Design of chemically inducible split Cas13
We set out to generate an inducible system by splitting the Cas13 protein. From among the diverse orthologs of the CRISPR-Cas13 system, we selected PspCas13b, because it is known as the most efficient Cas13 protein for fluorescent labeling of specific target RNA, and exhibits better target specificity than other Cas13 proteins6. In addition, the crRNA of PspCas13b has a relatively long spacer sequence (at least 30 nucleotides) that could provide better specificity than Cas13 orthologs with shorter spacer sequences.
To split the protein, we needed to select solvent-accessible loops whose cleavage would not disturb any secondary structure element such as an α-helix or β-sheet (Fig. 1a). Because the three-dimensional structure of PspCas13b has not yet been revealed, we performed a deep learning-based prediction of the protein structure using AlphaFold213,14. To improve the prediction and build a structure with a high per-residue confidence score called predicted local distance difference test (pLDDT) indicating reliable folding results, we conducted up to 12 additional recycling processes, as guided by performance improvements observed in previous research13. The average pLDDT score was 89.04, suggesting that there was a high confidence in the predicted structure. We further reconstructed functional domains with reference to the structure of Prevotella buccae Cas13b (PbuCas13b)15, and then examined the final predicted protein structure to identify candidate split sites (Fig. 1b).
We chose eight candidate split sites based on their localization in loop regions and their amino acid properties, taking care to avoid the terminal HEPN domains so as to maintain the nuclease activity of Cas13 (Fig. 1a). To develop the chemically inducible system, which was used to test the candidates, we coupled each split fragment pair to the interaction partners of the rapamycin-inducible dimerization system, FKBP and FRB, using a relatively long GS linker to avoid steric hinderance of nuclease activity (Fig. 1c).
To investigate the efficiency of each candidate chemically inducible split Cas13 system, we conducted an inducibility test. To minimize well-to-well variations, we used a dual-luciferase assay comprising a single plasmid containing the firefly (fluc) and Renilla (Rluc) luciferase genes controlled by distinct promoters (Fig. 1c), and normalized luciferase activity as the fluc/Rluc ratio. The split Cas13 mainly functions in the cytosol due to the presence of a nuclear export signal (NES) tag, which mediates its transport from the nucleus. The cytosol is abundant with RNA-binding proteins that interact dynamically with RNAs. In an effort to improve the accuracy of our screening of split Cas13 candidates, we incorporated a λ2 bacteriophage sequence16, which can be targeted for RNA degradation and is not present in human cells, into the 3’ UTR after the stop codon. This strategy aimed to minimize the potential for interference from endogenous RNA-binding proteins and thereby enhance the precision of our screening in terms of coding region translation and RNA degradation (Fig. 1c). We then applied rapamycin for 24 hours and assessed rapamycin-induced nuclease activity. Our results revealed that the split site N351/C352 could function effectively for rapamycin-inducible RNA degradation, while the split sites N550/C551, N624/C625, and N725/N726 exhibit background activity (Fig. 1d and Supplementary Table 1).
To characterize background activity, we used both experimental and computational approaches. We transfected individual dimerization domain-fused split Cas13 fragments with the targeted crRNA. We found that the background activity was unrelated to the residual nuclease activity of the split Cas13 fragments (Supplementary Fig. 1a). In addition, a split-firefly luciferase assay was conducted to clarify whether the fragments could spontaneously reconstitute to show activity17. From our previous screening under rapamycin conditions, we selected three distinct split sites based on their RNA degradation activity: N272/C273 with lower inducible activity, N351/C352 with ligand-inducible activity, and N550/C551 with spontaneous background activity. These experiments were conducted without crRNA to specifically avoid crRNA-dependent effects. Each split fragment was fused with split-firefly luciferase and both were expressed together in HEK 293 T cells. The results demonstrated that the split sites N272/C273 and N351/C352 with inducible activity exhibited no signal of firefly luciferase without dimerization domain, but the split site N550/C551 with background activity exhibited a luciferase signal (Fig. 1f). Therefore, we suggest that the background activity of split Cas13 fragments is caused by spontaneous reconstitution of split sites.
Second, to determine whether the structural interaction of split Cas13 could affect background activity, we analyzed the experimentally validated combinations and all candidate split fragments using AlphaFold2-based structural prediction13. Granted, AlphaFold2 is not fully reliable in predicting ribonucleoprotein structures, and Cas13 undergoes significant conformational changes upon crRNA binding15. However, AlphaFold2 is a valuable tool for bridging the gap between experimental and computational insights, such as in our experimental context. The surface interactions of these fragments were further investigated using MaSIF-site18, which is a geometric deep learning-based tool that provides molecular surface interaction fingerprinting (Supplementary Fig. 2a). When combined, the results from the experimental and computational analyses suggested that there is a relationship between auto-assembly activity and the extent of the labeled interface. Specifically, N550/C551, N624/C625, and N725/C726 exhibited higher auto-assembly activity in the experimental results (Fig. 1d) and displayed more extensive labeled interfaces and open structural interaction pockets in the computational analysis (Supplementary Fig. 2b, c). Notably, these high-activity combinations are located within the crRNA binding region of PspCas13b. This region exhibits elevated positive electrostatic potential, which may underlie the enhanced auto-assembly activity observed at the high-activity split sites (Supplementary Fig. 2d).
We further screened Cas13 split sites around amino acid 351 to identify the most effective split-site (Fig. 1e). Based on our collective results, we selected inducible split pairs, split Cas13-1 (N351/C352) and split Cas13-2 (N351/C350), for use in subsequent experiments. Thus, we herein developed a chemically inducible RNA degradation system that functions by reassembling Cas13 from split fragments, which appears to be the most effective strategy in terms of providing a low background activity and high inducibility.
Development and characterization of a photoactivatable Cas13
After establishing a chemically inducible RNA degradation system by reassembling Cas13 from split fragments, we placed this system under optogenetic control. The protein sizes are similar between the rapamycin-inducible dimerization system and the Magnet system used for photoinducible dimerization, allowing us to smoothly transition to a light-controlled approach. To produce a light-inducible Cas13 system, we fused the Magnet proteins, which undergo photoinducible dimerization, with fragment pairs having the same configuration as our chemically inducible split Cas13 (Fig. 2a). The Magnet system consists of a positively charged Magnet (pMag) and a negatively charged Magnet (nMag), which quickly heterodimerize under light illumination12. There are several engineered Magnet systems based on kinetic mutations within the Per-Arnt-Sim (PAS) core12. Because previous studies of photoactivatable Cas9 utilized the combination of pMag and nMagHigh1 (which exhibit high affinity) and provided information on the photoswitchable properties of the nuclease19, we used pMag and nMagHigh1 (nMagH) in developing our photoactivatable Cas13 system (named paCas13).
To validate the light-inducible dimerization of paCas13, we evaluated the target RNA knockdown activity using the dual-luciferase system and a light-emitting diode (LED) plate emitting 488-nm blue light. The split sites selected from the above-described experiments were assessed in the paCas13 system to identify whether its inducibility was consistent with that of the split Cas13 system. We transfected HEK 293 T cells with paCas13 fragments and target crRNA along with a dual-luciferase reporter containing the λ2 sequence, and then evaluated the light inducibility for 24 hours. We observed light-inducible RNA degradation mediated by reassembly of paCas13 and found that the luciferase activity patterns paralleled those obtained using the chemically inducible split Cas13 assay (Fig. 2b). To determine whether any fragment exhibited nuclease activity when expressed individually, we screened background activity by transfecting cells with vectors encoding each fragment, the crRNA, and a dual-luciferase reporter (Supplementary Fig. 1b). We found that background activity was independent of the remaining nuclease activity of each fragment. In accordance with the above results of the split Cas13, two light-inducible candidates were designated paCas13-1 (N351/C352) and −2 (N351/C350). The two differed by only one amino acid, but paCas13-1 had a lower background signal and paCas13-2 had greater RNA targeting activity, indicating that their light-inducible activity was substantially different (Fig. 2c). To determine the efficacy of RNA targeting with the paCas13, we performed a time-course assay. We observed that paCas13-2 displayed faster and more stable RNA degradation activity than paCas13-1 (Fig. 2d). This suggests that paCas13-1 and paCas13-2, which differ by a single amino acid, exhibit markedly distinct light-inducible activities. While paCas13-1 demonstrated lower background activity, making it a potentially safer choice for certain applications, paCas13-2 demonstrated better RNA targeting efficiency to achieve more rapid and stable RNA degradation.
As we sought to optimize the paCas13 system, we examined the alternative split Cas13 sites reported in recent studies10. To enable a more precise evaluation, we also targeted the λ2 sequence for luciferase readout; this provided an exogenous indicator and allowed us to monitor the system’s activity more clearly under induced and non-induced states. We compared the performance of the N351/ C350 and N761/C762 split sites under distinct inducible conditions, and we found that the N351/C350 split site exhibited high inducibility with substantially lower background activity than the previously characterized N761/C762 split site (Supplementary Fig. 3).
Since full-length PspCas13b was constructed with a C-terminal nuclear export sequence (NES)3, the cellular localization of each Cas13 fragment was investigated to confirm their stable expression. Different fluorescent proteins were fused to the C-terminus of each paCas13 pair, and their co-expression was examined (Fig. 2e). In the same cell under dark conditions, the EGFP-fused C-fragment displayed robust cytoplasmic translocation, while the mCherry-fused N-fragment was expressed throughout the entire cell, but displayed a predominant nuclear expression.
To investigate the factors that could influence cellular distribution, we performed a comprehensive investigation across all split candidates (Supplementary Fig. 4a). To determine the subcellular distribution of the HEPN domain, which is responsible for RNA cleavage, we constructed a fusion protein consisting of the HEPN domain and the mCherry fluorescent protein, and then performed real-time cellular monitoring. The expression was generally uniform across transfected cells, but we observed cellular toxicity and aggregation at 48 hours post-transfection in cells expressing the HEPN1 domain (Supplementary Fig. 4b). When split candidates were expressed individually, we observed nuclear localization patterns specifically for the only these N286, N351, and N624 fragments. The N286 fragment, the leading sequence to contain the positively charged inter-domain linker (IDL)15, was also the initial fragment to display a nuclear pattern (Supplementary Fig. 4a, c). This suggests that the IDL could be involved in directing N351 to the nucleus. The N624 fragment, meanwhile, demonstrated the distinct behavior of localizing in the nucleus when expressed alone and in the cytosol when co-expressed with its C625 partner (Supplementary Fig. 4c, d). This observation aligns with the experimental and computational data, which indicated that the N624/C625 combination has high auto-assembly activity and wide interfaces (Supplementary Fig. 2c). Interestingly, other fragments with high auto-assembly activity tended to form aggregates, possibly due to instability (Supplementary Fig. 4c, d). These observations suggest that electrostatic properties and protein-protein interactions contribute to the subcellular localization of the paCas13 system, which could affect its inducibility and background activity.
Building on these insights into the subcellular localization of these fragments, we examined how light stimulation could further modulate their interactions and subsequent functionality. After 24 hours of light stimulation, we observed a significant increase in the correlation between the N-terminal and C-terminal fragments, suggesting that the cytoplasmic recruitment of N351 was enhanced (Fig. 2e–g). These results support our contention that the paCas13 system functions as intended and provides further insights into its underlying mechanism. Although the N-fragment showed predominantly nuclear expression, enough was translocated to the cytoplasm to enable them to interact in that compartment.
Next, to confirm the ability of the paCas13 system to apply RNA interference, we targeted endogenous transcripts that had been previously targeted by a CRISPR-Cas13 system3,8. Although the paCas13 system does not reach the nuclease activity level of full-length Cas13, we observed significant light-induced RNA interference compared to dark conditions using paCas13-1 and −2 (Fig. 2h and Supplementary Tables 1, 2). To validate the robustness of these results, we included additional controls, including a cell viability test and an assessment of how light affects endogenous transcripts (Supplementary Fig. 1c, d). These controls confirmed that no significant changes were attributable to light exposure alone. Together, our results indicate that the paCas13 system can perturb endogenous transcripts through light illumination, with different levels of efficiency seen for paCas13 systems utilizing different split sites.
Optogenetically inducible RNA base editing by the padCas13 editor
Several techniques that use base editing composed of catalytically inactive Cas13 (dCas13) to deliver targeted RNA for therapeutic applications have been reported for certain diseases3,20. In these systems, dCas13 fused with the regulatory protein was generally expressed, such that the targeted RNAs were continuously affected with no spatiotemporal resolution.
To develop a light-inducible base editing system, we fused catalytically inactive paCas13 fragments with the ADAR2 (adenosine deaminase acting on RNA type 2) domain (ADAR2DD), which deaminates adenosine to inosine (which pairs with cytosine) in duplex RNAs3,21; this generated a system that we called the padCas13 editor (Fig. 3a). To characterize the activity of the padCas13 editor, we fused the wild-type ADAR2DD to the C-terminal ends of the C-fragments and constructed an RNA-editing reporter from the firefly gene of the dual-luciferase plasmid by generating a nonsense mutation, W417X (UGG to UAG). Unlike the wild-type reporter, the designed RNA-editing reporter contained a premature stop codon and exhibited no luciferase signal (Fig. 3a and Supplementary Fig. 5a). This mutation could be restored to the wild-type codon through A-to-I (adenine-to-inosine) editing, which would recover the firefly luciferase signal. To identify the optimal guide position and design for padCas13 editors, we also designed targeted crRNAs by tiling with 30- or 50-nucleotide spacers across the target region based on mismatch distance (Fig. 3b). The chemical and light-inducible editor systems both showed restoration of the luciferase signal by RNA editing effects at 24 hours post-induction in the tiling assay (Fig. 3c, Supplementary Fig. 5c, and Supplementary Table 3). The most effective tiled crRNAs in our system were those with a 50-nucleotide spacer and a 40-nucleotide mismatch distance. The padCas13-1 and padCas13-2 editors displayed similar editing patterns, with padCas13-2 exhibiting more efficient RNA editing (Fig. 3c). The crRNA-tiling-dependent editing patterns of the padCas13 editors were highly correlated with one another when compared to full-length Cas13 (Supplementary Fig. 5d, e). This was especially evident when the editing effects were normalized to a 0-1 scale; we obtained Pearson R values of 0.76 and 0.91 for padCas13-1 and −2, respectively.
To determine the efficacy and reversibility of RNA base editing with the padCas13 editor, we performed a time-course assay. The restoration of luciferase activity by padCas13 editors increased as the time of light stimulation increased without affecting cell proliferation and viability (Fig. 3d, g). To further validate the restoration of luciferase activity, we directly validated the level of RNA base editing by analyzing sequencing chromatograms generated for the targeted RNAs by reverse transcription-polymerase chain reaction (RT-PCR)22. We conducted time-course light stimulation experiments, measured RNA editing levels for up to 48 hours. The level of A-to-I RNA editing at the targeted region increased gradually over time, and the flat, dark-colored G (I) chromatogram signal increased in a time-dependent manner (Fig. 3e, f). Additionally, a comparative analysis under only dark conditions between non-targeted (NT) crRNA and targeted crRNA revealed a measurable increase in basal editing levels, indicating that additional factors, such as the crRNA itself, may contribute to the background editing activity (Supplementary Fig. 5b). Following these findings, we next investigated the reversibility of RNA base editing mediated by the padCas13 editor. When the light was turned off after 6 hours, luciferase activity returned to baseline within 24 hours (Fig. 3h). Further analysis of RNA editing levels also demonstrated reversibility, with editing levels aligning with the dark state within 24 hours (Fig. 3i). These results suggest that the padCas13 editor enables reversible RNA base editing under light with different kinetics observed between RNA and protein synthesis and translation.
Next, to improve the base editing effect of padCas13 editor, we replaced the ADAR2 deaminase domain with one harboring hyperactive mutations23. We observed that ADAR2DD with E488Q and T490A mutations exhibited an increased restoration rate in our light-inducible system compared to other ADAR2DD variants3,23, but had a higher base-editing signal in the dark state. To investigate the background activity of base editing, we transfected only the ADAR2DD (E488Q and T490A)-fused C-fragment of the padCas13 editor. Our results showed that the increased background signal appeared to depend on the ADAR2 variant but was unrelated to the specific C-fragment of padCas13 (Supplementary Fig. 6a, b). To verify this, we generated an additional group that carried stdMCP-ADAR2DD (E488Q and T490A) but not Cas13. Our results confirmed that the luciferase activity correlated with the ADAR2DD (Supplementary Fig. 6c). Because double-stranded RNA is required for the function of base editing, we also transfected the non-targeted and targeted crRNA. We found that the overexpression of the targeted crRNA had slight effects on both the luciferase restoration activity and the RNA editing rate (Supplementary Fig. 6c). Therefore, ADAR2DD protein overexpression can affect RNA editing in the absence of a crRNA.
To further confirm the base editing capabilities of our padCas13 editor, we compared our selected split site, N351/C350, with the previously characterized N761/C76210 under various induction conditions. This comparison demonstrated that our chosen site has lower background activity and is more suitable for RNA base editing (Supplementary Fig. 7). We also examined more cell types for the potential application of padCas13 editor. We observed light-inducible RNA base editing in human cervical carcinoma HeLa, breast cancer MCF7, fibrosarcoma HT1080, and mouse neuroblasts Neuro-2a cells (Supplementary Fig. 8). Across the alternative split site and different cell lines examined, the padCas13 editor demonstrated consistent light-inducible RNA base editing capabilities, underscoring its versatility and broad potential for diverse therapeutic applications.
To expand the range of disease mutations and protein modifications that can be targeted, we developed a light-inducible C-to-U (cytidine-to-uridine) RNA editing system by fusing padCas13 fragments to an evolved ADAR2DD capable of cytidine deamination20. To validate the activity of the padCas13 editor for C-to-U base editing, we designed a fluorescent protein-based indicator by introducing the Y66H green-to-blue mutation, Y66H (UAC to CAC), into GFP and developed a crRNA that would directly target the CAC codon with a 50-nucleotide spacer that included a 34-nucleotide mismatch distance and a U flip, resulting in a C-to-U conversion (Fig. 4a, b). Under light conditions, the padCas13 editor would correct the Y66H mutation to generate GFP fluorescence, which could be measured by cell imaging and flow cytometry. We transfected this padCas13 editor with cytidine deaminase and a targeted crRNA harboring the GFP indicator, and then evaluated the light inducibility after 24 hours of blue-light stimulation (Fig. 4c). Confocal microscopy revealed that the padCas13 editor-induced C-to-U RNA editing, thereby generating GFP-positive cells under the light condition (Fig. 4d and Supplementary Fig. 9). Moreover, flow cytometry-based quantification demonstrated that the light-induced C-to-U editing of the padCas13 editor increased the population of cells positive for the GFP signal from 6% (dark condition) to 18% (under light exposure) (Fig. 4e, f, and Supplementary Fig. 10). Additionally, we employed a photomask to demonstrate that we could spatially control padCas13 editor-mediated RNA editing (Fig. 4g–i). Strong GFP fluorescence was observed only on the light-exposed side after 24 hours of light stimulation. These results suggest that the padCas13 editor system enables light-induced, spatiotemporal RNA base editing and can be expanded by replacing the regulatory protein.
padCas13 editor enables endogenous RNA base editing
To determine whether our padCas13 editor system could be used to edit endogenous transcripts in mammalian cells, we examined previously identified disease-relevant target transcripts under light conditions, and measured the editing results via analysis of sequencing chromatograms3,7,8,20,22. The padCas13 editing system and targeted crRNA were transfected to HEK 293 T cells, and blue-light illumination was carried out for 24 hours. The results revealed that the padCas13 editors could mediate A-to-I and C-to-U editing of all evaluated endogenous targets with significant editing levels (Fig. 5a, b and Supplementary Tables 5, 6). Furthermore, we compared the editing efficiency of the padCas13 editors and full-length Cas13 (Fig. 5a, b). While the padCas13 system exhibited slightly lower editing efficiency, the RNA editing activity obtained under light conditions remained significantly higher than that seen in the dark group and was within a range similar to that achieved with full-length Cas13.
Next, to demonstrate the ability of the padCas13 editor to edit the phenotypic function of specific targets, we used our system to adjust a post-translational modification (PTM) in the Wnt/β-catenin pathway. More specifically, we performed light-induced base editing of the phosphorylation status of key residues of β-catenin whose phosphorylated forms inhibit protein degradation24. To design our PTM reporter system, we integrated a TOPFlash system based on firefly luciferase genes20,25 into a dual-luciferase assay system that enabled monitoring of signal induction. Under light conditions, the PTM reporter will be induced by padCas13 editor-programmed alteration of phosphorylation among residues of β-catenin (Fig. 5c). We evaluated both padCas13-1 and padCas13-2 editors, as a means to assess how the editing efficiency affected the phenotypic change at the protein level. Similar to our luciferase reporter results, we found that padCas13-2 editor was more effective than padCas13-1 editor. We further confirmed light-inducible RNA editing by observing PTM reporter activation, which reflected Wnt/β-catenin pathway activation downstream of effective targeting of β-catenin transcripts (CTNNB1) (Fig. 5d). These results demonstrate that the use of the padCas13 editor can be expanded to target PTMs of target proteins, such as those relevant to pathological conditions.
In vivo editing using the padCas13 editor
Several groups have recently reported in vivo genome editing with the CRISPR-Cas system26,27,28. This is important because the successful in vivo application of a system critically supports its potential for use in disease models. However, the vast majority of studies using CRISPR-based gene expression systems have been limited to the in vitro level, and no previous study has reported the use of a light-inducible CRISPR-Cas13 system in vivo. The latest developments in red-light inducible systems have expanded their in vivo applications29,30, but such work is still in its early stages. Moreover, little has been done using blue-light activation systems, which are commonly applied in optogenetics.
To determine whether our padCas13 editor could function in an animal model, we delivered padCas13 editor constructs to mice via hydrodynamic tail-vein injection with a luciferase-based RNA editing reporter and targeted crRNA plasmids. We then shaved the abdomen fur from the mice and randomly classified them into dark and light groups. An LED plate was positioned directly under the mouse cage for illumination purposes (Fig. 6a). Due to the anatomical arrangement of the mouse, where the liver is located closer to the skin surface than in a human, blue light can effectively penetrate the thin mouse skin and reach the liver tissues (Supplementary Fig. 11). Indeed, previous research demonstrated that blue light from an LED array can effectively reach mouse liver tissues31. Compared to mice injected with the padCas13-2 editor but not exposed to the light stimulation, the RNA editing-enabled mice exhibited a substantially greater luciferase signal (Fig. 6b–d and Supplementary Fig. 12a). Consistent with these findings, the padCas13 editor system successfully enabled RNA editing in vivo. The previous report of a Cas13-based RNA editing system (named REPAIR3) did not include information on its in vivo use. Here, we tested whether our padCas13 system was capable of RNA editing in vivo and observed the editing efficiency of the padCas13 editor was comparable to that of a full-length Cas13 system (Supplementary Fig. 12b). Moreover, to validate the efficacy of the padCas13 editor, we exposed transfected mice to blue light under various intensities (Dark, 0.25, 0.5, 1, and 5 mW/cm2). Our results indicated that the activity of the padCas13 editor was dependent on the intensity of illumination, with 1 mW/cm2 proving to be sufficient for inducing RNA base editing in vivo (Supplementary Fig. 13a, b). We also observed the intensity-dependent luminescence signal in the livers isolated from these mice, confirming the in vivo applicability of the padCas13 editor under blue light conditions (Supplementary Fig. 13c, d). These findings demonstrate that the padCas13 editor facilitates in vivo RNA base editing when activated by blue light. To further validate the robustness and safety of the padCas13 editor, we evaluated potential liver damage by measuring levels of alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), and total bilirubin. Biochemical analysis revealed that all of the measured parameters were within the clinically normal range, as defined by Charles River guidelines, for both the light-exposed and dark control groups (Supplementary Fig. 14a). Additionally, histological analysis via hematoxylin and eosin (H&E) staining of mouse liver tissues showed no observable differences between the light-exposed and dark control groups (Supplementary Fig. 14b). Our results suggest that the padCas13 editor system can facilitate the robust activation of target transcripts in a mouse model.