Adenine base editing in an adult mouse model of tyrosinemia

Abstract

Unlike traditional CRISPR-Cas9 homology-directed repair, base editing can correct point mutations without supplying a DNA-repair template. Here, we show in a mouse model of tyrosinemia that hydrodynamic tail-vein injection of plasmid DNA encoding the adenine base editor (ABE) and a single guide RNA can correct an A>G splice-site mutation. ABE treatment partially restored splicing, generated fumarylacetoacetate hydrolase (Fah)-positive hepatocytes in the liver, and rescued weight loss in the animals. We also generated Fah+ hepatocytes in the liver via lipid-nanoparticle-mediated delivery of chemically modified sgRNA and an mRNA of a codon-optimized base editor that displayed higher base-editing efficiency than the standard ABE. Our findings suggest that adenosine base editing can be used for the correction of genetic disease in adult animals.

Point mutations are a frequent cause of genetic disease, with ~50% of disease point mutations being G:C>A:T ¹. CRISPR genome editing, dependent on double-stranded DNA breaks (DSB) followed by nonhomologous end-joining or homology-directed repair (HDR), has been applied to many organisms^{2, 3}. However, major caveats of correcting point mutations through CRISPR genome editing include system efficiency, the introduction of DSBs and the need to provide DNA repair template. A new-generation of genome editing tool, Base Editing (BE) system, enables precise base changes (reviewed in ^{4, 5}), does not induce a high-level of DSBs, and finally does not require HDR donors for gene repair ^{4, 5}.

To date, two classes of BEs have been developed ^{1, 6}. Cytosine BEs that include cytosine deaminases, Cas9 nickases, and uracil glycosylase inhibitor, introduces C•G to T•A base changes⁶ and has been applied to successfully introduce a stop codon or correct T•A to C•G mutations in prokaryotes, fungi, plants, insects, amphibians, fish, and mammals ^{5, 7–11}. Adenine BEs comprise an evolved Escherichia coli TadA (ecTadA*, a tRNA adenosine deaminase evolved to accept DNA substrate) and Cas9 nickase, and mediate A•T to G•C substitutions¹. Local delivery of ABE by intramuscular injection of trans-splicing AAV was recently reported¹²; however, systemic delivery of ABE for correction of liver disease in adult animals has not been investigated.

To explore the therapeutic potential of ABE in the liver of adult animals, we chose a mouse model of hereditary tyrosinemia type I (HTI), a fatal genetic disease. HTI is caused by loss-of-function of fumarylacetoacetate hydrolase (FAH), a key enzyme of the tyrosine catabolic pathway ^{13, 14}. FAH deficiency leads to accumulation of toxic metabolic intermediates, causing apoptosis of mutant hepatocytes and severe liver damage. The Fah^mut/mut mouse model^{14, 15} harbors a homozygous G•C to A•T point mutation in the last nucleotide of exon 8, resulting in exon skipping and loss of FAH protein. 2-(2-nitro-4-trifluoromethylbenzoyl)-1,3-cyclohexanedione (NTBC)¹⁴ is an inhibitor of the tyrosine catabolic pathway upstream of FAH. Fah^mut/mut mice need to be kept alive on NTBC-supplemented water to prevent toxin buildup and liver damage¹⁴.

We and others recently reported that CRISPR can correct this Fah mutation through HDR^16–18 or allelic exchange¹⁹. Following correction by CRISPR, liver cells expressing the Fah enzyme, through their selective advantage, expand and repopulate the liver¹⁴. Thus, this mouse model is particularly suitable for testing ABE^{13, 15}.

Results

Use ABE to correct a Fah point mutation

The ABE editing window occupies positions ~4–9 with the 1^st, 5’ nucleotide (nt) of the candidate sgRNA counted as position 1 ¹. In order to edit the Fah mutation using ABE, we selected a previously validated Fah sgRNA ²⁰ (Fig. 1a). The disease associated point mutation is located at position 9 (A9). This selected sgRNA is the only sgRNA position holding the necessary “NGG” PAM and targeting the mutation “A” within the ABE editing window. Notably, ABE-mediated editing at position A6 will change a serine codon into alanine (S235A), but this edit will not restore splicing. Initially, we tested different dosages of two ABE enzymes with different base editing windows¹, ABE6.3 and 7.10, in Fah mutant mouse embryonic fibroblast cells (MEF). Consistent with the literature¹, ABE6.3 resulted in the highest A>G base editing efficiency at position A9 (6.9 ± 2.0%,2 µg ABE6.3) (Supplementary Fig. 1) and editing efficiency was dependent of the ABE enzyme concentration. We therefore chose to test ABE6.3 in mice.

Figure 1. ABE rescues liver disease phenotype in a mouse model of tyrosinemia.

(A) Fah^mut/mut mice harbor a G>A mutation (red) at the last nucleotide of exon 8, causing exon skipping. Exon sequences are in upper case. G>A mutation is at position 9 of the sgRNA target. (B) Hydrodynamic injection of ABE and sgRNA plasmids. (C) ABE6.3+sgRNA rescues body weight loss. Withdrawal of NTBC water is defined as Day 0. Error bars indicate s.d. (n=5 mice). In the ABE6.3+sgRNA group, the final body weight is significantly different from the lowest weight (P=0.02, one-tailed student t test). (D) ABE-treated mice show regions of Fah+ hepatocytes (n=3 mice, 32 days off NTBC). Scale bars = 75 μm.

ABE generates Fah⁺ hepatocytes in adult mouse liver

To deliver ABE to the liver of adult mice, we performed hydrodynamic tail vein injection²¹ with ABE6.3 and our validated Fah sgRNA plasmids (named ABE thereafter, Fig. 1b). We removed NTBC-supplemented water 6 days after hydrodynamic injection to initiate HTI symptoms in Fah^mut/mut mice. As shown in Fig. 1c, Fah^mut/mut mice injected with control PBS or ABE6.3 alone rapidly lost 20% of total body weight. In contrast, ABE6.3 and FAH sgRNA rescued weight loss in all five mice (Fig.1c).

To examine whether ABE generates Fah⁺ hepatocytes in vivo, we harvested three of the five mice at day 32 and stained liver sections with a Fah-specific antibody by immunohistochemistry (IHC). As shown in Fig. 1d and Supplementary Fig. 2, ABE generated widespread patches of Fah⁺ hepatocytes in the Fah^mut/mut livers due to the expansion of corrected hepatocytes¹⁴. Concordantly, ABE-treated mice showed improved liver histology compared to the liver damage in control Fah^mut/mut mice off NTBC water (Fig. 1d and Supplementary Fig. 3). Consistent with published literature^{14, 15}, the rapid expansion of Fah⁺ hepatocytes in the Fah^mut/mut liver contributed to the rescue of weight loss in mice treated with ABE¹⁴. These data indicate that ABE rescues the Fah disease phenotype in vivo.

To determine whether ABE successfully corrects the HTI Fah splicing mutation in exon 8, we performed reverse transcription PCR (RT-PCR) in liver mRNA using primers spanning exon 5 and 9. As shown in Fig. 2a, we found that wild-type liver had a 405 bp PCR band containing exon 8 and Fah^mut/mut liver had a 305 bp PCR band corresponding with truncated Fah mRNA lacking exon 8. In ABE-treated Fah^mut/mut mice (post NTBC withdrawal and hepatocyte expansion), we observed both 305 and 405 bp PCR bands, indicating that the exon 8 to exon 9 splicing was restored in a subset of hepatocytes. Sequencing of the 405 bp bands in ABE-treated mice confirmed the presence of the corrected “G” at position A9. (Fig. 2b). These data indicate that in vivo delivery of ABE corrects the G•C to T•A mutation in a subset of liver cells and generates the functional exon 8-containing Fah mRNA.

Figure 2. ABE partially corrects the Fah mutation in mouse liver.

(A) RT-PCR in a representative ABE mouse (4 liver lobes) using primers spanning exon 5 and 9. Wildtype Fah (+/+) amplicon is 405bp and mutant Fah (lacking exon 8) is 305bp. Gapdh, control gene. Images of the uncropped gels are provided in Supplementary Fig. 10. (B) Representative Sanger sequencing of the 405bp RT-PCR bands. (C) Deep sequencing of Fah genomic region in liver DNA. Error bars are s.d. (n=6 liver samples from two mice). (D) Long-term survival of two ABE-treated mice (a and b) without NTBC.

ABE partially corrects Fah mutation in the liver

To quantify ABE edited alleles post NTBC withdrawal, we amplified the Fah genomic region by PCR from liver DNA (using the same mice as in Fig. 1d), and performed amplicon deep sequencing. We observed that the A9 to G correction rate was 9.5±4.0%, (n = 6, Fig. 2c and Supplementary Table 1). Because many hepatocytes are 4N cells²⁰, it’s possible that A9 in one allele was converted into G in the edited 4N hepatocytes.

Because position A6 of the Fah site is within the editing window of ABE (Fig. 1a), we also evaluated the editing efficiency at this position. We detected 1.9±0.9% A6 to G editing in livers (Fig. 2c). Compared to in vitro editing in MEFs (Supplementary Fig. 1), the in vivo detection of relatively lower A6G editing is likely due to re-population of A9G edited healthy cells in the liver rather than to an in vivo preference to A9 editing. The rate of both A6 to G and A9 to G editing (Fig. 2c) is low (~0.1%). Editing position A6 to G will change a serine codon into alanine (S235A) in the FAH enzyme. Because S235 is near the active site of the Fah enzyme²², the A6G/A9G allele may compromise enzyme activity and impede functional rescue of edited hepatocytes.

We also measured the rate of insertions and deletions (indels) by deep sequencing. The detected indel rate was very low (0.05% ABE and 0.03% control) even after expansion of the edited hepatocytes (Fig. 2c and Supplementary Table 2). These data suggest that ABE can directly correct the Fah mutation without causing a high level of indel mutations. Two remaining ABE-treated mice were viable with normal body weight at 106 days without NTBC treatment (Fig. 2d and Supplementary Fig. 4), demonstrating the long-term viability and functionality of ABE treated hepatocytes.

Analysis of the off-target activity of ABE

To globally identify the off-target activity of ABE, we performed genome-wide Guide-Seq²³ in vitro followed by in vivo deep sequencing at selected off-target sites in mouse livers. First, we evaluated off-target activity in vitro using mouse Hepa1–6 cells stably expressing Cas9 and transfected with sgFah and Guide-Seq oligos¹⁶. Using the standard Guide-seq protocol, we detected only the Fah target site and one off-target site (defined as Guide-seq off-target site 1, or GOT1) (Supplementary Data Set 1). These data suggest that the Fah sgRNA used in this study is not associated with a large number of strong off-target sites. Next, we performed targeted deep sequencing in vivo in both control and ABE-treated livers. We did not detect increased A•T to G•C editing at GOT1 in ABE-treated livers compared to controls (n=4 mice, p>0.05) (Supplementary Fig. 5).

To expand our analysis, we sequenced four predicted top-ranking off-target sites in ABE-treated livers (n=6), and did not detect A•T to G•C editing above the background level in a control liver (all rates are <0.1%) (Supplementary Fig. 6). As the overall on-target ABE efficiency in our in vivo study is relatively low, detecting off-target effects is difficult. Future work is needed to further study potential off-target editing of ABE ^{24, 25}.

A codon-optimized RA6.3 improves ABE activity

New reports have indicated that efficiency of base editing can be improved with codon optimization of the Cas9 nickase and inclusion of an N-terminal nuclear localization sequence (NLS) ^{26, 27}. Optimized cytosine BEs and ABE7.10 have been reported ^{26, 27} but ABE6.3 has not been developed. To test if we could improve FAH base editing in our system, we cloned an optimized Cas9 nickase²⁶ into ABE6.3 (Fig. 3a) (named as reassembled ABE6.3 or RA6.3). When transiently transfected in cells, our optimized base editor, RA6.3, had substantially higher expression compared to ABE6.3 (Supplementary Fig. 7). To validate the efficiency of our optimized base editor, we transfected 293T cells with a selected sgRNA sequence with multiple “A” sites ¹ to evaluate A•T to G•C substitution by position (Fig. 3b). Deep sequencing showed an increase in base editing efficiency at all adenine positions (including A9) with use of our codon optimized ABE RA6.3 (Fig. 3b).

Figure 3. Optimizing the coding sequence of ABE6.3 and adding N-terminal NLS sequences improves base editing.

(a) Schematic representation of codon-optimized ABE 6.3. (b) Frequency of A-to-G editing in HEK293T cells 5 days after ABE and sgRNA transfection. Graphs show mean values. Error bars indicate s.d. (n = 3 biologically independent samples). P values determined by one-tailed student t tests.

To further confirm that RA6.3 improves editing efficiency, we compared the A-to-G conversion efficiency of ABE6.3 and RA6.3 at two other genomic sites in HEK293T cells (Fig. 4a, b). Our optimized RA6.3 increased A-to-G conversion rate at all the editable “A” sites within the editing window by an average of 2.1±0.9 fold (site1) and 4.8±1.8 fold (site2). Next, we compared the editing efficiency of HDR (100-nt single-stranded donor DNA) ¹, ABE6.3 and RA6.3, at two genomic sites in HEK293T cells (Fig. 4c, d). Compared with HDR using short homologous arms (site1:1.7±1.1%; site2: 1.8±0.8%), ABE6.3 mediated substantially higher A•T to G•C conversion rates at targeted sites (site1: 30±6.3%; site2: 7.3±1.6%), while RA6.3 further increased base editing efficiency (site1: 61±6.3%; site2: 34±3.2%). These results confirm that the optimized ABE6.3 variant improves editing efficiency in cultured cells.

Figure 4. RA6.3 shows a higher editing efficiency compared to ABE6.3 and Cas9-mediated HDR at two genomic sites in HEK293T cells.

(a,b) Frequency of A-to-G editing at different positions at two genomic sites (sequences as indicated). The “A”s within the editing window are in red. (c,d) Frequency of A-to-G conversion at a targeted A (in red) mediated by ABE6.3, RA6.3 or HDR at two genomic sites (sequences as indicated). Cas9 indicates the group transfected with Cas9 plasmid alone. All error bars indicate s.d. (n = 3 biologically independent samples). P values determined by one-tailed student t tests.

To compare RA6.3 and ABE6.3 in vivo, we measured the Fah editing efficiency following hydrodynamic injection of plasmids (Fig. 5a). IHC staining showed that RA6.3 generated more Fah-positive hepatocytes than ABE6.3 (Fig. 5b–c, p<0.05). This result was confirmed by deep sequencing of the Fah target site (Supplementary Fig. 8a). In addition, RA6.3 increased base editing at a second mouse genomic site in vivo compared to ABE6.3 (Supplementary Fig. 8b). Our combined in vitro and in vivo results indicate that RA6.3 is a robust base editor in mammalian cells and in the mouse liver.

Figure 5. RA6.3 increases editing efficiency in vivo compared to ABE6.3.

(a) Fah^mut/mut mice were injected with indicated plasmid combinations. To measure initial A->G conversion rate, mice were kept on NTBC to prevent expansion of corrected cells. (b) Quantification of Fah-positive hepatocytes by IHC at day 7. Error bars indicate s.d. (n=4 mice per group). *P = 0.0143 by one-tailed Mann-Whitney test. (c) Representative Fah IHC. Scale bars are 25um.

LNP-mediated delivery of ABE mRNA and sgRNA in vivo

Finally, we used lipid nanoparticle (LNP) delivery²⁸ through tail vein injection to deliver our optimized RA6.3 mRNA with chemically-modified sgRNA (sgRNA) to the liver of adult mice (Supplementary Fig. 9a). To measure initial ABE efficiency, we kept the mice on NTBC water after i.v. injection to prevent hepatocyte proliferation. As shown in Supplementary Fig. 9b, Fah^mut/mut mice injected with control PBS by IHC did not show any edited hepatocytes. In contrast, RA6.3 mRNA and FAH sgRNA showed 0.44% ± 0.28% edited hepatocytes (Supplementary Fig. 9c). The data suggest that non-viral delivery of ABE is possible in the adult mouse liver and indicate the need to improve ABE mRNA stability and delivery vehicles.

Discussion

In summary, our results demonstrate that transient delivery of ABE and sgRNA by non-viral hydrodynamic injection is sufficient to generate Fah⁺ hepatocytes and rescue the phenotypic weight loss of Fah mutant mice. We reported the first codon-optimized ABE6.3 and showed that in vivo delivery of ABE was able to correct the Fah mutation in vivo without inducing high-level indels. The relatively low rate of indel formation is one notable advantage of ABE-mediated gene correction compared to Cas9-mediated HDR. While we show phenotypic rescue in our FAH mice, the in vivo ABE efficacy in our proof-of-concept study may not suffice to treat other genetic disorders that require higher levels of correction and that do not selectively proliferate corrected cells. Thus, further improvement of ABE and delivery methods will continue to expand the utility of these powerful tools.

As an initial delivery vehicle, our study also showed that non-viral LNP delivery of ABE mRNA and sgRNA does generate FAH-positive hepatocytes in vivo, albeit with low efficiency. Because ABE mRNA is longer than Cas9 mRNA (5.2 kb vs 4.1 kb), it is possible that the delivery or translation of ABE mRNA is less efficient than Cas9 mRNA. Methods to improve delivery and to enhance mRNA stability/translation will be required to broaden the therapeutic application of ABE. This study demonstrates the potential application of ABE to correct disease genes with G•C to T•A point mutations in adult mammalian models.

METHODS

Construction of CRISPR vectors

sgRNA vector (U6_sgRNA_EFS_GFP, addgene #65656) expressing sgRNA²⁹ was digested with BsmBI. sgRNA Oligos were annealed, phosphorylated by T4 PNK, and ligated with linearized vector. RA6.3 vectors were cloned by Gibson assemble.

Mouse experiments

All animal study protocols were approved by the UMass IACUC. Fah^mut/mut mice¹⁴ were kept on 10mg/L NTBC water. Mice with more than 20% weight loss were humanely euthanized according to guidelines. Vectors for hydrodynamic tail vein injection were prepared using the EndoFreeMaxi Kit (Qiagen). For hydrodynamic liver injection, plasmids suspended in 2ml saline were injected via the tail vein in 5–7 seconds into 8–10 week old Fah^mut/mut mice. The plasmid doses were: Figs. 1–2 30μg ABE plasmid + 60μg sgFah plasmid; Fig. 5 and Fig. S8a: 30μg ABE plasmid (or 30 μg RA6.3) + 30μg sgFah plasmid; Fig. S8b 30μg ABE plasmid (or 30 μg RA6.3) + 15μg sgAbc plasmid. The ABE RA6.3 mRNA was synthesized by TRILINK and end-modified sgRNA was synthesized by Axolabs (Kulmbach, Germany). Lipid nanoparticle formulation and treatment was recently published²⁸. 1 mg/kg LNP RA6.3 mRNA and 0.5 mg/kg LNP Fah sgRNA were injected in 8–10 week old female Fah^mut/mut mice via tail vein injection. The mice were injected 3–4 doses (every 3 days) and kept on NTBC water. Mice were harvested at 5 days post the last injection.

Immunohistochemistry

Mice were sacrificed by carbon dioxide asphyxiation. Livers were fixed with 4% paraformaldehyde (PFA), embedded in paraffin, sectioned at 5 μm and stained with hematoxylin and eosin (H&E) for pathology. Liver sections were de-waxed, rehydrated and stained using standard immunohistochemistry protocols³⁰. The following antibody was used: anti-Fah (Abcam, 1:400). Representative images (of a total of >5 images) are shown.

Gene expression analysis and RT-PCR

RNA was purified using Trizol (Invitrogen) and reverse-transcribed using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Full gel scans are shown in Supplementary Fig. 10.

Illumina sequencing

Fah^mut/mut MEF cells were transformed by retroviral WZL-HRas^V12 and MSCV-shp53³¹ to enhance transfection efficiency. ABE plasmids were electroporated into Fah^mut/mut MEF. Genomic region was PCR amplified. Deep sequencing libraries were made from 1~100 ng of the PCR products. Libraries were normalized to approximately equal molar ratio, and sequenced on Illumina MiSeq machines (150bp, paired-end). Reads were mapped to the PCR amplicons as references using bwa with custom scripts (available in Supplementary Note 1 of ref. 1). Data processing was performed according to standard Illumina sequencing analysis procedures.

Comparison of ABE6.3, RA6.3 and HDR

HEK293T cells were seeded at a confluence of 60%. After 24h, cells were transfected with 1µg Cas9 or base editors, 300ng sgRNA expression plasmid, 6µl Lipofectamine 3000 (Thermo Fisher Scientific), and for HDR assays 0.7µg single-stranded donor DNA template (100nt, PAGE-purified from IDT) ¹. Genomic DNA was harvested 72h after transcription using quickextract buffer (Epicentre). Deep sequencing was performed to measure A to G conversion. Single-stranded 100-mer oligonucleotide donor templates and primers are listed in Table S3 and S4.

Guide-Seq

Deep-sequencing data from the GUIDE-seq experiment was analyzed using the GUIDEseq v1.10.0 Bioconductor Package³² after barcode deconvolution, adaptor removal and alignment to mouse genome (mm10). The default settings for SpCas9 are used except that min.reads is set to 2 and min.peak. score.1strandOnly is set to 2 as in ³³.