The evolution of sex differences in disease genetics

There are significant differences in the biology of males and females, ranging from biochemical pathways to behavioural responses, which are relevant to modern medicine. Broad-sense heritability estimates differ between the sexes for many common medical disorders, indicating that genetic architecture can be sex-dependent. Recent genome-wide association studies (GWAS) have successfully identified sex-specific and sex-biased effects, where in addition to sex-specific effects on gene expression, twenty-two medical traits have sex-specific or sex-biased loci. Sex-specific genetic architecture of complex traits is also extensively documented in model organisms using genome-wide linkage or association mapping, and in gene disruption studies. The evolutionary origins of sex-specific genetic architecture and sexual dimorphism lie in the fact that males and females share most of their genetic variation yet experience different selection pressures. At the extreme is sexual antagonism, where selection on an allele acts in opposite directions between the sexes. Sexual antagonism has been repeatedly identified via a number of experimental methods in a range of different taxa. Although the molecular basis remains to be identified, mathematical models predict the maintenance of deleterious variants that experience selection in a sex-dependent manner. There are multiple mechanisms by which sexual antagonism and alleles under sex-differential selection could contribute toward the genetics of common, complex disorders. The evidence we review clearly indicates that further research into sex-dependent selection and the sex-specific genetic architecture of diseases would be rewarding. This would be aided by studies of laboratory and wild animal populations, and by modelling sex-specific effects in genome-wide association data with joint, gene-by-sex interaction tests. We predict that even sexually monomorphic diseases may harbour cryptic sex-specific genetic architecture. Furthermore, empirical evidence suggests that investigating sex-dependent epistasis may be especially rewarding. Finally, the prevalent nature of sex-specific genetic architecture in disease offers scope for the development of more effective, sex-specific therapies. Funding This work was supported by the European Research Council (WPG and EHM; Starting Grant #280632), a Royal Society University Research Fellowship (EHM), the Swedish Research Council (JKA), and the Volkswagen Foundation (JKA). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests The authors declare that they have no competing financial interests.


47
Sex, and the existence of two sexes, has revolutionised life on Earth. The success of sexual reproduction is 48 attributed to recombination between parental chromosomes, which accelerates the loss of deleterious 49 alleles and the proliferation of advantageous ones [1,2]. The difference in gamete size between males and 50 females is a fundamental property of almost all sexual species. Yet sexual dimorphism extends far beyond 51 this, from cellular and anatomical specialisation to secondary sexual traits such as ornamentation and 52 behaviour. Furthermore, there are differences in gene co-expression and metabolome networks between 53 the sexes [3][4][5]. It is therefore not surprising that in the field of medicine, males and females frequently 54 differ in core features of disease [6].

55
The genetic basis of disease has been intensely researched, with the aim of providing improved 56 diagnosis and therapy. Heritable diseases can be classified as being rare with monogenic aetiology (caused 57 by a single mutation), or common (prevalence 0.1-1%), caused by multiple genetic variants, each with 58 high population frequency but small individual contribution to disease risk [7,8]. For these genetically 59 complex diseases and traits, genome-wide association studies (GWAS) have been successful at identifying 60 loci, but the heritability accounted for by main effects, and by polygenic risk score, remains conspicuously

82
Broad-sense heritability is the proportion of phenotypic variance in a population sample that can be 83 attributed to genetic variation [17]. With identical genetic architecture, and assuming a common 84 environment, trait heritability should be equal in male and female samples. However, in a study of twenty 85 quantitative traits in humans, eleven showed significant sex-bias in heritability [18]. Following a PubMed 86 literature search, we identified eighteen independent studies in humans (representing thirty-one traits) that 87 provided separate heritability estimates for males and females, and also stated whether the difference was 88 statistically significant. A summary of these data is presented in Figure 1, showing that while fifteen traits 89 did not exhibit significant sex-bias in heritability, thirteen had a higher heritability in females, and three a 90 higher heritability in males. Although there may be some bias in these studies (non-reporting, non-91 independence of traits or prior selection of traits with known sexual dimorphism), they illustrate that the 92 heritability of complex traits is commonly sex-biased across a range of phenotype classes.

93
It is well known that non-genetic factors influence differences in heritability, the most obvious 94 being sex hormones (androgens, oestrogens and progesterones, secreted from the gonads). These can 95 create systemic differences between males and females for trait expression, which in turn affects disease 96 risk and heritability, for example the protective effect of oestrogen on heart disease [19]. However, 97 experiments using hormone treatment and gonadectomy show that some gender differences in phenotypes, 98 such as immune response, behaviour, and toxin resistance, are not determined by sex hormones but by sex 99 chromosome dosage [20][21][22]. This implies that heritability differences are not always caused by sex 100 hormones, and can be caused by sex-specific differences in genetic architecture, whereby a genetic variant 101 has a different phenotypic outcome depending on whether it is expressed in a male or female environment.

102
The molecular genetic evidence for sex-specific genetic architecture is strong. For gene 103 expression in human cell lines, 15% of SNPs that control gene expression (expression quantitative trait 104 loci or eQTL) do so in a sex-specific manner, even in the absence of sex hormones [23]. For complex 105 traits, GWAS have identified many robust sex-specific loci across a range of human phenotypes. These 106 results are summarised Table 1, which shows thirty-two loci with sex-dependent effects in the twenty-two 107 traits studied. The majority of the effects were sex-specific (twenty-eight loci significant in one sex only) 108 although five sex-biased effects were also reported (significant in both sexes but different magnitude of 109 effect). One opposite effect direction locus has also been reported from a GWAS (for recombination rate

240
One implication of sexual antagonism is the maintenance of deleterious genetic variation at higher 241 population frequency than would be expected from mutation-selection balance [43,44]. This leads us to 242 consider its role in susceptibility to common, genetically complex disorders. Consistent with this 243 reasoning, mathematical simulation predicts that alleles that are under sex-differential selection (including

11
The same principle could be applied to resolved antagonism. For example, methylation is both   of pleiotropic genes in disease risk seems likely to be amplified by sex-specific selection.

300
For SNP-based association testing, the basic approach to identifying sex-specific effects is to analyse each 301 sex separately, i.e. sex-stratified. In comparison to joint tests, this approach is limited due to the loss in 302 power caused by partitioning of the sample [78]. A common follow-up to the sex-stratified tests is to 303 determine whether the association statistics for each sex are significantly different from one another.

304
Many main-effect studies incorporate sex as a covariate into the analysis, i.e. they are controlling for the 305 effect. However, whilst this approach acknowledges sex-effects it doesn't allow for their detection. For 306 binary traits with a prevalence of less than 1% inclusion of known covariates actually reduces power [79].

307
A joint analysis that incorporates a genotype-by-sex interaction term tests the difference in 308 allele frequencies between male and female cases, given their allele frequencies in controls. It is thus more 309 suited to identifying genetic differences in trait architecture between males and females rather than for

313
Other covariates, such as those used to correct for population stratification, can also be incorporated into

324
The statistical behaviour of genotype-by-sex tests must be assumed to be similar to genotype-

372
We anticipate that analysis of GWAS data with respect to sex, encouraged by both evolutionary 373 genetics and recent results presented in this review, will generate many more significant findings and 374 highlight the potential role of sex-specific and sexually antagonistic selection as a potent force in human 375 genetic architecture. Finally, we hope that the identification of sex-specific genetic aetiologies in what 376 otherwise appears to be the same disease will result in the development of more effective, sex-specific