Within-sibship GWAS improve estimates of direct genetic effects
Laurence J HoweiD, Michel Guillaume NivardiD, Tim MorrisiD, Ailin Falkmo Hansen, Humaira RasheediD, Yoonsu ChoiD, Geetha ChittooriD, Penelope A. LindiD, Teemu PalviaineniD, Matthijs D. van der ZeeiD, Rosa CheesmaniD, Massimo ManginoiD, Yunzhang WangiD, Shuai LiiD, Lucija KlarićiD, Scott M. RatliffiD, Lawrence F. BielakiD, Marianne NygaardiD, Chandra A. ReynoldsiD, Jared V. BalbonaiD, Christopher R. BaueriD, Dorret Irene BoomsmaiD, Aris BarasiD, Archie I CampbelliD, Harry J. CampbelliD, , Neil M DaviesiD
Abstract
Estimates from genome-wide association studies (GWAS) represent a combination of the effect of inherited genetic variation (direct effects), demography (population stratification, assortative mating) and genetic nurture from relatives (indirect genetic effects). GWAS using family-based designs can control for demography and indirect genetic effects, but large-scale family datasets have been lacking. We combined data on 159,701 siblings from 17 cohorts to generate population (between-family) and within-sibship (within-family) estimates of genome-wide genetic associations for 25 phenotypes. We demonstrate that existing GWAS associations for height, educational attainment, smoking, depressive symptoms, age at first birth and cognitive ability overestimate direct effects. We show that estimates of SNP-heritability, genetic correlations and Mendelian randomization involving these phenotypes substantially differ when calculated using within-sibship estimates. For example, genetic correlations between educational attainment and height largely disappear. In contrast, analyses of most clinical phenotypes (e.g. LDL-cholesterol) were generally consistent between population and within-sibship models. We also report compelling evidence of polygenic adaptation on taller human height using within-sibship data. Large-scale family datasets provide new opportunities to quantify direct effects of genetic variation on human traits and diseases.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 68
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (7 Mar 2021) · CC BY 4.0
In 159,701 siblings from 17 cohorts, population GWAS overstate direct effects on height, education, smoking, cognition, depressive symptoms and age at first birth; clinical traits mostly agree across models.
Associations from GWAS of unrelated people blend direct genetic effects with population stratification, assortative mating and indirect effects of relatives, which can bias heritability, genetic correlation, Mendelian randomization and selection analyses; sibling designs remove demography and parental indirect effects but had lacked large samples. Pooling sibships from 17 cohorts gives what the authors call the largest within-sibship GWAS to date, across 25 phenotypes.
Key results
ShownPopulation GWAS overstate direct effects for six phenotypes, most for age at first birth and educational attainment; C-reactive protein (CRP) shifted the other way, the other 18 (including body mass index and systolic blood pressure) changed little, and China Kadoorie Biobank showed no height shrinkage.
Score shrinkage (UK Biobank non-sibling variants, P < 1×10^−5): age at first birth 49% (95% CI 25% to 74%), educational attainment 46% (40% to 52%), depressive symptoms 39% (5% to 73%), ever smoking 19% (9% to 30%), cognitive ability 18% (2% to 35%), height 10% (8% to 12%), CRP −9% (−15% to −2%). China Kadoorie Biobank height −3% (−13% to 7%), heterogeneity vs European P = 0.005. · Genetic association estimates differ when…; Overview; Summary-based weighted scores; Estimating shrinkage from population to…
ShownWithin-sibship SNP heritability from LD score regression (LDSC) dropped for educational attainment, cognition and height, in line with the expected effect of the estimate shrinkage; the other 22 phenotypes gave similar estimates under both models.
Population vs within-sibship h²: educational attainment 0.14 vs 0.04 (71% attenuation, difference P = 1.5×10^−20), cognition 0.24 vs 0.13 (46%, P = 0.012), height 0.41 vs 0.34 (17%, P = 1.3×10^−3). With effective sample sizes, LDSC recovered a simulated h² of 0.30 (median 0.29 population, 0.32 within-sibship) in UK Biobank. · Within-sibship SNP heritability estimates; Within-sibship SNP heritability estimates
ShownEducational attainment's genetic correlations with height, body mass index (BMI) and CRP, clear in population data, were near zero within sibships, which the authors read as driven by demography and indirect genetic effects.
Population vs within-sibship rg: height 0.16 (95% CI 0.10 to 0.22) vs −0.02 (−0.15 to 0.10), difference P = 4.0×10^−3; BMI −0.32 vs 0.01 (−0.17 to 0.19), P = 6.9×10^−5; CRP −0.47 (−0.69 to −0.25) vs 0.04 (−0.34 to 0.42), P = 0.012. Computed for 22 phenotypes with h² > 0 in both models; jackknife over 100 blocks for differences. · Within-sibship genetic correlations with…; Within-sibship genetic correlations with…
ShownWithin-sibship Mendelian randomization (MR) of height and BMI on 23 outcomes mostly agreed with population MR, but the effects of taller height and lower BMI on educational attainment, and of BMI on age at first birth, were sharply attenuated.
Overall within-sibship vs population: height 0% (95% CI −12% to 12%), BMI −11% (−20% to −1%). Education per SD taller height: 0.05 SD (0.03 to 0.06) population vs 0.01 SD (−0.01 to 0.03) within-sibship, difference P = 2.9×10^−3; per SD higher BMI: 0.18 SD lower (0.14 to 0.21) vs 0.04 SD lower (0.00 to 0.08), P = 2.1×10^−7. P = 6.0×10^−4 for the gap in BMI's estimated effect on age at first birth. · Within-sibship Mendelian randomization: effects…; Within-sibship Mendelian randomization: effects…
ShownPolygenic adaptation toward taller height persists in within-sibship data, which resist population stratification, with weaker within-sibship hints of selection on more children and lower HDL-cholesterol.
Spearman r of tSDS (Singleton Density Scores aligned to the trait-raising allele) with GWAS signal, European meta-analysis: population 0.020 (95% CI 0.011 to 0.029), within-sibship 0.011 (0.003 to 0.020). Mean tSDS 0.23 (SE 0.06, P < 0.001) at 416 within-sibship height loci; 7-study pooled within-sibship r = 0.004 (0.002 to 0.007). Within-sibship P = 0.006 (children), 0.024 (HDL). · Polygenic adaptation; Polygenic adaptation
Methods, data and reproducibility
Each cohort fitted, on the same siblings, a principal-component-adjusted population GWAS and a within-sibship GWAS (genotype centred on the sibship mean, mean genotype as covariate). European-ancestry results were meta-analysed in METAL and compared through shrinkage of UK Biobank-weighted scores, LD score regression, Mendelian randomization and Singleton Density Score (SDS) selection tests.
- Genotyped individuals with at least one genotyped sibling, from 17 cohorts including UK Biobank, HUNT, Generation Scotland, QIMR, FinnTwin, Netherlands Twin Registry and DiscoverEHR · human · n = 159,701 individuals in 68,691 sibships
- European-ancestry meta-analysis of 25 phenotypes (largest: height, BMI, ever smoking, systolic blood pressure, educational attainment) · human · n = up to 137,776 (height)
- China Kadoorie Biobank siblings, analysed separately from the European meta-analysis · human · n = 13,856 individuals
- UK Biobank European-ancestry participants excluding the sibling sample; independent BOLT-LMM reference GWAS for variant selection and score weights · human · n = 422,730
- Precomputed Singleton Density Scores for UK10K data · human (UK10K)
- 1000 Genomes Phase 3 European LD scores used as the LDSC reference · human (1000 Genomes)
- HUNT individuals outside the sibling sample, used with simulated phenotypes to test LD score mismatch · human · n = 30,694
- Methods & tools
- R, METAL, BOLT-LMM, PLINK v2.0, LD score regression (LDSC), FastGWA, Singleton Density Score (SDS), seemingly unrelated regression, SiblingGWAS pipeline
- Code
- https://github.com/LaurenceHowe/SiblingGWAShttp://github.com/LaurenceHowe/SiblingGWASPost/blob/master/LDSCsimulations/CrypticRelatednessSims.R
- Data & accessions
- https://web.stanford.edu/group/pritchardlab/
- Software & environment
- PLINK v2.0 is the only software with a version given; GWAS ran through shared automated scripts (population model fitted by linear regression in R) under a preregistered analysis plan. No containers or requirement files are described.
- Reproducibility gaps
- GWAS summary statistics had not been released at posting (promised before publication); cohort genotype QC, imputation and phenotype definitions are in Supplementary Materials not included here; access to individual-level cohort data is not described.
Limitations (6)
- Within-sibship models do not remove indirect genetic effects of siblings acting through the shared environment, which could have affected the estimates; all phenotypes analysed were measured in adults. · stated by the authors
- Family-based GWAS are limited by available family data and are statistically inefficient (within-family homozygosity), so unrelated samples remain preferable for discovery; the authors call for larger non-European within-family studies, as non-European data here came only from China Kadoorie Biobank. · stated by the authors
- LDSC ratios, a confounding measure, were non-zero even in within-sibship data (median 21%), partly attributed to mismatch between 1000 Genomes LD scores and Scandinavian cohorts such as HUNT; simulations suggested heritability stayed unbiased, but the authors say more extensive simulation is needed. · stated by the authors
- Height adaptation correlations from individual studies were generally smaller than from the meta-analysis, and UK Biobank sibling data alone was inconclusive (r = 0.002, 95% CI −0.005 to 0.010). · stated by the authors
- Findings are reported at nominal P < 0.05 across 25 phenotypes with no stated multiple-testing correction, so weaker results (e.g. cognition h² difference P = 0.012, HDL adaptation P = 0.024) warrant caution. · noted in this summary
- Depressive-symptom shrinkage uses weights from a UK Biobank binary depression GWAS because no continuous measure was available there, and its interval is wide (5% to 73%). · noted in this summary
Who should read:Read this if you use GWAS summary statistics for heritability, genetic correlation, Mendelian randomization or selection analyses of behavioural or anthropometric traits and need to know how much of the signal reflects direct genetic effects.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Mendelian imputation of parental genotypes for genome-wide estimation of direct and indirect genetic effects
Alexander I. Young, Seyed Moeen Nehzati, … Augustine Kong·Nature GeneticsPeer-reviewed·Jun 1, 2022
- Estimating genetic nurture with summary statistics of multi-generational genome-wide association studies
Yuchang Wu, Xiaoyuan Zhong, … Qiongshi Lu·Proceedings of the National Academy of SciencesPeer-reviewed·Jun 15, 2021
- Interpreting population and family-based genome-wide association studies in the presence of confounding
Carl Veller, Graham Coop·PLoS BiologyPeer-reviewed·Apr 11, 2024
- Family-GWAS reveals effects of environment and mating on genetic associations
Tammy Tan, Hariharan Jayashankar, … Alexander I. Young·medRxiv·Oct 1, 2024
- Novel estimators for family-based genome-wide association studies increase power and robustness
Junming Guan, Seyed Moeen Nehzati, … Alexander I. Young·bioRxiv·Oct 24, 2022
- Estimating uncertainty in family-based GWAS
Xinyi Miao, Michael D. Edge, Arbel Harpak·bioRxiv·Aug 21, 2026