Authors: Mannan Luo (1Department of Psychiatry, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands), Victória Trindade Pons (1Department of Psychiatry, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands), Michael Zakharin (2Virginia Institute for Psychiatric and Behavioral Genetics, Virginia Commonwealth University, Richmond, VA, USA), Jean-Baptiste Pingault (3Department of Clinical, Educational and Health Psychology, Division of Psychology and Language Sciences, University College London, London, United Kingdom; 4Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom), Nathan A. Gillespie (2Virginia Institute for Psychiatric and Behavioral Genetics, Virginia Commonwealth University, Richmond, VA, USA), Hanna M. van Loo (1Department of Psychiatry, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands)
Categories: Article
Source: medRxiv
Authors: Mannan Luo, Victória Trindade Pons, Michael Zakharin, Jean-Baptiste Pingault, Nathan A. Gillespie, Hanna M. van Loo
Substance use disorders run in families, yet the mechanisms underlying intergenerational transmission remain unclear. We investigated indirect genetic effects, pathways through which parental genotypes influence offspring phenotypes via the family environment, for alcohol use disorder (AUD), nicotine dependence (ND), and related quantitative outcomes, and aimed to identify family environmental factors through which such effects may operate. Using transmitted and non-transmitted polygenic scores (PGS) constructed for problematic alcohol use, tobacco use disorder, and general addiction liability, we analyzed 5972 European-ancestry adult offspring with at least one genotyped parent from the population-based Lifelines cohort (Netherlands). Offspring outcomes included lifetime DSM-5 AUD diagnosis, AUD symptom count, maximum drinks in 24 hours, Fagerström Test for Nicotine Dependence score, and cigarettes per day. AUD findings were meta-analyzed with data from the Brisbane Longitudinal Twin Study (N = 1368; Australia). We also examined parent-of-origin effects and mediation by parental substance use and socioeconomic status using structural equation modeling. Transmitted PGS robustly predicted all AUD and ND outcomes (β = 0.07–0.16; OR = 1.20 for AUD diagnosis). Non-transmitted PGS, indexing indirect genetic effects, were negligible for all clinical syndrome outcomes. The only significant indirect genetic effect was on cigarettes per day (β = 0.03, p = 0.01), mediated by parental smoking behavior but not socioeconomic status. These findings indicate that intergenerational transmission of risk for AUD and ND is driven primarily by direct genetic effects, with modest indirect genetic effects on smoking quantity. Larger samples and cross-trait analyses are needed to further elucidate these mechanisms.
Substance use disorders (SUDs), including alcohol use disorder (AUD) and nicotine dependence (ND), run in families [1–3]. Offspring of parents with SUDs have a nearly threefold increased risk of developing the same disorder [4, 5]. However, understanding how this risk transmits across generations remains challenging because parents provide both genes and rearing environments [2]. Many putatively environmental risk factors, such as parental substance use or socioeconomic status, are themselves genetically influenced [6]. This potentially induces gene-environment correlation, making it difficult to disentangle the extent to which intergenerational transmission operates through inherited DNA, the rearing environment, or their interplay [7].
One pathway through which gene-environment correlation arises is indirect genetic effects, whereby parental genotypes influence offspring outcomes through the environments parents provide [8, 9]. Although often described as genetic nurture (e.g., parenting processes within the family), indirect genetic effects can also shape offspring outcomes through broader pathways that do not directly involve within-family nurturing [10]. For example, parental genetic liability to AUD may contribute to heavier drinking that normalizes alcohol use in the home [11], or to functional impairment that reduces socioeconomic status and affects neighborhood selection [12], both of which could increase offspring AUD risk independently of parentally transmitted alleles. The transmitted/non-transmitted polygenic score (PGS) design leverages parent-offspring genetic data to separate these pathways by partitioning parental alleles into (i) transmitted PGS (PGST), capturing the combined influence of direct and indirect genetic effects, and (ii) non-transmitted PGS (PGSNT), indexing indirect genetic effects independently of direct transmission [8, 9].
Previous studies, including our own, have detected modest indirect genetic effects on quantitative substance use outcomes such as cigarettes per day [13–15]. However, it remains unclear whether these effects generalize to clinically defined SUDs in the general population, specifically AUD and ND, two of the most prevalent and burdensome SUDs worldwide [16, 17]. To date, only one study (N = 4,806) has reported indirect genetic effects on AUD, in a high-risk sample enriched for familial alcohol problems [18]. It is also unknown whether such effects differ by parent-of-origin in clinical outcomes, as aggregating maternal and paternal PGS could obscure parent-specific direct [19] and indirect genetic effects [8]. Furthermore, parental substance use has been identified as a plausible mediating pathway for indirect genetic effects on offspring non-clinical outcomes such as smoking [13, 14], yet the mechanisms underlying such effects on clinical SUDs remain poorly understood. In particular, it is unclear whether they operate through parental substance use [20, 21] or socioeconomic disadvantage [22, 23], a transdiagnostic risk factor for SUDs.
To address these gaps, we constructed PGST and PGSNT to investigate direct and indirect genetic effects on offspring AUD, ND, and related quantitative phenotypes. Primary analyses were conducted in the population-based Lifelines cohort (the Netherlands). Given that indirect genetic effects are typically small [13, 14, 24, 25], we additionally meta-analyzed Lifelines with the Brisbane Longitudinal Twin Study (BLTS, Australia) for AUD outcomes, the only phenotype with comparable diagnostic assessments across cohorts, to increase power and assess cross-population generalizability. This study aimed (i) examine whether parental genotypes show indirect genetic effects (indexed by PGSNT) on AUD and ND, using both substance-specific PGS and a general addiction liability PGS; (ii) evaluate whether maternal and paternal indirect genetic effects differ in magnitude; and (iii) elucidate whether these effects are mediated by parental substance use, socioeconomic status (parental education and household income), or both.
The study protocol was preregistered with the Open Science Framework (https://osf.io/xu9w7/); deviations are described in the Supplementary Methods.
Primary analyses were conducted in the Lifelines cohort. Lifelines is a multi-disciplinary prospective population-based cohort study examining in a unique three-generation design, the health and health-related behaviors of 167729 persons living in the North of the Netherlands. It employs a broad range of investigative procedures in assessing the biomedical, socio-demographic, behavioral, physical and psychological factors which contribute to the health and disease of the general population, with a special focus on multi-morbidity and complex genetics. Data collection was accomplished across three assessments the baseline (2007–2013), wave 2 (2014–2017), and wave 3 (2019–2023). Detailed information on the study design and sample characteristics has been described elsewhere [26, 27]. For the present study, we included 19233 genotyped adult offspring with at least one genotyped parent, comprising 15966 parent-offspring pairs (father-child or mother-child) and 3267 complete trios (mother-father-offspring). Analytic sample sizes varied by outcome availability, ranging from 2083 to 5972.
For the AUD meta-analysis, we additionally included 1368 adult offspring with at least one genotyped parent from the BLTS [28], yielding a combined meta-analytic sample of 6229 offspring (Lifelines: n = 4861; BLTS: n = 1368). All other analyses, including ND-related outcomes, parent-of-origin effects, and mediation analyses, were conducted in Lifelines only due to limited data availability in BLTS.
In Lifelines, lifetime AUD was assessed at wave 3 (2019–2023) using DSM-5 criteria among ever-drinkers who reported retrospectively on their period of heaviest use. Lifetime AUD diagnosis was defined as endorsement of at least two DSM-5 symptoms [29]. To handle missing items while preserving diagnostic certainty, we applied a conservative PhenX Toolkit algorithm (Supplementary Methods). We also examined two continuous AUD-related outcomes to capture dimensional variation and improve statistical AUD symptom count (AUDsx) and maximum alcoholic drinks consumed in 24 hours (MaxDrinks), a proxy for binge drinking propensity [18]. In BLTS, lifetime AUD was assessed via self-report questionnaire using the same DSM-5 criteria (Supplementary Methods).
Lifetime nicotine dependence was assessed at wave 3 using the Fagerström Test for Nicotine Dependence (FTND) [30] in ever-smokers meeting a lifetime exposure threshold for at least one product (≥100 cigarettes or e-cigarettes, ≥25 cigars, ≥50 cigarillos, or ≥40 pipe sessions). Participants reported retrospectively on their period of heaviest tobacco use (Supplementary Methods). Those not meeting any exposure threshold were not administered the FTND questionnaire. FTND scores range from 0 to 10, with higher scores indicating greater dependence. We included lifetime average cigarettes per day (CPD) assessed at baseline, as a complementary severity indicator that correlates strongly with FTND [31] and has been shown to index genetic liability to dependence [32].
Parental socioeconomic status was indexed by years of education and equivalized household income, assessed via self-report at baseline (2006–2013). Educational attainment was derived from participants’ highest obtained educational degree, converted to the corresponding number of years of education required to complete that degree, following previous research in Lifelines [33]. Monthly household income was equivalized by dividing the midpoint of each income category by the square root of household size, thereby accounting for differences in household composition. For interpretability, income was rescaled by dividing by 100, so that estimates reflect differences per 100-euro change in equivalized monthly household income at baseline.
Details on genotyping, quality control, imputation, inference of transmitted and non-transmitted alleles, and PGS construction are provided in the Supplementary Methods. Briefly, 79988 Lifelines participants were genotyped across three batches and imputed using the Haplotype Reference Consortium v1 panel. Analyses were restricted to participants of European ancestry to minimize population stratification.
We used Haplotype-based Inference of Non-Transmitted Alleles (HINTA) [33], a validated method that overcomes a key limitation of standard transmitted/non-transmitted PGS the requirement for complete parent-offspring trios. By inferring non-transmitted alleles from haplotype comparisons and recombination patterns, HINTA can reliably reconstruct parental non-transmitted alleles even when only one parent is genotyped, achieving 99.8% concordance with standard trio-based pseudo-control phasing approach [34]. This enables accurate distinction of transmitted and non-transmitted alleles in both parent-offspring duos and complete trios, increasing sample size, statistical power, and the generalizability of indirect genetic effect estimates.
Transmitted (PGST) and non-transmitted (PGSNT) polygenic scores were constructed using summary statistics from the largest independent genome-wide association studies (GWAS) of problematic alcohol use (PAU) [35], tobacco use disorder (TUD) [36], and a general addiction factor for substance use disorders (SUDs) derived from a multivariate GWAS of problematic alcohol use, problematic tobacco use, cannabis use disorder, and opioid use disorder [37]. Variant weights were estimated using LDpred2-auto [38] and applied in PLINK 1.9 [39] to construct PGSs based on the transmitted and non-transmitted allele datasets.
To maximize statistical power for estimating overall effects, maternal and paternal transmitted and non-transmitted PGS were summed into combined parental scores. The missing parental PGSNT values among parent-offspring pairs were imputed using the mean of observed parents, following previous work [8, 33]. Parent-of-origin and mediation analyses used separate non-imputed maternal and paternal scores (PGST_m, PGSNT_m, PGST_f, PGSNT_f), with missingness handled by full-information maximum likelihood (FIML) in lavaan [40]. All PGSs were residualized on the first ten genetic principal components and standardized within each genotyping array to control for population stratification and batch effects.
All analyses were conducted in R version 4.2.1 [41].
We used mixed-effects regression models to examine associations between parental PGST and PGSNT with five offspring outcomes, including AUD diagnosis, AUDsx, MaxDrinks, FTND sum score, and CPD. Continuous outcomes were analyzed using mixed-effects linear regression with lmerTest package [42], and dichotomous outcomes using mixed-effects logistic regression with GLMMadaptive [43]. Models were specified Outcomei=β1PGSTi+β2PGSNTi+β3Sexi+β4Agei+1∣FamilyID+ei. Family ID was included as a random effect (intercept) to account for sibling relatedness, with offspring sex and age as covariates. Consistent with Kong et al. [8], the direct genetic effect (βDGE) was estimated as the difference between the transmitted and non-transmitted PGS effects (i.e., βT−βNT). Because PGST captures both direct and indirect genetic effects, subtracting βNT isolates the direct genetic transmission component. To correct for multiple testing across the five correlated outcomes, the effective number of independent tests was estimated using Galwey’s method (Meff=4) with poolr [44], and a Šidák-corrected significance threshold of p < .013 was applied to balance family-wise error rate control with statistical power [45].
Given the modest sample size for clinical AUD in Lifelines and small effect sizes typically expected for indirect genetic effects, we meta-analyzed transmitted and non-transmitted PGS effects on AUD across Lifelines and BLTS. Regression models were fitted separately in each cohort using identical model specifications, as described above. Cohort-specific effect estimates and standard errors were combined using fixed-effects inverse-variance-weighted meta-analysis with metafor [46].
Unlike the mixed-effects regressions above, which combined maternal and paternal PGS to maximize power, we next fitted structural equation models (SEMs) in lavaan [47] to estimate all four PGS components simultaneously (PGST_m, PGSNT_m, PGST_f, PGSNT_f). This enabled formal testing of differences between maternal and paternal effects (both transmitted and non-transmitted) using likelihood ratio tests (lavTestLRT), comparing models in which maternal and paternal paths were constrained to equality against unconstrained models. Models were estimated using FIML, incorporating all available data without listwise deletion or prior imputation [40]. Offspring sex and age were included as covariates, and family ID was specified as a clustering variable with robust standard errors to account for sibling relatedness. All four PGS components were modeled simultaneously with freely estimated covariances, partially accounting for correlations between maternal and paternal PGS induced by assortative mating (non-random mate selection based on phenotypic similarity), and thereby reducing upward bias in indirect genetic effect estimates, although residual bias from multigenerational assortative mating cannot be fully excluded [48, 49].
To identify which environmental pathways mediated indirect genetic effects, we conducted SEM-based mediation analyses in lavaan, restricted to outcomes with significant indirect genetic effects. Parental substance use, parental education, and household income were examined as candidate mediators. Each mediator was initially evaluated in a separate single-mediator model to assess its contribution independently. Significant mediators were then entered into a joint parallel multiple-mediator model to estimate their unique contributions after mutual adjustment. Maternal and paternal pathways were modeled simultaneously to enable parent-specific comparisons, and assortative mating was partially accommodated as described above. Confidence intervals for mediation effects and maternal-paternal difference parameters were obtained using Monte Carlo simulation with semTools (2,000 replications) [50]. Statistical significance was inferred when the 95% confidence intervals excluded zero.
In Lifelines, the analytic sample comprised 5972 adult offspring (mean age = 42.5 years, SD = 10.4), of whom 64.1% were female. Among ever-smokers meeting a lifetime exposure threshold for at least one tobacco product (n = 2083), the mean FTND score was 2.11 (SD = 2.16), reflecting nicotine dependence severity during their period of heaviest use. Among all ever-smokers, mean lifetime average CPD was 10.24 (SD = 6.00). Among ever-drinkers, the mean AUD symptom count was 0.88 (SD = 1.31), and 20.6% met criteria for lifetime AUD diagnosis (≥2 symptoms). Sample characteristics did not differ substantially between parent-offspring pairs and complete trios (Table 1). Bivariate correlations among PGSs and offspring outcomes are reported in the Supplementary Tables 1 and 2.
We examined associations of transmitted (PGST) and non-transmitted (PGSNT) polygenic scores with offspring AUD- and ND-related outcomes, including clinical outcomes (lifetime AUD diagnosis and AUD symptom count) and quantitative measures (MaxDrinks, FTND sum score, and CPD) (Table 2).
Transmitted PGS for both substance-specific (PGS t_PAU, PGST_TUD) and general addiction PGS (PGST_SUD) showed robust associations with corresponding outcomes (βT = .061–.161 for continuous traits; OR = 1.113 – 1.196 for AUD diagnosis; all p<.001). In contrast, indirect genetic effects, indexed by PGSNT, were largely absent. Of all associations tested, only one survived multiple-testing parental PGSNT_TUD was modestly associated with offspring CPD (βNT = .033, p=.012). No other significant indirect genetic effects were observed for AUD diagnosis, AUDsx, MaxDrinks, or FTND sum score.
Direct genetic effects were quantified as βDGE = βT − βNT, following Kong et al. [8]. These effects were most pronounced for FTND sum score (βDGE = .130) and CPD (βDGE = .094) using PGSTUD. The sole significant indirect genetic effect, parental PGSNT_TUD on offspring CPD, was approximately 35% of the corresponding direct genetic effect.
To enhance statistical power and assess cross-population generalizability, PGST and PGSNT effects on AUD were meta-analyzed across Lifelines and BLTS (total N = 6229; Supplementary Table 3 and Figure 1). Indirect genetic effects remained nonsignificant for both PGSNT_PAU and PGSNT_SUD, consistent with Lifelines findings.
To examine whether maternal and paternal indirect genetic effects differed, transmitted and non-transmitted PGS were decomposed by parent-of-origin (Table 3). Both maternal and paternal PGST significantly predicted all offspring outcomes, with comparable effect sizes across parents (e.g., PGS T_TUD on FTND: βmaternal =.104; βpaternal = .119). For indirect genetic effects, only paternal PGSNT_TUD showed a nominal association with offspring CPD (β = .046, p = .017), which did not survive correction for multiple testing. No parent-specific indirect genetic effects were observed for ND or AUD outcomes using PGSNT_SUD. Likelihood ratio tests indicated no significant maternal–paternal differences in either transmitted or non-transmitted PGS effects for any outcome (all p > .25).
Given that significant indirect genetic effects were observed only for CPD, we examined whether these effects operated through parental smoking, parental education, and household income (Supplementary Table 4). In single-mediator models, parental smoking emerged as the primary mediator, with maternal PGSNT_TUD showing stronger mediation via maternal smoking (β = .014, 95% MC CI [.007, .021]) and paternal PGSNT_TUD via paternal smoking (β = .007 [.002, .011]). Significant mediation was also observed for transmitted PGST_TUD via maternal smoking (β = .022 [.015, .030]) and paternal smoking (β = .004 [.001, .009]). Maternal education showed weaker evidence of mediation for PGSNT_TUD (β = .004 [.001, .007]) but no significant mediation via paternal education. Household income did not mediate effects for either parent.
To assess whether maternal education represented an independent mechanism or was confounded by parental smoking, we fitted a parallel multiple-mediator model including both mediators simultaneously (Figure 1). Parental smoking estimates were unchanged, whereas mediation via parental education was attenuated to non-significance after adjustment for smoking. The transmitted mediation effects (PGST_TUD) via maternal smoking were significantly larger than the corresponding paternal effects (βdiff = .018 [.009, .027]); no other maternal-paternal differences were significant.
This study provides the first comprehensive population-based examination of indirect genetic effects on AUD and ND. Using transmitted and non-transmitted polygenic scores in 5972 parent-offspring pairs and trios, supplemented by a meta-analysis of AUD across cohorts (total N = 6229), we found that intergenerational transmission of AUD and ND occurs primarily through direct genetic effects. Indirect genetic effects were largely absent, with the exception of offspring smoking quantity (CPD). Mediation analyses suggested parental smoking as a plausible environmentally-mediated pathway, with minimal contribution from parental socioeconomic factors.
Our null findings for indirect genetic effects on AUD contrast with prior evidence from Thomas et al. [18], who reported such effects on AUD in a high-risk sample enriched for familial alcohol problems. This discrepancy may reflect methodological differences. Thomas and colleagues did not test the total effect of non-transmitted parental PGS on offspring AUD, the standard parameter for indirect genetic effects [8, 9], but instead examined specific mediated pathways (e.g., parental relationship discord). Such mediated effects are difficult to interpret without a significant overall PGSNT effect, as individual pathways can reach significance even when total indirect genetic effects does not [51]. Moreover, these mediator-outcome associations remain observational and susceptible to genetic confounding, so apparent mediation may arise spuriously in the absence of true indirect genetic effects [52]. Finally, their high-risk sample likely captured more extreme environmental adversity, thereby increasing sensitivity to detect indirect genetic effects relative to population-based cohorts.
We observed divergent findings across nicotine indirect genetic effects were detectable for lifetime average smoking quantity (CPD) but not for nicotine dependence (FTND). One explanation may be limited statistical power, as the FTND was assessed only at Wave 3 among a subset of ever-smokers meeting stricter inclusion criteria, reducing power to detect small indirect genetic effects relative to the broader CPD sample. Alternatively, this pattern may suggest that CPD and FTND capture distinct dimensions of nicotine-related risk. CPD quantifies cumulative smoking behavior across the lifespan and may be more sensitive to environmental influences such as parental behavioral modeling and household smoking norms [53, 54], whereas FTND captures physiological dependence during the period of heaviest use and may more strongly reflect neurobiological mechanisms and direct genetic liability [32]. Larger studies with more comprehensive smoking phenotypes will be needed to determine whether indirect genetic effects differentially influence behavioral versus physiological dimensions of nicotine dependence
A key contribution of this study is the dissection of intergenerational pathways using joint mediation models that allow parental phenotypes to mediate both transmitted and non-transmitted genetic effects on offspring. Although parental genetic liability for SUDs has been proposed to reduce socioeconomic status and thereby increase offspring substance use risk [13, 55], we found that the indirect genetic effect on offspring CPD operated primarily through parental smoking quantity, while the contribution of parental education was fully attenuated when parental smoking was included in the same model. This is consistent with social learning theory, which posits that transmission may occur through behavioral modeling, whereby children observe and imitate parental behaviors [56]. These findings should be interpreted cautiously, as the pathway from parental phenotype to offspring outcome remains observational and may be confounded by unmeasured environmental factors and residual genetic liability not captured by the PGS [52]. They therefore represent an exploration of plausible environmentally-mediated pathways rather than definitive evidence of causal mechanisms.
This study has several strengths, including its large population-based design, the separation of transmitted and non-transmitted alleles in genotyped parent-offspring pairs and trios, the inclusion of both clinical syndromes and dimensional traits, and the cross-cohort meta-analysis for AUD. Furthermore, SEM-based mediation modeling enabled simultaneous evaluation of multiple environmental pathways while partially accounting for assortative mating. However, several limitations should be noted. First, statistical power was constrained by modest sample sizes for certain outcomes, particularly FTND, and by limited predictive performance of current SUD PGSs [57]. Transmitted PGS effects were modest (β = .07−.16), suggesting that these scores capture only a fraction of heritable variance. Together, these factors reduced our ability to detect more nuanced indirect genetic pathways, including parent-of-origin patterns and socioeconomic mediation. Second, parental household income was reported when offspring were already adults, limiting its validity as a proxy for financial resources during critical developmental periods. Third, in addition to genuine indirect genetic effects, PGSNT may capture residual confounding from population stratification and assortative mating [52, 58]. Although we adjusted for population structure using principal components and partially accounted for assortative mating within the SEM, residual bias may remain. Finally, analyses were restricted to individuals of European ancestry, limiting generalizability to diverse populations.
In this large-scale, population-based study, the intergenerational transmission of risk for AUD and ND appears to be driven primarily by direct genetic effects, with minimal evidence for indirect genetic effects. The exception was a modest effect on offspring smoking quantity, operating through parental smoking rather than broader socioeconomic pathways. However, the absence of detectable indirect genetic effects does not preclude their role in SUDs, as such effects may be subtle and require greater statistical power to detect. Future research combining larger samples, more refined SUD measures, and increasingly predictive polygenic scores will be essential to detect and characterize indirect genetic effects in intergenerational transmission of SUDs.