In vivo genome editing in animals using AAV-CRISPR system: applications to translational research of human disease

Adeno-associated virus (AAV) has shown promising therapeutic efficacy with a good safety profile in a wide range of animal models and human clinical trials. With the advent of clustered regulatory interspaced short palindromic repeat (CRISPR)-based genome-editing technologies, AAV provides one of the most suitable viral vectors to package, deliver, and express CRISPR components for targeted gene editing. Recent discoveries of smaller Cas9 orthologues have enabled the packaging of Cas9 nuclease and its chimeric guide RNA into a single AAV delivery vehicle for robust in vivo genome editing. Here, we discuss how the combined use of small Cas9 orthologues, tissue-specific minimal promoters, AAV serotypes, and different routes of administration has advanced the development of efficient and precise in vivo genome editing and comprehensively review the various AAV-CRISPR systems that have been effectively used in animals. We then discuss the clinical implications and potential strategies to overcome off-target effects, immunogenicity, and toxicity associated with CRISPR components and AAV delivery vehicles. Finally, we discuss ongoing non-viral-based ex vivo gene therapy clinical trials to underscore the current challenges and future prospects of CRISPR/Cas9 delivery for human therapeutics.

Introduction CRISPR (clustered regulatory interspaced short palindromic repeat)/Cas9-based RNA-guided DNA endonuclease is transforming biomedical science research and has quickly become the preferred genome-editing platform for interrogating endogenous gene function in vivo 1,2 . The CRISPR-based genome-editing tool has revolutionized the gene-editing technique because of its simplicity in target design, affordability, high efficiency, versatility, and multiplexing capability 3 . The commonly used CRISPR system can be implemented in mammalian cells by co-expressing Cas9 nuclease along with chimeric guide RNA (gRNA), which is derived from a synthetic fusion of the CRISPR RNA array (crRNA) and trans-activating crRNA (tracrRNA) 3 . The target site sequence of gRNA needs to be immediately followed by an optimal protospacer adjacent motif (PAM) sequence according to the species-derived Cas9 nuclease used at the 3′ end 4,5 . The background of and the recent developments in the CRISPRbased targeted genome-editing toolboxes have been extensively reviewed recently 1,2,6 . Various applications of CRISPR technologies for genome engineering and medical research have also been reviewed recently 2,6,7 .
The CRISPR/Cas9 complex can be introduced into the cell in the forms of DNA, messenger RNA (mRNA), or protein 8 . The DNA encoding Cas9 and gRNA can be delivered into the cell using the plasmid and viral expression vectors 8 . Through microinjection, liposome-mediated transfection, electroporation, or nucleofection 9 , several recent studies have shown that the delivery format of active Cas9 protein/gRNA ribonucleoprotein (RNP) complex has lower off-target effects due to rapid clearance of RNPs from the cell 8-10 and rapid gene editing as they cleave chromosomal DNA almost immediately after delivery 9,10 as compared with plasmid DNA transfection. However, the delivery formats of mRNA and protein pose certain technical challenges in vivo. For example, many of the genetic brain disorders affect very large brain areas and more than a single structure in the brain 11 . Therefore, global gene delivery to the central nervous system is a key to achieve effective therapies for neurological disorders. Although a recent study has successfully demonstrated genome editing in the mouse brain by local delivery of RNPs 12 , it is technically challenging to deliver the RNPs globally to the central nervous system. Thus, the viral-based in vivo genome editing remains a popular choice to achieve stable or elevated expression of Cas9 and single-guide RNA (sgRNA) required for establishing animal disease models and therapeutic gene editing in animals. Indeed, it has been demonstrated that systemic delivery of adeno-associated virus (AAV) vectors enabled uniform and broad vector distribution, and subsequently led to extensive and widespread transgene expression in the adult mouse central nervous system [13][14][15] .
Animal models are preferred over cell models, as they help understand disease mechanisms at the physiological and systemic levels. Somatic mutagenesis via in vivo genome editing provides an ideal platform to accelerate the generation of transgenic animals for rapid exploration of human diseases and correcting genetic defects in gene therapy. In vivo genome editing also avoids laborious germline targeting and high costs of maintaining transgenic neonates through adulthood in animals with a long life span. Obviously, somatic mutagenesis in adult animals has proven technically more challenging than gene editing of one-cell-stage embryos or zygotes or pluripotent stem cells. The major hurdle is to efficiently deliver the CRISPR components in vivo. To date, various viral, non-viral (for example, lipid nanoparticles), and physical (for example, hydrodynamic injection) based delivery approaches of the CRISPR/Cas9 complex have been adopted for in vivo genome editing. The pros and cons of these delivery methods have been extensively reviewed recently [16][17][18] .
Given the great potential of viral vectors in gene and cell therapy, five major classes of viral vectors-retroviruses 19 , lentiviruses 20,21 , adenoviruses 22,23 , AAVs 24,25 , and baculoviruses 26,27 -have been employed to deliver CRISPR components into mammalian cells for targeted genome editing. The advantages and disadvantages of using these viral vectors for in vivo delivery of the CRISPR transgenes have been extensively reviewed 24,28-30 . In Table 1, we listed general characteristics and applications of various viral delivery vectors. Among these, the AAV vector is the overall focus of this review. The AAV system provides major advantages for research and therapeutics, including a very mild immune response and toxicity elicited by AAV in animal models. Moreover, AAVs remain primarily episomal upon transduction, avoiding random integration of the viral genetic materials into the host genome that can disrupt neighboring gene function and cause insertional mutagenesis 31 . Indeed, there has been no reported case of disease caused by AAV in humans 31 . Additionally, AAV can exist longterm as concatemers in non-dividing cells for stable transgene expressions 31 . Given a good safety profile of AAV and therapeutic efficacy in a wide range of animal models and human clinical trials (ClinicalTrials.gov), AAV is thought to be one of the most suitable viral vectors for gene therapeutic applications and gene transfer in vivo. Furthermore, there is a wide range of AAV serotypes that can be selected to infect specific tissues in vivo. For these reasons, here, we provide an overview of the state of the art of various AAV-CRISPR systems as well as their principal vector designs for in vivo genome editing in animals. Senís and colleagues were amongst the first to exploit and demonstrate the use of the AAV vectors to package, deliver, and express CRISPR components for targeted gene editing in hard-to-transfect cells and the liver of adult mice 32 . Subsequently, an increasing number of the studies used AAV vectors to deliver the CRISPR components into animals for in vivo genome editing. Of note, for certain applications such as packaging of a very large transgene 29,33 , genome-wide screening 20,34,35 , and production of vaccines and recombinant proteins 36,37 , the other viral vectors may be a better option. However, in-depth discussions of the other viral delivery vehicles for the CRISPR components are beyond the scope of this review.
Here, we aim to review various AAV-CRISPR systems recently demonstrated in mice and to discuss how the combined use of tissue-specific minimal promoters, AAV serotypes, different routes of administration, and small Cas9 orthologues has enabled investigators to achieve maximal efficiency and specificity for in vivo genome editing. In addition, we discuss the clinical implications and potential strategies to overcome off-target effects, immunogenicity, and toxicity associated with CRISPR components and AAV delivery vehicles. Finally, we discuss the promises and hurdles associated with ongoing ex vivo gene therapy clinical trials.

AAV-CRISPR-based in vivo genome editing in mice
At least 33 published studies successfully used CRISPR/Cas9 alongside AAV vectors for in vivo genome editing in mice ( Table 2). The AAV-CRISPR system is particularly useful for editing disease-associated genes in the brain or central nervous system of mice because of the inability of most cationic nanocarriers to cross the blood-brain barrier, high transduction efficiency of AAV vectors in the brain, and non-dividing properties of neurons for long-term therapeutic effects. The AAV-CRISPR system has also been successfully used to restore the gene function in muscle-associated diseases such as Duchenne muscular dystrophy 54-57 and diseases associated with the eye 5,58-60 , liver 4,49,61 , heart 62-64 , and lung 65 . This AAV-CRISPR-mediated gene editing in mice provided proof-of-principle studies for human disease modeling, gene therapy, or gene functional characterizations. We discuss recent developments and advancements in four critical aspects of the AAV-CRISPR system that have enabled efficient and precise in vivo genome editing in mice.

Small Cas9 orthologues
Even though the AAV vector is an extremely attractive delivery vehicle for CRISPR, its relatively small viral genome-packaging capacity has limited its use for delivering large transgenes. Using dual-vector AAV system or smaller Cas9 orthologues along with truncated regulatory elements (promoter and polyadenylation signal) has greatly circumvented the transgene packaging issue ( Figure 1A). Owing to a small AAV viral genome-packaging capacity (~4.8 kb), it has been technically challenging to co-package Streptococcus pyogenes-derived Cas9 (SpCas9) (4.1 kb) and multiple sgRNAs into all-in-one AAV vectors for multiplex genome editing. Recently, it has been demonstrated that when very small promoters were used (for example, miCMV promoter and H1 promoter to drive the expressions of SpCas9 and its gRNA, respectively), it was possible to package both SpCas9 and its gRNA into a single vector 32 . However, the number of promoters that can be selected for tissue-specific transduction is limited. To circumvent this issue, the previously reported dual-vector system was adopted to express SpCas9 nuclease from one vector and to express its gRNAs together with a fluorescent reporter gene from another vector 25 . Recently, St1Cas9 (~3.3 kb) from Streptococcus thermophilus 66 and a rationally designed truncated form of SpCas9 67 were developed to fit SpCas9 and its gRNA into a single AAV construct. Unfortunately, St1Cas9 requires a more complex PAM sequence that limits the number of targetable loci 66 , and truncated SpCas9 has a much lower efficiency than its wild-type counterpart 67 .
To overcome these drawbacks, other recently discovered small Cas9 orthologues, including Staphylococcus aureus-derived Cas9 (SaCas9, 3.16 kb) 4 and Campylobacter jejuni-derived Cas9 (CjCas9, 2.95 kb) 5 , have been used to package the Cas9 and its gRNA into a single AAV delivery vehicle for in vivo genome editing. To date, at least 11 independent in vivo studies have used the AAV-SaCas9 system to edit disease-associated genes in a variety of tissues, including brain 68 , muscle 54,55,57,69 , retina 58 , heart 70 , and liver 4,49,71 . More recently, the quadruplex gRNAs/SaCas9 vector consisting of SaCas9 and multiplex sgRNAs was successfully delivered using AAV-DJ/8 for in vivo excision of HIV-1 proviral DNA in various solid tissues/organs via a single intravenous injection in humanized bone marrow/liver/thymus (BLT) mice with chronic HIV-1 infection 72 . In the all-in-one vector system, CjCas9 can be packaged with multiple gRNAs and even a fluorescent reporter gene into a single AAV vector 5 . Even though CjCas9 represents the smallest Cas9 orthologue characterized to date, only one in vivo study has demonstrated the use of AAV vectors to deliver CjCas9 into the mouse's retina for the treatment of age-related macular degeneration because of the recent development of CjCas9 5 . Excitingly, AAV-CjCas9 offers the possibility of editing multiplex genes and tracking the expression of fluorescent reporter genes at high efficiency by simply introducing a single type of AAV vector into the mouse body.
The CRISPR-mediated genome-editing specificities can be further improved by adopting truncated gRNAs 73 and small chemical molecules to regulate Cas9 activity 74 or to enhance doublestranded break-induced homologous recombination efficiency 75 without sacrificing on-target genome-editing efficiencies.
Natural and engineered AAV serotypes Multiple naturally occurring AAV serotypes have been harnessed to deliver the CRISPR/Cas9 complex into different tissues and organs in mice ( Figure 1B). AAV1 is particularly efficient to drive CRISPR transgene expression in the brain 25,65,76 , whereas AAV8 appears well suited for the transduction of liver 4,49,61,71,77 and muscle 55 in mice. AAV9 has the general ability to transduce all major tissues, including muscle 54,56,69 , retina 5 , heart 62-64,78 , and lung 65 , in mice. Recently, the AAV9-CRISPR system has been successfully used to identify functional tumor suppressors in glioblastoma by performing high-throughput mutagenesis in the brain of conditional-Cas9 mice to recapitulate human glioblastoma 79 . AAV2 59 and AAV5 58 have also been used to transduce retina, whereas AAV6 57 enables CRISPR-mediated gene editing in muscle tissue.
In order to alter tropism, reduce blockage by neutralizing antibodies, or enhance transduction efficiency, AAV capsids have been chemically or genetically modified to produce hybrid capsids by combining the properties of multiple serotypes, capsid shuffling, directed evolution, rational mutagenesis, or carrying peptide insertions that introduce the novel receptor-binding activity. AAV variants with engineered capsids (AAV2-retro, AAV2g9, AAV-DJ, and AAV-DJ/8) have been used to package CRISPR for transduction in the mouse brain. An in vivo-directed evolutionengineered AAV variant, rAAV2-retro, permits robust retrograde access to projection neurons in functionally connected and highly distributed networks with a high efficiency for neural circuit dissection and in vivo genome editing 68 . An AAV chimera derived from AAV2 and AAV9, AAV2g9, enables preferential, robust, and widespread neuronal transduction within the mouse brain for CRISPR/Cas9-mediated gene deletion for the treatment of neurological disorders 80 . To provide substantially higher infectivity rates across a broad range of tissues than any known native serotypes,  the AAV-DJ was engineered by DNA family shuffling to create a hybrid capsid from eight different native serotypes 87 . AAV-DJ/8 was created by introducing two point mutations in the heparinbinding domain of AAV-DJ to increase uptake in brain tissue in vivo, leading to a similar ability to infect heart and brain tissues with AAV8 and AAV9 87,88 . In fact, AAV-DJ/8 has been successfully used alongside the CRISPR/Cas9 system with doxycyclinedependent gRNA expression for inducible genome editing of neurons in vivo within the mouse brain 81 . AAV-DJ/8 has also been successfully used to package and deliver the CRISPR/SaCas9 vectors for in vivo excision of HIV-1 proviral DNA in animal models 72 .
Recently, the capsids of AAV9, which is the most efficient native serotype characterized to date for in vivo transduction of the brain, have been further engineered for better performance in transductions and tissue targeting specificity. These include AAV-PHP.B (a chimeric AAV9 variant generated by inserting a 7-mer sequence, TLAVPFK, on AAV9 capsid) 13 , AAV-AS (a chimeric AAV9 variant generated by inserting a poly-alanine peptide on AAV9 capsid) 14 , and AAV9/3 (double tyrosine-mutant form of AAV9) 89 , all of which were shown to be more efficient in transductions of the adult mouse central nervous system after intravenous injection compared with the wild-type AAV9. Furthermore, some of these engineered AAV capsids could de-target the native AAV tropism without affecting vector production, capsid assembly, infectivity, and gene transfer. Therefore, these novel engineered vectors offer promising options to package the CRISPR/Cas9 complex to efficiently edit disease-associated genes in the mouse brain or any other tissues. Additionally, self-complementary AAV9 (scAAV9) can be used to significantly decrease the time required for the onset of CRISPR expression in the central nervous system of mice 84 . Second-strand synthesis is a rate-limiting step for efficient transduction by recombinant AAV vectors 31 . The use of double-stranded self-complementary AAV (scAAV) has improved the transduction efficiencies by bypassing the secondstrand synthesis step, but cloning capacity was further reduced to accommodate the complementary strand 90 .
Tissue-specific minimal promoters A tissue-specific promoter is important to ensure the expression of the CRISPR transgene in a tissue-or organ-specific manner. Various truncated or minimal tissue-specific promoters have been used to drive the expression of CRISPR in specific tissues upon injection of the AAV vectors into the mouse body ( Figure 1C). For example, the 235 base pair (bp) mouse pMecp2 (methyl CpG binding protein 2) and 485 bp human hSyn1 (human synapsin I) promoters have been used to efficiently and specifically drive the expression of CRISPR and other transgenes in the mouse brain 25 and retina 59 . A striated muscle-restricted promoter, CK8, enabled specific expression of SpCas9 and SaCas9 for gene editing in the muscle of the mdx mouse model of Duchenne muscular dystrophy upon intramuscular (tibialis anterior) or systemic (retro-orbital) injection of AAV6 vectors 57 . The liver-specific TBG (thyroxine binding globulin) 4,49 and cardiac-specific Myh6 (myosin heavy polypeptide 6) 63 promoters have been successfully used to drive the expression of CRISPR for in vivo genome editing in mouse liver and heart, respectively. More general promoters such as CMV (cytomegalovirus) 54,58,60,62 and EFS (elongation factor-1 alpha short) 5,65 can be used to express the CRISPR in multiple tissues, including muscle, retina, heart, and lung, for simultaneous gene editing of various tissues in mice. Compared with the CMV promoter, the elongation factor-1 alpha (EF1a) promoter tends to give more consistent and prolonged expression regardless of cell type or physiology 91,92 . EF1a is more stable in long-term culture because it maintains high transgene expression and stability and the copy number of episomal vectors 91 . Thus, despite a stronger gene expression driven by the CMV promoter in many mammalian cell types, EF1a is preferred for efficient transgene expression in mouse embryonic stem cell lines 93 and for engineering stable tumor cell lines 92 . The library of minimal tissue-specific promoters alongside AAV vectors used for controlling the expression of CRISPR or other transgenes in vivo is expanding with the development of more tightly regulated synthetic hybrid promoters, such as widely used CAG, CAGGS, CASI, CBh, GFAP, TRE3G, SMVP, Spc512, H1/T0, CK7-miniCMV, pICAM2, HCRhAATp, and Tight promoters ( Table 2).
Chimeric gRNA expression is frequently driven by an RNA polymerase III promoter, most often by the human U6 small nuclear (U6) promoter. The human U6 promoter has been shown to be more efficient than its murine homolog for short hairpin RNA (shRNA) expression in both human and murine cells 94 . The human U6 promoter was also proven more efficient than the human H1 promoter in driving shRNA expression for the longterm inhibition of gene expression in vitro and in vivo 95 . However, the use of the U6 promoter to overexpress shRNAs in vivo can induce severe hepatotoxicity and inflammatory side effects due to abundant shRNA production 96 . The weaker H1 promoter may offer a safer option in this case, but its therapeutic efficacy is dependent on the sequence of the shRNA 96 . Some promoters require a particular nucleotide as the transcription start site (TSS) to initiate transcription. For instance, the human or mouse U6 promoters require a guanine nucleotide (G) at the 5′ end of the transcript for effective transcription 97 . To use the U6 promoter for driving gRNA expression, the first nucleotide of the transcribed gRNA should be a G to maximize U6 promoter activity 97 . If the G is absent at the TSS of the gRNA sequence, the human RNA polymerase III promoter H1 can be used to drive the expression of the gRNA 76 . The use of H1 promoter-expressed gRNAs could enhance the versatility of the CRISPR technology by expanding the targetable genome 98 . For multiplex CRISPR/Cas9-based genome editing such as large genomic DNA deletion or synergistically enhanced gene activation and repression, four independent polymerase III promoters (human U6 promoter, murine U6 promoter, 7SK, and H1) can be used to express four gRNAs in a single expression cassette 21 . The use of distinct polymerase III promoters can avoid the problem of genetic recombination in a viral vector and the loss of the proximal gRNA due to identical DNA sequences of the promoters 99 . The doxycycline-dependent promoters such as H1/TO and U6/TO can also be used for inducible in vivo genome editing by supplying animals with Dox-containing food to regulate the expression of the gRNA in a Dox-dependent manner 81 . The introduction of Dox inhibits Tet repressor (TetR) binding and induces gRNA expression 81 .

Local and global delivery of AAV vectors
The route of administration of AAV-CRISPR vectors into the mouse body also determines the efficacy and specificity of gene editing for in vivo therapeutics ( Figure 1D). Given the inability of the most native AAV serotypes (except for AAV9 100 , AAVrh8 101 , and AAVrh10 102 ) to cross the blood-brain barrier, stereotactic 25,65,76 and intrathecal 84 injections are still the preferred routes to systemic injection in order to introduce the AAV-CRISPR vectors into specific regions of the mouse brain or central nervous system for therapeutic gene editing. For the stereotactic injection, AAV-CRISPR vectors are intracranially injected into the brain by using a stereotaxic apparatus and the injection micropipette. In the intrathecal injection approach, AAV-CRISPR vectors are injected into the spinal canal or subarachnoid space to reach cerebrospinal fluid. For efficient therapeutic gene editing in the mouse muscle tissue, intramuscular (via tibialis anterior), intraperitoneal (via intraperitoneal space), or intravenous (via the tail vein, temporal vein, and retro-orbital) injections are the preferred methods to introduce AAV-CRISPR vectors, as have been demonstrated for the mdx mouse model of Duchenne muscular dystrophy 55-57 . Subretinal 58,60 or intravitreal 59,85 injections are used to deliver AAV-CRISPR vectors into the mouse retina for gene editing. As most of the AAV vectors will accumulate in the mouse liver tissue upon systemic injection, intravenous injection by tail vein or temporal vein can efficiently edit the genes associated with liver diseases 4,49,71 . Systemic injection remains preferred to local injection in the liver tissue because the delivery is more uniform and the distribution of viral vectors is broader. It is also a good delivery mode for gene editing in multiple tissues, as it is almost non-invasive to the body. On the other hand, intracardiac 64 , subcutaneous 78 , intraperitoneal 63 , or systemic 62,70 injection has been used to introduce AAV9 vectors carrying CRISPR transgene into the mouse heart for gene editing. Intranasal and intratracheal injections are used to locally deliver the AAV-CRISPR vehicles into the mouse lung tissue 65 . Therefore, the selection of an effective route of injections is dependent on the target tissue, AAV serotypes, and tissue-specific promoters used.

CRISPR/Cas9-based human gene and cell therapies
In August 2016, the first human clinical trial using a CRISPR-based gene-editing technique was started for cancer immunotherapy of metastatic non-small cell lung cancer 103 . Subsequently, 10 more early phase clinical trials are underway (ClinicalTrials.gov). One of these clinical trials has proceeded to phase 2 for the treatment of advanced esophageal cancer. Six of these early phase clinical trials are using the CRISPR/Cas9 system to create genetically altered immune cells, which are infused back into patients with an advanced stage of lung, bladder, prostate, renal, gastric, esophageal, or nasopharyngeal carcinoma or lymphoma to target and eradicate cancer cells (Table 3). In general, the programmed cell death protein 1 gene (PDCD1) will be knocked out in autologous T cells. These engineered T cells then will be selected and expanded ex vivo before infusion back into the patients. PDCD1, more commonly known as PD-1, is an inhibitory cell surface receptor involved in the regulation of T-cell function during immune response and tolerance. Knockout of PD-1 in T cells extracted from the patient's blood could prevent cells' immune response from switch-off after re-introduction of the gene-edited cells into the patient's bloodstream and attack and defeat the cancer cells 104 . In fact, antibodies (for example, nivolumab and pembrolizumab) that neutralize PD-1 with high response and low adverse effect rates have been successfully used for cancer immunotherapy in human clinical trials 105,106 . Gene editing is expected to inhibit PD-1 with a greater certainty, and by multiplying the cells in the laboratory, scientists can enhance the efficacy of triggering an immune response against tumors 104 .
These interventional studies aimed to evaluate the clinical response, safety, and maximal tolerant dose of the CRISPRmediated PD-1 knockout T cells in treating advanced-stage malignancies upon being infused back into patients. An anticancer chemotherapy drug, cyclophosphamide, will be administered for a few days to modify the immune micro-environment before cell infusion or to deplete regulatory T cells (Tregs) before collecting peripheral blood. After cell infusion, the interleukin-2 cytokine that has both immune-modulating and anti-tumor properties will be administered to sustain the survival of infused T cells. For cancer immunotherapy of lymphoma or leukemia, an anti-metabolite anti-cancer chemotherapy drug, fludarabine, will be co-administered with the cyclophosphamide before cell infusion. Biomarkers and immunological markers will be closely monitored after cell infusion to determine whether the injections are causing any serious adverse effects and to evaluate whether the patients will receive benefits from the treatment.
In another CRISPR/Cas9-based ex vivo gene therapy clinical trial, the CRISPR/Cas9 complex will be used to disrupt the C-C chemokine receptor type 5 (CCR5) gene in the CD34 + hematopoietic stem or progenitor cells from donors (NCT03164135). Then, these CCR5 modified CD34 + cells from donors will be infused into the HIV-infected patients with AIDS and hematological malignancies. In principle, HIV-1 virus requires CCR5 receptor on the surface of host immune cells (for example, T cells) to infect the cell; hence, disrupting or blocking the CCR5 receptor may make the immune cell resistant to the virus infection. When the CCR5-modified immune cells from the donor are mixed inside the body with the CCR5-positive immune cells from HIV-infected patients, the HIV-1 virus will infect and kill the CCR5-positive immune cells and the CCR5-negative immune cells will survive and eventually take over the immune function. The primary objectives of this clinical trial are to determine the safety of the allotransplantation of these CD34 + cells and to evaluate the resistance to the HIV-1 virus in HIV-infected patients after infusion of these CD34 + cells. Prior to the cell infusion, the patients will be treated with anti-retroviral therapy to achieve undetectable HIV-1 virus in the peripheral blood. During the follow-up, HIV-1 viral load, CD4 + T cells, and CCR5-negative cell counts in the peripheral blood will be monitored in HIV-infected patients.
Another promising ex vivo gene therapy clinical trial is the combined use of CRISPR/Cas9 gene-editing technology and chimeric antigen receptor (CAR)-based cancer immunotherapy (NCT03166878). The CAR-expressed T cells can be engineered to recognize leukemia antigens such as CD19 on B cells for the treatment of relapsed or refractory B-cell malignancies. However, the cost and technical difficulties in the production and expansion of CAR-expressed T cells have hindered the wide application of personalized autologous CAR-T cell therapy. Thus, universal CAR-T cells derived from healthy unrelated donors are developed to overcome some of these drawbacks. To evade hostmediated immunity and deliver anti-leukemic effects without graft-versus-host disease in patients with relapsed or refractory B-cell malignancies, gene-disrupted allogeneic CD19-directed BBζ CAR-T cells (known as UCART019) are developed by combining the lentiviral delivery of CAR and CRISPR mRNA electroporation to disrupt both of the endogenous T-cell receptor (TCR) and beta-2 microglobulin (B2M) genes. The TCR and B2M play important roles in triggering the immune response; hence, disrupting both of these genes could minimize immunogenicity associated with allotransplantation. The primary goal of these allogeneic CD19 CAR-T cells is to evaluate the feasibility, safety, and in vivo persistence of UCART019 adoptively transferred T cells in patients with relapsed or refractory CD19 + leukemia and lymphoma. The amount of UCART019 cells (engraftment), humoral host immunity, and anti-tumor response upon UCART019 cell infusions will be monitored.
Owing to the tissue complexities in humans, toxicity concern, and potential complications such as host immune response following a high dose of AAV vectors used to achieve significant therapeutic efficacy 107 , the AAV-CRISPR viral delivery system is restricted to ex vivo gene editing or genetic manipulation in animals. One of the promising strategies to overcome this issue was to mutagenize the surface-exposed tyrosine residues on the AAV capsid in order to avoid AAV degradation by the host cell proteasome machinery and to improve AAV intracellular trafficking to the nucleus, which can lead to high transduction efficiency at lower vector doses 107 . The combined use of capsidmodified and genome-modified next-generation AAV vectors has allowed higher transduction efficiency and transgene expression at further reduced doses 108 . To date, the AAV-CRISPR geneediting system was tested in non-human animals only, but the first in vivo gene therapy using CRISPR/Cas9 technology will be carried out in human clinical trials soon. In January 2018, an open-label and triple-cohort study will be conducted to evaluate the safety of therapeutic doses and the dosing regimen of CRISPR/ Cas9 plasmid to treat human papillomavirus (HPV) persistency and HPV-related cervical intraepithelial neoplasia. Given the important roles that E6 and E7 play in HPV-driven carcinogenesis, CRISPR/Cas9-mediated disruption of HPV16 and HPV18 E6/E7 DNA could be an attractive approach for therapeutic interventions by significantly downregulating the expression of E6/E7 in order to induce HPV-associated cell apoptosis and to inhibit cell growth.

Promises and hurdles associated with the AAV-CRISPR system for future clinical applications
An increasing number of studies in mice have clearly demonstrated that the combined use of tissue-specific minimal promoters, natural and engineered AAV serotypes, different routes of administration, and small Cas9 orthologues enables efficient packaging and precise delivery of AAV-CRISPR vectors for targeted in vivo genome editing in specific tissues with minimized side effects. Nevertheless, special considerations are required when selecting tissue-specific promoters, natural and engineered AAV serotypes, or routes of administration to avoid non-specific delivery and transgene expression of the AAV. In addition, the off-target effects, toxicity, and immunogenicity associated with CRISPR/Cas9 delivery remain to be fully resolved.

Non-specific delivery and transgene expression
Tissue-specific promoters and AAV serotypes might still be able to transduce and induce transgene expression in healthy tissues or other non-target organs if a strong promoter or a high dose of viral vectors is introduced into the body by intravenous injection. Despite a significant improvement in the efficiency and specificity of newly discovered natural and engineered AAV variants for systemic delivery in mice, the toxicity and side effects associated with the non-specific delivery of the transgenes to the non-target tissues and organs remain a concern 13,89 . It is easier to get delivery vehicles taken up by liver than other organs upon systemic injection. Therefore, it is more challenging to specifically target non-liver organs than liver via a systemic delivery approach. For instance, recently discovered synthetic vectors, AAV-PHP.B 13 and tyrosine-mutant AAV 89 , were found to transduce the adult mouse central nervous system more efficiently and widely than the natural AAV9 after intravenous injection. However, these synthetic vectors can also transduce the liver and other peripheral organs upon systemic delivery 13,89 . In this case, local injections such as stereotaxic injection into the brain can minimize the possible complications or adverse side effects associated with non-specific expression of CRISPR in healthy tissues or other non-target organs.
Off-target effects Even though CRISPR/Cas9 predominantly recognizes the intended target sites, a series of high-throughput genome-wide methods such as multiplex Digenome-seq 109,110 , ChIP-seq 111,112 , GUIDE-seq 113 , and whole-genome sequencing 114 as well as targeted deep sequencing 112 , Cas9 toxicity screens 115 , and SITE-seq biochemical methods 116 have revealed evidence of off-target effects due to target mismatch tolerance of CRISPR/Cas9. As inter-individual natural genetic variation can affect CRISPR/Cas9 specificity 117 , a recently developed CIRCLE-seq approach could be used to identify genome-wide off-target mutations of CRISPR/Cas9 that are associated with cell type-specific single-nucleotide polymorphisms to provide personalized specificity profiles 118 . Notably, given the same target sequence of gRNA, off-target sites of Cas9 before and after being fused to a catalytic enzyme (for example, cytidine deaminase base editor and chromatin modifiers) could be different; therefore, independent assessment of their genome-wide specificities is recommended 119 . Imprecise repair of Cas9-induced DNA double-stranded breaks can give rise to deleterious structural chromosomal rearrangements such as deletions, inversions, and translocations, which in turn may activate oncogenes or cause genome instability 120 . Hence, high-throughput screenings of CRISPR/Cas9 off-target activity and further improvement in the fidelity of CRISPR/Cas9 on-target activity are essential for safety in clinical gene transfer applications.
There are many ways to minimize CRISPR/Cas9 off-target effects in the human genome. For example, a recently developed RNA-targeting Cas9 (RCas9) system could avoid permanent off-target genetic lesions in DNA-mediated CRISPR-based therapeutics 121 . The specificity of the gRNA designed can be improved by selecting a target sequence with two Gs at the 5′ terminus 131 and avoiding potential mismatches at the seed sequence or basepairing adjacent to the PAM 147 . Another innovative way to improve the specificity and reduce the toxicity of the CRISPR/ Cas9 is co-delivery of DNA decoys or competitive inhibitor oligonucleotides bearing all possible off-target sequences that can sequester and prevent the CRISPR/Cas9 from binding to the off-target sites within a host genome. A similar concept has been successfully demonstrated to alleviate the off-target effects of RNA interference by using RNA decoys to reduce the sense strand activity of shRNAs 148 , while artificial microRNA (miRNA) sponges have been used to inhibit miRNA function and its ability to regulate natural mRNAs 149 . In addition, a recently developed miRNA-responsive CRISPR/Cas9 switch could be useful for cell type-specific genome editing by sensing endogenous miRNA activities 150 .

Immunogenicity
Owing to the exogenous nature of AAV and CRISPR components, host immune responses can attenuate therapeutic effects and cause side effects. Thus, the toxicity and immunogenicity associated with AAV and CRISPR components should be circumvented for safer and higher efficacy in clinical gene therapy applications. Even though the AAV generally elicits a very mild immune response and does not induce extensive cellular damage in vivo, certain AAV serotypes such as AAV9 may evoke humoral immune responses, as indicated by the presence of capsid-specific antibodies 50 . While transient immunosuppression is one of the possible ways to mitigate the host immune response following the delivery of AAV vectors 151 , it is not feasible for long-term therapeutic treatments of chronic diseases and is prone to adverse complications such as infections and malignancies. Empty capsid mutants can be used as decoys to overcome pre-existing humoral immunity by adsorbing antibodies in the bloodstream upon systemic delivery of both empty and functional AAVs 152 . Alternatively, the AAV capsids can be genetically engineered or mutated to reduce the binding affinity and the neutralizing effects of AAV antibodies 15,[153][154][155] . In addition to AAVs, the presence of Cas9-specific antibodies in Cas9-exposed animals indicated that Cas9 could evoke deleterious cellular and humoral immune responses in vivo 50 . Expression of Cas9 in vivo could affect the transcription of the genes associated with the immunological processes. This in turn may destabilize the host immune system, elicit significant cellular infiltration or expansion, and induce enlargement of the draining lymph nodes with increased immune cell counts 50 . The Cas9-responsive T-cell clonotype described previously could serve as a distinctive biomarker to assess Cas9-specific immunity before clinical implementation of the CRISPR system 50 . To minimize the immune response due to prolonged expression of Cas9, conditional genome editing with the self-limiting CRISPR/Cas9 system 58 or light-inducible 124,125 or chemical-inducible 126,127 Cas9 can be used to minimize the duration of Cas9 expression in the body.

Conclusions and perspectives
The AAV-CRISPR system holds enormous translational potential to develop curative therapeutic options for patients with severe and life-threatening genetic diseases by permanently editing disease-causing or risk genes in the human body. The delivery, efficacy, and safety issues in treating complex heritable and somatic disorders have to be fully resolved to fulfill this promise. Thus far, the ex vivo approach has been adopted to overcome the technical challenges associated with the in vivo delivery of the AAV-CRISPR viral vectors in humans. The ex vivo gene-editing approach is commonly used for the therapeutic treatment of blood disorders, cancers, and immune-related diseases. In the near future, in vivo genome editing is expected to offer better avenues to treat a wide range of human hereditary diseases in adults. In vivo gene therapy in humans provides several advantages over the ex vivo approach, and sometimes a combination of the two is necessary to achieve a good therapeutic outcome. For example, in vivo photoreceptor cell rescue can be used to halt retinal degeneration to preserve existing vision, while ex vivo photoreceptor cell replacement can be used to restore lost vision in patients with retinal dystrophy 156 . After in vivo photoreceptor cell rescue, the retinal environment may become more permissive for transplanted photo-receptor survival and de novo synaptogenesis via the ex vivo approach 156 .
The AAV vector already has a long history of success in clinical trials in vivo, but owing to the relatively recent arrival of CRISPR technology, the AAV-CRISPR system has yet to be tested in vivo for human gene therapy trials. Nevertheless, three ongoing human clinical trials (NCT03041324, NCT02702115, and NCT02695160) have used AAV vectors to deliver zinc finger nucleases, an earlier and well-established gene-editing tool, to the liver tissue for the treatment of hemophilia B and mucopolysaccharidosis. Similarly, the AAV vectors may be used for CRISPR delivery for human clinical trials in the future. Excitingly, the first in vivo human gene therapy trials using CRISPR/Cas9 plasmid administration to treat human HPVrelated malignant neoplasms will be initiated in January 2018 (NCT03057912).
Although it is more feasible to deliver the transgene ex vivo than in vivo, the in vivo approach eliminates the need for cell transplants. Therefore, in vivo therapy can avoid potential graft-versus-host disease and immunosuppression-related complications such as infections and malignancies, as seen in cellular therapy. Recently, various genetically engineered AAV variants or structure-guided derivations of AAV mutants have been developed to significantly improve the efficiency and specificity for in vivo delivery 157 . Compared with existing AAV serotypes, these newly engineered AAV capsids enabled higher transduction efficiency at a targeted tissue by altering the native tropism of the AAV capsid 13-15,158 and had low immunogenicity by evading preexisting anti-AAV capsid neutralizing antibodies in the human body 159,160 . Given the encouraging results obtained with the next-generation synthetic AAV capsids in animals, this synthetic AAV vector may also be used to deliver CRISPR/Cas9 transgene in humans for in vivo genome editing in the near future. Further improvement in the performance of engineered AAV variants and CRISPR components is necessary to realize the full potential of the AAV-CRISPR system for in vivo genome editing.

Competing interests
No competing interests were disclosed.

Grant information
This work was funded by National Institutes of Health grants AG017242, GM104459, and CA180126 (to Yousin Suh).
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.