The cilantro gene is eight genes and half a per cent

In 2012 the genetics company 23andMe asked its customers a question with a very short answer: does fresh cilantro taste like soap to you? Yes, no, or not sure. 18,495 people answered it. ¹ The paper that came out of that is the source of everything you have ever read about the cilantro gene, and it is a good paper. It also says, in its own words, several things that the story built on top of it does not.

What the study actually found

The study everyone is thinking of came out in 2012 in a journal called Flavour. A genome-wide scan across 14,604 people who had answered the soap question turned up one variant, rs72921001, on chromosome 11, at a p value of 6.4 x 10^-9. It held up in a second group of 11,851 asked a different question, whether they liked cilantro at all, at p = 0.0057. ¹ That is a real finding, cleanly done, and it has been repeated in every article about cilantro since.

Then come the sentences that do not get repeated. Each copy of the associated allele multiplies the odds of saying soap by 0.81. Despite the strength of the association, the authors write, it explains only about 0.5 per cent of the variance in perceiving that cilantro tastes soapy. ¹ When they estimated how much of the trait all their genotyped markers together could account for, they got about 0.087, and noted that the calculation does not exclude a heritability of zero. ¹ A famous genetic finding, and the study reporting it says its own genome-wide heritability estimate cannot rule out none.

The gene name is the next thing to go. Nothing in that work identified OR6A2 as the cause. The variant sits near a cluster of eight olfactory receptor genes, OR2AG2, OR2AG1, OR6A2, OR10A5, OR10A2, OR10A4, OR2D2 and OR2D3, and OR6A2 was singled out because it was known to respond to aldehydes of the kind cilantro is full of. ¹ It was the most plausible candidate in a neighbourhood, which is not the same as the answer.

Fourteen years, no retest

A finding this famous ought to have been retested since. Search the life-science literature for rs72921001 and seven things come back, every one of them a review or a commentary, and not one a study that went and looked again. The original paper is candid about how little else held: two variants reached genome-wide significance and only one replicated, the other sitting in a gene involved in membrane trafficking with no known function in taste or smell, which the authors allow may be a false positive. Three variants an earlier group had reported for cilantro dislike came back at p values of 0.53, 0.41 and 0.53. ¹ The largest genome-wide study of food liking published since cites the cilantro paper in its introduction and does not include coriander among its own traits. ¹²

The twin evidence is thinner than its abstract suggests too. A 2012 study of 286 twin pairs put the additive genetic component of cilantro odour pleasantness at 0.38 and of cilantro flavour pleasantness at 0.52, which sounds substantial until you reach the footnote attached to both. It reads: significant for familiality, heritability significant if shared environmental effects are assumed to be zero. ² The authors say it plainly in the text as well, that for most of the traits they modelled they could not unequivocally separate the genetic contribution from the shared-environment one. Only two traits in the study earned heritability outright, the intensity of PTC at 0.71 and the burn of quinine at 0.41. Basil odour, rated by the same people, scored 0.39, which is cilantro's 0.38 with the decimal moved. ² Families resemble each other about herbs. Separating why is the hard part, and only 35 of those 286 pairs were fraternal, which is the comparison the whole method rests on.

The sentence that settles it

The strongest correction to the popular story is a sentence in the paper itself. Eriksson and colleagues broke their respondents down by ancestry and found large differences: African-American, Latino, East Asian and South Asian participants were all significantly less likely than Europeans to detect a soapy taste, with odds ratios of 0.676, 0.637, 0.615 and 0.270, at p below 0.003. ¹ Then they checked whether the variant accounted for that spread, and reported that the differences in allele frequency across populations do not explain the differences in soapy-taste detection. ¹

South Asians, whose cooking uses cilantro by the bunch, were the least likely group of all to call it soapy, and the allele they carry does not account for that. The paper's own conclusion is hedged twice over: one of a cluster of olfactory receptor genes, perhaps OR6A2, may be the receptor that contributes to the detection of a soapy smell from cilantro in European populations. ¹ Perhaps, may, and in European populations. A separate survey, asking about disliking rather than soapiness, found 21 per cent of East Asian and 3 per cent of Middle Eastern respondents disliked it, against 0 to 6 per cent for a lettuce control. ³

Now the soap

The chemistry is not in dispute, and has not been since Potter and Fagerson worked out the leaf volatiles in 1990. ⁶ Coriander leaf is overwhelmingly aldehydes. Across four cultivars, (E)-2-decenal ran from 15.7 to 25.1 per cent of the essential oil, decanal from 14.4 to 18.8 per cent and (E)-2-dodecenal from 12.6 to 13.7 per cent, with six aldehydes among the thirteen compounds detectable over freshly chopped leaf. ⁵ The plant is not making these to be pleasant. Alkenals of this class are defensive compounds, and they activate TRPA1, the receptor that deals with irritants. ¹⁰ The insect association is literal too: (E)-2-decenal is the alarm pheromone of the brown marmorated stink bug. ⁹

The soap is the part that will not stand up. Trace the claim backwards and it does not end at a measurement. The peer-reviewed prevalence paper mentions soap twice, and both times the reference is a newspaper: Harold McGee in The New York Times in 2010, and a Wall Street Journal piece from 2009. ³ Go to McGee's column and he is careful in a way his readers were not. What he wrote was that the same or similar aldehydes are also found in soaps and lotions and the bug family of insects. ¹¹ The hedge is his. The version that spread deleted it.

He also gave a mechanism, and that part almost never travels. Soap is made by breaking fat molecules apart with lye, and aldehydes are a byproduct of that process, as they are when air attacks the fats and oils in cosmetics. ¹¹ The published chemistry agrees on the general point: 2-alkenals of this class are products of the oxidation of fats. ⁸ So the resemblance is a genuine family resemblance between two things made out of broken-down fat, arrived at from opposite directions. Nobody appears to have gone and measured a bar of soap for these compounds, which is the step everyone assumes happened.

What the receptor turned out to be doing

In 2026 a group at ShanghaiTech got the first structures of OR6A2, having had to engineer a more tractable version of the receptor to do it. It does selectively sense medium-chain aldehydes, so the fourteen-year-old guess about why it was the interesting gene in that cluster was a good guess. The binding turned out to be stranger than expected: the aldehyde forms a reversible covalent bond, a Schiff base, with a lysine sitting in the pocket, confirmed by mass spectrometry. ⁴ Smelling a coriander leaf involves briefly assembling a new molecule inside a protein and then taking it apart again.

And in a paper devoted entirely to this receptor, in Cell, the authors still describe OR6A2 as belonging to a cluster of olfactory receptors genetically associated with the soapy perception of cilantro. ⁴ Fourteen years and a solved structure later, the field's own language has not moved past association to cause. The internet made that jump in about a fortnight.

Can you get over it

The advice that circulates is to crush the leaves, or chop them hard, or make a pesto. There is a real paper behind it, and the paper did something more specific than that. In the January 2010 study McGee cited, coriander leaves were cut and then held in boiling water for two minutes, after which the C10 to C14 alkanals and 2-alkenals had fallen and the corresponding alcohols had risen. The authors went looking for the reason and found strong aliphatic aldehyde reductase activity, reported in this herb for the first time. ⁷ The enzyme is real. What was tested was cut leaf in boiling water, which is a different instruction from chop it finely. And McGee described the products as substances with no aroma, where the paper describes them as alcohols, which have plenty.

As for training yourself out of it, there appears to be no study at all. Repeated-exposure research is a large and healthy field, full of trials on vegetables and cheese and low-salt soup, and cilantro is simply absent from it. The claim traces to a single anecdote from one neuroscientist, quoted in that same 2010 column. ¹¹ It may well be true. It has never been tested.

So the fair account is smaller and more interesting than the meme. There is a variant, it is real, it nudges the odds a little, and it lives beside a receptor that grabs aldehydes with a lysine. Everything else, the certainty about the gene, the soap that nobody assayed, the promise that a food processor will fix it, was built on top of one honest newspaper column by people who kept quoting each other. If cilantro tastes of soap to you, that is genuinely how it tastes, and no amount of allele frequency will talk you out of it. Somebody at the table is going to want it in the salad anyway.