Genome editing that finds and removes DNA sequences to select specific cells

While knock-in of genes or changes of multiple bases can fully disrupt the gRNA recognition site, single-nucleotide changes can also prevent cleavage and are thus suited for the application of SNIPE. This is particularly the case for nucleotide changes located in the protospacer adjacent motif (PAM) or close to the PAM (Fig. 1b). For Cas9, changing a guanine in the 3' NGG PAM sequence, or changing a single position in a core sequence located at +4 to +7 position upstream of the PAM, strongly prevents cleavage (77–100%)³⁰. For Cas12a (Cpf1), this is achieved by changing a thymine in the 5'-TTTN PAM sequence, or by changing a single position located at +2 to +7 position downstream of the PAM (78–100%)³¹. Single nucleotide mismatches at other positions in the gRNA recognition site only moderately prevent cleavage.

We compare SNIPE to state-of-the-art antibiotic selection, which requires the introduction of a drug resistance gene alongside the intended mutation during the editing experiment. In the next step, the cells are then treated with the appropriate antibiotic that allows only cells that have incorporated the DNA donor to survive (Fig. 1a). For this approach, puromycin is the drug of choice in mammalian cells^32–34. We used prime editing (PE) in the 409B2 human induced pluripotent stem cell (hiPSC) line carrying a puromycin resistance gene with a premature stop codon that can be activated by editing of this stop codon (TAG) to a threonine codon (ACG) (Fig. 1c). PE is a widely used genome editing tool, in which a reverse transcriptase uses the nicked strand of the target site as a primer to introduce edits from PE gRNA (pegRNA)³⁵. However, PE efficiency strongly varies among target sites, calling for enrichment of desired editing events.

We stably introduced an inducible PE system carrying a human codon optimized RT linked to iCRISPR–Cas9H840A (iPrime)² in the hiPSC line carrying the inactivated puromycin resistance gene. We then delivered pegRNAs that were designed to revert the stop codon to a threonine codon (Fig. 1d) and carried 5'- and 3'-end phosphorothioate bonds and 2′-OMe residues to prevent degradation³⁶. We achieved 25% correction of the stop codon as assessed by Illumina sequencing of the target site (Fig. 1e). An additional 3% of editing events were due to imperfect PE, intended edits with additional unintended mutations, and indels. We then divided the cells for subsequent selection by SNIPE and by puromycin.

To enrich for edited cells by SNIPE, we electroporated a recombinant high fidelity Cas9 variant (Cas9-HiFi)³⁷ complexed with Cas9 SNIPE-gRNA targeting the sequence with the puromycin stop codon as ribonucleoprotein (RNP) while simultaneously inhibiting NHEJ and MMEJ². This resulted in that 84% of the cells carried the correction of the stop codon (3.2-fold increase) (Fig. 1f). Due to the presence of imperfect PE and indels before selection, SNIPE also increased the proportion of unintended editing events to 10%. We then generated 41 single-cell-derived cellular clones to assess the extent of large deletions or rearrangements after SNIPE and found no evidence for unintended copy number changes of the target locus (Supplementary Fig. 1).

Selection using 1 µg/ml puromycin for 3 days rather than by SNIPE resulted in 87% cells carrying activated puromycin resistance, comparable to SNIPE (Fig. 1g). SNIPE resulted in the death of 27% of unedited cells as estimated by a resazurin assay 3 days after editing. A comparable percentage of cell death of 37% was observed for 1 µg/ml puromycin treatment (Fig. 1h). Stop codon correction could be further increased to 96% by applying 1.5 µg/ml puromycin, but at the cost of very high cell death (96%) due to puromycin toxicity (Fig. 1g, h).

After having established that M3814 and POLQ siRNAmix is the best formulation for SNIPE, we tested additional targets by PE in the genes VEGFA ( + 5G to T) and HEK3 ( + 1T to A) for which pegRNAs have been optimized³⁵. We obtained PE efficiencies of 3% for VEGFA and 8% for HEK3. The outcome purities, i.e., the proportion of editing events that were the intended ones, were 69–80% (Fig. 2a). SNIPE increased PE outcomes to 22% for VEGFA, 17% for HEK3 and 77% for FANCF (mean 4.2-fold) and increased purities to 81–95%. Re-cutting the targets in the edited bulks using Cas9-HiFi RNP without inhibition of the two end-joining pathways moderately increased outcomes for two out of three targets (mean 2-fold), but drastically reduced purity to a mean of 25%. In addition to PE, we also employed base editing using recombinant ABE8e⁴⁰ and could increase base editing efficiency from 23% to 53% at the AHR target site using SNIPE (Supplementary Fig. 3).

To assess how SNIPE performs for heterozygous or homozygous genotypes, we used PE to edit FANCF ( + 5G to T) and isolated single cell-derived clones with the genotypes: WT/WT (wild type), PE/PE, PE/WT, and PE/indel (Supplementary Fig. 4). Subsequent SNIPE targeting the wildtype allele drastically killed both WT/WT (99% cell death) and PE/WT (98%), but hardly killed PE/PE (9%) and PE/indel (0%). Interestingly, SNIPE of the PE/WT clone increased PE sequencing reads from 50% to 75%, suggesting that in the few surviving cells, DSBs introduced by SNIPE were repaired by HR using the PE allele as a template. Consequently, SNIPE enriches for biallelic editing and can remove cells carrying at least one wild type allele from the cell population.

Next, we edited three targets where HDR efficiencies are low: CDH16 (T342A), AHR (V381A), and PRDM10 (N1129T). We used Cas9, gRNA, and exogenous single-stranded (ss) DNA donors to introduce the missense mutations with or without dual end-joining repair pathway inhibition (HDRobust) (Fig. 2b). We also tested SNIPE with GOLD-gRNAs, which we have previously developed to increase cleavage efficiency⁴¹. HDRobust increased HDR efficiency and purity for all targets (from 9–32% to 81–84% purity) relative to standard HDR editing with a DNA donor. SNIPE further increased HDRobust outcomes to 39% for CDH16, 33% for AHR and 49% for PRDM10 and increased purities (89–93%). Efficiencies were even better with GOLD-SNIPE (49–56%). We furthermore tested two targets where HDR efficiencies are high, C3 (A1286V) and DCHS1 (D777N). We achieved 84% and 94% HDR when using HDRobust followed by GOLD-SNIPE (Fig. 2c).

As expected, SNIPE is compatible with Cas12a editing since we could enrich desired edits after HDRobust from 20–59% to 43–91% for editing of NCOA6 (I823M), IZUMO4 (R185r.185_197del), and GLDC (L220F) with Cpf1-Ultra⁴² RNP (Fig. 2d). We also used SNIPE in a chimpanzee iPSC line, using M3814 to inhibit NHEJ and a siRNA designed to target the chimpanzee POLQ to inhibit MMEJ. We achieved 80% precise editing of DCHS1 with SNIPE (Fig. 2e).

Across the tested targets, SNIPE increases the median editing efficiency 7-fold compared to normal editing and, surprisingly, also increases outcome purity (Fig. 2f). The increase in purity is likely due to that SNIPE can target editing-induced indels that do not destroy the gRNA recognition site. Indeed, SNIPE often alters the indel pattern generated in the initial editing experiment (Supplementary Fig. 5).

To also analyze cellular clones after SNIPE in chimpanzee cells, we generated 24 single-cell-derived clones from the chimpanzee DCHS1 edit and quantified the number of homozygous and heterozygous edits. 79% of the precisely edited clones were homozygous, indicating a preferential selection of biallelic intended edits by SNIPE (Supplementary Fig. 6). Four clones (17%) have co-converted a SNP 52 bp downstream from the intended edit, but none showed copy number loss. To assess off-target editing, we scored the editing efficiency of the two most likely predicted off-target sites for each of the targets for which Cas9 and Cas12a were used in this study^43,44. Mean off-target editing with standard editing or SNIPE was 2.7% (0–47.2%) or 0.2% (0–2.1%), respectively. The CDH16 edit, which stood out with 47.2% combined off-targets with standard editing, was reduced to 2.1% with SNIPE (22-fold) (Supplementary Figs. 7 and 8).

Motivated by the potential of precision editing and subsequent SNIPE to change many positions in the same cell, we decided to introduce a combination of five missense mutations in five genes back to the ancestral state not observed to date in any present-day human but observed in all Neandertal and Denisovan genomes available^45,46. These changes have thus occurred and reached very high frequencies during the past half million years of human evolution. To avoid DSBs and genomic instability, we used PE and designed pegRNAs and respective nicking gRNAs, which only rarely introduce DSBs. After expression of the prime editor in the 409B2 iPrime stem cell line, we introduced the five pegRNA/nicking gRNA pairs by electroporation. This resulted in editing efficiencies of 5–12% for IZUMO4 (R185r.185_197del), CDH16 (T342A), and PROM2 (D458E), while no editing was observed for AHR (V381A) and NCOA6 (I823M) (<0.02%) (Fig. 3b). Three additional sequential PE rounds for the former three genes increased the percentage of edited chromosomes in the mixed cell population to 24% for IZUMO4, 30% for CDH16, and 57% for PROM2 (Fig. 3c). Subsequent single HDRobust edits for AHR and NCOA6 on this mixed cell population resulted in precise editing of 45% and 26%, respectively. We then applied SNIPE of one target, two targets, and three targets of the three genes with the lowest editing frequencies of precise edits to increase the percentage of cells precisely edited for all genes (Fig. 3d). After these experiments and assuming that the edits occur independently of each other, the best cell pool had an estimated 10% of cells carrying all five edits (Fig. 3e). From this pool, we sorted single cells to generate clones. Of the 116 cellular clones, 25 (22%) carried the desired five edits on all 10 chromosomes (Fig. 3f). We then used dd PCR to estimate the copy numbers of the five edited targets in the cellular clones. Four clones had an apparent copy number loss in the IZUMO4 target, and of these clones, one had an additional copy number loss in NCOA6 while the other six targets had two copies of all five targets (Fig. 3g). In addition, a previously heterozygous SNP 21 kb downstream of the IZUMO4 target had become homozygous in all ten clones, suggesting it had been converted by the sister chromatid in the process of editing and SNIPE. One out of ten clones also became homozygous for positions flanking the NCOA6 target site, while no loss of heterozygosity was observed for SNPs in the three other targets (Supplementary Fig. 9).

We then applied SNIPE to generate 19 cell lines carrying cDNA expression constructs encoding transcription factors (TF) or TF combinations described to be able to directly generate various cell types^50,51 (Fig. 4d). The TF inserts ranged in size from 0.9 to 4.5 kb. The mean knock-in efficiency after SNIPE, determined by PCR of the target loci from single cell-derived clones across the TF inserts, was 70% (s.e.m. 6.5%). For five transcription factors, all of the generated clones carried the desired knock-in. Across all tested knock-in edits, SNIPE increased median efficiency 3.4-fold compared to HDRobust (Fig. 4e). This strong enrichment can easily be seen when comparing the intensity of PCR products from cell populations before cloning that had not been edited, edited with HDRobust, and enriched with SNIPE, respectively (Fig. 4f).

We furthermore generated a hiPCS line that can express both Cas9 and MAD7 (ErCas12a)^52,53 in order to be able to edit genomic sites adjacent to both NGG and TTTN PAMs⁵⁴. Introduction of MAD7 in a Cas9-containing cell line by HDRobust followed by SNIPE resulted in 95% of single cell-derived clones carrying a 4 kb DNA fragment encoding MAD7 (Fig. 4g). We confirmed that both Cas9 and MAD7 function in these cells by using gRNAs designed for two different targets and observing ≥30% indels when Cas9 and MAD7 were used (Fig. 4h). We also used induced MAD7 to edit a site in the gene KNL1 (S1086G) followed by subsequent SNIPE with MAD7. Standard editing with a ssDNA donor resulted in 8% precise editing by HDR. Editing with HDRobust and subsequent SNIPE increased efficiency to 20% and to 38%, respectively (Fig. 4i).

We tested the ability of SNIPE to kill chronic myeloid leukemia (CML) K562 cells that carry a Philadelphia chromosome, i.e., a reciprocal translocation of chromosome 9 and 22 that results in the fusion oncogene BCR-ABL1^56,57 (Fig. 5e). This translocation generates a novel gRNA recognition site in K562 cells that is not present in normal cells (Fig. 5f). We targeted the BCR-ABL1 junction in K562 cells with SNIPE, and used THP-1 CML cells, where the translocation is absent, as a control. SNIPE did not kill THP-1 cells, but killed approximately half of the K562 cells (Fig. 5g). CRISPR-targeting the BCR-ABL1 locus without SNIPE killed 16% of cells, which is consistent with the toxicity resulting from the creation of multiple DNA double-strand breaks when targeting BCR-ABL1 in K562 cells⁵⁸. We used PCR amplification by primer pairs that amplify either 297 bp of healthy chromosome 22 or 267 bp of the Philadelphia chromosome to gauge the proportion of cells carrying the Philadelphia chromosome. This suggests a reduction of BCR-ABL1 containing chromosomes from 66% to 13% among the surviving K562 cells, which is comparable to the reduction seen with standard CRISPR-targeting (Fig. 5h, i).

Finally, we investigated the impact of SNIPE on cell survival and genotype after targeting disease alleles in the heterozygous cancer cell lines RD-ES (p53 R273C), HT-29 (APC E853*), and HuCC-T1 (KRAS G12D). Only 2–17% cells died after SNIPE (Fig. 5j) while surprisingly disease mutations were reduced from 41% to 5% for RD-ES, 33% to 8% for HT-29 and 65% to 15% for HuCC-T1 and wild type alleles increased from 58% to 90%, 65% to 84% and 33% to 56%, respectively (Fig. 5k). Similar to our observations from Supplementary Fig. 4, we hypothesize that the DSBs introduced in the disease alleles by SNIPE were repaired by HR using the wild type chromosome as a template, resulting in the creation of cells without cancer mutations.

We used a 409B2 human induced pluripotent stem cell line (Riken BioResource Center, catalog no. HPS0076, GMO permit AZ 54-8452/26), as well as a variant carrying an iCRISPR-Cas9⁹⁴. Furthermore, we modified iCRISPR 409B2 to carry a reverse transcriptase adjacent to Cas9H840A³⁵ (iPrime) as described in Riesenberg and Maricic 2018⁹⁴. The chimpanzee stem cell line Sandra A was generated in a previous study by reprogramming of isolated leukocytes⁹⁵. Cells were cultured at 37 °C and 5% CO2 in mTeSR1 medium (StemCell Technologies, catalog no. 05851) with supplement (StemCell Technologies, catalog no. 05852) on Matrigel Matrix (Corning, catalog no. 35248). Medium was changed daily. Cells were passaged 1:5 to 1:10 when they reached ∼80% confluence by dissociation with EDTA (VWR, catalog no. 437012 C) for 6 min at room temperature and plated in medium containing 10 μM Rho-associated protein kinase (ROCK) inhibitor Y-27632 (Calbiochem, catalog no. 688000) for 24 h. For cryopreservation, detached cells were resuspended in mFresR (StemCell Technologies, catalog no. 5854), gradually frozen to –80 °C in an isopropanol box and stored in liquid nitrogen. For the generation of cellular clones, stem cells were treated with TrypLE (Gibco, catalog no. 12605010) for 5 min at 37 °C and single cells were sorted by a single cell sorter (Cytena Single-Cell Printer) into 96-well plates containing conditioned StemFlex medium (Gibco, catalog no. A3349401) supplemented with 10% CloneR (StemCell Technologies, catalog no. 05888). After 10–14 days, colonies could be passaged and genotyped.

Human colorectal adenocarcinoma cells HT-29 (Cytion, catalog no. 300215) were cultured in EMEM (Sigma-Aldrich, catalog no. M4655) supplemented with 10% FBS (Gibco, catalog no. 10270-106) and human CML cells K562 (ECACC, catalog no. 89121407) were cultured in IMDM (Gibco, catalog no. 12440053) with 10% FBS. Human intrahepatic cholangiocarcinoma cells HuCC-T1 (Cytion, catalog no. 300469), the human bone sarcoma cell line RD-ES (Cytion, catalog no. 300410) and the human leukemia cell line THP-1 (Cytion, catalog no. 300356) were cultured in RPMI 1640 (ThermoFisher, catalog no. 11875-093) with 10% FBS. Medium was refreshed every 2–3 days, and cells were passaged using Accutase (Sigma, catalog no. A6964) once per week in a 1:3 to 1:8 ratio. For cryopreservation, medium with 10% DMSO (Life Technologies, catalog no. D12345) was used.

All cell lines were tested negative for mycoplasma contamination and authenticated by the supplier via certificate of analysis as well as in-house morphology check.

All gRNAs, DNA donors and primers were ordered from Integrated DNA Technologies (see Supplementary Data: List of oligonucleotides). Alt-R CRISPR-Cas9 tracrRNA was purchased as well from Integrated DNA Technologies (catalog no. 1072534). Small molecules M3814 (catalog no. HY-101570), AZD7648 (catalog no. HY-111783) and PolQi2 (catalog no. HY-150279) were ordered from MedChemExpress.

409B2 iPrime cells were treated with 2 μg ml⁻¹ doxycycline (Clontech, catalog no. 631311) 3 days prior to editing to express the prime editor. For electroporation, single cell suspensions were obtained by treatment of adherent cells with TrypLE (Gibco, catalog no. 12605010) for 5 min at 37 °C and trituration. Preheated medium was added before cells were counted using the Countess Automated Cell Counter (Invitrogen). One million cells per electroporation were added into 5 ml PBS (Gibco, catalog no. 14190250) and centrifuged at 200 × g for 5 min at room temperature. For all cells except the 409B2 iPrime cell line, Streptococcus pyogenes Cas9-HiFi (R691A) (IDT, catalog no. 1081061), Acidaminococcus sp. BV3L6 Cas12a (Cpf1-Ultra) (IDT, catalog no. 10001273), or the Streptococcus pyogenes Cas9 adenine base editor ABE8e (Protein Production Sweden PPS) were used. Single cell suspensions were transferred into cuvettes containing 252 pmol CRISPR enzyme if applicable, 100 μl Human Stem Cell nucleofection buffer (Lonza, catalog no. VVPH-5022), 100 pmol electroporation enhancer (IDT, catalog no. 1075916 or 1076301), 320 pmol gRNA (crRNA/tracR duplex for Cas9 and crRNA for Cas12a) (or 640 pmol pegRNA and 214 pmol nicking gRNA for prime editing) and 200 pmol of single-stranded DNA donor or 10 μg of ds DNA donor if applicable. For edits with HDRobust, 320 pmol of POLQ siRNA - SMART 10721 and 640 pmol of POLQ siRNA 765 were added into the cuvette and 2 µM M3814 was added to the medium after nucleofection for 2 days. Where applicable, 3 µM of PolQi2 and 1 µM of AZD7648 were added³⁸. For all cell types and edits, program B-16 of the Nucleofector 2b device (Lonza) was used for the electroporation and ROCK inhibitor Y-27632 was added for the first 24 h.

At least 3 days after editing when cells reached confluency they were dissociated using EDTA (VWR, catalog no. 437012 C), pelleted and resuspended in 30–50 μl QuickExtract DNA extraction solution (Lucigen, catalog no. QE09050). Single-stranded DNA was extracted from the cells by incubation at 65 °C for 10 min, 68 °C for 5 min and 98 °C for 5 min. Target PCRs were performed in a total volume of 25 μl using the KAPA2G Robust PCR Kit (Sigma, catalog no. KK5024) with supplied buffer B and 3 μl of cell extract and incubated in a T100 Thermal Cycler (Bio-Rad) at 95 °C 3 min; 34× (95 °C 15 s, 65 °C 15 s, 72 °C 15 s); 72 °C 60 s. In a second PCR, Illumina adapters (P5 and P7) with sample-specific indices were added using Phusion HF MasterMix (Thermo Scientific, catalog no. F-531L) and 0.3 μl of the first PCR product. After cycling at 98 °C 30 s; 25× (98 °C 10 s, 58 °C, 10 s, 72 °C 20 s); 72 °C 5 min, amplicons were checked on a 2% EX agarose gel (Invitrogen, catalog no. G4010–11), pooled and purified using solid phase reversible immobilization beads in a 1:1 ratio of beads to PCR solution⁹⁶. Libraries were sequenced as paired-end sequences of 2 × 151 bp (+7 bp index) on a MiSeq (Illumina) and base called with Bustard (Illumina). Adapter trimming was done using leeHom⁹⁷.

After sequencing, demultiplexing and conversion of bam files into fastq files was done using SAMtools v.1.12⁹⁸. Sequencing read percentage of targeted nucleotide substitution (HDR or PE for prime editing), indels (NHEJ and MMEJ) and mix of both (imperfect HDR or imperfect PE for prime editing) were calculated on the fastq files using CRISPResso1⁹⁹. The analysis was restricted to amplicons with a minimum of 70% similarity to the wild type sequence and to a window of 20 bp from each gRNA. Substitutions were ignored to avoid false characterization of sequencing errors as NHEJ and sequence similarity for HDR occurrence was set to 95%.

To measure viability, a resazurin assay was performed. The amount of living cells can be determined by fluorescent measurements as cellular dehydrogenases convert resazurin into fluorescent resofurin (excitation: 530–570 nm, emission: 590–620 nm). Cells were cultured for 3 days after editing, then fresh medium containing 10% resazurin solution (Cell Signaling, catalog no. 11884) was added. After 2 h, fluorescence was measured using a CLARIOstar imager (BMG Labtech). To detect background fluorescence, empty wells were filled with medium containing resazurin.

To genotype cellular clones of large knock-ins, the inserts were amplified with the KAPA2G Robust PCR Kit (Sigma, catalog no. KK5024) with supplied buffer B and 3 μl of cell extract. The cycling protocol for the PCR was: 95 °C 3 min; 34× (95 °C 15 s, 65 °C 15 s, 72 °C 5 min); 72 °C 60 s. Afterwards, the PCR products were analyzed in an automated parallel capillary electrophoresis using the Fragment Analyzer PROSize 2.0 (Agilent) with the dsDNA 920 Reagent Kit (Agilent, catalog no. DNF-920). The wild type cell line was taken along as a control. Successful knock-ins were defined if a peak with the expected size could be detected.

A quantitative droplet digital PCR was performed to measure the copy number of single-cell-derived cellular colonies. Primers were designed to flank the cut site while the probe coupled to FAM was designed to exclude the edited site. FOXP2 coupled to HEX was used as a reference gene. For the reaction, 1× ddPCR Supermix for probes (no dUTP, Bio-Rad, catalog no. 1863024) was mixed with 0.2 μM primers and 0.2 μM probe for target and reference and with 1 μl genomic DNA in QuickExtract DNA extraction solution (Lucigen, catalog no. QE090500). After droplet generation, the following cycling protocol was run: 5 min at 95 °C, 40 cycles 30 s at 95 °C (ramp rate 2 °C s⁻¹), 60 s at 61 °C for DCHS1 and 58 °C for the puromycin resistance gene, AHR, CDH16, IZUMO4, NCOA6 and PROM2 (ramp rate 0.4 °C s⁻¹) and 5 min at 98 °C. Afterwards, droplets were analyzed in a QX200 Droplet reader (Bio-Rad). For Supplementary Fig. 2, the QIAcuity Digital PCR System (Qiagen) was used according to the manufacturer's instructions with the following cycling protocol: 2 min at 95 °C, 40 cycles 15 s at 95 °C and 30 s at 58 °C. Copy numbers were determined by the ratio of fluorescent signal between the target and the reference gene.

To analyze off-targets of gRNAs used for SNIPE, CRISPOR⁴⁴ was used to predict the top two off-targets for every gRNA based on the CFD off-target score for Cas9 Hifi gRNAs. For Cpf1 gRNAs, due to the absence of a score, the top two candidates of the list of off-targets were chosen. Primers were designed for both off-targets per gRNA, and the loci were sequenced for all edited bulks following the Illumina library preparation and sequencing as well as analysis part of the methods section.

Bar graphs in figures were plotted and s.e.m. error bars were calculated using GraphPad Prism 10 software. The number of replicates is stated in the respective figure legends. No statistical method was used to predetermine sample size. The experiments were not randomized. Samples were prepared unblinded but in parallel. Analysis was performed on the basis of numerical sample names, without the identity of the samples being known during the analysis.

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