Skip to main content
Dryad logo

Systematic review reveals multiple sexually antagonistic polymorphisms affecting human disease and complex traits

Citation

Harper, Jon Alexander (2021), Systematic review reveals multiple sexually antagonistic polymorphisms affecting human disease and complex traits, Dryad, Dataset, https://doi.org/10.5061/dryad.rv15dv48k

Abstract

An evolutionary model for sex differences in disease risk posits that alleles conferring higher risk in one sex may be protective in the other. These sexually antagonistic (SA) alleles are predicted to be maintained at frequencies higher than expected under purifying selection against unconditionally deleterious alleles, but there are apparently no examples in humans. Discipline-specific terminology, rather than a genuine lack of such alleles, could explain this disparity. We undertook a two-stage review of evidence for SA polymorphisms in humans using search terms from (i) evolutionary biology and (ii) biomedicine. While the first stage returned no eligible studies, the second revealed 51 genes with sex-opposite effects, 22 increased disease risk or severity in one sex but protected the other. Those with net positive effects occurred at higher frequencies. None were referred to as SA. Our review reveals significant communication barriers to fields as a result of discipline-specific terminology.

Methods

For this systematic review we followed PRISMA guidance where possible (Moher et al., 2009). PubMed (https://pubmed.ncbi.nlm.nih.gov/) was searched for articles on 2nd December 2020 with no time limit. The searches were carried out in two Stages, with the organism filter set to human in both stages. In Stage 1, eligible studies were required to report specific genetic variants or haplotypes that were referred to as sexually antagonistic or were an example of intralocus sexual conflict. To achieve this, we conducted a Boolean search for articles that used the terms “sexual antagonism” OR “sexually antagonistic” OR “intralocus sexual conflict” AND “locus” OR “loci”, “gene” OR “snp” OR “polymorphism” OR “variant” OR “allele” in their abstract or title. The Stage 1 search returned 34 articles in total (full search term in the supplementary material; search output is accessible at https://pubmed.ncbi.nlm.nih.gov/collections/60255050/?sort=pubdate).

In Stage 2, studies were required to report specific genetic variants or haplotypes in humans with opposite effects in the two sexes on either complex traits, the outcome of a medical intervention, or disease risk/severity. We define complex traits as likely with a polygenic genetic architecture but are not directly related to a disease phenotype. In this second stage search terms were specifically designed to include papers from the biomedical literature that may have been missed in the first stage because they do not report their findings with terms normally found within the evolutionary biology literature. Again, we conducted a Boolean search for articles that used terms in their title or abstract to describe an opposite or different effect in the two sexes (“sex dependent”, “sex different”, “gender-dependent”, “sex AND opposite”, or “gender AND opposite”), or that capture this concept with alternative words for sex ((“male AND female AND opposite” OR “men AND women AND opposite” OR “boys AND girls AND opposite”) AND (“locus” OR “loci” OR “gene” OR “snp” OR “polymorphism” OR “variant” OR “allele”)). Full details of the search terms used and the numbers of articles returned are provided in the supplementary material. The Stage 2 search returned 881 papers (Figure 1) (full search term in the supplementary material; search output is accessible at https://pubmed.ncbi.nlm.nih.gov/collections/60254985/?sort=pubdate).

The abstracts of the papers from Stage 1 and Stage 2 were then examined, and any papers that had the possibility of reporting an opposite effect of a specific genetic locus on a complex trait, medical intervention or on disease risk/severity were considered for further screening. From Stage 1, no articles passed the screening. From Stage 2, this screening produced a shortlist of 70 candidate papers (https://pubmed.ncbi.nlm.nih.gov/collections/57906298/?sort=pubdate). Full texts of these candidate papers were then reviewed in detail. Papers were included in the final list if they described a sex-opposite or SA effect linked to a specific genetic locus or loci, and reported the effect to be statistically significant (at a P-value cut-off of <0.05, or with 95% confidence intervals not overlapping 1). One additional paper was considered from an outside source. Studies that only reported significant sex-by-variant effects were not automatically included unless they also satisfied the criteria above.

We converted all reported sex-specific effects into a standard effect size (Cohen’s d) quantifying the magnitude of how a given variant affects the studied trait expressed in the given sex. Specifically, Cohen’s d was computed based on the reported descriptive statistics (N, mean, standard error) or by conversion from other effect sizes (Odds ratio) and test statistics (F-values, t-values) using formulas reported elsewhere (Borenstein, 2009; Gurevitch et al., 2013, pp. 195–206; Lajeunesse, 2013). We sought information directly from the authors where these metrics were not possible to extract from the papers themselves (12 authors contacted, 5 responded, 3 responded with data, all later excluded as they did not fulfil the criteria for eligibility). We also recorded the gene name, locus (with accession/rs number where possible), trait affected and the frequency of the focal allele having the effect, hereafter referred to as effect allele frequency. Identification for the variants were taken from the studies where possible, but others necessitated searching the National Centre for Biotechnology Information (NCBI) to find the Reference SNP cluster ID for the variants described. Where effect allele frequencies were not reported, we used  genotype frequencies to calculate effect allele frequency. Not all studies reported allele or genotype frequencies and so we attempted to supplement these data with allele frequencies from the 1000 genomes database (Auton et al., 2015), since these show a strong correlation with effect allele frequencies reported in the studies reviewed (Pearson correlation: N = 25 r = 0.94, P > 0.001; Figure S1). However, using this approach we were only able to supplement our allele frequency data for one additional locus (rs7341475 in the RELN gene). In some cases, loci that affect multiple different traits were found. In such cases we calculated the geometric mean of their effect sizes to account for possible pseudoreplication.