Polygenic prediction of human complex traits using ancient DNA

Abstract

Author Manuscript Ancient DNA has revolutionized our understanding of human history and clarified many aspects of human evolution on a molecular level. In this article, I describe recent efforts to translate this into descriptions of phenotypic change over time, and to predict phenotypes of ancient groups and individuals. I do not discuss the more challenging problem of distinguishing between adaptive and neutral evolution and instead focus entirely on whether phenotypes, and their evolution, can be accurately reconstructed. I begin by describing the conceptual and technical limitations of current approaches, and then discuss efforts to reconstruct various phenotypes and the extent to which they are reliable. The fundamental challenges of phenotypic prediction Author Manuscript Several authors have developed theoretical approaches to model genetic prediction accuracy in ancient samples [1,2]. On a technical level, however, this problem is no different than predicting traits in present-day populations, which has been heavily studied [3–8]. The only fundamental difference is that we generally lack phenotypic information for ancient individuals with which to test our predictions. Any attempt to predict phenotypes from ancient DNA therefore faces three immediate challenges, with only the last being unique to ancient samples: Author Manuscript * • Environment: Phenotypes depend on both genetics and environment; any prediction based entirely on genetics makes an implicit assumption about the environment. Put another way, a change in genetic value is neither a necessary nor sufficient condition for phenotypic change [9]. • Transferability: The relationship between genetics and phenotype can be different in different populations [10]. This can be due to genetic or environmental interactions [11,12], the design limitations of association studies [13,14], or differences in allele frequencies and patterns of linkage between populations [5,7]. • Validation: Outside of a small number of traits that can be observed in the skeleton (e.g. stature [15]) or otherwise preserved there is no way of testing phenotypic predictions in ancient individuals. Most predictions cannot, therefore, be tested. Correspondence to mathi@pennmedicine.upenn.edu. Mathieson Page 2 Author Manuscript We generally have very little idea of which aspects of environment are relevant for any particular trait, and what their levels in ancient populations would have been. …