Bi-PE: bi-directional priming improves CRISPR/Cas9 prime editing in mammalian cells

PE2 plasmid was obtained from addgene (#132775). sgRNA plasmids were constructed through inserting oligos containing desired spacers into Bbs1 digested empty plasmids. pegRNA plasmids were generated by PCR amplification of existing sgRNA using reverse primers containing PBS, edits and HA sequences. Sequences of pegRNAs were listed in. Oligos used to generate sgRNAs were listed in. All plasmids were verified by Sanger sequencing. Supplementary Table S1 Supplementary Table S2

HEK293T were cultured in Dulbecco's modified Eagle's medium (Thermo Fisher Scientific), supplemented with 10% (v/v) fetal bovine serum (Life Technologies) and 1% penicillin/streptomycin (Boster Biological Technology Co. Ltd.), and maintained at 37°C with 5% CO₂. Cells were seeded onto 96-well plate (BIOFIL). Twenty-four hours after seeding, cells at a confluence of ∼70–80% were transfected with 0.7 ul of Transeasy™ (Forgene) and 276 ng of PE2 plasmid DNA, 62 ng of pegRNA1 plasmid DNA and 62 ng of pegRNA2 plasmid DNA (for Bi-PE and Cas9 nickase transfections); 276 ng of PE2 plasmid DNA, 93 ng of pegRNA plasmid DNA and 31 ng of sgRNA plasmid DNA (for PE3 transfections); or 276 ng of Cas9 plasmid DNA and 62 ng of sgRNA 1 plasmid DNA and 62 ng of sgRNA2 plasmid DNA (for paired Cas9 nuclease and Cas9 nickase transfections), according to the manufacturer's protocol.

Seventy-two hours post-transfection, genomic DNA was extracted by the addition of 50 μl of freshly made lysis buffer into each well of the 96-well plate. The lysate was incubated at 55°C for 10 min and was heat-inactivated at 95°C for another 10 min. Then the genomic DNA was subjected to PCR analysis using Phanta^® Max Super-Fidelity DNA Polymerase (Vazyme). The PCR reaction was performed with 1μl genomic DNA, and 0.2 μM of forward and reverse primers in a final volume of 30 μl. Primers flanking the deletion region were listed in (Supplementary Table S3). The amplicons were separated by agarose gel electrophoresis and gray values of armed bands were analyzed by Adobe Photoshop CC (2019). Editing efficiency was calculated as shown in Equation 1.

For capillary electrophoresis (.), the amplicons were subjected to Bio-Fragment Analyzer (Bioptic, Qsep1, C100001) with an S2 Cartridge. The DNA size marker (20–5000-bp) from Bioptic was used as internal control. Data were analyzed by Q-Analyzer software to calculate the peak area. Editing efficiency was calculated as shown in Equation. Supplementary Figure S5 2

For analyzing double LoxP insertions, the putative amplicons containing LoxP were purified with GeneJET Gel Extraction Kit (Thermo scientific) and ligated into a blunt-end vector using the pESI-blunt kit (YEASEN). DH5α-competent cells were then transformed with the ligation product. Colonies containing perfect double-LoxP insertions were recognized as accurate editing, and the ones containing double-LoxP insertions but harboring indels were recognized as inaccurate editing.

Genomic DNA (gDNA) was extracted 72 h after transfection and subjected to PCR analysis using Phanta^® Max Super-Fidelity DNA Polymerase (Vazyme). The PCR reaction included 1 μl of cell lysate, and 0.2 μM of forward and reverse primers in a final reaction volume of 30 μl. Genomic regions of interest were amplified by PCR with primers flanked with different barcodes (Supplementary Table S4). PCR reactions were performed as follows: 95°C for 3 min, then 35 cycles of (95°C for 15 s, 60°C for 15 s, and 72°C for 10 s), followed by a final 72°C extension for 10 min. The PCR products were purified with GeneJET Gel Extraction Kit (Thermo scientific) and quantified with NanoDrop (Thermo Fisher). Samples were sequenced commercially using the Illumina Novaseq6000 platform (150 bp, paired-end, Personal Biotechnology, Shanghai, China). A custom python script provided in Supplementary Note 1 was used to analyze and quantify the efficiency of the desired edits and indels produced by Bi-PE, PE3 and WT-Cas9. Substitution and indel frequencies were quantified as the percentage of total sequencing reads, and the threshold of editing activity was set to above 0.01%.

GraphPad Prism 8 software was used to analyze the data. All values are presented as mean ± s.d. of at least two independent biological replicates. Differences among groups were tested using two-tailed Student's t-tests. A P value < 0.05 was considered statistically significant. ***P < 0.001, **P < 0.01, *P < 0.05.

Previous studies have characterized the performance of PE3 in targeted small fragment deletions, while large fragment deletion remains a big challenge. To test the ability of PE3 in targeted deleting relatively large genomic fragment, ∼100–1000 bp, we designed a set of four pegRNAs targeting HEK3 loci to delete 198-bp, 372-bp, 530-bp and 654-bp fragment, respectively, all of which share a common spacer and prime binding site (PBS). An sgRNA that breaks the un-edited strand at + 90 was chosen as nick sgRNA (Supplementary Figure S1a). It was noteworthy that the spacers of the two sgRNAs and the PBS of the pegRNA had been demonstrated to be functional ((6) and data not shown). However, transfection of these individual set of pegRNA and sgRNA, together with PE2, into HEK293T cells did not result in any detectable desired genomic deletion (Supplementary Figure S1b).

In prime editing system, the newly reverse transcribed ssDNA was thought to invade into the downstream double strand DNA through homologous DNA pairing and ssDNA invasion, thereby inducing DNA repair mechanisms to introduce reverse transcribed ssDNA into the genome (6,18,19). We reasoned that the failure of those PE3s was partially due to inefficient ssDNA invasion and that repositioning the nicks to the region pairing with HA should facilitate the invasion and improve targeted deletion. To test this hypothesis, we designed three types of nicks, which were located inside (Type I), adjacent to the 3′ end of (Type II) or downstream (Type III) of the fragment to be knocked out (Figure 1A, B). The pegRNA was designed to delete a 654-bp fragment from HEK3 locus and totally six nicks were designed, including 4 Type I, 1 Type II and 1 Type III nick sgRNAs (Figure 1C). Before PE experiments, all the nick sgRNAs were validated to be functional by wild-type Cas9 in HEK293T cells, leading to the formation of indels within target positions of each sgRNAs (Supplementary Figure S2). However, PE experiments revealed that only Type II nicks produced detectable targeted deletion of the 654-bp fragment in HEK293T cells, as evidenced by agarose gel analysis of the amplicons flanking the target deletion (Figure 1B, C and Supplementary Figure S3). This phenomenon was further confirmed by four additional target editing, in which a 600-bp fragment in β-Actin locus, a 400-bp fragment in VEGFA locus, a 481-bp in AAVS1 locus and a 482-bp in DMD locus could only be deleted by PE3 with Type II nick (Figure 1C, D and Supplementary Figure S3). Moreover, this phenomenon was also demonstrated to be true in additional cell types, including HeLa (epithelial cell) and K562 (myelogenous leukemia) cells (Supplementary Figure S4). Together, these results supported a mode involving ssDNA invasion during the repair of PE induced lesion and suggested that positioning the nick of the non-edited strand to the homologous region achieved large fragment deletion.

Inspired by above findings, we hypothesized that engineering the nick sgRNA to pegRNA would allow bi-directional priming, thereby doubling the efficiency of large fragment deletion. Henceforth, the design of bi-directional priming was referred to as Bi-PE. PE3 using upstream sgRNA as pegRNA was referred to as left PE3 (L-PE3) and the one using downstream sgRNA as pegRNA was referred to as right PE3 (R-PE3) (Figure 2A). To compare the efficiency of Bi-PE and PE3, we designed a panel of Bi-PEs and corresponding PE3s targeting HEK3 loci to delete 372-bp, 530-bp, 654-bp and 861-bp fragment, respectively and determined the presence of targeted deletions by agarose gel analysis of the amplicons flanking the target region (Figure 2B). We observed a significantly improvement of the deletions by Bi-PE as compared to PE3 in two out of four deletions in HEK293T cells (654-bp and 861-bp, Figure 2C). In the deletion of 861-bp, the editing efficiency of Bi-PE achieved ∼31.5%, which was 5 times and 3 times higher than that of L-PE3 and R-PE3 respectively (Bi-PE: L-PE3: R-PE3 = 31.5%: 5.3%: 7.3%). In the deletion of 654-bp, the efficiency of Bi-PE was 9 and 12 times higher than that of L-PE3 and R-PE3 respectively (Bi-PE: L-PE3: R-PE3 = 23.4%: 2.4%: 1.8%). The efficiency of Bi-PE in deleting 530-bp was 1.5 and 1.4 times higher than that of L-PE3 and R-PE3 respectively (Bi-PE: L-PE3: R-PE3 = 3.3%: 1.3%: 1.4%). And the efficiency of Bi-PE in deleting 372-bp was 0.36 and 0.52 times higher than that of L-PE3 and R-PE3 respectively (Bi-PE: L-PE3: R-PE3 = 7.6%: 5.6%: 5.0%) (Figure 2C). To confirm the accuracy of above data determined by agarose gel analysis, we conducted capillary electrophoresis analysis. As shown in Supplementary Figure S5, results of capillary electrophoresis were consistent with those obtained from agarose gel analysis. Overall, Bi-PE exhibited a higher activity than either L-PE3 and R-PE3 (Figure 2D).

Because Bi-PE produced double nicks in the target region, it is possible that these nicks would induce fragment deletion via NHEJ or HDR (the 3′ end of pegRNA serving as donor template) pathways. To test such possibility, we coupled paired pegRNAs or their corresponding sgRNAs to Cas9 nickase. However, neither combination produced detectable targeted deletions (), suggesting that these targeted deletions were PE-specific. Supplementary Figure S6a

To test if the improvement of Bi-PE in deleting large genomic fragment was general, we included additional nine targeted deletions in four different loci, β-Actin, VEGFA, AAVS1 and DMD in HEK293T cells (Figure 2B and C). In β-Actin locus, the average efficiency of Bi-PE in deleting 600-bp fragment was ∼11.3%, which was 0.53 times higher than that of L-PE3 and 2.3 times higher than that of R-PE3 (Bi-PE: L-PE3: R-PE3 = 11.3%: 7.4%: 3.4%). In the same locus, Bi-PE had an average efficiency of 23.2% in deleting 1025-bp fragment, which was 1.5 times higher than L-PE3 and 0.39 times higher than R-PE3 (Bi-PE:L-PE3:R-PE3 = 23.2%:9.2%:16.7%). Similar improvement of Bi-PE over PE3 was also observed in VEGFA locus, in which the efficiency of Bi-PE in deleting 400-bp fragment was 6 times and 16 times higher than that of L-PE3 and R-PE3 respectively (Bi-PE:L-PE3:R-PE3 = 36.7%:5.3%:2.2%). In the deletion of 1522-bp fragment, Bi-PE had an average editing efficiency of 64.4%, an increase of 0.1 and 1.4 times, respectively, over L-PE3 and R-PE3 (Bi-PE:L-PE3:R-PE3 = 64.4%:58.7%:26.6%). In the rest five targeted deletions, Bi-PE outperformed both L-PE3 and R-PE3 in 3 deletions (926-bp in AAVS1 locus and 482-bp and 906-bp in DMD locus) in terms of deletion efficiency. In the deletion of 302-bp in DMD locus, Bi-PE showed similar efficiency to both PE3s. And only in one deletion (241-bp in AAVS1), Bi-PE was less efficient than both PE3s (Figure 2C). In summary, Bi-PE showed a higher editing efficiency in large fragment deletion than PE3 in 11 out of 13 deletion events (Figure 2C).

Then we tested if the efficiency of Bi-PE could be further improved by optimizing the HA length. We varied the length of HA from 8-bp to 50-bp. The results showed that HA length did influence the efficiency of deletion, but the effect was case-dependent and not strictly predictive (). Therefore, we recommend optimizing the length of HA when editing efficiency of Bi-PE is valued. Supplementary Figure S7

Next, we tested if Bi-PE also outperformed PE3 in large fragment deletion in other cell types, including K562, HeLa and B16 (mouse melanoma cell) cells. We found that the efficiency of Bi-PE was generally higher than that of PE3 in these cells. Strikingly, PE3 did not produce any detectable deletions in three events (600-bp in β-Actin locus in K562 cell, and 254-bp and 336-bp in Hoxd locus in B16 cell). By contrast, Bi-PE produced obvious deletions in the same events (Supplementary Figure S8). Together, these results demonstrated a significant improvement of Bi-PE over PE3 in targeted deletion of large DNA fragment (Figure 2B–D).

It has been observed that PE3 produced considerable level of indels in generating point mutations or small fragment insertions/deletions, which was likely due to PE3 producing two nicks in each strand of its target loci (6). An examination of the HTS data revealed that PE3 also produced undesired indels in alleles with aimed large fragment deletions, leading to inaccurate edits (Figure 2E and Supplementary Figure S9). For example, in HEK3 loci, L-PE3 mediated deletion of 654-bp fragment produced 4.7% undesired indels and R-PE3 produced 8.2% indels. By contrast, Bi-PE mediated same deletion only produced 0.35% indels. In the same loci, when deleting 372-bp fragment, L-PE3 and R-PE3 produced 2.2% and 1.74% indels, respectively, while Bi-PE only produced 0.85% indels (Figure 2E). This phenomenon was further confirmed by additional 11 targeted deletions (Figure 2E). On average, Bi-PE produced more than 2 times lower indels than PE3 (Figure 2F).

Finally, we sought to compare Bi-PE to wildtype Cas9 coupled with paired sgRNAs in targeted deletion since the latter has been shown to be efficient in such editing. We designed 6 targeted deletions and found that although wildtype Cas9 with paired sgRNAs achieved higher deletion efficiency compared to Bi-PE (on average, wildtype Cas9:Bi-PE = 68.3% versus 38.1%) (), wildtype Cas9 produced deletions were accompanied by significantly higher levels of un-intended indels compared to Bi-PE (on average, wildtype Cas9:Bi-PE = 17.2% versus 5.3%) (). Supplementary Figure S6b Supplementary Figure S6c

Above results suggested that Bi-PE significantly improved the ability of PE system in targeted deleting large DNA fragment as compared to PE3. Next, to further explore its versatility, we examined if Bi-PE system can insert a small fragment while deleting large fragments from the target loci (fragment replacement). As shown in Figure 3A, we designed two Bi-PE strategies, Bi-PE-2 and Bi-PE-3, both of which encoded the same replacement sequences (the edits). The pegRNAs of Bi-PE-3 contain both the edit and the HA that is homologous to the DNA sequence downstream of the deletion, while the ones of Bi-PE-2 contain only the edits (Figure 3A). Bi-PE-2 strategy was designed to test whether HA is necessary for the targeted replacement.

To assess the editing efficiency of the two Bi-PE and PE3 strategies, we tested their performance in 8 replacement events located on three different loci in the HEK293T cells (Figure 3B). On HEK3 locus, the efficiency of Bi-PE-2 in the deletion of 530-bp fragment while simultaneous insertion of 18-bp fragment (–530 + 18) was lower than that of Bi-PE-3 (4.8% versus 17.3%). While Bi-PE-2 showed significantly lower activity than Bi-PE-3 in the deletion of 654-bp (19.3% versus 45.8%) or 861-bp fragments (14.7% versus 60.8%) (Figure 3C). On other editing, Bi-PE-2 also showed lower activity than Bi-PE-3 (β-Actin –315 + 18, Bi-PE-2: Bi-PE-3 = 40.3%: 44.1%; β-Actin –600 + 18, Bi-PE-2: Bi-PE-3 = 32.3%: 34.6%; β-Actin –1025 + 18, Bi-PE-2:Bi-PE-3 = 36.8%:39.9%; VEGFA –400 + 18, Bi-PE-2:Bi-PE-3 = 8.4%: 48.8%; VEGFA –700 + 18, Bi-PE-2:Bi-PE-3 = 29.4%:60.0%) (Figure 3C). As a control, Cas9 nickase (H840A) coupled with pegRNAs did not produce detectable deletion and replacement, ruling out the possibility that pegRNAs acted as templates for HDR repair to induce such editing events (Supplementary Figure S10). Collectively, these observations demonstrated Bi-PE-3 strategy was more efficient than Bi-PE-2 in fragment replacement (Figure 3D), suggesting that although HA was not necessary for the replacement, but its presence guaranteed the editing efficiency (Figure 3B–D). Then we focused on Bi-PE-3 to test if its efficiency could be further improved by optimizing the HA length. We varied the length of HA from 8-bp to 50-bp. Similar to the results obtained from fragment deletion, we found that the effect of HA length was case-dependent and not strictly predictive (Supplementary Figure S11).

Next, we compared the efficiency of the two Bi-PE strategies to that of PE3s. We found that Bi-PE-3 outperformed both sides of PE3s in seven out of eight editing events, while Bi-PE-2 only outperformed both PE3s in four out of eight events (Figure 3B and C). And on average, Bi-PE-3 achieved a replacement efficiency of 43.9%, which is the highest one among the four strategies (Bi-PE-2:Bi-PE-3:L-PE3:R-PE3 = 23.3%:43.9%:17.2%: 24.7%) (Figure 3D).

Then we focused on Bi-PE strategy and tested whether the editing efficiency was correlated with the length of replacement sequence. We varied the replacement length from ∼10-bp to ∼100-bp. The results showed the editing efficiency seemed not decrease as the length of replacement increase. On VEGFA and β-Actin loci, the efficiencies of ∼100-bp replacement were even slightly higher than those of ∼10-bp replacement (Supplementary Figure S12).

In addition to editing efficiency, we also observed that Bi-PE-2 and Bi-PE-3 produced a lower level of indel than PE3 (Supplementary Figure S13). The most prominent example is the deletion of 861-bp fragments at the HEK3 locus, where Bi-PE-3 produced about 0.19% indels, while Bi-PE-2 produced about 0.72% indels, L-PE3 produced about 12.8% indels, and R-PE3 produced about 1.6% indels (Figure 3E). Overall, Bi-PE-2 and Bi-PE-3 produced less indel than PE3 in most editing events (six out of seven events) (Figure 3F). Together, these results suggested Bi-PE-3 strategy, but Bi-PE-2 outperformed PE3 in both efficiency and accuracy in fragment replacement.

After demonstrating that Bi-PE is more efficient than PE3 in targeted manipulations of large fragments, we next evaluate whether Bi-PE can also outperform PE3 in base conversions. We compared the performance of Bi-PE and PE3 in the edition of single base conversion in HEK293T cells (Supplementary Figure S14a). On FANCF locus (+5 G·C to T·A), the average editing efficiency of Bi-PE-2 and Bi-PE-3 was 9.7% and 12.7%, respectively, while the efficiency of L-PE3 and R-PE3 was 20.7% and 2% respectively (Supplementary Figure S14b). On β-Actin locus (+6 G·C to T·A), Bi-PE-2 and Bi-PE-3 achieved an average efficiency of 21% and 20.5% respectively. And L-PE3 and R-PE3 achieved an average editing efficiency of 27.7% and 3%, respectively (Supplementary Figure S14b). These data suggested that Bi-PE did not outperform PE3 in base conversion editing.

Next, we compared the performance of Bi-PE and PE3 on simultaneous conversion of two bases. The pegRNAs of each direction (L- and R-pegRNA) were designed to convert bases within each corresponding PAMs (Supplementary Figure S14c). On FANCF locus (+5 G·C to T·A, +44 C·G to A·T), Bi-PE-2 and Bi-PE-3 had average editing efficiency of 7.2% and 20.2% respectively. And the average editing efficiencies of L-PE3 and R-PE3 were 13.2% and 4% respectively. On β-Actin locus (+6 G·C to T·A, +44 C·G to A·T), the average editing efficiencies of Bi-PE-2 and Bi-PE-3 were 18.8% and 16% respectively, while the ones of L-PE3 and R-PE3 were 13% and 8.8%, respectively (Supplementary Figure S14d). These comparisons collectively suggested that Bi-PE-3 strategy was generally more efficient than PE3 in simultaneous conversion of two bases. Overall, the improvement of Bi-PE-3 over L- or R-PE3 ranged from 0.8 to 4.1 times (Supplementary Figure S14d).

We then examined if Bi-PE was also efficient in multiplex base conversions. We designed pegRNAs of each direction to simultaneously introduce 5-base pairs conversions on FANCF, β-Actin and RUNX1 loci, and 4-bp conversions on RNF2 and HEXA loci (Figure 4A and B). We observed robust multiplex base conversions across all editing strategies. And similar to the findings in two-base conversions, we found that Bi-PE-3 but not Bi-PE-2 strategy was more efficient than either L- or R-PE3 in multiplex base conversions (Bi-PE-2:Bi-PE-3:L-PE3:R-PE3 = 5.3%:9.3%:7.1%:6.6%) (Figure 4B and C). Noteworthy, in the editing products, there were a small fraction containing imperfect conversions (at least one target bases not being converted). The ratio of imperfect conversion ranged from 0.05% to 13% (Supplementary Figure S15).

Previous study had demonstrated that PE3 strategy was able to insert single LoxP site into the genome. However, the most widely application of Cre/LoxP system, conditional loss- or gain-of-function experiments, require the insertion of paired LoxP sites. To test if Bi-PE strategy could achieve targeted insertion of paired LoxP sites in the same allele, we design a pair of pegRNAs aiming to insert paired LoxP sites with same orientation and flanking a 90-bp segment into the HEK3 locus (Figure 5A). We transfected those pegRNAs together with PE2 into HEK293T cells and determined the presence of LoxP insertions by PCR. For the PCR reaction, we designed paired primers flanking outside the PBS to avoid amplifying randomly integrated LoxP. As shown in Figure 5B, we did detect a band corresponding to the size of double LoxP insertions. To better quantify the efficiency of double LoxP insertions, we cloned the PCR products into plasmids and performed single colony analysis (Supplementary Figure S16a). The analysis revealed that the efficiency of double LoxP insertions was about 13.2% (Figure 5C and Supplementary Figure S16b). Sanger sequencing of these colonies harboring LoxP sites confirmed that these events were actually targeted but not random insertion (Supplementary Figure S16c). Among the alleles harboring double LoxP sites, 72% were accurate and the rest alleles were found to contain fragment deletions between the LoxP sites (Figure 5D, Supplementary Figure S16c and Supplementary Figure S16d). Then we tried to lengthen the segment flanked between LoxP sites to 198-bp (Figure 5B). To facilitate the detection of LoxP insertion, we added EcoRV sequences to each LoxP sites, so that the presence of the insertions could be detected by EcoRV digestion. As shown in Figure 5B, we observed alleles containing single or double EcoRV sites, indicating single or double LoxP insertions respectively. However, the level of the insertions was significantly lower than that observed in 90-bp segment. Single colony analysis revealed that the efficiency of double LoxP sites was 2.9%, in which 52% were accurate (Figure 5C, D and Supplementary Figure S17). When we lengthened the segment to 372-bp, we did not detect obvious LoxP insertions (data not shown). Together, these results suggested that Bi-PE strategy was able to achieve double fragment insertion, but its efficiency attenuated with the lengthening of the segment in-between the insertions.

All data supporting the findings of this study are available in the article and its supplementary figures and tables or can be obtained from the corresponding authors upon reasonable request. Deep-sequencing data are available under BioProject ID PRJNA809555.