A Bayesian mixed modeling approach for estimating heritability

Haakon E Nustad,Marissa Leblanc,Christian M Page,Manuela Zucknick,Andrew H Reiner

doi:10.1186/s12919-018-0131-z

Abstract

BackgroundA Bayesian mixed model approach using integrated nested Laplace approximations (INLA) allows us to construct flexible models that can account for pedigree structure. Using these models, we estimate genome-wide patterns of DNA methylation heritability (h2), which are currently not well understood, as well as h2 of blood lipid measurements.MethodsWe included individuals from the Genetics of Lipid Lowering Drugs and Diet Network (GOLDN) study with Infinium 450 K cytosine-phosphate-guanine (CpG) methylation and blood lipid data pre- and posttreatment with fenofibrate in families with up to three-generation pedigrees. For genome-wide patterns, we constructed 1 model per CpG with methylation as the response variable, with a random effect to model kinship, and age and gender as fixed effects.ResultsIn total, 425,791 CpG sites pre-, but only 199,027 CpG sites posttreatment were found to have nonzero heritability. Across these CpG sites, the distributions of h2 estimates are similar in pre- and posttreatment (pre: median = 0.31, interquartile range [IQR] = 0.16; post: median = 0.34, IQR = 0.20). Blood lipid h2 estimates were similar pre- and posttreatment with overlapping 95% credibility intervals. Heritability was nonzero for treatment effect, that is, the difference between pre- and posttreatment blood lipids. Estimates for triglycerides h2 are 0.48 (pre), 0.42 (post), and 0.21 (difference); likewise for high-density lipoprotein cholesterol h2 the estimates are 0.61, 0.68, and 0.10.ConclusionsWe show that with INLA, a fully Bayesian approach to estimate DNA methylation h2 is possible on a genome-wide scale. This provides uncertainty assessment of the estimates, and allows us to perform model selection via deviance information criterion (DIC) to identify CpGs with strong evidence for nonzero heritability.

Highlights

A Bayesian mixed model approach using integrated nested Laplace approximations (INLA) allows us to construct flexible models that can account for pedigree structure
DNA methylation (DNAm) is an epigenetic mark that is implicated in many heritable diseases and traits
We further investigate if it is reasonable to estimate h2 for each CpG site given the data by performing model selection, to assess if the genetic component should be included at all

Summary

Introduction

A Bayesian mixed model approach using integrated nested Laplace approximations (INLA) allows us to construct flexible models that can account for pedigree structure Using these models, we estimate genome-wide patterns of DNA methylation heritability (h2), which are currently not well understood, as well as h2 of blood lipid measurements. Narrow-sense heritability (h2), traditionally estimated using twins or other constrained family relationships, can be estimated in wider pedigrees using a linear mixed-model approach [1, 2]. This approach is well-established for traits having moderate to high h2 [2, 3]. There are multiple reports of DNAm h2 with CpG-specific estimates ranging from 0 to 1, and with genome-wide mean or median estimates ranging from approximately 15 to 30% [5,6,7].

Objectives

Methods

Results

Conclusion