Comparing allele specific expression and local expression quantitative trait loci and the influence of gene expression on complex trait variation in cattle

Abstract

Background: The mutations changing the expression level of a gene, or expression quantitative trait loci (eQTL), can be identified by testing the association between genetic variants and gene expression in multiple individuals (eQTL mapping), or by comparing the expression of the alleles in a heterozygous individual (allele specific expression or ASE analysis). The aims of the study were to find and compare ASE and local eQTL in 4 bovine RNA-sequencing (RNA-Seq) datasets, validate them in an independent ASE study and investigate if they are associated with complex trait variation.

Results: We present a novel method for distinguishing between ASE driven by polymorphisms in cis and parent of origin effects. We found that single nucleotide polymorphisms (SNPs) driving ASE are also often local eQTL and therefore presumably cis eQTL. These SNPs often, but not always, affect gene expression in multiple tissues and, when they do, the allele increasing expression is usually the same. However, there were systematic differences between ASE and local eQTL and between tissues and breeds. We also found that SNPs significantly associated with gene expression (p < 0.001) were likely to influence some complex traits (p < 0.001), which means that some mutations influence variation in complex traits by changing the expression level of genes.

Conclusion: We conclude that ASE detects phenomenon that overlap with local eQTL, but there are also systematic differences between the SNPs discovered by the two methods. Some mutations influencing complex traits are actually eQTL and can be discovered using RNA-Seq including eQTL in the genes CAST, CAPN1, LCORL and LEPROTL1.

Keywords: ASE analysis; Allele specific expression; Genetic variation of complex traits; Genome-wide association study; RNA-sequencing; eQTL mapping.

Fig. 1

Allele specific expression (ASE), parent-of-origin allele specific expression (PO-ASE) and local expression quantitative trait loci (eQTL) diagram. a The gene in the figure has two SNPs in the exons (tSNP₁ and tSNP₂), and their expression can be measured using RNA-Seq and a SNP within 50 kb of the gene which is driving expression of the gene (dSNP). b In the case of PO-ASE, the allele inherited from the dam (in this example) increases the expression of the tSNP allele which is on the maternal chromosome (♀). c When the dSNP has an ASE effect, allele A of the dSNP (in this example) triggers the expression of tSNP allele on the same chromosome. d In local eQTL mapping, the expression of dSNP allele is correlated with total exon expression, where the heterozygote individual for dSNP shows medium expression (A/T, individual 2) and the homozygotes show either high (A/A, individual 1) or low (T/T, individual 3) expression

Fig. 2

GWAS results for tenderness (MQLDPF) and gene expression results, using RNA-Seq meta-analyses of all datasets, within 50 kb of calpastatin gene (CAST). The Manhattan plots show the GWAS results for MQLDPF in grey. The common GWAS suggestive line at p < 10^− 5 and the genome-wide threshold line at p < 5 × 10^− 8 are shown in blue and red, respectively. The dark shaded box indicates the location of CAST and light shaded boxes show 50 kb upstream and downstream of the gene. For ease of comparison, the common SNPs from both the GWAS and the gene expression study are plotted in each graph. The SNPs within 50 kb of calpastatin gene significantly associated with MQLDPF (p < 0.001) are in blue. The SNP significantly associated with expression of the gene (p < 0.001) as detected by ASE (top), local eQTL mapping (middle) and meta-analysis of ASE and local eQTL (bottom) are in red. Where the same dSNP was tested for multiple tSNP, the lowest p-value for each dSNP is shown in the graphs. Although ASE and local eQTL mapping are different measurements they show similar results

Fig. 3

The overlap between QTL and gene expression within 50 kb of calpain (CAPN1), leptin receptor transcript-like 1 gene (LEPROTL1) and ligand dependent nuclear receptor corepressor-like (LCORL) genes. The Manhattan plots show the GWAS results for MQLDPF (top), PW_HIP (middle) and the multi-trait test (bottom) in grey. The common GWAS suggestive line at 10^− 5 and the genome-wide threshold line at 5 × 10^− 5 are shown in blue and red, respectively. The dark shaded boxes and light shaded boxes indicate the location and 50 kb upstream and downstream of the genes. For ease of comparison, the common SNPs from both the GWAS and the gene expression study are plotted in each graph. The SNPs within 50 kb of CAPN1, LEPROTL1 and LCORL genes, tested in gene expression analyses and significant GWAS (p < 0.001) are in blue. The SNPs significantly influenced expression of CAPN1 and LEPROTL1 in ASE measurement, and LCORL in combined ASE and eQTL meta-analysis of all RNA-Seq datasets (p < 0.001) are shown in red

Fig. 4

The distribution of the Χ² statistic for the association between QTL and eQTL from a permutation test. This example is from the comparison of eQTL from a meta-analysis of all tissues and the multi-trait GWAS. The enriched and impoverished overlapping QTL and eQTL are shown in positive and negative values, respectively. The theoretical thresholds for a Χ²_df = 1 test is 3.84 (enrichment) and − 3.84 (impoverishment) for significance level p < 0.05 as are shown by red dotted lines. The empirical thresholds based on the highest and lowest 5% of values in 10,000 permutation tests, are shown with blue dotted lines. Therefore, in the eQTL meta analysis test, the X² for an enrichment test should be greater than 6.64 for significance level p < 0.05