Main

Drugs and antisense oligonucleotides have been identified that unsilence paternal UBE3A by downregulating UBE3A-ATS3,4. However, these potential therapeutics have a short half-life and require repeated invasive injections, which is not desirable for a paediatric-onset disorder that lasts a lifetime. CRISPR–Cas9 has the potential to permanently reduce the expression of non-coding RNAs via mutagenesis5. To determine whether Cas9 could be used to reduce Ube3a-ATS and unsilence paternal UBE3A, we designed 260 Staphylococcus pyogenes Cas9 (SpCas9) guide RNAs (gRNAs) that target putative regulatory regions and genes in or near Ube3a-ATS (Fig. 1a, Supplementary Table 1). Each gRNA was cloned into an expression plasmid containing SpCas9 and was transiently transfected into cultured cortical neurons from Ube3am+/pYFP mice, which harbour a paternal Ube3a-YFP knock-in allele6. Several gRNAs unsilenced paternal UBE3A–YFP as effectively as topotecan, the positive control3, while gRNAs targeting Ube3a decreased paternal UBE3A–YFP below its low baseline level (Fig. 1b, c, Supplementary Table 1). gRNAs that were most effective at unsilencing paternal UBE3A–YFP were located in or near the Snord115 and Snord116 genes (also known as H/MBII-52 and H/MBII-85, respectively), which are two clusters of C/D box small nucleolar RNAs that are processed from introns of Ube3a-ATS7. Since deletion of SNORD116 genes causes Prader–Willi syndrome8,9,10, we did not pursue gRNAs that targeted Snord116 genes or upstream regions of Ube3a-ATS.

Fig. 1: Screen to identify SpCas9 gRNAs that unsilence paternal Ube3a.
figure 1

a, The paternal Ube3a-ATS locus showing the locations of genes, the transcription collision region (dashed lines) and the gRNA library target sites. b, The percentage of Ube3am+/pYFP neurons co-stained for UBE3A–YFP and tdTomato, co-transfected with Camk2a promoter:tdTomato, SpCas9 and one gRNA (circles) or treated with 300 nM topotecan for 72 h. Black circles indicate P < 0.005 and fold change > 4. c, Ube3am+/pYFP neurons co-transfected with Camk2a promoter:tdTomato, SpCas9 and negative control gRNA or Spjw33 gRNA. Nuclei are shown in blue (4′,6-diamidino-2-phenylindole (DAPI)). Scale bar, 100 μm. d, Expression of the indicated genes 7 days after transducing Ube3am+/pYFP cortical neurons with lentiviral SpCas9 + Spjw33, relative to SpCas9 + negative control gRNA (dashed line). qPCR, normalized to Eif4a2; n = 3. All error bars show the s.e.m. unless otherwise noted. *P < 0.01.

Snord115 genes are located 3′ of the Prader–Willi syndrome critical region, in the distal portion of Ube3a-ATS that is expressed only in neurons11. Snord115 genes are highly similar, and in some cases identical, at the sequence level. One of the most effective gRNAs (Spjw33) is predicted to target 76 sites in the Snord115 cluster (Extended Data Fig. 1a, b, Supplementary Table 1), and was far more effective than gRNA pairs that were predicted to delete defined regions of Ube3a-ATS (Extended Data Fig. 1c).

To evaluate how targeting multiple Snord115 genes affected expression across the Ube3a-ATS locus, we transduced Ube3am+/pYFP cortical neurons with a lentivirus carrying SpCas9 and Spjw33 and quantified gene expression via quantitative PCR with reverse transcription (RT–qPCR). We found that expression of the paternal Ube3a-YFP allele was increased in neurons transduced with Spjw33 (Extended Data Fig. 1d–f). Furthermore, Spjw33 reduced the expression of Snord115 as well as regions of Ube3a-ATS that are immediately upstream (Ipw) and downstream (Ube3a-ATS 3′) of the Spjw33 target sites (Fig. 1d). This contrasted with topotecan, which decreased the expression of all transcripts derived from Ube3a-ATS (Extended Data Fig. 1g, h). A gRNA targeting human SNORD115 (negative control) had no effect in mouse neurons (Extended Data Fig. 1i). One or more base mismatches in the gRNA eliminated unsilencing activity (Extended Data Fig. 1j). We detected indel mutations at a low frequency at the Spjw33 target site (Extended Data Fig. 1k, Supplementary Table 2), but not at imperfectly matched (by 1–4 base pairs) sites (Extended Data Fig. 1k, Supplementary Table 2). These low-frequency indels are unlikely to disrupt the Ube3a-ATS non-coding RNA.

Since double-strand breaks are resolved in around 24 h, the likelihood that a break will be present at any point in time and physically impede transcription is expected to increase as a function of the number of target sites. Thus, we tested an additional 66 gRNAs that target different numbers of repetitive sites in Ube3a-ATS (Supplementary Table 1). We observed a correlation between the number of genomic target sites and the unsilencing efficiency of paternal Ube3a (Extended Data Fig. 1l, Supplementary Table 1). These data suggest that transcription through Ube3a-ATS is terminated at transient double-strand breaks in these short-term (7 day) cell culture experiments.

We next evaluated the extent to which Cas9-directed targeting of SNORD115 can unsilence paternal UBE3A in primary human neural progenitor-derived (phNPC) neurons12. We used a single-nucleotide variant (SNV) in UBE3A exon 5 to quantify expression of the presumed maternal and paternal UBE3A alleles (Extended Data Fig. 2a, b). As expected, the presumed maternal UBE3A allele was predominantly expressed in differentiated neurons, coincident with the emergence of paternal UBE3A-ATS expression (Extended Data Fig. 2c). We then transduced phNPC neurons with lentivirus carrying SpCas9:mCherry and adeno-associated virus type 2 (AAV2) carrying eGFP driven by the human SYN1 (hSYN1; encoding synapsin 1) neuron-specific promoter, and evaluated three SpCas9 gRNAs that target a different number of sites in the SNORD115 cluster (Extended Data Fig. 2d, e). To quantify gene expression in different cell types, we FACS-isolated (1) hSYN1–eGFP progenitors, (2) SpCas9/hSYN1–eGFP+ neurons (internal negative control), and (3) SpCas9+/hSYN1–eGFP+ neurons (Extended Data Fig. 2e–h). We confirmed that SpCas9 transcripts were expressed in SpCas9+-sorted cells, and that genes indicative of neuronal differentiation, specifically RBFOX3, SNORD115 and the UBE3A-ATS 3′ region, were expressed in hSYN1–eGFP+-sorted neurons (Extended Data Fig. 2g–n). We used SNV-specific RT–qPCR probes to quantify allelic expression of UBE3A. As expected, UBE3A was expressed biallelically in progenitors and, in SpCas9 neurons, the maternal UBE3A allele was expressed ninefold higher than the paternal allele (Extended Data Fig. 2o, p). In SpCas9+/hsa#3+ neurons, the expression of the paternal UBE3A allele was increased to near maternal UBE3A levels (relative to neurons transduced with the negative control gRNA; Extended Data Fig. 2o, p). Furthermore, there was a correlation between the number of gRNA target sites and the magnitude of paternal UBE3A unsilencing (Extended Data Fig. 2p).

Total UBE3A levels increased in human neurons transduced with SpCas9 and hsa#3, consistent with increased expression of the paternal allele, while the expression of SNORD115 and UBE3A-ATS 3′ was significantly decreased (Extended Data Fig. 2q). IPW levels did not change in human neurons, which contrasted with results in mouse neurons (Fig. 1d), possibly because hsa#3 target sites are about 50 kb 3′ of human IPW, whereas Spjw33 target sites are immediately downstream of mouse Ipw (compare Extended Data Figs. 1a and 2d).

The restrictive packaging limit of AAV prevented us from incorporating SpCas9 and a gRNA into a single AAV vector. Staphylococcus aureus Cas9 (SaCas9) is 1 kb smaller and can be incorporated into a single AAV vector with a gRNA13,14,15. The protospacer adjacent motif sites for SpCas9 and SaCas9 differ, so we designed and tested several SaCas9 gRNAs, settling on one (Sajw33) that was nearly identical to Spjw33 at the sequence level and that effectively unsilenced paternal Ube3a in cultured neurons (Extended Data Fig. 3a, b).

We packaged hSYN1 promoter-SaCas9 and U6 promoter-Sajw33 into AAV9 (Extended Data Fig. 3c), and injected this vector intracerebroventricularly (i.c.v.; bilaterally) into embryonic day 15.5 (E15.5) or postnatal day 1 (P1) Ube3am+/pYFP mice. When examined histologically at P30, animals injected at E15.5 showed biased unsilencing of paternal UBE3A–YFP in lower-layer cortical neurons, while animals injected at P1 showed biased unsilencing in upper-layer cortical neurons (Extended Data Fig. 3d). We therefore opted to use a dual E15.5 + P1 injection strategy to maximize the number of transduced neurons. Following dual injections in AS mice (Ube3am−/p+), paternal UBE3A was unsilenced throughout the P30 and P90 brain, including all layers of the cerebral cortex, all regions of the hippocampus, some cerebellar Purkinje neurons, and the spinal cord (Fig. 2a, b, Extended Data Fig. 4a–g). On the basis of western blot analyses, paternal UBE3A protein was restored to about 37%, 38% and 40% of wild-type levels in the cerebral cortex, hippocampus and spinal cord, respectively, of AS model mice at P90 (Fig. 2c, d), but showed no significant change in the cerebellum. Moreover, paternal UBE3A was unsilenced in about 58% of all NEUN+ cortical neurons at P90, with median intensity levels of 63% of maternal UBE3A (based on staining intensity in >9,000 cortical neurons), and was primarily localized to the nucleus, indicative of proper isoform expression16. We did not observe evidence of neuroinflammation or loss of hippocampal neural progenitors17 (Extended Data Fig. 4h–o). We also did not detect changes in expression or splicing patterns of known Snord115 target genes in AS mice treated with Sajw33 (Extended Data Fig. 5), suggesting that transcriptional downregulation and/or mutagenesis of some Snord115 genes does not compromise putative Snord115 functions18,19. Remarkably, paternal UBE3A was unsilenced 17 months after a single E15.5 viral injection into AS mice (Extended Data Fig. 6a–c), revealing an extremely long duration of action.

Fig. 2: AAV delivery of SaCas9 and Sajw33 unsilences paternal Ube3a throughout the nervous system.
figure 2

a, b, Ube3am−/p+ (AS model) mice were injected i.c.v. bilaterally at E15.5 and P1 with 1.5 × 1010 AAV particles per ventricle (6 × 1010 AAV particles per mouse) containing SaCas9 and either the negative control gRNA (a) or Sajw33 (b) and immunostained for UBE3A and NEUN at P90. Lower magnification images are shown in Extended Data Fig. 4f, g. Scale bars, 50 μm. c, d, Western blot (c) and quantification (d). UBE3A levels in the cerebral cortex, hippocampus (hip.), cerebellum (cblm.) and spinal cord of wild-type (WT) and AS (Ube3am−/p+) mice treated with control gRNA or Sajw33. n = 5 per group. **P < 0.001, not significant (NS). For confocal image and gel source data, see Supplementary Figs. 1 and 2.

It is currently unknown whether AAV can pass from treated fetuses to the mother, which is a potential safety concern for in utero gene therapies. To address this issue, and to evaluate tissue distribution of the SaCas9 vector more broadly, we quantified viral DNA (vDNA) in the cerebral cortex and the liver of dams and their P60 offspring (all dual injected). In the treated offspring, vDNA levels were higher in the cerebral cortex than in the liver (Extended Data Fig. 7a). No vDNA was detected in the cerebral cortex or the liver of the dams (Extended Data Fig. 7a), indicating that AAV did not transfer to the dams during pregnancy. Furthermore, SaCas9 RNA expression was largely restricted to the brain of treated offspring (Extended Data Fig. 7b), consistent with neuron-specific expression of the hSyn1 promoter. Finally, organ morphology was normal and no tumours were detected in 10-month-old treated mice (Supplementary Table 3).

Restoring Ube3a function during the embryonic and early postnatal period is predicted to be more efficacious at treating AS than restoring Ube3a in adulthood1,2. Thus, we next evaluated the extent to which dual E15.5 + P1 i.c.v. injection of the SaCas9 AAV vector rescued anatomical and behavioural phenotypes in AS model mice that are reproducible20,21 and that interrogate core symptoms of AS, including microcephaly and deficits in proprioception and motor function (Fig. 3a, raw data presented in Supplementary Table 4). AS mice also become obese as they age4. Body weight returned to wild-type levels in female, but not in male, AS mice that were treated with the SaCas9 + Sajw33 vector (Fig. 3b, Extended Data Fig. 8a). Brain weight at 10 months of age was also increased in AS mice treated with the SaCas9 + Sajw33 vector (Fig. 3c), suggesting that microcephaly can be partially rescued. AS mice treated with the SaCas9 + Sajw33 vector also demonstrated behavioural improvements in hindlimb clasping and centre time in the open field, but showed no rescue in the distance travelled in the open field or marble burying (Fig. 3d–g, Extended Data Fig. 8b, c). Improvements in the rotarod test were evident at 2 months of age and endured at 7 months of age (Fig. 3h, i, Extended Data 8d, e). In our hands, AS mice showed no deficits in contextual-based and cue-based learning and memory tasks (Extended Data Fig. 8f, g), consistent with several other labs (reviewed in ref. 20), so we were unable to evaluate the rescue of cognitive deficits. In summary, our data suggest that neonatal reactivation of paternal Ube3a can enduringly treat many symptoms of AS.

Fig. 3: The SaCas9 gene therapy vector rescues phenotypes in the AS mouse model.
figure 3

a, Timeline of assays performed on WT and AS mice injected i.c.v. bilaterally at E15.5 + P1 with 1.5 × 1010 AAV particles per ventricle (6 × 1010 AAV particles per mouse) containing SaCas9 and the negative control gRNA or Sajw33. b, Body weight of female mice measured monthly over 9 months. c, Brain weight measured at 40 weeks of age. d, Hindlimb clasping assay in 4-week-old mice. e, f, Open-field data for minutes 15–30 in 11-week-old mice. The time spent in the centre (e) and the distance travelled (f) are shown. g, Marble burying task in 12-week-old mice during a 30-min trial. h, i, Rotarod testing data (average of two trials) at 8 weeks of age (h) and 28 weeks of age (i). *P < 0.05, **P < 0.01, ***P < 0.001. All boxplot whiskers extend to the highest and lowest values within ±1.5× the interquartile range, and the boxes extended from quartile 1 to quartile 3, with the middle line showing the median.

We next sought to determine how the gene therapy vector disrupted Ube3a-ATS in vivo by focusing on brain samples from 10-month-old AS mice and controls. Using high-throughput DNA sequencing, we detected indels and substitutions in 0.3% of the Sajw33 target site amplicons (Extended Data Fig. 9a, b) and no evidence for mutations at predicted off-target sites (Supplementary Table 5; see Methods). Moreover, the genomic copy number for Snord115 was not increased or decreased (Extended Data Fig. 9c). Thus, indels and deletions/duplications were unlikely to contribute to widespread unsilencing of paternal Ube3a in mice.

AAV integrates into SaCas9-generated double-strand break sites at a relatively high frequency22,23. To identify AAV integration events in Snord115 genes, we performed PCR from cortical genomic DNA (gDNA) using primers that bind in the Snord115 locus and in the AAV vector (Extended Data Fig. 9d). We detected forward and reverse integration events in Sajw33-treated animals, but not in animals treated with the negative control gRNA vector (Extended Data Fig. 9e; a primer). Using primers that anneal farther within the AAV vector (Extended Data Fig. 9e; b primer), we detected additional faint bands that corresponded to truncated AAV vectors, which we validated by Sanger sequencing and high-throughput amplicon sequencing (Extended Data Fig. 9f, g). No AAV integration events were detected at predicted off-target sites (Extended Data Fig. 9h). Using qPCR, we detected approximately one AAV integration event per diploid genome (Extended Data Fig. 9i). Assuming that 50% of all diploid cells in the brain are neurons24, and since ~58% of all cortical neurons showed paternal Ube3a unsilencing, our data suggest that there are approximately four integrated viral vectors per paternal Ube3a unsilenced neuron.

In addition, we performed RT–PCR with primers that amplify across AAV integration sites. We detected fusion RNAs between Ube3a-ATS and the AAV vector only in Sajw33-treated animals (Extended Data Fig. 9j). In the forward orientation, fusion transcripts contained the polyadenylation sequence element from the 3′ untranslated region of SaCas9, resulting in premature transcription termination of Ube3a-ATS (Fig. 4a). Thus, AAV vector integration is functionally analogous to a gene trap4,25. In the reverse orientation, fusion transcripts were not detected beyond SaCas9 using a primer walking strategy (Fig. 4b), suggesting that convergent RNA polymerase II transcription blocks Ube3a-ATS, analogous to the paternal Ube3a/Ube3a-ATS collision model4. Together, these data suggest that AAV integration can disrupt Ube3a-ATS (Extended Data Fig. 10) and occurs frequently enough in vivo to account for the long-term, probably permanent, unsilencing of paternal Ube3a.

Fig. 4: AAV integration traps Ube3a-ATS.
figure 4

a, Identification of polyadenylated Ube3a-ATS/AAV fusion transcripts. (1) Total RNA was extracted from cortices of 10-month-old mice that were injected at E15.5 and P1 with AAV containing SaCas9 and the negative control gRNA or Sajw33. cDNA was synthesized using a polydT primer with a 3′ adaptor sequence (green line). (2) RT–PCR with forward primer, which anneals in Ube3a-ATS upstream of the Cas9 target site, and reverse primer annealing in polydT 3′ adaptor. PCR resulted in a smear (not shown) due to the repetitive binding site of the forward primer and complex splicing patterns over the Snord115 locus, as previously described26. (3) Product from (2) was purified and used as template in a second PCR with primers to detect polyadenylated Ube3a-ATS (3a), and polyadenylated Ube3a-ATS/AAV fusion RNA (3b). Minus (–) Taq polymerase with Ube3a-ATS primers was the negative control. b, Primer walking strategy to detect Ube3a-ATS/AAV fusion mRNAs. One forward primer (F) was used in combination with reverse primers that anneal progressively further into the AAV genome. The largest band in each lane is the correct predicted size based on integration in the inverted terminal repeat (ITR); smaller bands reflect integrations of truncated AAV genomes.

Methods

Animals

C57BL/6 mice, Ube3am−/p+ mice and Ube3am−/pYFP mice on the C57Bl/6 background and genotyping procedures were previously described3. All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of North Carolina at Chapel Hill, and in accordance with the US NIH guidelines.

gRNA cloning and Cas9 expression plasmids

To identify possible gRNA target sites, we analysed RNA sequencing (RNA-seq) data from cortical neuron cultures27, and publicly available data sets such as conservation (PhyloP), CTCF-binding sites, DNase hypersensitivity sites, CHiP–seq, polyadenylation sites and predicted RNA secondary structure. All gRNAs were cloned using Golden Gate assembly. Experiments using SpCas9 described in Fig. 1 and Extended Data Fig. 1 used lentiCRISPR v228. Experiments using SaCas9 used pX601-AAV-CMV::NLS-SaCas9-NLS-3×HA-bGHpA;U6::BsaI-sgRNA13, in which the CMV promoter was replaced with the hSyn1 promoter using XbaI and AgeI restriction sites. Experiments using mCherry-tagged SpCas9 described in Extended Data Fig. 2 used lentiCRISPRv2-mCherry (99154, Addgene). The pLenti-Camk2a-tdTomato vector was previously described29.

gRNA library screen

Primary mouse neuronal cultures were prepared as described3,27,30. Briefly, cortical neurons were dissected from E15.5 Ube3am+/pYFP embryos and plated in 384-well poly-d-lysine-coated plates. On days in vitro 3 (DIV3), each well was transfected with 50 ng Camk2a:tdTomato and 50 ng of lentiCRISPR:gRNA plasmid using Lipofectamine 2000 (Thermo Fisher). Each gRNA was transfected in quadruplicate. On DIV10, cells were fixed with 4% phosphate-buffered paraformaldehyde (PFA), and stained with primary rabbit anti-GFP antibody (NB600-308, Novus), secondary donkey anti-rabbit IgG Alexa 647 (A31573, Thermo Fisher) and DAPI. Images were acquired using the GE IN CELL Analyzer 2200 high-content imager. YFP expression was quantified in tdTomato+ nuclei using a custom Cell Profiler pipeline. The screen was performed in triplicate, presented as the average of the three replicates. All subsequent experiments using mouse primary cortical neuron cultures followed the same culture protocol and timeline.

RNA extractions, qPCR and RT–PCR

All RNA and DNA extractions were performed using TRIzol (Thermo Fisher). Total RNA was treated with DNase (New England Biolabs (NEB)) for all qPCR experiments. cDNA synthesis for experiments that involved standard qPCR and RT–PCR were performed using SuperScript IV VILO (Thermo Fisher). All standard qPCR experiments were performed using SsoAdvanced Universal SYBR Green Supermix (NEB) on the QuantStudio3 or Quantistudio5 (Applied Biosystems). RT–PCR experiments from P60 mouse cortical RNA (Extended Data Fig. 5) were performed with Platinum Taq polymerase (Invitrogen) and primers from Kishore et al.31. Primers that were used in all experiments are listed in Supplementary Table 6.

Spjw33 target site mutation analysis

E15.5 WT primary mouse neuron cultures were treated on DIV3 with lentivirus carrying SpCas9 and either Spjw33 or a scrambled gRNA. gDNA was extracted at DIV12, and the Spjw33 target site was amplified using nine primer sets designed to amplify nearly all copies of Snord115: forward primers: Snord115 1-3 F; reverse primers: Snord115 1-3 R. Amplicons were cloned via TopoTA cloning (Invitrogen) and individual colonies were analysed for mutations by Sanger sequencing.

Use of gRNA pairs to delete defined regions of Ube3a-ATS

gRNAs were selected based on target location in the Ube3a-ATS locus and lack of efficacy when tested alone in the primary screen (Supplementary Table 1). Of each gRNA, 50 ng was transfected either alone (−) or in combination with one other gRNA in E15.5 Ube3am+/pYFP embryos as described above. Unsilencing of paternal UBE3A-YFP was determined as previously described, with the only exception being the replacement of the CamK2a:td:tomato reporter plasmid with immunofluorescent labelling of SpCas9+ cells with an SpCas9 antibody (clone 7A9, BioLegend).

Analysis of UBE3A expression in differentiated phNPCs

Human fetal brain tissue was obtained from the UCLA Gene and Cell Therapy Core following IRB regulations. phNPCs were grown and differentiated as previously described12. Cells were mycoplasma tested and confirmed to be mycoplasma free. Briefly, cells were thawed and plated in 10-cm plates with proliferation media (Neurobasal A supplemented with primocin, BIT9500, glutamax, heparin, EGF, FGF, LIF and PDGF) in a humid incubator at 37 °C with 5% (vol/vol) CO2. After two passages, cells were transferred into six-well plates, 4 × 105 cells per well, and changed to differentiation media (Neurobasal A, primocin, B27+, glutamax, NT3 and BDNF) 24 h post-plating. A 50% media change was performed every 2–3 days. On day 14, 1.6 × 1010 AAV2 particles carrying hSyn1:eGFP (pAAV-hSyn1-eGFP, UNC Vector Core) were added to each well. On day 42, 4 × 108 lentivirus particles carrying SpCas9–mCherry and gRNAs were added to cultures, six wells per gRNA. hsa#1: 5′-CCTCTCTTCAGAACAATATA, hsa#2: 5′-TGGTCTCCTGCACTGAGCTG and hsa#3: 5′-TGCTCAATAGGATTACGCTG. On day 56, cells were lifted, six wells for each gRNA were pooled, stained with DAPI for live/dead discrimination, and sorted using the FACSAria II cell sorter. Cells were collected and total RNA was extracted using TRIzol. RNA was treated with DNase I (NEB). For allele-specific qPCR, cDNA was generated from 100 ng total RNA using SuperScript III (Thermo Fisher) and 2 pM of UBE3A-specific primer (5′-TCCTTTAGATCATACATCATTGG). Allele-specific expression levels were determined using TaqMan Universal Master Mix II (Thermo Fisher), and TaqMan genotyping probes for rsID:61734190 (Applied Biosystems). Quantification of allele-specific expression was calculated by assessing the distance between Ct values for each allele within each replicate at ∆Rn = 0.5 (n = 4). For qPCR of genes in the UBE3A-ATS locus, cDNA was synthesized from 100 ng total RNA using the VILO SSIV reverse transcriptase (Thermo Fisher). Quantification was performed using the ∆∆Ct method, normalized to EIF4A2. Presumption of the parental allele was based on stranded RNA-seq reads, with the paternal allele assigned to the SNV (A) that was observed from the UBE3A-ATS strand (which was observed only in neurons), and the maternal allele assigned to the alternate SNV (T).

Brain injections

Ube3am−/p+ dams were anaesthetized under 2% isoflurane throughout the procedure. Stock virus was diluted in PBS and a final concentration of 0.05% fast green was added immediately before the procedure. 1.5 × 1010 particles per ventricle of virus was delivered into each embryo via bilateral i.c.v. injection at E15.5. Standard protocol for in utero injection was followed with no electroporation step32. The dams were then individually housed and delivered the pups naturally. At P1, neonatal pups were placed on ice to induce hypothermia anaesthesia and bilaterally injected with virus at 1.5 × 1010 particles per ventricle following a previously published protocol33. Injected pups were allowed to recover on a heated pad before returning to their dam.

Immunohistochemistry

After behavioural experiments, mice were anaesthetized with sodium pentobarbital (60 mg/kg). Each brain was cut sagittally through the midline. One-half was flash frozen on dry ice for use in western blotting analysis, while the other half was immersion fixed in 4% PFA in PBS, pH 7.4, before cryoprotecting by incubation in 30% sucrose in PBS (for 48 h at 4 °C) and sectioning them to a thickness of 60 μm (P30 and P90 samples) or 100 μm (17-month-old samples) on a vibratome (Thermo Fisher Scientific). For immunostaining, we permeabilized sections with 0.01 M PBST (PBS containing 0.2% Triton X-100) for 30 min and then blocked with 5% donkey serum in PBST for 1 h at room temperature. We incubated blocked sections with mouse IgG2A anti-UBE3A (1:1,000; clone 3E5, SAB1404508, Sigma-Aldrich) and guinea pig anti-NEUN (1:1,000; ABN90P, EMD Millipore) overnight at 4 °C with gentle shaking. For the AAV toxicity study, we used goat anti-GFAP (1:1, 000; ab53554, abcam), rabbit anti-Iba1 (1:300; 019-19741, Wako), rat anti-TBR2 (1:300; 14-4875-82, eBioscience) and goat anti-DCX (1:500; sc-8066, Santa Cruz). The next day, we washed sections several times with PBST, and incubated them with Alexa Fluor 488-conjugated anti-mouse IgG2A (1:500; A21131, Invitrogen), Alexa Fluor 568-conjugated anti-rabbit (1:500; A10042, Invitrogen), Alexa Fluor 568-conjugated anti-rat (1:500; A21208, Invitrogen), Alexa Fluor 568-conjugated anti-goat (1:500; A11057, Invitrogen), Alexa Fluor 647-conjugated anti-guinea pig (1:500; A21124, Invitrogen) and DAPI (7 mg/ml; D1306, Invitrogen) during the secondary antibody incubation. We imaged sections via laser-scanning confocal microscopy (Zeiss LSM 710 and 780). Images were quantified with Fiji image processing software and Cell Profiler.

Western blotting

Brain regions were dissected from one hemisphere and protein lysates were prepared in RIPA lysis and extraction buffer (Thermo Scientific) containing protease inhibitor (P8340, Sigma) and phosphatase inhibitor (78426, Thermo Fisher). Total proteins (60 μg) were fractionated by SDS–PAGE and transferred to a nitrocellulose membrane. Membranes were blotted with rabbit anti-UBE3A (1:500; 10344-1-AP, Proteintech) and rabbit anti-GAPDH (1:2,000; 2118, Cell Signaling) in Intercept Blocking Buffer (927-70001, LI-COR) and secondary anti-rabbit antibody IRDye680RD (1:10,000; C60813-05, LI-COR). Membranes were imaged with the ODYSSEY CLx Infrared Imaging System (LI-COR). Relative UBE3A protein level was measured based on the loading control (GAPDH) with Fiji image processing software.

Behavioural assays

All behavioural assays were carried out blind to genotype and treatment of animals. Power analyses for the behavioural tests were previously reported20, and is reflected in our sample sizes. All mice were randomly assigned treatment conditions.

Hindlimb clasping assay

At P30, each mouse was held by the tail at least 10 inches away from a hood bench (measured from the tip of the tail) for 30 s. Each mouse received two trials, with 5 min inter-trial intervals. Trials were recorded with a Samsung HMX-F80 camera and scored offline. Clasping behaviour was defined by movement of the hindlimbs (either one or both) curling inward towards the belly of the animal. Data are presented as the total time spent clasping across two trials.

Accelerating rotarod

For the accelerating rotarod (3–30 r.p.m. over 5 min; model 47600, Ugo Basile Biological Research Apparatus) experiment, mice were given three trials on the first day with a 10 min inter-trial interval (training session) and re-tested on day three with two trials (testing session). For each day, the average time spent on the rotarod was calculated, or the time until the mouse made three consecutive wrapping/passive rotations on the rotarod (latency in seconds). The maximum duration of a trial was 5 min.

Open field

Mice were individually placed in a 45 × 45 cm square open field and allowed to explore for 30 min. The total distance moved by each mouse in the open arena was recorded by camera (Noldus Wageningen) connected to the EthoVision software (Noldus Wageningen). The centre zone was defined as a 35 × 35 cm square field in the middle of the arena.

Marble burying

Open polycarbonate cages (50 × 26 × 18 cm) were filled with 3 litres of bedding material (1/8” Corn Cob Bedding, irradiated, CC8-IRR, Lab Supply). On top of the bedding material, 20 black glass marbles were arranged in an equidistant 5 × 4 grid and the animals were given access to the marbles for 30 min. After the test, the mice were gently removed from the cage. Marbles covered by more than 50% by bedding were scored as buried.

Fear conditioning

All mice were brought into the experiment room at least 20 min before testing. On day 1 of conditioning, each mouse was placed in an individual chamber with the house light on and given three presentations of a 30-s tone, paired with a 2-s foot shock of 0.4 mA after 2 min of acclimation, with 80 s between the first and second pairing and then 120 s between the second and third pairing. On day 2, for the context-based learning experiment, each mouse was put back into the chamber from day 1 with the house light on and filmed for 5 min. On day 3, for the cue-based learning experiment, each mouse was put in a chamber with a new floor panel, a new black triangular insert. Vanilla extract (0.1 ml) was also placed in the chamber. The house light was turned off, and after 2 min of acclimation the tone was presented for 3 min. All activities were videotaped, the number and length of freeze were called using the Near-Infrared image tracking system (MED Associates).

vDNA qPCR

Organs from treated mice or dams were extracted at P30. Tissue (0.5–0.8 g) was isolated from each organ and DNA/RNA was extracted using TRIzol. DNase (100 ng; NEB)-treated total RNA was used in cDNA synthesis using the VILO SSIV reverse transcriptase mix (Thermo Fisher). SaCas9 levels were quantified in gDNA and cDNA samples using TaqMan probes and normalized to Eif4a2 or Actb, respectively.

In silico off-target prediction

Cas-OFFinder34 was used to identify Sajw33 off-target binding sites in the mm10 reference genome with the following parameters: Pam type = SaCas9 (5′-NNGRRT-3′), target genome = Mus musculus (mm10), seq = 5′-CTGAGGCCCAACCAGGGCGA, mismatch number = 6, DNA bulge size = 0 and RNA bulge size = 0. Primers were designed to amplify the loci flanking the top ten matches.

Next-generation amplicon sequencing

PCR and library preparation

We performed PCR from 5 ng gDNA isolated (TRIzol) from the cortex of 10-month-old mice dual injected with either negative gRNA (n = 3) or Sajw33 (n = 3). Multiple primer sets were designed to amplify all Sajw33 target sites across the Ube3a-ATS locus, and AAV integration events (Snord115 1 F, Snord115 4 F, Snord115 5 F, Snord115 1 R, Snord115 2 R, Snord115 3 R, ITR C R, ITR 2 R and AAV hSyn R). Off-target primers are listed in Supplementary Table 6. PCRs were performed on the QuantStudio5 qPCR machine and halted once all samples reached exponential growth phase (cycle 26) to avoid over-amplification bias. Each PCR was performed in triplicate to reduce random PCR sampling bias (360 PCRs total). PCRs from individual mice were pooled, PCR purified (Qiagen), end repaired and A-tailed (KAPA HyperPlus kit), and adaptors were ligated using KAPA dual-indexed adaptors ligation kit (KAPA Biosystems). Libraries were purified using AMPure XP beads (Beckman Coulter), and sequenced using the Illumina Miniseq system with the Miniseq Mid Output kit (Illumina) (300 cycles), 150-bp paired-end reads. Reads were filtered for adaptor contamination using cutadapt35 and filtered such that at least 90% of bases of each read had a quality score of >20.

Off-target

Reads from the top ten predicted off-target sites were analysed for mutations using the following approach: (1) reads were aligned to mm10, retaining reads that did not align perfectly. (2) Reads were removed that were present in negative samples, retaining putative off-target mutations that contained at least one read in each of the three Sajw33 replicates. Remaining reads were manually inspected for mismatches to the mm10 reference within 10 bp of the protospacer adjacent motif site. No reads survived these criteria (966,058 total reads from Sajw33-treated animals were analysed).

AAV integration PCR

gDNA PCR

gDNA was isolated from total cortical lysate stored in RIPA buffer using ethanol precipitation. gDNA (5 ng) was used in all PCRs. Primers in Extended Data Fig. 9d, e: A: 5F, Z: 1R, a: ITR 1 C R, b: AAVhSyn Rev, SaCas9: SaCas9 1 For/Rev. AAV integration events in Extended Data Fig. 9i were normalized to a region of Ube3a-ATS that exists only once in each allele, representing signal obtained for one diploid genome. Primers in Extended Fig. 9j: Ube3a-ATS 3′ 1 F/R, priwalk F/ITR 2 R. PCRs were performed using Platinum Taq polymerase (Invitrogen). qPCRs were performed using SsoAdvanced Universal SYBR Green Supermix (NEB).

RT–PCR

RNA was extracted from the cortex of 10-month-old E15.5 + P1 AAV-treated mice using TRIzol. RNA was DNase I treated (NEB) and extracted using TRIzol a second time. To identify Ube3a-ATS/AAV fusion transcripts (Extended Data Fig. 9j), 500 ng total was used to synthesize cDNA using SuperScriptIV VILO (Thermo Fisher). cDNA (0.05 µl) was loaded into a PCR using primers to amplify Ube3a-ATS (Ube3a-ATS 3′ 1 F/R), SaCas9 (SaCas9 1 F/R) and Ube3a-ATS/AAV fusion transcripts (priwalk F, ITR 1 R).

To identify polyA-trapped Ube3a-ATS/AAV fusion transcripts (Fig. 4a), cDNA was synthesized from 500 ng treated RNA using SuperScript IV (Invitrogen) and 3′RACE poly-dT primer36. cDNA (0.05 µl) was loaded into a PCR using primers to amplify polyadenylated Ube3a-ATS transcripts (forward: Snord115 HG, reverse: 3′RACE anchor). The resulting PCR product was PCR purified (PCR Purification Kit, Qiagen). Purified PCR (1 pg) was used as template with primers to amplify Ube3a-ATS (positive control, primers Ube3a-ATS 3′ 1 For/Rev), and Ube3a-ATS:AAV:pA fusion transcripts (forward: SaCas9 2 F, reverse: 3′RACE anchor). PCR products were purified and Sanger sequenced to confirm identity.

For the primer walking strategy (Fig. 4b), cDNA was synthesized from 500 ng total RNA using SuperScriptIV VILO (Thermo Fisher), which synthesizes cDNA using a mixture of random hexamers and polydT primers. RT–PCR primers are listed in Supplementary Table 6. RT–PCR was performed using Platinum Taq polymerase (Invitrogen), 5.5′ extension time, 3% KB extender, 40 cycles and touchdown PCR annealing temperatures (67 °C–60 °C for 15 cycles, −0.5 °C per cycle) to ensure optimum specificity of primer annealing.

Statistical analyses

All results from analyses are presented as the mean ± s.e.m. and differences were considered significant when P < 0.05. Two-tailed, unpaired Student’s t-test was used for comparisons between two normally distributed groups. For behavioural data consisting of more than two groups, varying in a single factor, one-way analysis of variance (ANOVA) and Kruskal–Wallis test were used. Comparison of non-normally distributed behavioural data were performed using a non-parametric Kruskal–Wallis test.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.