Genetic inheritance of gene expression in human cell lines

Abstract

Combining genetic inheritance information, for both molecular profiles and complex traits, is a promising strategy not only for detecting quantitative trait loci (QTLs) for complex traits but for understanding which genes, pathways, and biological processes are also under the influence of a given QTL. As a primary step in determining the feasibility of such an approach in humans, we present the largest survey to date, to our knowledge, of the heritability of gene-expression traits in segregating human populations. In particular, we measured expression for 23,499 genes in lymphoblastoid cell lines for members of 15 Centre d'Etude du Polymorphisme Humain (CEPH) families. Of the total set of genes, 2,340 were found to be expressed, of which 31% had significant heritability when a false-discovery rate of 0.05 was used. QTLs were detected for 33 genes on the basis of at least one P value <.000005. Of these, 13 genes possessed a QTL within 5 Mb of their physical location. Hierarchical clustering was performed on the basis of both Pearson correlation of gene expression and genetic correlation. Both reflected biologically relevant activity taking place in the lymphoblastoid cell lines, with greater coherency represented in Kyoto Encyclopedia of Genes and Genomes database (KEGG) pathways than in Gene Ontology database pathways. However, more pathway coherence was observed in KEGG pathways when clustering was based on genetic correlation than when clustering was based on Pearson correlation. As more expression data in segregating populations are generated, viewing clusters or networks based on genetic correlation measures and shared QTLs will offer potentially novel insights into the relationship among genes that may underlie complex traits.

Figure 1

Effect sizes for heritable genes. The figure shows a histogram for estimates of heritability for those genes that are both differentially expressed and significantly heritable, when a false-discovery rate of 0.05 is used.

Figure 2

Linkage results for 2,430 differentially expressed genes. Multipoint linkage analysis was conducted every 4 cM over the autosomal genome. This figure summarizes the results by counting the number of genes with P values ⩽.0005, ⩽.00005, and ⩽.000005, for each of the 4-cM locations.

Figure 3

Effect sizes for QTLs detected at a pointwise significance level of .000005

Figure 4

A, Two-dimension agglomerative hierarchical cluster constructed in the experiment (Y-axis) and gene-expression (X-axis) dimensions, using PC as the similarity measure, on the 574 genes described in the text. Several clusters in the gene-expression dimension are apparent from this color matrix display. However, two genes, Pip5k1a and Pip5k2a, known to function in the phosphatidylinositol signaling pathway, are seen here to cluster into two completely separate clusters in the gene-expression tree. B, Two-dimension agglomerative hierarchical cluster constructed in the experiment (Y-axis) and gene expression (X-axis) dimensions, using GC as the similarity measure, on the 574 genes described in the text. Although there are clearly patterns of expression that are highly similar to those shown in panel A, there are differences that serve to highlight the information that can be derived from the genetics dimension. In this instance, the two genes indicated in panel A as operating in the same pathway but clustering far away from each other cluster relatively closely together.