The alcohol flush and ALDH2: what one variant does to acetaldehyde

Two enzymes, two steps

Ethanol does not leave the body as it arrives. Most of it is oxidised in the liver, first by the alcohol dehydrogenases, of which ADH1B does much of the work, into acetaldehyde, and then by the aldehyde dehydrogenases into acetate^[1]. Of those, the mitochondrial ALDH2 does most of the clearing at the concentrations a drink produces^[1].

Acetaldehyde is the problem molecule. It is reactive, it binds to DNA and proteins, and when it builds up after a drink it causes the flushing, nausea and racing heart that give the reaction its name^[3]. In 2009 the International Agency for Research on Cancer (IARC) classified it on its own: acetaldehyde associated with the consumption of alcoholic beverages is carcinogenic to humans, Group 1^[4]. The working group leaned on genetics to reach that conclusion, and the genetics is the variant this article is about^[5].

ALDH2*2: one letter, four subunits

The variant changes one base on chromosome 12, G to A, and with it amino acid 504 of the protein, from glutamate to lysine^[6]. Older papers call the same change Glu487Lys, counting from the start of the mature protein instead of the first amino acid the gene encodes^[8]. The finding that made it famous came in 1989, when Crabb and colleagues genotyped Japanese liver samples and showed that people with a single copy had lost most of their ALDH2 activity along with people who had two. The inactive allele was dominant^[2].

Why one bad copy is enough comes down to how the enzyme is built. Working ALDH2 is a tetramer, four protein subunits assembled into one unit, and a variant subunit spoils the unit it joins: this is what geneticists call a dominant-negative effect^[8]. If a cell makes the two kinds of subunit in equal amounts and they assemble at random, only one enzyme in sixteen is made entirely of normal subunits.

Blood alcohol itself does not differ. In the study that compared all three genotypes after a measured dose, ethanol levels were the same across them; acetaldehyde was what separated them^[9]. The variant does not change how drunk a person gets. It changes how much of the toxic intermediate is left over.

ADH1B: a faster first step

The first step has its own common variant. rs1229984 changes amino acid 48 of ADH1B from arginine to histidine (written Arg47His in older papers), and the histidine form, ADH1B*2, turns ethanol into acetaldehyde much faster than the usual European form^[1]. Faster production can push acetaldehyde up even when ALDH2 works normally, and the allele is consistently associated with drinking less and with lower rates of alcohol dependence^[1].

The letters are a trap. ADH1B is read off the minus strand of chromosome 4, so the gene's own sequence and a DNA file disagree on how to write the change. Papers describe the fast allele as A; a chip export or a VCF, which use the forward strand, write it as T, and write the usual European allele as C^[7]. The reference genome GRCh38 happens to carry the fast T, so a typical European genome shows up in a VCF as a homozygous "variant", 1/1, at a position where nothing unusual is going on.

Common in East Asia, almost absent in Finland

ALDH2*2 is a recent East Asian allele. The China Kadoorie Biobank genotyped 150,722 adults from ten areas of China and found rs671 A at a frequency of 0.21; the fast ADH1B allele was at 0.69^[11]. The East Asian genomes in gnomAD put rs671 A at 0.2249^[12]. In Europe the allele is a rarity. The 1000 Genomes Project found no copy in 1,006 European chromosomes, a sample that includes Finns, and gnomAD's much larger European exome set puts it at about 4 in 100,000^[6].

The flush follows the allele. About 36 percent of East Asians report it^[3]. That is close to the share of people who carry at least one copy when the allele frequency is 0.21: about 38 percent, by our arithmetic of one minus 0.79 squared, if partners pair up regardless of genotype. For a Finn, the expected genotype at rs671 is G/G, and for a Finnish reader who flushes, ALDH2*2 is an unlikely explanation.

Acetaldehyde and oesophageal cancer

The cancer link was first seen in Japanese case-control studies of drinkers and became the textbook case of what epidemiologists call Mendelian randomisation. Genotypes are handed out at conception and do not depend on lifestyle, so a variant that changes exposure to a substance can test whether the substance itself does harm. Lewis and Davey Smith pooled the published studies of ALDH2 and oesophageal cancer in 2005 and found two effects pulling in opposite directions^[10].

Compared with people carrying two working copies, people with one copy of ALDH2*2 had an odds ratio of 3.19 for oesophageal cancer. People with two copies had an odds ratio of 0.36^[10]. The second number is not a protective gene at work. People with two copies drink very little because drinking makes them ill, so they receive little acetaldehyde from alcohol at all. People with one copy can drink, and when they do, more of the acetaldehyde stays in their tissues.

The China Kadoorie Biobank tested the same idea prospectively, following about half a million adults for around ten years. Among people who drank less than weekly, carrying ALDH2*2 was not associated with more oesophageal cancer. Among men who drank weekly, carriers had about three times the rate of non-carriers, and the excess was clear in men drinking 30 grams of alcohol a day or more^[13]. A later analysis of the 150,722 genotyped participants found that, among male drinkers, the cancer risk that rose with intake rose more steeply in carriers of one copy than in non-carriers, for head and neck, oesophageal and lung cancers^[11].

These numbers describe groups. A hazard ratio of 3.31 says that, across thousands of Chinese men who drank weekly, carriers were diagnosed about three times as often as non-carriers over the follow-up; it does not say what will happen to any one of them, and it says nothing about a European population where the variant barely exists. We do not turn it into a personal percentage, and the report does not either.

Why a European chip may not carry rs671

A genotyping chip reads a fixed list of positions chosen by its designer, between 563,320 and 955,958 rows in the ten consumer exports we measured^[14]. A marker that almost nobody in the target market carries is the kind a design can leave out. Whether a given export has a row for rs671 depends on the array version, and we have not audited every vendor's current list, so the only reliable answer is the file itself.

That distinction matters for an allele this rare. If the row is missing, the honest result is unknown, not G/G. Our report used to blur the two on chips, printing the common genotype for markers an array never carried, and five exports of one customer's DNA disagreed with each other as a result. It now reads the export's own list of rows and, when the marker is not there, says that it was never read.

A plain VCF has the opposite problem. It lists only positions where a genome differs from the reference, and for nearly every Finn rs671 is not one of them, so the file says nothing there. The report reads that silence as the reference genotype, G/G, and labels it an inference. A gVCF, which records where the reads matched, and a BAM or CRAM, which holds the reads themselves, can show the position was actually read. We explain the difference between those files in our guide to genome files.

What our report shows, file by file

The cards state the biology and the population findings. They give no drinking guidance, and they do not translate a genotype into a cancer probability. The ALDH2 card for one or two copies says that acetaldehyde is a recognised carcinogen and that, with this genotype, drinking raises the risk of oesophageal and head-and-neck cancer above average, which is what the studies above measured in populations. You can see the card as it appears on a chip export in the sample report.