Decomposing heritability and genetic covariance by direct and indirect effect paths

Jie Song; Yiqing Zou; Yuchang Wu; Jiacheng Miao; Ze Yu; Jason M. Fletcher; Qiongshi Lu

doi:10.1371/journal.pgen.1010620

Abstract

Estimation of heritability and genetic covariance is crucial for quantifying and understanding complex trait genetic architecture and is employed in almost all recent genome-wide association studies (GWAS). However, many existing approaches for heritability estimation and almost all methods for estimating genetic correlation ignore the presence of indirect genetic effects, i.e., genotype-phenotype associations confounded by the parental genome and family environment, and may thus lead to incorrect interpretation especially for human sociobehavioral phenotypes. In this work, we introduce a statistical framework to decompose heritability and genetic covariance into multiple components representing direct and indirect effect paths. Applied to five traits in UK Biobank, we found substantial involvement of indirect genetic components in shared genetic architecture across traits. These results demonstrate the effectiveness of our approach and highlight the importance of accounting for indirect effects in variance component analysis of complex traits.

Author summary

Complex human traits are influenced by both genetic and environmental factors. Recent evidence suggests that genetic effects on human traits can sometimes be mediated by the environment. One example is that parental genomes could influence parental behavior and family environment, which, in turn, shape the phenotypes of their children. Since parent and offspring genotypes are correlated, these environmentally mediated effects will be captured by conventional genome-wide association studies and may potentially bias our interpretation of genotype-phenotype associations. In this work, we propose a principled statistical framework named PARSEC to partition the genetic components of single traits and the shared genetic components between traits into direct and indirect effect paths. It marks an important methodological innovation by resolving a limitation in almost all existing genetic correlation estimation approaches and may have broad and impactful applications in future GWAS analysis for complex traits that are affected by parental genetics and family environment.

Citation: Song J, Zou Y, Wu Y, Miao J, Yu Z, Fletcher JM, et al. (2023) Decomposing heritability and genetic covariance by direct and indirect effect paths. PLoS Genet 19(1): e1010620. https://doi.org/10.1371/journal.pgen.1010620

Editor: Heather J. Cordell, Newcastle University, UNITED KINGDOM

Received: September 16, 2022; Accepted: January 16, 2023; Published: January 23, 2023

Copyright: © 2023 Song et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: This research has been conducted using the UK Biobank resource under Applications 42148 and 57284. The data underlying the results presented in the study can be requested from UK Biobank (https://www.ukbiobank.ac.uk). Implemented software of our approach is freely available at https://github.com/qlu-lab/PARSEC.

Funding: This work was supported by the National Institute on Aging (P30 AG017266 to JF and QL) and the Wisconsin Alumni Research Foundation (QL). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Genetic correlation is an effective metric for quantifying the shared genetic architecture of multiple complex traits [1,2] and has quickly gained popularity in genome-wide association studies (GWAS) in the past few years. Genetic correlation is typically defined as the correlation between additive genetic components of two complex traits and can be estimated from genome-wide data of millions of single nucleotide polymorphisms (SNPs). Several methods have been developed to improve genetic correlation estimation using individual-level GWAS data [3,4], GWAS summary statistics [5,6], or both [7]. Recent studies have also expanded this concept to quantify genetic correlations in local genomic regions [8–11], between human ancestral populations [12,13], and using other types of genetic variations [14]. Overall, these methods have become a routine component of complex trait genetic studies and provided insights into the genetic basis of numerous human traits.

As genetic correlation analysis becomes standard in GWAS, challenges arise in interpreting the pervasive correlations observed across many phenotypes. We now know that parents’ genomes can influence parental behaviors and family environment which, in turn, shape the phenotypes of their children. Additionally, because parent and offspring genotypes are correlated, typical GWAS can be severely confounded by parental genotypes, and effect estimates obtained in GWAS are mixtures of direct genetic effects (i.e., how one’s own genotypes affect his/her phenotype) and indirect genetic effects (i.e., how parents’ genotypes affect their children’s phenotypes; this is also referred to as genetic nurture) [15,16]. Therefore, interpretation of genetic correlations involving human behavioral phenotypes is particularly challenging because most methods for estimating genetic correlations [6,17] ignore these apparent gene-environment correlations underlying human behavior [18–25]. Being able to decompose shared genetics between two traits by direct and indirect effect paths is crucial for understanding the respective roles of genetics and environments. Although methods have been developed for separating SNP effects into direct/indirect paths, and follow-up genetic correlation analyses performed on direct/indirect association statistics can be a feasible strategy in some cases [26–29], these GWAS typically require data from a large number of families which can be difficult to obtain even in large biobanks.

In the past few years, an extensive literature has investigated the decomposition of phenotypic variance based on gene-environment relations. Methods like M-GCTA [30,31], Trio-GCTA [32], RDR [33], and GREML-KIN [34] have leveraged family structures in genetic samples to improve heritability estimation and quantify variance components for maternal, paternal, offspring, and sibling genetic effects. Several methods have also allowed imputation of parental genotypes which enables estimation of parental and offspring genetic effects in downstream GWAS applications [25,35,36]. These methods decompose the phenotypic variance of a single complex trait into multiple variance components according to direct and indirect effect paths, but none of them can be applied to partition genetic covariance–the shared additive genetic component between a pair of complex traits.

In this paper, we introduce a statistical framework named PARSEC (PARental Stratification of genetic Effect Components) to estimate the variance and covariance components of direct/indirect genetic effects on trait pairs using data from a limited number of families. This framework enables us to partition both heritability and genetic covariance into direct and indirect pathways, and gain knowledge of how genetics and environment jointly contribute to complex traits as well as the correlation between trait pairs. We employ the method of moments to obtain accurate estimates of variance and covariance parameters from family-based data and use a Jackknife approach to obtain confidence intervals. We apply our framework to five complex traits in UK Biobank (UKB) [37] and partition the trait-specific and shared genetic components by direct and indirect effect paths.

Results

Method overview

Our approach is built on a pair of linear models for a pair of standardized traits Y₁ and Y₂, standardized own- genotypes X, and standardized parental genotypes X_p+X_m:

Error terms ϵ₁ and ϵ₂ are assumed to have mean 0 and variance-covariance matrix Σ. α_k and β_k (k = 1, 2) are random effect vectors denoting direct (of own- genotypes) and indirect effects (of parental genotypes) on two traits. They share the variance-covariance structure below with M being the number of SNPs and I_M being the identity matrix.

Here, ρ_α and ρ_β are two key parameters we focus on, although other cross terms are also estimated and may be of interest in practice. These two parameters quantify the covariance of direct effects and covariance of indirect effects on two traits, respectively. We apply the method of moments to produce parameter estimates and employ a Jackknife approach to quantify the standard error of parameter estimates. More statistical details are described in the Methods section.

Simulation results

We first performed simulations to demonstrate that our method produces unbiased estimates with well-calibrated confidence intervals. We randomly selected 2000 families from UKB with data on full sibling pairs and imputed parental genotypes to perform simulations. The imputation is based on expectation of the average of parental genotypes using two or more sibling genotypes [36]. We explored multiple parameter settings. Each setting contained 200 repeats. Details on simulation settings are described in the Methods section and in S1 Table. Results related to covariance of direct effects and covariance of indirect effects are shown in Fig 1. Results for the estimates of all 14 parameters are shown in S1 Fig. We observed unbiased estimates and well-controlled confidence interval coverage rates for all parameters across various settings.

Download:

Fig 1. Simulation results on point estimates, type-I error, and statistical power for covariance of direct/indirect effects under different simulation settings.

The first row and the second row are respect to two simulation settings respectively. Panels A and C: box plots for point estimates of ρ_α (i.e., covariance of direct effects) and ρ_β (i.e., covariance of indirect effects). Red dotted lines indicate the true values of ρ_α. Blue dotted lines show the true value of ρ_β. Panels B and D: proportion of p-values < = 0.05 across 200 repeats. In panel B, the dotted line highlights the type-I error threshold 0.05.

https://doi.org/10.1371/journal.pgen.1010620.g001

Partitioning heritability by direct/indirect effect paths for five complex traits

We applied our model to 12,571 families in UKB with full sibling data available and parental genotypes imputed (Methods). A focal trait we studied in this paper is educational attainment due to its known substantial component in the indirect genetic effect path [15]. In our previous work [26], we partitioned direct and indirect genetic components for educational attainment and estimated their genetic correlations with many traits. Notable findings from that study include correlations with anthropometric traits (e.g. height) as well as numerous health outcomes. In addition, income GWAS is known to have a very high genetic correlation with education [38]. Thus, we chose educational attainment (EA) with height, body mass index (BMI), overall health, and income as the phenotypes of interest in this study. For each trait pair, we estimated 14 parameters including trait-specific parameters (which partition heritability), trait pair parameters (which partition genetic covariance), and parameters for error terms. The detailed data processing procedure is described in the Methods section.

First, we examined parameter estimates for the variance of direct genetic effects, variance of indirect genetic effects, and covariance between direct and indirect effects on the same trait. These parameters are the partitioned terms of trait heritability; therefore, we can reconstruct total heritability estimates by linearly combining these three terms. For comparison, we also calculated LDSC heritability estimates using GWAS performed on the same data. We observed a high correlation (correlation = 0.962; Fig 2A).

Download:

Fig 2. Total heritability and genetic covariance can be recovered from partitioned parameter estimates.

A: estimates for total heritability. B: estimates for genetic covariance. X-axis: LDSC estimates. Y-axis: Estimates reconstructed from PARSEC estimates in our analysis.

https://doi.org/10.1371/journal.pgen.1010620.g002

Results of these three parameters are shown in Fig 3A and S2 Table. We found a substantial and statistically significant direct effect component on BMI (p = 4.9×10⁻⁵). The BMI indirect effect was close to zero and not significant. Similarly, we found a significant direct genetic component (p = 4.8×10⁻³) and a weaker indirect component for height. Notably, the direct and indirect genetic components for height showed a significant positive covariance (p = 7.7×10⁻³). The direct and indirect genetic components for EA were similar in size but only the indirect component reached statistical significance. Consistent with multiple previous studies [26,36], we did not find evidence for a genetic correlation between the EA direct and indirect effects, although the reason for the lack of correlation remains unclear. We also found null results for household income and overall health.

Download:

Fig 3. Estimation results for 5 complex traits in UKB.

Each interval shows the point estimate and standard error for one parameter. Circles denote significant parameter estimates at the 0.05 level. Other estimates are denoted by solid circles. A: variance of direct effects, variance of indirect effects, and covariance of direct and indirect effects for height, BMI, overall health, EA and income. B: covariance of error terms for sibling pairs for each trait.

https://doi.org/10.1371/journal.pgen.1010620.g003

Some other parameters in PARSEC also have important implications. For example, we obtained the estimated covariance between siblings’ error terms for each trait which quantifies the degree to which siblings’ phenotypic correlation is explained by their shared environment (Fig 3B and S2 Table). Error term covariances are significant for all 5 traits in our analysis. Branigan et al. [39] previously measured the contribution of shared environment between twins on EA using ACE model. They obtained an estimate of around 0.3 from two UK cohorts. However, this strategy does not distinguish environmental contributions from the factors that can be explained by parental genetics. PARSEC provides a useful alternative strategy to estimate the environmental contribution between full siblings and tease part the contribution of parental genomes. We obtained an estimate of 0.22 for EA.

Partitioning genetic covariance among five complex traits

PARSEC also partitions the total genetic covariance between two traits into direct and indirect effect paths. Similar to the analysis for total heritability, we reconstructed the total genetic covariance estimates based on partitioned parameters (Methods) and compared them to LDSC estimates. We found a high correlation between these estimates (Fig 2B, correlation = 0.977).

Estimation results for ρ_α (i.e., genetic covariance of direct effects on two traits) and ρ_β (i.e., genetic covariance of indirect effects on two traits) are shown in Fig 4A and 4B (also see S3 Table). We found a significant direct effect covariance (ρ_α = 0.044) between BMI and worse overall health (a higher value in this trait indicates worse health; see Methods). This contributes to 69% of the total genetic covariance between BMI and overall health. Several other trait pairs showed significant covariances of indirect genetic components, including height and BMI (ρ_β = -0.02), height and income (ρ_β = 0.032), and EA and overall health (ρ_β = -0.018). EA and income have substantial and statistically significant covariances in both direct and indirect genetic components (ρ_α = 0.044 and ρ_β = 0.032). Across all trait pairs, direct effects explain an average of 79.6% of total genetic covariance, suggesting substantial contribution of indirect effect components and their correlation with direct effects.

Download:

Fig 4. Estimation results for covariance of direct/indirect effects on five complex traits.

A: upper triangle: heatmap of direct effect genetic covariance (i.e., ρ_α); lower triangle: heatmap of indirect effect genetic covariance (i.e., ρ_β). B: covariance between direct effects of trait 1 (rows) and indirect effects of trait 2 (columns). Size of the circles visualizes false discovery rate (FDR). Smaller size refers to larger FDR. Significant correlations with FDR<0.05 are marked by asterisks.

https://doi.org/10.1371/journal.pgen.1010620.g004

We also estimated other cross-term covariances, including the covariance of direct effect on one trait and indirect effect on another trait (Fig 4C and 4D). We identified a number of significant correlations. For example, we found highly significant correlations between the indirect effect of BMI and the direct effect of both EA ( = -0.037) and income ( = -0.028). The direct effect of income is also negatively correlated with the indirect effects of overall health ( = -0.02). The direct effect of height is correlated with the indirect effect of better overall health ( = -0.02), higher EA ( = 0.02), and lower income ( = - 0.029).

Discussion

Genetic correlation/covariance analysis has rapidly gained popularity in the past few years. Almost all GWAS papers these days estimate genetic correlation to gain insights into the shared genetics across multiple complex traits. However, interpretation of these correlations can be unsatisfying. Recent efforts in dissecting the direct and indirect trait genetic components have advanced this topic. For example, Warrington et al. estimated the direct and indirect (maternal) effects on birth weight, then correlated them with other complex traits [40]. They concluded that the genetic correlation between birth weight and higher educational attainment is mostly explained by the maternal effect on birthweight, not the direct effect. That is, mothers with higher genetic propensity for EA tend to have children with higher birth weight, while the fetus’ genetic propensity on education and birth weight seem uncorrelated. Similarly, in our recent study, we estimated direct and indirect genetic effects on EA leveraging multi-generational GWAS data [26]. We found that the well-documented genetic correlation between lower EA and higher depression risk is mostly explained by the indirect component of EA. That is, parents who have depression risk may negatively impact children’s school performance. Although inference of causality remains challenging, partitioning shared genetic components into direct and indirect paths has led to some incredibly interesting hypotheses that were previously neglected in the literature.

In these examples, people were able to partition the direct/indirect effects of one trait, and then quantify how they correlate with the combined effects of a second trait. In this paper, we introduced PARSEC, a new statistical framework to decompose heritability and genetic covariance of complex traits by direct and indirect effect paths. Our approach provides a more refined partition of the shared genetic components, allowing both traits’ direct/indirect effects to be investigated which provides a more complete picture of the complex genetic sharing underlying multiple phenotypes. Through simulations and application to five traits using full siblings and their imputed parental genotypes in UKB, we demonstrated that our model produces unbiased estimates with well-controlled confidence interval coverage.

Partitioned variance and covariance components can be combined to recover traditional heritability and genetic covariance estimates. But stratified heritability and genetic covariance estimates provided important new insights into the genetic basis of complex traits. First, our analysis confirmed that traits like EA have a substantial indirect genetic component. One interpretation is that a substantial proportion of trait heritability is mediated through parents and the family environment they create although other mechanisms remain possible [41]. For some other traits (e.g., BMI), the contribution of indirect genetic effect is negligible, and heritability is primarily explained by the direct genetic component. More importantly, our approach allowed decomposing genetic covariance, which is novel compared to other approaches in the literature. Similar to findings in single-trait analysis, our results highlight the importance of considering indirect genetic components when estimating and interpreting the shared genetic basis between complex traits. Several trait pairs in our analysis (e.g., height and income) only showed significant genetic covariance in their indirect genetic components. Some other traits (e.g., BMI and overall health) have substantial genetic sharing of the direct effect component. For EA and income, genetic covariances of both direct and indirect genetic components were substantial and highly statistically significant. Omitting indirect effect components in genetic correlation analysis fails to identify the family environment as a key factor contributing to cross-trait genetic correlation and will likely lead to biased interpretation. With our approach, it is now possible to carefully decompose heritability and shared genetic components into direct and family-mediated paths, which will benefit post-GWAS analysis of human socio-behavioral phenotypes.

Our study also has several limitations. First, parental genotypes used in our analysis were not directly measured but imputed based on siblings’ genotypes. The imputation accuracy depends on several factors including the degree of assortative mating in the population and number of siblings in each family. These may affect the accuracy of parameter estimates in PARSEC. But based on analysis that was conducted by SNIPar, the estimates for multiple regression were proved to remain unbiased and consistent. On the other hand, while our approach can be applied to measured genotypes and phenotypes in parent-offspring trios, available data in existing population cohorts often have smaller sample sizes which may lead to reduced statistical power. Second, we focused on genetic covariance instead of genetic correlation in our cross-trait analysis. Although it is also possible to calculate genetic correlations by combining stratified genetic covariance and heritability estimates, these correlation estimates tend to be numerically unstable at the sample size in our analysis. Third, our analyses were limited to quantitative traits. In our current analysis, multiple traits (i.e., EA, income, and overall health) were converted into continuous values from categorical input. For applications involving binary or categorical outcomes, future work is needed to expand the statistical model. Fourth, our analysis does not distinguish maternal and paternal indirect effects due to constraints in the data (i.e., only average parental genotype can be imputed), although we note that our model can be easily extended to estimate maternal and paternal effects separately if genotype data are available for both parents. We also assumed absence of sibling effects and indirect genetic effects from other distant family members. This assumption is partly supported by evidence in the literature [36], but these effects may be of interest for specific traits. An interesting future direction is to include sibling effects using cohorts with larger samples. Fifth, methods have been developed to leverage multi-generational GWAS summary statistics to quantify direct and indirect genetic effects at the SNP level [26]. While heritability and genetic covariance estimation based on individual-level data is known to be statistically more efficient compared to estimates produced from GWAS summary statistics [6,42], summary statistics-based analysis has the advantage of combining data from many studies while a single cohort of individual-level information is unavailable. Future studies need to quantitatively compare different analytical strategies in realistic data settings in order to understand the pros and cons of each approach and guide future studies. Finally, we assumed random mating and a purely additive genetic architecture in our model. The influence of assortative mating on heritability and genetic correlation estimation has been convincingly demonstrated in recent studies [43,44]. Genome-wide dominant effects [45] and epistasis [46] have also been studied for various complex traits with mixed results. It is an important future direction to expand the PARSEC model and develop estimators robust to these issues.

Taken together, our method is a first step to explore the covariance structure of direct and indirect effects for complex human traits. It marks an important methodological innovation and may have broad and impactful applications in future GWAS analysis for traits that are affected by parental genetics and family environment.

Methods

PARSEC model details

We assume a pair of linear genetic models as follows: where Y₁ and Y₂ denote two complex traits, X denotes own genotypes, and X_p and X_m denote paternal and maternal genotypes. Since we used sibling pairs data as input (that will be described later in this section), we assume the total number of sibling pairs is N. Then Y₁ and Y₂ are vectors with length 2N, and X, X_p and X_m are 2N-by-M matrices, where M is the total number of SNPs. We assume both genotypes and phenotypes (Y₁, Y₂, X, X_p, and X_m) to be standardized. α₁ and α₂ are random effect terms that quantify the direct genetic effects on Y₁ and Y₂, i.e., how someone’s own genotypes affect their own phenotype. Following conventions in the linear mixed model literature on modeling heritability [47], we also assume these effects to follow normal distributions.

Here, and are the direct genetic variance components for two traits respectively. Similarly, β₁ and β₂ are random effect terms for quantifying indirect genetic effects on two traits, i.e., how the parents’ genotypes affect someone’s phenotype. These random effects also follow normal distributions.

Importantly, direct and indirect effects on the same trait (i.e., α_k and β_k) can be correlated. Direct effects on two different traits (i.e., α₁ and α₂), or indirect effects on two traits (i.e., β₁ and β₂), can both be correlated. Fairly generally, we assume α₁, α₂, β₁, and β₂ to have a joint distribution:

Here, ρ_α quantifies the covariance of direct effects on two traits and ρ_β is the covariance of indirect effects on two traits. Similarly, , and are pairwise covariance parameters between α and β. We note that this model is similar to what is used in the genetic nurture literature [15,26] in that it quantifies how parental genotypes shapes children’s phenotypes. But a difference is that it is a polygenic model that uses random variables α_k and β_k to characterize genome-wide effects. This is motivated by the linear mixed model literature on heritability and genetic covariance estimation.

Based on the model formulation, naturally one would assume that the data required to fit this model would be genotypes of parent-offspring trios (i.e., X, X_p, and X_m) and offspring phenotypes on two traits (i.e., Y₁ and Y₂). While this is true, the number of trios available even in large population biobanks such as UKB is limited. Therefore, we made two important adjustments. One is that in our analyses, we leveraged the relatively large number of full sibling pairs available in UKB and recent statistical genetic advances in parental genotype imputation [36]. We begin with N sibling pairs (thus the total sample size is 2N), then impute the sum of parental genotypes X_p+X_m. These become the input data in our analysis. This also leads to the second adjustment we made which has also been shown above–we do not distinguish paternal and maternal indirect effects and instead focus on their average. This is similar to what was done in the indirect effect GWAS literature when sample size was small [26]. In practice, in cases where a large number of trios are available, paternal and maternal indirect effects may be denoted as separate random variables in this framework. As we mentioned, the indirect effect component quantifies the averaged maternal and paternal effects. If one dominates the other (such as maternal indirect effects dominates paternal effects), the estimate still has a clear interpretation. We illustrate this in more details as below:

Suppose the true model for children’s phenotype Y in a family is Where α, β_p, and β_m are the direct, indirect paternal, and indirect maternal effects, respectively, and ϵ denotes the noise term. Without loss of generality, we assume Y, X, X_p, and X_m are standardized to have mean 0 and variance 1. If we further denote transmitted alleles and non-transmitted alleles as T and NT respectively, then the model above can be rewritten as:

We also note that the child’s genotype is the sum of maternally and paternally transmitted alleles, i.e., X = T_p+T_m.

Now, recall that the actual model we fit in the PARSEC approach is

Here, maternal and paternal genotypes are combined and β quantifies the indirect genetic component in this framework. Fitting this model on data generated from the true model we introduced above will give us a β estimate that is the projection of T_pβ_p+T_mβ_m on X_p+X_m. It can be further shown that this projection is:

We have previously derived this relationship in an earlier paper [26]. Therefore, the estimated indirect effect β is the average of indirect maternal genetic effect β_m and indirect paternal genetic effect β_p, which also means . Even if maternal indirect effect dominates paternal indirect effect, our approach still provides an unbiased estimate for the variance of average indirect effect.

One implication of having sibling pairs instead of independent samples or trios in the analysis is that we also need to consider the shared environmental effects between siblings. Error terms on two traits (i.e., ϵ₁ and ϵ₂) are both normally distributed random variables. But we allow 1) correlation of errors on two traits for the same individual, which is a common assumption in the genetic correlation estimation literature [5,6], and importantly, we also allow 2) correlation of error terms between siblings due to their shared environments. More specifically, we assume:

We denote ϵ_1i, ϵ_2i as such family i sibling pair error terms for trait 1 and 2 respectively, assume , where .

Parameter estimation

For simplicity, we denote the whole set of 14 parameters as , and use Θ_s to denote the kth parameter. Rather than utilizing an iterative restricted maximum likelihood algorithm to solve the linear mixed model, we employ the method of moments to improve computational efficiency. This approach produces parameter estimates by minimizing the distance between the model-derived variance-covariance matrix and cross-product matrix of phenotypes, i.e., cov(Y) where . Since we assumed an additive penetrance model, we can show that the model-derived covariance also follows an additive form: where denote the sample relatedness matrix corresponding to s-th parameter. For example, and . We list all the relatedness matrices in S4 Table. Notice that the variances of error terms are not included in the parameter list. We replace them by one minus the rest of the variance terms since the phenotypic variance is standardized to be one. The cross-product matrix of phenotypes is the estimate of cov(Y) based on real data, i.e., . We estimate all parameters by minimizing the following function where ‖∙‖_F is the Frobenius norm. To minimize this function, we use the gradient descent method. That is, we let for all k and obtain a linear system of the form AΘ = B, where A is a matrix and B is a vector. Then we could obtain parameter estimates by solving this linear equation.

While the variance of each parameter estimate can be derived under this statistical framework, they can be numerically unstable (and even turn negative) when sample size is limited. Therefore, we applied a resampling-based block Jackknife approach [48] to quantify the variance of parameter estimates. Block Jackknife generates variance estimates by dividing the whole dataset into B blocks, holding out one block and producing parameter estimates using the other B-1 blocks in each time, then repeating this procedure for B times. Finally, we estimate variances using following formula: where denotes the estimate for s-th parameter in b-th repeat, and denotes the averaged value across B estimates for parameter Θ_s. To account for multiple testing in real data applications, we calculated false discovery rate (FDR) based on all estimates for all pairs of traits.

Our model can also incorporate fixed effect covariates [49]. For simplicity, we ignore the subscripts in the proposed model and add fix effects as follows: where Z denotes the covariate data matrix and γ denotes their fixed effects. To account for fixed effects, we can multiply a matrix Q that satisfies QZ = 0 to both sides of the equation. Matrix Q could be obtained by deriving the orthogonal vector spaces of the singular value decomposition of matrix Z. After multiplying matrix Q on both sides, the equation becomes the following without a fixed effect term.

Then, the estimation procedure for all parameters is the same as before.

Reconstructing heritability and genetic covariance using decomposed variance components

Following the model described above, total heritability h² and genetic covariance ρ_gc can be recovered from the decomposed variance component parameters as follows:

By plugging in empirical estimates on the right-hand side of these equations, we could obtain estimates for total heritability and genetic covariance (Fig 2).

Data processing

UKB samples with European ancestry were identified from principal component analysis (data field 22006). We used KING [50] to infer the pairwise family kinship and created family identifiers in UKB. Sum of parental genotypes were imputed by SNIPar [36]. Only autosomal SNPs with minor allele frequencies (MAF) greater than or equal to 0.05 and missing rate less than 0.01 were used for the analysis. Genotypes were standardized to have mean 0 and variance 1 using estimated minor allele frequencies, and missing values were imputed as 0. The sum of parental genotypes was standardized to have a variance of 2.

For phenotypes, participants were selected as overlapping samples of five phenotypes: height, BMI, EA, income, and overall health. We obtained height and BMI phenotypes from UKB data fields 12144 and 21001. Following previous work [51], we used data field 6138 to compute the EA phenotype as years of schooling. Household income data were obtained from UKB data field 738. The answers were coded as follows: (1)–Less than 18,000 (Pounds), (2)– 18,000 to 30,999, (3)– 31,000 to 51,999, (4)– 52,000 to 100,000, (5)–Greater than 100,000. Overall health was defined based on data field 100508. The answers were coded as follows: (1)–Excellent, (2)–Good, (3)–Fair, (4)–Poor. Note that a higher value indicates poorer health. We removed individuals with missing phenotype values and standardized all 5 traits to have mean 0 and variance 1. The final dataset we used for the analysis includes 4,748,473 SNPs and 12,571 sibling pairs (25,142 individuals and their imputed parents).

Simulation settings

We utilized imputed genotypes in UKB as input. SNPs with missing rate greater than 5% and minor allele frequency less than 5% were removed from the dataset. 2000 full sibling pairs were randomly selected in UKB. We simulated phenotypes using real genotypes and preset parameters. A total of four settings (i.e. four sets of preset parameter values) were explored. The parameter values for each setting are listed in S1 Table. All settings were repeated 200 times. In each repeat, we simulated phenotypes by generating indirect effects, direct effects, and error terms based on the proposed penetrance model using the true parameters. Then, we produced parameter estimates using our statistical framework, and obtained variance estimates using block Jackknife.

Supporting information

S1 Fig. Parameter estimates in simulations.

Panels A and B show results for simulation settings 1 and 2, respectively. Red dots denote pre-determined true parameter values in simulations.

https://doi.org/10.1371/journal.pgen.1010620.s001

(TIF)

S1 Table. Parameter settings for simulations.

https://doi.org/10.1371/journal.pgen.1010620.s002

(XLSX)

S2 Table. Parameter estimates for single traits analysis.

https://doi.org/10.1371/journal.pgen.1010620.s003

(XLSX)

S3 Table. Parameter estimates for trait pair analysis.

https://doi.org/10.1371/journal.pgen.1010620.s004

(XLSX)

S4 Table. Sample relatedness matrices for cov(Y).

https://doi.org/10.1371/journal.pgen.1010620.s005

(PDF)

Acknowledgments

We thank Dr. Yiliang Zhang and members of the Social Genomics Working Group at University of Wisconsin-Madison for helpful comments.

References

1. van Rheenen W, Peyrot WJ, Schork AJ, Lee SH, Wray NR. Genetic correlations of polygenic disease traits: from theory to practice. Nature Reviews Genetics. 2019;20(10):567–81. pmid:31171865
- View Article
- PubMed/NCBI
- Google Scholar
2. Zhang Y, Cheng Y, Jiang W, Ye Y, Lu Q, Zhao H. Comparison of methods for estimating genetic correlation between complex traits using GWAS summary statistics. Briefings in bioinformatics. 2021;22(5):bbaa442. pmid:33497438
- View Article
- PubMed/NCBI
- Google Scholar
3. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76–82. Epub 2010/12/17. pmid:21167468.
- View Article
- PubMed/NCBI
- Google Scholar
4. Loh P-R, Bhatia G, Gusev A, Finucane HK, Bulik-Sullivan BK, Pollack SJ, et al. Contrasting genetic architectures of schizophrenia and other complex diseases using fast variance-components analysis. Nature Genetics. 2015;47(12):1385–92. Epub 2015/11/02. pmid:26523775.
- View Article
- PubMed/NCBI
- Google Scholar
5. Lu Q, Li B, Ou D, Erlendsdottir M, Powles RL, Jiang T, et al. A Powerful Approach to Estimating Annotation-Stratified Genetic Covariance via GWAS Summary Statistics. Am J Hum Genet. 2017;101(6):939–64. pmid:29220677.
- View Article
- PubMed/NCBI
- Google Scholar
6. Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh P-R, et al. An atlas of genetic correlations across human diseases and traits. Nature Genetics. 2015;47(11):1236–41. pmid:26414676
- View Article
- PubMed/NCBI
- Google Scholar
7. Zhang Y, Cheng Y, Ye Y, Jiang W, Lu Q, Zhao H. Estimating genetic correlation jointly using individual-level and summary-level GWAS data. bioRxiv. 2021:2021.08.18.456908.
- View Article
- Google Scholar
8. Guo H, Li JJ, Lu Q, Hou L. Detecting local genetic correlations with scan statistics. Nature Communications. 2021;12(1):2033. pmid:33795679
- View Article
- PubMed/NCBI
- Google Scholar
9. Werme J, van der Sluis S, Posthuma D, de Leeuw CA. LAVA: An integrated framework for local genetic correlation analysis. bioRxiv. 2021:2020.12.31.424652.
- View Article
- Google Scholar
10. Shi H, Mancuso N, Spendlove S, Pasaniuc B. Local genetic correlation gives insights into the shared genetic architecture of complex traits. The American Journal of Human Genetics. 2017;101(5):737–51. pmid:29100087
- View Article
- PubMed/NCBI
- Google Scholar
11. Zhang Y, Lu Q, Ye Y, Huang K, Liu W, Wu Y, et al. SUPERGNOVA: local genetic correlation analysis reveals heterogeneous etiologic sharing of complex traits. Genome Biology. 2021;22(1):262. pmid:34493297
- View Article
- PubMed/NCBI
- Google Scholar
12. Brown BC, Ye CJ, Price AL, Zaitlen N. Transethnic Genetic-Correlation Estimates from Summary Statistics. Am J Hum Genet. 2016;99(1):76–88. Epub 20160616. pmid:27321947; PubMed Central PMCID: PMC5005434.
- View Article
- PubMed/NCBI
- Google Scholar
13. Miao J, Guo H, Song G, Zhao Z, Hou L, Lu Q. Quantifying portable genetic effects and improving cross-ancestry genetic prediction with GWAS summary statistics. bioRxiv. 2022:2022.05.26.493528.
- View Article
- Google Scholar
14. Guo H, Hou L, Shi Y, Jin SC, Zeng X, Li B, et al. Quantifying concordant genetic effects of de novo mutations on multiple disorders. eLife. 2022;11:e75551. pmid:35666111
- View Article
- PubMed/NCBI
- Google Scholar
15. Kong A, Thorleifsson G, Frigge ML, Vilhjalmsson BJ, Young AI, Thorgeirsson TE, et al. The nature of nurture: Effects of parental genotypes. Science. 2018;359(6374):424–8. pmid:29371463
- View Article
- PubMed/NCBI
- Google Scholar
16. Wang B, Baldwin JR, Schoeler T, Cheesman R, Barkhuizen W, Dudbridge F, et al. Robust genetic nurture effects on education: A systematic review and meta-analysis based on 38,654 families across 8 cohorts. The American Journal of Human Genetics. 2021;108(9):1780–91. pmid:34416156
- View Article
- PubMed/NCBI
- Google Scholar
17. Lee SH, Yang J, Goddard ME, Visscher PM, Wray NR. Estimation of pleiotropy between complex diseases using single-nucleotide polymorphism-derived genomic relationships and restricted maximum likelihood. Bioinformatics. 2012;28(19):2540–2. pmid:22843982
- View Article
- PubMed/NCBI
- Google Scholar
18. Bates TC, Maher BS, Medland SE, McAloney K, Wright MJ, Hansell NK, et al. The Nature of Nurture: Using a Virtual-Parent Design to Test Parenting Effects on Children’s Educational Attainment in Genotyped Families. Twin Res Hum Genet. 2018;21(2):73–83. Epub 20180313. pmid:29530109.
- View Article
- PubMed/NCBI
- Google Scholar
19. Trejo S, Domingue BW. Genetic nature or genetic nurture? Introducing social genetic parameters to quantify bias in polygenic score analyses. Biodemography Soc Biol. 2018;64(3–4):187–215. pmid:31852332.
- View Article
- PubMed/NCBI
- Google Scholar
20. Willoughby EA, McGue M, Iacono WG, Rustichini A, Lee JJ. The role of parental genotype in predicting offspring years of education: Evidence for genetic nurture. Molecular psychiatry. 2021;26(8):3896–904. pmid:31444472
- View Article
- PubMed/NCBI
- Google Scholar
21. de Zeeuw EL, Hottenga J-J, Ouwens KG, Dolan CV, Ehli EA, Davies GE, et al. Intergenerational transmission of education and ADHD: effects of parental genotypes. Behavior Genetics. 2020;50(4):221–32. pmid:32026073
- View Article
- PubMed/NCBI
- Google Scholar
22. Cheesman R, Hunjan A, Coleman JR, Ahmadzadeh Y, Plomin R, McAdams TA, et al. Comparison of adopted and nonadopted individuals reveals gene–environment interplay for education in the UK Biobank. Psychological science. 2020;31(5):582–91. pmid:32302253
- View Article
- PubMed/NCBI
- Google Scholar
23. Domingue BW, Fletcher J. Separating measured genetic and environmental effects: Evidence linking parental genotype and adopted child outcomes. Behavior genetics. 2020;50(5):301–9. pmid:32350631
- View Article
- PubMed/NCBI
- Google Scholar
24. Young AI, Benonisdottir S, Przeworski M, Kong A. Deconstructing the sources of genotype-phenotype associations in humans. Science. 2019;365(6460):1396–400. pmid:31604265
- View Article
- PubMed/NCBI
- Google Scholar
25. Hwang L-D, Tubbs JD, Luong J, Lundberg M, Moen G-H, Wang G, et al. Estimating indirect parental genetic effects on offspring phenotypes using virtual parental genotypes derived from sibling and half sibling pairs. PLoS genetics. 2020;16(10):e1009154. pmid:33104719
- View Article
- PubMed/NCBI
- Google Scholar
26. Wu Y, Zhong X, Lin Y, Zhao Z, Chen J, Zheng B, et al. Estimating genetic nurture with summary statistics of multigenerational genome-wide association studies. Proc Natl Acad Sci U S A. 2021;118(25). pmid:34131076; PubMed Central PMCID: PMC8237646.
- View Article
- PubMed/NCBI
- Google Scholar
27. Warrington NM, Freathy RM, Neale MC, Evans DM. Using structural equation modelling to jointly estimate maternal and fetal effects on birthweight in the UK Biobank. International Journal of Epidemiology. 2018;47(4):1229–41. pmid:29447406
- View Article
- PubMed/NCBI
- Google Scholar
28. Warrington NM, Hwang L-D, Nivard MG, Evans DM. Estimating direct and indirect genetic effects on offspring phenotypes using genome-wide summary results data. Nature Communications. 2021;12(1):5420. pmid:34521848
- View Article
- PubMed/NCBI
- Google Scholar
29. Moen G-H, Nivard M, Bhatta L, Warrington NM, Willer C, Åsvold BO, et al. Using Genomic Structural Equation Modeling to Partition the Genetic Covariance Between Birthweight and Cardiometabolic Risk Factors into Maternal and Offspring Components in the Norwegian HUNT Study. Behavior Genetics. 2022. pmid:36322199
- View Article
- PubMed/NCBI
- Google Scholar
30. Qiao Z, Zheng J, Helgeland Ø, Vaudel M, Johansson S, Njølstad PR, et al. Introducing M-GCTA a Software Package to Estimate Maternal (or Paternal) Genetic Effects on Offspring Phenotypes. Behavior Genetics. 2020;50(1):51–66. pmid:31493278
- View Article
- PubMed/NCBI
- Google Scholar
31. Eaves LJ, Pourcain BS, Smith GD, York TP, Evans DM. Resolving the effects of maternal and offspring genotype on dyadic outcomes in genome wide complex trait analysis ("M-GCTA"). Behav Genet. 2014;44(5):445–55. Epub 20140725. pmid:25060210; PubMed Central PMCID: PMC4174369.
- View Article
- PubMed/NCBI
- Google Scholar
32. Eilertsen EM, Jami ES, McAdams TA, Hannigan LJ, Havdahl AS, Magnus P, et al. Direct and Indirect Effects of Maternal, Paternal, and Offspring Genotypes: Trio-GCTA. Behavior Genetics. 2021;51(2):154–61. pmid:33387132
- View Article
- PubMed/NCBI
- Google Scholar
33. Young AI, Frigge ML, Gudbjartsson DF, Thorleifsson G, Bjornsdottir G, Sulem P, et al. Relatedness disequilibrium regression estimates heritability without environmental bias. Nature Genetics. 2018;50(9):1304–10. pmid:30104764
- View Article
- PubMed/NCBI
- Google Scholar
34. Hill WD, Arslan RC, Xia C, Luciano M, Amador C, Navarro P, et al. Genomic analysis of family data reveals additional genetic effects on intelligence and personality. Molecular Psychiatry. 2018;23(12):2347–62. pmid:29321673
- View Article
- PubMed/NCBI
- Google Scholar
35. Tubbs JD, Hwang L-D, Luong J, Evans DM, Sham PC. Modeling Parent-Specific Genetic Nurture in Families with Missing Parental Genotypes: Application to Birthweight and BMI. Behavior Genetics. 2021;51(3):289–300. pmid:33454873
- View Article
- PubMed/NCBI
- Google Scholar
36. Young AI, Nehzati SM, Benonisdottir S, Okbay A, Jayashankar H, Lee C, et al. Mendelian imputation of parental genotypes improves estimates of direct genetic effects. Nature Genetics. 2022;54(6):897–905. pmid:35681053
- View Article
- PubMed/NCBI
- Google Scholar
37. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. 2018;562(7726):203–9. Bycroft2018.
- View Article
- Google Scholar
38. Hill WD, Davies NM, Ritchie SJ, Skene NG, Bryois J, Bell S, et al. Genome-wide analysis identifies molecular systems and 149 genetic loci associated with income. Nature Communications. 2019;10(1):5741. pmid:31844048
- View Article
- PubMed/NCBI
- Google Scholar
39. Branigan AR, McCallum KJ, Freese J. Variation in the Heritability of Educational Attainment: An International Meta-Analysis. Social Forces. 2013;92(1):109–40.
- View Article
- Google Scholar
40. Warrington NM, Beaumont RN, Horikoshi M, Day FR, Helgeland Ø, Laurin C, et al. Maternal and fetal genetic effects on birth weight and their relevance to cardio-metabolic risk factors. Nature Genetics. 2019;51(5):804–14. pmid:31043758
- View Article
- PubMed/NCBI
- Google Scholar
41. Nivard M, Belsky D, Harden KP, Baier T, Ystrom E, Lyngstad TH. Neither nature nor nurture: Using extended pedigree data to elucidate the origins of indirect genetic effects on offspring educational outcomes. 2022.
- View Article
- Google Scholar
42. Evans LM, Tahmasbi R, Vrieze SI, Abecasis GR, Das S, Gazal S, et al. Comparison of methods that use whole genome data to estimate the heritability and genetic architecture of complex traits. Nat Genet. 2018;50(5):737–45. Epub 20180426. pmid:29700474; PubMed Central PMCID: PMC5934350.
- View Article
- PubMed/NCBI
- Google Scholar
43. Border R, Athanasiadis G, Buil A, Schork AJ, Cai N, Young AI, et al. Cross-trait assortative mating is widespread and inflates genetic correlation estimates. Science. 2022;378(6621):754–61. pmid:36395242
- View Article
- PubMed/NCBI
- Google Scholar
44. Border R, O’Rourke S, de Candia T, Goddard ME, Visscher PM, Yengo L, et al. Assortative mating biases marker-based heritability estimators. Nature Communications. 2022;13(1):660. pmid:35115518
- View Article
- PubMed/NCBI
- Google Scholar
45. Okbay A, Wu Y, Wang N, Jayashankar H, Bennett M, Nehzati SM, et al. Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nature Genetics. 2022;54(4):437–49. pmid:35361970
- View Article
- PubMed/NCBI
- Google Scholar
46. Sinnott-Armstrong N, Naqvi S, Rivas M, Pritchard JK. GWAS of three molecular traits highlights core genes and pathways alongside a highly polygenic background. eLife. 2021;10:e58615. pmid:33587031
- View Article
- PubMed/NCBI
- Google Scholar
47. Yang J, Benyamin B, McEvoy BP, Gordon S, Henders AK, Nyholt DR, et al. Common SNPs explain a large proportion of the heritability for human height. Nature Genetics. 2010;42(7):565–9. pmid:20562875
- View Article
- PubMed/NCBI
- Google Scholar
48. Efron B. The jackknife, the bootstrap, and other resampling plans: Philadelphia, Pa.: Society for Industrial and Applied Mathematics, 1982.; 1982.
49. Ahn M, Zhang HH, Lu W. MOMENT-BASED METHOD FOR RANDOM EFFECTS SELECTION IN LINEAR MIXED MODELS. Stat Sin. 2012;22(4):1539–62. pmid:23105913; PubMed Central PMCID: PMC3480741.
- View Article
- PubMed/NCBI
- Google Scholar
50. Manichaikul A, Mychaleckyj JC, Rich SS, Daly K, Sale M, Chen W-M. Robust relationship inference in genome-wide association studies. Bioinformatics. 2010;26(22):2867–73. pmid:20926424
- View Article
- PubMed/NCBI
- Google Scholar
51. Lee JJ, Wedow R, Okbay A, Kong E, Maghzian O, Zacher M, et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nature genetics. 2018;50(8):1112–21. pmid:30038396.
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. van Rheenen W, Peyrot WJ, Schork AJ, Lee SH, Wray NR. Genetic correlations of polygenic disease traits: from theory to practice. Nature Reviews Genetics. 2019;20(10):567–81. pmid:31171865
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Zhang Y, Cheng Y, Jiang W, Ye Y, Lu Q, Zhao H. Comparison of methods for estimating genetic correlation between complex traits using GWAS summary statistics. Briefings in bioinformatics. 2021;22(5):bbaa442. pmid:33497438
View Article
PubMed/NCBI
Google Scholar

[6] View Article

[7] PubMed/NCBI

[8] Google Scholar

[ref3] 3. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76–82. Epub 2010/12/17. pmid:21167468.
View Article
PubMed/NCBI
Google Scholar

[10] View Article

[11] PubMed/NCBI

[12] Google Scholar

[ref4] 4. Loh P-R, Bhatia G, Gusev A, Finucane HK, Bulik-Sullivan BK, Pollack SJ, et al. Contrasting genetic architectures of schizophrenia and other complex diseases using fast variance-components analysis. Nature Genetics. 2015;47(12):1385–92. Epub 2015/11/02. pmid:26523775.
View Article
PubMed/NCBI
Google Scholar

[14] View Article

[15] PubMed/NCBI

[16] Google Scholar

[ref5] 5. Lu Q, Li B, Ou D, Erlendsdottir M, Powles RL, Jiang T, et al. A Powerful Approach to Estimating Annotation-Stratified Genetic Covariance via GWAS Summary Statistics. Am J Hum Genet. 2017;101(6):939–64. pmid:29220677.
View Article
PubMed/NCBI
Google Scholar

[18] View Article

[19] PubMed/NCBI

[20] Google Scholar

[ref6] 6. Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh P-R, et al. An atlas of genetic correlations across human diseases and traits. Nature Genetics. 2015;47(11):1236–41. pmid:26414676
View Article
PubMed/NCBI
Google Scholar

[22] View Article

[23] PubMed/NCBI

[24] Google Scholar

[ref7] 7. Zhang Y, Cheng Y, Ye Y, Jiang W, Lu Q, Zhao H. Estimating genetic correlation jointly using individual-level and summary-level GWAS data. bioRxiv. 2021:2021.08.18.456908.
View Article
Google Scholar

[26] View Article

[27] Google Scholar

[ref8] 8. Guo H, Li JJ, Lu Q, Hou L. Detecting local genetic correlations with scan statistics. Nature Communications. 2021;12(1):2033. pmid:33795679
View Article
PubMed/NCBI
Google Scholar

[29] View Article

[30] PubMed/NCBI

[31] Google Scholar

[ref9] 9. Werme J, van der Sluis S, Posthuma D, de Leeuw CA. LAVA: An integrated framework for local genetic correlation analysis. bioRxiv. 2021:2020.12.31.424652.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref10] 10. Shi H, Mancuso N, Spendlove S, Pasaniuc B. Local genetic correlation gives insights into the shared genetic architecture of complex traits. The American Journal of Human Genetics. 2017;101(5):737–51. pmid:29100087
View Article
PubMed/NCBI
Google Scholar

[36] View Article

[37] PubMed/NCBI

[38] Google Scholar

[ref11] 11. Zhang Y, Lu Q, Ye Y, Huang K, Liu W, Wu Y, et al. SUPERGNOVA: local genetic correlation analysis reveals heterogeneous etiologic sharing of complex traits. Genome Biology. 2021;22(1):262. pmid:34493297
View Article
PubMed/NCBI
Google Scholar

[40] View Article

[41] PubMed/NCBI

[42] Google Scholar

[ref12] 12. Brown BC, Ye CJ, Price AL, Zaitlen N. Transethnic Genetic-Correlation Estimates from Summary Statistics. Am J Hum Genet. 2016;99(1):76–88. Epub 20160616. pmid:27321947; PubMed Central PMCID: PMC5005434.
View Article
PubMed/NCBI
Google Scholar

[44] View Article

[45] PubMed/NCBI

[46] Google Scholar

[ref13] 13. Miao J, Guo H, Song G, Zhao Z, Hou L, Lu Q. Quantifying portable genetic effects and improving cross-ancestry genetic prediction with GWAS summary statistics. bioRxiv. 2022:2022.05.26.493528.
View Article
Google Scholar

[48] View Article

[49] Google Scholar

[ref14] 14. Guo H, Hou L, Shi Y, Jin SC, Zeng X, Li B, et al. Quantifying concordant genetic effects of de novo mutations on multiple disorders. eLife. 2022;11:e75551. pmid:35666111
View Article
PubMed/NCBI
Google Scholar

[51] View Article

[52] PubMed/NCBI

[53] Google Scholar

[ref15] 15. Kong A, Thorleifsson G, Frigge ML, Vilhjalmsson BJ, Young AI, Thorgeirsson TE, et al. The nature of nurture: Effects of parental genotypes. Science. 2018;359(6374):424–8. pmid:29371463
View Article
PubMed/NCBI
Google Scholar

[55] View Article

[56] PubMed/NCBI

[57] Google Scholar

[ref16] 16. Wang B, Baldwin JR, Schoeler T, Cheesman R, Barkhuizen W, Dudbridge F, et al. Robust genetic nurture effects on education: A systematic review and meta-analysis based on 38,654 families across 8 cohorts. The American Journal of Human Genetics. 2021;108(9):1780–91. pmid:34416156
View Article
PubMed/NCBI
Google Scholar

[59] View Article

[60] PubMed/NCBI

[61] Google Scholar

[ref17] 17. Lee SH, Yang J, Goddard ME, Visscher PM, Wray NR. Estimation of pleiotropy between complex diseases using single-nucleotide polymorphism-derived genomic relationships and restricted maximum likelihood. Bioinformatics. 2012;28(19):2540–2. pmid:22843982
View Article
PubMed/NCBI
Google Scholar

[63] View Article

[64] PubMed/NCBI

[65] Google Scholar

[ref18] 18. Bates TC, Maher BS, Medland SE, McAloney K, Wright MJ, Hansell NK, et al. The Nature of Nurture: Using a Virtual-Parent Design to Test Parenting Effects on Children’s Educational Attainment in Genotyped Families. Twin Res Hum Genet. 2018;21(2):73–83. Epub 20180313. pmid:29530109.
View Article
PubMed/NCBI
Google Scholar

[67] View Article

[68] PubMed/NCBI

[69] Google Scholar

[ref19] 19. Trejo S, Domingue BW. Genetic nature or genetic nurture? Introducing social genetic parameters to quantify bias in polygenic score analyses. Biodemography Soc Biol. 2018;64(3–4):187–215. pmid:31852332.
View Article
PubMed/NCBI
Google Scholar

[71] View Article

[72] PubMed/NCBI

[73] Google Scholar

[ref20] 20. Willoughby EA, McGue M, Iacono WG, Rustichini A, Lee JJ. The role of parental genotype in predicting offspring years of education: Evidence for genetic nurture. Molecular psychiatry. 2021;26(8):3896–904. pmid:31444472
View Article
PubMed/NCBI
Google Scholar

[75] View Article

[76] PubMed/NCBI

[77] Google Scholar

[ref21] 21. de Zeeuw EL, Hottenga J-J, Ouwens KG, Dolan CV, Ehli EA, Davies GE, et al. Intergenerational transmission of education and ADHD: effects of parental genotypes. Behavior Genetics. 2020;50(4):221–32. pmid:32026073
View Article
PubMed/NCBI
Google Scholar

[79] View Article

[80] PubMed/NCBI

[81] Google Scholar

[ref22] 22. Cheesman R, Hunjan A, Coleman JR, Ahmadzadeh Y, Plomin R, McAdams TA, et al. Comparison of adopted and nonadopted individuals reveals gene–environment interplay for education in the UK Biobank. Psychological science. 2020;31(5):582–91. pmid:32302253
View Article
PubMed/NCBI
Google Scholar

[83] View Article

[84] PubMed/NCBI

[85] Google Scholar

[ref23] 23. Domingue BW, Fletcher J. Separating measured genetic and environmental effects: Evidence linking parental genotype and adopted child outcomes. Behavior genetics. 2020;50(5):301–9. pmid:32350631
View Article
PubMed/NCBI
Google Scholar

[87] View Article

[88] PubMed/NCBI

[89] Google Scholar

[ref24] 24. Young AI, Benonisdottir S, Przeworski M, Kong A. Deconstructing the sources of genotype-phenotype associations in humans. Science. 2019;365(6460):1396–400. pmid:31604265
View Article
PubMed/NCBI
Google Scholar

[91] View Article

[92] PubMed/NCBI

[93] Google Scholar

[ref25] 25. Hwang L-D, Tubbs JD, Luong J, Lundberg M, Moen G-H, Wang G, et al. Estimating indirect parental genetic effects on offspring phenotypes using virtual parental genotypes derived from sibling and half sibling pairs. PLoS genetics. 2020;16(10):e1009154. pmid:33104719
View Article
PubMed/NCBI
Google Scholar

[95] View Article

[96] PubMed/NCBI

[97] Google Scholar

[ref26] 26. Wu Y, Zhong X, Lin Y, Zhao Z, Chen J, Zheng B, et al. Estimating genetic nurture with summary statistics of multigenerational genome-wide association studies. Proc Natl Acad Sci U S A. 2021;118(25). pmid:34131076; PubMed Central PMCID: PMC8237646.
View Article
PubMed/NCBI
Google Scholar

[99] View Article

[100] PubMed/NCBI

[101] Google Scholar

[ref27] 27. Warrington NM, Freathy RM, Neale MC, Evans DM. Using structural equation modelling to jointly estimate maternal and fetal effects on birthweight in the UK Biobank. International Journal of Epidemiology. 2018;47(4):1229–41. pmid:29447406
View Article
PubMed/NCBI
Google Scholar

[103] View Article

[104] PubMed/NCBI

[105] Google Scholar

[ref28] 28. Warrington NM, Hwang L-D, Nivard MG, Evans DM. Estimating direct and indirect genetic effects on offspring phenotypes using genome-wide summary results data. Nature Communications. 2021;12(1):5420. pmid:34521848
View Article
PubMed/NCBI
Google Scholar

[107] View Article

[108] PubMed/NCBI

[109] Google Scholar

[ref29] 29. Moen G-H, Nivard M, Bhatta L, Warrington NM, Willer C, Åsvold BO, et al. Using Genomic Structural Equation Modeling to Partition the Genetic Covariance Between Birthweight and Cardiometabolic Risk Factors into Maternal and Offspring Components in the Norwegian HUNT Study. Behavior Genetics. 2022. pmid:36322199
View Article
PubMed/NCBI
Google Scholar

[111] View Article

[112] PubMed/NCBI

[113] Google Scholar

[ref30] 30. Qiao Z, Zheng J, Helgeland Ø, Vaudel M, Johansson S, Njølstad PR, et al. Introducing M-GCTA a Software Package to Estimate Maternal (or Paternal) Genetic Effects on Offspring Phenotypes. Behavior Genetics. 2020;50(1):51–66. pmid:31493278
View Article
PubMed/NCBI
Google Scholar

[115] View Article

[116] PubMed/NCBI

[117] Google Scholar

[ref31] 31. Eaves LJ, Pourcain BS, Smith GD, York TP, Evans DM. Resolving the effects of maternal and offspring genotype on dyadic outcomes in genome wide complex trait analysis ("M-GCTA"). Behav Genet. 2014;44(5):445–55. Epub 20140725. pmid:25060210; PubMed Central PMCID: PMC4174369.
View Article
PubMed/NCBI
Google Scholar

[119] View Article

[120] PubMed/NCBI

[121] Google Scholar

[ref32] 32. Eilertsen EM, Jami ES, McAdams TA, Hannigan LJ, Havdahl AS, Magnus P, et al. Direct and Indirect Effects of Maternal, Paternal, and Offspring Genotypes: Trio-GCTA. Behavior Genetics. 2021;51(2):154–61. pmid:33387132
View Article
PubMed/NCBI
Google Scholar

[123] View Article

[124] PubMed/NCBI

[125] Google Scholar

[ref33] 33. Young AI, Frigge ML, Gudbjartsson DF, Thorleifsson G, Bjornsdottir G, Sulem P, et al. Relatedness disequilibrium regression estimates heritability without environmental bias. Nature Genetics. 2018;50(9):1304–10. pmid:30104764
View Article
PubMed/NCBI
Google Scholar

[127] View Article

[128] PubMed/NCBI

[129] Google Scholar

[ref34] 34. Hill WD, Arslan RC, Xia C, Luciano M, Amador C, Navarro P, et al. Genomic analysis of family data reveals additional genetic effects on intelligence and personality. Molecular Psychiatry. 2018;23(12):2347–62. pmid:29321673
View Article
PubMed/NCBI
Google Scholar

[131] View Article

[132] PubMed/NCBI

[133] Google Scholar

[ref35] 35. Tubbs JD, Hwang L-D, Luong J, Evans DM, Sham PC. Modeling Parent-Specific Genetic Nurture in Families with Missing Parental Genotypes: Application to Birthweight and BMI. Behavior Genetics. 2021;51(3):289–300. pmid:33454873
View Article
PubMed/NCBI
Google Scholar

[135] View Article

[136] PubMed/NCBI

[137] Google Scholar

[ref36] 36. Young AI, Nehzati SM, Benonisdottir S, Okbay A, Jayashankar H, Lee C, et al. Mendelian imputation of parental genotypes improves estimates of direct genetic effects. Nature Genetics. 2022;54(6):897–905. pmid:35681053
View Article
PubMed/NCBI
Google Scholar

[139] View Article

[140] PubMed/NCBI

[141] Google Scholar

[ref37] 37. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. 2018;562(7726):203–9. Bycroft2018.
View Article
Google Scholar

[143] View Article

[144] Google Scholar

[ref38] 38. Hill WD, Davies NM, Ritchie SJ, Skene NG, Bryois J, Bell S, et al. Genome-wide analysis identifies molecular systems and 149 genetic loci associated with income. Nature Communications. 2019;10(1):5741. pmid:31844048
View Article
PubMed/NCBI
Google Scholar

[146] View Article

[147] PubMed/NCBI

[148] Google Scholar

[ref39] 39. Branigan AR, McCallum KJ, Freese J. Variation in the Heritability of Educational Attainment: An International Meta-Analysis. Social Forces. 2013;92(1):109–40.
View Article
Google Scholar

[150] View Article

[151] Google Scholar

[ref40] 40. Warrington NM, Beaumont RN, Horikoshi M, Day FR, Helgeland Ø, Laurin C, et al. Maternal and fetal genetic effects on birth weight and their relevance to cardio-metabolic risk factors. Nature Genetics. 2019;51(5):804–14. pmid:31043758
View Article
PubMed/NCBI
Google Scholar

[153] View Article

[154] PubMed/NCBI

[155] Google Scholar

[ref41] 41. Nivard M, Belsky D, Harden KP, Baier T, Ystrom E, Lyngstad TH. Neither nature nor nurture: Using extended pedigree data to elucidate the origins of indirect genetic effects on offspring educational outcomes. 2022.
View Article
Google Scholar

[157] View Article

[158] Google Scholar

[ref42] 42. Evans LM, Tahmasbi R, Vrieze SI, Abecasis GR, Das S, Gazal S, et al. Comparison of methods that use whole genome data to estimate the heritability and genetic architecture of complex traits. Nat Genet. 2018;50(5):737–45. Epub 20180426. pmid:29700474; PubMed Central PMCID: PMC5934350.
View Article
PubMed/NCBI
Google Scholar

[160] View Article

[161] PubMed/NCBI

[162] Google Scholar

[ref43] 43. Border R, Athanasiadis G, Buil A, Schork AJ, Cai N, Young AI, et al. Cross-trait assortative mating is widespread and inflates genetic correlation estimates. Science. 2022;378(6621):754–61. pmid:36395242
View Article
PubMed/NCBI
Google Scholar

[164] View Article

[165] PubMed/NCBI

[166] Google Scholar

[ref44] 44. Border R, O’Rourke S, de Candia T, Goddard ME, Visscher PM, Yengo L, et al. Assortative mating biases marker-based heritability estimators. Nature Communications. 2022;13(1):660. pmid:35115518
View Article
PubMed/NCBI
Google Scholar

[168] View Article

[169] PubMed/NCBI

[170] Google Scholar

[ref45] 45. Okbay A, Wu Y, Wang N, Jayashankar H, Bennett M, Nehzati SM, et al. Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nature Genetics. 2022;54(4):437–49. pmid:35361970
View Article
PubMed/NCBI
Google Scholar

[172] View Article

[173] PubMed/NCBI

[174] Google Scholar

[ref46] 46. Sinnott-Armstrong N, Naqvi S, Rivas M, Pritchard JK. GWAS of three molecular traits highlights core genes and pathways alongside a highly polygenic background. eLife. 2021;10:e58615. pmid:33587031
View Article
PubMed/NCBI
Google Scholar

[176] View Article

[177] PubMed/NCBI

[178] Google Scholar

[ref47] 47. Yang J, Benyamin B, McEvoy BP, Gordon S, Henders AK, Nyholt DR, et al. Common SNPs explain a large proportion of the heritability for human height. Nature Genetics. 2010;42(7):565–9. pmid:20562875
View Article
PubMed/NCBI
Google Scholar

[180] View Article

[181] PubMed/NCBI

[182] Google Scholar

[ref48] 48. Efron B. The jackknife, the bootstrap, and other resampling plans: Philadelphia, Pa.: Society for Industrial and Applied Mathematics, 1982.; 1982.

[ref49] 49. Ahn M, Zhang HH, Lu W. MOMENT-BASED METHOD FOR RANDOM EFFECTS SELECTION IN LINEAR MIXED MODELS. Stat Sin. 2012;22(4):1539–62. pmid:23105913; PubMed Central PMCID: PMC3480741.
View Article
PubMed/NCBI
Google Scholar

[185] View Article

[186] PubMed/NCBI

[187] Google Scholar

[ref50] 50. Manichaikul A, Mychaleckyj JC, Rich SS, Daly K, Sale M, Chen W-M. Robust relationship inference in genome-wide association studies. Bioinformatics. 2010;26(22):2867–73. pmid:20926424
View Article
PubMed/NCBI
Google Scholar

[189] View Article

[190] PubMed/NCBI

[191] Google Scholar

[ref51] 51. Lee JJ, Wedow R, Okbay A, Kong E, Maghzian O, Zacher M, et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nature genetics. 2018;50(8):1112–21. pmid:30038396.
View Article
PubMed/NCBI
Google Scholar

[193] View Article

[194] PubMed/NCBI

[195] Google Scholar

Figures

Abstract

Author summary

Introduction

Results

Method overview

Simulation results

Partitioning heritability by direct/indirect effect paths for five complex traits

Partitioning genetic covariance among five complex traits

Discussion

Methods

PARSEC model details

Parameter estimation

Reconstructing heritability and genetic covariance using decomposed variance components

Data processing

Simulation settings

Supporting information

S1 Fig. Parameter estimates in simulations.

S1 Table. Parameter settings for simulations.

S2 Table. Parameter estimates for single traits analysis.

S3 Table. Parameter estimates for trait pair analysis.

S4 Table. Sample relatedness matrices for cov(Y).

Acknowledgments

References