We have performed quantitative genetic analyses for genome-wide expression traits in a connected-cross set between three accessions in the model organism A. thaliana. Utilising the R/eqtl package in both crosses, we observed a 52.3% difference in the number of significant eQTLs between sets, with the Bur-0 × Col-0 cross being the one with the highest count (Figure 1). Regardless of this difference, the conservative figures of eQTLs obtained in both sets resemble the one previously described in RILs between accessions Ler-0 and Cvi-0 , but they significantly differ from a previous study involving Bay-0 and Shahdara accessions, where more than 36,000 eQTLs were detected . In our study, the use of a common approach for both sets allowed us to perform straightforward comparisons between them, drawing conclusions not necessarily affected by differences in statistical power . We observed that 17.4% of all local eQTLs detected were shared between crosses, likely due to polymorphisms solely present in Col-0 or shared by Cvi-0 and Bur-0 (although independent SNPs or some allelic heterogeneity pattern could also lead to this observation). An argument to support this hypothesis is the high level of shared eQTLs with an additive effect in the same direction (88.4%) and of similar strength (Figure 2). However, the low level of overlapping eQTLs demonstrates the greater number of cross-specific eQTLs, possibly due to the great complexity of expression traits [1, 11, 29] and their high potential for evolution . Our results are consistent with a recent survey in several Arabidopsis accessions that identified high allelic heterogeneity within local regulatory eQTLs [6, 12], suggesting the existence of many private polymorphisms between all three accessions extending the eQTL landscape.
Although cis-acting regulation among local eQTLs remains to be confirmed, the co-localisation of the eQTL and the controlled gene provides a robust probabilistic way of classifying eQTLs into their mode of action. We observed that, albeit the two crosses shared a common accession, the eQTL distribution among these types is very dissimilar. The BurCol set shows three times more distant than local eQTLs. At the same FDR, the CviCol set has a very different balance, with an equivalent number of both types of eQTLs (50% distant and 50% local, Figure 1c) and a lower number of hotspots (Figure 3) encompassing 44.2% of the trans eQTLs. Nevertheless, the distribution pattern of local eQTLs on each chromosome was conserved between crosses, likely due to a strong correlation with gene density and chromosome structural features. Also, we found that local eQTLs were overrepresented at higher significance in all crosses (Figure 1c-d), explaining a greater phenotypic variance per trait compared to distant eQTLs. Our result agrees with another study in Mouse, where many highly significant local-eQTLs and moderately significant trans-acting associations were observed . Furthermore, the BurCol recombinant population exhibited a higher number of distant eQTLs, concentrating 75% of them in 24 bins classified as hotspots (Figure 3). One of the most interesting intervals was the hotspot at bin #8, which contained the highest number of genes whose function is related to the regulation of transcription (52 genes), including the strong candidate GIGANTEA , a gene known to be involved in diverse developmental processes . It is difficult to guess whether a hotspot could be the expression of a locus actually controlling the transcription of many genes, or that of a major (for example, developmental) player that has indirect effects on many genes' expression (as secondary consequences of a strong phenotype for example). The presence of pleiotropic hotspots affecting the expression of many transcripts has already been described in yeast , where no more than 200 trans-eQTLs explained transcript variation for 1,716 traits. Similarly, many trans-hotspots were described in previous A. thaliana studies [3, 10]. In agreement with these studies, the hotspots detected here explained approximately 10% of the individual traits' variation, demonstrating their milder effect on transcript abundance compared to local eQTLs . Interestingly, only ~7% of the hotspots overlapped between crosses, suggesting either alternative master regulators or little conservation in complex regulatory networks . These conserved regions could also suggest the presence of polymorphisms in Col-0 as major contributors for the large number of trans eQTLs within these shared regions.
To increase the power in detecting eQTLs and identify milder genetic variations, we have utilised the bayesian framework VBQTL, a model designed to dissect gene expression variation in order to account for confounding factors . The implementation of this model in the CviCol set allowed us to detect twice as many eQTLs compared to the standard approach (Figure 4), without affecting the ratio of local versus distant eQTLs. Subsequently, we used this extended set and focused the posterior analyses on local eQTLs since they likely represent independent alleles, only affecting a single transcript and explaining a greater phenotypic variance, in contrast to distant eQTLs that may affect many genes and show minor phenotypic effects. The use of this new set of local eQTLs allowed us to look for potential traces of selection where Col alleles were overexpressed in a set of genes with related function. Assuming neutrality, no over-representation of either up-regulated or down-regulated alleles from the same accession is expected within a cluster of genes, unless a directional allelic drift has occurred, which may represent selection . We detected an overrepresentation of eQTLs in several GO functional categories and a significant skew within GO terms (Table 1, 2). The accumulation of local regulatory polymorphisms up-regulating Col alleles in all cases, suggests dissimilar patterns of responses to the environment and an advantage in a particular niche  or, else, a specific lack of cost in loosing this response. For example, one of the clusters of genes globally up-regulated in Col with respect to Cvi relates to hypoxia response, which is a major determinant of submergence tolerance. It was indeed recently shown that Col and Cvi contrasts for this trait, with Cvi being one of the least submergence-tolerant accessions . Interestingly, the entire set of significant categories here described are related to response to stress or biotic and abiotic response, highlighting its potential link with environmental and niche-adaptation changes.