Heritability and tissue specificity of expression quantitative trait loci

Abstract

Variation in gene expression is heritable and has been mapped to the genome in humans and model organisms as expression quantitative trait loci (eQTLs). We applied integrated genome-wide expression profiling and linkage analysis to the regulation of gene expression in fat, kidney, adrenal, and heart tissues using the BXH/HXB panel of rat recombinant inbred strains. Here, we report the influence of heritability and allelic effect of the quantitative trait locus on detection of cis- and trans-acting eQTLs and discuss how these factors operate in a tissue-specific context. We identified several hundred major eQTLs in each tissue and found that cis-acting eQTLs are highly heritable and easier to detect than trans-eQTLs. The proportion of heritable expression traits was similar in all tissues; however, heritability alone was not a reliable predictor of whether an eQTL will be detected. We empirically show how the use of heritability as a filter reduces the ability to discover trans-eQTLs, particularly for eQTLs with small effects. Only 3% of cis- and trans-eQTLs exhibited large allelic effects, explaining more than 40% of the phenotypic variance, suggestive of a highly polygenic control of gene expression. Power calculations indicated that, across tissues, minor differences in genetic effects are expected to have a significant impact on detection of trans-eQTLs. Trans-eQTLs generally show smaller effects than cis-eQTLs and have a higher false discovery rate, particularly in more heterogeneous tissues, suggesting that small biological variability, likely relating to tissue composition, may influence detection of trans-eQTLs in this system. We delineate the effects of genetic architecture on variation in gene expression and show the sensitivity of this experimental design to tissue sampling variability in large-scale eQTL studies.

Figure 1. Genetic Architecture of Genetic Variation in Gene Expression

For each considered transcript the major eQTL was identified by linkage analysis (genome-wide significance, p = 0.05) and characterised as cis or trans. Additive allelic effect and heritability (h ² _QTL) for each cis-eQTL (black symbol) and trans-eQTL (grey symbol) were plotted for LV, fat, kidney, and adrenal tissues.

Figure 2. Proportion of eQTLs That Are Excluded when Transcripts are Filtered Based on h ² _trait

Percentages were calculated as follows: (1 − (eQTL_TOT − eQTL_{h² cut-off})/eQTL_TOT) × 100, where eQTL_TOT is the total number of eQTLs detected at p = 0.05 without filtering based on h ² _trait (see Table 1), and eQTL_{h² cut-off} is the total number of eQTLs observed when transcripts were filtered based on a given trait heritability cut-off. For any trait heritability cut-off, the number of excluded eQTLs is higher for trans-acting eQTLs, since they are usually detected for transcripts with low heritability of gene expression.

Figure 3. Power to Detect a Major eQTL in the RI Strains

(A) Statistical power to detect an eQTL of given allelic effect is shown for various estimates of heritability of gene expression in 30 RI strains with four biological replicates. The eQTL allelic effect is an estimate of the absolute change in the transcript abundance that would be produced by substituting a single allele of one type with that of another type in the population (see Materials and Methods). Absolute FCs, corresponding to the allelic effects on the primary x-axis, are also reported on the secondary x-axis. (B) Statistical power to detect cis-eQTLs (solid line) and trans-eQTLs (dashed line) (detected with genome-wide significance, p = 0.05) is shown for various heritabilities of gene expression (h ² _trait) in LV, fat, kidney, and adrenal tissues. Specific allelic effects for cis- and trans-eQTLs were defined in accordance with those observed in this study at p = 0.05 (reported in Table 1).

Figure 4. FDR in LV, Fat, Kidney, and Adrenal Tissues

(A) For each major eQTL detected in LV, fat, kidney, and adrenal tissue, the expected FDR was calculated and plotted against different p-value thresholds in the range 10⁻⁶–0.05. Insert: FDR for various p-values in the range 10⁻⁶–10⁻³. (B) Vase box-plots for the FDRs of the cis- and trans-acting eQTLs detected at p = 0.05 in LV, fat, kidney, and adrenal tissue. Vase box-plots are box-plots where the width of the box at each point is proportional to the density of the data there. The thick line indicates the median FDR for each distribution.