The eQTL Mapping in African Americans With Local Ancestry

Expression quantitative trait loci (eQTL) are genetic variants that are associated with gene expression. Since the majority of trait-associated variants identified by genome-wide association study (GWAS) reside in non-coding regions, eQTLs are useful to explain the underlying mechanism of GWAS variants because of their regulatory effects. Unfortunately, recent eQTLs mappings are mostly performed in continental populations and ignore other minor populations. The eQTL mapping in admixed population (i.e. African Americans) is not only urgently needed to decrease the disparity in knowledge of gene expression regulatory mechanisms across populations, but also has great potential to reveal novel regulatory variants because the admixed population usually harbor more variants than founding populations. However, since the effect size of individual eQTL is usually small, the minor variation in allele frequency may lead to spurious association. For admixed population, in which the genome is mosaic of different origins, the global adjustment such as principle components is shown to be insufficient to remove the population stratification at local regions, and the adjustment of local ancestry at each test SNP is needed. However, whether the local ancestry adjustment is essential for controlling false positive rate in eQTL mapping where the trait is quantitative has not been studied. Here we investigated what are the effects of local ancestry adjustment on the control of false positive rate in eQTL mapping for admixed population. Because the eQTL mapping involves the association test of millions of SNPs with tens of thousands genes, we first developed a high efficient algorithm to make this task computational manageable and then performed extensive simulations and applied our method on an African American lymphoblastoid cell line (LCL) data. Our algorithm can finish 2,576,024 association tests in 10 minutes running on Quest.. In the simulation for the comparison of type-1 error rate, differential gene expression was assumed between CEU and African American populations. The genomic control factor is largely inflated before adjusting for population stratification (15.41) and is well-controlled after adjusting for principal components (1.02) and local ancestry (1.02). However, we did not notice a significant difference between two adjustment methods. The similar scenarios were observed in the AA LCL data. The genomic control factor is 1.041 before adjustment and decreases to 1.014 and 1.023 after the global adjustment and local adjustment respectively. Even though the inflation is enlarged for SNPs of Fst larger than 0.5, the relative magnitude remains. Our results suggest the principal components and local ancestry are both powerful to control the population stratification and the differential expression across population is mostly due to the overall population genomic structure instead of the ancestry at local genomic regions.