Design and in vitro validation of siABE
The inherited DNA and RNA off-target activities of BEs are of concern for clinic applications. As the deaminase domains used in the BEs largely contribute to the observed off-target activities24–26, we hypothesize that providing a negative-feedback gRNA targeting the deaminase domain used in BE would help to minimize such off-target activities. A previous study showed that the cysteine residue at position 90 (C90) residing near the enzymatic pocket of TadA* (Fig. 1A) appeared to be critical for its enzymatic function27. We first engineered ABE8e_V106W28 with the C90R mutation and confirmed that the C90R mutation completely abolished its editing activity (Fig. 1B). We then designed a gRNA to install the C90R mutation in ABE8e_V106W. At three days after co-transfection of the C90-gRNA and ABE8e_V106W plasmids into Neuro-2a cells, the C90R editing efficiency was estimated to be around 67% by BEAT analysis of the ABE8e_V106W transcripts (Fig. 1C).
We previously developed a Gp41-1 intein split strategy to enable highly efficient assembly of full-length iABE-NGA and AAV packaging for in vivo base editing5. To test if incorporation of the C90-gRNA into the intein split would allow simultaneous editing of the target gene while disabling the ABE enzyme itself, we inserted the C90-gRNA expressing cassette into the N-terminal half of intein split plasmids along with a targeting gRNA for the gene of interest (eg. the mdx4cv-gRNA) (Fig. 1D). To test the performance of this two-component siABE system, we co-transfected the HEK293 cells with the N-terminal, C-terminal and the BEON fluorescent reporter construct29, which carries the mdx4cv targeting sequence disrupting the downstream EGFP expression. The GFP fluorescence can be rescued by successful editing of ABE and corresponding target gRNA. As shown in Fig. 1E, transfection with the reporter alone resulted in minimal background fluorescence, while the cells co-transfected with all three plasmids, either with or without C90R gRNA, showed robust GFP fluorescence. FACS analysis showed that EGFP expression was restored to similar levels in the cells transfected with or without C90-gRNA (Fig. 1F). The editing rate was analyzed by BEAT analysis of the Sanger sequencing traces of the transcript amplicons. We found that the C90-gRNA did not significantly compromise the editing rate of the mdx4cv target sequence as compared with the positive control group without the C90-gRNA (Fig. 1G). However, there was a time-dependent increase in the C90R editing efficiency in the siABE-N/C transfected group (Fig. 1H), indicating that the negative feedback editing increasingly disables the ABE enzyme. We further tested the performance of siABE-N/C to edit an endogenous gene Apoc3, which encodes an inhibitor for lipoprotein lipase, in Neuro-2a (N2a) cells. Again, we observed no significant difference between ABE-N/C and siABE-N/C groups (Fig. 1I), suggesting that the C90-gRNA offers a simple approach to restrict the ABE activity without substantially compromising the editing of the target gene.
Design and in vitro validation of siCBE
Next, we took a similar approach to design a self-inactivating CBE (siCBE) system. We reasoned that CBE-mediated installation of a premature stop codon within the cytosine deaminase domain would switch off the CBE, thus providing a negative feedback loop for regulating the CBE activity. We designed a gRNA (Q125-gRNA) to install a premature stop codon at Q125 within the AncBE4 enzyme (Fig. 2A). Transfection of AncBE4 with the Q125-gRNA led to an average Q125X editing efficiency of 57.9% ± 5.5% (Fig. 2B). The installation of Q125X is predicted to gradually turn off the expression of AncBE4. To test this, we harvested the cells at 24, 48, 72 and 96 hours after transfection and analyzed the time-course expression of AncBE4 using the anti-Cas9 antibodies. The expression levels of both N-terminal and full-length AncBE4 were not significantly different between CBE and siCBE groups when examined at 24 hours (Fig. 2C, D). However, both the full-length and N-terminal AncBE4 was decreased more rapidly in cells transfected with siCBE starting from 48 hours when compared with the CBE group (Fig. 2C, D). Consistently, there was a time-dependent increase in the Q125X editing in the AncBE4 transcript (Fig. 2E). To test whether the addition of the Q125-gRNA in siCBE impairs the editing activity at an endogenous locus, we compared the editing efficiency at HEK4 site between the CBE and siCBE in HEK293 cells. The siCBE showed only a slight, but statistically insignificant, decrease in the editing efficiency as compared with CBE (Fig. 2F).
As different cytosine deaminase domains have been utilized in engineered CBEs, we tested a similar approach to develop siCBE-evoFERNY that carries a smaller evolved deaminase evoFERNY30. A W65-gRNA was designed in order to introduce a premature codon into tryptophan at 65 (W65) within evoFERNY. As described above for siCBE, we observed a time-dependent increase of editing efficiency at W65 of evoFERNY in siCBE-evoFERNY transfected cells (Fig. 3A). Western blot showed that the full-length CBE-evoFERNY expression decreased more rapidly in the siCBE-evoFERNY group than the canonical CBE-evoFERNY (Fig. 3B). To test if introduction of the self-targeting W65-gRNA may affect the performance of the target gRNA for the gene of interest, we inserted a gRNA targeting mouse Asgr1 to install a loss-of-function (LoF) mutation at the codon W158. It was recently shown that the LoF variants of human ASGR1 are associated with lower levels of non-HDL cholesterol31. The siCBE-evoFERNY achieved an average editing rate of 16.9% ± 1.0%, a small but significant decrease when compared with canonical CBE-evoFERNY (23.1% ± 2.0%, p = 0.003) (Fig. 3C).
Next, we attempted to further design the C-terminal half construct by adding the self-inactivating gRNA target sequence into the coding region of siCBE-C (siCBE-v2) in order to achieve simultaneous inactivation of both N- and C-terminal fragments (Supplementary Figure S1A). After transfection of the siCBE-v2 into HEK293 cells, we performed Western blot analysis at different time points to examine the inactivation of N-, C- and full-length CBE expression. As shown in Supplementary Figure S1B, we found that the C-terminal half and the full-length CBE were efficiently switched off. But surprisingly, we observed that inactivation of N-terminal half was not as efficient as the siCBE. We thus chose siCBE for further in vivo studies.
Design and in vitro studies of gRNA targeting Angptl3
Genome-wide association studies (GWAS) have linked a number of LoF variants in several genes to favorable lipid profile and reduced cardiovascular diseases, such as PCSK9, ANGPTL323, APOC324 and ASGR126, making them attractive therapeutic targets for hypercholesterolemia. To assess the in vivo performance of siBE, we chose ANGPTL3 as a therapeutic target with the goal to develop a “one shot, one cure” treatment for hypercholesterolemia as a monoclonal antibody against ANGPTL3 has recently been approved by FDA. We first designed several gRNAs to install either a premature stop codon or disrupt the canonical splicing donor (SD, e.g. GT dinucleotide) or acceptor (SA, e.g. AG dinucleotide) sequences into mouse Angptl3 using CBE or ABE. To install the Q135X premature stop codon (Q135X) into the coding sequence of mouse Angptl3, two overlapping gRNAs (gRNA1 and gRNA2) with different PAM sequences (GGG or GGC) were constructed (Fig. 4A). Transfection of N2a cells with the gRNAs and their corresponding CBEs (e.g. AncBE4 for gRNA1 and CBEmax-SpG for gRNA2) induced efficient installation of Q135X mutation in Angptl3 (Fig. 4B). The gRNA1 and AncBE4 combination induced 51.5 ± 8.0% conversion of C to T at the Q135 codon, while the efficiency of gRNA2 and CBEmax-SpG was slightly lower (40.0 ± 0.8%). We also designed a gRNA (E4-gRNA) to mutate the exon 4 SD site of mouse Angptl3 in combination with either CBEmax-SpG (IVS4 + 1G > A) or ABE8eV106W-SpG (IVS4 + 2T > C) (Fig. 4A). While CBEmax-SpG showed little editing efficiency to induce IVS4 + 1G > A conversion, ABE8eV106W-SpG induced 19.0 ± 3.9% editing to install IVS4 + 2T > C mutation at this SD site (Fig. 4C). Based on these data, we chose the gRNA1, which showed the highest editing efficacy, for the downstream experiments. We inserted two copies of the gRNA1 into the siCBE-N construct and compared the editing efficiency of siCBE with that of the canonical CBE for editing Angptl3. There was no significant different between siCBE and CBE groups (Fig. 4D).
In vivo performance of AAV9-siCBE/Angptl3 in mice
To test the performance of siCBE in vivo, we chose AAV9 as a delivery method and Angptl3 as the target gene. We packaged the two Gp41-1 intein split halves of the siCBE/Angptl3 into AAV9 (Fig. 5A) and tested if in vivo delivery of AAV9-siCBE could reduce the lipid profile in mice. The N-terminal vector carries two copies of Angptl3 gRNA1 under the control of human U6 promoter and AncBE4 split at 513 fused with the C-terminus of Gp41-1 intein driven by a liver-specific promoter human alpha-1-antitrypsin [hAAT]32–34. The C-terminal vector carries the self-targeting Q125X-gRNA and the N-terminal Gp41-1 intein fused with the second half of AncBE4 split driven by hAAT promoter. Both vectors are within the packaging capacity of AAV and packaged efficiently.
We treated six wild-type (WT, C57BL/6J) mice with AAV9-siCBE/Angptl3 (a total of 2 x 1014 vg/kg, 1:1 of the N and C-terminal half) through the tail vein injection at 6 weeks of age. Two additional mice receiving either the N-terminal half or the C-terminal half alone were used as negative control. We measured serum levels of Angptl3, triglyceride (TG) and total cholesterol (TC) at different time points after AAV9-siCBE/Angptl3 administration. Compared with the control mice, all AAV9-treated animals exhibited dramatically decreased ANGPTL3 protein levels at one week after the treatment (after: 15.4 ± 6.3 ng/ml vs control: 100.7 ± 24.8 ng/ml), which remained low at 4 and 8 weeks after treatment (Fig. 5B). Similarly, AAV9-siCBE/Angptl3 treatment dramatically reduced the TG levels in all animals at 1–8 weeks after treatment (Fig. 5C). The TC levels were also significantly reduced by AAV9-siCBE/Angptl3 treatment (Fig. 5D).
We quantified the gene editing efficiency in the liver samples from the treated mice at 8 weeks after treatment. Sequencing of the Angptl3 transcript amplicons from the treated animals showed an average of 56.9% ± 18.0% C to T editing at the codon Q135 (Fig. 5E). Sequencing of the genomic DNA from the liver samples showed a similar editing efficiency (57.8% ± 3.8%) at the target site (Fig. 5F). We also amplified the AncBE4 transcript from the treated animals and sequencing showed an average of 77.7% ± 4.0% editing while the mouse treated with the N-terminal half alone showed undetectable editing as expected. To further examine whether the editing of AncBE4 disrupted the expression of AncBE4, we performed Western blot analysis of the liver tissues at 8 weeks after AAV9 treatment with the anti-Cas9 antibody, which recognizes the N terminus of Cas9. As shown in Fig. 5H, both the N-terminal and full-length AncBE4 were not detected in mice treated with AAV9-siCBE/Angptl3, whereas the N-terminus of AncBE4 was readily detectable in the mouse treated with the N-terminal half alone. Taken together, these results suggest that the AAV9-delivered siCBE/Angptl3 induced highly efficient editing of Angptl3 and its own transcript.
A major advantage of the siBE system versus the conventional BE is that it would eliminate the deaminase domain-mediated nonspecific RNA editing as it efficiently switched off the expression of deaminase domain-containing fragment (see Fig. 5). We reasoned that the reduced expression duration of the functional base editor in the siBE system may also reduce the off-target DNA editing activity. To test this, we chose several previously validated off-target sites conferred by the HEK4 gRNA30. CBE or siCBE plus HEK4 gRNA were co-transfected into HEK293 cells and the off-target sites were amplified by PCR. As shown in Fig. 6A, while there was no significant difference for on-target editing of HEK4 site between the CBE and siCBE groups, siCBE exhibited significantly lower C5 editing at the off-target site 1 (OT1) (16.4 ± 0.7%, p < 0.001) compared to CBE (22.7 ± 1.3%). At OT2, both siCBE and CBE showed background levels of editing activities. Similarly, we analyzed the HEK4-OT4 site and found siCBE generated slightly lower, insignificant editing rate at C6 compared to CBE group.
The siBE system requires both a therapeutic gRNA and a self-targeting gRNA, which may lead to additional off-target activities in the genome20, 22, 35. To test the off-target activities of both Angptl3-Q135X gRNA and the self-targeting Q125X gRNA in mouse genome, we examined the top putative off-target sites predicted by the Cas-OFFinder36. We transfected N2a cells with CBE or siCBE constructs, amplified the off-target sites by PCR and subjected the amplicons to Sanger sequencing. None of the top predicted off-target sites of Angptl3-Q135X gRNA and the self-targeting Q125X gRNA showed detectable off-target activities (Fig. 6B&C). Similarly, we also measured the off-target activities in mouse livers following AAV9-siCBE/Angptl3 treatment. We observed undetectable off-target activities at four of the five off-target sites in the treated mice (Fig. 6D&E). These results suggest that the siCBE conferred highly specific editing in in vitro and in vivo settings.