CRISPR is derived from the prokaryotic adaptive immune system (Fig. 1) [4-36]. The distinctive clustered repeats were originally recognized in Escherichia coli by Ishino et al. [5] in 1987, and were later found to include unique barcode-like sequences of viral or plasmid origin, termed spacers (Fig. 2) [6-8]. In 2007, the hypothesized role of the repeats in adaptive defense was confirmed by experimental demonstration of spacer integration into the bacterial genome following phage challenge, as well as alteration of sensitivity to subsequent phage infection dependent upon the spacer content [4]. Subsequent studies revealed that CRISPR works in sync with the Cas gene, in the vicinity of the CRISPR locus, to cleave DNA or RNA sequences [9,10] targeted by a small guide RNA [11]. Based on these findings, multiple studies sought to identify the components of the CRISPR-Cas system and apply this knowledge to sequence-specific gene engineering.

There are six putative CRISPR systems; the three main types (types I to III) were discovered first, with three additional types (types IV to VI) being identified more recently [37,38]. During the processes of immunity, adaptation, expression, and interference, each type acts according to distinct mechanisms to ensure DNA recognition and cleavage [39].

Type I uses a large complex of Cas proteins, encoded by the Cas3 gene, which show separate helicase and DNase activities. Similarly, type III uses repeat-associated mysterious proteins, which constitute a large superfamily of Cas proteins. Types I, III, and IV are categorized as class 1 based on their multi-subunit effector complexes. By contrast, class 2 systems (comprising types II, V, and VI) each have a single-subunit effector. Type II uses only a single protein (Cas9) for its nuclease activity; the same is true for types V and VI, but with Cas9-like proteins. Owing to their simplicity, class 2 systems have been adopted for genome engineering [40-42], and only the bacterial type II CRISPR-Cas9 system has been utilized for RNA-guided engineering nucleases [43,44].

As noted above, type II CRISPR-Cas9 systems use a single endonuclease, Cas9. This enzyme acts in concert with two guide RNAs: CRISPR RNA (crRNA) and trans-acting CRISPR RNA (tracrRNA) (Fig. 3) [45]. To simplify the system and improve its utility, scientists employed a linker loop to engineer a dual tracrRNA:crRNA (called single guide RNA [sgRNA]), which participates in sequence-specific DNA cleavage with Cas9 [12]. Short 2 to 5 bp conserved sequences known as proto-spacer adjacent motifs, located on the side opposite to that of the RNA-DNA hybridization, are required for Cas9-DNA recognition [6]. Once recognition occurs, double-stranded DNA cleavage is performed by two Cas9 domains: the HNH domain, which cleaves the strand complementary to the crRNA-guide sequence, and the RuvC-like domain, which cleaves the noncomplementary strand [12]. Via this mechanism, programmed nucleases with customized sgRNA can cleave genomic DNA at specific loci, enabling precise genome editing.

Cas9 protein can be easily re-targeted to new DNA sequences by changing a small portion of the sequence of the accompanying RNA guide, which base-pairs directly with target DNA [46]. Another potential advantage of Cas9 is its ability to introduce several double-strand breaks (DSBs) within the same genome (also referred to as multiplexing) via the expression of multiple guide RNAs [13,14].

Prior to the advent of CRISPR technology, biologists used several generations of tools, all of which employed site-specific DNA DSBs for genome editing. To date, four types of DNA-binding nuclease have been developed: meganucleases, zinc finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs), and the most recently identified nuclease, Cas9.

Each of the previous tools had unique limitations. Meganucleases are like restriction enzymes but are programmed to target DNA sequences 14 to 40 bp in length. Owing to several shortcomings, including their lack of specificity in DNA recognition and the need to fuse the recognition and cleavage domains, meganucleases were used only briefly [47]. ZFNs and TALENs function according to similar principles, but differ in that their binding domains consist of three- and one-nucleotide recognition modules, respectively. These enzymes have separate DNA-binding domains and nonspecific cleavage domains, namely, FokI endonucleases, making them more efficient than meganucleases [48,49].

However, the construction of ZFNs remains a challenge due to the need to account for context-dependent binding preferences [50], notwithstanding previous efforts to circumvent this shortcoming. TALENs, despite having the advantage of one-to-one binding between the Transcription activator-like effector (TALE) unit and each base pair [51], require painstaking molecular biology cloning methods to synthesize highly conserved and repetitive TALE structures [52]. Consequently, the comparatively facile protein engineering of CRISPR makes this approach much more affordable and practical compared with precursor technologies (Table 1).

When DSBs are introduced, the lesion may be corrected by one of two major repair pathways: homology-directed repair (HDR) or nonhomologous end joining (NHEJ). HDR allows the exchange of genetic information between DNA molecules with similar sequences, whereas NHEJ forms short insertions or deletions (indels) in the target sequence. NHEJ does not require a repair template, but the resultant indels can cause frameshift mutations that lead to the production of nonfunctional, incomplete proteins, or to micro-RNA degradation by nonsense-mediated decay. On the other hand, the HDR machinery can repair DNA using exogenous single- or double-stranded DNA templates with sequence similarity to the DSB site. Thus, exploitation of HDR has allowed researchers to insert new genetic information at a target site, or to perform direct replacement of a mutated gene [45,46].

Although this approach was revolutionary, the natural process of HDR is inefficient because it requires selection and screening to identify the one-in-a-million cell in which homologous recombination has exchanged the wild-type gene for the desired modified version. However, CRISPR-Cas9 technology allows the inducible formation of DSBs, so that scientists can modify gene expression at the repair sitex; thus, opening a new avenue in genome editing [42].

The applications of CRISPR-Cas9 have expanded into fields such as agricultural products, livestock, disease modeling, and therapeutics. In this section, we focus on the therapeutic aspects of gene-based diseases, especially monogenic disorders (Fig. 4).

In gene therapy, genes in diseased cells and tissues can be corrected by two approaches: ex vivo and in vivo editing [46]. In ex vivo therapy, the target cell population is removed from the body, modified using a programmable nuclease, and then transplanted back into the original host; thereby, preventing complications due to immunological rejection. By contrast, in vivo editing therapy involves direct transfer of genome-editing reagents, such as a programmable nuclease and donor templates, into the human body [53]. Each approach has advantages and disadvantages, and they are implemented differently to treat particular disorders. There has been examples of gene-editing techniques applied in disease cell lines (Table 2) [54-77] and in disease mouse models (Table 3) [60,63-66,78-88]. Furthermore, scientists have reported series of therapeutic applications with genome editing using stem cell (Table 4) [89-111].

All cancers harbor multiple mutations that cause cells to grow progressively and express malignant phenotypes. These mutations can be categorized into four types: oncogenes, tumor suppressors, epigenetic factors and control loci, and chemoresistance genes. The CRISPR-Cas9 system represents a powerful, highly specific and adaptable tool for correcting such mutations and treating the cancers that contain them [118]. While oncogenic changes occur in many cancers and play important roles in malignant cell proliferation, oncogenes such as the receptor tyrosine kinase Erb2 can be targeted directly by CRISPR-Cas9 [119].

From another perspective, it is possible to utilize CRISPR-Cas9 to introduce cancer-causing mutations in human cell lines and animal models. In this context, the following cell lines have been constructed to date: lung cancer [120], acute myeloid leukemia [121], liver cancer [122], and pancreatic cancer [123].

CRISPR-Cas9 technology can be applied to animal models for the study of both cancers and other inherited diseases. Heritable gene modification can be achieved by injecting CRISPR-Cas9, targeting one or multiple alleles, directly into fertilized zygotes [42]. Among transgenic animal models, mice are the most widely used in experiments because of the relatively short time required to generate mutants; however, non-human primate models have been created successfully by multiplex gene targeting, potentially generating superior systems for the study of complex human diseases, for example, neurodegenerative disorders [124]. Nonetheless, mouse models remain the most cost-effective. Moreover, mice are amenable to large-scale in vivo mutagenesis studies, especially when highly specific targeted editing can avoid the confounding effects of off-target mutagenesis [125].

The applications of the CRISPR-Cas9 system to synthetic biology include all concepts related to synthetic gene circuits in living cells. Because synthetic gene circuits consist of sensors, processors, and actuators, synthetic biology has the potential not only to advance basic research, but also to enable practical applications in medicine, biofuel production, and synthesis of commodity chemicals [126]. The most practical applications of the CRISPR-Cas9 system have been in plants, especially crops such as rice, wheat, sorghum, and tobacco. For example, CRISPR-Cas9 was used to target and knock out the mildew-resistance locus (MLO) genes, which encode proteins that repress the defense against powdery mildew disease in hexaploid bread wheat [127].

CRISPR-Cas9 can also be applied to the systematic analysis of gene functions in human cells. A lentiviral sgRNA library was developed against genes identified by functional screening and high-throughput sequencing analysis. This powerful loss-of-function library screening is expected to facilitate discovery of genes that participate in various biological processes, including drug targeting, toxicity, and expression of certain phenotypes [128].

iPSCs, which are very similar to embryonic stem cells, are pluripotent cells with a high self-renewal rate that can differentiate into almost all cell types; however, their utilization is associated with significantly less ethical controversy than that of their embryonic counterparts. Recent advances in stem cell technology are likely to provide great benefits to the clinical use of iPSCs in clinical applications [129].

As mentioned above, iPSCs have a major advantage for personalized medicine because they can be derived from the patients themselves, and can therefore avoid immune rejection when transplanted. Ex vivo therapy includes correction of patient-derived iPSCs through gene editing, as well as differentiation into nonrenewable cell types such as neurons and cardiomyocytes (Fig. 5) [100].

In addition to this type of precision therapy, human iPSC lines with genotypes characteristic of specific diseases could be used to understand pathogenic mechanisms. Disease modeling and drug efficiency/toxicity testing with iPSCs not only increase the accuracy of disease simulation, but are also less expensive than generating animal models. However, care must be taken when interpreting the results of phenotypic comparisons between patient iPSC-derived cells and healthy control cells. Specifically, the results are vulnerable to confounding variables that might influence the phenotypes of interest, including epigenetic status and unmatched age, gender, and ethnicity. In this respect, gene editing is the only way to distinguish changes that are specifically relevant to a given disease [130].

The CRISPR-Cas9 system enables simultaneous knockout of multiple genes, as well as knock-in of specific alleles in iPSCs, distinguishing it from earlier gene-editing technologies such as ZFNs and TALENs. An isogenic human iPSC cell line precisely corrected by the CRISPR-Cas9 system was recently constructed, despite the handling difficulties associated with gene editing of human stem cells [131]. In the future, the use of CRISPR-Cas9 with iPSCs will lead to novel combinations of gene and cell therapies [132].

Prior to the clinical application of CRISPR-Cas9 in human patients, the safety and efficacy of the system must be validated. The specificity and efficiency of genome-editing tools can be improved by targeting DSB repair pathways, modifying nucleases, and changing the mode of delivery. We briefly discuss each topic below.

As promising as CRISPR sounds, a variety of concerns have been expressed about this technique. In early 2015, a Chinese research group used CRISPR-Cas9 to perform editing on nonviable human trinuclear zygotes, stimulating a vigorous discussion of the ethical implications. The International Summit on Human Gene Editing in Washington was convened in late 2015, providing a forum in which scientists could achieve a consensus on human germline engineering.

The CRISPR-Cas9 toolbox has many advantages and it can be used to correct many defects that occur systemically or from birth, including cystic fibrosis and Huntington’s disease. However, it remains unclear how we should set boundaries regarding which human traits are appropriate for editing. In addition, few would argue that a number of safety and efficacy concerns about CRISPR-Cas9 remain to be resolved. Especially in light of the possibility that undesirable parties could use this technology for eugenics, it would be irresponsible to allow modification of human embryos. In this respect, we must confront the need for robust regulation, even as arguments rage between the advocates of caution and progress.