Cell-specific CRISPR/Cas9 activation by microRNA-dependent expression of anti-CRISPR proteins

The rapid development of CRISPR/Cas technologies brought a personalized and targeted treatment of genetic disorders into closer reach. To render CRISPR-based therapies precise and safe, strategies to confine the activity of Cas(9) to selected cells and tissues are highly desired. Here, we developed a cell type-specific Cas-ON switch based on miRNA-regulated expression of anti-CRISPR (Acr) proteins. We inserted target sites for miR-122 or miR-1, which are abundant specifically in liver and muscle cells, respectively, into the 3’UTR of Acr transgenes. Co-expressing these with Cas9 and sgRNAs resulted in Acr knockdown and correspondingly in Cas9 activation solely in hepatocytes or myocytes, while Cas9 was efficiently inhibited in off-target cells. We demonstrate control of genome editing and gene activation using a miR-dependent AcrIIA4 in combination with different Streptococcus pyogenes (Spy)Cas9 variants (full-length Cas9, split-Cas9, dCas9-VP64). Finally, to showcase its modularity, we adapted our Cas-ON system to the smaller and more target-specific Neisseria meningitidis (Nme)Cas9 orthologue and its cognate inhibitors AcrIIC1 and AcrIIC3. Our Cas-ON switch should facilitate cell-specific activation of any CRISPR/Cas orthologue, for which a potent anti-CRISPR protein is known.

With respect to in vivo application of CRISPR technologies, strategies to confine CRISPR/Cas9 activity to selected cells and tissues are highly desired. For genetic studies in animals, for instance, confining perturbations to selected cells is critical when aiming at disentangling the role of selected cell types in a particular phenotype or simply to avoid negative side-effects and/or artefacts that would arise from unspecific perturbations. Moreover, in the context of therapeutic genome editing within human patients, ensuring maximum specificity and hence safety of a treatment is absolutely critical.
Until today, however, virtually any mode of efficient in vivo delivery of the CRISPR/Cas components (e.g. via viral vectors, nanoparticles, lipophilic complexes etc.) is likely to affect many cell types and tissues beyond the one of actual (therapeutic) interest. This limited specificity, in turn, causes substantial risks of (treatment) side-effects (14,15).
One strategy to address this limitation would be to render the activity of the CRISPR components dependent on endogenous, cell-specific signals, so that the genetic perturbation is induced solely in the target cell population, but not in off-target cells. One such signal are mi(cro)RNAs, i.e. small, regulatory and non-coding RNAs that are involved in eukaryotic gene expression control (16,17).
Being part of the RNA-induced silencing complex (RISC), miRNAs recognize sequence motifs present on m(essenger)RNAs that are complementary to the miRNA sequence. The RISC then typically mediates mRNA degradation, or translation inhibition or both, thereby causing a gene expression knockdown (16,17).
More than 1000 miRNAs have been described in humans (http://www.mirbase.org), and many miRNAs or miRNA combinations have been identified, which occur exclusively in selected cell types or disease states (18)(19)(20)(21)(22)(23). These include, for instance, miR-122, which is selectively expressed in hepatocytes (18), or miR-1, which is highly abundant in myocytes (22,23). Such unique signatures have in the past been successfully harnessed for cell-specific expression of transgenes in cultured cells and mice (24,25). Adapting this strategy to CRISPR/Cas would thus offer an effective means to confine CRISPR-mediated perturbations to selected cell types.
We have previously shown that integrating miRNA-122 binding sites into the 3'UTR (3' untranslated region) of a CRISPR/Cas9 transgene can be used to de-target Cas9 expression from hepatocytes (26). A subsequent study by Hirohide Saito's group expanded this approach to further miRNA candidates (miR-21 and miR-302a) (27). Moreover, they added a negative feedback loop to the system, thereby establishing a positive relation between miRNA abundance and Cas9 activity (27). To this end, the authors expressed Cas9 from an mRNA harbouring an L7Ae binding motif (K-turn), while co-expressing the L7Ae repressor from an mRNA carrying miRNA binding sites in its 3'UTR (27). The resulting Cas-ON switch enabled miRNA-dependent Cas9 activation. The system was leaky, however, and showed a less than 2-fold dynamic range of regulation, thereby limiting its utility for in vivo applications.
Here, we created a novel, robust and highly flexible cell type-specific Cas9-ON switch based on anti-CRISPR proteins (28-32) expressed from miRNA-dependent vectors. We placed AcrIIA4, a recently discovered Streptococcus pyogenes (Spy)Cas9 inhibitor, under miR-122 and miR-1 control, thereby enabling hepatocyte-or myocyte-specific activation of various (Spy)Cas9 variants (full-length Cas9, split-Cas9, dCas9-VP64) with an up to ~100-fold dynamic range of regulation. Finally, to demonstrate its modularity, we expanded our Cas-ON approach to the smaller and more target-specific Neisseria meningitidis (Nme)Cas9 and its cognate inhibitors AcrIIC1 and AcrIIC3.

Cloning
A list of all constructs used and created in this study is shown in Supplementary Table 1. Annotated vector sequences are provided as Supplementary Data (GenBank files). Plasmids were created using classical restriction enzyme cloning, Golden Gate Assembly (33) or Gibson assembly (New England Biolabs). Oligonucleotides were obtained from Integrated DNA Technologies (IDT) or Sigma-Aldrich.
Luciferase knockdown reporters carrying miRNA binding sites within the 3'UTR of the Renilla luciferase gene (pSiCheck-2 no miR target site/ 2xmiR-122 target sites) were generated by inserting a DNA fragment encoding two miRNA target sites followed by a bovine growth hormone (BGH) polyA signal into the psiCheck2 vector (Promega) via XhoI/NotI. The CMV promoter-driven SpyCas9 expression vector (Addgene plasmid # 113033) was previously developed by us (34). A SpyCas9-GFP fusion was cloned by PCR-amplifying EGFP from vector pSpCas9(BB)-2A-GFP (Addgene plasmid# 48138, which was a kind gift from Feng Zhang) followed by insertion of the PCR amplicon into the SpyCas9 vector via EcoRI/HindIII. The AcrIIA4 coding sequence and the mCherry-AcrIIA4 coding sequence were obtained as human codon-optimized, synthetic DNA fragments from IDT and cloned into pcDNA3.1 (-) (ThermoFisher) via NheI/NotI. 2xmiR-122 target sites or a scaffold sequence identical in length but lacking the miR target sites were inserted into the resulting vectors by oligo cloning via EcoRI/HindIII, yielding vectors CMV-(mCherry)-AcrIIA4-2xmiR-122 and CMV-(mCherry)-AcrIIA4-scaffold.
The luciferase cleavage reporter for measuring SpyCas9 activity was previously reported by us (34). It comprises an SV40 promoter-driven Renilla luciferase gene, a TK promotor-driven Firefly luciferase gene, and an H1 promoter-driven sgRNA targeting the Firefly luciferase gene. The pRL-TK vector encoding Renilla luciferase was obtained from Promega. AAV vectors encoding (i) SpyCas9 (Addgene #113034) or (ii) an H1 or U6 promoter-driven sgRNA (F+E scaffold (35)) and a RSV promoter-driven EGFP (Addgene #113039) were previously reported by us (36). Annealed oligonucleotides corresponding to the genomic target site were cloned into the sgRNA AAV vector via BbsI using Golden Gate cloning (33). All sgRNA target sites relevant to this study are shown in Supplementary

Cell culture
Cells lines were cultured at 5 % CO 2 and 37 °C in a humidified incubator and passaged when reaching 70 to 90 % confluency. HeLa and HEK293T cells were maintained in 1x DMEM without phenol red (ThermoFisher) supplemented with 10 % (v/v) fetal calf serum (Biochrom AG), 2 mM Lglutamine and 100 U per ml penicillin/100 µg per ml streptomycin (both ThermoFisher). Huh-7 medium was additionally supplemented with 1 mM non-essential amino acids (ThermoFisher). HeLa, HEK293T and Huh-7 cells were authenticated and tested for mycoplasma contamination prior to use

Luciferase assays
For luciferase experiments, HeLa and Huh-7 cells were seeded at a density of 6,000 cells per well, HEK293T cells were seeded at a density of 12,500 cells per well, and HL-1 cells were seeded at a density of 12,000 cells per well into 96-well plates (Eppendorf) using 100 µl culture medium per well. Renilla photon counts (as Cas9 is targeting the Firefly luciferase gene in these assays).

AAV lysate production
For production of AAV-containing cell lysates, low-passage HEK293T cells were seeded into 6-well

Design of miRNA-dependent anti-CRISPR vectors
To generate a cell-specific Cas9-ON switch, we aimed at rendering the activity of CRISPR/Cas9 dependent on the presence of cell-specific miRNAs, i.e. miRNAs that are abundant solely within the target cell type. Translating the abundance of a miRNA, which typically is a negative stimulus (causing gene expression knockdown), into a positive output (CRISPR activation) requires a negative feedback.
We hypothesized that anti-CRISPR proteins, a recently discovered class of phage-derived CRISPR/Cas inhibitors (28,30,38,39), would be an ideal mediator to establish this negative feedback.
Due to their small size (~80-150 amino acids), Acrs can be expressed quickly and efficiently from plasmids or viral vectors. More importantly, anti-CRISPR proteins block CRISPR/Cas9 DNA targeting, Cas9 nuclease function or both by directly binding to the Cas9/sgRNA complex. This posttranslational inhibitory mechanism enables a complete shutdown of CRISPR/Cas9 activity even upon simultaneous delivery of Cas9, a sgRNA, and an anti-CRISPR-encoding vector (29)(30)(31)40). Coupling the expression of anti-CRISPR proteins to selected miRNAs could therefore be a suitable approach for an efficient, cell type-specific Cas9-ON switch ( Figure 1A).
To validate this concept, we created a modular vector encoding a CMV promoter-driven AcrIIA4, an anti-CRISPR protein efficiently inhibiting the most widely employed Cas9 orthologue from S.
pyogenes (30). BsmBI (Esp3l) sites present in the 3'UTR of the AcrIIA4 gene enable the introduction of miRNA target sites via Golden Gate cloning, so that AcrIIA4 expression can be set under control of any abundant, cell-specific miRNA (or set of miRNAs, see discussion).
A prominent example is miR-122, which is highly expressed in the liver, but not in any other tissue (18). Using a luciferase reporter knockdown assay, we could confirm a strong miR-122 expression in human hepatocellular carcinoma cells (Huh-7), while human cervix carcinoma (HeLa) or embryonic kidney (HEK293T) control cells showed comparably low miR-122 levels (Supplementary Figure 3).
To place AcrIIA4 under miR-122 regulation, we inserted a concatemer of two miR-122 target sites into our modular AcrIIA4 construct. We further added an N-terminal mCherry to enable fluorescencebased detection of AcrIIA4 expression ( Figure 1B). Then, we co-transfected the resulting vector (mCherry-AcrIIA4-2xmiR-122) or a control vector lacking the miR-122 target site (mCherry-AcrIIA4scaffold) alongside a SpyCas9-GFP vector into Huh-7 and HeLa cells. Live-cell fluorescence microscopy and complementary Western blot analysis revealed an efficient knockdown of mCherry-AcrIIA4-2xmiR-122 expression in Huh-7, but not in HeLa cells, thereby indicating a successful coupling of miRNA-122 abundance to AcrIIA4 expression ( Figure 1C, D). As expected, SpyCas9-GFP expression was not affected by the AcrIIA4 knockdown ( Figure 1C, D).

Hepatocyte-and myocyte-specific activation of SpyCas9
To investigate whether the observed, miR-122-dependent knockdown of AcrIIA4 would be sufficient to release CRISPR/Cas9 activity specifically in hepatocytes, we performed a luciferase reporter cleavage assay. We co-transfected Huh-7 or HeLa cells with vectors encoding SpyCas9, AcrIIA4-  Figure 4).
Next, we tested whether our Cas9-ON strategy would also enable cell type-specific editing of endogenous genomic loci. To deliver the different components of our system efficiently, we chose to employ Adeno-associated virus (AAV) vectors. AAVs are highly efficient, safe (AAVs are nonpathogenic in humans), and-very importantly-one can re-target AAVs to specific cell types or tissues by modifying the viral capsid (41)(42)(43). These properties render AAV a prime vector candidate for therapeutic gene delivery.
We packaged (i) SpyCas9, (ii) sgRNAs targeting the human EMX1, CCR5 or AAVS1 locus as well as (iii) AcrIIA4-2xmiR-122 or AcrIIA4-scaffold into AAV2, a well-studied AAV serotype known for its ability to efficiently transduce various cell lines (42). We then co-transduced Huh-7 or HeLa cells with combinations of these vectors, while varying the AcrIIA4 vector dose, and measured the frequency of insertions and deletions at the target loci using a T7 endonuclease assay and TIDE sequencing (37).
We observed potent, miRNA-122-dependent gene editing at all three target loci in Huh-7 cells with a dynamic range of regulation of up to 16-fold ( Figure 2B-D, Supplementary Figure 5). The editing efficiency and leakiness of the system in the OFF-state depended on the used AcrIIA4 vector dose ( Figure 2B-D). Importantly, in the HeLa control cells, Cas9 activity was equally suppressed in the presence of the AcrIIA4-2xmiR-122 or AcrIIA4-scaffold vector ( Figure 2B), indicating that our miRNA-122-dependent Cas9-ON switch is indeed hepatocyte-specific.
The large size of CRISPR/SpyCas9 (~ 1,300 amino acids) poses a challenge with respect to its efficient delivery and expression, in particular when using vectors with a constrained packaging capacity. To circumvent this problem, several groups have recently developed split-SpyCas9 variants, which comprise an N-and C-terminal Cas9 fragment that reconstitute a functional Cas9 when coexpressed within the same cell (44)(45)(46)(47). To test whether our anti-CRISPR-based Cas9-ON strategy would also work for split-SpyCas9s, we employed an intein-based split-SpyCas9 system recently reported by the lab of George Church (48). We co-transfected plasmids encoding the N-and Cterminal SpyCas9 fragments alongside an AcrIIA4 vector (with or without miR-122 sites in the 3'UTR) and the aforementioned luciferase cleavage reporter into Huh-7 cells (or HeLa cells as control). In Huh-7 cells, the split-SpyCas9 remained fully active in the presence of the AcrIIA4-2xmiR-122 vector as indicated by potent luciferase knockdown, but was completely impaired if we co-administered the AcrIIA4-scaffold construct (Supplementary Figure 6). In HeLa cells, in contrast, split-SpyCas9 activity was impaired upon co-delivery of both, the AcrIIA4-2xmiR-122 or the AcrIIA4-scaffold vector (Supplementary Figure 6), demonstrating that our Cas9-ON switch can also be applied to control split-

SpyCas9.
MiR-1 plays an important role in muscle cell differentiation (22,23) and remains highly expressed in mature muscle cells (49). Using a luciferase knockdown assay, we confirmed high miR-1 levels in HL-1 cells, a widely employed murine cardiac muscle cell model (Supplementary Figure 7). We hypothesized that, similarly to miR-122 in hepatocytes, miR-1 could be harnessed for myocytespecific activation of CRISPR/Cas9 using the identical, Acr-based strategy. We therefore exchanged the two miR-122 binding sites in our AcrIIA4 constructs by two miR-1 target sites ( Figure 2E). Then, we packaged the resulting AcrIIA4-2xmiR-1 construct or the AcrIIA4-scaffold construct (as control), a SpyCas9 transgene, and a sgRNA targeting the murine Rosa-26 locus into AAV serotype 6, which is known to efficiently transduce a wide spectrum of tissues in vitro and vivo, including myocytes (50)(51)(52).
Upon co-infection of HL-1 cells with these vectors, we observed potent editing of the Rosa-26 locus in the presence of the AcrIIA4-2xmiR-1 vector, but not when using the AcrIIA4-scaffold control vector ( Figure 2F,G and Supplementary Figure 8), thereby demonstrating successful miR-1-dependent release of SpyCas9 activity in myocytes.

miRNA control of dCas9-effector fusions
So far, we have demonstrated the power of our Cas9-ON system for cell type-specific genome editing.
However, the CRISPR/Cas9 system offers many applications that go beyond a targeted introduction of double-strand breaks. These are typically based on catalytically inactive d(ead)Cas9 mutants employed as programmable DNA binding domain to recruit effector domains to selected genomic loci.
To test this hypothesis, we co-transfected vectors encoding a SpydCas9-VP64 transcriptional activator, a Tet operator (TetO) targeting sgRNA, a luciferase reporter driven from a TetO-dependent promoter, and an AcrIIA4-2xmiR-122 or AcrIIA4-scaffold vector into Huh-7 cells ( Figure 3A). We observed a potent, miR-122-dependent release of luciferase reporter expression with a dynamic range of regulation of up to 114-fold ( Figure 3B). Remarkably, the leakiness and dynamic range of the system could be tuned over a wide range by varying the AcrIIA4 vector dose ( Figure 3B). These results illustrate that our SpyCas9-ON system can also be applied for cell type-specific activation of SpydCas9-effector fusions.

Hepatocyte-specific activation of NmeCas9
Although the SpyCas9 remains the most widely employed CRISPR/Cas orthologue, the ongoing discovery and characterization of novel CRISPR/Cas effectors from various species rapidly expands the CRISPR toolbox. For many of these novel type I and II CRISPR/Cas effectors, corresponding anti-CRISPR proteins have already been found or are likely to be discovered in the near future (28)(29)(30)38,40,60,61), suggesting that our Cas9-ON approach might be easily transferable to many other CRISPR/Cas orthologues. One such orthologue is the Neisseria meningitidis (Nme)Cas9, which is not only ~ 300 amino acids smaller than SpyCas9, but also shows a far lower activity on off-target loci, presumably due to its extended protospacer sequence (62)(63)(64). Two anti-CRISPR proteins have recently been described, which efficiently inhibit NmeCas9 via distinct mechanisms. AcrIIC1 perturbs the NmeCas9 nuclease function, while AcrIIC3 induces NmeCas9 dimerization, thereby impairing its DNA binding (29,32).
We speculated that, similar to SpyCas9 control via miR-dependent AcrIIA4, placing AcrIIC1 and AcrIIC3 under miRNA regulation would enable cell-specific activation of NmeCas9. To test this hypothesis, we codon-optimized the AcrIIC1 and AcrIIC3 genes, introduced miR-122 target sites into their 3'UTRs and packaged them into AAV2. Then, we co-transduced Huh-7 or HEK293T control cells with the AcrIIC1-2xmiR-122 or AcrIIC3-2xmiR-122 vector (or AcrIIC1-scaffold or AcrIIC3-scaffold as control) alongside a vector co-encoding NmeCas9 and a sgRNA targeting the human VEGFA locus In this study, we employed cellular miRNA signatures to control the expression of anti-CRISPR proteins, thereby creating a synthetic circuit efficiently confining Cas9 activity to selected target cells.
Liver and muscle are interesting target tissues for CRISPR-mediated gene therapy approaches, e.g.
for treatment of hemophilia or Duchenne muscular dystrophy, respectively (66)(67)(68)(69)(70) In any case, the Acr vector dose is a crucial parameter to consider. We showed that too high a dose of Acr vector may cause significant suppression of CRISPR/Cas9 activity, even within the target cell population, presumably due to a limited capacity of the RNAi machinery. In contrast, too low a dose of Acr vector leads to insufficient Cas9 inhibition in off-target cells. Importantly, by modulating the dose of the supplied Acr vector or the strength of the Acr-driving promoter, one can easily tune the system toward the optimal switching behaviour for a particular application.
Another important feature of our Cas9-ON switch is its modularity, i.e. it should be compatible with any Cas9 orthologue, for which a potent anti-CRISPR protein is known (as exemplified here for SpyCas9 and NmeCas9). In light of the ongoing, rapid discovery and characterization of novel Acrs, the application spectrum of our switch is likely to further expand in the near future. Importantly, our Cas-ON strategy is also applicable to dCas9-effector fusions, provided an Acr is employed, which impairs Cas9 DNA binding. This is the case, e.g. for AcrIIA4 and AcrIIC3, which block DNA targeting of SpyCas9 and NmeCas9, respectively, but not for AcrIIC1, which impairs the NmeCas9 nuclease function, but does not interfere with its DNA binding. Therefore, the underlying, inhibitory mechanism can be an important parameter to consider when selecting Acrs to be used in our Cas-ON system. Importantly, our Cas-ON is compatible with many existing strategies for tissue-specific gene delivery and expression, such as engineered or evolved AAV vectors (41,42,74) or tissue-specific promoters (75,76). Thus, combining these approaches, potentially with additional layers of CRISPR/Cas control via chemical triggers (44,77,78) or light (34,46,79,80), will likely enable highly specific genome perturbations.
While we foresee that the most relevant applications of our approach will be in animal models and, in the long run, potentially in human patients, we reckon that a careful investigation of toxicity or immune reactions that might result from Acr overexpression should precede in vivo translation of our Cas9-ON strategy. Moreover, to avoid continuous sequestration of the endogenous miRNA pool within the target cells, it could be advisable to couple our Cas9-ON strategy to vector self-inactivation (81).
Taken together, our work demonstrates the power of miRNA-dependent anti-CRISPR transgenes to confine CRISPR/Cas9 activity to selected cells types and facilitate safe and precise genome perturbations in animals and patients.

SUPPLEMENTARY DATA
Supplementary Figures 1-9 and Supplementary Tables 1-3 are provided as Supplementary Information.