Global skin colour prediction from DNA

Susan Walsh,Krystal Breslin,Julia Koller,Fan Liu,Andreas Wollstein,Manfred Kayser,Lakshmi Chaitanya,Charanya Muralidharan,Leda Kovatsi,Ewelina Pospiech,Agnieszka Bronikowska,Wojciech Branicki

doi:10.1007/s00439-017-1808-5

Abstract

Human skin colour is highly heritable and externally visible with relevance in medical, forensic, and anthropological genetics. Although eye and hair colour can already be predicted with high accuracies from small sets of carefully selected DNA markers, knowledge about the genetic predictability of skin colour is limited. Here, we investigate the skin colour predictive value of 77 single-nucleotide polymorphisms (SNPs) from 37 genetic loci previously associated with human pigmentation using 2025 individuals from 31 global populations. We identified a minimal set of 36 highly informative skin colour predictive SNPs and developed a statistical prediction model capable of skin colour prediction on a global scale. Average cross-validated prediction accuracies expressed as area under the receiver-operating characteristic curve (AUC) ± standard deviation were 0.97 ± 0.02 for Light, 0.83 ± 0.11 for Dark, and 0.96 ± 0.03 for Dark-Black. When using a 5-category, this resulted in 0.74 ± 0.05 for Very Pale, 0.72 ± 0.03 for Pale, 0.73 ± 0.03 for Intermediate, 0.87±0.1 for Dark, and 0.97 ± 0.03 for Dark-Black. A comparative analysis in 194 independent samples from 17 populations demonstrated that our model outperformed a previously proposed 10-SNP-classifier approach with AUCs rising from 0.79 to 0.82 for White, comparable at the intermediate level of 0.63 and 0.62, respectively, and a large increase from 0.64 to 0.92 for Black. Overall, this study demonstrates that the chosen DNA markers and prediction model, particularly the 5-category level; allow skin colour predictions within and between continental regions for the first time, which will serve as a valuable resource for future applications in forensic and anthropologic genetics.

Highlights

Predicting phenotypes from genotypes is a component of complex genetics that has etched its way into many disciplines including personalized medicine, forensic genetics, anthropological genetics, and consumer genetics, depending on the particular phenotype that is predicted from DNA information
In the case of eye colour, one of the first physical appearance traits to be studied for predictability from DNA, elucidation of its associated DNA markers (Duffy et al 2007; Eiberg et al 2008; Frudakis et al 2003, 2007; Graf et al 2005; Han et al 2008; Kanetsky et al 2002; Kayser et al 2008; Liu et al 2010; Posthuma et al 2006; Rebbeck et al 2002; Sturm et al 2008; Sulem et al 2007, 2008; Zhu et al 2004), and subsequent step-wise ranking on how suitable they were for phenotype prediction (Liu et al 2009) led to the introduction, further development, and forensic validation of the IrisPlex system (Chaitanya et al 2014; Walsh et al 2011a, b, 2012)
It should be noted that considering the highest two categorical probabilities seem to best reflect the colour palette of that particular individual. These preliminary data indicate that the DNA markers and the prediction model we have developed in this study may achieve DNA-based global skin colour prediction regardless of bio-geographic ancestry, which, requires further investigation in additional individuals from around the world

Summary

Introduction

Predicting phenotypes from genotypes is a component of complex genetics that has etched its way into many disciplines including personalized medicine, forensic genetics, anthropological genetics, and consumer genetics, depending on the particular phenotype that is predicted from DNA information. In the case of eye colour, one of the first physical appearance traits to be studied for predictability from DNA, elucidation of its associated DNA markers (Duffy et al 2007; Eiberg et al 2008; Frudakis et al 2003, 2007; Graf et al 2005; Han et al 2008; Kanetsky et al 2002; Kayser et al 2008; Liu et al 2010; Posthuma et al 2006; Rebbeck et al 2002; Sturm et al 2008; Sulem et al 2007, 2008; Zhu et al 2004), and subsequent step-wise ranking on how suitable they were for phenotype prediction (Liu et al 2009) led to the introduction, further development, and forensic validation of the IrisPlex system (Chaitanya et al 2014; Walsh et al 2011a, b, 2012). The HIrisPlex DNA markers and prediction models were used in what has been referred to as the oldest forensic case to date—King Richard III (King et al 2014) as well as in anthropological estimations of ancestral physical appearance (Cassidy et al 2016; Gallego-Llorente et al 2016; Gamba et al 2014; Jones et al 2015; Martiniano et al 2016; Olalde et al 2015)

Methods

Results

Conclusion