A UK-led study tracking thousands of children from age four to sixteen has found distinct patterns in how genetic influences on thinking, learning and problem solving shift over development. The researchers report that common genetic variants linked to education and test performance become more predictive with age, particularly for high achievers, while rare, gene-disrupting mutations exert their strongest negative effects in early childhood and weaken by adolescence.
Cognitive ability refers to a person’s capacity to reason, solve problems, think abstractly and learn, and the study looked at two broad classes of genetic variation that relate to these traits alongside environmental factors. Common genetic variants are frequent DNA differences found in more than 1% of the population, each with a tiny association to traits, while rare damaging variants are uncommon mutations that can disrupt how a gene builds a protein.
The research joins a broader body of work showing that the genetic contribution to intelligence changes across the lifespan. Prior findings suggested genetics explain about 20% of intelligence differences in infancy, rising to 60% or more in adulthood. Building on this, the authors highlight that a polygenic score aggregating many common variants can explain roughly 7 to 10% of variation in cognitive test performance. The current study places these dynamics within a developmental framework, tracking how associations evolve from childhood into adolescence.
The investigators also touch on neurodevelopmental conditions, noting that rare deletions on chromosome 16 have been linked to autism and associated reading, speech and motor-skill challenges. They cite a 2024 study from some of the same researchers indicating that common variants account for about 10% of the variation in risk for rare neurodevelopmental conditions.
The authors, Daniel S. Malawsky and Hilary C. Martin of the Wellcome Sanger Institute in the UK, led a team that tracked common and rare variants side by side to observe their changing relationships with cognitive performance over time.
To map development, the team used data from three British cohorts: the Avon Longitudinal Study of Parents and Children, the UK Millennium Cohort Study and the UK Biobank. Their main longitudinal analysis focused on 6,495 unrelated children of European ancestry from Avon, whose IQ was tested at ages four, eight and sixteen. A subset of 701 had a score at age four, 5,134 at age eight and 3,774 at age sixteen, with 3,895 also having linked national school exam scores in English, maths and science at ages eleven and fourteen.
Because some children missed one or more assessments, researchers employed statistical imputation to estimate missing IQ scores for a broader group of eight thousand two hundred and seven children who had at least one IQ measurement. They found that at age four, about nine in ten scores were imputed estimates, and a validation check yielded a median correlation of 0.61 between imputed and actual scores.
In addition to Avon, the study drew on the Millennium Cohort Study for a composite cognitive score spanning ages three to seven and on UK Biobank data for a large sample of adults to compare patterns into adulthood.
To measure genetic variation, the researchers computed three polygenic scores per participant: overall educational attainment, its cognitive component, and a non-cognitive component capturing traits like perseverance and social factors linked to schooling. They also generated new exome sequencing data — examining the protein-coding portion of the genome — for 16,103 children and 10,140 parents across the two birth cohorts to assess rare variant burden. A rare variant burden was calculated by summing uncommon mutations present in less than 0.1% of the population and weighting each by how strongly natural selection has weeded out mutations in that gene.
The analysis included both peer-to-peer comparisons and family trio models (where both parents’ genetics were taken into account) to separate direct child genetics from the broader family environment.
Across the Avon cohort, the share of IQ variation explained by common DNA variants rose from 46% at age four to 56% at age sixteen, though that increase did not reach statistical significance. Focusing on the trio analyses, the educational and cognitive polygenic scores showed a growing relationship with IQ as children aged, with the scores explaining about 11% of the variation at age four, rising to 14% by sixteen. Scores tied to non-cognitive traits did not maintain a significant link once parental genetics were considered.
By contrast, the pattern for rare damaging variants moved in the opposite direction. A higher burden of rare gene-disrupting or protein-altering mutations correlated with lower IQ and poorer school performance, with disabling mutations showing a stronger effect than those that alter a protein’s structure but do not completely disrupt it. Yet this negative association weakened with age, and the pattern held in the family-trio analyses. Inherited rare damaging variants explained more of the childhood IQ variation than new mutations arising in the germ cells, and a larger share of children carried inherited burdens (12%) compared with new mutations (1.8%). The authors caution that their sample may overstate the contribution from new mutations and that the findings contrast with clinical cohorts where de novo mutations play a larger role.
The researchers also used quantile regression to explore whether these genetic associations differed across the IQ spectrum. They found that the strengthening link between educational and cognitive polygenic scores and age concentrated in the upper half of the IQ distribution, while the bottom five per cent showed no significant age-related change. In UK Biobank adults, the direct association of the educational polygenic score with IQ was about 1.62 times stronger at the 95th percentile than at the 5th percentile.
For rare disabling variants, negative associations were strongest among the lowest-scoring children in early life and diminished by adolescence. By age sixteen, differences in effect sizes across IQ percentiles were no longer statistically significant. Among children in the top five per cent for rare variant burden, roughly seven per cent had an IQ below 70 at age eight, compared with about three per cent of the rest.
When set against environmental and birth factors — parental education, maternal health complications during pregnancy and prematurity — the genetic measures added substantial explanatory power. Together, the polygenic scores, parental education and pregnancy factors explained about 22.8% of IQ variation at age four and 25.6% at age sixteen, in a model excluding gestational age due to limited data. The direct link between a child’s educational polygenic score and IQ was similar in magnitude to that of a parent’s actual years of education.
Rare damaging variants accounted for around 1.40% of IQ variation at age four, roughly four to eight times more than prematurity or maternal illness during pregnancy. Both birth factors, like the genetic measures, tended to lose their explanatory power with age.
Experts emphasise that these figures describe population-level trends and cannot predict an individual child’s trajectory. The authors also note the study’s limitations: it focused on British participants of European ancestry to avoid population confounding and because existing polygenic scores were developed in European samples, so the findings may not generalise to other groups. The rare variant scores examined only a specific portion of the genome and the observational design means the work shows associations, not proven biological mechanisms.
The study’s co-author, Liana Machado of the University of Otago, commented on the imputation caveat: “Confidence in the results is somewhat limited by the analyses involving substantial imputation related to missing data, which in some cases altered the results, leaving me with some uncertainty about the interpretation.” She added that the authors reported pre-imputation results and that the rare-variant analyses were limited by smaller samples, but that the overall methods strengthened confidence in the findings.
Looking ahead, the researchers suggest revisiting the questions with much larger whole-genome sequencing datasets and alongside clinical cohorts of children with neurodevelopmental conditions. They propose that early cognitive delays in some carriers of rare variants could reflect health issues that improve with medical care or brain development mechanisms that compensate for disrupted genes, and that future work should examine whether access to health care or other environmental factors drives the observed changes with age.
The study, Common and rare genetic variant associations with cognitive performance across development in British birth cohorts, was led by Daniel S. Malawsky and Hilary C. Martin, and published with colleagues including Mahmoud Koko, Petr Danacek, Wei Huang, Olivia Wootton, Qin Qin Huang, Emma E. Wade, Sarah J. Lindsay, Rosalind Arden, and Matthew E. Hurles.
