Prenatal Genomics Can Detect Autism-Linked Variants but Cannot Predict Autism

A genetic variant linked to autism can now be found in fetal DNA with remarkable technical precision, yet finding it is not the same as predicting autism. A review published in Frontiers in Psychiatry examines why the chain from prenatal genetic testing to a prediction of autism in an individual fetus breaks down, and it concludes that current prenatal sequencing may establish a molecular diagnosis of a genetic condition associated with an increased likelihood of autism — but that this is not prenatal diagnosis of autism itself.

The review, authored by Ainaz Shateri of Johns Hopkins University, organizes the evidence into four sequential steps: analytic validity, meaning whether a variant can be detected in fetal DNA; variant interpretation, meaning whether it can be classified under ACMG/AMP and ACMG–ClinGen standards; clinical validity, meaning how far a detected variant shifts the probability of autism; and clinical utility, meaning whether knowing the result improves outcomes. This framework makes explicit the classification step within the conventional three-stage sequence of analytic validity, clinical validity and clinical utility, and it locates precisely where translation from detection to prediction fails.

The technical side of the story is genuinely advanced. Prenatal genetics has moved from karyotyping to chromosomal microarray analysis, exome and genome sequencing, and screening based on cell-free DNA in maternal plasma. Against this backdrop, recurrent copy-number variants such as 16p11.2 and 15q11–q13 rearrangements, and pathogenic variants in genes such as CHD8, SCN2A, SHANK3 and PTEN, are technically detectable by currently available prenatal diagnostic genomic platforms when such testing is clinically indicated. The analytic step, in other words, is largely solved for these known variants.

The problems begin at interpretation and intensify at prediction. The interpretation of detected variants is frequently incomplete, and the clinical validity of these findings for autism specifically is limited by incomplete penetrance, variable expressivity and pleiotropy — meaning a pathogenic variant may never produce the associated condition, may produce it in wildly different forms, or may affect multiple organ systems rather than neurodevelopment alone. The clinical utility of these variants for neurodevelopmental prediction is unestablished. A separate distinction runs through the entire analysis: the difference between a molecular diagnosis of a genetic condition and a diagnosis of autism itself, which remains a behaviorally defined diagnosis.

Autism spectrum disorder is described in the review as a behaviorally defined, biologically heterogeneous neurodevelopmental condition with high heritability, estimated at 0.83 (95% CI 0.79–0.87) in Swedish population data. Large-scale exome sequencing, copy-number variant analysis and genome-wide association studies have implicated a growing number of genes and genomic loci in autism liability, and functional genomic work has repeatedly converged on mid-gestational cortical development. That convergence gives the genetics biological plausibility at the population level, but it does not convert population-level liability into individual-level prediction.

The polygenic side of the picture illustrates this gap most clearly. Common polygenic variation contributes substantially to population-level liability for autism, yet individual-level discrimination from polygenic scores is poor and ancestry portability is limited — a score built in one ancestry group does not transfer reliably to others. A measure that explains meaningful variance across a population can still leave any single fetus’s outcome largely uncertain, which is precisely the distinction the review draws between population liability and individual prediction.

There is also a structural constraint on who gets tested at all. Prenatal diagnostic genomic testing is largely indication-driven rather than a population screen for neurodevelopmental risk. Even a perfect analytic platform therefore reaches only a selected subset of pregnancies, and the review frames this as a further reason the current system cannot function as autism prediction.

The practical consequence is a careful reframing of what a positive prenatal genomic result means. When testing is clinically indicated and a pathogenic variant is found, the finding supports a molecular diagnosis of a genetic condition associated with an increased likelihood of autism — a statement about the variant and its known associations, not about the child’s future neurodevelopment. The review argues that counseling standards should resist deterministic interpretation of such results, given incomplete penetrance and variable expressivity.

Progress, the review argues, will require ancestrally diverse cohorts, population-based penetrance estimates, longitudinal prenatal-to-childhood follow-up, and probabilistic multimodal models that combine genetic and non-genetic information. Until those pieces exist, the gap the review identifies remains: susceptibility detectable before birth, prediction of autism in an individual fetus not validated, and clinical utility unestablished.

Subject of Research: The gap between detecting autism-associated genetic variants prenatally and validated individual-level prediction of autism

Article Title: Prenatal genomics and autism: detectable susceptibility without validated individual-level prediction

Article References: Shateri, A. (2026). Prenatal genomics and autism: detectable susceptibility without validated individual-level prediction. Frontiers in Psychiatry, 17, Article 1973963. https://doi.org/10.3389/fpsyt.2026.1973963

Image Credits: AI Generated

DOI: 10.3389/fpsyt.2026.1973963

Keywords: autism spectrum disorder, prenatal genomics, copy-number variants, clinical validity, polygenic risk score, genetic counseling, variants of uncertain significance, incomplete penetrance, 16p11.2, CHD8, heritability, cell-free DNA