Optimized prime editing of the Alzheimer’s disease-associated APOE4 mutation

Summary

Gene editing strategies to safely and robustly modify the Alzheimer’s disease-associated APOE4 isoform are still lacking. Prime editing (PE) enables the precise introduction of genetic variants with minimal unintended editing and without donor templates. However, it requires optimization for each target site and has not yet been applied to APOE4 gene editing. Here, we screened PE guide RNA (pegRNA) parameters and PE systems for introducing the APOE4 variant and applied the optimized PE strategy to generate disease-relevant human induced pluripotent stem cell models. We show that introducing a single-nucleotide difference required for APOE4 correction inhibits PE activity. To advance efficient and robust genome engineering of precise genetic variants, we further present a reliable PE enrichment strategy based on diphtheria toxin co-selection. Our work provides an optimized and reproducible genome engineering pipeline to generate APOE4 disease models and outlines novel strategies to accelerate genome editing in cellular disease model generation.

Highlights

• •PE screening optimized the introduction of the AD-associated APOE4 mutation
• •Optimized prime editing efficiently generated APOE4-carrying human iPSCs
• •PE efficiency is reduced for APOE4 to APOE3 correction compared to its introduction
• •Diphtheria toxin co-selection strategy significantly enriches precise prime editing

In this article, Sienski and colleagues applied a prime editing (PE) screening approach to optimize the introduction of the Alzheimer’s disease-associated APOE4 mutation, enabling efficient generation of APOE4-carrying human iPSCs. Despite a single-nucleotide difference, PE efficiency was reduced for APOE4 to APOE3 correction compared to its introduction. A diphtheria toxin co-selection strategy further enriches PE to enhance model generation.

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disorder associated with progressive cognitive decline and accounts for the majority of people living with dementia worldwide (Alzheimer’s Association, 2024; Pinals and Tsai, 2022). Polymorphisms in the apolipoprotein E (APOE) gene are the major genetic risk factor for late-onset AD, which accounts for more than 95% of the total AD population (Corder et al., 1993; Lambert et al., 2013; Yamazaki et al., 2019). The three major alleles of APOE, ε2 (APOE2), ε3 (APOE3), and ε4 (APOE4), only differ at two amino acid positions (112 and 158). Compared to the common allele APOE3, APOE4 is associated with an earlier age of onset and increased risk of late-onset AD while APOE2 has a protective effect on disease risk (Farrer et al., 1997; Li et al., 2020; Lumsden et al., 2020).

Studying APOE’s function and role in AD risk and pathogenesis can provide a better understanding of the underlying disease mechanisms and holds promise to facilitate the development of novel therapeutic strategies. However, those in-depth studies require suitable, disease-relevant, and reliable cellular models containing the APOE variants of interest. Previously, APOE4 disease models have been obtained through patient-derived cellular models or by introducing the variant through homology-directed repair (HDR)-based CRISPR editing which is often associated with low efficiencies, thus resulting in laborious clone screening (Lin et al., 2018; Wadhwani et al., 2019; Nelson et al., 2023). Unintended deleterious on- or off-target effects can also frequently occur after DNA double-strand breaks (DSBs) induced by SpCas9-based CRISPR editing (Fu et al., 2013; Pattanayak et al., 2013; Kosicki et al., 2018; Weisheit et al., 2020). Additionally, genome engineering and sequence validation at the APOE locus have also been further complicated by its guanine and cytosine (GC)-rich sequence composition.

The recent prime editing (PE) technology, however, can overcome unintended editing effects and facilitate disease model generation while also holding the potential to become a next-generation therapeutic strategy. In contrast to HDR-based editing by Cas9, PE uses a Cas9-nickase fused to an engineered reverse transcriptase (RT), introducing the modification of interest not through the repair of a DSB but instead only through a nicked DNA strand (Anzalone et al., 2019). PE guide RNA (pegRNA) directs the complex to the target site while simultaneously encoding the desired edit. To achieve maximum efficiency, pegRNAs must be optimized for each target site.

While PE has been recently applied to efficient and precise disease model generation in human induced pluripotent stem cells (iPSCs), optimized and reproducible strategies to introduce the APOE4 variant into human cells for AD modeling are still lacking (Sürün et al., 2020; Chemello et al., 2021; Li et al., 2022; Tremblay et al., 2022). Additionally, cell lines containing the variants of interest from large-scale initiatives are often only available to not-for-profit entities, creating the need for robust protocols to generate the cells in-house for commercial organizations. An optimized PE approach at the APOE4 target site would be fundamental to enable future therapeutic approaches to precisely correct the APOE4 variant, similar to recently reported systematic PE optimization strategies for the most common cystic fibrosis causing mutation CFTR F508del (Sousa et al., 2024).

Here, we developed a PE strategy using an unbiased pegRNA optimization approach to precisely introduce the APOE4 variant achieving editing efficiencies up to 28.2% in human iPSCs and up to 42.3% in HEK293T. However, optimized pegRNA parameters to introduce the APOE4 variant did not correlate with efficient correction of APOE4. This platform to optimize PE approaches can be broadly applied across disease areas to significantly simplify and improve disease model generation. It further provides a resource to model the APOE4 variant efficiently and thus enables future studies of AD pathogenesis.

Results

PE is a superior strategy to introduce the APOE4 variant

Introducing the APOE4 and APOE3 variants into cells of interest is crucial to gain a better mechanistic understanding of the strongest genetic risk factor for AD. A common strategy to introduce precise mutations is based on harnessing the cells’ own HDR pathway by providing a single-stranded DNA template together with SpCas9 and a corresponding single-guide RNA (sgRNA) (Figure 1A). Alternatively, the more efficient but error-prone non-homologous end joining (NHEJ) pathway can be exploited by performing a dual sgRNA edit while providing a double-stranded DNA (dsDNA) knockin fragment containing the mutation. However, using either SpCas9-mediated HDR- or NHEJ editing achieved less than 1% precisely introduced APOE4 mutations in HEK293T (Figure 1B). Unintended on-target insertion and deletions (InDels) on the other hand reached up to 60% of sequencing reads (Figure 1C). Additionally, SpCas9-based editing was associated with a high frequency of unspecific off-target editing of up to 5.7% at predicted top off-target sites (Figure 1D). Replicating the same strategies with the recently reported more specific nuclease PsCas9, which generates 3 nt DNA overhangs at the cut site, significantly improved the introduction of precise APOE4 mutations up to 3.9%, reduced InDels, and abolished off-target editing (Figures 1B–1D) (Bestas et al., 2023). The efficiency of introducing APOE4 further increased to 4% when applying a PE-based editing approach with recommended but not optimized parameters (Figure 1D). PE-based editing was not associated with any unintended on- or off-target editing (Figures 1C and 1D).

Figure 1.

Editing approaches to introduce the APOE4 variant

(A) APOE4 targeting using a homology-directed repair (HDR, blue), non-homologous end joining (NHEJ, red), or prime editing approach (PE, orange). sgRNA target sequences are indicated in their respective color (blue, red, and orange, respectively), PAM sequences are indicated in light blue and cuts occur at triangle positions.

(B–D) Precise editing efficiencies to introduce the APOE4 mutation (B), simultaneous non-specific InDel (C), and unintended off-target edits (D) using editing approaches relying on SpCas9 (gray), PsCas9 (light blue), or PE (dark blue) in HEK293T (n = 3 independent experiments). Data are shown as mean ± SEM. ^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001. ssODNs, single-stranded oligodeoxynucleotides; dsODNs, double-stranded oligodeoxynucleotides; PE, prime editing; InDels, insertions and deletions.

A high-throughput pipeline to optimize the 3′ extension of an APOE4-introducing pegRNA

While a PE-based strategy is a highly precise approach to model the APOE4 polymorphism that is superior compared to alternative editing approaches, its efficiency remains low. Improving the applied pegRNA has the potential to significantly enhance PE efficiencies, more specifically, refining its 3′ extension consisting of a primer binding site (PBS) and an RT template (RTt) containing the desired edit (Figure 2A) (Anzalone et al., 2019). Optimizing PE becomes particularly crucial when the priming region exhibits a G/C content outside the range of 40%–60%. Notably, APOE, being a GC-rich gene (with 73.7% G/C content in the priming region), necessitates a focused investigation. We thus employed an unbiased and large-scale screening strategy to identify the ideal pegRNA parameters to introduce the APOE4 mutation. The assay tested 96 pegRNAs with varying PBS and RTt length combinations in the context of the PE2 system while keeping the same predicted highest scoring spacer sequence. It was conducted in a 96-well-based format throughout, from initial cloning through transfection in HEK293T up to amplicon sequencing to assess the editing efficiency (Figure 2B). PBS length was assessed from 8 to 19 nt while RTt was tested in a range from 11 to 18 nt. Optimized PE increased more than 2.5-fold from the initial recommended parameters by Anzalone et al. up to 10.2% at an RTt length of 15 and PBS of 14 (RTt15 PBS14, Figure 2C) (Anzalone et al., 2019). No precise editing was observed when the RTt was smaller than 14 nt, highlighting the need for target-specific optimization. Next to the optimal combination of RTt and PBS lengths, the highest editing efficiencies were seen when only making slight changes to these parameters. Additionally, the editing efficiency declined with a long PBS, consistent with the energetic requirements associated with hybridizing a priming region with high G/C content. All pegRNA resulted in similarly low levels of InDels (<2%) at the target site, irrespective of the achieved precise editing (Figure S1).

Figure 2.

Optimized prime editing to introduce the APOE4 variant in HEK293T

(A) Prime editing schematic showing Cas9 nickase fused to RT and pegRNA with various 3′ extension lengths with highlighted APOE4 mutation.

(B) High-throughput 3′ extension pegRNA optimization platform in a continuous 96-well pipeline from cloning to amplicon sequencing.

(C) pegRNA optimization platform with varying PBS and RTt length of the APOE pegRNA in HEK293T for precise editing to introduce the APOE4 mutation (n = 1 independent experiment).

(D and E) Precise APOE4 editing (D) or InDels (E) across different prime editor variants for RTt15PBS14 and RTt18PBS16 with (+nick) or without nicking sgRNA (n = 3 independent experiments).

(F) Precise APOE4 editing across 3′ extension parameters using pegRNA (gray) or epegRNA (blue) (n = 3 independent experiments). Data are shown as mean ± SEM. PE, prime editing; PBS, primer binding site; RTt, reverse transcriptase template; InDels, insertions and deletions; epegRNA, engineered pegRNA. ^∗∗∗∗p < 0.0001. Significance levels were assessed as indicated or between nicking sgRNA samples and their respective sample without nicking sgRNA.

Enhanced PE systems to introduce the APOE4 variant

Since the development of the initial PE2 system, improved PE systems have been established to enhance PE efficiency and specificity. To assess if our APOE editing approach could be further improved, we applied the PEmax (PE3max), PE4max (PE5max), and PE6d systems in combination with additional nicking of the non-edited strand (Anzalone et al., 2019; Chen et al., 2021; Doman et al., 2023). PEmax is an enhanced prime editor architecture based on PE2. PE4 involves the co-expression of a dominant-negative MLH1 protein to temporarily inhibit DNA mismatch repair, while PE6d includes an engineered Moloney murine leukemia virus RT. Furthermore, PE3max and PE5max are modifications of PEmax and PE4max, respectively, incorporating nicking sgRNA. None of the advanced PE systems alone significantly exceeded PE2 efficiency in installing the APOE4 edit as tested for two different pegRNAs (Figure 2D). In combination with a nicking sgRNA on the other hand, precise editing was equally increased across PE systems up to 42.3% (3.7-fold). Nicking of the non-edited strand 46 bp upstream of the target site, however, significantly increased undesired InDels, primarily deletions induced from double nicks (Figure 2E). The level of InDels was nevertheless relatively low at less than 1.5% across editors, with the highest unintended on-target editing being induced by PE6d for both pegRNAs.

Next, we assessed if applying engineered pegRNA (epegRNA) containing structured RNA motifs at the 3′ terminus to increase pegRNA stability (tevopreQ1) could further improve PE efficiency at the APOE4 site (Nelson et al., 2022). For three out of the four tested pegRNA parameters, including the optimized RTt15PBS14 epegRNA, the editing outcome was not changed (Figure 2F). A shorter PBS length (RTt15PBS9) did substantially benefit from the epegRNA context while editing was not rescued for RTt12PBS12. Having optimized PE for introducing the APOE4 variant, we next aimed to (1) apply the optimized PE approach to generate disease models in relevant cell types and (2) assess if we can transfer our optimizations from introducing the APOE4 variant to correcting it.

An optimized pegRNA to introduce the APOE4 variant in pluripotent stem cells

iPSCs can be differentiated into a wide variety of cell types and model human diseases by prior introduction of genetic variants of interest. Applying PE to edit iPSCs allows for a more precise engineering and avoids or reduces common side effects such as DSB-associated toxicity or unintended on- and off-target editing due to a nick-based edit instead of relying on DSB (Ihry et al., 2018; Kosicki et al., 2018; Weisheit et al., 2020). To efficiently generate APOE4-based models of AD, we applied PE and our optimized pegRNA to introduce the APOE4 variant in iPSC and generated clonal cell lines (Figure 3A). Clones were verified by Sanger sequencing, and screening of 192 clones resulted in 9 APOE4/APOE4 homozygous clones and 35 APOE3/APOE4 heterozygous clones, equivalent to 13.8% precise editing in iPSC (Figure 3B). Integration of amplicon sequencing of DNA extracted from the edited pool into the screening pipeline accurately predicted the number of edited alleles (56 vs. 53 alleles). It would further reduce the workload in future screenings by predicting the number of clones required to screen to achieve a certain number of specific clones (Figure 3C). The clones were screened for markers of pluripotency after editing and were karyotypically normal (Figure S2). APOE3 and APOE4 protein expression was unaffected after introduction of the disease variant (Figure 3D). Ultimately, we have shown that our optimized PE strategy efficiently introduces the APOE4 allele in human iPSC and outlined a strategy for how to generate clonal and isogenic APOE4 iPSC lines.

Figure 3.

Introduction and correction of the APOE4 variant in human iPSC

(A–C) Pipeline to introduce the APOE4 mutation in human iPSC (A) using PE2 and RTt15PBS14 followed by single-cell cloning and Sanger sequencing of the individual clones (APOE3/3, APOE3/4, or APOE4/4) for quality control (B). Amplicon sequencing of DNA extracted from the edited pool is a precise estimator for the edited alleles and the number of clones required (C) (n = 1 independent experiment).

(D) APOE protein in unedited cells (APOE3) and two selected iPSC clones with introduced APOE4 mutations relative to housekeeper β-tubulin (n = 3 independent experiments, representative image shown).

(E) Precise editing efficiencies to introduce (gray) or correct (blue) the APOE4 mutation in iPSC across different 3′ extension length in APOE pegRNA (n = 3 independent experiments).

(F) Precise editing efficiencies to correct the APOE4 mutation in iPSC using either PEn (gray) or PE (blue) approach using the optimized RTt15 PBS14 pegRNA (n = 3 independent experiments).

(G and H) Predicted secondary structure of APOE3to4 (G) and APOE4to3 (H) pegRNA. APOE mutations in the RTt, U59 flipped nucleobase, and stem loop 1 sequence are highlighted. Colors indicate base-pair probabilities, from blue: low to green: mid to red: high. Data are shown as mean ± SEM. ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001. iPSCs, induced pluripotent stem cells; PE, prime editing; PEn, prime editing nuclease; PBS, primer binding site; RTt, reverse transcriptase template; InDels, insertions and deletions.

Optimized pegRNAs for APOE4 introduction are not efficient for APOE4 correction

We next assessed if these optimized PE parameters would also be applicable to correct the APOE4 variant as a potential therapeutic approach. The APOE4to3 correction PE approach can be based on the same spacer and PBS sequence as the previous APOE3to4 editing approach while only the RTt has to be adapted to include the single-nucleotide difference (C/T instead of T/C). We aimed to correct the APOE4 variant in the previously generated human iPSCs and assessed five pegRNAs from the least efficient RTt12 PBS12 to the best-performing RTt15 PBS14. All pegRNA to introduce the APOE4 variant in iPSC displayed the same order of editing efficiencies as previously observed in the large-scale assay in HEK293T. The achieved maximum editing efficiency of 28.2% in iPSCs even exceeded the previously obtained efficiencies in HEK293T (Figure 3E). However, none of the PE strategies to correct the APOE4 variant achieved an efficiency beyond 1.2% (RTt15 PBS14), a significant reduction of more than 23-fold compared to the efficiency obtained when introducing APOE4.

We next investigated if the poor performance of the APOE4-correcting pegRNA could be rescued when applying the prime editor nuclease (PEn) genome editor, a nuclease-based PE approach that can outperform conventional PE especially at hard-to-edit target sites (Adikusuma et al., 2021; Peterka et al., 2022). However, PEn only marginally increased APOE4 correction compared to PE (Figure 3F). Although different editing strategies relying on distinct DNA repair mechanisms were applied, the efficiency to correct APOE4 does not reach the same efficiency as for introducing APOE4.

Next, we assessed if the efficiency to modify the APOE4 variant differs in (1) a different cellular background due to their unique chromatin context or (2) with APOE4 as a naturally occurring variant in contrast to being introduced through genome editing. We assessed HeLa cells, which have a heterozygous APOE3/APOE4 genotype, allowing to test the efficiency to introduce and correct APOE4 in the same cellular model (Schaffer et al., 2014). The low efficiency of APOE4 correction could not be rescued either (Figure S3). On the contrary, although RTt15 PBS14 was still the most efficient pegRNA, the efficiency to introduce APOE3 was reduced to less than 1%. The simultaneous editing control at an independent locus demonstrated robust editing, indicating that only the applied PE approach at the APOE locus did not induce a high degree of editing for either correction or introduction (Figure S3). In contrast to the iPSC model, APOE is not expressed in HeLa cells and the obtained low editing efficiency might be directly correlated to the associated chromatin state (Schaffer et al., 2014; Li et al., 2023).

We next investigated if the single-nucleotide difference between both pegRNAs for either introducing or correcting APOE4 might result in any structural changes in their unbound state, potentially compromising their activity and explaining the drastic reduction in editing efficiency. Secondary structure prediction revealed that the APOE4 mutation in the APOE3to4 pegRNA is predicted to form a base pairing with the terminal guanine of the stem loop 1 sequence (Figures 3G, 3H, and S3) (Andronescu et al., 2007). The base pairing facilitates the formation of a bulge loop, allowing for the crucial recognition by SpCas9 through the flipped out U59 nucleobase (Nishimasu et al., 2014). APOE3 in the APOE4to3 pegRNA on the other hand remains unpaired, which does not allow for a protruding stem loop 1 and results in a base-paired and inward facing U59 nucleobase. Stem loop 1 formation has been shown to be essential for a functional SpCas9-sgRNA complex, and an abolished stem loop 1 cannot induce any significant editing, thus highlighting the incomplete stem loop 1 formation as a potential problem in the APOE4to3 pegRNA (Nishimasu et al., 2014).

A co-selection approach increases precise PE efficiency

To further benefit from our optimized PE approach to introduce the APOE4 variant, we aimed at providing a streamlined and robust strategy with high efficiencies for researchers aiming to generate APOE cell models. While we had already optimized the 3′ extension, we now focused on further enriching edited cells by using a co-selection method. Such a strategy is based on an increase in editing at the target site if another site is edited simultaneously (Kim et al., 2011; Ramakrishna et al., 2014). An enrichment of target edits can then be achieved if co-editing is relying on introducing a cellular resistance mutation to a selection reagent. We previously reported a robust and efficient co-selection method through simultaneously introducing resistance mutations to diphtheria toxin (DT) in the HBEGF gene (Figure 4A) (Li et al., 2021). Similar to other co-selection systems, we aimed to firstly extend our DT-based co-selection method to PE and, secondly, to enrich our APOE4 variant engineering system further (Levesque et al., 2022).

Figure 4.

DT-based enrichment of APOE4 genetic engineering

(A) Pipeline to enrich target locus (APOE4, red) PE through simultaneous PE at the HBEGF locus (blue) to introduce DT resistance followed by DT selection.

(B and C) Precise editing efficiencies (B) and simultaneous non-specific InDel (C) to introduce specific mutations in EMX1, FANCF, RNF2, and HBEGF with (blue) or without (gray) additional DT selection (n = 3 independent experiments).

(D and E) Precise editing efficiencies to introduce the DT resistance mutation in HBEGF (D) and simultaneous non-specific InDel (E) across varying PBS and RTt length of the HBEGF pegRNA (n = 1 independent experiment).

(F and G) Precise editing efficiencies to introduce the APOE4 mutation with (blue) or without (gray) additional DT selection using a 1:1 APOE:HBEGF pegRNA ratio (F) and across different ratios of APOE:HBEGF pegRNA (G) for APOE RTt15 PBS14 pegRNA (n = 3 independent experiments). All experiments were performed in HEK293T. Data are shown as mean ± SEM. ^∗∗p < 0.01; ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001. DT, diphtheria toxin; PE, prime editing; PBS, primer binding site; RTt, reverse transcriptase template; InDels, insertions and deletions.

We demonstrated a proof of concept for DT-based co-selection enrichment of PE by targeting the previously established EMX1, FANCF, and RNF2 as well as our DT-selection resistance gene HBEGF in HEK293T (Figure 4B). Targeting HBEGF with PE to introduce the resistance mutation increased from 5.3% to 95.8% after DT selection. Editing at target sites increased significantly up to 3-fold in all genes upon co-editing of HBEGF and selection with DT. Enrichment of precise target site editing either significantly reduced unintended on-target InDels (EMX1) or did not significantly affect InDel frequencies (FANCF and RNF2) (Figure 4C). Overall, we have shown that a DT-based co-selection system is applicable to PE and can enrich editing at target sites further.

An optimized DT co-selection approach to increase APOE4 variant introduction

To further optimize the selection process and increase the pre-selection precise editing efficiency at the HBEGF locus, we subjected our PE strategy for HBEGF to our previously applied unbiased pegRNA optimization protocol of PBS and RTt length. Optimized HBEGF editing would allow for a higher proportion of cells surviving the selection process and thus a smoother transition into routine culture and experiments. Similar to the previous APOE optimization, HBEGF editing efficiencies varied from 0% to 14.8% across pegRNA with only specific PBS or RTt length allowing for high editing (Figure 4D). An HBEGF RTt length of 12, 16, or 17 in combination with a PBS length of 11 achieved maximum efficiencies. Imprecise and unintended InDel frequencies followed the obtained pattern for precise editing, peaking at 2.2% (Figure 4E).

Lastly, we applied the optimized co-selection PE strategy for HBEGF (RTt17PBS12) to enrich APOE (RT15PBS14) editing in HEK293T cells. To obtain the highest proportion of precise edits in the enriched cell pool, we did not apply the nicking sgRNA for the enrichment strategy. Co-editing and selection significantly increased precise editing to introduce the APOE4 variant by 2.5-fold up to 23.8% (Figure 4F). We assessed if altering the prevalence of target-site pegRNA would also further increase overall precise target editing. Compared to an initial 1:1 ratio of APOE to HBEGF pegRNA ratio, a ratio of 10:1 or higher significantly enhanced precise editing up to 40.4% across different APOE pegRNAs, irrespective of the initial pegRNA efficiency (Figures 4G and S4). Overall, we have outlined and optimized a PE-based editing approach to introduce the AD-associated APOE4 mutation, which can further be enriched through a co-selection method.

Discussion

PE to precisely and efficiently model and correct the disease-associated APOE4 mutation holds the potential to substantially advance the current therapeutic opportunities, cellular model generation, and functional disease studies for AD. We outline an unbiased strategy for PE optimization, present an optimized pegRNA for APOE4 genetic engineering and an additional PE-compatible co-selection method to further increase editing efficiencies.

We employed a pegRNA screening approach to identify the most efficient 3′ extension to introduce the APOE4 mutation. Interestingly, our most efficient pegRNA did not match the pegRNA design recommendations and prediction algorithms, thus highlighting the strength of such an unbiased locus-specific experimental optimization to obtain the best pegRNA. The 3′ extension of the experimentally tested best APOE4 pegRNA (RTt15 PBS14) falls within the predicted pegRNA optimization starting window recommended by Anzalone et al. with a PBS length of around 13 nt and an RTt within 10–16 nt (Anzalone et al., 2019). RTt15 PBS14’s first base of the 3′ extension is a C, however, which has been advised against in pegRNA design as it can disrupt guide RNA structure. A recently developed prediction model of PE efficiency (PRIDICT) further predicted our applied spacer but in combination with RTt17 PBS9, which achieved only a modest editing efficiency of 6.7% in our pegRNA screen (Mathis et al., 2023).

While we have screened a wide range of pegRNA parameters and PE systems, our study did not yet include a comprehensive assessment of all available PE systems, all combinations of the various parameters or other nickases instead of SpCas9 such as PsCas9 with a different site preference (Bestas et al., 2023; Doman et al., 2023). We also only tested a single spacer sequence in combination with the varying 3′ extensions, which was chosen due to its proximity to the targeted mutation and favorable targeting scores.

Unexpectedly, changing a single nucleotide within the RTt to mediate APOE4 correction instead of introducing the APOE3 mutation surprisingly drastically reduced the editing efficiency, which could not be rescued by mediating a DSB either. As a result, it is inevitable to screen the editing parameters for the actual therapeutic edit, as the most efficient pegRNA identified through a mutation-introducing screening approach may not be universally transferable. To allow for APOE4 correction using PE, a different spacer sequence could further be tested. Additionally, engineering the current pegRNA to specifically rectify the compromised secondary structure would potentially enable efficient APOE4 editing using the same optimized pegRNA parameters. Complementary approaches which apply the sgRNA and a circularized RTt-PBS separately, might also circumvent the problem of disrupted pegRNA folding (Liu et al., 2022). Beyond PE, base editing has been successfully applied to target APOE4 (Arg112 and Arg158), although not mediating the here attempted APOE4to3 conversion (Cys112 and Arg158) but instead introducing an APOE4to3r (Arg112 and Cys158) conversion, an allele whose physiological role remains to be resolved (Komor et al., 2016).

While PE has shown to be a powerful and precise tool, achieving a substantial level of editing remains a key challenge for therapeutic applications or disease model generation in pre-clinical development. We addressed that challenge by expanding our recent DT-based selection approach to the PE technology and achieved a notable increase in precise on-target editing. This approach complements other PE co-selection tools that have demonstrated superior performance compared to antibiotic or fluorescent reporter-based selection (Agudelo et al., 2017; Levesque et al., 2022). By further employing our DT co-selection to introduce clinically relevant disease-associated APOE mutations, we have generated a valuable strategy for disease cell line generation in pre-clinical and fundamental AD research.

To conclude, we have here outlined a generalizable PE optimization and enrichment strategy and have highlighted the importance for specific pegRNA optimization. We also present an optimized pegRNA for APOE4 genetic engineering as applied to human iPSC and demonstrate how a DT co-selection approach can drastically enrich PE efficiencies. The strategies presented in this work are going to simplify APOE-associated disease studies and provide the tools to further streamline and improve cellular disease model generation across genetic variants.

Experimental procedures

Cell culture

HEK293T cells (GenHunter Corporation) and HeLa (ATCC) were routinely passaged in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum. Human iPSCs were routinely passaged in the Cellartis DEF-CS 500 Culture System (Takara Bio), according to manufacturer’s instructions. All cell lines were maintained in 37°C with 5% CO₂, mycoplasma free and authenticated using STR profiling. Transfected HEK293T cells were selected in DMEM supplemented with 20 ng/mL DT from day 3 post transfection until negative control cells were dead.

Karyotyping

Karyotyping was assessed after generating a genetically modified iPSC line by g-banding analysis by Cell Guidance Systems LLC. Sample preparation was performed according to service instructions. In short, iPSCs were plated and prepared for karyotyping at 80% confluency. Colcemid (Gibco) was added for 60 min at 10 μg/mL. Cells were dissociated using TryPLE (Gibco) and incubated in KCl (Gibco) for 25 min at 37°C. Cells were resuspended in fixative solution (3:1 methanol:acetic acid, both Merck) for 30 min at room temperature before sending them for analysis.

Plasmids

Cas9 plasmids were generated by gene synthesis (GenScript). SpCas9-expressing plasmids consisted of CMV- or EF1α-controlled codon-optimized SpCas9 fused to nuclear localization signal and a self-cleaving enhanced green fluorescent protein (2A-EGFP) peptide in the pVAX1 backbone (Invitrogen). EF1α-driven PE2 plasmids were based on the previously published CMV-PE2 construct (Addgene #132775). Expression plasmids for PsCas9, SpCas9, and PsCas9 sgRNA expression plasmids were constructed as previously described (Bestas et al., 2023). SpCas9 sgRNA expression plasmids were constructed using previously reported sgRNA scaffolds by ligating annealed complementary primer pairs containing the protospacer sequence and suitable overhangs (5′-AAAC-XX-3′ and 5′-ACCG-YY-3′) into digested plasmid template using T4 ligase (NEB) (Chen et al., 2017).

pegRNA expression plasmids were constructed using two overlapping oligos: duplex A including protospacer and duplex B including the 3′ extension (5′-ACCG-Oligo1-CGTT-3′, 5′-revOligo1-3′ and 5′-Oligo2-3′, 5′-AAAA-revOligo2-AACG-3′). U6 cloning vector was digested with AarI restriction enzyme generating ACCG and TTTT overhangs compatible with the oligo duplexes. The two oligos for each duplex were annealed through lowering the temperature from 95°C to 25°C at a rate of 0.3°C/s. Duplex A, B and the U6 plasmid were ligated at a 3:3:1 ratio using T4 ligase (NEB) for 30 min at room temperature. In total, 0.001 pmol ligation mix was heat-shock transformed into 20 μL DH10B competent cells (Thermo Scientific). Inoculation and carbenicillin (100 μg/mL, Gibco) selection was performed in 96 deep-well plates containing SOC media. Plasmid DNA was then isolated according to QIAGEN plasmid plus 96 miniprep kit (QIAGEN). All sgRNA and pegRNA sequences used in this work are listed in Table S4.

Cell transfections

HEK293T and HeLa were seeded at 0.20 × 10⁵ per well in 96-well plates at 24 h prior to transfections. Media was changed 6 h prior to transfections on iPSC and the cells were reverse transfected and seeded at 0.12 × 10⁵ per well in 96-well plates. Transfections were performed using FuGENE HD transfection reagent (Promega) at a 3:1 ratio of transfection reagent to plasmid DNA. The amount and ratios of transfected plasmid and donor DNA for all respective experiments are listed in Tables S1–S3. Donor dsDNA was prepared by annealing the single-stranded oligonucleotides in Duplex Buffer (IDT) with a final concentration of 2 μM. The mixed oligonucleotides were incubated for 10 min at 95°C followed by decreasing the temperature stepwise to 25°C at 0.3°C/min.

Genomic DNA extraction and sequencing analysis

Genomic DNA was extracted by adding QuickExtract DNA extraction solution (Lucigen) to the plated cells and following manufacturer’s instructions. Amplicons of interest for NGS analysis were prepared in a two-step PCR-based amplification approach. Briefly, target sites were amplified with site-specific primers containing NGS adapters in a 25 μL PCR reaction consisting of 2 μL genomic template DNA, 12.5 μL KAPA HiFi HotStart ReadyMix, and 0.3 μM primers for 95°C for 3 min; 30 cycles of 98°C for 20 s, 64°C for 15 s, and 72°C for 10 s; and a final extension at 72°C for 30 s. PCR products were purified using the HighPrep PCR Clean-up System (MagBio Genomics) followed by assessment on a fragment analyzer (Agilent) to determine and validate correct product size and determine DNA purity and concentration. A second round of PCR amplification using KAPA HiFi HotStart ReadyMix, 1 ng PCR template, and unique indexing primers was performed for 72°C for 3 min; 98°C for 30 s, 10 cycles of 98°C for 10 s, 63°C for 30 s, and 72°C for 3 min; and a final extension at 72°C for 5 min. Amplicons were again purified, analyzed on a fragment analyzer, and further quantified using a Qubit 4 Fluorometer (Life Technologies). Sequencing was performed on an Illumina NextSeq system according to manufacturer’s instructions. All primers used in this work are listed in Table S4.

Bioinformatic analysis

NGS sequencing data were demultiplexed using bcl2fastq software. Paired reads were merged using FLASH and mapped to the amplicon reference sequence using the Burroughs-Wheeler aligner. The sequence alignment map files were processed to determine the mutations (SNP, InDels) occurring in each read along with their locations in the reference amplicon by parsing the cigar and MD:Z attributes. All mutations with a phred quality score of 25 or over, present in at least 10 reads, and with a minimum frequency of 0.1% were taken forward. Protospacer sequences were designed and putative off-targets were identified using CRISPOR. Off-target sites per protospacer were ranked by Cutting Frequency Determination score, and the top six candidate sites were assessed for off-target editing using NGS as described earlier (Doench et al., 2016; Concordet and Haeussler, 2018). RNA secondary structure predictions were performed using Geneious Prime 2022.2.2 and the Vienna RNA Websuite, energy model RNA by Andronescu (Andronescu et al., 2007; Gruber et al., 2008).

Western blot

Cell pellets were lysed in RIPA buffer (Thermo Scientific) containing 1x protease inhibitor cocktail (abcam) to obtain whole-cell protein lysates. Protein quantification was performed using the DC protein assay (Bio-Rad), and 10 μg of total protein was denatured in laemmli buffer (Bio-Rad) containing 10% β-mercaptoethanol (Sigma-Aldrich) at 80°C for 10 min. The samples were run on a 4%–20% Criterion TGX Stain-Free Precast Gel for 15 min at 300 V in Tris-glycine-SDS buffer and transferred to a 0.2 μm polyvinylidene difluoride membrane using the Trans-Blot Turbo Transfer System (all Bio-Rad). The membrane was blocked through incubation with 3% bovine serum albumin in Tris-buffered saline with 1% Tween 20 (TBST) for 1 h at room temperature. Primary antibody staining for APOE (SAB2701946, Sigma-Aldrich) or β-tubulin (86298, Cell Signaling Technology) was performed overnight at 4°C followed by staining with HRP-coupled secondary antibody (31450/31460, Invitrogen) for 1 h at room temperature with TBST washes in-between. The blots were imaged on a ChemiDoc MP Imaging System (Bio-Rad) followed by staining for loading controls of appropriate sizes through reprobing of the membrane. Image Lab 6.1 software (Bio-Rad) was used to quantify protein bands, and proteins of interest were normalized to their respective loading control on the same blot.

Flow cytometry analysis

Pluripotency assessment of iPSC clones was performed using BD Stemflow kit (BD Biosciences) according to manufacturer’s instructions. In short, dissociated cells were fixed using Cytofix (BD) followed by permeabilization and staining for 30 min at room temperature with Alexa Fluor 647 mouse anti-SSEA-4 (clone: MC813) and PerCP-Cy5.5 mouse anti-Oct3/4 (clone: 40/Oct-3) including isotype controls Alexa Fluor 647 mouse IgG3, anti-IgG3 isotype control (Clone: J606), and PerCP-Cy5.5 mouse IgG1 isotype control (Clone: X40). Cells were washed with PBS and resuspended in PBS + 2% BSA (Merck) and analyzed using BD LSRFortessa (BD Biosciences).

Statistical analysis

Data are presented as mean ± standard error of the mean (SEM), independent experiments are shown as individual data points, and statistical analysis was performed using GraphPad Prism 9 (GraphPad Software). Statistically significant differences between normally distributed variables were assessed between two or more groups using two-sample Student’s t test or one-way analysis of variance followed by Sidak’s multiple comparison test, respectively. p ≤ 0.05 was considered statistically significant.

Resource availability

Lead contact

Information and requests should be directed to the corresponding author, Grzegorz Sienski (grzegorz.sienski@astrazeneca.com).

Materials availability

All unique and stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

Data and code availability

Data supporting the findings of this study are presented within the article and supplementary figures. The accession number for the NGS data reported in this paper is NCBI Sequence Read Archive database: PRJNA1143179.

Acknowledgments

We thank Steve Rees and Mike Snowden for supporting this work. We thank AstraZeneca’s NGS facility for their support. A.L. and A.K.R. are previous Postdoctoral fellows of the AstraZeneca R&D Postdoc program. This work was funded by Discovery Sciences and the Postdoctoral program at AstraZeneca. Figures were created with BioRender.com.

Author contributions

A.K.R., A.L., and G.S. conceived the study. A.K.R. and A.L. performed most of the experimental work. M.F. performed bioinformatic analyses. S.L. performed proof-of-concept DT enrichment experiments. A.K.R. prepared the manuscript with input from G.S., A.L., and M.M. All authors approved the final draft of the manuscript. G.S. and M.M. supervised the study.

Declaration of interests

A.K.R., A.L., S.L., M.F., M.M., and G.S. are employees and shareholders of AstraZeneca. M.M. and S.L. are listed as co-inventors in an AstraZeneca patent application (application number: WO2019099943A1, WO2020208185A1, and WO2021204877A2) related to this work. S.L. is now an employee of AccurEdit Therapeutics.

Published: December 5, 2024