Fig. 1

Summary of analysis workflow. a A uniform computational pipeline to analyze data from four immune RNA-seq datasets (DICE, BLUEPRINT, GEUVADIS, and DGN). The same pipeline for genotype imputation, expression and splicing quantification and QTL-mapping were applied to the four datasets. Sharing of QTLs among celltypes were quantified using mash [19], a statistical method that leverages the correlation structure of QTL effect sizes across multiple samples to re-estimate QTL effect in each sample. Colocalization analyses were performed for 72 GWAS of immune-related and non-immune traits. b Total number of genes and intron clusters with a significant QTL identified in DICE (left) and the other three studies (right) as a function of sample sizes. QTLs are considered significant when Storey’s q-value is below 0.05. c Studies with larger effective sequencing depth (BLUEPRINT and GEUVADIS EUR) have more sQTLs comapred to other studies. Effective sequencing depth = library size × read-length. Red line represents the fitted line in a simple linear model. d An eQTL at the gene CDK10 that is shared by all 15 cell types in DICE despite large differences in baseline expression levels across cell types. e An eQTL at the IL15RA gene that is shared across immune cell types but show cell type-specificity according to linear regression. Sharing of QTLs among cell types were quantified using mash [19], a statistical method that leverages the correlation structure of QTL effect sizes across multiple samples to re-estimate QTL effect in each sample. lm Z: Z-scores of linear model from FastQTL, mash Z: Z-scores estimated by mash (red). The lm Z-scores were colored in gray when the Z-score did not pass statistical significance after FastQTL permutation and in black when they were determined to be significant