Base editing of key residues in the BCL11A-XL-specific zinc finger domains derepresses fetal globin expression

Abstract

BCL11A-XL directly binds and represses the fetal globin (HBG1/2) gene promoters, using 3 zinc-finger domains (ZnF4, ZnF5, and ZnF6), and is a potential target for β-hemoglobinopathy treatments. Disrupting BCL11A-XL results in derepression of fetal globin and high HbF, but also affects hematopoietic stem and progenitor cell (HSPC) engraftment and erythroid maturation. Intriguingly, neurodevelopmental patients with ZnF domain mutations have elevated HbF with normal hematological parameters. Inspired by this natural phenomenon, we used both CRISPR-Cas9 and base editing at specific ZnF domains and assessed the impacts on HbF production and hematopoietic differentiation. Generating indels in the various ZnF domains by CRISPR-Cas9 prevented the binding of BCL11A-XL to its site in the HBG1/2 promoters and elevated the HbF levels but affected normal hematopoiesis. Far fewer side effects were observed with base editing- for instance, erythroid maturation in vitro was near normal. However, we observed a modest reduction in HSPC engraftment and a complete loss of B cell development in vivo, presumably because current base editing is not capable of precisely recapitulating the mutations found in patients with BCL11A-XL-associated neurodevelopment disorders. Overall, our results reveal that disrupting different ZnF domains has different effects. Disrupting ZnF4 elevated HbF levels significantly while leaving many other erythroid target genes unaffected, and interestingly, disrupting ZnF6 also elevated HbF levels, which was unexpected because this region does not directly interact with the HBG1/2 promoters. This first structure/function analysis of ZnF4–6 provides important insights into the domains of BCL11A-XL that are required to repress fetal globin expression and provide framework for exploring the introduction of natural mutations that may enable the derepression of single gene while leaving other functions unaffected.

Graphical abstract

Mohankumar and colleagues suggest that the alteration by base editor of key nucleotides at the zinc finger domains present exclusively in the BCL11A-XL isoform significantly elevates fetal hemoglobin without affecting terminal maturation in erythroid cells derived from human hematopoietic stem cells, unlike that in CRISPR-Cas9-mediated editing.

Introduction

BCL11A is a transcription factor that directly binds the fetal globin (HBG1/2) promoters and represses gene expression. It contains several classical C₂H₂-type zinc finger (ZnF) domains and is expressed in multiple tissues, such as the brain, liver, and bone marrow (BM), and in erythroid cells, hematopoietic stem cells (HSCs), and B cells.^1,2 Alternative splicing generates 4 major isoforms: BCL11A-XL, BCL11A-L, BCL11A-S, and BCL11A-XS, which retain a variable number of the ZnF domains. Only the longest isoform, BCL11A-XL, is implicated in fetal globin silencing. The shorter isoforms S and XS are expressed in the yolk sac and fetal liver, whereas the longer isoforms XL and L are more highly expressed in the adult BM.^3,4 The differential expression of BCL11A isoforms in the definitive erythroid compartment fits with a role in fetal globin silencing during development.³ The functional importance of the BCL11A-XL isoform in fetal globin repression was firmly established by short hairpin RNA (shRNA)-mediated BCL11A-XL knockdown for the treatment of β-hemoglobinopathies, and BCL11A-XL is now regarded to be a promising target for β-hemoglobinopathy treatments.^5,6,7,8

Genome-wide association studies initially implicated the BCL11A locus as a major quantitative trait associated with variation in fetal hemoglobin (HbF) levels.^9,10,11 Three sets of genetic variants are associated with high HbF, including single nucleotide variants (SNVs) in the intronic BCL11A enhancer,^12,13,14,15 regulatory SNVs in the HBG1/2 promoters that disrupt the binding site (BS) of BCL11A-XL, and the exonic SNVs in the ZnF4 and ZnF5 domains contained specifically in the BCL11A-XL isoform.¹⁶ Recent studies revealed that targeting the SNVs located in the +58 DNase hypersensitive site of the BCL11A intronic enhancer downregulates BCL11A expression in the erythroid lineage, which causes simultaneous derepression of HbF clinically, without altering the expression of BCL11A in non-erythroid cells.^{14,15,17,18,19,20,21} The regulatory SNVs located in the HBG1/2 promoters at the −115 site alter the core binding consensus motif of BCL11A-XL and substantially elevate the expression of HbF.^22,23 Ongoing clinical trials for the reactivation of HbF by genome editing emphasize the significance of the SNVs in the BCL11A intronic enhancer and BCL11A-XL BS in the HBG1/2 promoters.²⁴ Structural studies have indicated that the BCL11A ZnF domains ZnF4 and ZnF5 interact directly with the HBG1/2 promoter; however, the potential role of targeting these BCL11A-XL-specific ZnF domains on HbF regulation and other functions of BCL11A remains unexplored.²⁵

In this study, we test the effect of targeting different ZnF domains specific to the BCL11A-XL isoform: ZnF4, ZnF5, and ZnF6. We initially disrupted each ZnF using CRISPR-Cas9-based cutting. Next, taking cues from the available BCL11A SNVs^10,16,26 in exon4 and the structural studies depicting the specific amino acid residues involved in the interaction of BCL11A-XL with DNA,^25,27 single nucleotide substitutions were generated using base editors. We compared our disruptions of the XL-specific ZnF domain with the other BCL11A-associated HbF-inducing targets such as the erythroid-specific enhancer and −115 site in HBG1/2 promoters, and ablation of the BCL11A gene, both in vitro and in vivo. Furthermore, we analyzed the transcriptomic variations caused by the specific ZnF alterations and compared the effects with the other BCL11A-associated approaches. Our results provide important insights about the ZnF domains of BCL11A-XL required for fetal globin repression and other functions such as hematopoietic stem and progenitor cell (HSPC) engraftment and B cell production.

Results

CRISPR-Cas9-mediated disruption of BCL11A-XL specific ZnF domains in adult erythroid cells derepresses fetal globin expression

Previous studies have shown that BCL11A-XL interacts with the HBG1/2 promoters in adult erythroid cells using its C-terminal ZnF domain, which contains ZnFs 4, 5, and 6 (Figure S1A)²⁸. However, the roles of each ZnF domain specific to BCL11A-XL in HbF regulation and erythroid maturation have not been established. Therefore, we performed CRISPR-Cas9-mediated disruption of each ZnF (guide RNA (gRNA): ZnFs 4, 5, and 6), and compared the effects with the other BCL11A-associated targets, such as disruption of the intronic erythroid-specific BCL11A enhancer , and its −115 cluster binding site (BS) in the HBG1/2 promoters. In addition, gRNAs designed to ablate BCL11A (targeting BCL11A exon2) and adeno-associated virus integration site1 (AAVS1) were chosen as positive and negative controls, respectively (Figure 1A). Immortalized human umbilical cord derived erythroid progenitor 2 (HUDEP2) cells stably expressing Cas9 were transduced with the specific gRNAs (Figure S1B). A total of 70%–90% indels were achieved at all of the intended target sites (Figure 1B). Editing at ZnFs 4, 5, and 6 (amino acids 759, 785, and 810, respectively) did not alter BCL11A levels but resulted in the truncation of BCL11A-XL proteins to various lengths due to frameshift mutations (Figures 1C, S1C, and S1D). In contrast, significant downregulation of BCL11A protein was observed when targeting BCL11A exon2 (BCL11A knockout) or the enhancer. These results indicate that targeting BCL11A exon2 (N-terminal) may also reduce expression by nonsense-mediated RNA decay, whereas targeting the XL-specific ZnF domains (C-terminal) produced truncated variants of BCL11A.²⁹ As expected, the truncated BCL11A variants failed to repress HBG1/2 at the adult stage. After erythroid differentiation (Figure S1E), the ZnF alterations produced a higher frequency of HbF⁺ cells (Figure 1D) and higher fetal globin derepression than seen in control cells (Figure S1F). The HbF⁺ cells produced were noticeably more numerous than observed after modification of the enhancer (Enh vs. ZnF, p < 0.0001) and at a level comparable to BS disruption at the HBG1/2 promoters. Despite previous reports demonstrating that ZnF6 has a weak direct interaction with the HBG1/2 promoters compared to ZnF4 and ZnF5²⁸, we observed a significant elevation in HbF upon ZnF6 disruption. These results highlight the importance of each ZnF-domain specific to BCL11A-XL in the repression of the fetal globin genes.

Figure 1.

Targeted disruption of BCL11A-XL specific ZnF domains using Cas9 to upregulate fetal globin expression in adult erythroid cells

(A) Schematic representation of the target location of gRNA in different BCL11A-XL-specific ZnF domains (ZnF4, ZnF5, and ZnF6), Exon2, Enhancer (Enh), −115 cluster of HBG1/2 promoters (BS), and the AAVS1 site. (B) The frequency of indels was measured by ICE analysis at the respective target site in HUDEP2 cells during expansion. Mean ± SEM of n = 3 independent experiments. (C) The protein expression analysis by Western blot on the edited cells during differentiation (day 4) depicts the generation of BCL11A variants protein on targeting the XL-specific ZnF domains in contrast to the reduction of BCL11A protein while targeting Exon2 and the enhancer. (D) Intracellular HbF analysis using flow cytometry demonstrates the potential of ZnF mutants on HbF reactivation at the terminal stage of erythroid differentiation (day 8). Mean ± SEM of n = 3 independent experiments. (E) The percentage of gene modifications during HSPCs expansion after 48 h of electroporation and after erythroid differentiation (day 21), respectively. Mean ± SEM of n = 2 experiments on one donor. (F) The change in erythroid differentiation ability of the BCL11A-modified HSPCs was analyzed by flow cytometry using 2 erythroid-specific surface markers (CD71-transferrin; CD235a-glycophorin A) at the terminal stage of differentiation (day 21). Mean ± SEM of n = 2 experiments of one donor. (G) The measure of mature enucleated cells formed by the edited HSPCs during the terminal stage of erythroid differentiation (day 21) using NucRed stain and CD235a marker. Mean ± SEM of n = 2 experiments on one donor. (H ) The number of HbF⁺ cells expressed, and (I) the total HbF tetramer in edited cells formed on BCL11A ZnF domain modification by flow cytometry and HPLC, respectively, measured during terminal erythroid differentiation (day 21). Mean ± SEM of n = 2 experiments on one donor. (J) The long-term engraftment of BCL11A total knockout, BS and ZnF domain-modified HSPCs after 16 weeks of transplantation in NBSGW mice. Mean ± SEM of n ≥ 3 mice per group. (K) The persistence of gene modification in the edited cells is depicted by editing during both pre-infusion and after 16 weeks of transplantation.

BCL11A ZnF editing in CD34⁺ HSPCs through CRISPR-Cas9 elevates HbF but also affects erythroid maturation

In addition to its major role in globin regulation, BCL11A is essential for human HSC maintenance and erythropoiesis. Therefore, we assessed the effect of BCL11A-ZnF disruption on cell survival and erythroid maturation in CD34⁺ HSPCs obtained from a healthy donor by nucleofecting the CRISPR-Cas9-gRNA complex. Based on our HUDEP2 results in which enhancer disruption showed moderate HbF elevation, we determined the impact of the generating BCL11A-ZnF variants in erythroid cells derived from HSPCs, along with the BS disruption at the HBG1/2 promoters, and BCL11A knockout at exon2. HSPCs exhibiting efficient gene modifications (Figure 1E) at the respective target sites were subjected to erythroid differentiation to assess the impact of the modifications on functions beyond HbF elevation during in vitro erythropoiesis.

In line with previous research demonstrating the importance of BCL11A in cell maintenance and proliferation³⁰, the growth kinetics of committed erythroid progenitors from the edited samples were analyzed. The total BCL11A knockout and ZnF editing in CD34⁺ HSPCs affected cell proliferation, whereas editing other targets did not affect survival (Figure S1G). The frequency of gene perturbations was assessed at the erythroid terminal differentiation stage. This revealed that except for ZnF4 and total knockout edited cells, gene modifications in all of the other targets persisted during differentiation (Figure 1E). This indicates that cells carrying alterations at the terminal sequences coding for ZnF5 and ZnF6 retain edits and persist across multiple cell divisions. A previous study reported that disruption of BCL11A exon4 (ZnF) did not impede early erythroid differentiation²⁹, whereas in our study, targeting the ZnF4, ZnF5, and ZnF6 domains showed a decline in the percentage of terminal-stage erythroid cells (CD235a⁺CD71⁻) (Figure 1F), with a significant reduction in the enucleation (Figure 1G). However, a remarkable level of fetal globin reactivation was seen in adult erythroid cells on targeting the BCL11A ZnF domains (Figures 1H and 1I), which was equivalent to levels obtained with the BS disruption, despite the differences observed in cell proliferation and differentiation. Overall, these results demonstrate the in vitro consequence of disrupting either at the exon2 (N-terminal) or the ZnF domains (C-terminal) of BCL11A on cell survival and erythroid maturation along with globin regulation.

BCL11A isoforms have been implicated in HSC engraftment and B lymphocyte maturation.^31,32 Therefore, we investigated the engraftment potential of BCL11A-XL ZnF-modified cells in the NBSGW mouse model. Since ZnF5 had exhibited significant erythroid maturation defects during in vitro experiments, we infused only the ZnF4 and ZnF6-modified HSPCs. As controls, we also infused BCL11A BS-disrupted HSPCs and total BCL11A knockout (exon2-edited) HSPCs. All of the transplanted animals showed chimerism of human cells (hCD45⁺) in the BM after 16 weeks of infusion, even the cohorts with total BCL11A knockout cells (Figure 1J). Therefore, we genotyped the engrafted cells to study the long-term survival of gene-modified HSPCs, which revealed that the indel frequency in the HBG1/2 promoter (BCL11A BS) edited cells remained unchanged, whereas HSPCs with BCL11A complete knockout did not survive in mouse BM, consistent with the previous findings.³⁰ We had speculated originally that BCL11A ZnF domain modification would have lesser impacts on HSPCs, but we observed a drastic decrease in ZnF4 and ZnF6 gene modifications in the BM cells (Figure 1K). Similar to total BCL11A knockout, we observed that the CRISPR-Cas9-mediated disruption of BCL11A-XL-specific ZnF domains does not support HSC engraftment. Furthermore, due to the inadequate engraftment potential of the BCL11A ZnF mutants, we were unable to investigate its multilineage differentiation ability. These findings demonstrate that even a minimal loss of the BCL11A-XL C-terminal coding region functions similar to total protein loss and can negatively affect the engraftment of HSCs.

Specific base substitutions in the BCL11A-XL C-terminal ZnF domains reactivate HbF expression in adult erythroid cells

CRISPR-Cas9-mediated targeting of BCL11A-XL-specific C-terminal ZnF domains resulted in multiple truncated variants hampering the concise understanding of the role of BCL11A-XL ZnF domains in HbF repression, engraftment, and erythroid maturation. To address this, we introduced more subtle mutations at specific amino acid residues in the ZnFs using an adenosine base editor in HUDEP2 cells anticipating HbF elevation without compromising other functional roles of BCL11A.

The key nucleotides at ZnF4 (amino acids 742–765) and ZnF5 (amino acids 772–795) were targeted to alter crucial motifs required for DNA interaction and where possible to resemble as closely as possible known missense variants associated with higher HbF levels in patients with neurodevelopmental disorders but essentially normal hematology (Figure S2A)^{10,16,25,26,27}. With the current base editing technology, it is not possible to recreate the exact mutations, but we targeted mutations as closely as possible to the sites found in patients. No naturally occurring mutations have been reported in ZnF6 (amino acids 803–827), but we designed a gRNA that would result in a base substitution at a potential DNA contact region in the ZnF6 domain (Figure 2A). The fetal globin reactivation potential of base substitutions in each ZnF domain was compared with the same BCL11A control targets used in previous experiments, along with the gRNA at the initiation codon for the total knockout of BCL11A. Subsequently, all of these gRNAs were individually transduced into HUDEP2 cells with more than 90% efficiency (Figure S2B).

Figure 2.

The key base substitutions in BCL11A-XL-specific ZnF domains reactivate HbF expression in adult erythroid cells

(A) A diagrammatic representation of each of the target sites to induce nucleobase modifications by using ABE. BCL11A-gRNA targeting initiation codon of BCL11A, Enh-gRNA targeting the BCL11A erythroid-enhancer region, 2 gRNAs each targeting ZnF4 (ZnF4-1 and -2) and ZnF5 (ZnF5-1 and -2), respectively, and 1 gRNA targeting ZnF6 to produce missense mutations. BS gRNA to alter the BCL11A-XL BS in the −115 cluster of HBG promoter. AAVS1-gRNA targeting the safe harbor site at the AAVS1 locus for negative control. (B) Base conversion efficiency at all of the target sites on day 8 of expansion was analyzed using the EditR in silico tool after Sanger sequencing. Mean ± SEM of n = 3 independent experiments. (C) Close-up views of the BCL11A-XL-specific ZnF (ZnF4, ZnF5, and ZnF6) structures, illustrating the amino acids targeted in each of the indicated mutants. Yellow dashed lines indicate hydrogen bonds, and bases are named according to the HBG1/2 promoter numbering. (D) The protein expression analysis by western blot confirms no significant change in BCL11A expression on ZnF editing compared to the reduction of BCL11A protein observed on disrupting the initiation codon and the BCL11A-enhancer; 35 μg of protein used for loading. (E) Analysis of the erythroid differentiation profile (CD71 and CD235a) of the base-edited HUDEP2 cells on the end stage of differentiation (day 8) showed no significant variation among the edited samples. Mean ± SEM of n = 3 independent experiments. (F) Increase in percentage of intracellular HbF levels measured at the end stage of differentiation (day 8) by flow cytometry (HbF⁺ cells) on generated BCL11A ZnF mutants. Mean ± SEM of n = 3 independent experiments. (G) The elevation in the total HbF tetramer formation in adult erythroid cells in the BCL11A-modified cells was analyzed at the end stage of differentiation (day 8). Mean ± SEM of n = 3 independent experiments.

After 8 days of expansion, the efficiency of base conversion at the target sites was evaluated by Sanger sequencing (Figure 2B). In the case of the ZnF4 domain, the ZnF4-1 gRNA specifically converted the nucleotides that encode the amino acids S755 and N756 approximately with 40% and 90% base conversion efficiency, respectively. S755 makes van der Waals interactions with the −119 (T:A) base and the DNA backbone at the HBG1/2 promoters, whereas N756 forms a specific double hydrogen bond with −118 (A:T) (Figure 2C, ZnF4-1). The ZnF4-2 gRNA generated novel mutants H760R and R761G with ∼80% efficiency. Although these amino acids do not contact DNA, H760 is one of the Zn-ligating residues in ZnF4, and its mutation likely perturbs the structure of the domain (Figure 2C, ZnF4-2). Alteration of the ZnF5 domain was achieved using ZnF5-1 gRNA to efficiently convert bases that encode Q781 (∼90%), S782 (∼75%) and S783 (∼55%). Q781 forms a hydrogen bond with −116 (T:A) (Figure 2C, ZnF5-1). Similarly, the ZnF5-2 gRNA generated specific nucleotide perturbations that mutates R787 (∼55%), and H788 (∼70%); R787 forms hydrogen bonds with −114 (G:C), whereas H788 is one of the Zn-coordinating residues in ZnF5 (Figure 2C, ZnF5-2). The ZnF6 gRNA exhibited a less efficient (∼40%) conversion of bases and encoded changes to K817; this residue is predicted to make nonspecific electrostatic interactions with the DNA backbone (Figure 2C, ZnF6).

The control sites including the −115 cluster in the HBG1/2 promoter, BCL11A initiation codon and the BCL11A enhancer were substantially modified, with overall efficiencies greater than 70%. Unintended indels were less than 5%, despite the high editing efficiency at all of the target sites. The consequence of the generated BCL11A-XL mutants on protein expression was investigated to analyze the impact of pathogenic SNVs on the stability of the BCL11A protein. Interestingly, the XL-specific ZnF mutants exhibited no effect on protein synthesis and were expressed similarly to the AAVS1 negative control editing, whereas the alteration of initiation codon affected the BCL11A protein expression mimicking BCL11A knockout as expected (Figure 2D).

We assessed the functional consequence of the BCL11A mutations on fetal globin expression during terminal erythroid differentiation (Figure 2E). Interestingly, the HbF reactivation observed in all of the ZnF mutants was comparable to complete BCL11A knockout. The increase in the number of HbF⁺ cells ranged from 60% to 90% across the ZnF mutants, which may be due to a difference in editing efficiency or the impact of the different mutations (Figure 2F). The ZnF4 mutations increased the number of HbF⁺ cells more than the ZnF5 or ZnF6 mutations. Although the editing frequency at the ZnF6 target site was lower, significant HbF elevation was observed, so we hypothesize that ZnF6 also influences BCL11A binding to the HBG1/2 promoters. Consistent with the increase in HbF⁺ cells, the ZnF4 mutants produced higher levels of total HbF relative to mutations in other ZnF domains (ZnF5 and ZnF6).

Importantly, the majority of ZnF-specific mutations outperformed the disruption of the BCL11A BS in the HBG1/2 promoters in HbF reactivation (Figure 2G). It may be that BCL11A binds at additional sites in the globin locus; therefore, mutating the DNA-binding domain has a more profound effect than mutating a single binding motif. These findings confirm that the induction of HbF does not require haploinsufficiency or the total loss of BCL11A but can be mediated by targeted mutations.

Together, both the in vitro and in silico results imply that altering the C-terminal XL-specific ZnF domains involved in binding to the target DNA (e.g., the motifs in the HBG1/2 promoters) exhibited an HbF upregulation similar to that observed with total BCL11A loss. Furthermore, unique base conversations that alter key amino acid residues are sufficient to significantly affect ZnF domain function.

Generation of BCL11A-ZnF mutants in HSPCs leads to HbF upregulation during ex vivo erythropoiesis with reduced hematological effects

A previous study reported that eliminating hydrogen bonds formed between N756 and the −118 (A:T) base by creating N756A mutation reduces the affinity of BCL11A for DNA by 8-fold.²⁵ To confirm that changing N756 to glycine as achieved via base editing had the same effect, we validated the loss of binding of the ZnF4-1 mutant to the −115 site in the HBG1/2 promoters by gel shift assays using proteins containing S755G, N756G, and S755G and N756G double mutants (Figures 3A and S3A). We observed that the N756G mutant, both individually and in combination with S755G, abrogated the binding to the target site, whereas the S755G mutant alone did not significantly affect binding. Furthermore, to phenotypically evaluate the contribution of each mutant on fetal globin derepression, we performed single-cell sorting in HUDEP2 cells and observed that N756G contributed significantly to the HbF induction, consistent with the gel shift assay results (Figure S3B).

Figure 3.

Generation of BCL11A ZnF mutants in HSPCs exhibit elevated levels of HbF on erythroid differentiation

(A) Electrophoretic mobility shift assay depicting the loss of interaction between the ZnF4-1 (N756 mutants) mutants to a TGACC motif. Nuclear extract of COS-7 cells expressing epitope tagged fused ZnF4–6 of wild-type BCL11A and mutants with radiolabeled fetal globin promoter BCL11A binding motif TGACC. The black arrow shows the BCL11A ZF4-6 binding to the probe, and super-shifted bands are observed when antibody to the V5 epitope is added. (B) Target base perturbation efficiency at each target site was analyzed after 2 days of electroporation in EditR software. Mean ± SEM of n = 2 experiments of one donor. (C) The reactivation of total HbF measured by HPLC at the terminal stage of erythroid differentiation (day 21). Mean ± SEM of n = 2 experiments of one donor. (D) The percentage of edited cells capable of maturating into enucleated reticulocytes in vitro measured by CD235a expression and NucRed stain. Both erythroid differentiation and maturation analysis were measured at the terminal stage of erythroid differentiation (day 21). (E) The erythroid differentiation potential of base-edited HSPCs was examined using CD235a and CD71 surface markers at the terminal stage of erythroid differentiation (day 21). (F) Engraftment percentage of base-edited HSPCs and multilineage repopulation in NBSGW mice after 15 weeks of transplantation. (G) The percentage of base editing observed before infusion, after 15 weeks in the mouse BM and in the B cell subset. B cells were flow sorted using human CD45 and CD19 surface markers. (H) NGS analysis depicting the multiple editing pattern within engrafted ZnF4-1 edited cells derived from the mouse BM after 15 weeks of engraftment. The first A denotes the adenine base in the S755 position and subsequent AA denotes the dual adenine bases present in the N756 position, which is converted to G with different frequencies. Others in the graph indicate the summation of all of the nonspecific mutations. (I) Percentage of HbF⁺ cells in the CD235a⁺ cell population within the NBSGW BM after 15 weeks. n = 3; mean ± SEM.

Given the results in HUDEP2 cells, we next tested the effect of base editing at BCL11A-XL ZnFs in healthy donor CD34⁺ HSPCs, using adenine base editor 8e (ABE8e) mRNA and again compared them with the BCL11A associated targets (Figure S3C). We obtained base conversions at the respective target sites with overall efficiencies of 60%–70% in HSPCs (Figure 3B). As expected, the ZnF4-1 gRNA generated 2 missense mutations at N756 and S755 at ZnF4, whereas the gRNAs targeting the other BCL11A-associated targets altered either the BCL11A start codon (to achieve total knockout) or the intronic enhancer or the −115 site in the HBG1/2 promoters. Notably, all of the base modifications persisted throughout erythroid differentiation (Figure 3B).

To investigate the effect of the BCL11A ZnF4 mutant HSPCs on cell survival and differentiation, we monitored the base-modified HSPCs throughout differentiation to the erythroid lineage. Cell proliferation in the BCL11A knockout mutant was considerably lower than that in the ZnF4 mutant, confirming that the BCL11A-XL isoform has a role in erythroid cell proliferation and survival (Figure S3D). More important, the specific mutations in ZnF4 domain did not impair cell viability.

We anticipated that disrupting multiple binding of BCL11A in the globin locus could lead to stronger upregulation than the ablation of the single BCL11A BS at HBG1/2 promoters. Therefore, we assessed the potency of the BCL11A ZnF4 missense mutant on the derepression of fetal globin expression in erythroid progenitors derived from base-edited HSPCs. Similar to the total knockout, the BCL11A ZnF4 mutation resulted in high HbF⁺ cells (Figure S3E) and high total HbF production (Figure 3C).

We then investigated the effect of the BCL11A ZnF4 missense mutant on terminal erythropoiesis because our CRISPR-Cas9 cutting study showed a significant erythroid maturation delay in BCL11A ZnF truncated protein-expressing erythroid cells during the terminal stages. Interestingly, BCL11A ZnF4 base-edited cells underwent terminal erythroid maturation (CD235a⁺/NucRed⁻) more efficiently than total knockout cells and truncated proteins generated in the CRISPR-Cas9 cutting experiments on day 21 of HSPC differentiation (Figure 3D). Specifically, we also observed higher HbF-expressing enucleated cells at the erythroid terminal maturation phase in the BCL11A ZnF mutant (Figure S3F). Notably, the BCL11A ZnF4 mutant exhibited a modest decrease in the frequency of cells expressing erythroid differentiation markers (CD235a⁺/CD71⁻) with normal enucleation (Figure 3E). However, this effect was modest when compared to CRISPR-Cas9-mediated truncation of the BCL11A ZnF4 domain. Altogether, these data demonstrate that the fetal globin reactivation induced by altering the individual amino acids specific to the BCL11A-XL isoform involved in binding to the HBG1/2 promoters is preferable to indel formation via CRISPR-Cas9, and if refined, could ultimately serve as a strategy for fetal globin induction.

We then investigated the repopulation ability of the BCL11A ZnF4 base-edited human CD34⁺ HSPCs by transplanting them into immunodeficient NBSGW mice, noting that BCL11A truncated proteins from the CRISPR-Cas9 experiment exhibited defective engraftment. We observed efficient chimerism of human cells (hCD45⁺) in the control edited cells, whereas the ZnF4 mutant exhibited reduced engraftment (Figure 3F), but more important, the ZnF4-1 mutant cells were able to survive and repopulate over time. Because the neurodevelopmental patients harboring certain missense mutations in BCL11A-XL-specific ZnF domain exhibit essentially normal hematopoiesis, we analyzed the impact of the BCL11A ZnF4 mutant on multilineage repopulation potential. Consistent with a moderate reduction in engraftment, ZnF4-1 mutants also showed altered levels of lineage distribution, notably a significant decline in B cells, with a bias toward myeloid cells (Figure 3F). Although re-creating the exact patient SNVs using base editor is not feasible, we aimed to mimic the heterozygous nature of these BCL11A-ID patients by reducing the editing efficiency at the target site. Genotypic analysis of the engrafted human cells showed an evident drop in the editing frequency of the ZnF4-1 mutant within both the BM (2.5-fold) and total loss in the B cell engrafted compartment (Figure 3G). We observed that N756G amino acid production was prevalent over N756D or N756S amino acids. Furthermore, N756G amino acid also persisted at the 15 weeks of in vivo engraftment in NBSGW mice (Figure 3H). The CD235a⁺ erythroblasts present in the mouse BM of the ZnF4-1 mutant engrafted mice showed increase in HbF⁺ cells compared to edited controls (Figure 3I).

Furthermore, the persistence of single mutants in the BM population suggests that creating monoallelic single mutation resembling the patient SNVs at the ZnF domains elevates HbF, but the data also suggest that C-terminal ZnF domains have an important function in HSPC engraftment and B cell production outside the globin regulation within the erythroid lineage.

Global transcriptomic analysis of the BCL11A ZnF4 base-edited erythroid cells derived from CD34⁺ HSPCs

Since BCL11A regulates multiple pathways involved in the development and differentiation of various cell types, any alteration of BCL11A requires extensive characterization.^30,33 Therefore, we compared genome editing at the XL-specific ZnF domain along with the other BCL11A-associated targets. A total of 691 genes were differentially expressed after the complete loss of BCL11A, whereas disruption of ZnF4 (123 genes) and BS (94 genes) showed significantly fewer differentially expressed genes. The expression of only 9 genes was observed to be significantly altered in the case of erythroid-enhancer disruption, consistent with the modest HbF induction observed in vitro after erythroid differentiation (Figures 4A and 4B). The transcriptional repressor activity of BCL11A is evident based on the fact that more genes were upregulated on modifying BCL11A compared to the BS disruption in the HBG1/2 mutation. This clearly demonstrates that BCL11A ZnF4 mutant generation has fewer effects beyond the upregulation of fetal globin expression than are observed with the total ablation of BCL11A expression.

Figure 4.

Transcriptomic analysis of ZnF4-1 mutant in erythroid cells derived from HSPCs

(A) Total number of differentially expressed genes in total knockout (BCL11A), enhancer-edited (Enhancer), ZnF4-1 modification, and −115 site at HBG1/2 (BS) disruption compared to the edited control. (B) Correlation between the different transcriptome among upregulated and downregulated genes in total knockout, enhancer edited, ZnF4-1 modification, and BS disruption compared to the edited control. Significant gene expression is maintained at false discovery rate (FDR) 5% and −1>fold change >+1. (C) Correlation showing the significantly expressed (FDR 5% and −1>fold change >+1) primary targets among the different edited groups. (D) Heatmap showing the BCL11A primary target gene counts among all of the edited HSPCs.

Next, we examined the known primary targets³⁴ of BCL11A exhibiting significant differential expression among our study group (Figure 4C). In all of the target sites, HBG2 was the commonly upregulated gene, as expected. BGLT3 and HBG1 are the 2 other genes commonly upregulated among the ZnF4-1, BS, and BCL11A edited cells. Interestingly, the ZnF4 editing shared upregulation of the HBZ, HBBP1, and KLHDC8B genes with BCL11A total knockout. Consistent with the whole transcriptome profile, the disruption of BCL11A BS generates no other significant alterations in the primary targets (Figure 4D). Moreover, the significant increase in the expression of the cell-cycle-associated genes, such as CDKN2B, PPP2R5B, and BCL6 occurs only in BCL11A complete knockout cells and not in the ZnF4 mutant cells. This may explain why the ZnF4 mutation has modest effects on erythroid function other than globin regulation.

We further validated the expression of these globin locus genes by real time qRT-PCR (Figure S4A). These results confirm that specifically disrupting the ZnF4 domain has fewer effects on other erythroid genes compared to the total ablation of BCL11A.

Next, to evaluate whether the observed induction of HbF is specific to the on-target editing activity, we performed targeted deep sequencing at the potential off-target (OT) sites (with up to 3 mismatches) predicted by the in silico algorithm COSMID (Figure S4B). The top 9 sites were deep sequenced in the base edited HSPCs and we observed no significant DNA variation between the control and the edited samples indicating that the gRNA used is highly specific to the BCL11A ZnF4 locus. The CIRCOS plot depicting the predicted OT locations shows that none of the OT sites are in the coding region, providing further evidence that the gRNA is highly specific (Figure S4C). To further rule out the impact of unintended RNA deamination caused by ABE8e, we analyzed the whole transcriptome RNA sequencing (RNA-seq) information of the ZnF4-1 edited HSPCs during erythroid differentiation and compared it with the other BCL11A-associated targets. We observed that no significant difference in the adenosine-to-inosine (A-to-I) conversion among the gRNAs used in this study (Figure S4D). These results confirm the specificity of targeting the ZnF4 using ABE8e.

Discussion

From the time that BCL11A was implicated in fetal globin regulation, most of the gene therapy clinical trials for β-hemoglobinopathies have focused on the modulation of BCL11A through various approaches. The erythroid-specific enhancer has been targeted, and shRNA-mediated downregulation of BCL11A expression specifically within the erythroid lineage also results in the derepression of fetal globin. Disrupting the BCL11A BS at the −115 site in the HBG1/2 promoters by both CRISPR-Cas9-mediated disruption and base editing is being pursued. In addition to these studies that target BCL11A at the genomic or transcript level, small molecules such as nanobodies have been developed to selectively reduce BCL11A-XL at the protein level.^35,36,37 However, the spatiotemporal window for optimal nanobody delivery is still being explored. Although these studies emphasized the clinical importance of BCL11A in the treatment of β-hemoglobinopathies, they also provide insights into the role of BCL11A-XL protein in blood cell gene expression.

Here, we investigated a novel strategy to specifically alter the DNA-binding domains of BCL11A-XL to functionally mimic naturally occurring mutations affecting the BCL11A-XL isoform found in patients with normal or near-normal hematopoiesis. One of the major advantages of this strategy is that it generates very high levels of HbF induction, presumably because it impairs the binding of BCL11A-XL, not only to its site at −115 in the fetal globin promoters but also to other sites that may lie elsewhere in the globin locus and be required for complete fetal globin silencing.³⁸ Specifically, modification of ZnF4 led to very high levels of HbF in the adult stage, similar to those seen with the total loss of BCL11A protein.

BCL11A-XL has been demonstrated to interact with the TGACCA site in the HBG1/2 promoters (−115 cluster) via primarily the ZnF4 and ZnF5 domains.^25,28 Several microRNAs and transcriptional regulators, including LIN28B, have been shown to affect the production of the XL isoform by binding to BCL11A exon4.³⁴ Many studies have focused on targeting the BCL11A recognition motif at the HBG promoter rather than the endogenous BCL11A-XL ZnF domains to eliminate the BCL11A:HBG interaction.

To refine our understanding of ZnF domains in globin regulation, we targeted the endogenous BCL11A XL-specific ZnF (ZnF4, ZnF5, and ZnF6) domains in adult erythroid cells using CRISPR-Cas9-mediated editing and indel formation. Gene perturbations at the ZnF4, ZnF5, and ZnF6 domains efficiently generated ZnF truncated mutants that produced elevated levels of HbF expression, comparable with changes seen by targeting other BCL11A-associated regions such as the enhancer or the −115 cluster in the HBG1/2 promoters. Although ZnF6 has not been observed to interact directly with the DNA BS, we observed HbF upregulation on ZnF6 disruption, suggesting that it makes an indirect contribution and is required for proper fetal globin silencing. The truncated BCL11A-XL mutants in HSPCs were defective in cell survival and erythroid maturation, and they also failed to survive in an in vivo mouse BM environment.

To further explore the role of specific subdomains within each ZnF and to attempt to reduce side effects, we used a high-fidelity base editor to alter the key bases, present in XL-specific ZnF domains, that are known to interact with the target DNA strand. Together with the available database on SNVs, we particularly aimed to target the reported BCL11A SNVs associated with neurodevelopmental disorders, in which patients show derepressed fetal globin but otherwise have normal hematology. The current base editors have limited capability to generate desired heterozygous mutations without any bystander conversion. Nevertheless, we generated different ZnF-specific missense mutants in BCL11A-XL using an ABE8e and confirmed the reduced binding of BCL11A-XL to its DNA recognition site in a gel shift assay, demonstrating that specific mutations, with the expected compromised functions, could be made.

Among the ZnF domains, the reactivation of HbF was highest with ZnF4 and ZnF5 mutants, which was equivalent to total BCL11A suppression, followed by BS disruption at the HBG1/2 promoters, and the enhancer-mediated BCL11A downregulation. This emphasizes that base modifications at the ZnF domains are sufficient to produce significant levels of HbF. However, we also observed a reduction in engraftment and B cell production. These data are in line with another study that reported the role of ZnF domains in the immunomodulatory activity of B cells against Hirame novirhabdovirus in flounder.³⁹ The exact effect of the ZnF4-1 mutation in the functional activity of B cells and HSPC engraftment need to be studied to extend this approach clinically. However, the high efficiency of ABE8e resulted in more than 1 missense mutation in each of the ZnF domains, creating double mutants, unlike the natural observed variants. Nevertheless, the ZnF4 mutant resulted in effective HbF reinduction with near-normal erythroid maturation in vitro and thus generated fewer side effects than CRISPR-Cas9-mediated disruption of the domain or other major perturbations of the BCL11A gene.

Overall, altering the BCL11A ZnF4 domain may disrupt all of the potential interactions with the globin cluster, leading to effective induction of HbF compared to the previously reported multiplex editing at both the BCL11A enhancer and BS.⁴⁰ Furthermore, we observed minimal transcriptomic changes on altering the ZnF4 domain compared to total BCL11A loss. The transcriptomic profiling clearly demonstrated the upregulation of globin locus genes other than HBG1/2, such as HBBP1 and BGLT3, consistent with reports that the latter genes are also regulated by BCL11A.

A recent study highlighted the homology between the BCL11A and BCL11B proteins,^35,37 but due to the codon degeneracy, the sequences encoding the 2 distinct proteins varied. It is interesting that the nonoccurrence of the predicted OT site at BCL11B reconfirmed that the target is specific to BCL11A-XL and does not affect the function of BCL11B. Collectively, at both DNA and RNA levels, the occurrence of negligible levels of OTs suggests that the observed fetal globin derepression is solely due to the base alterations at the ZnF4 domain.

To conclude, our approach has shown that one can achieve HbF reactivation through minimal genetic alterations at key bases specifically in the BCL11A-XL with fewer transcriptomic changes. We believe that our approach of altering the DNA recognition domain could be extended further to other transcription factors that use multiple ZnFs to bind different sets of target genes. Our structure/function analysis of ZnF4–6 using base editor approaches provides important insights into the domains of BCL11A-XL required to repress fetal globin expression and its role in HSPCs. Furthermore, refining our editing strategy would act as a framework for exploring other natural mutations that may selectively reverse the repression of individual genes while preserving its other functions.

Materials and methods

Plasmids and lentivirus production

The target guide RNAs (Table S1) were cloned into the pLKO5.sgRNA.EFS.GFP plasmid (a gift from Benjamin Ebert, Addgene no. 57822). The gRNA-containing plasmids were then cotransfected with the VSV-G (vesicular stomatitis virus G protein) plasmid, pMD2.G (a gift from Didier Trono, Addgene no. 12259) and the envelope plasmid psPAX2 (a gift from Didier Trono, Addgene no. 12260) at a 2:1:1 ratio in HEK293T cells using the Mirus transfection reagent (1:3 ratio of DNA:reagent). Viral supernatants were collected at 48 and 72 h posttransfection, concentrated using an in-house preparation of PEG-6000 concentrator, and stored at −80°C until use.

Cell culture

HUDEP2 cells were cultured and differentiated as described previously. HUDEP2 cells, stably expressing Cas9 or ABE8e, were generated and validated.²² HUDEP2 lines were transduced with the gRNA lentivirus, and the transduction efficiency was determined by flow cytometry analysis of GFP expression after 48 h.

CD34⁺ HSPCs were isolated from the leftover G-CSF (granulocyte-colony-stimulating factor) mobilized blood of healthy donors after infusion to patients with informed consent as per clinical guidelines authorized by the institutional review board of Christian Medical College. The peripheral blood mononuclear cells were separated using density gradient centrifugation (Lymphoprep Density Gradient Medium, STEMCELL Technologies) and washed with 1× red blood cell (RBC) lysis buffer to remove any residual RBCs present. CD34⁺ HSPCs were purified using EasySep Human CD34 Positive Selection Kit II (STEMCELL Technologies). The isolated CD34⁺ cells were expanded and differentiated into the erythroid lineage as reported earlier, and the growth kinetics were measured by cell counting at each media change.⁴¹

COS-7 cells were cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum (Gibco) and 1% penicillin-streptomycin-glutamine (Gibco) and incubated at 37°C in 5% CO₂. Cells were lifted for passaging by incubation in 0.05% TrypLE Express Enzyme (Gibco) at 37°C for 5 min.

ABE8e mRNA production

To produce ABE8e mRNA, we used the ABE8e plasmid (a gift by David Liu, Addgene no. 138495) as the template and performed in vitro transcription as previously described.⁴² The resulting mRNA was stored in smaller aliquots at −80°C until use.

Nucleofection

For Cas9 experiments, 0.2 million HSPCs were electroporated with a Cas9:gRNA ratio of 50:100 pmol using a Lonza 4D-Nucleofector (pulse code: DZ100). For the base-editing experiments, we nucleofected 1 million HPSCs with 5 μg ABE8e mRNA with 100 pmol gRNA using a GTx electroporator (MaxCyte, pulse code: HSC-3).

Gene modification analysis

Genomic DNA from the edited and unedited cells was isolated and the target regions were PCR amplified for Sanger sequencing using the primers listed in Table S2. The efficiency of gene modification generated by Cas9 was analyzed using the Synthego Inference of CRISPR Edits (ICE) program.⁴³ To estimate the base conversion percentage, the Sanger sequencing files were analyzed using the EditR analysis software.⁴⁴

COS-7 cell transfections and nuclear extractions

COS-7 cells were transiently transfected with pcDNA3-based plasmids to overexpress BCL11A ZnF4–6 and mutants. Cells were transfected in 10-cm plates with 5 μg plasmid using FuGENE 6 Transfection Reagent (Promega). The mammalian expression plasmids used are listed in Table S3. Cells were washed in PBS after 48 h incubation at 37°C. Cells were harvested and resuspended in hypotonic lysis buffer (10 mM HEPES-KOH, pH 7.9, 1.5 mM MgCl₂, 10 mM KCl, 5 mM DTT, 1 mM PMSF, 0.01 mg/mL aprotinin, and 0.01 mg/mL leupeptin) and incubated on ice for 10 min. Cells were spun down and the pellets were resuspended in extraction buffer (20 mM HEPES-KOH, pH 7.9, 25% glycerol, 420 mM NaCl, 1.5 mM MgCl₂, 0.2 mM EDTA, 5 mM DTT, 1 mM PMSF, 0.01 mg/mL aprotinin, and 0.01 mg/mL leupeptin) on ice for 20 min. The suspension was centrifuged at 13,000 rpm for 3 min at 4°C and the supernatant containing nuclear extracts recovered.

Electrophoretic mobility shift assay

Oligonucleotides used in the synthesis of radiolabeled probe are listed in Table S2. The sense oligonucleotide was labeled with ³²P from γ-³²P ATP (PerkinElmer) using T4 PNK (New England Biolabs) before the antisense oligonucleotide was annealed by slow cooling from 100°C to room temperature. The probe was purified using a Quick Spin Column (Roche). The nuclear extracts were harvested from COS-7 cells. Empty extract from COS-7 cells was used as a control to check endogenous protein binding. Antibody against V5 (Invitrogen) was used to identify and super-shift of BCL11A ZnF4–6 and mutations. Samples were complexed in gel shift buffer (10 mM HEPES, pH 7.8, 50 mM KCl, 5 mM MgCl₂, 1 mM EDTA, 0.05 mg/mL poly(dI-dC), 1 mM DTT, 0.1 mg/mL BSA, and 5% glycerol) with and without V5 antibody and loaded on 6% native polyacrylamide gel in Tris-borate-EDTA buffer (45 mM Tris, 45 mM boric acid, and 1 mM EDTA). Electrophoresis was performed at 4°C for 105 min at 250 V followed by drying under vacuum. The gel was exposed to a FUJIFILM BAS CASSETTE2 2025 phosphor screen overnight and visualized using a GE Typhoon FLA 9500 fluorescent image analyser.

OT analysis

We used the COSMID web tool to predict Cas-dependent DNA OT sites with a stringency of no DNA and RNA bulges. Using the primers listed in Table S2, we amplified the predicted OT sites and carried out 150-bp paired-end Illumina sequencing with the specified adapter sequences. The resulting data were analyzed using the default settings of CRISPRESSOv2.⁴⁵ The transcriptome-wide A-to-I (or threonine-to-cysteine [T-to-C]) conversion in the ABE8e base-edited erythroid cells was calculated by REDItools version 2. All of the nucleotides other than A were removed from the analysis with the previously reported coverage and quality criteria.⁴⁶ The frequency of A converted to I (or T to C) was calculated by dividing the total number of edited A by the overall counts of A after filtering (A-to-I)/A × 100. The experiment was carried out as 2 biological experiments.

Flow cytometry analysis

The erythroid cells were stained with antibodies for 15 min, washed twice with PBS, and analyzed. The intracellular HbF staining and NBSGW mouse BM cells were processed as previously described⁴¹ in accordance with the Institutional Animal Ethics Committee. All of the antibodies used are listed in Table S4. BD Celesta and CytoFLEX-LX (Beckman Coulter) flow cytometers were used for sample analysis.

Western blot

Whole-cell protein lysates were prepared by resuspending the cell pellet with 1× radioimmunoprecipitation assay buffer along with Halt Protease Inhibitor Cocktail (Thermo Fisher Scientific). Approximately 35 μg protein was used for western blot analysis using the antibodies listed in Table S4.

High-performance liquid chromatography (HPLC)

At the terminal stage of erythroid differentiation, the differentiated cells were sonicated and centrifuged with maximum revolutions per minute at 4°C for 15–20 min to obtain the clear lysate.⁴⁷ Analysis of total hemoglobin (TOSOH HPLC variant analyser) was performed using the lysates. The percentage of total globin was calculated as [HbF/(HbF+HbA)] × 100.

In silico modeling

The BCL11A protein-DNA structure obtained from RCSB PDB-6U9Q was used as a reference.²⁵ We used Pymol software to visualize the amino acids involved in interactions with the DNA bases.

RNA-seq

Total RNA isolated during the second phase of erythroid differentiation (day 8) of edited HSPCs was processed for RNA-seq using Illumina (Novaseq 6000). The raw reads were filtered using Trimmomatic for quality scores and adapters. Filtered reads were aligned to the human genome (hg38) using splice aware aligners such as HISAT2 to quantify reads mapped to each transcript. The alignment percentage of reads ranged between 95.28% and 98.52% for all of the samples. The total number of uniquely mapped reads was counted using feature counts. The uniquely mapped reads were then subjected to differential gene expression using DeSeq2. Real time qRT-PCR was performed to validate the results using primers from Table S2.

Statistical analysis

All of the graphs were generated using GraphPad Prism version 8.1. One-way ANOVA was used in all of the analyses, and the comparison is with the control unless otherwise noted.

Data and code availability

Next-generation sequencing (NGS) data were deposited to the NCBI Sequence Read Archive using accession no. PRJNA952324. Transcriptome analysis raw files can be accessed through GSE: 229989.