Single nucleotide polymorphisms affect both cis- and trans-eQTLs

Abstract

Single nucleotide polymorphisms (SNPs) between microarray probes and RNA targets can affect the performance of expression array by weakening the hybridization. In this paper, we examined the effect of the SNPs on Affymetrix GeneChip probe set summaries and the expression quantitative trait loci (eQTL) mapping results in two eQTL datasets, one from mouse and one from human. We showed that removing SNP-containing probes significantly changed the probe set summaries and the more SNP-containing probes we removed the greater the change. Comparison of the eQTL mapping results between with and without SNP-containing probes showed that less than 70% of the significant eQTL peaks were concordant regardless of the significance threshold. These results indicate that SNPs do affect both probe set summaries and eQTLs (both cis and trans), thus SNP-containing probes should be filtered out to improve the performance of eQTL mapping.

Figure 1

Distributions of SNP-containing probes and SNP-containing probe sets on the Affymetrix arrays used in our study. The human data were generated using the Affymetrix Human Focus Arrays containing probes for 8,500 transcripts. The mouse data were generated using the Affymetrix Mouse U74Av2 microarrays, which contain 12488 probe sets. The SNP-containing probes of the human array in the studied population were established based on the HapMap Phase 3 data of CEPH population. The mouse SNP-containing probes were established based on the SNPs between the two parental inbred lines for the BXD RI population in the mouse SNP database. SNP-probe, SNP-containing probe

Figure 2

Examples of false positive and false negative cis-QTLs caused by SNP-containing probes. The LOD scores are plotted at the regions surrounding the target gene locations in the genome for each probe in the probe set. The red solid and black dashed lines represent SNP-containing and SNP-free probes, respectively. LOD curves of probe set 104225_at show that 11 SNP-free probes have a cis-QTL at 90cM on chromosome 2 while the SNP-containing probe 104225_at1 does not have the QTL on that location, which illustrates the false negative caused by SNP-containing probe. LOD curves of probe set 95010_at show that SNP-containing probe 95010_at16 has a cis-QTL at 92cM on chromosome 12 while the SNP-free probes in the same probe set do not.

Figure 3

Box plots of Pearson correlation coefficients of probe set summaries. The coefficients were obtained from correlating the original probe set summaries with those after filtering out probes. The black plots are for correlations between the original summaries and those after filtering out the SNP-containing probes. The red plots are for correlations between the original summaries and those from random probe filtering. Correlation coefficients were calculated after normalizing all individuals in the population together.

Figure 4

Histograms of p values obtained from testing the effect of removing SNP-containing probes on probe set summaries. The extra small p values compared with a uniform distribution indicate that in some probe sets correlation obtained from filtering out the SNP-containing probes is significantly weaker than that obtained from randomly filtering out equal number of probes. SNP-probe, SNP-containing probe.

Figure 5

Histograms for the numbers of overlapping eQTLs between randomly removing probes and the original mouse data. Probes were randomly filtered out of the SNP-containing probe sets to generate new datasets. The same preprocess and QTL mapping methods were applied to the new datasets. The eQTLs from the new datasets were compared to the eQTLs obtained from the same probe sets in the original dataset. The triangles point to the number of overlapping eQTLs between the original dataset and the dataset with all the SNP-containing probes filtered out. Ps, probe sets; SNP-probe, SNP-containing probe