Introduction
Conduct disorder (CD) is a pediatric behavioral disorder with a prevalence of approximately 4–16% in boys and about 1–9% in girls [
1]. Hallmark of CD are antisocial behaviors, that is, serious violations of basic rights of other people and/or age-appropriate societal norms resulting in severe aggression, deceitfulness, and rule-breaking behavior. The severe nature of these symptoms gives rise to a significant burden for affected patients, families, and societies at large [
2]. About 40% of boys and 25% of girls with CD are estimated to display antisocial behaviors persisting into adulthood and develop antisocial personality disorder [
3,
4].
CD symptomatology or antisocial behaviors can be considered as continuous traits that are caused by both genetic and environmental risk factors. More specifically, the interplay between genes and environment, also known as gene–environment (G × E) interactions, can provide insight into why some individuals are more susceptible to certain adverse genetic or environmental factors than others. These G × E interactions are assumed to be of great importance in multifactorial traits such as antisocial behavior [
5‐
7].
Recent insights suggest that the classic candidate G × E literature holds important limitations. Most notable, the use of poorly replicated candidate genes, underpowered samples, and inappropriate correction for multiple comparisons are suspected to have resulted in an inflated rate of false-positive findings across studies [
8‐
10]. Instead, hypothesis-free, genome-wide association studies (GWASs) can overcome these issues and thereby provide more robust candidates for both gene discovery and G × E research [
8,
11]. Regarding GWAS literature, two large studies identified a number of novel, sex-stratified susceptibility loci for antisocial behavior and antisocial personality disorder recently [
12,
13]. Moreover, a G × E interaction was suggested between one of these loci and childhood familial difficulties in males within the general population [
13].
So far, one of the most studied G × E interactions in relation to antisocial behavior involves a 30 bp length polymorphic region (LPR) in the
monoamine oxidase A (MAOA) gene and exposure to childhood maltreatment. The LPR affects the functionality of the MAOA enzyme resulting in alleles with lower (
MAOA-L) and higher (
MAOA-H) activity [
5,
6,
14]. Since the
MAOA gene is located on the
X chromosome, males have only one copy of the gene, whereas females have two, supporting sex-stratified analyses. Indeed, meta-analytic evidence has suggested that males with the
MAOA-L genotype were more susceptible to effects of maltreatment than males with
MAOA-H, while females with the
MAOA-H genotype appeared to be more susceptible to maltreatment effects, albeit weakly and less consistent than in males [
15]. Similar sex-stratified G × E interaction patterns in relation to antisocial behavior have been reported for
MAOA × maternal smoking during pregnancy (i.e., males with
MAOA-L were more susceptible to effects of smoking than males with
MAOA-H, whereas females with
MAOA-H were more susceptible to effects of smoking than females with
MAOA-L) [
16]. Thus, given location on the
X chromosome and (meta-analytic) implication of different functional alleles, more sex-stratified research is needed, taking into account limitations in candidate G × E research.
Regarding sex differences related to antisocial behavior, higher rates of antisocial behavior and crime have been reported in males compared to females [
17]. Furthermore, males appear to be over-represented in clinical samples [
1]. Considering these observations, sex-stratified investigation of potential risk factors is very much needed. Further reasons for conducting sex-stratified analyses include recent GWAS results pointing to different susceptibility loci for antisocial behavior in males and females [
12,
13].
Moreover, another important yet frequently overlooked limitation of a substantial part of the G × E literature arises from a lack of covariate interaction modelling in the G × E analyses [
18]. Modelling covariate interactions is important, because both the genetic and environmental factor of interest might be moderated by control variables and the G × E interaction should be adjusted accordingly. Another point of consideration is the inconsistent control for the highly comorbid attention-deficit/hyperactivity disorder (ADHD), which may actually drive part of the associations reported with antisocial behavior [
1,
19]. Similarly, further improvements could also be made by adjusting for frequently comorbid internalizing problems [
1]. Furthermore, gene–environment correlations (i.e., genetic confounding of the environment) should also be taken into account as a potential driving force behind apparent G × E interactions [
20].
In this study, we aimed to address the aforementioned issues concerning the existing G × E literature and investigated G × E interactions in relation to childhood antisocial behavior in the well-powered Avon Longitudinal Study of Parents and Children (ALSPAC). We focused on two key environmental risk factors for antisocial behavior, namely maternal smoking during pregnancy and childhood maltreatment [
7,
15,
16,
21], in the interplay with recently identified genetic variants from GWASs of antisocial behavior, while controlling for potential confounding by comorbid ADHD and addressing specific statistical concerns. Furthermore, we aimed to replicate previously reported G × Es for the much studied
MAOA candidate gene.
Discussion
In this study, we performed sex-stratified analyses of G × E interactions in relation to childhood antisocial behavior in a large population cohort for recent GWAS-implicated SNPs and MAOA with two well-known environmental risk factors, namely maternal smoking during pregnancy and childhood maltreatment. Regarding males, our most important findings are that G-allele homozygotes of the rs4714329 SNP and A-allele homozygotes of the rs9471290 SNP appeared to be more susceptible to effects of smoking during pregnancy in relation to antisocial behavior. Regarding females, we found that heterozygotes of the rs11215217 SNP appeared to be less susceptible, and carriers of both low- and high-activity allele of the MAOA-LPR appeared to be more susceptible to effects of childhood maltreatment in relation to antisocial behavior.
In males, the related SNPs rs4714329 and rs9471290 appeared to moderate the relation between smoking during pregnancy and antisocial behavior in such a way that risk allele homozygotes appeared to be more vulnerable to effects of maternal smoking than the other genotypes. More specifically, in risk allele homozygotes, antisocial behavior scores were more than twice as high in smoking-exposed subjects compared to unexposed subjects. By using the open-access GTEx database (available at
https://www.gtexportal.org/home/), the SNP rs4714329 was linked to the expression of nearby genes
LINC00951 and
LRFN2 in the brain [
13].
LRFN2 encodes a protein suggested to be involved in neural developmental processes such as neurite outgrowth and synaptic plasticity [
42]. LRFN2 is part of a larger protein class characterized by a leucine-rich repeat domain. Many leucine-rich repeats containing transmembrane proteins are thought to be involved in nervous system development and neurodevelopmental disorders [
43,
44]. LRFN2 regulates the post-synaptic PSD-95 complex, and has also been implicated in erythropoiesis, working memory, and autistic features [
42,
45‐
48].
LINC00951 is an intergenic, long non-protein coding RNA gene, which is also expressed in the brain [
13]. While many of these RNAs remain to be characterized, in general, they are assumed to be involved in gene expression regulation at both epigenetic and (post) transcriptional levels as well as other processes such as genomic imprinting [
49]. In addition, these RNAs may play an important role in neurodevelopmental disorders [
50].
Smoking during pregnancy has been one of the more strongly associated prenatal risk factors in relation to CD [
7,
51], although this may, in part, be due to genetic and/or familial confounding [
20,
52,
53]. Tobacco smoke consists of a mixture of many chemicals including nicotine, carbon monoxide, polycyclic aromatic hydrocarbons, and heavy metals, all of which may affect the developing fetus by various mechanisms [
54‐
57]. A number of recent studies investigating gene expression patterns in relation to smoking reported the gene
LRRN3 among their top hits of smoking-related differentially expressed genes [
58‐
60]. Similar to
LRFN2, LRRN3 is a leucine-rich repeat domain containing transmembrane protein expressed in the brain, and suggested to play a role in the development and maintenance of the nervous system [
61]. Functionally, LRRN3 has been implicated in autism, antidepressant action, and cortical thickness (alterations of which are associated with conduct and psychopathic features) [
62‐
64]. Although the before mentioned studies of smoking did not specifically target effects of smoking during pregnancy and gene expression alterations might be reversible, the reported results suggest that smoking might exert effects on pathways that are also affected by genetic risk factors related to antisocial behavior. Conversely, G × E interplay might be expected, i.e., moderation effects among genotype and environment such as observed in the present study.
Furthermore, as mentioned before, the use of smoking during pregnancy as an exclusively and independent environmental factor has been a point of discussion. As confounding by both genetic and socio-environmental factors has been suggested [
20,
52,
53], this could indicate that the observed G × E with smoking during pregnancy may at least in part be a proxy for a gene–gene interaction and/or G × E interaction with the other environmental factors. However, as we did not observe gene–environment correlations between the selected genetic variants and smoking during pregnancy and controlled our analyses for covariate interactions, we at least addressed part of these confounding issues.
Therefore, although the exact nature of the identified G × E interaction with smoking during pregnancy is not clear, both the genetic and environmental factors in this G × E may affect brain development through effects on leucine-rich repeat protein interaction networks thought to be involved in functions such as synapse and neural circuit formation, and thereby predispose offspring for antisocial behavior [
43,
44]. This also implies that future studies should also take into account related neural leucine-rich repeat protein (regulatory) genes when attempting to replicate or extent present findings.
A G × E interaction between the SNP rs11215217 and childhood maltreatment was observed in relation to offspring antisocial behavior in females. The nearest gene to this SNP is a non-coding, uncharacterized RNA gene (
LOC105369506). As before mentioned, multiple (regulatory) functions of non-coding RNA genes have been described and their role in neurodevelopmental disorders highlighted [
49,
50]. Of note, when adjusted for comorbid ADHD symptoms, the interaction became only nominally significant, which might indicate that the effect could be partially driven by comorbid ADHD.
Furthermore, the GWAS in relation to antisocial personality disorder by Rautiainen et al. [
13] suggested a male-specific interaction between the SNP rs4714329 and childhood familial difficulties (severe conflicts and/or economic difficulties) in the general population [
13]. Since we did not find any (male) G × E interactions between maltreatment and rs4714329 (or the related SNP rs9471290), we conclude that this suggested interaction does not appear to extend to childhood maltreatment in relation to pediatric antisocial behavior.
In addition, while interactions between the near-promoter LPR in
MAOA and childhood maltreatment in relation to antisocial behavior have been reported for both sexes previously [
15], we only observed a G × E interaction in females. More specifically, we observed a disadvantage mostly for maltreatment-exposed females with both low- and high-activity alleles (showing antisocial behavior scores more than twice as high compared to unexposed females), which is slightly different from the (additive) H-allele effect suggested in a previous meta-analysis [
15]. Furthermore, in males with a low-activity allele, we did not observe any interaction with maltreatment. While this null finding does not replicate previous meta-analytic results [
15], the largest study in the aforementioned meta-analysis also failed to find any interaction between
MAOA and stressful life events in relation to conduct problems, both in males and females [
65]. This study was also conducted within ALSPAC; however, important differences with the current study include the use of childhood life event scores instead of a specific measure of maltreatment, and the use of more general behavioral questionnaire data rather than diagnostic assessments of antisocial behavior. In addition to emphasizing our null finding in males, these differences may also explain the different female G × E results compared to the current study. Regarding smoking during pregnancy, we also failed to replicate the previous G × E findings for
MAOA [
16] in both sexes. Therefore, to conclude, while we reported a G × E between
MAOA-HL and maltreatment in females, we consider our other negative results regarding
MAOA as a sign to be slightly cautious when interpreting the earlier candidate gene-based G × E studies in this area [
8,
10,
18].
Strengths and limitations
Strengths of the current study have been the use of well-powered GWAS-implicated variants as novel targets for G × E research, the use of a large, ethnically homogeneous population sample with prospective measurements of smoking during pregnancy and childhood maltreatment, and more robust confounding control through modelling of covariates in interaction with both the genetic and environmental factors. Another strength has been the use of diagnostic interview data to measure childhood behavior. Moreover, we also performed adjustments for comorbid ADHD and internalizing problems, which is frequently lacking in both G × E and main effect studies. While we did not find main effects of the genetic variants (which may be due to methodological and/or clinical differences with the original studies), we did observe clear G × E interactions, which points to the importance of this field of study and implies that G × E’s (as part of the broad sense heritability model) might be able to explain part of the so-called ‘missing heritability’ [
66,
67]. Of note, ALSPAC is one of the samples used in the GWAS meta-analysis of antisocial behavior by Tielbeek et al. [
12]. However, since we failed to replicate the genetic main effect of the female-only SNPs implicated by that study, the meta-analytic genome-wide signals for these SNPs may be driven by the other cohorts in that study. While, on average, antisocial behavior levels were low (as expected in a population cohort), we observed relative effect sizes of moderate-to-large magnitude resulting from common genetic variants and environmental exposures, emphasizing the clinical relevance of these results.
Nevertheless, we need to acknowledge limitations of the present study. First, the use of singular genetic variants does not necessarily provide a comprehensive picture of G × E interactions as the genetic architecture of antisocial behavior is expected to be of a complex nature [
6,
12]. Alternative approaches to address this issue include the use of polygenic risk scores, gene-set (for example combining all genetic variants of a specific pathway), or gene-based (i.e., combining all variants related to a gene) analyses rather than singular variants. Nevertheless, we were able to identify different genetic loci that are likely to be of relevance given their implication as GWAS top hits. Furthermore, the top SNPs identified by the Rautiainen et al. GWAS were located only about 8 Mb distance (6p21.2) from the major histocompatibility complex (MHC) region at chromosome 6. The MHC region is highly polymorphic, displays extended LD structures and numerous disease associations have been reported for this region [
68]. However, as reported by Rautiainen et al., there was no LD between the identified top SNPs at 6p21.2 and SNPs showing up at the MHC region [
13]. Finally, maternal self-report measures of smoking during pregnancy and maltreatment, although measured prospectively may be subject to underreporting due to social desirability bias, which may affect the accuracy of effect estimates.
Conclusions
We studied sex-stratified G × E interactions in relation to antisocial behavior in a large population cohort and found interactions between recently (GWAS-)implicated variants and well-known environmental adversities. In males, G × E interactions with smoking during pregnancy were observed, which may be related to specific leucine-rich repeat protein networks involved in neurodevelopment. In females, G × E interactions with childhood maltreatment were found for one GWAS top SNP and MAOA. We were, however, unable to replicate other previously reported G × E interactions involving the MAOA gene. We conclude on a more general level that G × E studies do, indeed, contribute valuable information about the multifactorial nature of antisocial behavior, and we support the notion that well-powered GWASs provide more robust variants for G × E studies than classical candidate genes. Future studies should, in addition to GWAS top hits, incorporate polygenic, multimarker approaches, while addressing statistical robustness and potential sex differences when studying G × E interactions related to antisocial behavior.