Abstract
Background and aims
Alcohol use behaviors (AUBs) manifest in a variety of normative and problematic ways across the life course, all of which are heritable. Twin studies show that genetic influences on AUBs change across development, but this is usually not considered in research identifying and investigating the genes linked to AUBs. Understanding the dynamics of how genes shape AUBs could point to critical periods in which interventions may be most effective and provide insight into the mechanisms behind AUB‐related genes. In this project, we estimated how genetic influences on AUBs unfold across development using longitudinal modelling of polygenic scores (PGSs).
Design
Using results from genome‐wide association studies (GWASs), we created PGSs to index individual‐level genetic risk for multiple AUB‐related dimensions: Consumption, Problems, a temporally variable pattern of drinking associated with a preference for beer (BeerPref) and externalizing behavior (EXT). We created latent growth curve models and tested PGSs as predictors of latent growth factors (intercept, slope, quadratic) underlying trajectories of AUBs.
Setting
PGSs were derived in six longitudinal epidemiological cohorts from the United States, United Kingdom and Finland.
Participants
Participant data were obtained from the longitudinal studies AddHealth, ALSPAC, COGA, FinnTwin12, the older Finnish Twin Cohort and Spit for Science (total n = 19 194). These cohorts included individuals aged 14 to 67, with repeated measures collected over a span of 4 to 36 years.
Measurements
Primary measures included monthly frequency of typical alcohol consumption (CON) and heavy episodic drinking (HED).
Findings
When drinking behaviors were averaged across time, higher Consumption, Problems and EXT PGSs were robustly associated with higher levels of CON and HED (βs ranged from 0.105 to 0.333, P < 3.09E‐04) and higher BeerPref PGSs with higher HED (β = 0.064, P = 3.65E‐05). However, these PGSs were largely not associated with drinking trajectories in the latent growth curve models. In the meta‐analysis, only PGSs for chronic alcohol Problems consistently predicted a steeper slope (increasing trajectory) of CON across time (B = 0.470, P = 4.20E‐06). Other PGSs were associated with latent growth factors in some individual cohorts, but there was a large degree of heterogeneity.
Conclusions
Genetic associations appear to differ not only between alcohol use behaviors, but also across developmental time points and across cohorts, highlighting the need for genetic studies to take such heterogeneity into account. Individual‐level genetic profiles may be useful to point to personalized intervention timelines, particularly for individuals with high genetic risk scores for alcohol problems.
Keywords: AddHealth, ALSPAC, COGA, Finnish twin cohorts, gene‐by‐development interaction, polygenic scores, Spit for Science, trajectories
INTRODUCTION
Alcohol use behaviors (AUBs) encompass a wide variety of normative and problematic aspects of alcohol consumption, including typical drinking frequency and quantity, heavy episodic (binge) drinking, alcohol‐related problems and clinically significant alcohol use disorders (AUDs). These behaviors are heritable [1, 2, 3], with a complex polygenic architecture influenced by many genetic variants primarily of small individual effect [4, 5]. Genes exert their effects through both broad influences on individual predispositions toward impulsive, rewarding behaviors [‘externalizing’ (EXT)], as well as processes specific to AUBs such as alcohol metabolism [2, 6, 7, 8]. By amassing very large samples, genome‐wide association studies (GWASs) have begun to identify the genes that influence both broad EXT and alcohol‐specific risk [4, 5, 9]. However, these studies usually investigate static AUB measures, neglecting how the associated genetic variants may interact with developmental processes.
AUBs are dynamic phenomena that unfold in different ways across the lifespan, with strong age‐related trends in prevalence as well as substantial variability in these trends [10, 11, 12, 13, 14, 15, 16, 17, 18]. Alcohol addiction is similarly dynamic and thought to involve three distinct stages of neurobiological adaptation that correspond to changes in motivations for drinking and compulsivity of alcohol use [19]. Genetic epidemiology studies in twins have established that the heritability of AUBs changes across the lifetime, with the relative importance of genes increasing until a plateau in middle adulthood [20], followed by a slight drop‐off in older adulthood [21]. Some evidence suggests that there are also qualitative changes in which genes affect AUBs at different life stages [15, 22, 23].
Gene identification studies, however, typically focus on a snapshot of behavior at a single time point, such as a composite measure of current consumption quantity, past‐year alcohol problems or lifetime AUD diagnoses [4, 5, 24]. This approach, while aiming to maximize statistical power, risks missing out on differences critical to the etiology of AUBs such as longitudinal variation or changes in the relative importance of broad versus specific risk processes across time. Existing evidence indicates that the genetic influences on AUBs are not homogeneous, but can be separated into distinct factors related to consumption, problems and patterns of AUBs [25, 26, 27], as well as effects that are shared with EXT or are specific to alcohol use [8]. EXT‐related genes are correlated with early initiation of substance use, whereas alcohol‐specific genes are more relevant for increasing alcohol use/problems over time [8, 28]. The precedence of different neurobiological systems at different stages of the addiction cycle [19] also suggests that different sets of genes may be important for initiation, acceleration and/or persistence of alcohol use. Such different sets of genes have been observed for different stages of smoking behavior and nicotine dependence [4] and are likely to exist for other addictive behaviors.
A recent study [25] used a genomic structural equation modeling approach to identify four genetically distinct factors underlying 18 varied AUBs. These included (1) chronic and severe alcohol ‘Problems’; (2) ‘BeerPref’, a decreasing pattern of alcohol use/problems later in life marked by a preference for drinking beer and drinking without meals; (3) overall quantity and frequency of ‘Consumption’ of varied alcoholic beverage types; and (4) ‘AtypicalPref’, consumption of less common alcoholic beverages such as fortified wine and spirits. Notably, a large proportion of novel genomic loci were identified for the BeerPref factor, which indexed measures such as a reported reduction in drinking over the past 10 years as well as—like the Problems factor—clinically significant consequences of alcohol use. This suggests important genetic distinctions between persistent versus transient measures of alcohol use/problems—distinctions that may not be accurately captured by typical static GWAS phenotypes. However, this interpretation is only speculative because of the cross‐sectional nature of the study. It is as yet unknown whether these genetic factors are prospectively predictive of longitudinal changes in AUBs or how their AUB‐specific effects compare to that of broader EXT risk factors.
Understanding the dynamics of how genes and developmental trajectories interact to shape AUBs could point to critical time points in which interventions would be most effective, as well as improving our understanding of the mechanisms driving statistical genetic associations. In this study, we leverage longitudinal samples to investigate the dynamic effects of genes on AUBs across the lifespan, from early adolescence to middle and older adulthood. Using polygenic scores (PGSs), we examine how the aggregate genetic influences representing EXT‐ and AUB‐related processes shape trajectories of alcohol use across multiple environmental contexts.
METHODS
This study was pre‐registered under the Open Science Framework (http://osf.io/qg8sa). After registration, two cohorts (described below) were added to the analysis plan as additional suitable datasets were identified after study registration.
Participants
Data were obtained from six longitudinal studies designed for genetic epidemiology research:
n = 5107 participants from AddHealth, a United States (US) national study of adolescents followed into adulthood [29];
n = 5214 participants from the Avon Longitudinal Study of Parents and Children (ALSPAC), a birth cohort in the United Kingdom (UK) [30, 31, 32];
n = 1955 participants from the prospective subset of the Collaborative Studies on the Genetics of Alcoholism (COGA), a US study of families densely affected with AUDs [33];
n = 1219 participants from Finnish twin cohort (FinnTwin12) and n = 6257 from the older Finnish twin cohort (FTC), two twin birth cohorts in Finland [34, 35]; and
n = 4549 participants from Spit for Science (S4S), a cohort of US university students followed throughout college [36].
Data collection included DNA samples for genotyping in addition to interviews or self‐report surveys of behavior and mental health, collected at repeated intervals. In the current study, we use data from participants with ancestry similar to European reference panels (hereafter referred to as ‘EUR’) to match the genetic ancestry background of the discovery GWAS [9, 25]. Full details on data collection for these cohorts have been published previously and are described in the Supplementary Methods. A glossary of commonly used abbreviations is provided for reference in Appendix 1.
Measures
In each cohort (except FTC), we examined measures of typical frequency of alcohol consumption (CON) and heavy episodic drinking (HED), measured in days per month. Each instance of a repeated measure was recorded along with participants' age at the time of assessment, rounded to the nearest year, with ages ranging from 14 to 38 years old. In FTC, measures included typical frequency of alcohol use (Freq), grams of ethanol consumption per month (gEtOH), a dichotomous indicator of heavy drinking (Heavy) and past year frequency of passing out because of alcohol consumption (PassOut), measured at 18 to 67 years old. Additional details about measure derivations and data cleaning procedures can be found in the Supplementary Methods. A timeline of the assessments for each cohort can be found in Figure 1 and Tables S1–S2.
FIGURE 1.
Mean levels per age of (a) typical consumption (CON) and (b) heavy episodic drinking (HED) frequencies in five longitudinal cohorts and (c) multiple alcohol use behaviors in the older FinnTwin cohort.
PGSs
PGSs were derived based on GWAS summary statistics of three latent genetic factors described above [25]: Problems, BeerPref and Consumption. This discovery GWAS was carried out in 387 105 EUR individuals from the UK Biobank dataset. A fourth factor, AtypicalPref, was excluded based on prior evidence of its weak genetic signal [25, 37]. To examine broad as well as alcohol‐specific influences, an additional PGS was derived to index EXT based on a prior GWAS [9], which also used a genomic structural equation model approach to aggregate seven impulsivity‐related and substance use phenotypes in a total of 1.5 million EUR samples. The choice of GWAS phenotypes for deriving PGSs was based on their relevance to longitudinal patterns of AUBs, as discussed above, as well as to ensure that there was no overlap with our six study cohorts, which have been included in prior large AUB GWASs (e.g. Saunders et al.) [4].
We created these four PGSs in each cohort based on the previously published GWAS summary statistics. Summary statistics were weighted using PRS‐CS (‘auto’ version) and 1000 Genomes EUR reference panels for estimating linkage disequilibrium, as provided with the software [38]. PGSs were calculated in PLINK2 [39] using the ‘‐‐score’ method. Each PGS was entered as a dependent variable in a regression model with the first 10 ancestry principal components and genetic sex as predictors. The standardized PGS residuals, after removing the effects of these covariates, were used for all further analyses to reduce model complexity while accounting for potential confounding effects.
Data analysis
Data analyses were carried out using R statistical software [40]. First, to examine the overall level of association between the PGSs and AUBs, mean levels of CON and HED were calculated in each sample, averaging across all available waves of data per individual. Each PGS residual was used to predict mean CON and HED in a linear regression model with the lmer() or lm() functions, for cohorts with (COGA, FinnTwin12 and FTC) or without (AddHealth, ALSPAC and S4S) relatives, respectively. Regression coefficient estimates were meta‐analyzed using random effects inverse variance weighting and restricted maximum likelihood estimation with the metafor package [41]. Heterogeneity in the effect size was estimated using the I2 statistic, a value from 0% to 100%, which reflects the degree of variability accounted for by between‐study heterogeneity [42].
The primary analyses involved latent growth curve (LGC) models, structural equation models that use latent growth factors (intercept, slope and quadratic) to represent the mathematical function underlying the covariance between repeated measures. Here, we fit LGCs for each AUB in each cohort using the OpenMx package [43]. First, a baseline model containing either intercept + slope (IS) or intercept + slope + quadratic (ISQ) parameters was fit to determine the general shape of the longitudinal trajectories. After comparing the difference in −2*loglikelihood between models with a χ2 test, the best fitting model was selected. Finally, this model was fit sequentially with the addition of each PGS residual (controlling for the same covariates listed above) as an individual‐level predictor of the latent growth factors. This allows us to determine how genetic influences captured by the PGSs shape AUB trajectories (Figure 2). Full information maximum likelihood estimation was used to account for missingness and non‐normality. A multi‐level application (adapted from https://github.com/OpenMx/OpenMx/blob/master/inst/models/nightly/mplus-ex9.12.R) was implemented to account for the non‐independence of observations in samples that included relatives by design (COGA, FinnTwin12 and FTC). Because >90% of participants in each cohort had initiated alcohol use before or during the assessment timeframe, no observations were removed for missing phenotypic values. Individuals were coded with a frequency of zero at each non‐drinking wave.
FIGURE 2.
Schematic of the latent growth curve (LGC) model used to estimate the effects of polygenic scores (PGSs) on trajectories of alcohol use behaviors. Measurements of alcohol use are collected at multiple ages, shown in squares. The within‐person covariance between these measures is captured by latent growth factors, shown in circles, which represent the intercept (Int), Slope and quadratic (Quad) parameters that create a line of best fit for an individual's alcohol use trajectory over time. Factor loadings are fixed to 1 for Int, to intervals corresponding with the period between measurements (years) for Slope and to the square of the slope values for Quad. Variance and covariance of the latent growth factors are estimated (Ψ), as are the residual, unexplained variances of the individual measurement occasions (δ/Δ). Of most interest here are the predicted effects (β) of the PGSs on the latent growth factors, which estimate whether an individuals' genetic risk for a particular trait is associated with differences in their values of the latent growth factors and, therefore, point to genetic effects on differential trajectories of alcohol use.
Models were tested individually in each cohort and meta‐analyzed, as above. The results from FTC were not included in the meta‐analysis because of the difference in age and available measures, but were compared qualitatively with the LGC meta‐analysis results. In FTC, only IS models were selected as many participants only had data for three or fewer measurement occasions, and quadratic effect estimates were unreliable. We applied a study‐wide Bonferroni correction for 56 PGS‐outcome pairs [(5 cohorts * 4 PGSs * 2 outcomes) + (4 PGSs * 4 outcomes in FTC)], resulting in an adjusted α = 0.05/56 = 0.0009.
RESULTS
Across cohorts, there was a consistent trend of CON and HED increasing sharply from adolescence into emerging adulthood and then leveling off around 22 to 24 years old (Figure 1(a),(b); Table S1). This trend was also apparent in the LGC analyses. Comparing ISQ models, which allow for a quadratic growth factor, to linear IS models showed that the ISQ models provided a better fit in all cases (−2*loglikelihood ratio P < 6.34E−08 in all cohorts). Similar longitudinal patterns from 18+ years old were observed in the older FTC sample, although there also appeared to be an uptick in AUBs after 55 years [Figure 1(c)].
All PGSs were robustly associated with higher mean levels of CON and HED when these were averaged across time (Table 1; Figure S1). The only exception was BeerPref PGSs, which were associated with a higher level of HED (meta β = 0.064, P = 3.65E−05) but not CON. Consumption PGSs were the strongest predictor of mean CON (meta β = 0.333) while Problems and EXT PGSs were the strongest predictors of mean HED (meta β = 0.154–0.178). In FTC (Table 2), Problems, Consumption and EXT PGSs were significantly associated with higher mean Freq (β = 0.215–0.317), while Problems, BeerPref and EXT PGSs were all associated with higher mean levels of gEtOH, Heavy drinking and PassOut frequency (β = 0.025–0.155). Across all analyses, the within‐sample variance in AUBs accounted for by each PGS was less than 2%.
TABLE 1.
PGS prediction of mean levels of typical CON and HED frequencies, meta‐analyzed across five longitudinal cohorts.
| AddHealth β | ALSPAC β | COGA β | FinnTwin12 β | S4S β | Meta β | Meta P | 95% CI | I2 | |
|---|---|---|---|---|---|---|---|---|---|
| Mean CON | |||||||||
| Problems | 0.331* | 0.286* | 0.236 | 0.156 | 0.170 | 0.237 | 4.01E−10 * | 0.163–0.312 | 54.194 |
| BeerPref | −0.006 | −0.282* | −0.135 | 0.084 | 0.119 | −0.040 | 0.598 | −0.190 to 0.109 | 88.634 |
| Consumption | 0.386* | 0.588* | 0.443* | 0.115 | 0.146 | 0.333 | 3.09E−04 * | 0.152−0.513 | 92.451 |
| EXT | 0.171* | 0.127 | 0.050 | 0.245* | 0.309* | 0.195 | 7.94E−07 * | 0.118−0.272 | 57.910 |
| Mean HED | |||||||||
| Problems | 0.238* | 0.135* | 0.259* | 0.081 | 0.117 | 0.154 | 2.41E−06 * | 0.090−0.218 | 77.209 |
| BeerPref | 0.100 | 0.037 | 0.101 | 0.058 | 0.117 | 0.064 | 3.65E−05 * | 0.034−0.095 | 13.200 |
| Consumption | 0.157* | 0.106* | 0.206 | 0.053 | 0.086 | 0.105 | 2.16E−06 * | 0.062−0.149 | 51.385 |
| EXT | 0.240* | 0.140* | 0.107 | 0.158* | 0.243* | 0.178 | 1.10E−13 * | 0.131−0.225 | 58.501 |
Note: Columns show the per‐cohort estimates of a linear regression coefficient (β) for the effect of the PGS on mean CON or HED levels, as well as the meta‐analysis of this effect across cohorts (meta β) and its associated P‐value and 95% CI. The I2 statistic is a metric of between‐cohort heterogeneity in the effect size. Additional abbreviations are found in the glossary (Appendix 1).
Abbreviations: AddHealth cohort, Add Health; ALSPAC, Avon Longitudinal Study of Parents and Children; BeefPref, preference for drinking beer and drinking without meals; COGA, Collaborative Studies on the Genetics of Alcoholism; CON, consumption; EXT, externalizing behaviour; FinnTwin12, FinnTwin12 cohort; FTC, older Finnish twin cohort; HED, heavy episodic drinking; PGS, polygenic score; S4S, Spit for Science.
Significant after Bonferroni correction for 56 PGS‐outcome pairs, P < 0.0009 (bolded values are significant meta‐analysis statistics).
TABLE 2.
PGS prediction of mean levels of alcohol use behaviors in the older FTC sample.
| PGS | β | P | 95% CI |
|---|---|---|---|
| Freq | |||
| Problems | 0.215 | 3.70E−06* | 0.124–0.306 |
| BeerPref | 0.118 | 1.14E−02 | 0.027–0.210 |
| Consumption | 0.317 | 8.81E−12* | 0.226–0.408 |
| EXT | 0.250 | 7.72E−08* | 0.159–0.342 |
| gEtOH | |||
| Problems | 0.089 | 1.03E−04* | 0.044–0.134 |
| BeerPref | 0.105 | 5.89E−06* | 0.060–0.150 |
| Consumption | 0.081 | 4.46E−04* | 0.036–0.126 |
| EXT | 0.155 | 1.65E−11* | 0.110–0.201 |
| Heavy | |||
| Problems | 0.025 | 4.88E−07* | 0.015–0.034 |
| BeerPref | 0.034 | 9.54E−12* | 0.024–0.043 |
| Consumption | 0.016 | 1.03E−03 | 0.007–0.260 |
| EXT | 0.046 | 1.01E−20* | 0.036–0.056 |
| PassOut | |||
| Problems | 0.086 | 4.70E−10* | 0.059–0.113 |
| BeerPref | 0.069 | 7.50E−07* | 0.041–0.096 |
| Consumption | 0.016 | 2.57E−01 | −0.012 to 0.043 |
| EXT | 0.084 | 1.10E−09* | 0.057–0.112 |
Note: Columns show the estimates of a linear regression coefficient (β) for the effect of the PGS on mean levels of the specified alcohol use behavior and its associated P‐value and 95% CI. Additional abbreviations are found in the glossary (Appendix 1).
Abbreviations: BeefPref, preference for drinking beer and drinking without meals; EXT, externalizing behaviour; FTC, older Finnish twin cohort; HED, heavy episodic drinking; PGS, polygenic score; Freq, frequency; gEtOH, grams of ethanol consumption per month; Heavy, heavy drinking; PassOut, past year frequency of passing out because of alcohol consumption.
Significant after Bonferroni correction for 56 PGS‐outcome pairs, P < 0.0009.
In the LGC meta‐analysis (Table 3; Figures S2–S3), Problems PGSs were significantly associated with a higher slope value for CON (meta β = 0.470, P = 4.20E−06), meaning that higher genetic risk predicted an accelerated trajectory of increasing drinking frequency over time. There was a general trend that higher PGSs were associated with higher intercept and slope values and lower quadratic factor values (i.e. less leveling off of drinking behaviors over time) for both CON and HED. Some of these were significant in individual cohorts, but these trends were not statistically significant in the combined meta‐analysis. Substantial heterogeneity was observed in the estimates across samples, with I2 values of up to 92%. Again, an exception was seen for Problems PGSs predicting CON trajectories, which had not only the strongest effects, but also the most consistent ones (I2 = 0%–2%). Given this between‐sample variability, we carried out an exploratory secondary meta‐analysis that (1) excluded the S4S cohort, which had a different measurement time scale than the other cohorts and demonstrated the greatest deviation in estimates; (2) used sample‐size rather than inverse‐variance weighting; and (3) based the effect size on the t test statistics of β instead of raw β values, in case measurement scale differences between cohorts were driving the high heterogeneity. In this exploratory analysis (Table S3), heterogeneity remained high. Results were consistent with the primary meta‐analysis, although the effects of EXT PGSs on the CON intercept and Problems PGSs on the HED intercept were now also significant (meta standardized β = 0.276–0.404, P < 8.60E−4).
TABLE 3.
PGS prediction of latent growth factors underlying trajectories of typical CON and HED frequencies, meta‐analyzed across five longitudinal cohorts.
| PGS | Factor | AddHealth β | ALSPAC β | COGA β | FinnTwin12 β | S4S β | Meta‐analysis | |||
|---|---|---|---|---|---|---|---|---|---|---|
| Meta β | Meta P | 95% CI | I2 | |||||||
| CON | ||||||||||
| Problems | Int | 0.036 | 0.129 | 0.021 | 0.013 | 0.600 | 0.031 | 0.064 | −0.002 to 0.064 | 0.43 |
| Problems | Slope | 0.462 | 0.599 | 0.828 | 0.290 | −2.142 | 0.470 | 4.20E−06 * | 0.270–0.670 | 2.05 |
| Problems | Quad | −0.077 | −0.343 | −0.349 | −0.092 | 2.481 | −0.132 | 0.021 | −0.245 to −0.020 | 0.02 |
| BeerPref | Int | 0.028 | −0.048 | −0.003 | 0.035 | 0.269 | 0.015 | 0.375 | −0.018, 0.049 | 0.00 |
| BeerPref | Slope | 0.159 | −0.731 | 0.429 | 0.115 | −0.617 | −0.035 | 0.877 | −0.482 to 0.412 | 74.25 |
| BeerPref | Quad | −0.206 | 0.359 | −0.424 | −0.021 | 0.678 | −0.052 | 0.707 | −0.326 to 0.221 | 70.43 |
| Consumption | Int | −0.009 | 0.257* | −0.014 | 0.004 | 0.465 | 0.062 | 0.318 | −0.059 to 0.183 | 88.55 |
| Consumption | Slope | 0.266 | 1.281* | 0.866 | 0.173 | −1.635 | 0.563 | 0.034 | 0.043–1.083 | 79.22 |
| Consumption | Quad | 0.257 | −0.709* | −0.142 | −0.053 | 1.874 | −0.105 | 0.596 | −0.494, 0.284 | 84.54 |
| EXT | Int | 0.146 | 0.287* | 0.063 | 0.028 | 0.587 | 0.127 | 0.022 | 0.018–0.235 | 85.92 |
| EXT | Slope | 0.303 | −0.438 | 0.786 | 0.857* | −0.828 | 0.361 | 0.220 | −0.216 to 0.938 | 83.83 |
| EXT | Quad | −0.229 | 0.134 | −0.714 | −0.330* | 0.583 | −0.268 | 0.070 | −0.559 to 0.022 | 73.74 |
| HED | ||||||||||
| Problems | Int | 0.055 | 0.090* | 0.025 | 0.007 | 0.432 | 0.042 | 0.025 | 0.005–0.079 | 64.97 |
| Problems | Slope | 0.299 | 0.095 | 0.892* | 0.134 | −1.414 | 0.285 | 0.070 | −0.023 to 0.593 | 82.97 |
| Problems | Quad | −0.060 | −0.025 | −0.486* | −0.023 | 1.547 | −0.103 | 0.244 | −0.277 to 0.071 | 76.52 |
| BeerPref | Int | 0.030 | 0.085* | 0.011 | −0.008 | 0.737 | 0.029 | 0.161 | −0.012 to 0.070 | 70.25 |
| BeerPref | Slope | 0.155 | −0.290 | 0.592 | 0.192 | −2.404 | 0.105 | 0.538 | −0.229 to 0.438 | 85.49 |
| BeerPref | Quad | −0.065 | 0.241 | −0.353 | −0.064 | 2.282 | −0.029 | 0.794 | −0.249 to 0.190 | 85.14 |
| Consumption | Int | 0.002 | 0.008 | 0.000 | −0.005 | −0.016 | 0.000 | 0.971 | −0.020 to 0.019 | 0.00 |
| Consumption | Slope | 0.088 | 0.519* | 0.661 | 0.104 | 0.199 | 0.314 | 0.022 | 0.045–0.583 | 77.67 |
| Consumption | Quad | 0.118 | −0.374* | −0.333 | −0.023 | −0.003 | −0.140 | 0.236 | −0.373 to 0.092 | 86.77 |
| EXT | Int | 0.141* | 0.174* | 0.010 | 0.024 | 0.576 | 0.091 | 0.029 | 0.009–0.172 | 92.11 |
| EXT | Slope | 0.399* | −0.176 | 0.699 | 0.422* | −1.384 | 0.287 | 0.110 | −0.065 to 0.639 | 87.05 |
| EXT | Quad | −0.232 | 0.111 | −0.498* | −0.154* | 1.381 | −0.162 | 0.157 | −0.387 to 0.062 | 85.99 |
Note: Columns show the per‐cohort estimates of a linear regression coefficient (β) for the effect of the PGS on latent growth factors (Int, slope and Quad) underlying longitudinal trajectories of CON and HED, as well as the meta‐analysis of this effect across cohorts (meta β) and its associated P‐value and 95% CI. The I2 statistic is a metric of between‐cohort heterogeneity in the effect size. Additional abbreviations are found in the glossary (Appendix 1).
Abbreviations: AddHealth, Add Health cohort; ALSPAC, Avon Longitudinal Study of Parents and Children; BeefPref, preference for drinking beer and drinking without meals; COGA, Collaborative Studies on the Genetics of Alcoholism; CON, consumption; EXT, externalizing behaviour; FinnTwin12, FinnTwin12 cohort; FTC, older Finnish twin cohort; HED, heavy episodic drinking; Int, intercept; PGS, polygenic score; Quad, quadratic; S4S, Spit for Science.
Significant after Bonferroni correction for 56 PGS‐outcome pairs, P < 0.0009 (bolded values are significant meta‐analysis statistics).
In comparison, LGC models within the older FTC cohort (Table 4) demonstrated that BeerPref and EXT PGSs were associated with a higher intercept for Heavy drinking (β = 0.047–0.059, P < 1.03E−05). EXT PGSs also predicted higher intercept values for gEtOH (β = 0.197, P = 1.12E−06). These differences likely reflect the older age of the sample, as the intercept now references drinking behaviors in young adulthood, after the period of sharp increase observed in adolescence in other cohorts. Again, no significant prediction from the PGSs was seen for the latent growth factors representing longitudinal changes in AUBs.
TABLE 4.
PGS prediction of latent growth factors underlying trajectories of alcohol use behaviors in middle and older adulthood in the older FTC sample.
| PGS | Factor | β | P | 95% CI |
|---|---|---|---|---|
| Freq | ||||
| Problems | Int | 0.218 | 0.073 | −0.020 to 0.456 |
| Problems | Slope | −0.004 | 0.908 | −0.072 to 0.064 |
| BeerPref | Int | 0.207 | 0.092 | −0.034 to 0.447 |
| BeerPref | Slope | −0.027 | 0.446 | −0.095 to 0.042 |
| Consumption | Int | 0.007 | 0.956 | −0.234 to 0.247 |
| Consumption | Slope | 0.087 | 0.013 | 0.019–0.155 |
| EXT | Int | 0.324 | 0.009 | 0.081–0.566 |
| EXT | Slope | −0.020 | 0.565 | −0.089 to 0.049 |
| gEtOH | ||||
| Problems | Int | 0.091 | 0.025 | 0.011–0.171 |
| Problems | Slope | −0.001 | 0.932 | −0.019 to 0.017 |
| BeerPref | Int | 0.134 | 0.001 | 0.054–0.214 |
| BeerPref | Slope | −0.008 | 0.412 | −0.026 to 0.011 |
| Consumption | Int | 0.003 | 0.937 | −0.077 to 0.083 |
| Consumption | Slope | 0.019 | 0.041 | 0.001–0.037 |
| EXT | Int | 0.197 | 1.12E−06* | 0.118–0.277 |
| EXT | Slope | −0.012 | 0.184 | −0.031 to 0.006 |
| Heavy | ||||
| Problems | Int | 0.025 | 0.017 | 0.004–0.046 |
| Problems | Slope | 0.000 | 0.931 | −0.005 to 0.004 |
| BeerPref | Int | 0.047 | 1.02E−05* | 0.026–0.068 |
| BeerPref | Slope | −0.004 | 0.117 | −0.008 to 0.001 |
| Consumption | Int | 0.017 | 0.112 | −0.004 to 0.038 |
| Consumption | Slope | 0.000 | 0.870 | −0.005 to 0.004 |
| EXT | Int | 0.059 | 2.14E−08* | 0.038–0.080 |
| EXT | Slope | −0.004 | 0.101 | −0.008 to 0.001 |
| PassOut | ||||
| Problems | Int | 0.250 | 0.005 | 0.076–0.424 |
| Problems | Slope | −0.037 | 0.057 | −0.075 to 0.001 |
| BeerPref | Int | 0.243 | 0.008 | 0.063–0.423 |
| BeerPref | Slope | −0.040 | 0.049 | −0.079 to 0.000 |
| Consumption | Int | −0.118 | 0.194 | −0.297 to 0.060 |
| Consumption | Slope | 0.032 | 0.106 | −0.007 to 0.072 |
| EXT | Int | 0.172 | 0.051 | −0.001 to 0.344 |
| EXT | Slope | −0.020 | 0.312 | −0.057 to 0.018 |
Note: Columns show the estimates of a linear regression coefficient (β) for the effect of the PGS on latent growth factors (Int, slope and Quad) underlying longitudinal trajectories of CON and HED and its associated P‐value and 95% CI. Additional abbreviations are found in the glossary (Appendix 1).
Abbreviations: BeefPref, preference for drinking beer and drinking without meals; EXT, externalizing behaviour; FTC, older Finnish twin cohort; HED, heavy episodic drinking; Int, intercept; PGS, polygenic score; Freq, frequency; gEtOH, grams of ethanol consumption per month; Heavy, heavy drinking; PassOut, past year frequency of passing out because of alcohol consumption; Quad, quadratic.
Significant after Bonferroni correction for 56 PGS‐outcome pairs, P < 0.0009.
DISCUSSION
Combining multiple international, longitudinal cohorts, we investigated genetic associations with stability and change in alcohol use behaviors across the lifespan. PGSs were used to index multiple AUB‐related dimensions, including EXT, alcohol Problems, alcohol Consumption and a preference for drinking beer and drinking without meals (BeerPref). Despite their robust associations with AUBs on average, most PGSs were not associated with the dynamics of AUB trajectories. However, Problems PGSs significantly predicted a steeper slope of consumption frequency, indicating a genetically influenced acceleration of drinking that is relevant to the development of alcohol problems.
AUB trajectories
Consistent with previous longitudinal studies [11, 17, 44], we observed strong age‐related trends in prevalence, with AUBs peaking/plateauing in young adulthood. These trends were stable across cohorts from multiple study designs, separated by continents and decades. Genetic associations with these trajectories, on the other hand, were highly variable, largely because of between‐cohort heterogeneity in the effect sizes [45]. This highlights an ongoing challenge in using results from GWASs, in which the hundreds of AUB‐related genes identified through enormous, pooled samples are still only able to account for a small fraction of the known heritability of AUBs [4, 5], likely because of inter‐individual variability in the specific causal genes and processes [45]. Despite focusing on AUB dimensions, defined with a bottom‐up strategy based on their genetic architecture [25], and despite our best efforts to harmonize measures across cohorts, between‐sample heterogeneity remained high in our study.
This likely reflects the variability in sample characteristics of the cohorts included in the current study. As such, meta‐analysis may not be the best approach, because PGSs may index biological processes that are variably relevant for different groups of people or in specific environmental contexts. For example, the Problems PGS seemed to be especially relevant for predicting an increasing, persistent trajectory (Slope and Quad factors) of HED frequency in the high‐risk COGA sample, but not other cohorts. The Consumption PGS, derived from a UK Biobank discovery GWAS, was uniquely associated with latent growth factors underlying CON frequency in the UK‐based ALSPAC sample, which suggests that this factor may reflect predispositions to drinking that are specific to or enhanced by the UK (social) environment. Future research should strive to catalog how the effects of genetic risk and protective factors differ systematically across cohorts and environmental moderators.
Interpretation of genetic effects
In light of this heterogeneity, it is especially notable that the Problems PGS consistently predicted a steeper slope of consumption frequency—the only significant effect in the LGC meta‐analysis. This is indicative of a robust biological process driving acceleration of alcohol consumption and one that may be specific to the peak in emerging adulthood since the same effect was not observed in the older FTC sample. This difference may, however, be a reflection of the lower heritability of AUBs found in older cohorts [18], likely because of secular changes in social constraints on the expression of genetic predispositions. Although the genes captured by this latent factor include some in common with other AUBs, such as the well‐known alcohol metabolism genes [6], it is only moderately correlated with Consumption, BeerPref or EXT [25]. Further investigation of the genetic influences specific to Problems may, therefore, be helpful to illuminate the causal processes underlying young adult acceleration of alcohol consumption and their links to later alcohol problems.
Results from mean level prediction analyses demonstrated that all four PGSs harbor genetic variants relevant to AUBs, but the growth model results indicated that these may not be the same genes shaping dynamic changes in drinking. Some evidence suggests that, despite the relatively high heritability of AUBs, the genetic influences on trajectories of heavy (episodic) drinking are quite low [17]. If so, this may reflect genetic equifinality, in which high genetic risk leads to high AUB manifestation at some point in life, although the specific timing is shaped by the environment or stochastic effects. For example, if certain genes shape one's susceptibility to drinking in social environments or under stress, these genes will be linked to higher average AUBs when collapsing across a lifetime's worth of social/stressful experiences, but the timing and tempo of these experiences may be driven by the environment or by a wholly distinct set of genes. Other studies, however, have demonstrated that the heritability of latent growth factors for alcohol frequency and quantity is similar to that of cross‐sectional measures of AUBs, approximately 30% to 50% [46, 47]. Further research is needed to resolve this question, and it may be necessary to incorporate longitudinal phenotypes into gene discovery efforts to determine whether the same genes are associated with dynamic versus static AUB measures [48]. The continued low variance in mean AUB levels explained by PGSs suggests that other avenues are worth pursuing in an effort to recapture the ‘missing heritability’, including deep phenotyping [25], longitudinal measures [48] and expanding coverage of the genome to rare and structural variants [49].
AUB dimensions
By including multi‐dimensional PGSs, we obtained insight into the breadth and specificity of genetic effects. PGSs indexing EXT and Problems were broadly associated with average consumption frequency/quantity as well as heavy (episodic) drinking, while Consumption and BeerPref PGSs were more specifically associated with average frequency or heavy drinking, respectively. BeerPref was even inversely related to consumption frequency in some samples, supporting the hypothesis that this construct represents a genetic predisposition involving a low level of response to alcohol that prompts heavier drinking quantity [37], rather than the social influences that are more relevant to drinking frequency [26]. Our results highlight the need for genetic studies of the diverse heritable risk processes, both broad and specific, which are important for AUBs.
Limitations and strengths
This study's findings should be viewed in the context of several limitations, including the lack of sample ancestral diversity, small number of AUB measures that could be harmonized across cohorts/waves (lacking important measures like the quantity and types of alcohol consumed, which may be especially relevant to the BeerPref and Consumption factors) and subtle cohort differences in how AUB measures were assessed. We dissect PGS effects that are bound by GWAS discovery power and capture only a small proportion of the variance of AUBs. Nonetheless, these are balanced by strengths of the meta‐analytic approach drawing from complementary cohort study designs, employment of multi‐dimensional PGSs and longitudinal design that explores the changing dynamics of AUBs across the lifespan.
CONCLUSIONS
We highlight the importance of genetic influences on alcohol Problems in shaping an accelerating trajectory of drinking frequency, which may already be useful for early identification of at‐risk individuals. Multiple other genetic and environmental risk dimensions are, of course, also important in shaping lifelong alcohol use behaviors. Further longitudinal research, especially within gene discovery efforts, is needed to illuminate the mechanisms linking genetic variation to the development of alcohol use behaviors across the lifespan.
AUTHOR CONTRIBUTIONS
Jeanne Savage: Conceptualization (lead); data curation (equal); formal analysis (equal); funding acquisition (equal); visualization (lead); writing—original draft (lead). Fazil Aliev: Data curation (equal); formal analysis (equal); writing—review and editing (equal). Peter Barr: Data curation (equal); formal analysis (equal); writing—review and editing (equal). Maia Choi: Data curation (equal); formal analysis (equal); writing—review and editing (equal). Gabin Drouard: Data curation (equal); formal analysis (equal); writing—review and editing (equal). Megan E. Cooke: Data curation (equal); writing—review and editing (equal). Sally I. Kuo: Data curation (equal); writing—review and editing (equal). Mallory Stephenson: Data curation (equal); writing—review and editing (equal). Sarah J. Brislin: Data curation (equal); writing—review and editing (equal). Zoe E. Neale: Data curation (equal); writing—review and editing (equal). Spit for Science Working Group: Data curation (equal); funding acquisition (equal); project administration (equal); resources (equal). COGA Investigators: Data curation (equal); funding acquisition (equal); project administration (equal); resources (equal). Antti Latvala: Supervision (equal); writing—review and editing (equal). Richard J. Rose: Funding acquisition (equal); supervision (equal); writing—review and editing (equal). Jaakko Kaprio: Funding acquisition (equal); supervision (equal); writing—review and editing (equal). Danielle M. Dick: Funding acquisition (equal); supervision (equal); writing—review and editing (equal). Jacquelyn L. Meyers: Funding acquisition (equal); supervision (equal); writing—review and editing (equal). Jessica Salvatore: Funding acquisition (equal); supervision (equal); writing—review and editing (equal). Danielle Posthuma: Supervision (equal); writing—review and editing (equal).
DECLARATION OF INTERESTS
D.M.D. is a co‐founder of Thrive Genetics, a member of the advisory board of Seek Health Group, and owns stock in both companies.
Supporting information
Table S1. Descriptive statistics of alcohol use behaviors measured in five longitudinal cohorts.
Table S2. Descriptive statistics of alcohol use behaviors measured in the FTC cohort.
Table S3. Secondary meta‐analysis of the polygenic score (PGS) prediction of latent growth factors underlying trajectories of typical consumption (CON) and heavy episodic drinking (HED) frequencies.
Figure S1. Forest plots of polygenic score (PGS) prediction of mean levels of typical consumption (CON) and heavy episodic drinking (HED) frequencies, meta‐analyzed across five longitudinal cohorts.
Figure S2. Forest plots of polygenic score (PGS) prediction of latent growth factors underlying trajectories of typical consumption (CON) frequency, meta‐analyzed across five longitudinal cohorts.
Figure S3. Forest plots of polygenic score (PGS) prediction of latent growth factors underlying trajectories of heavy episodic drinking (HED) frequency, meta‐analyzed across five longitudinal cohorts.
ACKNOWLEDGEMENTS
This research was funded by a grant to J.E.S. (VI.VENI.201G‐064) from The Netherlands Organization for Scientific Research (NWO). Additional support for data analysis was provided by the US National Institute on Drug Abuse (NIDA) grant T32DA55569 (M.C.) and for each study cohort as described below.
AddHealth: Add Health is directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Waves I–V data are from the Add Health Program Project, grant P01HD31921 (Harris) from Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), with cooperative funding from 23 other federal agencies and foundations. Add Health was designed by J. Richard Udry, Peter S. Bearman and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill.
ALSPAC: We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. The UK Medical Research Council and Wellcome (grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors and M.C. and D.D. will serve as guarantors for the contents of this paper. A comprehensive list of grants funding is available on the ALSPAC website (http://www.bristol.ac.uk/alspac/external/documents/grant-acknowledgements.pdf). Genomewide genotyping data was generated by Sample Logistics and Genotyping Facilities at Wellcome Sanger Institute and LabCorp (Laboratory Corporation of America) using support from 23andMe.
COGA: We continue to be inspired by our memories of Henri Begleiter and Theodore Reich, founding PI and Co‐PI of COGA, and also owe a debt of gratitude to other past organizers of COGA, including Ting‐ Kai Li, P. Michael Conneally, Raymond Crowe and Wendy Reich, for their critical contributions. This national collaborative study is supported by National Institutes of Health (NIH) grant U10AA008401 from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) and the NIDA.
FinnTwin12 is supported by the NIAAA (award numbers R01AA015416, R01AA09203, K02AA018755 and K01AA024152) and the Academy of Finland (grants 100499, 205585, 118555, 141054, 265240, 263278 and 264146).
Finnish twin cohort: older twins: Phenotype and genotype data collection in the FTC has been supported by the Wellcome Trust Sanger Institute, the Broad Institute, ENGAGE‐European Network for Genetic and Genomic Epidemiology, FP7‐HEALTH‐F4‐2007, grant agreement number 201413 and the Academy of Finland Center of Excellence in Complex Disease Genetics (grant 352792).
Spit for Science has been supported by Virginia Commonwealth University, P20AA017828, R37AA011408, K02AA018755, P50AA022537 and K01AA024152 from the NIAAA, UL1RR031990 from the National Center for Research Resources and National Institutes of Health Roadmap for Medical Research, as well as support by the Center for the Study of Tobacco Products at VCU. REDCap support provided by CTSA award UM1TR004360 from the National Center for Advancing Translational Sciences. We thank Dr. Danielle Dick for founding and directing the S4S Registry from 2011 to 2022, and the S4S participants for making this study a success, as well as the many University faculty, students and staff who contributed to the design and implementation of the project.
The Externalizing Consortium: Principal Investigators include D.M.D., Philipp Koellinger, K. Paige Harden, and Abraham A. Palmer. The lead analysts include Richard Karlsson Linnér, Travis T. Mallard, Peter B. Barr, and Sandra Sanchez‐Roige; and significant contributors include Irwin Waldman. The Externalizing Consortium has been supported by the NIAAA (R01AA015416—administrative supplement to D.M.D.), and the NNIDA (R01DA050721 to D.M.D.) Additional funding for investigator effort has been provided by K02AA018755, U10AA008401 and P50AA022537 to D.M.D., as well as a European Research Council Consolidator Grant (647648 EdGe to Koellinger). The Externalizing Consortium thank the following groups for making the research possible: 23andMe, Add Health, Vanderbilt University Medical Center's BioVU, COGA, the Psychiatric Genomics Consortium's Substance Use Disorders working group, UK10K Consortium, UK Biobank and Philadelphia Neurodevelopmental Cohort.
The content is solely the responsibility of the authors and does not necessarily represent the views of the funding agencies. The funding agencies had no role in the study design, data analysis, manuscript preparation, or decision to submit for publication.
The Spit for Science Working Group: Director: Karen Chartier Co‐Director: Ananda Amstadter. Past Founding Director: Danielle M. Dick. Registry management: Emily Lilley, Renolda Gelzinis, Anne Morris. Data cleaning and management: Katie Bountress, Amy E. Adkins, Nathaniel Thomas, Zoe Neale, Kimberly Pedersen, Thomas Bannard & Seung B. Cho. Data collection: Amy E. Adkins, Kimberly Pedersen, Peter Barr, Holly Byers, Erin C. Berenz, Erin Caraway, Seung B. Cho, James S. Clifford, Megan Cooke, Elizabeth Do, Alexis C. Edwards, Neeru Goyal, Laura M. Hack, Lisa J. Halberstadt, Sage Hawn, Sally Kuo, Emily Lasko, Jennifer Lend, Mackenzie Lind, Elizabeth Long, Alexandra Martelli, Jacquelyn L. Meyers, Kerry Mitchell, Ashlee Moore, Arden Moscati, Aashir Nasim, Zoe Neale, Jill Opalesky, Cassie Overstreet, A. Christian Pais, Kimberly Pedersen, Tarah Raldiris, Jessica Salvatore, Jeanne Savage, Rebecca Smith, David Sosnowski, Jinni Su, Nathaniel Thomas, Chloe Walker, Marcie Walsh, Teresa Willoughby, Madison Woodroof & Jia Yan. Genotypic data processing and cleaning: Cuie Sun, Brandon Wormley, Brien Riley, Fazil Aliev, Roseann Peterson & Bradley T. Webb.
The Collaborative Study on the Genetics of Alcoholism (COGA): Principal Investigators B. Porjesz, V. Hesselbrock, A. Agrawal; Scientific Director, A. Agrawal; Translational Director, D. Dick, includes ten different centers: University of Connecticut (V. Hesselbrock); Indiana University (H.J. Edenberg, T. Foroud, Y. Liu, M.H. Plawecki); University of Iowa Carver College of Medicine (S. Kuperman, A. Anderson); SUNY Downstate Health Sciences University (B. Porjesz, J. Meyers); Washington University in St. Louis (L. Bierut, A. Agrawal, S. Hartz); University of California at San Diego (M. Schuckit); Rutgers University (D. Dick, R. Hart, J. Salvatore, J. Tischfield); The Children's Hospital of Philadelphia, University of Pennsylvania (L. Almasy); Icahn School of Medicine at Mount Sinai (A. Goate, P. Slesinger); and Howard University (D. Scott). Other COGA collaborators include: C. Holzhauer, M. Hesselbrock (University of Connecticut); D. Lai, J. Nurnberger Jr., L. Wetherill, X., Xuei, S. O’Connor, (Indiana University); J. Kramer (University of Iowa), G. Chan (University of Iowa; University of Connecticut); C. Kamarajan, A. Pandey, D.B. Chorlian, P. Barr, S. Kinreich, G. Pandey, Z. Neale, S., C. Chatzinakos, J. Zhang, Saenz deViteri, R. Christian, A. Bingly (SUNY Downstate); G. Pathak (Icahn School of Medicine at Mount Sinai); A. Anokhin, K. Bucholz, F. Dong, A. Hatoum, E. Johnson, V. McCutcheon, J. Rice, S. Saccone (Washington University); F. Aliev, Z. Pang, S. Kuo, S. Brislin, J. Moore (Rutgers University); A. Merikangas (The Children's Hospital of Philadelphia and University of Pennsylvania); M. Gitik, NIAAA Staff Collaborator.
Savage JE, Aliev F, Barr PB, Choi M, Drouard G, Cooke ME, et al. Trajectories of genetic risk across dimensions of alcohol use behaviors. Addiction. 2026;121(5):1153–1165. 10.1111/add.70292
Funding information This research was funded by a grant from The Netherlands Organization for Scientific Research, VI.VENI.201G‐064 to J.E.S; United States (US) National Institute on Drug Abuse (NIDA), grant T32DA55569; National Institute on Aging cooperative agreements U01AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill; the Add Health Program Project, grant P01HD31921 (Harris) from Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), with cooperative funding from 23 other federal agencies and foundations; United Kingdom (UK) Medical Research Council and Wellcome (grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC; National Institutes of Health, grant U10AA008401 from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) and the NIDA support COGA; FinnTwin cohorts are supported by the NIAAA (award numbers R01AA015416, R01AA09203, K02AA018755 and K01AA024152); Academy of Finland (grants 100499, 205585, 118555, 141054, 265240, 263278 and 264146; Wellcome Trust Sanger Institute, the Broad Institute, ENGAGE‐European Network for Genetic and Genomic Epidemiology, FP7‐HEALTH‐F4‐2007, grant agreement number 201413 and the Academy of Finland Center of Excellence in Complex Disease Genetics (grant 352792); Spit for Science has been supported by Virginia Commonwealth University, P20AA017828, R37AA011408, K02AA018755, P50AA022537 and K01AA024152 from the NIAAA, UL1RR031990 from the National Center for Research Resources and National Institutes of Health Roadmap for Medical Research, as well as support by the Center for the Study of Tobacco Products at VCU; REDCap support provided by CTSA award UM1TR004360 from the National Center for Advancing Translational Sciences; the Externalizing Consortium has been supported by the NIAAA (R01AA015416—administrative supplement to D.M.D.), and the NIDA (R01DA050721 to D.M.D.); additional funding for investigator effort has been provided by K02AA018755, U10AA008401 and P50AA022537 to D.M.D., as well as a European Research Council Consolidator Grant (647648 EdGe to Koellinger).
Contributor Information
Jeanne E. Savage, Email: j.e.savage@vu.nl.
Spit for Science Working Group:
Karen Chartier, Ananda Amstadter, Danielle M. Dick, Emily Lilley, Renolda Gelzinis, Anne Morris, Katie Bountress, Amy E. Adkins, Nathaniel Thomas, Zoe Neale, Kimberly Pedersen, Thomas Bannard, Seung B. Cho, Peter Barr, Holly Byers, Erin C. Berenz, Erin Caraway, James S. Clifford, Megan Cooke, Elizabeth Do, Alexis C. Edwards, Neeru Goyal, Laura M. Hack, Lisa J. Halberstadt, Sage Hawn, Sally Kuo, Emily Lasko, Jennifer Lend, Mackenzie Lind, Elizabeth Long, Alexandra Martelli, Jacquelyn L. Meyers, Kerry Mitchell, Ashlee Moore, Arden Moscati, Aashir Nasim, Jill Opalesky, Cassie Overstreet, A. Christian Pais, Tarah Raldiris, Jessica Salvatore, Jeanne Savage, Rebecca Smith, David Sosnowski, Jinni Su, Chloe Walker, Marcie Walsh, Teresa Willoughby, Madison Woodroof, Jia Yan, Cuie Sun, Brandon Wormley, Brien Riley, Fazil Aliev, Roseann Peterson, and Bradley T. Webb
COGA Investigators:
Bernice Porjesz, Victor Hesselbrock, Arpana Agrawal, Danielle Dick, Howard J. Edenberg, Tatiana Foroud, Yunlong Liu, Martin H. Plawecki, Samuel Kuperman, Allan Anderson, Jacquelyn Meyers, Laura Bierut, Sarah Hartz, Marc Schuckit, Ronald Hart, Jay Tischfield, Laura Almasy, Alison Goate, Paul Slesinger, Denise Scott, Cathryn Holzhauer, Michie Hesselbrock, Dongbing Lai, John Nurnberger, Jr., Leah Wetherill, Xiaoling Xuei, Sean O'Connor, John Kramer, Grace Chan, Chella Kamarajan, Ashwini Pandey, David B. Chorlian, Sivan Kinreich, Gayathri Pandey, Chris Chatzinakos, Jian Zhang, Stacey Saenz deViteri, Christian Richard, Arjun Bingly, Gita Pathak, Andrey Anokhin, Kathleen Bucholz, Fanghong Dong, Alexander Hatoum, Emma Johnson, Vivia McCutcheon, John Rice, Scott Saccone, Zhiping Pang, Sarah Brislin, Jennifer Moore, Alison Merikangas, and Miri Gitik
DATA AVAILABILITY STATEMENT
Raw data from this study are available to qualified researchers. AddHealth: dbGaP accession no. phs001367.v1.p1. ALSPAC: https://www.bristol.ac.uk/alspac/researchers/access/. COGA: https://cogastudy.org/. Finnish twin cohort (FinnTwin12 and older cohort): data used in the analysis is deposited in the Biobank of the Finnish Institute for Health and Welfare (https://thl.fi/en/research-and-development/thl-biobank/for-researchers) and is available to researchers after written application and following the relevant Finnish legislation. Spit for Science: dbGaP accession no. phs001754.v4.p2.
REFERENCES
- 1. Dick DM, Meyers JL, Rose RJ, Kaprio J, Kendler KS. Measures of current alcohol consumption and problems: two independent twin studies suggest a complex genetic architecture. Alcohol Clin Exp Res. 2011;35(12):2152–2161. 10.1111/j.1530-0277.2011.01564.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Hart AB, Kranzler HR. Alcohol dependence genetics: lessons learned from genome‐wide association studies (GWAS) and post‐GWAS analyses. Alcohol Clin Exp Res. 2015;39(8):1312–1327. 10.1111/acer.12792 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Wu SH, Guo Q, Viken RJ, Reed T, Dai J. Heritability of usual alcohol intoxication and hangover in male twins: the NAS‐NRC twin registry. Alcohol Clin Exp Res. 2014;38(8):2307–2313. 10.1111/acer.12487 [DOI] [PubMed] [Google Scholar]
- 4. Saunders GRB, Wang X, Chen F, Jang SK, Liu M, Wang C, et al. Genetic diversity fuels gene discovery for tobacco and alcohol use. Nature. 2022;612(7941):720–724. 10.1038/s41586-022-05477-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Zhou H, Kember RL, Deak JD, Xu H, Toikumo S, Yuan K, et al. Multi‐ancestry study of the genetics of problematic alcohol use in over 1 million individuals. Nat Med. 2023;29(12):3184–3192. 10.1038/s41591-023-02653-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Edenberg HJ. The genetics of alcohol metabolism: role of alcohol dehydrogenase and aldehyde dehydrogenase variants. Alcohol Res Health. 2007;30(1):5–13. [PMC free article] [PubMed] [Google Scholar]
- 7. Ksinan AJ, Su J, Aliev F, Dick DM. Unpacking genetic risk pathways for college student alcohol consumption: the mediating role of impulsivity. Alcohol Clin Exp Res. 2019;43(10):2100–2110. 10.1111/acer.14157 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Barr PB, Mallard TT, Sanchez‐Roige S, Poore HE, Linnér RK, Waldman ID, et al. Parsing genetically influenced risk pathways: genetic loci impact problematic alcohol use via externalizing and specific risk. Transl Psychiatry. 2022;12(1):420. 10.1038/s41398-022-02171-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Karlsson Linnér R, Mallard TT, Barr PB, Sanchez‐Roige S, Madole JW, Driver MN, et al. Multivariate analysis of 1.5 million people identifies genetic associations with traits related to self‐regulation and addiction. Nat Neurosci. 2021;24(10):1367–1376. 10.1038/s41593-021-00908-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Barr PB, Silberg J, Dick DM, Maes HH. Childhood socioeconomic status and longitudinal patterns of alcohol problems: variation across etiological pathways in genetic risk. Soc Sci Med. 2018;209:51–58. 10.1016/j.socscimed.2018.05.027 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Britton A, Ben‐Shlomo Y, Benzeval M, Kuh D, Bell S. Life course trajectories of alcohol consumption in the United Kingdom using longitudinal data from nine cohort studies. BMC Med. 2015;13(1):47. 10.1186/s12916-015-0273-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Harrington M, Velicer WF, Ramsey S. Typology of alcohol users based on longitudinal patterns of drinking. Addict Behav. 2014;39(3):607–621. 10.1016/j.addbeh.2013.11.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Holton A, Boland F, Gallagher P, Fahey T, Kenny R, Cousins G. Life course transitions and changes in alcohol consumption among older Irish adults: results from the Irish longitudinal study on ageing (TILDA). J Aging Health. 2019;31(9):1568–1588. 10.1177/0898264318783080 [DOI] [PubMed] [Google Scholar]
- 14. Leggat G, Livingston M, Kuntsche S, Callinan S. Alcohol consumption trajectories over the Australian life course. Addiction. 2022;117(7):1931–1939. 10.1111/add.15849 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Long EC, Lonn SL, Sundquist J, Sundquist K, Kendler KS. A National Swedish Longitudinal Twin‐Sibling Study of alcohol use disorders among males. Addiction. 2017;112(8):1378–1385. 10.1111/add.13833 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Mezquita L, Ibáñez MI, Moya J, Villa H, Ortet G. A longitudinal examination of different etiological pathways to alcohol use and misuse. Alcohol Clin Exp Res. 2014;38(6):1770–1779. 10.1111/acer.12419 [DOI] [PubMed] [Google Scholar]
- 17. Stephenson M, Barr P, Thomas N, Cooke M, Latvala A, Rose RJ, et al. Patterns and predictors of alcohol misuse trajectories from adolescence through early midlife. Dev Psychopathol. 2024;1‐17(2):734–750. 10.1017/S0954579424000543 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Virtanen S, Kaprio J, Viken R, Rose RJ, Latvala A. Birth cohort effects on the quantity and heritability of alcohol consumption in adulthood: a Finnish longitudinal twin study. Addiction. 2019;114(5):836–846. 10.1111/add.14533 [DOI] [PubMed] [Google Scholar]
- 19. Koob GF, Volkow ND. Neurobiology of addiction: a neurocircuitry analysis. Lancet Psychiatry. 2016;3(8):760–773. 10.1016/S2215-0366(16)00104-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Kendler KS, Schmitt E, Aggen SH, Prescott CA. Genetic and environmental influences on alcohol, caffeine, cannabis, and nicotine use from early adolescence to middle adulthood. Arch Gen Psychiatry. 2008;65(6):674–682. 10.1001/archpsyc.65.6.674 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Gustavson DE, Bell TR, Buchholz EJ, Zellers S, Luczak SE, Reynolds CA, et al. Genetic and environmental influences on alcohol consumption in middle to late life. Psychol Addict Behav. 2025;39(7):591–606. 10.1037/adb0001052 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Waaktaar T, Kan KJ, Torgersen S. The genetic and environmental architecture of substance use development from early adolescence into young adulthood: a longitudinal twin study of comorbidity of alcohol, tobacco and illicit drug use. Addiction. 2018;113(4):740–748. 10.1111/add.14076 [DOI] [PubMed] [Google Scholar]
- 23. Edwards AC, Kendler KS. Alcohol consumption in men is influenced by qualitatively different genetic factors in adolescence and adulthood. Psychol Med. 2013;43(9):1857–1868. 10.1017/S0033291712002917 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Walters RK, Polimanti R, Johnson EC, McClintick JN, Adams MJ, Adkins AE, et al. Transancestral GWAS of alcohol dependence reveals common genetic underpinnings with psychiatric disorders. Nat Neurosci. 2018;21(12):1656–1669. 10.1038/s41593-018-0275-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Savage JE, Barr PB, Phung T, Lee YH, Zhang Y, McCutcheon VV, et al. Genetic heterogeneity across dimensions of alcohol use behaviors. Am J Psychiatry. 2024;181(11):1006–1017. 10.1176/appi.ajp.20231055 [DOI] [PubMed] [Google Scholar]
- 26. Mallard TT, Savage JE, Johnson EC, Huang Y, Edwards AC, Hottenga JJ, et al. Item‐level genome‐wide association study of the alcohol use disorders identification test in three population‐based cohorts. Am J Psychiatry. 2022;179(1):58–70. 10.1176/appi.ajp.2020.20091390 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Sanchez‐Roige S, Palmer AA, Fontanillas P, Elson SL, Adams MJ, Howard DM, et al. Genome‐wide association study Meta‐analysis of the alcohol use disorders identification test (AUDIT) in two population‐based cohorts. Am J Psychiatry. 2019;176(2):107–118. 10.1176/appi.ajp.2018.18040369 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Meyers JL, Salvatore JE, Vuoksimaa E, Korhonen T, Pulkkinen L, Rose RJ, et al. Genetic influences on alcohol use behaviors have diverging developmental trajectories: a prospective study among male and female twins. Alcohol Clin Exp Res. 2014;38(11):2869–2877. 10.1111/acer.12560 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Harris KM, Halpern CT, Whitsel EA, Hussey JM, Killeya‐Jones LA, Tabor J, et al. Cohort profile: the National Longitudinal Study of adolescent to adult health (add health). Int J Epidemiol. 2019;48(5):1415–1415k. 10.1093/ije/dyz115 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)‐‐a metadata‐driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42(2):377–381. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Boyd A, Golding J, Macleod J, Lawlor DA, Fraser A, Henderson J, et al. Cohort Profile: the 'children of the 90s'‐‐the index offspring of the Avon Longitudinal Study of Parents and Children. Int J Epidemiol. 2013;42(1):111–127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Fraser A, Macdonald‐Wallis C, Tilling K, Boyd A, Golding J, Davey Smith G, et al. Cohort profile: the Avon longitudinal study of parents and children: ALSPAC mothers cohort. Int J Epidemiol. 2013;42(1):97–110. 10.1093/ije/dys066 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Begleiter H, Reich T. The collaborative study on the genetics of alcoholism. Alcohol Health Res World. 1995;19(3):228. [PMC free article] [PubMed] [Google Scholar]
- 34. Rose RJ, Salvatore JE, Aaltonen S, Barr PB, Bogl LH, Byers HA, et al. FinnTwin12 cohort: an updated review. Twin Res Hum Genet. 2019;22(5):302–311. 10.1017/thg.2019.83 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Kaprio J, Bollepalli S, Buchwald J, Iso‐Markku P, Korhonen T, Kovanen V, et al. The older Finnish twin cohort ‐ 45 years of follow‐up. Twin Res Hum Genet. 2019;22(4):240–254. 10.1017/thg.2019.54 [DOI] [PubMed] [Google Scholar]
- 36. Dick DM, Nasim A, Edwards AC, Salvatore JE, Cho SB, Adkins A, et al. Spit for Science: launching a longitudinal study of genetic and environmental influences on substance use and emotional health at a large US university. Front Genet. 2014;5:47. 10.3389/fgene.2014.00047 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Savage JE, Spit for Science Working G , Dick DM, Posthuma D. Drinking motives and alcohol sensitivity mediate multidimensional genetic influences on alcohol use behaviors. Alcohol Clin Exp Res. 2025;49(5):1001–1012. 10.1111/acer.70045 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Ge T, Chen CY, Ni Y, Feng YA, Smoller JW. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nat Commun. 2019;10(1):1776. 10.1038/s41467-019-09718-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second‐generation PLINK: rising to the challenge of larger and richer datasets. Gigascience. 2015;4:7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. R Core Team . R: a language and environment for statistical computing Vienna, Austria: R Foundation by Statistical Computing; 2017. [Google Scholar]
- 41. Viechtbauer W. Conducting Meta‐analyses in R with the metafor package. J Stat Softw. 2010;36(3):1–48. 10.18637/jss.v036.i03 [DOI] [Google Scholar]
- 42. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta‐analysis. Stat Med. 2002;21(11):1539–1558. 10.1002/sim.1186 [DOI] [PubMed] [Google Scholar]
- 43. Boker S, Neale M, Maes H, Wilde M, Spiegel M, Brick T, et al. OpenMx: an open source extended structural equation modeling framework. Psychometrika. 2011;76(2):306–317. 10.1007/s11336-010-9200-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Lee MR, Sher KJ. "maturing out" of binge and problem drinking. Alcohol Res. 2018;39(1):31–42. 10.35946/arcr.v39.1.06 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Huedo‐Medina TB, Sánchez‐Meca J, Marín‐Martínez F, Botella J. Assessing heterogeneity in meta‐analysis: Q statistic or I2 index? Psychol Methods. 2006;11(2):193–206. 10.1037/1082-989X.11.2.193 [DOI] [PubMed] [Google Scholar]
- 46. Zellers SM, Iacono WG, McGue M, Vrieze S. Developmental and etiological patterns of substance use from adolescence to middle age: a longitudinal twin study. Drug Alcohol Depend. 2022;233:109378. 10.1016/j.drugalcdep.2022.109378 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Drouard G, Silventoinen K, Latvala A, Kaprio J. Genetic and environmental factors underlying parallel changes in body mass index and alcohol consumption: a 36‐year longitudinal study of adult twins. Obes Facts. 2023;16(3):224–236. 10.1159/000529835 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Thomas NS, Gillespie NA, Chan G, Edenberg HJ, Kamarajan C, Kuo SIC, et al. A developmentally‐informative genome‐wide association study of alcohol use frequency. Behav Genet. 2024;54(2):151–168. 10.1007/s10519-023-10170-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Jang SK, Evans L, Fialkowski A, Arnett DK, Ashley‐Koch AE, Barnes KC, et al. Rare genetic variants explain missing heritability in smoking. Nat Hum Behav. 2022;6(11):1577–1586. 10.1038/s41562-022-01408-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1. Descriptive statistics of alcohol use behaviors measured in five longitudinal cohorts.
Table S2. Descriptive statistics of alcohol use behaviors measured in the FTC cohort.
Table S3. Secondary meta‐analysis of the polygenic score (PGS) prediction of latent growth factors underlying trajectories of typical consumption (CON) and heavy episodic drinking (HED) frequencies.
Figure S1. Forest plots of polygenic score (PGS) prediction of mean levels of typical consumption (CON) and heavy episodic drinking (HED) frequencies, meta‐analyzed across five longitudinal cohorts.
Figure S2. Forest plots of polygenic score (PGS) prediction of latent growth factors underlying trajectories of typical consumption (CON) frequency, meta‐analyzed across five longitudinal cohorts.
Figure S3. Forest plots of polygenic score (PGS) prediction of latent growth factors underlying trajectories of heavy episodic drinking (HED) frequency, meta‐analyzed across five longitudinal cohorts.
Data Availability Statement
Raw data from this study are available to qualified researchers. AddHealth: dbGaP accession no. phs001367.v1.p1. ALSPAC: https://www.bristol.ac.uk/alspac/researchers/access/. COGA: https://cogastudy.org/. Finnish twin cohort (FinnTwin12 and older cohort): data used in the analysis is deposited in the Biobank of the Finnish Institute for Health and Welfare (https://thl.fi/en/research-and-development/thl-biobank/for-researchers) and is available to researchers after written application and following the relevant Finnish legislation. Spit for Science: dbGaP accession no. phs001754.v4.p2.