, Baker, et al 2023] that seems to have triggered your own further analyses.
While I've not formally logged my mental state for fine-grained changes, I don't personally recall – nor have any family/friends/professionals/etc mentioned – any mental disturbances around the time of highest abundance.
I'll be retesting soon, so may then have a better idea how much of a brief outlier that 7.735% reading may have been. That sample was a couple weeks after a surgical procedure & recovery involving antibiotics – it was my curiousity about such effects that motivated that test, which in general showed very mild changes from pre-procedure biome composition.
Fantastic, thank you for weighing in. To know how useful of a result that is, I'll have to do a bit of a dive on TinyHealth's methods. (For one thing: how is sampling done?) But as you mentioned in the tweet thread, lots of potential for interactions where trace amine production is necessary but not sufficient.
The detail & level of explanation/qualification in their "deep shotgun metagenomic sequencing" reports has impressed me: https://www.tinyhealth.com/technology
For understandable reasons, their 'action plans' emphasize more-anodyne interventions like diet changes, supplements, & "get dirty - exposure to nature can benefit gut health" – rather than, well, the title of your blog. :)
1. Re the simulation - I explained that this was just an intuition pump for why 80% genetic shouldn't equal exactly 80% twin concordance. I linked a paper which did the full calculation and got a pretty reasonable result - https://genepi.qimr.edu.au/staff/nick_pdf/Classics/1970_Smith_AHG_Dorret.pdf . As far as I can tell, you haven't rebutted the claim (either from my discount simulation or the real paper) that this establishes "low twin concordance = non-genetic" is a bad way of looking at things - you've just pointed out unrelated errors in the hope that people confuse this with the main point being wrong. If we continue this discussion, I'd prefer you to either assert that you still think low twin concordance has to mean non-genetic, or else agree that you've abandoned this argument.
2. The argument here is that, since there is missing heritability in everything, schizophrenia is not special, and does not require a special explanation.
If there is some reasonable explanation for missing heritability (hard-to-find genes, GxE interaction), then there is no puzzle around missing heritability in schizophrenia, and no motivation to search for a gut bacterium that causes it.
If there is no such reasonable explanation, then you either need every trait to be caused by gut microbes, or by something equally weird and mysterious and hard to find. And if there are so many different weird/mysterious/hard-to-find causes, again, there's no motivation to go searching in the gut for a cause of schizophrenia.
This doesn't prove that schizophrenia isn't a gut microbe, it just dismantles one argument that might motivate people to think it is, bringing this back to the prior.
3. You don't really seem to deny that it's weird that schizophrenia starts with a trigger episode between 18 - 30, you just say that sometimes bacterial composition in the gut changes (but why would it happen between 18 - 30 so often, and so often change in one direction?) I agree that the neurodevelopmental theory of schizophrenia hasn't been completely fleshed out (although it is fleshed out more than you seem to think, see for example the section What's Happening In Adolescence at https://psychiatryonline.org/doi/10.1176/appi.ajp.20240305 ) but most complex diseases aren't completely understood.
4. You're putting in all this work to show that it isn't genetic, and then when I show that it behaves like a genetic disorder, retreating to "okay, it's genetic, but the genes cause the microbiomes". Because of how concordance declines from identical to fraternal twins, it's not just vaguely genetic, in some sense where genes maybe influence it a little. It has to have 80% of the variance explained by genes! If you want to say those genes all code for microbiota, fine, but then you've abandoned all the stuff you were saying earlier about how you think there are holes/inconsistencies in the genetic picture. Instead you're trying to have it both ways - claiming it's not genetic so you can recycle the missing heritability arguments in favor of there having to be some other cause, then claiming it *is* genetic when I point out how it behaves in every way like a genetic disorder!
4b. Regarding the lack of an adoptive-child-to-adoptive-parent schizophrenia correlation, you only say:
"Can we think of any reason why there might not be a lot of studies showing that children adopted out from mentally healthy parents to parents with schizophrenia acquire the schizophrenia risk of their adoptive families? Any reason at all."
I'm not sure what you're implying here, but I think maybe you're saying nobody would ever adopt a healthy child out to schizophrenic parents?
But see any of the studies that have been done on this issue, for example https://www.sciencedirect.com/science/article/abs/pii/0022395687900896 , for how they handle it. I don't understand what you think you're doing here, when the studies have been done and come up with the result you're mocking as impossible.
You continue to assert that "practically all the major schizophrenia risk genes are immune genes", which is false, and which many commenters and I called you on the last time you asserted it.
5. I didn't mention the clozapine dose-dependence at all and it's not a big part of my argument, so I don't understand why you're so interested in refuting it.
In any case, you only provide a possibility proof that the direction of causality *could be* microbes -> clozapine.
My argument was that it's equally possible (and, in fact, more plausible according to everyone who has studied this), that the direction is antipsychotics -> microbes, therefore the presence of unusual microbes in the gut of people with schizophrenia provides basically no evidence that schizophrenia is a microbial condition.
This is a big part of my complaint in general. You have two main pieces of evidence in favor of your microbial hypothesis - the missing heritability, and the presence of weird microbes in schizophrenics' guts. I've shown that there are alternative better explanations for both of them. You instead try to argue that it's still *sort of possible* that microbes could be an explanation. But even if this is true, so what? We're just back to our prior that any given condition could be caused by microbes. And in this case we have many many other pieces of evidence that this one isn't - for example, all the risk factors, the lack of response to antibiotics (which IMHO you fail to defuse above), the developmental course, the non-resemblance in non-genetic relatives, etc. So it just doesn't seem interesting at all that, by stretching really hard, you can sort of trace a course by which if a lot of things go perfectly it's still possible that there are microbes involved.
6. I feel like you've tried to dance around all of my points and are still confused about basics of how genetics work. If I'm going to believe crazy hypotheses about schizophrenia, I'm going with E. Fuller Torrey's thing about how it's all cat feces, at least that one is funny.
Obviously there are genetic factors involved in disease risk. I am arguing that, to the extent that the genome is useful in predicting risk, this is mediated by its impact on the microbiome.
I am saying this is the most likely-seeming "environment" component of "gene-environment interactions".
>You continue to assert that "practically all the major schizophrenia risk genes are immune genes", which is false
Are you arguing that the MHC is not an immune complex? Or are you arguing that it's worth considering things outside the MHC, when the MHC signal is TEN SEPTILLION TIMES stronger than the next-strongest signal?
Did you think I was taking poetic license, with the title of the article?
Do the math. How far would we have to move the *actual sun* away from the Earth for it to be ten septillion times less bright? It's further than alpha Centauri.
Spending time thinking about the influence of these other genes on schizophrenia risk is, very literally, a bigger waste of time than studying the impact of starlight on skin cancer risk, or putting out a solar panel at night.
Thank you for bringing Kety 1987 to my attention. I looked for anything like that, but couldn't find it. But it does not support your claim that a child adopted out from mentally healthy parents to parents with schizophrenia is not at increased risk.
It suggests that a child adopted out from parents with schizophrenia to mentally healthy parents is still at increased risk. Pretty sure the mistake you're making here is called "confusing the inverse". https://en.wikipedia.org/wiki/Confusion_of_the_inverse
Oh, DID you use an LLM in writing "Contra Skolnick"? Or did you come up with that ending all on your own
I'm not yet convinced of his specific proposed mechanism, but I do think it is highly likely that schizophrenia and a whole host of poorly understood diseases will turn out to have microbial causes. I think you're being a bit too dismissive of this point.
Loads of infectious diseases have high heritability, plenty of times in the realm of 50-80% (https://www.mdpi.com/2073-4425/16/2/177). I suppose you could describe tuberculosis or polio or measles as a complex disorder with many genetic, environmental, and microbial factors influencing them, but I think you'd agree it would be strange to do so. A clearer description is that they are caused by microbes, and there are genetic and environmental factors that affect one's risk.
I agree that low twin concordance doesn't mean that genes are unimportant drivers of trait variation. However, identical twin concordance of less than 100% means that some sort of environmental factor is necessary, wouldn't you agree? If you did a twin study of sickle-cell anemia or cystic fibrosis, you'd find 100% identical twin concordance. Microbes are not the only relevant environmental factor that can affect disease risk, but microbes are the cause of the vast majority of diseases we understand the causes of.
As I understand it, the possible explanations for missing heritability basically come down to some combination of gene-environment interactions, gene-gene interactions, rare genetic variations, or violations of the equal environment assumption. Disease mediation by microbes would fit the category of gene-environment interactions. I can also imagine a role for more similar environmental exposures of identical twins causing increased concordance for microbes compared to for some other possible traits.
A key area I would disagree with you is that we should expect similar reasons for missing heritability for diseases compared to most other traits. Something like height or IQ varies continuously in a more-or-less normal distribution, which makes sense if they are the sum of a large number of small genetic and environmental effects all adding up. Most diseases are not like that, though. Yes, severity of symptoms can vary, but you pretty much either have the disease or you don't. I know you talk about modelling it as a continuous trait that gets expressed once it crosses some kind of threshold. What does this actually mean in the real world, though? What causes this threshold? What makes some people schizophrenic when their identical twins aren't? What makes people with schizophrenia exhibit psychotic symptoms sometimes, but then these go away? It seems like this would necessitate some kind of environmental trigger, and a microbe would serve that role well. Do you have any alternative hypotheses of what it would be? Do you have any other examples of a binary trait outcome affected by a large number of genes that we actually understand how it works (as opposed to poorly understood diseases that might still turn out to be caused by microbes)?
I think a lot of the missing heritability of diseases (that strike young and have a big fitness effect, specifically) will turn out to be microbes, which I don't think will necessarily be the case for most other traits. In addition to the threshold effect making diseases very different from continuously-varying traits, natural selection ought to make genetic diseases rare under most circumstances. Exceptions would include traits with a heterozygote advantage, those that only cause a disease under evolutionarily novel circumstances, or those that mostly strike in old age. Do you think that these explanations apply to schizophrenia? If not, why do you think natural selection has not make it substantially rarer?
What are your main criticisms of Torrey's hypothesis about Toxoplasma causing a subset of schizophrenia cases? Is it just the same points about it being highly heritable, or is there something about the hypothesis you feel doesn't fit with the data? Toxoplasma famously evolved to get into rodents' brains and manipulate their behavior to increase their chances of getting caught by cats. It can even do that to our closest relatives, chimpanzees (https://www.cell.com/current-biology/fulltext/S0960-9822(15)01517-1?). If it can affect chimpanzee behavior, it doesn't seem far-fetched to me that it could get into human brains and start messing things up. Do you have an alternative explanation for the association between cat exposure and schizophrenia?
I wouldn't be incredibly surprised if a microbe was involved in some way. I disagree with Skolnick on two points:
1. Skolnick dismisses the genetic evidence as false. I think the genetic evidence is very strong. As you point out, genetic evidence is potentially compatible with it being a microbe; I think if it's a microbe, it will be be a microbe with strong genetic liability, and Skolnick's arguments against the genetics will still be wrong. I think this means that one of Skolnick's affirmative arguments for it being a microbe is false.
2. I don't think there's a strong case that it's a gut microbe in particular, let alone the specific gut microbe he mentions. R. gnavus is in lots of people, there's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes, and it doesn't really behave the way I would expect a schizophrenia cause to behave.
Maybe a good example of this is rubella. We know prenatal rubella infection can cause schizophrenia! There's your microbial cause, hooray! But nobody cares about this because it's a long and complicated pathogenesis process and rubella infections only fully explain a ~single digit percent of schizophrenia cases. I think other microbial causes will end up pretty similar - anything that damages the brain increases liability, lots of different microbes can damage the brain, but we still need to figure out overall what the disease *is* and what these lots of different forms of damage are doing.
This is also why I don't think Torrey's toxoplasma open-and-shut solves the case. We've checked what percent of schizophrenics have toxoplasma. It's both more than you would expect by chance, and much less than 100%. I think in the end this will turn out to be something like rubella - yes, it makes the odds worse, yes, it causes some single digit percent of cases, but we still need to figure out what the disease is and why there are all these non-necessary-and-non-sufficient risk factors.
For comparison, consider something like heart attacks. What *causes* heart attacks? Is it eating donuts? Certainly those don't help. But lots of people get heart attacks after never having eaten any donuts in their lives, and lots of people eat donuts but don't get heart attacks. There are actually plenty of infections that increase risk of heart attack by damaging heart tissue, but again, lots of people get heart attacks without getting any of those.
When we try to figure out "what heart attacks are", the goal isn't to find the One Food That All Heart Attack Victims Eat But Nobody Else Has Ever Eaten. It's to understand what the heart is and how it works, identify the specific disease process going on, and then in retrospect it will be obvious that donuts and infections and everything else all contribute a bit to that disease process.
This is quite different from what we're doing in eg AIDS, where we did discover that it was all just one virus. I think the fact that we discovered that pretty quickly in AIDS, but have been studying schizophrenia for centuries with only these suggestive leads, should make us expect schizophrenia to be more like heart attacks than like AIDS.
>There's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes
The things I disagree with you on are you saying that the low twin concordance rate disproves anything, and you saying that missing heritability disproves anything. If you have dropped those two claims, I think you should say so explicitly (or if you did say that somewhere, I missed it - please point it out).
>> "There's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes"
It doesn't seem circular to me. Suppose you said "I believe in Bigfoot because people report seeing large footprints". I demonstrate that large footprints can also be caused by bears, and say "There's no particular footprint evidence for Bigfoot beyond that which can be explained by bears". Is this circular reasoning?
The low twin concordance rate is not proof or disproof of anything. It is evidence that, if we seek a satisfactory explanation for this disease, we must look beyond the genome. Happy?
Regarding Bigfoot: the reason I say it's circular is that, in this analogy, the only evidence you have for the existence of these so-called "bears" is the footprints!
If you could produce a bear—if we had, say, microbiome data from a study of clozapine in healthy volunteers which showed an increase in the relative abundance of Lachnospiraceae, then I would happily grant you bears. If there were animal studies showing something similar, I would grant you bears. I would even take something as tenuous as an in vitro effect, e.g. if it had been shown that clozapine inhibits the growth of Agathobacter rectalis, because it's well known that A. rectalis abundance is inversely associated with that of Ruminococcus gnavus. But you do not have so much as a bear claw donut; you have not "demonstrated" anything, merely asserted it.
So we are both proposing hypothetical megafauna, to explain these footptints—the association between abundance of trace amine-producing bacteria and schizophrenia. The difference is that your model ascribes to coincidence the fact that these organisms are tightly associated with serum tryptamine and phenethylamine in the general population.
> "The low twin concordance rate is not proof or disproof of anything. It is evidence that, if we seek a satisfactory explanation for this disease, we must look beyond the genome. Happy?"
No! We're still talking past each other. My claim is "the low twin concordance rate is exactly what you would expect from a ~80% heritable disease, and provides exactly zero evidence for any additional non-genetic cause". If you disagree with that claim, I continue to think you're dismissing the genetic evidence. If you agree with it, I think you shouldn't say things like "it's evidence that we must look beyond the genome".
> "Regarding Bigfoot: the reason I say it's circular is that, in this analogy, the only evidence you have for the existence of these so-called "bears" is the footprints!"
Let's tighten the analogy. Suppose we're talking about Local Forest. There are big footprints. I'm not sure that there are bears in Local Forest in particular, you're not sure that there are Bigfeet. But there is a high prior on bears (because they're a common well-known animal and likely to be in any given forest), and a low prior on Bigfeet (since they are previously unknown to science and there are lots of reasons to expect we would have seen them if they existed).
If our only evidence beyond priors is something which is explained by bears and Bigfeet equally well, then we should be almost certain that there are bears and that there aren't Bigfeet.
Antipsychotics disrupting gut bacteria is the most plausible thing in the world, like bears being in the woods. But schizophrenia being a gut microbe requires every past scientist to have missed something obvious, contradicts much of what we know about schizophrenia (like its non-response to antibiotics and its association with various neurodevelopmental insults), and there's no other reason to believe it besides the claims about twin concordance/missing heritability (which I say are also wrong, and rightly understood provide no evidence) and a very speculative theory about how the disease sort of resembles tryptamines (which I don't think it does). Like Bigfoot, it's something we wouldn't believe unless we had great evidence. The bears explain the gut microbe prevalence evidence just fine on their own, so merely proving bears plausible should be enough to make us forget about Bigfoot.
> "If you could produce a bear—if we had, say, microbiome data from a study of clozapine in healthy volunteers which showed an increase in the relative abundance of Lachnospiraceae, then I would happily grant you bears."
On this single tangent: "but why would [bacterial composition in the gut change] happen between 18 - 30 so often, and so often change in one direction?"
Ages 18-30 is likely a period of especially-high variability in all infectious & biome-changing categories: it's a time of relocations, new living arrangements, new workplaces, new relationships, new diets. And: probably a few courses of antibiotics for various conditions.
And simply lots of "re-roll" random walks, where *some* imbalances prove both persistent & noticeable, could be enough to create the impression of things happening "in one direction".
Screening the gut biomes of early onset (teen or before) & atypically-late (40yo+) cases could provide a strong signal whether any bacteria might be relevant there.
Thank you for the feedback, both on the argument and the tone.
I can afford a leeeeeetle snark (I think a gentle ribbing about the length of ACX's posts etc. is pretty harmless) and I try to spend within my means. I don't think I accused him of arguing in bad faith at any point. Did I come across that way? Or was this just a preemptive warning?
I think the main area where you come across as making accusations of bad faith is when you imply he is lying (your meme even says "lie about simulation results being in range") and then kind of walk it back to saying he had AI write the post and didn't check it for accuracy. I obviously don't know what he did, but people honestly misremember details all the time, so I'd probably give him the benefit of the doubt there. Plus, importantly, the main point he is trying to make is that something can be quite highly heritable while still having lower identical twin concordance than most people would expect, which is true. This in no way disproves your hypothesis, but it does argue against the claims you sound like you are making during your first post. For that reason, he probably thought it made sense as a rebuttal regardless of the precise numbers, which are only tangentially relevant.
If I may offer you some advice, I think you need to rethink the way you communicate the role of microbes and the relationship to genetics. You're doing a better job of this in your second post, but in your first post, you make it sound like all the evidence for a genetic role in schizophrenia is just noise, which is not true. In your first post, it sounds like you are saying that GWAS results can be dismissed as spurious correlations from overfitting, and that many genes with small effects can be dismissed as meaningless. You also only talked about inheriting microbes from your parents as a way to explain why schizophrenia runs in families, which as Scott Alexander notes is not the reason people think it is genetic. Coupled with some of the comments you have made (that you could predict someone's name with 32% accuracy from genes, or that someone would be essentially the same person if all their genes were swapped out, both of which are quite far-fetched claims), people are coming away thinking you don't understand the genetic evidence, and they are dismissing you for this reason.
What I think you really need to argue is that your hypothesis is compatible with the existing evidence, including the genetic evidence. You are doing a better job of this in this post, such as by mentioning how a person's gene's affect their microbiome. The much higher concordance of identical twins than fraternal ones makes it clear that, in the current environment, genes are a major factor in why some people get schizophrenia and some people do not. The important points to stress are that these genes interact with the environment to produce the phenotype, and microbes are a possible environmental factor. The most basic twin study model of heritability sorts variation into genes, shared environment, and nonshared environment. The tricky thing about gene-environment interactions (such as genetic susceptibility to microbes) is that they can be captured in any of these buckets depending on the nature and distribution of the relevant environmental variables. If it varies a lot from person to person, it will show up mostly as nonshared environment. If it varies a lot between families, it will show up mostly as shared environment. If it is fairly ubiquitous, it will show up mostly as genes.
You've got to keep in mind that most people are open to the idea that microbes cause disease (after all, they cause the vast majority of diseases that we fully understand the causes of), but people are also used to a very different model of infectious disease than what you are talking about. They will think of things like respiratory infections that quickly sweep through whole regions, or STDs that spread from person to person in an easy to understand way. They then see that schizophrenia doesn't exhibit those patterns, so they dismiss it. A disease model where the majority of people have a latent infection and only occasionally do symptoms emerge is a very different pattern. If most people are exposed to it, then a lot of the variation will be explained by genes. For instance, the leprosy in India example I gave below. Or think about AIDS. Right now your risk mostly comes down to the environment: are you exposed to HIV or not? But imagine suddenly everyone was injected with HIV. There are genes that make you resistant, so suddenly the variation in AIDS would mostly be explained by genes. Since, as you say, R. gnavus is in half of our guts, a big part of people's risk will be how genetically resistant they are to its effects.
The genetic evidence is probably telling us something important about how schizophrenia works, but you appear to be mostly ignoring it. People kept bringing up the relevance of genes related to synaptic pruning, and I think you were too dismissive of them. It's clearly not the whole explanation: people with identical genomes don't always get schizophrenia, and people with schizophrenia aren't always in psychosis, so there has to be some sort of environmental trigger, and your hypothesis fills that role well. But don't be too quick to dismiss past work on understanding schizophrenia. For instance, maybe R. gnavus causes schizophrenia in people with synaptic pruning problems, but most people are more resistant to its effects. I don't know the answer. But if your hypothesis is right, it should be able to explain all the existing evidence while also explaining stuff the existing hypotheses can't.
>he probably thought it made sense as a rebuttal regardless of the precise numbers, which are only tangentially relevant.
I mean, they're relevant in the sense that, whatever the gap is between the predicted and observed MZ twin concordance, that's the portion that we have to ascribe to nongenetic factors. And so even on his own premise, his definition of schizophrenia being "80% heritable" still ends up being support for the hypothesis that it's mostly something else.
And I guess that's the thing that annoys me here, and the reason I didn't just jump straight to "Good, we're saying the same thing, I bring you the fabled gene-environment interaction, the 2nd hit of the 2-hit model".
I see a guy very good at convincing people of things by leading them through a labyrinth of logic. But the one he's built here is so complex and circuitous that even he has lost himself in it, and sort of admits this--but still doesn't consider the possibility that there is no center to find.
What I mean is that the MZ twin concordance is not very meaningful on its own. Anything below 100% shows that the environment matters to some extent, but it doesn't show how much. What is meaningful is a comparison of the MZ and DZ concordance rates. Using Falconer's formula (not the only method of estimating heritability, but the simplest for illustration purposes), 80% heritability is what you would get from 90% MZ twin concordance and 50% DZ twin concordance, but also from 50% MZ twin concordance and 10% DZ twin concordance (whereas a 50% MZ twin concordance would imply 0% heritability if the DZ concordance were also 50%). Your original post talked about the microbiome being heritable only in the sense of getting it from your parents, and your only mentioned identical twins only in the context of having less than 100% concordance, with no comparison to fraternal twins. This is why Scott Alexander spends the first 4 of 6 points saying you don't understand the evidence for the role of genes (as he notes, "This doesn’t describe a process which could be expected to differ between identical and fraternal twins!"). You make the point much more strongly in this post by showing that a person's genetics do affect their microbiome.
Another important point in your favor, which I have not seen you bring up, is that genes can affect someone's response to a microbe by making one person have a latent infection and another person exhibit symptoms. For example, tuberculosis and leprosy are infectious diseases that have been with us for thousands of years, and still exist today. There has been strong selection on genes that cause resistance to these diseases (https://www.biorxiv.org/content/10.1101/2024.09.14.613021v1.full), so plenty of people can have a latent infection and show no symptoms. However, in the evolutionary arms race, the microbes keep evolving, too, so they haven't fully disappeared, even though they are less common now. Interestingly, that same paper shows that the polygenic score for schizophrenia risk has also substantially decreased over thousands of years. I would guess that, like leprosy and tuberculosis, the microbe(s) causing schizophrenia have been with us a long time, so there has been time for people to evolve resistance to them, which is probably part of why there is a genetic signal to schizophrenia. People can evolve either to fight off the microbes entirely or evolve to tolerate them without any ill effects.
Thank you for educating me on this! An interesting point to consider: leprosy and tuberculosis appear to be straight negatives. Schizophrenia, in this model, is just the tail end of the bell curve for abundance of a common microbial gene, or perhaps a two-hit involving e.g. a para-endogenous MAOI, and perhaps there is a significant advantage to having a small amount of such a gene, or one without the other.
Is R. gnavus beneficial for the host in small doses? Maybe, maybe not. I don't know the answer to that. Do you? You should probably figure that out before you start trying to cure people of it.
Yeah, I didn't buy his rebuttal against the ideal of schizophrenia being microbially caused, for some of the reasons you mentioned. An illustrative example that gives some context is that the heritability of leprosy is 79% in this twin study (https://www-jstor-org.ezaccess.libraries.psu.edu/stable/2396811?seq=7), yet no one would call leprosy a polygenic disorder.
I think it would be helpful to clarify your stance on a couple of questions, as it is hard to tell exactly what you are arguing:
1. Approximately what proportion of schizophrenia cases do you think are caused by microbes (of any kind)?
2. Approximately what proportion of schizophrenia cases do you think are caused by R. gnavus specifically?
In your original post, it comes across like you think the answer to #2 is 100%, but in the comments on Scott Alexander's article, you seem to walk it back and imply that you think the answer to #2 is fairly high but less than 100%. Personally, I think #1 is well above 0%, and I could potentially see it being 100% but possibly somewhat lower. I don't have enough information to make a good estimate for #2, but my suspicion is that it is lower than you seem to think it is.
You have done a great job synthesizing several pieces of evidence into a plausible-sounding hypothesis. However, I have found that it is very easy to construct a plausible story but then have it turn out not to hold up. What will really convince me one way or the other is how well your hypothesis makes predictions that hold up. Obviously the best way will be the clinical trials you say you are going to do. Still, there are various other testable predictions that could help support or refute your hypothesis (depending on what your answers to questions #1 and #2 above are). I'll put a few questions I have in the comments to this comment. I'm curious to hear your answers to any or all of those (probably some are not answerable given the available information, but some may be).
There are various other known risk factors for schizophrenia. A true explanation for the cause should be able to unify all of these. The Scott Alexander article mentions a few at the end: "Other risk factors include urban environment, winter birth, prenatal stress, and maybe cat exposure." Ironically, even though he seems dismissive of a microbial explanation, these seem to me to point to one. Most infectious diseases are more common in urban environments. Plenty of illnesses (especially viral) peak in the winter. Prenatal stress could be explained in other ways, but it might also impair the immune system. Cat exposure has been suggested to cause schizophrenia via Toxoplamsa gondii exposure. My intuition is that these risk factors point to various microbes causing schizophrenia; however, if there is a single cause, or a most common one, then they should all be linked to that particular microbe. Is R. gnavus abundance linked to any or all of these risk factors?
How robust is the schizophrenia-R. gnavus association across individuals and studies? In the study you cited, I can't tell how susceptible their analyses are to outliers. Is the result driven by, say, a few people with 1000x enriched R. gnavus, and the rest had normal levels of it? Or was the elevated abundance seen in all or almost all people with schizophrenia? How many other studies have looked at schizophrenia and the gut microbiome, and did they find the R. gnavus association or not?
Schizophrenia is negatively correlated with IQ (from before symptoms set in; https://psychiatryonline.org/doi/full/10.1176/appi.ajp.2008.07081242). Does your hypothesis have a good explanation of this association? For instance, would the hallucinogenic by-products of R. gnavus interfere with brain function or development even at low doses, or would a larger brain require a higher dosage to exhibit symptoms?
THANK YOU! This is the violence I was asking for!! On this point: consider phenethylamine, and specifically the brain's reaction to it. As mentioned in the original post, PEA induces neuroinflammation, and the brain responds by upregulating MAO-B.
This degrades phenethylamine, but also dopamine and norepinephrine, which are critical for learning and memory.
Phenethylamine is static on the broadcast channel. The listener cranks down the volume on his receiver, to avoid having his ears blasted out, but now he's going to miss the quieter parts in the symphony that he was hoping to hear.
I will be reaching out to the authors of that study (Vasileva et al) for their raw data, so we can look at the question about outliers. My feeling is that, if it had just been a couple guys, they would have mentioned it, but you never know.
Replication: Ioannou et al did a similar study in schizophrenia a few years back. They used mOTUs, which is a sort of sloppy metagenomics approach, but they found Eggerthella and Lachnoclostridium elevated in 3 cohorts. Lachnoclostridium is a now-defunct genus which has been mostly moved to Enterocloster, the other gut bacterial genus known to have an aromatic decarboxylase. See the "FKA Lachnoclostridium" tab on my "Big Beautiful Metabolomics Table"; I'll add the taxa reported as elevated in schizophrenia by Vasileva (2024) so you can compare.
I think some of the strongest evidence for a genetic cause of schizophrenia is that having an older father is linked to increased risk (https://jamanetwork.com/journals/jamapsychiatry/fullarticle/207596). Men keep producing new sperm throughout their lives, so older fathers pass on more new mutations. The association of paternal age with schizophrenia (and many other disorders) is used as evidence that de novo mutations play a role in these diseases. Does your hypothesis have a way of explaining this association?
Thanks for writing this rebuttal, was wondering if you were going to give a response in a larger post. Whilst I have no horse in this race, I was fairly annoyed at comments pointing out your bias as a microbiome scientist, whilst ignoring Scott Alexander (and the rationalist movement as a wholes') long and sordid obsession with DNA and genetics. Great post
So special pleading, moving the goal posts, taunts, no acknowledgements of the bad arguments you made. Odd that you didn't explain about the 10 minute limit truncating your post over at ACX, or why you didn't make the same justifications. No acknowledgement that the bone marrow transplant article goes directly against a claim you made to me in the comments of your last post.
I'd like to to take you seriously but not only am I now skeptical of this claim, but all your others as well because who knows what you've missed but are prepared to double down on.
My daily mood stabilizer has been life-changing, especially for mental noise and executive dysfunction (which is not what it's indicated for). A psychiatrist would also have me on daily antipsychotics for better protection against mania, but they work by blocking dopamine receptors, so you can imagine how that would suck.
So far every 2-3 months, I have a slow build of energy, confidence, sense of deep meaning in life... which I have to nip in the bud lmao.
That's early hypomania, which could potentially lead to a full episode- but generally there are a few weeks to catch it. When that happens, I take a fast-acting antipsychotic and am a bit sleepy and flat for the course (~a week). Taking them only as-needed is risky and not what a psychiatrist would tell you to do.
But I meticulously track my state and behavior (like, with a daily spreadsheet that has 20+ rows), meditate (helps with self-awareness), and have my husband here who knows me well. I feel confident a big episode won't happen again.
Sorry, not funny, but I had to seize that thread because it's the perfect example of something that came up in the comments last time, about why we haven't seen a case report of home FMT in an attempt to treat schizophrenia the way we have with bipolar.
Because any disorder which involves impaired judgement or capacity for self-awareness and reflection gets really slippery really quickly, particularly in the case when there's the potential for *advocating for yourself in an attempt to manage the condition* to be mistaken for *a symptom of the condition*. You'd think that'd be a narrow band, and yet...
Funny, I'd think this concern would apply even more to bipolar because a self-report of I CURED MYSELF is classic hyper-confident, grandiose, feel-amazing manic behavior.
Though the main reason I'm taking the reports I read with a grain of salt is because people can be euthymic for years between episodes with no treatment at all.
Good point! Doubly tricky because someone might get "cured", go 5 years, then have another episode...but if it's tough to say what those 5 years would have been like without the FMT.
> he used an LLM to write his rebuttal post, and this was sort of a “hallucination”.
The more I think about the latter possibility, the more likely it seems. It fits the pattern of confidently wrong assertion-with-citation that you get from current models, Scott just mentioned dealing with o3 hallucinations in the writing of a recent piece, and he seems like the kind of guy who would have a custom GPT trained on his corpus.
----
I believe he's said previously that he does this. I can't search my old comments on Substack, but I had a conversation ~6 months ago with someone after he issued a correction like "o3 said this so I included it, but it's actually not true at all, my bad" for something that was obviously not true if you spent more than ~20 seconds thinking about it. I'm really worried that AI is slowly replacing the necessity of struggling with hard or controversial concepts, and taking the benefits of that struggle: understanding, along with it.
The interesting thing about your stuff is that it actually seems testable in-vivo. Unlike other crackpot theories, which are usually unfalsifiable or make no clear hypothesis at all, it seems relatively straightforward to me, a layman, that you can just act on someone's gut microbiome to introduce strains of bacteria that aren't currently there. Add the cholesterol-eating bacteria (or whatever the mechanism is, I remember it increases cholesterol in urine or something like that), and see if someone's cholesterol decreases. If it does, then that's pretty strong evidence that the theory is sound, and repeated intervention over a larger group would slowly turn it from hypothesis to fact. I'm sure it's actually much more complicated than this, but whatever the specifics, it seems like there's a broad-strokes path to falsifiability, so your stuff is therefore quite interesting.
This guy fuckin' gets it! Thanks dude.
Y'know a few weeks ago Freddie de Boer put out a live debate challenge to Matt Yglesias, and one of the comments was like "I wanna see this debate but is there a reason it has to be live, like what's wrong with written exchange" and one answer that I didn't see brought up is that it at least lets you know nobody's serving you LLM slop
> The more I think about the latter possibility, the more likely it seems. It fits the pattern of confidently wrong assertion-with-citation that you get from current models,
to be fair, Scott was like this long before modern LLMs were developed.
lol.
I'm no microbiologist, as a lowly purely clinical MD this is helpful in spelling out your positions... Seems to me this is absolutely worth further research, though if I were consenting a treatment-resistant schizophrenic for an FMT trial, it would sound something like "there is a pretty bizarre and unproven theory this might cure you, (or maybe possible make you less treatment resistant?), but we think it's worth a try, even if unlikely. Do you want to be the one to try?" Not sure how the IRB will like the proposal but I'm for it. If you are not already in touch with people doing psych clinical research I could *attempt to connect you but not at all sure if I'll get traction with that department - no close contacts!
As others have commented I think you probably oversell your case a bit which leaves it a bit vulnerable, though perhaps it is a useful strategy to draw in attention, too. I think there is probably more compatibility between your theories then acknowledged -- you are not claiming that our genetic immune factors do not play a role here, and the prevailing theory already includes hand waving about environmental triggers. We don't have much of a disease model to contextualize this within other "dysbioses" and so your theory might sound as if you are calling it an "infection" and I think it is fair to object this topic would seem to be more complex. I hope we DO start to have an understanding of that as a disease class to maybe overturn some of the "functional" disease classes which have been growing!
For a relatively healthy person, I'm starting to appreciate that I would Not want to monkey around on the cutting edge of biohacking around this topic. I might be way off base here but if this is part of our genetic inheritance as you say, and something more mutable then the rest, is it possible that globalization in addition to antibiotics is actually part of the origins of the modern maladies? Like, for me, as someone with a family history of both MS and AS, I really do not want to accidentally run into any ETX or any Dialister either, if I'm somehow lucky enough not to have those lurking in my system so far... and maybe my family members who did encounter it came from an inheritance where that was not the prevailing immune problem they were set up to handle... maybe thats way off but it's a question I have and suddenly makes me have some Chesterton fence timidness, despite being bullish on the field generally.
>you are not claiming that our genetic immune factors do not play a role here, and the prevailing theory already includes hand waving about environmental triggers
So that's the funny thing, right? Like, if your definition of "80% genetic" leaves 75% of real-world MZ twin concordance unexplained...fine, I guess we're saying the same thing, so I think I've found the other "20%" and can we please focus on that.
>For a relatively healthy person, I'm starting to appreciate that I would Not want to monkey around on the cutting edge of biohacking around this topic
It's very funny when people e.g. try a new vitamin regimen and call it "biohacking"—like I'm sorry dude but if there's no risk of corrupting the filesystem you're not really "hacking" anything. But to your point: yeah I am happy and healthy and while I'm willing to test stuff on myself because that's the only really ethical way to do a first-in-humans for most things, I wouldn't be doing it just to try and "optimize" anything. My goal is of course to make it so that it's less like hacking and more like the App Store, where if you want an upgrade, you can just install the thing and trust that it's going to work.
The Dialister thing is interesting, I wanted to ramble on in the post about it but the example had gone on long enough lol. Thing is: correlation is not causation! I said Dialister and Phascolarctobacterium are the only two that really eat succinate, and this is true, but some other taxa use succinate in auxiliary ways; it's not their main fuel but a metabolic accelerant for things like C. diff and Veillonella (jury is out on Veillonella but personally I'm sus of her).
So it may be that Dialister isn't really bad, just a poor man's Phascolarctobacterium, because it's not very good at its job; apparently it doesn't take up and consume succinate as quickly, which might make it easier for an opportunist like C. diff to get a foothold. I think part of ecosystem resilience is having every niche well-filled, so that everything you eat breaks down into stuff that either you or a symbiont can use.
I've had an idea in my mind for a while now, to make a deckbuilding game to learn/teach this stuff (things like Magic the Gathering and Balatro are the only way I've ever seen people learn the names and properties of 400 different lil guys by heart). The core mechanic would be about building combos—like, if lunch today features arabinose and glucose, your Bacteroides turns arabinose into acetate+succinate, and then a Faecalibacterium can use the acetate + glucose to make butyrate, and maybe that's points. But I think one of the core mechanics would have to be about making sure there's nothing left "on the table", so you'd wanna have a Phascolarcto in there too, to mop up the succinate. I doubt anyone would actually want to play such a game, and there's probably a way to have a computer do the metabolic network modeling, but where's the fun in that?
>is it possible that globalization in addition to antibiotics is actually part of the origins of the modern maladies?
Have you seen my piece on celiac? A great example of how precisely that could be the case. https://stephenskolnick.substack.com/p/celiac-disease-and-the-gluten-intolerance
The deck building game sounds... so fabulous but entirely unapproachable lol. An interesting way of trying to map this though, for the dedicated. Yes I've been following your blog for years, actually I got to meet you briefly back in '23 at vibecamp :) ... and to be fair I suppose the concept is really not novel - in med school we learned the model of autoimmunity that a susceptible host meets a bug and triggers and aberrant response. Amazing you've been identifying real targets now. I've just been in the mode of focusing on getting diverse plant fiber to support a diverse microbiome as the best current practice - and it's a good reminder that it is not really the more diverse the better. As well as getting an infectious side effect you could give yourself the ticking time bomb. I suppose I was just having a brief daydream of getting out of the sick care system and getting more involved with this stuff myself, but it's not quite ready for broad clinical use, best to wait for your app store to develop ;)
Also yeah, I know it'd be a tough sell to literally anyone lol. But yes please put me in touch with your clin psych research contacts if you don't mind. (shoot me an email at SDSkolnick@gmail.com)
I'm on your side, and this looks like a convincing rebuttal.
But if you're a lone outsider trying to single-handedly force a paradigm shift, you can't afford snark, put-downs, or unprovable accusations of bad faith. These are usually good excuses to dismiss someone as a credible actor, and a man in your position doesn't get any excuses. A single lapse in message discipline can tank your entire project.
> But if you're a lone outsider trying to single-handedly force a paradigm shift, you can't afford snark, put-downs, or unprovable accusations of bad faith.
I'm not sure that's true anymore. Before the post-Manhattan project bureaucratization of science snark and put-downs were rather common on scientific writing. Now between the rise of the internet and the current administration's attacks of the science bureaucracy we may be getting back to the old system.
https://www.experimental-history.com/p/the-rise-and-fall-of-peer-review
Thanks for the feedback on both the content and the approach.
>you can't afford snark, put-downs, or unprovable accusations of bad faith
I can afford a leeeeetle snark. But I try to spend within my means.
Did it come off like I was accusing him of arguing in bad faith? Or was that a pre-emptive caution?
The speculation that he might have used an LLM to write the post would have been fightin' words to me.
I just wanted to note that, regarding concordance, there's an environmental term that I'm pretty sure Scott is implying, given his spreadsheet. It can be additive or multiplicative, depending on the mechanism. Critically: if the environment changes the prevalence of the disease changes, and so do the concordance rates.
So you can't calculate concordance rates from just knowing the percent of the variance that is explained by genes.
I'm not saying that any of this is material to your argument.
Also, I'm a huge fan of phage therapy which has always seemed on the cusp of fruition for the last 20 years. So given it's perpetual nascence maybe it's a bad thing to consider, but it seems like the perfect tool for launching a targeted strike against a particular strain of gut bacteria. (Several phages might be needed, since some strains of e.coli are immune to a given phage that others are susceptible to.)
My personal assumption is that other microbes, like T. gondii, also contribute to schizophrenia. And the effects of T. gondii are pretty well known. But if R. gnavus abundance contributes to a schizophrenia risk score then this should be one way to test the matter. Right?
yes! Subtracting is always gonna be more difficult than adding to a microbiome, but phage therapy is probably the way. The reason it's perpetually on the cusp is that, exactly as you point out, it doesn't scale easily. But as a species we're putting in a lot more squeeze for a lot less juice on plenty of fronts. Just about convincing the right people to allocate the resources effectively.
I am glad you wrote this, since the rebuttal + further comments exchange on ACX seemed like a lot of Bayes damage. And it was generally well-written and entertaining! But I notice that I am still not convinced - that these answers to most, but not all, of Scott's objections feel unsatisfying. There's a reciprocity between the two pieces, of both parties swatting away trivially inconvenient annoyances to their preferred models and then going on to further explicate them. I do take your point that the original Holy Grail post was just the tl;dr version, and it's to your credit that you include some of the missing pieces here. Strong claims require strong evidence! But the bar is really quite high, and while stronger, as a layman I am not sure even this 2.0 corpus clears it. (It is entirely possible that a combination SME and math deficit means I'm missing the meat of the argument, and from an expert point of view it's a slam-dunk...yet it's part of the Rationalist training to notice that gut feeling* of "this argument doesn't seem fully coherent", and not be intellectually satisfied until that itch is properly scratched.)
Probably gonna flesh this out later into a point-by-point. Preferably before Scott does a Contra Contra ("No, Rescuing The President Is Not Manliness-Dependent"), those never go anywhere and just lead to mutual retrenchment with no Yes, And.
*pune, or play on words, not intended, I am definitely stealing Branch-COVIDIAN though
Thank you I was particularly proud of Branch-COVIDians lol.
This is, as you point out, not proof. No amount of spreadsheet clickery or essay can be. Gotta go do things in the real world for that!
Ended up procrastinating on a response due to a bad bacterial infection (yeah, ironic) and am ashamed at now having been scooped by Scott anyway. Lucky Hunter made a lot of great points too, in both directions. Since most of the technical minutiae have been addressed one way or another...
1) Concordance and missing heritability: I wish to note that Scott didn't write just the one article on schizo genetic inheritence that he linked in the rebuttal (which, yes, he did goof bigly on + got called + added to Official List Of Mistakes). It was a whole series of posts over a good length of time, going back to SSC, complete with other interlocutors that he also Contra'd, oblique references to Hey Guys Remember When I Demonstrated Schizophrenia Is Mostly Genetic, and recurring callbacks like the toxoplasma thing. It's "part of the brand", so to speak. Therefore I don't find it surprising at all that he responded quickly and strongly to your post - it's good grist material to springboard one of his own, you're a known reader, author and audience have overlapping interest in the subject, etc.
2) Accusations of LLM use: come on, this is silly. For starters, it's an obvious Don't Threaten Me With A Good Time. As an AI booster, the valence on "he sometimes uses LLMs!" against Scott is...probably not what you'd intend? It'd be like making the claim that Zvi Mowshowitz uses Twitter to find source material. Scott also explicitly claimed he didn't "[make] extensive use of AI" for any posts prior to Missing Heritability: MMTYWTK* , and that it was employed for research rather than writing. Now, Contra Skolnick did come out a few days later, that's true...but it seems really dumb to take flak for using o3 (both on that post and in the earlier Contra Cowen On USAID), write a good defense laying out that Actually, No, I Know What I'm Doing With These Tools, and then hypothetically have ChatGPT slop out a pro forma reply with such poor QC that he'd get called on it. Does Scott sometimes do dumb things? Sure! But since I'm not convinced on the speed argument (the man is not exactly known for the slowness of his writing), and you and I are both unwilling to assume malicious lying...
...that leaves us with basically a vibes argument, the penumbras and enumerations of LLM. Once upon a time, if a writer made a mistake citing his own corpus, we'd call that "making a mistake". In current_year, it's a harbinger of LLM slop, because...uh...humans are usually infallible at recollection of noncentral facts that only need to be directionally correct to support their arguments? (Again, valence issue, Scott fucks up way more often than just this one time!) If we're armchair quarterbacking, then I'd counterpropose that that was a particularly egregious and embarrassing error for Scott, so he'd be more rather than less liable to conveniently forget the correction. Likewise, I get the sense from some of your other comments that you're particularly skeptical of AI, at least when it comes to writing. Are you sure this isn't just reaching for a convenient Axe of Grinding? Are there any other smoking finger gunprints of LLM malfeasance in the post? If he's hypothetically got a custom-trained one, it sure does a damn good job of sounding just like Scott...(and as another commenter notes, it's not like "confidently wrong assertion" is a novel tone either)
3) Fuzzy gnavus: look, not to pull a tu quoque, but it's weird to do a bunch of throat-clearing about one's empiricist boner fides and then just go ahead and Roll To Disbelieve against the conclusion of the paper's actual authors + sundry others. They already did the experiment! And as others have pointed out, it's weird that we haven't had an antibiotics -> gnavus wipe -> miraculous scizo recovery case study to date, outside of like one maybe-sort-of-if-you-squint-and-play-reference-class-tennis case. You claim this would be very difficult and/or unethical to intentionally test with confidence, and thus probably won't happen...but that's the same Fully General Counterargument of disprovability you complain about wrt particle physicists' hedging! I do think you're on solid ground with your more general claim that Antibiotic Overuse Is Really Bad, Acktually, and it seems easy enough to directly test gnavus-cide if one bothers to try. But this really feels like a lamplight key search situation (though it's not like keys are never found near lampposts). Even if one accepts the Model 2 of many gnavus, now handle it -> severe symptoms -> clozapine, it still feels like one of those "necessary but not sufficient" constructions. Increased gnavus prevalence being found in all sorts of other diseases, few if any of which seem to have similar relation to tryptamine/phenethylamine levels, makes it seem kind of like the "vitD gets depleted while sick but we can't really prove D supplementation cures people faster" thing.
4) Adopt-A-Microbiome: Scott says no correlation between adopter and adoptee, instead it's between adoptee and bio parents. You clap back that There Is No Evidence that non-schizo adoptees acquire the schizo risk of their schizo adopters. Which...what? I don't understand if you're making the assertion that there *is* evidence, we just don't find it because sample size is too small to establish base rates (I'd imagine schizophrenia is a good way to get screened out of adopting *and* adoption? But surely it's happened a nonzero number of times in both directions?). Either way, I'm confused about how this is a particularly objectionable objection. Someone putting their bet on Genetic should be expected to bring out a strong line of argument like correlation with bio parents, no? Likewise, I'd have expected a counterargument from you more along the lines of "actually, the microbiomes of adopters and adoptees are more similar than adoptees and bio parents". Which would dovetail nicely with Scott's link that spouses have more similar microbiomes than siblings! But that shot wasn't taken, and instead it was...somewhere between a non-sequitur and a tone argument. Why?
Do wish to reiterate humility here: almost all the actual science parts of this go way over my head, I'm probably missing a lot, this is just me nitpicking weird structural deficiencies in the elephant rather than helping flesh it out. I am glad it is being fleshed out though! Even if you or Scott turn out to be dead-wrong about everything, it's important to test hypotheses (and existing consensus!) against Worthy Opponents. A productive and entertaining exchange of ideas, with even technically illiterate idiots like me leaving a bit more informed. Would that some of your other spicy takes got the same sort of trial-by-fire...you've got so many ideas that are just on the cusp of being actionable, and "sound right", but need more refinement.
*https://www.astralcodexten.com/p/using-ai-to-research-the-missing
Re: whether psychosis feels like a psychedelic or an amphetamine, I don't have schizophrenia, but have bipolar I and have gone through mania/psychosis 3 times.
My mania *absolutely* resembles a psychedelic / stimulant experience - to the extent that the first time I had a psychotic break, for part of the lead-up I was convinced I'd been dosed with LSD.
Imagine being on a kind of euphoric psychedelic + ritalin (or what I'd imagine meth to be like) for days/weeks/a month straight, and completely losing touch with reality as the trip gets deeper- transforming into an agitated, 'bad trip' over time.
It's unique, but it doesn't feel categorically different. The closest analogue I have to early mania was when I tried DMT and didn't "break through," plus a stimulant.
The things that do stand out are the hyper-confidence, the grandiosity, the feeling of the conscious experience radically speeding up, and the partial memory blackout.
It just so happens that I – a fiftysomething male who's never had symptoms of schizophrenia – had my gut biome profiled 3 times in 2024. All samplings showed a relative abundance of R gnavus significantly above the 0.1% to 0.3% loosely conjectured as 'usual': 4.000% in March, 1.820% in September, 7.735%(!) in October.
That last value is potentially near the 6+ "log fold change" differential abundance that's associated with tough schizophrenia cases in the paper [Vasileva, Yang, Baker, et al 2023] that seems to have triggered your own further analyses.
While I've not formally logged my mental state for fine-grained changes, I don't personally recall – nor have any family/friends/professionals/etc mentioned – any mental disturbances around the time of highest abundance.
I'll be retesting soon, so may then have a better idea how much of a brief outlier that 7.735% reading may have been. That sample was a couple weeks after a surgical procedure & recovery involving antibiotics – it was my curiousity about such effects that motivated that test, which in general showed very mild changes from pre-procedure biome composition.
More comments & a screenshot of my results in my X replies: https://x.com/gojomo/status/1942721390806982850
Fantastic, thank you for weighing in. To know how useful of a result that is, I'll have to do a bit of a dive on TinyHealth's methods. (For one thing: how is sampling done?) But as you mentioned in the tweet thread, lots of potential for interactions where trace amine production is necessary but not sufficient.
They send you a swab & tube to return a small dab of feces through the mail: https://www.tinyhealth.com/how-it-works
The detail & level of explanation/qualification in their "deep shotgun metagenomic sequencing" reports has impressed me: https://www.tinyhealth.com/technology
For understandable reasons, their 'action plans' emphasize more-anodyne interventions like diet changes, supplements, & "get dirty - exposure to nature can benefit gut health" – rather than, well, the title of your blog. :)
1. Re the simulation - I explained that this was just an intuition pump for why 80% genetic shouldn't equal exactly 80% twin concordance. I linked a paper which did the full calculation and got a pretty reasonable result - https://genepi.qimr.edu.au/staff/nick_pdf/Classics/1970_Smith_AHG_Dorret.pdf . As far as I can tell, you haven't rebutted the claim (either from my discount simulation or the real paper) that this establishes "low twin concordance = non-genetic" is a bad way of looking at things - you've just pointed out unrelated errors in the hope that people confuse this with the main point being wrong. If we continue this discussion, I'd prefer you to either assert that you still think low twin concordance has to mean non-genetic, or else agree that you've abandoned this argument.
2. The argument here is that, since there is missing heritability in everything, schizophrenia is not special, and does not require a special explanation.
If there is some reasonable explanation for missing heritability (hard-to-find genes, GxE interaction), then there is no puzzle around missing heritability in schizophrenia, and no motivation to search for a gut bacterium that causes it.
If there is no such reasonable explanation, then you either need every trait to be caused by gut microbes, or by something equally weird and mysterious and hard to find. And if there are so many different weird/mysterious/hard-to-find causes, again, there's no motivation to go searching in the gut for a cause of schizophrenia.
This doesn't prove that schizophrenia isn't a gut microbe, it just dismantles one argument that might motivate people to think it is, bringing this back to the prior.
3. You don't really seem to deny that it's weird that schizophrenia starts with a trigger episode between 18 - 30, you just say that sometimes bacterial composition in the gut changes (but why would it happen between 18 - 30 so often, and so often change in one direction?) I agree that the neurodevelopmental theory of schizophrenia hasn't been completely fleshed out (although it is fleshed out more than you seem to think, see for example the section What's Happening In Adolescence at https://psychiatryonline.org/doi/10.1176/appi.ajp.20240305 ) but most complex diseases aren't completely understood.
4. You're putting in all this work to show that it isn't genetic, and then when I show that it behaves like a genetic disorder, retreating to "okay, it's genetic, but the genes cause the microbiomes". Because of how concordance declines from identical to fraternal twins, it's not just vaguely genetic, in some sense where genes maybe influence it a little. It has to have 80% of the variance explained by genes! If you want to say those genes all code for microbiota, fine, but then you've abandoned all the stuff you were saying earlier about how you think there are holes/inconsistencies in the genetic picture. Instead you're trying to have it both ways - claiming it's not genetic so you can recycle the missing heritability arguments in favor of there having to be some other cause, then claiming it *is* genetic when I point out how it behaves in every way like a genetic disorder!
4b. Regarding the lack of an adoptive-child-to-adoptive-parent schizophrenia correlation, you only say:
"Can we think of any reason why there might not be a lot of studies showing that children adopted out from mentally healthy parents to parents with schizophrenia acquire the schizophrenia risk of their adoptive families? Any reason at all."
I'm not sure what you're implying here, but I think maybe you're saying nobody would ever adopt a healthy child out to schizophrenic parents?
But see any of the studies that have been done on this issue, for example https://www.sciencedirect.com/science/article/abs/pii/0022395687900896 , for how they handle it. I don't understand what you think you're doing here, when the studies have been done and come up with the result you're mocking as impossible.
You continue to assert that "practically all the major schizophrenia risk genes are immune genes", which is false, and which many commenters and I called you on the last time you asserted it.
5. I didn't mention the clozapine dose-dependence at all and it's not a big part of my argument, so I don't understand why you're so interested in refuting it.
In any case, you only provide a possibility proof that the direction of causality *could be* microbes -> clozapine.
My argument was that it's equally possible (and, in fact, more plausible according to everyone who has studied this), that the direction is antipsychotics -> microbes, therefore the presence of unusual microbes in the gut of people with schizophrenia provides basically no evidence that schizophrenia is a microbial condition.
This is a big part of my complaint in general. You have two main pieces of evidence in favor of your microbial hypothesis - the missing heritability, and the presence of weird microbes in schizophrenics' guts. I've shown that there are alternative better explanations for both of them. You instead try to argue that it's still *sort of possible* that microbes could be an explanation. But even if this is true, so what? We're just back to our prior that any given condition could be caused by microbes. And in this case we have many many other pieces of evidence that this one isn't - for example, all the risk factors, the lack of response to antibiotics (which IMHO you fail to defuse above), the developmental course, the non-resemblance in non-genetic relatives, etc. So it just doesn't seem interesting at all that, by stretching really hard, you can sort of trace a course by which if a lot of things go perfectly it's still possible that there are microbes involved.
6. I feel like you've tried to dance around all of my points and are still confused about basics of how genetics work. If I'm going to believe crazy hypotheses about schizophrenia, I'm going with E. Fuller Torrey's thing about how it's all cat feces, at least that one is funny.
Obviously there are genetic factors involved in disease risk. I am arguing that, to the extent that the genome is useful in predicting risk, this is mediated by its impact on the microbiome.
I am saying this is the most likely-seeming "environment" component of "gene-environment interactions".
>You continue to assert that "practically all the major schizophrenia risk genes are immune genes", which is false
Scott. Look at the Manhattan plot.
https://pmc.ncbi.nlm.nih.gov/articles/PMC5918692/figure/F1/
It's log10 scale.
Are you arguing that the MHC is not an immune complex? Or are you arguing that it's worth considering things outside the MHC, when the MHC signal is TEN SEPTILLION TIMES stronger than the next-strongest signal?
Did you think I was taking poetic license, with the title of the article?
Do the math. How far would we have to move the *actual sun* away from the Earth for it to be ten septillion times less bright? It's further than alpha Centauri.
Spending time thinking about the influence of these other genes on schizophrenia risk is, very literally, a bigger waste of time than studying the impact of starlight on skin cancer risk, or putting out a solar panel at night.
Thank you for bringing Kety 1987 to my attention. I looked for anything like that, but couldn't find it. But it does not support your claim that a child adopted out from mentally healthy parents to parents with schizophrenia is not at increased risk.
It suggests that a child adopted out from parents with schizophrenia to mentally healthy parents is still at increased risk. Pretty sure the mistake you're making here is called "confusing the inverse". https://en.wikipedia.org/wiki/Confusion_of_the_inverse
Oh, DID you use an LLM in writing "Contra Skolnick"? Or did you come up with that ending all on your own
I'm not yet convinced of his specific proposed mechanism, but I do think it is highly likely that schizophrenia and a whole host of poorly understood diseases will turn out to have microbial causes. I think you're being a bit too dismissive of this point.
Loads of infectious diseases have high heritability, plenty of times in the realm of 50-80% (https://www.mdpi.com/2073-4425/16/2/177). I suppose you could describe tuberculosis or polio or measles as a complex disorder with many genetic, environmental, and microbial factors influencing them, but I think you'd agree it would be strange to do so. A clearer description is that they are caused by microbes, and there are genetic and environmental factors that affect one's risk.
I agree that low twin concordance doesn't mean that genes are unimportant drivers of trait variation. However, identical twin concordance of less than 100% means that some sort of environmental factor is necessary, wouldn't you agree? If you did a twin study of sickle-cell anemia or cystic fibrosis, you'd find 100% identical twin concordance. Microbes are not the only relevant environmental factor that can affect disease risk, but microbes are the cause of the vast majority of diseases we understand the causes of.
As I understand it, the possible explanations for missing heritability basically come down to some combination of gene-environment interactions, gene-gene interactions, rare genetic variations, or violations of the equal environment assumption. Disease mediation by microbes would fit the category of gene-environment interactions. I can also imagine a role for more similar environmental exposures of identical twins causing increased concordance for microbes compared to for some other possible traits.
A key area I would disagree with you is that we should expect similar reasons for missing heritability for diseases compared to most other traits. Something like height or IQ varies continuously in a more-or-less normal distribution, which makes sense if they are the sum of a large number of small genetic and environmental effects all adding up. Most diseases are not like that, though. Yes, severity of symptoms can vary, but you pretty much either have the disease or you don't. I know you talk about modelling it as a continuous trait that gets expressed once it crosses some kind of threshold. What does this actually mean in the real world, though? What causes this threshold? What makes some people schizophrenic when their identical twins aren't? What makes people with schizophrenia exhibit psychotic symptoms sometimes, but then these go away? It seems like this would necessitate some kind of environmental trigger, and a microbe would serve that role well. Do you have any alternative hypotheses of what it would be? Do you have any other examples of a binary trait outcome affected by a large number of genes that we actually understand how it works (as opposed to poorly understood diseases that might still turn out to be caused by microbes)?
I think a lot of the missing heritability of diseases (that strike young and have a big fitness effect, specifically) will turn out to be microbes, which I don't think will necessarily be the case for most other traits. In addition to the threshold effect making diseases very different from continuously-varying traits, natural selection ought to make genetic diseases rare under most circumstances. Exceptions would include traits with a heterozygote advantage, those that only cause a disease under evolutionarily novel circumstances, or those that mostly strike in old age. Do you think that these explanations apply to schizophrenia? If not, why do you think natural selection has not make it substantially rarer?
What are your main criticisms of Torrey's hypothesis about Toxoplasma causing a subset of schizophrenia cases? Is it just the same points about it being highly heritable, or is there something about the hypothesis you feel doesn't fit with the data? Toxoplasma famously evolved to get into rodents' brains and manipulate their behavior to increase their chances of getting caught by cats. It can even do that to our closest relatives, chimpanzees (https://www.cell.com/current-biology/fulltext/S0960-9822(15)01517-1?). If it can affect chimpanzee behavior, it doesn't seem far-fetched to me that it could get into human brains and start messing things up. Do you have an alternative explanation for the association between cat exposure and schizophrenia?
I wouldn't be incredibly surprised if a microbe was involved in some way. I disagree with Skolnick on two points:
1. Skolnick dismisses the genetic evidence as false. I think the genetic evidence is very strong. As you point out, genetic evidence is potentially compatible with it being a microbe; I think if it's a microbe, it will be be a microbe with strong genetic liability, and Skolnick's arguments against the genetics will still be wrong. I think this means that one of Skolnick's affirmative arguments for it being a microbe is false.
2. I don't think there's a strong case that it's a gut microbe in particular, let alone the specific gut microbe he mentions. R. gnavus is in lots of people, there's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes, and it doesn't really behave the way I would expect a schizophrenia cause to behave.
Maybe a good example of this is rubella. We know prenatal rubella infection can cause schizophrenia! There's your microbial cause, hooray! But nobody cares about this because it's a long and complicated pathogenesis process and rubella infections only fully explain a ~single digit percent of schizophrenia cases. I think other microbial causes will end up pretty similar - anything that damages the brain increases liability, lots of different microbes can damage the brain, but we still need to figure out overall what the disease *is* and what these lots of different forms of damage are doing.
This is also why I don't think Torrey's toxoplasma open-and-shut solves the case. We've checked what percent of schizophrenics have toxoplasma. It's both more than you would expect by chance, and much less than 100%. I think in the end this will turn out to be something like rubella - yes, it makes the odds worse, yes, it causes some single digit percent of cases, but we still need to figure out what the disease is and why there are all these non-necessary-and-non-sufficient risk factors.
For comparison, consider something like heart attacks. What *causes* heart attacks? Is it eating donuts? Certainly those don't help. But lots of people get heart attacks after never having eaten any donuts in their lives, and lots of people eat donuts but don't get heart attacks. There are actually plenty of infections that increase risk of heart attack by damaging heart tissue, but again, lots of people get heart attacks without getting any of those.
When we try to figure out "what heart attacks are", the goal isn't to find the One Food That All Heart Attack Victims Eat But Nobody Else Has Ever Eaten. It's to understand what the heart is and how it works, identify the specific disease process going on, and then in retrospect it will be obvious that donuts and infections and everything else all contribute a bit to that disease process.
This is quite different from what we're doing in eg AIDS, where we did discover that it was all just one virus. I think the fact that we discovered that pretty quickly in AIDS, but have been studying schizophrenia for centuries with only these suggestive leads, should make us expect schizophrenia to be more like heart attacks than like AIDS.
>Skolnick dismisses the genetic evidence as false
I literally don't.
>There's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes
This is circular.
>> "I literally don't."
The things I disagree with you on are you saying that the low twin concordance rate disproves anything, and you saying that missing heritability disproves anything. If you have dropped those two claims, I think you should say so explicitly (or if you did say that somewhere, I missed it - please point it out).
>> "There's no particular evidence that it's overrepresented in schizophrenics beyond the level that could be explained by antipsychotic-induced gut changes"
It doesn't seem circular to me. Suppose you said "I believe in Bigfoot because people report seeing large footprints". I demonstrate that large footprints can also be caused by bears, and say "There's no particular footprint evidence for Bigfoot beyond that which can be explained by bears". Is this circular reasoning?
The low twin concordance rate is not proof or disproof of anything. It is evidence that, if we seek a satisfactory explanation for this disease, we must look beyond the genome. Happy?
Regarding Bigfoot: the reason I say it's circular is that, in this analogy, the only evidence you have for the existence of these so-called "bears" is the footprints!
If you could produce a bear—if we had, say, microbiome data from a study of clozapine in healthy volunteers which showed an increase in the relative abundance of Lachnospiraceae, then I would happily grant you bears. If there were animal studies showing something similar, I would grant you bears. I would even take something as tenuous as an in vitro effect, e.g. if it had been shown that clozapine inhibits the growth of Agathobacter rectalis, because it's well known that A. rectalis abundance is inversely associated with that of Ruminococcus gnavus. But you do not have so much as a bear claw donut; you have not "demonstrated" anything, merely asserted it.
So we are both proposing hypothetical megafauna, to explain these footptints—the association between abundance of trace amine-producing bacteria and schizophrenia. The difference is that your model ascribes to coincidence the fact that these organisms are tightly associated with serum tryptamine and phenethylamine in the general population.
> "The low twin concordance rate is not proof or disproof of anything. It is evidence that, if we seek a satisfactory explanation for this disease, we must look beyond the genome. Happy?"
No! We're still talking past each other. My claim is "the low twin concordance rate is exactly what you would expect from a ~80% heritable disease, and provides exactly zero evidence for any additional non-genetic cause". If you disagree with that claim, I continue to think you're dismissing the genetic evidence. If you agree with it, I think you shouldn't say things like "it's evidence that we must look beyond the genome".
> "Regarding Bigfoot: the reason I say it's circular is that, in this analogy, the only evidence you have for the existence of these so-called "bears" is the footprints!"
Let's tighten the analogy. Suppose we're talking about Local Forest. There are big footprints. I'm not sure that there are bears in Local Forest in particular, you're not sure that there are Bigfeet. But there is a high prior on bears (because they're a common well-known animal and likely to be in any given forest), and a low prior on Bigfeet (since they are previously unknown to science and there are lots of reasons to expect we would have seen them if they existed).
If our only evidence beyond priors is something which is explained by bears and Bigfeet equally well, then we should be almost certain that there are bears and that there aren't Bigfeet.
Antipsychotics disrupting gut bacteria is the most plausible thing in the world, like bears being in the woods. But schizophrenia being a gut microbe requires every past scientist to have missed something obvious, contradicts much of what we know about schizophrenia (like its non-response to antibiotics and its association with various neurodevelopmental insults), and there's no other reason to believe it besides the claims about twin concordance/missing heritability (which I say are also wrong, and rightly understood provide no evidence) and a very speculative theory about how the disease sort of resembles tryptamines (which I don't think it does). Like Bigfoot, it's something we wouldn't believe unless we had great evidence. The bears explain the gut microbe prevalence evidence just fine on their own, so merely proving bears plausible should be enough to make us forget about Bigfoot.
> "If you could produce a bear—if we had, say, microbiome data from a study of clozapine in healthy volunteers which showed an increase in the relative abundance of Lachnospiraceae, then I would happily grant you bears."
I assume there's some reason why case-control microbiome data from a study of risperidone which shows a decrease in the relative abundance of a bunch of species won't make you happy, but in case I'm wrong, https://www.sciencedirect.com/science/article/abs/pii/S0920996418302743?via%3Dihub .
On this single tangent: "but why would [bacterial composition in the gut change] happen between 18 - 30 so often, and so often change in one direction?"
Ages 18-30 is likely a period of especially-high variability in all infectious & biome-changing categories: it's a time of relocations, new living arrangements, new workplaces, new relationships, new diets. And: probably a few courses of antibiotics for various conditions.
And simply lots of "re-roll" random walks, where *some* imbalances prove both persistent & noticeable, could be enough to create the impression of things happening "in one direction".
Screening the gut biomes of early onset (teen or before) & atypically-late (40yo+) cases could provide a strong signal whether any bacteria might be relevant there.
Thank you for the feedback, both on the argument and the tone.
I can afford a leeeeeetle snark (I think a gentle ribbing about the length of ACX's posts etc. is pretty harmless) and I try to spend within my means. I don't think I accused him of arguing in bad faith at any point. Did I come across that way? Or was this just a preemptive warning?
I think the main area where you come across as making accusations of bad faith is when you imply he is lying (your meme even says "lie about simulation results being in range") and then kind of walk it back to saying he had AI write the post and didn't check it for accuracy. I obviously don't know what he did, but people honestly misremember details all the time, so I'd probably give him the benefit of the doubt there. Plus, importantly, the main point he is trying to make is that something can be quite highly heritable while still having lower identical twin concordance than most people would expect, which is true. This in no way disproves your hypothesis, but it does argue against the claims you sound like you are making during your first post. For that reason, he probably thought it made sense as a rebuttal regardless of the precise numbers, which are only tangentially relevant.
If I may offer you some advice, I think you need to rethink the way you communicate the role of microbes and the relationship to genetics. You're doing a better job of this in your second post, but in your first post, you make it sound like all the evidence for a genetic role in schizophrenia is just noise, which is not true. In your first post, it sounds like you are saying that GWAS results can be dismissed as spurious correlations from overfitting, and that many genes with small effects can be dismissed as meaningless. You also only talked about inheriting microbes from your parents as a way to explain why schizophrenia runs in families, which as Scott Alexander notes is not the reason people think it is genetic. Coupled with some of the comments you have made (that you could predict someone's name with 32% accuracy from genes, or that someone would be essentially the same person if all their genes were swapped out, both of which are quite far-fetched claims), people are coming away thinking you don't understand the genetic evidence, and they are dismissing you for this reason.
What I think you really need to argue is that your hypothesis is compatible with the existing evidence, including the genetic evidence. You are doing a better job of this in this post, such as by mentioning how a person's gene's affect their microbiome. The much higher concordance of identical twins than fraternal ones makes it clear that, in the current environment, genes are a major factor in why some people get schizophrenia and some people do not. The important points to stress are that these genes interact with the environment to produce the phenotype, and microbes are a possible environmental factor. The most basic twin study model of heritability sorts variation into genes, shared environment, and nonshared environment. The tricky thing about gene-environment interactions (such as genetic susceptibility to microbes) is that they can be captured in any of these buckets depending on the nature and distribution of the relevant environmental variables. If it varies a lot from person to person, it will show up mostly as nonshared environment. If it varies a lot between families, it will show up mostly as shared environment. If it is fairly ubiquitous, it will show up mostly as genes.
You've got to keep in mind that most people are open to the idea that microbes cause disease (after all, they cause the vast majority of diseases that we fully understand the causes of), but people are also used to a very different model of infectious disease than what you are talking about. They will think of things like respiratory infections that quickly sweep through whole regions, or STDs that spread from person to person in an easy to understand way. They then see that schizophrenia doesn't exhibit those patterns, so they dismiss it. A disease model where the majority of people have a latent infection and only occasionally do symptoms emerge is a very different pattern. If most people are exposed to it, then a lot of the variation will be explained by genes. For instance, the leprosy in India example I gave below. Or think about AIDS. Right now your risk mostly comes down to the environment: are you exposed to HIV or not? But imagine suddenly everyone was injected with HIV. There are genes that make you resistant, so suddenly the variation in AIDS would mostly be explained by genes. Since, as you say, R. gnavus is in half of our guts, a big part of people's risk will be how genetically resistant they are to its effects.
The genetic evidence is probably telling us something important about how schizophrenia works, but you appear to be mostly ignoring it. People kept bringing up the relevance of genes related to synaptic pruning, and I think you were too dismissive of them. It's clearly not the whole explanation: people with identical genomes don't always get schizophrenia, and people with schizophrenia aren't always in psychosis, so there has to be some sort of environmental trigger, and your hypothesis fills that role well. But don't be too quick to dismiss past work on understanding schizophrenia. For instance, maybe R. gnavus causes schizophrenia in people with synaptic pruning problems, but most people are more resistant to its effects. I don't know the answer. But if your hypothesis is right, it should be able to explain all the existing evidence while also explaining stuff the existing hypotheses can't.
All very good points, well taken.
>he probably thought it made sense as a rebuttal regardless of the precise numbers, which are only tangentially relevant.
I mean, they're relevant in the sense that, whatever the gap is between the predicted and observed MZ twin concordance, that's the portion that we have to ascribe to nongenetic factors. And so even on his own premise, his definition of schizophrenia being "80% heritable" still ends up being support for the hypothesis that it's mostly something else.
And I guess that's the thing that annoys me here, and the reason I didn't just jump straight to "Good, we're saying the same thing, I bring you the fabled gene-environment interaction, the 2nd hit of the 2-hit model".
I see a guy very good at convincing people of things by leading them through a labyrinth of logic. But the one he's built here is so complex and circuitous that even he has lost himself in it, and sort of admits this--but still doesn't consider the possibility that there is no center to find.
What I mean is that the MZ twin concordance is not very meaningful on its own. Anything below 100% shows that the environment matters to some extent, but it doesn't show how much. What is meaningful is a comparison of the MZ and DZ concordance rates. Using Falconer's formula (not the only method of estimating heritability, but the simplest for illustration purposes), 80% heritability is what you would get from 90% MZ twin concordance and 50% DZ twin concordance, but also from 50% MZ twin concordance and 10% DZ twin concordance (whereas a 50% MZ twin concordance would imply 0% heritability if the DZ concordance were also 50%). Your original post talked about the microbiome being heritable only in the sense of getting it from your parents, and your only mentioned identical twins only in the context of having less than 100% concordance, with no comparison to fraternal twins. This is why Scott Alexander spends the first 4 of 6 points saying you don't understand the evidence for the role of genes (as he notes, "This doesn’t describe a process which could be expected to differ between identical and fraternal twins!"). You make the point much more strongly in this post by showing that a person's genetics do affect their microbiome.
Another important point in your favor, which I have not seen you bring up, is that genes can affect someone's response to a microbe by making one person have a latent infection and another person exhibit symptoms. For example, tuberculosis and leprosy are infectious diseases that have been with us for thousands of years, and still exist today. There has been strong selection on genes that cause resistance to these diseases (https://www.biorxiv.org/content/10.1101/2024.09.14.613021v1.full), so plenty of people can have a latent infection and show no symptoms. However, in the evolutionary arms race, the microbes keep evolving, too, so they haven't fully disappeared, even though they are less common now. Interestingly, that same paper shows that the polygenic score for schizophrenia risk has also substantially decreased over thousands of years. I would guess that, like leprosy and tuberculosis, the microbe(s) causing schizophrenia have been with us a long time, so there has been time for people to evolve resistance to them, which is probably part of why there is a genetic signal to schizophrenia. People can evolve either to fight off the microbes entirely or evolve to tolerate them without any ill effects.
Thank you for educating me on this! An interesting point to consider: leprosy and tuberculosis appear to be straight negatives. Schizophrenia, in this model, is just the tail end of the bell curve for abundance of a common microbial gene, or perhaps a two-hit involving e.g. a para-endogenous MAOI, and perhaps there is a significant advantage to having a small amount of such a gene, or one without the other.
Is R. gnavus beneficial for the host in small doses? Maybe, maybe not. I don't know the answer to that. Do you? You should probably figure that out before you start trying to cure people of it.
Yeah, I didn't buy his rebuttal against the ideal of schizophrenia being microbially caused, for some of the reasons you mentioned. An illustrative example that gives some context is that the heritability of leprosy is 79% in this twin study (https://www-jstor-org.ezaccess.libraries.psu.edu/stable/2396811?seq=7), yet no one would call leprosy a polygenic disorder.
I think it would be helpful to clarify your stance on a couple of questions, as it is hard to tell exactly what you are arguing:
1. Approximately what proportion of schizophrenia cases do you think are caused by microbes (of any kind)?
2. Approximately what proportion of schizophrenia cases do you think are caused by R. gnavus specifically?
In your original post, it comes across like you think the answer to #2 is 100%, but in the comments on Scott Alexander's article, you seem to walk it back and imply that you think the answer to #2 is fairly high but less than 100%. Personally, I think #1 is well above 0%, and I could potentially see it being 100% but possibly somewhat lower. I don't have enough information to make a good estimate for #2, but my suspicion is that it is lower than you seem to think it is.
You have done a great job synthesizing several pieces of evidence into a plausible-sounding hypothesis. However, I have found that it is very easy to construct a plausible story but then have it turn out not to hold up. What will really convince me one way or the other is how well your hypothesis makes predictions that hold up. Obviously the best way will be the clinical trials you say you are going to do. Still, there are various other testable predictions that could help support or refute your hypothesis (depending on what your answers to questions #1 and #2 above are). I'll put a few questions I have in the comments to this comment. I'm curious to hear your answers to any or all of those (probably some are not answerable given the available information, but some may be).
There are various other known risk factors for schizophrenia. A true explanation for the cause should be able to unify all of these. The Scott Alexander article mentions a few at the end: "Other risk factors include urban environment, winter birth, prenatal stress, and maybe cat exposure." Ironically, even though he seems dismissive of a microbial explanation, these seem to me to point to one. Most infectious diseases are more common in urban environments. Plenty of illnesses (especially viral) peak in the winter. Prenatal stress could be explained in other ways, but it might also impair the immune system. Cat exposure has been suggested to cause schizophrenia via Toxoplamsa gondii exposure. My intuition is that these risk factors point to various microbes causing schizophrenia; however, if there is a single cause, or a most common one, then they should all be linked to that particular microbe. Is R. gnavus abundance linked to any or all of these risk factors?
How robust is the schizophrenia-R. gnavus association across individuals and studies? In the study you cited, I can't tell how susceptible their analyses are to outliers. Is the result driven by, say, a few people with 1000x enriched R. gnavus, and the rest had normal levels of it? Or was the elevated abundance seen in all or almost all people with schizophrenia? How many other studies have looked at schizophrenia and the gut microbiome, and did they find the R. gnavus association or not?
Schizophrenia is negatively correlated with IQ (from before symptoms set in; https://psychiatryonline.org/doi/full/10.1176/appi.ajp.2008.07081242). Does your hypothesis have a good explanation of this association? For instance, would the hallucinogenic by-products of R. gnavus interfere with brain function or development even at low doses, or would a larger brain require a higher dosage to exhibit symptoms?
THANK YOU! This is the violence I was asking for!! On this point: consider phenethylamine, and specifically the brain's reaction to it. As mentioned in the original post, PEA induces neuroinflammation, and the brain responds by upregulating MAO-B.
This degrades phenethylamine, but also dopamine and norepinephrine, which are critical for learning and memory.
Phenethylamine is static on the broadcast channel. The listener cranks down the volume on his receiver, to avoid having his ears blasted out, but now he's going to miss the quieter parts in the symphony that he was hoping to hear.
I will be reaching out to the authors of that study (Vasileva et al) for their raw data, so we can look at the question about outliers. My feeling is that, if it had just been a couple guys, they would have mentioned it, but you never know.
Replication: Ioannou et al did a similar study in schizophrenia a few years back. They used mOTUs, which is a sort of sloppy metagenomics approach, but they found Eggerthella and Lachnoclostridium elevated in 3 cohorts. Lachnoclostridium is a now-defunct genus which has been mostly moved to Enterocloster, the other gut bacterial genus known to have an aromatic decarboxylase. See the "FKA Lachnoclostridium" tab on my "Big Beautiful Metabolomics Table"; I'll add the taxa reported as elevated in schizophrenia by Vasileva (2024) so you can compare.
I think some of the strongest evidence for a genetic cause of schizophrenia is that having an older father is linked to increased risk (https://jamanetwork.com/journals/jamapsychiatry/fullarticle/207596). Men keep producing new sperm throughout their lives, so older fathers pass on more new mutations. The association of paternal age with schizophrenia (and many other disorders) is used as evidence that de novo mutations play a role in these diseases. Does your hypothesis have a way of explaining this association?
Thanks for writing this rebuttal, was wondering if you were going to give a response in a larger post. Whilst I have no horse in this race, I was fairly annoyed at comments pointing out your bias as a microbiome scientist, whilst ignoring Scott Alexander (and the rationalist movement as a wholes') long and sordid obsession with DNA and genetics. Great post
So special pleading, moving the goal posts, taunts, no acknowledgements of the bad arguments you made. Odd that you didn't explain about the 10 minute limit truncating your post over at ACX, or why you didn't make the same justifications. No acknowledgement that the bone marrow transplant article goes directly against a claim you made to me in the comments of your last post.
I'd like to to take you seriously but not only am I now skeptical of this claim, but all your others as well because who knows what you've missed but are prepared to double down on.
¯\_(ツ)_/¯
https://youtu.be/-Jtpm0uWG38?si=qOVyjT4pfOOey94F
Thank you for sharing! That's possibly the most horrifying thing I've ever heard. Are you still dealing with those episodes?
Were you replying to my comment?
If so, since diagnosis last year, no. :)
My daily mood stabilizer has been life-changing, especially for mental noise and executive dysfunction (which is not what it's indicated for). A psychiatrist would also have me on daily antipsychotics for better protection against mania, but they work by blocking dopamine receptors, so you can imagine how that would suck.
So far every 2-3 months, I have a slow build of energy, confidence, sense of deep meaning in life... which I have to nip in the bud lmao.
That's early hypomania, which could potentially lead to a full episode- but generally there are a few weeks to catch it. When that happens, I take a fast-acting antipsychotic and am a bit sleepy and flat for the course (~a week). Taking them only as-needed is risky and not what a psychiatrist would tell you to do.
But I meticulously track my state and behavior (like, with a daily spreadsheet that has 20+ rows), meditate (helps with self-awareness), and have my husband here who knows me well. I feel confident a big episode won't happen again.
Oh you feel confident, huh? 🧐
Sorry, not funny, but I had to seize that thread because it's the perfect example of something that came up in the comments last time, about why we haven't seen a case report of home FMT in an attempt to treat schizophrenia the way we have with bipolar.
Because any disorder which involves impaired judgement or capacity for self-awareness and reflection gets really slippery really quickly, particularly in the case when there's the potential for *advocating for yourself in an attempt to manage the condition* to be mistaken for *a symptom of the condition*. You'd think that'd be a narrow band, and yet...
All good, I lol'd.
Funny, I'd think this concern would apply even more to bipolar because a self-report of I CURED MYSELF is classic hyper-confident, grandiose, feel-amazing manic behavior.
Though the main reason I'm taking the reports I read with a grain of salt is because people can be euthymic for years between episodes with no treatment at all.
Good point! Doubly tricky because someone might get "cured", go 5 years, then have another episode...but if it's tough to say what those 5 years would have been like without the FMT.
Are there populations that don't use antibiotics maybe the Amish? I wonder if they would have different schizophrenia risks