Two decades ago, several groups started to associate genetic variants with measures of brain structure and function rather than clinically diagnosed disease categories [1,2,3,4]. For example, it was assumed that variance in the genetic constitution of monoamine transporters should have a stronger impact on in vivo transporter availability and, consequently, brain function than on subjective mood states [1, 2]. Likewise, genetic variants associated with the dopamine D2 receptor were associated with in vivo receptor availability [3, 5]. Shortly thereafter, genotype effects on MRI-derived functional brain activation rather than protein expression were studied, with a focus on working memory-dependent brain activation [6]. A meta-analysis of effect sizes reflecting the association of genetic variance with brain versus behavioral data confirmed that the assessed genetic variants displayed stronger associations with brain function than with cognition or clinical symptoms [7, 8]. However, many early candidate gene study findings failed to replicate and meta-analyses showed that the observed associations between genotype and functional activation were smaller than originally assumed [9]. In fact, most early studies were not sufficiently powered to reliably produce large effect sizes. Moreover, a variety of disease-related as well as comorbid factors including smoking or stress hormone activation can interact with genotype effects on brain correlates [10, 11], requiring adequate sample sizes to address complex interactions.

To address the issue of statistical power and to confirm whether brain structures and functions are directly associated with genetic variance or are a secondary consequence of the disorder (e.g., due to pharmacological drug effects), prospective long-term studies are required, which assess young participants before the manifestation of symptoms or clinical disorders. For this purpose, the IMAGEN consortium was established in 2010. It includes 8 European centers; each of them recruited at least 250 healthy adolescents aged 14, who have been followed up at ages 16, 19, and 22 [12].

Another important benefit of the shift towards large multidisciplinary collaborations such as IMAGEN and ABCD [13] is that it allows for data driven discovery science and out-of-sample prediction for in vivo imaging genetics researchers. The initial IMAGEN sample size (>2000) was large enough to assess the effect of previously identified SNPs or polygenic risk scores on functional activation and their interaction with additional factors including environmental measures [14,15,16] and comorbid factors such as smoking or stress hormone activation [10, 11]. Although not large enough for candidate gene or polygenic score construction [17], a phenotypically rich longitudinal dataset like IMAGEN is uniquely placed to identify how genes and behavior relate via psychological or neurobiological intermediate phenotypes.

Since its inception, the IMAGEN consortium has published a number of significant papers investigating how genetic and imaging findings contribute to specific traits, behaviors, symptoms, and disorders, for example with respect to impulsive decision making and drug consumption [18, 19]. Despite the 100+ IMAGEN publications in the last decade, no systematic review of IMAGEN findings has been published to date. Here we present all original IMAGEN imaging genetics papers, i.e., papers assessing effects of genetic variation on brain structure or functional brain activation during either reward anticipation, behavioral inhibition or processing of affective faces, and discuss their respective behavioral correlates with a focus on drugs of abuse including alcohol, tobacco, and cannabis.


Data sources

A systematic search for all IMAGEN publications was carried out by LMM and AR between October 1st, 2016, and December 31st, 2018. Relevant studies were identified using the list of published papers provided online by the IMAGEN consortium ( and by systematically searching the PUBMED database using the search terms “IMAGEN” and “consortium.” Study references of identified articles were additionally reviewed and taken into account. Altogether, we identified a total number of 110 papers from the IMAGEN consortium including manuscripts with IMAGEN as a contributor. Among these, 62 publications report interactions between genotype and functional or structural brain data.

Study selection

To provide a systematic account of all papers published by the IMAGEN consortium, we reviewed all identified papers. Abstracts were screened for relevance, and all identified articles were discussed by LMM, HW, AR, and AH. We excluded studies that were systematic literature reviews and animal studies that were not using the IMAGEN sample of participants (Fig. 1).

Fig. 1: Selection process.
figure 1

Flow-chart of selection process of identified publications.

We categorized papers from the IMAGEN consortium with respect to whether genetic variations were associated with (1) structural brain measures or (2) with one of the three tasks applied for functional imaging, i.e., the stop signal task (SST), the monetary incentive delay task (MID) and the Emotional faces (EF), or (3) intake of drugs of abuse. The variety of genetic variants addressed in IMAGEN studies was classified into four categories with respect to the main function of putatively associated genes: (1) neurodevelopment, (2) apoptosis and cell cycle, (3) neurotransmission, and (4) metabolic or endocrine function [20, 21].

To facilitate further use of the papers by the scientific community, we created three tables with key findings of all imaging genetics papers published by the IMAGEN consortium. The tables provide a brief overview of every paper, listing the title of the paper, authors, journal, year, the single nucleotide polymorphisms (SNPs) investigated and the key imaging variable (brain volume, structure, or functional task) addressed by the study. Genotype effects are discussed with respect to findings of other papers by the IMAGEN consortium and by independent groups.

From the 110 papers, 7 were reviews that did not contain original data or focused exclusively on animal data; these 7 papers are not further discussed in this review. The remaining 103 studies reviewed here include 41 papers that focused on methodological issues or behavioral features, while 62 manuscripts reported genotype associations with brain structure or function or correlations between behavioral variables and functional brain activation.


Description of identified studies

We identified 103 relevant original papers including published abstracts. Among these papers, 12 manuscripts reported associations between genetic variations and brain volume [4, 14, 19, 22,23,24,25,26,27,28,29,30] (Table 1). Furthermore, 16 manuscripts reported associations of genetic variation on functional brain activation elicited by the MID task [15, 31,32,33,34,35,36,37,38,39,40,41,42,43,44,45]; 6 papers described functional brain activation elicited by the SST [16, 18, 46,47,48,49], and 4 papers showed functional brain activation elicited by the EF [50,51,52,53] (Table 2). We identified 24 papers that did not address imaging genetics but reported correlations between behavioral variables and functional brain activation elicited by (1) the MID task (n = 13) [54,55,56,57,58,59,60,61,62,63,64,65,66], (2) the SST (n = 3) [67,68,69], and (3) the FRT (n = 8) [70,71,72,73,74,75,76,77] (Table 3).

Table 1 Influence of gene on brain structure volumes.
Table 2 Genetic effects on functional brain activation elicited by MID, SST and EF task.
Table 3 Functional activation elicited by MID, SST, and EF task.

Another 41 papers investigated methodological or statistical approaches or the associations of some other variables not listed above (e.g., maternal smoking and video gaming) on brain imaging parameters and behavioral variables and therefore were not included in the supplementary tables [12, 78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117].

From the multitude of findings, we here focus on brain regions relevant for drug use, including the hippocampus, striatum and frontal cortex. Drug use frequently starts during adolescence and has been a focus of IMAGEN research [55,56,57,58,59]. Figure 2 illustrates the potential of a simultaneous assessment of the effect of genetic variations on brain structure, function and behavior, using as an example the effects of genetic variation on (1) the volume of frontal, hippocampal, amygdala and striatal brain areas, (2) functional activation during reward anticipation and feedback, behavioral inhibition, and (3) alcohol, tobacco, and cannabis consumption. We also indicate whether structure or function of the respective brain regions were themselves directly associated with drug consumption.

Fig. 2: Effect of genetic variations on brain structure, function and behavior.
figure 2

Genes and epigenetic modifications associated with brain structure (hippocampus and putamen volume), functional activation (reward anticipation and feedback as elicited by the MID task; SST task) and behavior (alcohol consumption, nicotine and cannabis consumption).

Genotype effects on brain structure relevant for drug use

With respect to brain structure, frontocortical, hippocampal, and striatal (especially putamen) volumes were associated with variation in genes that contribute to metabolic and endocrine function, cell cycle and apoptosis, neurodevelopment and neurotransmission (Table 1). Notably, only variation in the HRK (“Between Harakiri, BCL2 Interacting Protein”) gene, which has been associated with cell cycle and apoptosis [20], contributed to both hippocampus and putamen volume [4, 25] (Fig. 2), indicating that the genetic contribution to brain volumes in adolescence may be rather site-specific. The other polymorphisms associated with brain volume at age 14 contribute to metabolic and endocrine function (APOE), cell cycle and apoptosis (BABAM1, FBXW8, HMGA2, DDR2, HRK, BCL2L1, HMGA2, and Efhd2), neurodevelopment (NR2F6, USHBP1, DCC, and FAT3), and neurotransmission (DLG2, CHR1, CNR1, and NPTN). The longitudinal design of the IMAGEN study will help to assess whether the effects of these genetic variations on brain volume are also age dependent [4, 14, 19, 22,23,24,25,26,27,28].

Genotype effects on functional brain activation relevant for drug use

With respect to reward anticipation and feedback (Table 2 and Fig. 2), functional activation elicited by the MID task was associated with polymorphisms in genes influencing neurodevelopment (TNM4 and BDNF) [15, 39] and neurotransmission (DRD2/ANKK1and MAOA) [36, 38]. Further, lower striatal activation during reward anticipation was associated with a high risk-taking bias (i.e., a strong tendency to engage in risky behavior) [59], while no significant difference was observed between adolescents with and without family history of alcohol use disorder [56].

With respect to the SST, distinct brain networks were associated with drug use versus attention deficit-hyperactivity symptoms; this study also reported that genetic variation in the norepinephrine transporter gene SLC6A2 was associated with the use of illegal substances but not functional activation elicited by behavioral inhibition [18]. The Arf6 activator Efa6/PSD3 was associated with ethanol-induced sedation and reduced tolerance development in drosophila and with altered prefrontal cortex activation during behavioral inhibition in the IMAGEN sample [46]. Moreover, epigenetic variation in the PPM1G gene locus was associated with increased functional activation of the right subthalamic nucleus during behavioral inhibition [16], and epigenetic modification of the OPRL1 gene mediated the effect of psychosocial stress and neutral striatal activation in the MID task on binge drinking in adolescence [31].

Regarding the EF paradigm, genetic variation in CB1R was associated with activation of the bilateral amygdala to angry faces, which in turn was correlated with higher drug intake [18].

Genotype and other associations with drug use

With respect to alcohol consumption, Fig. 2 shows genes associated with alcohol intake in the 30 days before study inclusion. The EHD4 gene, which has been associated with the cell cycle and apoptosis [21], showed a statistically significant association with alcohol consumption at age 14 [34]. Neuroimaging data additionally showed that low functional activation of the ventral striatum and of the putamen during reward anticipation at age 14 predicted high alcohol intake at age 16, with the association between functional putamen activation and alcohol intake being mediated by the BDNF Val66Met polymorphism [19]. The above mentioned epigenetic variation in the PPM1G gene was associated with high impulsiveness and early escalation of alcohol use [16]. Also, the human ortholog sof the Arf6 and Erf6 genes were associated with increased frequency of drinking and binge drinking episodes [46].

With respect to cannabis and other illegal drugs consumption, genetic variation in the norepinephrine transporter gene was associated with hypoactivation of the right inferior orbitofrontal network and speed of motor inhibition during SST, which was correlated with higher drug intake [18].

With respect to nicotine consumption, lower striatal activation during reward anticipation correlated with prenatal exposure to maternal smoking [57]. Risk variant rs578776, a variant in the CHRNA5–CHRNA3–CHRNB4 gene cluster, influences susceptibility to nicotine dependence by dampening the response of the anterior cingulate cortex to reward feedback, without recruiting the striatum or orbitofrontal cortex during feedback or anticipation [37].


The IMAGEN study has identified several genetic polymorphisms that interact with adolescent brain function and behavior, thus helping to disentangle their general functional roles [16, 46, 50]. With 2000 adolescents aged 14 when included in the study, the IMAGEN cohort was the largest sample available to date for imaging genetics analyses, thus improving power compared with previous studies with smaller sample sizes [118]. Due to its longitudinal design, the IMAGEN study can help to reveal gene-environment interactions in the manifestation of mental disorders from adolescence to and young adulthood.

From the large body of IMAGEN publications, we focused on findings of genes interacting with (1) brain volume, (2) functional activation, and (3) drug intake, which tends to start during adolescence and has repeatedly been attributed to gene-environment interactions [18, 55]

To explore psychopathology-relevant brain functions, we selected tasks that probe key aspects of reinforcement-related behaviors and emotional processing implicated in frequent neuropsychiatric disorders: response inhibition, emotional reactivity, and reward sensitivity. The neuroimaging tasks were chosen because they reliably elicit strong activation in functional networks underlying inhibitory control (SST) [119], emotional reactivity to social stimuli (EF) [120] and reward anticipation/outcome (MID) [121]. Some limitations of using these three tasks include incomplete assessment of different aspects of behavioral constructs, e.g., SST measuring motor impulsivity but not delay discounting, which was assesed behaviorally; passive reception of socially salient stimuli (faces) outside of social context; lack of possibilities for computational modeling of reward-related decisions in the MID task.

The IMAGEN study has assessed a wide range of environmental factors including childhood trauma, bullying, stressful life events, family mental health, pre- and peri-natal events, and family conflict. For example, peer victimization was indirectly associated with increased anxiety via decrease in left putamen and caudate volume [114]. Moreover, stressful life events were associated with the interaction between functional amygdala activation elicited by the EF task in adolescents with conduct and hyperactivity symptoms [71]. We here describe genetic and environmental effects on brain structure, function, and behavior assessed in the IMAGEN study.

Regarding reward anticipation and feedback, the IMAGEN consortium focused on polymorphisms implicated in both impulsive and addictive behavior. Reduced ventral striatal activation during reward anticipation has repeatedly been observed in alcohol-dependent patients and may be attributed to dopamine dysfunction following detoxification [2, 122]. However, it was unclear whether reduced activation during reward anticipation is present prior to the development of alcohol use or alcohol use disorder. Here, Büchel et al. [55] observed that reduced functional activation of the ventral striatum, midbrain, and prefrontal cortex at age 14 predicts drug use at age 16. Environmental factors that may contribute to such a blunted ventral striatal activation include maternal smoking during pregnancy [57]. Blunted ventral striatal activation in adolescents has been associated with increased impulsivity [60], a finding that replicates a similar observation in adult alcohol-dependent patients and controls [123]. Moreover, impulsivity has been associated with early life stress, [2] and methylation of the OPRL1 gene was shown to mediate the effect of psychosocial stress on impulsive alcohol intake (binging) and the associated ventrostriatal activation in adolescence [31]. The OPRL1 genotype has repeatedly been replicated on alcohol use disorders [124, 125].

Beyond environmental factors, genetic variance contributes to reduced ventral striatal activation during reward anticipation elicited by cues that predict reward. In humans, Stacey et al. [40] observed an association between reduced functional activation of the ventral striatum and a haplotype block containing the Rasgrf2 polymorphism rs26907, which was associated with alcohol intake in a previous meta-analysis [126]. In animal experiments, Stacey et al. [40] observed that Rasgrf2 knockout mice displayed a significant reduction in ethanol intake relative to wild type controls, lower intake at the highest ethanol concentrations, and blunted diurnal drinking. Rasgrf2 knockout mice also showed a significant reduction in ethanol preference, which was measured by the percentage of ethanol relative to water intake [40]. In addition, Rasgrf knockout mice showed a diminished noradrenergic and serotonergic response in the ventral striatum/nucleus accumbens and a decrease in β1 adrenoreceptor gene expression [127]. Easton and colleagues have shown that Rasgrf2 mediates the presence of noradrenaline (NA) and serotonin (5-HT) in the synaptic cleft at both basal level and after acute alcohol exposure. After subchronic alcohol exposure, the NA system is modified, and Rasgrf2 regulates alterations in expression levels of adrenoceptor mRNA. Rasgrf2 may be a mechanism by which the NA system is prevented from adapting to subchronic alcohol intake, which could in turn influence vulnerability to the effects of repeated alcohol exposure [128]. Moreover, it was shown that Rasgrf2 controls dopaminergic signaling and adaptations to alcohol also in other brain regions, beyond the nucleus accumbens [127]. Another Rasgrf2 polymorphism (rs2369955) was associated with current and future binge drinking in the IMAGEN cohort [19]. Therefore, the IMAGEN findings regarding Rasgrf2 are neurobiologically plausible, however, replication in an independent imaging sample has yet to be done.

These IMAGEN findings can be explained by the fact that Rasgrf2 interacts with activation of the MAPK/ERK pathway, which is involved in neurotransmission through dopamine receptors and transporters and thus potentially associated with dopamine-dependent reward mechanisms in alcoholism [129, 130]. In mouse models, ethanol administration increased ERK activity in the nucleus accumbens, and inhibition of ERK activity influenced ethanol self-administration [19].

Further genetic variance impacting on ventral striatal activation was discovered by Nees et al. [15], who showed that BDNF Val homozygotes compared with Met carriers had lower putamen reactivity during reward anticipation. This is in partial accordance with results from Pecina et al., who also observed effects of BDNF ValMet genotype on functional activation during the anticipations of rewards and losses, albeit only significant during loss anticipation [131]. Lower striatal activation during reward anticipation may impact on personality traits reflecting the orientation towards positive reinforcers. Benzerouk et al. [132] indeed observed that BDNF Val carriers with a positive family history for alcohol use disorders displayed lower levels of reward dependence compared to probands without such a family history. ValMet genotype was also associated with risk of relapse in alcohol dependence, however, no functional brain imaging was performed in the study [133]. Furthermore, the BDNF Val68Met polymorphism regulates BDNF expression and was also implicated in rodent models of uncontrolled and excessive alcohol intake [134]. BDNF expression was lower in the nucleus accumbens [135] and in the central and medial nucleus of the amygdala of alcohol preferring versus non-preferring rats [136], and short versus long-term alcohol intake differentially regulate BDNF mRNA levels in the nucleus accumbens in rats actively versus passively consuming alcohol [137].

With respect to a dimensional approach towards mental disorders [138], the IMAGEN consortium also assessed whether blunted ventral striatal activation elicited by reward indicating cues is associated with psychotic or affective symptoms and disorders. Previous studies showed that low functional activation of the ventral striatum during reward anticipation in patients with depression and schizophrenia was related to the severity of negative mood states including depression and anhedonia [139, 140]. In the IMAGEN adolescents, Stringaris et al. [54] observed that low ventral striatum activation during reward anticipation at age 14 predicted transition to subthreshold or clinical depression in previously healthy adolescents at age 16. Vulser et al. [89] observed that adolescents with subthreshold depression also had smaller gray matter volumes in the ventromedial prefrontal and rostral anterior cingulate cortex as well as in the putamen [25]. In 2018, Vulser et al. showed that early fractional anisotropy variations in tracts projecting from the corpus callosum to the anterior cingulate cortex may denote a higher risk of transition to depression in adolescents [112]. Findings regarding the cingulate cortex volume are interesting in light of their association with major depression [141] and the role of this brain area in error detection and behavioral control [142], which may also play a role in recovery from affective disorders. Indeed, patients with sustained activation elicited by reward feedback across task runs in the anterior cingulate cortex were more responsive to behavioral activation therapy [143].

With respect to higher cognitive functions, Nymberg et al. [36] showed that higher ventral striatal and caudate activation during reward feedback was associated with higher working memory performance and that this interaction was limited to A allele carriers of the DRD2/ANKK1 polymorphism. Also Taq1A of the ANKK1 gene interacted with a lateral orbitofrontal activation during reward anticipation [45] and predicted alcohol drinking 2 years later [105]. These findings provide evidence for an interaction between reward processing and complex cognitive capacities [144, 145]. Differences in striatal dopaminergic neurotransmission have also been associated with differences in working memory performance [146, 147], in line with findings concerning dopaminergic markers and cognitive capacity [65].

Beyond reward anticipation and feedback, impulsivity as operationalized by behavioral inhibition has been implicated in the development and maintenance of substance use disorders [148]. The subthalamic nucleus modifies information processing in fronto-striatal networks relevant for impulse control [148, 149]. Increased functional activation of the subthalamic nucleus was associated with hypermethylation in the PPM1G gene, which also correlated with increased impulsiveness and alcohol use in adolescence [16]. Further brain areas implicated in behavioral inhibition are the prefrontal cortex [18], whose activation was associated with the human ortholog of Arf6 and orthologs of Efa6 (PSD1-4) [46].

Regarding the processing of affective faces, the study of Spechler et al. shows that stronger amygdala activation to signals of threat was associated with high cannabis intake. This observation points to another neurobiological dimension contributing both to negative mood states and drug intake [72]. Indeed, increased amygdala activation towards aversive stimuli has been shown to contribute to anxiety, feelings of being threatened and aggression, which in return predict excessive alcohol use [148]. Stronger amygdala activation during emotional processing was also observed in adolescents who experienced both a high number of stressful life events and strong symptoms of conduct disorder and hyperactivity [71], highlighting how experience can modulate brain-behavior relationships.

Overall, the IMAGEN consortium was able to detect a number of genetic polymorphisms with specific effects on brain structure and function and their association with symptoms of mental disorders in adolescents. IMAGEN’s study design has supported the elucidation of biological and environmental mechanisms of substance use-related behaviors, but we note limitations and ways to address them. The rather large-sample size overcomes some of the limitations of previous genetic imaging studies, however, independent replications are required. As studies like IMAGEN have their data repeatedly used by researchers, which—if systematic errors existed in the data would then be propagated into all research output, strict quality assurance and quality control measures are taken by the central database team that has overseen data management across all study time points. This allows for quick identification and resolution of errors that could occur before data are released for analyses (i.e., data entry and data upload/transfer).

The very nature of multivariate multimodal approaches bears a multiple comparison issue. Fortunately, clinical samples and other large phenotypically rich studies such as UK Biobank provide the opportunity to externally validate IMAGEN findings. Resampling methods such as bootstrapping and cross-validation (e.g., leave-one-out and split-half) are techniques we have used to internally validate model fit and estimate a model’s generalizability [41]. Moreover, the IMAGEN consortium asks for research proposals before data access is granted, and thus prevents unethical data dredging. Nonetheless, exploratory data analysis can be performed and has to be clearly described as such. Replication of such a cohort might be very costly, that is why rigorous statistics including the above mentioned internal cross-validation for estimation of predictive power and the reporting of confidence intervals in order to describe a range of plausible estimates should be promoted. With respect to intraindividual variance, reliability measures as well as developmental changes replications can be carried out within the IMAGEN sample, e.g., by assessing whether the same polymorphisms impact on subcortical brain volumes at age 14 and 16 or in young adulthood. Given the multisite design, variance from different acquisition sites were considered based on a central protocol.

The issue of participant attrition over time may limit some longitudinal multimodal statistical modeling and internal validation methods (i.e., due to reduced power and incomplete data), but with more than 1300 participants assessed at age 22, there is still great opportunity to understand the influence of biology and the environment on developmental trajectories of substance use. Relevant to identifying gene-environment interactions on brain and behavior, the relatively homogeneous socioeconomic background of the Western European participating families meant IMAGEN was not well placed to investigate particular environmental influences such as nutrition or hazardous exposures. Furthermore, no resources were available to independently confirm the reliability of the self-reported environmental data.

For the purpose of genetic analyses, the IMAGEN participants are of European ancestry—as a result the generalizability of findings to other cultures and stress/environmental exposures is unknown. Therefore, future replications should also be carried out outside of Europe, requiring careful harmonization of clinical tools and paradigms. As recently reported [150], the IMAGEN study has aligned itself with other global neuroimaging-genetics adolescent cohorts that will further enhance and clarify the work presented here.