Expression variation in connected recombinant populations of Arabidopsis thaliana highlights distinct transcriptome architectures

Abstract

Background: Expression traits can vary quantitatively between individuals and have a complex inheritance. Identification of the genetics underlying transcript variation can help in the understanding of phenotypic variation due to genetic factors regulating transcript abundance and shed light into divergence patterns. So far, only a limited number of studies have addressed this subject in Arabidopsis, with contrasting results due to dissimilar statistical power. Here, we present the transcriptome architecture in leaf tissue of two RIL sets obtained from a connected-cross design involving 3 commonly used accessions. We also present the transcriptome architecture observed in developing seeds of a third independent cross.

Results: The utilisation of the novel R/eqtl package (which goal is to automatize and extend functions from the R/qtl package) allowed us to map 4,290 and 6,534 eQTLs in the Cvi-0 × Col-0 and Bur-0 × Col-0 recombinant populations respectively. In agreement with previous studies, we observed a larger phenotypic variance explained by eQTLs in linkage with the controlled gene (potentially cis-acting), compared to distant loci (acting necessarily indirectly or in trans). Distant eQTLs hotspots were essentially not conserved between crosses, but instead, cross-specific. Accounting for confounding factors using a probabilistic approach (VBQTL) increased the mapping resolution and the number of significant associations. Moreover, using local eQTLs obtained from this approach, we detected evidence for a directional allelic effect in genes with related function, where significantly more eQTLs than expected by chance were up-regulated from one of the accessions. Primary experimental data, analysis parameters, eQTL results and visualisation of LOD score curves presented here are stored and accessible through the QTLstore service database http://qtlstore.versailles.inra.fr/.

Conclusions: Our results demonstrate the extensive diversity and moderately conserved eQTL landscape between crosses and validate the utilisation of expression traits to explore for candidates behind phenotypic variation among accessions. Furthermore, this stresses the need for a wider spectrum of diversity to fully understand expression trait variation within a species.

Figure 1

Genetic landscape for transcript accumulation variation. a. eQTL heatmap for CviCol population significant at a 5% FDR. Each horizontal bar represents an eQTL mapped on the x-axis and controlling the accumulation of a transcript expressed from the locus indicated on the y-axis. The colour of the bar indicates the direction and strength of the eQTL additive effect, and its length along the x axis encompasses the eQTL support interval. Local eQTLs form the diagonal, while distant eQTLs fall elsewhere in the map. b. eQTL heatmap for BurCol as described in a. c. Bar plot indicating the proportion of local and distant eQTLs for increasing LOD value intervals in CviCol set. d. Bar plot as in c. for BurCol set.

Figure 2

Correlation between relative expression levels for potentially shared local eQTLs between the CviCol and BurCol sets. The relative allelic expression level at the eQTL was plotted for transcripts sharing local eQTLs with additive effects in the same direction in both populations. The linear regression and correlation is indicated with a dashed line.

Figure 3

Distribution of distant eQTLs across the genome and detection of hotspots. The number of distant eQTLs (y-axis) is plotted against the physical position of the 1Mb-window where they peak (x-axis). In each cross, intervals with an excess of eQTLs relative to the threshold estimated by permutation (dashed lines) were classified as hotspots. This figure refers to Additional file 5: Table S3b.

Figure 4

Bar plot showing overlap and specific eQTL detections when comparing the VBQTL and standard approach. Linkage mapping methods were compared and blue regions denote the percentage of common eQTLs mapped at a 5% FDR using both approaches. eQTLs solely mapped in one or the other strategy are depicted at the bar-extremes (in green and brown). Ten factors were included in the VBQTL analysis. a. CviCol set, b. BurCol set.