Estimation and partition of heritability in human populations using whole-genome analysis methods

Abstract

Understanding genetic variation of complex traits in human populations has moved from the quantification of the resemblance between close relatives to the dissection of genetic variation into the contributions of individual genomic loci. However, major questions remain unanswered: How much phenotypic variation is genetic; how much of the genetic variation is additive and can be explained by fitting all genetic variants simultaneously in one model, and what is the joint distribution of effect size and allele frequency at causal variants? We review and compare three whole-genome analysis methods that use mixed linear models (MLMs) to estimate genetic variation. In all methods, genetic variation is estimated from the relationship between close or distant relatives on the basis of pedigree information and/or single nucleotide polymorphisms (SNPs). We discuss theory, estimation procedures, bias, and precision of each method and review recent advances in the dissection of genetic variation of complex traits in human populations. By using genome-wide data, it is now established that SNPs in total account for far more of the genetic variation than the statistically highly significant SNPs that have been detected in genome-wide association studies. All SNPs together, however, do not account for all of the genetic variance estimated by pedigree-based methods. We explain possible reasons for this remaining "missing heritability."

Figure 1

A. Distributions of the off-diagonal elements for the genetic relationship matrix G in the pedigree design using MZ and DZ twin pairs (green), the within-family design using full sibling pairs (pink), and the population using ‘unrelated’ individuals (blue). B. Examples of the genetic relationship matrix G for 4 individuals in the pedigree design (upper matrix), the within-family design(middle matrix), and the population design (lower matrix). Notes Figure 1: A. The distribution of off-diagonal elements for the genetic relationship matrix G for the pedigree design represents the expected proportion of genome wide IBD for MZ (=1) and DZ (= 0.5) twin pairs. The distribution of actual proportion of the genome that is shared IBD within full sibling pairs in the within the family design varies around 0.5 whereas the distribution of the actual proportion of the genome that is shared IBS for individuals not knowingly related to each other varies around zero. IBD and IBS estimates for the within-family design and the population design, respectively, are derived from data from the Framingham Heart Study (FHS). Mean IBD was estimated at .5028 with a standard deviation of 0.0368 for full siblings. Mean IBS was estimated at −.0002 with a standard deviation of 0.0046 for pairs of unrelated individuals. Only full sibling pairs were selected for the IBD estimation in the within-family design while for the IBS estimation in the population design only one sibling per family was selected and one member of each of the remaining pairs of individuals that had an estimated genetic relationship of more than 0.025 was removed. Estimates of IBS are relative to an arbitrary base population with an average relationship between all pairs of individuals of zero. In this analysis, the sample under study is used as the base population, consequently the average relationship between all pairs of individuals is zero and the average relationship of an individual with him- or herself is one. B. Diagonal elements in matrix G represent an individual's estimated genetic relatedness with him- or herself. Off-diagonal elements in matrix G represent genetic similarity between individuals. Note that matrix G is symmetrical and that for clarity only elements in the lower triangular are provided. In the pedigree design, off-diagonal elements represent expected genome wide IBD sharing for 4 individuals (one MZ pair and one DZ pair) from two independent families in which parents are assumed to be unrelated. In the within-family design, the off-diagonal elements represent actual variation around the expected genome wide IBD sharing [$E\left(\widehat{\pi }\right)=0.5$ = 0.5] for two independent full sibling pairs. In the population design, the off-diagonal elements represent actual genome wide IBS for 6 pairs of not knowingly related individuals. An individual's genetic relatednes with him- or herself (diagonal elements) is an estimate of 1+F, with F being the inbreeding coefficient relative to the base population.