A CRISPR Approach to Gene Targeting

It is getting easier and easier to determine complete genome sequences—of model organisms, animals and plants of commercial importance, and humans: Craig Venter, Jim Watson, the 1000 Genome Project, soon you and me. Now that researchers have all this information at hand, the focus has shifted in many cases to manipulating particular sequences to determine their function or to alter their impact. A new study by Jinek et al.¹ proposes a new approach—based on the oldest of DNA recognition principles—to the design of reagents that can target specific genomic sequences.

Precision genome engineering has been enhanced substantially in recent years by the development of targetable DNA cleavage reagents.² A double-strand break (DSB) made at a specific genomic location by, for example, zinc-finger nucleases (ZFNs) will often be repaired inaccurately by nonhomologous end joining (NHEJ), creating a targeted mutation (Figure 1). When a modified donor DNA is also provided, repair by homologous recombination will lead to introduction of donor sequences at the target. These break-induced modifications can be very efficient, in the range of 10% or more of all targets in a single treatment. Gene-editing nucleases such as ZFNs, not only have been used for engineering precise genomic changes in experimental organisms but are being tested in current clinical trials.³

Figure 1.

Consequences of targeted genomic cleavage. A double-strand break made by any type of cleavage reagent can be repaired by error-prone nonhomologous end joining (NHEJ), leaving small insertions and/or deletions at the site. An alternative mode of repair is homologous recombination (HR), which can use a manipulated donor DNA as a template, resulting in replacement of genomic sequences. The break can be made by any targetable nuclease: zinc-finger nucleases (ZFNs), transcription activator–like effector nucleases (TALENs), homing endonucleases (HEs), or, perhaps, the new CRISPR reagents.

ZFNs are hybrid proteins that have several favorable properties as targeting reagents.² The zinc-finger modules that comprise their DNA-binding domain can be assembled in many combinations to recognize a wide range of genomic sequences (Figure 2). The FokI-derived cleavage domain is not active as a monomer, so the nuclease is assembled only when two ZFNs bind at the designed target. The binding and cleavage domains can be manipulated separately to alter recognition and cleavage properties independently.

Figure 2.

ZFNs and TALENs. (Top) Each zinc finger (small ovals) in a zinc-finger nuclease (ZFN) binds primarily to three consecutive base pairs; a minimum of three fingers is required to provide sufficient affinity. Different colors indicate fingers recognizing different DNA triplets. Each set of fingers is joined to a FokI-derived cleavage domain (large ovals) by a short linker. (Bottom) In transcription activator–like effector nucleases (TALENs), each module (small ovals) binds a single base pair; the four colors indicate modules for each of the four base pairs. The minimum effective number of modules is 10–12, but more are typically used. The linker to the FokI domain (large ovals) is longer than for ZFNs and contains additional TALE-derived sequences.

A problem with ZFNs has been the unpredictability of their recognition capabilities. Some fingers apparently bind their corresponding DNA triplet (or quartet) reliably in different contexts, but others do not. Even procedures that select finger combinations explicitly for new targets are not always successful, and they can be dauntingly laborious.⁴

This design challenge has recently been addressed with the adoption of an alternative set of DNA-binding modules derived from Xanthomonas, a genus of proteobacteria.^5,6 Each transcription activator–like effector (TALE) module recognizes a single base pair, and standard modules for each of the four possibilities seem to behave well in essentially any sequence context. TALENs (TALE nucleases) (Figure 2) consist of multiple TALE domains fused to the FokI cleavage domain, and they have outperformed ZFNs in many early trials.

Although TALE modules make design for new targets much easier and apparently more reliable, some questions about specificity remain. Ask any biochemist or molecular biologist what the gold standard is for DNA sequence recognition and the answer will be: Watson–Crick base pairing. This is the key to the proposal by Jinek et al.¹

The reagents described in the new article derive from a novel system of adaptive immunity, found in many species of bacteria and archaea, called CRISPR (for the clustered regularly interspersed short palindromic repeats that characterize the genomic loci involved). This system is rather complex, and several variations exist, but the common features can be outlined as follows.⁷

When a viral genome or plasmid enters one of these microbial hosts, a few fragments of the invading DNA are captured as “spacers” between identical “repeats” that are specific to the particular CRISPR system (Figure 3). Both the repeats and spacers are typically a few tens of base pairs in length. Transcription of the locus produces a precursor RNA that is processed into smaller fragments, each carrying one spacer linked to a portion of a repeat. When the same viral or plasmid sequence invades the host again, the corresponding spacer RNA guides destruction of the invading RNA or DNA, depending on the particular system. Cas (CRISPR-associated) proteins mediate both production of the spacer RNAs and cleavage of the invading target.

Figure 3.

The type II CRISPR system. In a bacterial genome, identical repeats flank virus- or plasmid-derived spacer sequences in tandem arrays (blue). Long transcripts (green line) are processed into short RNAs containing a single spacer and a partial repeat. These short RNAs form partial duplexes with tracrRNAs and are bound by the Cas9 protein (orange oval). The complex then cleaves invading viral or plasmid DNA directed by the spacer RNAs. tracrRNA, trans-activated CRISPR RNA.

In the type II CRISPR systems the Cas9 protein forms a complex with the spacer-containing RNA and a second RNA, trans-activating CRISPR RNA (tracrRNA), that is partially complementary to the repeat sequence, and this complex catalyzes destruction of the invading DNA (Figure 3). Jinek et al. have shown that this DNA cleavage reaction can be recapitulated in vitro with purified Cas9 and a single RNA molecule that has the minimal required features of both spacer and tracr (Figure 4). Only a target DNA that matches the spacer RNA sequence is cleaved. Different spacer RNAs direct cleavage to different DNA sequences, and both strands of the target are cut.

Figure 4.

The CRISPR minimal-cleavage elements described by Jinek et al.¹ A single RNA (green lines) with the critical elements of spacer and tracrRNA binds Cas9 protein (orange oval) and directs cleavage (arrowheads) to a sequence in DNA (blue) that has homology to the spacer. The region of RNA–DNA base pairing provides cleavage specificity. The target must also have a particular two– to three–base pair sequence adjacent to the region of homology, called PAM, which is recognized by the complex. PAM, protospacer adjacent motif; tracrRNA, trans-activated CRISPR RNA.

Jinek et al. used spacers of 20–30 nucleotides to demonstrate the efficiency and specificity of cleavage by Cas9–RNA complexes. Both supercoiled plasmid DNA and short, double-stranded oligonucleotides are good substrates. Each DNA strand is cut by one of the two separate nuclease domains of Cas9; mutation of either active site leads to single-strand cleavage. The critical region of RNA–DNA duplex is at the downstream end of the spacer DNA, which corresponds to the 3′ side of the RNA in the match. A minimum of 16 base pairs is required. In addition, Cas9 recognizes 2 or 3 base pairs in the DNA just to the right of the hybrid region, called PAM (protospacer adjacent motif), which is probably also recognized during the establishment phase of immunity. Finally, a region of RNA duplex between the repeat segment and its complement in tracrRNA is necessary for cleavage. Using information about all these requirements, Jinek et al. produced a single RNA molecule (Figure 4) that guides cleavage in conjunction with Cas9.

All the experiments described above were performed in vitro with purified components, but several aspects have been confirmed in bacteria. The authors make the bold prediction that this system can potentially be used in place of ZFNs or TALENs for targeted genomic cleavage in higher organisms. Let's think about how this might work.

Cas9 protein and the targeting RNA would need to be expressed in the cells or organism of interest. Presumably both could be produced from DNA vectors with appropriate promoters; Cas9 messenger RNA and the targeting RNA could be produced in vitro and introduced into cells; or purified protein and synthetic targeting RNA could be introduced. The optimal choice would depend on the experimental situation.

Recognition specificity is provided by the match between the targeting RNA and the DNA target. Watson–Crick pairing can be very specific, and a match of 16–20 base pairs is sufficient to ensure recognition of a unique sequence in a complex genome. Discrimination could therefore be more precise than with either zinc fingers or TALE modules.

A key issue for all gene-targeting reagents is how delivery to the target cells or organisms will be accomplished. In many animals, direct injection of nuclease-encoding messenger RNAs into early embryos has proved quite effective in generating germline modifications. For example, this approach has added a very welcome tool to the arsenal of rat geneticists.⁸ For human somatic therapy, targeting is most easily applied to situations that allow ex vivo treatment of cells before reinfusion. Cells of the hematopoietic lineages are obvious targets, and as more pluripotent cell types are identified or generated, the applications will expand.

What about activity of the system in eukaryotic cells? Both zinc fingers and TALE modules come from natural transcription factors that bind their targets in a chromatin context. This is not true of the CRISPR components. There is no guarantee that Cas9 will work effectively on a chromatin target or that the required DNA–RNA hybrid can be stabilized in that context. This structure may be a substrate for RNA hydrolysis by ribonuclease H and/or FEN1, both of which function in the removal of RNA primers during DNA replication. Only attempts to apply the system in eukaryotes will address these concerns.

Intriguingly, some eukaryotic cells appear to have an inherent system to make double-strand breaks in the region of DNA–RNA hybrids. This was revealed by disabling ribonuclease H in yeast.⁹ Such a maneuver might enhance the activity of CRISPR cleavage as well, but with the potential side effect of inducing breaks at multiple regions of transcription.

Gene editing through base pairing has been attempted many times and is still being pursued. The efficiency of modification by introduction of simple oligonucleotides, chemically modified oligos, or oligo mimics such as peptide nucleic acids remains discouragingly low in most cases.^10,11 Triplex-forming oligonucleotides¹² have shown activity, but with a limited range of targets and less efficiency than ZFN or TALEN cleavage. Whether the CRISPR system will provide the next-next generation of targetable cleavage reagents remains to be seen, but it is clearly well worth a try. Stay tuned.