Abstract
Custom RNA base editing exploiting the human Adenosine Deaminase Acting on RNA (ADAR) enzyme may enable therapeutic gene editing without DNA damage or use of foreign proteins. ADAR’s adenosine-to-inosine (effectively A-to-G) deamination activity can be targeted to transcripts using an antisense guide RNA (gRNA), but efficacy is challenged by limits of in vivo delivery. Embedding gRNAs into a U7 small nuclear RNA (snRNA) framework greatly enhances RNA editing with endogenous ADAR, and a 750-plex single-cell mutagenesis screen further improved the framework. An optimized scaffold with a stronger synthetic U7 promoter enables 76% RNA editing in vitro from a single DNA construct per cell, and 75% editing in a Hurler syndrome mouse brain after one systemic AAV injection, surpassing circular gRNA approaches. The technology also improves published DMD exon-skipping designs 25-fold in differentiated myoblasts. Our engineered U7 framework represents a universal scaffold for ADAR-based RNA editing and other antisense RNA therapies.
Similar content being viewed by others
Introduction
Editing nucleic acid sequences in vivo holds massive potential for precisely treating genetic diseases and beyond. While CRISPR/Cas9 nucleases and DNA base editors have revolutionized gene and cell therapies, drawbacks remain, such as the potential for DNA damage, permanent off-targets, or germline heritability, as well as delivery challenges and immunogenicity from foreign proteins. In contrast to DNA editing with bacterial enzymes, custom RNA editing using endogenous human ADAR enzymes may eliminate these concerns1,2. ADAR binds double-stranded RNA (dsRNA) structures and deaminates adenosines to inosine, which the cellular translation machinery interprets as guanosine. Natural ADAR deamination plays many important functions: besides editing mRNA coding sequences, it regulates RNA splice sites, modulates RNA interference, and shields endogenous dsRNA including Alu repeat sequences from immune sensors of viral dsRNA3,4. For therapeutic purposes, ADAR can edit additional adenosines using a custom antisense gRNA that anneals to the target RNA sequence to form the dsRNA5.
For decades, custom antisense RNA sequences for therapeutic exon skipping have been substantially enhanced using a U7 small nuclear RNA (snRNA) scaffold6,7,8. The human U7 snRNA (RNU7-1) normally binds histone pre-mRNA through a complementary 20nt Histone Downstream Element (HDE) and directs endonucleocytic cleavage of the histone pre-mRNA; this process also depends on the Sm-protein binding region and the U7 hairpin. By replacing the HDE with a 20−30 nt custom antisense oligo (ASO) sequence, optimizing the Sm-binding domain (SmOPT), and driving expression of the construct by the U7 promoter, exon skipping of a therapeutic target RNA could be substantially increased9,10. Associated studies revealed that the assembled Sm protein complex protects the short antisense RNAs from degradation, and the standard 5′ RNA cap becomes hypermethylated to a 2,2,7-trimethyl guanosine (3mG) cap for nuclear localization11. Importantly, the engineered antisense SmOPT U7 RNAs did not adversely affect natural U7 snRNA histone processing, the broader transcriptome, or display other toxicities12,13. This strategy has now been tested against a wide variety of diseases14,15,16,17,18,19. Notably, a clinical trial skipping a duplicated DMD exon 2 showed a sustained benefit in patients up to 18 months post treatment20,21,22. Since ADAR editing and splicing occur co-transcriptionally in the nucleus4,23, the same advantage might extend to ADAR guide RNAs.
Here we show that the SmOPT U7 hairpin framework dramatically increases the potency of antisense RNAs for ADAR-mediated RNA editing. The greatly increased efficiency of ADAR editing using the SmOPT U7 system removes the need for ADAR enzyme overexpression and allows detectable editing from a single DNA copy of the gRNA construct. This breakthrough enables a pooled library screen exploring the effect of mutations along the original SmOPT and U7 hairpin sequence. This screen sheds light on potential mechanisms by identifying critical residues required for RNA editing, and reveals improved SmOPT U7 hairpin variants that further boost RNA editing. In particular, an improved triple-variant SmOPT U7 hairpin, with an engineered enhanced U7 promoter, results in 76% targeted RNA editing from a single integrated DNA copy of the gRNA construct, and 75% editing in total, unsorted mouse brain from a systemic AAV PHP.eB injection. Furthermore, when these optimized SmOPT U7 hairpin variants and promoters are coupled to previously published antisense constructs for DMD, exon skipping increases up to 25-fold in differentiated myoblasts.
Results
Expression of ADAR gRNAs in a SmOPT U7 format boosts RNA editing
Guide RNAs for ADAR editing have traditionally been expressed by the U6 snRNA promoter, as for short hairpin RNAs and CRISPR gRNAs. We previously reported1,24 that a U6-driven 100 nt antisense RNA possessing an A-C mismatch in the center (denoted 100.50) enables editing of the RAB7A transcript by endogenous ADAR, although editing efficiencies are higher upon ADAR overexpression. When this antisense gRNA was coupled to an Optimal Sm-binding (SmOPT) sequence [AAUUUUUGGAG]9 and a human U7 (hU7) or mouse U7 (mU7) snRNA hairpin25,26, and expressed via the hU7 or mU7 promoters, RNA editing with endogenous ADAR was greatly increased to levels comparable to those with ADAR overexpression (Fig. 1a), thus removing the need for additional enzyme. Excitingly, similar patterns of robust RNA editing were seen for antisense gRNAs targeting FANCC 3’UTR, SMAD4 coding sequence (CDS), and SOD1 translation initiation site (TIS) (Fig. 1a), as well as splice acceptor adenosines for DMD exons 71 and 74 (Supplementary Fig. 1). While the RAB7A, SOD1, and DMD antisense sequences comprised a simple A-C mismatch, the FANCC and SMAD4 antisense sequences also contained mismatch loops at the −5 and +30 positions, previously shown to improve editing specificity1. Both elements of the U7 snRNA framework were necessary – neither the SmOPT U7 hairpin nor the promoter alone increased RNA editing (Fig. 1a). Further testing showed that the SmOPT sequence was required; substituting it with the wild-type U7 or U1 Sm binding domains sharply reduced activity (Supplementary Fig. 2). We confirmed that our longer antisense guide RNAs possessing the SmOPT U7 scaffold localize to the nucleus, similar to previous exon skipping ASOs (Supplementary Fig. 3).
a 100 nt antisense guide RNAs targeting adenosines in the RAB7A 3’UTR, FANCC 3’UTR, SMAD4 coding sequence, or SOD1 start codon (TIS) were expressed using the hU6, hU7, or mU7 promoters, either alone or possessing an additional SmOPT sequence and human or mouse U7 hairpin on the 3’end. Plasmids also contained the CMV promoter to overexpress ADAR2 (gray) or GFP (endogenous ADAR, yellow). RNA was measured from HEK-293T cells 2 days post transfection. Mean values ± SD are shown; n = 3 biological replicates. P values were calculated by Mixed-effects analysis using Dunnett’s multiple comparisons test (two-sided). b Sequences containing two or three copies of the hnRNP A1 binding motif (UAGGGW, underlined) were added onto the 5′ end of antisense guide RNAs expressed using a human U1 promoter and possessing a SmOPT mU7 hairpin. RNA was measured from HEK-293T cells with endogenous ADAR levels 2 days after plasmid transfection. Mean values ± SD are shown; n = 5 biological replicates. P values were calculated by Mixed-effects analysis using Dunnett’s multiple comparisons test (two-sided). c Landing Pad system for testing guide RNAs at a single copy per cell. A Landing Pad construct allowing doxycycline-inducible expression of FusionRed, BxB1 integrase, and a blasticidin-resistance marker was integrated into a single AAVS1 safe harbor genomic locus in HEK-293T cells. When the resulting Landing Pad cell line was transfected with plasmids expressing a guide RNA and selectable marker, B × B1 integrase recombines the attB2 and attP2 sites to incorporate exactly one guide RNA construct into a single site in the genome. d Antisense guide RNAs possessing various 5′ hnRNP A1 domains were constructed using the hU6 promoter with no 3’hairpin or the hU1 promoter with a SmOPT mU7 hairpin as described above (Linear) or containing additional circularizing ribozyme sequences flanking the guide RNA (Circular). Constructs were transfected into HEK-293T Landing Pad cells; RNA from successfully integrated single-copy gRNA constructs was measured 13 days after the initial plasmid transfection. For all panels, negative controls were measured from samples where the gRNA targeted a different gene. Mean values ± SD are shown; n = 3 biological replicates. Source data are provided as a Source Data file.
The hnRNP A1 binding site further enhances editing
Like the U7 promoter, the human U1 promoter is another pol II-type snRNA promoter previously used to express antisense snRNAs for exon skipping27; we confirmed that it also expresses effective guide RNAs for ADAR editing. To further enhance editing, we added recruitment domains for Heterogenous nuclear ribonucleoprotein A1 (hnRNP A1). One of the most abundant nuclear proteins, hnRNP A1 has multiple RNA processing functions, including splicing modulation, transcriptional regulation, and intracellular localization28. Attaching hnRNP A1 binding motifs onto the 5′ end of antisense oligos for exon skipping, opposite the SmOPT and U7 hairpin, enhanced activity29,30. Using the same 5′ motifs, we observed an increase in RNA editing up to two-fold (Fig. 1b), although the effect varied depending on the antisense length and target.
SmOPT-U7 gRNAs enable detectable editing from a single copy per cell
While a transient transfection may deliver hundreds of plasmids to each cell, current levels of AAV gene delivery in vivo are often limited to a few copies per cell. To compare our gRNAs in a stringent, controlled manner, a Landing Pad 293T cell line was created which uses doxycycline-inducible BxB1 integrase to insert exactly one copy of plasmid construct into the AAVS1 safe harbor genomic locus31,32 (Supplementary Fig. 4 and Fig. 1c). Upon transfection, an attB-containing plasmid integrates into the attP site in a directed fashion within the Landing Pad construct. Successful integrations are enriched by antibiotic selection, and RNA editing can be measured in the single-copy cell lines. With the U7 snRNA framework, we observed detectable RNA editing from a single genomic copy of gRNA under endogenous ADAR levels (Fig. 1d). As before, adding a 5′ hnRNP A1 binding motif often further increased RNA editing. In contrast to SmOPT U7 gRNAs, antisense gRNAs expressed by the U6 promoter, or lacking the SmOPT U7 hairpin, showed no editing at this lower dose. Previous publications have shown that circular ADAR-recruiting gRNAs have improved durability and RNA editing activity1,2. When antisense gRNAs are flanked by Twister ribozyme sequences, the ribozymes undergo autocatalytic cleavage and the RNA ligase RtcB inside the cell joins the RNA ends together to form a circle33. When antisense gRNAs with a SmOPT U7 hairpin and 5′ hnRNP A1 binding motif were flanked by Twister ribozymes, circular gRNAs were formed, yet RNA editing was diminished compared to gRNAs with a linear U7 snRNA framework (Fig. 1d).
High throughput split-pool single cell screen of SmOPT-U7 variants
One benefit afforded by detectable RNA editing from a single copy is the ability to perform a high throughput pooled screen on gRNAs. Using a single cell sequencing method, each cell’s RNA editing can be matched to the one gRNA present within that cell, even though the gRNA-target RNA interaction occurs in trans. Previous pooled ADAR gRNA screens have only sequenced in-cis hairpin structures, where the target RNA is physically connected to the gRNA34,35. First, we customized a system to link endogenous mRNA editing to the individual gRNA responsible in a single-cell context. Most methods of single cell RNA sequencing (scRNA-seq) only capture the 5′ or 3′ ends of transcripts and would miss an edited site within a transcript. A recent method of split-pool single-cell barcoding used in situ reverse transcription to detect specific transcripts within a cell, assisted by a Splint scaffold to organize the cell-specific ligated barcodes36. We adapted this method to sequence longer (>100 nt) stretches with sufficient transcript capture efficiency and resolution to measure RNA editing at a specific target site. Protocol improvements included: moving the universal molecular identifier (umi) onto the reverse transcription (RT) primer for better fidelity; an alternative RT enzyme active at hotter temperatures; reducing four rounds of subcode ligation to three, with Illumina indexing supplying an additional level of barcode diversity; and an improved computational analysis pipeline.
Using this method, we performed a mutagenesis screen across the SmOPT and U7 hairpin sequences. Since it was developed in 19939, the SmOPT sequence has been used with a U7 hairpin on antisense short RNAs to target a wide range of diseases. However, while many groups have exchanged the ASO portion, the SmOPT U7 hairpin sequence has remained unchanged6. Mutagenesis screens on natural U7 snRNA have uncovered some required nucleotides but also some flexibility within the Sm-binding site and U7 hairpin37. To discover which residues were essential for RNA editing, and possible improved variants, we generated a library of 252 mutations along the Sm binding or U7 hairpin sequences, comprising: single base substitutions across the SmOPT sequence with a constant mU7 or hU7 hairpin; a constant SmOPT sequence with paired base substitutions across the mU7 or hU7 hairpins; and single base substitutions across the natural U7 Sm or U1 Sm binding domains with a constant mU7 hairpin (Fig. 2a, Supplementary Data 1). A mutated Sm domain (SmMUT) with mU7 hairpin served as a negative control9. This SmOPT U7 library was cloned onto three gRNAs possessing a variety of structural features: 100.50 antisense against the RAB7A 3’UTR with a double hnRNP A1 domain, 100.50 antisense against the SNCA 3’UTR with no 5′ domain, and a 115.80 antisense against the GAPDH CDS possessing symmetrical loops at the −5 and +30 positions, previously shown to improve editing specificity1. All gRNAs were expressed using the mU7 promoter.
a Mutagenesis screen workflow. 1) Three antisense guide RNAs targeting the RAB7A 3’UTR, SNCA 3’UTR, or GAPDH CDS were each cloned onto 252 mutagenesis variants along the Sm binding or U7 hairpin sequences, in duplicate pools. 2) Plasmid libraries were sequenced to match the guide RNA with a 14 N random GuideID barcode located in the GFP 5’UTR. 3) Plasmid libraries were integrated into the Landing Pad 293T cell line. 10 M cells, each containing a single copy of construct, were pooled, fixed, and permeabilized. 4) Four custom primers reverse transcribed GFP, RAB7A, SNCA, or GAPDH transcripts within each fixed cell. Each RT primer possessed an additional 10 N umi and an anchor sequence, which was annealed to a Splint scaffold. 5) Cells were split equally across 60 different wells, each containing a different 7mer subcode that was ligated onto the reverse transcribed cDNA. Cells were then washed, pooled, and split again across 60 new subcode wells. For 60 subcodes, 3 rounds of split-pool ligation yields 216,000 possible combinations. Finally, labeled cells were split into separate wells of 10,000 cells each, lysed, and amplified with dual index primers for Illumina sequencing. b For each of the 750+ guide RNAs, RNA editing was measured for all three transcripts. Editing results comparing the duplicate plasmid libraries (Replicate Pools A versus B) are shown. Guide RNAs targeting RAB7A, SNCA, or GAPDH are shown in red, yellow, or blue, respectively. Positive and negative controls for editing are marked by black symbols. c For each gene target, the effect on RNA editing of single nucleotide substitutions across the SmOPT sequence or paired substitutions across the mU7 hairpin is shown. Underscores indicate locations where extra nucleotides were inserted. In some cases, two nucleotides were inserted. The multicolor dashed line marks the percent editing from the original SmOPT mU7 hairpin construct. Left panels show RNA editing rates from the pooled screen; right panels show RNA editing rates when desired mutations were individually validated at a single copy in the Landing Pad cell line. Source data are provided as a Source Data file.
For the split-pool screen, plasmid libraries were first sequenced to match the gRNA with a 14 N random GuideID barcode located in the GFP 5’UTR; across the six pools, 14,870 total unique GuideIDs were identified. (Supplementary Data 2). After these plasmid libraries were integrated at a single copy into Landing Pad 293T cells, 10 million cells were fixed and permeabilized. Custom primers annealed to and reverse transcribed GFP, RAB7A, SNCA, and GAPDH mRNA within the fixed cell. Each RT primer also included an additional 10 N umi and an anchor sequence, which was then annealed to a Splint scaffold. Fixed, reverse-transcribed cells were subjected to three rounds of Split-pool barcoding, wherein each cell may travel down any one of 216,000 potential combinations of wells, yet every transcript within that cell will be tagged identically. As a final round, 10,000 cells per well were dispensed across a PCR plate, lysed, and the released barcoded cDNA was amplified with dual index primers for Illumina sequencing. For each gRNA, transcripts from all cells possessing the corresponding 14 N GuideID barcodes were aggregated, and the percent of RNA editing was calculated for RAB7A, SNCA, and GAPDH.
Our split-pool screen successfully detected RNA editing across all three endogenous transcripts (Fig. 2b). Furthermore, cell barcodes associated with editing of RAB7A, SNCA and GAPDH were exclusively linked to gRNAs designed to the corresponding target gene. A strong correlation was observed between the two replicate pools. Negative control gRNAs possessing the non-functional SmMUT sequence did not show any editing. Thus, the pooled screening method accurately reports RNA editing enabled by gRNAs, and was used to analyze the effects of SmOPT and hairpin mutations on editing.
Split-pool screen identified improved SmOPT-U7 hairpin variants
Among all three transcripts, any mutation within the first seven nucleotides of the SmOPT sequence eliminated RNA editing, using either the mU7 or hU7 hairpin (Fig. 2c and Supplementary Fig. 5). Neither the wild-type U1 Sm or U7 Sm binding domains showed RNA editing (consistent with the individual transfections in Supplementary Fig. 2), nor any of their mutated variants, except the one which forms the SmOPT sequence. While less pronounced, some mutations along the U7 hairpin sequences also consistently reduced RNA editing across all targets. Overall, the hU7 hairpin produced less RNA editing than the mU7 hairpin, despite HEK-293T being a human cell line.
Many mutations in the SmOPT or U7 hairpin sequence increased RNA editing on all three transcripts. The top 23 mutations were individually cloned as plasmids and single-copy integrated into Landing Pad 293T cells (Fig. 2c). These individual validations largely confirmed the pooled screen results, with most variants outperforming the original SmOPT with mU7 hairpin. Although many mutations which inserted extra nucleotides before the SmOPT sequence increased editing, and were individually reproducible, follow-up experiments showed that these effects depended on the adjoining antisense sequence rather than representing a universal improvement to the SmOPT.
The top five individually validated mutations were then tested in every possible combination on the three target genes (Supplementary Fig. 6). Three variants stood out: SmOPT-11A + mU7-3GG; SmOPT-11A + mU7-12C; and SmOPT-11A + mU7-3GG + mU7-12C. These combinations were examined, via single-copy integration, on an additional set of antisense gRNAs, with or without the 5′ double hnRNP A1 binding motif (Fig. 3 and Supplementary Fig. 7). The FANCC 3’UTR, SMAD4 CDS, and SOD1 TIS were targeted with the same 100.50 antisense sequence in Fig. 1. To confirm robustness of the improved SmOPT U7 hairpin sequences, the RAB7A 3’UTR, SNCA 3’UTR, and GAPDH CDS were now targeted with a shorter 80.40 antisense gRNA. Guide RNAs were expressed using the human U1 snRNA promoter27. Overall, these SmOPT U7 hairpin variants significantly increased gRNA editing. A similar pattern was observed for antisense gRNAs targeting splice acceptor adenosines for DMD exon 71 or 74 skipping (Supplementary Fig. 8).
The original SmOPT mU7 hairpin sequence (white) or three combination variants (SmOPT-11A + mU7-3GG, SmOPT-11A + mU7-12C, and SmOPT-11A + mU7-3GG + mU7-12C) were cloned onto six antisense guide RNAs, with or without a 5′ double hnRNP A1 binding motif, and individually transfected into Landing Pad 293T cells. Following puromycin selection for successfully integrated single-copy guide RNA constructs, RNA was measured 13 days post transfection. Negative controls consist of samples where the transfected plasmid targeted a different gene. Mean values are shown for 2 biological replicates. Results across all antisense guide RNAs (n = 12; 6 gene targets +/− hnRNP) were analyzed in aggregate by two-way ANOVA; P values comparing each combination variant to the original SmOPT mU7 hairpin were calculated using Dunnett’s multiple comparisons test. Source data are provided as a Source Data file.
An improved U7 promoter further increases RNA editing
In addition to improving the SmOPT U7 hairpin structure, we reasoned that the gRNA promoter could be engineered for higher gRNA expression. While pol II-type U7 or U1 promoters could mediate robust RNA editing, the pol III-type U6 promoter could not (Fig. 1). Even though RNA polymerase II also transcribes mRNA, the pol II snRNA genes lack a TATA box or polyadenylation sequence. Instead, full expression and maturation of pol II snRNAs requires Distal and Proximal Sequence Elements (DSE, PSE) within their promoters, and a 3’box signal before the termination sequence that recruits the Integrator complex for post-transcriptional processing10,38. We therefore tested synthetic versions of the mouse U7 snRNA promoter and terminator by duplicating or replacing the DSE, PSE, and 3’box with alternative motif sequences (Fig. 4a)39.
a Arrangement of the synthetic guide RNA expression cassette; locations of the modified DSE, PSE, and 3’box elements within the mouse U7 promoter and terminator are highlighted. b Antisense guide RNAs targeting the RAB7A 3’UTR (100.50) or SNCA 3’UTR (80.40), with or without a 5′ double hnRNP A1 binding motif, were cloned onto the original SmOPT U7 hairpin sequence (white) or the triple variant (SmOPT-11A + mU7-3GG + mU7-12C, purple). Guide RNAs were expressed using either the natural hU1 promoter and mU7 terminator (open bars) or an improved synthetic mU7 cassette (striped bars). Plasmids were individually transfected into Landing Pad 293T cells. RNA editing was measured 13 days post transfection, following puromycin selection for cells containing single-copy integrations of the guide RNA construct. Expression levels of guide RNA and target mRNA were measured using ddPCR and normalized to U1 snRNA or HPRT1 mRNA housekeeping genes, respectively (ratio of copies per µl). Mean values ± SD are shown; n = 3 biological replicates. Results were analyzed in aggregate by three-way ANOVA: for RNA editing, P < 0.0001 for all three main effect variables (promoter, SmOPT U7 hairpin, and guide RNA); for guide RNA quantification, P < 0.0001 for the promoter and guide RNA main effect variables, but not significant for the SmOPT U7 hairpin. For each antisense guide RNA, P values are indicated for pairwise comparisons using one-way ANOVA with Tukey’s test (ns, not significant). Source data are provided as a Source Data file.
The top synthetic mU7 promoter increased RNA editing from a single integrated cassette when compared to the human U1 promoter, our top natural promoter. After a two-day plasmid transient transfection, every combination of promoter and SmOPT mU7 hairpin reached a saturating level of RNA editing (Supplementary Fig. 9). However, under single copy integration, the synthetic mU7 promoter and the triple-variant SmOPT U7 hairpin (SmOPT-11A + mU7-3GG + mU7-12C) each independently significantly increased RNA editing (P < 0.0001 for each main effect variable) (Fig. 4b). Combined, they achieved 76% editing of the RAB7A 3’UTR and 68% editing of the SNCA 3’UTR. The improved promoter significantly increased gRNA expression, and target mRNA expression levels were unchanged.
The modified SmOPT-U7 hairpin outperforms alternative gRNA designs
Several recent publications have described alternative approaches to ADAR gRNA design using gene-encoded vectors and endogenous ADAR enzyme levels. We sought to compare – and possibly combine – our improved promoter and scaffold against these other methods. The CLUSTER approach appends a multivalent series of short antisense binding sequences following a 20.8 target sequence attached to an ADAR recruiting domain40. Two groups have published circular gRNAs formed by flanking Twister ribozyme sequences, termed cadRNA and LEAPER 2.01,2. These circular gRNAs contain a longer antisense domain (100−200 nt) interspersed with uridine deletions, G mismatches, or loops to reduce bystander editing. Fortunately, all of these publications tested the RAB7A 3’UTR using plasmid transient transfection of HEK-293 cells, which allows for a direct comparison.
The top ADAR gRNAs against the RAB7A 3’UTR were chosen from each publication: CLUSTER 3 × RS 19-13-11-20p8 (35% editing); cadRNA 200.100.loops.interspersed.v3 (54% editing); and LEAPER circ-arRNA151 (editing value not listed). We expressed these gRNAs using either the hU6 promoter, as originally reported, or our improved mU7 promoter (Fig. 5). To investigate whether our scaffold could synergize with these other methods, we appended our triple-variant SmOPT U7 hairpin after the antisense gRNA (but within the circularizing ribozymes for circular gRNAs). For plasmid transient transfection, all constructs successfully edited the RAB7A 3’UTR, reproducing the published editing rates and often approaching a saturating level. Both CLUSTER and circular RNA constructs retained activity when expressed by the improved U7 promoter. The LEAPER circ-arRNA151 antisense sequence possesses a stretch of 5 uridines, which triggers premature termination from the U6 promoter; expressing this gRNA from the improved mU7 promoter raised RAB7A editing rates equivalent to the other circular RNAs. Conversely, circularizing the hnRNP 100.50 antisense sequence did not increase editing, with or without the triple-variant SmOPT U7 hairpin.
Antisense guide RNAs targeting the RAB7A 3’UTR (100.50, 80.40, or 60.30) with a 5′ double hnRNP A1 binding motif were compared to three previously published guide RNAs: CLUSTER 3xRS 19-13-11-20p8; cadRNA 200.100.loops.interspersed.v3; and LEAPER circ-arRNA151. Published RAB7A RNA editing values for transient transfection of HEK-293 cells are marked. Guide RNAs were expressed using either the hU6 promoter (open bars) or an improved synthetic mU7 promoter (striped bars). Antisense guide RNAs possessed no SmOPT U7 hairpin (gray), or the triple-variant SmOPT U7 hairpin (purple). In some cases, the guide RNA sequences were flanked by circularizing ribozymes to form circular guide RNAs (as noted). Plasmids were individually transfected into Landing Pad 293T cells. For transient transfection, RNA editing was measured 2 days later. For single-copy genomic integration, RNA editing was measured 13 days post transfection, following puromycin selection for cells containing the guide RNA construct. Mean values ± SD are shown; n = 3 biological replicates. Source data are provided as a Source Data file.
When these same constructs were tested under the more stringent single-copy integration condition, the linear hnRNP SmOPT U7 hairpin constructs retained most of their RNA editing activity (Fig. 5). RNA editing was mostly preserved when the antisense footprint was reduced from 100 nt down to 60 nt, demonstrating a compatibility for shorter antisense sequences that were previously not possible with gene-encoded gRNAs and endogenous ADAR1. In contrast, the CLUSTER and circular gRNAs showed a dramatic decline in RNA editing under single copy levels, showing that these approaches require much higher levels of delivered DNA construct.
The engineered promoter and SmOPT U7 hairpin improve ASOs for exon skipping
We next tested whether our improved mU7 promoter, 5′ hnRNP domain, and triple-variant SmOPT U7 hairpin would also enhance published antisense oligos for exon skipping. Separate from ADAR RNA editing, multiple groups have designed ASOs with the original SmOPT U7 hairpin to cover splicing elements and block the splicing machinery, causing exons to be excluded from the mature mRNA. In Duchenne’s Muscular Dystrophy, an exon skipping strategy can restore translation of the DMD gene by removing duplicated or mutated exons, or shifting a codon reading frame. For DMD exon 2, we tested two published antisense sequences termed A and C, which cover the splice acceptor and splice donor sites, respectively22. The scAAV9.U7-ACCA vector, which expresses two copies each of the A and C short RNAs, has shown efficacy and safety in mice, non-human primates, and human clinical trials (NCT04240314) up to 18 months post treatment13,20,21. We also tested a 75 nt combined antisense sequence (A + C) that covers the entirety of DMD exon 2 (Fig. 6). For DMD exon 51, we tested two published antisense sequences, a 45 nt (long) and a discontinuous 43 nt (split), which anneal within the exon and have also shown long-term efficacy in mice18.
a Antisense Oligo RNA expression constructs alongside a constitutive GFP and puromycin resistance marker were flanked by insulator sequences (ins) and terminal repeats (TR) for the piggybac transposase. b Placement of the ASO binding sites on DMD exon 2 or 51, which were expressed with or without a 5′ double hnRNP A1 binding motif. c ASOs were expressed with the original SmOPT U7 hairpin (white) or the triple combination variant (SmOPT-11A + mU7-3GG + mU7-12C, purple), using either the natural hU1 promoter and mU7 terminator (open bars) or an improved synthetic mU7 cassette (striped bars). Constructs were randomly integrated into the genome of RD rhabdomyosarcoma cells using piggyBac transposase, followed by 7 days of puromycin selection for successful integrants. GFP+ myoblasts were then differentiated for 10 days to induce expression of the full-length DMD Dp427m transcript. Exon skipping was measured using droplet digital PCR. Top: Representative ddPCR plots with probes specific for exon retained or exon skipped mRNA isoforms. Bottom: Percent exon skipping. Empty vector control is a piggyBac integration construct which contains the GFP and puromycin markers but lacks an antisense RNA. Mean values ± SD are shown; n = 3 biological replicates. Results were analyzed in aggregate by three-way ANOVA with P < 0.0001 for all three main effect variables (promoter, SmOPT U7 hairpin, and guide RNA). For each ASO, P values are indicated for pairwise comparisons using one-way ANOVA with Dunnett’s test. Source data are provided as a Source Data file.
We used piggyBac transposase to randomly integrate the ASO SmOPT U7 constructs into the genome of RD human rhabdomyosarcoma cells along with a constitutively expressed GFP and puromycin resistance marker (Fig. 6a). After selection for successfully integrated cells, GFP+ myoblasts were differentiated for 10 days to induce expression of the full-length DMD transcript41. Differentiated myocytes exhibited even GFP expression levels, indicating equal levels of piggyBac transduction (Supplementary Fig. 10). Exon skipping was measured by droplet digital PCR using probes specific for the exon-skipped or exon-retained isoforms (Fig. 6c).
Across all ten ASO variations, the improved synthetic mU7 promoter and triple-variant SmOPT U7 hairpin greatly increased the frequency of exon skipping over the original scaffold (P < 0.0001 for either variable), enabling up to 47% skipping for DMD exon 2 and 43% for DMD exon 51 (Fig. 6c). Adding the hnRNP A1 domain also increased exon skipping for most—although not every—ASO. Combining all three improvements resulted in a 25-fold increase for exon 2 skipping (A + C) and a tenfold increase for exon 51 skipping (long).
The engineered promoter and SmOPT U7 hairpin produce high in vivo RNA editing
Finally, to demonstrate the utility of this RNA editing platform in vivo, we chose the mouse IDUA-W392X model of Hurler syndrome, also used in previous ADAR studies1,2. This mouse model replicates a W402X premature stop codon mutation in human alpha-L-iduronidase (IDUA), an enzyme essential for breaking down glycosaminoglycans. Hurler syndrome is currently treated with a weekly enzyme infusion to reduce GAG accumulation; however, this has limited effect in the central nervous system (CNS) and does not cure the disease. Recent efforts have focused on engineering an IDUA fusion protein capable of crossing the blood-brain barrier and preventing cognitive decline42. RNA editing approaches may be particularly suitable for the CNS, which expresses transcripts that require ADAR editing (GRIA2), and high levels of ADAR2 in addition to ADAR143.
We adapted our Landing Pad system for single-copy integration to include a segment of the mouse Idua gene (Supplementary Fig. 11). The W392X mutation lies near the edge of exon 9, and circular gRNAs designed against the mRNA or pre-mRNA have been shown to work equally2. Thus, we first tested 100.50 antisense gRNAs targeting the mRNA or pre-mRNA. Again, the combination of a 5′ hnRNP domain, the improved synthetic mU7 promoter, and the triple-variant SmOPT U7 hairpin produced the highest RNA editing rates. A second round of gRNA optimization shifting the length and position of the antisense region and introducing mismatch loops further increased editing. A 90.65 antisense Hurler gRNA with loops targeting the mouse Idua W392X pre-mRNA was chosen for in vivo testing.
AAV vectors were cloned containing two copies of the Hurler gRNA or a Scramble negative control gRNA; both vectors included 5′ hnRNP domains, improved mU7 and hU1 synthetic promoters, and triple-variant SmOPT U7 hairpins (Fig. 7a). To target the brain, 2E12 scAAV PHP.eB44 were injected systemically into mIDUA-W392X mutant mice. Additional controls were mutant mice injected with saline only, or wild-type C57BL/6 mice. Four weeks post injection, the anterior cerebrum (largely cortex) was harvested. Mice receiving the Hurler gRNA showed 75% average editing of the W392X stop codon, which significantly rescued IDUA enzyme activity, and reduced GAG accumulation down to wild-type levels (Fig. 7b). Hurler or Scramble AAV constructs generated equivalent levels of gRNA expression. We confirmed previous reports that W392X mice have lower Idua mRNA expression levels compared to wild-type mice45; surprisingly, Idua transcript levels remained low after editing. In the absence of other reports showing restored Idua mRNA levels in this mouse model1,2,46,47,48, we can only posit that other genetic differences beyond nonsense-mediated decay may be responsible, such as a loxP site in the preceding intron, or other SNPs. No exon skipping was observed (Supplementary Fig. 12). ADAR editing with this gRNA design was highly specific, with editing observed only at one bystander adenosine, which results in a synonymous mutation that continues to code for alanine (Fig. 7c). We conclude that the modified SmOPT U7 hairpin framework retains its remarkable potential for enabling high efficiency RNA editing, even when delivered to deep brain regions with systemic injection of AAV.
a AAV vector used for guide RNA delivery. A 90.65 antisense Hurler guide RNA with −6 + 30 loops targeting the mouse Idua W392X pre-mRNA or a 100 nt antisense Scramble negative control were combined with a 5′ hnRNP domain and the triple variant SmOPT U7 hairpin. 2E12 scAAV PHP.eB were retroorbitally injected into mIDUA-W392X mutant mice. In addition, mIDUA-W392X mice injected with saline only or wild-type C57BL/6 mice were included as controls. Brain tissue (anterior cerebrum) was harvested 4 weeks later. b In vivo cerebral cortex samples were analyzed for: RNA editing of the target Idua-W392X mutation; IDUA enzyme activity; Glycosaminoglycan accumulation; Guide RNA expression level relative to the native U1 snRNA; and Idua transcript levels relative to Gapdh mRNA. Mean values ± SD are shown; n = 3 mice. Results were analyzed by one-way ANOVA with multiple pairwise comparisons. c Bystander A > G editing along the region of Idua exon 9 targeted by the AAV PHP.eB Hurler guide. The W392X mutation is highlighted in red. Editing of the subsequent adenosine results in a synonymous GCA > GCG mutation (both alanine). Mean values ± SD are shown; n = 3 mice. Source data are provided as a Source Data file.
Discussion
ADAR-mediated RNA editing is a growing field in gene therapy, since exploiting the pre-existing ADAR machinery eliminates risks from foreign or overexpressed enzymes. Multiple groups have reported editing with endogenous ADAR in vivo1,2,40,49. However, recent clinical trials all deliver gRNAs directly as chemically modified synthesized ASOs50, which require artificial nucleobases and continual redosing, and face challenges with tissue penetrance and biochemical toxicity51. In contrast, an AAV-delivered payload naturally expresses the gene-encoded gRNA within the target cell, and can persist for decades with a single treatment. A rapidly expanding repertoire of engineered AAV serotypes now allows specific delivery to a variety of organs using a non-invasive systemic injection, e.g. PHP.eB for mouse brain, and emerging primate-targeting options44,52,53,54.
One challenge for AAV gene therapies, however, is that current transduction levels may be limited to a few copies of episome per cell. Our breakthrough scaffold achieves high RNA editing from a single DNA copy of construct (Fig. 5), which represents the minimal possible cellular dose of a gene therapy. Importantly, the high potency of RNA editing afforded by this engineered scaffold was observed in vivo as well as in vitro, and represents a substantial advance over previous approaches. For example, optimized circular gRNAs delivered with the same PHP.eB serotype to a Hurler syndrome mouse model reported up to 30% editing in the brain stem, but only 5–10% editing in the cortex2,49. In contrast, our SmOPT U7 gRNAs achieved 75% editing in anterior cerebral cortex (Fig. 7).
Many of the advances reported here were enabled by our single copy integration system, which screens gRNAs under more stringent physiological levels with more uniform expression and kinetics than a plasmid transient transfection. It enabled a pooled gRNA library screen in cells, since editing of the endogenous transcript can be traced back to the cell’s single gRNA. Our assay can currently screen a thousand gRNA constructs in parallel, which we anticipate could be scaled up further. The library mutagenesis screen not only revealed stronger SmOPT U7 hairpin variants, but also identified which residues are responsible for the gain in editing. The Optimized Sm domain, which binds the standard seven-protein Sm ring and leads to accumulation in the nucleus55, is required for activity, especially the first 7 nucleotides (AAUUUUU) (Fig. 2). In contrast, the wild-type U7 Sm domain, which recruits two alternative Lsm proteins to the Sm ring, does not produce RNA editing. Mutating residues along the U7 hairpin—particularly near the front of the hairpin—also reduced RNA editing activity, although the detriment was less pronounced. The mutations enhancing RNA editing activity (mU7-3GG, 10 A, and 12 C), also increased the annealing strength of the hairpin. Notably, the mouse U7 hairpin consistently outperformed the human U7 hairpin, even though the screen was performed in human cells, indicating some sequence-specific hairpin preferences.
We found that the SmOPT U7 scaffold’s advantages for ASOs for exon skipping also apply to antisense gRNAs for ADAR editing (a finding also recently reported in a preprint56). Guide RNAs gain qualitative benefits from being processed through the natural snRNA maturation machinery, which is well-described in the literature9,11,37,38,55. Hypermethylation of the 5′ RNA cap and assembly with the Sm protein ring is critical for shuttling snRNAs into nuclear areas of pre-mRNA splicing and ADAR editing. We show that guide RNAs with our SmOPT U7 scaffold also localize to the nucleus (Supplementary Fig. 3), keeping them in closer association with the target mRNA. Notably, splicing and ADAR editing frequently occur co-transcriptionally, which likely explains why SmOPT U7 antisense RNAs enhance custom RNA editing as well as exon skipping.
Given this finding, how is it possible that our improved gRNA constructs can skip exons when desired (DMD exons 2, 51, 71, and 74), yet avoid exon skipping when unwanted and instead enable editing (RAB7A, SNCA, Idua)? The placement of the gRNA relative to the splicing elements on the target mRNA is likely an important factor. Altered splicing may be more likely in short coding exons versus a long 3’UTR. Future applications of the split-pool assay could allow gRNAs to be simultaneously screened for RNA editing and splicing, since the sequencing primers can be designed to reverse transcribe either outcome on the endogenous target mRNA.
Previous publications used linear antisense guide RNAs 100 nt or longer to recruit endogenous ADAR24,57, although ADAR can edit cis-hairpin dsRNAs down to 20 nt58. With the SmOPT U7 hairpin, antisense sequences < 30 nt can bind to target mRNA for exon skipping22, although longer sequences may be beneficial (Fig. 6 and59). Using our system, antisense sequences down to 60 nt could still mediate robust editing (Fig. 5); these shorter sequences also reduced off-target editing (Supp. Fig. 13). The usefulness of adding a 5′ hnRNP A1 binding domain also depended on the gRNA – while it often enhanced RNA editing, in some cases it reduced editing. The optimal design will likely depend on each target sequence.
Even though our improved U7 scaffold and promoters were developed in HEK-293 cells, we demonstrate that their activity extends to specialized cell types, including muscle and brain. ASOs possessing the original SmOPT U7 hairpin sequence have been effective across a wide range of tissue types6,7,8,14,15,16,17,18,19,22. Likewise, the U7 and U1 promoters drive ubiquitous expression of housekeeping snRNAs25,27, although they can be boosted by tissue-specific enhancers10. A pol II-type snRNA promoter, like U7 or U1, is required for full maturation of the guide RNAs; the pol III-type U6 promoter failed to produce editing activity (Fig. 1). Our synthetic promoters were designed for universal expression (Fig. 4); however, we acknowledge that the optimal combination of scaffold and promoter could vary by cell type. All of these variables may need to be empirically optimized for each target transcript, assessing both RNA editing and splicing outcomes, for clinical application of this technology.
In this study, we employed a simple method of antisense design that uses a single A-C mismatch at the target base, sometimes combined with two mismatched loops up- and downstream of the target base. Even though the Hurler guide RNAs were optimized in HEK-293 cells, which express ADAR1 but not ADAR2, they were active in the mouse brain, which expresses both ADAR1 and ADAR2. We observed no off-target edits that would affect the protein sequence beyond the corrected mutation itself (Fig. 7).
However, this antisense design strategy may be insufficient for many clinical applications, where greater specificity is desired, and relevant adenosines may appear in sequence contexts disfavored by ADAR60. Bystander editing of nearby adenosines can be reduced by including A-G mismatches, uridine deletions, or additional loops in the antisense sequence1,2. Recent progress has been made in the design, high-throughput screening and machine learning-guided generation of ADAR gRNA sequences, wherein the secondary structure of a gRNA: target complex can be customized to focus ADAR activity on almost any desired adenosine and account for the different editing preferences of ADAR1 versus ADAR261,62.
In summary, our engineered U7 hairpin scaffold dramatically increases the potency of antisense RNAs so that they can work at one vector copy per cell – the lowest gene therapy dose possible – for ADAR editing. This builds on decades of success using AAV-delivered antisense SmOPT U7 hairpin RNAs, which have proven to be safe, long-lasting, and not perturb the transcriptome12,13,18,21. Beyond editing, making simple changes to the promoter and hairpin sequences of existing exon-skipping ASOs can improve their efficacy over 10-fold (Fig. 6). Using this expression system, we achieved an unprecedented level of CNS RNA editing in vivo (Fig. 7). Unlike chemical oligonucleotide delivery, AAV delivery of gene-encoded gRNAs has demonstrated very high durability in non-dividing cells like neurons, offering the possibility of one-time cures for serious diseases. Applications exist beyond correcting single point mutations: custom ADAR editing can target stop codons, start codons, splice sites, regulatory elements, and even synthetic biology logic circuits63.
Methods
Ethics statement
This research complies with all relevant ethical regulations. All animal procedures were performed in accordance with protocol S16003 approved by the Institutional Animal Care and Use Committee of the University of California, San Diego.
Cell culture
HEK-293T cells (ATCC CRL-3216) and RD rhabdomyosarcoma cells (ATCC CCL-136) were cultured in DMEM with Glutamax and 10% FCS (Thermo) according to ATCC instructions. Plasmids (1 µg transient or 0.5 µg integration) were transfected using TransIT-293 or TransIT-LT1 reagent (Mirus). After integration, cells were selected for 7 days with puromycin (Sigma). RD cells were differentiated for 10 days on collagen-coated plates using DMEM supplemented with 2% horse serum, 0.1 µM TPA, and 5 µM GSK126 (Sigma).
To generate the Landing Pad cell line, first, an attP1 site was inserted into the AAVS1 genomic locus (PPP1R12C) of HEK-293T cells using a Cas9-expressing plasmid64. Single cell clones were screened by PCR for heterozygous knockins. Then, a plasmid containing the Landing Pad cassette and an attB1 site (Fig. 1c) was incorporated into these cells using a B × B1 integrase-expressing plasmid31,32. Cells containing the Landing Pad cassette were selected using blasticidin (Sigma) and monitored for FusionRed expression. For single copy integration experiments, cells were cultured with 2 µM doxycycline (Sigma) to induce expression of the BxB1 integrase.
Molecular biology
Plasmids were purified by anion-exchange columns (Qiagen, Macherey-Nagel) and quantified on a Qubit fluorometer (Invitrogen). RNA was purified using the Qiagen RNeasy Mini kit with DNase (cell culture) or RNeasy Plus Universal Mini kit (organs). cDNA was reverse transcribed using oligo d(T)23VN, random hexamers, or custom primers using SuperScript IV (Thermo). Control reactions with no reverse transcriptase were included to confirm the absence of genomic or plasmid DNA. Target sequences were PCR amplified using KAPA HiFi HotStart DNA polymerase, labeled with NEB Next dual index primers, and sequenced on Illumina iSeq or MiSeq platforms. Exon skipping and gene expression were quantified using the Bio-Rad QX200 ddPCR Supermix for Probes (no dUTP). Primer (IDT) and gRNA sequences are listed in Supplementary Data 3.
RNA-FISH
HEK-293T cells were seeded in 96-well glass bottom plates at 50,000 cells/well and transfected with 300 ng of plasmid using TransIT-293. After 2 days, cells were washed with 1× DPBS and fixed with 4% formaldehyde in 1 × PBS for 10 min at room temperature. They were then washed with DPBS, permeabilized with 0.1% Triton X-100 in PBS for 10 min, and washed again to prevent over-permeabilization. Cells were equilibrated in a solution of 40% deionized formamide, 0.1% Tween-20, and 2 × SSC for five minutes. A probe mix containing 0.1 µM primary oligo probes (/5Alex647N/ttCAGCTGGATTTCCCAATTCTGA, /5Alex647N/ttCCTGTTTGGATTGCAGAGTGTTAC; IDT) in 40% formamide, 2 × SSC, 0.1% Tween-20, and 10% dextran sulfate was added, and the plate was incubated for 16 h at 42 °C. Post-hybridization, cells underwent four washes of five minutes each in pre-warmed 2 × SSCT at 60 °C, followed by a two-minute wash at room temperature with a buffer containing 30% deionized formamide, 2 × SSC, and 0.1% Tween-20. This was followed by two additional two-minute washes in 2 × SSCT at room temperature. Finally, cells were counterstained with SlowFade Gold Antifade Mountant with DAPI and imaged on a Zeiss Axio Observer with 20x/0.8 objective using Filter Set 112 HE LED.
Single cell sequencing
A library containing 252 mutations across the Sm binding and U7 hairpin sequences was synthesized (Twist Biosciences, Supplementary Data 1) and cloned onto three antisense gRNA plasmids in duplicate pools (NEB Golden Gate Assembly kit), each comprising approximately 3000 bacterial colonies. Plasmid libraries were sequenced to confirm diversity and match the gRNA with a 14 N random barcode before the GFP gene (Supplementary Data 2).
The Split pool barcoding method was adapted from O’Huallachain, et al.36 After formaldehyde fixation and methanol permeabilization, cDNA was reverse transcribed within the fixed cell using phosphorylated custom primers for GFP and the target mRNAs and SuperScript IV (Thermo). To retain cDNA within the cell, amino-allyl dUTP was included in the reverse transcription reaction, and the resulting cDNA was covalently attached to the cell with a DTSSP crosslinker (Thermo). Each RT primer possessed an additional 10 N umi and an anchor sequence on the 5′ end, which was annealed to a Splint scaffold. Cells were subjected to three rounds of Split-pool barcode ligation (NEB T4 DNA ligase). For each round, cells were split equally across 60 different wells, each containing a different 7mer subcode, which was ligated onto the reverse transcribed cDNA inside the fixed cells. Cells were then washed to remove unligated subcodes, pooled, and split again across 60 new wells. Finally, 10,000 labeled cells were aliquoted into each well of a PCR plate and broken apart with proteinase K, RNAse H, and DTT (to reduce the cross-linker) (NEB, Thermo). The released barcoded cDNA was amplified with dual index primers for Illumina sequencing.
Our computational analysis pipeline first deduplicated sequencing reads by umi (filtering out any reads appearing less than thrice) and grouped transcripts according to their ligated series of subcodes CellID. To be included, cells needed at least 5 unique (by umi) transcripts overall, at least 3 unique GFP transcripts, and with at least 77% of the GFP transcripts sharing the same 14 N GuideID. For each gRNA, transcripts from all cells possessing the corresponding GuideIDs were aggregated, and the percent of RNA editing was calculated for each gene target.
AAV vectors
AAV PHP.eB was produced by triple transfection of plasmids encoding the recombinant transfer vector (Supplementary Data 3), pHelper, and PHP.eB RepCap44 into HEK-293T cells using PEI (Polysciences). Virus was harvested, purified using an iodixanol density gradient ultracentrifugation method, concentrated using ultrafiltration spin columns (Cytiva), resuspended in buffered saline, and quantified using the Takara AAVpro Titration kit.
Mouse experiments
Male C57BL/6J and IDUA-W392X (B6.129S-Iduatm1.1Kmke/J) mice45 were purchased from Jackson Labs. Male mice were chosen due to limited availability of homozygous mice. 5−6-week-old mice were injected retro-orbitally with 2.0 × 1012 AAV PHP.eB or a saline-only control; brain tissue was harvested 28 days later. Mice were group-housed in a temperature (18−26 °C) and humidity (40−60%) controlled vivarium with a 12:12 light/dark cycle and food and water ad libitum.
IDUA enzyme assay
The IDUA enzyme assay was performed as previously described65. Mouse tissues were homogenized in PBS with 0.1% Triton X-100 using Qiagen QiaShredder columns and held on ice for 30 min. Protein was quantified using the Qubit Protein BR Assay (Thermo). 25 µl lysate was combined with 25 µl of 360 µM 4-methylumbelliferyl-α-L-iduronidase (Glycosynth) in 0.4 M sodium formate (pH 3.5, Sigma) and incubated at 37 °C for 30 min. The reaction was stopped by adding 200 µl 0.5 M glycine carbonate buffer (pH 10.4, Sigma). Fluorescence was measured at 355 nm excitation and 460 nm emission and compared to a standard curve of 4-methylumbelliferone (Sigma) in glycine carbonate buffer.
GAG assay
The GAG assay was performed as previously described1. Briefly, mouse tissues were homogenized in PBS with 0.05% Tween-20 using Qiagen QiaShredder columns. Samples were digested with 0.1 mg/ml proteinase K (NEB) at 56 °C for 3 h and then 95 °C for 10 min to inactivate the enzyme. Soluble protein was clarified by centrifugation and quantified using the Pierce Quantitative Colorimetric Peptide Assay (Thermo). GAG was measured using the Blyscan GAG assay kit (Biocolor).
Statistics and reproducibility
Experiments were performed with at least n = 3 biological replicates or target genes, consistent with standard practice; no statistical method was used to predetermine sample size. Results were reproduced in at least two independent experiments. No data were excluded from the analyses, other than transfected cell cultures which failed to meet a >95% GFP+ threshold and were excluded before RNA harvest. Mice were randomly allocated into experimental groups. Investigators were not blinded to allocation during experiments and outcome assessment; however, samples were processed in parallel with all sequencing and measurements recorded before analysis. GraphPad Prism version 10 was used for plotting figures and computing associated P values. Geneious Prime version 2023 was used for analyzing single nucleotide variation in NGS amplicon data.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Source data are provided with this paper. NGS raw data files generated in this study have been deposited in the Sequence Read Archive with accession code PRJNA1251206. Sequence and barcode information is listed in the Supplementary Data files. Source data are provided with this paper.
References
Katrekar, D. et al. Efficient in vitro and in vivo RNA editing via recruitment of endogenous ADARs using circular guide RNAs. Nat. Biotechnol. 40, 938–945 (2022).
Yi, Z. et al. Engineered circular ADAR-recruiting RNAs increase the efficiency and fidelity of RNA editing in vitro and in vivo. Nat. Biotechnol. 40, 946–955 (2022).
Mendoza, H. G. & Beal, P. A. Structural and functional effects of inosine modification in mRNA. RNA 30, 512–520 (2024).
Savva, Y. A., Rieder, L. E. & Reenan, R. A. The ADAR protein family. Genome Biol. 13, 252 (2012).
Booth, B. J. et al. RNA editing: expanding the potential of RNA therapeutics. Mol. Ther. 31, 1533–1549 (2023).
Imbert, M., Dias-Florencio, G. & Goyenvalle, A. Viral vector-mediated antisense therapy for genetic diseases. Genes 8, 51 (2017).
Gadgil, A. & Raczyńska, K. D. U7 snRNA: a tool for gene therapy. J. Gene Med. 23, e3321 (2021).
Lesman, D., Rodriguez, Y., Rajakumar, D. & Wein, N. U7 snRNA, a small RNA with a big impact in gene therapy. Hum. Gene Ther. 32, 1317–1329 (2021).
Grimm, C., Stefanovic, B. & Schümperli, D. The low abundance of U7 snRNA is partly determined by its Sm binding site. Embo J. 12, 1229–1238 (1993).
Eckenfelder, A. et al. The cellular processing capacity limits the amounts of chimeric U7 snRNA available for antisense delivery. Mol. Ther. - Nucleic Acids 1, e31 (2012).
Stefanovic, B., Hackl, W., Lührmann, R. & Schümperli, D. Assembly, nuclear import and function of U7 snRNPs studied by microinjection of synthetic U7 RNA into Xenopus oocytes. Nucleic Acids Res. 23, 3141–3151 (1995).
Domenger, C. et al. RNA-seq analysis of an antisense sequence optimized for exon skipping in Duchenne patients reveals no off-target effect. Mol. Ther. Nucleic Acids 10, 277–291 (2017).
Gushchina, L. V. et al. Lack of toxicity in nonhuman primates receiving clinically relevant doses of an AAV9.U7snRNA vector designed to induce DMD exon 2 skipping. Hum. Gene Ther. 32, 882–894 (2021).
Marquis, J. et al. Spinal muscular atrophy: SMN2 pre-mRNA splicing corrected by a U7 snRNA derivative carrying a splicing enhancer sequence. Mol. Ther. 15, 1479–1486 (2007).
Biferi, M. G. et al. A new AAV10-U7-mediated gene therapy prolongs survival and restores function in an ALS mouse model. Mol. Ther. 25, 2038–2052 (2017).
Rashnonejad, A., Amini-Chermahini, G., Taylor, N. K., Wein, N. & Harper, S. Q. Designed U7 snRNAs inhibit DUX4 expression and improve FSHD-associated outcomes in DUX4 overexpressing cells and FSHD patient myotubes. Mol. Ther. Nucleic Acids 23, 476–486 (2020).
Asparuhova, M. B. et al. Inhibition of HIV‐1 multiplication by a modified U7 snRNA inducing Tat and Rev exon skipping. J. Gene Med. 9, 323–334 (2007).
Aupy, P. et al. Long-term efficacy of AAV9-U7snRNA-mediated exon 51 skipping in mdx52 mice. Mol. Ther. - Methods Clin. Dev. 17, 1037–1047 (2020).
Goyenvalle, A. et al. Engineering multiple U7snRNA constructs to induce single and multiexon-skipping for Duchenne muscular dystrophy. Mol. Ther.: J. Am. Soc. Gene Ther. 20, 1212–1221 (2012).
Waldrop, M. A. et al. Sustained dystrophin expression in an infant with a DMD exon 2 duplication following dosing with scAAV9.U7-ACCA. Mol. Ther. 31, 396 (2023).
Gushchina, L. V. et al. Persistence of exon 2 skipping and dystrophin expression at 18 months after U7snRNA-mediated therapy in the Dup2 mouse model. Mol. Ther. Methods Clin. Dev. 31, 101144 (2023).
Wein, N. et al. Translation from a DMD exon 5 IRES results in a functional dystrophin isoform that attenuates dystrophinopathy in humans and mice. Nat. Med. 20, 992–1000 (2014).
Licht, K. et al. A high resolution A-to-I editing map in the mouse identifies editing events controlled by pre-mRNA splicing. Genome Res. 29, 1453–1463 (2019).
Katrekar, D. et al. In vivo RNA editing of point mutations via RNA-guided adenosine deaminases. Nat. Methods 16, 239–242 (2019).
Goyenvalle, A. et al. Rescue of dystrophic muscle through U7 snRNA-mediated exon skipping. Science 306, 1796–1799 (2004).
François, V. et al. Selective silencing of mutated mRNAs in DM1 by using modified hU7-snRNAs. Nat. Struct. Mol. Biol. 18, 85–87 (2011).
Gorman, L., Mercatante, D. R. & Kole, R. Restoration of correct splicing of thalassemic β-globin pre-mRNA by modified U1 snRNAs. J. Biol. Chem. 275, 35914–35919 (2000).
Jean-Philippe, J., Paz, S. & Caputi, M. hnRNP A1: the swiss army knife of gene expression. Int. J. Mol. Sci. 14, 18999–19024 (2013).
Dickson, A., Osman, E. & Lorson, C. L. A negatively acting bifunctional RNA increases survival motor neuron both in vitro and in vivo. Hum. Gene Ther. 19, 1307–1315 (2008).
Goyenvalle, A., Babbs, A., Ommen, G.-J. Bvan, Garcia, L. & Davies, K. E. Enhanced exon-skipping induced by U7 snRNA carrying a splicing silencer sequence: promising tool for DMD therapy. Mol. Ther. 17, 1234–1240 (2009).
Duportet, X. et al. A platform for rapid prototyping of synthetic gene networks in mammalian cells. Nucleic Acids Res. 42, 13440–13451 (2014).
Matreyek, K. A., Stephany, J. J., Chiasson, M. A., Hasle, N. & Fowler, D. M. An improved platform for functional assessment of large protein libraries in mammalian cells. Nucleic Acids Res. 48, e1 (2020).
Litke, J. L. & Jaffrey, S. R. Highly efficient expression of circular RNA aptamers in cells using autocatalytic transcripts. Nat. Biotechnol. 37, 667–675 (2019).
Liu, X. et al. Learning cis-regulatory principles of ADAR-based RNA editing from CRISPR-mediated mutagenesis. Nat. Commun. 12, 2165 (2021).
Park, S. et al. High-throughput mutagenesis reveals unique structural features of human ADAR1. Nat. Commun. 11, 5130 (2020).
O’Huallachain, M. et al. Ultra-high throughput single-cell analysis of proteins and RNAs by split-pool synthesis. Commun. Biol. 3, 213 (2020).
Kolev, N. G. & Steitz, J. A. In vivo assembly of functional U7 snRNP requires RNA backbone flexibility within the Sm-binding site. Nat. Struct. Mol. Biol. 13, 347–353 (2006).
Ezzeddine, N. et al. A subset of drosophila integrator proteins is essential for efficient U7 snRNA and spliceosomal snRNA 3′-end formation. Mol. Cell. Biol. 31, 328–341 (2011).
Faresse, N. J. et al. Genomic study of RNA polymerase II and III SNAPc-bound promoters reveals a gene transcribed by both enzymes and a broad use of common activators. PLoS Genet. 8, e1003028 (2012).
Reautschnig, P. et al. CLUSTER guide RNAs enable precise and efficient RNA editing with endogenous ADAR enzymes in vivo. Nat. Biotechnol. 40, 759–768 (2022).
Marchesi, I. et al. 12‐O‐tetradecanoylphorbol‐13‐acetate and EZH2 inhibition: a novel approach for promoting myogenic differentiation in embryonal rhabdomyosarcoma cells. J. Cell Physiol. 233, 2360–2365 (2018).
Kida, S. et al. Enzyme replacement with transferrin receptor-targeted α-L-iduronidase rescues brain pathology in mucopolysaccharidosis I mice. Mol. Ther. - Methods Clin. Dev. 29, 439–449 (2023).
Tan, M. H. et al. Dynamic landscape and regulation of RNA editing in mammals. Nature 550, 249–254 (2017).
Chan, K. Y. et al. Engineered AAVs for efficient noninvasive gene delivery to the central and peripheral nervous systems. Nat. Neurosci. 20, 1172–1179 (2017).
Wang, D. et al. Characterization of an MPS I-H knock-in mouse that carries a nonsense mutation analogous to the human IDUA-W402X mutation. Mol. Genet. Metab. 99, 62–71 (2010).
Keeling, K. M. et al. Attenuation of nonsense-mediated mRNA decay enhances in vivo nonsense suppression. PLoS ONE 8, e60478 (2013).
Wang, D. et al. Ataluren suppresses a premature termination codon in an MPS I-H mouse. J. Mol. Med. 100, 1223–1235 (2022).
Bose, S. K. et al. In utero adenine base editing corrects multi-organ pathology in a lethal lysosomal storage disease. Nat. Commun. 12, 4291 (2021).
Yi, Z. et al. Utilizing AAV-mediated LEAPER 2.0 for programmable RNA editing in non-human primates and nonsense mutation correction in humanized Hurler syndrome mice. Genome Biol. 24, 243 (2023).
Mullard, A. RNA-editing drugs advance into clinical trials. Nat. Rev. Drug Discov. 23, 323–326 (2024).
Lauffer, M. C., Roon-Mom, Wvan, Aartsma-Rus, A. & Collaborative, N. Possibilities and limitations of antisense oligonucleotide therapies for the treatment of monogenic disorders. Commun. Med. 4, 6 (2024).
Campos, L. J. et al. Advances in AAV technology for delivering genetically encoded cargo to the nonhuman primate nervous system. Curr. Res. Neurobiol. 4, 100086 (2023).
Moyer, T. C. et al. Highly conserved brain vascular receptor ALPL mediates transport of engineered AAV vectors across the blood-brain barrier. Mol. Ther. https://doi.org/10.1016/j.ymthe.2025.04.046 (2025).
Packard, T. A. et al. Engineering AAV capsids with CNS targeted biodistribution from massively diverse libraries using machine learning. Mol. Ther. 30, 130 (2022).
Schümperli, D. & Pillai, R. S. The special Sm core structure of the U7 snRNP: far-reaching significance of a small nuclear ribonucleoprotein. Cell Mol. Life Sci. 61, 2560–2570 (2004).
Smargon, A. A., Pant, D., Glynne, S., Gomberg, T. A. & Yeo, G. W. Small nuclear RNAs enhance protein-free RNA-programmable base conversion on mammalian coding transcripts. bioRxiv 2024.06.12.598766 (2024).
Qu, L. et al. Programmable RNA editing by recruiting endogenous ADAR using engineered RNAs. Nat. Biotechnol. 37, 1059–1069 (2019).
Thomas, J. M. & Beal, P. A. How do ADARs bind RNA? New protein‐RNA structures illuminate substrate recognition by the RNA editing ADARs. Bioessays 39, 1600187 (2017).
Shimo, T., Ueda, O. & Yamamoto, S. Design and evaluation of antisense sequence length for modified mouse U7 small nuclear RNA to induce efficient pre-messenger RNA splicing modulation in vitro. PLoS ONE 19, e0305012 (2024).
Eggington, J. M., Greene, T. & Bass, B. L. Predicting sites of ADAR editing in double-stranded RNA. Nat. Commun. 2, 319 (2011).
Quiroz, J. F. D. et al. Development of a selection assay for small guide RNAs that drive efficient site-directed RNA editing. Nucleic Acids Res. 51, e41 (2023).
Jiang, Y. et al. Generative machine learning of ADAR substrates for precise and efficient RNA editing. bioRxiv 2024.09.27.613923 (2024).
Kaseniit, K. E. et al. Modular, programmable RNA sensing using ADAR editing in living cells. Nat. Biotechnol. 41, 482–487 (2023).
Mali, P. et al. RNA-guided human genome engineering via cas9. Science 339, 823–826 (2013).
Ou, L., Herzog, T. L., Wilmot, C. M. & Whitley, C. B. Standardization of α-L-iduronidase enzyme assay with Michaelis–Menten kinetics. Mol. Genet. Metab. 111, 113–115 (2014).
Acknowledgments
We thank the Shape Therapeutics Vector Production Team for generating AAV; Nicole Enger for assisting the synthetic promoter development; and Dhruva Katrekar and Jacob Tome for technical advice. UCSD work was generously supported by Institutional Funds to P.M.
Author information
Authors and Affiliations
Contributions
S.M. Byrne conceived the project with assistance from A.W.B. and P.M. S.M. Byrne devised gRNAs with the hnRNP and SmOPT U7 domains and adapted the split-pool library screening method. S.M. Burleigh designed and developed the synthetic promoter and AAV constructs. R.F. designed and built the Landing Pad system for single-copy integrated barcoded gRNAs. Y.J. developed the computational analysis pipeline. Y.S. advised on the placement of mismatch loops. S.M. Byrne and R.P. designed and performed experiments to test individual gRNAs. E.K. conducted FISH experiments. J.R. and A.P. conducted mouse experiments. S.M. Byrne, A.W.B. and P.M. wrote the paper with input from other authors.
Corresponding author
Ethics declarations
Competing interests
S.M. Byrne, S.M. Burleigh, R.F., Y.J., Y.S., R.P., E.K., and A.W.B. are current or former employees of Shape Therapeutics, Inc. P.M. is a scientific co-founder of Shape Therapeutics, Navega Therapeutics, Pi Bio, Boundless Biosciences, and Engine Biosciences. S.M. Byrne, S.M. Burleigh, Y.S., Y.J., R.F., and A.W.B. are named inventors on Shape Therapeutics patent applications relating to this work, including: WO 2021/216853, WO 2023/077013, and WO 2024/044282. The remaining authors declare no competing interests. The terms of these arrangements have been reviewed and approved by the University of California, San Diego in accordance with its conflict of interest policies.
Peer review
Peer review information
Nature Communications thanks Victorio Jauregi Matos and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Source data
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Byrne, S.M., Burleigh, S.M., Fragoza, R. et al. An engineered U7 small nuclear RNA scaffold greatly increases ADAR-mediated programmable RNA base editing. Nat Commun 16, 4860 (2025). https://doi.org/10.1038/s41467-025-60155-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-025-60155-z
This article is cited by
-
Enhancing RNA base editing on mammalian transcripts with small nuclear RNAs
Nature Chemical Biology (2025)
