Aggression is a common type of human behaviour (Tuvblad and Baker, 2011) and is considered a characteristic that is shared by all humans (Veroude et al., 2016). The propensity for aggression, however, varies considerably between individuals. This article addresses the question to what extent the variation that is seen for aggression has a genetic cause. Broadly, aggression can be defined as a behaviour that intends to cause physical or emotional harm to others (Anderson and Bushman, 2002). High levels of aggression are also seen in individuals with severe mental disorders (e.g., autism, bipolar disorder, and schizophrenia) as well as in patients with (rare) Mendelian disorders (Zhang-James and Faraone, 2016). Because of the large impact of aggression on the affected individual, their families, their environment, and society as a whole, there is a substantial interest in studying aggression from a wide range of disciplines. In this context, one goal is to unravel the aetiology of aggression by identifying environmental exposures and biomarkers, including genetic factors, epigenetic marks, and metabolites, that could function as predictors of (excessive) aggression (Boomsma et al., 2015).
Research often focuses on the pathological aspects of aggressive behaviour, while aggression does not solely have negative consequences or outcomes. Under certain circumstances, aggressive behaviour is beneficial to individuals, for example when competing for limited resources, like food or mates (Lindenfors and Tullberg, 2011), or achieving social dominance (Hawley et al., 2007). Aggression can further be a powerful deterrent against aggressive behaviour from others. Because both high and low levels of aggression can be detrimental to survival and procreation, it has been postulated that aggression is under stabilizing selection, implying that variation in aggression should show significant heritability. Substantial heritability estimates have indeed been reported in animals (Anholt and Mackay, 2012) and humans, as reviewed below.
Benefits of aggressive acts depend on the type of aggression, its success, environmental circumstances and also vary across cultures (Bukowski et al., 2011). For example, predatory goal-oriented aggression has been associated with social dominance in some instances (Dodge et al., 1997; Hawley, 2003; Voulgaridou and Kokkinos, 2015), but this association seems to vary between groups that are more prosocial and groups that consist predominantly of individuals with disruptive behaviour problems (Wright et al., 1986). A decrease in social status can also result from aggression, in particular from reactive aggression, which is an uncontrolled type of aggression stemming from internal or external frustration. In reverse, after a conflict, proactive aggression is increased in the victorious party while the losing party is less likely to engage in another aggressive act (Polman et al., 2007; Penn et al., 2010). To differentiate between different outcomes of aggression, researchers have distinguished aggression subtypes (e.g. reactive vs. proactive; overt vs. covert), developmental stages (childhood vs. adolescent onset), and comorbidities (e.g., with internalizing problems or with attention deficit hyperactivity disorder (ADHD)). In summary, the outcomes and types of aggressive acts can differ greatly between persons and circumstances, and need not always be dysfunctional.
At the start of the 1990s, research on aggressive behaviour was given a new impulse by a seminal paper of Brunner et al. (1993), in which a Dutch pedigree was described where men exhibited impulsive aggression, arson, violence, and borderline mental retardation. The family appeared to have a rare point mutation in the structural gene for monoamine-oxidase-A (MAOA) – which codes for an enzyme that is involved in the oxidative deamination of neurotransmitters like dopamine, serotonin, and norepinephrine – resulting in a deficiency of the MAOA enzyme. A study, by Caspi et al. (2002), compared variants of the MAOA gene in children who experienced maltreatment and showed that children with the variant resulting in lower levels of the MAOA enzyme were more likely to develop antisocial behaviour (ASB). Efforts to replicate the latter finding have been contradictory, either without replication (Haberstick et al., 2005; Young et al., 2006) or with replication (Foley et al., 2004; Kim-Cohen et al., 2006; Nilsson et al., 2018). Nevertheless, the studies of Brunner and Caspi stressed the importance of biological factors in the development of aggression and ASB. This instigated extensive efforts to study the genetic basis of aggression.
Enormous progress has been made with respect to technology in molecular biology and large-scale genotyping, as well as in the development of statistical methods for genetic association studies and polygenic scores for individual risk assessment, once sufficiently large genetic-association studies are available (Dudbridge, 2016). Costs for genotyping and sequencing of DNA, the epigenome and of RNA, and biomarker assessment, such as metabolomics, have steadily decreased, allowing for large studies, relating aggressive behaviour to genome, epigenome, transcriptome, and other biomarkers (Hagenbeek et al., 2016). Progress also has been made in characterizing the exposome, which reflects the totality of a person’s environmental exposures in space and time (Wild, 2005).
Genome-wide association studies (GWASs) provide a conceptual framework to examine whether individual differences in aggression are associated with allelic differences in millions of single-nucleotide polymorphisms (SNPs) across the genome (Visscher et al., 2017). Because a GWAS targets the entire human genome, it enables a data-driven approach to identify loci of interest. This hypothesis-free approach could potentially help researchers to overcome limits imposed by multifactorial nature of a trait and incomplete understanding of its physiological basis.
Here, we synthesise knowledge deriving from studies on genetics of human aggression and variance in liability to aggression-related traits. Our review has three foci: (1) to give a comprehensive overview of reviews already done on genetics of human aggression, (2) to carry out a systematic review of GWAS studies on human aggression, and (3) to introduce an automated systematic review for the selection of relevant literature, based on supervised machine learning. For consistency, in this review, we will use the general term ‘aggression’ (or ‘aggressive behaviour’, or ‘aggression-related traits’) to refer to the terminologies used by different authors (see Supplement S1, Supplemental digital content 1, http://links.lww.com/PG/A223), including anger, hostility dimensions, parent-reported child aggressive behaviour, physical aggression, ASB, violent offending, conduct disorders (CD), oppositional defiant disorder (ODD), and antisocial personality disorder (ASPD).
To optimize detection of the relevant literature for our review, we incorporated two strategies:
- A ‘traditional’ (manual) search strategy where search terms were used to extract the relevant articles from literature databases.
- An automated screening with Automated Systematic Review Software (ASR) where relevant articles were detected via the utilization of machine learning algorithms and a software development platform.
Search terms were developed by the authors based on prior literature and discussions with an expert librarian (J.W.S) from the LUMC. A literature search was performed in PubMed, Embase, Web of Science, Cochrane library, PsychInfo and Academic Search Premier with a comprehensive list of general search terms and medical subject headings (Supplement S2, Supplemental digital content 2, http://links.lww.com/PG/A224). Searches were conducted separately for reviews/meta-analyses and GWA studies. Searches included literature without a specific time limit and were conducted in mid-April 2019.
A selection was made from all titles and abstracts that were found in the databases using prespecified inclusion and exclusion criteria (Table 1). Articles were included if they (1) were written in English and (2) focused on human aggression. Studies were excluded if (1) they focused on animals, or (2) general terms linked to ‘aggression/violent etc.’ did not refer to a psychological/ psychiatric perspective but rather to characteristics of disease (e.g. aggressive cancer), or (3) articles discussed only a single gene. Psychiatric disorders, which incorporate acts of aggression and are highly correlated to aggression and antisocial lifestyles, like ODD, CD, and ASPD, were included. Articles referring to associations between genetic data and other (neuro)psychiatric disorders as main outcome (e.g. psychosis, borderline personality disorders, schizophrenia, bipolar disorder, anxiety, major depression, intellectual disability, Alzheimer’s disease, autism, ADHD, and addictions) were excluded. This increased the probability that the genetic profile that we examined was not confounded due to high comorbidity of aggression with other psychiatric disorders. Articles referring to aggression from the perspective of victimization and bullying were excluded. The publications were reviewed independently by two authors (V.V.O and P.J.R.), and when in doubt other coauthors were consulted until consensus on inclusion was reached.
Selection procedure and analyses
The search on review/meta-analyses resulted in 1713 records (Fig. 1). Duplicate entries were removed (N = 27). Next, 1660 records were excluded based on screening the titles and abstracts. In total, 26 potentially relevant reviews were retrieved for a full-text screening. Studies that did not fulfil or only partially fulfilled our criteria were excluded from the analysis (N = 12), leading to the inclusion of 14 articles. Four additional reviews were added through the automated selection, leading to a total of 18 articles – 13 targeted and 5 systematic reviews. These were organized into the following categories: review type (targeted or systematic), definition of aggression, type of reviewed studies (heritability, candidate gene, GWAS), population (children, adolescents, adults), quantity and period of the publications included in the reviews (parameters are made on the basis of reference lists with inclusion of publications on the aggression-related traits), described genes and main conclusions.
The search for GWASs on aggression resulted in 356 records. A total of 331 were excluded based on screening of the titles and abstracts. This led to the retrieval of 25 potentially relevant studies for full-text screening. Studies that did not fulfil or only partially fulfilled our criteria were excluded (N = 8), leading to the inclusion of 17 GWAS articles. Three additional studies were selected from the automated selection, including one SNP-heritability and two linkage studies. The studies were analysed by phenotype, sample characteristics, SNPs, or genetic variants associated with aggression-related traits at P < 1E−05, genetic variants position in genes and chromosomes.
Several GWAS articles report findings on multiple (stratified) GWASs. Tielbeek et al. (2017) adjusted for the fact that they performed three genome-wide association meta-analyses (GWAMA) by setting the genome-wide significance threshold at P = 1.67E−08, whereas others did not apply such a correction. This threshold might be overly conservative as the GWAMAs are stratified, which makes the P-values nonindependent across GWAMA. Therefore, we maintained a significance threshold of P = 5.0E−08 for all studies, and denote any SNP with a P-value below this threshold as genome-wide significant. While the traditional threshold might be too lenient in this context, we note that, when discussing GWASs, the P-value of a SNP in any given study is of less relevance than replication across GWASs.
Automated titles and abstracts screening
In parallel with the manual selection of titles and abstracts, another selection was made with the use of an automated selection tool ‘Automated Systematic Review’ (ASR) – software hosted at https://github.com (Automated systematic reviews by using Deep Learning and Active Learning, 2019). This software allows for automated in- and exclusion of articles for systematic reviews based on the titles and abstracts of articles. This enabled a comparison between ‘traditional’ manual selection and the automated screening on performance characteristics (e.g. time spent on selection and false-negative results). Furthermore, an additional selection was performed with the ASR on a large dataset of references to retrieve any new additional articles to our review, which would have been missed in the traditional search strategy (see Supplement S3, Supplemental digital content 3, http://links.lww.com/PG/A225).
We trained a model using ASR. To do so, the model requires a training set based on expert knowledge, consisting of articles that are either labelled relevant or nonrelevant (labels 1 = included; 0 = not) (see Supplement S3: Figure S3.1, Supplemental digital content 3, http://links.lww.com/PG/A225). To study the operating characteristics of the ASR, we used a dataset (N = 2955) consisting of relevant and nonrelevant articles on the genetics of human aggression, as labelled by researchers. From this labelled dataset of N = 2 955 500 records were repeatedly drawn at random as training sets. The number of relevant records in the training sets varied between 10 and 80 (e.g. 10 relevant records vs. 490 nonrelevant records), in increments of 10. These sets were used to train models to include relevant records and exclude nonrelevant records. For each model, we computed receiver operating characteristic parameters that were then used to select the optimal model (see Supplement S1: Table S3.1, Figure S3.2, Supplemental digital content 1, http://links.lww.com/PG/A223). We selected the model that returned the lowest false-positive rate (FPR) while allowing for a maximum false-negative rate of FNR = 0.03 at most. Note that FNR = 0.03 corresponds with a true positive rate of TPR = 0.97.
We applied the optimal model to predict classification in different searches: (1) reviews of genetics of human aggression (1713 records); (2) GWASs on human aggression (356 records); (3) searches 1 and 2 combined (2069 records) to analyse parameters of automated selection in comparison to manual selection.
Training sets were provided to the ASR for the reviews on aggression [26 relevant records out of 1713 (1.5%)] and the GWASs on aggression [25 relevant records out of 356 (7.0%)] (see Supplement S3: Table S3.2, Supplemental digital content 3, http://links.lww.com/PG/A225). The automated selection predicted 1018 records out of 1713 (59.4%) as relevant for reviews (including all prelabelled positives: TPR = 1.0; FPR = 0.59) and 243 records out of 356 (68.3%) for GWAS (including 24 prelabelled positives: TPR = 0.96; FPR = 0.66). Automated selection predicted 1261 records out of 2069 (60.9%) as important (including 50 prelabelled positives: TPR = 0.98; FPR = 0.60). The workload for manual selection was ~60 hours. This means that for the applied model and these set(s), the reduction in workload is expected to be ~23.5 hours. By allowing for a higher FNR in model selection, the workload could be reduced even further, although at the expense of missing more true positives.
Our automated selection repeated the traditional manual search with inclusion rates [100% for reviews (58.8% false positives), 96.0% for GWASs (66.2% false positives), 98.4% for reviews and GWASs combined (60.0% false positives)], 0 cases were false negatives for reviews, 1 case for GWASs, and 1 case for reviews and GWASs combined.
A new search on ‘human aggression genes’ was performed in the same databases without additional search terms and time limitation (14 400 records) to detect new contributions to the systematic review, resulting in 55.8% included records. Exclusion of duplicate records resulted in 6469 records. From these, four reviews were added to the overview of reviews on aggression, and one SNP-heritability and two linkage studies were added to the GWASs review as additional information for the interpretation of GWAS findings. These seven studies were detected only by the ASR approach and did not appear in the traditional approach.
We included 18 reviews on the genetics of human aggression in our analyses, each covering different periods and including varying numbers of studies (Table 2). The reviews cover more than 2000 studies on aggression.
What is considered to be aggression?
Reviews indicate that the phenotypic definitions of aggression vary considerably, and heterogeneity of the phenotypic definition is mentioned as a major hurdle in aggression research by multiple articles. Definitions of aggression, as well as the focal points of reviews, range from broadly defined externalizing and ASBs (see Supplement S1, Supplemental digital content 1, http://links.lww.com/PG/A223), which also include potentially nonaggressive behaviours like rule-breaking behaviour (Fernandez-Castillo and Cormand, 2016), to a narrow focus on chronic physical aggression (Tremblay et al., 2018). Other reviews and studies focus more explicitly on psychiatric classifications like ODD, CD, and ASPD, which encompass aggressive acts and are correlated to ASB (Veroude et al., 2016; Raine, 2019). One review incorporated the analysis of genetics of aggression in suicidal behaviour (Baud, 2005). Classifications, which are useful in clinical practice, tend to consist of constellations of heterogeneous ASBs (e.g. ‘often initiates physical fights’ vs. ‘is often truant from school’) and personality characteristics (e.g. ‘having difficulty sustaining long-term relationships’ vs. ‘lacks concern, regret, or remorse about other people’s distress’ (American Psychiatric Association, 2013)).
Several reviews proposed a focus on more homogeneous or dimensional constructs of aggression (Fernandez-Castillo and Cormand, 2016; Tremblay et al., 2018). A dimensional construct is in line with the conceptualization that pathological aggression is situated on the extreme ends of a normal distribution (Veroude et al., 2016). Some authors see a risk in the dimensional approach and note that findings might become predominantly driven by variations within normal, adaptive levels of aggression (Ferguson, 2010). However, if pathological levels of aggression are indeed the extreme end of a continuous phenotype, the same genetic and environmental factors should apply to both the normal range and extremes of the distribution.
In the end, concerns regarding heterogeneity and the impact of different phenotype definitions are empirical questions, which are currently also being asked in other GWASs of psychiatric disorders such as depression (Cai et al., 2019). Such questions can be resolved, once well-powered GWASs are available, by estimation of genetic correlations among different phenotype definitions of aggression and can also be addressed through genetic modelling of twin and family data. For example, Hendriks et al. (2019, submitted) analysed twin data collected by multiple instruments, commonly employed to measure aggression in children. While phenotypic correlations between different aggression scales could be low, a genetic multivariate analysis of these data showed high genetic correlations among different instruments. Such observations mean that different instrument tap into the same genetic liability and could be analysed simultaneously in GWAS.
Reviews that propose some sort of differentiation among aggressive behaviours often return to a distinction between reactive and proactive aggression. Reactive aggression is commonly described as impulsive and defensive, while proactive aggression is considered predatory and premeditated. Both types of aggression may involve similar biological systems. The aminergic systems (e.g. serotonergic and dopaminergic) have been proposed as likely to regulate both forms of aggression (Waltes et al., 2016). Interestingly, Runions and colleagues (2019) argue that researchers studying reactive and proactive forms of aggression have conflated motivation (aversive vs. appetitive) and implementation (impulsive vs. premeditated) and propose that predatory aggression can also be impulsive in nature, defined as recreation instead of rage, while reactive aggression could also be delivered after a longer period of time, referring to reward instead of revenge.
The developmental aspect of aggression is a major theme in reviews (Moffitt, 2005; Tuvblad and Baker, 2011; Provencal et al., 2015; Veroude et al., 2016; Waltes et al., 2016; Davydova et al., 2018). Age of onset is often mentioned as an important differentiating factor for subtypes of ASB, with aggression usually already present in early childhood, while rule-breaking behaviour and delinquency usually develop during adolescence. Tremblay (2010) proposes a developmental framework of aggression among a covert/overt axis and a second destructive/nondestructive axis as the most viable constructs to subtype disruptive behaviour (aggression, opposition-defiance, rule breaking, and stealing-vandalism). Children who display destructive and overt disruptive behaviours, especially those exhibiting chronic physical aggression, experience more risk factors early in life, engage in aggression from a young age, and have a more persistent developmental course of aggression and ASB. A differentiation on age of onset is considered especially relevant in reviews, which include epigenetics. Epigenetic changes may be triggered by early life adversity (Provencal et al., 2015; Manchia and Fanos, 2017; Tremblay et al., 2018; Curry, 2019), although variation in epigenetic marks can also reflect influences of DNA polymorphisms (van Dongen et al., 2016).
In research, aggressive behaviour often is measured by questionnaires, such as the Achenbach System of Empirically Based Assessment scales (ASEBA; Achenbach et al., 2017), the Strengths and Difficulties Questionnaire (SDQ; Goodman et al., 2010), or the Buss Durkee Hostility Inventory (BDHI; Buss and Durkee, 1957). Aggression scales in such instruments may include items which reflect behaviour that is related to aggression, but would not be considered aggression based on item content. For example, the ASEBA Aggressive Behaviour scale for children contains items like ‘Argues a lot’ or ‘Gets in many fights’, but also ‘Unusually loud’ or ‘Suspicious’. Measures can also derive from observational studies, especially in younger children, and some experimental paradigms are available to measure aggression in across wider age ranges. Such experiments can, however, not cover the full spectrum of aggressive behaviour and, perhaps even more critically, cannot be applied in epidemiological samples.
There is a divergence between measurement of aggression in research projects compared to how (pathological) aggression is defined in clinical practice. Questionnaires are used as tools by clinicians, but the presence of these behaviours is mostly determined by interviews with the patient, and others who know the person (e.g. parents and teachers), by observation, and by the patient’s (criminal) records. Psychiatric disorders that include aggressive behaviours or disorders, which are correlated to aggressive and antisocial lifestyles, are dependent on classification systems like the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Diseases (ICD). In these classifications, a dichotomy is applied in which a disorder is either present or absent, largely ignoring the dimensional nature of human behaviour. In genetic studies, a focus on the dichotomy rather than on continuous variation may lead to a loss of statistical power (van der Sluis et al., 2013).
Another important question, especially in clinical settings, is when aggression becomes pathological. Some aggressive behaviours are clearly defined as pathological, like aggressive behaviours that define CD (e.g., ‘Has used a weapon that can cause serious physical harm to others) or ASPD (e.g., ‘Irritability and aggressiveness, as indicated by repeated physical fights or assaults’). In contrast, other aggressive behaviours are less clearly considered pathological, because they occur to some extent in all individuals, like anger or hostility. This even is the case for some aggressive behaviours, which are part of disruptive behaviour disorders (e.g., ODD: often argues with authority figures). For aggression to be pathological, it is essential that aggressive behaviours cause clinically significant impairment in social, academic, or occupational functioning.
Approaches in genetics of aggression studies and the current status quo
There are several designs to study the genetic aetiology of aggression, with the two major ones being genetic epidemiological/behavioural genetic approaches on the one hand and molecular genetic approaches on the other (Fig. 2). Behavioural genetic studies have a long and successful history (Loehlin, 2009). More recently, molecular genetic studies have seen enormous breakthroughs with the development of techniques like GWASs (Visscher et al., 2017).
Behavioural genetic approaches
Numerous studies focused on explaining the aetiology of aggression and ASB through family, twin, and adoption studies, which can disentangle genetic and environmental influences. Twin models enable researchers to divide the variance for a trait, or the liability to a disorder, into genetic and nongenetic components. The genetic variance component often is defined as the additive (A) effects of many genes. Environmental variance components consist of environmental influences common to siblings from the same family (C), creating resemblance of family members through environment rather than through genetics, and a unique or nonshared environmental component (E). Unique environmental influences affect family members in different ways (Boomsma et al., 2002). Unsystematic influences such as measurement error also are included in the E component, unless explicitly modelled. In general, reviews indicate that additive genetic factors explain around 50% of the variability of aggressive behaviour (Craig and Halton, 2009; Rhee and Waldman, 2011; Tuvblad and Baker, 2011; Fernandez-Castillo and Cormand, 2016). The estimate varies around 50% across studies, with some reviews reporting somewhat higher heritability estimates (65%) and others giving estimates for aggression and ASB that vary more [e.g. 38–88% (Veroude et al., 2016); 28–78% (Tuvblad and Baker, 2011)]. Physical aggression seems to show larger heritability estimates (65%) than reactive (20–43%) and proactive aggression (32–48%), while rule-breaking behaviour, which is often aggregated with aggression indices, also shows a heritability around 50% (Waltes et al., 2016; Gard et al., 2018). Heritability estimates of aggressive behaviour were higher in children with stable callous unemotional traits (81%) compared to children low in callous unemotional traits (30%) (Gard et al., 2018). This suggests a larger influence of genes on children with more severe aggressive tendencies (Gard et al., 2018). Contributions of shared environment are relatively small and decrease with age, with the vast majority of adult studies not reporting any shared environmental influences (Tuvblad and Baker, 2011; Veroude et al., 2016; Waltes et al., 2016). Thus, research in behaviour genetics clearly indicates that there is a substantial genetic component to aggressive behaviour in humans. In longitudinal studies, heritability estimates of aggression and ASB increase somewhat from childhood through adulthood (Tuvblad and Baker, 2011; Veroude et al., 2016; Waltes et al., 2016). Genetic factors also contribute to the stability of aggressive behaviour during preschool and school age, and puberty (Porsch et al., 2016; Waltes et al., 2016). Measurement instrument, and also rater, seem to influence heritability estimates, with heritability based on parent-report and teacher-report estimated as higher than those based on self-report and observational studies. Studies based on self-report tend not to find any shared environmental influences (Tuvblad and Baker, 2011), but such studies are not available for younger children. Unlike parent or teacher reports, observational studies more often give an assessment of aggression at one particular moment in time only. Parent- and teacher-reports tend to provide phenotype information that is more averaged over longer periods of time and are similar in terms of heritability estimates. Parent-report leads to higher estimates of shared environmental influences than teacher-report, when parental characteristics that influence ratings of multiple children (e.g. twins or siblings) are not taken into account. When twins have different teachers, similarities between them tend to decrease. This may reflect actual differences in aggressive behaviour with different teachers and/or different settings, but may also reflect teacher characteristics that influence assessments of multiple children.
In summary, heritability is estimated consistently around 50%, with some variation that may be due to different conceptualization of aggressive and ASBs, with more severe types of aggression showing higher heritability.
Heritability estimates of aggression and ASB may differ between environments suggesting an interaction between genes and environment (GxE). Proposed putative environmental moderators are familial adversity (e.g. maltreatment and parental delinquency), social disadvantage (e.g., poverty and bad neighbourhoods), violent media exposure, and alcohol use. Tuvblad and Baker (2011) argue that, compared to genetic factors, environmental influences are relatively more pronounced for ASBs in the presence of high environmental risk and disadvantaged environments. Conversely, genetic influences will be more pronounced when environmental risk factors are absent or less prominent. In one study, the moderating effects of neighbourhood seemed to be specific to the heritability of nonaggressive ASB, while heritability estimates of aggressive ASB were not influenced by neighbourhood disadvantage (Burt et al., 2016). Such findings underscore the differential influence of environmental adversity on certain types of ASB, with aggressive behaviour showing less sensitivity to environmental influences than other types of ASB. Later reviews, however, indicate mixed findings. Some reported an increase in genetic variance in the presence of environmental risk. To illustrate, when young children were subjected to high levels of maternal disengagement, genetic factors explained more variance in later conduct problems (Boutwell et al., 2012; Waltes et al., 2016). An increase in heritability of externalizing disorders was also found when young adults were exposed to a combination of risk factors [e.g. antisocial or lack of prosocial peers and relationship problems with parents (Hicks et al., 2009; Veroude et al., 2016)].
Depending on the type of aggression, mean levels of aggression often are higher in males than in females. Differences in heritability estimates, however, between males and females are modest or absent. According to Tuvblad and Baker (2011), heritability did not differ significantly between genders across different twin studies, either quantitatively or qualitatively [see also (Vink et al., 2012)]. These studies mainly included mother-reports of childhood aggression and heritability estimates were higher in males than in females when self-report data were analysed (Waltes et al., 2016). It has been suggested that gender differences in heritability become more pronounced from adolescence, which could be indicative of the ‘Young Male Syndrome’, in which the onset of puberty and increasing levels of testosterone are related to increases in aggression in 12- to 25-year-old males (Craig and Halton, 2009). This would also suggest a possible role of genes related to androgen synthesis and function in the development of aggression from puberty onwards.
In summary, twin studies highlight the importance of genetic influences, with estimates of the heritability of aggression and ASB often reported to be around 50% (Moffitt, 2005), without much evidence for sex differences in heritability estimates. Such significant heritability is a first requirement for initiating studies that aim to find molecular signatures in the DNA sequence that are associated or causally related to the phenotype.
Integrating data on genetics of aggression from molecular genetic studies
Geneticlinkage and candidate gene studies: Molecular genetic studies include genetic linkage and association studies, either genome-wide or with a focus on a limited number of candidate genes or candidate regions. In linkage studies, DNA markers are assessed in related individuals to investigate the inheritance of markers with known chromosomal locations together with aggression in pedigrees. Sometimes candidate regions to be investigated are suggested from studies in other species. With the arrival of large-scale association studies, linkage studies, which require family-based designs, have become less common, but early studies have suggested regions on three chromosomes that could be associated with aggression. Dick et al. (2004) analysed retrospectively reported childhood CD in an adult sample from COGA (Collaborative Study on the Genetics of Alcoholism). Regions on chromosomes 19 and 2 may contain genes associated with risk of CD. The same region on chromosome 2 has been linked do alcohol dependence in this sample. Criado et al. (2012) in a linkage study of cortical even-related oscillations associated with ASPD and CD suggested that chromosome 1 may contain a genetic locus for ASPD/CD.
Genetic association studies initially were candidate gene studies. These require a priori knowledge of or hypotheses about which genes are implicated in the aetiology of the trait of interest. For aggression, associations were considered for genes from the serotoninergic [5-HTTLPR (5-hydroxytryptamine (serotonin) receptors), SLC6A4 (solute carrier family 6 member 4)], dopaminergic [dopamine receptors genes DRD4, DRD2, DRD5, and SLC6A3 (solute carrier family 6 member 3)] and GABAergic systems [e.g. genes that code GABA (gamma-aminobutyric acid) receptors, like GABRA2 (gamma-aminobutyric acid type A receptor alpha2 subunit)], as well as genes related to catecholamine catabolism [MAOA (monoamine oxidase A), COMT (catechol-O-methyltransferase)] (Provencal et al., 2015; Fernandez-Castillo and Cormand, 2016; Veroude et al., 2016; Davydova et al., 2018; Gard et al., 2018). Other studies focused on associations with the genes involved in stress response pathways (Craig and Halton, 2009; Waltes et al., 2016); hormone regulation [e.g., AVPR1A (argenine vasopressin receptor 1A)] (Fernandez-Castillo and Cormand, 2016; Veroude et al., 2016; Waltes et al., 2016; Salvatore and Dick, 2018); hypoglycaemia and insulin secretion (Craig and Halton, 2009); and neuronal transcripts and brain expression patterns (Craig and Halton, 2009; Anholt and Mackay, 2012; Waltes, Chiocchetti and Freitag, 2016; Gard et al., 2018). Candidate gene studies have been criticised (e.g. Duncan and Keller, 2011), since it became clear that findings for candidate genes are often not replicated in well-powered GWASs (e.g. Bosker et al., 2011; Luo et al., 2016). It is likely that this also extends to studies of aggression, but the status of the candidate genes for aggression must await well-powered GWASs.
Many reviews agree that aggression is a polygenic trait influenced by many genes and that each explains a small proportion of the phenotypic differences. However, there may be an overlap between genes of large effect underlying monogenic disorders and those affecting continuous variability of related quantitative traits. Extending the idea of a shared genetic basis between Mendelian disorders and polygenic traits, one alternative approach based on the search for genes for aggression in studies of rare, functional genetic variants associated with aggression phenotypes catalogued in Online Mendelian Inheritance in Man [OMIM; (Zhang-James and Faraone, 2016)]. Most of these genes had not been implicated in human aggression before, but the most significantly enriched pathways (e.g. serotonin and dopamine signalling) had been previously implicated in aggression. Among these genes, only two were previously related to aggression [MAOA, GRIA3 (glutamate ionotropic receptor AMPS type subunit 3)]. New associations were found with genes [e.g. CAMTA1 (calmodulin binding transcription activator 1), APBB2 (amyloid beta precursor protein binding family B member 2), DISC1 (DISC1 scaffold protein), and others], which implicated in cell-to-cell signalling and interaction, nervous system development and function, and behaviour. The novel genes and pathways identified in this study suggested additional mechanisms underlying aggression.
Genome-wide association studies: GWASs investigate millions of SNPs, under a continuous or dichotomous, case/control model. The result is a list that, for every variant, indicates the expected increase in a trait (continuous) or genetic liability (dichotomous) for every copy of an effect allele. Due to the large number of tests, the genome-wide significance level is set at P = 5.0E−08 (Sham and Purcell, 2014), to properly control for the type I error rate. This adjusted threshold already considers the fact that neighbouring SNPs are not inherited independently from one another. However, the nonindependent inheritance of SNPs indicates that association tests between noncausal SNPs and the trait of interest contain a part of the polygenic signal (Bulik-Sullivan et al., 2015). As such, even when only a limited number of SNPs reach this stringent significance level, there is signal in the other association tests. The weighted effects of all the genetic variants involved in aggression could produce a polygenic risk score with a certain predictive value (Beaver et al., 2018).
Many reviews discussed a whole-genome approach to understanding aggression, but only three have done so in a systematic manner (Fernandez-Castillo and Cormand, 2016; Veroude et al., 2016; Waltes et al., 2016). We will summarize findings for genes harbouring, or in proximity to, variants that reached genome-wide (P ≤ 5.0E−08) or nominal (P ≤ 1.0E−05) significance levels in all GWAS of aggression phenotypes to date. These include aggression-related phenotypes, i.e. anger, hostility dimensions, aggressive behaviour, physical aggression, ASB, violent offending, CD, ODD, and ASPD.
To provide a complete picture of the GWAS literature available, we chose to include phenotypes, which clearly include aggression, but are sometimes conflated with other ASBs (e.g. rule breaking) or personality characteristics (e.g. being suspiciousness and being loud). These phenotypes can be found in Supplement S4, Supplemental digital content 4, http://links.lww.com/PG/A226. Most GWASs on aggression were performed in child and adolescent samples of European ancestry, in which aggression was assessed using rating scales (Table 3).
GWAS studies have mainly resulted in nominal associations between genetic variants and aggression-related traits and disorders. Collectively, these studies reported 10 genome-wide significant findings (Dick et al., 2011; Rautiainen et al., 2016; Tielbeek et al., 2017; Montalvo-Ortiz et al., 2018). Five of these variants are located inside or close to four genes: LINC00951 (long intergenic nonprotein coding RNA 951) (Rautiainen et al., 2016), C1QTNF7 (C1q tumor necrosis factor-related protein 7) (Dick et al., 2011), PSMD1 (proteasome 26S subunit, non-ATPase 1), and HTR2B (5-hydroxytryptamine receptor 2B) (Montalvo-Ortiz et al., 2018). Lastly, the five remaining significant SNPs are located on chromosomes 11 (Dick et al., 2011; Tielbeek et al., 2017), 13 (Dick et al., 2011), 1, and X (Tielbeek et al., 2017).
In a mixed sample of subjects from European and African-American ancestry, three SNPs inside C1QTNF7 were significantly associated with CD symptoms in adults with substance dependence (Dick et al., 2011). When the sample was split on the basis of ancestry, no SNPs reached suggestive levels in the European-American sample. In the African-American sample, one out of the three SNPs reached suggestive levels (minimum P = 4.35E−06), along with two additional suggestive findings (minimum P = 2.67E−07). C1QTNF7 is less expressed in the brain, compared to such tissues as endometrium, gall bladder, lungs, ovaries and 18 other tissues, and has a potential role in maintaining energy balance (Kaye et al., 2017).
In a study focusing on ASPD in Finnish criminal offenders, Rautiainen and colleagues (2016) found one hit (rs4714329, P = 1.6E−09) in the cross-sex meta-analysis. This variant is in proximity to LINC00951 (long intergenic nonprotein coding RNA 951). The same SNPs returned suggestive associations in the male-specific GWAMA of ASPD (P = 1.38E−07). The signal from these variants was specific for ASPD, and did not cover a broader range of criminal behaviour. Montalvo-Ortiz and colleagues (2018) found that SNPs located in the HTR2B (P = 2.16e−08) and PSMD1 (P = 1.79e−08) genes were significantly associated with cannabis-related physical aggression in African-Americans, but these SNPs did not reach even suggestive significance in European-Americans. Cannabis use has been associated with greater impulsive decision-making and increased aggressive behaviour. Notably this is the only GWAS study which focused purely on physical aggression.
Anney and colleagues (2008) listed 54 SNPs nominally associated with conduct problems. These SNPs tagged 41 genes, three of which are with known functions and are involved in the regulation of dopamine receptor D2 signalling [PAWR (proapoptotic WT1 regulator)], synaptic plasticity [KIRREL3 (kirre like nephrin family adhesion molecule 3)], and neuronal development [RBFOX1 (ral guanine nucleotide dissociation stimulator like 1)]. Sonuga-Barke and colleagues (2008) analysed interactions between CD symptoms and maternal warmth. Nominal effects were found for SNPs located in genes involved in brain maturation, neurotransmission, neuronal development, and regeneration. Viding and colleagues (2010) examined teacher-reported conduct problems in children and found no suggestive SNPs (minimum P = 4.6E−05).
For adult ASB (Tielbeek et al., 2012), the strongest signal was for a SNP (rs346425; P = 2.51E−07) located on chromosome 5. Salvatore and colleagues (2015) in an adult ASB sample observed the strongest association for rs4728702 (P = 5.77e−07), located in ABCB1 (ATP binding cassette subfamily B member 1) on chromosome 7 that may confer general risk across a wide range of externalizing behaviours. Enrichment analyses further indicated involvement of immune-related pathways. Two GWASs compared cohorts of Finnish violent offenders to the general population (Tiihonen et al., 2015; Rautiainen et al., 2016), and obtained association signals at genes involved in neuronal development (Tiihonen et al., 2015) and adaptive immunity (Rautiainen et al., 2016).
Aebi and colleagues (2016) hypothesized that BCL2L1 (BCL2 like 1) is likely associated with oppositional behaviour, because of its influence on presynaptic plasticity through regulation of neurotransmitter release and retrieval of vesicles in neurons. Brevik and colleagues (2016) applying gene-based tests observed NTM (neurotrimin) as the top gene, which is differentially expressed in aggression-related structures of the amygdala and the prefrontal cortex in early stages of brain development.
Merjonen and colleagues (2011) saw suggestive associations for SNPs that lie inside genes involved in the maintenance of high frequency synaptic transmission at hippocampal synapses, and regulating synaptic activation [SHISA6 (shisa family member 6)] in a Finnish population sample. Mick and colleagues (2011) found associations for SNPs that lie inside or close to multiple genes, including LRRC7 (leucine-rich repeat containing 7), involved in neuronal excitability and used as postsynaptic marker of hippocampal glutamatergic synapse integrity, and STIP1 (stress-induced phosphoprotein 1), involved in astrocyte differentiation and highly expressed in the brain. A second GWAS by Mick and colleagues (2014) observed a nominal association of proneness to anger with the gene, involved in calcium influx and release in the postsynaptic density, and in long-term potentiation [FYN (FYN proto-oncogene, Src family tyrosine kinase)]. McGue et al. (2013) reported four SNPs associated with behavioural disinhibition including symptoms of CD and aggression, one of which (rs1368882; P = 1.90E−06) was located inside the GLIS1 (GLIS family zinc finger 1) gene responsible for a transcription factor that is involved in regulating the expression of numerous genes.
Recently, two larger studies attempted to identify genes associated with aggression or ASB by increasing power through the inclusion of multiple cohorts. Pappa and colleagues (2016) collected a sample of 18 988 children 3–15 years for meta-analysis and reported a near genome-wide significant locus on chromosome 2p12 (P = 5.3E−08). This locus is in proximity to two genes: LRRTM4 (leucine-rich repeat transmembrane neuronal 4), which regulates excitatory synapse development, and SNAR-H (small NF90 (ILF3) associated RNA H), which is implicated in the transcription process and is expressed in neurons. They found 19 genes nominally related to aggression from gene-based tests, which include LRRTM4, PDSS2 (decaprenyl diphosphate synthase subunit 2), TRIM27 (tripartite motif containing 27), MRC1 (mannose receptor C-type 1), MECOM (MDS1 and EVI1 complex locus), and CASC17 (cancer susceptibility 17).
Another larger study by Tielbeek and colleagues (2017) focused on the broader ASB phenotype in 16 400 individuals. The overall GWAMA found no hits, but sex-stratified GWAMAs returned three genome-wide significantly associated SNPs (minimum P = 1.95E−08), but failed to identify significant genes. This suggested that there might be sex-specific genetic effects on ASB and focusing on a more specific phenotype could improve chances of findings significant results.
Thus, nominal genome-wide associations (P < 1E−05) have been found in genes involved in a wide variety of biological systems: the immune system, the endocrine system, pathways involved in neuronal development and differentiation and synaptic plasticity. These findings have not been replicated across GWASs, but some studies reported the same genes independently: NTM (Tiihonen et al., 2015; Brevik et al., 2016) and RBFOX1(A2BP1) (Anney et al., 2008; Sonuga-Barke et al., 2008).
In summary, the 17 GWASs in our review show that genome-wide significant and/or suggestive associations between aggression-related traits and SNPs are found on all chromosomes (range: 1–63; see Supplement S5-6, Supplemental digital content 5, http://links.lww.com/PG/A227; Supplemental digital content 6, http://links.lww.com/PG/A228). As shown in Fig. 3, nearly 55% of suggestive associations were found on chromosomes 1, 2, 5, 6, 7, 9, 10, and 11, with the majority of suggestive SNPs on chromosome 7 reported in the sample of African ancestry (Montalvo-Ortiz et al., 2018). The genome-wide significant associations are located on chromosomes 1, 2, 4, 6, 11, 13, and X.
Aggression has a considerable genetic component, as indicated by decades of behaviour genetics research. However, no genomic variants have (yet) been identified. In our review covering GWASs on human aggression, only 4 out of 17 studies reported genome-wide significant hits in primary or replication samples (Dick et al., 2011; Rautiainen et al., 2016; Tielbeek et al., 2017; Montalvo-Ortiz et al., 2018). In the reviews on aggression and GWASs, several explanations are offered for the discrepancy between heritability estimates in behavioural and molecular genetic studies; for example, the heterogeneous, context-dependent, and developmental nature of aggression, but foremost, small sample sizes. Fortunately, these limitations can be remedied and provide future directions for research.
Most of the reviews covered mention the often cited heritability estimates of 50% for aggression by Miles and Carey (1997), and 41% for ASB by Rhee and Waldman (2002) and these estimates are confirmed in more recent empirical studies. Moderation, or any genotype × environment effects seem small, and most pronounced for nonaggressive ASB (Burt et al., 2016).
How to address nonsignificant findings in GWAS studies on psychiatric problems is a pressing issue. Opinions are divided on what approach is most optimal to define phenotypes for GWAS analyses. Some believe that reduction of phenotypic heterogeneity could lead to more genome-wide significant findings (Anholt and Mackay, 2012; CONVERGE consortium et al., 2015; Runions et al., 2019). This view is supported by the GWASs covered in this review that did find genome-wide significant hits. These relatively underpowered studies (Nrange = 2185–6220 participants) focus on individuals with severe ASB and specific types of aggression: individuals with DSM-defined CD symptoms (Dick et al., 2011), cannabis-induced physical aggression (Montalvo-Ortiz et al., 2018), and criminal offenders with ASPD (Rautiainen et al., 2016). Two studies were conducted in specific samples; exclusively male, with associations only in African-American subgroup (Montalvo-Ortiz et al., 2018), and predominantly male (89% of cases) and ethnically homogeneous (Rautiainen et al., 2016).
In contrast, other researchers propose a broader approach, which includes more lenient phenotypes (Vassos, Collier and Fazel, 2014; Ormel et al., 2019). This lenient phenotyping approach has already achieved success in depression research; for example, although here the value of minimal versus broader phenotyping is debated as well (Cai et al., 2019). The two largest GWASs on aggression that were covered by this review used broad, lenient measures of childhood aggression (Pappa et al., 2016) and ASB (Tielbeek et al., 2017). Pappa and colleagues (2016) found no significant hits, but several promising loci on chromosomes 2, 3, 6, and 17 (minimum P = 5.3E−08). Tielbeek and colleagues (2017) reported three significant hits for the sex-stratified GWAMAs.
Early linkage studies on aggression indicated chromosomes 1 (Criado et al., 2012), 2, and 19 (Dick et al., 2004) as potential loci. GWAS findings in our review confirm loci on chromosomes 1 and 2, which gave more associated variants and significant results. The X- and Y-chromosomes did not give evident results, even if one significant sign was reported in X-chromosome (Tielbeek et al., 2017).
To identify 80% of all causal SNPs, depending on the extent of SNP heritability, between 105 and 107 (100 000–10 000 000) independent subjects would be required (Holland et al., 2019). This means that, with sample sizes l0 time less than the lower bound, current GWASs were clearly underpowered. At present, several initiatives are under way to collaborate in achieving larger sample sizes. One example of a large collaborative project is the ACTION consortium (Aggression in Children: Unraveling gene-environment interplay to inform Treatment and InterventiON strategies: http://www.action-euproject.eu/), which has brought together over 30 cohorts with childhood data on aggression for GWAS, EWAS, and biomarker studies.
As mentioned, multiple reviews suggest that heterogeneity of aggression is a problem in research, with several reviews suggesting some kind of distinction between subtypes, subgroups, or developmental stages. Standardized phenotypic and environmental assessments are proposed as a solution (Craig and Halton, 2009). Although this standardization of assessment could be an option, recent advances in multivariate modelling allow for exploration of other potential avenues (e.g. Baselmans et al. 2019). This approach is also discussed in the meta-analyses of Zhang-James and Faraone (2016), in which aggression might be considered a multidimensional trait consisting of distinct, but related, constructs with shared aetiologies (Zhang-James and Faraone, 2016). In other words, although some individuals show different problem behaviours, including aggression, they all share a common genetic vulnerability. Taking a multivariate, approach would allow the inclusion of large cohorts with existing phenotypic (Bartels et al., 2018) and SNP data. However, the focus on ever broader phenotypes and bigger samples raises the question how to translate results into practice, to alleviate problems of individuals.
We should recognize that the nature-nurture debate has moved on from the question whether aggressive behaviour is heritable to the discovery of the biological bases of aggression. This is currently achieved by investigating aggression’s relation to genes, SNPs, and relevant biological pathways. It is expected that GWASs with larger or combined datasets will improve our understanding of the mechanisms of gene regulation of aggression. Individual GWASs on aggression and aggression-like traits are still limited in terms of explaining variation in the population, but ongoing GWASs and other efforts, e.g. in epigenetics and biomarker studies are likely provide insight into the aetiology of aggressive behaviour. Expansion of disease gene maps (Goh et al., 2007) by including aggression-related traits into, for example, OMIM datasets can help in future analyses of underlying cellular network-based relationships between genes and functional modules of aggressive behaviour, and future work should determine whether genes mediating aggression pathways are enriched in the polygenic background of disorders associated with aggression.
Also, leveraging on genotype-tissue expression [GTEx; (eGTAxProject, 2017)] GWAS findings can be annotated with additional information and thereby identify biologically relevant systems. One particularly interesting source of biological annotation revolves expression quantitative trait loci (eQTL), i.e. SNPs that have been associated with gene expression levels. Once genome-wide hits are found, overlapping these with known eQTLs could identify genes that are of biological interest (Lowe et al., 2015; Gusev et al., 2016; Zhu et al., 2016).
Systematic reviews with automated functions
The workload on selection process of researchers in our systematic review was around 60 hours (screening and selecting relevant articles from list of 2069 records). By using automated procedures to screen for relevant literature for inclusion in systematic reviews, it was possible to save 39.1% (23.5 h) of reading/scanning time. The downside of automated methods is that relevant literature can be missed. On the contrary, even an expert reviewer might omit studies that the automated procedures include. Optimization of the expert reviewer is covered by education and training, whereas optimization of automated selection is under active development (Cohen et al., 2006; Khabsa et al., 2016; Borah et al., 2017). We opted for a recent approach that utilizes a machine learning algorithm to obtain a selection of articles that could be relevant for this systematic review.
Although the ASR tool we applied is quite new and is still under active development, we found that applying the machine learning approach as implemented in the software hosted at https://github.com (Automated systematic reviews by using Deep Learning and Active Learning, 2019) could be indeed of considerable aid to the researcher performing a systematic review solving problems of missed literature in screening phase due to human errors or excluded by searching algorithms.
For the benefit of further developments in automated selection approaches aiding the review process, we advise review authors to supply their search results as additional information to their work. These results can then serve for further refinement of literature search models. This would avoid double work across research groups, create a comprehensive overview of aggression literature, and increase our understanding of the genetic nature of human aggression.
Aggression in humans is a heritable trait, whose genetic basis largely remains to be uncovered. No sufficiently large GWASs have been carried out yet. With increases in sample size, we expect aggression to behave like other complex human traits for which GWAS has been successful. There are several ongoing efforts to achieve genome-significant GWAS findings – merging samples in consortia, replication strategies, searching for close phenotypes from other domains associated with aggression for sample extension, developing new approaches of partitioning genetic heterogeneity and sample stratification. Automated tools for systematic review, which are based on machine learning, could be used to optimize the integration of research findings from different studies.
The authors thank the librarian J.W. Schoones, MA (Walaeus Library, Leiden University Medical Center) for valuable help with refining the search methodology and extracting the relevant literature, and project members of A.G.J. van de Schoot team (Utrecht University) for guidance related to the automated selection tool.
D.I.B. was supported by Royal Netherlands Academy of Science Professor Award (PAH/6635); P.J.R., H.F.I. and D.I.B. were supported by the European Community’s Seventh Framework Program Grant 602768: ACTION (Aggression in Children: Unraveling gene-environment interplay to inform Treatment and InterventiON strategies); C.M.v.d.L. was supported by the Amsterdam Law and Behavior Institute (A-LAB; Vrije Universtiteit, Amsterdam).
Conflicts of interest
There are no conflicts of interest.
Achenbach TM, Ivanova MY, Rescorla LA. Empirically based assessment and taxonomy of psychopathology for ages 1½-90+ years: developmental, multi-informant, and multicultural findings. Compr Psychiatry. 2017; 79:4–18
Aebi M, van Donkelaar MM, Poelmans G, Buitelaar JK, Sonuga-Barke EJ, Stringaris A, et al. Gene-set and multivariate genome-wide association analysis of oppositional defiant behavior subtypes in attention-deficit/hyperactivity disorder. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:573–588
American Psychiatric Association; American Psychiatric AssociationDiagnostic and Statistical Manual of Mental Disorders. 20135th, Arlington, VA: American Psychiatric Publishing
Anderson CA, Bushman BJ. Human aggression
. Annu Rev Psychol. 2002; 53:27–51
Anholt RR, Mackay TF. Genetics of aggression. Annu Rev Genet. 2012; 46:145–164
Anney RJ, Lasky-Su J, O’Dúshláine C, Kenny E, Neale BM, Mulligan A, et al. Conduct disorder and ADHD: evaluation of conduct problems as a categorical and quantitative trait in the international multicentre ADHD genetics study. Am J Med Genet B Neuropsychiatr Genet. 2008; 147B:1369–1378
Emmeke A. Automated Systematic Reviews by Using Deep Learning and Active Learning. 2019https://github.com/msdslab/automated-systematic-review
. [Accessed 18 January 2019]
Bartels M, Hendriks A, Mauri M, Krapohl E, Whipp A, Bolhuis K, et al. Childhood aggression and the co-occurrence of behavioural and emotional problems: results across ages 3-16 years from multiple raters in six cohorts in the EU-ACTION project. Eur Child Adolesc Psychiatry. 2018; 27:1105–1121
Baud P. Personality traits as intermediary phenotypes in suicidal behavior: genetic issues. Am J Med Genet C Semin Med Genet. 2005; 133C:34–42
Boomsma D, Busjahn A, Peltonen L. Classical twin studies and beyond. Nat Rev Genet. 2002; 3:872–882
Boomsma DI. Aggression in children: unravelling the interplay of genes and environment through (epi)genetics and metabolomics. J Ped Neo Ind Med. 2015; 4:e040251
Borah R, Brown AW, Capers PL, Kaiser KA. Analysis of the time and workers needed to conduct systematic reviews of medical interventions using data from the PROSPERO registry. BMJ Open. 2017; 7:e012545
Bosker FJ, Hartman CA, Nolte IM, Prins BP, Terpstra P, Posthuma D, et al. Poor replication of candidate genes for major depressive disorder using genome-wide association data. Mol Psychiatry. 2011; 16:516–532
Boutwell BB, Beaver KM, Barnes JC, Vaske J. The developmental origins of externalizing behavioral problems: parental disengagement and the role of gene-environment interplay. Psychiatry Res. 2012; 197:337–344
Brevik EJ, van Donkelaar MM, Weber H, Sánchez-Mora C, Jacob C, Rivero O, et al; IMAGE Consortium. Genome-wide analyses of aggressiveness in attention-deficit hyperactivity disorder. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:733–747
Brunner HG, Nelen M, Breakefield XO, Ropers HH, van Oost BA. Abnormal behavior associated with a point mutation in the structural gene for monoamine oxidase A. Science. 1993; 262:578–580
Bukowski WM, Laursen BP, Rubin KH. Handbook of peer interactions, relationships, and groups. 2011, New York, NY: Guilford Press
Bulik-Sullivan BK, Loh PR, Finucane HK, Ripke S, Yang J, Patterson N, et al; Schizophrenia Working Group of the Psychiatric Genomics Consortium. LD score regression distinguishes confounding from polygenicity in genome-wide association studies
. Nat Genet. 2015; 47:291–295
Burt SA, Klump KL, Gorman-Smith D, Neiderhiser JM. Neighborhood disadvantage alters the origins of children’s nonaggressive conduct problems. Clin Psychol Sci. 2016; 4:511–526
Buss AH, Durkee A. An inventory for assessing different kinds of hostility. J Consult Psychol. 1957; 21:343–349
Cai N, et al. Minimal phenotyping yields GWAS hits of low specificity for major depression. bioRxiv. 2019440735
Caspi A, McClay J, Moffitt TE, Mill J, Martin J, Craig IW, et al. Role of genotype in the cycle of violence in maltreated children. Science. 2002; 297:851–854
Cohen AM, Hersh WR, Peterson K, Yen PY. Reducing workload in systematic review preparation using automated citation classification. J Am Med Inform Assoc. 2006; 13:206–219
CONVERGE Consortium; CONVERGE Consortium. Sparse whole-genome sequencing identifies two loci for major depressive disorder. Nature. 2015; 523:588–591
Craig IW, Halton KE. Genetics of human aggressive behaviour. Hum Genet. 2009; 126:101–113
Criado JR, Gizer IR, Slutske WS, Phillips E, Ehlers CL. Event-related oscillations to affective stimuli: heritability
, linkage and relationship to externalizing disorders. J Psychiatr Res. 2012; 46:256–263
Curry A. A painful legacy. Science. 2019; 365:212–215
Davydova JD, et al. Recent advances in genetics of aggressive behavior. Vavilov Journal of Genetics and Breeding. 2018; 22:716–725
Dick DM, Aliev F, Krueger RF, Edwards A, Agrawal A, Lynskey M, et al. Genome-wide association study of conduct disorder symptomatology. Mol Psychiatry. 2011; 16:800–808
Dick DM, Li TK, Edenberg HJ, Hesselbrock V, Kramer J, Kuperman S, et al. A genome-wide screen for genes influencing conduct disorder. Mol Psychiatry. 2004; 9:81–86
Dodge KA, Lochman JE, Harnish JD, Bates JE, Pettit GS. Reactive and proactive aggression in school children and psychiatrically impaired chronically assaultive youth. J Abnorm Psychol. 1997; 106:37–51
Dudbridge F. Polygenic epidemiology. Genet Epidemiol. 2016; 40:268–272
Duncan LE, Keller MC. A critical review of the first 10 years of candidate gene-by-environment interaction research in psychiatry. Am J Psychiatry. 2011; 168:1041–1049
Ferguson CJ. Genetic contributions to antisocial personality and behavior: a meta-analytic review from an evolutionary perspective. J Soc Psychol. 2010; 150:160–180
Fernàndez-Castillo N, Cormand B. Aggressive behavior in humans: genes and pathways identified through association studies. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:676–696
Foley DL, Eaves LJ, Wormley B, Silberg JL, Maes HH, Kuhn J, Riley B. Childhood adversity, monoamine oxidase a genotype, and risk for conduct disorder. Arch Gen Psychiatry. 2004; 61:738–744
Gard AM, Dotterer HL, Hyde LW. Genetic influences on antisocial behavior: recent advances and future directions. Curr Opin Psychol. 2019; 27:46–55
Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabási AL. The human disease network. Proc Natl Acad Sci U S A. 2007; 104:8685–8690
Goodman A, Lamping DL, Ploubidis GB. When to use broader internalising and externalising subscales instead of the hypothesised five subscales on the strengths and difficulties questionnaire (SDQ): data from British parents, teachers and children. J Abnorm Child Psychol. 2010; 38:1179–1191
Gusev A, Ko A, Shi H, Bhatia G, Chung W, Penninx BW, et al. Integrative approaches for large-scale transcriptome-wide association studies. Nat Genet. 2016; 48:245–252
Haberstick BC, Lessem JM, Hopfer CJ, Smolen A, Ehringer MA, Timberlake D, Hewitt JK. Monoamine oxidase A (MAOA) and antisocial behaviors in the presence of childhood and adolescent maltreatment. Am J Med Genet B Neuropsychiatr Genet. 2005; 135B:59–64
Hagenbeek FA, Kluft C, Hankemeier T, Bartels M, Draisma HH, Middeldorp CM, et al. Discovery of biochemical biomarkers for aggression: a role for metabolomics in psychiatry. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:719–732
Hawley PH. Strategies of control, aggression, and morality in preschoolers: an evolutionary perspective. J Exp Child Psychol. 2003; 85:213–235
Hawley PH, Little TD, Rodkin PC. Aggression and Adaptation: The Bright Side to Bad Behavior. 20071st ed, New York, New York: L. Erlbaum Associates
Hendriks AM, et al. Content, Diagnostic, Correlational, and Genetic Similarities Between Common Measures of Childhood Aggressive Behaviors and Related Psychiatric Traits. 2019submitted
Hicks BM, South SC, Dirago AC, Iacono WG, McGue M. Environmental adversity and increasing genetic risk for externalizing disorders. Arch Gen Psychiatry. 2009; 66:640
Holland D, et al. Beyond SNP heritability
: polygenicity and discoverability of phenotypes estimated with a univariate gaussian mixture model. bioRxiv. 2019133132doi: 10.1101/133132
Kaye S, Lokki AI, Hanttu A, Nissilä E, Heinonen S, Hakkarainen A, et al. Upregulation of early and downregulation of terminal pathway complement genes in subcutaneous adipose tissue and adipocytes in acquired obesity. Front Immunol. 2017; 8:545
Khabsa M, Elmagarmid A, Ilyas I, Hammady H, Ouzzani M. Learning to identify relevant studies for systematic reviews using random forest and external information. Mach Learn. 2016; 102:465–482doi: 10.1007/s10994-015-5535-7
Kim-Cohen J, Caspi A, Taylor A, Williams B, Newcombe R, Craig IW, Moffitt TE. MAOA, maltreatment, and gene-environment interaction predicting children’s mental health: new evidence and a meta-analysis. Mol Psychiatry. 2006; 11:903–913
Lindenfors P, Tullberg BS. Evolutionary aspects of aggression the importance of sexual selection. Adv Genet. 2011; 75:7–22
Loehlin JC. Kim YK. History of behavior genetics. Handbook of Behavior Genetics. 2009, New York, New York: Springer New York.3–11
Lowe WL Jr, Reddy TE. Genomic approaches for understanding the genetics of complex disease. Genome Res. 2015; 25:1432–1441
Luo X, Stavrakakis N, Penninx BW, Bosker FJ, Nolen WA, Boomsma DI, et al. Does refining the phenotype improve replication rates? A review and replication of candidate gene studies on major depressive disorder and chronic major depressive disorder. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:215–236
Manchia M, Fanos V. Targeting aggression in severe mental illness: the predictive role of genetic, epigenetic, and metabolomic markers. Prog Neuropsychopharmacol Biol Psychiatry. 2017; 77:32–41
McGue M, Zhang Y, Miller MB, Basu S, Vrieze S, Hicks B, et al. A genome-wide Association Study of behavioral disinhibition. Behav Genet. 2013; 43:363–373
Merjonen P, Keltikangas-Järvinen L, Jokela M, Seppälä I, Lyytikäinen LP, Pulkki-Råback L, et al. Hostility in adolescents and adults: a genome-wide association study of the young finns. Transl Psychiatry. 2011; 1:e11
Mick E, McGough J, Deutsch CK, Frazier JA, Kennedy D, Goldberg RJ. Genome-wide association study of proneness to anger. Plos One. 2014; 9:e87257
Mick E, McGough J, Loo S, Doyle AE, Wozniak J, Wilens TE, et al. Genome-wide association study of the child behavior checklist dysregulation profile. J Am Acad Child Adolesc Psychiatry. 2011; 50:807–17.e8
Miles DR, Carey G. Genetic and environmental architecture on human aggression
. J Pers Soc Psychol. 1997; 72:207–217
Moffitt TE. Genetic and environmental influences on antisocial behaviors: evidence from behavioral-genetic research. Adv Genet. 2005; 55:41–104
Montalvo-Ortiz JL, Zhou H, D’Andrea I, Maroteaux L, Lori A, Smith A, et al. Translational studies support a role for serotonin 2B receptor (HTR2B) gene in aggression-related cannabis response. Mol Psychiatry. 2018; 23:2277–2286
Neumann A, Pappa I, Lahey BB, Verhulst FC, Medina-Gomez C, Jaddoe VW, et al. Single nucleotide polymorphism heritability
of a general psychopathology factor in children. J Am Acad Child Adolesc Psychiatry. 2016; 55:1038–1045.e4
Nilsson KW, Åslund C, Comasco E, Oreland L. Gene-environment interaction of monoamine oxidase A in relation to antisocial behaviour: current and future directions. J Neural Transm (Vienna). 2018; 125:1601–1626
Ormel J, Hartman CA, Snieder H. The genetics of depression: successful genome-wide association studies
introduce new challenges. Transl Psychiatry. 2019; 9:114
Pappa I, St Pourcain B, Benke K, Cavadino A, Hakulinen C, Nivard MG, et al. A genome-wide approach to children’s aggressive behavior: the EAGLE consortium. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:562–572
Penn JKM, Zito MF, Kravitz EA. A single social defeat reduces aggression in a highly aggressive strain of Drosophila. Proc Natl Acad Sci U S A. 2010; 107:12682–12686
Polman H, Orobio de Castro B, Koops W, van Boxtel HW, Merk WW. A meta-analysis of the distinction between reactive and proactive aggression in children and adolescents. J Abnorm Child Psychol. 2007; 35:522–535
Porsch RM, Middeldorp CM, Cherny SS, Krapohl E, van Beijsterveldt CE, Loukola A, et al. Longitudinal heritability
of childhood aggression. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:697–707
Provençal N, Booij L, Tremblay RE. The developmental origins of chronic physical aggression: biological pathways triggered by early life adversity. J Exp Biol. 2015; 218:123–133
Raine A. A neurodevelopmental perspective on male violence. Infant Ment Health J. 2019; 40:84–97
Rautiainen MR, Paunio T, Repo-Tiihonen E, Virkkunen M, Ollila HM, Sulkava S, et al. Genome-wide association study of antisocial personality disorder. Transl Psychiatry. 2016; 6:e883
Rhee SH, Waldman ID. Shaver PR, Mikulincer M. Genetic and environmental influences on aggression. Human aggression
and violence: Causes, manifestations, and consequences. 2011, Washington, DC: American Psychological Association143–163
Runions KC, Morandini HAE, Rao P, Wong JWY, Kolla NJ, Pace G, et al. Serotonin and aggressive behaviour in children and adolescents: a systematic review. Acta Psychiatr Scand. 2019; 139:117–144
Salvatore JE, Edwards AC, McClintick JN, Bigdeli TB, Adkins A, Aliev F, et al. Genome-wide association data suggest ABCB1 and immune-related gene sets may be involved in adult antisocial behavior. Transl Psychiatry. 2015; 5:e558
Salvatore JE, Dick DM. Genetic influences on conduct disorder. Neurosci Biobehav Rev. 2018; 91:91–101
Sham PC, Purcell SM. Statistical power and significance testing in large-scale genetic studies. Nat Rev Genet. 2014; 15:335–346
Stranger BE, Brigham LE, Hasz R, Hunter M, Johns C, Johnson M, et al. Enhancing GTEx by bridging the gaps between genotype, gene expression, and disease The eGTEx Project. Nat Genet. 2017; 49:1664–1670
Sonuga-Barke EJ, Lasky-Su J, Neale BM, Oades R, Chen W, Franke B, et al. Does parental expressed emotion moderate genetic effects in ADHD? An exploration using a genome wide association scan. Am J Med Genet B Neuropsychiatr Genet. 2008; 147B:1359–1368
Tielbeek JJ, Medland SE, Benyamin B, Byrne EM, Heath AC, Madden PA, et al. Unraveling the genetic etiology of adult antisocial behavior: a genome-wide association study. Plos One. 2012; 7:e45086
Tielbeek JJ, Johansson A, Polderman TJC, Rautiainen MR, Jansen P, Taylor M, et al; Broad Antisocial Behavior Consortium Collaborators. Genome-wide association studies
of a broad spectrum of antisocial behavior. JAMA Psychiatry. 2017; 74:1242–1250
Tiihonen J, Rautiainen MR, Ollila HM, Repo-Tiihonen E, Virkkunen M, Palotie A, et al. Genetic background of extreme violent behavior. Mol Psychiatry. 2015; 20:786–792
Tremblay RE. Developmental origins of disruptive behaviour problems: the ‘original sin’ hypothesis, epigenetics and their consequences for prevention. J Child Psychol Psychiatry. 2010; 51:341–367
Tremblay RE, Vitaro F, Côté SM. Developmental origins of chronic physical aggression: a bio-psycho-social model for the next generation of preventive interventions. Annu Rev Psychol. 2018; 69:383–407
Trivedi DK, Hollywood KA, Goodacre R. Metabolomics for the masses: the future of metabolomics in a personalized world. New Horiz Transl Med. 2017; 3:294–305
Tuvblad C, Baker LA. Human aggression
across the lifespan: genetic propensities and environmental moderators. Adv Genet. 2011; 75:171–214
van Dongen J, Nivard MG, Willemsen G, Hottenga JJ, Helmer Q, Dolan CV, et al; BIOS Consortium. Genetic and environmental influences interact with age and sex in shaping the human methylome. Nat Commun. 2016; 7:11115
van der Sluis S, Posthuma D, Nivard MG, Verhage M, Dolan CV. Power in GWAS: lifting the curse of the clinical cut-off. Mol Psychiatry. 2013; 18:2–3
Vassos E, Collier DA, Fazel S. Systematic meta-analyses and field synopsis of genetic association studies of violence and aggression. Mol Psychiatry. 2014; 19:471–477
Veroude K, Zhang-James Y, Fernàndez-Castillo N, Bakker MJ, Cormand B, Faraone SV. Genetics of aggressive behavior: an overview. Am J Med Genet B Neuropsychiatr Genet. 2016; 171B:3–43
Viding E, Hanscombe KB, Curtis CJ, Davis OS, Meaburn EL, Plomin R. In search of genes associated with risk for psychopathic tendencies in children: a two-stage genome-wide association study of pooled DNA. J Child Psychol Psychiatry. 2010; 51:780–788
Vink JM, Bartels M, van Beijsterveldt TC, van Dongen J, van Beek JH, Distel MA, et al. Sex differences in genetic architecture of complex phenotypes? Plos One. 2012; 7:e47371
Visscher PM, Wray NR, Zhang Q, Sklar P, McCarthy MI, Brown MA, Yang J. 10 years of GWAS discovery: biology, function, and translation. Am J Hum Genet. 2017; 101:5–22
Voulgaridou I, Kokkinos CM. Relational aggression in adolescents: a review of theoretical and empirical research. Aggression and Violent Behavior. 2015; 23:87–97
Waltes R, Chiocchetti AG, Freitag CM. The neurobiological basis of human aggression
: a review on genetic and epigenetic mechanisms. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:650–675
Wild CP. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Prev. 2005; 14:1847–1850
Wright JC, Giammarino M, Parad HW. Social status in small groups: individual–group similarity and the social "misfit". J Pers Soc Psychol. 1986; 50:523–536
Young SE, Smolen A, Hewitt JK, Haberstick BC, Stallings MC, Corley RP, Crowley TJ. Interaction between MAO-A genotype and maltreatment in the risk for conduct disorder: failure to confirm in adolescent patients. Am J Psychiatry. 2006; 163:1019–1025
Zhang-James Y, Faraone SV. Genetic architecture for human aggression
: a study of gene-phenotype relationship in OMIM. Am J Med Genet B Neuropsychiatr Genet. 2016; 171:641–649
Zhang-James Y, Fernàndez-Castillo N, Hess JL, Malki K, Glatt SJ, Cormand B, et al. An integrated analysis of genes and functional pathways for aggression in human and rodent models. Mol Psychiatry. 2018[Epub ahead of print]. doi: 10.1038/s41380-018-0068-7 [doi];10.1038/s41380-018-0068-7 [pii]
Zhu Z, Zhang F, Hu H, Bakshi A, Robinson MR, Powell JE, et al. Integration of summary data from GWAS and eqtl studies predicts complex trait gene targets. Nat Genet. 2016; 48:481–487