Precise, predictable genome integrations by deep-learning-assisted design of microhomology-based templates
Cell culture
HEK293T (American Type Culture Collection (ATCC), CRL-11268) were cultured as recommended by the ATCC. Cell lines tested negative for Mycoplasma and were authenticated by the suppliers.
Modeling of gene-editing outcomes
The inDelphi model was obtained from GitHub (https://github.com/maxwshen/inDelphi-model) and deployed in a suitable Python virtual environment (https://github.com/XenoThomasNaert/Pythia-Editing). To investigate the impact of the number of tandem repeats on the expected percentage of perfect DNA repair, we developed custom Python code. The percentage repair by µH is defined as the sum of all repair outcomes that use at least one µH tandem repeat. The code iteratively analyzes µH tandem repeat lengths ranging from two to six and the number of tandem repeats from one to eight. This analysis was conducted using the inDelphi HEK293T or mouse embryonic stem cell (mESC) predictive model for the first 250,000 gRNA sites identified by presence of an NGG PAM, encountered in the human gencode v43 transcript sequences. For all HEK293T experiments, predictive modeling was performed using inDelphi’s HEK293T mode, whereas, for predictive modeling in Xenopus and mice, the mESC mode was used as it was validated as predictive in early-dividing Xenopus embryos14.
Cloning and in vitro linearization of PaqMan repair plasmids and PCR generation of repair templates
Donor plasmid was assembled in a pUC19 backbone using Gibson cloning (NEBuilder HiFi DNA assembly master mix) and featured a pCMV-eGFP transgenic cassette flanked by zero, four or five µH tandem repeat repair arms and inverted PaqCI restriction enzyme sites. The insert was obtained from AAV-CMV-GFP, which was a gift from C. Cepko (Addgene, plasmid 67634; RRID:Addgene_67634). The pUC19 destination vector was commercially purchased (N3041S, New England Biolabs (NEB)). Inverted PaqCI sites and µH tandem repeat repair arms were added by overhang PCR before Gibson assembly. Linearization was performed by overnight digest at 37 °C of 10 µg of donor plasmid using 20 U of PaqCI (R0745, NEB) in 1× rCutSmart buffer (B6004S, NEB). Complete linearization was ensured using classical agarose gel electrophoresis.
Alternatively, repair templates containing µH tandem repeat repair arms were generated by overhang PCR using Phusion polymerase (Thermo Fisher, F530S) with primers designed to contain an overhang sequence containing the µH tandem repeat repair arms (listed in Supplementary Table 1). For in vitro use, PCR products were cleaned using a MinElute PCR purification kit (28004, Qiagen) and eluted in ultrapure water. For in vivo use, PCR products were cleaned by classical phenol–chloroform extraction with sodium acetate–ethanol precipitation and quantified using Nanodrop (ThermoFisher).
µH tandem repeat-mediated integration in vitro
AAVS1 gRNA was assembled using Alt-R CRISPR–Cas9 IDT CRISPR RNA (crRNA) and Alt-R CRISPR–Cas9 trans-activating crRNA (tracrRNA), according to the manufacturer’s instructions, by heating it to 95 °C and cooling it to room temperature, yielding a duplex at a final concentration of 1 μM. Cas9 protein (PNABio, CP01) was diluted to 166.67 ng μl−1 in 1× PBS. HEK293T cells were reverse-transfected using Lipofectamine CRISPRMAX (Thermo Fisher, CMAX00003) as follows. RNP was assembled by incubation for 5 min at room temperature of 1 μM gRNA duplex, 250 ng of Cas9 protein and 0.6 μl of Cas9 PLUS reagent (from CRISPRMAX kit) in 23 μl of Opti-MEM (Thermo Fisher, 31985070). Then, 200 ng of PaqCI (R0745, NEB) digest product was added to the RNP. Transfection complexes were generated by incubation at room temperature for 20 min of 25 μl of RNP repair template, 1.2 μl of CRISPRMAX transfection reagent and 23.8 μl of Opti-MEM medium. Resulting transfection complexes were mixed with 40,000 HEK293T cells (suspended in a total volume of 100 μl of DMEM) and plated on 96-well Nunclon plates (Thermo Fisher, 167008). Cells were cultured for 25 days and cell sorting for GFP+ cells was performed.
For TRAC CAR knock-in, gene editing was performed identically to above, with some exceptions. Specifically, a CD19-specific CAR expression construct based on pUC19-HDRT-TRAC-CD19.CAR-Cas12a.PAM.mutated (Addgene, plasmid 215769; RRID:Addgene_215769)30 was ordered synthetically. The construct consisted of P2A, CD19-Car, bHg poly(A) and 400 bp of classical HDR homology arms. For targeting the TRAC locus, we used the following gRNA 5′-AGCTGGTACACGGCAGGGTC-3′. Repair templates were generated containing classical HDR homology arms (400 bp), no repair arms (HITI) or tandem repair arms by PCR. We used 100 ng of repair template (instead of 200 ng) and transfection was performed 1 day after seeding of 20,000 cells in a 96-well plate.
For the 32-target experiment, gRNAs were designed for the coding sequence (CDS) from human genome assembly GRCh38 using a custom python script, identifying gRNAs with each permutation of strong (S) and weak (W) bases at positions −6 to −4 and AGG at positions −3 to −1 with NGG as the PAM at positions 0 to 2. Identified gRNAs were filtered for those with CRISPRScan scores73 exceeding 80. To avoid negative selection because of gene essentiality when targeting CDS, we filtered the gRNA list to exclude any gene occurring in DEG15 (ref. 32), a database of essential genes as determined from shRNA and CRISPR–Cas screens. Next, the eight classes of permutations involving S and W bases were sorted into bins. For each class, one gRNA was selected per bin, arranged according to the degree of sequence context µH, ranging from low to high. For each gRNA, the off-target profile was determined and deemed acceptable using Cas-OFFinder69 (list of gRNAs in Supplementary Table 2).
Gene editing was performed identically to above, with some exceptions. Specifically, we used 100 ng of repair template (instead of 200 ng), generated by overhang PCR as described above from AAV-CMV-GFP (Addgene, plasmid 67634; RRID:Addgene_67634) (Supplementary Table 3). Transfection was performed 1 day after seeding of 25,000 cells in a 96-well plate. Cells were sorted on days 2 and 15. Here, integration efficiency was defined as follows. All cells were pregated on live cells, using SYTOX deep red nucleic acid stain (1 µM final) (Thermo Fisher, S11380). Then, the percentage of GFP+ cells on day 15 was calculated as a proportion of the percentage of GFP+ cells at day 2, thus accounting for differences in initial transfection efficiency by transient expression of the pCMV:eGFP cassette on day 2. On-target efficiency was defined as the difference between the integration efficiency of on-target gRNA and that of negative control gRNA on day 15, thus identifying the level of true on-target integration.
For ssDNA donor experiments, dsDNA repair templates were generated by overhang PCR as described above from AAV-CMV-GFP (Addgene, plasmid 67634; RRID:Addgene_67634). ssDNA was generated and quality-controlled by the Guide-it Long ssDNA production system v1 (Takara Bio, 051818) according to the manufacturer’s instructions. Here, 50 ng of ssDNA repair template was used. On day 15, gene integration by GFP+ cells was quantified using flow cytometry. Living cells were pregated before gating for SYTOX deep red nucleic acid stain (1 µM final) (Thermo Fisher, S11380).
µH tandem repeat-mediated integration and gene tagging in vivo in Xenopus
X. tropicalis animals were kept according to Swiss law for care and handling of research animals. Husbandry and treatment were approved by the local authorities (Veterinäramt Zurich). Gene symbols follow Xenbase (http://www.xenbase.org/, RRID:SCR_003280). For Xenopus experiments, repair templates for pCMV:eGFP, pax8-CNS1:eGFP and CarAct:dsRed2 were generated by overhang PCR as described above. For pCMV:eGFP (5× 3-bp µH tandem repeats) and pax8-CNS1:eGFP experiments, h11-α and h11-β gRNAs were assembled as follows: 1 µl of Alt-R CRISPR–Cas9 IDT crRNA (100 µM stock) and 1 µl of Alt-R CRISPR–Cas9 tracrRNA (100 µM stock) were mixed with 3 µl of nuclease-free duplex buffer (IDT) and heated at 95 °C for 5 min and allowed to cool to room temperature. For RNP assembly, 1.8 μl of Cas9 protein (1 μg μl−1, PNABio CP01) was mixed with 0.2 μl of gRNA and heated to 37 °C for 5 min, before adding repair template. The final injection mix consisted of 1 μl of h11-α RNP, 1 μl of h11-β RNP and 1 μl of repair template (stock concentration: 10 ng μl−1), thus yielding a final repair template concentration of 3.33 pg nl−1. Embryos were injected unilaterally at the two-cell stage. For pCMV:eGFP (4× 3-bp µH tandem repeats) and CarAct:dsRed2, we mixed Cas9 protein (3 μl at 1 μg μl−1, PNABio CP01) with gRNA (1 μl) and incubated for 5 min at 37 °C to assemble RNP. RNP was mixed with repair template at a ratio of 4:1; thus, adding 1 μl of repair template (10 ng μl−1) to the mix yielding a final repair template concentration of 2 pg nl−1. Embryos were injected at the one-cell stage immediately after cortical rotation, targeting the gray sperm entry point with 5–10 nl of injection mix.
Embryo development was monitored and, at Nieuwkoop–Faber stage 40, embryos were lysed (50 mM Tris pH 8.8, 1 mM EDTA, 0.5% Tween-20 and 2 mg ml−1 proteinase K) overnight at 55 °C. Three classes of embryos were lysed as follows: embryos with unilateral or bilateral nonmosaic fluorophore expression, embryos with mosaic expression often restricted to a subset of the muscle cells and control embryos of the same clutch that were not microinjected. After proteinase K inactivation, junction products between the h11 locus and transgene cassette were picked up using PCR and subjected to Sanger sequencing. Whole-embryo bleaching, staining and clearing were performed as previously described74 using 1:250 anti-GFP (Aves, GFP-1020) and 1:250 anti-RFP (Rockland, 600-401-379-RTU).
For tagging, the repair templates including the homology arms (Supplementary Table 4) were ordered from Twist Biosciences, PCR-amplified, and phenol–chloroform-purified. RNP was assembled as described above and was coinjected with repair template (2–8 ng μl−1) and TRITC–dextran (0.5 ng μl−1; Sigma-Aldrich). Embryos were sorted for TRITC fluorescence the next morning and mBaoJin fluorescence was assessed at the tailbud stage. Stage 45 tadpoles were anesthetized in 0.02% MS-222 (Sigma-Aldrich, A5040) for confocal live-cell imaging. Then, tadpoles were tail-clipped for genomic DNA extraction. Tails were lysed (50 mM Tris pH 8.8, 1 mM EDTA, 0.5% Tween-20 and 2 mg ml−1 proteinase K) at 55 °C overnight and heat-inactivated at 98 °C. Boundary products were amplified using phusion polymerase (NEB, M0530L) and sent for commercial Sanger sequencing (Microsynth). Sequence alignments were performed in Benchling. The tadpoles were fixed in 4% PFA (Merck, 158127) at 4 °C overnight and permeabilized in PBS–Tween-20 (0.1%; PanReac AppliChem, A4974). Immunostaining for myh9-tagged, acta2-tagged and ncam1-tagged animals was performed by incubation at 4 °C overnight in 1:100 anti-myosin IIA (Sigma-Aldrich, M8064), 1:100 phalloidin–FluoProbes 647 (Interchim, FP-BA0320) or 1:10 anti-Ncam1 (DSHB, supernatant, XAN-3 (clone 6F11)). The myh9 and ncam1 animals were further incubated at 4 °C overnight with 1:200 goat anti-rabbit IgG (H + L) DyLight 633 (Thermo Fisher, 35562) and 1:200 goat anti-mouse IgG (H + L) Alexa Fluor 633 (Thermo Fisher, A21050) respectively.
Immunoprecipitation of endogenously labeled Myh9, Ncam1 and Acta2 in Xenopus embryos
X. tropicalis embryos displaying unilateral or bilateral mBaoJin expression in a tissue-restricted manner were snap-frozen in liquid nitrogen at NF stage 42–45. Protein extraction was performed by placing 2–5 embryos in a 1.5-ml tube containing 500 µl of immunoprecipitation lysis buffer (Pierce, 87787) solution with freshly added protease and phosphatase inhibitor cocktail (Thermo Fisher, 78440). Homogenization of the tissue was achieved by 15 strokes of a 21G needle, followed by 10 strokes of a 26G needle. After 15-min incubation on ice, lysates were centrifuged at 16,000g for 30 min at 4 °C. Protein lysates were transferred to a fresh 1.5-ml tube. Then, 1% of the input samples were used for western blotting and the remaining lysate was subjected to immunoprecipitation. For precipitation, the protein lysates were incubated with anti-DYDDDK magnetic agarose beads (Thermo Fisher, A36797) for 3 h at 4 °C. Resin-associated proteins were washed four times with lysis buffer and eluted with 4× Laemmli buffer (Bio-Rad, 1610747). Samples were subjected to western blotting as described below. Anti-FLAG (Sigma-Aldrich, F3165; 1:1,000) was used as the primary antibody and anti-mouse IgG–peroxidase (Sigma-Aldrich, A8924; 1:5,000) was used as the secondary antibody for protein detection.
µH tandem repeat-mediated integration in vivo in mouse brain
pAAV-mTubb3 and pAAV-nEFCas9 were gifts from J. Belmonte (Addgene, plasmid 87116; RRID:Addgene_87116; Addgene, plasmid 87115; RRID:Addgene_87115). The sequence between AgeI and NdeI restriction sites was exchanged for a synthetic DNA fragment containing a gRNA targeting mTubb2a (5′-GGGCGAGTTCGAGGAGGAGG-3′), the 3′ Tubb2a sequence context, µH tandem repeat repair arms and eGFP to generate pAAV-mTubb2a. pAAV-nEFCas9 and pAAV-mTubb2a were packaged with serotype 8 and were generated by the Viral Vector Facility at the University of Zurich.
All procedures of mouse animal experimentation were carried out according to the guidelines of the Veterinary Office of Switzerland and following approval by the Cantonal Veterinary Office in Zurich (license 008/2022). Four C57BL/6 mice were used for virus injections. Mice were housed on a 12-h reversed light–dark cycle at an ambient temperature of between 21 °C and 23 °C, with the humidity level between 55% and 60%. Mice were anesthetized with 1.5–2% isoflurane mixed with oxygen and were head-fixed in a stereotactic frame (Kopf Instruments). Body temperature was maintained at ~37 °C using a heating pad with a rectal thermal probe. Vitamin A cream (Bausch & Lomb) was applied over the eyes to avoid dry eyes. After analgesia treatment (extended buprenorphine release EthiqaXR, 3.25 mg kg−1, subcutaneous; lidocaine over scalp), an incision was made on the scalp and small holes were drilled over bilateral visual cortex using the following coordinates: 3.5 mm caudal, 2.5 mm lateral relative to bregma and 0.5 mm ventral from the pia. We used 1:1 mixture of AAV-Cas9 (1.5 × 1013 genome copies (GC) per ml) and AAV-mTubb3 (2.3 × 1013 GC per ml) and injected 600 nl of AAVs in each hemisphere. To prevent virus backflow, the pipette was left in the brain for 5–10 min after completion of injection. Mice were housed for 3 weeks to allow for gene knock-in. Next, animals were killed and perfused using 4% PFA; brains were dissected and postfixed for 2 h in 4% PFA. Whole-brain staining was performed, adapted from previously described WildDisco48,75. Whole-mount brains were dehydrated to 100% methanol (high purity throughout this adapted wildDISCO, Supelco Emplura; Merck, 8.22283), delipidated with dichloromethane and bleached in 3% hydrogen peroxide prepared by diluting 30% H₂O₂ 1:10 in 100% methanol. Then, brains were permeabilized and blocked for 3 days using 10% donkey serum and 2% Triton X-100 in 1× PBS. Antibody staining was performed with 1:250 anti-GFP (Aves, GFP-1020) and 1:250 anti-RFP (Rockland, 600-401-379-RTU) in 5 ml of immunostaining buffer containing 3% donkey serum (Jackson ImmunoResearch, 017-000-121), 10% CHAPS (BioChemica, A1099), 10% dimethyl sulfoxide (DMSO), 1% glycine (Sigma-Aldrich, G7126) and 1% CD5 (Santa Cruz; sc-215141B) in 0.1× PBS for 7 days at 37 °C on a rotating wheel. After 3 days of washing, 1:400 of donkey anti-rabbit–Cy3 antibody (Jackson ImmunoResearch, 711-165-152) and donkey anti-chicken–AlexaFluor594 antibody (Jackson Immuno Research, 703-585-155) in immunostaining buffer was added for 7 days at 37 °C on a rotating wheel. Brains were washed extensively, dehydrated to 100% methanol in steps and then cleared overnight in BABB (high-purity solvents required; two parts benzyl benzoate (Sigma-Aldrich, 8.18701) and one part benzyl alcohol (Sigma-Aldrich, 108006)). For colocalization of Tubb2a and GFP, we performed standard immunofluorescence76, with 1:250 anti-GFP (Aves, GFP-1020) and 1:300 anti-tubulin βII antibody (Abcam, ab179512/EPR16773). As secondary antibody, we used 1:500 Alexa Fluor 594 AffiniPure donkey anti-chicken IgY (IgG) (H + L) (Jackson ImmunoResearch, 703-585-155) and 1:500 goat anti-rabbit IgG (H + L) secondary antibody, DyLight 488 (Thermo Fisher, 35552).
Immunoprecipitation of endogenously labeled Tubb2a derived from mouse brain tissue
Mice were injected with AAV-mTubb2a and AAV-Cas9 or ssAAV-8/2-hCMV-chI-EGFP-WPRE-SV40p(A)) as a control as described above, with the exception that the the AAV mixture was injected at eight locations (four locations per hemisphere) with 400 nl of AAV per injection site. After 21 days, mice were killed and brain halves were extracted and snap-frozen in liquid nitrogen. One brain half was used per immunoprecipitation reaction. Protein extraction was performed by placing thawed brain halves in 2-ml bead-beating tubes filled with 1.4-mm ceramic beads (Revvity, 19-627D) and 500 μl of 0.32 M sucrose containing immunoprecipitation lysis buffer (Pierce, 87787) and freshly added protease and phosphatase inhibitor cocktail (Thermo Fisher, 78440). Homogenization was achieved by shaking at 6 m s−1 for one cycle for 30 s using Bead Raptur Elite (Omni International, 19-042E). Brain lysates were incubated on ice for 5 min before being centrifuged at 500g for 5 min at 4 °C. The homogenized brain solution was carefully transferred to a fresh, prechilled 1.5-ml Eppendorf tube containing a layer of 1.2 M and a layer of 0.84 M sucrose lysis buffer solution. Gradient centrifugation was carried out at 21,000g for 60 min at 4 °C and deceleration set to 3 out of 10. The myelin-free fraction, which was found below the 0.84 M layer, was transferred to a fresh tube. An equal amount of lysis buffer was added, followed by a last centrifugation at 16,000g for 15 min at 4 °C. Then, 1% of input sample was taken from the clarified lysates. For immunoprecipitation of GFP or endogenous GFP-labeled Tubb2a protein, 25 μl of GFP-trap magnetic agarose beads (Proteintech, gtma) were used per reaction and incubated for 3 h at 4 °C on an overhead shaker. Resin-associated proteins were washed four times with lysis buffer and eluted with 4× Laemmli sample buffer (Bio-Rad, 1610747). Samples were loaded on 10% SDS polyacrylamide gels followed by western blotting on a PVDF membrane (Roth, T830.1). Blots were stained with Ponceau S (Roth, 5938.1) and blocked in 2.5% BSA for 1 h at room temperature. Incubation with primary antibody was performed overnight at 4 °C using anti-GFP (Thermo Fisher, MA5-15256; 1:1,000) followed by a 1.5-h incubation at room temperature with secondary antibody anti-horseradish peroxidase (Sigma-Aldrich, A8924, 1:5,000). The enhanced chemiluminescence detection system (Thermo Fisher, 32209) was used to visualize proteins using the Vilber Fusion FX machine.
Imaging methods
For stereomicroscopy imaging, a SteREO Discovery.V8 from Zeiss and Zen2011 Blue Edition was used. In toto cleared X. tropicalis embryos and mouse brains were imaged using mesoSPIM75,77. For all mesoSPIM recordings, fluorophores were excited with the appropriate laser lines and a quadband emission filter (ZET405/488/561/640, Chroma) was used. Imaging was performed using dibenzyl ether as the immersion medium. Two-photon imaging was performed using a custom-built system with a reflective multi-immersion Schmidt objective78. A femtosecond Ti:sapphire laser (Chameleon Ultra II, Coherent) tuned to 980 nm provided excitation. A 720-nm short-pass filter (ET720SP, AHF) placed in front of the photomultiplier tube blocked excitation light and custom interchangeable filter cubes were used to select the GFP emission channel. Live time-lapse imaging for Xenopus embryos was performed on a widefield Thunder imager (Leica) and on a LSM980 Airyscan 2 (Zeiss). For widefield epifluorescence, a Leica DMi8 with a Leica K3M camera using a widefield light-emitting diode was applied for colocalization of Tubb2a and GFP. Stitching was performed with BigStitcher79. Data were rendered using Fiji80, Imaris (Oxford Instruments) or Napari (https://github.com/napari/napari)81. Segmentation was performed with U-Net74,82.
DNA preparation, Sanger sequencing and NGS
Cells, Xenopus embryos or mouse AAV-injected hemispheres were lysed (50 mM Tris pH 8.8, 1 mM EDTA, 0.5% Tween-20, 2 mg ml−1 proteinase K) at 55 °C overnight. After proteinase K inactivation (10-min incubation at 98 °C), PCRs were performed using GoTaq G2 (Promega, M7845), Q5 (NEB, M0491L) or Phusion polymerase (ThermoFisher, F530S) (primers listed in Supplementary Table 1). For sequencing, amplicons were cleaned using nucleoSpin gel and PCR cleanup (Machery-Nagel, 740609) and sent for commercial sequencing (Microsynth). For NGS, amplicons were generated by PCR with appropriate adaptor sequences and commercially sequenced (INVIEW CRISPR Check (size: 450–500 bp, Illumina PE sequencing 2× 300 bp), Eurofins Genomics). Data analysis was performed using CRISPResso2 (ref. 83) and/or custom data processing.
Pythia in silico modeling
The Pythia Python script is designed to simulate CRISPR–Cas-mediated gene-editing efficiencies using a given wild-type and mutant DNA sequence. It iteratively constructs potential editing templates by varying the lengths of the left and right homology arms and uses the inDelphi tool to predict repair outcomes and their frequencies. The results, including the predicted repair outcomes and their corresponding frequencies, are stored and reported to identify the most effective repair template for achieving the desired genomic modification.
We modeled the optimal ssODN repair template length, with the maximal Pythia score, across clinically relevant point mutations in RPE65, involved in retinitis pigmentosa and Leber congenital amaurosis, among others. For this, we obtained all RPE65 ClinVar (accessed at January 6, 2024) single-nucleotide missense variants. For each missense variant, we calculated the minimal number of base changes required to change the codon usage from the human missense variant amino acid toward the restoration of the wild-type amino acid at that location. Next, Pythia code was used to compute the optimal ssODN repair template with the maximal Pythia score to establish this base point mutation, thus reverting the clinically relevant mutation at the amino acid level.
Pythia editing in vitro
Potential ssODN repair templates were designed for three independent GFP gRNAs to establish two point mutations to convert eGFP to eBFP. Pythia scores were calculated with repair arm length set at 1 to 24, both left and right. From these, we performed a binning from 0 to 100 across the scores and randomly selected 30 repair templates for each gRNA, selecting three repair templates per decile bin and, thus, 90 in total. ssODN repair templates were ordered as desalted nonmodified primers from Microsynth (Supplementary Table 6). HEK293T-AAVS1(CMV:eGFP), featuring a stable one-copy integration of a pCMV:eGFP construct, was seeded at a density of 10,000 cells in a 96-well plate in 150 μl of standard DMEM. Then, 24 h later, cells were transfected using Lipofectamine CRISPRMAX (Thermo Fisher, CMAX00003) and Lipofectamine 3000 (Thermo Fisher, L3000001). gRNA was assembled using Alt-R CRISPR–Cas9 IDT crRNA and Alt-R CRISPR–Cas9 tracrRNA, according to the manufacturer’s instructions, by heating it to 95 °C and cooling it to room temperature, yielding a duplex at a final concentration of 1 μM. RNP was assembled by incubation for 5 min at room temperature of 1 μM gRNA duplex, 250 ng of Cas9 protein (Alt-R S.p. Cas9 Nuclease V3, IDT) and 0.6 μl of Cas9 PLUS reagent (from CRISPRMAX kit). Transfection complexes for RNP were generated by incubation at room temperature for 20 min of 25 μl of RNP repair template, 1.2 μl of CRISPRMAX transfection reagent and 23.8 μl of Opti-MEM medium. Transfection complexes for ssODN were generated using Lipofectamine 3000 (Thermo Fisher, L3000001) according to the manufacturer’s instructions. In brief, 1 μl of 20 nmol ml−1 of ssODN repair template was packaged in a final volume of 10 μl. Both RNP transfection (50 µl final per well) and ssODN transfection (10 µl final per well) reagents were added to the 96-well plate. On the next day (approximately 20 h later), the medium was refreshed and cells were split and maintained according to standard HEK293T principles until analysis using flow cytometry at day 18.
Pythia editing in vivo in Xenopus
A gRNA targeting the X. tropicalis gene tyr was designed and the Pythia software was deployed to identify the optimal repair template to generate a double point mutation. Repair template was ordered as desalted ssODN from Microsynth. gRNA was assembled using Alt-R CRISPR–Cas9 IDT as described above for Xenopus. For RNP assembly, 3 μl of Cas9 protein (1 μg μl−1, PNABio CP01) was mixed with 1 μl of gRNA and incubated at 37 °C for 5 min. Next, 1 µl of ssODN repair template (5 µM stock, 1 µM final concentration) was added. Embryos were microinjected in the one-cell stage immediately after cortical rotation, targeting the gray sperm entry point with 5–10 nl of injection mix. Restriction digests of PCR products were performed with BsrDI (NEB, R0574S) overnight at 37 °C in NEB buffer r2.1 and with BtsI-v2 (NEB, R0667S) overnight at 37 °C in NEB rCutSmart.
For viability testing, RNP was assembled as described above and mixed with ssODN templates (35 bp, 48 bp and 66 bp, 1 µM final concentration) and TRITC–dextran (0.5 ng μl−1, Sigma-Aldrich). Then, 16 h after injections, embryos were sorted for TRITC+ fluorescence, followed by live–dead sorting.
To compare Pythia scores to editing outcomes in the generation of a point mutation, the Pythia software was used to identify the optimal repair template for a new locus on the tyr gene. Two repair templates of decreasing length and Pythia score and two templates of increasing length but decreasing Pythia score were designed. Injections were performed as described above but with a fixed DNA concentration of 10.8 ng μl−1 (below the identified toxicity limit). The experiment was split into two injection rounds with a different mating pair for each. The optimal repair template and the two shorter-than-optimal templates or the two longer-than-optimal templates were injected per injected round. After 40 h, the embryos were pooled into groups of 75 per condition and lysed (50 mM Tris pH 8.8, 1 mM EDTA, 0.5% Tween-20 and 2 mg ml−1 proteinase K) at 55 °C overnight, followed by 10 min of heat inactivation at 98 °C. After centrifugation, DNA in the aqueous middle phase was PCR-amplified using Phusion polymerase (NEB, M0530L) with overhang primers containing NGS adaptor sequences (Supplementary Table 1). The product was purified (Macherey Nagel, 740609) and analyzed by NGS (INVIEW CRISPR Check (size: 450–500 bp), Eurofins Genomics) using CRISPR-GRANT84.
Statistics and reproducibility
Statistical analyses are described in detail throughout the manuscript. For Xenopus, stereomicroscopy and mesoSPIM light-sheet imaging were performed on multiple embryos obtained from injected clutches or natural matings. The images shown are representative examples that reflect the consistent expression patterns as observed across positive embryos (efficiencies reported throughout the manuscript) of the same reporter genotype or injection condition. These imaging experiments were designed to qualitatively assess spatial expression of tagged proteins or reporter constructs and no statistical analysis was applied.
For the mouse experiments, mesoSPIM light-sheet imaging was performed on a single injected brain hemisphere. These datasets provide near full-tissue views that are representative of outcomes observed in our injection. Histological immunofluorescence staining of gene-edited mouse brains was performed on serial sections from individual animals. The images presented are representative of expression patterns reproducibly observed across multiple sections and animals processed under the same conditions. These experiments were intended to provide qualitative spatial validation rather than quantitative comparisons.
Immunoprecipitation of endogenously tagged Myh9, Ncam1 and Acta2 in Xenopus was performed once using pooled lysates from 2–5 representative embryos per condition. Western blotting and GFP immunoprecipitation from mouse brain tissue were carried out once, using lysate from a single brain hemisphere of one injected animal per condition. These experiments served as qualitative validations and were not designed for direct statistical comparison.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
