In vivo base editing by a single i.v. vector injection for treatment of hemoglobinopathies

Individuals with β-thalassemia or sickle cell disease and hereditary persistence of fetal hemoglobin (HPFH) possessing 30% fetal hemoglobin (HbF) appear to be symptom free. Here, we used a nonintegrating HDAd5/35++ vector expressing a highly efficient and accurate version of an adenine base editor (ABE8e) to install, in vivo, a –113 A>G HPFH mutation in the γ-globin promoters in healthy CD46/β-YAC mice carrying the human β-globin locus. Our in vivo hematopoietic stem cell (HSC) editing/selection strategy involves only s.c. and i.v. injections and does not require myeloablation and HSC transplantation. In vivo HSC base editing in CD46/β-YAC mice resulted in > 60% –113 A>G conversion, with 30% γ-globin of β-globin expressed in 70% of erythrocytes. Importantly, no off-target editing at sites predicted by CIRCLE-Seq or in silico was detected. Furthermore, no critical alterations in the transcriptome of in vivo edited mice were found by RNA-Seq. In vitro, in HSCs from β-thalassemia and patients with sickle cell disease, transduction with the base editor vector mediated efficient –113 A>G conversion and reactivation of γ-globin expression with subsequent phenotypic correction of erythroid cells. Because our in vivo base editing strategy is safe and technically simple, it has the potential for clinical application in developing countries where hemoglobinopathies are prevalent.


Introduction
Autologous hematopoietic stem cell (HSC) gene therapy for hemoglobinopathies has shown promising clinical efficacy (1)(2)(3)(4). However, current protocols involving isolation of patient HSCs, their in vitro genetic modification with integrating vectors, and reinfusion of the modified HSCs following myelotoxic BM conditioning are technically complex and expensive. We are attempting to develop an in vivo HSC gene therapy approach that does not require myeloablation and integrating vectors, and that is technically easier. In this approach, we are using capsid-modified, helper-dependent HDAd5/35++ vectors (1,2). These vectors target CD46, a receptor that is expressed on primitive HSCs (2,3). HDAd5/35++ vectors are injected i.v. after mobilization of HSCs from the BM by agents routinely used for HSC mobilization/ harvesting. Mobilized HSCs are transduced while they are in the periphery. A large fraction of HSCs returns to the BM. Mobilization of HSCs is critical for in vivo transduction because, in the BM, they are surrounded by extracellular stroma proteins (4) and are not accessible to gene transfer vectors (2). To expand in vivo-transduced HSCs, we currently use an in vivo selection mechanism based on a mutant O 6 -methylguanine-DNA methyltransferase (mgmt P140K ) gene that confers resistance to O 6 -BG/BCNU Individuals with β-thalassemia or sickle cell disease and hereditary persistence of fetal hemoglobin (HPFH) possessing 30% fetal hemoglobin (HbF) appear to be symptom free. Here, we used a nonintegrating HDAd5/35++ vector expressing a highly efficient and accurate version of an adenine base editor (ABE8e) to install, in vivo, a -113 A>G HPFH mutation in the γ-globin promoters in healthy CD46/β-YAC mice carrying the human β-globin locus. Our in vivo hematopoietic stem cell (HSC) editing/selection strategy involves only s.c. and i.v. injections and does not require myeloablation and HSC transplantation. In vivo HSC base editing in CD46/β-YAC mice resulted in > 60% -113 A>G conversion, with 30% γ-globin of β-globin expressed in 70% of erythrocytes. Importantly, no off-target editing at sites predicted by CIRCLE-Seq or in silico was detected. Furthermore, no critical alterations in the transcriptome of in vivo edited mice were found by RNA-Seq. In vitro, in HSCs from β-thalassemia and patients with sickle cell disease, transduction with the base editor vector mediated efficient -113 A>G conversion and reactivation of γ-globin expression with subsequent phenotypic correction of erythroid cells. Because our in vivo base editing strategy is safe and technically simple, it has the potential for clinical application in developing countries where hemoglobinopathies are prevalent.
β-Thalassemia and SCD are the most common inherited diseases in humans worldwide (18). β-Thalassemia is caused by mutations in the β-globin gene that result in absent (β 0 /β 0 ) or deficient (β + /β + , β + /β 0 ) β-globin chain synthesis. Patients with β-thalassemia have a multitude of pathological symptoms and die prematurely (18). Most patients with SCD are homozygous for a Glu6Val substitution, resulting in the production of sickle β-globin (β S -globin). Clinically, SCD is characterized by recurrent episodes of severe bone pain, multiorgan failure, and early mortality. Individuals with hereditary persistence of fetal hemoglobin (HPFH) and thalassemic β-globin mutations exhibit few or no pathological effects because fetal hemoglobin (HbF) inhibits hemoglobin precipitation by reversing the globin chain imbalance and improving erythropoiesis (19).
This led to a number of HSC gene therapy approaches aimed toward the reactivation of fetal/ γ-globin -e.g., targeting 2 transcriptional repressors, ZBTB7A and BCL11A, that participate in the developmental silencing of the 2 γ-globin genes (HBG1/Gγ and HBG2/Aγ) through binding to their respective promoter cis-regulatory elements (20). An approach based on RNA interference to silence bcl11a expression (20) or reactivation of γ-globin expression by CRISPR-Cas9 editing of the erythroid bcl11a enhancer (21) have delivered promising clinical data. Furthermore, preclinical studies to reactivate γ-globin by CRISPR/Cas9 editing of the BCL11A and ZBTB7A binding motifs within the HBG1/2 promoters have shown efficient HbF reactivation (22). Along this line, we used an in vivo HSC transduction approach to express a CRISPR/Cas9 specific to the BCL11A binding site within the HBG1/2 promoters and showed efficient γ-globin reactivation in transgenic mice carrying the WT 248 kb β-globin locus yeast artificial chromosome (β-YAC mice) (23). However, we also detected an undesired deletion of the HBG1 gene due to CRISPR/Cas9-mediated double-stranded DNA breaks (DSBs) in both the HBG1 and the HBG2 promoter. There is also accumulating evidence that DSBs catalyzed by CRISPR/Cas9 can result in large genomic deletions and chromosomal rearrangements, as well as p53-induced cell cycle arrest and apoptosis (24)(25)(26). To address these problems, we focused our work on base editors, enzymes that are capable of introducing precise cytidine or adenine substitutions with minimal occurrence of DSBs and indels at the target site (27)(28)(29)(30). Here, we employed an advanced adenine base editor version (ABE8e). ABE8e contains additional mutations that increase on-target activity and greatly decrease off-target activity (31).
We targeted ABE8e to the BCL11A binding site in the HBG1/2 promoters ( Figure 1A) to mediate a A>G conversion at the -113 position. The corresponding sgRNA was selected in an earlier study from a series of sgRNAs targeting the BCL11A binding motif as the most efficient in γ-globin reactivation (12). It is thought that the -113 A>G mutation does not disrupt BCL11A binding, but rather creates a de novo binding site for the transcriptional activator GATA1 ([T/A]GATA[A/G]) that outcompetes binding of the BCL11A (32). However, as a result of the relatively wide editing window of ABE8e, in addition to the -113 A>G conversion, bystander editing at neighboring adenines is expected ( Figure 1A, right panel). Specifically, the -116A>G conversion would destroy the BCL11A binding motif.
Here, we evaluated an HDAd-EF1α.ABE8e vector in vitro in CD34 + cells from patients with β-thalassemia and SCD, as well as in ex vivo and in vivo HSC gene therapy settings in the β-YAC mouse model. In this context, we tested a regimen of base editing that does not require the integration of the mgmt P140K cassette for in vivo selection.

Results
HBG1/2 promoter editing in healthy CD34 + cells. We used HDAd5/35++ vectors to express ABE8e either under the control of the phosphoglycerate kinase (PGK) promoter (HDAd-PGK.ABE8e) or the strong EF1α promoter (HDAd-EF1α.ABE8e) ( Figure 1B). The HDAd vectors also contained a human mgmt P140K expression cassette that allows for enrichment of edited cells after treatment with O 6 -BG/ BCNU. To produce HDAd-EF1α.ABE8e, suppression of ABE8e expression on HDAd producer cells was required. This was achieved by placing the ABE8e gene under the regulation of miR-183-5p, miR-218-5p   33), and dual miR/virus-associated RNAI (miR-vaRNAI) (34). The first strategy makes use of 2 miRNAs, hsa-miR-183-5p and hsa-miR-218-5p, expressed exclusively by the HDAd producer cells. The second strategy makes use of a miRNA expressed by the helper virus called miR-vaRNAI. Because these miRNAs are only present during vector production, transgene expression from the HDAd is unimpeded in the transduced target. In addition, we further modified the helper vector by inserting expression cassettes for anti-CRISPR peptides capable of blocking ABE8e activity (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.162939DS1) (35). The editing rate of HDAd-ABE8e vectors was tested in human CD34 + cells from 2 healthy donors at day 3 after transduction ( Figure 1C). HDAd-EF1α.ABE8e catalyzed, on average, approximately 26% of target site -113 A>G conversions, while HDAd-PGK.ABE8e and HDAd-ABEmax (used in an earlier study; ref. 12) conferred 6.6% and 1.3% editing, respectively. Subsequent studies, therefore, focused on HDAd-EF1α.ABE8e. Due to the wide editing window of ABE8e, in addition to the -113 adenine (shown as A8 at the bottom of Figure 1D), neighboring adenines (A5, A9, A11) can also be converted to guanines. To analyze the effect of editing on reactivation of γ-globin/HbF expression, HDAd-EF1α.ABE8e-transduced CD34 + were subjected to 18 days of erythroid differentiation (ED) with and without O 6 -BG/BCNU in vitro selection ( Figure 1E). Target site conversion rates were measured at days 7, 13, and 18 of ED ( Figure 1F). Starting at about 20% of editing, one-time treatment with of O 6 -BG plus 25 or 35 μM BCNU increased the editing rate to ~60% by day 18 of ED.
In ED cultures, the percentages of HbF + cells in total erythroid cells and enucleated erythroid (NucRed -) cells were significantly higher in the HDAd-EF1α.ABE8e-transduced over the untransduced cells (Figs.2C, S3A). Notably, background γ-globin expression in the control vector-transduced samples is triggered by cytokines present in the culture medium (36). In vitro selection at day 3 further increased the percentage of γ-globin-expressing RBCs. Reactivation of γ-globin protein expression was also shown by HPLC, where γ-globin chains reached levels of 30% relative to β-and α-globin chains ( Figure 2D).
Efficient γ-globin reactivation resulted in a significant reduction of reactive oxygen species (ROS) levels, a hallmark of oxidative stress seen in β-thalassemia and SCD (37) ( Figure 3A and Supplemental Figure 3B). γ-Globin reactivation by HDAd-EF1α.ABE8e also improved ED/expansion of βthalassemia and SCD CD34 + cells. The total number of erythroid cells increased 2-and 7-fold, respectively, for β-thalassemia and SCD samples between days 7 and 18 in settings with in vitro selection (Figure 3B). Improved erythropoiesis is also supported by microscopic analyses of cells at the end of ED, and this analysis shows more differentiated cells (maturing orthochromatic erythroblasts, reticulocytes/ pyrenocytes) in the β-thalassemia ( Figure 4A) and SCD ( Figure 4B) samples.
HDAd-EF1α.ABE8e transduction and base editing did not cause cytotoxicity as demonstrated by (a) no reduction in the percentage of primitive HSCs (CD34 + CD38 -CD90 + ) (Supplemental Figure 4A) and (b) no significant effect on progenitor colony formation (Supplemental Figure 4B).
In summary, transduction of CD34 + cells from patients with β-thalassemia and SCD with HDAd-EF1α.ABE8e mediated efficient editing of the HBG1/2 target site and the reactivation of γ-globin expression, resulting in phenotypic amelioration of erythroid cells after in vitro ED of HSCs.
The vectors also contain a PGK-mgmt P140K cassette for O 6 -BG/BCNU selection. The IR and frt sequences are remnants from previous vectors that were integrated by SB100x transposase. They are irrelevant for this study, which does not employ the SB100x integrating system. (C) Editing rate in CD34 + cells from 3 healthy, G-CSF-mobilized donors. Cells were transduced at an MOI of 2,000 viral particles (vp)/cell, and 3 days later, genomic DNA was analyzed by Sanger sequencing for the -113 A>G conversion. ABEmax is an early adenine base editor version (12). (D) Sequences of the top 12 edited alleles in CD34 + cells from donor #1 after transduction with HDAd-EF1α.ABE8e. The target site -113 A (A8 in the sequence shown in the lower panel) is marked by a green box. Editable window in the spacer by ABE8e is highlighted in orange. (E and F) In vitro transduction studies with donor #1 CD34 + cells that were subjected to erythroid differentiation (ED) and 1 cycle of O 6  ABE8e editing approach. BM lineage -(Lin -) cells (a fraction enriched for HSCs) from β-YAC/CD46 mice were transduced with HDAd-EF1α.ABE8e and transplanted into lethally irradiated C57BL/6 mice (Figure 5A). Engraftment, measured based on human CD46 expression on PBMCs, was > 95% after week 8, indicating that ex vivo HSC transduction and editing did not affect HSC functions ( Figure 5B). After transplantation, A>G conversion rates measured in PBMCs by Sanger sequencing were > 90% at the first time point analyzed (4 weeks after transplantation) and remained stable over the 16-week (week 16 primary [week 16-P]; see Figure 5A) observation period ( Figure 5C). As seen in in vitro studies with human CD34 + cells ( Figure 1D), substantial bystander A>G editing was observed, specifically at position A5. A>G conversion rates in splenocytes, BM mononuclear cells (MNCs), and Lincells, as well as CFUs were in agreement with the PBMC data ( Figure 5D). The comparison of editing rates in the transplanted cells (~38% -113 A>G) and in week-16-P mice (>95% -113 A>G) likely indicates preferential transduction of repopulating HSCs and expansion of edited cells after transplantation ( Figure 5E). The vast majority of edits were A>G conversions ( Figure 5E). The rate of A>T and A>C conversions was ~3 and ~4 orders of magnitude lower, respectively. NGS of the target area confirmed the A>G conversion rates detected by Sanger sequencing and determined that indels (mostly deletions) were < 1.5% of all reads ( Figure 5F and Supplemental Figure 5). HSCs contain 2 HBG promoters (HBG1 and HBG2) on each allele. To assess editing of the 4 target sites on a single-cell level, we isolated individual progenitor colonies and sequenced their DNA (n = 20 colonies). The data show 100% biallelic editing of both promoters ( Figure 5G). γ-Globin protein was detected by flow cytometry in 80%-90% of peripheral RBCs starting after week 8 ( Figure 6A), in accordance with the HSC engraftment/expansion kinetics ( Figure 5B) and previous studies with CRISPR/Cas9 (37). Relative γ-globin expression levels were measured on the mRNA and protein levels. HBG mRNA levels were approximately 35% of human HBB mRNA levels and approximately 22% of mouse HBA and HBB mRNA ( Figure 6B). In mice transplanted with ex vivo-edited Lincells, human γ-globin levels were about 30% of human β-globin chains and approximately 20% of mouse α-or β-globin chains ( Figure 6C). Editing rates and γ-globin reactivation were stable in secondary recipients, demonstrating that editing occurred in long-term repopulating HSCs (Supplemental Figure 6).
In summary, HDAd-EF1α.ABE8e confers efficient ex vivo HSC editing without toxicity associated with the transduction and editing processes. After editing, γ-globin levels were 35% of human β-globin levels without any O 6 -BG/BCNU in vivo selection.
In vivo HSC genome editing. One of our hypotheses was that, after in vivo HSC transduction, HDAd genomes and, consequently, ABE8e and mgmt P140K expression would be gradually lost due to cell division and vector genome degradation (Supplemental Figure 7). We, therefore, designed a new in vivo HSC transduction/selection regimen shown in Figure 7A. After mobilization of CD46/β-YAC transgenic mice and i.v. injection of a single HDAd-EF1α.ABE8e vector, O 6 -BG/BCNU treatment was started at day 2 and repeated at days 12 and 26 to capitalize on mgmt P140K expression from episomal HDAd-EF1α. ABE8e genomes and to dilute out the vector by stimulating cell division. The latter was demonstrated by qPCR that measured vector copy numbers (VCN) in BM MNCs. In in vivo transduced primary mice, during the 16-week experiment, the VCN declined by 2-4 orders of magnitude ( Figure 7B), resulting in undetectable ABE8e and mgmt P140K mRNA levels at week 16 after transduction (data not shown). Base editing was measured by Sanger sequencing in PBMCs ( Figure 7C). A>G conversion rates increased after selection to 50%-90% (average 67%) for the -113 target site with a similar pattern of bystander adenine editing seen in the ex vivo study. Furthermore, conversion rates were similar in PBMCs, splenocytes, and BM MNCs ( Figure 7D). Editing rates were comparable for CD3 + (T cell), CD19 + (B cell), Gr-1 + (myeloid), and Ter-119 + (erythroid) lineages in the BM, indicating that editing occurred in multipotential HSCs and that HBG1/2 editing (and γ-globin reactivation) did not provide a proliferative  advantage or disadvantage for any specific lineage in β-YAC mice ( Figure 7E). NGS demonstrated again a high specificity for A>G conversions with less than 1.5% of indels (Figure 7, F-H, and Supplemental Figure 8). Editing of the 4 HBG target sites on a single-cell level was less efficient than in the ex vivo setting ( Figure 7I). In total, 66.7% and 71.4% of colonies had biallelic conversion in the HBG1 and HBG2 promoters, respectively. A total of 21% (HBG1) and 23.6% (HBG2) of colonies had monoallelic -113 A>G conversions. The percentages without HBG1 and HBG2 editing were 21% and 23.8%, respectively.
Flow cytometry measuring γ-globin in peripheral RBCs showed a gradual increase from 30% (average at week 4) to 70% γ-globin + RBCs (average at week 16-P) ( Figure 8A). A similar kinetic has been observed in our previous studies (12,38). Levels of human γ-globin (hHBG) mRNA relative to human HBB mRNA were 30%, on average, at week 16 after in vivo transduction ( Figure 8B). Relative to mouse HBA and HBB mRNA, the level of hHBG mRNA was 12% and 15%, respectively. Human γ-globin protein levels, measured by HPLC of erythrocyte lysates, were about 25% of human β-globin chains and approximately 15% or 20% of mouse α-or β-globin chains, respectively ( Figure 8C). The presence of HbF in peripheral RBCs of in vivo-treated CD46/β-YAC mice was also shown by staining of cytospins with a human HbF specific antibody (Figure 8, D and E). The signals show a spectrum of intensity, most likely related to how many of the 4 target sites were edited. HbF staining on human peripheral blood and cord blood RBCs were used as negative and positive controls, respectively. Editing rates and percentages of γ-globin + RBCs were maintained in secondary recipients (Supplemental Figure 9).
In summary, in vivo HSC transduction by a single i.v. HDAd injection into mobilized β-YAC/CD46 mice followed by early in vivo selection resulted in > 60% editing at the -113 A site. In total, 25% γ-globin of human β-globin is expressed in > 60% of erythrocytes in healthy mice.
Off-target editing. Using nonintegrating vectors eliminates the risk of insertional mutagenesis. However, the application of genome editing enzymes bears the possibility of genotoxicity due to off-target editing. To assess this, we screened potential off-target editing sites using Circularization for In vitro Reporting of Cleavage Effects by sequencing (CIRCLE-Seq), a highly sensitive in vitro method capable of interrogating genome-wide off-target activity (39) (Figure 9). CD46/β-YAC genomic DNA was cleaved with SpCas9 coupled with sgHBG#2 guide RNA for CIRCLE-Seq. The sgHBG#2 was identical to that used in the HDAd-EF1α.ABE8e vector. CIRCLE-Seq identified a total of 272 candidate off-target sites (OTS) (Supplemental Table 1). To investigate off-target editing in in vivo-transduced mice, we performed targeted deep sequencing for the top 20 of the 272 sites based on the CIRCLE-Seq read count. Genomic DNA from the mouse with the highest on-target editing (65%) (week 16 secondary [week 16-S]) and a naive mouse was used. Results showed that no significant off-target base editing in the editable window (positions 3-14) occurred at these 20 sites (Figure 9, middle panels). Notably, approximately 4% of alleles with single nucleotide variants in OTS #11 were found in both naïve and treated animals. Indel frequencies around the predicted nicking sites (70 bp at each side) of the 20 candidates showed no significant difference between naive and treated animals (Figure 9, bottom panel). We conducted further off-target analyses using Cas-OFFinder, an in silico algorithm to predict potential OTS. A total of 76 and 89 sites with 4 or fewer mismatches to sgHBG#2 guide RNA were nominated in mouse and human genome, respectively ( Figure 10, A and B, and Supplemental Tables 2 and  3). Twenty-four of the 76 Cas-OFFinder sites in the mouse genome overlapped with the CIRCLE-Seq OTS list (31.6%). The top 5 ranked sites from each list were amplified from the in vivo-transduced mouse (analyzed above) or in vitro-transduced β-thalassemia CD34 + cells for amplicon NGS. Results showed no significant editing at these sites.
Effect on transcriptome. We compared RNA-Seq data of total RNA from BM MNCs of CD46/ β-YAC mice before treatment and mice at week 16 after in vivo HSC transduction with HDAd-EF1α. ABE8e (n = 3 animals) (Figure 11 and Supplemental Tables 5 and 6) Of the 2,742 genes annotated to the human genome, only 13 were considered differentially expressed (adjusted P < 0.01)  Table 5). Six of them were upregulated and were clustered based on their biological pathways, into 3 groups: (a) hemoglobin genes, specifically HBG2, HBG1, and HBD, whose upregulation is most likely the result of in vivo editing of the β-globin locus present in CD46/β-YAC mice, and (b) mgmt, as a result of residual expression of human mgmt P140K mRNA from the HDAd vector. Based on the mouse annotation process, we mapped 9,037 genes in total, from which 7 were considered differentially expressed (adjusted P < 0.01) (Supplemental Table 6). However, the level of upregulation was barely above the threshold set at 2.0-fold log. Of the 7 genes, 1 gene is listed in the Catalogue of Somatic Mutations in Cancer (COSMIC). This gene, Creg1, is upregulated in treated animals. Creg is considered to be a tumor suppressor, and its upregulation should not be associated with cancer (40). Additional studies are required to discover whether the modest upregulation of these genes is due to O 6 -BG/BCNU treatment.
Hematological analyses. There were no clinical side effects of the in vivo HSC transduction and selection approaches. Histological analyses of major organs did not show abnormalities. To further assess the safety of in vivo base editing with HDAd-EF1α.ABE8e, we performed hematological analyses of blood and BM samples of week 16 in vivo-transduced CD46/β-YAC mice. No significant differences between treated and untreated mice in blood cell counts (Supplemental Figure 10A), erythroid parameters (Supplemental Figure 10B), platelet counts (Supplemental Figure 10C), and reticulocyte counts (Supplemental Figure 10D) were found.
Furthermore, there were no significant differences in the lineage composition of BM cells (Supplemental Figure 10E). Moreover, in agreement with the disappearance of vector DNA over time, ABE8e was expressed only transiently, which triggered IgM but not a stable IgG anti-Cas antibody response (Supplemental Figure 10F).
In summary, in vivo HSC editing with HDAd-EF1α.ABE8e was found to be safe and yielded γ-globin reactivation at levels that are therapeutically relevant.

Discussion
The approach described here is aimed toward the reactivation of γ-globin/HbF for the treatment of β-thalassemia and SCD. HbF has powerful antipolymerization properties because its γ-globin subunits form mixed hybrid tetramers of 2 α-globin chains with 1 γ-globin and 1 mutated β-globin chain that are largely excluded from the polymer (41). We used a nonintegrating HDAd5/35++ vector that would trigger permanent genetic edits in HSCs through the transient expression of ABE8e and a sgRNA that targets the distal BCL11A binding motif in the HBG1/2 promoters. Studies were performed in ex vivo and in vivo settings in CD46/β-YAC transgenic mice. The ex vivo approach yielded > 80% HbF + RBCs with γ-globin levels that were 30% of human β-globin levels. Notably, this study was done without any in vivo O 6 -BG/BCNU selection. More importantly, we demonstrated that in vivo HSC transduction of CD46/β-YAC transgenic mice with a single HDAd-EF1α.ABE8e vector and early treatment with O 6 -BG/BCNU resulted in > 60% HbF + RBCs with human γ-globin levels that were 30% of those of human β-globin, in the absence of a competitive background (healthy mice). This outcome is potentially curative in patients with β-thalassemia and SCD. In individuals with HPFH, HbF makes up approximately 30% of total hemoglobin and is homogeneously distributed in the RBC population (42,43). Furthermore, murine ex vivo gene therapy studies with γ-globin lentivirus vectors demonstrated that long-term amelioration of disease of murine β-thalassemia and SCD occurred when approximately 20% gene-modified HSCs repopulated the marrow, when HbF levels were approximately 20%, and when HbF + cells constituted two-thirds of the circulating RBCs (44)(45)(46)(47).
In the absence of β-YAC mice with a thalassemia or SCD genotype or phenotype, we tested disease correction using CD34 + cells from patients with β-thalassemia and SCD. We have demonstrated that HDAd-EF1α.ABE8e transduction mediated efficient target site editing and reactivation of γ-globin after ED. This resulted in the reduction of ROS to background level and improved ED/expansion of thalassemia and SCD CD34 + cells; it also suggests that the in vivo base editing could be curative in patients with β-thalassemia and SCD.
Several factors could have contributed to the high efficacy of in vivo editing. The greatly enhanced catalytic activity of ABE8e (31) and the high level of ABE8e expression in HSC mediated by the EF1α promoter could have contributed. This is supported by comparative studies with HDAd-PGK.ABE8e, where ABE8e expression was driven by the relatively weak PGK promoter and where rates of editing (both ex vivo and in vivo) and γ-globin reactivation were significantly lower (Supplemental Figure 11). The prolonged expression of the base editing machinery from episomal HDAd vector in some transduced HSCs may also have contributed to the high efficacy of in vivo editing. The VCN declined only by 1 order of magnitude within 6 weeks, which implies that the editing machinery is active during that time. This is, in part, supported by the ex vivo study (performed without O 6 -BG/BCNU selection), where an increase of editing rates from the transplant (~38%) to > 95% in week 16 mice was observed. We speculate that the extended presence of the HBG1/2 sgRNA and ABE8e in HSCs could increase the probability for editing of poorly accessible target sites (i.e., sites blocked by heterochromatin at a certain stage of HSC differentiation). In this context, stimulation of cell division by O 6 -BG/BCNU selection could have further increased target site accessibility. As the data in secondary recipients indicate, editing must have occurred in primitive long-term repopulating HSCs (before cell division and differentiation).
Bystander editing of adenines near the -113 target site -specifically -116 A>G, which would destroy the BCL11A binding motif and, most likely, further inhibit the binding of the repressor -may also have contributed to the high efficacy of in vivo editing. The -116 A>G bystander conversion was found in > 99% of -113 A>G edited alleles (Supplemental Figure 12). Also, in the absence of the -110 A>G substitution, a GATA binding motif would be created similar to the -113A>G HPFH (32). This would enable binding of the transcriptional activator GATA1 to the destroyed BCL11A motif and could further increase γ-globin reactivation. Expansion of edited HSPCs with episomal HDAd-EF1α.ABE8e genomes by early treatment with O 6 -BG/BCNU, capitalizing on the prolonged presence of episomal vector genomes, may also have contributed to the high efficacy of in vivo editing. While we did not perform in vivo studies without O 6 -BG/BCNU treatment, the in vitro data with CD34 + cells from healthy donors ( Figure 1F) and patients with β-thalassemia (Figure 2) illustrate the effect of selection.  Notably, in patients with β-thalassemia and SCD, γ-globin-expressing erythroid cells would have a survival advantage created by the disease background. In this context, we expect that rates of γ-globin reactivation achieved with our approach could be even higher in patients with hemoglobinopathies and the in vivo selection with O 6 -BG/BCNU on demand.
Safety concerns with our approach are related to the i.v. injection of HDAd5/35++ vectors and genotoxicity due to off-target base editing. HDAd vectors are sensed by the RES and can trigger the release/ production of proinflammatory cytokines. We have developed pharmacological approaches to avoid these responses in mice (2) and nonhuman primates (NHPs) (8). Furthermore, i.v.-injected HDAd5/35++ vectors preferentially transduce mobilized HSCs (8,48), minimizing base editing in nonhematopoietic tissues.
Recently, virus-like particles (VLPs) have been used for in vivo editing of murine liver or retina (49). Furthermore, nonviral vehicles based on lipid nanoparticles (LNPs) for mRNA delivery (50, 51) hold great promise for in vivo HSC genome editing. However, these new approaches also face the hurdles accounted with intravascular administration of viral vectors -namely, the unproductive sequestration, which requires high vector doses, and the associated acute toxicity. Furthermore, while viruses have evolved highly efficient mechanisms for cellular uptake and intracellular trafficking, it is not straightforward to incorporate these functions into nonviral delivery vehicles.
Clearly, prolonged expression of the editing machinery could increase the frequency of off-target editing and the genotoxicity risk. Our data indicate that off-target DNA or RNA editing is minimal, while on-target activity is high. No hematological or histological abnormalities were found in primary and secondary mice over a total period of 32 weeks. More than 98% of edits at the target site were the desired A>G conversions; less than 1.5% of reads had small deletions or insertions in the observation window. Interestingly, the indel frequency in secondary mice (at week 16) was significantly lower than in primary mice, indicating a selection against cells with indels over time. NGS of OTS predicted by CIRCLE-Seq or Cas-OFFinder did not show editing in the quantification window. Off-target base editing can manifest as guide-independent, spurious editing of RNA (52,53) or genomic DNA (31). This, as well as clonal expansion of malignant cells, could result in alterations in the transcriptome. While RNA-Seq shows the expected activation of human γ-globin genes (10-to 100-fold log 10 ) and residual human mgmt P140K mRNA, changes in the mouse transcriptome were minimal (2.0to 2.15-fold log 10 ), and it is debatable whether these small alterations have biological consequences.
Extensive off-target analyses are crucial for a potential clinical translation of the approach. Therefore, in upcoming NHP studies with an improved HDAd-EF1α.ABE8e vector, great attention will be given to genomic analyses and potential long-term effects of in vivo base editing.
Our current attempts are focused on further improvement of the safety and efficacy of the approach, including new HDAd vector platforms (54), alternative mobilization regimens (55), and in vivo selection strategies. The latter includes an in vivo selection strategy that would not involve low-dose BCNU (56).
We believe that our study is an additional step toward the vision to treat hemoglobinopathies by a simple i.v. injection of a nonintegrating vector. (A) Analysis based on in silico OTS prediction using Cas-OFFinder software for the mouse genome. Seventy-four sites were nominated, from which the top 5 were subjected to NGS. Amplicons were prepared using genomic DNA template from a naive mouse and an in vivo-transduced mouse with the highest on-targeting editing rate. The middle and lower panels show the frequency of A>G conversions and indels, respectively. (B) Analysis based on in silico OTS prediction using Cas-OFFinder software for the human genome. Eighty-nine sites were nominated, from which the top 5 were subjected to NGS. Amplicons were prepared using genomic DNA from erythroid cells derived from CD34 + cells of a thalassemia patient that were in vitro transduced with HDAd-EF1α.ABE8e.

Methods
Supplemental Methods are available online with this article. CD34 + cell culture CD34 + cells from G-CSF-mobilized healthy adult donors were provided by the Fred Hutch Cell Processing Facility. CD34 + cells from patients with β-thalassemia and SCD were previously collected during mobilization clinical trials conducted at George Papanikolaou Hospital, Thessaloniki, Greece (57). CD34 + cells were recovered from frozen stocks and incubated overnight in StemSpan H3000 medium (Stemcell Technologies) supplemented with penicillin/streptomycin, Flt3 ligand (Flt3L, 25 ng/mL), IL-3 (10 ng/mL), thrombopoietin (TPO) (2 ng/mL), and stem cell factor (SCF) (25 ng/mL). Cytokines and growth factors were from Peprotech. Cytokines and growth factors were from Peprotech. CD34 + cells were transduced with HDAd5/35++ vectors in low-attachment 12-well plates. Figure 11. Differentially expressed genes after in vivo base editing HDAd-EF1α.ABE8e. RNA-Seq analysis was performed with BM MNCs collected before treatment and at week 16 after in vivo HSC transduction/selection with HDAd-EF1α.ABE8e (n = 3 females, age 8-12 weeks). The genome of CD46/β-YAC mice contains the WT 248 kb β-globin locus (β-YAC) including the LCR, γ-and β-globin genes. Shown are genes with altered mRNA expression based on their adjusted P value after annotation with the human genome. (A) Altered mRNA expression after annotation with a reference human genome. Statistically significant changes (P < 0.01) are highlighted in red (upregulated). (B) Volcano plot of mRNA data after annotation with the mouse reference genome are shown. Statistically significant changes (P < 0.01) are highlighted in red (upregulated) or blue (downregulated). Statistical analyses were performed using 2-way ANOVA.