Umwelten

Genetic heterogeneity in autism

🔬ResearchPeer-reviewed

4 min read

Summary

Autism is not one thing genetically. Hundreds of genes are associated with autism, no single gene explains more than a tiny fraction of cases, and the genetic architecture varies widely between individuals. This heterogeneity has been one of the central frustrations of autism genetics: if every autistic person has a different genetic profile, how can genetic findings inform care?

Recent work — particularly Litman, Sauerwald et al. 2025 — Genetic programs underlying autism phenotypic heterogeneity — is beginning to answer this by moving from gene-level to person-level analysis. Instead of asking “which genes are associated with autism?”, these studies ask “do autistic people cluster into groups with distinct genetic and phenotypic profiles?” The emerging answer is yes — and the groups align with clinically meaningful differences, including the presence or absence of intellectual disability and developmental delays.

What the evidence shows

Hundreds of genes, no master switch

Over 100 genes have been confidently identified as autism-associated, with hundreds more implicated at lower confidence levels. These include genes involved in synaptic function, chromatin remodelling, transcription regulation, and neuronal development. The pattern is one of extreme polygenicity: many genes of small effect, plus rare de novo mutations of larger effect in a subset of individuals.

Person-centred clustering reveals structure

Litman, Sauerwald et al. (2025) used a generative mixture model across 239 phenotypic features in 5,392 autistic individuals from the SPARK cohort and identified four latent classes:

  1. Social/behavioral — high social-communication and behavioural difficulties, no developmental delay, enriched for ADHD, anxiety, depression.
  2. Mixed ASD with developmental delay — strong developmental delays, language delay, intellectual disability, motor disorders. Lower psychiatric comorbidity. Earlier diagnosis. This class maps most closely to autistic people with intellectual disability.
  3. Moderate challenges — lower scores across all categories; still significantly above non-autistic siblings.
  4. Broadly affected — high scores across everything, highest intervention burden.

These are not just phenotypic clusters. Each class maps to distinct patterns of common genetic variation (polygenic scores), de novo mutations, and rare inherited variation. The genetic programs are genuinely different.

Developmental timing is class-specific

The genes disrupted in each class are expressed at different points in brain development. The “Mixed ASD with DD” class shows disruption in genes active during earlier developmental windows — suggesting that the developmental pathway to autism-with-ID may diverge from the pathway to autism-without-ID at a very early stage. This is not the same thing as saying they are different conditions, but it is consistent with that interpretation.

ICD-11 already moves in this direction

The ICD-11 distinguishes ASD with and without intellectual disability as separate diagnostic entities, each with different functional implications and care needs. The genetic evidence increasingly supports this distinction being more than administrative convenience.

Open questions

  • Where does sensory processing fit? The Litman/Sauerwald phenotypic features did not include direct sensory processing measures. Adding sensory data to the model might further differentiate the classes — or might cross-cut them. This is a testable and important prediction.
  • Are four classes the right number? The model finds four as the statistical optimum, but the underlying biology may be more continuous. The classes are a useful simplification, not a definitive taxonomy.
  • How stable are the classes across populations? The SPARK cohort is US-based. Replication in European, East Asian, and Global South populations is needed.
  • Do the classes predict response to intervention? If different genetic programs produce different phenotypic presentations, they might also respond differently to sensory-processing interventions. No study has tested this yet.

Relevance to this work

The practical relevance is indirect but important:

  1. The population of focus here has a genetic identity. The “Mixed ASD with DD” class — characterised by developmental delays, language delay, intellectual disability, and earlier diagnosis — has a distinct genetic architecture. This strengthens the case for population-specific knowledge and interventions rather than generic “autism” approaches.
  2. The “spectrum” metaphor may be misleading. If autistic people fall into genuinely distinct genetic groups, the metaphor of a single spectrum from “mild” to “severe” obscures more than it reveals. The emphasis on individual profiles over generic categories is validated by the genetics.
  3. Future work should integrate sensory data. If sensory processing measures were added to a SPARK-like cohort, the resulting class descriptions would be directly useful — telling practitioners not just that a person is in a particular class, but what their sensory processing profile is likely to look like.

Key sources

Graph View

Backlinks