Alcohol misuse in early adolescence is a risk factor for alcohol use disorders (AUD) in adulthood (1, 2). AUD is a heterogeneous and heritable reward-deficit disorder that transitions from controlled to uncontrolled, compulsive drinking (3, 4). AUD negatively impacts socioeconomic aspects by increasing the burden of disease, injuries, and mortality (5). The risk of early-onset alcohol use is higher among the offspring of alcoholic parents, partly due to heritability and partly due to environment, as alcoholic parents are likely to maltreat or have a poor relationship with their children (6, 7). Furthermore, adolescents misusing alcohol may also indulge in nicotine use, illicit drug use or violent and non-violent delinquencies (8), further elevating the risk for other mental illnesses in adulthood. Therefore, understanding the predictors of the early onset of alcohol use can help develop strategies to prevent early AUD and promote potential positive health outcomes. However, understanding the predictors of early-onset alcohol use can be intricate as the risk for AUD is moderated by the complex interaction between genes and environment (G×E), both negative and positive, thereby contributing to individual variation in progression from initial alcohol use to AUD (9).

The quality of the early environment contributes to risk and resilience for AUD in adulthood. Children may get adverse effects on behaviour in a proximal environment by parents when they are younger and distally by their neighborhood or peers when they grow older (10). Such negative environmental influences may affect their behaviour in adulthood via the effect on gene expression in the brain. Our preclinical studies in rats show the effect of early life stress on high alcohol intake in adulthood through changes in epigenetics and gene expression of genes involved in the reward systems (11–15). Childhood maltreatment (by a parent or any adult) or poor parent-child relationship are risk factors for early alcohol use, binge drinking, and AUD (7, 16). Childhood maltreated females show a 16-25 fold increased risk of AUD than maltreated males (17) and transition faster from initial alcohol use to AUD (18, 19). In Sweden, maltreatment during adolescence is more common than in childhood, with adolescent males and females differing in their maltreatment experience and the perpetrators (20). Adolescent females reporting poor parent-child relationships are more likely to have early-onset of alcohol use than males (7). In contrast to negative environmental influences, a positive environment, such as a good parent-child relationship, might protect adolescents against maltreatment (21) and early onset of substance use, especially in females (7, 22). Furthermore, a good parent-child relationship is protective against self-motivated alcohol use and attenuates peer-induced drinking behaviour in adolescents (23). Based on the evolutionary perspective of human development, Belsky proposed an operational perspective to study G×E called differential susceptibility to environmental influences, which hypothesizes that individuals differ in their susceptibility to environmental influences, with some being more susceptible to both negative and positive developmental experiences (especially those of rearing environments) and environmental exposures than others (24–26). Recently, a study based on the differential susceptibility framework assessed the interplay between the negative and positive environmental exposures and their effect on alcohol-related problems among youths, where they observed that adolescents and young adults carrying a susceptible allele of the SNP rs2290045 in the VGLUT2 gene exposed to stressful life events reported more alcohol-related problems if they were also exposed to poor parenting, whereas, in contrast, those exposed to good parenting reported lower alcohol-related problems (27). Another study in adolescents supported the differential susceptibility model by showing that impulsive children exposed to coercive parenting style were at higher risk of alcohol use than those not exposed to coercive parenting style (28). Thus, given the aggravating effect of a negative environment and the mitigating effect of a positive environment on alcohol drinking behaviour, it is critical to include both negative and positive environments to elucidate the effect of G×E on alcohol use behaviour (29).

For the genetic aspect of G×E, we are interested in the Monoamine Oxidase A (MAOA), an X-chromosome-linked gene implicated in AUD (30). MAOA encodes the monoamine oxidase A enzyme that metabolizes monoamine neurotransmitters (31), which are crucial for reward (3, 32) and stress regulation (33). A functional upstream variable number tandem repeat polymorphism in the promoter region (MAOA-uVNTR) is known to regulate MAOA gene expression in cell lines, where the short repeats (2 and 3 repeats) underexpress, whereas the long repeats (3.5 and 4 repeats) overexpress the MAOA gene (34, 35). MAOA-uVNTR is widely studied in the context of G×E, predominantly through the ‘diathesis-stress model’ perspective, associating MAOA-uVNTR risk alleles with mental health problems in the presence of a stressful environment in a sex-dependent manner. For example, the short-allele (MAOA-S) in men or the long-allele (MAOA-L) in womenin the presence of negative environmental factors has been associated with antisocial outcomes (36–39), conduct disorder (40), panic disorder (41), aggression (42–44), anxiety (45), mood disorders (46–49), impulsivity (50); schizophrenia (51) or attention-deficit/hyperactivity syndrome (39). Altogether, the above-mentioned studies based on the classical diathesis–stress model suggest that the MAOA-S in males and the MAOA-L in females are “ risk alleles” that render the carriers susceptible to the effects of the negative environment (52). A neurobiological explanation is that they have a hyperresponsivity of the amygdala during emotional arousal and reduced amygdala-orbitofrontal connectivity resulting in heightened emotional reactivity to a negative environment (53–56).

Alternatively, studying MAOA-uVNTR through the ‘differential susceptibility to environmental influences’ perspective would define the conventional MAOA-risk alleles as “plasticity alleles” that render the carriers sensitive to the effects of both negative and positive environments resulting in worst or beneficial outcomes, respectively (57). However, differential susceptibility properties of MAOA-uVNTR are not yet investigated. Previous MAOA×E studies regarding alcohol consumption have either studied negative or positive environments separately but never together (i.e. in interaction with each other), given that one’s life is a blend of negative and positive experiences. The sex-specific MAOA-risk alleles in the presence of a neutral or positive environment have been associated with lower alcohol use, whereas in the presence of a negative environment, has been associated with alcohol misuse and AUD (58–62). However, the picture is incomplete since it is unknown whether the MAOA-risk allele carriers experiencing a negative environment benefit from a positive environment compared with non-carriers to protect from alcohol misuse.

Hence, the present study aimed to assess whether the quality of a positive parent-child relationship (a positive environmental factor, Epos) moderates the association of interaction of MAOA genotype (candidate gene, cG) and familial or non-familial maltreatment (a negative environmental factor, Eneg) with alcohol misuse in adolescent males and females (i.e. cG×Eneg×Epos). Our second aim was to assess the robustness of the three-way interaction to covariates, such as co-existing negative and positive environments and behavioural outcomes, that can be a risk or resilience factor for harmful alcohol consumption patterns in adolescents. The environmental covariates included parental alcohol use (7), being a victim of bullying in school (63, 64), negative-life events (for example, breaking-up of a romantic relationship with a boyfriend or girlfriend, divorce of a parent, death of a family member, etc.) and positive-life events (for examples, achievements in school or competitions, performed well in exams, etc.) (65). Also, we adjusted the three-way interaction to the negative aspects of a parent-child relationship, such as rejection, chaos and coercion; because adolescents, especially those carrying genetic susceptibility to their environment, could be even more sensitive to the effects of the negative parent-child relationship than the non-carriers (66). Next, because alcohol use, nicotine use (snus or cigarette smoking), illicit drug use and delinquent behaviour are related, and they typically co-occur in adolescents (8, 67–71), we adjusted the three-way interaction to these co-occurring behavioural outcomes to isolate the association of the three-way interaction with alcohol use.

We hypothesize that the MAOA-risk alleles would display differential susceptibility to the environment; MAOA-risk allele carriers maltreated in early life would report higher or lower alcohol consumption depending on poor or good parent-child relationships, respectively, and benefit more from having a good parent-child relationship than the non-carriers.

The present cG×Eneg×Epos study demonstrates for the first time the differential susceptibility of the MAOA-uVNTR genotype to negative and positive environments constituting risk or protection against alcohol misuse among Swedish adolescents. We used a differential susceptibility to environmental influences framework. We used separate constructs for FM, NFM (negative environment), and positive-parent child relationship (positive environment) reported at wave-2. Using separate constructs for correlated parent-related variables, FM and positive-parent child relationship, enabled us to isolate the moderating effect of positive-parent child relationship on effect of FM or NFM on alcohol use. However, taking into consideration a possible correlation between various negative environmental factors, in the post-hoc analyses, we have adjusted the three-way interactions to the maltreatments reported at wave-1, negative-parent-child relationship and other covariates ( Supplementary Tables S1 , S2 , S8A, B , S10A, B ), and we have reported both unadjusted and adjusted three-way interactions. In the unadjusted analyses, we observed that the females carrying the MAOA-L allele (MAOA-SL/LL genotype) with FM or NFM reported lower alcohol use in the presence of a good parent-child relationship (specifically, warmth dimension), whereas, reported higher alcohol use in the presence of a poor parent-child relationship compared to their counterparts carrying MAOA-SS genotype ( Figures 1 – 4 ). The MAOA-L allele carriers benefited from a good parent-child relationship to counteract the negative effects of maltreatment on alcohol use, whereas MAOA-SS genotype carriers did not. Thus, the MAOA-L allele is sensitive to negative and positive environmental factors, thereby displaying characteristics of a “plasticity” allele rather than a “risk” allele. Thus, MAOA plasticity allele carriers in a negative environment may not always be at risk of alcohol misuse and develop AUD in adulthood because the positive environment may mitigate the risk. Hence, promoting a positive parent-child relationship could be a preventive strategy for mitigating the risk of AUD amongst MAOA plasticity allele carriers. Among MAOA-S allele-carrying females, the opposing effects of a good parent-child relationship on their alcohol consumption need further investigation, although such an unexcepted pattern may exist in some families (78).

The present study’s findings cannot be directly compared with earlier MAOA×E studies on alcohol because they have only accounted for the negative environment and have reported higher alcohol consumption among females carrying the MAOA-L allele (58, 62) with some discrepancies (61, 79). Nevertheless, further replications of the findings are needed in different populations with varying degrees of negative and positive environmental factors. The findings of the present study underscore the importance of studying a positive environment to help explain inconsistencies in earlier MAOA×Eneg studies. Also, the differential susceptibility of the MAOA-L allele could explain the lack of association between the in vivo transcriptional activity of MAOA-alleles (80), otherwise clearly demonstrated in in-vitro (34). Additionally, understanding the biological “how” and “why” underlying differential susceptibility of the MAOA-L allele may help develop preventive intervention strategies for AUD treatment.

The intermediate phenotype of emotional regulation and epigenetic mechanisms such as DNA methylation might help to explain the observed differential susceptibility of the MAOA-L allele. A study in females carrying the MAOA-L allele exposed to childhood maltreatment observed heightened emotional reactivity during adolescence that predicted pathological personality in adulthood, suggesting emotional reactivity as a mediator between MAOA×Eneg and psychiatric illness (56). A good parent-child relationship, however, facilitates effective emotion regulation among adolescents in high-risk environments (81). Thus, we speculate that the protective effect of a good parent-child relationship (specifically, warmth dimension) in MAOA-L allele carriers might have mitigated the heightened emotional reactivity to the negative environment, resulting in lower alcohol consumption. Besides, DNA methylation is implicated in emotion regulation (82, 83), and it might translate environmental cues to genes and influence the development of psychiatric disorders (84). In general, DNA methylation is sensitive to the early environment (85), and alcohol consumption (13, 86), and specifically, MAOA methylation and transcription factors regulate the MAOA allelic expression in response to environmental changes (87). Hence, the parent-child relationship-induced changes in MAOA methylation may have contributed to emotional reactivity and differential susceptibility of plasticity alleles. However, further studies are needed to elucidate such a molecular mechanism.

The cG×Eneg×Epos was associated with alcohol consumption in females but not males ( Supplementary Tables S9 , S11 , Supplementary Figures S4 , S5 , S7 , S8 ) probably because of a higher proportion of high-alcohol-drinkers in females than males (25.23% vs 15.49%). In the present study, the lower number of male drinkers in the statistical models might have too few cases to detect the association of cG×Eneg×Epos interaction with alcohol consumption. The observed drinking trend, however, aligns with the recent reports among Swedish adolescents, where alcohol drinking is more common among girls than boys (88, 89). Moreover, a higher proportion of girls than boys in Sweden reported alcohol consumption at least once during their lifetime or intoxication during the past 30 days (90). Another reason for associations in females could be because, in our study, the high-alcohol-drinking females reported higher frequencies of psychological maltreatment, bullying in school and negative life events relative to the high-alcohol-drinking males ( Supplementary Table S5 ). Presumably, similar to females, the interaction effect on alcohol consumption would be evident in males at a higher severity of maltreatment, as shown by a study concerning antisocial outcomes (91). Also, in males, co-occurring nicotine usage and involvement in delinquent behaviours might have interfered with the associations with alcohol use. Furthermore, in contrast to females, males, irrespective of genotype, did not show sensitivity to the quality of the parent-child relationship; they instead displayed a trend of higher alcohol consumption in the presence of a good parent-child relationship ( Supplementary Figures S4 , S7 ). Females have a stronger perception of stress, are strongly affected by the quality of their relationship with their parents, and prefer to seek emotional support as a coping mechanism relative to males (7, 92, 93). Thus, having a good parent-child relationship (warmth, autonomy support, and structure) might have emotionally supported the females to mitigate stress and alcohol consumption.

Moreover, we adjusted the three-way interactions for predictor- and outcome-intersection problems, which are common in G×E studies. Predictor-intersection means the intersection of co-occurring and re-occurring maltreatment that cumulatively may predict mental health outcome(s) (94). For example, we found a considerable overlap between FM and NFM measured at two-time points; in females who experienced FM at wave-1, 19.9% also experienced NFM, while 75.9% experienced FM and 38.2% experienced NFM at wave-2 ( Supplementary Figure S9 ). The observed overlap between maltreatment reported at wave-1 and wave-2 may intersect, and wave-1 data may confound the findings at wave-2 on alcohol consumption ( Supplementary Figure S9 ). Outcome-intersection means the intersection of several mental health outcomes, possibly stemming from a single intermediate phenotype, such as a deficit in emotion regulation (94). For example, we observed that outcomes such as high alcohol drinking co-occur with other outcomes such as nicotine use, illicit drug use, and delinquency ( Tables 1 , 2 ). In both predictor and outcome intersections, isolating the effect of a single predictor or outcome is challenging, respectively. We handled the predictor-intersection problem by adjusting the significant three-way interactions for co-occurring and re-occurring environmental factors to isolate the effect of single maltreatment. The outcome intersection was illustrated by adjusting the three-way interactions to co-occurring and re-occurring behavioural outcomes to isolate the effect of alcohol consumption. Using the “all-in-one” approach, both the three-way interactions with the FM and NFM were non-significant, but this could probably be due to the overfitting of the model ( Supplementary Tables S8A , S10A ). However, using the single approach, we observed that the three-way interaction with NFM was robust to all the intersecting predictor environmental and behavioural covariates, except for the parent’s alcohol use, negative parent-child relationship and adolescents’ nicotine use reported at wave-2 ( Supplementary Tables S8B , S10B ). The findings indicate that having a good parent-child relationship protects robustly against the negative effect of NFM on alcohol consumption in MAOA-L allele-carrying females. Another way to handle the predictor intersection problem could be to create a composite variable to merge all the potential predictors into one. However, we did not choose it due to our goal of characterizing the sample. Moreover, the outcome intersection may obscure potential G×E associations, and thus, co-occurring outcomes like substance use and antisocial outcomes should be considered together with alcohol consumption in future G×E studies.

Indeed, alcohol use disorder is a genetically complex psychiatric disorder which includes genetic and neurobiological heterogeneity, polygenicity, and interaction of genes with the environment (95). Thousands of genes and their common and rare variants with small effect sizes contribute to the complexity of AUD. However, identifying the contribution of a single candidate gene (MAOA), selected on the basis of biological plausibility, despite its small effect size, is important to understand the small part of the genetic puzzle of AUD. Our group has studied MAOA in interaction with environment in four cohorts and have found relatively good explained variation in the model of alcohol consumption (MAOA-uVNTR × poor family relation or maltreatment/abuse/sexual abuse explained 23-25% variation in alcohol consumption and alcohol-related problematic behaviour among 16- and -19 year Swedish adolescent females (60), MAOA-uVNTR × poor family relations or sexual abuse explained 7-10% variance in alcohol consumption among Swedish 17-18 year adolescent males and females (58), Interaction of MAOA-uVNTR with maltreatment moderated by MAOA DNA methylation explained 9.4% variance in alcohol consumption among mean age 22.1 year old Swedish young adult males, who sought treatment for substance misuse (72), and in the present study, interaction of MAOA-uVNTR × Family or non-family maltreatment moderated by positive parent-child relationship explained 5 to 6% variation in alcohol consumption among Swedish 16-19 year adolescent females). Replication of the association of MAOA-uVNTR with alcohol consumption in interaction with negative and/or positive environment by our present and previous findings together with others (61, 62, 79), strengthens the individual contribution of MAOA-uVNTR, not alone but in interaction with the environment, to explain a complex polygenic trait of alcohol use disorder. Although a single locus might not fully explain the heterogenous and polygenic trait, the interplay between the specific genetic variants and environmental factors can provide additional predictive power. Indeed, given the role of the MAOA gene in encoding the MAOA enzyme that metabolizes monoaminergic neurotransmitters involved in mediating reward, polymorphisms in the genes in dopaminergic (DRD2, DRD4 and DAT1 genotypes), or serotonergic (5-HTTLPR) neurotransmission cumulatively in the interaction of environment could explain more variation in the alcohol consumption and can better predict the risk of alcohol consumption (66, 96). However, these candidate genes have been first individually associated with alcohol consumption and related phenotypes in interaction with either a negative or positive environment [see review (97)].

Alternative to using genetic risk or plasticity scores on the basis of a single or combination of several candidate genes, Genome-Wide Association Studies (GWAS) derived polygenic scores emerged in gene-environment research to improve the prediction accuracy of complex traits by considering the cumulative impact of multiple loci. Polygenic scores are based on SNPs typically linked to multiple genes across the genome that may collectively influence the trait. Despite hundreds of genes, all measured SNPs together typically explain not more than 10% variances in substance use (98); specifically, polygenic scores explained around 1.9% variance in alcohol consumption (99) or <0.1% variance in alcohol consumption upon interaction with the environment (100). However, as recently pointed out by Zhang and Belsky (97), it is important to note that i) GWAS derived polygenic scores are exclusively based on SNPs and do not include any other types of genetic polymorphisms such as variable number tandem repeats as in MAOA gene, thereby providing limited information to G×E research, ii) These are hypothesis free and polygenic score are derived on the basis of mere correlation of SNPs with phenotype and not on the basis of biological plausibility, thereby making predictions of complex traits and replication of results difficult in samples with heterogenous phenotypes iii) SNPs identified in GWAS are not the true causal elements but are proxies that correlate with the true causal variants, whereas, the in case of variable number tandem repeats, the risk allele itself is considered as the causal element, iv) GWAS-derived polygenic score are sensitive to variation in racial and ethnic patterns of linkage disequilibrium and therefore, polygenic scores based on one race or ethnicity may be difficult to use in another. For example, the results from a recent G×E study that predicted alcohol use among adults using polygenic scores composed merely of European ancestry may not be generalizable to other ancestry (99). In addition, the G×E studies with polygenic scores are still weak to robustly predict complex traits like substance use (98), and they are not suitable for G×E inquiries testing differential susceptibility hypothesis because the SNP having a main effect on the phenotype are included in polygenic risk scores and not the SNPs that have an effect on differential susceptibility, as they may not have large effects to be detected in GWAS (101).

The strengths of the present study are: 1) Large sample size- which gave us adequate power to detect moderate to large effect size (MAOA× FM×positive parent-child relationship: R² increase = 0.5%, observed power = 0.541; and MAOA×NFM×positive parent-child relationship: R² increase = 1%, observed power = 0.832). Large sample size also helped us overcome the problem of false-negative/positive findings observed in small sample-sized G×E studies (102). 2) We used a differential susceptibility framework incorporating a more holistic range of risk and protective environmental factors than older studies, which often overlook factors mitigating the risk. 3) Collecting maltreatment self-reports at two-time points allowed us to account for instances of maltreatment that participants were hesitant to report earlier. 4) The sample was representative of a segment of the general adolescent population, leading to the generalizability of the findings to a non-clinical, non-refereed population. The limitations of the study are 1) Reliability on retrospective self-reports, which involve a risk of recall bias. 2) As we measured the parent-child relationship only at a one-time point, we could not establish a causal relationship. 3) Using an at-home-based postal questionnaire, presumably overseen by the parent, may have prevented the adolescents from the most adverse social environments from participating in the present study. 4) The association of cG×Eneg×Epos with alcohol consumption in females carrying the non-risk allele might not be accurately represented due to a small sample size ( Figures 2 , 3 ). Therefore, the findings in females carrying the non-risk allele should be cautiously interpreted and need replication in larger samples.

The present study provides the first evidence that the MAOA-L allele (SL/LL genotype) in females is sensitive to negative and positive environments concerning their alcohol use. Furthermore, the study indicates that a good parent-child relationship, especially the warmth dimension, protects against alcohol consumption among the MAOA-L allele-carrying females exposed to maltreatment. However, the interactions were not significant after adjusting to several environmental and behavioural covariates, especially parent’s alcohol use, negative parent-child relationship, and nicotine use (smoking and/or snus) suggesting predictor and outcome intersection. Thus, the results indicate that alcohol use and nicotine use are intercorrelated and that future studies and frameworks for preventive strategies should target these co-occurring behaviours together. Insights into differential susceptibility of the MAOA-L allele to the environment may help to explain the lack of association between in-vivo and in-vitro allele-specific transcriptional activity. The study underscores the need to re-evaluate the diathesis-stress-based MAOA×environment research strategy and recommends including negative and positive environments in the analyses. The study also emphasizes the importance of considering early life events and parent-child relationships to design personalized intervention strategies for AUD. The study calls for future studies to effectively assess and address the predictor- and outcome-intersection phenomenon. As it is the first study to address differential susceptibility of MAOA-uVNTR concerning alcohol use, replication in larger samples and molecular research assessing underlying epigenetic mechanisms will be beneficial for future MAOA×E studies regarding alcohol use.

The raw data supporting the conclusions of this article will be made available by the authors upon request, without undue reservation.

The studies involving humans were approved by Regional Board for Research Ethics in Uppsala, Sweden (Ref: 2012/187). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation in this study was provided by the participants’ legal guardians/next of kin.

MB: Conceptualization, Formal Analysis, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review & editing. DC: Methodology, Writing – review & editing. AT: Writing – review & editing. CA: Conceptualization, Formal Analysis, Funding acquisition, Investigation, Methodology, Validation, Writing – review & editing. SH: Methodology, Writing – review & editing, Formal Analysis. KN: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – review & editing.