Autism’s “four subtypes” – what the 2025 study actually found

A short, plain-language take on Litman et al. (2025)

You may have seen the headlines. A study in Nature Genetics has, we are told, identified four genetic subtypes of autism. Friends are emailing it to me. Colleagues are asking what I think. So let me set out, in plain language, what Litman and colleagues (2025) actually did – and what they did not.

What they did

Researchers took data from over 5,000 Autistic children whose parents had filled in a battery of standard questionnaires: how their child communicated, what repetitive behaviours they did, what emotional difficulties parents perceived, when they first walked, when they first used words. A computer model sorted those parent responses into four groups. Then the researchers looked at the genetics of children in each group and reported that the groups had different genetic patterns.

What they did not do

They did not identify four genetic subtypes of autism. Quite simply, the four groups were built from parent-reported behaviour first. Genetics was looked at second. The genetics did not create the groups. The behaviour questionnaires did.

And here is the line buried in the paper that you will not see in the press release: none of the four groups had a significantly different polygenic score for autism. The genetic differences they did find were mostly in scores for ADHD, IQ, educational attainment, and depression. The two groups with the strongest rare-variant signals are essentially the groups of children with co-occurring learning disability and developmental delay – something we have known is associated with rare high-impact genetic variants for years.

The questionnaires they used are not neutral

The Social Communication Questionnaire asks parents to rate how “abnormal” their Autistic child’s eye contact is. The Repetitive Behavior Scale rates how “problematic” stimming and routines are. The Child Behavior Checklist measures “internalising” and “externalising problems”. Every one of these tools asks an outsider – a parent – to interpret an Autistic child’s behaviour through a deficit lens. None asks the Autistic child anything. None considers monotropism, sensory difference, the Cullen Autistic pragmatic language hypothesis, the double empathy problem, or the three-dimensional Autistic space framework (more on Autistic theories and research). None considers that what looks like a “problem” from the outside might be a coping mechanism, a preference, a strength, or a perfectly reasonable response to an overwhelming environment.

When the categories themselves are loaded – “Broadly affected”, “Moderate challenges” – we should ask: affected by what? Challenges according to whom?

Why this matters

Autistic people and our communities have spent decades dismantling the crude tiers that came before: high versus low functioning, Asperger’s versus autism, severe versus mild. We did this work because those binaries flattened us into hierarchies and decided in advance who got to speak, who deserved support, and whose autism was even “real”. We do not need a new four-tier vocabulary, wrapped in the authority of computational genetics and a Nature journal’s reputation, to bring those binaries back through the side door. Particularly when our profiles are never fixed ([Farahar TDAS link]), and subtyping historically and in the recent past has been weaponised against Autistic people with and without traditional speech to deny support and services.

A more honest summary

Here is what Litman et al. actually showed. Parent-reported behaviour patterns can be statistically grouped. The groups correlate with genetic variation, mostly for co-occurring conditions and learning disability. That is a useful contribution to research methods. It is not the discovery of autism’s genetic subtypes.

Until autism research starts with Autistic experience – until it asks us – every “subtype” it produces will tell us more about the people doing the looking than about the people being looked at.

Reference: Litman, A., Sauerwald, N., Green Snyder, L., Foss-Feig, J., Park, C. Y., Hao, Y., Dinstein, I., Theesfeld, C. L., & Troyanskaya, O. G. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics, 57, 1611–1619.

A new study about autism – what it really found

📖  An Easy Read explanation

By Dr Chloe Farahar

📄  About this Easy Read •  This Easy Read is about a new science paper. •  The paper is about autism. •  It came out in July 2025. •  Some people said the paper found new types of autism in our genes. •  I do not think that is what the paper found. •  This Easy Read explains why.

🔬  What is the new paper? •  A team of scientists wrote the paper. •  The lead writer is called Litman. •  It is in a science journal called Nature Genetics. •  The scientists wanted to understand why autism looks different in different people.

👨‍👩‍👧  What did the scientists do? •  They looked at more than 5,000 Autistic children. •  They asked parents to fill in questionnaires. •  The questionnaires asked about the child’s behaviour. •  A computer sorted the children into 4 groups. •  The 4 groups were based on the parents’ answers. •  Then the scientists looked at the children’s genes.

📰  What did the news say? •  Some people reported this as though the scientists found 4 genetic types of autism. •  “Genetic” means caused by genes. •  That sounds like the scientists found the genes that make someone Autistic. •  But that is NOT what they found.

❌  What the study did NOT find •  The 4 groups were made from parent answers – not from genes. •  The genes were only looked at AFTER the groups were made. •  This matters a lot. •  The scientists also checked autism gene scores in each group. •  The autism gene scores were the SAME in all 4 groups. •  This means the groups were NOT different in their autism genes.

🧬  What about the gene differences they did find? •  The scientists did find some gene differences. •  But those differences were for OTHER things. •  The differences were for ADHD, IQ, school grades, and depression. •  These are NOT autism. •  Two of the groups had more rare gene changes. •  Those two groups were the children who also had a learning disability. •  We already knew learning disability is linked to rare gene changes. •  So this is not really new.

❓  Were the questions in the study fair? •  The questionnaires see autism as something “wrong” or “broken”. •  One asks parents to rate how “abnormal” their child’s eye contact is. •  Another asks if stimming is a “problem”. •  None of the questionnaires ask the Autistic child anything. •  They do not think about monotropism. •  Monotropism means deep focus on one thing at a time. •  They do not think about sensory needs. •  They see autism through a “deficit” lens. •  “Deficit” means looking only at what is missing or wrong.

🏷️  The names of the groups are loaded •  One group is called “Broadly affected”. •  We can ask: affected by what? •  Another group is called “Moderate challenges”. •  We can ask: challenges to whom? •  These names hide a judgement inside them. •  They compare Autistic children to non-Autistic children.

⚠️  Why does this matter? •  Autistic people fought for years to stop being put in boxes. •  Old boxes included “high functioning” and “low functioning”. •  Those boxes were unfair. •  They decided who could speak. •  They decided who got help. •  This new study could bring those old boxes back. •  That would be a step backwards.

✅  The main message •  The study did NOT find genetic types of autism. •  It found that parent answers can be sorted into groups. •  The groups link to genes for OTHER things – not autism itself. •  The headlines have got it wrong. •  We should be careful before we believe them. •  Until autism research asks Autistic people about our own lives, every “type” it finds will tell us more about the people doing the looking than about us.

Behaviour mistaken for biology

A critical reading of Litman et al. (2025) – Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs

In July 2025, Litman and colleagues published Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs in Nature Genetics. Public coverage has, predictably, distorted what the study actually did. Headlines and social media posts have suggested that researchers have at last identified genetic subtypes of autism. Quite simply, they have not. This essay sets out what Litman et al. did, what they did not, and why the distinction matters.

What the study did

The authors took 5,392 Autistic children from the SPARK cohort and applied a generative finite mixture model to 239 phenotypic features. Those features came from four parent-completed instruments: the Social Communication Questionnaire-Lifetime (Rutter et al., 2003), the Repetitive Behavior Scale-Revised (Lam & Aman, 2007), the Child Behavior Checklist 6–18 (Achenbach & Edelbrock, 1979), and a developmental milestones form. From this, they generated four latent classes (“Social/behavioral”, “Mixed ASD with DD” (developmental delay), “Moderate challenges”, and “Broadly affected”). They then examined genetic data and reported class-specific patterns of common, de novo, and inherited variation.

What the study did not do

The study did not identify genetic subtypes. The classes were constructed entirely from parent-reported behavioural questionnaires before any genetic analysis took place. Once the classes were fixed, the authors looked for genetic patterns within them. This is a fundamentally different claim from “we found genetic subtypes.” The genetic signal here is downstream of an observer-rated behavioural classification, not the basis for it.

Equally important, the autism polygenic score showed no statistically significant difference between any of the four classes. The genetic differences the authors did find were in polygenic scores for ADHD, IQ, educational attainment, and major depression – co-occurring conditions and contested cognitive constructs, not autism itself. The two classes with the strongest rare-variant signals (“Mixed ASD with DD” and “Broadly affected”) are, on inspection, the classes capturing co-occurring learning disability and developmental delay. That learning  disability has stronger associations with high-impact rare variants is already well-established (Satterstrom et al., 2020). The paper has, in effect, rediscovered this and reframed it as four “classes of autism”.

The framework is value-laden, not neutral

Each instrument used to construct the classes is built on the deficit-medical model. The SCQ asks parents to rate “abnormal” eye contact and “failure” to seek shared enjoyment. The RBS-R rates how “problematic” repetitive behaviours are. The CBCL frames distress, withdrawal, and emotional regulation as “internalising” and “externalising problems”. None of these instruments asks what an Autistic child’s experience is. None engages with monotropism (Murray, Lesser, & Lawson, 2005), sensory difference, the double empathy problem (Milton, 2012), pragmatic Autistic language (Cullen, 2018), or the variability of our Autistic experiences explained by the three-dimensional Autistic space framework (Farahar, 2018/2026) (more detail on these Autistic theories and research). The class names embed value judgements as well: “Broadly affected” – affected by what, and compared to whom? “Moderate challenges” – challenges defined by whose expectations of how a child should be?

Replication is not validation

The authors replicated the four classes in the Simons Simplex Collection. This is methodologically reassuring but conceptually limited. Both cohorts use the same observer-rated, deficit-framed instruments. Reproducing the same classification in another sample tells us the model is consistent; it does not tell us the underlying construct is valid. A consistently flawed measurement remains flawed.

Why this matters

Behaviour observed by parents has been reified into genetic destiny in public conversation. The risk is real. Autistic communities have spent decades dismantling the binaries – high versus low functioning, Asperger’s versus autism, severe versus mild – that reduced individuals to crude tiers (Kapp et al., 2013; den Houting, 2019). A new four-tier vocabulary, dressed in the authority of computational genetics, threatens to bring those binaries back through a side door. Binaries that are illogical given the knowledge that our profiles are never fixed (Farahar, 2018/2026).

The honest contribution of Litman et al. (2025) is narrower than the headlines: parent-reported behavioural patterns cluster in statistically describable ways, and those clusters correlate with genetic variation associated mainly with co-occurring conditions and learning disability. That is interesting. It is not the discovery of autism’s genetic subtypes.

References

Achenbach, T. M., & Edelbrock, C. S. (1979). The Child Behavior Profile: II. Boys aged 12–16 and girls aged 6–11 and 12–16. Journal of Consulting and Clinical Psychology, 47(2), 223–233.

den Houting, J. (2019). Neurodiversity: An insider’s perspective. Autism, 23(2), 271–273.

Kapp, S. K., Gillespie-Lynch, K., Sherman, L. E., & Hutman, T. (2013). Deficit, difference, or both? Autism and neurodiversity. Developmental Psychology, 49(1), 59–71.

Lam, K. S. L., & Aman, M. G. (2007). The Repetitive Behavior Scale-Revised: independent validation in individuals with autism spectrum disorders. Journal of Autism and Developmental Disorders, 37(5), 855–866.

Litman, A., Sauerwald, N., Green Snyder, L., Foss-Feig, J., Park, C. Y., Hao, Y., Dinstein, I., Theesfeld, C. L., & Troyanskaya, O. G. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics, 57, 1611–1619.

Milton, D. E. M. (2012). On the ontological status of autism: the ‘double empathy problem’. Disability & Society, 27(6), 883–887.

Murray, D., Lesser, M., & Lawson, W. (2005). Attention, monotropism and the diagnostic criteria for autism. Autism, 9(2), 139–156.

Rutter, M., Bailey, A., & Lord, C. (2003). The Social Communication Questionnaire: Manual. Western Psychological Services.

Satterstrom, F. K., et al. (2020). Large-scale exome sequencing study implicates both developmental and functional changes in the neurobiology of autism. Cell, 180(3), 568–584.e23.