Enhancers predominantly regulate gene expression during differentiation via transcription initiation

Abstract

Gene transcription occurs via a cycle of linked events, including initiation, promoter-proximal pausing, and elongation of RNA polymerase II (Pol II). A key question is how transcriptional enhancers influence these events to control gene expression. Here, we present an approach that evaluates the level and change in promoter-proximal transcription (initiation and pausing) in the context of differential gene expression, genome-wide. This combinatorial approach shows that in primary cells, control of gene expression during differentiation is achieved predominantly via changes in transcription initiation rather than via release of Pol II pausing. Using genetically engineered mouse models, deleted for functionally validated enhancers of the α- and β-globin loci, we confirm that these elements regulate Pol II recruitment and/or initiation to modulate gene expression. Together, our data show that gene expression during differentiation is regulated predominantly at the level of initiation and that enhancers are key effectors of this process.

Keywords: Poll II recruitment; enhancers; gene regulation; promoter proximal pausing; transcription.

Figure 1. scaRNA-seq maps transcription initiation and pausing at a single molecule level in vivo.

(A) Total counts of scaRNA molecules in promoter proximal regions (0-300bp, relative to observed TSS) stand proxy for levels of initiation and pausing. Nascent gene expression was measured using reads mapping to introns from RNA-seq (excluding the first 300bp of a given gene) (la Manno et al., 2018). (B) Overview of data analysis: scaRNA counts (scatter plot with green background) and intronRNA counts (scatter plot with orange background) are analyzed to find significant differential changes in expression (red dots) out of all genes (grey dots). Datasets are compared in situations when gene expression changes (0h and 24h of erythropoiesis). The change in scaRNA versus the change in intronRNA can be plotted as a scatter plot where each quadrant identifies a distinct class of regulation; 1 (pausing gain), 2 (initiation gain), 3 (initiation loss), 4 (pausing loss). (C) scaRNA-seq in K562 cells to validate the assay with MYC (left) and HSPA1A (aka. Hsp70) (right). The figure shows (from top to bottom) annotated UCSC gene isoforms, human expressed sequence tags (ESTs), polyA+ RNA-seq in K562, the density of reconstructed RNA molecules and distributions of their RNA 5’ and 3’ ends. RNA 5’ and 3’ distributions indicate focused initiation of transcription and promoter proximal pausing downstream of the site of initiation (+20-60nt). All data are displayed as raw (unnormalized) counts. scaRNA-seq data were derived from 5x10⁶ K562 cells. polyA+ RNA-seq was downloaded from Encode and an isogenic duplicate merged prior to display. Data are displayed raw (without normalization). Human ESTs were obtained as a UCSC genome browser track. See Figure S1 and S2

Figure 2. Annotated TSS positions are systematically skewed upstream of their in vivo locations.

(A) Plots of the distance between the position of observed TSSs (called from scaRNA-seq data) and annotated TSSs associated with UCSC, Gencode or Refseq gene annotations. (B) A meta-analysis of scaRNA-seq data around observed TSS positions and associated annotated TSSs from three different annotations (Gencode, Refseq and UCSC). scaRNA observed and Refseq meta profiles show punctate initiation around the TSS, Gencode and UCSC show dispersed. 3’ RNA ends indicative of Pol II pausing peak between +50-100 bps and their distribution extends to around +150bps relative to the TSS. Antisense transcription and Pol II pausing are visible in all profiles, scaling on the Y axis gives the impression that antisense transcription occurs to a lower degree in Refseq and scaRNA observed than Gencode or UCSC. Data displayed as raw (unnormalized) coverage of RNA ends (reads per bp per TSS). (C) Heatmaps of RNA 5’ ends (purple), RNA 3’ ends (orange). 0 bp position is highlighted with purple dotted line. Coverage is per bp from the observed TSS on the sense strand. Coverage is log (coverage +1) and capped at 55 reads per bp per TSS (log4) to aid in visualization of a wide dynamic range of coverage values as a heatmap. Distance to TSS is shown in bp.

Figure 3. Transcription initiation is the predominant point of regulation in the transcription cycle.

(A) A scatter plot of log2 fold change in scaRNA molecules versus intronRNA for all detectably expressed genes (grey dots, n=2713). Genes which show significantly differential expression (red dots, n=327). The data show a positive correlation (R² of 0.62), with concordant changes in scaRNA and intronRNA indicating that regulation occurs predominantly via initiation. Each quadrant is marked with the correlation coefficient and percentages of genes (out of a total of 2713), all genes (grey) and significantly differentially expressed genes (red). (B) Tfrc exemplifies increases in scaRNA across the region of promoter proximal transcription and intronRNA from the remainder of the gene, between 0h and 24h. This indicates an increase in initiation rather than a decrease in pausing (see Figure 1B). Set inset right for a zoomed view of scaRNA-seq data over promoter (grey box), with Y-axis scaled for visualization. (C) Npm1 exemplifies decreases in intronic RNA and scaRNA at 0h and 24h across the region of promoter proximal transcription. The level of scaRNA (5’ and 3’ end) transcripts decrease in parallel with intronic RNA at 24h indicating a decrease in initiation rather than increase in pausing which would be associated with an increase in scaRNA (see Figure 1B). Set inset right for a zoomed view of scaRNA-seq data over promoter (grey box), with Y-axis scaled for visualization. All data were derived from biological triplicates at 0h and 24h of erythroid differentiation. scaRNA tracks were scaled to the sample with the lowest number of millions of mapped reads and a biological triplicate merged for visualization. IntronRNA tracks were normalized by reads per million and a biological triplicate merged for visualization. Differential count analyses were performed on raw unnormalized data using DEseq2 (Anders and Huber, 2010). See Figure S3 and 4.

Figure 4. Enhancers regulate target gene expression via transcription initiation.

(A) The α- and β-globin loci represented to scale. Regulatory elements (enhancers) are indicated as orange boxes and the two copies of α-globin at the α-locus are highlighted in red. Two copies of β-globin at the β-globin locus are highlighted in blue. At the α-globin locus, deletion of the R1 and R2 enhancers (ΔR1R2) reduce nascent α-globin expression by 95% (Hay et al., 2016). At the β-globin locus, deletion of the HS2 and HS3 (ΔHS2HS3) reduce nascent β-globin expression by 70% (Bender et al., 2012). (B) The α-globin locus in wildtype and ΔR1R2 cells, with the two α-globin genes (Hba-a1 and Hba-a2). scaRNA-seq and mNET-seq signals across the gene are both reduced showing that in ΔR1R2 there is a loss of transcription initiation. Two zoomed in views, highlight each gene’s promoter proximal region, with scaRNA-seq and mNET-seq. mNET-seq data represents the distribution of 3’ most base of each read (representing Pol II active site) with a 2nt window, to smooth data for visualization. scaRNA-seq data on zoomed view uses a scaled Y-axis (red values) to aid in visualization of the profile of Pol II pausing (3’ scaRNA ends). (C) The β-globin locus in wildtype and ΔHS2HS3 cells, with the two β-globin genes (Beta-S and Beta-T). scaRNA-seq and mNET-seq signals across the gene are both reduced showing that in the ΔR1R2 model there is a loss of transcription initiation. Beta-T appears more affected than Beta-S. Two zoomed in views highlight each gene’s promoter proximal region. scaRNA-seq and mNET-seq. mNET-seq data represents the distribution 3’ most base of each read (representing Pol II active site) with a 2nt window, to smooth data for visualization. scaRNA-seq data on zoomed view uses a scaled Y-axis (red values) to aid in visualization of the profile of Pol II pausing (3’ scaRNA ends). scaRNA-seq and mNET-seq tracks were scaled to the sample with the lowest number of millions of mapped reads and a biological triplicate merged for visualization. mNET-seq data were normalized by merging a replicate (n=2 for each genotype) and subsampling to the data set with the lowest total read count and visualization. See Figure S4 and 5.

Figure 5. Promoter hypersensitivity and 3D chromatin structure are unperturbed by enhancer deletion.

(D) α-globin locus with two copies of α-globin (Hba-a1 and Hba-a2 – grey boxes) and regulatory elements (R1-R4 and Rm) highlighted as orange boxes. Deleted enhancers (R1 and R2) are shown in red text and deletions are visible in the ATAC-seq data as lack of coverage compared to wildtype. NG-Capture-C data show chromatin interactions from the viewpoint of the α-globin promoters (which appear as gaps in the Capture C track) interactions between the remaining enhancers and the genes persist in the ΔR1R2 deletion suggesting chromatin hub formation or maintenance is not affected by their deletion. Below, a zoomed in view of the α-genes shows that the promoters remain hypersensitive in the absence of the R1 and R2 enhancers. ATAC-seq signal in the 3’ UTR of the α-globin genes is also decreased in the ΔR1R2 mouse. (E) β-globin locus with two copies of β-globin genes (Beta-S and Beta-T marked as grey boxes) and regulatory elements (R1-R6) highlighted as orange boxes. Deleted enhancers (HS2 and HS3) are shown in red text and deletions are visible in the ATAC-seq data as lack of coverage compared to wildtype. NG-Capture-C data show chromatin interactions from the viewpoint of the β-globin gene promoters (which appear as gaps in the Capture C track). Interactions between the enhancers and the genes persist in the ΔHS2HS3 deletion suggesting chromatin hub formation or maintenance is not affected by their deletion. A known structural polymorphism in the locus specific to this mouse strain is shown as a brown box and results in no NG-Capture-C coverage for the ΔHS2HS3 over this region but does not affect interpretation of the genes or enhancers. Zoomed view of the genes shown highlighting that promoters remain hypersensitive in the absence of R2 and R3 enhancers. All ATAC-seq data performed in triplicate were merged and normalized by reads per million. NG-Capture-C was also performed in triplicate and tracks show the mean interaction profile. NG-Capture C and ΔR1R2 ATAC-seq data for the α-globin locus are from published work (Hay et al., 2016). See Figure S5.

Figure 6. Enhancers stimulate Pol II recruitment to target genes

(F) Pol II ChIP-seq in wildtype and ΔR1R2 shows a total loss of Pol II across the α-globin genes (Hba-a1 and Hba-a2) consistent with defects in initiation not Pol II pause release. (G) Pol II ChIP-seq in Wildtype and ΔHS2HS3 shows a total loss of Pol II across the β-globin genes (Beta-S and Beta-T) consistent with a defect in initiation not Pol II pause release All ChIP-seq experiments were performed in biological triplicate and each replicate normalized by millions of mapped reads under Pol II peaks before being merged for visualization.