COMT and the 'warrior vs worrier' gene: a real enzyme, an oversold story

What COMT does

Dopamine, noradrenaline and adrenaline belong to a chemical family called catecholamines. COMT inactivates them by attaching a methyl group, and it does the same to catechol drugs such as L-DOPA, the Parkinson's medicine^[3]. Its products are measurable in blood and urine; one of them, 3-methoxytyrosine, shows up again later in this article^[4].

The brain region the story cares about is the prefrontal cortex, the part behind the forehead that handles planning, holding things in mind and switching between rules. Dopamine there modulates how prefrontal neurons respond during working memory^[5], and COMT is a key enzyme in clearing it^[1]. So a variant that slows COMT should, in principle, leave a little more dopamine around in exactly the place where thinking happens. That chain of reasoning is sound. Everything that follows is about how much it matters.

One letter, two versions of the enzyme

At rs4680 the DNA reads either G or A. With G, codon 158 of COMT is GTG and codes for valine; with A it is ATG and codes for methionine^[2]. Everyone carries two copies of the gene, so there are three genotypes: Val/Val, Val/Met and Met/Met.

The number usually quoted for the difference is three- to fourfold. It comes from the 1996 paper that tied the variant to a long-known variation in COMT activity between people^[3], and it was repeated as a fourfold variation in the 2001 study that started the cognitive story^[5]. When Chen and colleagues measured the enzyme directly in 108 postmortem brains, the gap in the prefrontal cortex was about 38 percent in favour of Val/Val, which the authors noted was much lower than the two- to fourfold differences other groups had reported from blood, liver and purified enzyme^[1]. The Met protein was less abundant while the messenger RNA was not, which points to a less stable protein rather than a gene that is read less often^[1].

Where the warrior and the worrier came from

In 2001 Egan and colleagues tested 175 people with schizophrenia, 219 of their unaffected siblings and 55 controls on the Wisconsin Card Sorting Test, a task in which the sorting rule changes without warning and the score counts how often a person keeps using the old one^[5]. The more Met copies a person had, the fewer of those perseverative errors they made. Genotype explained 4 percent of the variation in the score, and small brain-imaging subgroups of 11 to 16 people showed a more efficient prefrontal response with more Met^[5].

Two years later a brain-imaging study in Science reported the other half. During a sustained pain challenge, Met/Met volunteers showed a weaker response of the brain's own opioid system than heterozygotes, rated the pain as worse and reported a more negative mood, and Val/Val volunteers showed the opposite^[8]. Put the two together and you have a trade: Met for better thinking when things are calm, Val for coping when they are not.

By 2006 the trade had its name. A review titled 'Warriors versus worriers' proposed that Val carriers might have an advantage in processing aversive stimuli, a 'warrior strategy', while Met carriers might have an advantage in memory and attention, a 'worrier strategy', and noted some evidence linking Met to anxiety^[9]. The review hedged each step with 'may be associated'. The idea underneath is the inverted-U model of dopamine's effect on the frontal cortex^[10], in which function is best at a middle level of dopamine, so a slower enzyme helps or hurts depending on the starting point and the moment. The label turned a hypothesis about averages in small groups into a personality type before anyone had tested it at scale.

What the larger studies found

Cognition was tested first and most. In 2007 Barnett and colleagues pooled every study up to August 2006 that reported Wisconsin Card Sorting errors by genotype: 12 studies and 1,910 people^[6]. In healthy controls, Met/Met did better than Val/Val with an effect size of 0.29, a confidence interval from 0.02 to 0.55 and a p value of 0.03. In patients with schizophrenia there was no effect. The authors also noted that the effect was larger in studies published earlier^[6].

A year later the same first author, with Munafò, widened the net to six measures: the card-sorting test, the n-back working-memory task, trail making, verbal recall, verbal fluency and IQ^[11]. Most showed no association. IQ showed one with an effect size of 0.06, six hundredths of a standard deviation, and there was evidence of publication bias, with effect sizes falling as later studies came in.

Large single cohorts agree. In the Avon Longitudinal Study of Parents and Children in England, between 1,857 and 2,659 eighteen-year-olds took an n-back test, and genotype predicted neither accuracy nor discrimination^[12]. The same cohort's children gave the most generous result in this area: at ages 8 and 10, a three-variant COMT haplotype that includes rs4680 predicted verbal inhibition and working memory in a curved, inverted-U pattern^[10]. That is a finding about a haplotype in children, and it falls well short of a temperament.

The 'worrier' half fares worse. Lee and Prescott pooled 27 studies of anxiety-related traits in 15,979 people^[13]. Where they found anything, it ran against the label: Val/Val White men scored slightly higher on neuroticism than Met/Met White men, with an effect size of 0.13, and Val/Val Asian men higher on harm avoidance. Women showed no association, and the effects shrank when studies were pooled across ethnic groups or across anxiety measures.

Pain is where the variant comes closest to a replicated behavioural effect. A 2012 meta-analysis found that fibromyalgia or chronic widespread pain was the only chronic pain condition that could be linked to rs4680, with Met as the risk allele, and found no link with migraine-type headache or chronic musculoskeletal pain^[14]. A 2023 meta-analysis of 31 samples and 4,631 people found small effects of rs4680 on pain thresholds in fibromyalgia, headache and across chronic pain conditions^[15]. Small effects on pain thresholds in patients are a long way from a coping style.

Psychiatric genetics, candidate genes and genome-wide scans

COMT was one of the most studied candidate genes in psychiatry, schizophrenia above all. The 2016 meta-analysis of 67 case-control studies, 15,565 people with schizophrenia and 17,251 controls, found an association only under some genetic models: an odds ratio of 1.08 under a recessive model, 1.21 in what the paper groups as Caucasian samples, and none in Asian samples^[16]. Small, model-dependent odds ratios are exactly what genome-wide studies were built to test.

The test came. A 2022 genome-wide study compared 76,755 people with the condition and 243,649 controls and found 287 distinct loci^[17]. Not one of them sits near COMT. The GWAS Catalog lists two weaker signals from that study within a megabase of the gene, in its neighbours ARVCF and ZDHHC8, at p values of 1 and 5 in a million, well short of the genome-wide threshold^[4]. The same search across six more large studies of depression, bipolar disorder, ADHD, neuroticism, intelligence and anxiety found nothing on that stretch of chromosome 22 at all^{[18, 19, 20, 21, 22, 23, 4]}.

There is one real link between this part of chromosome 22 and schizophrenia, and it is a different thing. The 22q11.2 deletion removes a stretch of 0.7 to 3 million bases that includes COMT, and about a quarter of people with the deletion are diagnosed with schizophrenia^[24]. Losing a whole block of genes is not the same as carrying the Met letter, and in adults with the deletion, multiple studies have found no association between the Val/Met variant and schizophrenia^[24].

The GWAS Catalog lists 39 associations for rs4680^[4]. Seven are protein measures, six of them of COMT itself, some with p values below 10 to the power of minus 1,000, and 29 are metabolite levels, many of them COMT's own products such as 3-methoxytyrosine. Two are body size, a body mass index result and a body-shape score. The one behavioural entry is a 2023 genome-wide study of internet addiction in Iran^[25]. Genome-wide studies have found the enzyme many times over and the personality not at all.

Supplements matched to a genotype

The commercial version of the story adds a shopping list. Forums pair Met/Met with one set of supplements, Val/Val with another, and both with lists of things to avoid, usually reasoning from the inverted U: a slow enzyme means more dopamine, so lower it; a fast one means less, so raise it. The biochemistry supplies the motive. It does not supply evidence that the plan does anything.

The largest supplement trial we found that took COMT genotype into account shows what such a trial looks like. The Minnesota Green Tea Trial randomised 937 postmenopausal women to green tea extract or placebo for a year, stratified by COMT genotype^[26]. A report on 237 of the women measured weight, waist and hormones such as insulin and adiponectin; the extract did not change weight, and genotype made a difference to insulin and adiponectin whichever capsule the women took. It measured nothing about mood, focus or stress. In the trial literature we searched on 10 October 2026, we found no randomised trial that tested whether choosing a supplement by COMT genotype improves any of those.

How common each genotype is

A personality type ought to be rare enough to say something. In the Finnish genomes in gnomAD, 55.3 percent of the chromosomes carry Met^[7]. Working from the allele and homozygote counts, that puts 30.3 percent of Finns at Met/Met, 49.9 percent at Val/Met and 19.8 percent at Val/Val. The supposed worrier is common in Finland and the supposed warrior is the rarest of the three.

Met/Met is about four times as common in Finland as in East Asian samples. Taken literally, the label would make Finland a nation of worriers.

What the Aimosti report shows

rs4680 has a card in the Frontier section, the exploratory part of the report that sits behind a divider marked as a credibility firewall and opens with a warning that nothing in it is medical, diagnostic or a basis for any decision. The card names the genotype as Val/Val (faster clearance), Val/Met (intermediate) or Met/Met (slower clearance), labels the evidence 'Mixed evidence', and cites its sources, among them the 2007 card-sorting meta-analysis and the GWAS Catalog. Frontier cards are left out of the clinician hand-off page. The fitness cards in the same section are covered in our piece on sports gene tests.

How the genotype is read depends on the file. A chip export from 23andMe, AncestryDNA or MyHeritage is read by the rs number. If the export has no called row for rs4680, the card stays unresolved and says the marker was not read, rather than showing a default. A VCF or gVCF is read by the rs number or the position. A plain VCF lists only differences from the reference, and the reference letter here is G, Val, so a Val/Val genome usually has no rs4680 line at all; the report then shows Val/Val and says plainly that this is an inference from silence. A gVCF can do better: when its reference blocks show the COMT region was read, at least 90 percent of it, the report treats the missing line as a reading. A BAM or CRAM is used only for the Deep Read panel, which does not call rs4680, so it adds nothing to this card.

# rsid    chromosome  position   genotype      (23andMe-style chip export, GRCh37)
rs4680    22          19951271   AG

#CHROM  POS        ID      REF  ALT  ...  GT        (VCF, GRCh38)
22      19963748   rs4680  G    A    ...  0/1