Effect of RNA half-life on mRNA levels. (A) Example genes (short-lived histone gene Hist1h2bb (orange) and the stable gene Adck5 (purple)) illustrating the inference of mRNA half-lives from expression data. Data points correspond to measured mRNA abundance at various time points after inhibition of transcription (time zero). (B) Half-life distribution of RefSeq genes with estimated mRNA decay rates. Half-lives of very stable genes were set to 21 h (the maximal inferable half-life given the experimental set-up) (see Supplementary information section 5 for details). (C) Genes are classified into five equal groups according to predicted transcription rate (0–100%), and within each group measured mRNA levels are shown as boxplots separately for genes with different mRNA half-life (color coded). Within a transcription group with a sufficient number of genes, short-lived genes show less measurable mRNA than long-lived genes. In the two low transcription bins (0–40%), mRNA levels are less well modeled and they are depleted of short-lived mRNAs (see Supplementary Figure 7 for illustration).

Focus on high-confidence microRNA target genes. (A) Same plot as in Figure 1C. All RefSeq genes are classified into three color-coded groups according to upregulated in Dicer^−/− (orange), downregulated in Dicer^−/− (purple), and unchanged (gray). (B) Subset of all the genes in (A), where the absolute log fold-change (logFC) between Dicer^−/− and Dicer^+/− is >0.5. This subset contains likely targets and non-targets. (C) Pearson correlation (r) between the residual of the linear model and the logFC between Dicer^−/− and Dicer^+/− (see Supplementary information section 7 for details) as a function of cutoff in absolute logFC between Dicer^−/− and Dicer^+/−. A logFC cutoff of zero corresponds to (A), and a cutoff of 0.5 (solid vertical line) corresponds to (B). Correlations are shown for subsets of genes for logFC cutoffs incremented in 0.1 intervals. The point size illustrates the number of genes at each cutoff. At 0.8, the subset contains 1000 genes (dashed line), the subset at 1.9 contains 100 genes (dotted line). Increasing logFC cutoffs select higher confidence microRNA-target genes that can explain the residual of the linear fit increasingly better.

Post-transcriptional regulation in tissue-specific and ubiquitously expressed genes. (A) Histogram of the number of cell or tissue types with detectable expression of the analyzed genes. Genes are grouped in tissue-specific (expressed in 1–5 tissues, purple), intermediate (gray), and ubiquitously expressed (expressed in 70 or more tissues, green). (B) Genes are classified into five equal groups according to predicted transcription rate (0–100%), and within each group measured mRNA levels are shown as boxplots separately for genes with different tissue expression (as in (A), color coded). At a given level of transcription, tissue-specific genes have on average less measured mRNA than ubiquitously expressed genes, suggesting that the degree of post-transcriptional regulation is higher in tissue-specific genes.

Using histone marks and RNA polymerase II to model mRNA levels. (A) Metagene plot showing the distribution of histone marks along the gene body of genes aligned at their TSS with low, intermediate, and high expression levels. (B) Scatter plot of RNA polymerase II (Pol-II, green) and three histone marks H3K36me3 (dark blue), H3K4me2 (light blue), H3K27me3 (orange) versus mRNA levels on the vertical axis. The number of reads aligned to either gene body (H3K36me3, mRNA) or at the TSS (H3K4me2, H3K27me3, Pol-II) is shown in logarithmic scale. (C) Predicted transcription rate combining the four measures in a linear model versus mRNA level. Axes as in (B). (D) Bar plot showing the fraction of total variance in mRNA levels that is explained by each single histone mark, Pol-II occupancy or a linear combination of them (dark gray). The maximally explainable variance (light gray) is limited by the amount of measurement noise (see Supplementary information section 4 for details). Error bars indicate 95% confidence interval.

Effect of targeting by microRNAs on mRNA levels. (A) Scatter plot of Dicer^+/− versus Dicer^−/− ES cells inferred by microarray measurement (Sinkkonen et al, 2008). Genes with increased mRNA levels in Dicer^−/− are enriched for putative microRNA targets (orange), while genes with decreased mRNA levels are possibly affected by secondary effects (purple). (B) Distribution of the log fold-change (logFC) between Dicer^−/− and Dicer^+/−. (C) Genes are classified into five equal groups according to predicted transcription rate (0–100%), and within each group measured mRNA levels are shown as boxplots separately for genes with different log fold-change between Dicer^−/− and Dicer^+/− (color coded). Within the same transcription group, putative microRNA target genes (orange) show insignificantly different mRNA levels as non-target genes (P-value=0.303).

H3K36me3 explains most of the variance in mRNA level. (A) Scatter plot of predicted transcription rate versus measured mRNA level for the ES, NP, and TN. (B) Distribution of mRNA levels in ES, categorized into low and high expression groups. (C) Correlation (r) between H3K36me3 and mRNA for genes in expression groups from (B) in ES, NP, and TN. The correlation of H3K36me3 with mRNA level differs between dividing cells and postmitotic TN cells: In diving cells (ES and NP), it is best for high expressed genes, while in the postmitotic TN, it is best for low expressed genes. Error bars indicate 95% confidence interval.