Monthly Archives: July 2020

INTRODUCTION

Fusing a deaminase with the Cas9 nickase (nCas9) forms cytosine base editors (CBEs), which enable programmable conversion of cytidine-to-thymidine (C-to-T) mutations within a specific region of the genomic DNA without causing double-stranded breaks (13). CBEs have displayed substantially higher editing efficiency than the conventional Cas9 endonuclease-mediated homology-directed repair method for installing point mutations (4, 5). In addition, recent protein engineering efforts have improved their product purities and efficiencies (6, 7), greatly expanded the genome targeting scope (8), and minimized the undesirable RNA off-target effects (911). CBEs are important genetic tools and could potentially correct more than 5000 pathogenic single-nucleotide polymorphisms (SNPs) associated with human-inherited diseases caused by T-to-C (or G-to-A) mutations (3, 12, 13).

The presence of multiple targets within the CBEs activity window [e.g., the editing window of BE4max is approximately from positions 4 to 8 of the protospacer, counting the protospacer adjacent motif (PAM) as positions 21 to 23] can introduce unwanted bystander editing, resulting in deleterious multiC-to-T conversions (14). Earlier studies have shown that the activity window size can be narrowed using strategies such as modulating the catalytic activity of deaminase (15), using more rigid linkers between Cas9 and deaminase, or deleting nonessential deaminase sequences (16, 17). These approaches can systematically enhance precision for position-dependent single-nucleotide editing irrespective of nearby sequence contexts, although the genome targeting scope might be compromised because of the requirement that the target nucleotide needs to be placed at a specific position relative to an available PAM. Alternatively, sequence context-specific CBE can avoid bystander editing without sacrificing the activity window size (3). The engineered APOBEC3A (A3A) enzyme preferentially deaminates in the TCR motif (target C underlined), which has been exploited for more precise base editing, and the resulting eA3A-BE3 base editor exhibited high on-target precision with minimized bystander editing (18). However, in the most challenging case, when editable Cs are located consecutively within the activity window, especially in the case of CC dinucleotides when a bystander C is located right upstream of the target C, the existing CBEs nonselectively edit both of the Cs. Nearly 38% of the human pathogenic SNPs that are caused by T-to-C disease point mutations lie in the context of CC, followed by AC (29%), GC (21%), and TC (13%) (see data file S1) (1, 12), necessitating the development of new CBEs that can precisely discriminate between the target and bystander Cs.

Various APOBEC enzymes in vertebrates mediate defense against infections from retroviruses or retrotransposons by deaminating C to U in the viral complementary DNA (cDNA) (19, 20), suggesting that these cytosine deaminases could have unique preferences for particular sequence motifs to distinguish DNA sequences from the native host (2123). In this study, we identified human APOBEC3G (A3G) as a candidate for developing sequence-specific BEs in multiple C contexts. We characterized and engineered A3G-BE variants to efficiently edit a single C at various endogenous sites in human embryonic kidney293T (HEK293T) cells. By introducing mutations that improve catalytic activity, solubility, and overall protein scaffold, we obtained and characterized three novel variants (A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14) that exhibit high editing efficiencies and precision in the context of the CC motif. A3G-BE variants have broader activity windows than BE4max that could expand the targeting scope for precision base editing. We also demonstrated that these variants could efficiently and precisely generate or correct mutated alleles associated with the known pathogenic phenotypes, illuminating A3G-BEs potential application in treating human genetic diseases. Last, we performed whole-genome sequencing (WGS) by using the most recently developed genome-wide off-target analysis by two-cell embryo injection (GOTI) method to detect DNA off-targets and used RNA sequencing (RNA-seq) to examine the RNA off-targets of cells treated with A3G-BE5.13. Our results showed that the most active A3G-BE5.13 induces baseline levels of the genome- and transcriptome-wide off-target mutations, suggesting high editing fidelity for future clinical applications.

Previous studies have demonstrated that A3G predominantly deaminates the third C in the 5-CCC-3 motif of a single-stranded DNA (ssDNA) substrate (24). To test whether this motif preference could be preserved when A3G is fused to nCas9 as A3G-BE, we replaced the rAPOBEC1 deaminase domain of BE4max with the full-length, human codon-optimized A3G to construct A3G-BE2.1 (6). Since it has been reported that the N-terminal domain (NTD) could mediate aggregation of A3G monomers to impede A3Gs mobility (25) and because the C-terminal domain (CTD) of A3G is sufficient for deamination activity in vitro (26, 27), we therefore truncated the NTD of A3G to construct A3G-BE4.4, which only contains the CTD of A3G (Fig. 1A). HEK293T cells were then transfected with plasmids expressing BE4max, A3G-BE2.1, and A3G-BE4.4 with single-guide RNAs (sgRNAs) targeting EMX1 #1 and FANCF #a3 sites, which contain dinucleotide Cs (C5 and C6 of EMX1 #1 and C6 and C7 of FANCF #a3) within the canonical BE4max activity window. We extracted the genomic DNA after 72 hours and amplified the target regions for high-throughput sequencing (HTS). Analysis of the C-to-T editing efficiencies of the dinucleotide Cs showed that A3G-BE2.1 and A3G-BE4.4 edited 21 to 42% of the cognate Cs (C6 of EMX1 #1 and C7 of FANCF #a3) but only 1 to 3% of the bystander Cs (C5 of EMX1 #1 and C6 of FANCF #a3), while BE4max edited 47 to 62% of both the cognate and bystander Cs without obvious selectivity (Fig. 1B). No significant difference was observed between A3G-BE2.1 and A3G-BE4.4 for editing efficiencies of the cognate Cs, suggesting that the CTD itself adequately determines the enzymatic activity and sequence specificity of A3G.

(A) Schematic showing the protein architecture of base editors. BE4max is used to replace the rAPOBEC1 with either full-length (NTD + CTD) or CTD-only human A3G to construct A3G-BE2.1 or A3G-BE4.4, respectively. Linkers between functional domains are shown as horizontal blue lines. NLS, nuclear localization signal; UGI, uracil glycosylase inhibitor. (B) C-to-T editing efficiency and specificity of A3G-BE2.1 and A3G-BE4.4 at EMX1 #1 and FANCF #a3 sites bearing the CC motif (red). (C) Nine endogenous sites of HEK293T bearing either CC or CCC motif (red) within the canonical BE4max activity window. Each PAM and the sequence motif identifying the nucleotides at +1 and 2 positions from the target C (underlined) are shown. (D) C-to-T editing efficiency and specificity of BE4max and A3G-BE4.4 at the endogenous sites listed in (C). Bar figures of (B) and (D) show means and error bars representing SD of n = 2 and n = 3 independent biological replicates performed on different days, respectively. Statistical significance shown on top of each bar using two-tailed Students t test compares to editing efficiency of the preceding bystander C of the same BE. For example, t test was performed between the BE4max editing efficiencies of C8 and C9 at DMD #1 site. ns (not significant), *P < 0.05, ***P < 0.001, ****P < 0.0001.

Because the wild-type A3G in nature preferentially deaminates in the C2C1C0A+1 sequences of the HIV-1 genome (28), we next examined whether nucleotides at positions 2 and +1 of the cognate C0 also affect the base editing efficiency and specificity. We tested BE4max and A3G-BE4.4 at nine different loci containing the dinucleotide Cs motif with different combinations of nucleotides placed at the 2 and +1 positions (N2C1C0D+1, where D denotes A, T, and G) (Fig. 1C). HTS analysis confirmed that A3G-BE4.4 showed selective editing of the cognate Cs across all the sites. At six of the nine sites, A3G-BE4.4 reached at least 79% of the editing efficiencies of the cognate Cs of those of BE4max (Fig. 1D). Notably, at DMD #1 site, which contains the ACCA motif, similar to the native CCCA, and harbors the cognate C9 outside the canonical BE4max activity window, A3G-BE4.4 induced threefold higher editing of the cognate C9 compared to BE4max. However, although being selective, A3G-BE4.4 displayed very low cognate C editing efficiencies with only 13, 6, and 3% C-to-T conversion rates at the remaining three PPP1R12C #a3, BCS1L #1, and EMX1 #a18 sites, respectively. These results may have occurred because the wild-type A3G disfavors deamination of certain motifs such as GCC, suggesting that the motif-dependent deamination activity of A3G could influence the efficiency of the selective base editing (29). We then quantified the specificity by dividing the editing efficiency of the cognate C by that of bystander C (cognate-to-bystander editing ratio). Across the nine sites, A3G-BE4.4 recorded the editing ratios ranging from 11 to 290, while BE4max achieved a maximum ratio of 6 at EMX1 #a18 and less than 2 at all other sites (fig. S1A). Non-T by-products generated by A3G-BE4.4 averaged slightly higher than BE4max in most of the sites (fig. S1B), consistent with previous observations that generally lower product purity is generated by editing of a single C versus multiple Cs (6). A3G-BE4.4 also showed significantly fewer indels than BE4max at three of the nine sites (HEK3 #1, HEK4 #a1, and EMX1 #a3), supporting an earlier study suggesting that single-nucleotide and multiple base editing have no significant correlation in terms of indel generation (fig. S1C) (18). Together, these results indicated that A3G-BE4.4 has sufficient editing efficiency to precisely edit the second C in the sequence context of 5-CC-3 dinucleotides.

Given the relatively low base editing efficiencies of A3G-BE4.4 for cognate Cs observed from the PPP1R12C #a3, BCS1L #1, and EMX1 #a18 sites, we envisioned that the wild-type A3G-CTD activity could be further improved. We devised three subsets of mutations that could be introduced into the A3G-CTD of A3G-BE4.4 based on different possible functional effects, including set A (P200A + N236A + P247K + Q318K + Q322K) to improve catalytic activity, set B (partial replacement of A3Gs loop 3 with A3As, that is H248N + K249L + H250L + G251C + F252G + L253F + E254Y) to increase ssDNA binding affinity, and set C (L234K + C243A + F310K + C321A + C356A) to enhance protein solubility (Fig. 2A and fig. S2A) (27, 30, 31). We first introduced set A to A3G-BE4.4 to construct A3G-BE5.1 and introduced sets B and C mutations to A3G-BE5.1 to construct A3G-BE5.3 and 5.4, respectively (fig. S2B and table S1). To further maximize A3Gs potential deamination activity, two additional mutations, T311A + R320L, were introduced to A3G-BE5.3 to construct A3G-BE5.10 (fig. S2B and table S1) (27, 31). We tested A3G-BE4.4, A3G-BE5.1, A3G-BE5.3, A3G-BE5.4, and A3G-BE5.10 at EMX1 #1 and FANCF #a3; all of the further improved mutants showed substantially higher editing efficiency than A3G-BE4.4 did on both the cognate Cs and the bystander Cs (Fig. 2B and fig. S2C). Notably, when the loop 3 of A3G was partially replaced with A3As by set B mutations, A3G-BE5.3 and A3G-BE5.10 exhibited substantial loss of the motif preference, and both Cs were efficiently edited. Structural alignment of the wild-type A3A, wild-type A3G, and the A3G containing the set A mutations, among which P247K lies in loop 3, showed that loop 3 of the wild-type A3A, as well as the A3G with set A mutations, exhibits greater proximity to the ssDNA substrate, suggesting that the observed increase in the editing efficiency and relaxation of the sequence specificity might be partly due to the stronger nonspecific binding to the ssDNA substrate (fig. S2D).

(A) Set of residue mutations of A3G for improving catalytic activity (set A), ssDNA binding (set B), and protein solubility (set C) listed on each row. Counting of the residue number starts with the first residue of the original full-length A3G. (B) Screening of A3G-BE mutants at EMX1 #1 site to determine variants with enhanced editing efficiency and retained sequence specificity. C-to-T editing efficiencies are represented as bidirectional bars with values for the cognate C6 (blue) on the right and the bystander C5 (red) on the left. (C) An enlarged view of the interactions of Tyr315 (green sticks) with the ssDNA substrate (yellow sticks). The hydrogen bond between the 5 phosphate group of the DNA backbone and the hydroxyl group of Tyr315, and the interaction between the rings of the target cytidine (dC0) and Tyr315 are represented as dashed lines. (D) C-to-T editing efficiency and specificity of A3G-BE5.13 and A3G-BE5.14 at three endogenous sites previously poorly edited by A3G-BE4.4. Panels (B) and (D) show means and error bars representing SD of n = 3 independent biological replicates performed on different days. For (D), statistical significance shown on top of each bar using two-tailed Students t test compares to editing efficiency of the preceding bystander C of the same BE. ns (not significant), **P < 0.01, ***P < 0.001, ****P < 0.0001.

We hypothesized that modulating the nonspecific binding to DNA could restore the sequence specificity. Using structure-guided analysis, Tyr315 of A3G was identified as a key residue that interacts with both the DNA backbone and the target C (Fig. 2C). We speculated that changing Tyr315 to Phe, which lacks only the hydroxyl group from Tyr, could remove the hydrogen bond with the 5 phosphate group of ssDNA while maintaining the - interaction with the target C. We introduced Y315F to A3G-BE5.1, A3G-BE5.3, A3G-BE5.4, and A3G-BE5.10 to construct A3G-BE5.12, A3G-BE5.13, A3G-BE5.14, and A3G-BE6.11, respectively (fig. S2B and table S1). Y315W (to provide steric hindrance) and Y315L (to remove both the hydrogen bond and the - interaction) were also introduced into A3G-BE5.10, resulting in A3G-BE6.16 and A3G-BE6.17, respectively. Additional mutations to further reduce the nonspecific binding, including N244Q, S286A, and R313A, were introduced into A3G-BE6.11 to construct A3G-BE6.18, A3G-BE6.19, and A3G-BE6.20, respectively. Last, we reverted the replacement of the A3Gs loop 3 with A3As from A3G-BE6.11 to construct A3G-BE6.21 (fig. S2B and table S1). Testing all the above variants at EMX1 #1 and FANCF #a3 showed that A3G-BE6.11 induced higher selectivity than A3G-BE5.10 by moderately reducing editing of the bystander Cs. At the same time, A3G-BE6.16 and A3G-BE6.17 displayed markedly reduced editing efficiencies of the cognate Cs, even below those of A3G-BE4.4 (Fig. 2B and fig. S2C). Although all A3G-BE6.18, A3G-BE6.19, A3G-BE6.20, and A3G-BE6.21 showed improved editing ratios of the cognate to bystander Cs compared with A3G-BE6.11, their cognate C editing efficiencies did not outperform A3G-BE4.4. Nevertheless, A3G-BE5.13 and A3G-BE5.14, both of which contain Y315F, exhibited greater cognate C editing efficiency than A3G-BE4.4 did and demonstrated appreciable restoration of the sequence specificity (Fig. 2B and fig. S2C).

We further tested A3G-BE5.13 and A3G-BE5.14 at the PPP1R12C #a3, BCS1L #1, and EMX1 #a18 sites at which the editing efficiencies of A3G-BE4.4 were previously low (Fig. 1D). HTS analysis showed that both A3G-BE5.13 and A3G-BE5.14 gained superior editing efficiency for the cognate Cs as compared to A3G-BE4.4 (Fig. 2D). Moreover, bystander editing of A3G-BE5.13 and A3G-BE5.14 remained substantially lower than that of BE4max, resulting in significant improvement of base editing efficiency while maintaining the specificity. Together, these results suggested that through rational engineering, A3G-BE5.13 and A3G-BE5.14 overcame the low editing drawbacks of A3G-BE4.4 on discrete sequence contexts.

To comprehensively understand the capability of sequence-specific base editing of A3G-BE5.13 and A3G-BE5.14, we tested them at eight other endogenous sites with the dinucleotide Cs motif positioned across the whole protospacer. HTS analysis confirmed that all A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 selectively edited the second C within the CC motifs across all the sites. The cognate-to-bystander editing ratios were calculated to be up to 186 (A3G-BE5.14 at EMX1 #c16 site), while BE4max either nonselectively edited both Cs or failed to perform outside its canonical activity window (Fig. 3A and fig. S3A). At BCS1L #6 and RNF2 #2 sites, which contained the cognate Cs at positions 12 and 15 of the protospacers, respectively, highly efficient and selective editing for the cognate Cs were only observed when using A3G-BE5.13 and A3G-BE5.14, while A3G-BE4.4 and BE4max did not yield efficient C-to-T editing (Fig. 3A). Notably, at both BCS1L #6 and RNF2 #2 sites, the single C located at the fifth position was not efficiently edited by all A3G-BE variants, probably due to lack of the CC dinucleotide sequence context. Both A3G-BE5.13 and A3G-BE5.14 displayed efficient editing up to C15 of RNF2 #2 but not C18 of FANCF #2 (Fig. 3A). For the two cognates Cs existing in EMX1 #b1 (C7 and C15) and FANCF #2 (C6 and C10) sites, A3G-BE4.4 efficiently edited only the ones residing closer to the 5 end (C7 of EMX1 #b1 and C6 of FANCF #2), indicating a possible narrower window size compared with A3G-BE5.13 and A3G-BE5.14. The lowest cognate-to-bystander editing ratios for all three A3G-BEs occurred at EMX1 #b1, which bears three consecutive Cs of the CCCA motif, suggesting that the requirement for single-nucleotide editing within more than two consecutive Cs might need to be more stringent. We did not find a consistent trend in the product purity following the treatment of all BEs, which might be due to the discrepancies among distinct properties of BEs that have different activity windows, deamination activities, and sequence specificities (fig. S3B) (6). We also observed indels being generated with varying frequencies across the sites without apparent correlation among BEs (fig. S3C).

(A) Heat maps are showing average C-to-T editing efficiencies of n = 3 independent biological replicates of BE4max, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 at eight endogenous sites containing the preferential CC or CCC motif across the whole region within the protospacers. The cognate Cs predicted to be preferentially editable by A3G-BEs are indicated by the black triangles. (B) Average C-to-T base editing frequencies at each protospacer position from the six poly-C endogenous sites shown in fig. S4. Bidirectional arrows in between vertical dashed lines show the base-editable ranges within the protospacer region by the indicated A3G-BEs (C) Schematic representation of the activity window sizes of A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14, with NGG PAM shown as positions 21 to 23. Standard, light, and near-transparent green represent the predicted relative base editing activity within the approximate regions of the protospacer.

To determine the sizes of the activity window of A3G-BEs, we tested A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 at six endogenous genomic sites, which contain consecutive Cs within the protospacer, and analyzed their C-to-T editing efficiencies. For all the tested sites, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 revealed consistent and broad base editing activity window but differed mainly in their relative editing efficiencies, for which A3G-BE5.13 showed the highest followed by A3G-BE5.14 and A3G-BE4.4 (fig. S4). We observed that A3G-BE4.4 displayed comparatively lower editing efficiencies around positions 8 to 15 compared with those in positions 5 to 7 at four sites (VEGF #2, EMX1 PolyC #1, EMX1 PolyC #1, and HEK4 PolyC #1), suggesting that editing toward the 3 end of the protospacer, although targetable, could have lower editing efficiency. Next, we compared the average editing frequencies of Cs at each protospacer position from all the six sites. We found that the activity windows of A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 span from positions 5 to 15, 3 to 15, and 4 to 15 of the protospacer, respectively (Fig. 3, B and C). Together, these data indicated that A3G-BEs enable sequence-specific editing with broadened targeting ranges.

Given that the preferential motif of A3G extends to three consecutive Cs, C2C1C0, we hypothesized to test whether the sequence specificity could be maintained when the middle C, the 1 position of the target, is altered to other nucleotides. To assess this possibility, we selected five endogenous sites that contained a T or A at the 1 position (C2TC0 or C2AC0 motifs) and, now, counting editing of the C at 2 position to be the bystander incidence (fig. S5A). We transfected HEK293T with BE4max, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 with sgRNAs targeted to the selected sites and performed HTS. After quantifying the C-to-T editing efficiencies, we found that, compared to BE4max, A3G-BEs indeed displayed significantly higher editing of the cognate Cs over bystander Cs within these altered sequence contexts (fig. S5B). A3G-BE5.14, among other A3G-BEs, exhibited the highest specificities (up to 89 cognate-to-bystander editing ratio) at four of the five sites (fig. S5B). While A3G-BE5.13 and A3G-BE5.14 have comparable or higher cognate C editing efficiency than BE4max, A3G-BE4.4 editing efficiencies of the cognate Cs were below 9% at four of the five sites, indicating that the absence of C at the 1 position might restrain A3G-BE4.4 from efficient editing. In addition, we observed relatively higher bystander C2 editing from A3G-BE5.13 at HEK3 #b1 and HEK3 #b2 sites, which contained T immediately upstream of the bystander C2. Since C and T are structurally similar compared to the other two nucleotides, we speculated that this sequence context might be more prone to bystander editing. These findings indicated that A3G-BEs could selectively edit a target C in the CTC and CAC motifs and therefore can further expand the targeting scope for precision base editing in broader sequence contexts.

To test A3G-BEs in disease-relevant contexts, we sought to precisely generate SNPs of reported human pathogenic diseases (32). Three genetic variants caused by C-to-T (or G-to-A) substitution in which the wild-type sequences lie within the preferential 5-CC-3 motif of A3G-BEs were selected, including cystic fibrosis (model 1), hypertonic myopathy (model 2), and transthyretin amyloidosis (model 3) (Fig. 4A). Individual sgRNAs targeted to these disease-associated sites were constructed and cotransfected into HEK293T with BE4max, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14. Genomic DNA was harvested after 72 hours and prepared for HTS to quantify the percentage of alleles perfectly modeled and of those that were imperfectly modified because of bystander editing. Direct comparison with BE4max of the modified allele frequencies demonstrated that A3G-BEs induced a substantially higher proportion of perfectly modified alleles for all three models (Fig. 4B). Despite the previous observations in which A3G-BE5.13 displayed more relaxed base-editing sequence specificity among other selected A3G-BEs, it achieved the highest percentage here of the perfectly modified alleles for hypertonic myopathy (model 2) (36%). For transthyretin amyloidosis (model 3), in which the target C lies at position 11 of the protospacer, all A3G-BEs produced the desired allele with high efficiencies (>35%), while BE4max failed to edit the target C (<0.1%) because of its inability to edit outside its activity window (fig. S6A). As a result, A3G-BE5.14 accomplished 613-fold higher correct modeling of transthyretin amyloidosis than BE4max did, highlighting the advantage of precise editing with an expanded activity window. Similarly, for cystic fibrosis (model 1), all A3G-BEs induced more than 50% of the perfectly modified alleles, while BE4max averaged 0.6%.

(A) Sequences of the protospacers and PAMs (blue) for model 1 (cystic fibrosis), model 2 (hypertonic myopathy), and model 3 (transthyretin amyloidosis). Position of the disease-relevant C>T (or G>A) point mutations are red and indicated by black triangles shown with the nucleotide numbers within the disease-associated genes. (B) Percent of alleles modified to the indicated genotypes following the treatment of BE4max and A3G-BEs for generating the three models presented in (A). (C) Sequences of the protospacers and PAMs (blue) for correction 1 (hereditary pyropoikilocytosis), correction 2 (cystic fibrosis), and correction 3 (holocarboxylase synthetase deficiency), bearing T>C (or A>G) point mutations for which the positions are indicated with black triangles showing the nucleotide numbers within the disease-associated genes. (D) Percent of alleles modified to the indicated genotypes following the treatment of BE4max and A3G-BEs for correcting the three disease-associated variants presented in (C). Panels (B) and (D) show means and error bars representing SD of n = 3 independent biological replicates performed on different days. Statistical significance shown on top of each bar using two-tailed Students t test compares to the percentages of perfectly generated/corrected alleles by BE4max. ns (not significant), *P < 0.05, ****P < 0.0001.

Next, to examine the therapeutic applicability of A3G-BEs, we selected three reported human pathogenic SNPs caused by T>C (or A>G) mutations, which can be preferentially targeted by A3G-BEs, including hereditary pyropoikilocytosis (correction 1), cystic fibrosis (correction 2), and holocarboxylase synthetase deficiency (correction 3) (Fig. 4C) (32). We generated three HEK293T lines containing 200 base pair (bp) of each disease-relevant sequence integrated into the genome (see Materials and Methods). Codelivery of the BEs and sgRNAs targeted to the disease-associated sites and analysis of the HTS data to quantify the perfectly corrected alleles verified that all A3G-BEs significantly outperformed BE4max by a minimum of threefold in corrections 1 and 2. In addition, A3G-BE4.4 exclusively induced more than 50% of perfectly corrected alleles among other BEs and accomplished 6496-fold higher correction than BE4max in correction 3 (Fig. 4D). Correction 3, in which the protospacer contained two motifs preferred by A3G-BEs, CC and CTC, interfered with the precise single C-to-T editing by A3G-BE5.13 and A3G-BE5.14 and resulted in substantial dual C editing due to their wide activity window sizes and high efficiencies (fig. S6A). Collectively, these comparisons indicated that A3G-BEs have higher targeting precision than BE4max for reversing pathogenic SNPs within their preferred sequence contexts.

We further investigated the editing efficiency of A3G-BEs in therapeutically more relevant cell types, including the induced pluripotent stem cells (iPSCs) and human embryonic stem cells (hESCs). We nucleofected iPSC and ESI-017 hESC lines with BE4max, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 with sgRNA targeting the hypertonic myopathy (model 2)associated site and performed clonal expansion of the successfully nucleofected cells for 10 to 14 days before analysis. In the iPSCs, analysis of the sequencing chromatograms revealed that A3G-BEs more efficiently edited the cognate C7 than the bystander C6, which were 10, 46, and 34% at C7 and 2, 15, and 5% at C6 by A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14, respectively. In contrast, BE4max nonselectively edited both Cs, 39 and 50% at C7 and C6, respectively (fig. S6B). The observed trend was consistent with the ESI-017 hESCs (fig. S6C), indicating the utility of A3G-BEs to serve as important tools to precisely model genetic variants in clinically relevant cell types.

Several CBEs were reported to generate genome- and transcriptome-wide off-target editing, which became a major concern for their clinical uses (9, 10, 33, 34). We then examined the propensity of A3G-BEs to cause deamination on off-target loci by performing orthogonal R-loop assay (35). Briefly, the nuclease-dead SaCas9 (dSaCas9) sgRNA complex creates an R-loop, recapitulation of a stochastic ssDNA exposure in the genome, at a DNA locus unassociated with the on-target site. Base editing mediated by cytosine deaminase in the off-target R-loop independently of SpCas9 nickase and its sgRNA is detected via targeted HTS (fig. S7A). We assessed six off-target loci (Sa #1 to #6 sites) by cotransfecting SpCas9-derived CBE (BE4max or A3G-BEs), on-target SpCas9 sgRNA, dSaCas9, and off-target dSaCas9 sgRNA into HEK293T (table S2). For the on-target editing at EMX1 #1 site, specificities and efficiencies of all CBEs exhibited consistent results with our previous observations without the dSaCas9 system (fig. S7B). We then quantified the editing activities of 18 cytosines within those six off-target loci. We found that A3G-BEs show substantially reduced off-target editing compared with BE4max, except at those cytosines lying within the 5-CC-3 motif, e.g., C10 and C15 at Sa #2, C11 at Sa #5, and C8 at Sa #6 sites (fig. S7C). A3G-BE4.4 showed no significant off-target editing at 10 of the 18 cytosines. A3G-BE5.13 induced higher off-target mutations than both A3G-BE4.4 and A3G-BE5.14 at all cytosines but still significantly lower than BE4max at 11 of the 18 cytosines. Together, these results suggested that A3G-BEs generally exhibit lower propensities to cause Cas9/sgRNA-independent off-target mutations. We then selected A3G-BE5.13, the most active variant among the three selected ones, for further whole-genome off-target characterization.

To comprehensively understand the capability of A3G-BE5.13 to generate Cas9/sgRNA-independent DNA off-target mutations, we performed WGS using the most recently established GOTI method (33). A blastomere of two-cell embryos derived from Ai9 (CAG-LoxP-Stop-LoxP-tdTomato) mice was injected with Cre mRNA, A3G-BE5.13 mRNA, and sgRNA. At embryonic day 14.5 (E14.5), progeny cells were FACS (fluorescence-activated cell sorting)sorted on the basis of tdTomato expression, and WGS was separately performed for the resulting two cell populations with (tdTomato+) and without (tdTomato) the tdTomato expression (Fig. 5A) (33). Using the WGS data obtained from the tdTomato sample as the reference, single nucleotide variants (SNVs) for the tdTomato+ sample were called via three different algorithms, and the overlapping SNVs detected from all the three algorithms were counted as the true off-target variants. Notably, we detected only 17 and 24 SNVs per embryo in each replicate from those treated by A3G-BE5.13, similar to the spontaneous mutation rate found from embryos delivered with Cre alone, as compared to the average of 283 SNVs per embryo by BE3 as previously detected (Fig. 5B and fig. S8A) (33). The mutation patterns of A3G-BE5.13 only showed a slight bias toward C-to-T or G-to-A compared with BE3 (Fig. 5C). We also tested the on-target Tyr-C site used in the GOTI experiments, which harbors both C3C4 and C4TC6 motifs. The WGS results showed that the editing only happened at the C6 in the C4TC6 motif, which is consistent with our previous data that A3G-BEs could selectively edit a target C in the CTC motif. (fig. S8B). Collectively, these data indicated that A3G-BE5.13 induces minimum DNA off-target SNVs across the genome while maintains highly efficient and selective editing at the on-target position.

(A) Scheme of the experimental workflow of GOTI. (B) Comparison of the total number of detected DNA off-target SNVs using the GOTI method. The number of SNVs identified in Cre-, BE3-, and A3G-BE5.13treated embryos were 14 12 (SD; n = 2), 283 32 (SD; n = 6), and 20 5 (SD; n = 2), respectively. (C) Distribution of DNA mutation types in each group. (D) Scheme of the experimental workflow of identifying transcriptome-wide off-target SNVs through RNA-seq. (E) Comparison of the total number of detected RNA off-target SNVs. The number of SNVs identified in nCas9-, BE4max-, A3G-BE5.13treated cells were 2669 712 (SD; n = 2), 198,688 37,775 (SD; n = 2), and 1410 39 (SD; n = 2), respectively. (F) Distribution of RNA mutation types in each group. For (C) and (F), the number in each cell indicates the percentage of a certain type of mutation among all mutations. For (B) and (E), each data point represents independent biological replicates performed on different days.

Last, we characterized the transcriptome-wide off-target effect of A3G-BE5.13. We transfected HEK293T with sgRNA and nCas9, BE4max, or A3G-BE5.13 encoded in plasmid as cotranslational P2A fusion to green fluorescent protein (GFP). After 48 hours, we sorted cells with the top 5% GFP signal to isolate the high-expression population (Fig. 5D). We first confirmed the robust on-target efficiency of DNA editing by BE4max and A3G-BE5.13 in these cells using HTS (fig. S8C). We then performed RNA-seq and analyzed the sequencing data to call SNVs in each replicate sample according to the method described previously (10). Our results showed that the engineered A3G-BE5.13 did not induce significant RNA SNVs as compared to the control treated by the nCas9 (Fig. 5E). However, BE4max caused a substantial amount of off-target mutations, in line with the previous studies of the wild-type rAPOBEC1-based CBEs (911). Distribution of mutation types of the detected SNVs of A3G-BE5.13 was similar to that of the nCas9 control, indicating a minimum disturbance on the transcriptome despite the high expression of intracellular A3G-BE5.13 proteins (Fig. 5F). These results further demonstrate that the A3G-BEs developed in this study are with high precision and markedly reduced RNA editing activity (9, 10) and indicate that A3G-BE5.13 could serve as a promising CBE variant with high fidelity and minimum risk of off-target effects.

Here, we developed and characterized three new base editors using the A3G deaminase that is capable of recognizing the unique natural motif of CCCA. A3G-BE4.4 displays considerable editing efficiency and selectivity when the target motif lies within around positions 5 to 11 of the protospacer. In most of the sites, A3G-BE4.4 exhibited remarkable sequence specificity by discriminating between two consecutive Cs. However, we also observed that A3G-BE4.4 editing efficiency was poor at certain sites, probably due to the presented motifs being disfavored by the wild-type A3G and its naturally moderate catalytic activity, which could be improved by our engineered A3G-BE5.13 and A3G-BE5.14 variants (36). Both A3G-BE5.13 and A3G-BE5.14 displayed high efficiency across broader activity windows, from positions 4 to 15, with slightly relaxed CC selectivity. An initial screening of these three A3G-BEs could be conducted to determine which one performs the best for the selective editing of a single desired C.

We estimated the scope of base-editable disease variants that could be corrected by using A3G-BEs. Among the total of 1515 pathogenic SNPs identified within the BEable-GPS (Base Editable prediction of Global Pathogenic-related SNVs) entries (12), 61% (929 of 1515) were found to lie within the CC or CNC sequence context preferred by A3G-BEs (18). We then identified 540 human pathogenic SNPs that could be precisely correctable by our A3G-BEs, occupying 36% of the total number (see data file S1). Manual filtering was conducted to ensure that neighboring bystander Cs within the activity window did not exist along with the target motif of A3G-BEs. This indicates that our engineered A3G-BEs greatly expand the number of precisely targetable genetic variants for potential therapeutic applications.

WGS and RNA-seq analysis suggested that our A3G-BEs variants induce minimum levels of both DNA and RNA off-target SNVs. A3Gs intrinsically high sequence specificity could reduce the probability of deaminating Cs other than its preferential motif. Our orthogonal R-loop assay showed that A3G-BEs exhibit a greater propensity to edit cytosines lying within the CC motif (fig. S7C). Apart from this reason, an earlier study indicated that mutations in the conserved zinc-coordinating, or catalytic, residues of either the NTD or CTD of the full-length A3G nearly abolished its capability to edit RNA and demonstrated that both domains are essential for optimal RNA editing (37). We speculate that the high fidelity of our engineered A3G-BEs could be due to the lack the NTD so that their ability to cause mutations in the transcriptome might be impaired (Fig. 5, D to F). These findings greatly mitigate the concerns about the off-target issues associated with A3G-BEs, showing great potential for their future therapeutic applications.

It is imperative that we develop genome editing tools that have the ability to produce anticipated results with the highest probability with minimum errors. Bystander editing is a major factor giving rise to imprecision, a limitation that should be improved for future clinical usage. Our engineered A3G-BEs here that recognize a specific CC motif could offer a toolkit to precisely edit a target C. These toolkits, if expanded, could allow versatile and precise editing of single nucleotides from various other distinct motifs. We envision that the continued development of novel base editing technology could facilitate the precise conversion of cytosines and treatment of human genetic diseases.

HEK293T cells (American Type Culture Collection, CRL-3216) were cultured in the T-75 flask (Corning) using high-glucose Dulbeccos modified Eagles medium (DMEM) with GlutaMAX and sodium pyruvate (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific) and 1 penicillin-streptomycin (Thermo Fisher Scientific) at 37C with 5% CO2. Upon reaching 80 to 90% confluency, cells were dissociated using TrypLE Express (Life Technologies) and passaged at a ratio of 1:3. Cells were verified mycoplasma-free using a mycoplasma detection kit (abm). ESI-017 hESCs (ESI BIO, CVCL_B854) and iPSCs (Coriell Institute, AICS-0058-067) were maintained in mTeSR1 (STEMCELL Technologies) in tissue culture dish coated with Matrigel (1:200; Corning). Dispase (STEMCELL Technologies) was used for routine passage. To perform nucleofection, a single-cell suspension was prepared using Accutase (Innovative Cell Technologies). The pluripotency of those cells was confirmed via staining of Oct4, Sox2, and Nanog. Both ESI-017 and iPSC lines were routinely tested for mycoplasma contamination and found negative.

A3G-BE2.1 was constructed by amplifying the BE4max plasmid (Addgene) outside the rAPOBEC1 region and In-Fusion cloning (Takara) with the synthesized human codon-optimized A3G fragment (Integrated DNA Technologies). Deletion of the NTD of A3G to construct A3G-BE4.4 was performed by polymerase chain reaction (PCR) amplification of A3G-BE2.1 outside the NTD region using Q5 High-Fidelity 2X Master Mix (New England Biolabs) and recloning the linearized fragment. Sets of mutations introduced into A3G-BE variants for enhancing editing efficienciesincluding A3G-BE5.1, A3G-BE5.3, A3G-BE5.4, and A3G-BE5.10were constructed using gBlocks (Integrated DNA Technologies) that contain the desired mutations and cloned with the remaining backbone of the A3G-BE4.4 plasmid. Other variants for introducing individual mutations, including Y315F, were constructed by site-directed mutagenesis using the general PCR method. Gibson assembly was used to attach P2A-GFP fragment to the C-terminal ends of nCas9, BE4max, and A3G-BE5.13 for the RNA-seq experiment that requires sorting of the transfected cells with the top 5% GFP signal. Similarly, the P2A-PuroR fragment was attached to the C-terminal ends of BE4max, A3G-BE4.4, A3G-BE5.13, and A3G-BE5.14 through Gibson assembly to select puromycin-resistant cells after nucleofection of iPSCs and hESCs. All assembled constructs were transformed into Stellar competent cells (Takara). Plasmids were extracted using either the QIAprep Spin Miniprep Kit (Qiagen) or the ZymoPURE II Plasmid Midiprep Kit (Zymo Research), and concentrations were measured using NanoDrop One (Thermo Fisher Scientific). sgRNAs were constructed by using the previous method (38). Briefly, a pair of primers for top and bottom strands encoding the 20-bp target sequence were 5 phosphorylated using T4 polynucleotide kinase (New England Biolabs) and annealed by heating the oligos to 95C and cooling down to room temperature at 5C/min1. The mixture was diluted 1:25 using water and ligated into a sgRNA expression vector using T4 DNA ligase (New England Biolabs) and BsaIHF v2 (New England Biolabs) following the manufacturers instructions.

The HEK293T stable cell line was constructed by cloning a 200-bp fragment of disease-associated gene upstream of an EF1 promoter to drive the expression of the puromycin-resistant gene in a lentiviral vector. The single-base mutation of a disease-associated gene was inserted by PCR and In-Fusion cloning (Takara). The lentiviral vector was transfected into HEK293T cells in a 24-well plate (Olympus) at 80 to 90% confluency. For each well, 288 ng of the plasmid containing the vector of interest, 72 ng of pMD2.G, and 144 ng of psPAX2 were transfected using 1.0 l of Lipofectamine 2000 and 25 l of Opti-MEM I reduced serum medium (Life Technologies). Viral supernatant was harvested 48 hours after transfection, filtered with a 0.45-m polyvinylidene difluoride filter (Millipore), and then serially diluted to add into a 24-well plate cultured with 5 104 HEK293T cells per well. After 24 hours, cells transduced with lentivirus were split into new plate wells supplemented with puromycin (3 g/ml1). Seventy-two hours after the puromycin selection, cells were harvested from the well with the fewest surviving colonies to ensure single-copy integration and were then further cultured for expansion.

Transfection and extraction of the genomic DNA were adopted from the previous method (7). Briefly, HEK293T cells were counted using Countess II FL (Thermo Fisher Scientific) and plated into a poly-d-lysinecoated 48-well plate (Corning) under 250 l of the cell culture medium with a density of 4.5 104 cells per well. After ~16 hours, cells were transfected using 1.2 l of Lipofectamine 2000 (Thermo Fisher Scientific) with 750 ng of base editor, plasmid and 250 ng of sgRNA plasmid per well following the manufacturers protocol. For orthogonal R-loop assay, 300 ng of BE, 300 ng of dSaCas9, 200 ng of SpCas9 sgRNA, and 200 ng of SaCas9 sgRNA plasmids were cotransfected per well using 1.2 l of Lipofectamine 2000. After incubation at 37C for 72 hours, the medium was aspirated and incubated under 100 l of lysis buffer [10 mM tris-HCl (pH 7.5), 0.05% SDS, and proteinase K (25 g/ml1) (Fisher BioReagents)] for 1 hour at 37C. The lysed mixture was heat inactivated at 80C for 30 min and stored at 4C until use. For preparing RNA-seq samples, 7.5 106 cells were seeded in 10-cm culture dish and transfected after 20 hours with 22.5 g of base editor P2A-GFP expression plasmid and 7.5 g of EMX1 #1targeting sgRNA plasmid mixed with 90 g of PEI MAX (Polysciences) in 1.0 ml of Opti-MEM I. The mixture was incubated for 30 min in room temperature and applied to the cells dropwise before cell sorting after 48 hours.

The HTS library was prepared using two rounds of PCR. For the first round, a 200-bp DNA fragment of the target region was amplified in a total volume of 25 l mixed with 12.5 l of the Q5 High-Fidelity 2X Master Mix, 1 l of the extracted genomic DNA, and a pair of primers (see the Supplementary Materials). Successful amplification of individual samples was checked using 1% agarose gel. For the second round, combinations of different Illumina indexes were attached at each 5 and 3 end of the first PCR products using the same total PCR volume. The PCR products were combined and column purified using a QIAquick PCR Purification kit (Qiagen) and further gel extracted to remove nonspecific amplifications. The final mixture of the library was quantified using the Qubit dsDNA HS Assay Kit (Life Technologies) and prepared for loading into a 150-cycle MiSeq reagent kit v3 (Illumina) according to the manufacturers protocol.

FASTQ files were generated by demultiplexing total sequencing reads using the MiSeq Reporter or Illuminas bcl2fastq 2.17 software. CRISPResso2 (available in GitHub; https://github.com/pinellolab/CRISPResso2) was used with the batch mode function to quantify the base editing conversion rates, indel frequencies, and product purities of the aligned reads (39). Heat maps displaying average base editing frequencies at each nucleotide position of three independent biological replicates were generated by running the CRISPResso2 analysis.

The use and care of animals followed the guidelines of the Biomedical Research Ethics Committee of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. GOTI experiments were performed according to the previous method (33). Briefly, mRNA of A3G-BE5.13 or Cre was generated by attaching the T7 promoter to the coding region through PCR amplification and using its purified PCR product as the template for in vitro transcription (IVT) using the mMESSAGE mMACHINE T7 ULTRA Kit (Invitrogen). Similarly, for sgRNA, the T7 promoter was attached, and the MEGAshortscript T7 Transcription Kit (Invitrogen) was used for IVT. mRNA and sgRNA products were purified using the MEGAclear Transcription Clean-Up Kit (Invitrogen). Fertilized embryos were obtained from C57BL/6 females (4 weeks old) mated to heterozygous Ai9 males (JAX strain 007909). A3G-BE5.13 mRNA (50 ng/l), Cre mRNA (2 ng/l), and sgRNA (50 ng/l) were mixed and injected using a FemtoJet microinjector (Eppendorf) into the cytoplasm of one blastomere of the two-cell embryo in a droplet of Hepes-CZB (Chatot-Ziomek-Bavister) medium containing cytochalasin B (5 g/ml). The embryos were incubated at 37C with 5% CO2 under KSOM (Potassium simplex optimized medium) medium for 2 hours and transferred into oviducts of ICR (Institute for Cancer Research) females at 0.5 days post coitum.

WGS and data analysis were performed according to the previous method (33). Briefly, at E14.5, prepared fetal tissues were dissociated using trypsin-EDTA (0.05%) and homogenized by passing through pipette tips multiple times. Cells were centrifuged, and the resulting pellet was resuspended in DMEM supplemented with 10% FBS before filtering through a 40-m cell strainer. tdTomato and tdTomato+ cells were isolated through FACS, and their genomic DNA were each extracted using the DNeasy Blood and Tissue Kit (Qiagen). WGS was performed at mean coverages of 50 by Illumina HiSeq X Ten. Burrows-Wheeler Aligner (version 0.7.12) was used to map qualified sequencing reads to the reference genome (mm10), and then the mapped BAM files were sorted and marked using Picard tools (version 2.3.0). SNVs were called from three algorithms, Mutect2 (version 3.5), LoFreq (version 2.1.2), and Strelka (version 2.7.1) with default parameters, separately (4042). Using the tdTomato sample from the same embryo as the reference, only variants shown to be mutated in the tdTomato+ at the same coordinate were counted within the mapped BAM file. SNVs overlapping from all the three algorithms were considered as the true variants.

Forty-eight hours after transfection, HEK293T cells cultured in 10-cm dish were washed with phosphate-buffered saline (Thermo Fisher Scientific) and dissociated by TrypLE Express. Cells were centrifuged, and the resulting pellet was resuspended in 5 ml of normal culture medium. Cells (0.5 to 0.7 106) with the top 5% GFP signal were sorted using SH800S cell sorter (Sony). Approximately a quarter of the sorted cells were collected in separate tubes for genomic DNA extraction and HTS analysis of the on-target base editing. For the remaining cells, the RNeasy Plus Mini Kit (Qiagen) was used to purify the total RNA. RNA library preparations and sequencing reactions were conducted at GENEWIZ LLC. (South Plainfield, NJ, USA). RNA samples were quantified using Qubit 2.0 fluorometer (Life Technologies), and RNA integrity was checked using Agilent TapeStation 4200 (Agilent Technologies). Sequencing libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina following the manufacturers instructions (New England Biolabs). Briefly, mRNAs were enriched with Oligo(dT) beads and were fragmented for 15 min at 94C. First- and second-strand cDNAs were subsequently synthesized. cDNA fragments were end-repaired and adenylated at 3 ends, and universal adapters were ligated to cDNA fragments, followed by index addition and library enrichment by limited-cycle PCR. The sequencing libraries were validated on the Agilent TapeStation (Agilent Technologies) and quantified by using Qubit 2.0 fluorometer and by quantitative PCR (Kapa Biosystems). The sequencing libraries were clustered on one lane of a flowcell and loaded on the Illumina HiSeq 4000 to be sequenced using a 2 150-bp paired-end configuration.

RNA-seq data analysis was carried out using the previous method (10). Qualified reads obtained from FastQC (version 0.11.3) and Trimmomatic (version 0.36) were aligned to the reference genome (Ensembl GRCh38) using STAR (version 2.5.2b) in two-pass mode with default parameters (43). Picard tools (version 2.3.0) were applied to sort and mark duplicates of the mapped BAM files. The refined BAM files were subject to split reads that spanned splice junctions, local realignment, base recalibration, and variant calling with SplitNCigarReads, IndelRealigner, BaseRecalibrator, and HaplotypeCaller tools from GATK (version 3.5), respectively (44). Clusters of more than four SNVs identified within a 35-bp window were filtered to maintain high-confidence variants, and found variants with base quality of >25, mapping quality score of >20, Fisher strand values of >30.0, qual by depth values of <2.0, and sequencing depth of >20 were counted.

For nucleofection of iPSCs and hESCs, cells were detached by using Accutase. For each reaction, 1.0 106 cells were resuspended in 82 l of P3 Primary Cell Nucleofector Solution and 18 l of supplement 1 using the P3 Primary Cell 4D-Nucleofector X Kit L (Lonza). Three micrograms of base editor P2A-PuroR expression plasmid and 1 g of sgRNA plasmid were added in the single-cell suspension and mixed well. The single-cell suspension was then transferred into a Nucleocuvette. Nucleofection was carried out in 4D-Nucleofector X Unit (Lonza) using code CB200, and cells were immediately plated on a Matrigel-coated 35-mm dish in mTeSR supplemented with 1 CloneR (STEMCELL Technologies). After 24 hours, puromycin (1.0 g/ml1) was supplemented into the medium for 1 day selection, and the surviving colonies were expanded for 10 to 14 days until extraction of the genome using the DNeasy Blood and Tissue Kit (Qiagen). The target region was PCR amplified using 30 cycles and sent for Sanger sequencing. EditR (baseeditr.com) was used to quantify the mutation peaks of Sanger chromatograms for analyzing the base conversion.

Bioinformatic analysis of pathogenic SNPs obtained from the BEable-GPS database (https://picb.ac.cn/rnomics/BEable-GPS/) was performed by finding correctable pathogenic SNPs that contain the target C located within the activity window of positions 4 to 8 of the protospacer, with NGG PAM positioned 21 to 23 (12). We then manually filtered the list on the basis of the sequence contexts containing the CC and/or CNC motif preferred by A3G-BEs. We counted precisely correctable pathogenic SNPs by manually filtering each disease on the basis of whether another base-editable bystander C was present within the activity window. For example, variant NM_012203.1(GRHPR): c.84-2A>G (protospacer; 5-TCACAGCCGCGGGGAAAGGG-3), in which the target C lies in the CC context but has a nearby bystander C lying in a CAC context potentially editable by A3G-BEs was removed from counting. The summarized list of SNPs can be found in data file S1.

Three biologically independent replicates performed on different days were used to calculate means and SD unless stated otherwise. All bar plots and figures except for heat maps were generated using Prism 8 (GraphPad). P values were calculated using Prism 8 by performing two-tailed Students t test, with a statistical significance level represented on each figure as ns (not significant), *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Acknowledgments: We thank D. Zhangs NABLab (Rice University) and G. Baos laboratory (Rice University) for providing the usage of the MiSeq Sequencing System. Funding: This work was supported by the Robert A. Welch Foundation (C-1952 to X.G. and C-1559 to A.B.K.), the NIH grant (HL151545 to X.G.), the Rice University Creative Ventures Fund (to X.G. and A.B.K.), the NSF grants (CHE-1664218 to A.B.K. and PHY-1427654 to the Center for Theoretical Biological Physics), the National Natural Science Foundation of China (31922048 to E.Z.), and the Agricultural Science and Technology Innovation Program (to E.Z.). Author contributions: S.L., N.D., and X.G. designed the study. S.L. and N.D. constructed plasmids, performed FACS, and prepared the HTS library. S.L. performed transfection, HTS, and HTS data analysis. S.L. and Q.Y. maintained HEK293T cells and created disease-associated stable cell lines. Y.S., T.Y., and E.Z. performed GOTI, WGS, and software analysis of the off-target SNVs. J.L. and I.B.H. helped with RNA-seq sample preparation. S.L. and L.L. performed nucleofection and clonal expansion of iPSCs and ESI-017 hESCs. N.D. performed the analysis of pathogenic SNPs statistics. S.L. and J.Y. performed statistical analysis. S.L., N.D., Q.W., and A.B.K. provided structural insights into A3G. All authors wrote and edited the manuscript. Competing interests: S.L., N.D., and X.G. are inventors on a pending provisional patent application submitted by the William Marsh Rice University related to this work. The authors declare that they have no other competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. FASTQ files containing HTS reads have been deposited in the National Center for Biotechnology Information, NIH Sequencing Read Archive and are available with accession number PRJNA623461. Additional data related to this paper may be requested from the authors.

Read the rest here:
Single C-to-T substitution using engineered APOBEC3G-nCas9 base editors with minimum genome- and transcriptome-wide off-target effects - Science...

INTRODUCTION

Many studies have shown that the CRISPR-Cas9 system is a revolutionary technology (1, 2). This relatively easy-to-use technology has provided unprecedented opportunities for scientific research and disease treatments, including applications in high-throughput screening and functional genomics research and treatment of virus infections (3), genetic diseases (4), and cancer (5). Nevertheless, there are now several well-known disadvantages with the CRISPR-Cas9 system, including the fact that single guide RNAs (sgRNAs) can sometimes lead to off-target effects such as double-strand breaks in untargeted genome regions, which can cause unintended adverse consequences such as gene mutations, insertions, deletions, and even tumorigenic events (6). Seeking to overcome these challenges, several strategies have been developed to improve the precision of CRISPR-Cas9 gene editing, including Cas9 modifications (e.g., Cas9 nickase and high-fidelity variants), prime editors, base editors, and selecting sgRNAs with minimal off-target capacity (7, 8). Recently, some inducible Cas9 expression systems have been developed to limit the activity or lifetime of Cas9, thereby lowering the probability of off-target effects by reducing the exposure time of a cells genome to the Cas9 nuclease (9).

There are a variety of chemically induced CRISPR-Cas9 systems, including doxycycline-regulated Cas9 (10), trimethoprim (TMP) (11) and 4-hydroxytamoxifen (4-OHT)controlled Cas9 (12), rapamycin-inducible split-Cas9 (13), 4-OHTresponsive inteindependent Cas9 (14), and 4-OHTresponsive nuclear receptors split-Cas9 (15), among others. However, a notable adverse effect of these systems is the potential for cytotoxicity from the chemical inducers: Doxycycline can negatively affect cell numbers and colony formation (16), TMP can inhibit uptake of folic acid by the cells (17), 4-OHT can increase cytosolic levels of autophagosomes and cause irregularly clumped chromatin in the nuclei (18), and rapamycin can perturb the endogenous mammalian target of rapamycin pathway (19). Moreover, once these agents are inside the cells or present in an in vivo context, these inducer chemicals can diffuse freely, limiting the spatial resolution of editing induction. In addition, it is difficult to rapidly remove the inducer compounds, so they can persist for a long time, making it difficult to turn Cas9 activity on and off quickly and precisely.

These limitations have helped motivate the development of multiple systems based on the optical control of Cas9 activity because light is a reversible and noninvasive inducer modality that potentially offers fine precise spatiotemporal resolution. The first reported example of a photoactivatable Cas9 system was paCas9 system based on blue light (20). In the paCas9 system, Cas9 nucleases are fragmented into two nonfunctional fragments that can be reconstituted as an active nuclease under blue light illumination based on dimerization of their respective fusion domains, the positive Magnet (pMag) or negative Magnet (nMag) proteins from the filamentous fungus Neurospora crassa (21). Later studies reported the ultraviolet (UV) lightmediated cleavage of a synthesized complementary oligonucleotide element that normally inactivates the editing-guiding function of sgRNAs (22).

There is also a recently reported blue lightbased anti-CRISPR system comprising AcrIIA4 (23) (a potent Cas9 inhibitor) and the LOV2 blue-light photosensor (24). Without illumination, the AcrIIA4-LOV2 complex remains bound to Cas9, inhibiting its nuclease activity. Under blue light illumination, the AcrIIA4-LOV2 complex is separated from Cas9 and its editing activity can be restored (25). However, neither UV nor blue light is able to penetrate deeply into the body, owing to the strong absorption and scattering of these light energies by biological tissues (26). UV light hardly penetrates the skin and blue light does merely by 1 mm (27, 28). This substantial limitation, viewed alongside the fact that UV and prolonged blue light exposure can cause cytotoxicity (29, 30), highlights the difficulty of applying these light-induced Cas9 systems for in vivo research applications and clinical translation.

We have, for some time, been investigating far-red light (FRL)inducible genetic systems due to the deep tissue penetration of FRL with above 5 mm beneath the surface of skin (27, 28). We here report our development of an FRL-activated split-Cas9 (FAST) system that can be used to noninvasively induce gene editing activity in cells located deep inside animal tissues. The FAST system relies on two split-Cas9 fusion proteins with high-affinity binding domains: One half of Cas9 is constitutively expressed, while the other is under the FRL-inducible control of the bacterial phytochrome BphS optical controllable system previously established by our group (31). We initially assembled the FAST system components in human embryonic kidney (HEK)293 cells and used light-emitting diode (LED)based FRL illumination to demonstrate successful activation of targeted genome editing. Next, after achieving FRL-inducible editing in diverse human cell lines, experiments with implants confirmed that FAST was able to robustly activate editing in cells positioned in subdermal animal tissues. Experiments with the transgenic tdTomato reporter mouse line established FRL-induced FASTmediated editing of mouse somatic cells (hepatocytes in the liver), and work with cell cycleinactivating gene edits of cancer cells in xenograft tumor mice demonstrate how FAST can be deployed against disease. Thus, beyond extending the optogenetic toolbox for gene editing of mammalian cells to include induction by the highly in vivocompatible and deep tissuepenetrating energies of FRL, our study extends this initial technology to demonstrate applications relevant for basic biological and biomedical research.

To develop an optogenetically controlled device for genome editing with deep tissuepenetrative capacity and with negligible phototoxicity in vivo, first, we constructed an FRL-controlled full-length Cas9 system based on our previously reported orthogonal FRL-triggered optogenetic system (FRL-v2) (31). However, there was serious background leakage in dark state with low-induction performance under illumination. Therefore, we focused on building a FAST system based on split-Cas9 (13) and FRL-v2, which comprises the bacterial FRL-activated cyclic diguanylate monophosphate (c-di-GMP) synthase (BphS) and a c-di-GMPresponsive hybrid transactivator, p65-VP64-BldD. For the FAST system, we then fused the N-terminal Cas9 fragment [Cas9(N)] to the Coh2 domain from Clostridium thermocellum (32) and fused the C-terminal Cas9 fragment [Cas9(C)] to the DocS domain from the same bacterium. Expression of the NLS-Cas9(N)-Coh2 fusion protein is driven by the FRL-v2specific chimeric promoter (PFRL), while expression of the DocS-Cas9(C)-NES fusion protein is driven by a constitutive promoter (PhCMV). A complete Cas9 protein can be reconstituted upon FRL illumination because of the high-affinity interaction of the Coh2 and DocS domains (Fig. 1). Confirming the editing activity of the reconstituted Cas9, we found that HEK-293 cells cotransfected with pXY137 (PhCMV-p65-VP64-BldD-pA::PhCMV-BphS-P2A-YhjH-pA, 100 ng), pYH20 [PFRL-NLS-Cas9(N)-Linker-Coh2-pA, 50 ng], pYH102 [PhCMV-DocS-Linker-Cas9(C)-NES-pA, 100 ng], and pYW57 [PU6-sgRNA (CCR5)-pA, 50 ng] successfully edited the targeted human CCR5 locus (11.9% indel frequency) upon FRL illumination (1 mW/cm2; from an LED source, 730 nm); no editing was detected for dark control cells (Fig. 2, A and B). These detected edits were analyzed by the mismatch-sensitive T7 endonuclease I (T7E1) assay. We further used Sanger sequencing to confirm that the FRL-induced, FAST-mediated edits (indel mutations) occurred in the targeted region of the human CCR5 locus at a frequency of ~20% using the tracking of indels by decomposition (TIDE) analysis (fig. S1).

(A) Schematic of the split-Cas9 fusion protein components of the FAST system. Coh2 and DocS are two C. thermocellum proteins that interact with high affinity. Cas9 is formed from two separate (N- and C-terminal) Cas9 fragments that individually lack nuclease activity. When Cas9s two fragments Cas9(N) and Cas9(C) are respectively fused with Coh2 and DocS, they readily combine to reconstitute a nuclease-active form of Cas9. (B) Schematic of the FAST system, as deployed in mammalian cells, based on the fragments detailed in (A). FRL (~730 nm) activates the engineered bacterial photoreceptor BphS, which converts guanosine triposphate (GTP) into c-di-GMP. c-di-GMP can bind to BldD (derived from sporulating actinomycete bacteria) and be translocated into the nucleus. This induces dimerization of the synthetic transcriptional activators p65-VP64-BldD [BldD fused with p65 (the nuclear factor Btransactivating domain) and VP64 (a tetramer of the herpes simplex virusderived VP16 activation domain)], after which they bind to PFRL to activate expression of the N-terminal fusion fragment of split-Cas9. The other (C-terminal) fusion fragment is constitutively expressed, as driven by the human cytomegalovirus promoter (PhCMV). DNA double-strand breaks are formed by Cas9 after the Coh2-DocS heterodimerizationmediated reconstitution of the two fusion fragments.

(A) Time schedule of FRL-controlled gene editing in HEK-293 cells. Cells were illuminated (1 mW/cm2; 730 nm) for 4 hours once a day for 2 days and were collected at 48 hours after the first illumination for further analysis. (B) A mismatch-sensitive T7 endonuclease I (T7E1) assay to test HEK-293 cells (6 104) transfected with full-length Cas9 (pHP1) or the FAST system (pXY137, pYH20, and pYH102), together with the sgRNA targeting to CCR5 locus (pYW57). FRL-mediated editing (indel deletions) of the human EMX1, CXCR4, and VEGFA loci by FAST was performed using the same experimental procedure as that used when targeting the CCR5 gene. (C) FRL-mediated multiplex editing of the human CCR5 and CXCR4 loci. (D) FAST-mediated DNA insertion via homology-directed repair (HDR), achieved by adding a single-stranded oligodeoxynucleotide (ssODN) template (10 M), bearing a HindIII restriction endonuclease site. Homologous arms are indicated in red. The target sites of sgRNA (EMX1) are marked in blue. HEK-293 cells (6 104) were cotransfected with full-length Cas9 (pHP1) or the FAST system (pXY137, pYH20, and pYH102) and the sgRNA targeting to EMX1 locus (pYH227) via a nucleofection method. In (B) to (D), n = 2 from two independent experiments. Red arrows indicate the expected cleavage bands. Detailed description of genetic components and transfection mixtures are provided in tables S1 and S5. N.D., not detectable.

We next confirmed that the FAST system can cleave different targeted endogenous genomic loci and induce indel mutations via nonhomologous end joining (NHEJ) in an FRL-dependent manner by designing sgRNAs targeting three additional human genes (EMX1, CXCR4, and VEGFA), and these induced indel mutations were detected by T7E1 assay. With each of these sgRNAs, FRL-induced but not dark-induced indel mutations were observed (Fig. 2B). We also confirmed that the FAST system can cleave targeted exogenous d2EYFP reporter efficiently (fig. S2). In addition to single gene targeting, we also tested whether our FAST system can simultaneously edit multiple target sites. Using one sgRNA targeting CCR5 and another sgRNA targeting CXCR4, the FAST system was capable of inducing the desired indel mutations at the two target sites upon FRL illumination (Fig. 2C), demonstrating optogenetic multiplexed control of NHEJ-mediated indel mutations in mammalian cells.

We further investigated whether FAST can be used for homology-directed repair (HDR)mediated genome editing. The FAST system components and a donor template (single-stranded oligodeoxynucleotide containing a HindIII site) were electroporated into HEK-293 cells. Assessment of HDR events at the EMX1 locus using restriction endonuclease assays showed that the FAST system induced HindIII site integration at the EMX1 locus at a frequency of 5.7% under FRL illumination; no HDR events were detected in dark controls (Fig. 2D). Together, these results establish that the FAST system can be deployed for optogenetic control of NHEJ-/HDR-mediated indel mutations.

To demonstrate photoactivatable regulation of gene editing in diverse mammalian cell lines, we introduced the FAST system into four different human cell lines, and it achieved successful FRL-induced gene editing (CCR5 locus) in each of them (Fig. 3A). Next, experiments testing the FRL illumination intensity and duration-dependent activity of the FAST system showed that the frequency of edits (indel mutations at CCR5) increased along with illumination intensity and with illumination time (Fig. 3, B and C), indicating the tunability of the FAST system. We also used a photomask to establish proof of principle for spatially controlled gene editing with the FAST system (Fig. 3, D and E). We also conducted an experiment with two rounds of FRL illumination to verify repeated induction cycles of the FAST system wherein the first round of illumination achieved indel mutations guided by an sgRNA targeting CXCR4 locus, followed by transfection of a second sgRNA targeting the CCR5 locus, which guided successful indel mutations after the second FRL illumination. However, engineered cells shifted to the dark did not have indel mutations in CCR5 locus (fig. S3, A and B). This result indicates that the FAST system is reusable and reversible.

(A) FAST-mediated gene editing in four human cell lines. (B) Illumination intensitydependent FAST gene editing. In (A) and (B), cells were collected for mismatch-sensitive T7E1 assays, as indicated in the time schedule of Fig. 2A. (C) Evaluation of exposure timedependent FAST system gene editing performance. Cells were collected for T7E1 assays at 24 hours after the start of the second illumination. (D) Schematic of the photomask device used to demonstrate the spatial regulation of FAST-mediated gene editing. Cells were illuminated through a photomask containing a 7-mm line pattern. (E) Spatial control of FRL-dependent gene editing mediated by the FAST system. HEK-293 cells (3 106) were cotransfected with the FAST system, sgRNA (pYW57), and a frameshift enhanced green fluorescent protein (EGFP) reporter containing a CCR5 locus (pYH244) and were illuminated with FRL (0.5 mW/cm2; 730 nm; 2-min on, 2-min off) for 48 hours. EGFP is not expressed without Cas9 activity because the EGFP sequence is out of frame. Upon double-strand cleavage by Cas9, the frameshifts caused via DNA repair by NHEJ enable EGFP expression. The fluorescence of EGFP was assessed via fluorescence meter ChemiScope 4300 Pro imaging equipment (Clinx) at 48 hours. In (A) to (C), n = 2 from two independent experiments. Red arrows indicate the expected cleavage bands. Detailed description of genetic components and transfection mixtures are provided in tables S1 and S5. SEAP, human placental secreted alkaline phosphatase.

We then evaluated the photocytotoxicity of FRL (730 nm) or blue light (470 nm) illumination on mammalian cells. When HEK-293cells were transfected with human placental secreted alkaline phosphatase (pSEAP2)-control-and then exposed to FRL or blue light for different intensity, the SEAP expression demonstrated that the FRL exposure resulted in negligible cytotoxicity. However, a marked difference was observed from the blue light illumination, which substantially reduced cell viability (fig. S4, A and B). Moreover, we did not observe substantially increased cytotoxicity with FRL illumination of cells engineered with the FAST system (fig. S4, C and D), indicating the inertness and noncytotoxicity of the system constituents. In short, neither FRL illumination nor the ectopic presence of FAST system constituents was verified to influence the gene expression capacity of the engineered cells. In addition, we also compared the controllable gene editing performance of our FAST system with the rapamycin-responsive split-Cas9 system (13) and the blue lightcontrolled paCas9 system (20) that have been reported. The results showed that the genome editing efficiency of rapamycin-responsive split-Cas9 system was lower than the FAST system (fig. S5, A and B), and the paCas9 system had relative higher background leakage in the dark. Our FAST system showed notable induction of indel mutations under FRL illumination but with negligible background in the dark (fig. S5, C and D). Off-target activity of the FAST system was also assessed simply. We checked a potential off-target site of human BMP1 locus, as reported previously (33). The indel frequencies were determined through T7E1 assay at the on-target and potential off-target sites of BMP1. As a result, no mutations were detected at the potential off-target site after editing by our FAST system (fig. S6, A and B). This is probably due to the FAST-mediated transient expression of split-Cas9 that lowered the probability of off-target effects by reducing the exposure time of a cells genome to the Cas9 nuclease (79). However, there might be off-target effects that can still occur in illuminated cells.

Having established the basic performance characteristics of the FAST system in human cells, we next conducted experiments with mice to verify the systems capacity to induce gene editing based on the tissue-penetrating capacity of FRL. Specifically, we conducted an experiment using hollow fiber implantation of HEK-293 cells equipped with the FAST system into the dorsum of mice and exposed to FRL illumination (10 mW/cm2; alternating 2-min on/off for 4 hours) (Fig. 4A). Notably, the FRL illumination of the FAST cell-bearing mice induced notable activation of gene editing (~11.4% of the cells retrieved from the implant fibers was edited at the CCR5 locus versus not detectable for dark control cells) (Fig. 4B). These results demonstrate that the FAST system can be used to activate gene editing inside animal tissues, exploiting the physical properties of FRL as an inducer modality.

(A) Schematic for the time schedule and experimental procedure for FRL-controlled gene editing in mice harboring hollow fiber implants with HEK-293 cells. Pairs of 2.5-cm hollow fibers containing a total of 5 106 transgenic HEK-293 cells (engineered with FAST system) were subcutaneously implanted on the dorsum of wild-type mice and illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off) for 4 hours each day for 2 days. Cells were collected from the hollow fiber implants at 48 hours after the first illumination and assessed with mismatch-sensitive T7E1 assay to assess targeted gene editing efficiency (CCR5 locus). (B) Representative T7E1 assay for FAST-mediated indel mutations. n = 3 mice. The red arrow indicates the expected cleavage bands. Detailed description of genetic components and transfection mixtures are provided in table S1 and S5.

We obtained transgenic mice harboring a homozygous Rosa26 CAG [cytomegalovirus (CMV) enhancer fused to the chicken beta-actin] promoter loxP-STOP-loxP-tdTomato cassette present in all cells. In this model, tdTomato is silent because of the stop signal [three repeats of the simian virus 40 (SV40) polyadenylate (polyA) sequence], but the deletion of the stop cassette allows transcription of the tdTomato gene, resulting in fluorescence expression. The Cas9-mediated DNA cleavage of the stop sequence guided by sgRNAs can initiate CAG promoter to drive tdTomato expression (34). Therefore, we used this mouse model to examine the in vivo genome editing performance of the FAST system in mice somatic cells (Fig. 5A). We used hydrodynamic injection to introduce the FAST system components, along with an sgRNA designed to target the deletion of the SV40 polyA stop cassette, which should activate tdTomato reporter protein expression upon successful editing. Note that it is difficult to activate tdTomato expression by Cas9 system as the desired edit requires two cuts on the same allele; we eventually achieved the desired edit, but it required optimization of the delivery mode for the FAST components. Briefly, we chose hydrodynamic injection because it is known to result in enrichment of plasmids (and thus, transgene expression) in liver cells (35). We reduced the overall number of plasmids by combining some constructs (fig. S7, A and B) and explored a number of different injection time and illumination schedules (Fig. 5A), but we only detected weak tdTomato signals in the FRL-illuminated FAST mice (fig. S8).

(A) Schematic showing the time schedule and experimental procedure for assessing in vivo gene editing. The minicircle iteration of the FAST system pYH412, pYH413, and pYH414 at a 7:15:4 (w/w/w) ratio were injected hydrodynamically via tail vein. Twenty-four hours after injection, mice were illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off) for 4 hours per day for 3 days. A second injection of the minicircle-based FAST system components was performed on the fifth day, followed by 4 hours daily illumination for three additional days. In our design, the tdTomato reporter protein was expressed after a stop cassette was destroyed by Cas9 editing. (B) Fluorescence IVIS image of mouse livers. (C) The frequency of edits (targeting the aforementioned stop cassette) by monitoring fluorescence intensity of the tdTomato reporter in Gt(ROSA)26Sortm14(CAG-tdTomato)Hze mice. (D) Representative fluorescence microscopy images of tdTomato and tdTomato+ hepatocytes present in frozen liver sections from FRL-illuminated mice. Blue indicates 4,6-diamidino-2-phenylindole (DAPI) staining nuclei; red indicates endogenous tdTomato expression. The images represent typical results from three independent measurements. Scale bar, 100 m. Data in (C) are means SEM; n = 3 mice. P values were calculated by Students t test. ****P < 0.0001 versus control.

We speculated that this apparently weak induction of editing activity may result from rapid degradation of the plasmids, so we constructed minicircle (36) iterations of our FAST system. Minicircle DNA vectors without the bacterial backbone of the plasmid, markedly reducing the possibility of random integration of bacterial DNA sequences into the genome, have been shown to maintain gene expression in cells for long durations because these molecules are resistant to degradation (37). We delivered the minicircle iterations of the FAST via hydrodynamic injection and used FRL illumination schedules as follows: alternating 2-min on/off for 4 hours, once each day for 3 days; we then monitored the fluorescence signal intensity in livers. FRL illumination of the mice bearing the FAST system resulted in strong editing and thus, tdTomato reporter expression (Fig. 5, B and C). We also detected strong tdTomato expression in liver sections prepared from the FRL-illuminated FAST mice (Fig. 5D), and Sanger sequencing of genomic DNA extracted from the livers verified the success of the targeted excision of the SV40 polyA stop cassette in the FRL-induced FAST mice (fig. S9). Collectively, these results demonstrate that the FAST system can be used for in vivo editing of the genomes of somatic cells located in the internal organs of mice.

We further investigated the optogenetic activation of the FAST system in tumor models as proof-of-concept examples for therapeutic genome editing. The polo-like kinase (PLK1) protein is a highly conserved serine-threonine kinase that promotes cell division, and strong PLK1 expression is a marker in various types of tumor (38). Extensive work has established that inhibition or depletion of PLK1 leads to cell-cycle arrest, apoptosis, and a so-called mitotic catastrophe in cancer cells, which provides a promising modality for anticancer therapy (39, 40). After initially confirming that the FAST system can edit the PLK1 locus (indel mutations and extensive apoptosis) in the FRL-illuminated human lung cancer A549 cells in vitro (fig. S10, A to D), we then evaluated the tumor therapy application of our FAST system by testing the in-tumor editing performance of the FAST system for the disruption of the PLK1 locus in mice bearing A549 xenograft tumors.

We first delivered the minicircle iterations of the FAST system alongside a PLK1-targeting sgRNA minicircle vector when the tumors had reached 80 to 100 mm3; note that we also injected transfection reagent, a cationic polymer-coated nanoparticle (APC), (41) to facilitate the transfection of tumor cells in situ. Subsequently, FRL illumination was delivered to the xenograft-bearing mice via LED for 4 hours each day for 7 days (Fig. 6A), and tumor development was monitored by measuring the sizes of the tumors every 2 days. Notable inhibition of tumor growth was observed for the FAST mice that received FRL illumination; no such inhibition was observed for the dark control FAST or FRL-illuminated vehicle control mice (Fig. 6, B to D). Mismatch-sensitive T7E1 assays confirmed that the FRL-induced FAST system achieved the desired genome disruption of PLK1 gene in the tumor tissue (Fig. 6E) at a frequency of ~21.5% detected by TIDE analysis (Fig. 6F). Moreover, quantitative real-time polymerase chain reaction (qRT-PCR) verified the expected reductions in tumor PLK1 mRNA expression upon FRL illumination (Fig. 6G). Consistent with the observed antitumor efficacy, subsequent histologic analysis of tumor sections revealed extensive cancer cell necrosis (Fig. 6H) and very extensive cell apoptosis [via both terminal deoxynucleotidyl transferasemediated deoxyuridine triphosphate nick end labeling (TUNEL) and caspase-3labeling assays; Fig. 6, I and J]. Thus, FRL-triggered FAST-mediated gene editing can inhibit cancer cell growth in xenograft tumors in mice. These results further indicate that our FAST system could be deployed for deep tissue gene editing.

(A) Schematic showing the time schedule and experimental procedure for the in-tumor FAST-mediated gene editing. The minicircle iteration of the FAST system targeting to PLK1 locus pYH412, pYH420, and pYH414 at a 7:15:4 (w/w/w) ratio were injected intratumorally. Twenty-four hours after per injection, mice were illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off) for 4 hours per day totally for 7 days. (B) Images of tumor tissues from the different treatments. (C) Tumor growth curves for the different treatments. (D) The weight of tumor tissues after the different treatments. (E) Indel mutations in the tumor tissues detected via mismatch-sensitive T7E1 assays. Red arrows indicate the expected cleavage bands. (F) The gene editing efficacy quantified by the TIDE analysis. (G) Relative mRNA expression levels of the PLK1 gene quantified by quantitative real-time polymerase chain reaction (qRT-PCR). The data are means SEM; n = 5 mice. P values were calculated by Students t test. ****P < 0.0001 versus control. (H) Representative fluorescence microscopy images of hematoxylin and eosin (H&E) staining of tumor tissues. The images represent typical results from three independent measurements. Scale bar, 100 m. Representative fluorescence microscopy images of TUNEL staining (I) and caspase-3 (J) staining of tumor tissues. The images represent typical results from three independent measurements. Scale bars, 100 m. Photo credit: Yuanhuan Yu, East China Normal University.

CRISPR-Cas9 is an undeniably revolutionary technology that is changing biological and medical research (4, 5, 42), and several innovative extensions of the basic CRISPR-Cas9 concept have enabled a new era of conditional genome editing activation iterations with chemical (1015) and UV/blue light inducers (20, 22, 25). Nevertheless, limitations with these systems warrant the development of alternatives that exploit different induction sources. The FAST system we developed in the present study opens the door for spatiotemporally selective induction of Cas9 genome editing deep inside animal tissues. It bears emphasis that our induction uses LED lights rather than lasers or optical fibers, highlighting that FAST should be very easy to deploy in a wide range of experimental contexts. Although we did face initial hurdles with induction efficiency for in vivo applications, our development of a minicircle-based iteration of the FAST system easily overcame this and permitted robust editing in mouse livers. The deep tissuepenetrating utility of the FAST system was applied to achieve anticancer therapy by disrupting PLK1 gene in mice bearing A549 xenograft tumors. In this way, we could greatly reduce side effects of the anticancer drugs and promote the precision treatment of cancers. We also envision that the FAST system can be used to study the function of cancer-associated genes during tumor development process by controlling gene knockout or interference in specific tissues at different time nodes.

While we do demonstrate FAST system applications for biological research and the treatment of disease, the present paper merely reports the initial proof-of-principle study. Given that FAST is a fully genetically encoded system, a variety of vectors, alternative plasmids, and tissue-specific promoters could be used to selectively deliver FAST system components to diverse tissues, and we fully anticipate that adeno-associated virus vectors will become a popular modality for this task. Moreover, there is no obvious factor to prevent the deployment of FAST as a genome-integrated stable system, which should enable researchers to selectively activate targeted editing anywhere that they are able to supply sgRNAs and FRL illumination from an LED.

We anticipate that the combination of precise temporal control and deep tissue penetration will enable rapid-uptake FAST in a variety of research communities. Chemical inducers can cause adverse effects in cells and can diffuse freely, and the complexity of cellular and organismal metabolism makes it exceedingly difficult to precisely control the spatiotemporal dynamics of inducible gene editing systems (1619). In this light, perhaps researchers can deploy FAST and FRL induction strategies to explore the development, basic biology, or etiopathological basis of diverse processes that occur in animal internal organs such as the heart, lungs, liver, kidneys, etc., and in tissues, including muscles and bone marrow. In theory, the FAST system should give researchers previously unattainable precise control of conditional genetic knockout and knock-in experiments. A huge variety of temporal illumination schemes should be feasible with FAST because FRL has low phototoxicity, representing a clear advantage over UV- and blue lightbased Cas9 induction systems. Moreover, FAST may offer neuroscientists an alternative to the presently popular optical fiber implantationbased approaches for optogenetic-based gene editing research.

In summary, we have developed a FAST system that is apparently safe (negligible phototoxicity to mammalian cells, high tissue permeability, and noninvasiveness). With FRL as its fundamental basis, the FAST system offers excellent tunability (robust induction of gene editing and almost negligible background activity) and precise controllability (illumination intensity dependent, exposure time dependent, and strong spatiotemporal specificity), making it suitable and practical for the many biological and biomedical applications that require gene editing in vivo, especially for processes that occur within animal tissues.

The FAST system consists of the following main components: the FRL sensors (BphS and p65-VP64-BldD) (31), interacting proteins (cohesion Coh2 and dockerin DocS from C. thermocellum) (32), and the N- and C-terminal fragments of Streptococcus pyogenes Cas9 [Cas9(N) (residues 2 to 713) and Cas9(C) (residues 714 to 1368)] (13). Complementary DNAs (cDNAs) encoding BphS and p65-VP64-BldD were prepared, as previously described (31). cDNAs encoding Coh2 and DocS were chemically synthesized by the company Genewiz Inc. cDNAs encoding the N- and C-terminal fragments of Cas9 fused with a nuclear localization signal from SV40 T antigen were amplified from the Addgene plasmid 42230. The inducible Cas9 was constructed on the basis of the Cas9(N) and Cas9(C) fragments fused with Coh2 and DocS, respectively, which were cloned through Gibson assembly according to the manufacturers instructions [Seamless Assembly Cloning Kit; catalog no. BACR(C) 20144001; OBiO Technology Inc.]. All genetic components have been validated by sequencing (Genewiz Inc.). Plasmids constructed and used in this study are provided in table S1.

The sgRNAs targeting CCR5, EMX1, CXCR4, VEGFA, BMP1, tdTomato stop cassette, and PLK1 were generated by annealed oligos and cloned into the BbsI site of a constitutive mammalian PU6-driven sgRNA expression vector (pYH49). The PU6-sgRNA fragment was PCR amplified from the Addgene plasmid 58767 and then cloned into the corresponding sites (MluI/XbaI) of pcDNA3.1(+) to obtain the pYH49 expression vector. The target sequences and oligonucleotides used for sgRNA construction are listed in table S2.

All cell types {HEK-293 [CRL-1573; American Type Culture Collection (ATCC)], HeLa (CCL-2; ATCC), telomerase-immortalized human mesenchymal stem cells (43), and HEK-293derived Hana3A cells engineered for constitutive expression of RTP1, RTP2, REEP1, and Go} were cultured at 37C in a humidified atmosphere, containing 5% CO2 in Dulbeccos modified Eagles medium (DMEM; catalog no. C11995500BT; Gibco) supplemented with 10% fetal bovine serum (FBS; catalog no. 16000-044; Gibco) and 1% (v/v) penicillin/streptomycin solution (catalog no. ST488-1/ST488-2; Beyotime Inc.). All cell lines were regularly tested for the absence of mycoplasma and bacterial contamination. Cells were transfected with an optimized polyethyleneimine (PEI)based protocol (44). Briefly, cells were seeded in a 24-well cell culture plate (6 104 cells per well) 18 hours before transfection and were subsequently cotransfected with corresponding plasmid mixtures for 6 hours with 50 l of PEI and DNA mixture [PEI and DNA at a ratio of 3:1 or 5:1 (w/w)] (PEI molecular weight, 40,000; stock solution of 1 mg/ml in ddH2O; catalog no. 24765; Polysciences Inc.). At 12 hours after transfection, the culture plate was placed below a custom-designed 4 6 LED array (1 mW/cm2; 730 nm) for illumination.

For HDR-mediated genome editing experiments, 6 105 HEK-293 cells were nucleofected with the FAST system plasmids (pXY137, 200 ng; pYH20, 100 ng; and pYH102, 200 ng), sgRNA expression vector (pYH227, 100 ng; targeting EMX1), and 10 M single-stranded oligonucleotide donor using the SF Cell Line 4D-Nucleofector X Kit L (catalog no. V4XC-2024; Lonza) and the CM-130 program (4D-Nucleofector System; Lonza). At 24 hours after nucleofection, cells were illuminated by FRL (1 mW/cm2; 730 nm) for 4 hours once a day for 2 days, and then cells were collected at 48 hours after the first illumination for analysis. Genomic DNA was isolated using a TIANamp Genomic DNA Extraction Kit (catalog no. DP304; TIANGEN Biotech Inc.) according to the manufacturers instructions.

Genomic DNA was extracted from cells or tissues using the TIANamp Genomic DNA Extraction Kit (catalog no. DP304; TIANGEN Biotech Inc.) according to the manufacturers instructions. The genomic region containing the target sites was PCR amplified using the 2 Taq Plus Master Mix II (Dye Plus) DNA polymerase (catalog no. P213; Vazyme Inc.). The primers used for PCR amplification are listed in table S3. The PCR amplicons were purified using HiPure Gel Pure Micro Kits (catalog no. D2111-03; Magen Inc.) according to the manufacturers protocol. Purified PCR products (300 ng) were mixed with 1.5 l of 10 M buffer for restriction enzyme (catalog no.1093A; Takara Bio) and ultrapure water to a final volume of 15 l and reannealed (95C, 5 min; 94C, 2 s, 0.1C per cycle, 200 times; 75C, 1 s, 0.1C per cycle, 600 times; and 16C, 5 min) to form heteroduplex DNA. After reannealing, the heteroduplexed DNA was treated with 5 U of T7E1 (catalog no. M0302; New England BioLabs) for 1 hour at 37C and then analyzed by 1.5% agarose gel electrophoresis. Gels were stained with GelRed (catalog no. 41003; Biotium) and imaged with Tanon 3500 gel imaging system (Tanon Science & Technology Inc.). Relative band intensities were calculated by ImageJ software. Indel percentage was determined by the formula 100% [1 (1 (b + c)/(a + b + c))1/2], in which a is the integrated intensity of the undigested PCR product, and b and c are the integrated intensities of each cleavage product.

Sequence of the gene region containing the target sequence was amplified by PCR. Purified PCR amplicons from the nuclease target site were cloned into the T-vector pMD19 (catalog no. 3271; Takara Bio). Thirty clones were randomly selected and sequenced using each genes PCR forward primers by the Sanger method (45). Primers used for PCR amplification are listed in table S3.

Target regions were amplified by PCR. Purified PCR samples were analyzed by Sanger sequencing. The sequencing data files (.ab1 format) were imported into the TIDE Web tool (https://tide.nki.nl/) (46) to quantify nature and frequency of generated indels.

The genomic PCR and purification were performed, as described above. Purified PCR products were mixed with 15 U of HindIII (catalog no. 1060B; Takara Bio), 2 l of 10 M buffer for restriction enzyme, and ultrapure water to a final volume of 20 l and then incubated at 37C for 3 hours. The digested products were analyzed by agarose gel electrophoresis. Gel staining and imaging were performed, as described above. Quantification was calculated on the basis of relative band intensities. The HDR percentage was determined by the formula 100% (b + c)/(a + b + c), in which a is the intensity of the undigested PCR product, and b and c are the intensities of each HindIII-digested product.

HEK-293 cells (6 104) were cotransfected with the FAST system (pXY137, 100 ng; pYH20, 50 ng; and pYH102, 100 ng), the sgRNA targeting d2EYFP (pYH410, 50 ng), and the d2EYFP reporter plasmid (pYW110, 200 ng). At 12 hours after transfection, cells were illuminated (1 mW/cm2; 730 nm) for 4 hours once a day for 2 days and were harvested after trypsinization and washed in phosphate-buffered saline (PBS) for three times. About 10,000 events were collected per sample and analyzed with a BD LSRFortessa cell analyzer (BD Biosciences) equipped for d2EYFP [488-nm laser, 513-nm longpass filter, and 520/30 nm emission filter (passband centered on 530 nm; passband width of 30 nm)] detection. Data were analyzed using the FlowJo V10 software.

The production of human placental SEAP in cell culture medium was quantified using a p-nitrophenylphosphatebased light absorbance time course assay, as previously reported (31). Briefly, 120 l of substrate solution [100 l of 2 SEAP buffer containing 20 mM homoarginine, 1 mM MgCl2, and 21% (v/v) diethanolamine (pH 9.8) and 20 l of substrate solution containing 120 mM p-nitrophenylphosphate] were added to 80 l of heat-inactivated (65C, 30 min) cell culture supernatant. The time course of absorbance at 405 nm was measured by using a Synergy H1 hybrid multimode microplate reader (BioTek Instruments Inc.) installed with the Gen5 software (version 2.04).

Cell viability was assayed using an MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide] cytotoxicity assay kit (catalog no. E606334-0250; Sangon Biotech Inc.) according to the manufacturers instructions. Briefly, 10 l of MTT reagent (5 mg/ml) was added to each well of 96-well plates. The samples were mixed gently and incubated for 4 hours in a CO2 incubator. Formazan solubilization solution (100 l) was added into each well. The plate was put on a shaker to mix gently for 10 min to dissolve the formazan crystals, and then the plate was read with a Synergy H1 microplate reader (BioTek Instruments Inc.) at 570 nm.

The off-target sites of the BMP1 gene were examined according to the previously reported procedure (33). Genomic DNA was extracted, as described above, and the region of genome containing the possible nuclease off-target sites was PCR amplified using appropriate primers (table S3). The following procedures were similar to those of on-target examination by T7E1 assay, as described above.

Minicircles are episomal DNA vectors that allow sustained transgene expression in quiescent cells and tissues. Minicircle DNA vectors were prepared, as previously described (36). Minicircle-producing system contains the Escherichia coli strain ZYCY10P3S2T (a genetically modified minicircle-producing bacterial strain) and the empty minicircle-producing plasmid pMC.BESPX (gene of interest would be cloned into this plasmid). Briefly, ZYCY10P3S2T competent cells prepared with standard protocol, as previously described (36), were transformed with the minicircle-producing plasmid pMC.BESPX carrying the gene of interest. The transformed cells were cultured and induced by 0.01% l-arabinose to produce minicircle DNA vectors that were devoid of the bacterial plasmid DNA backbone and contain only genes of interest.

The in vivo DNA delivery reagent APC is a cationic polymer-coated nanoparticle composed of biocompatible polystyrene sulfonate and -cyclodextrinPEI (Mw, 25 kDa) and prepared, as previously reported (41). First, the seed solution was prepared by adding freshly prepared 600 l of NaBH4 (10 mM) into 5-ml mixture of HAuCl43H2O (0.5 mM) and cetyltrimethylammonium bromide (CTAB; 0.1 M) and incubated at 30C for 30 min. Ten milliliters of HAuCl43H2O (1 mM), 10 ml of CTAB (0.2 M), 120 l of AgNO3 (0.1 M), and 600 l of hydroquinone (0.1 M) were mixed together as growth solution. When the color of the growth solution turned from yellow to colorless, 320 l of seed solution was added. The desired longitudinal surface plasmon resonance peak was obtained after keeping the reaction mixture undisturbed in dark at 30C for 12 hours. The products were then gathered by centrifugation at 7000 RCF (relative centrifugal force) for 10 min at 30C. The supernatant was removed, and the precipitate was resuspended in 2 ml of 30C ultrapure water. Furthermore, 1 ml of the products from last step [Au (0.2 mg/ml)] was added to 10 ml of polysodium 4-styrenesulfonate (2 mg/ml) dissolved in NaCl (1 mM) solution and stirred for 1 hour at 30C. The solution was centrifuged at 7000 RCF for 10 min, and the residue was resuspended to obtain 2 ml of biocompatible polystyrene sulfonatecoated nanoparticle solution. Last, 1 ml of biocompatible polystyrene sulfonatecoated nanoparticles was added to 10 ml of -cyclodextrinPEI (2 mg/ml) dispersed in NaCl (1 mM) solution and stirred for 1 hour at 30C to obtain APC.

Apoptosis analysis at the cellular level was assessed using the Annexin Vfluorescein isothiocyanate (FITC)/propidium iodide (PI) Apoptosis Detection Kit (catalog no. E606336; Sangon Biotech Inc.). Briefly, A549 cells (3 104) cotransfected with the minicircle iterations of the FAST system and the sgRNA targeting PLK1 {pYH412 (PhCMV-p65-VP64-BldD-pA::PhCMV-BphS-P2A-YhjH-pA, 135 ng), pYH414 [PFRL-NLS-Cas9(N)-Linker-Coh2-pA, 77 ng], and pYH420 [PU6-sgRNA (PLK1)::PhCMV-DocS-Linker-Cas9(C)-NES-pA, 288 ng]} were illuminated by FRL (1 mW/cm2; 730 nm) for 4 hours once a day for 2 days and were then collected at 48 hours after the first illumination for analysis. The subsequent procedures were performed according to the manufacturers instructions and analyzed by flow cytometry (BD LSRFortessa cell analyzer; BD Biosciences). The LSRFortessa was equipped with green fluorescence channel (488-nm laser, 530/30 nm emission filter, 505 nm longpass dichroic mirror) and red fluorescence channel (561-nm laser, 610/20 nm emission filter, 595 nm longpass dichroic mirror). A gate was applied on forward scatter and side scatter to remove debris from cell populations. Data were analyzed using the FlowJo V10 software.

Total RNA of cells or tissues was extracted using the RNAiso Plus kit (catalog no. 9109; Takara Bio). A total of 500 ng of RNA was reverse transcribed into cDNA using a PrimeScript RT Reagent Kit with the genomic DNA Eraser (catalog no. RR047; Takara Bio). Quantitative PCR (qPCR) reactions were performed on the LightCycler 96 real-time PCR instrument (Roche Life Science) using the SYBR Premix Ex Taq (catalog no. RR420; Takara Bio). Program for qPCR amplifications were as follows: 95C for 10 min, followed by 40 cycles at 95C for 10 s, 60C for 15 s, and 72C for 10 s, and then 95C for 10 s, 60C for 60 s, 97C for 1 s, and last, 37C for 30 s. The qPCR primers used in this study are listed in table S4. Samples were normalized to the housekeeping gene glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as the endogenous control. Standard Ct method was used to obtain relative mRNA expression level.

Wild-type mice [8 week old, male, C57BL/6J, East China Normal University (ECNU) Laboratory Animal Center] were randomly divided into two groups. The semipermeable KrosFlo polyvinylidene fluoride hollow fiber membrane (Spectrum Laboratories Inc.; notably, the light-absorption properties of this material to lights of 300 to 1000 nm are almost the same) implants containing optogenetically engineered HEK-293 cells (pairs of 2.5-cm hollow fibers containing a total of 5 106 engineered cells) were subcutaneously implanted beneath the dorsal skin of the mice under anesthesia (two 2.5-cm hollow fibers in each mouse). At 1 hour after implantation, the mice were illuminated by FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off, alternating, to avoid the thermal discomfort in mice caused by continuous illumination) for 4 hours once a day for 2 days. The control mice were kept in dark. Cells were then collected from the implanted hollow fibers at 48 hours after the first illumination, and the genomic DNA was extracted for mismatch-sensitive T7E1 assay to quantify the indel mutations of the endogenous gene CCR5.

The transgenetic Ai14 tdTomato reporter mice [6 week old, female, Gt(ROSA)26Sortm14(CAG-tdTomato)Hze, from the Jackson laboratory; Ai14 is a Cre reporter allele designed to have a loxP-flanked stop cassette, preventing the transcription of a CAG promoterdriven red fluorescent tdTomato, all inserted into the Gt(ROSA)26Sor locus] were randomly divided into three groups (vehicle, FAST without illumination, and FAST with FRL). The minicircle DNA vectors encoding the FAST system {pYH412 (PhCMV-p65-VP64-BldD-pA::PhCMV-BphS-P2A-YhjH-pA, 81 g), pYH413 [PU6-sgRNA (tdtomato stop cassette)::PhCMV-DocS-Linker-Cas9(C)-NES-pA, 173 g], and pYH414 [PFRL-NLS-Cas9(N)-Linker-Coh2-pA, 46 g]} were dissolved in Ringers solution [NaCl (8.6 g/liter), KCl (0.3 g/liter), and CaCl2 (0.28 g/liter)] and injected into mices tail vein by hydrodynamic injection. The injection volume of the DNA mixture solution was 100 l per mouse weight (gram). Twenty-four hours after injection, mice were illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off, alternating, to avoid the thermal discomfort in mice caused by continuous illumination) for 4 hours per day for 3 days (according to the time schedule in Fig. 5A). A second-round injection of the minicircle-based FAST system was performed on the fifth day, followed by 4 hours of daily illumination for three additional days. On the 15th day after the first hydrodynamic injection, mice were euthanized, and the livers were isolated for fluorescence imaging or histological analysis. The tdTomato signal from isolated liver was detected using IVIS Lumina II in vivo imaging system (PerkinElmer, USA) and frozen tissue section histological analysis.

First, dissected liver tissue blocks were soaked in 4% paraformaldehyde for 2 hours. Subsequently, the tissue blocks were stepwise dehydrated with 15% sucrose solution overnight and then soaked in 30% sucrose solution for another 3 hours. After being washed three times with PBS, freshly dissected tissue blocks (<5 mm thick) were placed on to a prelabeled tissue base mold and embedded in Tissue-Tek optimal cutting temperature (O.C.T.) compound (catalog no. 4583; Sakura). These tissue blocks were stored at 80C for freezing until ready for sectioning. The tissues were sliced into frozen sections with 5-m thickness using Cryostat Microtome (Clinical Cryostat; CM1950; Leica) for further processing or stored at 80C ultralow-temperature freezer.

A total of 5 106 of A549 cells were suspended in 0.2 ml of sterile PBS and subcutaneously injected onto the back of the 6-week-old female BALB/c nude mice (ECNU Laboratory Animal Center). When the tumor size reached about 80 to 100 mm3, APC/FAST complex containing 20 l of APC and the minicircle iteration of the FAST system {pYH412 (PhCMV-p65-VP64-BldD-pA::PhCMV-BphS-P2A-YhjH-pA, 2.7 g), pYH414 [PFRL-NLS-Cas9(N)-Linker-Coh2-pA, 1.5 g], and pYH420 [PU6-sgRNA (PLK1)::PhCMV-DocS-Linker-Cas9(C)-NES-pA, 5.8 g]} were injected intratumorally. These injected mice were randomly divided into two groups (dark and illumination). Injections were conducted under anesthesia once every 2 days for five times. Twenty-four hours after every injection, mice were illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off, alternating, to avoid the thermal discomfort in mice caused by continuous illumination) according to the time schedule in Fig. 6A or kept in dark. Mice of the vehicle control group were intratumorally injected with 20 l of APC and 50 l of PBS and were then illuminated with FRL (10 mW/cm2; 730 nm; 2-min on, 2-min off), as indicated in Fig. 6A. The tumor sizes and the body weights of mice were measured every 2 days. On the 15th day after the first intratumor injection, all mice were sacrificed and tumor weights were recorded. The tumor volumes were measured using a digital caliper and calculated by the following formula: tumor volume = [length of tumor (width of tumor)2]/2. Then, tumors were isolated for indel mutation analysis and tumor apoptosis detection by hematoxylin and eosin (H&E) staining, TUNEL, and caspase-3labeling assays.

Glass slides that hold the frozen tissue sections were washed with PBS three times for 5 min each time, transferred to 0.5% Triton X-100 (dissolved in PBS; Sigma-Aldrich) for 10 min, and washed with PBS twice for 5 min each time. The slides were rinsed in running tap water at room temperature for 1 min. The samples were then stained in hematoxylin staining solution (catalog no. E607317; Sangon Biotech Inc.) for 8 min and washed in running tap water for 10 min. Next, the samples were differentiated in 1% acid alcohol for 10 s, washed in running tap water for 30 min, and were then counterstained in eosin staining solution (catalog no. E607321; Sangon Biotech Inc.) for 30 s to 1 min and washed in running tap water for 10 min. Last, the tissue sections were sealed by a drop of mounting medium over the tissue and then covered by a coverslip. The prepared slides were then observed by a microscope (DMI8; Leica) equipped with an Olympus digital camera (Olympus DP71; Olympus).

A TUNEL Apoptosis Assay Kit (catalog no. 30063; Beyotime Biotechnology Inc.) was used to evaluate tumor tissue apoptosis according to the manufacturers instructions. After washing three times with PBS, the slides were incubated with 4,6-diamidino-2-phenylindole (DAPI) solutions (5 g/ml; catalog no. C1002; Beyotime Inc.) for 2 to 5 min at room temperature. The slides were further washed three times with PBS and mounted with the antifade mounting media. Last, the slides were sealed and observed by a fluorescence microscope (DMI8; Leica) equipped with an Olympus digital camera (Olympus DP71; Olympus). TUNEL-positive nuclei were stained green, and all other nuclei were stained blue.

Isolated tumor frozen tissue sections were thawed at room temperature for 15 min and rehydrated in PBS for 10 min. The tissue samples were surrounded with a hydrophobic barrier using a barrier pen after draining the excess PBS. Then, the slides were soaked in 0.5% Triton X-100 (dissolved in PBS; catalog no. 9002-93-1; Sigma-Aldrich) for 20 min. Nonspecific staining between the primary antibodies and the tissue samples was blocked by incubating sections in the block buffer (1% FBS in PBS) for 1 hour at room temperature. After incubating with the anticaspase-3 antibody (1:100; catalog no. ab32351; Abcam) overnight at 4C, the slides were washed three times for 15 min each time in PBS and then incubated with the Alexa Fluor 555 goat anti-rabbit immunoglobulin G antibody (1:500; catalog no. ab150078; Abcam) for 1 hour at room temperature. After washing three times with PBS, the slides were incubated with DAPI solutions (5 g/ml; catalog no. C1002; Beyotime Inc.) for 2 to 5 min at room temperature. The slides were further washed three times with PBS and mounted with the antifade mounting media. Last, the slides were sealed and observed by a fluorescence microscope (DMI8; Leica) equipped with an Olympus digital camera (Olympus DP71; Olympus). Caspase-3positive cytoplasm was stained red, and all nuclei were stained blue.

All experiments involving animals were conducted in strict adherence to the guidelines of the ECNU Animal Care and Use Committee and in direct accordance with the Ministry of Science and Technology of the Peoples Republic of China on Animal Care. The protocols were approved by the ECNU Animal Care and Use Committee (protocol IDs, m20180105 and m20190607). All mice were euthanized after the termination of the experiments.

All in vitro data represent means SD and are described separately in the figure legends. For the animal experiments, each treatment group consisted of randomly selected mice (n = 3 to 5). Comparisons between groups were performed using Students t test, and the results are expressed as means SEM. GraphPad Prism software (version 6) was used for statistical analysis.

Acknowledgments: We are grateful to all the laboratory members for cooperation in this study, especially J. Jiang, S. Zhu, and X. Yang. Funding: This work was financially supported by the grants from the National Key R&D Program of China, Synthetic Biology Research (no. 2019YFA0904500), the National Natural Science Foundation of China (NSFC; no. 31971346 and no. 31861143016), the Science and Technology Commission of Shanghai Municipality (no. 18JC1411000), the Thousand Youth Talents Plan of China, and the Fundamental Research Funds for the Central Universities to H.Y. This work was also partially supported by NSFC no. 31901023 to N.G. We also thank the ECNU Multifunctional Platform for Innovation (011) for supporting the mouse experiments and the Instruments Sharing Platform of School of Life Sciences, ECNU. Author contributions: H.Y. conceived the project. H.Y. and Y.Y. designed the experiment, analyzed the results, and wrote the manuscript. Y.Y., X.W., J.S., H.L., and Y.C. performed the experimental work. Y.P., D.L., and N.G. analyzed the results and revised the manuscript. All authors edited and approved the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. All genetic components related to this paper are available with a material transfer agreement and can be requested from H.Y. (hfye{at}bio.ecnu.edu.cn).

See original here:
Engineering a far-red lightactivated split-Cas9 system for remote-controlled genome editing of internal organs and tumors - Science Advances