What Insights Can Be Gained Through Comparisons of Broad-Sense Heritability Estimates in Isolated and Outbred Populations?

Inconsistencies have been observed between published estimations of non-additive heritability for many phenotypes between studies of isolated populations and of unrelated individuals. This prompted us to investigate the methods used for estimating broadsense heritability in order to discern whether such inconsistencies could be indicative of particular trait ætiologies, result from specific population characteristics, or stem from non-equivalence between interpretations of heritability in differing study settings. To achieve this, we analyse simulated data mimicking either an isolate or a large sample of unrelated individuals, as well as data from the isolates of Cilento, with a range of cardiovascular related traits. We focus on linear mixed modelling techniques for trait variance decomposition. We discuss the use of identity by descent (IBD) based relatedness matrices or genetic correlation matrices (GRMs) and show the strengths of GRMs against even explicit knowledge of IBD-sharing in an isolate. In the case of trait architectures that involve many low-frequency causal variants, we demonstrate that analysing population isolates avoids the downward bias of heritability estimates that is known to occur in studies of outbred populations. Our analyses of the Cilento isolates produced significant estimates for non-additive components for traits such as body-mass index and low-density lipoprotein level. These analyses, in conjunction with our simulation study, have aided us firstly to clarify the disparities in published heritability results from differing populations and estimation methods and secondly to suggest how such arrays of results should be interpreted. Furthermore, we advocate that for many traits, non-additive genetic models should be revisited.