Silencing of BCL11A by disrupting enhancer-dependent epigenetic insulation

Key Points

• •BCL11A, a key therapeutic target for hemoglobinopathies, is transcriptionally regulated by enhancer-dependent epigenetic insulation.
• •Disrupting epigenetic insulation silences BCL11A and reactivates HbF in erythroid cells, serving as a targetable vulnerability.

Visual Abstract

Abstract

The transcription factor BCL11A is a genetically and clinically validated regulator of the fetal-to-adult hemoglobin switch in human erythroid cells. CRISPR editing of an intronic enhancer within the BCL11A gene reactivates fetal hemoglobin (HbF) in adult erythroid cells, serving as the first CRISPR-based therapy for β-hemoglobinopathies. However, the molecular basis for the remarkable efficacy of CRISPR-mediated enhancer ablation remains elusive. Here, we describe a new genome architecture, an enhancer-dependent chromatin rosette, that is essential for epigenetic insulation and the developmentally regulated, hematopoietic lineage–specific expression of BCL11A. CRISPR-mediated disruption of the BCL11A erythroid enhancer impairs transcription of enhancer-driven RNAs and NIPBL-dependent cohesin loading, leading to the destabilization of the rosette structure, loss of chromatin insulation, and epigenetic silencing of BCL11A. Moreover, targeted depletion of enhancer RNAs using antisense oligonucleotides silences BCL11A by disrupting epigenetic insulation, causing HbF reactivation in adult erythroid cells. These findings uncover an essential role for enhancer-driven epigenetic insulation in transcriptional control, presenting a new strategy for the therapeutic targeting of BCL11A.

The deceptively named transcription factor B-cell lymphoma/leukemia 11A (BCL11A) is a central regulator of the fetal-to-adult hemoglobin switch in erythroid cells and a therapeutic target in hemoglobinopathies. Wang et al elucidate the underlying mechanism by which CRISPR-based enhancer editing silences BCL11A, and they show that it is regulated by enhancer-dependent epigenetic insulation. Disrupting epigenetic insulation silences BCL11A and reactivates fetal hemoglobin in erythroid cells, revealing a targetable vulnerability.

Introduction

The developmental transition from human fetal hemoglobin (HbF, α2γ2) to adult hemoglobin (HbA, α2β2), known as the fetal-to-adult hemoglobin switch, is genetically controlled during ontogeny. Elevated γ-globin (HBG1 and HBG2) expression can ameliorate the severity of major β-hemoglobinopathies, including sickle cell disease (SCD) and β-thalassemia, caused by mutations or deletions in the adult β-globin (HBB) gene. Identifying molecular targets and developing broadly applicable strategies to reactivate HbF remains a longstanding goal for treating these disorders.

The transcription factor BCL11A is a genetically and clinically validated regulator of the fetal-to-adult hemoglobin switch in erythroid cells. Common genetic variants, which are identified through genome-wide association analysis, within erythroid-specific enhancers of BCL11A are associated with elevated HbF in both the general population and patients with hemoglobinopathies.^1,2 BCL11A directly represses γ-globin transcription in adult erythroid cells^3,4 and its inactivation alone is sufficient to reactivate HbF to therapeutic levels, ameliorating disease phenotypes in humanized SCD mouse models.⁵ CRISPR/Cas9-mediated editing using a single-guide RNA (sgRNA) targeting the +58 BCL11A enhancer, located 58 kb downstream of its transcriptional start site (TSS), markedly suppresses BCL11A expression.^{6, 7, 8} This strategy has since been translated into clinical trials,⁹ yielding transformative clinical outcomes,^{10, 11, 12} including the elimination of vaso-occlusive crises in 97% of patients with SCD¹⁰ and resolution of transfusion dependence in 91% of patients with β-thalassemia.¹¹ These successes culminated in the recent regulatory approval of exagamglogene autotemcel (Casgevy), the first CRISPR-based gene therapy for SCD and β-thalassemia.

Despite its therapeutic benefits, the precise mechanism by which CRISPR editing of the +58 enhancer with a single sgRNA silences BCL11A remains unclear. Moreover, the high costs, limited availability, and potential risks of current gene therapies make them inaccessible to most patients. Hence, alternative, scalable, and more affordable approaches to reactivate HbF levels are needed to alleviate the overall disease burden. In studying the developmental regulation of BCL11A during hematopoiesis, we uncovered new insights into its transcriptional control through enhancer-mediated epigenetic insulation. The +58 enhancer is essential for orchestrating chromatin rosette architecture and establishing the epigenetic insulation necessary for BCL11A expression. These new aspects of BCL11A regulation not only explain the efficacy of CRISPR editing at the BCL11A enhancer in hemoglobin disorders but also reveal new vulnerabilities for improved therapeutic targeting.

Materials and methods

Cells and CRISPR/Cas9 editing

HUDEP-2 cells were cultured in StemSpan SFEM (Stemcell Technologies, catalog no. 09655) with 0.1mM dexamethasone (Sigma, catalog no. D4902), 100 ng/mL of human stem cell factor (PeproTech, catalog no. AF-300-07), 3 IU/mL of erythropoietin (Epogen; Amgen), 1 μg/mL of doxycycline (Sigma, catalog no. D5207), and 1% penicillin-streptomycin (Gibco, catalog no. 15140122).^13,14 CD34⁺ hematopoietic stem/progenitor cells (HSPCs) from healthy donors or patients with SCD were cultured and induced for erythroid differentiation.¹⁵ In phase 1 (days 0-7), cells were maintained at 0.1 × 10⁶ to 0.4 × 10⁶/mL in Iscove modified Dulbecco medium with 2% AB plasma (SeraCare, catalog no. 1810-0013), 3% AB serum (GeminiBio, catalog no. 50-753-3010), 1% penicillin-streptomycin, 3 IU/mL of heparin, 200 μg/mL of holo-transferrin (Sigma, catalog no. T0665), 1 IU of erythropoietin, 10 ng/mL of stem cell factor, and 1 ng/mL of interleukin-3 (PeproTech, catalog no. 200-03). In phase 2 (days 8-12), cells were maintained at 0.2 × 10⁶ to 0.8 × 10⁶/mL without interleukin-3. HPSCs were electroporated with Cas9-sgRNA ribonucleoprotein (RNP) (40 pmol of 3× NLS-Cas9 and 120 pmol of sgRNA) using a Lonza 4D-Nucleofector. Genomic DNA was isolated for amplicon sequencing and analyzed using CRIS.py.¹⁶

CUT&RUN, CUT&Tag, and ATAC-seq

Cleavage under targets and release using nuclease (CUT&RUN) and cleavage under targets and tagmentation (CUT&Tag) were performed with modifications as previously described.^17,18 Dead cells were removed using the EasySep Dead Cell Removal Kit (Stemcell Technologies, catalog no. 17899). Reads were processed using Cutadapt (version 4.4 [v4.4])¹⁹ and aligned to hg38 using bowtie2 (v2.4.2).²⁰ Peaks were called using MACS2 (v2.1.1).²¹ CTCF motif analysis was performed using GimmeMotifs (v0.18.0)²² with MA0139.1 from JASPAR.²³ BigWig files were visualized in the WashU Epigenome Browser. Assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) was performed using Zymo-Seq ATAC Library Kit (Zymo, catalog no. D5458). Reads were aligned to hg38 using bowtie (v1.1.2).²⁴ Coverage tracks were generated by converting BED files to BigWig.

Nascent RNA sequencing

Cells were treated with 0.5μM 5-ph-IAA (5-phenyl-1H-indole-3-acetic acid) and incubated with 5 μL of 200mM EU (5-ethynyl uridine) from the Click-iT Nascent RNA Capture Kit (Thermo Fisher Scientific, catalog no. C10365). RNA was subjected to a click reaction with biotin azide and CuSO₄, followed by rRNA depletion and library preparation using NEBNext Ultra II RNA Library Prep Kit (New England Biolabs; catalog no. E7770). Reads were trimmed using Cutadapt (v4.4)¹⁹ and aligned to hg38 using STAR (v2.5.2b).²⁵

PRO-seq and PRO-cap

Precision run-on sequencing (PRO-seq) was performed with modifications as previously described.²⁶ Cells were permeabilized and subject to nuclear run-on with 40μM biotin-11-CTP (PerkinElmer, catalog no. 3354), biotin-11-UTP (PerkinElmer, catalog no. 2716), ATP (Thermo Fisher Scientific, catalog no. 18330019), and GTP (Thermo Fisher Scientific, catalog no. 18332015). RNA was extracted, fragmented, and ligated to a 3′ adapter. Biotinylated RNA was captured by Dynabeads streptavidin C1 (Thermo Fisher Scientific, 65002), treated with T4 polynucleotide kinase (PNK), and ligated to a 5′ adapter. Reads were aligned to hg38 using BWA-MEM (v0.7.16a). Strand-specific BigWig files (1 bp bin, counts per million normalized) were generated using SAMtools (v1.17)²⁷ and deepTools (v3.5.4).²⁸ Precision run-on and capped RNA sequencing (PRO-cap) was performed with optimizations.²⁶ Cells were processed by nuclear run-on with 40μM biotin-11-CTP, biotin-11-UTP, biotin-11-ATP (Revvity, catalog no. NEL544001EA), and biotin-11-GTP (Jena Bioscience, catalog no. NU-971-BIOX-S). TSS were identified from 5′ read ends, strand-separated with SAMtools (v1.17),²⁷ converted to BEDgraph, and analyzed using TSScall.²⁹

Capture-Pore-C (CPC) and data analysis

Cells were crosslinked and digested with NlaIII, followed by ligation and digestion by proteinase K and pronase. DNA was hybridized with biotin-labeled probes using the KAPA HyperCapture Reagent Kit (Takara, catalog no. 09075828001) and size-selected. Libraries were prepared using the Ligation Sequencing Kit V14 (Oxford Nanopore Technologies, catalog no. SQK-LSK114) and sequenced using PromethION. POD5 files were base-called using Dorado (v0.5.1), trimmed using Cutadapt (v4.4),¹⁹ mapped to hg38, and analyzed with the Nextflow Pore-C pipeline (v23.10.0).³⁰ Sample correlations were assessed using HiCExplorer (v3.7.2).³¹ Pairwise interactions were identified using CHiCANE (v0.1.8, P < .05).³² Multiway interactions were defined by overlapping distinct probe regions using pairtools (v0.3.0).³³ Virtual 4C was performed using Cooler (v0.8.11),³⁴ and read counts were normalized as (reads × 10⁹)/(length × total PETs)³⁵.

ASO

Antisense oligonucleotides (ASO) with 2′-MOE (2′-O-methoxyethyl) or locked nucleic acid modifications were designed and synthesized by Integrated DNA Technologies. Electroporation was performed using 2 μL of 100μM ASO mixed with 18 μL of P3 buffer containing 0.4 × 10⁶ HUDEP-2 cells using the Lonza 4D-Nucleofector P3 kit.

Results

BCL11A is regulated by lineage-specific enhancers during hematopoiesis

To understand how BCL11A is transcriptionally regulated during hematopoiesis, we analyzed its epigenetic state and messenger RNA (mRNA) expression in human and mouse hematopoietic lineages using publicly available epigenomic data sets (supplemental Table 1, available on the Blood website). The lineage-specific and developmental stage–specific expression of BCL11A is associated with putative enhancer elements characterized by chromatin accessibility (ATAC-seq) and activating histone modifications, including H3K4me1 and H3K27ac (Figure 1A; supplemental Figures 1 and 2). This analysis further identified the +55, +58, and +62 enhancers of BCL11A (hereafter referred to as E+55, E+58, and E+62)⁶ as erythroid lineage–specific and developmental stage–specific, with ATAC-seq signals first appearing in lineage-committed colony-forming unit–erythroid cells and peaking in proerythroblasts (Figure 1A).

Figure 1.

BCL11A is developmentally regulated by lineage-specific enhancers and epigenetic silencing. (A) Chromatin accessibility profiles based on publicly available ATAC-seq data sets across human hematopoietic lineages are shown for the BCL11A locus at chr2:60,319,000-60,600,000 (hg38). The dashed lines indicate the BCL11A TSS and erythroid-specific enhancers. The right panel displays BCL11A mRNA expression levels (TPM), with each dot representing an independent replicate. (B) Chromatin immunoprecipitation sequencing (ChIP-seq) profiles of H3K27ac and H3K27me3 in primary human hematopoietic stem cell (HSCs) (blue), erythroblasts (red), and T lymphocytes (black). H3K27me3 deposition is observed at the BCL11A promoter and gene body in T lymphocytes. (C) ChIP-seq profiles of H3K27ac and H3K27me3 in primary mouse HSCs, erythroblasts, and T lymphocytes, revealing conserved patterns of Bcl11a repression in T lymphocytes via H3K27me3-mediated silencing. The dashed line indicates the Bcl11a TSS. The sources of the public data sets are listed in supplemental Table 1. (D) Schematic representation of BCL11A gene expression and enhancer activity in HSCs, erythroblasts, and T lymphocytes, illustrating a shared CTCF binding profile but distinct epigenetic landscapes that correlate with lineage-specific gene expression. Erythroid-specific enhancers (E+55, E+58, and E+62) are highlighted in red, whereas the regions of active and repressive chromatin are indicated in green and gray, respectively. BasoE, basophilic erythroblast; BFU-E, burst-forming unit–erythroid; CFU-E, colony-forming unit–erythroid; CLP, common lymphoid progenitor; CMP, common myeloid progenitor; DC, dendritic cell; GMP, granulocyte-monocyte progenitor; MEP, megakaryocyte-erythroid progenitor; MPP, multipotent progenitor; MK, megakaryocyte; NK, natural killer; orthoE, orthochromatic erythroblast; PolyE, polychromatic erythroblast; ProE, proerythroblast; TPM, transcripts per million.

Notably, despite the presence of active chromatin at the promoter and lineage-specific enhancers, the BCL11A locus is enriched with H3K27me3, a hallmark of Polycomb-mediated repressive chromatin,³⁶ in human erythroid, B-lymphoid, and myeloid lineage cells (Figure 1B; supplemental Figures 1 and 2). In BCL11A-expressing erythroid cells, a distinct H3K27me3 chromatin immunoprecipitation sequencing (ChIP-seq) peak is located at 13 to 19 kb upstream of the TSS, along with broader peaks located downstream of the BCL11A locus. However, in BCL11A-nonexpressing cell types, such as T-lymphoid cells, the H3K27me3-marked repressive chromatin extends to encompass the promoter and gene body (Figure 1B; supplemental Figure 2B). The BCL11A locus contains 7 distinct CTCF-binding sites (CBS); however, the binding of CTCF remains largely consistent between BCL11A-expressing and BCL11A-nonexpressing cell types (supplemental Figure 3). Similar patterns are observed in mouse hematopoietic lineages (Figure 1C; supplemental Figure 4). These findings suggest that although BCL11A is regulated by lineage-specific enhancers in BCL11A-expressing hematopoietic lineages, it undergoes developmentally regulated and evolutionarily conserved epigenetic silencing in BCL11A-negative lineages (Figure 1D).

BCL11A enhancers are required for maintaining epigenetic insulation

To investigate how CRISPR editing of the complex, multicomponent BCL11A enhancer with an sgRNA (sg1617)^{9, 10, 11, 12} effectively disrupts its expression in relevant cellular contexts, we employed the immortalized human erythroid progenitor cell line HUDEP-2 and primary erythroblasts derived from bone marrow CD34⁺ HSPCs.^3,13 Both cell models exhibit active BCL11A transcription driven by the intronic erythroid–specific enhancers, an adult-stage phenotype characterized by high β-globin, low γ-globin expression, and H3K27me3-marked repressive chromatin flanking the BCL11A locus (Figure 2A).

Figure 2.

BCL11A enhancers are required for epigenetic insulation. (A) Epigenetic landscape of the BCL11A locus in human primary erythroblasts and HUDEP-2 cells, showing chromatin accessibility (ATAC-seq) and the occupancy of active histone marks (H3K27ac and H3K4me3), the repressive histone mark (H3K27me3), and CTCF. The dashed lines indicate the BCL11A TSS and erythroid-specific enhancers. (B) BCL11A and HBG mRNA expression levels, along with the percentage of HBG and the levels of HbF by high-performance liquid chromatography and HbF⁺ cells, in WT and CRISPR-edited single-cell HUDEP-2 clones. Each dot represents an independent single-cell–derived clone. Data are presented as mean ± standard deviation (SD) and analyzed by 1-way analysis of variance (ANOVA). ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001. (C) Chromatin accessibility, CUT&Tag (H3K27ac), and CUT&RUN (H3K27me3) profiles in WT HUDEP-2 cells and cells with BCL11A exon 2 mutation, enhancer KO, or sg1617-mediated enhancer editing (sg1617). The right panel shows a representative genomic region with no detectable chromatin state changes. (D) Quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis of CD34⁺ HSPCs from healthy donors and patients with SCD edited with Cas9-RNP targeting E+58 (sg1617) or a control locus (sgAAVS). Samples were collected at differentiation day 8. Results are shown as mean ± SD (n ≥ 3 replicates) and analyzed by 1-way ANOVA. ∗∗∗P < .001; ∗∗∗∗P < .0001. (E) Genome browser tracks of ATAC-seq, H3K27ac, and H3K27me3 in primary erythroblasts derived from CD34⁺ HSPCs from healthy donors or patients with SCD, showing H3K27me3 deposition at the BCL11A promoter and gene body following Cas9-RNP-mediated E+58 enhancer editing (sg1617) or a control locus (sgAAVS), as indicated by the log₂ fold change of H3K27me3 CUT&RUN signals. CD34⁺ HSPCs were induced for erythroid differentiation for 8 days (healthy donors) or 12 days (SCD). (F) A model depicting the critical role of the erythroid enhancer E+58 in BCL11A gene regulation and the maintenance of epigenetic insulation. Enh KO, enhancer knockout; Exon2 Mut, BCL11A exon 2 mutation; NS, not significant; sgAAVS, sgRNA targeting the adenovirus-associated virus integration site.

Using CRISPR/Cas9 and sg1617 RNPs, the same approach employed by exagamglogene autotemcel,⁹ we generated 4 independent single-cell–derived HUDEP-2 lines with homozygous 1-bp insertion (+1 bp) and 13- or 15-bp deletions (−13 or −15 bp) at the targeted E+58 enhancer, respectively (supplemental Figure 5A). For controls, we generated 3 single-cell–derived HUDEP-2 lines carrying homozygous deletions of a 10-kb genomic region containing all 3 enhancers (supplemental Figure 5B). In sg1617-edited or enhancer knockout (KO) HUDEP-2 clones, we confirmed significant BCL11A downregulation and γ-globin (HBG1/2, hereafter HBG) reactivation by quantitative reverse transcription polymerase chain reaction, resulting in marked increases in HbF levels as measured by high-performance liquid chromatography and HbF⁺ F cells (Figure 2B; supplemental Figure 5C-E). Strikingly, H3K27me3-marked repressive chromatin extended to the BCL11A promoter and gene body in both sg1617-edited and enhancer KO HUDEP-2 cells (Figure 2C; supplemental Figure 6A). This was accompanied by decreased or loss of chromatin accessibility and diminished H3K27ac at the BCL11A promoter and enhancers. The marked increase in H3K27me3 was specific to the BCL11A locus, with no consistent changes at other genomic loci or across the genome (Figure 2C; supplemental Figure 6A-B).

To assess whether the observed effects on H3K27me3 resulted from enhancer disruption, we generated 2 single-cell–derived HUDEP-2 lines harboring frameshift mutations in exon2 of BCL11A without affecting its intronic enhancers (hereafter Exon2 Mut). These mutations abolished BCL11A protein expression and reactivated HBG, but they did not affect BCL11A mRNA levels or enhancer activity (Figure 2B-C; supplemental Figure 7A-B). Unlike sg1617-edited and enhancer KO HUDEP-2 cells, BCL11A Exon2 Mut HUDEP-2 cells exhibited no detectable changes in H3K27me3 levels or distribution (Figure 2C), suggesting that enhancer inactivation drives the epigenetic silencing of BCL11A in erythroid cells.

To confirm these findings in primary erythroblasts, we introduced RNPs containing CRISPR/Cas9 with sg1617 or control adenovirus-associated virus integration site–targeting sgRNA (sgAAVS) into bone marrow CD34⁺ HSPCs from healthy donors or patients with SCD, followed by 8 days (healthy) or 12 days (SCD) of ex vivo erythroid differentiation.^8,37 This approach achieved INDEL (insertion-deletion) frequencies of 96.5% and 96.0% at the sg1617-targeted E+58 enhancer in healthy and SCD HSPCs (supplemental Figure 8A-B), respectively, resulting in significant BCL11A downregulation and HBG upregulation in primary erythroblasts (Figure 2D). Notably, E+58 enhancer editing in healthy or SCD HSPCs also markedly increased H3K27me3 at the BCL11A promoter and gene body in primary erythroblasts, with no detectable effects at other genomic loci or across the genome (Figure 2E; supplemental Figure 9A-C). These findings demonstrate the essential role of the E+58 enhancer in maintaining chromatin insulation, which facilitates transcription while preventing H3K27me3-mediated epigenetic silencing of BCL11A in erythroid cells (Figure 2F).

Given the role of CTCF-mediated boundary function in regulating epigenetic spreading,³⁸ we investigated the contribution of CBS to BCL11A regulation in erythroid cells. Using CRISPR-mediated KO, we generated single-cell–derived HUDEP-2 lines with homozygous deletions of each CTCF-binding site at the BCL11A locus (supplemental Figure 10A). The KO of individual CBS had no effect on BCL11A mRNA expression (supplemental Figure 10B-C). The combined KO of 2 CBS, C−7 and C−12, located between the H3K27ac-marked active chromatin and H3K27me3-marked repressive chromatin upstream of the BCL11A promoter also did not affect BCL11A expression or H3K27me3 distribution (supplemental Figure 10D-F). Similarly, the combined KO of 2 CBS downstream of the BCL11A locus (C+114 and C+271) or 3 sites (C−12, C−7, and C+114) had no effect on BCL11A expression (supplemental Figure 10D-F). Furthermore, acute CTCF protein degradation via the auxin-inducible degron system³⁹ caused no detectable changes in BCL11A or HBG expression in proliferating HUDEP-2 cells (supplemental Figure 11A-D). In contrast, combining enhancer editing with sg1617 and CTCF degradation resulted in BCL11A downregulation and HBG reactivation (supplemental Figure 11E).

BCL11A locus is organized into enhancer-dependent chromatin structures

We hypothesized that E+58 orchestrates the formation of high-order genome architecture essential for chromatin insulation, thereby preventing BCL11A from epigenetic silencing. To test this hypothesis, we first analyzed locus-specific 3D genome conformation using short-read sequencing-based chromatin conformation capture (3C) assays in erythroid cells. Hi-C, which provides genome-wide chromatin interaction maps,⁴⁰ revealed that the topologically associated domain (TAD) and sub-TAD structures encompassing the BCL11A locus remain largely unchanged across BCL11A-expressing HSPCs, primary erythroblasts, and B-lymphoid cells (GM12878), as well as BCL11A-negative T-lymphoid cells (supplemental Figure 12). To achieve higher resolution, we performed Micro Capture-C⁴¹ in HUDEP-2 targeting 7 CBS at the BCL11A locus (supplemental Figure 13A) and identified pairwise interactions among discrete CTCF sites, the BCL11A promoter, and erythroid enhancers (supplemental Figure 13B). However, due to the use of small DNA fragments and short-read sequencing, Micro Capture-C did not capture multiway interactions, which are essential for resolving more complex high-order chromatin structures.

To this end, we developed a new method based on long-read Nanopore sequencing, which we named Capture-Pore-C (CPC) (Figure 3A; supplemental Methods). The initial steps resemble those of conventional 3C-based assays, involving chromatin crosslinking, in situ restriction digestion, proximity ligation, and de-crosslinking. However, instead of fragmenting the chromatin, we isolated large chromatin concatemers (∼10 kb) containing ligated DNA fragments that represent chromatin interactions. Biotin-labeled probes targeting specific regions of interest were used for 2 rounds of hybridization capture and size selection, followed by library preparation and sequencing on the Nanopore PromethION platform. The key advantages of CPC are its ability to capture not only pairwise interactions between different regions but also bona fide multiway interactions involving multiple regions simultaneously, providing a more comprehensive view of chromatin architecture (Figure 3A).

Figure 3.

BCL11A locus is organized into enhancer-dependent chromatin structures. (A) Schematic representation of the CPC workflow. Cells were crosslinked with formaldehyde, followed by in situ digestion with a restriction enzyme and chromatin fragment ligation. The chromatin was then de-crosslinked, and proteins and peptides were removed. Long DNA concatemers underwent 2 rounds of hybridization capture using biotinylated probes, followed by PCR amplification and size selection. Libraries were prepared and sequenced on the Oxford Nanopore PromethION platform. CPC data were analyzed to identify pairwise and multiway chromatin interactions. (B) Pairwise chromatin interactions at the BCL11A locus in WT, enhancer KO, and sg1617-edited HUDEP-2 cells. The upper panel shows genome browser tracks for ATAC-seq, H3K27ac, H3K27me3, and CTCF binding at the BCL11A locus, with the dashed lines indicating CBS serving as CPC baits. The bottom panel displays the significant chromatin interactions (P < .05) identified using CHiCANE³² for all the CTCF and promoter baits within chr2:60,319,000-60,600,000. Specifically, the first panel shows the combined interactions from all the baits, whereas the subsequent panels display the interactions for each bait. Within each panel, the virtual 4C track anchored at the bait region and the interaction arcs are shown, with a color scale indicating the interaction frequency. (C) CPC analysis of the pairwise interactions (P < .05) in WT and enhancer-edited (sg1617) cells, with enhancers E+55, E+58, and E+62 serving as baits (dashed lines). The results show a reduction in enhancer-mediated pairwise interactions across all 3 enhancers following editing. (D) A model illustrating chromatin interactions at the BCL11A locus in WT and enhancer-edited cells. CPC analysis revealed both E-P interactions and CTCF-mediated chromatin loops, both of which were weakened upon enhancer editing, leading to altered chromatin organization and epigenetic silencing of BCL11A. E-P, enhancer-promoter.

Using CBS as bait regions, we identified pairwise interactions between each CTCF site and other genomic regions. Notably, BCL11A enhancer KO or editing with sg1617 decreased CTCF-mediated interactions (Figure 3B; supplemental Figure 14A-B). When using BCL11A enhancers as baits, we detected pairwise interactions mediated by each enhancer (Figure 3C; supplemental Figure 14C). E+58 editing disrupted not only enhancer-promoter looping and E+58-mediated interactions but also interactions involving the neighboring E+55 and E+62 enhancers (Figure 3C). Therefore, our analysis of pairwise interactions using CPC revealed the formation of highly interactive chromatin loops between critical cis-regulatory elements (CRE) at the BCL11A locus, which are impaired by enhancer KO or editing (Figure 3D).

Enhancer-dependent chromatin configuration is required for epigenetic insulation

We next analyzed multiway interactions identified by CPC, representing them as line-dot graphs, where each line corresponds to an independent Nanopore sequencing read and each dot represents 1 captured bait region (Figures 3A and 4A). Using CBS as baits, we observed that most multiway interactions involved both upstream and downstream CTCF sites. BCL11A enhancer KO or editing with sg1617 not only decreased the number of multiway interactions but also reduced their complexity and distribution (Figure 4A-B; supplemental Figure 15A-C). Notably, the multiway interactions involving the BCL11A promoter and upstream CBS were among the most significantly affected chromatin interactions following E+58 editing (Figure 4C).

Figure 4.

Enhancer-dependent chromatin configuration is required for epigenetic insulation. (A) Multiway chromatin interactions at the BCL11A locus in WT, enhancer KO, and sg1617-edited HUDEP-2 cells. The upper tracks show chromatin accessibility (ATAC-seq), H3K27ac, H3K27me3 signals, and CTCF binding at the BCL11A locus. In the lower graphs, each line represents an independent Nanopore sequencing read, with the dots indicating bait regions used in CPC. The number of multiway interactions is reduced following enhancer KO or sg1617 editing. (B) The distribution of promoter-associated multiway interactions with individual CBS in WT, enhancer KO, and sg1617-edited HUDEP-2 cells based on the total number of paired-end tags (PETs) for each condition. The pie charts illustrate the changes in the number and distribution of promoter-CTCF contacts following enhancer KO or editing, as indicated by the sizes of the pie charts. (C) Top-ranked multiway interactions affected by enhancer editing relative to WT. Interaction frequencies were normalized to the total PETs per sample. The bar graph highlights the most significantly altered interactions, with the dots and lines below each column indicating the involved genomic regions. (D) CPC analysis of multiway interactions using BCL11A enhancers as baits. Multiway interactions associated with enhancers E+55, E+58, and E+62 are significantly diminished upon sg1617 editing, illustrating the role of enhancers in maintaining higher-order chromatin architecture. (E) Pie chart analysis of enhancer-associated multiway interactions with other enhancers, the promoter, or CBS. The upper panels illustrate global changes in enhancer connectivity, whereas the lower panels depict enhancer-CBS interactions at individual CBS across different conditions. Enhancer editing leads to a marked reduction in interaction frequency and complexity. (F) Top-ranked multiway interactions altered upon enhancer editing relative to WT. The bar graph represents the most significant changes normalized to the total PET numbers. The dots and lines below each column indicate the involved genomic regions. (G) A model depicting the 3D chromatin organization of the BCL11A locus in WT and enhancer-edited erythroid cells. In normal erythroid cells, CREs and CBS within the BCL11A locus form clusters, termed BCL11A enhancer-dependent chromatin rosette, which facilitate multiway chromatin looping, enhancer-promoter interactions, and epigenetic insulation to sustain BCL11A expression. Following enhancer editing via CRISPR-Cas9, epigenetic insulation and enhancer-promoter interactions are disrupted, accompanied by the loss of CTCF-mediated looping, leading to the epigenetic silencing of BCL11A. Enh, enhancer.

When using BCL11A enhancers as baits, we found that they form multiway interactions with either the upstream promoter and CTCF sites or the downstream CTCF sites (Figure 4D). Following CRISPR editing of E+58, we observed a near-complete loss of intraenhancer interactions among the E+55, E+58, and E+62 enhancers, particularly those involving the E+58 enhancer (Figure 4D-E; supplemental Figure 15D-E). Moreover, BCL11A enhancer-mediated multiway interactions with other critical CRE, including the BCL11A promoter and CBS, were markedly reduced or entirely lost. We also noted reorganization of multiway interactions between the BCL11A enhancers and CBS, with markedly decreased interactions involving the upstream C−28 and C−12 sites and the downstream C+114 site (Figure 4E-F).

All these results reveal lineage-specific assemblies of high-order chromatin structures that involve both pairwise and multiway interactions among key CRE, including the BCL11A promoter, enhancers, and CBS. These interactions collectively segregate the BCL11A locus into chromatin domains associated with distinct epigenetic states (Figure 4G). This structure, which we term the “enhancer-dependent chromatin rosette,” is essential for chromatin insulation, protecting the locus from epigenetic silencing. The disruption of enhancer function impairs this insulation, leading to epigenetic spreading and gene silencing.

BCL11A E+58 eRNA transcription is disrupted by CRISPR editing

Our analysis of chromatin architecture raised an important question regarding how E+58 enhancer editing via exagamglogene autotemcel affects epigenetic insulation. To address this question, we performed PRO-cap and PRO-seq, which provide base-pair resolution of nascent transcripts by mapping the location of active RNA polymerases.²⁶ In wild-type (WT) HUDEP-2 cells, the reverse strand exhibits transcription of BCL11A mRNA, whereas the forward strand shows strong signals at the enhancers with the highest signals observed at the E+58 enhancer (Figure 5A), consistent with the production of enhancer RNAs (eRNA).⁴² E+58 editing with sg1617 resulted in the near-complete loss of eRNA transcription, with no detectable effects on eRNA production at other loci (Figure 5A-B; supplemental Figure 16A-D). Notably, both PRO-cap and PRO-seq identified a strong TSS signal at chr2:60,495,297 within the E+58 enhancer, just 13 bp downstream of the sg1617-targeted DNA sequences (chr2:60,495,264-60,495,283) (Figure 5A).

Figure 5.

BCL11A eRNA production is disrupted by CRISPR editing. (A) Genome browser tracks showing PRO-cap in WT HUDEP-2 cells and PRO-seq in both WT HUDEP-2 cells and single-cell clones following sg1617-mediated enhancer editing. Nuclear run-on transcripts are shown separately for the forward and reverse strands, with BCL11A transcribed on the reverse strand and eRNA on the forward strand. The right panel shows a representative control region where transcription remains unchanged after sg1617 editing. The middle zoomed-in region highlights transcriptional signals at E+55, E+58, and E+62, whereas the bottom zoomed-in panel shows the precise location of sg1617 relative to the E+58 TSS and its directionality. (B) qRT-PCR analysis of E+58 eRNAs in HUDEP-2 cells (top) and primary erythroblasts (bottom). A significant reduction in eRNA expression is observed following enhancer disruption. Results are shown as mean ± SD (n ≥ 3 replicates) and analyzed by 1-way ANOVA or t test. ∗∗P < .01; ∗∗∗∗P < .0001. (C) Effects of sg1617 editing on GATA-binding motifs (highlighted in red) and the E+58 eRNA TSS (blue box). (D) INDEL-induced changes in eRNA transcription in primary erythroblasts derived from SCD CD34⁺ HSPCs. The line plots illustrate the predicted TSS signals for WT (blue) and mutant (orange) sequences, with the lower panels showing the differential signal (mutant WT). The gray- and blue-shaded boxes denote the locations of the CRISPR-induced INDELs and the TSS of E+58 eRNAs, respectively. (E) The effects of the top-ranked INDELs on eRNA transcription in primary erythroblasts derived from SCD CD34⁺ HSPCs. The bar graph depicts the top INDEL-induced changes in transcriptional activity, showing that the CRISPR editing–induced INDELs impair BCL11A E+58 eRNA expression. Enh KO, enhancer knockout; Fwd, forward; Mut, mutant; Rev, reverse.

CRISPR editing with sg1617 generates a diverse set of INDELs within the E+58 enhancer, including a 1-bp insertion and 13- or 15-bp deletions, at various frequencies in CD34⁺ HSPCs from healthy donors and patients with SCD (supplemental Figure 8A-B). The sg1617-mediated INDELs overlap with a GATA1 binding site predicted to be essential for E+58 activity (Figure 5C; supplemental Figure 17A)^7,8; however, the precise mechanism by which these INDELs effectively disrupt enhancer function remains unclear. To investigate this, we employed Puffin-D, a deep learning–based model,⁴³ to characterize TSS strength and location based on sequence features. Notably, Puffin-D identified a strong TSS signal from the reference genome sequence at the A nucleotide (chr2:60,495,297), which precisely aligns with the PRO-cap–identified TSS within the E+58 enhancer (Figure 5D; supplemental Figure 17B). More importantly, sg1617-induced INDELs substantially decreased the TSS signal or shifted its location (Figure 5D), with the strongest effects observed for deletions that removed the GATA1 binding site, such as 37-, 15-, and 13-bp deletions (Figure 5E). The most frequent INDEL in both healthy and SCD HSPCs, characterized by a single A nucleotide insertion into the GATA1 binding site, results in a marked reduction of the TSS signal for the E+58 eRNA (Figure 5C-E; supplemental Figure 17A-C). Moreover, CRISPR-mediated deletion or mutation of the TSS at chr2:60,495,297 reduced E+58 eRNA and BCL11A expression in HUDEP-2 cells (supplemental Figure 17D-E). Therefore, E+58 editing with sg1617 generates INDELs that impair eRNA production by disrupting its transcriptional initiation.

E+58 eRNA is essential for epigenetic insulation and BCL11A transcription

Three-dimensional genome organization requires the cohesin complex to be loaded onto chromatin by the cohesin-loading factor NIPBL, which initiates and propagates loop formation to establish chromatin architecture.⁴⁴ To investigate the roles of NIPBL and cohesin in BCL11A regulation, we performed CUT&RUN of the cohesin subunit SMC1, NIPBL, and CTCF in WT and sg1617-edited HUDEP-2 cells (Figure 6A). In WT HUDEP-2 cells, SMC1 binds strongly to the promoter-proximal CTCF sites, whereas NIPBL exhibits the highest binding affinity at the BCL11A enhancers, particularly at E+58, suggesting its role in facilitating cohesin loading at this site. E+58 editing by sg1617 abolished NIPBL binding at the BCL11A enhancers and diminished cohesin occupancy at the BCL11A promoter and enhancers, but not other genomic loci (Figure 6A; supplemental Figure 18A-B). Crosslinking immunoprecipitation (CLIP) analysis further revealed that NIPBL interacts strongly with eRNA transcribed from the E+58 and E+55 enhancers (Figure 6B). These findings suggest that NIPBL associates with eRNA in erythroid chromatin, highlighting a crucial role for eRNA in NIPBL-mediated cohesin loading at lineage-specific enhancers.

Figure 6.

BCL11A eRNAs orchestrate NIPBL and cohesin binding for epigenetic insulation. (A) Genome browser tracks showing chromatin accessibility (ATAC-seq) and the occupancy of SMC1, NIPBL, and CTCF at the BCL11A locus in WT HUDEP-2 cells and single-cell clones following sg1617-mediated enhancer editing. The dashed lines indicate the BCL11A TSS and erythroid-specific enhancers. The right panel shows a representative genomic region with no detectable changes in chromatin state or NIPBL/cohesin binding. (B) CLIP-qPCR analysis of NIPBL and eRNA interactions in HUDEP-2 cells. A schematic of the workflow is shown on the left. Results are presented for 2 independent CLIP experiments using NIPBL or IgG (negative control) antibodies. The known NIPBL-interacting small nuclear RNA, RNU-1, was analyzed as a positive control, whereas PAPOLG, a transcribed genomic region upstream of the BCL11A locus, served as a negative control. Two independent PCR primers were used to detect E+58 and E+55 eRNA transcripts. Results are mean ± SD (n ≥ 3 replicates) and analyzed by 1-way ANOVA for each primer. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001. (C) qRT-PCR and F-cell analysis of HUDEP-2 cells transduced with nontargeting control (control ASO) or E+58 eRNA-targeting ASO (ASO12 and ASO23). ASO treatment resulted in a significant reduction in E+58 eRNA and BCL11A mRNA expression, accompanied by increased HBG expression and F cells. Results are shown as mean ± SD (n ≥ 3 replicates) and analyzed by 1-way ANOVA. ∗∗P < .01; ∗∗∗P < .001. (D) E+58 eRNA-targeting ASO reduced NIPBL binding at the BCL11A enhancers and increased H3K27me3-associated repressive chromatin at the BCL11A promoter in HUDEP-2 cells, as indicated by the log₂ fold change of H3K27me3 CUT&RUN signals. The right panel shows a representative genomic region with no detectable changes in chromatin state. (E) A model illustrating the 3D chromatin organization of the BCL11A locus in WT and enhancer-edited erythroid cells. In normal erythroid cells, the E+58 enhancer maintains chromatin insulation and prevents epigenetic silencing by serving as a TSS for eRNA production and as a NIPBL/cohesin landing site for chromatin loop formation. CRISPR-mediated enhancer editing disrupts both functions, leading to destabilized chromatin architecture, impaired chromatin insulation, and epigenetic silencing of BCL11A. CLIP, crosslinking immunoprecipitation; Ctrl ASO, control antisense oligonucleotide; IgG, immunoglobulin G; NS, not significant.

To assess the functional role of E+58 eRNA in epigenetic insulation, we used ASOs to deplete E+58 eRNA in erythroid cells. Two independent ASO pools (ASO-12 and ASO-23), each containing equal molar amounts of 2 independent E+58 eRNA-targeting ASOs, achieved 62% and 63% depletion of E+58 eRNA, respectively (Figure 6C). E+58 eRNA depletion led to significant BCL11A downregulation, HBG reactivation, and increased F cells in HUDEP-2. ASO-mediated BCL11A silencing was associated with reduced NIPBL binding at the BCL11A enhancers and increased H3K27me3 deposition at the BCL11A promoter (Figure 6D), highlighting the critical role of E+58 eRNA in maintaining enhancer-mediated epigenetic insulation. Notably, E+58 eRNA depletion by ASO led to less pronounced HbF reactivation and H3K27me3 accumulation than enhancer KO or editing, likely due to incomplete eRNA depletion and/or the requirement for extended cell propagation to facilitate Polycomb-mediated silencing.

Collectively, these findings demonstrate that the E+58 enhancer maintains chromatin insulation and prevents epigenetic silencing through 2 key functions: (1) serving as a TSS for eRNA production and (2) acting as a NIPBL and cohesin landing site for chromatin loop formation (Figure 6E). CRISPR editing of the BCL11A enhancer disrupts both functions, causing destabilized chromatin configuration, impaired chromatin insulation, and epigenetic silencing of BCL11A.

Discussion

CRISPR editing with an sgRNA targeting the E+58 enhancer introduces small INDELs that effectively silence BCL11A expression, inducing therapeutic levels of HbF in adult erythroid cells.^{6, 7, 8} This discovery was unexpected because prior studies of enhancer mutations or enhancer-mutant mice typically reported only mild or negligible phenotypes, suggesting widespread enhancer redundancy during metazoan development.⁴⁵ Moreover, despite the long-standing view of noncoding sequences as “undruggable,” this work highlights tissue-specific enhancers as viable therapeutic targets.^{6, 7, 8} This approach has progressed to clinical trials,⁹ where CRISPR/Cas9 editing of autologous CD34⁺ HSPCs at the E+58 BCL11A enhancer (exagamglogene autotemcel) normalized blood counts and eliminated vaso-occlusive crises in 97% (29/30) of patients with SCD.¹⁰ Similarly, exagamglogene autotemcel treatment achieved transfusion independence in 91% (32/35) of patients with transfusion-dependent β-thalassemia.¹¹ These outcomes led to the regulatory approval of exagamglogene autotemcel, the first CRISPR-based gene therapy for treating human disease.⁴⁶ However, the mechanism by which E+58 editing effectively silences BCL11A remains poorly understood.

Here, we demonstrate that CRISPR editing-mediated BCL11A silencing occurs by disrupting enhancer-driven epigenetic insulation. To conceptualize this mechanism, we propose an “enhancer-dependent chromatin rosette” model, in which critical CRE assemble into a higher-order structure maintained by highly interactive, multiway chromatin interactions in an enhancer-dependent manner (Figure 6E). To investigate this architecture, we developed CPC, a method capable of detecting both pairwise and multiway interactions between discrete chromatin regions. We identified preferential associations among the E+58 enhancer, its neighboring E+55 and E+62 enhancers, and promoter-proximal CBS. This single-molecule sequencing strategy, which enables in situ reconstruction of multiway chromatin concatemers, provided direct evidence for the formation of an enhancer-dependent rosette that mediates chromatin insulation. These findings reveal new insights into the enhancer-dependent epigenetic architecture regulating BCL11A and provide broader implications for developmental gene regulation.

CTCF is essential for 3D genome organization and epigenetic insulation³⁸; however, our CBS KO and degron studies revealed no detectable changes in BCL11A expression in proliferating HUDEP-2 cells. One possibility is that CBS at the BCL11A locus act collectively, with the loss of some compensated by others. CTCF motifs may differ in binding affinity or resistance to degradation, as observed for other transcription factors.^47,48 Prior studies showed that acute CTCF loss collapses chromatin loops and weakens insulation, although most transcriptional programs remain unchanged.^49,50 Notably, CTCF is largely displaced from chromatin during mitosis and rapidly rebinds upon mitotic exit to reestablish loops and TAD insulation, whereas its absence at this stage disrupts higher-order architecture.^{51, 52, 53} Therefore, defining CTCF’s role in BCL11A regulation will likely require cell cycle–resolved degradation to assess effects on chromatin rosette reassembly and transcription.

Our findings uncover a link between eRNA transcription and NIPBL-cohesin–mediated chromatin loop formation at the BCL11A enhancer and suggest that factors such as GATA1 also contribute to BCL11A enhancer activity and eRNA transcription. GATA1 is a key regulator of transcriptional initiation by RNA polymerase II at gene-proximal promoters.^54,55 Deep learning–based analyses revealed that sg1617-induced INDELs, which colocalize with the GATA1 binding site, profoundly alter the strength or position of E+58 eRNA initiation. Consistently, CRISPR editing of GATA1 binding sites within BCL11A enhancers efficiently depleted BCL11A, producing the strongest HbF-inducing effects.^7,8 These findings suggest that GATA1, together with other transcriptional machinery, supports transcriptional initiation and eRNA production, which in turn facilitate NIPBL binding and cohesin recruitment to establish chromatin architecture that protects the BCL11A locus from epigenetic silencing. Although the precise sequence of events remains unclear, our observations that E+58 eRNA directly binds NIPBL and that ASO-mediated eRNA depletion impairs NIPBL binding suggest that eRNA transcription may initiate this process. Furthermore, eRNA depletion triggered the epigenetic silencing of BCL11A, leading to HbF reactivation in adult erythroid cells. These new aspects of BCL11A regulation highlight eRNA production as a potential vulnerability for HbF reactivation and open new avenues for therapeutic intervention.

The high costs, limited accessibility, and potential clinical risks associated with current gene therapies make them unattainable for most patients. Therefore, developing alternative, scalable, and cost-effective strategies to reactivate HbF is crucial for reducing the overall disease burden. Given that BCL11A directly regulates HbF expression^3,56,57 and has been clinically validated as a therapeutic target,^{6, 7, 8} approaches aimed at modulating its activity or regulation hold substantial promise for mechanism-based therapy.

Our discovery of enhancer-dependent epigenetic insulation in BCL11A regulation reveals new, targetable vulnerabilities, presenting unexpected opportunities for therapeutic intervention. One key vulnerability is eRNA production mediated by the E+58 enhancer, which is critical for erythroid-specific BCL11A expression and HbF repression. Agents that inhibit eRNA transcription or deplete eRNA transcripts, such as ASOs, could selectively downregulate BCL11A in erythroid cells, thereby reactivating HbF expression. The success of ASO-based therapies in treating muscular and retinal dystrophies, hereditary amyloidosis, and central nervous system disease underscores their potential for safe and effective gene perturbation.^{58, 59, 60, 61} In addition, antibody-oligonucleotide conjugates offer a promising approach for enhancing on-target, tissue-specific delivery.^62,63 Another potential strategy involves disrupting NIPBL-cohesin–mediated chromatin loop formation to induce the epigenetic silencing of BCL11A. Importantly, these strategies circumvent the need for genomic sequence modifications or complex ex vivo cell manipulation, thereby mitigating clinical risks associated with bone marrow transplantation. Developing potent, specific inhibitors of BCL11A eRNAs, along with combinatorial approaches targeting multiple layers of BCL11A regulation and function, represents a promising path toward more accessible therapies for major hemoglobin disorders.

Conflict-of-interest disclosure: J.X. is a consultant for Agios Pharmaceuticals. J.X. and S.H.O. hold patents on the development of therapeutic methods for treating hemoglobinopathies. The remaining authors declare no competing financial interests.