Preliminary Evidence for Genetic Nurture in Depression and Neuroticism Through Polygenic Scores

Key Points

Question

Is there evidence for parental genetic nurture in the risk of depression and neuroticism?

Findings

In this cross-sectional study of 38 702 offspring, results from polygenic score (PGS) modeling provide limited preliminary evidence of genetic nurture in depression and neuroticism. Parental PGSs for depression were significantly associated with offspring neuroticism, with a regression estimate two-thirds that of the offspring’s own PGS, whereas results suggest that previous associations between cannabis use PGS and depression may be noticeably biased by parental genetic nurture.

Meaning

These findings suggest that genetic nurture may bias results from both epidemiologic and genetic studies on depression, but proper modeling should be used to identify potential avenues for future prevention and intervention efforts.

This cross-sectional study used linear mixed modeling of polygenic scores to estimate the association of genetic nurture across a range of parental traits with offspring lifetime depression and neuroticism.

Abstract

Importance

Modeling genetic nurture (ie, the effects of parental genotypes through influences on the environment experienced by their children) is essential to accurately disentangle genetic and environmental influences on phenotypic variance. However, these influences are often ignored in both epidemiologic and genetic studies of depression.

Objective

To estimate the association of genetic nurture with depression and neuroticism.

Design, Setting, and Participants

This cross-sectional study jointly modeled parental and offspring polygenic scores (PGSs) across 9 traits to test for the association of genetic nurture with lifetime broad depression and neuroticism using data from nuclear families in the UK Biobank, with data collected between 2006 and 2019. A broad depression phenotype was measured in 38 702 offspring from 20 905 independent nuclear families, with most of these participants also reporting neuroticism scores. Parental genotypes were imputed from sibships or parent-offspring duos and used to calculate parental PGSs. Data were analyzed between March 2021 and January 2023.

Main Outcomes and Measures

Estimates of genetic nurture and direct genetic regression coefficients on broad depression and neuroticism.

Results

This study of 38 702 offspring with data on broad depression (mean [SD] age, 55.5 [8.2] years at study entry; 58% female) found limited preliminary evidence for a statistically significant association of genetic nurture with lifetime depression and neuroticism in adults. The estimated regression coefficient of the parental depression PGS on offspring neuroticism (β = 0.04, SE = 0.02, P = 6.63 × 10⁻³) was estimated to be approximately two-thirds (66%) that of the offspring’s depression PGS (β = 0.06, SE = 0.01, P = 6.13 × 10⁻¹¹). Evidence for an association between parental cannabis use disorder PGS and offspring depression was also found (β = 0.08, SE = 0.03, P = .02), which was estimated to be 2 times greater than the association between the offspring’s cannabis use disorder PGS and their own depression status (β = 0.04, SE = 0.02, P = .07).

Conclusions and Relevance

The results of this cross-sectional study highlight the potential for genetic nurture to bias results from epidemiologic and genetic studies on depression or neuroticism and, with further replication and larger samples, identify potential avenues for future prevention and intervention efforts.

Introduction

Clinical depression, characterized by core symptoms of low mood and anhedonia, is a major public health burden, affecting approximately 290 million people globally in 2019 and accounting for approximately 2% of disability-adjusted life-years lost across all causes.¹ Lifetime depression is moderately familial, with a meta-analysis² of twin studies estimating additive genetic factors to account for 34% of phenotypic variance and shared environmental influences accounting for approximately 10%. The influence of additive genetic factors appears to increase throughout development, whereas common environmental effects wane.^2,3

An important limitation of the classic approach of separating genetic from environmental factors is the phenomenon of genetic nurture, which refers to the effects of parental genotypes on offspring traits over and above those of direct genetic inheritance, mediated through the parentally provided rearing environment. Offspring may inherit genetic risk factors for depression from their parents, whereas heritable traits in parents, such as personality, mental health, or parenting styles, could also impact the child’s overall depression risk. Genetic nurture refers to the heritable component of parental traits influencing an offspring’s phenotype.

One major motivation for modeling genetic nurture is that such effects may indicate potential avenues for family-based interventions. Another motivation is that genetic nurture may bias estimates of direct genetic effects.^4,5 For instance, genetic nurture effects can mask direct single-nucleotide polymorphism (SNP) effects acting in the opposite direction or bias effect estimates even when no direct effect exists. Such misidentification of SNP effects may interfere with the elucidation of biological mechanisms by genetic studies. Biased SNP effect estimates may also affect the estimation of the overall contribution of common SNPs to phenotypic variance (SNP heritability).

Genetic nurture can be conceptualized as a form of passive gene-environment correlation in intact (ie, nonadoptive) families. This gene-environment correlation can upwardly bias traditional twin estimates of shared environmental contributions.^6,7,8 Thus, the evidence for shared environmental influences on depression suggests a potential role for genetic nurture. Indeed, strong epidemiologic evidence supports an influence of the parentally provided environment on a child’s risk of depression, including heritable factors in the parental generation, such as maternal or paternal depression.^9,10,11 However, epidemiologic studies typically attribute parent-offspring correlations to familial environmental influences, ignoring the possibility that these correlations arise from shared genetic influences between parental traits and offspring depression.

Genetically informed designs are necessary for untangling genetic from familial environmental influences, including genetic nurture. These designs include adoption, assisted conception, children of twins, and matched sibling study designs, which are able to estimate unconfounded effects of genetic and environmental factors.¹² A recent systematic review examined evidence from genetically informed studies for several mental health–related traits.¹² Across multiple studies, evidence supports a role for both genetic and environmental transmission of depression from parents to offspring across the lifespan.¹² Because depression is itself heritable, some of the environmentally mediated effects may be attributable to parental genotypes. Overall, evidence also supports an environmental, but not shared genetic, pathway from a range of parenting behaviors to broad internalizing symptoms, which are closely related to depression, anxiety, and other mood disorders.¹² To the extent that these parenting behaviors are heritable, their effects can contribute to genetic nurture. Only 2 studies have thus far directly examined genetic nurture using molecular genetics.^13,14 One study estimated parental genetic nurture to explain up to 14% of the variance in offspring depression symptoms at 8 years of age,¹⁴ whereas the other found no evidence for maternal-specific genetic nurture.¹³ Preliminary evidence^13,14 supports a role for genetic nurture in depression. However, studies^12,13,14 to date have examined only a narrow range of parental traits and focused on childhood depression symptoms. In this study, we apply recently developed methods for genotype imputation of missing parents and linear mixed modeling of polygenic scores (PGSs) to estimate genetic nurture across a range of parental traits in offspring lifetime depression and neuroticism in adulthood using data from nuclear families in the UK Biobank (UKB).

Methods

Sample Description

The UKB is a population-based sample of approximately 500 000 individuals with detailed phenotypic information and biological measures, including genotype data.¹⁵ The UKB received ethical approval from the National Health Service National Research Ethics Service North West. The current study was approved by the UKB Data Access Committee. Written informed consent was obtained by UKB researchers from all participants. Given its size, the UKB has inadvertently sampled a number of closely related individuals, with 30% of sampled individuals being related (third degree or closer) to another person in the UKB.¹⁶ Although such close relatedness is often undesired in population genomic studies designed for unrelated individuals, such as genome-wide association studies (GWASs), they are useful for disentangling direct and genetic nurture influences. Because genetic nurture effects are mediated through the family environment, we excluded any individuals who ever reported being adopted as a child, leaving 38 702 individuals in the current study. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Phenotype Definitions

Our main offspring outcome of interest was broad depression as used by previous GWASs in the UKB,^17,18 which includes patients formally diagnosed with depression and individuals who have reported seeking help from a physician for nerves, anxiety, tension, or depression. This wider conception of depression has shown strong genetic correlations with more restrictive definitions in the UKB and appears to be more powerful in testing for genetic associations.¹⁸ The procedure for identifying broad depression case-control status is detailed elsewhere.¹⁸ After filtering for genotype data quality and availability of covariates, 37 590 offspring from 20 900 families were available for analysis.

As a secondary offspring outcome of interest, we also considered a continuous measure of neuroticism, which may increase statistical power. Neuroticism shares many core features with depression and is strongly associated with future depression risk. It is highly genetically correlated with depression and shows evidence for a causal role in depression.^17,19,20,21 Neuroticism was measured as the sum of 12 items, such as “Are you a worrier?” and “Do you often feel lonely?” with a complete description elsewhere.²² After filtering for genotype data quality and availability of covariates, 30 719 offspring from 19 273 families were available for PGS modeling. Table 1 gives the number of available offspring from each family type with data on broad depression and neuroticism.

Table 1. Offspring Sample Size by Family Type for Broad Depression and Neuroticism.

Family type	Broad depression		Neuroticism
	Cases	Controls
Nuclear family types, No.
No parental genotype data (sibships)	12 379	21 949	27 945
Genotype data on father only	270	449	603
Genotype data on mother only	1072	1688	2331
Genotype data on both parents	308	587	754
Total No.	14 029	24 673	31 633
Female	9531	12 907	18 114
Male	4498	11 766	13 519
Year of birth, mean (SD)	1953 (8)	1952 (8)	1953 (8)

Genotyping, Imputation, and Polygenic Scoring

A detailed description of the array genotyping and quality control for UKB participants is given elsewhere.¹⁶ We identified nuclear families of White British ancestry with available array genotyping data as described previously.²³ Only White British individuals were included in the study because of small sample size and limited statistical power to fit these models in participants of other ancestries. Briefly, average identity by descent sharing output by the KING software programs²⁴ along with information on year of birth and sex were used to identify nuclear family units. For some families, 1 or both parents are unobserved but can be partially recovered from information on their relatives. Thus, for sibling pairs and parent-offspring duos, we imputed the additive genotype for missing parents at all genotyped loci using 2 methods (IMPISH^23,25 and snipar²⁶) on unphased array genotyping data. For siblings, only the expectation of the sum of parental additive genotypes can be imputed, whereas for parent-offspring duos, we obtained the missing parent’s expected additive genotype value. For all individuals, including offspring and parents (observed and imputed), we constructed PGSs for 9 traits as the sum of genotypic dosages (observed or imputed) weighted by their posterior effect size estimated by the PRS-CS-auto method²⁷ with the default parameter settings. Finally, PGSs for each trait were mean centered and standardized relative to the genotyped individuals such that the genotyped PGS variance equals 1.

The 9 traits for which PGSs were calculated were broad depression, neuroticism, alcohol dependence,²⁸ attention-deficit/hyperactivity disorder,²⁹ bipolar disorder,³⁰ cannabis use disorder,³¹ educational attainment, schizophrenia,³² and smoking (cigarettes per day).³³ Details of their sample size and SNP heritability are provided in eTable 1 in Supplement 1. For broad depression, neuroticism, and educational attainment, the largest available GWAS included samples from the UKB, which partially overlapped with our target sample. To avoid overfitting, we performed our own GWAS in the UKB after excluding our target sample of related individuals and meta-analyzed these results with available non-UKB GWAS summary statistics^34,35,36 using the METAL software programs.³⁷ Summary statistics for other traits were chosen so that their sample did not include UKB participants.

Statistical Analysis

Data were collected between 2006 and 2019, and analyses were conducted between March 2021 and January 2023. For the broad depression and neuroticism outcomes, we fit (generalized) linear mixed models independently for each of the 9 PGSs using the lme4 package in R, version 4.1.0 (R Foundation for Statistical Computing).³⁸ All models were fit with a random-effect term to account for any residual variance shared among siblings. Across all models, the offspring outcome was modeled as a function of basic demographic covariates, including year of birth, sex, genotyping array, and the first 40 genetic principal components, with all continuous variables standardized. To control for potential bias due to population stratification or geographic gene-environment correlation,^39,40 we also controlled for the offspring’s reported longitude and latitude of birth in all PGS analyses. In addition to these basic covariates, we also jointly modeled the parental and offspring PGSs, which were our main independent variables of interest. Because only the mean (imputed) parental PGS was available for families with no observed parental data, models were fit separately for nuclear families with no parental genotypes measured and families in which at least 1 parent was genotyped. Thus, for models using data from families in which no parent was observed, the fitted model uses the imputed parental PGS and estimates the sum of parental coefficients. When fit using data where at least 1 parent has been observed, the fitted model includes separate terms for the maternal and paternal PGS coefficients. In summary, the following 2 models were fit separately for each PGS of interest:

Model 1: No parental genotypes observed:

Y_ij = C_i + PGSc_i + PGSp_i + s_j

Model 2: One or both parental genotypes observed:

Y_ij = C_i + PGSc_i + PGSm_i + PGSf_i + s_j

where Y is the phenotype of interest measured in individual i from family j, C are the covariates, PGS_c is the child’s PGS, PGS_m is the mother’s PGS, PGS_f is the father’s PGS, PGS_p is the sum of PGS_m and PGS_f, and s is the random effect shared among siblings within family j.

Regression coefficient estimates and P values for offspring (c), parental (p), maternal (m), and paternal (f) terms were combined across the 2 data types by applying the inverse-covariance weighted combination procedure as described by Tubbs et al²³ to the model results fit on the 2 data types. For significance testing, we considered both a nominal P < .05 and a Bonferroni-corrected P threshold of 0.05/9 = 5.6 × 10⁻³ to account for the 9 PGSs considered. Variance explained (R²) by the offspring, maternal, and paternal PGS were calculated as the square of the respective estimated coefficients (ie, c², m², and f², respectively). Similarly, the shared offspring-maternal R² and offspring-paternal R² were calculated as c²m² and c²f², respectively. Before calculating these R² components, coefficients from models of broad depression were transformed from the observed scale to the liability scale using a linear approximation,⁴¹ assuming a population prevalence of 35% as estimated in the entire population-based UKB sample.¹⁸

Results

This study of 38 702 offspring with data on broad depression (mean [SD] age, 55.5 [8.2] years at study entry; 58% female and 42% male) found limited preliminary evidence of a statistically significant association of genetic nurture with lifetime depression and neuroticism in adults. The association among broad depression PGSs calculated for trios is shown in Figure 1. As expected, the variance of PGS for parents whose genotypes have been imputed is lower than those for genotyped individuals. The variance of PGS_p imputed from siblings using IMPISH (0.34) was lower than that calculated by snipar (0.36), with a ratio significantly different from 0 indicated by an F test (P = 4.3 × 10⁻⁵). This finding suggests that snipar imputation recovered more information on the parental genotypes, which is expected given that it incorporates identity by descent information during imputation. Additionally, the covariance between broad depression PGSs of first-degree relatives are all close to the expected value of 0.5, whereas the correlation between observed and imputed pairs is inflated because of the reduced variance of imputed PGSs compared with observed PGSs. We performed several tests for both same-trait and cross-trait assortative mating (eMethods in Supplement 2 and eTable 2 in Supplement 1). We find some suggestive evidence that both same-trait and cross-trait assortative mating may be acting on depression and neuroticism, with an expected parental PGS correlation of 0.02, which was corrected for in subsequent regression analyses. As shown in Figure 1C, there is a slight correlation between maternal and paternal PGS for broad depression when one has been observed and one imputed, which is expected given the reduced variance of imputed PGSs and the dependency induced by conditioning on the observed offspring and parent during imputation.

Figure 1. Correlations Among Family Member Polygenic Scores (PGSs).

Each plot shows the correlation (R) and its 95% CI. Trio indicates families in which offspring and both parents have been genotyped. PGS_c indicates child’s PGS; PGS_f, father’s PGS; PGS_m, mother’s PGS; PGS_S1 and PGS_S2, sibling 1 and 2 from a sibling pair. Lines indicate the estimated regression of the y-axis variable on the x-axis variable. Shaded density plots show the distribution of the y variables and x variables by data type.

The estimated regression coefficients for PGS_c, PGS_m, PGS_f, and PGS_p on offspring broad depression and neuroticism are shown in Figure 2 and detailed in Table 2. In the main text, we present results from models that used snipar to impute missing parental genotypes, given its improved imputation routine providing greater power. However, as indicated in eTable 3 in Supplement 1, point estimates from models using parental genotypes imputed using IMPISH are similar.

Figure 2. Estimated Polygenic Score (PGS) Regression Coefficients.

Dots indicate the estimated regression coefficient of a given PGS, and bars represent its 95% CI. Vertical dotted line indicates 0 on the x-axis. Estimates are from mixed models in which offspring and parental PGS are jointly modeled with other covariates. ADHD indicates attention-deficit/hyperactivity disorder; BPD, bipolar disorder; OR, odds ratio; PGS_c, child’s PGS; PGS_f, father’s PGS; PGS_m, mother’s PGS; and PGS_p, sum of the parental PGS coefficients.

^aEstimates that are nominally significant with a P < .05.

^bEstimates that are significant after Bonferroni correction to account for the 9 PGSs considered, with a P < 5.6⁻³ (0.05/9).

Table 2. Estimated Regression Coefficients From Joint Modeling of PGSs^a.

Outcome by PGS	PGSc		PGSm		PGSf		PGSp
	β (SE)	P value	β (SE)	P value	β (SE)	P value	β (SE)	P value
Broad depression
Broad depression	0.17 (0.02)	1.13 × 10−16	0.04 (0.03)	1.96 × 10−1	0.01 (0.03)	7.18 × 10−1	0.05 (0.03)	1.16 × 10−1
Neuroticism	0.12 (0.02)	5.20 × 10 − 10	0.05 (0.03)	1.24 × 10−1	0.02 (0.03)	6.44 × 10−1	0.06 (0.03)	1.11 × 10−1
Schizophrenia	0.09 (0.02)	1.64 × 10−5	0 (0.03)	9.52 × 10−1	0.01 (0.03)	8.08 × 10−1	0.01 (0.03)	7.59 × 10−1
Bipolar disorder	0.1 (0.02)	1.62 × 10−6	0.02 (0.03)	5.97 × 10−1	−0.02 (0.03)	6.62 × 10−1	0 (0.03)	9.65 × 10−1
Cannabis use disorder	0.04 (0.02)	6.84 × 10−2	−0.01 (0.03)	7.90 × 10−1	0.09 (0.03)	1.30 × 10−2	0.08 (0.03)	2.11 × 10−1
Smoking	−0.02 (0.02)	2.58 × 10−1	0.01 (0.03)	8.31 × 10−1	0.01 (0.03)	6.55 × 10−1	0.02 (0.03)	5.12 × 10−1
Educational attainment	−0.04 (0.02)	3.41 × 10−2	0.04 (0.03)	2.16 × 10−1	−0.05 (0.04)	1.95 × 10−1	−0.01 (0.03)	8.73 × 10−1
ADHD	0.03 (0.02)	1.10 × 10−1	0.03 (0.03)	3.59 × 10−1	−0.01 (0.03)	7.70 × 10−1	0.02 (0.03)	5.76 × 10−1
Alcohol dependence	0.03 (0.02)	1.50 × 10−1	0.02 (0.03)	5.92 × 10−1	0 (0.03)	9.30 × 10−1	0.02 (0.03)	5.54 × 10−1
Neuroticism
Neuroticism	0.13 (0.01)	1.22 × 10−41	−0.01 (0.02)	4.04 × 10−1	0.03 (0.02)	1.19 × 10−1	0.01 (0.02)	4.02 × 10−1
Broad depression	0.06 (0.01)	6.13 × 10−11	0 (0.02)	8.42 × 10−1	0.05 (0.02)	5.37 × 10−3	0.04 (0.02)	6.63 × 10−3
Schizophrenia	0.02 (0.01)	1.37 × 10−2	−0.01 (0.02)	7.50 × 10−1	0.02 (0.02)	2.09 × 10−1	0.02 (0.02)	3.13 × 10−1
Bipolar disorder	0.03 (0.01)	5.72 × 10−3	0 (0.02)	9.34 × 10−1	−0.01 (0.02)	6.54 × 10−1	−0.01 (0.02)	6.92 × 10−1
Cannabis use disorder	0.01 (0.01)	4.39 × 10−1	−0.01 (0.02)	5.00 × 10−1	0.03 (0.02)	6.67 × 10−2	0.02 (0.02)	2.07 × 10−1
Smoking	−0.01 (0.01)	1.23 × 10−1	0.02 (0.02)	3.07 × 10−1	0 (0.02)	8.84 × 10−1	0.02 (0.02)	2.57 × 10−1
Educational attainment	−0.03 (0.01)	2.78 × 10−3	−0.03 (0.02)	6.96 × 10−2	0 (0.02)	9.23 × 10−1	−0.03 (0.02)	5.64 × 10−2
ADHD	0.02 (0.01)	4.29 × 10−2	0.02 (0.02)	2.30 × 10−1	−0.02 (0.02)	3.10 × 10−1	0 (0.02)	9.29 × 10−1
Alcohol dependence	0.02 (0.01)	6.63 × 10−2	0.01 (0.02)	4.64 × 10−1	−0.01 (0.02)	7.26 × 10−1	0.01 (0.02)	7.38 × 10−1

Abbreviations: ADHD, attention-deficit/hyperactivity disorder; PGS, polygenic score; PGS_c, child’s polygenic score; PGS_f, father’s polygenic score; PGS_m, mother’s polygenic score; PGS_p, sum of the PGS_m and PGS_f.

^a All the PGSs were jointly fitted using a mixed model to control for shared variance among siblings from the same family. Covariates include year of birth, sex, genotyping array, the first 40 genetic principal components, longitude of birthplace, and latitude of birthplace.

The regression coefficient of the broad depression PGS_c was estimated to be 0.17 (SE, 0.02; P = 1.13 × 10⁻¹⁶), whereas the PGS_p coefficient was 0.05 (SE, 0.03; P = .12) and not statistically significant. The PGS_c for bipolar disorder, neuroticism, and schizophrenia surpassed the Bonferroni-corrected P value threshold for an association with increased odds of offspring depression. The PGS_f, but not PGS_m, for cannabis use disorder (β = 0.09, SE = 0.03, P = .01) was also associated with greater risk of depression in offspring at a nominal P value threshold.

The neuroticism PGS_c was associated with offspring neuroticism (β = 0.13, SE = 0.01, P = 1.22 × 10⁻⁴¹), whereas the neuroticism PGS_p was not (β = 0.01, SE = 0.02, P = .40). However, the broad depression PGS_p was significantly associated with offspring neuroticism (β = 0.04, SE = 0.02, P = 6.63 × 10⁻³), largely driven by the broad depression PGS_f, which surpassed the Bonferroni-corrected P value threshold (β = 0.05, SE = 0.02, P = 5.37 × 10⁻³). Meanwhile, the regression coefficient of the offspring’s own broad depression PGS_c (β = 0.06, SE = 0.01, P = 6.13 × 10⁻¹¹) was approximately one-third higher than that of the PGS_p. Additionally, as found in a previous study,²⁶ the PGS_c for educational attainment (β = −0.03, SE = 0.01, P = 2.78 × 10⁻³) was significantly associated with offspring neuroticism at a Bonferroni-corrected significance level.

The variance explained in offspring broad depression and neuroticism by child, maternal, and paternal PGS, along with the shared variance explained by correlated child-mother and child-father factors, is shown in Figure 3 and detailed in eTable 4 in Supplement 1. Together, these broad depression PGSs explained approximately 1.1% of the variance in the offspring’s liability to broad depression. The offspring’s own PGS_c for broad depression explained the largest proportion of variance, accounting for 1.0%. The PGS_m explained approximately 0.06%, with the PGS_f explaining merely 0.01%. The remaining contributors, comprising shared offspring-parental influences, together explained less than 0.001% of variance in offspring depression liability. Apart from the neuroticism PGS, in which the total contribution to offspring depression variance was approximately 0.7%, the total variance explained by other PGSs were each less than 0.4%. Notably, the variance explained in offspring depression by the cannabis use disorder PGS_c (0.05%) was lower than the variance explained by the cannabis use disorder PGS_f (0.3%).

Figure 3. Variance Explained by Offspring and Parental Polygenic Score (PGS).

This figure shows the estimated variance explained in offspring broad depression and neuroticism by child’s PGS (PGS_c), mother’s PGS (PGS_m), and father’s PGS (PGS_f) along with the shared variance explained by correlated child-mother PGS (PGS_cm) and child-father PGS (PGS_cf) influences for 9 traits. Vertical lines indicate 95% CIs. ADHD indicates attention-deficit/hyperactivity disorder; BPD, bipolar disorder.

Discussion

Using data from nuclear families in the UKB, we estimated the relative contributions of genetic transmission and genetic nurture to depression and neuroticism through joint modeling of offspring and parental PGS across a number of traits. Overall, our results provide limited preliminary evidence that genetic nurture may exist for depression and neuroticism, but the influence is likely to be very small for most traits compared with the influence of genetic variants passed on to a child. Nevertheless, for some traits, we found that genetic nurture may heavily bias estimates of PGS association with depression from common modeling approaches in samples of unrelated individuals. However, given the relatively weak evidence of an association of genetic nurture with depression in this study, these results should be interpreted with caution as we await further replication in larger samples.

The regression coefficient of the parental depression PGS on offspring depression was estimated to be approximately one-third (29%) the size of the offspring’s own PGS regression coefficient, comparable to the ratio observed previously for educational attainment.^4,42 However, the association between parental depression PGS and offspring depression was not statistically significant. For offspring neuroticism, the ratio of parental to offspring depression PGS regression coefficients was larger, estimated at two-thirds (66%). Although the genetic nurture coefficient was statistically significant, the SEs are relatively large, ranging from 0.02 to 0.06, which is between 33% and 100% the size of the offspring’s estimated depression PGS coefficient (0.04). Notably, this finding means that in samples of unrelated individuals, the association between depression PGS and neuroticism may be overestimated by up to 50%. Our models suggested that the observed association between parental depression PGS and offspring neuroticism was largely driven by paternal contributions. Although our study is likely underpowered to reliably distinguish between maternal and paternal sources of influence, this result serves to highlight the importance of including data from fathers, who are not assessed as often as mothers in large studies of developmental health outcomes, so that these influences may be resolved.

Interestingly, a nominally significant association of parental cannabis use disorder PGS on offspring depression was observed, with an estimated regression coefficient twice as large as that of the offspring’s own cannabis use disorder PGS. Consistent evidence supports an epidemiologic association between cannabis use and depression, but the direction of causality and underlying mechanisms are still unclear.⁴³ Although our findings require further validation, they highlight the potential for genetic nurture to bias naive estimates of PGSs in studies of unrelated individuals. Crucially, prior estimates of the association between cannabis use PGS and risk of depression⁴⁴ may be up to twice as large as the unconfounded direct association. This consideration is especially important when attempting to glean biological insights from genetic studies because findings can be biased by genetic nurture, which is an environmental source of offspring phenotypic variation.

Despite finding statistically significant associations between certain parental PGSs with offspring depression and neuroticism, for most traits, the variance explained is still small compared with those of the offspring’s own PGS. For example, although the offspring’s neuroticism PGS accounted for 1.7% of the variance in neuroticism, the variance explained by maternal and paternal genetic nurture together through their neuroticism PGSs accounted for only 0.1%. One reason is that PGSs are not perfect, often capturing only a fraction of the SNP heritability. If the ratio of variance explained by direct genetic influences to variance explained by genetic nurture is correctly estimated and the narrow-sense heritability of neuroticism is assumed to be 0.25,⁶ genetic nurture through parental neuroticism could account for up to 1.5% of the variance in offspring neuroticism.

A previous study using the UKB to decompose direct and genetic nurture associations at the genome-wide SNP level²⁶ found that the genetic correlation between population and direct regression coefficients was not significantly different from unity but that the correlation between direct and genetic nurture coefficients was negative (−0.42, SE = 0.19, P = .03). Although these results do not discount the presence of genetic nurture, they imply that using population-based SNP association estimates as weights for parental PGS may underestimate the influence of genetic nurture on neuroticism in the UKB.

It is important to note that the PGSs used here are imperfect measures of genetic nurture, representing only a fraction of the heritable component of overall parental nurture. These nurturing influences also include nongenetic factors, such as parenting behavior, meaning that the overall impact of nurture may be much higher than the genetic nurture influences we have estimated using PGSs. Indeed, an extended adoption study showed that genes and rearing experience accounted for approximately the same variance in treated major depression.⁴⁵ Moreover, our modeling found that the variance explained by parental genetic nurture from some traits such as cannabis use disorder, educational attainment, and schizophrenia were comparable or even exceeded the variance explained by the offspring’s own PGS. Future research should consider additional offspring outcomes and a wider range of PGSs that may harbor genetic nurture influences. Notably, parental PGSs for a wide range of traits can be used as a proxy of the parentally provided environment, allowing researchers to test for associations with parental exposures even if they have not been directly measured or cannot be obtained.

Limitations

Although our results provide preliminary evidence for genetic nurture influences on depression and neuroticism, they await further replication, given some limitations of the current study. First, PGSs for most traits capture only a fraction of overall genetic contributors. Although this leads to relatively small absolute regression coefficients, the ratio of direct transmission to genetic nurture coefficients is expected to be relatively stable if a trait’s genetic architecture is stable across generations. Second, weights used to construct PGSs come from GWASs on unrelated individuals, meaning that SNP effect sizes themselves may be biased by genetic nurture. Future studies should compare regression of offspring and parent PGS using weights derived from population SNP estimates vs genetic nurture SNP estimates. Third, our study was restricted to individuals of White British ancestry living in the UK, limiting its generalizability. As cross-ancestry GWAS sample sizes increase, future studies should include more ancestrally and geographically diverse participants.

Conclusion

Overall, the results of this cross-sectional study highlight the importance of considering genetic nurture in epidemiologic and genetic studies of depression and suggest previous estimates of association between PGS and depression from unrelated samples are likely to be biased by the presence of genetic nurture. More generally, our approach demonstrates the unique advantages of applying low-cost genotyping to large samples of related individuals, allowing researchers to easily identify sources of parental nurture at the SNP level and to correct for their biasing influence on GWAS estimates from unrelated individuals.

Supplement 1.

eTable 1. GWAS Summary Statistics Used to Calculate PGS

eTable 2. Results of Tests for Assortative Mating

eTable 3. Results from Mixed Models of Depression and Neuroticism

eTable 4. Estimates of Variance Explained by PGS

Supplement 2.

eMethods. Supplementary Methods

Supplement 3.

Data Sharing Statement