Study finds no gay gene. Was there one to find?

Cecile Janssens
Sep 1 · 8 min read

When you take a guided tour in a museum, you walk from highlight to highlight listening to a few fascinating details and anecdotes about each. The critical questions often come later, after the tour.

The new article on the genetics of same-sex sexual behavior reads like a guided tour. The authors take us along the highlights of their diverse body of work that shines a light on the genetics of same-sex sexual behavior from various perspectives. They didn’t find a gay gene but do find evidence that genetics do play a role in sexual behavior.

Let me guide your tour.

It’s about sex.

The article opens with an introduction about same-sex sexual behavior, sexual preference, and sexual orientation, but we need to get straight from the beginning that the main analyses were based on the single question ‘Did you ever have sex with someone of the same sex? Yes/No”.

Most people who ever had sex with someone from the same sex mostly engage in opposite-sex behavior (Figure). If you had a same-sex experience only once, you are considered “nonheterosexual” in this study. The “nonheterosexuals” seem to be two distinct groups: women who (now?) mostly have sex with men and men who mostly have sex with men only.

Adapted from Ganna et al. Science 2019.

Can these two groups be combined as one? Do the authors expect that their sexual behavior is explained by the same genes? Wouldn’t it be more obvious to consider only those who mostly or always have sex with someone of the same sex?

(In one of their final analyses, they show that this group may not be one, as the genetic correlation between those with less than a third of their sexual partners being of the same sex and those who had same sex partners only was 0.13. Why didn’t they start with this analysis and reconsider the phenotype that the entire manuscript is based on?)

Five variants

The authors first perform genome-wide association analyses to find DNA variants (single nucleotide polymorphisms, SNPs) that are associated with same-sex sexual behavior. Out of millions tested, they identify two SNPs in men, one in women, and two in men and women combined.

Source: Ganna et al. Science 2019.

Remarkably, the observed associations were 10 times higher in 23andMe. If the estimates in the UK Biobank had been reported with only 2 decimal points, they would all be zero (0.00). If these observed betas are regular betas, then the corresponding odds ratios would all be 1.00, indicating that the SNPs were not associated with same-sex sexual behavior in the UK Biobank. Were they?

Three variants. Or one?

As is common practice, they then tested whether the SNPs were also associated with same-sex sexual behavior in other populations. They found that “three of the SNPs replicated at a nominal P value in the meta-analyzed replication datasets.”

Meta-analyzed replication datasets means that the authors had pooled the three independent populations into one. This is common, but deceiving when one dataset is larger than the two others combined. Two of the three SNPs only replicated in the largest population, not in the smaller ones. Only one SNP replicated in two out of three populations.

(The SNP found in women did not replicate, which should be no surprise if you remember that most of them were mostly having sex with men.)

Genetic variants accounted for 8 to 25% of variation in same-sex sexual behavior

The ability to identify genetic variants depends on the sample size of the study and even very large studies may not be able to find variants that have very small effects. Using a frequently used analytic technique, the authors estimated that the combined contribution of these undiscovered variants explained 8 to 25% of the variation in same-sex behavior. That is: 8% in the UK Biobank and 25% in the 23andMe dataset.

Of interest, but complex, is that the 8 to 25% range was when they analyzed same-sex sexual behavior on the so-called liability scale. The authors explain: “In the liability threshold model it is assumed that an observed dichotomous phenotype is the product of a latent liability, and that this latent liability has a standard normal distribution.” In other words, same-sex sexual behavior is assumed to have an underlying continuum. People vary in how likely they will engage in same-sex sexual behavior.

I know that the likelihood of engaging in same-sex sexual behavior is not the same as the proportion of same-sex partners, but the figure below illustrates what assuming a normal distribution may imply, namely that most people have no clear sexual preference. Does that sound likely?

Adapted from Ganna et al. Science 2019.

The authors don’t justify whether the normal distribution of the liability is an appropriate assumption here. If it is not, should we then conclude that the SNP heritability cannot be calculated on the liability scale and not consider this analysis?

Underlying biological mechanisms

The authors then moved on to seeking biological support for their identified SNPs. They first performed cell- and tissue-type enrichment analyses. The idea behind the analysis is that enriched gene expression in certain cell or tissue types provides hints for certain biological processes. For example in type 2 diabetes, you may expect enrichment in adipose tissue and in Alzheimer’s disease enrichment in the brain cells. But what cells do you expect to be enriched for having EVER had intercourse with someone of the same sex?

The authors found no evidence for enrichment.

Next, they investigated whether the five SNPs are associated with other quantitative traits. The article reports that one SNP was in a DNA region that contains a number of olfactory receptor genes and that another was associated with a higher rate of male balding pattern. For the latter, they did not report that the same SNP was associated with 22 other traits, half(!) of which were related to BMI and fat mass, and others to lower cheese intake, higher frequency of solarium/sunlamp use, and higher use of sun protection.

These were exploratory analyses, seeking hints for pathways, but it would have been appropriate to acknowledge that the reported traits were observed among many other (random?) signals.

Not a continuum

Towards the end of their article, the authors admit that their “primary phenotype as ever or never having had a same-sex partner … fails to capture the multifaceted richness and complexity of human sexual orientation. To explore the consequences of this simplification, we pursued genetic analyses across different aspects of sexual orientation and behavior.”

They then “performed a GWAS on the proportion of same-sex partners to total partners, with a higher value indicating a higher proportion of same-sex partners”. Did they identify the same SNPs? The article did not report.

“These findings suggest that the same-sex sexual behavior variable and the proportion of same-sex partners among nonheterosexuals capture aspects of sexuality that are distinct on the genetic level, which in turn suggests that there is no single continuum from opposite-sex to same-sex sexual behavior. Interpretations of any one set of results in our study must consider this complexity.” “Overall, our findings suggest that the most popular measures are based on a misconception of the underlying structure of sexual orientation and may need to be rethought.”

This reasoning doesn’t make sense. The study suffered from poor measurement of same-sex sexual behavior (that mostly are people who mostly have sex with opposite-sex partners) and poor measurement for the continuum of sexual preference (that largely depends on how many sexual partners someone had in their life). They found that the two measurements had no genetic overlap and had different patterns of correlations with other traits. It would be appropriate to doubt the measurements and leave inferences to further studies.

Genetic correlations

Lastly, they “showed strong genetic correlations (all rg ≥ 0.83) of same-sex sexual behavior with items assessing same-sex attraction, identity, and fantasies, suggesting that these different aspects of sexual orientation are influenced by largely the same genetic variants.”

While these high correlations should not be too surprising, the inference doesn’t seem correct. The authors do not mention that the phenotypic correlations were equally high. In other words, people tend to give the same answers irrespective of whether you ask them about sex, attraction, identity, or fantasies. The genetic correlation may be so high because the different questions are different ways of asking the same, not because they are distinct and largely independent characteristics of sexuality with the same genetic basis.

The high genetic correlation would imply that the identified SNPs are also associated with these other aspects as well. The authors don’t report about this.

Unusually high

The authors report that the 23andMe data had an “unusually high proportion of individuals who had had same-sex sexual partners in this sample (18.9%).” The authors didn’t provide data from other studies to show that this proportion was indeed much higher than commonly observed. Yet, the study is not about being “non-heterosexual”, but about having had sex with a same-sex partner, even if only once.

But was it so high? When researchers study diseases, they usually find that the disease predisposing allele is the least common allele of a SNP. Here they found the opposite. The effect alleles of the three SNPs that replicated in external populations were extremely common. So common that 90% of the people has at least one effect allele on each of the three SNPs and 31% has two on all of them.

Here begins my curiosity: Are these effect alleles common by chance, or is there an evolutionary explanation why these are common?

A great study?

The authors claim that they discovered 5 variants (which isn’t true), that genetic variants explained 8 to 25% (which may not be true), and that the continuum of sexual behavior needs to be rethought (which was concluded from the analyses of two poorly measured variables).

The praise for this study on Twitter from the GWAS community leaves me confused. Is this really an outstanding paper in GWAS research? Am I raising irrelevant methodological concerns? Or not enough? Or is it that the story-telling style of writing distracts from critical reflection and from checking what was not told?

I hope the latter. Although that too is worrisome.

A final word

I am surprised about the public response about this study. Surprised how easily and eager the media reports that there is no gay gene but that there is evidence that genetics do play a role.

I don’t think there is a gay gene. There never was. We don’t have major genes that guide our behavior. For no behavior. And neither for sex.

But why should same-sex sexual behavior be justified? Or sexual orientation or identity? It is nobody’s business with whom you have sex and why. Have sex with whomever you want if they want it too. It does not need to be in your genes. Your genes don’t have to justify anything.

Cecile Janssens

Written by

Professor of epidemiology | Emory University, Atlanta USA | Writes about (genetic) prediction, critical thinking, evidence, and lack thereof.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade