Dynamic genetic regulation of gene expression during cellular differentiation

Abstract

Genetic regulation of gene expression is dynamic, as transcription can change during cell differentiation and across cell types. We mapped expression quantitative trait loci (eQTLs) throughout differentiation to elucidate the dynamics of genetic effects on cell type-specific gene expression. We generated time-series RNA sequencing data, capturing 16 time points during the differentiation of induced pluripotent stem cells to cardiomyocytes, in 19 human cell lines. We identified hundreds of dynamic eQTLs that change over time, with enrichment in enhancers of relevant cell types. We also found nonlinear dynamic eQTLs, which affect only intermediate stages of differentiation and cannot be found by using data from mature tissues. These fleeting genetic associations with gene regulation may explain some of the components of complex traits and disease. We highlight one example of a nonlinear eQTL that is associated with body mass index.

Fig. 1. Gene expression trends throughout cardiomyocyte differentiation.

(A) The first two gene expression principal component loadings for all 297 RNA-seq samples across cell lines, where each sample is colored by day of collection. (B) Predicted cell line cluster expression trajectories for 20 gene clusters according to split-GPM. Many gene clusters (8, 11, 15, 16, and 20) exhibit periodic expression trajectories that correspond with cell culture media changes.

Fig. 2. eQTL patterns during cardiomyocyte differentiation.

We limit to genes with at least one significant eQTL (WASP combined haplotype test; eFDR <= .05) across time points. If a gene has more than one significant eQTL, we select a single variant for that gene with the smallest geometric mean p-value across all 16 time points. (A) Spearman correlation of p-values between eQTLs from each day (x-axis) and existing iPSC (grey) and iPSC-derived cardiomyocyte (red) eQTLs. (B) Spearman correlation of eQTL p-values for each pair of days. (C). Factors identified via sparse matrix factorization of eQTL -log₁₀ p-values using 3 latent factors and a L1 penalty of .5.

Fig. 3. Dynamic eQTLs detect genetic regulatory changes caused by cardiomyocyte differentiation.

(A) Linear interaction association between genotype (color) of rs11124033 and time point (x-axis) on residual gene expression (cell line effects regressed on expression) of FHL2 (y-axis). (B) Enrichment of dynamic eQTLs within cell type specific chromHMM enhancer elements relative to 1000 sets of randomly selected matched background variants. Dynamic eQTLs were classified as early or late (C) Nonlinear interaction association between genotype (color) of rs28818910 and time point (x-axis) on residual gene expression of C15orf39 (y-axis). (D) Nonlinear interaction association significance of all variants tested within 50 KB of the C15orf39 transcription start site with expression of C15orf39 (green) and GWAS significance for BMI of variants in the same window (blue). Vertical line depicts genomic location of the most significant nonlinear dynamic eQTL (rs28818910) for C15orf39.