Gene set enrichment in eQTL data identifies novel annotations and pathway regulators

Abstract

Genome-wide gene expression profiling has been extensively used to generate biological hypotheses based on differential expression. Recently, many studies have used microarrays to measure gene expression levels across genetic mapping populations. These gene expression phenotypes have been used for genome-wide association analyses, an analysis referred to as expression QTL (eQTL) mapping. Here, eQTL analysis was performed in adipose tissue from 28 inbred strains of mice. We focused our analysis on "trans-eQTL bands", defined as instances in which the expression patterns of many genes were all associated to a common genetic locus. Genes comprising trans-eQTL bands were screened for enrichments in functional gene sets representing known biological pathways, and genes located at associated trans-eQTL band loci were considered candidate transcriptional modulators. We demonstrate that these patterns were enriched for previously characterized relationships between known upstream transcriptional regulators and their downstream target genes. Moreover, we used this strategy to identify both novel regulators and novel members of known pathways. Finally, based on a putative regulatory relationship identified in our analysis, we identified and validated a previously uncharacterized role for cyclin H in the regulation of oxidative phosphorylation. We believe that the specific molecular hypotheses generated in this study will reveal many additional pathway members and regulators, and that the analysis approaches described herein will be broadly applicable to other eQTL data sets.

Figure 1. The global view of eQTL mapping results adipose tissue.

The x-axis shows the absolute genomic position on the SNP/QTL axis (chromosomes shown in alternating colors), and y-axis shows the absolute genomic position of the genes whose expression was measured. Each data point in the graph represents an association between the genotype at a specific genomic locus and the expression of a gene. The strong diagonal pattern is called the “cis-eQTL band” and represents an association between the expression level of a gene and the genotype at the gene's locus. In addition, multiple vertical bands (“trans-eQTL bands”) illustrate associations between the expression of many genes and the genotype at a single genomic locus. For clarity, only data points with association scores greater than 4.2 are shown (roughly 5000 data points).

Figure 2. A trans-eQTL band enriched in genes in the “Oxidative Phosphorylation” pathway.

(A) The expression patterns of 68 genes (“trans-band targets”) were all associated to the haplotype pattern at a region on chromosome 13 near 81.8 Mb (six strongest associations shown). Strains were assigned to haplotype groups based on local windows of genotype calls ,,. Each point represents the median-centered expression value on log₁₀-scale for a given strain in the haplotype group. (B) Trans-band targets were tested for enrichment in FGS from GO, KEGG, and Ingenuity pathways database (ING). The genes in this trans-eQTL band were found to be strongly associated with the Oxidative Phosphorylation pathway, with the most significant enrichment (S_e = −21.26) found among the top 40 annotated eQTL associations. For comparison, 100 random permutations of the eQTL association scores showed a maximum enrichment score of −4.13. Visualization of the enrichment analysis is shown as described in . (C) Expression profiles in the Gene Atlas data set are strongly correlated for the 19 genes (red) which are annotated as being involved in the Oxidative Phosphorylation pathway (“eQTL/FGS genes”). In addition, 10 genes (blue, “inferred”) were also highly correlated which were not annotated in this pathway, but whose role in oxidative phosphorylation could be inferred. The remaining 39 genes (gray, “unknown”) in the trans-eQTL band do not share a correlated expression pattern in the Gene Atlas data set. Tissues are sorted from lowest to highest average expression according to 19 eQTL/FGS genes (red). All expression values have been median-centered on log₁₀-scale. The “avg. r²” label indicates the average Pearson correlation to the genes in the eQTL/FGS group.

Figure 3. Use of the Gene Atlas reference data set to identify novel pathway members.

(A) Median pairwise correlation coefficients among trans-band targets in the Gene Atlas data set were calculated for all 1593 enriched pairs of trans-eQTL bands and functional gene sets (FGS). Trans-band target genes which were annotated in the enriched category (eQTL/FGS genes) were much more likely to share a correlated expression pattern in the Gene Atlas data set than randomly chosen genes from trans-band targets. The 440 trans-eQTL bands in which eQTL/FGS genes had median pairwise correlation greater than 0.5 were selected for further study. (B) Using eQTL/FGS genes as a set of true positives, a jackknife procedure was used to calculate the median expression of each eQTL/FGS gene to the remaining eQTL/FGS genes (solid line). For comparison, an analogous calculation was performed for a presumed set of true negatives (a random set of genes) of the same size as the non-FGS genes in the trans-eQTL band (dashed line). The ratio between these two distributions was used to define a median correlation threshold of R>0.59 at 20% false discovery rate (FDR). When applied to the set of all unannotated trans-band targets (dotted line), annotation for 350 genes could be inferred.

Figure 4. Validation of cyclin H as a regulator of oxidative phosphorylation.

(A) Three individual siRNAs against two candidate genes (CCNH and RASA1) were transfected into HIB1B preadipocytes. siRNAs against PGC-1α and functional partners of CCNH were used for comparison. Mitochondrial density and oxidative phosphorylation (OXPHOS) were evaluated by FACS (see Methods). (B) RNA isolated from cells treated as in (A) was used to verify knock-down of gene expression using qRT-PCR. (C) Overexpression of CCNH, but not RASA1, resulted in increased mitochondrial activity. HIB1B cells were transfected with expression vectors for the indicated genes, and mitochondrial parameters were measured 72 hr later. (D) To confirm the effect on oxidative phosphorylation measured using Mitotracker dyes, expression of six known OXPHOS genes was assayed using qRT-PCR in response to siRNAs targeting PGC-1α, CCNH, and RASA1. In all cases, knock down of CCNH, but not RASA1, resulted in decreased expression. (E) To confirm that the effect of CCNH was not due to general effects on transcription or cell viability, the expression of four control genes unrelated to OXPHOS was also measured by qRT-PCR. In all cases, siRNAs targeting CCNH produced no significant change in expression.