In Situ Gene Therapy via AAV-CRISPR-Cas9-Mediated Targeted Gene Regulation

In Vitro and In Vivo Genome Editing via a Dual-AAV Split-Cas9 System

We first confirmed that split-Cas9 constructs delivered as AAVs were functional in vitro and in vivo. Expression cassettes of split-Streptococcus pyogenes (SpCas9) and gRNA were delivered via a dual-AAV vector system (Figure S1A). The first AAV contains a gRNA driven by a human RNA polymerase III promoter, U6, and a N-terminal Cas9 (NCas9) fused to an N-intein driven by a cytomegalovirus (CMV) promoter, as well as a polyadenylation (polyA) signal.³⁰ The second AAV cassette contains a CMV-driven C-terminal Cas9 (CCas9) fused to a C-intein as well as a polyA signal. We confirmed targeted genome editing through next-generation sequencing (NGS)⁴⁰ across two distinct cell types in vitro (Figure S1C), notably also observing robust AAV6-mediated editing in human CD34+ hematopoietic stem cells. We additionally confirmed in vivo genome editing in adult C57BL/6J mice injected with 5E+11 vg/split-Cas9/mouse through the tail vein (Figure S1C).

As a hit and run approach suffices for genome editing and is preferable over long-term nuclease expression, we next engineered small-molecule inducibility32, 41, 42 of in vivo CRISPR-Cas9 editing activity. Specifically, we engineered one AAV construct to bear a minimal CMV promoter with a tetracycline response element (TRE) up-stream of the C-intein-CCas9 fusion, whereas the other bore a full-length CMV promoter that drives expression of the N-intein-NCas9 fusion and a tet-regulatable activator (tetA). The binding of tetA to the TRE, upon doxycycline addition, allows for inducible expression of the CCas9 and thereby temporal regulation of genome editing (Figure S1B). We confirmed robust functioning of this system both in vitro and in vivo (Figures S1D and S1E). Taken together, these studies confirmed the functionality of the dual-AAV split-Cas9 format for CRISPR-Cas9 delivery.

In Vitro and In Vivo Genome Regulation via a Dual-AAV Split-dCas9 System

Next, to engineer targeted genome repression and activation, we utilized a dead split-Cas9 (dCas9) protein and its fusion to repression and activation domains (specifically a KRAB and a VP64+rTA [VR] domain, respectively; Figures 1A and 1B).12, 13, 14, 15, 16 Utilizing the dual-AAV strategy enabled us to package these additional effector domains without exceeding the viral packaging capacity. We confirmed functionality via in vitro experiments in HEK293T cells targeting CXCR4 for repression and RHOX1 for activation utilizing the AAV-DJ serotype, with control non-targeting virus (equal viral titers). For in vivo AAV delivery, we performed tail vein injections at titers of 5E+11 vg/split-dCas9/mouse using the AAV8 serotype and harvested mice livers 4 weeks post-injection. We achieved 80% in vivo repression at the Cd81 locus (n = 4; p < 0.0001) and over 2-fold in vivo activation at the Afp locus (n = 4; p = 0.0117). Taken together, we confirmed targeted gene repression and activation, as assayed via qRT-PCR, in both in vitro and in vivo scenarios and across multiple genomic loci (Figures 1C–1F).

To see whether we could further improve targeted genome regulation, we screened additional repression and activation domains by taking advantage of the modular vector designs of our dual-AAV-CRISPR platform (Figure 2A). Specifically, for our activation system, we evaluated coupling of additional VP64 or P65 domains onto the N-terminal of the dCas9 vector (dNCas9). The additional domains indeed yielded enhanced activation of a target locus (ASCL1) in HEK293T cells, with a ∼17-fold higher activation with VP64 (n = 3; p = 0.0387) and ∼23-fold higher activation with P65 (n = 3; p = 0.0069; Figure 2B), implying that tethering of a VP64 or P65 domain on the N-terminal in addition to the existing VP64-RTA on the C-terminal led to improved gene activation. We further confirmed this improved architecture in vivo in mice, observing over 6-fold activation at the Afp locus (n = 4; p = 0.0271; Figure 2B). In addition, to evaluate the in vivo kinetics of CRISPR-based gene regulation, we performed a time course ELISA on mice injected with 5E+11 vg/split-dCas9/mouse of AAV8 VP64-dCas9-VR-Afp, which were bled weekly for 10 weeks. Control mice received equal titers of a non-targeting AAV8 virus. We observed that Afp activation peaks at week 6, with a 19 ng/mL concentration of Afp in the blood, from a baseline of 3.8 ng/mL (calculated based on an Afp protein standard curve; Figure 2C).

Next, we focused on optimizing targeted gene repression. Whereas dCas9 alone can cause repression, as it can halt RNA polymerase progression by steric hindrance when targeted downstream of the transcription start site (TSS), or can competitively interfere with transcription factor binding when targeted to promoter regions or regulatory elements,12, 44, 45 KRAB-dCas9 has been shown to be more potent for gene silencing than dCas9 alone.45, 46, 47 To determine whether we could further improve the repression system, we evaluated fusions of additional KRAB, DNA methyltransferase domains (DNMT3A or DNMT3L),48, 49 or FOG1⁵⁰ onto the C-terminal of the dCas9 vector (dCCas9) and also the use of single versus dual-gRNAs. To avoid repeat sequences in the AAV that can compromise vector stability and viral titers, we utilized a human U6 promoter and a mouse U6 promoter to drive each individual gRNA and also used non-homologous trans-activating small RNA (tracrRNA) sequences.⁵¹ However, we did not observe an increase in repression with the addition of repression domains, implying that a single KRAB domain suffices for our transient repression assays in HEK293T cells, but the use of dual-gRNAs consistently yielded enhanced gene repression (Figures 2D and S2).

In Situ Cellular Reprogramming of Rod Photoreceptors

Having established a robust in vivo genome regulation system, we next focused on applying the same in a therapeutically relevant mouse model of disease, specifically focusing on retinitis pigmentosa. For these studies, we utilized an AAV2 mutant containing a tyrosine to phenylalanine substitution (Y444F) due to its high retinal transduction efficiency.⁵² To further boost gene targeting, we utilized a dual-gRNA approach per above. We designed corresponding Cas9-based editing (split-Cas9-Nrl) and dCas9-based repression system (split-KRAB-dCas9-Nrl), where the KRAB repressor domain is fused to the N-terminal of the dCas9 protein (Figures 3A–3C). We first delivered the split-Cas9-Nrl vectors into mouse embryonic fibroblasts (MEFs) and assessed gene editing rates through a T7E1 assay, which cuts mismatched double-stranded DNA (dsDNA), and confirmed editing rates of about ∼25% (Figure S3A).

We next used qRT-PCR to measure the relative expression levels of photoreceptor-specific genes in reprogrammed retinas and controls. We confirmed downregulation of Nrl, with simultaneous upregulation of cone-specific genes, i.e., Arr3, Opnlmw, PDE6C, and GNAT2 (Figure S3B). To assay rod reprogramming into cone-like cells, we transduced transgenic Nrl-EGFP mice bearing GFP-labeled rod photoreceptor cells, with 2E+10 vg/split-Cas9/mouse of AAV-split-Cas9 or 2E+10 vg/split-dCas9/mouse of AAV-split-KRAB-dCas9 systems targeting Nrl into the subretinal space at postnatal day 7 (P7). These were then sacrificed for histology at P30 via flash freezing, sectioning, and staining of retinas for a cone marker, cone arrestin (mCAR). We indeed observed a reprogrammed photoreceptor phenotype with both our split-Cas9-Nrl and split-KRAB-dCas9-Nrl in vivo, with a decrease in Nrl-GFP+ rod photoreceptors (Figure S3C) and an increase in the number of mCAR-positive cells (Figure S3D). We next repeated the above experiments in normal wild-type C57BL6 mice. Consistent with the results in Nrl-EGFP transgenic mice, we again observed a reprogrammed photoreceptor phenotype in the retina, characterized by a significant increase in the number of mCAR- and mOpsin-positive cells (Figures S3E and S3F). Taken together, the above experiments confirmed the efficacy of our dual-AAV-CRISPR platform in engineering in situ cellular reprogramming of retinal rod cells.

Prevention of Photoreceptor Degeneration in a Retinitis Pigmentosa Mouse Model

Next, to validate our approach in a retinitis pigmentosa mouse model, we targeted Nrl in rd10 mice, a model of autosomal recessive retinitis pigmentosa. These mice carry a spontaneous mutation of the rod-phosphodiesterase gene and exhibit rod degeneration around P18. By P60, rods are no longer detectable, with accompanying cone photoreceptor degeneration.⁵³ To assess whether conversion of rods to cones is sufficient to reverse degeneration and rescue visual function, we subretinally injected split-Cas9-Nrl or split-KRAB-dCas9-Nrl in rd10 mice at P7. Whereas untreated eyes had sparsely distributed photoreceptor cell nuclei in the outer nuclear layer (ONL), split-Cas9-Nrl- or split-KRAB-dCas9-Nrl-treated eyes had ∼2 or 3 layers of ONL, indicating that the treatment prevented photoreceptor degeneration and preserved ONL (Figures 3D–3H). To determine the effect of the treatment on cone physiological function and visual acuity, we also measured optic kinetic nystagmus (OKN) to quantify visual acuity 6 weeks after injection (P50). All eyes treated with split-KRAB-dCas9-Nrl had improved visual function, as indicated by significantly higher visual acuity (Figure 3H). Taken together, our split-KRAB-dCas9 AAV system thus paves the way for fine control of in situ gene expression for gene therapy of retinitis pigmentosa and importantly also enables a scarless approach for in vivo genome engineering.

Vector Design and Construction

See Supplemental Notes for annotated sequence information on the modules used for the AAV vector constructions. Split-Cas9/dCas9 AAV vectors were constructed by sequential assembly of corresponding gene blocks (Integrated DNA Technologies) into a custom synthesized rAAV2 vector backbone. gRNA sequences were inserted into NCas9 or dNCas9 plasmids by cloning oligonucleotides (IDT) encoding spacers into AgeI cloning sites via Gibson assembly.

gRNA Designs

Editing gRNAs were designed utilizing the in silico tool: MIT CRISPR Design and Broad Institute CRISPRko (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). Regulation gRNAs were designed utilizing an in silico tool to predict gRNAs.⁵⁷

Mammalian Cell Culture

All HEK293T cells were grown in DMEM (10%) supplemented with 10% fetal bovine serum (FBS) and 1% Antibiotic-Antimycotic (Thermo Fisher Scientific) in an incubator at 37°C and 5% CO₂ atmosphere. HEK293T cells were plated in 24-well plates for AAV transductions. Hematopoietic stem cells expressing CD34 (CD34+ cells) were grown in serum-free StemSpan SFEM II with StemSpan CD34+ Expansion Supplement (10×; all from STEMCELL Technologies). CD34+ cells were plated in 96-well plates for AAV transductions.

Production of AAVs

AAV8 was utilized for in vivo studies in the liver, AAV2-Y444F was used for in situ studies in the eye, AAV6 was utilized for in vitro studies in CD34+ cells, and AAV-DJ was utilized for in vitro studies in HEK293T cells.

Lipid-Mediated Cell Transfections

One day prior to transfection, HEK293T cells were seeded in a 24-well plate at a cell density of 1 or 2E+5 cells per well. 0.5 μg of each plasmid was added to 25 μL of Opti-MEM medium, followed by addition of 25 μL of Opti-MEM containing 2 μL of Lipofectamine 2000. The mixture was incubated at room temperature for 15 min and then added to the cells. The entire solution was added to the cells in a 24-well plate and mixed by gently swirling the plate. Media was changed after 24 hr, and the plate was incubated at 37°C for 72 hr in a 5% CO₂ incubator. Cells were harvested, spun down, and frozen at −80°C.

T7E1 Assay

To examine the efficacy of the Nrl gRNAs, we performed T7E1 assay in immortalized mouse fibroblasts. Briefly, cells were transfected with pAAV-U6-gRNA and hCas9 (Addgene 41815) using Lipofectamine 2000 (Thermo Fisher Scientific). Two days after transfection, the cells were harvested and genomic DNA was extracted with DNeasy Blood & Tissue kit (QIAGEN), and a T7E1 (New England Biolabs) assay was done following manufacturer’s instructions. Primers to amplify genomic regions are listed as following: NRL-F: 5′-ACCTCTCTCTGCTCAGTCCC-3′ and NRL-R: 5′-GACATGCTGGGCTCCTGTC-3′. The cleavage frequency was calculated from the proportion of cut bands intensity to total bands intensity.

Animal Experiments

Gene Expression Analysis and qRT-PCR

RNA from cells was extracted using RNeasy kit (QIAGEN; 74104), from liver using RNeasy Plus Universal Kit (QIAGEN; 73442), and from eyeballs using AllPrep DNA/RNA Mini Kit (QIAGEN; 80204). cDNA was synthesized from RNA using Protoscript II Reverse Transcriptase Kit (NEB; E6560L). Real-time PCR (qPCR) reactions were performed using the KAPA SYBR Fast qPCR Kit (Kapa Biosystems; KK4601), with gene-specific primers (Table S2A) in technical triplicates and in biological triplicates. Relative mRNA expression was calculated using the comparative CT (ΔΔCT) method and normalized to β-actin or GAPDH. Mean fold change and SD were calculated using Microsoft Excel.

Genomic DNA Extraction and NGS Preps

Genomic DNA from cells and tissues was extracted using DNeasy Blood and Tissue Kit (QIAGEN; 69504), according to the manufacturer’s protocol. Next-generation sequencing libraries were prepared as follows. Briefly, 4–10 μg of input whole-liver gDNA was amplified by PCR with primers that amplify 150 bp surrounding the sites of interest (Table S2B) using KAPA Hifi HotStart PCR Mix (Kapa Biosystems; KK2602). PCR products were gel purified (QIAGEN; 28704) and further purified (QIAGEN PCR Purification Kit; 28104) to eliminate byproducts. Library construction was done with NEBNext Multiplex Oligos for Illumina kit (NEB; E7335S). 10–25 ng of input DNA was amplified with indexing primers. Samples were then purified and quantified using a qPCR library quantification kit (Kapa Biosystems; KK4824). Samples were then pooled and loaded on an Illumina Miseq (150 bp paired-end run or 150 single-end run) at 4 nM concentrations. Data analysis was performed using CRISPR Genome Analyzer.⁴⁰

ELISA

Levels of serum Afp were measured using the alpha-fetoprotein (Afp) mouse ELISA kit (Abcam; ab210905) according to manufacturer’s guidelines. First, 50 μL of 2 μg/mL capture antibody was added to each well of a 96-well of a high bind microplate (ab210903). The plates were sealed and incubated overnight at 4°C on a plate rocker. The plates were manually washed three times with 350 μL of 1× wash buffer (ab206977). To reduce non-specific binding, the plates were blocked by adding 300 μL of 1× blocking buffer (ab210904) to each well. The plates were then sealed and incubated at room temperature for 2 hr. Plates were washed as described above. The Afp protein standards were diluted in 1× blocking buffer (ab210904) and prepared for a two-fold diluted standard curve. Samples were diluted 1:20 in 1× blocking buffer (ab210904), and 50 μL of sample and standard (in duplicates) were added onto the plates and allowed to bind for 2 hr. Plates were washed as described above. Then, 50 μL of 0.5 μg/mL of detector antibody was added to each well and incubated for 1 hr at room temperature. The plates were washed as described above. Horseradish peroxidase (HRP)-streptavidin solution (ab20901) was added to each well at a 1:7,500 dilution and incubated at room temperature for 1 hr. Plates were washed as described above. Then, 100 μL of 3,3’,5,5’-tetramethylbenzidine (TMB) substrate was added to each well and incubated until optimal blue density was obtained. Finally, 100 μL of stop solution was added to each well. The absorbance was immediately determined on a microplate reader (BioRad iMark) at a wavelength of 450 nm.