Highly efficient generation of isogenic pluripotent stem cell models using prime editing

The recent development of prime editing (PE) genome engineering technologies has the potential to significantly simplify the generation of human pluripotent stem cell (hPSC)-based disease models. PE is a multicomponent editing system that uses a Cas9-nickase fused to a reverse transcriptase (nCas9-RT) and an extended PE guide RNA (pegRNA). Once reverse transcribed, the pegRNA extension functions as a repair template to introduce precise designer mutations at the target site. Here, we systematically compared the editing efficiencies of PE to conventional gene editing methods in hPSCs. This analysis revealed that PE is overall more efficient and precise than homology-directed repair of site-specific nuclease-induced double-strand breaks. Specifically, PE is more effective in generating heterozygous editing events to create autosomal dominant disease-associated mutations. By stably integrating the nCas9-RT into hPSCs we achieved editing efficiencies equal to those reported for cancer cells, suggesting that the expression of the PE components, rather than cell-intrinsic features, limit PE in hPSCs. To improve the efficiency of PE in hPSCs, we optimized the delivery modalities for the PE components. Delivery of the nCas9-RT as mRNA combined with synthetically generated, chemically-modified pegRNAs and nicking guide RNAs improved editing efficiencies up to 13-fold compared with transfecting the PE components as plasmids or ribonucleoprotein particles. Finally, we demonstrated that this mRNA-based delivery approach can be used repeatedly to yield editing efficiencies exceeding 60% and to correct or introduce familial mutations causing Parkinson’s disease in hPSCs.


SUMMARY
The recent development of prime editing (PE) genome engineering technologies has the potential to significantly simplify the generation of human pluripotent stem cell (hPSC)-based disease models. PE is a multi-component editing system that uses a Cas9nickase fused to a reverse transcriptase (nCas9-RT) and an extended PE guide RNA (pegRNA). Once reverse transcribed, the pegRNA extension functions as a repair template to introduce precise designer mutations at the target site. Here, we systematically compared the editing efficiencies of PE to conventional gene editing methods in hPSCs. This analysis revealed that PE is overall more efficient and precise than homology-directed repair (HDR) of site-specific nuclease-induced double-strand breaks (DSBs). Specifically, PE is more effective in generating heterozygous editing events to create autosomal dominant disease-associated mutations. By stably integrating the nCas9-RT into hPSCs we achieved editing efficiencies equal to those reported for cancer cells, suggesting that the expression of the PE components, rather than cellintrinsic features, limit PE in hPSCs. To improve the efficiency of PE in hPSCs, we optimized the delivery modalities for the PE components. Delivery of the nCas9-RT as mRNA combined with synthetically generated chemically-modified pegRNAs and nicking guide RNAs (ngRNAs) improved editing efficiencies up to 13-fold compared to transfecting the prime editing components as plasmids or ribonucleoprotein particles (RNPs). Finally, we demonstrated that this mRNA-based delivery approach can be used repeatedly to yield editing efficiencies exceeding 60% and to correct or introduce familial mutations causing Parkinson's disease in hPSCs.

INTRODUCTION
One of the current challenges of using hPSCs to model human diseases is to precisely and efficiently engineer the genome to introduce designer mutations (Hockemeyer and Jaenisch, 2016;Soldner and Jaenisch, 2018). Currently, the predominant approach in hPSCs is to induce targeted double-strand DNA breaks (DSBs) using highly active site-specific nucleases, such as the CRISPR/Cas9 system (Cong et al., 2013;Ding et al., 2013;Jinek et al., 2012;Jinek et al., 2013;Mali et al., 2013) or protein engineering platforms including zinc finger nucleases (ZFNs) (Hockemeyer et al., 2009;Soldner et al., 2011;Zou et al., 2009) and transcription activator-like effector nucleases (TALEN) (Boch et al., 2009;Cermak et al., 2011;Hockemeyer et al., 2011). Such targeted DSBs have been shown to substantially increase genome editing efficiency over conventional homologous recombination. However, since nuclease-induced DSBs are in most context preferentially repaired by non-homologous end joining (NHEJ) compared to homology-directed repair (HDR) mechanisms, DSB-mediated genome editing frequently generates undesirable compound heterozygous editing outcomes with one correctly targeted allele and an insertion or deletion (indel) on the second allele, causing the disruption of the protein coding sequence (Cox et al., 2015). Therefore, it has been challenging to generate disease-associated dominant mutations in a heterozygous setting. By contrast, PE has been shown to overcome this limitation, as it does not require a DSB but directly repairs a nicked DNA strand (Anzalone et al., 2019). PE is a multicomponent editing system composed of a Cas9-nickase fused to a reverse transcriptase (nCas9-RT) and an extended prime editing guide RNA (pegRNA) that is reverse transcribed and functions as a repair template at the target site (Anzalone et al., 2019).
Here, we systematically compare different genome editing methods and show that PE is overall more efficient and precise to introduce heterozygous point mutations into hPSCs.
Furthermore, by optimizing the delivery modality of the PE components, we were able to establish a highly efficient genome editing platform for hPSCs. By comparing plasmid, RNA-protein RNPs and in vitro transcribed mRNA delivery, we found that nucleofecting nCas9-RT as mRNA combined with synthetically generated and chemically-modified pegRNAs yielded editing efficiencies exceeding 60%, which is comparable to efficiencies observed in tumor cell lines (Anzalone et al., 2019;Nelson et al., 2021). Together, these data indicate that PE has the potential to greatly facilitate the generation of diseasespecific hPSC models.

RESULTS
To evaluate the potential use of PE to genetically modify hPSCs, we directly compared editing outcomes of PE to established CRISPR/Cas9 and TALEN targeting approaches with the goal of introducing disease-relevant point mutations. Initially, we chose to target the Leucine Rich Repeat Kinase 2 (LRRK2) gene to introduce the G2019S (G6055A) mutation (Gilks et al., 2005), which is one of the most frequent pathogenic substitutions linked to Parkinson's disease (PD). This mutation is found in approximately 4% of dominantly inherited familial PD cases, in both heterozygous and homozygous forms and around 1% of sporadic PD cases (Healy et al., 2008). To introduce the G2019S (G6055A) mutation into hPSCs, we generated plasmid-based CRISPR/Cas9, TALEN and PE reagents (without [PE2] or with secondary ngRNA [PE3]) using previously established optimized design and targeting procedures ( Figure 1A) (Anzalone et al., 2019;Hockemeyer et al., 2011;Hsu et al., 2021;Soldner et al., 2011;Soldner et al., 2016).
Briefly, we co-electroporated the human embryonic stem cell (hESC) line WIBR3 with the respective genome engineering components and an enhanced green fluorescent protein (EGFP)-expressing plasmid to allow for the enrichment of transfected cells by fluorescence activated cell sorting (FACS). Genome editing outcomes were evaluated in the transfected bulk cell population (EGFP-positive) 72 hours after electroporation and in single cell-derived subclones. Next generation sequencing (NGS) of amplicons spanning the G2019S (G6055A) region indicated that 2% to 4% of alleles carried the correct G2019S (G6055A) substitution ( Figures 1B and S1A). While we observed roughly comparable editing efficiencies using CRISPR/Cas9, TALEN and PE (with secondary ngRNA [PE3]) gene targeting, all HDR-based approaches generated a significantly higher number of undesired editing outcomes with 19.6% and 3.3% indels for CRISPR/Cas9 and TALEN, respectively, compared to less than 0.5% for PE ( Figure 1B and S1A).
Genotyping of expanded single cell-derived clones showed a higher efficiency in generating heterozygous correctly targeted cell lines using PE primarily due to a substantial higher number of compound heterozygous editing outcomes for CRISPR/Cas9 and TALEN targeting with the correctly inserted G6055A sequence variant on one allele and indels on the second allele ( Figure 1C, as identified by RFLP and Sanger sequencing analysis). Together, these data indicate that PE is overall more efficient and substantially more precise in generating heterozygous mutations in hPSCs, as compared to traditional DSB-based genome engineering approaches.
To scale the PE-based genome editing approach and streamline the derivation of correctly modified single cell clones, we applied a recently established genome editing platform, which employs multiplex low cell number nucleofection, limited dilution, and NGS-dependent genotyping instead of the time consuming and laborious FACS sorting and manual single cell expansion steps to isolate correctly edited hPSC lines (see Figure   S2 and Experimental Procedures for details). While this approach results in slightly lower overall bulk editing efficiencies (as determined by NGS), most likely due to the lack of FACS-based enrichment of transfected cells, the substantially reduced number of cells required and the streamlined workflow allows for highly efficient, multiplexed generation of genome-edited hPSC lines in parallel in less than 4 weeks ( Figure S2). Importantly for this work, the omission of FACS enrichment allowed us to systematically and simultaneously compare a larger number of delivery modalities, as described below.
To confirm the feasibility of using PE to introduce point mutations efficiently and robustly into hPSCs, we tested PE at additional genomic loci. We were able to introduce mutations into the previously published and commonly targeted HEK3 (CTT insertion) locus (Anzalone et al., 2019), as well as two additional PD-associated mutations into the SNCA (a-Synuclein) locus (A30P [G88C] (Krüger et al., 1998) and A53T [G209A] (Polymeropoulos et al., 1997); Figure S1B,C) with editing efficiencies comparable to the LRRK2 locus ( Figure 1D,E; quantified as pure prime editing efficiencies [PPE] as defined in Petri et al. (2021)). As described for LRRK2, the analysis of single cell-expanded clones revealed the efficient generation of heterozygous and homozygous hPSC lines carrying the dominant A30P mutation in the SNCA gene ( Figure S3A,B). Importantly, representative cytogenetic analysis of single cell-expanded LRRK2 (G2019S) and SNCA (A30P) clones showed normal karyotypes for 7 out of 7 tested cell lines. Together, these experiments demonstrate that PE can be used to robustly and efficiently introduce disease-associated mutations into hPSCs to generate isogenic disease models.
During these experiments, we noted that editing outcomes for both the PE2 and PE3 approaches appeared considerably lower than what was previously reported for a variety of human tumor cell lines (Anzalone et al., 2019;Nelson et al., 2021). Indeed, we found that plasmid-based targeting of the HEK3 (CTT insertion) locus resulted in only ~4.3% PPE in WIBR3 hESCs compared to ~12.7% PPE in HEK293T tumor cells using the PE3 strategy ( Figure 1E). Similar differences in gene editing efficiencies between hPSCs, primary cells and tumor cell lines have been commonly observed for other genome engineering approaches including CRISPR/Cas9 targeting (Bowden et al., 2020;Haapaniemi et al., 2018;Ihry et al., 2018). It remains unclear if this difference is the result of cell-intrinsic factors that restrict genome editing specifically in hPSCs or whether the low efficiency is a consequence of insufficient delivery of the PE components.
To test whether PE efficiencies could be increased by optimized delivery of the prime editor, we stably expressed the nCas9-RT protein (PE2 version of the prime editor [PE3] secondary ngRNA) resulted in editing efficiencies up to 12% and 27%, respectively ( Figure 2D). While these data do not exclude fundamental biological differences in the PE process between hPSCs and other cell types, these experiments demonstrate that the method of delivery of the PE components has a significant role in dictating the genome editing efficiency in hPSCs, which can be comparable to the efficiencies observed in tumor cell lines and primary cells.
To improve PE efficiencies using transient delivery of the PE components, we set out to optimize PE delivery conditions. Initially, we focused on delivering the PE components as RNPs, a highly efficient approach described for CRISPR/Cas9-mediated genome editing (Zuris et al., 2015), which was recently successfully adapted for PE in zebrafish and human primary T cells . Using recombinant nCas9-RT protein (PE2 version of the prime editor protein as described in Anzalone et al. [2019] purified from bacteria) ( Figure S4A) and the previously established protocols for RNPbased CRISPR/Cas9 editing Zuris et al., 2015), we nucleofected preassembled RNPs containing the recombinant nCas9-RT protein and chemically-modified synthetic pegRNAs (without [PE2] or with [PE3] secondary ngRNA) targeting the HEK3 (CTT insertion), LRRK2 (G2019S) and SNCA (A30P) loci. Consistent with previous reports in other cell types, we observed RNP-mediated editing outcomes in hPSCs with locus-dependent efficiencies between 1% and 6% ( Figure 2E and S4B,C). While these data clearly indicate the feasibility of RNP-based PE in hPSCs, the observed efficiencies are comparable to the plasmid-based approach and far below the efficiencies observed with stable expression of nCas9-RT from the AAVS1 locus. To exclude that RNP-based PE efficiencies were concentration-or batch-dependent, we repeated some of these experiments with protein from independently purified nCas9-RT batches ( Figure S4D) and used higher protein concentrations ( Figure S4E). However, none of these conditions resulted in substantially improved RNP-based editing efficiencies in hPSCs ( Figure   S4D,E).
An alternative approach, allowing highly efficient delivery of Cas9 for CRISPR/Cas9-based genome editing (Chang et al., 2013;Hwang et al., 2013;Wang et al., 2013), is to deliver the prime editor using in vitro transcribed mRNA Surun et al., 2020). To systematically compare plasmid-, RNP-and mRNA-based PE at Bulk NGS revealed that the combination of in vitro transcribed mRNA-based delivery of the nCas9-RT with chemically-modified synthetic pegRNAs and ngRNAs consistently increased editing efficiencies across all three tested loci up to 13-fold compared to plasmid-and 8-fold compared to RNP-delivery ( Figure 3A). Using this optimized in vitro transcribed mRNA-based delivery approach allowed us to achieve editing efficiencies up to 26.7% (for the SNCA locus). Surprisingly, when combined with secondary nicking of the non-edited strand (PE3), these editing efficiencies were even higher than using the stably nCas9-RT expressing cell lines and comparable to efficiencies commonly observed in human tumor cell lines (Anzalone et al., 2019;Nelson et al., 2021). While mRNA-based editing efficiencies seem to depend on the approach used to in vitro transcribe the nCas9-RT mRNA ( Figure S5A), we found that mRNA-based editing efficiencies are highly consistent across different nCas9-RT mRNA batches when using the best in vitro transcription conditions ( Figure S5B). Furthermore, we observed that using singlestranded mRNA compared to either RNPs or double-stranded plasmid DNAs resulted in improved overall health and increased survival of single cells as indicated by increased clonal survival following nucleofection ( Figure 3B). To test if the high efficiency of mRNAbased PE is a unique feature of the WIBR3 hESCs, we repeated some key experiments in a second human-induced pluripotent stem cell (hiPSC) line 8858, (Pasca et al., 2015) by targeting the LRRK2 (G2019S) and SNCA (A30P) loci and found comparable editing efficiencies ( Figure 3C). Importantly, we were able to establish single cell-derived clones carrying the correct SNCA (A30P) and LRRK2 (G2019S) mutations with high efficiency ( Figure 3D). Considering that potential therapeutic applications would require precise genome-editing of hPSCs in xeno-free conditions, we were able to show efficient and robust PE of the LRRK2 (G2019S) locus in WIBR3 hESCs using several commonly used feeder-free culture systems with comparably high PE efficiencies ( Figure S5C).
A major limitation of classical CRISPR/Cas9 targeting remains the high number and complexity of undesirable editing outcomes (indels). These alleles are resistant to targeting with the same reagents and thus limit the overall HR-editing efficiencies in the context of continued editing or retargeting. Given the much-reduced occurrence of indelcontaining alleles in mRNA-based PE, we hypothesized that this approach might allow efficient retargeting of the same locus. Indeed, we find for all tested loci (HEK3, LRRK2, SNCA) that additional rounds of mRNA-based prime editing of the same cell population could substatially increase overall editing efficiencies ( Figure 4A). This indicates that multiple rounds of mRNA-based PE could result in precise and nearly complete editing of a bulk hPSC population without any type of selection.
The data presented thus far describes the insertion of disease-associated mutations into a wild-type genetic background. To test whether mRNA-based PE can be used to correct disease-causing mutations, we designed pegRNAs to specifically target only the mutated SNCA (A30P) allele to revert this mutation back to wild-type in hPSCs.
When targeting a heterozygous SNCA (A30P) hESC line, bulk NGS indicated the correction of 31.0% of the mutated A30P alleles ( Figure 4B). Subsequent genotyping of single-cell derived clones indicated 26.2% precisely corrected clones without additional undesired modifications (indels) of the wild-type allele ( Figure 4C). Taken together, our data indicate that in vitro transcribed mRNA-based PE is a highly efficient gene editing approach in hPSCs that has the potential to greatly facilitate the generation of diseasespecific hPSC models.

DISCUSSION
The experiments performed here provide a detailed experimental road map for how to implement PE towards genome engineering of hPSCs. We show that mRNA transfection of the prime editor component (nCas9-RT) paired with the transfection of chemicallymodified guide RNAs is well tolerated and highly effective for introducing precise designer mutations in hiPSCs and hESCs. Considering that mRNA-based PE does not require specialized molecular or biochemical skills and consistently achieves high editing efficiency in hESC and hiPSC lines, we predict that this approach has the potential to greatly facilitate the generation of disease-specific hPSC models and will be widely adopted by researchers.
During the process of establishing this workflow, we made several key observations. We find that PE can be as efficient in hPSCs as has been reported for cancer cells (Anzalone et al., 2019;Nelson et al., 2021). We demonstrate that this approach efficiently allows to introduce or correct heterozygous disease-related mutations in hPSCs with base pair precision and without introducing undesired additional modifications on the second allele. The resulting cells showed a normal karyotype, consistent with low genotoxicity of PE due to the lack of DSBs (Anzalone et al., 2019). In the past, generating such isogenic sets cells that differ exclusively at individual diseasecausing sequence variants was highly laborious and an experimental bottleneck. Here we overcome this challenge by deploying PE via optimized delivery methods. We demonstrate that hPSCs can be subjected to several rounds of PE, eventually yielding up to 60% correctly targeted alleles. Importantly, PE efficiencies might be further increased by including mRNAs coding for DNA mismatch repair inhibiting proteins, a novel approach that has been recently shown to significantly improve the prime editing platform . These very high editing efficiencies without the need for selection of enrichment of targeted clones provide an intriguing platform to develop more robust in vitro disease models and potential therapeutic applications of PE in hPSCs or differentiated cell types.

Limitations of the study
In our study we successfully introduced three out of three familial PD point mutations into hPSCs using previously established algorithms to design peRNAs . In each case a classical protospacer adjacent motif (PAM) was present close to the intended amino acid substitution. Our work did not explore more complex or challenging genetic modifications, however we expect that systematic approaches that establish optimized design parameters for prime editing, as recently described for cancer cells (Kim et al., 2020;Nelson et al., 2021) and the development of Cas9 variants with non-classical PAMs (Chatterjee et al., 2020;Kleinstiver et al., 2015;Miller et al., 2020) combined with the optimized protocols reported here will allow PE to become a general method of choice for genome editing in hPSCs.

EXPERIMENTAL PROCEDURES hPSCs culture
All hESC and hiPSC lines were routinely maintained on irradiated or mitomycin Cinactivated mouse embryonic fibroblast (MEF) feeder layers as described previously (Soldner et al., 2016 Scientific]. Cells were collected for genomic DNA extraction and NGS-based allele quantification 3 days post transfection.

nCas9-RT protein purification
The nCas9-RT fragment from pCMV-PE2 was retrieved by BglII digestion then cloned into the pET30a(+) expression vector in frame between the NotI and NdeI sites using NEBuilder® HiFi DNA Assembly Master Mix (NEB) with bridging gblocks (Supplemental

Prime editing using mRNA
Human PSCs cultured on MEFs were harvested and nucleofected using the same procedure as described in the plasmid delivery section, except with 4µg in vitro transcribed nCas9-RT mRNA, 150pmol chemically modified synthetic pegRNA for PE2 strategy or 4µg in vitro transcribed nCas9-RT mRNA, 100pmol chemically-modified synthetic pegRNA and 50pmol chemically modified synthetic ngRNA for PE3 strategy.
For feeder feel culture, hPSCs were harvested using accutase. In multi-dosing experiments, after the 1st nucleofection, hPSCs were nucleofected for the 2nd and 3rd time at day 7 and day 14, respectively. Detailed protocols can be found on protocols.io (dx.doi.org/10.17504/protocols.io.b4qnqvve).

Genotyping of single cell expanded genome edited hPSCs clones by RLFP.
Restriction fragment length polymorphism (RFLP) analysis was performed as previously described (Hernandez et al., 2005)

Single cell survival assay
Human PSCs nucleofected with plasmid, RNP or mRNA were seeded to MEFs at 100 cells/cm 2 , then cultured for 14 days with media changed every other day. Cells were then stained for alkaline phosphatase as described above and the number of colonies in each condition were counted.

Bulk NGS and allele quantification
Edited bulk cells were collected using trypsin at days 5 post nucleofection, then DNA extracted, mutation-region amplified, NGS and analyzed as described above Detailed protocols can be found on protocols.io (dx.doi.org/10.17504/protocols.io.b4n3qvgn).

Software and Statistics
Bar graphs were drawn in Graphpad Prism 9. Error bars indicate the standard deviation

Declaration of interests
The authors declare no competing interests.