Genetic influence on behaviour and disease measured in 500,000 sibling pairs

One of the largest studies of its kind has delivered some of the most precise within-family estimates yet of how strongly genes shape human behaviour and disease risk, and it did so by exploiting a quirk of biology that most genetic studies try to ignore: the random lottery of which chromosomes siblings inherit from their parents. Analysing roughly 500,000 full sibling pairs drawn from more than one million research-consented participants of the consumer genetics company 23andMe, researchers report within-family heritability estimates for 14 traits, including smoking initiation, alcohol consumption, number of children, personality, type 2 diabetes and asthma. The findings, published in Nature Human Behaviour, suggest that genetic variation genuinely contributes to variation in these traits, but that the genetic influence is on average about 40 per cent smaller than twin studies have implied.

The technique at the heart of the study, known as within-family segregation inference or FSIR, rests on a simple observation. Full siblings do not inherit identical stretches of their parents’ DNA; the random shuffling and allocation of parental chromosomes during the formation of eggs and sperm, known as Mendelian segregation, means the amount of genome shared identical by descent varies from pair to pair. If genetic variants truly affect a trait, siblings who happen to share more of their genome should, on average, resemble each other more on that trait than siblings who share less. Crucially, this random sharing is independent of the family environment, population ancestry and most other confounders that plague conventional heritability studies. The researchers sharpened the method further by stratifying the genome into quartiles of recombination rate, fitting the identity-by-descent coefficients for each stratum jointly, and summing the contributions to obtain the final estimate.

Conventional estimates, whether from comparing unrelated individuals or from comparing twins raised together, can be inflated by indirect genetic effects, in which one person’s genes influence another person’s outcome through the environment, or by shared environmental factors such as household income, diet and neighbourhood. They can also be distorted by population stratification, the confounding that arises when genetic ancestry correlates with environmental exposures. By contrasting the genome-wide realized relatedness between siblings with their phenotypic similarity, the within-family design sidesteps most of these problems. What it measures is the direct effect of genetic variation on the trait itself. The study cohort was about 40 per cent male, with an average age of approximately 50 years, and sibling relationships were inferred from patterns of shared genomic segments rather than self-report.

To validate the approach, the team first turned to two well-studied ‘model traits’: height and body mass index. Across the four ancestry and ethnicity groups in the study, European, East Asian, African and Latino, height yielded a meta-analysed within-family heritability of 0.70 (standard error 0.03) and BMI 0.49 (standard error 0.03). Individual group estimates ranged from 0.68 in Europeans to 1.13 in East Asians for height, and from 0.47 to 0.68 for BMI, with no significant heterogeneity across groups. These figures align closely with previous sibling-based estimates, which reassured the researchers that self-reported measurements from the cohort were adequate for the design. Correcting for assortative mating, the well-known tendency of people to choose partners of similar height, the height estimate translates to a heritability of roughly 0.84 in the current population.

The behavioural results were among the study’s most striking. Smoking initiation showed a significant within-family heritability of 0.34 (standard error 0.05), alcohol consumption 0.18 (standard error 0.04), number of children 0.27 (standard error 0.10 in a meta-analysis across sexes) and the personality dimension ‘talk versus listen’ 0.48 (standard error 0.13). In other words, a substantial fraction of the variation in how people behave, from whether they pick up their first cigarette to how talkative they are, can be attributed to the direct effects of the genetic variants they inherited, independent of the family they grew up in. Statistical significance was declared at a Bonferroni-corrected threshold for the 42 trait–ancestry combinations analysed, corresponding to an estimate more than 3.2 times its standard error.

The study also quantified genetic influence on disease phenotypes whose risk factors include behaviour. Type 2 diabetes, linked to high-fat, high-carbohydrate diets, showed a heritability of 0.43 (standard error 0.06), while asthma, influenced by smoking, pet ownership and diet, showed 0.34 (standard error 0.06). Self-reported burden of heart-metabolic disease yielded 0.26 (standard error 0.05), general health score 0.15 (standard error 0.04) and physical activity 0.10 (standard error 0.04). For binary traits such as diabetes and asthma, the estimates were converted from an observed zero-to-one scale to an underlying liability scale using a new method-of-moments approach developed by the team and released as a publicly available R script, because the standard linear transformations used in previous studies can produce biased results when heritability is large and disease prevalence is small.

Perhaps the most consequential finding concerns what the design revealed about the environment. The method also estimates a residual sibling correlation, denoted c-squared, that reflects effects shared by siblings but independent of how much DNA they happen to share. A positive c-squared can indicate shared environmental effects, although it can also arise from assortative mating, and unmodelled non-additive genetic effects can push it negative. Among traits with significant heritability, the researchers found significant positive c-squared estimates in the European subcohort for smoking initiation (0.17), height (0.12), heart or metabolic disease burden (0.09) and general health score (0.08). For birth weight and physical activity, significant shared components emerged even though heritability itself did not reach significance. For type 2 diabetes, the meta-analysed c-squared was 0.25 (standard error 0.05). This suggests that shared environmental effects between siblings, or unaccounted population stratification, may have inflated earlier between-family estimates of diabetes heritability, which reached 0.62 in a non-twin pedigree study and 0.72 in twin studies, both significantly larger than the within-family figure of 0.43.

The comparison with twin studies was systematic. Across 12 phenotypes where precise estimates were available, defined as those with standard errors below 0.1, twin-based heritability estimates were on average 1.7-fold higher than the within-family estimates, while non-twin between-family studies produced figures largely consistent with the new results. The within-family estimates also exceeded what genome-wide association studies have accounted for so far. For BMI, for example, whole-genome-sequence-based SNP heritability is only about 0.34, well below the within-family estimate of roughly 0.5, although a between-family whole-genome-sequence analysis accounting for non-additive effects reported 0.39. The authors note that this gap could partly reflect non-additive genetic effects, such as gene-by-gene or gene-by-shared-environment interactions, which contribute to within-family estimates but are largely invisible to SNP-based methods. Separating additive from non-additive variation within families would require sibling samples in the millions.

One result stands out for its social significance: the heritability estimates were highly concordant across ancestry and ethnicity groups. For all traits except type 2 diabetes, heterogeneity tests found no significant differences between groups, and even for diabetes the heterogeneity was modest. This was not a foregone conclusion, since populations of African ancestry carry more genetic variation overall, and environments differ substantially between groups. The authors caution that the lack of detected heterogeneity may partly reflect limited statistical power in the undersampled non-European groups, where only two of seven significant traits reached significance in the Latino subcohort, but the broad consistency suggests that the genetic architecture of these traits, as measured within families, is broadly similar across human populations.

The study has clear limitations that the authors acknowledge. All phenotypes were self-reported, which introduces measurement error that biases heritability estimates downward, possibly explaining an unexpectedly low estimate for eye colour of 0.45 (standard error 0.13) when twin studies report nearly 100 per cent. Sample sizes in non-European ancestry groups remain limited, and the researchers could not analyse educational attainment or cognitive traits because of ethics restrictions, leaving unresolved a debate in which within-family estimates for educational attainment range from about 0.04 to 0.40. Still, with standard errors below 0.1 for nine of the traits, this analysis represents the most precise within-family picture to date of how much genes matter for behaviour and common disease, and it points to a substantial reservoir of genetic influence on complex traits that current genome-wide association studies have yet to explain.

Subject of Research: Within-family heritability of behavioural and disease traits estimated from about 500,000 sibling pairs of diverse ancestries

Article Title: Within-family heritability estimates for behavioural and disease phenotypes from 500,000 sibling pairs of diverse ancestries

Article References: Yengo, L., Liang, Y., 23andMe Research Team, Wang, X., Granka, J. M., Evans, D. M., Sidorenko, J., & Visscher, P. M. (2026). Within-family heritability estimates for behavioural and disease phenotypes from 500,000 sibling pairs of diverse ancestries. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02599-x

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02599-x

Keywords: heritability, sibling pairs, within-family estimation, behavioural genetics, type 2 diabetes, asthma, smoking initiation, alcohol consumption, BMI, height, population stratification, shared environment