Efficient CRISPR editing with a hypercompact Cas12f1 and engineered guide RNAs delivered by adeno-associated virus

Overall strategy for gRNA engineering

Efforts to engineer gRNAs have contributed to improving CRISPR tools with respect to efficiency of indel generation²⁷, simplicity⁴, multiplexing²⁸, imaging²⁹ and specificity^4,30. To optimize the gRNA of Cas12f, we used knowledge from past gRNA engineering efforts and identified five potential modification sites (MSs) throughout the tracrRNA and crRNA sequences as follows: MS1, an internal penta(uridinylate) (UUUUU) sequence in the tracrRNA; MS2, the 3′ terminus of the crRNA; MS3, the ‘stem 1’ region of the tracrRNA; MS4, the tracrRNA–crRNA complementary region; and MS5, the ‘stem 2’ region of the tracrRNA. The gRNA engineering steps proceeded sequentially from MS1 to MS5 (Fig. 1a). As the thermodynamically unfavorable base-paring of the natural tracrRNA and crRNA (Supplementary Fig. 1a) would nullify the effects of gRNA engineering, we used a single-guide RNA (sgRNA) connected with a GAAA loop as a precursor for the series of gRNA engineering steps.

MS1: correcting an internal penta(uridinylate) sequence

The canonical Cas12f1 tracrRNA contained an internal UUUUU sequence that spanned positions −24 to −20 (numbered 5′ to 3′), as reported previously²⁵. The five consecutive thymidinylates in a template would prevent the production of a full-length tracrRNA under the H1 and U6 promoters³¹. Therefore, we designated the penta(uridinylate) site (MS1) as the starting point for gRNA engineering.

To remove the termination cue, we replaced each U with a non-U nucleotide and investigated the indel frequencies at an endogenous target (target 1) in HEK293T cells (hereafter, please refer to Supplementary Table 1 for target information). Deep-sequencing analyses revealed that each substitution yielded at least a fourfold increase in indel frequency with a much higher increase (about 50-fold) by the substitution of U⁻²¹ with C (Fig. 1b). For further gain, we compared possible combinations for the penta(uridinylate) site by fixing the C substitution at the U⁻²¹ position, and substituting other uridines with the nucleotides, resulting in the highest indel frequencies, as shown in Fig. 1b. Comparative analysis revealed that substituting UUUUU with 5′-GUGCU in the tracrRNA further increased the efficiency of indel generation (Supplementary Fig. 1b). A similar screening and comparative analysis found that 5′-AGCAA in the crRNA was an optimal counterpart for the 5′-GUGCU in the tracrRNA (Supplementary Figs. 1c–e). The substitutions in the crRNA (that is, 5′-AGCAA) alone did not increase the indel frequency, but significantly improved indel efficiencies were achieved with the concomitant modification of crRNA and tracrRNA at MS1 (Fig. 1c).

MS2: adding 3′-poly(uridinylates) to the crRNA

Previously, we reported that a poly(uridinylated) (U-rich) 3′ overhang on the crRNA increased Cas12a-mediated indel frequencies, making them comparable to those of SpCas9 (ref. ³²). As Cas12f1 shows a similar domain architecture to Cas12a, we explored whether a similar U-rich crRNA modification (MS2) would affect Cas12f1-mediated indel frequencies. In line with our previous results, the addition of Ts stimulated Cas12f1 activity until a T₅ or T₆ termination sequence was generated (Supplementary Fig. 2a). An adenylate (A) was incorporated after TTTT to obviate the termination signal for the U6 promoter, and the number of uridines in the crRNA was further increased by adding thymidinylates next to the adenylate. When a 5′-TTTTATTTTTT sequence was added to the 3′ terminus of the crRNA to create a U-rich 3′ overhang, indel frequencies were maximized. We changed the intervening A into C or G and found that a 5′-U₄RU₄ (R = A or G) was an optimal overhang in that position (Supplementary Fig. 2b; please refer to our previous report³² for detailed information). When combined in an sgRNA, the MS1 and MS2 modifications showed synergistic effects, yielding significant increases in indel frequencies, by up to 1,148-fold for target 1 (Fig. 1d). In contrast, we observed only marginal levels of indel frequencies with a canonical gRNA (percentage indels ≤0.1%). The MS1-/MS2-modified sgRNA was then subjected to further rounds of gRNA engineering.

MS3: truncating the 5′ region of the tracrRNA

Cas12f1 has an exceptionally long gRNA due to an oversized tracrRNA²⁵. We hypothesized that the entire tracrRNA is unlikely to participate in interactions with the compact Cas12f1. A recent study also indicated that the stem 1 region is in a structurally disordered state^33,34. To test this hypothesis, we either extended or trimmed the tracrRNA in the stem 1 region. A tracrRNA with an 18-nt truncation caused a significant increase in indel frequency (Supplementary Fig. 3a). To determine the optimal truncation, we tested a set of tracrRNA truncations beginning at positions −149 to −137. The results confirmed that a 5′ truncation of 18–21 nt yielded a highly potent sgRNA (Fig. 1e) and used a tracrRNA with a 20-nt truncation as a basis for additional engineering steps.

MS4: truncating the tracrRNA–crRNA complementary sequence

We sought to further trim the sgRNA without compromising indel frequencies at the crRNA–tracrRNA complementary region. We generated sgRNAs with different lengths of the tracrRNA–crRNA complementary region (Supplementary Fig. 3b). An elongated crRNA was suggested to improve the function of the CRISPR–Cas12a system³⁵; thus, we tested an elongated sgRNA (+13 bp) in addition to four shortened sgRNAs. The elongated sgRNA showed a marked decrease in indel-generating efficiency, but all trimmed sgRNAs retained efficiency.

To pinpoint the optimally trimmed sgRNA, we tested various truncations at 1-bp resolution. Notably, a truncation of the entire complementary sequence resulted in even higher editing efficiency (Fig. 1f). The truncated region exactly matched the previously reported disordered region^33,34.

We tested substituting the GAAA tetraloop linking the tracrRNA and crRNA with a hammerhead ribozyme to produce an sgRNA at the expression stage, followed by generation of a dual gRNA with an overhang on either the tracrRNA or the crRNA after self-cleavage (Supplementary Fig. 3c,d). However, none of these possibilities showed increased efficiency compared with the sgRNA with a total of 59-nt truncation.

MS5: truncating the stem 2 region in the tracrRNA

We considered one more possible modification site, the tracrRNA stem 2 region, because the segment spanning from A⁻¹²⁹ to U⁻¹⁰³ was also reported to be disordered^33,34. Thus, we also trimmed this region at 1-bp resolution, keeping the 5′-UUAG loop preserved. Though the modified gRNAs did not further enhance the efficiency achieved by MS2/MS3/MS4 gRNA at target 2, deletion from C⁻¹³¹ to G⁻¹⁰¹ increased indel frequencies by about 2.7-fold at a GAK locus (Supplementary Fig. 3e). Validations using more targets revealed that the 27-nt truncation at the stem 2 region mediated increased indel frequencies, particularly for targets with relatively low indel frequencies generated by gRNA MS2/MS3/MS4 (Supplementary Fig. 3f). The truncation of the stem 4 region nullified the MS1–MS5 engineering (Supplementary Fig. 3g), and was not included for final engineering. Various combinations of MS modifications yielded different increases in indel frequencies (Fig. 1g). When MS1/MS2/MS3, MS2/MS3/MS4 or MS2/MS3/MS4/MS5 modifications are combined, the engineered Cas12f1 systems showed the highest genome-editing performance and thus are referred to as Cas12f_ge3.0, Cas12f_ge4.0 and Cas12f_ge4.1, respectively. It is noteworthy that the MS1 site was removed during MS4 engineering and, therefore, MS1 and MS4 engineering are mutually exclusive. The Cas12f1_ge4.1 system is characterized by highly efficient, very compact genome editors, with gRNA down-sized by almost 40% (Fig. 1h). Taken together, our extensive gRNA engineering efforts yielded a potent, extremely compact CRISPR–Cas12f1 system.

Finally, we sought to explain how each gRNA modification (MS1–MS5) contributes to increased indel frequencies using targeted RNA-sequencing (RNA-seq) analysis (Supplementary Fig. 4a). As expected, the MS1 engineering led to a drastic increase in the expression of the full-length sgRNA (Supplementary Fig. 4b,c). Besides increasing the affinity of the Cas–gRNA interaction as suggested previously³², the U-rich 3′ overhang appeared to stabilize the sgRNA transcript in cells. The MS3 engineering was not associated with changes in gRNA expression, but further increased the dsDNA cleavage activity of Cas12f1 when stacked to the MS1/2 modifications (Supplementary Fig. 4d). The MS4 and MS5 engineering further increased the cleavage activity. To validate the effects of the MS3 modification on indel frequencies, MS1/2- and MS1/2/3-modified gRNAs were compared with respect to indel-generating efficiency in vivo. Out of 19 targets tested, 17 (90%) showed increased indel frequencies, by at least twofold, with the average fold increase being 3.12 (Supplementary Fig. 4e). The structures of the Cas12f_ge3.0, Cas12f_ge4.0 and Cas12f_ge4.1 gRNAs are presented in Supplementary Fig. 5.

Large-scale validation of Cas12f

We next investigated whether the increased genome-editing efficiency of the engineered gRNAs can be validated at a wider range of targets. We searched in silico for endogenous targets containing the sequence 5′-TTTR-N₂₀-NGG-3′, which are targetable with SpCas9, AsCas12a and Cas12f1 (Fig. 2a). We randomly selected 88 such endogenous loci (for target information, please refer to Supplementary Table 2) and measured the SpCas9-, AsCas12a- and Cas12f-mediated indel frequencies in HEK293T cells. Cas12f with canonical gRNAs generated indel frequencies of <1.0% over all tested targets, with 91% (80 of 88) of targets showing frequencies of <0.1%. However, use of our engineered gRNAs led to significant increases in indel frequencies at most target sites (Fig. 2b). The average efficiency of Cas12f_ge4.1 was comparable to that of SpCas9 (P > 0.05) and was even higher than that of AsCas12a (Fig. 2c). The average increase in efficiency induced by Cas12f_ge4.1 sgRNA was 867-fold. Notably, Cas12f_ge4.1 had more targets with high indel frequencies (≥50%) than SpCas9 and AsCas12a (Fig. 2d). In addition, Cas12f_ge4.1 showed higher efficiencies for 76.1% (67 of 88) of targets, compared with the Cas12f_ge3.0 and Cas12f_4.0 versions, whereas the Cas12f_ge4.0 and Cas12f_ge3.0 versions were most effective for 17.0% and 6.8% of targets, respectively (Fig. 2e).

We then sought to refine the Cas12f system further, because there still remained targets resistant to genome editing by Cas12f (in fact, the situation is also true for Cas9 and Cas12a, but Cas12f1 showed more targets with indel frequencies <1% than SpCas9). We hypothesized that the low efficiency of Cas12f1 at certain sites may originate from different cleavage efficiency between target and nontarget strands, because the compact size of Cas12f might cause less efficient nontarget strand cleavage. To test this hypothesis, we selected targets that carry a 5′-TTTR-N₂₀-spacing-N₂₀-YAAA-3′ sequence, where ‘spacing’ is a 10- to 80-bp-long dsDNA segment (Fig. 2f). These sequences are targetable by a pair of gRNAs oriented in opposite directions; two dsDNA cleavage events occur in the spacing region. Although each gRNA alone mediated relatively low indel frequencies, targets in ten loci showed sharply increased indel frequencies with the paired gRNAs. The fold increase varied among targets, but all tested targets showed indel frequencies of >1%. Moreover, final indel frequencies were further improved by using MS1/MS2/MS3- versus MS1/MS2-modified gRNAs (Fig. 2g), mainly because indel-generating efficiencies of each gRNA were increased by MS3 engineering. However, the fold increase was more pronounced for MS1/MS2 engineering, compared with the MS1/MS2/MS3 and MS2/MS3/MS4/MS5 versions (Fig. 2h). This result would be explained by our hypothesis that Cas12f1 displays unequal cleavage kinetics for the target and nontarget strands, and that the degree of difference is reduced by MS3 engineering. A longer spacing region of ≥50 bp did not yield this pair gRNA-assisted increase in indel frequencies (Fig. 2i).

Favorable kinetic property for Cas12f-induced DNA cleavage

In addition to the compactness of Cas12f1, this system has an additional advantage for gene therapy: it induces dsDNA cleavages outside the protospacer sequence^33,36. This property implies that, even after the initial round of NHEJ-mediated indel mutations, the protospacer sequence is likely to remain unchanged. Then, further rounds of the dsDNA cleavage–NHEJ process can continue (Fig. 3a). This property is even more desirable for a large DNA-deletion strategy involving a pair of gRNAs. We analyzed the profile of indel mutations induced by Cas12f. Most mutation patterns included relatively long deletions that affected the protospacer sequence (Supplementary Fig. 6a,b). In contrast, indel mutations outside the protospacer were relatively rare. We interpreted these long deletions to be the products of multiple cutting-and-joining processes. In fact, this assumption was confirmed through a time-course investigation of indel patterns. In the early phase of transfection, deletions of <5 bp were dominant (Fig. 3b; the radius of a bubble indicates the mutation frequency). However, the frequency of long deletions increased over time until 4 d later. In contrast, the pattern of indel mutations was almost consistent over time for SpCas9 and LbCas12a. Moreover, Cas12f caused a more persistent increase in indel frequencies, compared with SpCas9 and LbCas12a (Fig. 3c).

A handful of genetic disorders can potentially be treated by deletion of pathogenic introns or exons using paired gRNAs and Cas proteins, including Duchenne muscular dystrophy³⁷, Leber congenital amaurosis 10 (LCA10)³⁸ and Usher’s syndrome type 2A³⁹. We explored the potential utility of the Cas12f system for those applications. As a case study, we selected a pair of sites in the vicinity of exon 51 of the human dystrophin gene that are common targets for LbCas12a and Cas12f. Screening experiments identified target sequences that show similar indel frequencies for LbCas12a and Cas12f. Despite the similar indel efficiencies of individual gRNAs, Cas12f resulted in a higher level of deletions, compared with LbCas12a (Fig. 3d). These results indicate that Cas12f might be particularly useful for AAV delivery in gene therapy applications that require deletions.

AAV delivery of the engineered Cas12f system

Next, we investigated the genome-editing performance of a recombinant AAV2 (rAAV2)–Cas12f vector. We constructed an rAAV vector carrying sequences encoding either Cas12f_ge4.1 or a control vector (scrambled sgRNAs). Cas12f1 and sgRNA expression were driven under the control of the chicken β-actin and the human U6 promoters, respectively (Supplementary Fig. 7a). The total length of these sequences (4.40 kb) fell within the permissive size for an AAV payload, even in the presence of two sgRNA sequences and an enhanced green fluorescent protein (eGFP)-encoding reporter sequence. The rAAV2 particles were produced in HEK293T cells after transfection with an rAAV vector, pAAVED2/2 and a helper plasmid. The sgRNAs respectively targeted an intergenic locus (target 1) and the KRT1 gene (target 2).

AAV delivery to HEK293T cells led to an increase of the frequencies of indel mutations over time (Supplementary Fig. 7b) and with increasing numbers of rAAV2 particles (Supplementary Fig. 7c). The infection was monitored by green fluorescence, which was persistent for 2 weeks post-transduction (Supplementary Fig. 7d).

Next, we explored the targeting of therapeutically useful loci for the deletion of a pathogenic cryptic exon in the CEP290 gene for the treatment of LCA10 (ref. ³⁸). We tested on both sides of the c.2991+1655A>G mutation site and identified a pair of highly potent sgRNAs (Fig. 3e). We then constructed an rAAV vector carrying the Cas12f1_ge4.1 system, and the deletion-inducing efficiency of Cas12f was compared with that of SaCas9 (EDIT101, a gene therapeutic agent under clinical trial) in HEK293T cells. We observed higher levels of deletions on agarose gels for the Cas12f system (Fig. 3f). Quantitative analysis using droplet digital PCR showed a 46% higher deletion rate of Cas12f, compared with EDIT101 (Fig. 3g). These results indicate that Cas12f might provide a versatile and valid genome-editing platform for gene therapy.

When using an elongation factor (EF)-1α core promoter, a bovine growth hormone (BGH) poly(A) signal sequence, a U6 promoter and an XTEN linker between sequences encoding Cas12f1 and a potential fusion partner are used, we have an upper limit of approximately 2.1 kb for a fusion partner gene for AAV delivery. Considering the sizes of genes encoding validated regulators, we propose that the Cas12f system could provide a scaffold for various applications including CRISPR interference (CRISPRi)⁴⁰, CRISPR activation (CRISPRa), base editing^8,9, prime editing¹⁰ and site-specific epigenetic regulations^5,6 (Fig. 3h). The possibility of such applications was explored in a CRISPRa strategy, where dCas12 (D510A) fused to VP64 activated transcriptional expression of OCT4 gene in a gRNA-dependent manner (Fig. 3i).

Genome-editing specificities of Cas12f

Considering the persistent activity of Cas12f in cells (Fig. 3c,f,g), it is particularly important to examine the specificity of this system. First, we assessed the activity of Cas12f when gRNA_ge4.1 contained single- or adjacent two-base mismatches with the protospacer complementary sequence. Certain levels of tolerance were observed for single-base mismatches, particularly at positions 1–3, 5 and 17–20 (Fig. 4a). To compare the results with that of Cas12a^32,41, Cas12f showed lower tolerance in the protospacer-adjacent motif (PAM)-proximal regions and similar or slightly higher tolerance in the PAM-distal regions (positions 17–20). However, Cas12f exhibited less tolerance for mismatches in the middle region (positions 6–16). Moreover, Cas12f showed negligible levels of tolerance for two-base mismatches, except for positions 19/20, again similar to Cas12a.

Next, we employed targeted approaches to assess specificity. Using Cas-OFFinder⁴², we selected potential off-target sites that contained three base mismatches, but no bulges, with a set of on-target sites in P2RX5-TAX1BP3, CLIC4, NLRC4 and an intergene region, for which Cas12f showed higher on-target efficiencies than Cas12a (Fig. 4b and Supplementary Table 3). Deep-sequencing analysis revealed that Cas12f was more specific than AsCas12a: whereas AsCas12a resulted in residual levels of indels (<0.1%) at two off-target sites and an indel frequency of 0.36% at one other site among a total of 26 potential off-target sites, Cas12f_ge4.0 and _ge4.1 resulted in an indel frequency of 0.04% at each one of the potential off-target sites. We also compared genome-editing specificity for targets in RPL32P3, PRKCH and EMX1, for which Cas12a was previously observed to induce off-target effects^41,43 (Supplementary Table 3). On the whole, Cas12f and AsCas12a induced similar off-target effects, except for the off-target sites that had a single mismatch in the PAM-distal region (OF1–3 for RPL32P3; Fig. 4c).

We next employed the Digenome-sequencing (Digenome-seq) analysis to further examine the specificity of Cas12f⁴⁴. Three targets (RPL32P3, CLIC4 and P2RX5-TAX1BP3) were selected to compare the specificity of AsCas12a and Cas12f_ge4.1. Analysis of the Integrative Genomics Viewer (IGV) files from the Cas12f_ge4.1 experiments shows a presence of gaps between forward- and reverse-strand reads at both on-target and off-target sites (Fig. 4d), which is assumed to arise from either the ssDNA cleavage activity by the cleavage-activated Cas12f²⁵ or a generation of 3′ overhang. The Digenome-seq analysis revealed that gRNAs targeting RPL32P3, CLIC4 and P2RX5-TAX1BP3 showed off-target activity for Cas12f at 57, 51 and 19 loci, respectively, which were similar to or smaller in number than 57, 87 and 27, respectively, for AsCas12a (Fig. 4e and Supplementary Table 4). Intrinsically, Cas12f would be expected to show fewer off-target sites than Cas12a because of the more restricted preference of PAMs³⁶. We then validated the nine potential off-target sites for RPL32P3 by measuring Cas12f- and AsCas12a-mediated indel frequencies. The indel frequencies at the on-target site were similar for Cas12f and AsCas12a. Similarly, the indel frequencies at off-target sites were not significantly different between the two CRISPR systems, although Cas12f showed slightly higher off-target activity at the sites with a mismatch in the PAM-distal region, in line with Fig. 4a. In addition, a certain level of indel frequencies was observed for noncanonical TTTR PAM, such as GTTG and ATTG, for both enzymes. However, the overall indel frequencies at the investigated sites were similar between Cas12f and AsCas12a, indicating that Cas12f shows high genome-editing specificity comparable to Cas12a (Fig. 4f). The Cas12f system not only recognized fewer off-target sites, but also resulted in lower off-target/on-target indel frequency ratios. Despite the lower off-target activity, Cas12f showed long deletions (up to ~10 kb), as is observed for SpCas9 and AsCas12a (Supplementary Fig. 8), which requires further scrutiny⁴⁵.