henhaobio
Sign in

medRxivGenetic & Genomic MedicinePreprint

Family-GWAS reveals effects of environment and mating on genetic associations

Tammy TaniD, Hariharan Jayashankar, Junming GuaniD, Seyed Moeen Nehzati, Mahdi Mir, Michael BennettiD, Esben AgerboiD, Rafael AhlskogiD, Ville Anapaz, Stefania BenónísdóttiriD, Dorret Irene BoomsmaiD, Archie I CampbelliD, Christopher F. ChabrisiD, Zhengming CheniD, China Kadoorie Biobank Collaborative Group, Eco J. C. de GeusiD, Erik A. EhliiD, Abdelrahman G. ElnahasiD, FinngeniD, Andrea GannaiD, Alexandros GiannelisiD, Liisa HakasteiD, Caroline HaywardiD, Jouke‐Jan HottengaiD, Mikkel Aagaard HoumarkiD, , Alexander I. YoungiD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Genome-wide association studies (GWAS) have discovered thousands of replicable genetic associations, guiding drug target discovery and powering genetic prediction of human phenotypes and diseases. However, genetic associations can be affected by gene-environment correlations and non-random mating, which can lead to biased inferences in downstream analyses. Family-based GWAS (FGWAS) uses the natural experiment of random assignment of genotype within families to separate out the contribution of direct genetic effects (DGEs) -- causal effects of alleles in an individual on an individual -- from other factors contributing to genetic associations. Here, we report results from an FGWAS meta-analysis of 34 phenotypes from 17 cohorts. We found evidence that factors uncorrelated with DGEs make substantial contributions to genetic associations for 27 phenotypes, with population stratification confounding -- a form of gene-environment correlation -- likely the major cause. By estimating SNP heritability and genetic correlations using DGEs, we found evidence that assortative mating has led to overestimation of SNP heritability for 5 phenotypes and overestimation of the degree of shared genetic effects (pleiotropy) between 22 pairs of phenotypes. Polygenic predictors constructed from DGEs are particularly useful for studying natural selection, assortative mating, and indirect genetic effects (effects of relatives genes mediated through the family environment). We validate our meta-analysis results by predicting phenotypes in hold-out samples using polygenic predictors constructed from DGEs, achieving statistically significant out-of-sample prediction for 24 phenotypes with little attenuation of predictive power within-families. We provide FGWAS summary statistics for 34 phenotypes that can be used for downstream analyses. Our study provides both a template for performing FGWAS and an argument for its value for debiasing inferences and understanding the impact of environment and mating patterns.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 3 (22 Jan 2026) · CC BY 4.0

Family-based GWAS meta-analysis of 34 phenotypes shows standard GWAS effects carry sizeable confounding uncorrelated with direct genetic effects, while assortative mating likely inflates SNP heritability and genetic correlations.

Standard GWAS associations mix direct genetic effects with gene-environment correlation, indirect genetic effects and non-random mating, which can bias downstream heritability, genetic correlation, Mendelian randomization and selection inferences. Compared with the sib-GWAS of Howe et al. (24 phenotypes, INFO >0.3), this work imputes missing parental genotypes with snipar to combine sibling and parent-offspring data, uses an INFO ≥0.99 threshold and adds 18 phenotypes.

Key results

  1. ShownDirect genetic effects (DGEs) and GWAS population effects are less than perfectly correlated genome-wide for most phenotypes, including several unrelated to education or cognition. LDSC-based correlations, mostly near 1, point to uncorrected population stratification as the main source.

    The snipar estimate was below 1 for 27/30 phenotypes (FDR<0.05, one-sided) and under 0.75 for 13. For educational attainment (EA) it was 0.689 (S.E.=0.023) from meta-analysed statistics vs 0.547 (S.E.=0.052) from pooling cohort-level estimates. LD score regression (LDSC) estimates were mostly close to 1, with height, EA and HDL cholesterol among those significantly below 1. · Correlations between direct genetic effects and…; Discussion

  2. ShownA substantial share of genome-wide population-effect variance is uncorrelated with DGEs: modest for height and BMI, near half for EA and a majority for depressive symptoms; LDSC intercepts and ratios also indicate less confounding in DGE statistics.

    Share uncorrelated with DGEs: 10.7% (S.E.=0.6%) for height, 10.2% (S.E.=1.0%) for BMI, 48.3% (S.E.=3.2%) for EA, 58.7% (S.E.=7.0%) for depressive symptoms. LDSC intercept above 1 for 29 phenotypes using population effects vs 15 using DGEs. Mean LDSC ratio 0.104 (DGE) vs 0.306 (population) over 18 adequately powered phenotypes. · Substantial contribution of confounding to GWAS…; LDSC Intercept Analyses

  3. ShownSNP heritability from DGEs is lower than from population effects for five phenotypes, all affected by assortative mating, which the authors see as the main cause. A previously reported gap for cognitive performance was not replicated.

    FDR<0.05 for age at first birth (women), EA, depression, ever-smoker and height. EA: 7.2% (S.E. 0.8%) from DGEs vs around 14% from population effects. Depression: 2.5% (S.E. 1.5%) vs 8.2% (S.E. 1.0%). Cognitive performance: 0.188 (S.E. 0.027) vs 0.186 (S.E. 0.016), whereas Howe et al. reported 0.14 vs 0.24. · Smaller SNP heritability from direct genetic…; Discussion

  4. ShownGenetic correlations (rg) from DGEs differ from those based on population effects for 22 trait pairs, and correlations involving EA are generally attenuated. The pattern tracks cross-mate phenotypic correlations, consistent with cross-trait assortative mating inflating population-effect rg.

    GenomicSEM equality tests, FDR<0.05. EA-BMI rg: -0.288 (S.E. 0.026) from population effects vs -0.068 (S.E. 0.045) from DGEs. In 39,710 putative UK Biobank mate pairs, cross-mate phenotypic correlations explained R = 40.77% (95% CI 34.44%-46.52%) of population-effect rg vs 15.37% (10.41%-20.71%) of DGE rg. · Robust estimation of pleiotropy using direct…; Investigation of the relationship between…

  5. ShownPolygenic indices (PGIs) built from DGE summary statistics predict out of sample for 24 phenotypes and lose little power within families. An EA PGI built from population effects loses much more power within families.

    Validated in the Millennium Cohort Study (primary) and UK Biobank (secondary). Height DGE PGI partial R² 17.9% (95% C.I. 15.8%-20.1%). Diastolic and systolic blood pressure, HDL, non-HDL cholesterol and EA DGE PGIs all exceeded 1%. EA PGI R² was 1.8% (DGE) vs 5.3% (population effect), yet the population-effect PGI showed far greater within-family attenuation. · PGIs based on direct genetic effects exhibit…; Validation Phenotypes; PGI Analyses

Methods, data and reproducibility

16 mostly European-ancestry and one East Asian cohort ran family-based GWAS in snipar (own plus parental genotypes; missing parents filled by Mendelian imputation from relatives; only variants with imputation INFO ≥0.99). Multivariate fixed-effect meta-analysis of European cohorts gave DGEs, non-transmitted coefficients and population effects, compared by moment estimators, LDSC, GenomicSEM and PRS-CS PGIs.

  • Family-based GWAS results contributed by 16 mainly European-ancestry cohorts (incl. FinnGen, Finnish Twin Cohort, Swedish Twin Register, Minnesota Twins, iPSYCH, Botnia Family Study) · human · n = 34 phenotypes; 30 retained after DGE effective sample size filter of 5000
  • China Kadoorie Biobank FGWAS summary statistics (East Asian genetic ancestry), analysed separately · human · n = 15 phenotypes
  • Millennium Cohort Study, primary PGI validation cohort (UK births around 2000; about half with both parents genotyped) · human
  • UK Biobank white British subsample, secondary PGI validation and mate-pair identification · human · n = 39,710 putative mate pairs
  • Sakaue et al. GWAS summary statistics used as reference for China Kadoorie Biobank comparison · human · n = 4 phenotypes
  • PRS-CS EUR LD reference panel built from UK Biobank HapMap3 SNPs · human · n = 1,117,425 SNPs
  • 1000 Genomes EAS allele frequencies for China Kadoorie Biobank quality control · human (1000 Genomes is a human reference panel)
Methods & tools
snipar, LDSC v1.0.0, stratified LDSC, GenomicSEM v0.0.5, PRS-CS, lavaan v0.6-15, stan v2.21.0, brms v2.17.0, lme4
Data & accessions
https://thessgac.com/
Software & environment
Software versions stated: LDSC v1.0.0, GenomicSEM v0.0.5, lavaan v0.6-15, stan v2.21.0, brms v2.17.0. No container or dependency file mentioned.
Reproducibility gaps
Summary statistics are promised via the SSGAC portal 'by the time of publication', though the Methods say DGE statistics are publicly available. MoBa and HUNT withdrew their summary statistics after a competing-interest disclosure. Cohort-level data access is not described.
Limitations (6)
  • Analyses were limited to common variants (minor allele frequency >1%), so the conclusions may not carry over to rare-variant GWAS. · stated by the authors
  • DGE effective sample sizes remain roughly an order of magnitude below the largest GWAS meta-analyses, so in the near term population-effect PGIs will likely predict better than DGE PGIs. · stated by the authors
  • The China Kadoorie Biobank showed the opposite pattern (higher DGE-based SNP heritability for BMI, EA and height). This may reflect ascertainment or negative gene-environment correlation, and more East Asian data are needed to tell whether it is cohort specific. · stated by the authors
  • The share of population effects uncorrelated with DGEs is attributed mainly to population stratification, but indirect genetic effects (IGEs) weakly correlated with DGEs or cross-trait assortative mating could contribute. The share is likely smaller at genome-wide-significant loci. · stated by the authors
  • The number of phenotypes with a DGE-population correlation below 1 is 27/30 in the Results but 26 in the Discussion, so the supplementary tables need checking. · noted in this summary
  • The cross-trait assortative mating analysis rests on 39,710 putative UK Biobank mate pairs inferred from shared household and location fields, not confirmed spouses. · noted in this summary

Read this if you use GWAS summary statistics for heritability, genetic correlation, Mendelian randomization or polygenic-score work and want to know how much population stratification and assortative mating may bias them.

Citations
47

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

Similarity Map
SelectedSame fieldCross-field
THISHerzig '26Veller '24Howe '21Abdellaoui '22Young '23Trejo '19
75% semantic matchGenetics
Anthony F. Herzig, Camille Noûs, Aude Saint Pierre+1 more·Theoretical Population Biology·Jun 15, 2026· 2 citations