All-in-one adeno-associated virus delivery and genome editing by Neisseria meningitidis Cas9 in vivo

Background Clustered, regularly interspaced, short palindromic repeats (CRISPR) and CRISPR-associated proteins (Cas) have recently opened a new avenue for gene therapy. Cas9 nuclease guided by a single-guide RNA (sgRNA) has been extensively used for genome editing. Currently, three Cas9 orthologs have been adapted for in vivo genome engineering applications: Streptococcus pyogenes Cas9 (SpyCas9), Staphylococcus aureus Cas9 (SauCas9), and Campylobacter jejuni (CjeCas9). However, additional in vivo editing platforms are needed, in part to enable a greater range of sequences to be accessed via viral vectors, especially those in which Cas9 and sgRNA are combined into a single vector genome. Results Here, we present in vivo editing using Neisseria meningitidis Cas9 (NmeCas9). NmeCas9 is compact, edits with high accuracy, and possesses a distinct protospacer adjacent motif (PAM), making it an excellent candidate for safe gene therapy applications. We find that NmeCas9 can be used to target the Pcsk9 and Hpd genes in mice. Using tail-vein hydrodynamic-based delivery of NmeCas9 plasmid to target the Hpd gene, we successfully reprogram the tyrosine degradation pathway in Hereditary Tyrosinemia Type I mice. More importantly, we deliver NmeCas9 with its sgRNA in a single recombinant adeno-associated vector (rAAV) to target Pcsk9, resulting in lower cholesterol levels in mice. This all-in-one vector yielded > 35% gene modification after two weeks of vector administration, with minimal off-target cleavage in vivo. Conclusions Our findings indicate that NmeCas9 can enable the editing of disease-causing loci in vivo, expanding the targeting scope of RNA-guided nucleases. Electronic supplementary material The online version of this article (10.1186/s13059-018-1515-0) contains supplementary material, which is available to authorized users.


Introduction
A major advance in the field of gene therapy has been the introduction of Cas9 nuclease-enabled genome editing (1). Clustered, regularly interspaced, short palindromic repeats (CRISPR) loci specify an adaptive immune pathway that evolved in bacteria and archaea to defend against mobile genetic elements (MGEs) (2,3). The effector complex in type II CRISPR systems includes the Cas9 nuclease, which is guided by a CRISPR RNA (crRNA) and a trans-activating RNA (tracrRNA). These dual RNAs can be fused to form a single-guide RNA (sgRNA) (4). Each crRNA contains a unique "spacer" sequence that can be programmed to cleave a DNA segment of interest. Cas9 scans DNA for a specific Protospacer Adjacent Motif (PAM), opens the duplex to form an RNA-DNA hybrid between the guide and the spacer, and introduces a double-strand break (DSB) in the DNA target (1,3). Cas9 and sgRNA have been adapted to enable genome editing in cultured cells following various modes of delivery including plasmid and RNA transfections, viral transduction, and ribonucleoprotein (RNP) electroporation. Precise and efficient in vivo editing is more difficult to achieve, largely due to the difficulties inherent in delivery.
Several methods have been developed to deliver Cas9 in vivo including viral and non-viral methods (5). These include the use of gold and lipid nanoparticles to deliver Cas9 in RNP or RNA form in mice. However, these methods present challenges for routine use including cost and tissue distribution (6)(7)(8). One of the more intriguing gene delivery vehicles that has emerged in recent years is recombinant adeno-associated virus (rAAV). This vector possesses several attributes that benefit gene therapy applications, including lack of pathogenicity and replication as well as an ability to infect dividing and non-dividing cells (9). In addition, rAAV is also capable of infecting a wide range of cells and maintain sustained expression (10,11). Compared to other viral vectors, rAAV persists in concatemeric, episomal forms, while eliciting mild immune responses (12)(13)(14). The usefulness of rAAV-based delivery for gene therapy is reflected in the number of clinical trials involving rAAV (15). One of the most exciting advancements for rAAV gene therapy field has been the FDA's recent market approval of a therapy for RPE65-mediated inherited retinal disease (IRD), the first of its kind in the United States (16). 5 More recently, several groups have focused their efforts on using this tool for in vivo delivery of Cas9 orthologs (17)(18)(19)(20). The majority of Cas9 genome editing efforts have been focused on the widely-used type II-A ortholog from Streptococcus pyogenes, SpyCas9. Although it exhibits consistently robust genome-editing activity, considerable effort has been required to overcome off-target editing activities of wild-type SpyCas9 (21-23) (Amrani et al., manuscript submitted).
Dual-rAAV delivery of SpyCas9 and sgRNA can be achieved (28), but it requires the usage of highly minimized promoters that limit expression and tissue specificity. Furthermore, dual rAAV formats carry significant costs as well as limitations in co-transduction.
Furthermore, off-target editing by SauCas9 is not unusual (18,30). For these reasons, many genomic sites of interest cannot be targeted by all-in-one rAAV delivery of the Cas9 genome editing machinery, and additional capabilities and PAM specificities are therefore needed.
In this study, we report the in vivo delivery of NmeCas9 and its guide by a single expression cassette that is sufficiently small for all-in-one rAAV vectors. Two disease genes were targeted separately to highlight the therapeutic potential of NmeCas9: the Hpd gene in a hereditary 6 tyrosinemia type I (HTI) mouse model (Fah neo ), and the Pcsk9 gene in C57Bl/6 mice. Hpd encodes the 4-hydroxyphenylpyruvate dioxygenase enzyme in the tyrosine metabolism pathway, and disrupting Hpd can lead to a decrease in the accumulation of toxic fumarylacetoacetate in tyrosinemia models (35). Separately, Pcsk9 encodes proprotein convertase subtilisin/kexin type 9 (PCSK9), an antagonist of the low-density lipoprotein (LDL) receptor (36,37). When PCSK9 is knocked out, more LDL receptors are available at the surface of hepatocytes to allow cholesterol binding and recycling towards the lysosomes for degradation (38,39). The alleviation of tyrosinemia symptoms upon Hpd disruption, as well as the reduced serum cholesterol levels that result from Pcsk9 disruption, provide convenient readouts for genome editing activity (18,35). We used these systems to validate all-in-one rAAV delivery of NmeCas9 as an effective in vivo genome editing platform in mammals. 7

Construction of All-in-One AAV-sgRNA-hNMeCas9 Plasmid and rAAV vector production
The human-codon-optimized NmeCas9 gene under the control of the U1a promoter, and a single-guide RNA cassette driven by the U6 promoter, were cloned into an AAV2 plasmid backbone. The NmeCas9 ORF was flanked by four nuclear localization signals -two on each terminus -in addition to a triple-HA epitope tag. Oligonucleotides with spacer sequences targeting Hpd, Pcsk9, and Rosa26 were inserted into the sgRNA cassette by ligation into a SapI cloning site (Supplementary Note).
AAV vector production was performed at the Horae Gene Therapy Center at the University of Massachusetts Medical School. Briefly, plasmids were packaged in AAV8 capsid by tripleplasmid transfection in HEK 293 cells and purified by sedimentation as previously described (40).
The off-target profiles of these spacers were predicted computationally using the Bioconductor package CRISPRseek. Search parameters were adapted to NmeCas9 settings as described Each well was transfected with 500 ng all-in-one AAV-sgRNA-hNmeCas9 plasmid, using Lipofectamine LTX with Plus Reagent (Invitrogen) according to the manufacturer's protocol.

DNA isolation from cells and liver tissue
Genomic DNA isolation from cells was performed 72 hours post-transfection cells using DNeasy Blood and Tissue kit (Qiagen) following the manufacturer's protocol. Mice were sacrificed and liver tissues were collected 10 days post-hydrodynamic injection or 14 and 50 days post-tail vein 8 rAAV injection. Genomic DNA was isolated using DNeasy Blood and Tissue kit (Qiagen) according to the manufacturer's protocol.

GUIDE-seq
GUIDE-seq analysis was performed as previously described (22). Briefly, 7.5 pmol of annealed GUIDE-seq oligonucleotides and 500 ng of all-in-one AAV-sgRNA-hNmeCas9 plasmids targeting Pcsk9, Rosa26 and Hpd were transfected into 1x10 5 Hepa 1-6 cells using Lipofectamine LTX with Plus Reagent (Invitrogen). At 72 hours post-transfection, genomic DNA was extracted using a DNeasy Blood and Tissue kit (Qiagen) per manufacturer protocol. Library preparations, deep sequencing, and reads analysis were performed as previously described (41,42). The Bioconductor package GUIDEseq was used for off-target analysis as described previously using maximum allowed mismatch of 10 nt between the guide and target DNA (41). For read alignment, mouse mm10 was used as a reference genome.

Indel analysis
TIDE primers were designed ~700 bp apart, with the forward primer at ~200 bp upstream of the cleavage site. 50 ng of genomic DNA was used for PCR amplification with High Fidelity 2X PCR Master Mix (New England Biolabs). For TIDE analysis, 30 μl of a PCR product was purified using QIAquick PCR Purification Kit (Qiagen) and sent for Sanger sequencing using the TIDE forward primer (Supplementary Table). Indel values were obtained using the TIDE web tool (https://tide-calculator.nki.nl/) as described previously (43).
Targeted deep-sequencing analysis was performed for Hepa 1-6 cells and mouse liver gDNA using a two-step PCR amplification approach as described previously (42) Table). In the second-step PCR, equimolar amounts of DNA were amplified with a universal forward primer and an indexed reverse primer using Phusion High Fidelity DNA Polymerase (98 °C, 15s; 61°C, 25s; 72 °C, 18s; 9 cycles) to ligate the TruSeq adaptors. The resultant amplicons were separated in a 2.5% agarose gel and the corresponding ~250 bp product bands were extracted using Monarch DNA Gel Extraction Kit (New England Biolabs). 9 The libraries were then sequenced on an Illumina MiSeq in paired-end mode with a read length of 150 bp. To analyze genome editing outcomes at genomic sites, the command line utilities of CRISPResso were used (44). Input parameters were adjusted to filter low-quality reads (-q 30 -s 20). Furthermore, the background was determined using the control sample (no guide) and subtracted from the experimental samples. The resulting indel frequencies, sizes and distributions were then plotted using Graphpad PRISM. Both sexes were used in these experiments. Mice were maintained on NTBC water for 7 days postinjection and then switched to normal water. Body weight was monitored every 1-3 days. The PBS-injected control mice were sacrificed when they became moribund after losing 20% of their body weight after removal from NTBC treatment.

Animals and liver tissue processing
Mice were euthanized according to our protocol and liver tissue was sliced and fragments stored at -80°C. Some liver tissues were fixed in 4% formalin overnight, embedded in paraffin, sectioned and stained with hematoxylin and eosin (H&E).

Serum analysis
Blood (~200 μL) was drawn from the facial vein at 0, 25 and 50 days post vector administration.
Serum was isolated using a serum separator (BD, Cat. No. 365967) and stored under -80 °C until assay.
Serum cholesterol levels were measured by Infinity™ colorimetric endpoint assay (Thermo-Scientific) following the manufacturer's protocol. Briefly, serial dilutions of Data-Cal™ Chemistry Calibrator were prepared in PBS. In a 96-well plate, 2 μL of mice sera or calibrator dilution was mixed with 200 μL of Infinity™ cholesterol liquid reagent, then incubated at 37 °C for 5 minutes. The absorbance was measured at 500 nm using a BioTek Synergy HT microplate reader.

Western blot
Liver tissue fractions were ground and resuspended in 150 μL of RIPA lysis buffer. Total

Humoral Immune Response
Humoral IgG1 immune response to NmeCas9 was measured by ELISA (Bethyl; Mouse IgG1 ELISA Kit, E99-105) following manufacturer's protocol with a few modifications. Briefly, expression and three-step purification of NmeCas9 and SpyCas9 was performed as previously described (4). 0.5 μg of recombinant NmeCas9 or SpyCas9 proteins suspended in 1x coating buffer (Bethyl) were used to coat 96-well plates (Corning), and incubated for 12 hours at 4 °C 11 with shaking. The wells were washed 3 times while shaking for 5 minutes using 1x Wash Buffer.
Plates were blocked with 1x BSA Blocking Solution (Bethyl) for 2 hours at room temperature, then washed three times. Serum samples were diluted 1:40 using PBS and added to each well in duplicate. After incubating the samples at 4 °C for 5 hours, the plates were washed 3x times for 5 minutes and 100 μL of biotinylated anti-mouse IgG1 antibody (Bethyl; 1: 100,000 in 1 x BSA Blocking Solution) was added to each well. After incubating for 1 hour at room temperature, the plates were washed 4 times, and 100 μL of TMB Substrate was added to each well. The plates were allowed to develop in the dark for 20 minutes at room temperate, and 100 μL of ELISA Stop Solution was then added per well. Following the development of the yellow solution, absorbance was recorded at 450 nm using a BioTek Synergy HT microplate reader.

Cells and in vivo by Hydrodynamic Injection
Recently, we have shown that the relatively compact NmeCas9 is active in genome editing in a range of cell types (Amrani et al., manuscript submitted). To exploit the small size of this Cas9 ortholog, we generated an all-in-one AAV construct with human-codon-optimized NmeCas9 under the expression of the mouse U1a promoter, and with its sgRNA driven by the U6 promoter ( Figure 1A; Supplementary Note).
Two sites in the mouse genome were selected initially to test the nuclease activity of NmeCas9 in vivo: the Rosa26 "safe-harbor" gene (targeted by sgRosa26), and the proprotein convertase subtilisin/kexin type 9 (Pcsk9) gene (targeted by sgPcsk9), a common therapeutic target for lowering circulating cholesterol and reducing the risk of cardiovascular disease (Figure 1B).
Genome-wide off-target predictions for these guides were determined computationally using the Bioconductor package CRISPRseek 1.9.1 (46) with N4GN3 PAMs and up to 6 mismatches.
Many N4GN3 PAMS are inactive, so these search parameters are nearly certain to cast a wider net than the true off-target profile. Despite the expansive nature of the search, our analyses revealed no off-target sites with fewer than four mismatches in the mouse genome (Supplementary Figure 1). On-target editing efficiencies at these target sites were evaluated in mouse Hepa 1-6 hepatoma cells by plasmid transfections and indel quantification was performed by sequence trace decomposition using the TIDE web tool. We found >25% indel values for the selected guides, the majority of which were deletions ( Figure 1C).
To evaluate the preliminary efficacy of the constructed all-in-one AAV-sgRNA-hNmeCas9 vector, endotoxin-free sgPcsk9 plasmid was hydrodynamically administered into the C57Bl/6 mice via tail-vein injection. This method can deliver plasmid DNA to ~40% of hepatocytes for transient expression (47). Indel analyses by TIDE using DNA extracted from liver tissues revealed 5-9% indels 10 days after vector administration (Figure 1D), comparable to the editing efficiencies obtained with analogous tests of SpyCas9 (48). These results suggested that NmeCas9 is capable of editing liver cells in vivo. 13

Phenotypes of Hereditary Tyrosinemia Type I Mice
Hereditary Tyrosinemia type I (HT-I) is a fatal genetic disease caused by autosomal recessive mutations in the Fah gene, which codes for the fumarylacetoacetate hydroxylase (FAH) enzyme.
Patients with diminished FAH have a disrupted tyrosine catabolic pathway, leading to the accumulation of toxic fumarylacetoacetate and succinyl acetoacetate, causing liver and kidney damage (49). Over the past two decades, the disease has been controlled by 2-  Three groups of mice were treated by hydrodynamic injection with either PBS, or with one of the two sgHpd1 and sgHpd2 all-in-one AAV-sgRNA-hNmeCas9 plasmids. One mouse in the sgHpd1 group and two in the sgHpd2 group were excluded from the follow-up study due to failed tail-vein injections. Mice were taken off NTBC-containing water 7 days after injections, and their weight was monitored for 43 days post-injection ( Figure 2B). Mice injected with PBS suffered severe weight loss (a hallmark of HT-I), and were sacrificed after losing 20% of their body weight 14 ( Figure 2C). Overall, all sgHpd1 and sgHpd2 mice successfully maintained their body weight for 43 days overall, and for at least 21 days without NTBC (Figure 2C). NTBC treatment had to be resumed for 2-3 days for two mice that received sgHpd1 and one that received sgHpd2 to allow them to regain body weight during the 3 rd week after plasmid injection, perhaps due to low initial editing efficiencies, liver injury due to hydrodynamic injection, or both. Conversely, all other sgHpd1 and sgHpd2 treated mice achieved indels with frequencies ranging from 35% to 60% ( Figure 2D). This level of gene inactivation likely reflects not only the initial editing events but also the competitive expansion of edited cell lineages (after NTBC withdrawal) at the expense of their unedited counterparts (51,52,54). Liver histology revealed that liver damage is substantially less severe in the sgHpd1 and sgHpd2 treated mice compared to Fah mut/mut mice injected with PBS, as indicated by the smaller numbers of multinucleated hepatocytes compared to PBS-injected mice (Supplementary Figure 3).

In vivo Genome Editing by NmeCas9 Delivered by a rAAV Vector
Although plasmid hydrodynamic injections can generate indels, therapeutic development will require less invasive delivery strategies, such as rAAV. To this end, all-in-one AAV-sgRNA-hNmeCas9 plasmids were packaged in hepatocyte-tropic AAV8 capsids to target Pcsk9 (sgPcsk9) and Rosa26 (sgRosa26) (Figure 1B) (55,56). Vectors were administered into C57BL/6 mice via tail vein (Figure 3A). We monitored cholesterol level in the serum, and measured PCSK9 protein and indel frequencies in the liver tissues 25 and 50 days post injection.
Using a colorimetric endpoint assay, we determined that the circulating serum cholesterol level in the sgPcsk9 mice decreased significantly (p < 0.001) compared to the PBS and sgRosa26 mice at 25 and 50 days post-injection ( Figure 3B). Targeted deep-sequencing analyses at Pcsk9 and Rosa26 target sites revealed very efficient indels of 35% and 55% respectively at 50 days post-vector administration ( Figure 3C). Additionally, one mouse of each group was euthanized at 14 days post-injection, and revealed on-target indel efficiencies of 37% and 46% at Pcsk9 and Rosa26, respectively ( Figure 3C). As expected, PCSK9 protein levels in the livers of sgPcsk9 mice were substantially reduced compared to the mice injected with PBS and sgRosa26 (Figure 3D). The efficient editing, PCSK9 reduction, and diminished serum cholesterol indicate the successful delivery and activity of NmeCas9 at the Pcsk9 locus. 15 SpyCas9 delivered by viral vectors is known to elicit host immune responses (19,57). To investigate if the mice injected with AAV8-sgRNA-hNmeCas9 generate anti-NmeCas9 antibodies, we used sera from the treated animals to perform IgG1 ELISA. Our results show that NmeCas9 elicits a humoral response in these animals (Supplementary Figure 4). Despite the presence of an immune response, NmeCas9 delivered by rAAV is highly functional in vivo, with no apparent signs of abnormalities or liver damage (Supplementary Figure 5).

NmeCas9 is Highly Specific in vivo
A significant concern in therapeutic CRISPR/Cas9 genome editing is the possibility of off-target edits. We and others have found that wildtype NmeCas9 is a naturally high-accuracy genome editing platform in cultured mammalian cells (32)  Several of the putative OT sites for sgPcsk9 and sgRosa26 lack the NmeCas9 PAM preferences (N4GATT, N4GCTT, N4GTTT, N4GACT, N4GATA, N4GTCT, and N4GACA) ( Figure 4B) and may therefore represent background. To validate these OT sites, we performed targeted deep-sequencing using genomic DNA from Hepa 1-6 cells. By this more sensitive readout, indels were undetectable above background at all these OT sites except OT1 of Pcsk9, which had an indel frequency less than 2% (Figure 4B). To validate NmeCas9's high fidelity in vivo, we measured indel formation at these OT sites in liver genomic DNA from the AAV8-NmeCas9treated, sgPcsk9and sgRosa26-targeted mice. We found little or no detectable off-target editing in mice liver sacrificed at 14 days at all sites except sgPcsk9 OT1, which exhibited fewer than 2% lesion efficiency (Figure 4C). More importantly, this level of OT editing stayed below <2% even 16 after 50 days, and also remained either undetectable or very low for all other candidate OT sites.
These results suggested that extended (50 days) expression of NmeCas9 in vivo does not compromise its targeting fidelity ( Figure 4C).

All-in-One rAAV delivery of hNmeCas9
Compared to transcription activator-like effector nucleases (TALENs) and Zinc-finger nucleases (ZFNs), Cas9s are distinguished by their flexibility and versatility (1). Such characteristics make them ideal for driving the field of genome engineering forward. Over the past few years, CRISPR-Cas9 has been used to enhance products in agriculture, food and industry, in addition to the promising applications in gene therapy and personalized medicine (58). Despite the diversity of Class 2 CRISPR systems that have been described, only a handful of them have been developed and validated for genome editing in vivo. In this study, we have shown that NmeCas9 is a compact, high-fidelity Cas9 that can be considered for future in vivo genome editing applications using all-in-one rAAV. Its unique PAM enables editing at additional targets that are inaccessible to the other two compact, all-in-one rAAV-validated orthologs (SauCas9 and CjeCas9).

Pathway Reprograming
Patients with mutations in the HPD gene are considered to have Type III Tyrosinemia and exhibit high level of tyrosine in blood, but otherwise appear to be largely asymptomatic (59,60).

Efficient, Accurate NmeCas9 Genome Editing with rAAV Delivery
To achieve targeted delivery of NmeCas9 to various tissues in vivo, rAAV vectors are a promising delivery platform due to the compact size of NmeCas9 transgene, which allows the delivery of NmeCas9 and its guide in all-in-one format. We have validated this approach for the targeting of Pcsk9 and Rosa26 genes in adult mice, with efficient editing observed even at 14 days postinjection. As observed previously in cultured cells (32) (Amrani et al., manuscript submitted), NmeCas9 is intrinsically accurate, even without the extensive engineering that was required to reduce off-targeting by SpyCas9 (21)(22)(23). We performed side-by-side comparisons of NmeCas9 OT editing in cultured cells and in vivo by targeted deep-sequencing, and we found that offtargeting is minimal in both settings. Editing at the sgPcsk9 OT1 site (within an unannotated locus) was the highest detectable at ~2%. Despite these promising results, more extensive and long-term studies, including in larger animals, will be needed to fully understand the long-term effects of Cas9 expression in tissues, as well as the development of approaches that clear viral vectors after editing is complete.
In conclusion, we demonstrate that NmeCas9 is amenable to in vivo genome editing using the highly desirable all-in-one rAAV platform. With its unique PAM preferences and high fidelity, this all-in-one AAV-sgRNA-hNeCas9 can be applied to a range of genome editing purposes in vivo.