Discover more from Traitwell
A review of the evidence for MAO-A
Like most human traits, anti-social behavior and aggression in particular are known to be heritable. This has now been established by large twin studies and other comparisons of relatives and adoptees. Without pretending to unwarranted precision, the known heritability lies somewhere between 40 and 50%, which is within the range of most traits (Patrick et al 2006; Flannery et al 2007; Knopick et al 2016). With age, heritability of anti-social behavior increases, just as it does for most other traits, due to the increased ability of people to seek out and shape the environments that suit them (earlier studies which ignored this grossly underestimated heritability for many traits). In a sense, life is the triumph of genes over experience. This implies that genes play an important role in the development of aggression and anti-social behavior. However it does not tell us which genes, specifically, are involved. But we have known for some time about the important role of at least one gene: MAO-A, the so-called “warrior gene”.
In 1978 a woman—let’s call her the complainant—approached the University Hospital of Nijmegen in the Netherlands for help. Many of her male relatives behaved oddly, including her brother. They had an unnerving aggressive look in their eyes. Now her 10 year-old son was behaving problematically. She was seen by Dr. Han G. Brunner. The doctor proved to be extraordinarily persistent in flaking open the details of her trouble, in a way seldom seen outside of fanciful television dramas.
For more than 15 years Dr. Brunner tracked down relatives, some of them in shelters, interviewing them and obtaining blood samples from 24 of them. This was made easier by the complainant’s grand-uncle, affable by comparison, who had compiled a tree of his relatives thirty years previously. It stretched to four generations. The male kindred had a history of violence to match their gaze. At least 14 males were affected. When they were not raping their sisters, groping them and forcing them to undress at knifepoint, or sinking pitchforks into opponents during brawls, they were setting things on fire, exposing themselves in supermarkets, or attempting suicide. Socially withdrawn and shy, once angered they overreacted dramatically. Episodes could last for three whole days, during which they slept little and had violent night terrors. Many were on intimate terms with the law. As a whole the affected members were borderline mentally retarded, with a median IQ of roughly 85, and mildly dysmorphic (that is, oddly shaped). Only one had completed primary education and obtained a job. By contrast the unaffected family members had completed school and had jobs. It was natural to suspect a genetic dimension to all this, as the original complainant had.
Brunner found definite evidence when analyzing urine from some of the impulsive family, five of which he examined personally. Three had a clear monoamine disorder, betrayed by chemical products in the urine. Only males were affected, suggesting something passed down through females but not developed by them. DNA analysis showed that the carrier had a point-sense mutation on the X chromosome. This exon mutation knocked out the ability of the MAO-A genes (located at region Xpl 1.23-11.4) to properly manufacture the oxidase enzyme which selectively metabolizes monoamines. That enzyme, located in the mitochondria, is thought to be critical for effective regulation of the serotonin, dopamine and noradrenaline neurotransmitters in the monoamine group. Transmitters that are in turn linked to behavior. In stark popular terms, the genetic defect causes the brain to “misfire”, and violent behavior is one of the results.
A full but extremely cautious description of some of these findings was published by Brunner and his colleagues in a series of papers (Brunner 1993a, 1993b). As a result, the condition in which the MAOA-A gene is knocked out is now known as Brunner syndrome. The authors were no doubt well aware that attributing aggression to genetic foundations would attract hostile criticism and perhaps even lower their status among colleagues. They took care to avoid “reductionism” of a “complex behavior” to an “aggression gene”. So they made the automatic call for further research.
One obvious way to proceed was to try animal experiments, where variables can be controlled at will in ways that human observations cannot. Many human genes have homologs in other species, a consequence of evolution’s frugality and make-shifts. The call was soon answered by successive teams involving Jean Shih, then at the University of Southern California, Los Angeles, working with inbred mouse strains. At first the researchers knocked out the mouse MAO-A gene explicitly (Cases et al 1995). Under close observation these mice proved to be much more aggressive, along with other severe behavioral abnormalities, than mice with functioning MAO-A genes. The aggressive behavior emerged soon after weaning, as shown by bite wounds. Later, in collaboration with a different team, Shih was able to find a strain of lab mice serendipitously, in which the homolog of the MAO-A gene was already disabled (Scott et al 2008). Dead mice would show up more frequently in the mornings when their cages were inspected. Both strains rapidly attacked intruders in controlled tests. They concluded that the evidence “supports the idea that the particularly aggressive behavior of the few known human males lacking MAOA is not fostered by an unusual genetic background or complex psychosocial stressors but is a more direct consequence of MAOA deficiency.”
More evidence among humans was reported by Avshalom Caspi and his colleagues in 2002, using data collected by the large Dunedin Longitudinal Study in New Zealand. Caspi, then at King’s College London, was perplexed by the fact that childhood maltreatment, supposed by the sympathetic to be a cause of aggressive behavior and other psychological abnormalities, appeared to have varying effects. Many children, indeed most, subjected to it did not develop those behaviors, only some did. It was natural to suspect an interaction effect involving some other factor. Knowing of Brunner’s MAO-A findings, Caspi and his collaborators chose it as a candidate gene possibly interacting with childhood abuse. This is not the idea that both factors may contribute (additively) to the outcome, but rather the idea that one factor moderates the outcome of the other, multiplicatively. If it is true then subjecting children to abuse alters the effect of the MAO-A gene, making it much more extreme than it is without abuse (see below for a helpful figure). This is known as a “GxE” effect, because it emerges as a multiplicative term in statistical models, which can estimate its relative weight based on data sets. (Note that “interaction” can also be used to mean other things, not relevant here).
The Dunedin longitudinal study had been running for many years with a representative cohort of 1,037 individuals, who were closely followed from age 3, as they grew up. The subjects were then 27 years old. The cohort had been largely preserved, with a very low attrition rate, maintaining representativeness. There was good data, from their case history interviews, about the treatment that the children had been naturally subjected to. Gene samples were taken to test for MAO-A activity. The children did not have that gene knocked out as Brunner’s subjects had, and the mice had, but some of them did have a low-functioning variant. This variant had been discovered by Sabol et al in 1998, and is caused by a particular variable number tandem repeat in the DNA. In particular 2, 3 and 5 repeats seem to produce low-functioning MAO-A, whereas 3.5 or 4 repeats produce high-functioning MAO-A (though Manca et al 2018 dispute this).
A composite index of anti-social behavior for each subject was constructed and validated. A statistical model was then used to determine how well the gene, the abusive environment and a multiplicative interaction of the two predicted anti-sociality in the data set. It turned out that both factors contributed separately, that is additively, but also that the interaction was statistically significant and could not be excluded. The figure below is from the resulting paper, Caspi et al 2002, and shows the purported interaction clearly. It demonstrates that the gene moderates the effect of the environment, but a similar figure could show the environment as a moderator instead, the ideas are interchangeable.
The authors, like Brunner before them, were appropriately cautious. “Until this study’s findings are replicated, speculation about clinical implications is premature.” Still, it looked like some hard data supported that idea of a GxE interaction on an important outcome, delinquency. The effect size was impressive: “although individuals having the combination of low-activity MAOA genotype and maltreatment were only 12% of the male birth cohort, they accounted for 44% of the cohort’s violent convictions, yielding an attributable risk fraction (11%) comparable to that of the major risk factors associated with cardiovascular disease.” And “85% of cohort males having a low-activity MAOA genotype who were severely maltreated developed some form of antisocial behavior.”
Though interaction effects are much beloved by social scientists—perhaps because they support a perceived need for “nuance” and “complexity” in explanations, and usually involve a favored factor, the “environment”—they are rarely substantiated using field or observational data. Consequently the Caspi study of 2002 received a lot of attention, and a slew of studies looking for interaction effects followed. Interest in the idea has only grown since then. Several of these subsequent studies attempted to replicate the Caspi results specifically. A number of them succeeded in replicating the result. A successful example is given by Fergusson et al 2012, who worked with 398 males from the Christchurch Health and Development Study, also in New Zealand, and took pains to conduct the study along similar lines. The authors concluded that “The present findings add to the evidence suggesting that there is a stable G×E interaction involving MAOA, abuse exposure and antisocial behaviour across the life course”. However not all the replications succeeded, and several found no definite evidence for an interaction effect—for a list, see Byrd & Manuck, 2014.
Before considering the replications further, it is worth considering the inherent ambiguities in such findings first, even if they do hold up. An abusive environment may be acting here as a proxy for other genes for aggression. Like most traits, aggression is likely to be polygenic with at the very least an additive effect. Both parents and offspring would have many such genes. It is certainly possible that a combination of “abusive environment” and “MAO-A low functioning” could just be code for “has multiple genes predisposing for aggression” in the parent, and therefore the children too. Possibly many of them, to cross the threshold in environmental classes. Without measuring those other genes somehow it is hard to say. Caspi et al did verify that MAO-A genes did not seem to be more prevalent in their sample among those subjected to abuse, so that the two factors were not correlated. (This is itself puzzling: surely one would expect them to be correlated to some extent, since it would tend to give parents a short fuse?) But that leaves the issue of other genes unresolved. Twins would help to resolve the question but their presence in the right design here would be a minor miracle.
This yo-yo effect, in which findings appear only to disappear then reappear later is well-known in the social sciences. It was already apparent by the 1960s (Hunter & Schmidt 2014). One major cause of this is sampling error. Most social science studies are conducted using relatively modest sample sizes, because of numerous constraints, including cost. The samples used may not be representative, merely by chance, because of low numbers or because of carelessness about restriction of range and sample selection. (Indeed a vast number of academic studies have blithely used students—a captive resource—as subjects. Students are highly atypical, being pre-selected for intelligence and other features.) Measurement error can also vary a lot between studies, so that one may find an effect and another may not.
To solve these problems, the technique called meta-analysis was invented in the mid-1970s, independently by two completely separate teams for slightly different purposes: John Hunter and Frank Schmidt, and Gene Glass et al. (see Hunter & Schmidt 2014; Glass 1981). The essence of the technique in either version is to pool the data from previous studies into a much larger study, the bigger the better, and correct for attenuation due to measurement error and other artifacts. This reduces the sampling variation that is the fundamental (certainly, the most interesting) cause of the yo-yo phenomenon, and increases the power of the study to find an effect. A study that fails to find an effect is much more impressive if there is a strong expectation that it was capable of detecting the effect in the first place, otherwise it was no more than a useless diversion.
The variable follow-up studies on MOA-A interactions have been subjected to more than one meta-analysis, ironically not always with the same result. The most comprehensive is that by two Pittsburgh psychologists, Amy L. Byrd and Stephen B. Manuck (see Byrd & Manuck, 2014) They include a review of the dueling meta analyses to that date, and present their own meta-analysis of fully 27 studies, for a total of 18,400 subjects. A subset of 20 studies yielded data explicitly for 11,064 males. “We found common regulatory variation in MAOA to moderate effects of childhood maltreatment on male antisocial behaviors, confirming a sentinel finding in research on gene-environment interaction.” Many others agree that the interaction seems established (Knopick et al 2016). Possibly.
The last qualifier is necessary for a good reason. Interaction effects are notoriously hard to find in observational (as opposed to experimental) data, for relatively intractable reasons (McGue and Carey, 2017). As mentioned above, measurement error is a significant attenuating influence. When interaction effects are modeled, these errors don’t just cumulate, they multiply. Likewise restrictions of range in the data, in which perhaps only a small amount of real world variation is captured, multiplies again in the interaction term. The sharply amplified attenuating effect can be shown algebraically (see McClelland & Judd 1993c) but the idea is easy to grasp intuitively. Moreover, the factors involved are often, perhaps usually, not independent. They co-vary for many reasons. Humans are not passive but active, seeking out niches and shaping them to their needs relentlessly, in close concert with their fellows. This means that observations do not necessarily contain the contrasts needed to tease apart and validate the supposed effects. When deliberate experiments are conducted, those contrasting combinations can be guaranteed, but experimentation on humans here would be unethical. Simulations have shown that most studies claiming to find interactions probably do not have large enough sample sizes to have sufficient power to do so (Tolan et al, 2017). This increases the odds of publication bias, in which we see more successes than failures published because resumes demand them.
It is as an interaction effect that the low-functioning variant of MAOA-A is currently believed to act. Although (generally disappointing and un-replicable) candidate gene studies have now been crowded out by genome-wide association studies (GWAS), and several of those have been conducted in search of traits allied to aggression, none of the broader studies have considered interactions. It is also true that none has identified MAO-A as one of the “hits” but this is not as interesting as it may appear, leaving aside the neglected interaction for the moment. As of 2019 no GWASs on aggression had been able to detect any specific genes, simply because they were not large enough and therefore underpowered (Odintsova et al 2019). The situation is more or less unchanged today, but given the known heritability of aggression it is bound to improve dramatically when sample sizes are raised to the millions and more, as it has for other traits.
By their very nature GWASs are not suited to the task of evaluating candidate genes anyway, being rather an hypothesis-free exercise in trawling for genes in bulk, using many independent models, one for each gene considered, and huge data sets to develop power. It is to fish with dynamite rather than a line, to see what floats up. Multiple genes may contribute an effect to the outcome, which is captured in a composite polygenic score. These protean polygenic scores (there are many methods choose from) are driven by their explanatory and predictive ability, and it is really not so important which individual genes contribute to them, as each gene typically has a tiny additive effect on the score overall anyway. Although the statistical significance of each gene found in this way is an additional outcome of the exercise, and useful for turning up completely unsuspected candidates, a stringent and very conservative statistical correction must be applied in order to account for the multiple comparisons made when trawling through many genes. Empirically, polygenic scores which simply ignore statistical significance have been found to have the best predictive power on hold-out test sets (likewise in the machine learning field, significance plays little role in selecting predictors). If, and it is a big if, you know what gene you are looking for, GWAS is not the method you should use to evaluate it since it provides no help in that exercise. There you want to fly-fish with lines and not with dynamite. Adequately powered replication studies are the method of choice, which must be subjected to meta-analysis as explained above to boost power even further. This allows subtleties like interactions to be taken into account, always bearing in mind the caveats given above.
For all these reasons, the effect of MAO-A on aggression and impulsiveness, and in particular its interactions with childhood abuse, should be accepted provisionally, until contrary data emerges. This is no different from any other scientific finding depending on inherently limited and noisy observational data.
Brunner syndrome, in which the gene is completely knocked out and borderline mental retardation is also present, is very rare. However the variant in which there is low rather than no MAO-A activity is present in about 30% of the members of European populations—globally, its prevalence appears to vary a lot in other sub-populations. As mentioned above, this low-activity version is caused by certain repeats in the related DNA sequence, which some people carry. These errors may have arisen independently multiple times, after which they were inherited. We may speculate that there could have been selective value to this variant in the evolutionary past, when aggression may have been useful for survival. This would explain why, like psychopathy and other anti-social disorders, it has not been entirely eliminated. Selection for the variant would have depended on it remaining relatively infrequent, otherwise it may have become counter-productive. A society dominated by people at each other’s throats all the time, each for his own, is unlikely to succeed, nor are the genes propagated within it (see Shackleford et al 2016 for extensive discussion and references). This sort of reasoning, for which there is a surprising amount of indirect evidence for many traits, is the source of the phrase “warrior gene”.
To definitely establish whether a person has the low-functioning MAO-A variant, a full genome sequence is required in order to detect the exact repeat number that the person carries. However there are some useful proxies available which can be used to suggest whether nor not a person may have the low-functioning variant. Detecting these proxies does not require a full genome sequence. They can be read from the common SNPs sequenced by popular DNA chipsets. In practice an association has been found between these proxies and repeat number variants, but it is by no means definitive. And note that even if you do have low-activity MAO-A, this does not imply that you are necessarily aggressive or impulsive. It is only one association.
If you are curious, you can use Traitwell’s app to first test for the proxies, then decide if you want to go further. Are you curious? We sure are. At Traitwell you can also order a complete DNA sequence for more definite results.
1981. Glass, Gene. Meta-Analysis in Social Research. Sage.
1993a. 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. Oct 22;262(5133):578-80.
1993b. Brunner HG, Nelen MR, van Zandvoort P, Abeling NG, van Gennip AH, Wolters EC, Kuiper MA, Ropers HH, van Oost BA. “X-linked borderline mental retardation with prominent behavioral disturbance: phenotype, genetic localization, and evidence for disturbed monoamine metabolism”. American Journal of Human Genetics Jun; 52(6): 1032-9.
1993c. McClelland, G. H., & Judd, C. M. “Statistical difficulties of detecting interactions and moderator effects”. Psychological Bulletin, 114(2), 376–390.
1994. Rowe, David C. The Limits of Family Influence. Guildford.
1995. Cases O, Seif I, Grimsby J, Gaspar P, Chen K, Pournin S, Müller U, Aguet M, Babinet C, Shih JC, et al. Aggressive behavior and altered amounts of brain serotonin and norepinephrine in mice lacking MAOA. Science. 1995 Jun 23;268(5218):1763-6.
1998. Sabol SZ, Hu S, Hamer D. “A functional polymorphism in the monoamine oxidase A gene promoter”. Hum Genet. Sep;103(3):273-9.
2002. Caspi A, McClay J, Moffitt TE, Mill J, Martin J, Craig IW, Taylor A, Poulton R. “Role of genotype in the cycle of violence in maltreated children”. Science. 2002 Aug 2; 297(5582): 851-4.
2008. Scott, A. L., Bortolato, M., Chen, K., & Shih, J. C. (2008). “Novel monoamine oxidase A knock out mice with human-like spontaneous mutation”. Neuroreport, 19(7), 739–743.
2006. 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”. Molecular Psychiatry. 2006 Oct; 11(10):903-13.
2006. Patrick, Chistopher J. et al (ed.) Handbook of Psychopathy. Guildford.
2007. Flannery, D. J. et al. (ed.) The Cambridge Handbook of Violent Behavior and Aggression. CUP.
2007. Taylor A, Kim-Cohen J. “Meta-analysis of gene-environment interactions in developmental psychopathology”. Developmental Psychopathology Fall;19(4):1029-37.
2012. Fergusson, D. M., Boden, J. M., Horwood, L. J., Miller, A. L., & Kennedy, M. A. (2011). “MAOA, abuse exposure and antisocial behaviour: 30-year longitudinal study”. The British journal of psychiatry 198(6), 457–463.
2013. Raine, Adrian. The Anatomy of Violence: The Biological Roots of Crime. Pantheon.
2014. Byrd, A. L., & Manuck, S. B. (2014). “MAOA, childhood maltreatment, and antisocial behavior: Meta-analysis of a gene-environment interaction.” Biological Psychiatry, 75(1), 9–17.
2014. Hunter, John and Frank Schmidt. Methods of Meta-Analysis: Correcting Error and Bias in Research Findings Third Edition. Sage.
2016. Knopik, Valerie S. and Jenae M. Neiderhiser, John C. DeFries, Robert Plomin. Behavioral Genetics. Seventh Edition. Worth.
2016. Shackleford, Todd et al The Evolution of Violence. Springer.
2017. Tolan, Patrick H. and Bennett L. Leventhal (ed.) Gene-Environment Transactions in Developmental Psychopathology: The Role in Intervention Research. Springer.
2017. McGue, Matt and B. Carey “Gene–Environment Interaction in the Behavioral Sciences: Findings, Challenges, and Prospects” in Tolan & Leventhal, 2017.
2018. Flannery, D. J. et al. (ed.) The Cambridge Handbook of Violent Behavior and Aggression. Second Edition. CUP.
2018. Manca, M., Pessoa, V., Lopez, A.I. et al. “The Regulation of Monoamine Oxidase A Gene Expression by Distinct Variable Number Tandem Repeats.” J. Mol. Neurosci. 64, 459–470 (2018).
2019. Odintsova VV, Roetman PJ, Ip HF, Pool R, Van der Laan CM, Tona KD, Vermeiren RRJM, Boomsma DI. Genomics of human aggression: current state of genome-wide studies and an automated systematic review tool. Psychiatr Genet. Oct;29(5):170-190.