CTCF-binding element regulates ESC differentiation via orchestrating long-range chromatin interaction between enhancers and HoxA

Abstract

Proper expression of Homeobox A cluster genes (HoxA) is essential for embryonic stem cell (ESC) differentiation and individual development. However, mechanisms controlling precise spatiotemporal expression of HoxA during early ESC differentiation remain poorly understood. Herein, we identified a functional CTCF-binding element (CBE⁺⁴⁷) closest to the 3'-end of HoxA within the same topologically associated domain (TAD) in ESC. CRISPR-Cas9-mediated deletion of CBE⁺⁴⁷ significantly upregulated HoxA expression and enhanced early ESC differentiation induced by retinoic acid (RA) relative to wild-type cells. Mechanistic analysis by chromosome conformation capture assay (Capture-C) revealed that CBE⁺⁴⁷ deletion decreased interactions between adjacent enhancers, enabling formation of a relatively loose enhancer–enhancer interaction complex (EEIC), which overall increased interactions between that EEIC and central regions of HoxA chromatin. These findings indicate that CBE⁺⁴⁷ organizes chromatin interactions between its adjacent enhancers and HoxA. Furthermore, deletion of those adjacent enhancers synergistically inhibited HoxA activation, suggesting that these enhancers serve as an EEIC required for RA-induced HoxA activation. Collectively, these results provide new insight into RA-induced HoxA expression during early ESC differentiation, also highlight precise regulatory roles of the CTCF-binding element in orchestrating high-order chromatin structure.

Embryonic stem cell (ESC), which are derived from the inner cell mass (ICM) of blastocysts, exhibit unlimited self-renewal and differentiation into the three embryonic germ layers (1, 2, 3). Therefore, ESC are of great utility to the fields of regenerative medicine and tissue engineering (4, 5). ESC pluripotency requires expression of master control transcription factors such as Oct4/Nanog/Sox2 (6, 7), while expression of HoxA/Gata4/Pax6 genes is essential for ESC differentiation (8, 9, 10). Recently, studies indicate that high-order chromatin structures play a critical role in regulating gene expression and maintaining ESC status (11, 12). Numerous studies report that topologically associated domains (TADs) are the basic unit of nuclear chromatin (13, 14). Functional DNA elements, such as enhancers, within the same TAD are more likely to promote target gene expression via direct long-range chromatin interactions (15, 16). Long-range chromatin interactions require proteins to mediate those interactions, among them CTCF. Using auxin-induced degradation of CTCF protein, Nora et al. (17) found that CTCF loss promoted disappearance of TADs, revealing CTCT as essential for maintaining TAD formation. CTCF forms a chromatin loop through directly binding CTCF-binding elements (CBEs) enriched at TAD boundaries and relies on binding polarity to maintain TAD formation (18, 19, 20). Although many report that CBEs at TAD boundaries play key regulatory roles in biological development (21), analysis of CBE activity inside a TAD is lacking.

Retinoic acid (RA) rapidly induces ESC differentiation (22, 23) and activates HoxA expression (24, 25). Recent studies show that CBE activity functions in tumorigenesis (26, 27), immunogenesis (28, 29), and stem cell development (21, 30) by controlling target gene expression through organizing high-order chromatin structures. We and others have found that multiple enhancers are required for RA-induced HoxA expression and early ESC differentiation via long-range chromatin interactions with HoxA genes (31, 32, 33, 34). Previous studies have also reported many regulatory patterns for HoxA activation during ESC differentiation; for example, several CBEs (also known as CBS or CTCF-binding sites) within the HoxA locus can organize HoxA chromatin structures and play significant regulatory roles in ESC differentiation and leukemogenesis (26, 35, 36). However, it is unknown whether CBEs function to regulate HoxA expression by orchestrating enhancer chromatin structures during RA-induced early ESC differentiation.

Here, we identified a novel functional CBE⁺⁴⁷ required for RA to activate HoxA expression and promote early ESC differentiation. Relevant to mechanism, we report that CBE⁺⁴⁷ regulates HoxA expression by organizing precise chromatin interactions between its adjacent enhancers and HoxA. Our study increases understanding of mechanisms underlying RA-induced HoxA activation and early ESC differentiation and highlights the unique and precise regulatory roles of CBE in high-order chromatin structure.

Results

Identification of a CTCF-binding element (CBE⁺⁴⁷) closest to the 3'-end of HoxA within the same TAD in ESC

TADs are basic structural units in high-order chromatin, and DNA functional elements such as enhancers or insulators in the same TAD likely regulate expression of neighboring genes (15, 16). To identify functional elements that regulate HoxA expression, we first analyzed Hi-C data in ESC. We found that the HoxA 5'-end was located at a TAD boundary region, while the 3'-end was within the TAD (Fig. 1A). Others had shown that CBEs located at the HoxA locus regulate HoxA expression and ESC differentiation (36, 37). Interestingly, we identified many significant CBEs in the same TAD at the HoxA 3'-end (Fig. 1B), but their function was not previously characterized.

Figure 1.

Identification of a CBE (CBE⁺⁴⁷) closest to the HoxA 3'-end within the same TAD in ESC.A, Hi-C interaction map of ∼0.5 Mb region surrounding the HoxA locus in undifferentiated ESC. Data were extracted from Bonev et al., 2017. B, IGV (Integrative Genomics Viewer) screenshots showing gene tracks of CTCF, MED1, MED12, and YY1 ChIP-seq signal occupancy at Skap2 and HoxA loci in ESC. CTCF ChIP-seq shows CTCF binding throughout the locus. Multiple CTCF sites are located downstream of the HoxA locus. C and D, IGV view of selected ChIP-seq tracks at Skap2 and HoxA loci in ESC. Shown are H3K27ac, H3K4me1, H3K4me2, H3K4me3 (C), and Pol ll, ATAC-seq and DNase (D). Vertical blue line indicates the CBE⁺⁴⁷ region. Gray shadowing indicates HoxA locus. In (A), blue dotted box area shows the TAD boundary region. Other results relevant to these findings are shown in Fig. S1.

Here, we focused on a significant CBE closest to the HoxA 3'-end, ∼47 kb from HoxA and located in the second intron of Halr1; we designate that element CBE⁺⁴⁷ (Fig. 1B and Fig. S1). We observed significant binding of MED1, MED12, and YY1 upstream of CBE⁺⁴⁷ (Fig. 1B) as well as enrichment for the epigenetic modification-related markers H3K27ac and H3K4me1/2/3 (Fig. 1C). We also observed significant chromatin accessibility (based on DNase and ATAC-seq analysis) and transcriptional activity marked by PolⅡenrichment in these regions (Fig. 1D). These data indicate that CBE⁺⁴⁷ is located between HoxA and potential regulatory elements. Others had shown that CBEs regulate target gene expression by organizing chromatin interactions between a functional element and a target gene (38, 39). Therefore, we postulated that CBE⁺⁴⁷ may play a regulatory role.

Gene expression analysis in cultured WT and CBE⁺⁴⁷-deleted cells grown in self-renewal conditions

To assess CBE⁺⁴⁷ function, we employed CRISPR-Cas9 knockout methodology (31). Specifically, we designed two sgRNAs, upstream and downstream of CBE⁺⁴⁷ (Fig. 2A) and after Cas9 cleavage and DNA recombination, obtained two homozygous CBE⁺⁴⁷ knockout lines (CBE⁺⁴⁷-KO). PCR of genomic DNA using specific primers and Sanger sequencing confirmed CBE⁺⁴⁷ deletion (Fig. 2B). AP staining of CBE⁺⁴⁷ knockout lines revealed no significant differences in cell morphology relative to WT cells in self-renewing culture conditions (Fig. 2C). Moreover, expression of Halr1 and Sakp2 mRNAs was comparable in WT and KO lines, based on qRT-PCR analysis (Fig. 2D). However, most of the 3'-end genes of HoxA, including Hoxa2-a6, were significantly upregulated, while the TAD boundary gene Hoxa9 was significantly downregulated in CBE⁺⁴⁷-KO relative to WT cells (Fig. 2E). These data show that CBE⁺⁴⁷ deletion significantly changes HoxA expression and indicate that even in an undifferentiated ESC state, CBE⁺⁴⁷ is required to maintain proper HoxA expression.

Figure 2.

Gene expression analysis in WT and CBE⁺⁴⁷-deleted cells grown in self-renewal culture conditions.A, schematic showing CRISPR/Cas9-mediated deletion of CBE⁺⁴⁷ (blue shadowing) using two sgRNAs. Indicated primers shown in blue were used to distinguish CBE⁺⁴⁷-KO (CBE⁺⁴⁷ knockout) from wild-type (WT) clones. B, validation of knockout lines by genomic DNA PCR and sequencing. Left, images of agarose gel showing PCR analysis of representative clones. Right, DNA sequencing of CBE⁺⁴⁷-KO cell clones #1 and #2 using the indicated primer. C, appearance of WT and CBE⁺⁴⁷-KO cells from lines 1 and #2 stained with alkaline phosphatase (AP). D–K, data derived from qRT-PCR analysis of WT and CBE⁺⁴⁷-KO cells showing indicated transcripts in undifferentiated ESC. Expression levels in WT ESC were set to 1 and shown as the blue dotted line. Data are represented as means ± SD. Significance was based on Student’s t-test (two-tailed; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). In all analyses, n = 4 (two CBE⁺⁴⁷-KO lines (#1 and #2) and two biological replicates per line). In (B): M, DNA marker. In (C): scale bar, 50 μm.

Given the critical function of HoxA in ESC differentiation (8, 32, 40) and the observation that CBE⁺⁴⁷ deletion significantly increased expression of a subset of HoxA genes, we asked whether ESC pluripotency was perturbed by CBE⁺⁴⁷ deletion by evaluating expression of pluripotency- and differentiation-related genes by qRT-PCR (Fig. 2, F–K). We observed no significant differences in expression of pluripotency-regulated genes in CBE⁺⁴⁷-KO versus WT cells, except for slight upregulation of Sox2. However, mesoderm and trophectoderm genes were significantly upregulated in CBE⁺⁴⁷-KO cells. Expression of other germ layer genes such as Otx2 (epiblast), Sox17 (endoderm), and Sox11 (ectoderm) also significantly increased in CBE⁺⁴⁷-KO cells. These results indicate that CBE⁺⁴⁷ deletion does not regulate ESC pluripotency but may alter its differentiation capacity.

CBE⁺⁴⁷ deletion significantly potentiates RA-induced HoxA expression and promotes abnormal differentiation of early ESC

During early RA-induced differentiation of ESC, HoxA genes become significantly activated (25, 32) (Fig. S2). Therefore, we asked whether CBE⁺⁴⁷ regulates those activities. To do so, we treated CBE⁺⁴⁷-KO and WT ESC with RA for 12, 24, and 48 h (Fig. S3A) and then compared HoxA expression levels by qRT-PCR. HoxA 3'-end genes (Hoxa2-a6) showed significantly higher expression in CBE⁺⁴⁷-KO cells, as did the TAD boundary genes Hoxa7 and Hoxa9. However, expression of HoxA 5'-end genes (Hoxa10, Hoxa11, and Hoxa13) was comparable in both genotypes (Fig. 3, A–C). Also, antisense long noncoding RNAs (such as Hoxaas3) at the HoxA locus were significantly upregulated in CBE⁺⁴⁷-KO relative to WT cells (Fig. S3, B–D). These data suggest overall that CBE⁺⁴⁷ functions to restrict RA-induced overexpression of HoxA.

Figure 3.

CBE⁺⁴⁷deletion potentiates RA-induced HoxA expression and perturbs ESC differentiation.A–C, qRT-PCR of WT and CBE⁺⁴⁷-KO cells showing HoxA gene expression in ESC after RA treatment for 12 h (A), 24 h (B), and 48 h (C). D–F, qRT-PCR of WT and CBE⁺⁴⁷-KO ESC showing transcripts of indicated genes in ESC after RA treatment for 12 h (D), 24 h (E), and 48 h (F). In A–F, expression levels in WT ESC were set to 1 and shown as the blue dotted line. Data are represented as means ± SD. Significance is based on Student’s t-test (two-tailed; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). In all analyses, n = 4 (two CBE⁺⁴⁷-KO lines (#1 and #2) and two biological replicates per line). Other results relevant to this figure are shown in Figs. S2 and S3.

To assess whether CBE⁺⁴⁷ knockout alters ESC differentiation, we used qRT-PCR to analyze potential changes in pluripotency- or differentiation-related master control genes in WT and CBE⁺⁴⁷-KO ESC under RA induction conditions. Expression of pluripotency genes (Nanog and Klf4) did not change significantly at 12 h after ESC differentiation but significantly decreased in CBE⁺⁴⁷-KO relative to WT cells by 24 and 48 h. The epiblast master control gene Otx2, endoderm master control gene Gata4, and ectoderm master control gene Pax6 showed significantly higher expression levels in CBE⁺⁴⁷-KO versus WT cells, while the mesoderm master gene T and the trophectoderm master gene Hand1 were significantly downregulated (Fig. 3, D and E). These findings suggest overall that CBE⁺⁴⁷ is essential to maintain RA-induced HoxA expression and proper early ESC differentiation.

Transcriptome analysis of WT and CBE^+47kb-deleted cells during RA-induced early ESC differentiation

To further assess effects of CBE⁺⁴⁷ loss on early (24 h) RA-induced ESC differentiation, we performed transcriptome analysis using RNA-seq after RA-induced differentiation of WT and CBE⁺⁴⁷-KO cells. Relative to WT cells, statistical analysis of CBE⁺⁴⁷-KO cells revealed 869 genes upregulated and 360 downregulated (fold-change ≥ 2, p < 0.05) (Fig. 4A). Among them, the most significantly upregulated genes were at the 3’-end of HoxA (Fig. 4B). Because these genes located in the same TAD with CBE⁺⁴⁷, we hypothesized that CBE+47 deletion significantly promotes gene expression restricted to that TAD. These findings reveal that CBE⁺⁴⁷ serves as a cis-regulatory element playing a local regulatory role (Fig. S4). Moreover, relative to WT cells, CBE⁺⁴⁷ KO cells expressed significantly higher levels of HoxB/C/D cluster genes (Fig. S5A). Genes relevant to RA signaling, such as Crabp2, Cyp26a1, and Stra6, were also significantly overexpressed in CBE⁺⁴⁷ KO cells, as were the differentiation-related genes Cbx4 and Foxa1 (Fig. S5B). By contrast, pluripotency genes such as Myc, Lin28a, and Etv4 were significantly suppressed in CBE⁺⁴⁷ KO relative to WT cells (Fig. S5C). These data suggest that in normal cells CBE⁺⁴⁷ functions to enable proper differentiation.

Figure 4.

Transcriptome analysis in WT and CBE⁺⁴⁷-deleted cells following RA-induced early ESC differentiation.A, heatmap depicting gene expression changes in WT and CBE⁺⁴⁷-KO cells treated 24 h with RA (Fold-change ≥ 2; p < 0.05, as determined by DESeq2). B, RNA-seq results shown heatmap of HoxA. C and D, GO-BP analyses indicating differentially expressed genes. E, GSEA plots for the top three KEGG signaling pathways showing log2 fold change for the entire transcriptome. NES, Normalized Enrichment Score. Other results related to this figure are shown in Figs. S4–S7.

To assess mechanisms underlying these activities, we performed bioinformatic analysis of up- and downregulated genes in CBE⁺⁴⁷ KO cells (Fig. S6). GO analysis indicated that biological processes (BP) relevant to upregulated genes include gland development and positive regulation of nervous system development (Fig. 4C). GO-BPs of downregulated genes mainly involved embryonic placenta development, positive regulation of vasoconstriction, and epithelial cell differentiation (Fig. 4D). KEGG pathway enrichment analysis revealed upregulated genes to be enriched in pathways including cGMP-PKG signaling, neuroactive ligand–receptor interaction, axon guidance, stem cell pluripotency, Hippo signaling, cAMP signaling, Wnt and Hedgehog signaling (Fig. S7A). Enrichment results relevant to downregulated genes revealed pathways related to proteoglycans in cancer, transcriptional dysregulation in cancer, arachidonic acid metabolism, HTLV-I infection, Hippo signaling, and thyroid hormone signaling (Fig. S7A). These findings, which were further validated by Gene Set Enrichment Analysis (GSEA) (Fig. 4E and Fig. S7B), suggest that CBE⁺⁴⁷ deletion promotes abnormal gene expression and perturbs proper RA-induced early ESC differentiation.

CBE⁺⁴⁷ is required for proper chromatin interactions between adjacent enhancers and HoxA

Previous studies report that CTCF regulates target gene expression by controlling chromatin interaction between enhancers and target genes and that this activity occurs within a TAD (41). Studies by us and others have identified three enhancers (E1, E2, and E3) at the HoxA 3'-end (Fig. 5A) required for HoxA expression following RA-induced ESC differentiation (31, 32, 33, 34). Among them, E1 and E3 reportedly mediate significant long-range interactions with HoxA chromatin. Remarkably, CBE⁺⁴⁷ is located between those three enhancers and HoxA (Fig. 5A). Therefore, we speculate that the CBE⁺⁴⁷ may be involved in organizing the interactions between the three enhancers and HoxA chromatin.

Figure 5.

CBE⁺⁴⁷is required for proper chromatin interactions between HoxA and adjacent enhancers.A, IGV (Integrative Genomics Viewer) screenshots showing gene tracks of H3K27ac (+RA, 24 h) and CTCF ChIP-seq signal occupancy at Skap2 and HoxAloci in WT ESC. B–D, chromatin interaction profiles using indicated enhancers as anchors in WT or CBE⁺⁴⁷-KO cells after 24 h of RA treatment. Numbers on Y-axis denote interaction frequency. Light green, light purple, light red, and light gray shadows mark E1, E2, E3, and HoxA regions, respectively. E, quantitative results showing interaction differences seen in WT and CBE⁺⁴⁷-KO cells.

Next we used Capture-C methodology, which is widely used to define chromatin interactions between enhancers and target genes (31, 42, 43, 44, 45, 46), to define interactions between E1, E2, and E3 and HoxA chromatin in WT and CBE⁺⁴⁷-KO cells. For this analysis we used cells induced 24 h with RA, with enhancers E1, E2, and E3 as respective anchor regions (Fig. 5, B–D). In WT cells, we observed significant interactions between all three enhancers and HoxA chromatin, with E1 interacting most strongly (Fig. 5B). However, in CBE⁺⁴⁷ KO cells, interactions between the E1 and HoxA chromatin were significantly reduced relative to WT cells. However, E2 and E3 interaction with HoxA chromatin increased significantly in CBE⁺⁴⁷ KO relative to WT cells, and the intensity of chromatin interactions shifted from the 3'-end to the 5'-end of HoxA (Fig. 5, C and D). In addition, the results also show that the interaction between enhancers decreases significantly except for the slight increase of E1 and E2 (Fig. 5E). These results indicate overall that CBE⁺⁴⁷ is essential to maintain these proper chromatin interactions during RA-induced early ESC differentiation.

Multiple enhancer deletions show synergistic effect on RA-induced HoxA expression

Next, we used CRISPR-Cas9 to delete enhancers individually or in groups to establish multiple homozygous enhancer knockout lines (namely, ΔE1, ΔE2, ΔE3, ΔE1/2, ΔE1/2/3) (Fig. S8). We then treated each line 24 h with RA and performed qRT-PCR to determine HoxA expression (Fig. 6, A–E). ΔE1 and ΔE2 cells showed similar phenotypes in terms of HoxA regulation, that is, both 3’-end (Hoxa1) and central (Hoxa5-a10) genes were significantly suppressed relative to WT cells (Fig. 6, A and B). Also relative to WT cells, in ΔE3 cells, Hoxa1-a7 were significantly inhibited, indicating that these enhancers have specific and unique regulatory effects (Fig. 6C). In ΔE1/2 and ΔE1/2/3 cells, HoxA expression was significantly reduced compared with ΔE1, ΔE2, or ΔE3 (Fig. 6, D and E and Fig. S9, A and B). In addition, considering the closest distance between E1 and CBE⁺⁴⁷, we also compared the expression differences of HoxA cluster gene between them. The results showed that expressions of HoxA 3'-end gene ware significantly different (Fig. S9C), which indicated that the regulation of HoxA by CBE⁺⁴⁷ and E1 was opposite. Over all, these findings indicate that interactions among E1, E2, and E3 enhancers with HoxA chromatin synergize to promote RA-induced HoxA activation.

Figure 6.

Deletions of multiple enhancers synergize to alter RA-induced HoxA gene expression.A–E, transcripts of HoxA genes were measured by qRT–PCR and normalized to WT levels in ΔE1 (A), ΔE2 (B), ΔE3 (C), ΔE1/2 (D), andΔE1/2/3 (E) cells following 24 h of RA treatment. Expression levels in WT ESC were set to 1 and shown as the blue dotted line. Data are represented as means ± SD. Significance is based on Student’s t-test (two-tailed; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.) In A–C and E, n = 4, including two enhancer knockout lines and two biological replicates per line. In D, n = 2, including one enhancer knockout line and two biological replicates per line. Other results relevant to these are shown in Figs. S8 and S9.

Discussion

Precise expression of HoxA genes is critical for ESC differentiation. Although others have shown that CBEs play an important role in ESC differentiation by regulating high-order chromatin structure (36, 47), identification of a functional CBE and mechanisms regulating HoxA expression remain largely unknown. Here, we identified a new functional CBE⁺⁴⁷ and propose a novel model based on our results. Specifically, in WT cells (Fig. 7, left), three enhancers near CBE⁺⁴⁷ act as an EEIC to interact with the 3'-end of HoxA chromatin to maintain normal HoxA expression and promote correct early differentiation of ESC induced by RA. In CBE⁺⁴⁷-KO cells (Fig. 7, right), enhancer interactions decrease, resulting in a relatively loose EEIC. Furthermore, EEIC interactions with central HoxA chromatin regions increase following CBE⁺⁴⁷ loss, promoting HoxA overexpression. These results indicate that a functional CBE⁺⁴⁷ maintains normal HoxA expression during ESC differentiation by organizing precise interactions between adjacent enhancers and HoxA chromatin. Our findings highlight indispensable fine-regulatory roles of CBEs as functional elements in high-order chromatin structure and reveal a direct regulatory effect of the CBE⁺⁴⁷ on RA-induced HoxA expression and early ESC differentiation.

Figure 7.

Schematic showing how CBE⁺⁴⁷regulates RA-induced HoxA expression and early ESC differentiation by orchestrating long-range chromatin interactions between HoxA and adjacent enhancers.Left, model showing CBE⁺⁴⁷ activity in WT cells. Three enhancers near CBE⁺⁴⁷ act as an EEIC and interact with the 3'-end of HoxA chromatin to maintain normal HoxA normal expression, allowing proper ESC differentiation after RA treatment. Right, in CBE⁺⁴⁷-KO cells interactions between enhancers decrease, resulting in a relatively loose EEIC and allowing increased interactions between that EEIC and the central region of HoxA chromatin. As a result, HoxA is overexpressed and early ESC differentiation proceeds abnormally.

Previously, Ferraiuolo et al. (48) showed that HoxA exhibits a special chromatin structure required for its expression and that CTCF regulates formation of that structure and thus gene expression. Recent studies focusing on CBEs within the HoxA locus, including the TAD boundary, have found that CBEs organize HoxA chromatin structure, maintain normal HoxA expression, and regulate ESC differentiation (36). Narendra et al. (36) showed that CBE5/6 knockout significantly promoted HoxA expression, altered HoxA chromatin structure, and enhanced ESC neural differentiation. Studies in other cells, such as AML cells, also reveal that CBEs located at the HoxA locus significantly regulate HoxA expression and affect tumorigenesis (26, 35). Here, we find that besides several CBEs in the HoxA locus, there are several CBEs downstream of HoxA within the same TAD (Fig. 1B). Among them, CBE⁺⁴⁷ KO significantly enhanced RA-induced HoxA expression and promoted early ESC differentiation, strongly suggesting that CBE⁺⁴⁷ is a functional CBE. However, other CBEs have not yet been evaluated and their potential activity requires future analysis. Recently, others developed a CRISPR-Cas9-based genetic screen method to detect functional CBEs in breast cancer and AML (49, 50). Comparable approaches may be taken in the future to screen for functional CBEs that regulate ESC differentiation.

Our transcriptome analysis showed that after CBE⁺⁴⁷ deletion, genes in the same TAD with CBE⁺⁴⁷ are significantly upregulated relative to genes in nearby TADs (Fig. S4), suggesting that CBE⁺⁴⁷ is a cis-regulatory element that plays a local regulatory role. We also observed upregulation of RA signaling-related genes (Crabp2, Cyp26a1, Stra6) when CBE⁺⁴⁷ was deleted, while pluripotency-related genes (Myc and Lin28a) were significantly downregulated (Fig. S5), supporting the idea that CBE⁺⁴⁷ loss promotes RA-induced early ESC differentiation. Recent studies show that Lin28a inhibits HoxA expression to maintain limb development (51), while De Kumar et al. (40) found that HOXA1 binds to the Myc gene near enhancers, suggesting positive feedback between HoxA and Myc or Lin28a in early ESC differentiation induced by RA. These possibilities require further analysis.

RA significantly induced HoxA expression during early ESC differentiation (Fig. S2). Others have shown that at least three enhancers (E1, E2, and E3) are needed for HoxA expression in this context (31, 32, 33, 34). CBE⁺⁴⁷ lies between these three enhancers and HoxA (Fig. 5A). As an insulator, CBE may block interaction between enhancers and target genes, exerting an inhibitory effect (52, 53, 54). Our analysis first reveals that the three enhancers interact significantly as an EEIC. Second, we find that the EEIC interacts significantly with HoxA. Third, the interaction of these enhancers with HoxA also has its own pattern. When CBE⁺⁴⁷ is deleted, enhancer interactions significantly decrease, resulting in formation of a relatively loose EEIC. We conclude that CBE⁺⁴⁷ is required to organize enhancer interaction. Following CBE⁺⁴⁷ loss, interaction of E1 becomes more concentrated in the middle of the HoxA locus, and interaction of E2 and E3 with HoxA moves from the 3'-end to middle regions. Furthermore, CBE⁺⁴⁷ deletion significantly promotes higher expression of Hoxa2-a9, which corresponds to changes in chromatin interactions between enhancers and HoxA. Thus CBE⁺⁴⁷ not only organizes enhancer interactions but orchestrates interactions between enhancers and HoxA, ensuring correct HoxA expression. Although the CBE⁺⁴⁷ organizes high-order chromatin structures and previous studies show that the binding direction of CTCF alters formation of a CTCF-mediated chromatin loop (19), other CBEs may pair with CBE⁺⁴⁷ to form chromatin loops, a possibility that requires further analysis.

Multiple enhancers reportedly maintain RA-induced HoxA expression through long-range chromatin interactions with HoxA (31, 32, 33, 34). Here, we show that knockout of multiple enhancers synergizes to alter HoxA expression (Fig. 6), a finding partially consistent with a previous report (33). Although E1/E2/E3 enhancers are known to have regulatory effects, we also observed other enhancers located in the Skap2 region in the same TAD with HoxA (Fig. 5A), although their functional effects are unknown. Furthermore, although CBE⁺⁴⁷ organizes interactions between E1, E2, and E3 enhancers, we did not observe significant CTCF-binding peaks in the E2 region, suggesting that other factors may regulate enhancer interactions. A recent study using the ChIA-PET method (Chromatin Interaction Analysis by Paired-End Tag sequencing) from Wang et al. (55) shows that RARα (Retinoic Acid Receptor alpha) mediates chromatin interactions in AML cells. Activation of target genes by RA signaling requires RAR/RXR (retinoic acid receptor/retinoid X receptor) signaling (56, 57). Remarkably, we also found significant RAR/RXR binding peaks in E1, E2, and HoxA regions over the course of RA-induced ESC differentiation (data not shown) (58), suggesting that RAR/RXR regulates these interactions, a possibility that needs further verification. We also found that chromatin interactions between a single enhancer and HoxA are different. Thus, mechanisms underlying maintenance of these different interactions require further analysis.

In summary, we identified a new functional CBE⁺⁴⁷ that regulates RA-induced HoxA expression and early ESC differentiation via orchestrating long-range chromatin interactions between its adjacent enhancers and HoxA. These findings further our understanding of intrinsic mechanisms governing RA-induced early ESC differentiation and highlight specific regulatory roles of CBE in high-order chromatin structure.

Experimental procedures

Embryonic stem cell culture

Mouse ESC E14 were cultured as previously described with some modifications (31). ESC were grown in culture dishes coated with 0.1% gelatin (Sigma, Lot # SLBQ9498V) in Dulbecco's Modified Eagle's Medium (DMEM, Gibco, Lot # 8119213) supplemented with 15% fetal bovine serum (FBS, AusGeneX, Lot # FBS00717-1), 1x nonessential amino acids (100x, Gibco, Lot # 2027433), 1x L-Glutamate (100x, Gibco, Lot # 2085472), 1x Penicillin-streptomycin (100x, Gibco, Lot # 2029632), 50 μM β-mercaptoethanol (Sigma, Cas # 60-24-2), 10 ng/ml LIF (ESGRO, Lot # ESG1107), 1 μM PD0325901 (a MEK inhibitor, MedChemExpress, Lot # HY-10254), and 3 μM CHIR99021 (a GSK inhibitor, MedChemExpress, Lot # HY-10182). The medium was replaced every 1 to 2 days. All cells were maintained at 37 °C in a 5% CO₂ incubator.

Retinoic acid (RA)-induced early ESC differentiation

Cells were gently washed with 1x phosphate buffer saline solution (PBS), dissociated, and plated at an appropriate density on gelatin-coated plates in LIF/2i withdrawal medium supplemented with 2 μm RA (Solarbio, Lot # 1108F031). The culture medium was replaced at the given point in time.

RNA extraction, reverse transcription, and quantitative real-time PCR (qRT-PCR)

Total RNA was extracted from differentiated or undifferentiated ESC using TRIzol Reagent (Life Technologies, Lot # 213504). cDNA synthesis was performed using a PrimerScriptTM RT reagent Kit with gDNA Eraser (TaKaRa, Lot # AJ51485A) according to the manufacturer’s instructions. PCR reactions were performed using HieffTMqPCR SYBR Green Master Mix (YEASEN, Lot # H28360) and a BioRad CFX Connect Real-Time system. PCR cycling conditions were as previously reported: 95 °C for 5 min, 40 cycles of 95 °C for 15 s, 60 °C for 15 s, and 72 °C for 30 s. We then constructed a melting curve of amplified DNA (31). Target gene values were normalized to Gapdh expression and the experimental control using ΔΔCt methods (59). Primer sequences used in this study are shown in Table S1.

CRISPR/Cas9-mediated CBE⁺⁴⁷ and enhancers deletion in ESC

The CRISPR/Cas9 system was used following published protocols (31, 60, 61). Briefly, target-specific guide RNAs (sgRNAs) were designed using an online tool (http://chopchop.cbu.uib.no/). sgRNAs of the appropriate site and score were selected. sgRNA sequences are shown in Table S2. For CBE⁺⁴⁷ and enhancer knockout, sgRNAs were cloned into a Cas9-puro vector using the Bsmb1 site. ESC were transfected with two sgRNA plasmids using Lipofectamine 3000 (Life Technologies, Lot # 2125386), and 24 h later, cells were treated with 5 μM puromycin (MCE, Lot # 64358) for 24 h and then cultured in medium without puromycin for another 5 ∼ 7 days. Individual colonies were picked and validated by genomic DNA PCR and subsequent Sanger DNA sequencing. PCR primers used for genotyping are listed in Table S3.

Alkaline phosphatase (AP) staining of WT and CBE⁺⁴⁷-deleted cells

ESC were plated at low density in 12-well plates coated with gelatin for 4 days and then washed twice with PBS and incubated with reagents from the AP Staining kit (Cat # AP100R-1, System Biosciences) following the manufacturer’s instructions. Digital images were taken using an Olympus Inverted Fluorescence Microscope.

Enhancer capture-C and data analysis

Capture-C probes were designed using an online tool within three enhancer regions (http://apps.molbiol.ox.ac.uk/CaptureC/cgi-bin/CapSequm.cgi): the E1 bait (Mouse, mm10, chr6: 52012160–52015360), the E2 bait (Mouse, mm10, chr6:52075358–52076639), and the E3 bait (Mouse, mm10, chr6: 52101884–52104217). Probe sequences are listed in Table S4.

Capture-C libraries were prepared as previously described with minor modifications (31, 42, 43). Briefly, RA-induced differentiated ESC (WT and CBE⁺⁴⁷-KO) were fixed with 1% (vol/vol) formaldehyde for 10 min at room temperature, quenched with 125 mM glycine in PBS, and then lysed in cold lysis buffer [10 mM Tris-HCl, pH7.5, 10 mM NaCl, 5 mM MgCl₂, 0.1 mM EGTA, 0.2% NP-40, 1× complete protease inhibitor cocktail (Roche, Lot # 3024150)]. Chromatin was digested with DpnII (New England Biolabs, Lot # 10014860) at 37 °C overnight. Fragments were then diluted and ligated with T4 DNA ligase (Takara, Lot # 1211707) at 16 °C overnight. Cross-linking was reversed by overnight incubation at 60 °C with proteinase K (Bioline, Lot # BIO-37037). Then 3C libraries were purified by phenol-chloroform followed by chloroform extraction and ethanol-precipitated at −80 °C overnight. Sequencing libraries were prepared from 10 μg of the 3C library by sonication to an average size of 200 ∼ 300 bp and indexed using NEBnext reagents (New England Biolabs, Lot # 0031607), according to the manufacturer's protocol. Enrichment of 2 μg of an indexed library incubated with 3 μM of a pool of biotinylated oligonucleotides was performed using the SeqCap EZ Hybridization reagent kit (Roche/NimbleGen, Lot # 05634261001), following the manufacturer's instructions. Two rounds of capture employing 48∼72 and 24 h hybridizations, respectively, were used. Correct library size was confirmed by agarose gel electrophoresis, and DNA concentration was determined using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Lot # 0000248352). All sequencing was performed on Hi-Seq 2500 platforms using paired 150 bp protocols (Illumina).

Capture-C data were analyzed using previously described methods (31, 42, 43). Briefly, clean paired-end reads were reconstructed into single reads using FLASH (62). After in silico DpnII digestion using the DpnII2E.pl script, reads were mapped back to the mm10 mouse genome using Bowtie1. Finally, chimeric reads containing captured reads and Capture-Reporter reads were analyzed using CCanalyser3.pl. Results were visualized using the Integrated Genome Browser (IGV) (63) and online tool (https://epgg-test.wustl.edu/browser/) (64).

RNA-seq and bioinformatics analysis

ESC were lysed with Trizol reagent (Life Technologies, Lot # 265709) and RNA was extracted based on the manufacturer's instructions. RNA was then sequenced by a company (Novogene). Clean reads were mapped to the Ensemble mm10 mouse genome using Hisat2 with default parameters. Gene reads were counted by Htseq (65). Fold changes were computed as a log2 ratio of normalized reads per gene using the DEseq2 R package (66). Gene expression with∣log2 (fold change)∣≥ 1 (p < 0.05) was considered significantly altered. Heatmaps were drawn using the heatmap.2 function. Two biological replicates were analyzed for each experimental condition.

Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) pathway analyses

GO and KEGG pathway analyses were performed using the following online tools: Metascape (http://metascape.org/gp/index.html#/main/step1) (67) and DAVID Functional Annotation Bioinformatics Microarray Analysis tool (http://david.abcc.ncifcrf.gov/) (68).

Gene set enrichment analysis (GSEA)

GSEA was carried out using the online tool (https://www.gsea-msigdb.org/gsea/index.jsp), as previously reported (69, 70). KEGG pathway-related gene sets were obtained from the KEGG database (https://www.genome.jp/kegg/) (71).

Statistical analyses

Data were analyzed by Student’s t-test (two-tailed) unless otherwise specified. Statistically significant p-values are indicated in Figures as follows: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Data availability

Data supporting this study are available from the corresponding author upon request. Raw data reported here (DNA-seq and RNA-seq) have been deposited in the NCBI Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE154495.

Published data sets were downloaded using online tools (http://cistrome.org/db/#/ and http://promoter.bx.psu.edu/hi-c/) and analyzed in this study (72, 73). These include: GSM2588420 for H3K27ac ChIP-seq analyses and GSE124306 for E3 enhancer Capture-C in mouse ESC (WT) 24 h after RA induction (31, 33); GSE96107 for Hi-C (74); GSM859491 for H3K27ac (75); GSM2630487 for H3K4me1 (76); GSM881353 for H3K4me2 (75); GSM1258237 for H3K4me3 (77); GSM699165 for CTCF (78); GSM2645432 for YY1 (79); GSM1439567 for Med1 (80); GSM560345 for Med12 (81); GSM1276711 for RNA PolⅡ (77); GSM1014154 for DNase (82); and GSM2651154 for ATAC-seq data (83) in mouse undifferentiated ESC.

Supporting information

This article contains supporting information.

Conflict of interests

The authors declare that they have no conflict of interest.

Edited by Ronald Wek