Letter to the Editor: “Genetic Link Between Gender Dysphoria and Sex Hormone Signaling”

Foreman et al. (1) tested the interesting hypothesis that transgender women (formerly called male-to-female transsexuals) have frequencies of polymorphisms in genes for several steroid metabolizing enzymes and hormone receptors that differ from those of male controls. They concluded that gender dysphoria, at least in natal males, may have a genetic basis.

The evidence does not justify the conclusions drawn because, in a surprising omission for a genetics article (2), no effort was made to control the type 1 error rate (α) for multiple hypothesis testing. Two ways of doing this are the familywise error rate (FWER; the likelihood of incorrectly rejecting a single true null hypothesis among m null hypotheses, i.e., false positives), most often controlled with a Bonferroni correction (α/m). In Tables 1 and 2, polymorphisms in 12 different genes were examined. For α = 0.05, the FWER was controlled at 0.05/12 = 0.004. Thus even the smallest P value, 0.015 for the STS gene, is >0.004 and thus not significant. Similarly, in Table 3, which tests 10 hypotheses, the smallest P value, 0.009 for SULT2A1, was >0.005 (0.05/10) and thus not significant. The P values in Table 4 cannot be interpreted without knowing how many interactions were examined; there are 66 possible pairwise interactions between 12 genes.

It is often argued that a Bonferroni correction is too stringent and thus is prone to type 2 errors (incorrectly accepting a false null hypothesis, i.e., false negatives). Therefore, many studies use the false discovery rate (FDR; the proportion of incorrectly rejected null hypotheses among all rejected null hypotheses, i.e., an FDR of 0.05 means that 5% of all positives at that statistical threshold are false positives), most often calculated via the Benjamini-Hochberg procedure (3). For a given α, that procedure ranks all P values from smallest to largest, finds the largest k such that P(k) ≤ (k/m)*α, and rejects all null hypotheses up to and including that one. But this is not much less stringent than the FWER when there are few tests with uncorrected P < α. For the data in Tables 1 and 2 combined, there are 3/12 such tests, the highest having P = 0.03. This is higher than the threshold of 0.05*3/12 = 0.0125. Accordingly, no association passes the Benjamini-Hochberg procedure either; the same is true for the data in Table 3. Moreover, of the three associations with the lowest P values in Tables 1 and 2, only one, Erα, shows an uncorrected P < 0.05 for genotype association in Table 3. Similarly, only one of the total of four genes with associations having uncorrected P < 0.05 in Tables 1–3, SRD5A2, is one of the six genes involved in putative two-locus gene interactions in Table 4. Thus, three different ways of analyzing the same data yield inconsistent results, supporting the idea that all the identified associations are due to chance.

Therefore, the data in this article do not demonstrate that any polymorphism in the examined genes, or any combination of polymorphisms, influences risk of gender dysphoria in transgender women.

Acknowledgments

Disclosure Summary: The author has nothing to disclose.

References