CRISPR-based therapeutic genome editing for inherited blood disorders

Abstract

Therapeutic genome editing promises to transform medicine. Pivotal discoveries have provided a diverse and versatile set of tools to correct pathogenic mutations or produce protective alleles using CRISPR-based technologies. These innovative therapies are especially adaptable for blood and immune disorders, where clinical methods allow hematopoietic stem cells (HSCs) to be mobilized, harvested, engineered ex vivo, and transplanted back into a patient to permanently replace their blood system. This paradigm has been exemplified with the first FDA-approved CRISPR-Cas9 therapy for sickle cell disease and β-thalassemia, exa-cel (Casgevy). Although promising, efficient delivery of gene edits involves complicated ex vivo manipulation and toxic myeloablative conditioning. The quiescent and elusive nature of HSCs also brings associated challenges. In this Review, we explore the state-of-the-art genome editing technologies of nucleases, base editors and prime editors, which hold promise to address unmet clinical needs for patients with inherited hematological disorders. We highlight the progress made for several disorders, and discuss the challenges that remain for ex vivo and in vivo targeting of HSCs for next-generation gene therapies.

Introduction

The discovery of programmable RNA-guided nucleases from prokaryotic antiviral systems^1-3 sparked a revolution in molecular biology and medicine. In their natural habitat, clustered regularly interspaced short palindromic repeat (CRISPR) systems provide acquired resistance to bacteriophages or mobile genetic elements via RNA-mediated recognition of invading nucleic acid sequences, leading to rapid cleavage by the CRISPR-associated (Cas) nuclease^2,4. Thanks to their ease of programmability, CRISPR-Cas enzymes were repurposed to precisely install DNA double-strand breaks (DSBs) at diverse genomic loci in human cells^5,6. The portability of CRISPR-Cas nucleases to eukaryotic cells offered a wealth of opportunities to permanently rectify diverse genetic disorders by therapeutic genome editing. To realize the therapeutic potential of genome editors, safe and efficient genetic modification must be completed with minimal genotoxic effects. Although efficient sequence disruption is readily achieved in human cells, reaching therapeutic levels of precise gene correction has been more challenging. New CRISPR-based technologies, such as base and prime editors, now offer the possibility of installing small genetic modifications without requiring DSBs^7-9. These technologies could prove to be more precise than nuclease-driven genome editing. Building on these pivotal scientific developments, the ever-growing CRISPR toolkit could address unmet clinical needs for patients afflicted with genetic disorders.

Inherited hematological disorders are leading targets among monogenic diseases that could be amenable to therapeutic genome editing. Blood is an accessible tissue where hematopoietic stem and progenitor cells (HSPCs) can be temporarily mobilized from the bone marrow to the peripheral blood, collected by apheresis and CD34⁺ selection, engineered ex vivo, and cryopreserved. After reinfusion to the blood, engineered hematopoietic stem cells (HSCs) can repopulate the bone marrow to provide a permanent source of healthy hematopoietic cells as a potential lifelong cure for diverse blood disorders. Among them, sickle-cell disease (SCD) and transfusion-dependent β-thalassemia (TDT), caused by mutations in the HBB gene, are the most common inherited blood disorders worldwide with more than 300,000 patients diagnosed annually^10-12. Erythroid-specific deficiency of BCL11A, a direct repressor of fetal hemoglobin (HbF) expression, has been shown to rescue pathologic features of β-hemoglobinopathies^11,13-17. These fundamental discoveries led to the development of a CRISPR-based strategy to disrupt the +58 erythroid-specific enhancer of BCL11A in HSPCs to restore HbF expression^16,18-20. This approach eliminated vaso-occlusive episodes or the need for transfusions in many SCD and TDT patients, respectively¹², leading to the first U. S. FDA-approved CRISPR-Cas9 therapy, exagamglogene autotemcel (exa-cel; Casgevy) (Box 1). This new paradigm in molecular medicine could potentially be extended to other blood and immune disorders, including but not limited to inherited bone marrow failure syndromes²¹, immunodeficiencies^22-25, lysosomal storage diseases²⁶, and even acquired diseases such as HIV infection^27,28.

Box 1 – Exa-cel.

In November 2023, the United Kingdom Medicines and Healthcare products Regulatory Agency (MHRA) approved exagamglogene autotemcel (exa-cel) for the treatment of sickle-cell disease and transfusion-dependent β-thalassemia, followed by the U.S. Food and Drug Administration (FDA), the European Commission, and Health Canada. These landmark approvals stemmed from decades of fundamental research into hemoglobin biology²³⁸. In the mid-20^th century, an astute clinician hypothesized that sickle cell anemia disease manifestations developed only after the age of 3-6 months because of the protective influence of HbF²⁸². 60 years later, BCL11A was identified as the key molecular regulator of fetal to adult hemoglobin switching^13,14,17,144. Common genetic variation at BCL11A associated with HbF level and sickle cell disease severity inspired the development of therapeutic gene editing strategies that disrupted regulatory sequences to restore the expression of HbF^16,18,20. A promising CRISPR-Cas9 guide RNA identified via saturation mutagenesis of an erythroid-specific BCL11A enhancer²⁰ was licensed non-exclusively to Vertex and CRISPR Therapeutics. Clinical trials of ex vivo gene editing showed efficient and persistent gene editing in circulating blood cells, reflecting frequent genetic modification of engrafting HSCs. Treated patients had substantially elevated HbF, reversal of anemia, and in many patients elimination of transfusion requirements (in TDT) and pain episodes (in SCD). The translation of exa-cel from the bench to the bedside was relatively swiftly and smoothly executed, laying a foundation for additional CRISPR-based medicines in the clinical investigation pipeline (see Table 1 for selected examples of clinical trials for inherited blood disorders using CRISPR-based technologies).

Open questions include thorough analysis of the distribution of gene edit alleles, including structural variants at the on-target locus and possible off-target effects associated with human genetic diversity¹⁹³; the clonal composition and dynamics of HSCs after the procedure; detailed hematological investigation of red blood cell maturation and survival after editing; and comprehensive and long-term clinical outcomes, especially for SCD where multi-organ insidious deterioration is the natural history and organ damage might have occurred prior to therapy. Broadening access of exa-cel to many patients faces a number of barriers, including patient awareness and acceptance of the toxicities, complexities and uncertainties of the therapy, high short-term financial costs, the need for logistically complex and cumbersome stem cell collections and ex vivo cell manufacturing, and availability of hospitals and blood banks to support myeloablative hematopoietic transplant. Of note, clinical efficacy has also been reported with zinc finger nuclease-mediated disruption of a BCL11A enhancer²⁸³, and Cas9- or Cas12a-mediated editing at the HBG1/HBG2 promoters^284,285. Although encouraging, these other clinical trials were suspended due to reprioritization by commercial sponsors, reflecting the challenging landscape for developing high cost personalized one-time gene therapies.

This review discusses recent advances towards the development of CRISPR-based therapies for inherited blood disorders, highlighting primary obstacles that hinder safe and efficient gene editing in HSPCs. Gene therapies of HSPCs for inherited blood disorders using integrating viral vectors have been reviewed elsewhere^29,30 and are minimally discussed. Although new genome editing platforms are emerging, such as recombinase-based^31,32 and epigenome editing^33,34, here we focus on nucleases, base editors, and prime editors, which have been successfully used in HSPCs and extensively characterized in terms of genotoxicity. We present selected examples of therapeutic genome editing strategies developed for three groups of inherited blood disorders. We note that many other inherited blood disorders, including primary immunodeficiencies and bone marrow failure syndromes, could also warrant consideration for therapeutic genome editing, but a detailed discussion of these conditions falls beyond the scope of this review. We point out potential pitfalls related to the complexity of HSC biology and attempt to connect lessons learned to overcoming these key challenges. Finally, we discuss potential directions to improve broad patient access to state-of-the-art gene and cell therapies.

Genome editing technologies

The first proof-of-concept for mammalian genome editing was demonstrated thirty years ago when Rouet et al. introduced DNA DSBs in mouse cells using the rare-cutting endonuclease I-SceI³⁵. Because chromosomal DSBs must be repaired, this landmark work provided evidence that targeted DSBs could generate small insertions and deletions (indels) via non-homologous end joining (NHEJ) and, in the presence of a DNA donor, could stimulate homologous recombination³⁵ (Fig. 1a). Two years later, the first designer nucleases were engineered by fusing zinc finger proteins to the cleavage domain of the FokI endonuclease³⁶, allowing genome editing in human cells^37,38. The therapeutic potential of zinc finger nucleases in hematopoietic cells was demonstrated by targeting a corrective complementary DNA to the IL2RG locus for X-linked severe combined immunodeficiency (X-SCID) in HSPCs³⁹, or disrupting CCR5 in T cells^40,41. After breaking the DNA-binding code of transcription-activator-like effectors (TALEs) from Xanthomonas^42,43, TALE proteins were also fused to the catalytic domain of FokI to generate designer nucleases for genome editing in human cells⁴⁴. Together, zinc finger and TALE nucleases paved the way for therapeutic genome editing of HSPCs to remedy severe blood disorders. While promising, these first- and second-generation technologies required extensive protein engineering to achieve efficient targeted genome modification. The discovery of readily programmable RNA-guided CRISPR-Cas nucleases ignited the field of therapeutic genome editing.

Figure 1. Precision genome editing in human cells.

The components of the nuclease, base editors, and prime editors are illustrated on the left, and the intended editing outcomes are on the right. (a) The Cas9 nuclease from Streptococcus pyogenes (SpCas9) is guided by a single-guide RNA (sgRNA) and introduces a DNA double-strand break (DSB) via its RuvC and HNH domains. In the absence of a DNA donor, DNA DSBs are repaired via non-homologous end joining (NHEJ) or microhomology-mediated end joining (MMEJ), generating small insertions and deletions (indels) at the target site. In the presence of a DNA donor, DNA DSBs can be repaired via homology-directed repair (HDR) to install precise modifications or integrate a transgene of interest. (b) Base editors are composed of a nickase fused to a deaminase. Guided by an sgRNA, cytosine base editing (CBE) installs C-to-T substitutions and adenine base editing (ABE) installs A-to-G substitutions. (c) Prime editors are composed of a nickase fused to a reverse transcriptase (MMLV RT), and they are guided by a prime editing guide RNA (pegRNA) harboring a 3’ extension encoding the modification of interest. The latest generation prime editor 7 (PE7) also contains a truncated La protein (residues 1-194) which interacts with and stabilizes pegRNAs. The different types of prime editing systems, their DNA intermediates, and strategies that promote permanent incorporation of the intended edit to the genome are illustrated. The standard PE2 system installs a single-strand break (SSB). The PE3 system uses an additional nick sgRNA to introduce a second SSB on the opposite DNA strand to bias DNA repair towards incorporating the modification of interest to the opposite DNA strand, although increasing undesired insertions and deletions on-target. Alternatively, the installation of C•C mismatches and/or additional silent mutations promotes mismatch repair (MMR) evasion and thus the permanent incorporation of the intended edit without staggered DSB. For twin prime editing, a pair of pegRNAs are used to synthesize complementary flaps on opposite DNA strands, which anneal together and replace the endogenous genomic sequence after the excision of the unedited duplex, fill-in synthesis, and ligation. The protospacer-adjacent motif (PAM) is illustrated in blue. UGI, Uracil Glycosylase Inhibitor.

CRISPR-Cas nucleases

The evolutionary arms race between bacteria and their viruses, also known as bacteriophages, provides powerful tools for molecular biology. CRISPR-Cas systems are adaptive immune systems present in ~85% and ~40% of archaeal and bacterial genomes, respectively⁴⁵. These diverse bacteriophage defense systems come with different flavors, and Class 2 systems have been widely engineered for biotechnology applications due to their single-protein effector module^45-47. Streptococcus pyogenes Cas9 (SpCas9) is the most widely used and characterized nuclease for therapeutic genome editing (Fig. 1a). SpCas9 genome binding depends on the initial recognition of its NGG protospacer adjacent motif (PAM), which triggers DNA strand separation to initiate base pairing between the 20 nucleotides single-guide RNA (sgRNA) spacer and the targeted genomic site^48,49. Upon sufficient Watson-Crick base pairing, the nuclease undergoes conformational activation and introduces a DNA DSB via its HNH and RuvC catalytic domains^50,51. Nuclease-driven DSBs are primarily repaired by end joining mechanisms via direct ligation of the two DSB ends with or without end processing⁵². Although the latter would restore the SpCas9 recognition sequence, end processing and repair through NHEJ or microhomology-mediated end joining (MMEJ) can lead to indels (Fig. 1a), disrupting the target site and preventing successive rounds of nuclease cleavage⁵². Since NHEJ takes place with rapid kinetics throughout the cell cycle^53,54, nuclease-driven genome editing enables highly efficient gene disruption in human cells, including in quiescent HSPCs^18,55.

Therapeutic genome editing employing the widely used SpCas9 nuclease is constrained by the requirement for an NGG PAM, and its large size (≈ 4.1 kb) also poses challenges for in vivo delivery. Smaller CRISPR-Cas nucleases, including but not limited to SaCas9, St1Cas9, NmeCas9, and Cas12a (Cpf1) have been adopted to facilitate in vivo delivery via a single adeno-associated virus (AAV) and to expand the targeting range of CRISPR-Cas systems through alternative PAM sequences^56-60. The expanded CRISPR toolkit could also offer alternatives in cases of preexisting adaptive immunity⁶¹. While the field has leveraged the biological diversity of natural CRISPR-Cas systems, the high efficiency of SpCas9 has also motivated engineering efforts to expand its targeting range. Rational structure-guided engineering and phage-assisted continuous evolution campaigns have led to the development of multiple SpCas9 variants with altered PAM sequences^62-65. Notably, these engineering efforts led to the SpG variant (NGN PAM) and SpRY, the first ‘deprogrammed’ SpCas9 variants that can accept almost all NNN PAMs, although NYN (where Y is C or T) PAMs are less effective^66,67. Finally, extensive efforts have enabled the generation of high-fidelity variants with improved specificity^68-70, including HiFi Cas9, which has enabled highly efficient and precise gene editing in HSPCs when delivered as a ribonucleoprotein complex⁷¹. Collectively, these scientific advances have significantly expanded the CRISPR toolkit available for therapeutic genome editing.

Targeting a therapeutic transgene of interest to a specific genomic locus to complement a defective gene holds promise for diverse blood disorders. Co-delivering a DNA donor with SpCas9 allows for precise modifications or the integration of a transgene through homology-directed repair (HDR) (Fig. 1a)⁵². One of the main limitations of HDR is its dependence on DNA end resection, which occurs with slower kinetics than NHEJ and is limited to the S and G2 phases of the cell cycle^52,53. Because long-term engrafting HSCs (LT-HSCs) are mostly quiescent, this cell cycle limitation poses challenges for maintaining long-term engraftment of HDR-edited HSCs^72-75. This obstacle among others, such as the toxicities of the template donor, the electroporation, and the cellular response to DSBs, was highlighted by the CEDAR clinical trial for SCD, which was paused following a serious adverse event of pancytopenia after the first patient was dosed⁷⁶. Of importance, the HDR donor was co-delivered via AAV6 transduction, which potentially increased cytotoxicity compared to NHEJ-based methods⁷⁷. Genome editing with AAV donor templates have also been linked to unintended on-target integrations of concatemers and inverted terminal repeats, which are challenging to detect with standard genotyping methods^78,79. Alternative HDR donor templates, such as linear single-stranded DNA (ssDNA)^73,80, circular single-stranded DNA (cssDNA)⁸¹, and integrase-defective lentiviral vector (IDLV)⁷⁷, have also been successfully used in HSPCs. Altogether, while progress has been made to improve HDR in HSPCs, technologies that take advantage of cellular DNA repair pathways active throughout the cell cycle could facilitate therapeutic genome editing of quiescent LT-HSCs in situ.

Base editors

Early in the CRISPR craze, it became evident that a catalytically inactive Cas9 (dCas9) could be fused to proteins of interest to target them to specific genomic loci without introducing DSBs^82-85. Building on this, two different groups engineered fusions of an APOBEC1 or PmCDA1 cytidine deaminase with dCas9 to perform targeted DNA base editing installing C•G to T•A substitutions without DSBs (Fig. 1b)^7,86. A key limitation of cytosine base editing (CBE) is the generation of a U•G intermediate recognized by Uracil DNA Glycosylase (UGI), which catalyzes removal of the U and initiates base excision repair (BER). To overcome this issue, a small UGI inhibitor protein from Bacillus subtilis bacteriophage PBS1 was fused to the cytosine base editor. Additionally, the HNH domain of Cas9 was restored (SpCas9-D10A nickase) so that it could nick the non-edited strand to preferentially resolve the U•G into the desired T•A substitution⁷. The resulting BE3 editor achieved efficient C•G to T•A editing by introducing a single-strand break (SSB).

Half of known pathogenic point mutations in humans are caused by spontaneous cytosine deamination resulting in C•G to T•A transitions⁸. To potentially correct these pathogenic mutations, researchers engineered an adenine deaminase for programmable A•T to G•C base editing⁸. Protein evolution of an E. coli transfer RNA adenosine deaminase (TadA) yielded an enzyme able to catalyze adenine deamination to inosine on DNA substrates (Fig. 1b)⁸. Whereas uracil intermediates are readily recognized and removed from DNA via BER, inosine intermediates are inefficiently repaired, and no BER inhibitors are needed for efficient adenine base editing (ABE)⁸.

A key feature of base editors is their editing window. Structural studies of SpCas9 primed for DNA cleavage revealed that the PAM-distal nucleotides of the non-targeted DNA strand are disordered⁸⁷, releasing a stretch of ssDNA accessible for deaminases upon target recognition. Standard-window base editors usually target nucleotides at positions 4-8 of the non-targeted DNA strand (based on the 5’ position of the protospacer numbered 1) and multiple variants with narrower or wider editing windows have been developed, as reviewed elsewhere⁸⁸. Therapeutic base editing thus depends on the identification of the right variant and spacer to install the substitution of interest without modifying proximal bystander bases, although additional silent bystander mutations can be tolerated in some cases.

A key challenge for therapeutic base editing is the evaluation of complex off-target activities. All base editors can introduce sgRNA-dependent off-targets, but CBE also induces genome-wide sgRNA-independent off-targets, likely caused by the overexpression of a cytidine deaminase^89-91. In addition, both CBEs and ABEs can induce transcriptome-wide guide-RNA-independent deamination of RNA^92,93. Also, nicks can occasionally be converted to DSBs⁹⁴. Further engineering efforts have improved base editors’ efficiency and specificity, yielding state-of-the-art CBEs and ABEs with optimized architectures, evolved deaminases, and lower RNA off-target activity^92,93,95-101. Overall, these engineered enzymes allow highly efficient base editing in HSPCs with few DSBs^19,102-104. Although promising, therapeutic base editing of HSPCs is limited by the editing window, the presence of bystander edits, and the nucleotide substitution capabilities of base editors. Hence, many inherited blood disorder mutations are not candidates for correction by current base editing platforms.

Prime editors

Rewriting genetic codes without DSBs could potentially correct pathogenic mutations caused by base transversions between purine and pyrimidines, as well as small insertions and deletions. To expand the genome editing toolkit, a genome rewriting enzyme was engineered by fusing the SpCas9-H840A nickase with the Moloney Murine Leukemia Virus (MMLV) reverse transcriptase (Fig. 1c)⁹. Appending a 3’ extension harboring a primer binding site (PBS) and a reverse transcriptase template (RTT) to the sgRNA allowed priming and targeted reverse transcription in human cells (Fig. 1c)⁹, a process reminiscent of template-primed reverse transcription¹⁰⁵. Known as prime editing, this versatile technology allows the installation of small indels and all possible nucleotide substitutions. One key challenge is the permanent incorporation of the edit encoded on the 3’ DNA flap intermediate into the genome. Since the prime editing-synthesized flap triggers DNA mismatch repair (MMR)^106,107, the original sequence is easily restored when using the strategy known as prime editing 2 (PE2) where only one SSB is introduced (Fig. 1c). Co-delivering an additional nick sgRNA to install a second SSB on the opposite DNA strand biases DNA repair towards incorporating the modification of interest, a strategy denoted as PE3⁹ (Fig. 1c). Although this approach substantially improves prime editing efficiency, it induces staggered DSBs, which come at the cost of increased small indels, tandem duplications, and rare events of translocations and LINE-1 retrotransposon insertions^{9,106,108,109}. Alternatively, the PE3b strategy⁹ uses a nick sgRNA that matches the edited strand rather than the original allele to prevent concurrent SSBs and reduce the unwanted consequences of DSBs.

During MMR, DNA mismatches and insertion–deletion loops are first recognized by the MutSα (MSH2–MSH6) or MutSβ (MSH2–MSH3) heterodimers, respectively^110,111. MSH2 then recruits the MutLα (PMS2-MLH1) heterodimer, an endonuclease that catalyzes the incision of the nicked DNA strand around the mismatched heteroduplex¹¹². In the final steps, EXO1 excises the heteroduplex, polymerase δ synthesizes the excised DNA strand, and Ligase 1 seals the nascent strand^113-115. A dominant-negative hMLH1 protein can be co-delivered to inhibit MMR with (PE5) or without (PE4) a nick sgRNA¹⁰⁶. However, co-delivering dominant-negative hMLH1 as mRNA does not appear to improve editing efficiency in HSPCs^116,117. Moreover, cellular inhibition of MMR could be mutagenic, as recently reported with increased off-target indels at low-complexity genomic regions in mouse embryos¹¹⁸. Alternatively, incorporating additional silent mutations near the intended edit^106,119 or introducing C•C mismatches, which are resistant to MMR^120-122, markedly improve prime editing efficiency at multiple loci in HSPCs¹¹⁷. Strikingly, editing efficiency at the B2M gene in HSPCs rose from an average of 2% to 25% and 57% with one and two C•C mismatches, respectively¹¹⁷. This simple MMR-evading approach allows highly efficient prime editing with prodigious product purity in HSPCs¹¹⁷, bypassing the need for staggered DSBs. These edits with several mismatches could be suitable for recoding regulatory elements and correcting mutations, along with silent mutations to boost efficiency, but they constrain flexibility to discretely edit single nucleotide variants throughout the genome.

The latest generation prime editor, PE7, a fusion between PEmax¹⁰⁶ and the small RNA-binding exonuclease protection factor La (Fig. 1c), also improves editing efficiency in HSPCs without an additional nick sgRNA by stabilizing the prime editing guide RNAs (pegRNAs)¹²³. Although the exact mechanism by which La enhances prime editing is not yet fully elucidated, PE7 functionally interacts with the 3’ polyuridine tracts and likely protects La-accessible pegRNAs from exonuclease degradation¹²³. Given the challenges in producing long GMP-grade synthetic RNAs, such as engineered pegRNAs (epegRNAs) harboring a structural protection motif¹²⁴, using PE7 in combination with shorter La-accessible pegRNAs could facilitate clinical translation.

The possibility of rewriting DNA offers opportunities to recode exon-sized sequences. Although standard prime editing is limited to small genetic changes, paired pegRNAs strategies, such as twin prime editing (TwinPE), have been developed to integrate larger modifications (30 – 250 bp) or deletions (< 10 kb) by synthesizing complementary flaps on opposite DNA strands (Fig. 1c)^125-129. Since TwinPE does not produce heteroduplexes that engage MMR¹²⁵, this approach likely evades MMR by design. Of importance, highly efficient TwinPE was observed when modulating nucleotide metabolism in quiescent HSPCs that were not stimulated with cytokine culture prior to electroporation¹¹⁷. Although the activity of prime editing systems during different phases of the cell cycle has yet to be determined extensively, TwinPE potentially occurs efficiency throughout the cell cycle, which could enable efficient therapeutic genome editing in LT-HSCs. However, the cellular response of HSCs to different prime editing systems is not fully understood. The genotoxic effects of TwinPE-induced staggered DSBs should be thoroughly characterized, because installing these larger modifications via concurrent SSBs could come at the cost of unwanted indels or larger deletions at the target site and inducing a p53 response.

Emerging CRISPR-based platforms for targeted transgene integration

Achieving safe and efficient targeted therapeutic transgene integration in HSCs is no easy task. Approaches that couple prime editing-mediated installation of a recombinase landing pad into the genome with the introduction of a recombinase to precisely integrate large DNA sequences hold potential for gene therapy. Denoted as PASSIGE^125,130 and PASTE¹³¹, these two-step methods are based on the installation of attB or attP landing pads via prime editing followed by transgene integration using Bxb1 recombinase, allowing the integration of large DNA cargoes over 5 kb in human cells. In PASTE, the Bxb1 recombinase is fused to the C-terminal region of the prime editor, whereas in PASSIGE the two components are delivered individually^{125,128,130,131}. Because Bxb1 tethering to PEmax decreases prime editing efficiency, co-delivering the prime editor and the recombinase separately improves site-specific integrations¹³⁰. Although progress has been made with evolved Bxb1 recombinases^130,132, therapeutic transgene integration has yet to be achieved in HSCs. Other emerging platforms include CRISPR-associated transposases^88,133. In their current form, recombinase and transposase modalities depend on plasmid donors, which are highly toxic in primary hematopoietic cells. Because naked double-stranded DNA (dsDNA) can trigger pattern recognition receptor pathways such as the cGAS–STING pathway, co-delivering a dsDNA donor induces cell death in HSCs^134,135. Alternatively, ssDNA and cssDNA donors have limited immunogenicity and are thus better tolerated in HSCs^73,80,81,134. Given this immunological barrier, genome editing modalities that rely on ssDNA donors or evade dsDNA sensing might improve targeted therapeutic transgene integration in HSCs.

Therapeutic opportunities for hematological disorders

β-hemoglobinopathies

Sickle cell anemia was dubbed the first molecular disease in 1949¹³⁶. Less than a decade later, the pathogenic amino acid substitution in sickle hemoglobin was identified as a gene mutation inherited in a strictly Mendelian manner¹³⁷. Inherited mutations in the β-globin gene (HBB) lead to the most common inherited blood disorders worldwide, with approximately 60,000 and 300,000 annual diagnoses for TDT and SCD, respectively^11,12,138. TDT mutations result in an imbalance between the α and β-globin chains of adult hemoglobin, causing ineffective erythropoiesis^12,138. The HBB-E6V SCD mutation results in the polymerization of deoxygenated sickle hemoglobin, causing erythrocyte deformation, hemolysis, and vaso-occlusive episodes^11,12. Matched sibling donor allogenic HSC transplantation (HSCT) offers a curative option with >90% survival rate, although this approach is unavailable to most patients^139,140. Alternative donor allogeneic transplantation remains an area of active clinical investigation¹⁴¹. Viral gene addition has also shown promising clinical efficacies, as reviewed elsewhere²⁹, and provided a foundation for gene therapy of β-hemoglobinopathies. Challenges related to integrating viral vectors include the requirement of multiple vector copies per cell to achieve high-level expression, with some patients developing imbalanced globin expression¹⁴², which indicates that endogenous gene regulation would be preferred.

Restoring HbF expression offers a universal therapeutic strategy for all mutations causative of β-hemoglobin disorders, bypassing the need to develop personalized gene editing approaches for each of more than 350 pathogenic mutations¹³⁸. Human genetics studies of rare and common variants associated with HbF level and with the clinical severity of TDT and SCD have helped pinpoint the genetic elements that silence HbF in adults, including the BCL11A erythroid enhancers and HBG1/2 promoters^{13,16,143-146}. Disrupting the +58 or +55 erythroid-specific enhancers of BCL11A¹⁶ via nuclease-driven indels or base editing has been shown to restore HbF expression and ameliorate disease phenotypes (Fig. 2a,b)^{12,18-20,147-149}. Furthermore, naturally occurring mutations in the promoters of HBG1 and HBG2 have been linked to hereditary persistence of HbF (HPFH)¹⁵⁰, and SCD patients who express high levels of HbF present with much milder symptoms¹⁵⁰. Therefore another approach has involved installing HPFH or HPFH-like mutations at the HBG1 and HBG2 promoters via base editing, which generates binding sites for the activating transcription factors TAL1, GATA1 or KLF1 and/or disrupts the binding of repressors such as BCL11A or ZBTB7A (Fig. 2b). This approach also allows potent HbF induction to alleviate TDT and SCD^103,151,152. Diverse approaches have also been developed to directly correct the SCD mutation using nuclease-driven HDR^73,153, base editing¹⁰², and prime editing^116,154 (Fig. 2b,c). Although the SCD HBB-E6V allele cannot be restored to wild-type using base editing, custom adenine base editor variants recognizing non-canonical PAMs were used to convert the SCD allele into the non-pathogenic Makassar allele (HBB-E6A) (Fig. 2b)^65,102,155. Finally, ex vivo and in vivo prime editing approaches were developed to restore the expression of WT β-globin (Fig. 2c)^116,154. Although clinical hurdles were encountered using an HDR approach to directly correct the SCD allele⁷⁶, these alternative base and prime editing strategies could overcome challenges related to the cell cycle dependence of HDR⁵³ and the toxicity of the donor template⁷⁷.

Figure 2. Selected examples of therapeutic genome editing strategies for β-hemoglobinopathies.

(a) Nuclease-driven editing at the erythroid-specific +58 and +55 enhancers of BCL11A. Targeted disruption of the GATA1 binding motif at BCL11A restores HbF expression. (b) Base editing strategies for sickle-cell disease and β-thalassemia. Targeting the +58 BCL11A enhancer or the HBG1/HBG2 promoters with cytosine base editing (CBE) or adenine base editing (ABE) restores the expression of HbF. Base editing at HBB with the protospacer-adjacent motif (PAM)-reprogrammed ABE-NRCH allows the correction of the sickle-cell disease HBB-E6V mutation to the benign Makassar allele (HBB-E6A). (c) Prime editing (PE) at HBB allows the correction of the sickle-cell disease HBB-E6V mutation. The primer binding site (PBS) and the reverse transcriptase template (RTT) from the pegRNA 3’ extension are shown. Additional nick sgRNAs for the PE3 or PE3b strategies are shown. The PE3b nick sgRNA only recognizes its target after the installation of the HBB-E6V correction edit via prime editing, decreasing concurrent single-strand breaks, and consequently, indels.

It can be anticipated that as novel therapies emerge with various delivery modalities (see below), patients with unique hematologic states will be described each containing varied mixtures of gene edit distributions across HSCs and hemoglobin distribution across erythroid cells. Future clinical research with detailed and long-term assessment of molecular, hematologic, and clinical outcomes will be required to determine minimal therapeutic thresholds for overall and per-cell editing and hemoglobin distribution based on each editing approach. However, encouraging results have been observed in clinical studies of exa-cel, with average ~75% indels at the BCL11A enhancer and ~40% total HbF in SCD patients, setting a point of comparison for future studies¹⁴⁸. Together, significant progress has been made towards therapeutic gene editing for β-hemoglobinopathies, leading to the first FDA-approved CRISPR-Cas9 therapy (Box 1) and paving the way for other inherited blood disorders.

Primary immunodeficiencies

Inborn errors of immunity are rare blood disorders characterized by defects in immune cell functions and susceptibility to severe infections. Among them, chronic granulomatous disease (CGD) is caused by defective function of the nicotinamide adenine dinucleotide phosphate (NADPH) oxidase complex responsible for phagocyte respiratory burst and superoxide production. Mutations in genes encoding components of the complex (such as the phox subunits) lead to inflammatory complications and recurrent invasive infections, the major cause of morbidity and mortality in CGD patients¹⁵⁶. HDR and ABE strategies have been developed to correct the prototypical X-linked CGD mutation in the CYBB gene (the R226X mutant of the p91^phox subunit)^104,157. In a side-by-side comparison, mice transplanted with ABE8e-SpRY-treated cells exhibited higher phenotypic correction compared to a nuclease-driven HDR method using an ssDNA donor¹⁰⁴. Although this ABE approach relies on a near PAMless variant (SpRY)⁶⁶, delivering the high-fidelity ABE8e-SpRY-HiFi variant as mRNA mitigated guide RNA (gRNA)-dependent off-target editing events observed with ABE8e-SpRY. Together, these investigational new drug-enabling studies — including the profile of potential off-target DNA edits, transcriptome-wide RNA edits, and chromosomal perturbations — provided support for initiating a first-in-human clinical trial for X-CGD using base editing (NCT06325709) (Table 1)¹⁰⁴. Nuclease-driven HDR approaches have also been developed to correct the common delGT mutations in NCF1 (p47^phox CGD), but frequently result in chromosomal deletions and rearrangements between NCF1 and its pseudogenes (NCF1B and NCF1C)^158,159. Although p47^phox CGD is not amenable to base editing, nucleotide insertion via prime editing offers a promising alternative. Indeed, the first-in-human clinical trial for prime editing (NCT06559176), sponsored by Prime Medicine, is designed to treat p47^phox CGD (Table 1).

Table 1. Selected examples of CRISPR-based clinical trials for inherited blood disorders.

Disease indication	Therapeutic strategy	Sponsor	Phase	Clinical trial
Transfusion-dependent β thalassemia	Nuclease-driven editing at the HBG1/2 promoters using CRISPR-Cas12b	Institute of Hematology and Blood Diseases, China	NA	NCT06041620
Sickle cell disease	Nuclease-driven correction (HDR) at HBB using CRISPR-Cas9	Kamau Therapeutics, USA	I, II	NCT04819841
Sickle cell disease	Nuclease-driven editing at BCL11A using CRISPR-Cas9	Bioray Laboratories, China	NA	NCT06287099
Hemophilia B	Nuclease-driven (HDR) coagulation factor IX gene insertion using CRISPR-Cas9	Regeneron Pharmaceuticals, USA	I, II	NCT06379789
Transfusion-dependent β thalassemia and sickle cell disease	Nuclease-driven editing at BCL11A using CRISPR-Cas9	Vertex Pharmaceuticals, USA	III	NCT05477563
Transfusion-dependent β thalassemia and sickle cell disease	Nuclease-driven combined enhancer editing at BCL11A using CRISPR-Cas9	Boston Children’s Hospital, USA	I	NCT06647979
X-linked chronic granulomatous disease (X-CGD)	Gene correction at CYBB using base editing	National Institute of Allergy and Infectious Diseases (NIAID), USA	I, II	NCT06325709
Sickle cell disease	Modification of the HBG1/2 promoters using base editing	CorrectSequence Therapeutics, China	I	NCT06565026
Sickle cell disease	Modification of the HBG1/2 promoters using base editing	Beam Therapeutics, USA	I, II	NCT05456880
Chronic granulomatous disease (CGD)	Gene correction at NCF1 using prime editing	Prime Medicine, USA	I, II	NCT06559176

The NIH National Library of Medicine clinical trials database (ClinicalTrials.gov) was searched on January 5^th, 2025, for active clinical trials using the keywords "CRISPR", "Hematopoietic stem cells", "Base editing", and "Prime editing". NA, not applicable.

Among inborn errors of immunity amenable to diverse CRISPR-based technologies, severe congenital neutropenia (SCN) is characterized by impaired neutrophil maturation, which leads to life-threatening infections in the absence of treatment. Mutations of the ELANE gene, encoding neutrophil elastase, account for around half of the cases of SCN as well as the related condition cyclic neutropenia (CyN)¹⁶⁰. The only curative treatment for this blood disorder, allogeneic HSCT, is limited by donor availability and immune compatibility, with risks of graft-versus-host disease or graft rejection¹⁶¹. A leading model of pathogenicity suggests that abnormal neutrophil elastase proteins cause an unfolded protein response and endoplasmic reticulum stress, resulting in promyelocyte apoptosis and differentiation arrest at the promyelocyte-to-myelocyte transition^162-164. We and others have previously dissected ELANE neutropenia pathogenicity and found that nuclease-driven indels triggering nonsense-mediated decay circumvent neutropenia mutations^24,165. The promising target sequences identified offer a potential universal nuclease-based approach for therapeutic gene editing of all pathogenic ELANE mutations^24,165 that could be portable to base and prime editing. Alternatively, inhibition of ELANE mRNA through the introduction of a staggered DSB initiated bv two gRNAs targeting the promoter TATA box using SpCas9 nickase (D10A) also restored neutrophil maturation in patients’ HSPCs¹⁶⁶.

Severe combined immunodeficiencies (SCIDs) are another group of disorders which cause profound lymphocytopenia and failure to thrive without treatment. Similar to β-hemoglobinopathies, the overall survival rate after HSCT from an HLA-matched sibling donor is greater than 90%, but available for less than 20% of patients^167-169. HSCT from alternative donors is associated with increased risks of graft-versus-host disease and incomplete immune reconstitution^167-169. Gene therapies have been developed to provide a functional copy of a defective gene causing SCID. However, therapeutic transgene integration via gamma-retrovirus transduction has been linked with insertional oncogenesis^170-173, although stem cell gene therapies for SCID have an improved safety profile with current-generation lentiviruses^169,174. Nonetheless, in certain contexts, such as with the use of strong, viral-derived enhancer/promoter elements, lentiviral therapies can still be associated with insertional mutagenesis risk^175,176. Recently, cases of hematologic cancer following gene therapy for cerebral adrenoleukodystrophy were reported in 7 out of 67 patients¹⁷⁷. In 6 of these patients, expanded clones were found to contain at least one confirmed insertion site in the MECOM oncogene. Importantly, an internal MNDU3 promoter-enhancer included in the lentiviral vector to drive therapeutic-level expression of the deficient ABCD1 gene product might have contributed to insertional oncogenesis¹⁷⁷. Extensive characterization of the mechanisms of clonal expansion and transformation is needed to better understand and mitigate these risks.

Targeted integration of a functional copy of a defective gene to its endogenous locus under physiological transcriptional control holds promise as an alternative to semi-random lentiviral integrations and could improve both the safety and efficacy of gene correction¹⁷⁸. This strategy has been used to functionally complement diverse SCID-causing genes, such as IL2RG^23,39,178, RAG1¹⁷⁹, and RAG2^180,181. These therapeutic avenues rely on HDR and could prove challenging to translate into the clinic⁷⁶. On the other hand, developing personalized base or prime editing therapies for every pathogenic mutation causing SCID would require an immense amount of time and resources. Despite this barrier, developing gene therapies for orphan diseases remains important. McAuley et al. developed an ABE strategy to correct a CD3δ SCID mutation prevalent in rural Mennonite communities of North America, comprising over 20% of SCID genotypes in Alberta, Canada²⁵. As CRISPR-based gene therapies progress towards the clinic, developing personalized treatments to address unmet clinical needs for patients afflicted with very rare diseases should remain a priority and could benefit from a streamlined process.

Bone marrow failure syndromes

Diverse defective cellular processes, including ribosomal biogenesis, DNA repair, and telomere maintenance, have been linked with failure of the bone marrow to produce blood. Collectively known as inherited bone marrow failure syndromes (IBMFS), patients afflicted with these syndromes are at increased risks of developing blood cancers or life-threatening infections. Among these syndromes, Fanconi anemia (FA) has been studied as a potential target for therapeutic genome editing. One fundamental obstacle lies in the mechanism of this disease. Normally implicated in DNA interstrand crosslink repair, the FA DNA repair pathway is also needed for Cas9-induced single-strand template repair (SSTR)¹⁸². While NHEJ-driven compensatory mutations could correct the phenotype in certain cases of FA¹⁸³, precise HDR gene correction using an ssDNA donor would be compromised in patient cells¹⁸². Alternatively, ABE has been used to functionally restore two prevalent mutant alleles of FANCA, which account for 60-65% of pathogenic FANCA mutations²¹. Of importance, base-edited HSPCs showed proliferative advantages²¹, corroborating previous findings of successful engraftment of gene-corrected HSCs in FA patients without prior cytotoxic conditioning regimen¹⁸⁴. Therapeutic genome editing of IBMFS, where CRISPR-engineered HSCs would have a selective advantage over non-corrected cells, holds potential for the development of chemotherapy-free gene and cell therapies.

Therapeutic challenges

Genotoxic risks

Engineering the human genome poses potential safety risks. High levels of genotoxicity could lead to oligoclonal hematopoiesis and compromised graft resilience, potentially elevating the risk of hematologic cancer development. Evaluating and mitigating genotoxic risks associated with genome editing procedures, as well as optimizing protocols to achieve higher yields of gene-edited HSCs, are of paramount importance for ensuring the safety of CRISPR-based gene therapies for severe blood disorders. Long-term clonal tracking of gene-edited HSCs through the installation of synthetic barcodes¹⁸⁵ and/or monitoring natural barcodes¹⁸⁶ could also serve as a valuable tool to follow clonal composition and dynamics. Although progress has been made to evaluate and mitigate genotoxic risks, assays to determine the biological and clinical importance of any genomic effects are often lacking.

Multiple research groups have demonstrated that programmable RNA-guided genome editing tools can introduce undesired modifications at off-target loci (Fig. 3) and have developed unbiased detection methods for these modifications based on searching the genome in silico, in vitro, or in cellulo^187-192. Complementary approaches have also been developed to evaluate genome-wide and transcriptome-wide sgRNA-independent off-target activities of base editors^89-93. One key consideration for clinical translation is human genetic diversity, as allele-specific off-targets can be concentrated in an ancestral group or be unique to an individual¹⁹³. In the case of BCL11A enhancer editing, we found a genetic variant at CPS1 that creates a likely off-target site and is concentrated in individuals of African ancestry. It is expected to be carried by approximately 10% of patients with SCD¹⁹³. Importantly, gene editing in HSPCs from a donor heterozygous for this off-target risk allele demonstrated that off-target indels and rearrangements were indeed produced, although they could be mitigated by using the HiFi Cas9 variant^71,193. This finding prompted the FDA to mandate a post-marketing surveillance study for exa-cel, to screen patients for the common CPS1 variant and test for allele-specific off-target editing in clinical samples. Although computational predictions allow variant-aware off-target nomination, novel methods are still needed to validate and mitigate off-target effects in cells of relevant genotypes¹⁹⁴ as well as to determine their biological consequences, if any. Combining computational predictions of human genome off-target potential accounting for genetic diversity together with biological validations should facilitate the development of safer gene therapies.

Figure 3. Potential genotoxic outcomes of genome editing in human cells.

Genome editors can perform their intended edit, or can install imprecise on-target edits, including unintended indels or bystander edits with base editing, and unintended indels or scaffold incorporation with prime editing. In presence of sufficient homology, genome editors can install unwanted edits at off-target genomic loci. Cytosine base editing can also induce gRNA-independent edits at off-target genomic loci. The generation of double-strand breaks can generate larger or complex rearrangements at on- and off-targets, such as a large deletion, chromosomal translocation, centromere distal chromosome loss, and chromothripsis.

Beyond off-target editing, additional genotoxic events include DSB-induced large deletions with extensive unidirectional end resection, and centromere-distal chromosome loss which can result in chromothripsis or the formation of micronuclei (Fig. 3)^195-198. We recently found that these unintended outcomes are a byproduct of cellular proliferation of HSPCs stimulated by ex vivo culture with cytokines⁵⁵. Similarly, chromosome loss was mitigated in patient T cells when introducing Cas9-mediated DSBs prior to T cell activation/stimulation¹⁹⁹. Micronuclei formation occurs in the M phase²⁰⁰ whereas extensive end resection occurs in S/G₂⁵³, suggesting that these genotoxic events are taking place in actively dividing cells. Increasing evidence indicates that HSCs, which are mostly quiescent in situ, possess unique DNA damage repair preferences compared to rapidly dividing progenitors^{18,39,73,75,201,202}. Given that extended ex vivo culture with cytokines increases genotoxicity and negatively correlates with HSC engraftment potential^203,204, targeting quiescent HSCs holds promise as an alternative to current clinical manufacturing protocols and could ultimately enable successful targeting of quiescent HSCs in situ.

Therapeutic gene editing using base or prime editors could mitigate DSB-induced genotoxic events. Nevertheless, using nickase-based systems decreases but does not completely abrogate large deletions and translocations⁹⁴, which are more frequent with cytidine base editors due to suboptimal inhibition of BER, likely generating DSB intermediates. Concurrent SSBs generated via the PE3 system also generate staggered DSBs. PE3-mediated paired nicking generates 5’ or 3’ ssDNA overhang intermediates, leading to deletions or tandem duplications of the sequence between the cut sites, respectively, including in HSPCs^{108,117,205,206}. Of importance, omitting the nick sgRNA (PE2 system) completely abrogated tandem duplications in HSPCs, and allowed highly efficient editing with high product purity with MMR-evading designs¹¹⁷. The PE3b system, where the nick sgRNA matches the edited strand and not the original allele⁹, minimizes the presence of concurrent nicks and offers a favorable compromise between efficiency and product purity in HSPCs when there is little flexibility for MMR evasion^116,117. Considering the genome-wide specificity of prime editors^9,207,208, these early pre-clinical results are encouraging for the goal of efficient therapeutic genome editing with minimal genotoxicity.

Targeting quiescent HSCs

In their bone marrow niche, HSCs establish a state of metabolic and cell cycle dormancy known as functional quiescence²⁰⁹. Through their unique capability to self-renew and differentiate, HSCs ensure the lifelong production of all blood cells by delicately balancing pro-survival and stress-response mechanisms to restrain lineage commitment and support functional G₀ phase cell cycle dormancy^209,210. Pharmacological regulation of mTOR and Wnt signaling allows the ex vivo maintenance of LT-HSCs, suggesting a crucial role for these pathways in regulating stem-cell renewal or lineage commitment²¹⁰. Activation of the mTOR and JAK pathways drives HSCs from quiescence to rapid cycling, promoting lineage commitment and HSC depletion^211-213. A recent study found that the c-Myc target MYCT1 limits environmental sensing to maintain HSC self renewal and engraftment ability²¹⁴. Interestingly, MYCT1 expression was markedly downregulated during ex vivo culture²¹⁴. In therapeutic gene editing settings, mobilized peripheral blood CD34⁺ HSPCs are predominantly in G₀ phase, and rapidly progress to G₁ and then S/G₂/M after 24 and 48 hours of ex vivo culture with cytokines, respectively^18,55,74. Inducing quiescence by inhibiting mTORC1 and GSK-3 after editing and a short period of cycling improves HDR in long-term engrafting HSCs⁷⁴. A chemically defined culture system has also been developed to support in vitro expansion of HSCs without conventional cytokines²¹⁵. Overall, ex vivo cell culture and cytokine stimulation can present challenges to maintaining the engraftment and self-renewal potential of gene modified HSCs.

The quiescent nature of HSCs also generates metabolic barriers for therapeutic gene editing. Ex vivo culture with cytokines drives HSCs from G₀ to G₁^18,55,74, inducing a reversible switch from metabolic dormancy to support the anabolic demands of cell division. This switch activates the translation machinery, allowing efficient genome editor expression via mRNA delivery. Since HSCs are predominantly quiescent in their niche, it is not clear whether cytokine stimulation is needed for maximal base or prime editor mRNA expression, but this could potentially limit efficient base or prime editing in vivo using current mRNA modalities. Another key metabolic obstacle is the level of nucleotides that are available for reverse transcription and thus prime editing. To support DNA replication and cell proliferation, the levels of dNTPs are orders of magnitude higher in actively dividing compared to quiescent cells^216-218. SAMHD1, a triphosphohydrolase enzyme, tightly regulates the level of dNTPs in a cell cycle-dependent fashion and acts as an antiviral factor to restrict lentiviral infection in hematopoietic cells^216-218. To counteract this defense mechanism, HIV-2 and SIV express the Vpx accessory protein, which associates with the CRL4-DCAF1 E3 ubiquitin ligase to target SAMHD1 for proteasomal degradation, thus raising dNTP levels to support viral infection. We demonstrated that modulating nucleotide metabolism by co-delivering Vpx mRNA and supplementing deoxynucleosides in the culture media enhance prime editing at the step of reverse transcription in HSPCs¹¹⁷. Altogether, future efforts at engineering HSCs in situ should consider the quiescent nature of HSCs and their metabolic constraints.

Chemotherapy-free cell therapies

Autologous gene therapies, including exa-cel, typically rely on myeloablative conditioning to facilitate the engraftment of engineered HSPCs^12,169,174. Although non-myeloablative chemotherapy can be sufficient for gene-corrected cells with a selective advantage to reach therapeutic levels, myeloablation is necessary for therapeutic indications dependent on high levels of engraftment, such as β-hemoglobinopathies and lysosomal storage disorders²¹⁹. HSC gene therapies are therefore limited to the most severe forms of blood disorders, because genotoxic conditioning prior to transplantation is associated with acute toxicities such as cytopenias, mucositis, infection and organ injury (requiring inpatient support for 1-2 months), as well as long-term toxicities such as infertility, organ toxicity, and risk of secondary malignancy^134,167. In addition, stem cell collection can be onerous. Beam Therapeutics acknowledged the death of a patient due to respiratory failure four months after BEAM-101 infusion during a phase I/II clinical trial of base editing for SCD (NCT05456880) (Table 1)²²⁰. The complications were likely related to the myeloablative conditioning step using the DNA alkylating agent busulfan²²⁰. Also, the death of a patient with SCD was reported on the exa-cel trial, and attributed to respiratory failure associated with coronavirus 2 (SARS-CoV-2) infection, along with busulfan-associated lung injury and preexisting lung disease¹⁴⁸.

Non-genotoxic conditioning regimens could improve the safety profile of autologous HSC-based therapies. Antibody-drug conjugates (ADCs) targeting antigens expressed on HSCs, like CD117/c-kit and CD45, may produce myeloablation and enable hematopoietic repopulation^221-224. Studies in non-human primates have shown that indeed such ADCs allow for durable hematopoietic replacement by gene-modified HSCs^225,226. A challenge is that no single antigen is expressed solely on HSCs—for example, CD117 is also expressed on mast cells, germ cells, and various other non-hematopoietic cells, and CD45 on mature hematopoietic cells. A clinical trial testing an anti-CD117 ADC was paused following the death of a patient from cardiorespiratory failure²²⁷, suggesting the clinical implementation of ADC-based myeloablative conditioning will require careful determination of safe and effective treatment regimens.

A promising alternative to toxic conditioning could be to mobilize existing HSPCs out of their niche while conferring a selective advantage to engineered cells, enabling efficient competitive engraftment. This paradigm has been exemplified by co-delivering mRNA to overexpress the homing receptor CXCR4 to repopulate bone marrow after prior mobilization treatment to produce niche space availability²²⁸. In addition, conferring selective advantage by multiplexed genome editing might help achieve therapeutic levels of editing without myeloablative conditioning. Targeting two endogenous loci with CRISPR-Cas9 results in double-editing events that are not statistically independent across a wide range of biological systems, including in Caenorhabditis elegans²²⁹, Drosophilia melanogaster^230,231, and human cells^108,232. Genome editing at a selectable endogenous gene enables the enrichment of a second genome modification of interest via marker-free co-selection, a process compatible with NHEJ, HDR, base editing, and prime editing^108,232. The mechanism of co-selection is not fully understood, but it likely involves a combination of factors. Co-editing events might occur in cells that have the highest editor expression while being in a similar cell cycle stage and expressing the most favorable DNA repair factors. Additionally, other metabolic features might make some cells more amenable to completing multiple editing events at different loci. Building on this, strategies can be envisioned to improve the enrichment and engraftment of gene-corrected HSCs through CRISPR-driven introduction of selectable alleles. For example, epitope base editing of surface receptors in HSPCs confers resistance to monoclonal antibodies and chimeric antigen receptor (CAR) T cells, while maintaining receptor expression and function, and has enabled targeted immunotherapies for acute myeloid leukemia, B cell lymphoma, and T cell leukemia^233,234. Multiplexed epitope and therapeutic editing is a promising approach to engineer resistance to clinically used antibodies, which could be leveraged to enrich HSCs that harbor a therapeutic edit at a second locus of interest. Considering the progress made towards delivery of genome editors to HSCs in vivo^235,236, marker-free co-selection strategies could facilitate gene therapy for severe blood disorders without ex vivo engineering and myeloablative conditioning.

Improving broad patient access

The current model for the development of gene and cell therapies poses translational challenges for their broad accessibility to patients with severe blood disorders. This is especially true for rare genetic diseases that are not viable for pharmaceutical commercialization. For common blood disorders with a universal therapeutic strategy, such as the restoration of HbF expression for β-hemoglobinopathies, the main contemporary barriers are patient acceptance of therapies that, while potentially transformational, are still associated with substantial toxicity and complexity, the healthcare infrastructure needed to administer the autologous cell therapy and the scientific and medical expertise needed to produce and administer gene therapies (Fig. 4). HSC mobilization, isolation, and ex vivo manufacturing requires extensive resources and scientific expertise. Additionally, administering the conditioning chemotherapy, transfusing engineered cells, and monitoring the patient post-transplantation necessitate at least four to six weeks of hospitalization (Fig. 4). Although potentially curative, the infrastructure and expertise needed to successfully provide the first CRISPR-based gene and cell therapies, as well as their cost^237,238, will restrict their widespread administration. Navigating the complex intellectual properties involved in therapeutic genome editing also generates an additional layer of complexity. Considering that more than 300,000 babies are born every year with β-hemoglobinopathies, developing safe and efficient delivery methods to engineer HSCs in vivo could drastically improve broad patient accessibility.

Figure 4. Overview of the ex vivo therapeutic gene editing journey for inherited blood disorders.

The administrative, clinical, and scientific steps required to administer a gene-edited hematopoietic stem cell (HSC) therapy for inherited blood disorders are summarized, from the patient preparation stage to the post-infusion monitoring and follow-up. The general manufacturing steps and examples of potential clinical complications and risks are indicated. During follow-up, detailed studies of the clonal composition of HSCs have yet to be described, but this could be a fruitful future research direction. BM, bone marrow.

In vivo delivery

The development of drug delivery systems for CRISPR-based genome editors is making remarkable progress²³⁹, but engineering HSCs in vivo generates an additional layer of complexity. A suitable delivery system would safely and efficiently deliver the genome editor transiently to HSCs with minimal off-target cellular uptake, cytotoxicity, and immunogenicity. To mitigate off-target effects, prolonged genome editor expression via integrating viral vector delivery might not be suitable for in vivo HSC engineering. In addition, vectors based on adenoviruses or adeno-associated viruses (AAVs) can pose safety risks with regards to immunogenicity, dose-limiting toxicity, broad tropism, viral integration and unwanted long-term expression, and limited redosing capacity^239-245. On the other hand, non-viral delivery systems offer promising alternatives to address these safety concerns.

The COVID-19 pandemic brought lipid nanoparticles (LNPs) (Fig. 5) and RNA therapeutics to the forefront of modern medicine. Pivotal discoveries, including the impact of mRNA nucleoside modification on the suppression of RNA recognition by Toll-like receptors²⁴⁶, enabled the development of life-saving mRNA vaccines^247,248. LNPs are an up-and-coming modality for the transient delivery of genome editors. FDA-approved LNPs are composed of an ionizable lipid, a helper lipid, a polyethylene glycol (PEG)-lipid, and cholesterol (Fig. 5). The nanoscale structure of LNPs dictates their tropism, efficiency, and safety, which motivated the development of a diverse suite of LNP chemistries to deliver genetic medicines^239,249. Standard four-component LNPs leverage the physiological cholesterol transport of lipoproteins by binding to serum ApoE, followed by hepatocyte uptake through low-density lipoprotein receptor (LDL-R)-mediated endocytosis. To overcome liver accumulation, selective organ targeting (SORT) LNPs have been developed via the addition of a defined amount of ionizable lipids, positively charged lipids, or negatively charged lipids to increase the selective delivery of therapeutic mRNA to the liver, lungs, and spleen, respectively^250,251.

Figure 5. Delivery particles for therapeutic genome editing.

(a) Adenoviral vectors are composed of an icosahedral capsid with fibers connected through a penton base and deliver a linear double-stranded DNA (dsDNA) payload. (b) Adeno-associated viral vectors are composed of an icosahedral capsid that packages and delivers a single-stranded DNA (ssDNA) payload flanked by two inverted terminal repeats. (c) Moloney murine leukemia virus (MMLV) virus-like particles (VLPs) are pseudotyped with VSV-G envelope glycoprotein, and formulated with MMLV-gag fused to MMLV-protease-polymerase (Gag-Pol) or loaded with genome editor ribonucleoprotein (RNP) complexes. Coupling a VSV-G mutant with a single-chain variable fragment (ScFv) targeting a cell-specific surface marker allows programmable genome editor delivery to specific cell types²⁷³. (d) Lipid nanoparticles are formulated with an ionizable lipid, a helper lipid, a polyethylene glycol (PEG) lipid, and cholesterol to deliver genome editor RNAs. Serum ApoE adsorption enables cellular uptake through the low-density lipoprotein (LDL) receptor. Lipid nanoparticles can also be coupled to a ScFv targeting a cell-specific surface marker to increase specificity²³⁵.

Different groups have demonstrated the portability of LNP-mediated mRNA delivery for HSC engineering ex vivo and in vivo^235,236,252. The addition of covalent lipid species to four-component LNPs created bone marrow (BM)-homing LNPs capable of transfecting 14 unique hematopoietic cell types, including HSCs²³⁶. Using a Townes (HBB^s/s) mouse model of SCD, two BM-homing LNPs doses resulted in 2.4% base editing at HBB²³⁶. Interestingly, mass spectrometry proteomics identified ApoE as the top enriched serum protein that binds BM-homing LNPs²³⁶. As quiescent HSCs do not express the ApoE-binding LDL receptor²⁵³, LNP formulations depending on ApoE-mediated cellular uptake might be challenging for in vivo editing of LT-HSCs. Alternatively, anti-CD117-conjugated LNPs targeting the stem cell factor receptor enabled in vivo mRNA delivery to LT-HSCs²³⁵. Although promising, high levels of nonspecific cellular uptake were observed in the liver and lungs after a single dose of CD117-LNPs²³⁵. Future research avenues could help LNPs reach therapeutic levels of editing while decreasing nonspecific cellular uptake through formulation engineering, repeated dosing, and stabilizing nucleic acid modifications. Among potential research directions, synthesis of therapeutic mRNA with altered topologies such as circularization^254,255 or branched chemically modified polyA tails²⁵⁶ could improve genome editor mRNA stability when higher levels of expression are needed for efficient editing.

One of the main challenges in genome editor delivery is the endosomal release of the payload, also known as endosomal escape, which remains a highly recalcitrant problem with less than 5% release efficiency^257,258. To complicate matters, the mechanism of endosomal escape is poorly understood^258,259. Among the hypotheses, temporary, short-lived, and small breaches in the endosomal lipid bilayer might allow RNA to reach the cytoplasm²⁵⁹. Some studies suggest that RNA payloads could leak out during endosome maturation as they move toward fusion with lysosomes^257,259,260. Future efforts to improve endosomal escape by disrupting the lipid bilayer locally without rupturing the endosome (which causes cytotoxicity by releasing compartmentalized proteins and molecules into the cytoplasm) should facilitate therapeutic RNA delivery²⁵⁹. The development of alternative delivery strategies to overcome this barrier, and potentially bypass endosomes altogether, such as proteolipid vehicles²⁶¹ or peptide-mediated ribonucleoprotein (RNP) complex delivery^262,263, also merits further exploration. Indeed, highly efficient delivery and genome editing have been achieved ex vivo in primary T cells and HSPCs using a cell-penetrating Cas9 RNP and a cell-penetrating endosomal escape peptide²⁶². Although efficient, further research is needed to evaluate the immunogenicity of peptide-mediated delivery systems.

Virus-like particles (VLPs) also hold potential to transiently deliver RNP complexes in vivo (Fig. 5)^264-270. These assemblies of viral proteins carry a therapeutic RNP complex cargo instead of a viral genome and are pseudotyped with an envelope glycoprotein, allowing tissue-targeting without the risks of viral genome integration. To generate VLPs, M-MLV Gag-pro-pol, Gag-linker-editor, and VSV-G envelope glycoprotein vectors are co-transfected in a producer cell line to package the therapeutic RNP cargo (Fig. 5). The VLP platforms enabled in vivo gene correction using base and prime editing^269,270, but in vivo HSC engineering has yet to be achieved. Given LDL-R serves as the cellular receptor for VSV-G²⁷¹, current VLP modalities would not be suitable to target quiescent LT-HSCs in vivo²⁵³, but alternative envelope glycoproteins could potentially overcome this limitation²⁷². Finally, antibody-targeted envelope delivery vehicles take advantage of a mutant VSV-G glycoprotein that maintains endosomal fusion activity while lacking native LDL-R binding affinity (Fig. 5)²⁷³. VLPs retargeted to primary T cells via CD3 and CD4 single-chain variable fragments (scFvs) allowed in vivo nuclease-driven editing, although at low efficiency²⁷³. Future improvements could redirect VLPs specifically to LT-HSCs, which lack the LDL receptor needed for current LNPs and VLPs.

Altogether, efficient in vivo targeting of LT-HSCs will likely pose challenges for diverse inherited hematological disorders, especially β-hemoglobinopathies where high levels of editing are needed. Endowing CRISPR-engineered HSCs with a selective advantage might be necessary to enrich gene-corrected cells. Using the Townes SCD mouse model and adenoviral vector delivery, therapeutic levels of in vivo prime editing were only achieved through low-dose chemotherapy selection of transduced cells¹⁵⁴. Coupling epitope editing^233,234 with gene correction could bypass the need for toxic chemotherapy to achieve therapeutic levels of editing via marker-free co-selection^108,232 using clinically relevant antibodies.

Simplified manufacturing

The current manufacturing process for most autologous gene and cell therapies involves the collection of patient cells, which are then engineered offsite (Fig. 4). Rapid manufacturing protocols for lentiviral transduction of unstimulated primary T cells and HSPCs in 24 hours or less have been developed^274,275. Minimizing ex vivo manipulation could reduce costs and resources with the potential to decentralize cell therapy manufacturing to local host laboratories. Because unwanted events such as large deletions and centromere-distal chromosome fragment loss are byproducts of cellular proliferation^55,199, rapid cell engineering without ex vivo culture could both facilitate manufacturing and decrease genotoxicity. Considering these recent developments, ex vivo culture could be minimized in clinical settings to mitigate nuclease-driven genotoxicity. To further simplify cell therapy administration, the purification and engineering of CD34⁺ HSPCs could be performed in a closed system without culture prior to infusion, and even potentially at the bedside in a circuit in continuity with the patient.

Summary and Perspective

The translation of CRISPR-Cas9 from bacterial cultures for dairy products to the clinic was relatively rapid. In less than two decades, microbiologists discovered and characterized a diverse set of bacteriophage defense mechanisms while biomedical scientists repurposed and engineered CRISPR effector proteins to develop gene therapies for inherited genetic diseases. Building from rich clinical knowledge of allogeneic HSCT and conventional gene therapy, the first CRISPR-Cas9 therapy to cross the finish line, exa-cel, has provided hope for patients afflicted with inherited hematological disorders. Whereas exa-cel offers a compelling present day therapeutic option, the development of more advanced and precise CRISPR-based technologies could further enhance the efficacy and safety profile of gene therapies for β-hemoglobinopathies. In addition, this landmark approval will undoubtedly set the stage for additional therapeutic indications. As the field moves forward, repurposing genome editing platforms to target new genes should streamline the expansion of CRISPR-based therapies to many more patients. FDA-approved therapies could be repurposed for new inherited disease indications by modifying the guide RNA and validating the efficiency and specificity of the strategy. Benchmarking these next-generation gene therapies should reduce the time and resources needed to translate a genome editing strategy from the bench to the bedside.

Although promising, the development of a gene correction approach for every pathogenic mutation causative of severe blood disorders would require immense effort, at least until novel platform-based regulatory mechanisms are available. The restoration of HbF expression through nuclease-driven sequence disruption offers a universal approach to remedy hundreds of pathogenic mutations causative of β-hemoglobinopathies. Nevertheless, this paradigm is not applicable to most inherited genetic disorders. Targeted complementation of a defective gene with a functional copy under physiological transcriptional control at its endogenous locus could also remedy diverse pathogenic mutations using a single genome editing strategy. Considering the quiescent nature of HSCs and the difficulties in achieving efficient HDR, new technologies are needed to achieve efficient targeted integration of therapeutic DNA in non-dividing cells. One key barrier for these emerging technologies is the toxicity of the dsDNA donors, which triggers an inflammatory transcriptional program through cytosolic cGAS-STING sensing^134,135. In addition, a recent study found that an endogenous RNase, SLFN11, triggers innate immune responses upon recognition of intracellular ssDNA²⁷⁶, generating an additional delivery barrier. Overcoming these DNA repair and immunological barriers should facilitate therapeutic genome editing of HSCs.

New challenges are emerging as the field moves towards in vivo genome editing. In their niche, quiescent HSCs are in metabolic dormancy. We recently found that the low levels of nucleotides available for reverse transcription restricts prime editing in HSPCs¹¹⁷. Co-delivering Vpx as mRNA using LNPs, or as a structural protein using VLPs, targeted SAMHD1 for proteasomal degradation and increased the level of nucleotide available for prime editing. Alternatively, increasing the affinity of reverse transcriptase for dNTPs^117,277 or using DNA polymerase systems²⁷⁸, such as click editing²⁷⁹, could facilitate in vivo HSC engineering. The metabolic dormancy of HSCs could also limit editor mRNA translation using LNPs delivery, whereas engineering mRNA with stabilizing modifications could improve editor expression^256,280,281. Considering the low levels of editing achieved in vivo with current modalities, endowing engineered HSCs with a selective advantage appears especially important. The development of marker-free co-selection strategies^108,232 using novel conditioning agents could address this limitation, although attention should be paid to maintaining clonal diversity.

Once optimized, therapeutic genome editing platforms could be expanded to new indications by developing target-to-therapy pipelines. Different indications will require adaptation, and the restoration of disease phenotypes should be assessed rigorously using appropriate models. Interactions with regulatory agencies are confidential, and there is no universal developmental path for therapeutic genome editing IND submission. Establishing transparent blueprints for IND-enabling studies should facilitate preclinical development. Novel regulatory paradigms might be needed to facilitate the implementation of individualized therapies, tailored to a patient’s genotype and clinical situation. Ultimately, gene editing strategies should be optimized for maximum editing efficiency with minimal genotoxicity, using healthy or patient HSPCs. Complementary assays should be performed to assess off-target editing using computational predictions and experimental validations. Human genetic diversity should be considered when assessing off-target activity¹⁹³, and population-specific off-targets should be incorporated into the validation pipeline. Larger unwanted events, such as large deletions, translocations, and centromere distal chromosome loss should also be quantified using complementary genotyping approaches. After on-target optimization, and thoroughly validating and mitigating genotoxicity, therapeutic strategies could move forward in the translational pipeline to expand existing platforms to new therapeutic indications. Monitoring patients and their blood cells longitudinally to observe molecular, cellular, and physiological outcomes of genome editing can both maximize knowledge and preserve patient trust.

The landmark approval of exa-cel is just the beginning, with more CRISPR-based therapies on the horizon. Streamlining the development of CRISPR-based therapies using approved platforms, along with ensuring freedom to operate after patent expirations, should improve broad patient access in the long term. For more common genetic diseases, such as SCD with approximately 300,000 new cases annually, commercial sponsors and academic investigators are working to develop the next frontier of safe and efficient delivery methods to engineer HSCs in vivo. Academic-led initiatives also strive to address unmet clinical needs for patients afflicted with orphan genetic conditions, including “n = 1” diseases. Achieving precise and safe gene correction to restore a diseased blood system with minimal healthcare infrastructure could be transformative.