Efficient gene base editing in living organisms using small single-virus carriers with compact adenine editors

Starting with the small, robust editor SaABE8e (3.9 kb)³¹ and its PAM-variant SaKKH-ABE8e^31,35, we set out to optimize the components of the AAV genome to yield size-minimized AAV-based delivery of adenine base editors in which the entire base editor, its guide RNA, and all necessary promoters and regulatory sequences are present in a single AAV (≤~5.0 kb, including ITRs). We first identified high-efficiency guide RNAs targeting Pcsk9 to allow evaluation of in vivo genome editing (irrespective of protein knockdown) in N2A and 3T3 cells by transfecting plasmids encoding SaABE8e or SaKKH-ABE8e and corresponding sgRNAs with spacers targeting the endogenous Pcsk9 gene. The editing efficiency of each sgRNA was analysed by targeted high-throughput DNA sequencing (HTS) (Supplementary Fig. 1). The most efficient guide RNA installed a W8R coding mutation in Pcsk9 using SaABE8e (Supplementary Fig. 1).

To encode the full-length SaABE8e protein on a single-AAV genome, we first used the small ubiquitous promoter EFS (EF-1α short) and a terminator we previously validated to yield efficient split-intein base editor delivery: the gamma portion of woodchuck hepatitis virus post-transcriptional regulatory element (WPRE, gamma subunit W3) with bovine growth hormone (bGH) polyadenylation signal³⁶. To simplify production when testing multiple AAV architectures, sgRNA targeting Pcsk9 W8 was provided on a separate AAV.

Both intact and intein-split SaABE8e AAVs resulted in dose-dependent and tissue-dependent editing activity consistent with the tropism of AAV9 across collected tissues^37,38, with liver showing the highest editing efficiencies, followed by heart, then skeletal muscle. Intact SaABE8e AAV yielded robust editing at both doses administered, reaching 59%, 12% and 5.3% editing of bulk liver, heart and skeletal muscle, respectively, comparable to or higher than the editing efficiencies achieved with intein-split SaABE8e in all tested tissues and doses (Fig. 1b). Decreasing the amount of sgRNA AAV to one half or one quarter of the amount of base editor AAV did not affect editing efficiency in the liver, but decreased editing efficiency by ~1.4-fold per 2-fold decrease in administered sgRNA AAV in both heart and muscle (Supplementary Fig. 2), indicating that the dose of guide RNA partially limits editing efficiency in extrahepatic tissues under these conditions.

Next, we sought to improve editing efficiency and minimize the size of ABE AAV by identifying minimal necessary elements on the AAV genome. We designed and compared five AAV genome architectures for delivery of SaABE8e to assess the impact of modifying the EFS promoter by adding a minimal minute virus of mice (MVM) intron³⁹, modifying the terminator by removing the truncated WPRE gamma subunit W3, or replacing the bGH polyadenylation signal with an SV40 late polyadenylation signal. We designed and produced the following AAV expression cassettes: (1) EFS-SaABE8e-W3bGH, (2) EFS-MVM-SaABE8e-W3bGH, (3) EFS-MVM-SaABE8e-bGH, (4) EFS-SaABE8e-bGH and (5) EFS-SaABE8e-W3-SV40. Each of the five AAV candidates was administered to 6–7-week-old wild-type C57BL/6J mice by retro-orbital injection at a high (4 × 10¹¹ vg editor AAV plus 4 × 10¹¹ vg sgRNA AAV) or low (4 × 10¹⁰ vg editor AAV plus 4 × 10¹⁰ vg sgRNA AAV) dose (Fig. 1c).

At 3 weeks post injection, we collected liver, heart and skeletal muscle, and analysed each tissue by HTS. Editing efficiencies followed a consistent pattern among architectures, with the highest efficiency construct of EFS promoter driving SaABE8e expression and a bGH polyadenylation signal without W3 (EFS-SaABE8e-bGH) outperforming the other architectures across all assessed doses and tissues. These data demonstrate that the cis-acting W3 element is not necessary for sufficient expression of SaABE8e from the EFS promoter in these tissues and cell types, and highlights the importance of assessing AAV elements in the context of a specific editing application.

Next we assessed the single-AAV ABE construct at a dose of 8 × 10¹¹ vg per mouse, equal in total AAV dose per mouse to that of the high-dose experiments described above requiring two AAVs. We observed further improvements in editing compared with the lower doses, especially in heart and muscle, which were now edited with 33% and 22% average efficiency, respectively. This level of editing corresponds to a 2.1-fold and 2.5-fold increase in editing in heart and muscle, respectively, compared with the highest observed level of base editing from dual-AAV SaABE8e with editor and sgRNA delivered on separate AAVs (Fig. 2b, P = 0.00048 and P = 0.0020, respectively, by unpaired t-test) at the same total dose of AAV. The relatively wide editing window of SaABE8e is maintained in vivo and indels remain low in each tissue (Supplementary Fig. 3).

We also quantified delivered editor AAV genomes from tissue by digital droplet PCR (ddPCR) and found that editing efficiencies correlate with the quantity of delivered genomes in each tissue (Extended Data Fig. 1), with liver being much more amenable to transduction with AAV9 than heart and muscle, consistent with previous analysis of the biodistribution of AAV9⁴⁰. We found that tissues with largest differences in editing between single- and dual-AAV strategies were less efficiently transduced than liver. Heart and muscle were similarly transduced, which may indicate that the EFS promoter is more active in heart or may reflect effects of tissue heterogeneity.

These levels of base editing achieved in the liver, heart and muscle from single-AAV ABE injection would be sufficient to offer therapeutic benefit for many disorders^21,41–43. These data also represent some of the highest reported somatic cell in vivo base editing in these tissues at clinically relevant doses^{21,26,36,44,45} of ≤10¹⁴ vg kg⁻¹. Together, these results demonstrate that this engineered single-AAV ABE architecture mediates robust genome editing in vivo and highlights the benefits of single-AAV systems over dual-AAV methods, especially in less well-transduced tissues or when using lower total doses of AAV.

To assess the collective targeting scope of this suite of four small ABE8e variants (SaKKHABE8e, Nme2ABE8e, CjABE8e and SauriABE8e), we determined the number of adenines in the entire hg38 human reference genome that are targetable in principle by at least one of these variants. We analysed the surrounding sequence context of each adenine for the presence of a small-ABE8e-targetable PAM that would place each adenine within an appropriate base editing window. This analysis revealed that 82% of all adenines in the human genome are theoretically targetable by at least one of these four small ABE8e variants (Fig. 4d). These data suggest that the single-AAV ABE system can potentially target the vast majority of adenines across the genome, although bystander editing in most cases will result in additional mutations, approximately half of which will be non-silent⁵⁴.

To test the in vivo therapeutic potential of the single-AAV ABE system, we installed in mice mutations that are associated with decreased cardiovascular disease risk in humans^55–57 by precisely knocking down Pcsk9 or Angptl3 protein levels. Knockdown of these proteins reduces levels of serum biomarkers including circulating protein and total cholesterol, as well as triglycerides when knocking down Angptl3, facilitating robust functional assessment of editing efficiency. We used SaABE8e, SaKKH-ABE8e and the newly designed SauriABE8e to disrupt start codons, splice donors and splice acceptors⁵⁸ to precisely block production of the targeted protein without relying on double-strand breaks or indel formation.

We first designed and measured editing activity of guide RNAs expected to disrupt production of Pcsk9 or Angptl3 by transfection of plasmids encoding a size-reduced ABE8e and sgRNA targeting sites throughout human PCSK9 in HEK293T cells (Supplementary Fig. 4a) or mouse Pcsk9 and Angptl3 in Neuro-2a cells (Supplementary Fig. 4b,c and Extended Data Fig. 2). Base editing efficiencies varied from undetectable to 89% as measured by deep sequencing of genomic DNA at the targeted loci. We advanced to in vivo experiments three highly efficient SaKKH-ABE8e guide RNAs targeting exon 1 splice donor of human PCSK9, and mouse Pcsk9 and exon 6 splice donor of mouse Angptl3, as well as a SauriABE8e sgRNA targeting the exon 1 splice donor of mouse Pcsk9.

We also compared the editing activity of single-AAV8 SaKKH-ABE8e (1 × 10¹¹ vg) targeting the Pcsk9 exon 1 splice donor to the previously optimized³⁶ dual-AAV split-intein SpABE8e architecture paired with an SpCas9 sgRNA validated to efficiently edit and knockdown Pcsk9 in vivo^44,45 by targeting the same splice donor. We administered single-AAV8 SaKKH-ABE8e and dual-AAV8 SpABE8e at the same total dose per mouse to 6–8-week-old C57BL/6J mice at 3 doses (1 × 10¹¹ vg, 1 × 10¹⁰ vg or 1 × 10⁹ vg total AAV per mouse) and measured disruption of Pcsk9 exon 1 spice donor in liver at 4 weeks after administration, by HTS. The maximum vg kg⁻¹ dose used (5 × 10¹² vg kg⁻¹ for a 20 g mouse) is comparable to or lower than those used in gene therapy non-human primate studies and human clinical trials^6,63. We observed that the single- and dual-AAV ABE systems performed similarly at each dose, with the single-AAV yielding 54%, 38% and 3.7% average editing in liver at a dose of 1 × 10¹¹ vg, 1 × 10¹⁰ vg and 1 × 10⁹ vg, respectively (Fig. 5c), and editing via dual-AAV SpABE8e in liver at the same dose of AAV averaging 57%, 35% and 1.0%, respectively (no significant difference at any dose by unpaired t-test). These results collectively show that single-AAV ABE performs comparably to the highly active previously optimized dual-AAV SpABE8e at a range of doses at a therapeutically relevant locus³⁶.

To investigate the modest apparent editing efficiency improvement of the single-AAV system compared with split SpABE8e in liver, we directly compared single-AAV8 SaKKH-ABE8e and dual-AAV8 intein-split SaKKH-ABE8e with matched promoter and polyA by systemic retro-orbital delivery to 6–8-week-old C57BL/6 mice at two doses (Fig. 5d). Single- and dual-AAV intein-split SaKKH-ABE8e show similar editing at the high dose of 1 × 10¹¹ vg (50% and 46%, respectively, not significant by unpaired t-test) but markedly lowered activity for the dual-AAV intein-split SaKKH-ABE8e compared with single-AAV SaKKH-ABE8e at a dose of 1 × 0¹⁰ vg (5.2% versus 29%, respectively, P = 0.0004). These data indicate that single-AAV SaKKH-ABE8e may perform similarly to optimized dual-AAV SpABE8e under conditions in which AAV transduction is already at or near saturation, perhaps because the increased activity of SpABE8e and/or increased activity afforded by the cis-regulatory elements that can be included with the extra space on two AAV genomes can overcome limitations of intein-splitting and dual transduction. For applications in which AAV transduction is well below saturation, however, single-AAV delivery can result in substantial editing efficiency improvements. We also assessed the integrity of AAV genomes packaged in single-AAV ABEs, by alkaline gel electrophoresis of single-AAV ABEs as well as smaller intein-split dual-AAV ABEs (Supplementary Fig. 5) and found that single-AAV ABEs package full-length and truncated genomic species, consistent with their size being at the AAV packaging limit¹⁹.

To assess whether the efficient editing observed in the liver can translate into efficient target gene knockdown and concomitant reduction in circulating lipid levels, we serially bled editor-treated mice and measured plasma levels of the targeted protein and total cholesterol. A non-targeting control of dual-AAV ABE7.10 targeting Dnmt1, an edit not expected to affect cholesterol or lipid metabolism, was included for comparison (Supplementary Fig. 6). We observed nearly complete protein knockdown in all experimental conditions at a dose of 1 × 10¹¹ vg (5 × 10¹² vg kg⁻¹) single-AAV SaABE8e or SaKKH-ABE8e by 4 weeks, with most knockdown evident by 2 weeks (Fig. 5e,f,g, and Supplementary Figs. 7a,c and 8a,c). On average, single-AAV ABE treatment at this dose resulted in 99%, 91% and 94% knockdown of human PCSK9, mouse Pcsk9 and mouse Angptl3 protein levels, respectively, compared with control animals treated with AAV encoding the Dnmt1-targeting guide RNA, again matching the knockdown of these proteins achieved by state-of-the-art LNP-mediated ABE mRNA delivery to the liver^45,62. This high level of protein knockdown is also consistent with observed editing levels (Fig. 5b) since the hepatocytic tropism of AAV8⁶⁴ and the fact that hepatocytes constitute roughly 70% of the murine liver⁶⁵ imply that 44–61% bulk liver editing corresponds to ~60–85% editing of hepatocytes.

Protein knockdown also resulted in decreased circulating cholesterol in all ABE-treated mice (Fig. 5h–j, and Supplementary Figs. 7b,d and 8b,d). Plasma total cholesterol in human PCSK9-targeted mice decreased by 24% from baseline levels to 45 mg dl⁻¹ after 4 weeks. At the highest AAV dose in Pcsk9-targeted mice, plasma cholesterol was lowered by an average of 25% compared with age-matched non-targeting controls to 53 mg dl⁻¹ after 4 weeks (Fig. 5i), similar to the degree of cholesterol lowering observed in liver-specific Pcsk9 knockout mice⁶⁶. For Angptl3-targeted mice, we observed a 38% decrease in plasma cholesterol compared with age-matched non-targeting controls to 44 mg dl⁻¹. These results demonstrate substantial lowering of cholesterol using single-AAV ABEs.

We assessed the dose-dependence of AAV dose and editing on circulating Pcsk9 and total cholesterol. Knockdown of mouse Pcsk9 and decrease in total plasma cholesterol were dose-dependent and closely reflected the level of editing observed at each dose (Supplementary Fig.). Dual SpABE8e, which effected editing to similar levels as single-AAV ABE8e, also resulted in decreased plasma cholesterol in a dose-dependent manner (Supplementary Fig.). For both single- and dual-AAV ABEs targeting mouse Pcsk9, cholesterol and protein knockdown correlated closely with editing percentage, regardless of editor type administered. 7 7

We also measured plasma triglycerides in Angptl3-targeted mice, as loss-of-function alleles of Angptl3 are known to reduce levels of both cholesterol and triglycerides⁶⁷. In the Angptl3-targeted mice, we observed a 45% decrease in circulating triglycerides compared with non-targeting control to 25 mg dl⁻¹ after 4 weeks (Fig. 5k and Supplementary Fig. 8e). The editing and reduction of circulating Angptl3, cholesterol and triglycerides achieved here with single-AAV ABE is, to our knowledge, the highest thus far reported upon targeted genome editing to knockdown Angptl3^62,68. Together, these results demonstrate robust base editing at multiple therapeutically relevant loci achieved with single-AAV ABEs, resulting in strong effects on target protein level and metabolic changes in adult mice.

Lastly, we assessed liver morphology and off-target editing in mice treated with single-AAV ABEs. We performed histology on livers from mice treated with single-AAV8 SaKKH-ABE8e and guide targeting PCSK9 exon 1 donor at 1 × 10¹¹ vg 4 weeks after administration and did not find evident morphological changes compared to untreated mice (Supplementary Fig. 9). Although off-target editing in cell culture has been characterized for SaCas9-based genome editing agents, no off-target editing in tissues treated with SaCas9-based editing agents in vivo has been reported^22,69. To assess single-AAV ABEs in vivo off-target editing, we sequenced the top three computationally predicted sites^70,71 from liver tissue of C57BL/6J mice treated with 1 × 10¹¹ vg single-AAV8 SaKKH-ABE8e, single-AAV8 SaKKH-ABE8e V106W, or dual-AAV8 intein-split SaKKH-ABE8e targeting mouse Pcsk9 exon 1 at 4 weeks after administration. We observed low but detectable (up to 0.45%) and dose-dependent editing at one off-target site in vivo (Extended Data Fig. 3a,b), suggesting the importance of considering off-target editing outcomes when using single-AAV ABEs, even though observed off-target editing was relatively rare at the sites examined. Off-target editing in vivo at this site was ameliorated by addition of the TadA V106W mutation, which has been reported to lower guide-independent DNA off-target editing and mRNA off-target editing^31,61. Off-target editing in vivo was also ameliorated by delivery via dual-AAV intein-split SaKKH-ABE8e, which may be due to inherent lower overall activity of intein-linked ABE8e, or the lower dose of complete base editors, although we did not observe any significant difference between full-length or intein-split SpABE8e off-target editing in N2A cells by plasmid transfection in vitro (Extended Data Fig. 3c). Finally, we assessed in vivo off-target mRNA editing in single-AAV8 SaKKH-ABE8e-treated mice by analysing complementary DNA amplicons of mouse homologues of demonstrated ABE mRNA off-target human transcripts⁶¹, some containing partial TadA recognition sequences, and observed no off-target mRNA editing compared with untreated mice (Extended Data Fig. 4). These results suggest that single-AAV ABEs maintain low levels of off-target DNA and RNA editing in vivo for the guide RNA tested, and that deaminase mutations can further minimize off-target editing in vivo.

Expression vectors for tissue culture were cloned using KLD, Gibson or USER assembly. sgRNA expression plasmids were cloned via KLD or Goldengate assembly to install protospacers as indicated in Supplementary Table. Plasmids were constructed via USER assembly or Gibson assembly of PCR-amplified fragments. Plasmids encoding recombinant AAV (rAAV) genomes were cloned by Gibson assembly of plasmid restriction fragments and PCR amplicons with Gibson-compatible overhangs. All plasmids for mammalian tissue culture experiments were purified using Plasmid Plus Maxiprep or Midiprep kits (Qiagen), ZymoPURE II Midiprep kit (Zymo Research) or PureYield Plasmid Miniprep kits (Promega). Key plasmids developed in this study will be available through Addgene. 1

HEK293T cells (ATCC CRL-3216) and Neuro-2A cells (ATCC CCL-131) were maintained in Dulbecco’s modified Eagle’s medium (DMEM) plus GlutaMax (Thermo Fisher) supplemented with 10% (v/v) fetal bovine serum (FBS) at 37 °C with 5% CO₂. At 16–24 h before transfection, HEK293T cells or N2A cells were seeded on 96-well plates (Corning) at 1.4 × 10⁴–2.0 × 10⁴ cells per well at >90% viability; for SauriABE transfections, HEK293T cells were seeded in 48-well plates (Corning) at 4.0 × 10⁴ cells per well at >90% viability. Cells in 96-well plates were transfected at approximately 70–85% confluency with 0.5 μl lipofectamine 2000 (Thermo Fisher), and 187.5 ng base editor plasmid and 37.5 ng sgRNA plasmid per well (180 ng editor and 60 ng sgRNA for SauriABE8e). Cells in 48-well plates were transfected with 1.5 μl lipofectamine 2000, with 750 ng editor and 250 μl sgRNA. Cells were cultured for 72 h after transfection, then media was removed, cells were washed with 1× PBS (Thermo Fisher), and genomic DNA was extracted by addition of 30–60 μl lysis buffer (10 mM Tris-HCl, pH 7.5–8.0, 0.05% SDS and 20 μg ml⁻¹ proteinase K (New England Biolabs)) per well for 96-well plates and 150 μl per well for 48-well plates. Genomic DNA was stored temporarily at 4 °C or longer term at −20 °C until further use.

Genomic DNA was amplified by PCR using Phusion Hot Start II DNA polymerase or Phusion U Hot Start DNA polymerase with 0%–3% dimethylsulfoxide added. Barcodes for Illumina sequencing were added via a second PCR step, using 1 μl of the first PCR as a template. Total PCR cycles were kept to a minimum to avoid PCR bias. Barcoded PCR products were pooled according to amplicon, gel extracted (MinElute; Qiagen) and quantified by qPCR (KAPA; KK4824) or Qubit dsDNA HS assay kit (Thermo Fisher). Sequencing of pooled libraries was performed using Illumina MiSeq according to the manufacturer’s instructions. Primers for amplification of each locus from genomic DNA are compiled in Supplementary Table. 1

Sequencing reads were demultiplexed using MiSeq Reporter (Illumina). Alignment of amplicon sequences to reference sequence was performed using CRISPResso2⁸² with ‘discard_indel_reads’ on. For quantification of base editing, efficiency was calculated as percentage of (reads containing an A to G edit at given position without indels)/(number of total reads). Indels were calculated explicitly as (discarded reads)/(total aligned reads) × 100. Base editing at a given position was calculated explicitly as: (frequency of specified point mutation in non-discarded reads) × 100 × (100 – (indel reads))/100)).

AAV was produced as previously described³⁶. HEK293T clone 17 cells (ATCC CRL-11268) were maintained in DMEM plus GlutaMax (Thermo Fisher) supplemented with 10% (v/v) FBS without antibiotic in 150 mm dishes (Thermo Fisher, 157150) at 37 °C with 5% CO₂ and passaged every 2–3 d. Cells were split 1:3 the day before polyethyleneimine transfection with 5.7 μg AAV genome plasmid, 11.4 μg pHelper (Chlontech) and 22.8 μg rep-cap plasmid per plate. Media were exchanged for DMEM with 5% FBS the day after transfection. Four days after transfection, cells and media were collected using a rubber cell scraper (Corning), pelleted by centrifugation at 2,000 g for 10 min, resuspended in 500 µl hypertonic lysis buffer per plate (40 mM Tris base, 500 mM NaCl, 2 mM MgCl₂ and 100 U ml⁻¹ salt active nuclease (ArcticZymes, 70910–202)) and incubated at 37 °C for 1 h to lyse the cells. The media were decanted and combined with a 5× solution of 40% poly(ethylene glycol) 8,000 (PEG 8k; Sigma-Aldrich, 89510) in 2.5 M NaCl for a final concentration of 8% PEG/500 mM NaCl, incubated on ice for 2 h and then centrifuged at 3,200 g for 30 min. The pellet was resuspended in 500 μl of hypertonic lysis buffer per plate and added to the cell lysate. Crude lysates were either incubated at 4 °C overnight or taken immediately for ultracentrifugation.

Cell lysates were clarified by centrifugation at 2,000 g for 10 min and added to Beckman Quick-Seal tubes via 16-gauge 5” disposable needles (Air-Tite N165). A discontinuous iodixanol gradient was formed by sequentially floating layers: 9 ml 15% iodixanol in 500 mM NaCl and 1× PBS-MK (1× PBS plus 1 mM MgCl₂ and 2.5 mM KCl), 6 ml 25% iodixanol in 1× PBS-MK, and 5 ml each of 40% and 60% iodixanol in 1× PBS-MK. Phenol red at a final concentration of 1 µg ml⁻¹ was added to the 15, 25 and 60% layers to facilitate identification. Ultracentrifugation was performed using a Ti 70 rotor in a Sorvall wX+ ultracentrifuge (Thermo Fisher) at 58,000 r.p.m. for 2 h and 15 min at 18 °C. Immediately following centrifugation, 3 ml of solution was withdrawn from the 40–60% iodixanol interface via an 18-gauge needle. The solution was exchanged into cold PBS containing 0.001% F-68 using PES 100 kD MWCO columns (Thermo Fisher, Pierce 88533) and concentrated. The concentrated AAV solution was sterile filtered using a 0.22 µm filter, quantified by qPCR (AAVpro titration kit version 2; Clontech) and stored at 4 °C until use.

All experiments in live animals were approved by the Broad Institute and University of Pennsylvania Institutional Animal Care and Use Committees and were consistent with local, state and federal regulations as applicable, including the National Institutes of Health Guide for the Care and Use of Laboratory Animals. C57BL/6J mice (000664) for use in experiments were purchased from The Jackson Laboratory. Humanized PCSK9 mice were reported previously⁶⁰. All mice were housed in a room maintained on a 12 h light and dark cycle with ad libitum access to standard rodent diet and water except for 4 h fasts just before bleeds for plasma analysis.

AAV was diluted into 100 µl of sterile 0.9% NaCl USP (Fresenius Kabi, 918610) before injection. Anaesthesia was induced with 2–4% isoflurane. Following induction, as measured by unresponsiveness to bilateral toe pinch, the right eye was protruded by gentle pressure on the skin, and an insulin syringe was advanced, with the bevel facing away from the eye, into the retrobulbar sinus where AAV solution was slowly injected. One drop of proparacaine hydrochloride ophthalmic solution (Patterson Veterinary, 07–885–9765) was then applied to the eye as an analgesic. At collection, mice were euthanized by carbon dioxide asphyxiation. Genomic DNA was purified from minced tissue using gDNAdvance kit (Beckman Coulter, A48705↗) according to the manufacturer’s instructions and used as template for high-throughput sequencing. RNA was purified from 30 mg of snap-frozen liver tissue with RNeasy Plus mini kit (Qiagen 74134) according to the manufacturer’s instructions, then reverse transcribed to cDNA using SuperScript III first-strand synthesis supermix (Invitrogen, 18080–450) with an oligo dT primer, which was used as template for high-throughput sequencing.

Genomic DNA was purified from tissue using Beckman gDNAdvance kit (Beckman Coulter, A48705↗) according to the manufacturer’s instructions and used as template for ddPCR. ddPCR was carried out using ddPCR Supermix for Probes (BioRad, 1863026) with 10 ng of genomic DNA as template and 3 units NEB EcoRI-HF (R3101S) per reaction. Droplets were autogenerated and PCR was performed at an annealing and extension temperature of 61 °C for 2 min for a total of 60 cycles. Droplets were analysed on a QX200 droplet analyser and droplet fluorescence was quantified using QuantaSoft (BioRad). The sequence of primers and probes used are listed in Supplementary Table 4.

A custom Python script (Supplementary) was used to analyse the targetability of all adenosines in the hg38 human reference genome. An adenosine was counted as targetable if the surrounding genomic sequence context contained a small ABE8e-targetable PAM that would place that adenosine within an appropriate base editing window. The PAM sequences, protospacer lengths and base editing windows associated with each small ABE8e variant are provided in Supplementary Table. The percentage of calculated genomic adenines on each chromosome is shown in Supplementary Table. Code 5a 5b

Initial blood samples were collected following a 4 h fast. Age-matched littermates/colonymates were randomly assigned to experimental groups and administered AAV particles (n = 5) via retro-orbital injection. Blood samples were collected following a 4 h fast at 1-week intervals via the tail tip. After 4 weeks, all mice were killed by carbon dioxide asphyxiation after a 4 h fast. Whole livers were collected for genomic DNA isolation and analysis and for hematoxylin/eosin staining, and terminal blood samples were collected.

Pre-treatment and post-treatment plasma human PCSK9, mouse Pcsk9 or mouse ANGPTL3 were measured using the human proprotein convertase 9/PCSK9 Quantikine ELISA kit, mouse proprotein convertase 9/PCSK9 Quantikine ELISA kit or human angiopoietin-like 3 Quantikine ELISA kit, respectively, according to the manufacturer’s instructions (R&D Systems). Total cholesterol or triglyceride levels were measured using the Infinity cholesterol reagent or Infinity triglycerides reagent, respectively, according to the manufacturer’s instructions (Thermo Fisher).

A portion of the left medial lobe was fixed in 4% paraformaldehyde at 4 °C overnight, washed with PBS, then dehydrated gradually by serial substitution of PBS for 30%, 50%, 70% and 100% ethanol. Samples were kept at −20 °C until analysis, when they were paraffinized by the Rodent Histopathology Core of Harvard Medical School. Liver paraffin block was then cut into 5 μm sections followed by hematoxylin and eosin staining for histopathological examination.

Alkaline agarose gel (1% agarose in water with 50 mM NaOH and 1 mM EDTA) was prepared by dissolving agarose in water, allowing to cool but not solidify, then adding a 50× solution of NaOH and EDTA. The formed gel was submerged in 1× alkaline running buffer (50 mM NaOH, 1 mM EDTA) in a submarine style gel electrophoresis setup at 4 °C. AAV (5 × 10¹⁰ vg) was treated with DNAse I (NEB, M0303S), lysed in 1× alkaline lysis buffer (50 mM NaOH, 1 mM EDTA, 0.3% SDS, 5% glycerol and 0.0025% xylene cyanol) for 3 min at 95 °C, then cooled on ice. Samples were loaded onto the gel, then electrophoresed at 20 V for 15 h. The gel was neutralized in 0.1 M Tris at pH 8 for 1 h at 4 °C with rocking. The gel was stained in 4× SYBR Gold in 0.1 M NaCl at 4 °C with rocking and protection from light. The gel was briefly washed with deionized water, then imaged on a UV transilluminator.

Data are presented as mean ± s.e.m. unless otherwise noted. The number of independent replicates and statistical tests are described in the figure legends. All statistical tests were conducted using GraphPad Prism 9.

Further information on research design is available in thelinked to this article. Nature Research Reporting Summary