A Family-Based Tool Splits Autism Risk Into What Children Inherit and What Parents Build Around Them

Every polygenic score carries a question it can’no’t answer alone. These scores, which tally thousands of tiny DNA variants into a single estimate of disease predisposition, are built by comparing large groups of unrelated people, and they can’t tell whether the risk they capture comes from the DNA itself or from the social and geographic environment that DNA tends to travel with. A new statistical framework called PGS-TRI, described in the journal Nature Genetics, settles that question by bringing the family into the calculation: it reads the genomes of a child and both parents at once, and uses the rules of inheritance to separate nature from nurture.

The tool was assembled by Ziqiao Wang, now an assistant professor of genome sciences at the University of Virginia School of Medicine, during her postdoctoral training at Johns Hopkins University, in a project overseen by biostatistician Nilanjan Chatterjee. Its first major test was autism: the team analyzed 18,383 case-parent trios, autistic children together with their mothers and fathers, drawn from the Simons Foundation’s SPARK consortium and its GEARS environmental study across diverse ancestral populations.

Reading the Family Instead of the Child

The logic rests on a simple fact of biology: a child inherits roughly half of each parent’s genetic variants, so the average of the parents’ scores is what the child’s score should be under chance transmission. When affected children consistently score above that parental midpoint, the excess is a direct effect, risk carried in the child’s own DNA. But scores can also act at a distance. “Traditionally, genetic studies focus on the ‘direct effects’ of genes passed from parent to child. However, a child’s health is also shaped by ‘indirect effects,'” said Wang, meaning how a parent’s own genetic makeup influences the environment they provide for their child. “PGS-TRI analyzes families, both parents and children, to figure out which health effects come directly from your genes versus which come from the environment your parents created.”

Those indirect effects are not speculation. Geneticists call the phenomenon genetic nurture, and a landmark 2018 study of Icelandic families showed that DNA variants parents never passed down still predicted their children’s educational attainment, presumably because those variants shaped the home the children grew up in. PGS-TRI detects such effects by comparing mothers’ and fathers’ scores: if a child’s risk tracks the gap between the two parents’ genetic predispositions, something about the parental environment, not the inherited sequence, is doing the work.

The Autism Score Survived the Family Test

The first result was a kind of vindication. Among the 12,813 European-ancestry families, PGS-TRI estimated that each standard-deviation rise in a child’s autism polygenic score came with about a 28 percent increase in autism risk, very close to the 33 percent reported by the large Danish iPSYCH study that trained the score on unrelated cases and controls. Because the family design is shielded from confounding by population structure and by the tendency of similar people to have children together, the agreement suggests those biases have not seriously inflated the score’s apparent power. Direct effects of similar strength also appeared in American and South Asian families.

Then the pattern broke. In families of African or East Asian ancestry, the autism score showed no significant direct effect at all, and across the full dataset its strength faded in a straight line as a family’s genetic distance from the European training population grew. That decay mirrors what a 2023 Nature study of two large biobanks documented across 84 traits: polygenic scores lose accuracy in proportion to how far an individual sits, genetically, from the people used to build them. The root cause is no mystery. As of 2021, about 86 percent of participants in genome-wide association studies were of European descent, and the PGS-TRI team argues that closing the gap requires recruiting far more diverse families into genetic research.

What Mothers Pass On Without Passing It Down

The indirect-effect analysis produced the study’s most provocative numbers. Mothers’ genetic scores for body mass index and for several neurocognitive traits, including ADHD, major depression, bipolar 2 disorder, and neuroticism, were associated with children’s autism risk beyond whatever DNA the children inherited. For some traits, these maternally mediated effects were larger than the scores’ direct effects. The body-mass finding fits a long epidemiological record: a 2024 meta-analysis of 42 studies covering 3.7 million mother-child pairs found that children of mothers with obesity before pregnancy had roughly 40 percent higher odds of an autism diagnosis. Notably, the child’s own BMI score showed no direct effect on autism risk, exactly as expected, since a toddler’s genetic predisposition to adult body weight should not shape brain development, which makes BMI a built-in sanity check for the method.

The researchers themselves apply the brakes here. The same traits whose scores show maternal effects, education among them, also influence which families volunteer for a study like SPARK, so a gap between mothers’ and fathers’ scores could arise from who enrolls rather than from biology in the womb. Sensitivity checks against UK Biobank data left most of the signals standing, some stronger, but the authors state that they can’no’t exclude participation bias. The method also measures only the difference between maternal and paternal indirect effects, not each parent’s contribution alone.

One Gene Emerged From 4,907 Candidates

As a discovery exercise, the team pointed PGS-TRI at 4,907 genetically predicted gene-expression traits. One gene cleared the statistical bar: CADM2, whose predicted expression showed a direct effect on autism risk in both cross-ancestry and European-only analyses, with nominal support in an independent brain-tissue dataset. The gene is a plausible actor. It encodes a synaptic adhesion protein enriched in the frontal cortex and striatum, and earlier genome-wide studies have tied it to impulsivity, cognition, body weight, and educational attainment. The university’s announcement describes CADM2 as a promising target for developing ways to prevent autism; the paper itself claims less, an association between genetically predicted expression and risk, the kind of lead that begins a research program rather than ends one. A second gene, LRRC37A4P, reached significance alongside it.

The framework also held up on a second condition. In 1,904 trios with orofacial clefts, PGS-TRI confirmed a strong direct effect of an established cleft-lip score across European and Asian families, detected an interaction with maternal smoking during pregnancy, and identified the gene TRAF3IP3 as a protective factor for cleft lip with or without cleft palate.

“We hope that PGS-TRI will empower researchers to look beyond just the DNA a child inherits and start understanding the broader family environment that shapes their health, leading to more personalized and effective ways to support children with complex conditions like autism,” Wang said. “This tool gives us opportunities to study the potential causal relationships of parents’ phenotype and their child’s disease risks and better understand how nature and nurture work together to influence a child’s health.”

For now, the tool’s sharpest message may be the map it drew of its own limits. The gene-expression scores were trained on adults, so their reach into fetal development is unproven, and the autism scores fail in exactly the families least represented in the data that built them. PGS-TRI can measure that failure family by family, which is more than the scores could say for themselves.

FAQ