The Genom of Homo sapiens.pdf

404 CHEUNG ET AL.netic analysis. Expression phenotypes are good modelsfor developing molecular and statistical tools for analyzingquantitative traits more generally. In comparison toother human phenotypes, it is relatively easy to measurea large number of expression phenotypes at once in thesame person. There are various mechanisms that mightregulate gene expression in humans. In some cases, theexpression level of a gene is regulated only by a cis-actingelement. In others, the expression level is regulated byseveral genes that act in trans or a combination of regulatoryelements. These levels of complexity make expressionphenotypes good models for the many componentsunderlying complex traits and diseases in humans.ANALYSIS OF NATURAL VARIATION INHUMAN GENE EXPRESSIONOur goals are (1) to define the extent of variation ingene expression in normal individuals and (2) to determinethe genetic basis of that variation. We analyzed theexpression levels of genes in lymphoblastoid cells from50 unrelated individuals from the Utah pedigrees of theCentre d’Etude du Polymorphisme Humain (CEPH) usingmicroarrays (Affymetrix Human Genome FocusChip) that contain about 8500 human genes. Expressionprofiling for each sample was performed in duplicate. Asa measure of variation in gene expression, for each gene,we calculated the variance ratio (variance among individualsdivided by mean variance between microarray replicateson same individual). This allows us to characterizevariation among individuals relative to measurementnoise (Cheung et al. 2003). We ranked the genes by varianceratio and focused on those that are the most variable(i.e., with the highest variance ratio) since they are likelyto be more amenable to genetic dissection. Then, we measuredthe expression levels of some of these “variable”genes in individuals in large families and also in sets ofmonozygotic twin pairs.the differences in expression levels of 15 highly variablegenes among unrelated individuals to those betweenmonozygotic (MZ) twins. We found that in these transformedcells, the variance within MZ twin pairs, as a fractionof variance among unrelated individuals, rangedfrom 0.002 to 0.57 (mean 0.19, median 0.19). These findingsindicated that the expression levels of genes arehighly correlated among monozygotic twins compared tounrelated individuals. Differences in expression levels ofgenes in lymphoblastoid cells reflect germ-line geneticdifferences, despite the transformation process.VARIABLE GENESWe analyzed data for ~3800 (45%) of the 8500 genes,restricting attention to those that were expressed in atleast 10 out of the 50 unrelated individuals who werestudied. Among these genes, the variance ratios rangedfrom 0.2 to 48.9 (mean 2.6, median 1.6). For most of thegenes, the variance of expression levels between individualsis higher than the variance between microarray replicates(Fig. 1). This shows the reproducibility of the microarraydata. In our data, some genes have varianceratios that are approximately 1; for these genes, there isno evidence for meaningful variation among individuals.Therefore, in order to maximize the chance to detect geneticdifferences that account for variation in gene expression,we focus on the genes with variance ratio appreciablygreater than 1. We expect that these are thegenes where variation is most likely to be biologicallymeaningful, and therefore potentially due to genetic differences.The points in red in Figure 1 represent the genes(top 5%) with the highest variance ratios.We examined the genomic location of these “variable”genes (Fig. 2). We found that they were not clustered inthe genome; instead, they mapped to many sites acrossthe genome. Figure 2 shows the chromosomal locationsof the 50 most variable genes. For these genes, the rangeLYMPHOBLASTOID CELLS AS RNA SOURCEFOR GENE EXPRESSION ANALYSISWe have chosen to use lymphoblastoid cells in our studyfor several reasons. First, we need an RNA source that canbe obtained from a large number of normal individuals inlarge pedigrees. Immortalized lymphocytes (transformedby EBV) are available from all the members of the CEPHpedigrees. These are exceptionally large three-generationfamilies that have been studied extensively. Genotypes forgenetic mapping are available for many of these families,which will facilitate our effort to map the genetic determinantsof variation in gene expression. Second, lymphoblastoidcells from our study subjects can be grown undernear-identical conditions. It has been shown that largedifferences are found in expression levels of genes that arestudied on different occasions in fresh blood samples fromthe same individual (Whitney et al. 2003). With the lymphoblastoidcells, we can control the growth conditions inorder to minimize environmental variation.We were concerned about how transformation may affectgene expression. To examine this issue, we comparedFigure 1. Scatter plot of variance in expression levels betweenindividuals and between microarray replicates for ~3800 genes.The genes with the highest variance ratios (top 5%) are highlightedin red. The solid line indicates a variance ratio of 1.0.