Predicting skeletal stature using ancient DNA

Abstract

Objectives: Ancient DNA provides an opportunity to separate the genetic and environmental bases of complex traits by allowing direct estimation of genetic values in ancient individuals. Here, we test whether genetic scores for height in ancient individuals are predictive of their actual height, as inferred from skeletal remains. We estimate the contributions of genetic and environmental variables to observed phenotypic variation as a first step towards quantifying individual sources of morphological variation. Materials and methods:We collected stature estimates and femur lengths from West Eurasian skeletal remains with published genome-wide ancient DNA data (n = 182, dating from 33,000-850 BP). We also recorded genetic sex, genetic ancestry, date and paleoclimate data for each individual, and δ 13 C and δ 15 N stable isotope values where available (n = 69). We tested different methods of calculating polygenic scores, using summary statistics from four different genome wide association studies (GWAS) for height, and three methods for imputing missing genotypes.Results: A polygenic score for height predicts 6.3% of the variance in femur length in our data (n = 132, SD = 0.0069%, p = 0.001), controlling for sex, ancestry, and date. This is consistent with the predictive power of height PRS in present-day populations and the low coverage of ancient samples. Comparatively, sex explains about 17% of the variance in femur length in our sample. Environmental effects also likely play a role in variation, independent of genetics, though with considerable uncertainty (longitude: R 2 = 0.033, SD = 0.008, p = 0.011). Genotype imputation did not improve polygenic prediction, and results varied based on the GWAS summary statistics we used.Discussion: Polygenic scores explain a small but significant proportion of the variance in height in ancient individuals, though not enough to make useful predictions of individual phenotypes. However, environmental variables also contribute to phenotypic outcomes and understanding their interaction with direct genetic predictions will provide a framework with which to model how plasticity and genetic changes ultimately combine to drive adaptation and evolution.

Population differentiation of polygenic score predictions under stabilizing selection

“…These scores have been used to provide a window into the phenotypic diversity of past populations (Mathieson et al, 2015;Berens et al, 2017;Martiniano et al, 2017) and to disentangle genetic and environmental contributors to temporal phenotypic variation (e.g. at the Neolithic transition; Cox et al, 2019Cox et al, , 2021Marciniak et al, 2021). Such studies are most convincing when there are relevant phenotypic measurements on at least some ancient individuals and polygenic prediction accuracies can be judged.…”

Population differentiation of polygenic score predictions under stabilizing selection

“…R. Soc. B 377: 20200416 transition; [109][110][111]). Such studies are most convincing when there are relevant phenotypic measurements on at least some ancient individuals and polygenic prediction accuracies can be judged.…”

Quantitative Human Paleogenetics: what can ancient DNA tell us about complex trait evolution?

“…They concluded that realized height changes are partially predicted by genetics, with both temporal trends having a period of constancy between the Mesolithic and Neolithic, and an increase between the Neolithic and Bronze Age. A follow-up study by Cox et al (2021) used polygenic scores for height to show that PRS predicts 6.8% of the observed variance in femur length in ancient skeletons, after controlling for other variables. This is approximately one quarter of the predictive accuracy of PRS in present-day populations, which the authors attribute to the low-coverage aDNA data used in their study.…”

“…When used to compute polygenic scores for ancient populations, only a subset of GWAS variants can be used, which substantially reduces predictive accuracy. Cox et al (2021) estimate that the combined effect of low-coverage and pseudo-haploid genotypes reduced their predictive accuracy by approximately 75%, when compared…”

“…Traits that are preserved in the fossil record can provide a degree of partial benchmarking (Cox et al, 2019(Cox et al, , 2021; however, the genetic components of variation are often only partially explained by polygenic scores, and environmental components almost always play large roles in realized trait variation, often dwarfing the contribution of polygenic scores. Furthermore, only a few-largely osteological-traits are well preserved over time, so these comparisons will always be limited in scope.…”