Detection of parent-of-origin specific expression quantitative trait loci by cis-association analysis of gene expression in trios

Abstract

Parent-of-origin (PofO) effects, such as imprinting are a phenomenon in which homologous chromosomes exhibit differential gene expression and epigenetic modifications according to their parental origin. Such non-Mendelian inheritance patterns are generally ignored by conventional association studies, as these tests consider the maternal and paternal alleles as equivalent. To identify regulatory regions that show PofO effects on gene expression (imprinted expression Quantitative Trait Loci, ieQTLs), here we have developed a novel method in which we associate SNP genotypes of defined parental origin with gene expression levels. We applied this method to study 59 HapMap phase II parent-offspring trios. By analyzing mother/father/child trios, rules of Mendelian inheritance allowed the parental origin to be defined for ~95% of SNPs in each child. We used 680,475 informative SNPs and corresponding expression data for 92,167 probe sets from Affymetrix GeneChip Human Exon 1.0 ST arrays and performed four independent cis-association analyses with the expression level of RefSeq genes within 1 Mb using PLINK. Independent analyses of maternal and paternal genotypes identified two significant cis-ieQTLs (p<10(-7)) at which expression of genes SFT2D2 and SRRT associated exclusively with maternally inherited SNPs rs3753292 and rs6945374, respectively. 28 additional suggestive cis-associations with only maternal or paternal SNPs were found at a lower stringency threshold of p<10(-6), including associations with two known imprinted genes PEG10 and TRAPPC9, demonstrating the efficacy of our method. Furthermore, comparison of our method that utilizes independent analyses of maternal and paternal genotypes with the Likelihood Ratio Test (LRT) showed it to be more effective for detecting imprinting effects than the LRT. Our method represents a novel approach that can identify imprinted regulatory elements that control gene expression, suggesting novel PofO effects in the human genome.

Figure 1. Example of determination of parental origin and parent-of-origin specific association testing for a hypothetical SNP.

(A) For the given SNP (alleles A and B), the homozygous mother can only contribute allele A to her offspring, which means that the child's A allele is maternally inherited, and child's B allele must therefore be paternally inherited. Maternal (_MAT) and paternal (_PAT) alleles are shown in pink and blue, respectively. (B) Example data showing annotation of parental origin of this SNP in 10 individuals. (C) cis-association study with SNPs within ±1 MB around the gene. (D) Comparison of standard eQTL study and imprinted eQTL study (see Methods for details).

Figure 2. Putative cis-ieQTLs detected by parent of origin specific association analysis in 59 HapMap trios.

Results for (A) SRRT and (B) SLC27A4 and (C) PEG10. In each plot, the panels show the distribution of p-values for each SNP within a ±1 Mb region around each gene based on separate association of the maternally- and paternally-inherited SNPs (left and middle panels respectively). The x-axis shows the genomic coordinates in Mb, and y-axis is the –log₁₀ p-value of association between each SNP allele of defined parental origin and gene expression level. Box plots show the distributions of gene expression for CEU (circles) and YRI offspring (crosses) with respect to maternally (_MAT) and paternally (_PAT) inherited SNPs. A clear difference between heterozygous individuals of reciprocal parental origin is also consistent with a PofO effect at these loci, as these individuals show a difference in gene expression despite having the same genotype (right panels).

Figure 3. Comparison of separate association using maternal and paternal alleles with the Likelihood Ratio Test (LRT) for putative 30 ieQTLs.

(A) Correlation between –log₁₀ p-values generated using a standard association test and the LRT (r² = −0.87) (B) Correlation between -log₁₀ p-values generated using the heterozygote test and the LRT (r² = 0.71). Note that 6 of the 30 SNP-gene pairs did not have a sufficient number of heterozygotes to perform the test.

Figure 4. The two most significant SNP-gene pairs identified using the Likelihood Ratio Test.

Results for (A) PSCD4 and (B) TBC1D7. In each plot the three panels show variation of gene expression with respect to maternally inherited SNPs, paternally inherited SNPs and two classes of heterozygotes, respectively. Box plots show the distributions of gene expression for CEU (circles) and YRI offspring (crosses) with respect to maternally (_MAT) and paternally (_PAT) inherited SNPs. Note that unlike loci identified using separate association with maternal and paternal alleles where a difference in gene expression level is seen for only one of the two parental alleles, for both genes the A and C alleles have opposite effects on gene expression when inherited maternally versus paternally.