Dynamic Control of Enhancer Repertoires Drives Lineage and Stage-Specific Transcription during Hematopoiesis

Abstract

Enhancers are the primary determinants of cell identity, but the regulatory components controlling enhancer turnover during lineage commitment remain largely unknown. Here we compare the enhancer landscape, transcriptional factor occupancy, and transcriptomic changes in human fetal and adult hematopoietic stem/progenitor cells and committed erythroid progenitors. We find that enhancers are modulated pervasively and direct lineage- and stage-specific transcription. GATA2-to-GATA1 switch is prevalent at dynamic enhancers and drives erythroid enhancer commissioning. Examination of lineage-specific enhancers identifies transcription factors and their combinatorial patterns in enhancer turnover. Importantly, by CRISPR/Cas9-mediated genomic editing, we uncover functional hierarchy of constituent enhancers within the SLC25A37 super-enhancer. Despite indistinguishable chromatin features, we reveal through genomic editing the functional diversity of several GATA switch enhancers in which enhancers with opposing functions cooperate to coordinate transcription. Thus, genome-wide enhancer profiling coupled with in situ enhancer editing provide critical insights into the functional complexity of enhancers during development.

Figure 1. Comparative Analysis of Enhancer Repertoires during Human Erythropoiesis

(A) Ex vivo erythroid differentiation of fetal liver (FL) or adult bone marrow (BM) CD34+ HSPCs. Cells at matched stages of differentiation (HSPC: F0 and A0; ProE: F5 and A5) were collected for transcriptomic profiling and ChIP-seq analyses. (B) Identification of lineage or developmental stage-specific enhancers. Venn diagram shows the overlap between HSPC and ProE, or fetal and adult enhancers. The numbers of lost, shared, or gained enhancers in each comparison are shown. (C) ChIP-seq density heatmaps for H3K4me1 and H3K27ac within lost, gained and shared enhancers in each comparison. (D) Representative lineage-specific enhancers are shown. The putative active enhancers are depicted by shaded lines. (E) Representative developmental stage-specific enhancers are shown. (F) Enrichment of lineage-defining TFs in lineage-specific enhancers. The top enriched TF motifs in lost or gained enhancers between HSPCs and ProEs at fetal or adult stage are shown. P values were calculated using the hypergeometric test. (G) Enrichment of distinct coregulators in stage-specific enhancers. The top enriched TF motifs in lost or gained enhancers between fetal and adult HSPCs (F0 versus A0) or ProEs (F5 versus A5) are shown. See also Figure S1.

Figure 2. Enhancers Control Lineage and Stage-Specific Transcription

(A) Enhancers positively associate with lineage or stage-specific gene expression changes. The enrichment significance of differentially expressed genes (columns) harboring different types of enhancers (rows) is calculated using Fisher's exact test. (B) Representative genes targeted by one, two, three, and four ‘A0-A5-Gained’ enhancers are shown, respectively. The putative enhancers are depicted by shaded lines. The mRNA expression of each gene is shown on the right. Values are shown as mean ± standard error of the mean (SEM) between replicates. (C) mRNA expression of genes harboring single versus multiple enhancers. (D) The correlation between mRNA expression and regular enhancers versus super-enhancers. (E) mRNA expression of genes harboring enhancers with varying distance.

Figure 3. CRIPSR/Cas9-Mediated Deletion Analysis of the SLC25A37 Super-Enhancer

(A) Chromatin signatures and TF occupancy within the human or mouse SLC25A37 locus in HSPC (A0) versus ProE (A5) or undifferentiated G1E versus differentiated G1ER cells are shown, respectively. The SLC25A37 constituent enhancers (E1, E2, and E3) and the proximal promoter (P) are depicted by shaded lines. Super-enhancers were called using Rose (Whyte et al., 2013) base on the H3K27ac ChIP-seq signal. The sequence conservation by PhastCons analysis is shown. (B) Schematic of CRISPR/Cas9-mediated genomic editing to dissect the SLC25A37 super-enhancer. The scheme is adapted from (Pott and Lieb, 2015). (C) Expression of Slc25a37 mRNA in unmodified (Control) and enhancer-deletion G1E/G1ER cells at various time points (0 to 48h) after β-estradiol treatment. The mRNA expression levels relative to GAPDH are shown. Each color circle represents an independent single-cell derived bi-allelic enhancer or promoter-deletion clone. Results are means ± standard deviation (SD) of multiple independent clones. P-value measures the statistical significance between control (C) and each experimental group. *P < 0.01; **P < 0.001. (D-G) ChIP-qPCR analysis of H3K4me1, H3K27ac, GATA1 and TAL1 in control and enhancer-deletion cells. Primers against Slc25a37 promoter (P) and each constituent enhancer (E1, E2, and E3) are used. Oct4 promoter is analyzed as a negative control. Results are means ± SD of multiple independent clones. *P < 0.01. (H) Schematic of the hierarchical structure of the SLC25A37 super-enhancer. See also Figures S2 and S3.

Figure 4. Combinatorial Control of Enhancers by Transcription Factors

(A) Enrichment of functionally relevant TF motifs. The numbers of identified motif-matched loci and the percent of motifs overlapped with ChIP-seq peaks are shown. (B) Hierarchical clustering of 86 TF motifs significantly enriched in the lineage and/or stage-specific enhancers. Heatmap shows the enrichment score of TF motifs. (C) Hierarchical clustering of ChIP-seq occupancy of the indicated TFs within enhancers. Heatmap shows the enrichment score based on the overlap significance calculated by Fisher's exact test. (D) Combinatorial TF modules within enhancers during erythropoiesis. TF modules were defined by hierarchical clustering of enriched TF motifs based on their enrichment score across all 12 enhancer types. Five distinct TF motif modules are shown for (1) GATA1-TAL1, (2) E2FSP1, (3) AP1, (4) RUNX1-FLI1-PU.1-ETS, and (5) NFE2-MYB. The color code indicates the Pearson Correlation Coefficient (PCC) of the enrichment scores. See also Figure S4.

Figure 5. TF Combinatorial Regulatory Networks within Enhancers

(A) Schematic of the construction of enhancer-mediated TF combinational regulatory networks. (B) Representative TF combinational regulatory networks in adult HSPCs (A0, left) and ProEs (A5, right). In the network, the nodes represent TF with ChIP-Seq data (red) or motif information (grey). The size of node represents the enrichment score. The color of edges represents different types of combination (red: TF-to-TF; green: TF-to-Motif; grey: Motif-to-Motif). The width of edges represents the combinatorial score. (C) Enrichment of known protein-protein interactions (PPIs) in network-predicted TF interactions. The “Network” in x-axis represents TF interactions predicted in A0 or A5 network, whereas “Random” represents all possible interactions among TFs shown in the network. (D) Enrichment of hematologic and erythroid trait-associated SNPs in enhancers containing network-predicted TF interactions. All, all A0-A5-Gained and A0-A5-Lost enhancers; TF, enhancers occupied by at least one TF based on ChIP-seq analysis; TF Pairs, enhancers occupied by TF pairs; TF Pairs in Network, enhancers occupied by TF pairs identified in the A0 or A5 network. P-value measures the statistical significance between all enhancers and each enhancer group. *P < 0.05; **P < 0.001. (E) A0-A5-Gained enhancers are highly enriched for erythroid SNPs. The y-axis shows the percentage of enhancers containing erythroid trait-associated SNPs. The x-axis is the same as in panel D. (F) ChIP-seq density heatmaps are shown for H3K4me1, H3K27ac, GATA1, GATA2, FLI1, PU.1, TAL1, KLF1, p300 and BRG1 within the indicated lost or gained enhancers. The percentage of GATA2-to-GATA1 switch enhancers is shown on the right. See also Figures S5 and S6.

Figure 6. GATA Switch Controls Enhancer Activities during Erythroid Specification

(A) Distribution of GATA switch and non-switch enhancers within the lost, gained or shared enhancers in adult HSPCs (A0) and ProEs (A5). The numbers of GATA1, GATA2, or both (GATA switch) occupied enhancers are shown. (B) Differential enriched TF motifs in GATA2-Only, GATA switch, or GATA1-Only enhancers. (C) mRNA expression of genes associated with GATA2-Only, GATA switch, or GATA1-Only enhancers during erythroid differentiation of adult HSPCs (A0) to ProEs (A5). The normalized expression level in particular cell types (z-score) were calculated using barcode (McCall et al., 2014). (D) GREAT analysis (McLean et al., 2010) of GATA2-Only, GATA switch, or GATA1-Only enhancers. Top enriched gene ontology (GO) biological processes, mouse phenotypes and pathways are shown, respectively. See also Figure S6.

Figure 7. Genome Editing of GATA Switch-Mediated Enhancer Turnover

(A) Chromatin signatures and TF occupancy within the human or mouse PINX1 locus. The putative GATA2-Only (E1) and GATA switch (E2) enhancers are shown. (B) Schematic of CRISPR/Cas9-mediated deletion of Pinx1 or Mrto4 enhancers. (C) Expression of Pinx1 mRNA in unmodified (Control) and enhancer-deletion cells at various time points (0 to 48h) after β-estradiol treatment. Each color circle represents an independent bi-allelic enhancer-deletion clone. Results are means ± SD of multiple independent clones. *P < 0.01; n.s. not significant. (D) Chromatin signatures and TF occupancy within the mouse Mrto4 locus. The putative GATA2-Only (E1) and GATA switch (E2) enhancers are shown. (E) Expression of Mrto4 mRNA in control and enhancer-deletion cells. (F) Chromatin signatures and TF occupancy within the human or mouse GATA2 locus. The putative GATA switch (E1, E2, and E3) enhancers are shown. (G) Schematic of CRISPR/Cas9-mediated deletion of Gata2 enhancers. (H) Expression of Gata2 mRNA in control and enhancer-deletion cells. See also Figure S7.