"Autism is not one condition. It is four." The headline is everywhere, and this time the science behind it is solid. In July 2025, a Princeton and Flatiron Institute team published in Nature Genetics the largest attempt yet to sort autism into subgroups using both behaviour and genes. The result is more interesting, and more careful, than the slogan.
The Study
Aviya Litman and colleagues, with Olga Troyanskaya, Natalie Sauerwald and Chandra Theesfeld, analysed 5,392 autistic children aged 4 to 18 from SPARK, the largest autism cohort in the United States, alongside their siblings. Instead of starting from a few scores, they took more than 230 traits per child: social interaction, repetitive behaviours, language, developmental milestones, anxiety, attention, mood, sleep. A "person-centred" model then looked for children who resemble each other across all of it, and only afterwards compared the groups genetically.
The Four Groups
Social and behavioural challenges, about 37 percent. Core autistic traits, but developmental milestones reached on time. Many co-occurring conditions: ADHD, anxiety, depression, OCD. The genes involved are mostly active after birth.
Mixed autism with developmental delay, about 19 percent. Walking and talking came later than in other children, but few psychiatric symptoms. More rare inherited variants, in genes mostly active before birth.
Moderate challenges, about 34 percent. The same kind of difficulties as the first group, but fewer of them and milder, with no delays and no co-occurring diagnosis.
Broadly affected, about 10 percent. Developmental delays, marked social and communication difficulties, repetitive behaviours, plus anxiety, depression and mood dysregulation. This group carries the highest share of damaging de novo mutations, changes present in the child but in neither parent.
The striking part is that the biological pathways affected barely overlap between groups. Different genetic routes, different timing of gene activity, different profiles at the end.
Why It Matters
Today, genetic testing explains autism in roughly one child in five. Studies that pool everyone together dilute signals that only exist in a subgroup. If the four groups hold up, research on causes, on prognosis and on which support works for whom can finally be done group by group rather than on an "average autistic child" who does not exist.
It also matches what families and clinicians see: an autistic child with early delays and no anxiety, and an autistic teenager with ADHD and depression who spoke on time, share a diagnosis but not a story.
What It Is Not
The authors say it themselves: there are at least four classes, this is a starting point, not a new classification. Three limits worth keeping in mind. The children come from one American cohort of families who volunteered for a genetic study. The traits were reported by parents. And the groups are statistical: a child sits closer to one profile, they do not carry a label. No clinician will diagnose "type 2 autism" from this paper, and no test on this site can place you in one of the four groups.
What It Changes For You
If you are an autistic adult, or wonder whether you are, the study is a reminder that autism with anxiety and ADHD, autism with early language delay, and autism with mild traits are different experiences that deserve different answers. A screening test tells you whether autistic traits are present and how strong they are; that is where the conversation starts. Our free AQ-50 autism test gives a validated score in 15 minutes, and the Neurotype test looks at autism, ADHD, sensory profile and mood together, which is exactly the combination this study says matters. For the molecular side of the story, see our article on the molecular map of autism.
Sources: Litman A, et al. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics; Princeton University news (July 2025); Simons Foundation (July 2025).