Efficient in vitro and in vivo RNA editing via recruitment of endogenous ADARs using circular guide RNAs

Abstract

Recruiting endogenous adenosine deaminases using exogenous guide RNAs to edit cellular RNAs is a promising therapeutic strategy, but editing efficiency and durability remain low using current guide RNA designs. In this study, we engineered circular ADAR-recruiting guide RNAs (cadRNAs) to enable more efficient programmable adenosine-to-inosine RNA editing without requiring co-delivery of any exogenous proteins. Using these cadRNAs, we observed robust and durable RNA editing across multiple sites and cell lines, in both untranslated and coding regions of RNAs, and high transcriptome-wide specificity. Additionally, we increased transcript-level specificity for the target adenosine by incorporating interspersed loops in the antisense domains, reducing bystander editing. In vivo delivery of cadRNAs via adeno-associated viruses enabled 53% RNA editing of the mPCSK9 transcript in C57BL/6J mice livers and 12% UAG-to-UGG RNA correction of the amber nonsense mutation in the IDUA-W392X mouse model of mucopolysaccharidosis type I-Hurler syndrome. cadRNAs enable efficient programmable RNA editing in vivo with diverse protein modulation and gene therapeutic applications.

Extended Data Fig. 1. Characterization of genetically encoded cadRNAs

(a) RNA editing efficiencies achieved 48 hours and 96 hours post transfection of circular.200.100 and ribozyme.mutant.200.100 plasmids. Ribozyme.mutant.200.100 was created by substituting two key residues in both twister ribozymes (P3 ribozyme: residue 15 G to U and residue 16 U to G; P1 ribozyme: residue 22 A to G and residue 26 C to U) of the construct circular.200.100^,. Values represent mean +/− SEM (n=3; p=0.0021, p=0.0112; unpaired t-test, two-tailed). (b) Schematic representation of various products detected by inward and outward binding primers used for quantification. The outward binding primers selectively amplify the cadRNA. The inward binding primers amplify uncleaved and cleaved-unligated fractions in addition to cadRNA. Values represent mean +/− SEM (n=3). (c) Cells transfected with circular.200.100 and ribozyme.mutant.200.100 plasmids were treated with actinomycin D for 1, 6 and 16 hours starting at 24 hours post transfections. qPCRs were carried out using inward binding primers from panel (b) and expression levels were normalized to untreated samples. (d) Levels of circular.100.50 and linear.100.50 adRNA were measured in the nucleus and cytoplasm. GFP transfected cells were included as controls. U1 snRNA and GAPDH were used to normalize for the nuclear and cytoplasmic compartments respectively. Relative U1 snRNA and GAPDH levels seen in the nuclear vs cytoplasmic fractions were consistent with other published work. Values represent mean +/− SEM (n=3). All experiments were carried out in HEK293FT cells.

Extended Data Fig. 2. Curbing bystander editing of the RAB7A transcript

Histograms of percent A-to-G editing within a 200 bp window around the target adenosine in the RAB7A transcript as quantified by Sanger sequencing. The target adenosine is located at position 0. The dsRNA stretch formed between the antisense and the target are shown below each histogram. Design 1 (cadRNA): Unmodified circular.200.100 antisense, in addition to the A-C mismatch at position 0, two mismatches are seen at positions +66 and +91 that were created to avoid a stretch of poly Us to allow for transcription from a U6 promoter. Design 2 (cadRNA.loops.interspersed.v1): Loops of size 8 bp created at position −5 and +30 relative to the target adenosine and additional 8 bp loops added at 15 bp intervals along the antisense strand. Design 3 (cadRNA.loops.interspersed.v2): As compared to v1, a G-mismatch was positioned opposite a highly edited A (at position +9), an additional 8 bp loop was added at position −81 and the loop at position +49 was changed to a 12 bp loop. Design 4 (cadRNA.loops.interspersed.v3): As compared to v1, the 8 bp loop at +30 was changed to a 12 bp loop starting at position +27, one additional 8 bp loop was added at position −81 and the loop at position +49 was changed to a 12 bp loop. Values represent mean % editing (n=2). All experiments were carried out in HEK293FT cells.

Extended Data Fig. 3. Characterization of IVT synthesized cadRNAs

qPCRs were carried out on cDNA synthesized from IVT-circular.200.100 adRNA and IVT-ribozyme.mutant.200.100 adRNA using primers binding to the ligation stem and ribozyme sequence. n.d.: not detected. Values represent mean +/− SEM (n=3).

Extended Data Fig. 4. In vivo specificity of cadRNAs

2D histograms comparing the transcriptome-wide A-to-G editing yields observed with an AAV delivered construct (y-axis) to the yields observed with the control AAV construct (x-axis). Each histogram represents the same set of reference sites, where read coverage was at least 10 and at least one putative editing event was detected in at least one sample. N_sig is the number of sites with significant changes in editing yield. Points corresponding to such sites are shown with red crosses. The on-target editing efficiency values obtained in the RNA seq are highly inflated due to a large number of reads coming from the cadRNAs mapping onto the target and thus have been omitted from the 2D histograms. The on-target editing values obtained via Sanger sequencing for the four samples analyzed by RNA seq were mCherry-M1: 0%, mCherry-M2: 0%, 2x.circular.200.100-M1: 42.94% and 2x.circular.200.100-M2: 41.32% respectively. M1 and M2 refer to injected mouse 1 and 2.

Extended Data Fig. 5. Transcriptomic changes associated with in vivo cadRNA expression

(a) qPCRs were carried out on IFN-inducible genes involved in sensing of dsRNA 2 weeks and 8 weeks post AAV injections. Values represent mean +/− SEM (n=3; p-values for 2 week long experiment, 2x.circular.200.100 vs mCherry, for genes from left to right p=0.0721, p=0.0353, p=0.8082, p=0.0748, p=0.0303; p-values for 8 week long experiment, 2x.circular.200.100 vs mCherry, for genes from left to right p=0.7276, p=0.6020, p=0.3838, p=0.3491, p=0.2746; unpaired t-test, two-tailed). (b) qPCRs were carried out on ADAR variants 2 weeks and 8 weeks post AAV injections. Values represent mean +/− SEM (n=3; p-values for 2-week long experiment, 2x.circular.200.100 vs. mCherry, for ADAR variants from left to right p=0.3165, p=0.1885, p=0.2815; p-values for 8 week long experiment, 2x.circular.200.100 vs. mCherry, for genes from left to right p=0.8150, p=0.1440, p=0.9532; unpaired t-test, two-tailed). (c) Transcriptome-wide differentially expressed genes in the two groups: 2x.circular.200.100 vs. mCherry are highlighted in red.

Figure 1:. Engineering circular ADAR recruiting guide RNAs (cadRNAs).

(a) A comparison of the RNA editing efficiencies in the 3’ UTR of the RAB7A transcript via various adRNA designs. Values represent mean +/− SEM (n=3; with respect to the linear.100.50, left-to-right, p=0.7289, p=0.0226, p=0.0019, p=0.0055, p=0.0027, and p=0.0006; unpaired t-test, two-tailed). In the schematics, the pink strand represents the antisense domain of the adRNA while the target mRNA is in blue. The bulge indicates the A-C mismatch between the target mRNA and adRNA. The adRNAs are labelled using the following convention: (domain name).(antisense length).(position of A-C mismatch from 5’ end of the antisense). (b) RNA editing efficiencies achieved 48 hours and 96 hours post transfection of various adRNA designs. Values represent mean +/− SEM (n=3; left-to-right, p=0.0019, p=0.0027, p=0.0006 and p=0.8488, p=0.0014, p=0.0077; unpaired t-test, two-tailed). The 48 hour panel data is reproduced from Figure 1a. (c) RT-PCR based confirmation of adRNA circularization in cells. (d) The ability of adRNAs to effect RNA editing of the cluc transcript was assessed in the presence of an siRNA targeting ADAR1. Values represent mean +/− SEM (n=3; left-to-right, p=0.0002, p=0.0216 and p=0.0001; unpaired t-test, two-tailed). All experiments were carried out in HEK293FT cells.

Figure 2:. Transcriptome-wide and target transcript-level specificity profiles of cadRNAs.

(a) (left-panel) 2D histograms comparing the transcriptome-wide A-to-G editing yields observed with a cadRNA construct (y-axis) to the yields observed with the control sample (x-axis). Each histogram represents the same set of reference sites, where read coverage was at least 10 and at least one putative editing event was detected in at least one sample. N_sig is the number of sites with significant changes in editing yield. Points corresponding to such sites are shown with red crosses. The on-target editing values obtained via Sanger sequencing for the samples are HEK293FT: 0%, circular.100.50: 40.47% and circular.200.100: 43.54% respectively. (right-panel) A comparison of the number of off-targets induced by delivery of circular adRNAs, linear adRNAs, and linear adRNAs with co-delivered ADAR2. (b) Engineered cadRNA designs for reducing bystander editing. Design 1 (cadRNA): Unmodified circular.200.100 antisense. Design 2 (cadRNA.bulges): Antisense bulges created by positioning guanosines opposite bystander edited adenosines. Design 3 (cadRNA.loops): Loops of size 8 bp created at position −5 and +30 relative to the target adenosine. Design 4 (cadRNA.loops.interspersed): Loops of size 8 bp created at position −5 and +30 relative to the target adenosine and additional 8 bp loops added at 15 bp intervals all along the antisense strand. Plots depicting the location and extent of all substitutions in the 200 bp dsRNA stretch (n=1 representative plot shown for each construct, analyzed via CRISPResso2). (c) Plots depict % of on-target edited or unedited reads with and without further A-to-G hyperedits in the 200 bp dsRNA stretch formed between the cadRNA and target RNA as observed with the various designs. Substitutions other than A-to-G were not considered for this analysis. Values represent mean % +/− SEM on-target editing in 200 bp long amplicons as quantified by NGS (n=3). (d) Heatmaps of percent editing within a 60 bp window around the target adenosine in the GAPDH and RAB7A transcripts. The positions of adenosines relative to the target adenosine (0) are listed to the left of the heatmap. Values represent mean (n=2). All experiments were carried out in HEK293FT cells.

Figure 3:. In vitro activity of cadRNAs.

(a) Plasmid delivered in situ cadRNA generation: RNA editing efficiencies across various transcripts observed in HEK293FT and K562 cells via plasmid delivered circular.200.100 adRNA, 48 hours post transfections are shown. Values represent mean +/− SEM (n=3). These experiments were carried out using either cadRNA or cadRNA.loops.interspersed from Figure 2b. Associated changes in expression levels of target transcripts as compared to levels seen in untransfected controls is also shown, 48 hours post transfections (p=0.2599, p=0.0135, p=0.1982, p=0.7871, p=0.0144, p=0.2674, p=0.1168, p=0.7852, p=0.5145; unpaired t-test, two-tailed). (b) In vitro transcribed (IVT) circular adRNA generation: Linear forms of twister ribozyme flanked circular adRNAs were transcribed in vitro using a T7 polymerase, purified using LiCl, and transfected into cells, where they circularize in situ by the endogenous RNA ligase RtcB. RNA editing efficiencies across various transcripts observed in HEK293FT and K562 cells via IVT circular adRNA, 24 hours post transfections are shown. Values represent mean +/− SEM (n=3). Associated levels of IVT and plasmid delivered circular.200.100 adRNA targeting RAB7A measured in transfected HEK293FT cells 24 hours post transfections are also shown. Values represent mean +/− SEM (n=3).

Figure 4:. In vivo activity of cadRNAs.

(a) (i) AAV vectors used for adRNA delivery. (ii) Schematic of the in vivo experiment. (b) In vivo RNA editing efficiencies of the mPCSK9 transcript in mice livers via systemic delivery of U6 transcribed linear (U6+27) and genetically encoded circular adRNAs packaged in AAV8. Values represent mean +/− SEM (n=3; p=0.0002; unpaired t-test, two-tailed). (c) Relative expression levels of circular adRNAs. Values represent mean +/− SEM (n=3; p=0.0305; unpaired t-test, two-tailed). (d) mPCSK9 transcript levels relative to GAPDH. Values represent mean +/− SEM (n=3; p=0.6179, p=0.6125, p=0.9323; unpaired t-test, two-tailed). (e) Schematic of the IDUA-W392X mRNA, and RNA editing experiment. (f) In vivo UAG-to-UGG RNA editing efficiencies of the IDUA transcript in mice livers via systemic delivery of genetically encoded circular adRNAs packaged in AAV8. Values represent mean +/− SEM (n=3). (g) IDUA transcript levels relative to GAPDH. Values represent mean +/− SEM (n=3; p=0.1185, p=0.3815, p=0.0042; unpaired t-test, two-tailed). (h) GAG content in mice livers of AAV8-scrambled.2x.circular.200.100 and AAV8-IDUA.2x.circular.200.100 injected IDUA-W392X mice. Wild type C57BL/6J mice were included as controls. Values represent mean +/− SEM (n=3; p=0.0285; unpaired t-test, two-tailed).