To leave no one behind, the microbiome is all the organisms (mostly bacterial, but viral, fungal, protozoal and God knows what else) that live in our gastrointestinal tract.
A tremendous article in the current issue of Neuron (vol. 114 pp. 196 – 211 ’26) dissects the literature on the relationship of the microbiome to autism. A lot of what follows is a direct lift from the article, modified for clarity and verbosity. I’ve put in my own comments, but it is not my work.
We now have an average of 102 articles each year on the relationship of the microbiome to autism since the first article on the subject appeared in 2011.
Two population studies have found no increase in rates of autistic symptoms in the population, even as rates of diagnoses became higher in younger cohorts (2011, 2015 — couldn’t they have come up with more recent references?)
Rates of GI symptoms in children with autism are consistently significantly higher than in typically developing (TD) children (e.g., 33% versus 15% (An. Pediatr. (Engl Ed). 2023; 99:102-110). GI comorbidities are also high in people with intellectual disability or developmental delays more generally, including Down syndrome, for example.
Human observational studies. The twelve most cited studies (107 – 1,219 citations which say there is something to the association are reviewed. The make up much of the evidence base of the meta-analyses. The sample sizes are small (7 to 43 people). No hypotheses were given at the start. There was failure to account for multiple statistical tests (in all?) and no replication.
The dangers of conducting exploratory analyses of high-dimensional data with small samples are now well known and have contributed to the replication crisis across many disciplines (particularly GWAS < Genome Wide Association Studies >).
Major league baseball with little statistical training figured this out a century ago. Joe Sixpack with 6 hits in 8 at bats with a batting average of .750 was not a candidate for the best hitter in baseball (now 502 at bats are needed). They knew statistical flukes were likely with small numbers.
As you might expect the findings of the 12 studies were inconsistent; which microbes in autistic kids differ from normals, how they differ from normals, which metabolites differ from normal, levels of microbial diversity (lower, the same or higher).
The only two studies using sibs as controls found no significant differences.
Even for factors with known impact on the microbiome, including age, body mass index (BMI), fruit consumption, or a range of metabolic conditions, require samples of several hundred to detect the largest effects and thousands for the smallest. These conclusions align with those of Rothschild et al who found robust microbiome associations with various phenotypes (e.g., age, BMI, diabetes, and smoking status) in very large samples (>34,000 individuals). But they report that “when sub-sampling our large cohort into smaller-sized cohorts, we found that even cohorts of 1,000 individuals have significantly lower average accuracy of associations between bacteria and phenotypes.” They conclude that “these results may explain the relatively low agreement that exists across studies in the set of bacteria associated with different traits and conditions, and they call for employing larger cohort sizes in microbiome studies.”
The inference seems to be that, even if specific findings are inconsistent, most of the (published) studies that have examined the question have found some differences in the microbiome of autistic people compared with TD controls. This is often couched in more general terms with claims of “dysbiosis” of the gut microbiome in individuals with autism. As pointed out by researchers in the field, this is a vague term, which effectively captures any profile of differences, regardless of consistency. Two studies showing directly contradictory profiles of change could both be classed as evidence of dysbiosis.
Even worse, the effect sizes, when found are small (3 – 5%).
The use of mouse models of autism is criticized — are there even ‘autisticLike behaviors in mice.
The most widely used mouse assays of behavior are (1) marble burying, (2) the three-chamber social approach, and (3) ultrasonic vocalization (USV) e.g. high pitched cries we can’t hear. In marble burying, a number of marbles are placed on top of the bedding in the mouse’s cage and the mouse’s tendency to dig in the bedding leads to them becoming buried. This is believed to tap into stereotyped and compulsive behavior and to be related to repetitive behaviors in people with autism. The three-chamber social approach assay measures the preference of an animal to explore another mouse compared with an inanimate object or a novel mouse compared with a familiar one. This is taken to be a proxy for social cognition and behavior in people with autism. Rodents emit USVs in diverse social situations across their lifespan, and these are taken to have a communicative function. USVs can be assayed in the lab under various conditions and are claimed to be relevant to the core symptoms of communication deficits in human patients. One issue with this particular assay is that both increases and decreases in USVs have been interpreted as “autistic-like” behaviors. In addition to these behavioral tests, a wide range of other assays may also be performed to assess arousal, locomotor activity, motor coordination, anxiety, stereotypy, learning and memory, and other parameters.
Attempts have been made to put the human fecal microbiome into mice and look for effects. But 85% of bacterial genera found in the mouse microbiome not present in man. A substantial proportion of the species that comprise the human gut microbiome fail to colonize in the recipient animals” and that “species that do colonize assemble into communities that differ substantially from those found in the human donors.
There are three seminal papers, prominently cited as providing evidence that intervening on the microbiome in mice can either cause or treat autism-related behaviors. According to Google Scholar, as of May 19th, 2025, these are the most highly cited studies in this field, with 3,739, 1,245, and 1,107 citations, respectively.
#1 paper. There is no good evidence that maternal immune activation has any causal involvement in autism etiology. A number of studies have found a small increased risk of autism in the offspring of mothers exposed to infections during pregnancy, sometimes only when subdividing cohorts by some covariates, such as the semester of pregnancy in which the infection occurred.
The final paper of this set (#3), by Sharon and colleagues, ostensibly offered the most direct test of the causal involvement of the microbiome in autism. The authors claimed that “colonization of mice with gut microbiota from human donors with ASD, but not from TD controls, is sufficient to promote behaviors in mice consistent with the core behavioral features of ASD.” A tiny sample set ( 8 ASD 5 normals). Pathetic ! ! !
Deeper problems with the statistical analyses were discovered post publication. In particular, the authors had treated each mouse as an independent observation when they should have been treated as experimental replicates of transplants from specific donors. To their credit, the authors made their primary data and code available for reanalysis. When the data were analyzed with the appropriate statistical model, the supposedly significant differences disappeared, with the exception of marble burying, where the remaining effect is statistically significant but very small. The main claims of the paper thus rest on flawed analyses and do not stand up to statistical scrutiny.
What a waste of taxpayer money. (my comment).
Two human studies of fecal transplants as therapy were large, 40 and 49 autistics, but the design was open label: not randomized controlled trials.
However one double-blind Randomized Clinical Trial of Fecal Microbiota Transplant (FMT) in children with Autism Spectrum Disorder (ASD) has been reported, randomizing 52 to FMT and 51 to placebo. The study failed to find any between-arm differences in improved clinical symptoms in children across three separate scales: SRS-2 T-score (primary), ABC scores, or Vineland-3 composite score. In an apparently post hoc analysis of outcomes, they did observe a significant between-arm difference in one domain of the Vineland-3 but did not account for multiple tests. Importantly, these authors observed outcome improvements in both arms of the study (i.e., participants in the placebo arm similarly improved to those in the FMT arm) and concluded that “it is essential to conduct efficacy evaluations in ASD research using randomised, double-blind, placebo-controlled trials.”
A typical study starts with the notion that there is some link between autism and the microbiome but is vague in terms of the specific hypothesis being tested, so that what we see is Hypothesizing After Results are Known (HARKing).
“We note that the gut microbiome has been similarly (and equally vaguely) “linked to” the risk of a large number of other conditions, including depression, anxiety, attention-deficit/hyperactivity disorder (ADHD), obsessive-compulsive disorder (OCD), schizophrenia, post-traumatic stress disorder (PTSD), Parkinson’s disease, Alzheimer’s disease, cardiovascular disease, obesity, eczema, asthma, arthritis, and many others. Given that the same methods have been used for most of these studies, it seems likely that they may suffer from the same methodological shortcomings.
Word to the wise: any time you see a paper saying that lab test/environmental value/life event has been ‘linked to’ some disease or other, ask to see the evidence and examine it carefully.
All is not lost. At least the problem has been recognized and work is being done to clean things up. This happened with functional MRI and neuropsychiatric disease 10 years ago. Here are two links. https://luysii.wordpress.com/2016/07/17/functional-mri-research-is-a-scientific-sewer/, https://luysii.wordpress.com/2020/06/07/functional-mri-research-is-a-scientific-sewer-take-2/
It’s hard to believe but most functional MRI work back then was being done without knowing whether the subject was awake or asleep. It took a lot of work but things have been cleaned up. Maybe a pony will be found in the roomfull of feces that is research on the microbiome and autism today.