Sports gene tests: what ACTN3 and ACE can and cannot say

What sports DNA tests read

The consensus group searched for companies marketing genetic tests for sport, exercise performance or injury, and found 39^[1]. More than half, 21 of the 39, published nothing about which DNA variants they tested; the authors read that silence as a tacit admission that the evidence behind the chosen variants is weak. Among the 18 that did publish, two genes dominated.

More has been published on these two variants than on any others in sport and exercise, which the authors took to be the reason for the choice^[1]. That makes them the fair test of the category. Beyond them, the statement says, the evidence for choosing any particular variant is extremely weak or non-existent.

ACTN3 R577X and the sprinters

Alpha-actinin-3 is a structural protein of the Z disc in fast-twitch muscle fibres, the ones that produce force at high speed^[3]. In 1999 a research group reported that a common nonsense variant in its gene, ACTN3, leaves people in the general population with none of it^[2]; in their later study it was absent in 18 percent of healthy white people^[3]. The variant, R577X, turns codon 577 from arginine, R, into a stop signal, X. In a raw data file it is rs1815739, and the X allele is written T^[4].

Two X copies cause no disease; the related protein alpha-actinin-2 is thought to stand in^[3]. Expecting any effect to show at the extremes of performance, the same group worked with the Australian Institute of Sport and genotyped 429 elite white athletes from 14 sports, split into sprint and power events and endurance events, against 436 white controls.

The sprint result was the striking one. Six of 107 sprint and power athletes had two X copies, against 80 of 436 controls, and none of the 35 female sprinters had them, nor did any of the 32 sprint athletes who had competed at Olympic level^[3]. The endurance athletes leaned the other way, significantly only in women, and the authors concluded that the effect differs between the sexes.

A Finnish study pointed the same way. Among 89 Finnish elite sprint athletes and 52 endurance athletes, the XX genotype was more frequent and RR less frequent in the endurance group, and none of the top Finnish sprinters had XX^[5].

How common XX is, and why Finland is different

The 18 percent figure that travels with this gene comes from white Australian controls^[3]. It is not a world figure, and it is not a Finnish one. Of all the genomes in gnomAD, 16.0 percent have two X copies, and the share runs from 3.0 percent in the African and African American group to 34.2 percent in the South Asian one^[4].

Finland is the outlier in Europe. 514 of 5,287 Finnish genomes have two X copies, 9.7 percent, against 19.4 percent of non-Finnish European genomes; the X allele itself is carried on 31.0 percent of Finnish chromosomes and 44.2 percent of other European ones^[4]. That puts about one Finn in ten in the group our own Frontier card labels endurance-leaning (our arithmetic on gnomAD's counts), and the population average says nothing about which ten.

How much of performance the variants explain

A 2013 meta-analysis pooled the athlete studies for both genes^[7]. Across all sports, RR athletes were no more common than expected: an odds ratio of 1.03, with a confidence interval from 0.92 to 1.15. Restricted to power events, the odds ratio was 1.21 (1.03 to 1.42). For ACE, the II genotype went with athletic status at 1.23 (1.05 to 1.45) and with endurance events at 1.35 (1.17 to 1.55).

The consensus statement turned that figure into people. An odds ratio of about 1.2 sounds like a 20 percent better chance of being an elite sprinter, but the UK has an estimated 20 million people with the RR genotype, and only a tiny fraction of them are elite athletes^[1]. A genotype held by about one Briton in three (our arithmetic on those two figures) cannot pick out the few.

Studies of ordinary people measured the effect directly. In 992 Greek adolescents, ACTN3 genotype accounted for 2.3 percent of the variation in 40 m sprint time in boys and was not associated with endurance^[8]. In 602 adults on a 12-week arm strength programme, about 2 percent of baseline strength and of the strength gained was attributable to the genotype^[9]. The consensus authors cite an estimate of 2 to 3 percent for sprint performance and judge the true figure, from the wider literature, as probably under 1 percent^[1].

The rest of the heritability does not sit in a handful of other genes either. A 2008 calculation took 23 polymorphisms then associated with endurance and found a 0.0005 percent chance that any one person in the world carried the favourable form of all of them^[11]. An international consortium ran genome-wide scans on two cohorts, then tested the 45 most promising markers in seven more, 1,520 endurance athletes and 2,760 controls in all. It found one significant marker, in GALNTL6, and no panel of variants common to the athlete groups^[12].

ACE: an insertion that chips do not read

ACE codes for angiotensin-converting enzyme, which is tied to the regulation of blood pressure^[7]. Its best-known variant is not a single letter. In 1990 it was described as the presence or absence of a DNA fragment of about 250 base pairs, a difference that accounted for 47 percent of the variance in the enzyme's level in serum^[13]. The fragment was later sized at 287 base pairs; it sits in intron 16, and the allele that has it is called I, for insertion, and the one without it D^[7].

A genotyping chip reads one letter at each of a fixed list of positions, as our piece on raw data false positives describes, so a 287-base insertion is not something it reports directly. What can be read is a nearby single-letter variant inherited along with it. For ACE that is rs4343 in exon 17: in 64 Europeans its A allele marked the insertion and its G allele the deletion, with a correlation (r²) of 0.88^[14]. That is good enough for research and still not a reading of the insertion. In gnomAD, the A allele is carried on 43.5 percent of Finnish chromosomes^[15].

The ACE performance story is weaker than ACTN3's to begin with. The 1.35 odds ratio comes from pooled case-control studies^[7], and in the training study below that typed both genes, muscle size and strength did not differ by ACE genotype before or after training^[16].

Training studies that sorted people by genotype

The boldest claim in this market is that a genotype shows which kind of training a person will respond to. Several studies trained people and then looked.

The two independent ACTN3 studies cut against the sales logic. If XX meant a poor response to strength training, XX women in the larger study should have gained least; they gained most, and about 2 percent of the gain was attributable to genotype^[9]. In the smaller one, the gain from nine weeks of training was the same whatever the genotype^[16].

The study in favour of matched training came from inside the industry. The corresponding author's address is a company called DNA Sports Performance, one of the listed affiliations is a DNAFit research centre, DNAFit provided all the genetic testing, and the paper declares no conflict of interest^[17]. Only 67 of the 123 athletes who started finished. With the seller supplying the testing and 56 of the 123 starters lost, the trial does not settle the question.

HERITAGE shows why the question keeps coming back: people differ in how much they gain, and the differences run in families^[10]. A 2011 follow-up found 21 variants that together accounted for 49 percent of the variance in that gain, but it found and measured them in the same 473 people, and its authors wrote that large-scale replication was warranted^[18].

PPARGC1A, the third gene in our Frontier cards, has the same shape. A 2005 study found the Ser482 allele of rs8192678 on 29.1 percent of chromosomes in 104 world-class Spanish endurance athletes and 40.0 percent in 100 unfit British men, whose mean age was 49 against the athletes' 27^[19]. In gnomAD the same allele is on 29.9 percent of Finnish chromosomes^[20], about the athletes' rate (our comparison of the two figures).

The 2015 consensus statement

The statement came out of a symposium of the International Federation of Sports Medicine's scientific commission in Santorini, and its 24 authors include the researchers behind several of the studies above, among them the senior author of the ACTN3 papers^[1]. They granted that ACTN3 and ACE have a little replicated evidence behind them. On prediction they were blunt.

On children they went further: in the current state of knowledge, no child or young athlete should be exposed to consumer genetic testing to define or alter training or to pick out gifted children^[1]. Their concern was that parents and coaches are led to believe a test lets them plan and invest in a child's sporting future, on results that cannot carry that weight.

What Aimosti's report shows, file by file

We read these variants because people ask about them, and label them by the evidence. ACTN3 appears as a muscle fibre trait card and as a card in Frontier, the section behind the report's credibility firewall, which opens by saying that nothing in it is medical, diagnostic or a basis for any decision. ACE and PPARGC1A have Frontier cards too.

A plain VCF lists only where a genome differs from the reference, and the reference letter at rs1815739 is C, the R allele, so a genome with no record there is shown as R/R with a note that this is an inference from silence. A missing VCF record is not proof of the reference explains why. The ACE insertion is not typed from any file, and every ACE outcome on the card carries the word inferred.

The ACTN3 card calls the effect a tiny influence on any one person next to training and says the result does not pick a sport; the ACE card calls itself a curiosity, never a talent verdict. The couple outlook, which works out the odds of a few simple traits in a child of two uploaded genomes, leaves ACTN3 out, because an athletic-child outlook is the kind of prediction that guidance on testing children warns against. Deep Read, the add-on for BAM and CRAM files, is a pharmacogenomic panel and does not run the trait or Frontier sections.

Single-gene claims for other traits fare much the same: see what the DNA diet trials found, the COMT warrior gene and the MTHFR myth. For why many small effects add up to a ranking and not a prediction, see polygenic scores in plain words.