Skip to main content

Animal Testing

Autism Mouse Models for the Microbiome?

Many readers will have seen the paper that just came out on a possible mouse-model demonstration of a connection between autism and the gut microbiome. It’s certainly generated a lot of headlines, and its very title guaranteed that it would: “Human Gut Microbiota from Autism Spectrum Disorder Promote Behavioral Symptoms in Mice“. I found that a pretty startling claim, but I’m well aware that GI function is disturbed in many patients with autism, and I’m open to speculative hypotheses. But I don’t think this paper proves its case.

I’m not alone. A number of other observers are raising questions. When you dig into the paper, you find that the statistical power of the study is. . .extremely low. It involves faecal samples from 5 control volunteers and 11 patients diagnosed with autism spectrum disorder, and even those latter donors are divided into three subgroups based on the ADOS severity scale. As many are pointing out, it gets down to the point of a 3-control 5-ASD-patient analysis, which just seems too small to reach a meaningful conclusion in a question this complicated. Especially a question this complicated that’s being addressed by mouse behavior observations. Anyone who’s done CNS drug research will tell you that rodent behavior is complex, difficult to interpret, and can be influenced by a lot of things other than what you’d like to imagine. And before you even get to that, the tiny number of human donors really makes me wonder. It should go without saying that the human microbiome is quite variable, person-to-person, and I have trouble believing that this is (or even can be) a representative sample.

What’s more, the effects were noted in second-generation mice: they took normal C57 rodents and colonized them with donor samples, then bred them, and re-colonized the offspring, and that’s where the effects were seen. The paper says that these mice were “either sampled (feces, serum, brains) or behavior-tested”, which makes me wonder if the animals used for the behavioral conclusions had their microbiomes examined directly at all. The mice and their behavior are correlated to the original donor metadata, but not (from what I can see) to what’s present in their own microbiota – I keep thinking that I’m not reading this correctly, but that’s what I’m getting.

And in those correlations with original donor data, they’re going so far as to mention a connection with donor age, but you’d need an even *larger* set of donors to make any such claim, wouldn’t you? When you look at the S1 supplementary data, those ages range from 3 to 11: how wide is that? Does anyone know?

Another problem I have is with the idea of there being mouse behaviors that can be tied to human autism spectrum behavior in the first place. These doubts creep into one’s consciousness after you’ve done CNS drug discovery work, and they can never be fully expunged. Anyway, here’s a neurogenetics guy, who works on rodent models, who has the same doubts, so it’s not just me. To the best of my knowledge, there are no generally accepted mouse models of autism.

The paper goes further, to identifying metabolites produced by the gut microbiota and testing these separately to see if they’re involved in behavioral effects. That seems like a rather broad jump, right to actionable small molecules. Such cases have of course been suggested, but even those are preliminary and have more statistical power behind them than this paper does. What small molecules do the authors propose? Taurine and 5-aminovaleric acid (5AV). Both of these are characterized in the paper as “weak GABA agonists”, which is true, especially the “weak” part, but note that the earlier work that I linked to is talking about gut production of GABA itself, and even that needs a lot more work. Interestingly, the authors also note changes in the gut levels of genistein and daidzein, which are isoflavones that are coming from the soybeans in the mouse chow (and are notorious frequent hitters in in vitro assays, for what that’s worth).

They go on to administer large amounts of taurine and 5AV to pregnant mice, and do behavioral tests on the offspring. And by this point, I think that things are pretty far out on a limb of loosely connected assumptions, with a great many alternate explanations available. The amounts of these compounds being given are *far* over any reasonable production from gut microbes – and indeed, if the paper tests for blood levels of these in their gut-altered mice, I have missed it. The authors also predict behavioral effects from the genistein and daidzein concentrations, too, and I really don’t know what to say about that.

There’s overall a cavalier attitude towards generating hypotheses in this paper and chaining them end-to-end. You have to believe that these effects are robust enough (large effect size) to show up in very small samples, that the mouse behavioral effects are indeed connected with human autism, that the metabolome changes seen are causative and that the correct ones have been identified, that the high-dose recapitulation experiments are connected to what’s seen in those earlier animal models, and on and on. . .and you also have to believe that (as mentioned) alternate explanations have been excluded, when to my mind, they haven’t been at all. From an outsider’s perspective, the whole paper seems aimed at showing that small-molecule gut metabolites are big players in autism, and that’s a huge conclusion to put on top of such a slender foundation.

And indeed, the end of the paper discloses that its various authors have filed patent applications related to just that hypothesis, and helped to found Axial Biotherapeutics, which is dedicated to that idea. Now, I work in the industry, of course, and I have no problem with industrial scientists publishing their data to make their case. But papers from any source need to back up what they claim. I am also open to the idea that the gut microbiota are involved in autism (and indeed, in other human CNS disorders), but I will accept or reject such ideas based on more evidence than is in this paper. In that context, it should be noted that there are reports of microbiota transplants directly on human patients already, with long-term follow-up and a call for larger, more controlled trials to test these effects. It would seem that such trials are certainly worth running. Taurine and 5-aminovaleric acid, though, have a long way to go.



53 comments on “Autism Mouse Models for the Microbiome?”

  1. Project Osprey says:


    Is Red Bull going to have to carry a black box warning for possible autism now?

    1. TB says:

      The paper suggests a protective effect of taurine on the offspring of mice treated during pregnancy. Not that taurine “causes” autism. No need for a black box warning on Red Bull. Your energy drinks are safe and not the reason for your autism.

  2. Lars says:

    Yeah, or cat food (it is an essential nutrient for cats).

    Also, this should probably be tagged with Autism?

    1. loupgarous says:

      Cat food, hell. Meat and fish deliver 58 mg/day to omnivores. But even vegans manage to show detectable (but very low) levels of taurine.

      1. DrOcto says:

        But one could argue that vegans display increased behavioural abnormalities.

        1. loupgarous says:

          A classic case of equivocal correlation, I’m afraid. 🙂

  3. johnnyboy says:

    ‘Behavioural symptoms’, in SECOND generation mice ? Without even checking that there is an actual difference in the mice’s microbiomes ? What complete and utter garbage. And in Cell, too. You can just publish any old crap these days if it’s got ‘microbiome’ in it. If only they’d have added some AI angle to this, they’d surely have made it into Nature.

    I’m sure they’re going to get absolutely smothered in VC millions.

    1. RM says:

      I could believe that behavioral symptoms only show in second generation mice, but my initial hypothesis there would be that it was the gut biome of the mother during pregnancy which caused the issue, rather than the gut biome of the children. (With the mode of action being some effect on fetal brain development. – Though not necessarily through direct action of a neurotropic small molecule. It could also be an indirect effect due to, e.g., modulation of the mother’s immune system.)

      But from what Derek says, it sounds unlikely that the authors have adequately investigated and tested for those sorts of effects. (Or even adequately considered them as a hypothesis.) So we’re still way off in the realm of hypothetical sky castles.

    2. Calvin says:

      They got $25M in a series A already. Note that the CEO is a partner at one of the VCs involved, Longwood Fund. And this is Christoph Westphal’s VC fund (Sirtris!!!!!!!). So if your bulls**t detectors weren’t already buzzing after reading the paper they are now! And to cap it off they’ve roped in “high net worth individuals in Southern California”. So most likely a bunch of rich parents with an affected child who are desperate and have fallen for the snake oil on sale here…..Sigh.

      1. loupgarous says:

        Apparently venture capitalists don’t read p-values on papers like this, or have heard that “extraordinary claims require extraordinary proof”.

    3. Irene S says:

      Not just because it says microbiome. Always depends on who says/reports/did the work. Many people work on it but don’t get Cell papers

    4. Shamus says:

      “‘Behavioural symptoms’, in SECOND generation mice ? Without even checking that there is an actual difference in the mice’s microbiomes ? What complete and utter garbage.”

      Of course you look at 2nd generation, because they are studying a developmental disorder.

      They also did check this generation’s microbiome and metabolome (in serum and colon contents). Figures 1I-K are the summary of microbiome sequencing from these offspring. All of Figure 2 is devoted to the microbiomes present in the offspring of the original recipients, and the correlation of specific microbes present in the offspring to the mouse behavior. Whereas, all of Figure 4 is displaying metabolomics data from these offspring.

      They sequenced at all levels- the original human donors (summarized in Figure 1C,D), the recipient mice receiving the donor microbiomes (summarized in Figure S1), and the offspring of these recipients (conceived and developed in a maternal metabolic environment influenced by the different microbiomes) Figure 1I-K, and Figure 2.

      All of this is in the paper, not sure why this commentary states otherwise.

  4. Wile E. Coyote, Genius says:

    Key information that I see left out is whether or not the mice were vaccinated.

    1. Anon says:

      You win today’s Interwebs Gold Medal for that comment. Reminds me of my CRO days when I sat through a client meeting where the SVP of QA for a pharma company held forth on how vaccines were evil since they caused autism in her child. We sat embarrassed and offended, while being kicked under the table by our SVP as a warning to keep quiet and listen to the crackpot theorizing since we didn’t want to lose the client!

  5. SirWired says:

    How did Cell allow a paper with such sweeping claims based on such thin data? This looks like poster-session material, at best…

  6. Kent G. Budge says:

    I have an autistic son and three autistic nephews.

    The notion that autism causes gut problems seems so much more plausible to me than that gut problems cause autism that I’m going to require some pretty remarkable evidence to take a paper making that claim seriously.

    From my perspective, this paper falls well short of that.

    1. Charles H. says:

      I think you’re theorizing in advance of facts. But so are they. From the evidence I’m aware of the causation could go in either direction, or could even be “both symptoms are derived from (the same) some other effect”.

      I don’t see why it isn’t as reasonable for gut bacteria to cause autism as the other way around. There are definitely known links in both directions.

    2. MMK says:

      @Kent I couldn’t agree more, being a father of an autistic teen, and a husband of a therapist.

  7. Isidore says:

    According to Wikipedia (yeah, I know, unfortunately the 2006 FEBS Microbiology reference is behind a paywall) taurine amounts to 0.1% of human body weight, so the average child who weighs, say, 30 kg, has 30 g of taurine in him/her, much of it in the large intestine. I wonder if there’s a significant difference between autistic and non-autistic children with regard to their taurine levels.

  8. BernYeeMetabolicProducts says:

    Its pretty well established that autism changes the metabolic profiles of its sufferers, producing low molecular weight carboxyxlic acids and oxidized species. There are also several publications on the role of HFCS, processing of HFCS and mercury contamination that can also change metabolic profiles- which was quietly published a decade ago.

    You are what you eat and if stressed you do change your metabolome and microbiome- in any which order you feel comfortable with.

    Its America’s Dirty Little Secret

  9. Louie A. Freedom says:

    Is the twitter-linked neurogenetist guy related to Dr. Freddie Mercury

  10. Bioscientist and Cynic says:

    Speaking as the parent of an autistic kid, autistic children are notoriously *funny* / picky eaters (often insisting on weirdly restricted diets, only eating one thing for days at a time, switching to refusing something they previously ate enthusiastically/exclusively, sometimes hardly eating at all)… so again, ‘autism changes your metabolic profile’, esp. in children. would make me think ‘Autism changes your eating habits which in turn changes your metabolic profile’ as a minimal hypothesis. Was deeply sceptical about the stuff in this paper just reading the abstract, let alone now I’ve read the blog.

    1. Carl Bar says:

      As someone recently diagnosed as ASD i can confirm the picky as all hell dietary thing.

  11. Simona says:

    ADOS is not a severity scale, it’s a diagnostic questionnaire that says nothing about ‘severity’ (whatever the that even means in the case of autism, if I engage in ~stereotyped behavior~ in public but can understand conversational nuance well, and my friend does the opposite, which one of us is more severely autistic?). The fact that it’s used as such proves how methodologically poor most autism research is.

  12. Lane Simonian says:

    Regardless of the problems with this study, there is mounting evidence of a connection between gut dysbiosis and various neurological conditions/diseases.

    Certain bacteria in the gut contribute to the production of glutathione–a critical antioxidant compound which is low in people with autism spectrum disorder. Certain neurotoxins such as mercury and various pesticides can lower glutathione levels and thus increase their damage to the brain. What is unclear at this point is the role that oxidative stress and concomittantly low levels of glutathione in the gut play in potentially increasing oxidative stress and declining levels of glutathione in the brain and whether this is a small or potentially large contributor to neurological damage.

  13. Bioscientist and Cynic makes a key point. Assuming the data presented in the paper mean something about the microbiota that were transferred (which I’m not convinced of), diet shapes microbiota structure to a huge degree. So, are the differences due to autism, or restricted diets? No increase of the n/group is going to tease this apart.

    What is often (always) lacking in these microbiota transplant experiments is a transplantation from an unrelated but also “non-healthy” group. If the microbiota of an obese individual or an individual with IBD transmits the same phenotype, then while we are potentially learning something about how the microbiota can modify phenotypes, it may be unrelated to the condition the non-healthy person suffers from.

  14. Not in neuro! says:

    It seems less important how many marbles a mouse hid under its bedding than how many marbles each mouse was able to recover afterwards as this would be a direct measure of how many mice lost their marbles.

    Seriously though, germ-free (GF) mice are anything but “normal C57 rodents”. The absence of all microbes causes some pretty significant changes in intestinal morphology and physiology that are still being described. Using second generation animals gets around this and allows for potential developmental effects to manifest. The group absolutely made the right call doing it this way.

    The thought of doing behavioral assays with gnotobiotic mice is enough to activate my fear-response. Since there’s no such thing as a germ-free behavioral suite, mice would be exposed to environmental microbes over the weeks of testing. I’m guessing this pretty well trashed the humanized portion of their microbiomes, but you don’t know what you don’t measure. Running microbiome analysis on (presumably) matched controls can suggest what was likely engrafted at the start. They could also have made the argument that any developmental effects could still be active, but that’s not the direction they ran.

    It’s sad to see ‘om-omics’ being used so poorly. Here they are generating piles of microbiome, metabolome and transcriptome data and their primary readout is a behavioral assay? There’s a reason I don’t work in neuro.

  15. loupgarous says:

    Another case of what Richard Feynman called “cargo cult science”. It seems rife in behavioral science of any kind (which is why Peter Boghossian of Portland State University sent his spoof articles mainly to sociology journals)

  16. Nameless says:

    On the one hand developing small molecules that act in the brain (brain cancer) or against CNS disease is extremely difficult. First the molecule has to absorbed, survive the liver intact then distribute into the CNS and all that in sufficient quantity that it achieves therapeutic effect.
    On the other hand these arguments go “these bacteria make stuff and poof, your brain melts”. I have trouble believing that animals as a whole have gut bacteria that will influence their behavious in such drastic ways. In the wild animals died when they showed abnormal behaviour.

  17. Trish says:

    Have you looked at any of the human studies related to the microbiome in ASD (Stanford, Baylor, ASU, Calgary…)?

    The FDA recently fast-tracked a breakthrough drug (CP101) for microbiota replacement therapy (MTT) in Autism, following compelling outcomes in a phase 1 MTT trial at ASU, where 44% of subjects no longer qualified for autism diagnosis, and follow-up paper showed results have held for 2 years following treatment(!).

    This mouse paper is far from perfect, as you point out, but fortunately there are many others demonstrating microbiota dysbiosis in ASD. These are exciting times!

    1. loupgarous says:

      Did you mean to link to a different press release?

      This one says:

      “Previously, Adams and Krajmalnik-Brown, Finch’s collaborators at Arizona State University treated 18 children affected by ASD with Full-Spectrum Microbiota in an open-label study. They found the treatment was well-tolerated and led to a 77% reduction of GI symptoms and a 24% reduction of core ASD symptoms at eight weeks post-treatment. A recently published study that followed these 18 children for two years after initial treatment reports sustained improvements in GI symptoms and core behavioral ASD symptoms.”

      While this is an encouraging outcome for the GI symptoms, it’s more equivocal for reduction of autism spectrum disorder symptoms than the outcome you mentioned in your post.

      I’d like to see the publication containing the results you mentioned, if you have a link to it. A 44% remission of autism spectrum disorder would be impressive indeed.

      1. Trish says:

        It is impressive, and I’m surprised it hasn’t been covered more in the press.

        From the paper:

        “At the beginning of the open-label trial, 83% of participants rated in the severe ASD diagnosis per the CARS (Fig. 2a). At the two-year follow-up, only 17% were rated as severe, 39% were in the mild to moderate range, and 44% of participants were below the ASD diagnostic cut-off scores…”

        1. loupgarous says:

          Thanks, I missed that part of the paper.

    2. Mikalobakus says:

      Might be a clue to what’s going on. But my immediate thought is, if the microbiome causes the problems, how/why did it get into the pathological state in the first place? Fixing it with pills may work for a while, but is it addressing the underlying cause?

      1. Trish says:

        Many studies have shown correlation of ASD regression with c-section birth, early antibiotic use, lack of prolonged breastfeeding, and other forms of microbiome disruption.

        The “pill” is actually able to reseed the gut with missing microbes. Along with appropriate diet and avoiding further broad spectrum antimicrobial use, this can establish a healthy microbiome that is capable of serving all of the necessary functions, including proper digestion, immune system support, and neurotransmitter synthesis.

        We only have 2 years’ worth of follow-up data, but so far results are holding well in the treatment group!

  18. Eugene says:

    Behavioral changes in rodents infected with Toxoplasma gondii indicate that micro-organisms can affect behaviour.

    1. loupgarous says:

      The literature on behavioral change in humans who are seropositive for Toxoplasma gondii but not suffering acute symptoms or evidence of central nervous system T. gondii cysts has yet to show a causal link, merely correlation. Some behaviors studied (such as lack of personal tidiness) may increase the likelihood of exposure to T. gondii, so those correlations could be equivocal evidence.

      1. Eugene says:

        Agreed, just an example of a micro-organism affecting mammalian behavior. Human behavior is more complex and it is tougher to untangle the knot of cause and effect. A lot of the ASD research sounds like the early days of Alzheimer’s research (Aluminum cookware). Humans have evolved to deal with these organism and even benefit from them, but there could be cases of where the immune does not keep them perfectly in check. These micro-organisms are also constantly evolving. The human micro-biota may have subtle interactions that are difficult to pick out from the background noise.

    2. johnnyboy says:

      Toxoplasma cysts actually lodge into the rodent brain. So there’s a bit less of a stretch for it to have a direct effect on behavior, than the bacterial composition of the gut flora.

  19. Nile says:

    As it is, and not as any wishful thinking on my part would see it, this is one more example of the ‘replicability crisis’ in the Life Sciences, and it’s going to go the same way as the ‘Depression Gene’ – a rigorous study will bebunk a vast edifice of irreproducable results and statistically-flawed conclusions, after dozens of startups and nearly-mature biotech companies burn their way through hundreds of millions of dollars of value destruction before the word gets out.

    The replicability crisis is not occurring in a vacuum; nor even in the oxygen-depleted atmosphere of academia. It is occurring in the real economy and in the real world of investments and markets.

    This is a matter of practical interest to wealthy individuals, institutional investors, and the regulatory authorities.

    You really, really don’t want the SEC to get involved in a regulatory push for minimum standards of replicability and rigour in published research used to support a biotech investment.

    But that external intervention is a likely outcome unless this mess gets sorted out within the community of scientists: the SEC care a lot more about honesty to investors and the integrity of the capital markets than we seem to care (or at least, more than we seem to be able to act) about the integrity of research.

  20. jbosch says:

    exaggerated claims = (# of MD’s/ # of PhD’s) x (# of omics studies)

    I think there’s a clear correlation in various publications.

    1. johnnyboy says:

      I like your formula, but you’d need to work in the numbers of MD-PhDs into it. In my experience, this double-diploma combo has severe inflationary effects on the diplomate’s belief in his own bullshit.

  21. Mark says:

    The elephant in the room with autism is that it’s poorly defined. Diagnosis is wobbly; at the moment we’re not even sure of the level of autism in girls because we think there may be under-recognition because autistic girls behave differently to autistic boys – in effect, we can’t even diagnose it reliably in half the population. Diagnosis depends exclusively on behaviour, and yet it’s accepted that autistic individuals’ behaviour varies enormously. It’s quite possible that a test group of autistic people actually consists of a hotch-potch of people who have radically different biologies, all of which we happen to label as ASD for want of any better way to classify. For this reason, study size is a real issue. If your underlying population actually contains 10 different underlying populations, the study size is 1/10th what you think (and the study is also confounded by different variations).
    Follow-up studies showing improvement have to be viewed with _extreme_ caution too; regression towards the mean will always look like improvement, and many families with autistic children will be doing their utmost to help their child regress towards the mean. Older kids are less often diagnosed than younger, because their behaviours are often more subtle, meaning that a few years after a study it’s quite likely some proportion of the study population will no longer be diagnosed. It’s pretty hard to design really reliable double-blind studies for many of the things people most associate with autism (double-blind dietary studies are atrociously hard).
    The whole area is difficult but worthy of attention. I admire those who put in the effort to carry out good studies. It’s sad to see things muddied by another poor-quality piece of work.

    1. Trish says:

      Your points are well taken, but It is important to note that this study was kids 7-17, and 83% of them were evaluated as “severe” at the start of the trial. These kids were much older and much more affected than those who might be able to improve symptoms through rigorous early intervention, etc. Kids in this older age group don’t make such drastic changes in 8 weeks.

      The current study (phase 2) is looking at adults with autism aged 18-60. Presumably, their parents are no longer rushing them around to 40 hours a week of intensive therapies. These results will be much more difficult for the “it’s all genetic” group to dismiss…but I’m sure they will still try!

  22. dave w says:

    Basically what Mark said above – all this assumes that “autism” is a definite neuro-clinical condition in the first place. I believe this is far from being established, and am skeptical that it is actually true.

  23. Anonymous Researcher snaw says:

    There are few, if any, mouse models of any human CNS disease that I would give much credence. Even when done well, with highly-experienced researchers and well designed experiments, Neuroscience research is very difficult to get right.

  24. chimaera says:

    If they’re interested a possible relationship between gut microbiome and autistic behaviour, it baffles me that they bothered with a mouse study at all. Recruit a sufficiently large number of neurotypical and neurodiverse people and take gut micriobiome samples. Job done! If there’s a meaningful correlation it should show up here without introducing the complication of dealing with mouse models.

    1. Trish says:

      There are many human studies taking place on the microbiome in ASD. Stanford, Baylor, UCLA…many institutions are contributing to this growing body of work.

    2. Craig says:

      Yes, a human study like you describe would be a great start, and are/have been performed.
      The question becomes what do those correlations to the microbiome mean?
      How do you test whether one microbiome has a differential outcome on any sort of host physiology without an experimental model?

  25. TommyBoy says:

    “The mice and their behavior are correlated to the original donor metadata, but not (from what I can see) to what’s present in their own microbiota – I keep thinking that I’m not reading this correctly, but that’s what I’m getting.”
    The nomenclature isn’t the most clear, but they do all of their testing and correlations with those 2nd generation mice (oTD and oASD).
    Yes, Figure 1H compares the 2nd generation recipient mice to metadata from the original donor. But Figure 2E compares the microbiome of the recipient mice to their own behavioral scores. (But this was only for male mice, as females did not show a robust change in behavior influenced by which microbiome they received).

  26. DTX says:

    It was interesting to see the notes implying an autism link with mercury exposure. In the US, exposure to environmental mercury peaked in ~1960. In the 1950s and before, the “pink disease” (acrodynia) was common in infants due to mercury overexposure. At that time, mercury-based calomel was used as a teething cream in infants, mercury-based diaper powders were used, and mercury had many other medical & other common uses. And we used to paint organic mercury on cuts of kids (mercurochrome). Yes, in the good old days…..

    Of course, those mercury proponents would note the above is based on history & fact and therefore can’t be true….

    1. TommyBoy says:

      What “notes” on mercury exposure? This commentary, nor the paper it is attempting to discuss, describes anything about mercury.

  27. TommyBoy says:

    “if the paper tests for blood levels of [taurine and 5-aminovaleric acid] in their gut-altered mice, I have missed it.”

    For 2nd gen. recipient mice, the authors did screen the serum for metabolites. So, the “gut-altered” mice were tested in the serum for these specific compounds, not just their colon contents. This is Figure 4C, F and Table S3. 5-AV was qualitatively decreased in serum in mice receiving microbes derived from ASD, but did not reach statistical significance (p=0.07), and no significant difference with taurine (p=0.11).

    In the mice that were given the metabolites during pregnancy (with the effect observed in their offspring), they did test serum as well. They also tested colon contents, amniotic fluid, and fetal brains. These data are shown in Figure 6B. As the authors go on to show that metabolite treatment was only effective during pregnancy (and not after), it stands to reason that these would be the appropriate compartments to test. Taurine administration did not change its concentrations in these compartments, but 5-AV treatment did increase in serum, amniotic fluid, and fetal brain.

    So you must have missed these data, as well as the microbiome comparison data in offspring/recipient mice.

Comments are closed.