Understanding and overcoming adverse consequences of genome editing on hematopoietic stem and progenitor cells

Abstract

Hematopoietic stem and progenitor cell (HSPC) gene therapies have recently moved beyond gene-addition approaches to encompass targeted genome modification or correction, based on the development of zinc finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs), and CRISPR-Cas technologies. Advances in ex vivo HSPC manipulation techniques have greatly improved HSPC susceptibility to genetic modification. Targeted gene-editing techniques enable precise modifications at desired genomic sites. Numerous preclinical studies have already demonstrated the therapeutic potential of gene therapies based on targeted editing. However, several significant hurdles related to adverse consequences of gene editing on HSPC function and genomic integrity remain before broad clinical potential can be realized. This review summarizes the status of HSPC gene editing, focusing on efficiency, genomic integrity, and long-term engraftment ability related to available genetic editing platforms and HSPC delivery methods. The response of long-term engrafting HSPCs to nuclease-mediated DNA breaks, with activation of p53, is a significant challenge, as are activation of innate and adaptive immune responses to editing components. Lastly, we propose alternative strategies that can overcome current hurdles to HSPC editing at various stages from cell collection to transplantation to facilitate successful clinical applications.

Graphical abstract

Precise genome editing of hematopoietic stem cells via programmable nucleases has developed rapidly, with great therapeutic potential. Several significant hurdles remain, including immune, DNA damage, and off-target effects, particularly challenging for gene correction via homology-directed repair.

Introduction

Hematopoietic stem cells (HSCs) carry out lifelong reconstitution of all component cells in the blood through their remarkable capacity for self-renewal and differentiation, and genetic modifications of hematopoietic stem and progenitor cells (HSPCs) for the treatment of a variety of congenital and acquired disorders have been pursued for over three decades, with accelerating clinical progress in recent years.¹ Gene therapy platforms have evolved steadily from relatively inefficient and genotoxic integrating replication-incompetent murine retroviral vectors to lentiviral vectors (LVs) able to transduce HSCs more efficiently.² Safety modifications have mitigated risks of genotoxicity due to semi-random integration into the genome. However, HSPC gene-addition therapies utilizing integrating vectors still have a finite risk due to unpredictable integration events. Therapeutic genes that need tightly controlled physiologic gene expression are challenging to deliver via gene-addition vectors, given the difficulty of reconstituting the complex epigenetic control found at the endogenous gene locus in a gene transfer vector. Finally, disorders resulting from production of an abnormal gene product are challenging to treat via simple gene addition.

Thus the advent of pioneering gene-editing approaches allowing targeted gene inactivation, gene insertion, or gene correction via programmable nucleases has been met with interest, acclaim, and rapid preclinical and clinical development.^3,4 Programmable nucleases such as zinc finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs), and CRISPR-CRISPR-associated protein (Cas) can produce double-stranded DNA breaks (DSBs) at precise genomic locations. Various outcomes arise from the different DNA repair mechanisms triggered by these DSBs. Canonical non-homologous end joining (NHEJ) is rapid, template free, error prone, and not dependent on cell cycling. It produces short nucleotide insertions or deletions, collectively known as “indels.” Thus, NHEJ-mediated gene editing is primarily used to knockout the function of a genomic locus by inducing frameshift or nonsense mutations at targeted sites. Microhomology-mediated end joining (MMEJ) or alternative end-joining (alt-EJ) occurs more slowly than NHEJ and is characterized by large deletions within targeted coding exons. Homology-directed repair (HDR) is a slow process that is facilitated when cells enter the S/G2 phases of the cell cycle and requires donor templates with homology arms to genomic sequences flanking the DSB. HDR enables precise genetic modifications, including correction of point mutations adjacent to DSBs or site-specific insertion of larger sequences.

The potential advantages of targeted gene editing of HSPCs over standard gene addition via integrating vectors have resulted in rapid preclinical development and initiation of clinical trials. For target diseases requiring precise gene expression or replacement of a deleterious gene product with a normal protein, editing has opened the door to clinical progress not conceivable with gene-addition approaches dependent on viral vectors. However, for many congenital immunodeficiency disorders, metabolic and storage disorders, and hemoglobinopathies, already enjoying clinical efficacy utilizing gene-addition therapies, it is challenging for clinicians, biopharmaceutical companies, regulators, and patients to weigh the reasonably well-understood risks and benefits of viral vector gene-addition therapies, studied intensively in clinical trials for 30 years, versus much less well-understood gene-editing technologies, with much more limited long-term animal model data, and very short follow-up from small pioneering clinical trials.⁵

Challenges to safe and efficient utilization of gene editing in engrafting long-term HSCs (LT-HSCs), particularly if HDR versus NHEJ is required, have already become apparent. These include achieving efficient and non-toxic delivery of editing machinery and repair cassettes; circumventing potent responses to nuclease-induced DSBs involving p53 activation, particularly in engrafting LT-HSCs; avoiding off-target (OT) genotoxic effects; and facilitating engraftment of edited cells. This review comprehensively covers the recent progress and challenges in genome editing specifically applied to HSPCs, both regarding editing efficiency and undesired cellular effects, and summarizes paths forward toward more widespread clinical utilization to treat human disease.

Platforms for targeted gene editing of HSPC

Three main editing platforms have been developed, each consisting of a nuclease capable of producing DSBs and a protein or RNA-targeting moiety able to position the nuclease at specific sites in the genome, resulting in NHEJ and creation of indels. For gene insertion or correction via HDR, each platform also requires simultaneous delivery of a cassette containing the sequence to be inserted, flanked by arms homologous to the surrounding DNA, either as an oligonucleotide cassette or via a non-integrating viral vector such as adeno-associated virus 6 (AAV6) (Figure 1).

Figure 1.

Gene-editing platforms, delivery approaches, and the possible detrimental effects of hematopoietic stem and progenitor cell (HSPC) genome editing

Programmable nucleases including the zinc finger nuclease (ZFN), transcription activator-like effector nuclease (TALEN), and CRISPR-CRISPR-associated nuclease (Cas) system facilitate targeted gene disruption, insertion, or correction at designated genomic sites of HSPCs via viral delivery or non-viral transfection of gRNAs, nucleases, and nucleic acid cassettes for homology-directed repair (HDR). These genome-editing techniques may achieve successful outcomes, whereas accompanied by detrimental adverse effects as indicated. IDLV, integration-deficient lentivirus; RNP, ribonucleoprotein; AAV6, adeno-associated virus type 6; ssODN, single-stranded oligodeoxynucleotide.

ZFNs consist of a pair of designer DNA-binding zinc fingers, each consisting of three or four DNA-binding domains that recognize specific 3 base pair motifs, flanking a 9- to 18-bp genomic target and fused to the Flavobacterium okeanokoites (FokI) endonuclease.^6,7 Dimerization via binding of both ZFNs to the target site is required for nuclease activity. ZFNs were the first editing platform to reach clinical trials, targeting inactivation of the CCR5 gene encoding an HIV-1 co-receptor, initially via editing of mature T cells and now in an ongoing trial targeting HSPCs.^8,9 ZFNs can also perturb other therapeutically relevant genes.¹⁰ In addition to targeted gene disruption, ZFNs enable site-specific insertion via HDR.¹¹ Due to inherent properties of transcription factors, ZFN can potentially target heterochromatic sites more effectively than CRISPR-Cas, even in transcriptionally repressed chromatin. However, the applicability of ZFNs is limited by the requirement to design and engineer a new set of zinc fingers for each target site. The complex ZFN design algorithms need to take into account both the DNA binding of each finger and the impact of nearby DNA sequences, poor target accessibility in non-G-rich sites, and the impossibility of designing efficient ZFN for some sites.¹²

The TALEN editing platform is based on a similar concept, fusing a DNA-binding domain consisting of arrays of repeated conserved 33−34 amino acid sequences with 2 variable amino acids at positions 12 and 13 conferring binding to specific nucleotides. Like ZFNs, pairs of these TALENs can be targeted to specific DNA sequences to induce DSBs and create indels via NHEJ or gene insertion/correction via co-delivery of a homologous cassette for HDR.¹³ Design of TALENs targeting specific sites is more straightforward than design of ZFNs. The clinical feasibility of TALEN-mediated editing of HSPCs targeting sites relevant for treatment of β-thalassemia or sickle cell disease (SCD) via fetal hemoglobin (HbF) reactivation has been demonstrated in humanized mice and nonhuman primates (NHPs).^14,15 Nevertheless, the applicability of TALENs has been limited by difficulties in non-toxic and efficient delivery of the relatively large TALEN components to HSPCs and requirement of complex molecular cloning methods to produce each TALEN.^12,16

CRISPR-Cas editing platforms derived from microbial adaptive immune systems have revolutionized gene editing due to simplicity, high throughput, versatility, and fidelity.¹⁷ In contrast to the requirement for protein engineering of ZFNs or TALENs for each new genomic target site, CRISPR-Cas utilizes a short guide RNA (gRNA) to target the Cas nuclease to the desired location. Upon binding to target DNA region and subsequent conformational change, Cas9 activates its nuclease domain and cuts with remarkable specificity even in the monomeric configuration, whereas ZFN and TALEN gain specificity via required dimerization to activate FokI nucleases. The CRISPR-Cas complex generates a DSB near the target sequence adjacent to a required protospacer adjacent motif (PAM), followed by NHEJ or by HDR if a homologous cassette is supplied.¹⁸ In less than 10 years, CRISPR-Cas editing systems have already been applied in numerous areas of biological investigation including lineage tracking,19, 20, 21 disease modeling,^22,23 agriculture, and animal husbandry. Extraordinarily rapid preclinical and clinical development of CRISPR-Cas approaches for numerous diseases has occurred,24, 25, 26 resulting in early evidence of benefit in a clinical trial for patients with thalassemia and SCD.²⁷ Despite these scientific breakthroughs, many issues regarding CRISPR-Cas editing remain, such as the restriction of editing to sites near PAM sequences and the undesired effects involving off-target effects, DNA damage responses (DDRs), and activation of immune pathways as discussed below.

Delivery of editing components to HSPCs

CRISPR-Cas systems were first discovered in bacteria, mechanistically dissected and initially developed for editing and utilizing purified DNA targets, optimized in mammalian immortalized cell lines, and only then moved forward to testing in primary cells.¹⁷ As exemplified by the prolonged struggle to successfully modify HSPCs with integrating viral vectors, primary cells are much more challenging targets than immortalized cell lines, having evolved mechanisms to block entry of exogenous nucleic acids or proteins recognized as potential pathogens and to protect their genomic DNA from potentially deleterious DSBs associated with viruses, inflammation, or environmental exposures such as irradiation. As will be described, true LT-HSC populations have developed, particularly robust defenses, given the necessity to preserve and protect self-renewal capacity and genomic integrity for the lifetime of the organism; thus approaches optimized in cell lines may prove less efficient and more toxic to engrafting LT-HSCs.

Lessons learned from over 30 years of gradual improvements in the efficiency of viral HSPC gene-addition therapies have facilitated the much more rapid development of HSPC gene editing, based on a better understanding of underlying biology and the development of predictive murine xenograft and NHP preclinical models. We will describe the advantages and disadvantages of various approaches for delivery of editing components to HSPCs and more generally the impact of editing on LT-HSC engraftment and function.

Nucleases and gRNAs

Given the efficiency and lack of toxicity of LVs for the introduction of exogenous material into HSPCs, one landmark paper utilized LVs to deliver both the Cas nuclease and a library of gRNAs into murine HSPCs, achieving stable LT-HSCs.²³ However, use of integrating LVs results in sustained nuclease and gRNA expression and thus carries the potential for continued off-target editing (OTE). In addition, semi-random integration of the LV in the context of clinical gene editing would obviate one major advantage of targeted gene editing, reduced genotoxicity.¹⁶ Moreover, sustained nuclease expression from an integrated LV is not tolerated in human HSPCs, likely due to ongoing DSBs resulting in p53 activation and apoptosis (see below).

Therefore, investigators next utilized engineered integration-deficient LVs (IDLVs) to deliver nucleases and targeting molecules, taking advantage of the cell-entry properties of LVs while avoiding genotoxicity related to integration and chronic nuclease expression. Early studies demonstrated feasibility of using IDLVs to deliver ZFNs and a corrective HDR cassette into human HSPCs capable of engrafting in immunodeficient mice, albeit with very low efficiency.28, 29, 30 IDLVs express transgenes in HSPCs for up to 7−10 days; thus inclusion of the ZFN nuclease in the IDLV likely resulted in DSB-mediated genotoxicity due to prolonged nuclease activity. Delivery of the ZFN via mRNA electroporation and optimization of ex vivo culture conditions and timing of exposure to editing components improved results using this platform (see below).³¹ Adenoviral vector platforms have also been developed for delivery of gene-editing components, utilizing specific subtypes able to enter HSPCs.^32,33 These vectors have very large capacities, allowing co-delivery of nucleases, targeting components, and a large corrective or gene knock-in cassette via a single viral particle. However, human HSPCs edited with an adenoviral delivery platform also showed poor xenoengraftment.³³ Recently, a non-integrating Sendai virus expressing Cas9 together with an LV expressing gRNAs achieved transduction of up to 95% in mobilized peripheral blood (mPB)-HSPCs.³⁴

Given the complexity, toxicity, and possible mutagenesis associated with viral delivery, non-viral transfection has been considered as an alternative. Transfection refers to the delivery of cargo molecules (such as nucleic acids, DNA or RNA, protein) across the cell membrane via physical or chemical stimulation. Chemical transfection has advantages in simplicity, cost effectiveness, and a wide range of targetable cell types and includes polymer, calcium phosphate, or lipid facilitated transfection.³⁵ Physical transfection encompasses microinjection, laser-fection, and electroporation. Electroporation permeabilizes the cellular membrane through transient formation of pores, allowing entry of nucleic acids and other materials, and has been the most used non-viral methodology for delivery of editing components.

Early studies utilized plasmid DNA encoding ZFNs targeting the CCR5 locus, with up to 20% allele disruption in engrafted human cells.³³ However, high concentrations of electroporated DNA have been found toxic to human HSPCs,³⁶ resulting in decreased engraftment.³⁷ Alternatively, mRNAs encoding the nuclease (together with gRNAs in the case of CRISPR-Cas platforms) can be electroporated with lower toxicity and high efficiency.^11,38 Recently, most investigators have settled on electroporation of Cas protein pre-complexed to gRNAs in a ribonucleoprotein (RNP) complex,³⁹ with efficiency improved by use of gRNAs with chemical modifications promoting stability.⁴⁰ One study systematically compared the efficiency and toxicity of plasmid DNA, mRNA, LVs, and RNPs for NHEJ induction in human HSPCs and found that RNPs achieved high efficiency with low toxicity.⁴¹ Many investigators have achieved 50%–90% NHEJ editing in human or NHP HSPCs and at least short-term engrafting cells utilizing the RNP CRISPR-Cas platform.^24,42, 43, 44 In addition, the short time period of intracellular active Cas nuclease activity when delivered via RNPs results in lower OTE as compared to other delivery approaches that express nucleases for longer time periods.⁴⁵

Delivery of gene correction or gene knock-in cassettes

Targeted gene correction, replacement, or addition via HDR also require co-delivery of a nucleic acid cassette containing the sequence to be inserted into a target site, flanked by homology arms. A variety of approaches have been utilized for delivery of these homology templates, which can vary in size from a few hundred base pairs to several kilobases. IDLV HDR cassette delivery resulted in low-level engraftment of corrected cells in xenografted mice.³¹ For targeting of point mutations, single-stranded reverse-strand DNA oligonucleotides with 60 base pair homology arms delivered via electroporation were equally effective to IDLVs and simpler to construct.²⁹

Delivery of larger cassettes is most frequently achieved via non-integrating AAV vectors. AAV vectors remain episomal in both dividing and non-dividing transduced cells but possess smaller cargo capacities than IDLV (AAV ∼ 4.7 kb versus IDLV ∼ 9 kb). Among multiple AAV serotypes, AAV6 is most tropic for HSPCs,^11,46,47 permitting high transduction efficiency, even of quiescent cells.⁴⁸ Recent preclinical studies with CRISPR/AAV6 systems achieved HDR correction in murine, NHP, and human HSPCs targeting SCD,^26,49,50 X-linked chronic granulomatous disease (X-CGD),⁵¹ and interleukin 2 receptor (IL-2R)gamma mutations.⁵² However, the higher efficiency of editing, as assessed in vitro comparing AAV6 to single-stranded oligodeoxynucleotide (ssODN) for HDR cassette delivery, was not maintained following xenotransplantation, with either equivalent or inferior HDR in engrafted cells long term comparing AAV6 to ssODNs.^26,53 Engraftment of human cells was also impaired by AAV6 exposure in one study⁵³ but was improved after extensive optimization.^51,54

Overcoming barriers to efficient, non-toxic, and safe HSPC gene editing

Off-target and deleterious on target effects

Nuclease-based genome editing can be accompanied various types of genotoxicity, as recently reviewed.⁵⁵ HSPCs are particularly susceptible to malignant transformation following integration of viral vectors, potentially based on lifelong self-renewal properties, an issue also relevant to HSPC editing. Utilization of gene editing for interpretable laboratory research and most importantly clinical applications requires prediction, detection, and avoidance of editing-related genotoxicity. OTE can occur via binding of the editing complex and nuclease cleavage at other than the desired target site, resulting in typical insertions, deletions, and point mutations.⁵⁶ Initial efforts to avoid OTE focused on algorithms to design ZFNs, TALENs, or gRNAs to target sites with minimal sequence homology to other sites in the genome. However, genome-wide technologies designed to retrieve actual off-target sites, including GUIDESeq and CIRCLE-Seq, reported discrepancies between prediction of off-target sites by computational methods and actual editing in experimental models.^57,58 Thus, systems to predict relevant OTE have been continuously improved.⁵⁹ Sites identified by these approaches have been detected in vivo long term following transplantation of CRISPR-Cas-edited HSPCs in a macaque model.⁶⁰ A second issue is chromosomal translocations resulting from fusion of on-target and/or off-target sites, shown to persist in vivo in both macaques and humans, although not yet associated with leukemic transformation.^60,61 Finally, a recent report documented chromothripsis, a concerning pattern of large chromosomal rearrangements, occurring as a consequence of on-target editing, due to structural changes in the nucleus related to CRISPR-Cas9 editing.⁶²

A more accurate prediction of OTE through evolving technological development has enabled the selection of optimal gRNAs, allowing for more accurate and safer NHEJ-based strategies. However, genome editing via HDR to precisely correct or insert DNA at an exact site has little ability to avoid OTE via choice of an editing target; thus different approaches are needed to improve specificity.⁶³ Modifications of Cas nucleases have mitigated OTE but also reduced on-target editing.^64,65 Introduction of a single point mutation in high-fidelity Cas9 achieved both improved specificity and retained high efficiency.⁶⁶

Immune responses

Both cellular innate immune pathways as well as potential pre-existing adaptive immunity to editing components are considerations for clinical development of HSPC gene editing.⁶⁷ Immune responses can be triggered by the presence of editing components such as Cas protein, mRNA, or gRNAs; delivery vectors such as IDLV, AAV, or adenoviruses; or even transgene products.^67,68 Once activated, these immune responses may compromise HSPC engraftment and persistence.

Intracellular innate immunity has evolved to protect mammalian cells from infection with pathogens, via recognition of foreign nucleic acids and production of cytokines such as interferons (IFNs) to induce adaptive immunity. Plasmid DNA can stimulate cyclic GMP/AMP synthase and activate type 1 IFN pathways.⁶⁹ In vitro-transcribed gRNAs triggered a type I IFN response via recognition by DDX58 (RIG-I).^70,71 RNP electroporation of HSPCs stimulated IFN-α secretion.⁷² Additional inflammatory genes such as IL-8, IL-6, and tumor necrosis factor (TNF)-α were upregulated in response to RNP electroporation.⁷³ Since inflammatory cytokines have profound effects on HSPC homing, self-renewal capacity, and lineage bias, there is a need for greater understanding of inflammatory consequences of gene editing and interventions to evade or suppress these undesirable immune reactions.⁷⁴

A number of chemical modifications to gRNAs have been shown to reduce innate immune activation and improve editing efficiency.^72,75,76 Innate immune barriers to editing via IDLVs can be overcome with cyclosporine H.⁷⁷ A comprehensive study of gene-expression changes in HSPCs linked to each CRISPR-Cas editing component was performed, including electroporation, Cas9 (mRNA or protein), chemically modified gRNAs, and AAV6.⁷⁸ Each component induced pathways associated with inflammation, stress, and apoptosis; mRNA delivery resulted in the most profound changes. Surprisingly, AAV6 exposure did not result in a detectable antiviral response. Overall, RNP delivery induced the fewest potentially detrimental pathways.

Pre-existing adaptive immunity to editing components and delivery vectors is likely a much less significant barrier to ex vivo HSPC gene editing as compared to the in vivo approaches required for altering liver, brain, muscle, or other solid organs. Not surprisingly, pre-existing B cell and T cell adaptive immunity against Staphylococcus aureus (staph aureus) and Streptococcus pyogenes (strep pyogenes) Cas nucleases (Sa- and SpCas9, respectively) was detected in a majority of normal human volunteers and model animals,^79,80 not surprising given the ubiquity of these micro-organisms and impairing in vivo gene editing.^81,82 Many humans also have pre-existing immunity to AAV6 or to vesicular stomatitis virus G protein (VSV-G)-pseudotyped IDLVs.^83,84 Engineered editing components with reduced immunogenicity are being developed.^81,82 Given that HSPCs are generally infused with 1 to 2days of ex vivo electroporation and/or transduction with editing components, residual foreign proteins could at least theoretically trigger an adaptive immune response, resulting in decreased engraftment.

Many preclinical studies of HSPC gene editing have been performed in profoundly immunodeficient mice unable to mount B cell, T cell, or natural killer (NK) cell responses, and therefore immune barriers to HSPC gene editing have not been extensively assessed. Most autologous HSPCs clinical gene therapies have utilized busulfan conditioning, which is profoundly myeloablative but does not significantly impair pre-existing adaptive immunity, resulting in demonstrable immune rejection of HSPCs transduced with foreign transgenes via integrating LVs.⁸⁵ In a clinical trial of gene therapy for Fanconi anemia, HSPCs were infused without conditioning, and VSV-G-specific cellular and humoral immunity could be detected following transplantation; thus immune rejection potentially explains the very low persistence of transduced cells in this trial.⁸⁴ On the other hand, successful and persistent engraftment of LV-transduced Fanconi anemia HSPCs has been achieved;⁸⁶ thus the risk of immune rejection of transduced or edited HSPC in the absence of immunoablative conditioning remains an open question. Reassuringly, the first clinical trial of CRISPR-Cas HSPC gene editing, utilizing RNP delivery of editing components, recently reported excellent engraftment and persistence following busulfan conditioning;²⁷ therefore this issue may not be significant for most ex vivo HSPC gene-editing clinical applications.

Activation of DDR and p53 pathways

Protection of the self-renewing LT-HSC pool from both depletion and mutational load over a lifetime requires a very finely tuned response to DNA damage, including DSBs.⁸⁷ ZFNs, TALENs, and Cas nucleases all induce DSBs, with repair occurring via NHEJ, HDR, or MMEJ, depending on the presence of a homology template, characteristics of the gRNA, and target cell type and cycling status. Cellular DDRs to DSBs vary markedly between progenitor cells and true self-renewing LT-HSCs.^88,89 These discrepancies may help explain the differences in editing frequencies between HSPCs assessed in vitro and results in vivo, with a marked decrease in the level of edited cells over time as short-term HSPC contributions to engraftment are replaced by output from true LT-HSCs (see below) and much lower rates of HDR as compared to NHEJ in LT-HSCs.⁷³

CRISPR-Cas9 induces p53 activation after DNA cleavage and subsequent cell-cycle arrest.⁹⁰ This toxicity is also observed when single guide RNA (sgRNA) and Cas9 protein are transiently introduced following electroporation. A recent study comprehensively examined DDRs to editing.⁷³ Using RNA sequencing and cellular imaging, it was observed that the signaling pathway related to programmed cell death was significantly activated in DSB-induced cells, suggesting that a DSB at a single target site is sufficient for triggering the DDR. Nuclease-induced DSBs were repaired within 24 h and were more frequent on a per-cell basis when using low-specificity gRNAs. DDR pathways involving TP53 were activated, with cell-cycle arrest via downstream activation of target genes such as p21.⁷³ Similar pathways were activated in NHP HSPCs.^24,91 Enache and colleagues⁹² reported that the presence of Cas itself facilitates activation of p53 pathway, even in the absence of gRNA and regardless of delivery methods. Delivery of a homology cassette via AAV6 resulted in more marked TP53 pathway upregulation, particularly in cells that have undergone successful HDR,⁷³ and cell-cycle arrest was higher following HDR. Exposure to lentiviruses, including IDLV, also activated p53.⁹³

p53 inhibition to enhance gene editing in HSPCs

Transient inhibition of TP53 or TP53-binding proteins during HSPC editing has been explored to neutralize DDR pathways that decrease HSPC function and impair HDR. Introduction of an mRNA encoding a dominant-negative (DN) form of TP53 (GSE56) reversed gene expression and functional changes associated with editing and resulted in improved engraftment of edited cells in immunodeficient mice, directly demonstrating the crucial role this pathway serves in mediating DSB-induced cytotoxicity and arrest of cell growth.⁷³ Addition of the adenoviral protein E4orf6/7, which has been shown to affect cell survival and cell-cycle progression, further improved HDR frequency. Cell-cycle analysis revealed a significant increase in the number of edited primitive HSPCs in the S/G2 phase treated with Ad5-E4orf6/7 in combination with GSE56.²⁰

53BP1 is a p53-binding protein that enhances p53 activity as well as specifically favoring NHEJ over HDR. A selective protein inhibitor of 53BP1 (i53) demonstrated significant improvement in HDR in cell lines and HSPCs.⁹⁴ De Ravin and colleagues^95,96 efficiently corrected the gp91^phox mutation in patient HSPCs using HDR-mediated gene-cassette insertion and transient 53BP1 suppression with i53. HDR editing via ssODN donor templates was improved 2-fold with inclusion of i53 in patient HSPCs. HDR enhancement with i53 was used for the development of an NHP model for SCD.²² Rhesus macaque CD34⁺ cells were electroporated with i53 mRNA along with an RNP complex and donor ssODN to induce β-to-βs globin conversion (20A > T). They observed a 93% increase in HDR (29% to 56%) and detected βs globin protein (∼100%) in differentiated erythroid cells studied in vitro. After transplantation of edited CD34⁺ cells, editing efficiency decreased over time and plateaued at 1% for β-to-βs globin conversion. These results emphasize the necessity of further optimization of HDR using relevant models such as in NHPs prior to clinical applications requiring efficient HDR and that i53 utilization may not fully rescue toxicity in LT-HSCs.

No increases in chromosomal translocations or activation of cancer-associated genes have been detected following transient TP53 inhibition, but the impact of global inhibition of p53 on oncogenesis has not been fully elucidated. To address this concern, Jayavaradhan and colleagues⁹⁷ developed a site-specific inhibitory strategy by tethering a DN form of 53BP1 to the Cas9 protein, resulting in enhancement of HDR efficiency and inhibition of NHEJ. The DN-53BP1-Cas9 complex inhibited NHEJ and downstream pathways only following editing at the specific target site, suggesting that DN-53BP1 does not disrupt naturally occurring cellular DNA repair foci.

Enhancing HDR versus NHEJ

In CRISPR-Cas systems, cleavage efficiency and predicted low off-target activities have been the main criteria for selection of gRNAs. In many studies investigating gene editing, HDR is the desired repair pathway for precise mutation correction or gene knock-in. Although gRNAs producing the highest total indel frequencies are generally assumed to be the best for maximizing HDR, accumulating evidence has revealed that extent of overall DNA cleavage does not necessarily correlate with relative HDR versus NHEJ frequency.^98,99 Recent analyses reveal that specific gRNAs exhibit preferences for certain DNA repair pathways. Interestingly, a preference for MMEJ, an alternative error-prone mechanism of DSB repair involving alignment of microhomologous regions, has been shown to be predictive of HDR frequency, which might be explained by shared components in MMEJ and HDR repair mechanisms. MMEJ and HDR are both characterized by Mre1-dependent DSB end resection.¹⁰⁰ Furthermore, individual gRNAs have been reported to produce indels of nonrandom sizes, or “dominant indels,” following editing.^101,102 In NHEJ, error-prone repair of DSBs results in small deletions or insertions. In contrast, MMEJ results in deletions larger than 3 bases. Recently, 51 gRNAs targeting 6 clinically relevant genes were studied and revealed that each gRNA had a distinctive indel signature comprising a specific proportion of NHEJ versus MMEJ indels in HSPCs.⁹⁹ Regardless of DNA template type (ssODN or AAV6 vector delivery), HDR rate correlated with MMEJ repair but not with total indel events. In other cell types, HDR outcompeted MMEJ but was unable to outcompete NHEJ.¹⁰³ Selection of gRNAs based on these findings can improve HDR rate by up to 90% without cell-cycle synchronization or NHEJ inhibitors.⁹⁹ Moreover, this property is conserved with nucleases of different origins (Sp- versus SaCas9).

Enrichment via selection of HDR-edited HSPCs has also been explored as a strategy. The Porteus group^49,104 has developed cassettes containing a fluorescent protein and enriched for cells undergoing HDR by flow cytometric sorting for the cassette transgene. Due to issues with introducing fluorescent proteins into HSPCs destined for clinical use and the difficulties of fluorescence-activated cell sorting (FACS) of large cell numbers for transplantation, a non-signaling truncated nerve growth factor receptor (tNGFR) has also been utilized, permitting selection via immunoabsorption, a more clinically feasible approach.⁴⁹ In addition, antibiotic-based surrogate reporter systems could be employed to enable the selection of HDR cells.¹⁰⁵ However, validation and safety testing of these types of complex pre-transplant processes are incomplete, and it is unclear whether these selected cells will retain engraftment potential in vivo in relevant preclinical models.

Engraftment and persistence of gene-edited cells

In addition to editing efficiency, clinical outcomes of HSPC gene editing are ultimately dependent on the long-term repopulating capacity of edited cells. Thus, the persistence of gene-edited HSPCs is frequently assessed in xenografted mice and NHP models. Without exception, hematopoietic output from edited HSPCs is far more robust in the first weeks to months following transplantation than at later time points,¹⁰⁶ in studies employing TALENs,¹⁴ ZFNs,^11,107 or CRISPR-Cas platforms.¹⁰² This drop in edited cells is far more marked than experienced following transplantation of lentivirally transduced HSPCs in NHP models and multiple human gene-addition clinical trials,^1,2,108,109 suggesting the issue is not ex vivo culture of target HSPCs or inadequate conditioning regimens but specific to nuclease-based editing. When HDR is required, the fall-off is even greater. Although similar levels of HDR were observed in infusion products and engrafted hematopoiesis at early time points, persistent marked reduction of HDR-edited cells occurs at later time points.⁹⁵ Pushing to a very high-level NHEJ in HSPCs has been feasible and has allowed sustained engraftment with high levels of NHEJ-edited cells in preclinical NHP models and now an early clinical trial, but there is a ways to go before clinical applications requiring HDR are feasible.

There are many potential contributors to these problems, including preferential toxicity and loss of LT-HSCs during editing due to innate immune pathways, p53 activation, electroporation or viral-vector-mediated cytotoxicity, and/or impaired competition with endogenous LT-HSCs surviving conditioning, in addition to the resistance of LT-HSCs to HDR as compared to NHEJ. Dissecting out contributions of each potential factor is very challenging, because analysis of human or NHP-purified LT-HSCs in vitro may not reflect LT-HSC function following transplantation, and immunodeficient murine models, while powerful, cannot assess immune effects and may not fully reflect physiologic human LT-HSC behavior. NHP models are likely more predictive¹¹⁰ but impractical for screening of multiple interventions.

Quantitating the impact of editing on LT- HSCs and their clonal output over time is challenging. Integrated viral vector integration sites can be used as tags for clonal tracking, or LVs can contain barcoded libraries allowing precise quantitation of clonal HSPC output.¹¹¹ Although NHEJ-induced “scars” are variable, they are not sufficiently diverse to allow for clonal tracking of thousands−millions of transplanted HSPCs. Two recent studies incorporated barcodes into HDR homology cassettes to allow analysis of the impact of editing on human HSPC clonal diversity and dynamics in immunodeficient mice.^20,112 Both studies documented long-term engrafting multilineage clones; however, the pattern became oligoclonal later after transplantation, suggesting very few active HDR-edited LT-HSCs. p53 inhibition increased the number of long-term persisting HSPC clones 5-fold, a very relevant finding that could not have been made without barcoding.

Many approaches have been employed to improve engraftment of gene-edited HSPCs. Increasing the dose of edited HSPCs might overcome competition with endogenous LT-HSCs surviving conditioning.⁴² However, obtaining much larger starting numbers of HSPCs from patients and editing them is expensive and impractical, and ex vivo expansion of human LT-HSCs is not yet robust. One clever strategy has been to directly manipulate the cellular compartment involved in engraftment and long-term repopulation. Shin and colleagues¹¹³ reported that brief and forced cell cycling prior to electroporation promoted accumulation of HDR alleles in an engraftment-enriched (EE) population (CD34⁺CD38⁻) and improved the survival of engrafted cells. Starting with a more enriched population of more primitive HSCs (CD34⁺CD90⁺) offers increased cost and resource effectiveness with comparable efficiency to bulk HSPC editing.^24,114

Studies employing HDR-mediated editing consistently show lower efficiency than NHEJ-mediated editing in LT-HSCs due to intrinsic quiescence and reduced engraftment following HDR. In a macaque model, there was a negative correlation between engraftment of edited HSPCs and HDR-mediated gene correction but not indel percentage.²² DNA template selection has also been found to influence engraftment.^26,95 Due to a relatively large cargo capacity and minimal integration, AAVs have been promising vectors for delivery of exogenous HDR DNA templates. However, AAV exposure results in DNA damage and up to 90% reduction in engraftment.^{26,49,51,53,96} A direct comparison of engraftment for mPB-HSPCs edited using AAV6 or ssODNs for cassette delivery showed that although HDR efficiency was greater with AAV6 than with ssODNs, engraftment following xenotransplantation was significantly lower, resulting in a 6-fold higher proportion of corrected cells with ssODNs. According to De Ravin and colleagues,⁹⁵ the presence of i53 mRNA did not rescue the dampened engraftment efficiency of CRISPR/AAV-edited cells. In contrast, GSE56 enhanced engraftment of HSPCs edited by ZFNs and CRISPR-Cas9 RNP in immunodeficient mice, even in the presence of AAV6.^20,73,115 Interestingly, addition of Ad5-E4orf6/7 protein reduced the engraftment effect of GSE56 but promoted a higher proportion of HDR-knock-in cells and enabled polyclonal reconstitution.²⁰

Editing of HSPCs without DSBs

DSB-mediated p53 activation, conditioning toxicities, insufficient HSPC doses following editing, and imperfect delivery of gene-editing materials are all obstacles to current HSPC gene-editing strategies. Increasing efficiency, particularly of HDR, results in loss of long-term-repopulating HSCs. To avoid surveillance by DNA repair systems and cellular deterioration, techniques that manipulate the genome without DSBs are desirable, with several already under active preclinical development.

Base editing

Base editors (BEs) are a set of programmable nucleotide deaminases that produce single-nucleotide conversions.¹¹⁶ Unlike nuclease-mediated editing, base editing does not rely on the creation of DSBs, avoiding activation of undesirable cellular DDR pathways, formation of unpredictable indels via NHEJ-mediated repair, and the risk of chromosomal translocations at the site of DSBs.¹¹⁷ Following a DSB, gene correction via HDR is always in competition with formation of indels via NHEJ, with NHEJ generally more efficient, particularly in LT-HSCs.

Base editors are programmable DNA-binding complexes complexing an inactive “dead” or single-strand Cas9n nuclease, together with a DNA-modifying deaminase and a targeting gRNA. Almost 60% of human pathogenic mutations are single-nucleotide changes, potentially correctable by a single-nucleotide conversion. To date, two classes have been developed: cytosine base editors (CBEs), which convert C:G to T:A, and adenine base editors (ABEs), which convert A:T to G:C.¹¹⁸ Upon recognition and binding to target DNA via a gRNA, a small “R loop” of single-stranded DNA is formed, allowing the ABE or CBE to mediate nucleotide deamination. A single-stranded nick in the non-edited strand is produced, which leads to cellular DNA repair of the site using the edited strand as a template. Together, CBE and ABE can generate all four transition mutations and offer the potential to correct a majority of pathogenic single-nucleotide polymorphisms.¹¹⁷ The ability to induce precise point mutations without the need for induction of DSBs presents a potentially safer gene-editing tool for editing of HSPCs and treatment of various blood disorders. Notably, all base-editing studies thus far have demonstrated a significant reduction but not complete elimination of DSBs as compared to conventional nuclease-mediated strategies.

Initial efforts to employ base editing in HSPCs have focused on treatment of SCD and β-thalassemia. Zeng and colleagues¹¹⁹ reported the efficient base editing of the +58 BCL11A erythroid enhancer in order to prevent BCL11A expression and thus enhance HbF production. After establishing efficient correction in vitro, they infused base-edited HSPCs into immunodeficient mice and observed multilineage hematopoietic reconstitution and functional LT-HSC self-renewal as documented upon secondary transplantation. Interestingly, engrafting LT-HSCs showed a preference for C > T rather than C > G/A substitutions, suggesting an intrinsic preference in DNA repair mechanisms in specific cell types.

Actual correction of the SCD mutation to a wild-type allele requires conversion of a GAG (Glu) codon to GTG (Val), not feasible with current ABEs. However, conversion of GAG (Glu) to GCG (Ala), feasible with ABEs, results in production of hemoglobin Makassar, a naturally occurring variant hemoglobin known to be non-pathogenic and non-sickling. This approach was recently shown to be effective in a murine model of SCD.¹²⁰ Base editing did not activate p53 pathways or result in the large deletions and translocations often seen with nuclease editing at hemoglobin loci.

A limitation of both CRISPR-Cas9 and base editing is the availability of a suitable PAM sequence at specific positions relative to the target nucleotide. However, use of engineered Cas variants to increase the range of targets and editing windows has also increased the likelihood for unwanted bystander mutations and OTE.¹²¹ Furthermore, overexpression of a constitutively active deaminase in base editors can mediate random genome-wide guide-independent OTE at locations sharing no similarity to the on-target sequence, particularly in transcribed genomic regions with R loops.^122,123 To address these challenges, Chu and colleagues¹²¹ replaced nonessential domains within the SpCas9 component of an ABE with bacterial tRNA adenosine deaminase (TadA), the DNA modifying enzyme in ABEs. These deaminase-inlaid base editors (IBEs) exhibited a shifted base-editing preference while reducing guide-independent DNA and RNA off-target deamination. In demonstration of the applicability of these IBEs for treating disease, the authors utilized the IBE to convert the sickle cell mutation to the Makassar codon, showing high efficiency and reduced off-target effects.¹²¹

Overall, base editing shows early promise for efficient and non-genotoxic correction of HSPCs. Delivery can be accomplished via electroporation of a single RNP consisting of fused inactive or Cas9n-deaminase-gRNA complex, without requirement for coincident delivery of a homology template, as is necessary for conventional nuclease-based gene correction. However, some loss of LT-HSC activity after base editing has been noted.¹¹⁹ Although not completely eliminated, DSBs may occur in the rare event that nicked single-stranded DNA undergoes DNA duplication before repair of the mismatch. Further optimization and testing in animal models, as well as analyses of cellular responses to base editing and the biological impact of Cas-dependent and independent OTE in HSPCs, are ongoing.

Additional gene-editing tools

Currently, base editors are largely limited to installation of transition mutations within a small editing window adjacent to a PAM sequence, preventing application to a wider range of genetic defects. To compensate for this, the Liu group¹²⁴ developed a “search-and-replace” prime-editing (PE) technology for versatile and precise genome editing without DSBs and without the need for a PAM sequence or an HDR template. PE systems consist of a PE gRNA (pegRNA), which both recognize the target site and encode the desired correction and a catalytically inactive Cas9 nuclease fused to a reverse transcriptase. Remarkably, PEs can install all 12 types of transition and transversion mutations (restricted to four in base editing), targeted insertions (up to 44 bp with 23% efficiency), and deletions (up to 80 bp with 50% efficiency) with few off-target effects and the potential to correct 89% of all pathogenic human genetic variants.¹²⁴ However, PE in primary cells has shown low editing efficiencies compared to nuclease-based editing and base editing.^125,126 Thus, further optimization of PE efficiency in HSPCs and strategies for enhancing delivery are necessary.⁹⁸

Other gene-editing platforms free of nuclease-mediated DSBs are clade F AAVs, which are novel AAV variants isolated from CD34⁺ HSPCs of healthy donors.¹²⁷ Interestingly, BRCA-dependent homologous recombination utilizing these viruses achieved almost 60% gene insertion into the AAVS1 site of human HSPC.¹²⁸ However, some studies have demonstrated inefficient gene editing with clade F AAV in comparison to nuclease-mediated editing and AAV6.^129,130 Alternatively, these issues might be overcome by using highly purified vectors, allowing less toxicity at the required high MOI.^48,131 This system is only useful for gene insertion into the natural target AAVS1 locus, which decreases the risk of genotoxicity associated with random gene insertion but does not allow mutation correction.

Approaches to improving engraftment

Ex vivo expansion or improved maintenance of true LT-HSCs during culture, if successful, might be able to produce a larger number of cells for genetic manipulation and thus overcome HSPC toxicity and impaired competition with endogenous non-edited HSPCs surviving conditioning. Molecules such as prostaglandin E₂ (PGE₂),132, 133, 134 aryl hydrocarbon receptor antagonist StemRegenin 1 (SR1),^135,136 UM171,^137,138 and valproic acid (VPA)^139,140 have been utilized to augment the survival and proliferation of CD34⁺ HSPCs ex vivo, particularly important for HDR applications, which require S/G2 cell-cycle transit. Many HSPC “enhancers” are being studied in various preclinical and clinical settings, generally focused on increasing cord blood HSPC numbers; however, systematic comparative analyses are lacking. Further studies, preferably in NHPs, will be needed to determine whether there is positive impact on LT-HSCs and whether the changes in cycling alter gene editing and engraftment efficiencies.^113,141

Intra-bone delivery of genetically engineered HSPCs has been explored to increase engraftment, based on the hope that fewer cells would be lost via lodgment outside the marrow. In murine xenograft models, higher engraftment has been achieved following intra-bone injection of human HSPCs; however, the efficiency of homing may be quite different in a xenograft.¹⁴² Conflicting data exist in human cord blood transplantation regarding whether intra-bone delivery is more efficient. HSPCs transduced with an LV have been directly infused into the marrow space to treat thalassemia, and the safety and efficacy of this delivery approach were demonstrated, although there was no comparison to standard intravenous delivery.¹⁴³ In rhesus macaques, LV-transduced autologous HSPCs were directly compared via both intra-bone and intravenous delivery into the same animal. Intra-bone-transduced HSPCs engrafted less well than intravenous-transduced HSPCs.¹⁴⁴ Accordingly, further optimization and study of intra-bone injection are necessary before applying this approach to edited HSPCs.

Avoidance of myelotoxic conditioning

Although necessary to decrease competition from endogenous HSPCs and allow efficient engraftment of edited cells, conditioning of recipients prior to HSCT using irradiation or chemotherapy is associated with significant organ toxicities and the need for prolonged hospitalization. In addition, the genotoxic stress of conditioning on endogenous surviving marrow HSPCs, particularly in patients with underlying chronic marrow stress and inflammation such as in SCD, may predispose to later hematologic malignancies. Indeed, two SCD patients that rejected donor allogeneic HSPC transplants were found to have TP53-mutated clones detectable pre-transplantation that later expanded and transformed to myeloid tumors.¹⁴⁵ A patient treated with LV gene-addition therapy later developed myelodysplasia due to transformation of non-vector-containing cells.¹⁴⁶ These cases stimulated the hypothesis that SCD patients may have prematurely aged/damaged HSPCs more susceptible to transformation, even in the absence of insertional oncogenesis, due to further damage from conditioning and post-transplantation proliferative stress. However, clonal transformation of an LV-containing HSPC in an SCD patient treated with gene-addition therapy has recently been presented at meetings, although full details have not yet been published.¹⁴⁷ Reportedly, the LV insertion site was not near any known cancer-related gene. These findings are concerning for gene editing of SCD, if these patients have pre-existing-compromised and pre-malignant HSPC clones, given the potentially increased proliferative and genotoxic stress associated with gene editing as compared to LV gene addition.¹⁴⁸

Antibody-mediated conditioning

Several non-myeloablative-targeted conditioning strategies have been proposed to allow engraftment without radiation or high-dose chemotherapy. Proteins found on the surface of HSPCs but not expressed in other critical tissues have been targeted by monoclonal antibodies that are either conjugated to toxins or able to facilitate antibody-dependent cellular cytoxicity.149, 150, 151 Antibody-based conditioning regimens targeting CD117 (c-KIT) present on HSPCs and erythroid precursors, or the pan-leukocyte protein CD45 resulted in efficient endogenous HSPC depletion and rapid donor engraftment in murine or xenograft models. Due to selective depletion of HSPCs, the pre-existing immune system remains intact after administration of anti-CD117, and thus this strategy would be valuable not only for gene editing but also for all HSPC gene therapies. However, CD117 is reportedly expressed in cells of the CNS, eye, and kidney,¹⁵² suggesting further examination of nonspecific toxicity to other systems. Moreover, the impact of pre-existing immunity against exogenous genome-editing materials remains an unresolved question with these non-immunosuppressive approaches.

In vivo genetic modification of HSPCs

An even more attractive approach would be the ability to correct or modify HSPCs in vivo, without the need for collection, ex vivo culture and manipulation, and engraftment following conditioning.¹⁵³ The Lieber laboratory^154,155 has carried out a long series of studies developing CD46-targeting adenoviral vectors able to transduce HSPCs in vivo, focused on delivery of Sleeping Beauty transposons for random integration of a therapeutic γ-globin gene, resulting in amelioration of a murine thalassemia intermedia phenotype. Administration of granulocyte-colony-stimulating factor (G-CSF) and plerixafor to release HSPCs from the marrow into the vasculature for improved access to intravenous infusion of CD46-targeting adenoviral HDAd5/35⁺⁺ vectors containing a human γ-globin gene cassette and an SB100x transposase allows integration-improved efficiency. More recently, this group has delivered CRISPR-Cas9 editing machinery or base editors to reactivate endogenous γ-globin in SCD mice, observing a substantial increase in γ-globin expression.^27,156 Importantly, the editing frequency in cells transduced with vectorized base editors was sustained throughout the period of observation in β-YAC mice, whereas CRISPR-Cas9 vector-transduced cells showed gradual decreases, possibly reflecting DSB-mediated LT-HSC impairment. The Kiem group¹⁵⁷ has demonstrated the feasibility of in vivo gene therapies utilizing intravenous foamy virus vector infusion in a canine model of severe combined immunodeficiency. None of these in vivo approaches are ready for human clinical trials, and questions remain regarding editing of germline tissues, potential acute toxicities of high doses of infused viral vectors, and relatively low efficiencies in large animal models.

Conclusions

The pace of new knowledge and initiation of clinical trials in the field of gene editing has been extraordinary, based on close collaborations among basic scientists in fields such as structural biology, biochemistry, and microbiology together with translational investigators focused on clinically relevant target cells such as HSPCs as well as gene engineers both in academia and industry. In this review article, we have reviewed the milestones reached and challenges uncovered specifically relevant to therapeutic HSPC gene editing and suggested likely ways forward. In addition to development of strategies to universally improve HSPC gene editing, researchers will need to take into account the unique characteristics and underlying genetic basis of each disease, which certainly hold their own specific thresholds for therapeutically relevant editing and target preferences in order to design the most ideal gene-editing strategy for a particular condition.^34,158 New engineering and delivery approaches are reported so frequently that a major challenge is choosing among so many innovative and promising options, but overall, this is the kind of problem that those developing gene therapies 30 years ago would have been jubilant to experience.