Facial attractiveness plays a crucial role in a variety of human interactions, including human mate choice. People prefer to date and marry facially attractive individuals. The preference for more attractive partners is warranted from an evolutionary perspective, given that facially attractive individuals have higher reproductive success than their less attractive counterparts and facial attractiveness is generally thought to indicate genetic quality in terms of disease resistance. People are also more likely to ascribe positive personality attributes to form same-sex appliances, employ and even vote for facially attractive individuals. Despite historical beliefs, facial attractiveness is not merely an arbitrary cultural convention. People from different cultures show strong agreement in what is considered facially attractive. Even young infants who have not been exposed to cultural norms, prefer to look at faces that adults describe as facially attractive. Previous studies have identified several facial cues (eg sexual dimorphism and symmetry) and hormones (eg testosterone and cortisol) that play a role in facial attractiveness. Yet despite the high estimated heritability of facial attractiveness, very few studeis have accessed the genetic variation underlying facial attractiveness. To our knowledge, only the human leukocyte antigen (HLA) genes have been investigated as candidate genes for facial attractiveness. Roberts et al found that HLA heterozygous men (ie men that have differenc copies of the HLA genes) were considered more attractive than HLA homozygous men (ie men that have similar copies of the HLA genes). Follow up studies have replicated this association in male, but not female subjects. Recent studies have shown that genome wide association (GWA) studies can successfully identify common genetic variants and genes which regulate quantitative heritable traits such as height and facial morphology. GWA studies therefore provide a more robust approach to identifying common genetic variants that are associated with facial attractiveness compared to candidate gene approaches that have been used for HLA).

The primary aimof this study is to identify common genetic variants, and ultimately putative genes, that are associated with facial attractiveness using GWA methodologies. We specifically chose the ALSPAC dataset because it is, to our knowledge, the largest dataset with both facial images and GWA data. To accomplish this aim, 3D facial images obtained from the ALSPAC image set will be standardised for size and orientation. 30 (15 male) caucasian students from the United Kingdom (UK) will rate all the images for attractiveness on a seven point Likert scale over eight one-hour sessions (rate calculated from previous work). 30 raters are sufficient to produce an accurate measure of facial attractiveness. We request permission to have the images rated at the Perception Lab, University of St Andrew's (UK), because of the well-established image rating facilities, large participation pool and streamlined workflow; images are rated in the UK to reduce cross-cultural variation in attractiveness judgements. Attractiveness ratings will be averaged for each image. The ALSPAC team have cleaned and imputed a GWA study dataset consisting of 8365 individuals with genotype calls for ~2.5 million common variants spread across the genome. We will use this resource to conduct a 2 stage (discovery and replication) genome-wide association study. Initially the discovery phase will include analysing ~5000 individuals that have both genotypic data and facial images. Power analysis (PowerGwas/QT version 1.0) indicate a sample size of 5000 is adequate to provide 80% statistical power to detect single nucleotide polymorphisms (SNPs) that explain as little as 0.8% of the variance in facial attractiveness. The association between each of the ~2.5 million SNPs (exposure variables) and facial attractiveness (outcome variable) will be independently tested in the ALSPAC cohort using linear additive regression, while controlling for pubertal development. SNP associations that exceed the standard significance threshold for genome-wide significance (pless than 5 x 10-8), will be identified and replicated independently in additional cohorts, making up the second replication phase for the GWA study. To determine which genes (and pathways) are most likely associated with facial attractiveness we will conduct a range of post-hoc analyses including (a) assigning SNPs to genes, (b) epistasis modelling and (c) pathway analyses. Briefly, multiple genes are ascribed to each SNP and these genes are then prioritised using epistasis modelling and pathway analysis, allowing us to further identify which biological processes regulate facial attractiveness. In addition, we will calculate a more accurate heritability estimate of facial attractiveness; do a GCTA analysis to estimate the amount of phenotypic variance in attractiveness common SNPs explains; test the relationship between admixture and attractiveness; test the relationship between genome-wide heterozygosity and attractiveness using the ~2000 ALSPAC individuals who have whole genome sequencing data; and test the association between previously imputed classical HLA alleles and facial attractiveness separately for males and females. Based on previous work we predict that HLA alleles will be associated with male, but not female attractiveness. All GWA analyses will be conducted at the University of Bristol.

The second aimis to determine the association between health measures (exposure variables) and facial attractiveness (outcome variable). The health measures will be divided into prenatal risk factors (eg parental age, presence of gestational diabetes) and childhood health measures (eg body mass index, blood pressure and self-reported health). Facial attractiveness is generally assumed to serve as a 'health certificate', but studies testing this assumption mostly utilise a few self-reported health measures and small sample size (~N=40-200). The size and quality of the ALSPAC dataset, especially the wide range of physiological measurements, provides us with the ideal opportunity to test the association between health indices and facial attractiveness in male and female faces respectively. Based on previous research and work currently under review, we predict that facial attractiveness will be significantly associated with health measures, but more so for male than for female subjects.

The third aimis to determine whether SNPs associated with facial attractiveness are also associated with other traits proposed to indicate overall quality, specifically increased height, body mass index (BMI) within the health BMI range, increased sexual dimorphism (eg facial masculinity/femininity) and facial symmetry. To do so we will calculate morphometric or perceptual measures of sexual dimorphism and facial symmetry before testing the relationship between allelic scores of SNPs for facial attractiveness and these traits.