Differentially expressed genes reflect disease-induced rather than disease-causing changes in the transcriptome

Comparing transcript levels between healthy and diseased individuals allows the identification of differentially expressed genes, which may be causes, consequences or mere correlates of the disease under scrutiny. Here, we propose a bi-directional Transcriptome-Wide Mendelian Randomization (TWMR) approach that integrates summary-level data from GWAS and whole-blood eQTLs in a MR framework to investigate the causal effects between gene expression and complex traits. Whereas we have previously developed a TWMR approach to elucidate gene expression to trait causal effects, here we are adapting the method to shed light on the causal imprint of complex traits on transcript levels. We termed this new approach reverse TWMR (revTWMR). Integrating bi-directional causal effects between gene expression and complex traits enables to evaluate their respective contributions to the correlation between gene expression and traits. We uncovered that whole blood gene expression-trait correlation is mainly driven by causal effect from the phenotype on the expression rather than the reverse. For example, BMI- and triglycerides-gene expression correlation coefficients robustly correlate with trait-to-expression causal effects (r=0.09, P=1.54x10-39 and r=0.09, P=1.19x10-34, respectively), but not detectably with expression-to-trait effects. Genes implicated by revTWMR confirmed known associations, such as rheumathoid arthritis and Crohns disease induced changes in expression of TRBV and GBP2, respectively. They also shed light on how clinical biomarkers can influence their own levels. For instance, we observed that high levels of high-density lipoprotein (HDL) cholesterol lowers the expression of genes involved in cholesterol biosynthesis (SQLE, FDFT1) and increases the expression of genes responsible for cholesterol efflux (ABCA1, ABCG1), two key molecular pathways in determining HDL levels. Importantly, revTWMR is more robust to pleiotropy than polygenic risk score (PRS) approaches which can be misled by pleiotropic outliers. As one example, revTWMR revealed that the previously reported association between educational attainment PRS and STX1B is exclusively driven by a highly pleiotropic SNP (rs2456973), which is strongly associated with several hematological and anthropometric traits. In conclusion, our method disentangles the relationship between gene expression and phenotypes and reveals that complex traits have more pronounced impact on gene expression than the reverse. We demonstrated that studies comparing the transcriptome of diseased and healthy subjects are more prone to reveal disease-induced gene expression changes rather than disease causing ones.