CRISPR RNA-guided autonomous delivery of Cas9

Abstract

Cas9 is an endonuclease that can be programed to autonomously deliver diverse effectors to specified genetic addresses. High-resolution structures of this protein and its associated CRISPR RNA (crRNA) guide explain the molecular mechanisms of crRNA-guided DNA recognition and provide a molecular blueprint that has facilitated structure-guided functional remodeling. Here, we retrace events that led from early efforts to understand the central role of Cas9 in CRISPR-mediated adaptive immunity to contemporary efforts aimed at developing and deploying this enzyme for programmable genetic editing.

Introduction

Recording information in DNA is the cornerstone of biology and the basis of genetic inheritance. In 2005, three groups independently reported that CRISPR (Clustered Regularly-Interspaced Short Palindromic Repeats) loci in bacterial and archaeal genomes often contain short fragments of DNA derived from “transmissible genetic elements such as bacteriophages and conjugative plasmids”^1–3. These observations led the authors to hypothesize that CRISPR loci were central components of a heritable, nucleic acid based immune system for protection from “genetic aggressors”. In 2007, Barrangou et al. tested this hypothesis by challenging Streptococcus thermophilus with lytic phages. The authors showed that the surviving bacteria often contained new phage-derived “spacer” sequences in the CRISPR locus. Deletion of specific spacers resulted in bacterial sensitivity to the corresponding phages. On rare occasion, phages were isolated that escaped CRISPR-mediated protection. These “escape” phages contained mutations in the complementary target sequence (protospacer) or in a sequence adjacent to the protospacer, the protospacer adjacent motif, PAM^4–7. Collectively, these experiments established a direct link between the DNA sequence at CRISPR loci and sequence-specific phage resistance (Box 1).

Box 1. An overview of CRISPR-mediated immunity and application in editing.

In Type II CRISPR-Cas systems, acquisition of new spacers requires Cas1 and Cas2 proteins, as well as Cas9. While the specific function of Cas9 in new sequence acquisition has not been well established, it does play a role in selecting pre-spacers that contain PAMs (protospacer adjacent motif)^101–103. During interference, CRISPR loci are transcribed and each repeat in the CRISPR RNA base-pairs with a small portion of the trans-acting RNA (tracrRNA). The resulting RNA duplexes are cleaved by RNaseIII and the mature crRNA-tracrRNA hybrids are loaded into Cas9⁹. Cas9 scans dsDNA for PAMs and PAM binding destabilizes the DNA duplex for crRNA-guided sampling of the adjacent sequence²⁹. Hybridization between the crRNA-guide and target DNA triggers a conformational change in Cas9 that activates the RuvC and HNH nuclease domains^{24,25,30,34,93}, leading to the cleavage of the invading phage DNA. Jinek et al showed that the tracrRNA and the crRNA hybrid could be covalently connected with a short linker and that the resulting single-guide RNA (sgRNA) could be used to target cleavage of any dsDNA adjacent to a PAM by design¹⁰. This programmable nuclease has now been used for a wide variety of genome engineering applications in cells from all domains of life.

CRISPR loci are critical, but insufficient, for defense against phages⁴. They are often flanked by a diverse cassette of CRISPR-associated (cas) genes. Early experiments performed in S. thermophilus demonstrated that mutations in cas9 (originally called cas5) result in loss of phage resistance, even when the CRISPR locus contains a “spacer” complementary to the invading phage⁴. This suggested that Cas9 is critical for CRISPR-mediated phage defense. To investigate the mechanism of protection and the fate of phage DNA, Garneau et al. sequenced viral DNA isolated from infected bacterial cells and showed that both stands of the target DNA were cleaved within the spacer sequence, resulting in a blunt-ended cleavage product⁸. While this work clarified the mechanism of defense, i.e. cleavage of invading phage DNA, the mechanism for loading CRISPR RNAs into Cas9 remained enigmatic. In 2011, Emmanuelle Charpentier’s laboratory reported the identification of a trans-activating crRNA (tracrRNA) with sequence complementarity to repeat sequences of the CRISPR RNA⁹.

The authors showed that processing of the long primary CRISPR transcript into spacers for Cas9 delivery was dependent on the tracrRNA, Cas9, and an endogenous RNase III enzyme (Box 1). Subsequently, Jinek et al. purified Cas9 protein from Streptococcus pyogenes and showed that Cas9-mediated cleavage of dsDNA relied on both the crRNA-guide and the tracrRNA¹⁰. To simplify this two-RNA system, Jinek et al. fused the 3’ end of the crRNA to the 5’ end of the tracrRNA to generate a single chimeric RNA and demonstrated that this ‘single guide’ RNA (sgRNA) could target Cas9 to cleave virtually any DNA sequence by design (Box1 and Figure 1). Similarly, Gasiunas et al. reported purification of the Cas9 protein from S. thermophilus and demonstrated programmable cleavage of dsDNA targets¹¹. Together, this work demonstrated that Cas9 is a programmable nuclease, with potential for applications in genome engineering.

Figure 1: Impact of sgRNA loading and target DNA binding on Cas9 Structure.

(a) Domain organization of the Cas9 protein. (b) Sequence and secondary structure of the sgRNA are shown on the left. The location bridge helix (BH) is highlighted in magenta. On the right, a ribbon diagram of the Cas9 RNP oriented to highlight RNA interactions with the bridge helix (magenta). (c) The Cas9 protein has a crescent-shape composed of two lobes. Three recognition domains (REC1– 3, gray) and the bridge helix (BH, magenta) form the REC-lobe. The NUC-lobe is composed of two nuclease domains (RuvC-light blue and HNH-royal blue) and a PAM interacting (PI) domain (green). Five crystal structures [Cas9 alone (4CMP), with sgRNA (4ZT0), with sgRNA/complementary ssDNA (4OO8), with sgRNA/complementary ssDNA/dsDNA PAM (4UN3), and sgRNA/dsDNA containing a PAM and protospacer (5F9R)] highlight major conformational changes at each stage of assembly.

Science generally advances at a steady pace, one result leading to the next experiment, and years of work eventually guides the field to fundamental new understandings. However, work on CRISPR systems, and Cas9 in particular, has been anything but “usual”. In 2013, less than six months after the two reports on programmable cleavage of dsDNA by Cas9^10,11, three papers provided in vivo proof that RNA-guided Cas9 nucleases could be used to edit genes in both mouse and human cell lines^12–14. Evidence of a new, easy-to-use genome editing tool captured the attention of scientists across a wide range of different disciplines and structural biologists raced to provide a molecular understanding of Cas9 function. In this review, we illustrate how mechanistic insights, particularly those gained from structures of S. pyogenes Cas9 (SpCas9), have facilitated engineering of the Cas9 ribonucleoprotein (RNP) to improve the genome editing process and to impart new functionalities on the protein.

Structural insights into the Cas9 catalytic mechanism

The first structures of Cas9 were published in 2014 and currently, there are 30 Cas9 structures from six different organisms deposited in the protein databank^15–22. Cas9 proteins are diverse, ranging in length from 950 to 1650 amino acids and some share as little as 6% amino acid sequence identity (12% similarity)²³. Despite this diversity, all Cas9 proteins share a bilobed structure, composed of nuclease (NUC) and recognition (REC) lobes (Figure 1)^15,19. The NUC lobe contains two nuclease domains (i.e., HNH and RuvC), and a C-terminal PAM interacting domain (PI). The REC lobe is composed of three α-helical domains (REC1–3), and a long arginine-rich helix (called the bridge helix, BH), which connects to the two lobes (Supplemental Video 1).

The structure of Cas9 alone was originally described as adopting an ‘auto-inhibited’ conformation, such that the nuclease domains are inaccessible or inactive¹⁵. However, dynamic measurements of Cas9 performed using Förster resonance energy transfer (FRET), high-speed atomic force microscopy (HS-AFM), and molecular dynamics simulations have shown that Cas9 is conformationally flexible prior to binding the RNA guide^24–26. This flexibility may explain recent biochemical results suggesting that Cas9 from both S. pyogenes (SpCas9) and Francisella novicida (FnCas9) exhibit non-sequence specific nuclease activity in the absence of an RNA guide²⁷.

Loading of either a single-guide RNA (sgRNA) or the tracrRNA+crRNA pair induces large-scale conformational changes that create extensive contacts between the RNA and the enzyme. These interactions coincide with a reduction in Cas9 flexibility and a significant reduction in non-sequence specific nuclease activity^17,27. RNA binding by Cas9 results in a 65Å rigid-body migration of the REC3 domain from one end of the REC lobe to the other, and a 60Å rigid-body migration of the entire NUC lobe (Figure 1c)^15,17. Movement of the NUC lobe orients the PI domain for PAM detection, and the bridge helix (BH) is wedged between the tracrRNA and the crRNA (Figure 1b–c). This positions the bridge helix perpendicular to the RNA duplex formed by complementary regions of the tracrRNA+crRNA pair and spatially separates the tracrRNA from the crRNA guide sequence. One side of the bridge helix interacts with the first 9-nucleotides of the crRNA-guide (i.e., the first part of seed region), while the other side of the bridge helix make contacts with the first stem-loop of the tracrRNA (Figure 1b). Collectively, the RNA forms a triangular structure that completely encases the bridge helix. The guide, tetra-loop, and stem-loop 1 of the RNA bind the REC lobe (K_d=0.7nM), while stem-loops 2 and 3 have high-affinity (K_d=0.2 nM) for the NUC lobe. RNA binding brings the two lobes together, restricting conformational flexibility²⁸.

To understand the mechanism of crRNA-guided DNA interrogation and nuclease activation, several labs have determined structures of the SpCas9 RNP bound to short fragments of DNA^16,19,20. A structure of the Cas9 bound to a 23-nucleotide fragment of single-strand (ss)DNA complementary to the crRNA-guide reveals rigid-body rotations in both REC2 and REC3 domains that are necessary to accommodate the complementary target stand, while avoiding steric clashes. These rotations are accompanied by a 25Å movement of the HNH domain toward the DNA-RNA duplex, although the active site is oriented away from the DNA (Figure 1c)^17,19. However, this ssDNA target did not contain a PAM. Anders et al. subsequently determined the crystal structure of SpCas9 bound to a partially duplexed DNA target containing a duplexed TGG PAM sequence providing insights into the molecular mechanism of PAM recognition. The guanine bases of the PAM are recognized by nucleotide specific hydrogen bonds to a pair of arginine residues (R1333/R1335) that extend into the major groove (Figure 1C and 2a)²⁰. Interestingly, FnCas9 also recognizes an NGG PAM and the structure reveals a similar mechanism that relies on a pair of arginines, but the arginine residues are on distinct secondary structural elements²¹. Although the PI domains of SpCas9 and FnCas9 share a conserved fold, the arginines responsible for PAM recognition in FnCas9 are separated by 29-residues and located on two different loops, while the arginines in SpCas9 are separated by a single amino acid^20,21.

Figure 2: Engineering Cas9 to change PAM specificity.

(a) A pair of arginine residues (R1333 and R1335) in the PI domain of SpCas9 extend into the major groove, making nucleotide specific hydrogen bonds with guanine bases on the non-complementary strand (PDB: 4UN3). (b) A screen for Cas9 mutants with altered PAM specificity using a library of Cas9 PI domain mutants identified the VQR, EQR, and VRER mutants, which recognize the NGAN, NGAG, and NGCG PAMs, respectively. Crystal structures 5B2R (VQR), 5B2S (EQR), and 5B2T (VRER) show the interactions between the mutant Cas9s and altered PAMs. (c) The ω-dCas9 fusion protein was programmed to bind a protospacer upstream of the phage pIII gene and a phage-assisted continuous evolution protocol was used to isolate a family of Cas9 mutants that bind to diverse PAM sequences (green “NNN”). Cas9 mutants that bind the protospacer activate expression of pIII, which encode an essential phage tail fiber protein (G3P). Phages expressing G3P will package mutant cas9 and amplify it by reinfection. Absence of G3P protein results in phage particles with attachment defects, which are lost from the library. Mutations with broadened PAM specificities can induce pIII expression from a greater number of the NNN PAM sequences.

Finally, binding a double-stranded DNA target containing a PAM and a protospacer results in a 110° rotation of the HNH active site toward the scissile phosphate on the complementary DNA, suggesting that crRNA-guided hybridization allosterically regulates the HNH domain (Figure 1c)¹⁶. However, in this structure, active sites of the HNH and RuvC nuclease domains remain 10Å and 5.5Å from their respective scissile phosphates¹⁶. Notably, both HNH and RuvC are metal-ion dependent phosphodiesterase domains (one-metal and two-metal mechanisms, respectively) and ethylenediaminetetraacetic acid (EDTA) was used to prevent DNA cleavage during complex reconstitution and subsequent crystallization. This locked the complex in a pre-cleavage state. However, the metal ions may also facilitate active site engagement with the DNA target, which cannot be observed under these conditions. A structure of Cas9 with engaged active sites will help clarify mechanisms of cleavage and we anticipate that non-cleavable modifications to the phosphate backbone may help trap the metal-dependent nuclease domains in action.

Target identification and nuclease activation proceed through a series of sequential steps. Cas9 searches the genomic landscape for PAMs via a three-dimensional diffusion process²⁹. PAM binding destabilizes the DNA duplex and facilitates directional crRNA-guided strand invasion that initiates at the PAM and proceeds toward the 5’ end of the guide²⁹. Mismatches between the crRNA-guide and the complementary DNA result in binding and cleavage defects, which are position-specific. PAM-proximal mismatches in the seed region lead to rapid dissociation, while mismatches at the opposite end (PAM-distal) have much less of an impact on binding. However, binding alone is insufficient for nuclease activation and recent work indicates that complete (or near complete) base pairing is necessary for nuclease activation^30–33. Target hybridization positions the HNH nuclease domain in the active conformation, which then allosterically activates the RuvC nuclease domain^30,34. Allosteric activation is transmitted by formation of an extended α-helix between the two domains (S909-N940) and mutations that inhibit helix formation block RuvC cleavage³⁰. Following cleavage of both DNA strands, Cas9 remains tightly bound to the two ends of the DNA. Thus, Cas9 is a single-turnover enzyme^29,34.

Engineering altered PAM specificity

The requirement for PAM binding increases target specificity, which limits off-target binding. However, strict PAM requirements also reduce programmable versatility of the enzyme. Several strategies have been used to alter PAM specificity. Different Cas9 proteins have distinct PAM recognition profiles and grafting the PI domain from S. thermophilus Cas9 (NGGNG PAM) onto SpCas9 (NGG PAM) resulted in a chimera with specificity for the desired NGGNG PAM¹⁹. However, these two proteins are closely related (60% identity) and a PAM swap strategy may not be generalizable to more diverse Cas9 proteins. To expand the PAM binding repertoire of SpCas9 from NGG to NAA, Anders et al. borrowed a trick previously used to alter binding sites of homing endonucleases²⁰. Since major groove recognition of adenosines often involves bidentate hydrogen bonding with either asparagine or glutamine, the authors mutated the SpCas9 arginines (R1333/R1335) to glutamines in an attempt to broaden targeting to an A-rich PAM²⁰. However, this structure guided engineering effort failed, resulting in Cas9 proteins with DNA binding defects.

As an alternative approach to expand the PAM binding repertoire, Kleinstiver et al. used a library encoding Cas9 proteins with random mutations in the PI domain to identify Cas9 mutants with distinct PAM preferences³⁵. The Cas9 library was programed to target a protospacer in the ccdB gene, which encodes a toxin targeting DNA gyrase, flanked by either an NGA or NGC PAM. Cells with a Cas9 mutant capable of cleaving the toxic DNA survive and sequencing cas9 alleles in the survivors revealed three predominate mutations (i.e.,VQR, EQR, and VRER), which recognize NGAN, NGAG, and NGCG PAMs, respectively (Figure 2b). Structures of these Cas9 mutants bound to their respective PAMs reveal multiple changes that synergistically displace the phosphodiester backbone of the PAM duplex in a way that reorients bases for sequence specific recognition in the major groove³⁶. A similar screening approach was used to identify Cas9 mutants from Staphylococcus aureus (SaCas9) with relaxed PAM specificity (NNGRRT to NNNRRT)³⁷.

To evolve SpCas9 variants with promiscuous PAM compatibility and high DNA specificity, Hu et al. recently performed a phage-assisted continuous evolution (PACE) experiment designed to derive variants of SpCas9 with an expanded PAM (xCas9) (Figure 2c)³⁸. In this setup, E. coli phages encode a mutant library of catalytically dead Cas9 proteins that are fused to the ω subunit of bacterial RNA polymerase (ω-dCas9). The ω-dCas9 is programmed to target a protospacer upstream of a gene encoding a phage attachment protein (i.e., G3P), which is necessary to form infectious virions. However, the PAM sequence is intentionally scrambled (NNN), so only phages carrying ω-dCas9 mutants capable of promiscuous PAM recognition are able to drive expression of the phage attachment protein. This approach identified a mutant (xCas9 3.7) with broadened PAM specificity (NGG, NG, GAA, and GAT). The E1219V mutation carried by this variant is near the PAM binding site and arose in the first round of evolution. However, it is difficult to rationalize the impact of many of the other mutations on PAM recognition, although some are near the RNA/DNA duplex and may have an impact on target binding.

Programmable repair

Cas9-induced DNA double-strand breaks (DSB) are often repaired by non-homologous end joining (NHEJ), an efficient repair mechanism that “stiches” the free ends of the DNA back together. However, NHEJ is error prone, often resulting in insertions or deletions, which can perturb gene function and generate functional gene knockouts. Knockouts are critical for determining the biological function of a specific gene, but the genetic basis of many plant and animal diseases are already known. Therefore, the next major challenge is to create methods that enable efficient and precise sequence modification to prescribed genes. Homology-directed repair (HDR) of double-stranded DNA breaks using DNA repair templates flanked by sequences homologous to regions on either side of the break has proven effective for introducing new sequences at specified locations, but the efficiency of HDR is low relative to NHEJ.

To alter the cellular NHEJ/HDR DNA repair ratio, Cas9 has been fused to various effector proteins (Figure 3). Genetically tethering Cas9 to E. coli Rec A, a single-strand DNA binding protein with a central role in homologous recombination, resulted in an unexpected increase in NHEJ in human cells³⁹. While Rec A from E. coli did not have the desired outcome, fusions to eukaryotic enzymes that play a critical role in HDR (i.e., Rad52 or CtIP) enhances HDR by 2–3 fold^40,41. Since many of the proteins necessary for HDR are only active in late S and G2 phases of the cell cycle, synchronizing Cas9 expression with activation of the HDR machinery also enhances templated repair. This has been accomplished by fusing Cas9 to the N-terminal region of geminin, a protein that is subject to ubiquitination and degradation in phases of the cell cycle during which the HDR machinery is not active⁴².

Figure 3: Cas9 for autonomous delivery of diverse effectors.

Cas9 activity can be enhanced or modified by association with various effectors. Effectors are fused directly to Cas9 or tethered to Cas9 through RNA aptamers integrated into the sgRNA. In general, effectors have been used to influence DNA repair, control the timing and location of Cas9 activity, regulate transcription or image chromosomes.

In addition to fusing Cas9 to proteins involved in HDR, DNA repair templates have also been hitched to Cas9 or the sgRNA for delivery to the cleavage location. Single-stranded DNA donors have been linked to synthetic crRNAs using ‘click’ chemistry⁴³. In addition, sgRNA containing a streptavidin binding aptamer designed to bind biotinylated DNA donors has been used⁴⁴. Alternatively, Cas9 has been fused to SNAP-tags, which covalently react with O⁶-benzylguanine (BG)-labeled DNA repair templates⁴⁵. This results in a 20-fold increase in the HDR/NHEJ ratio with a typical increase of 2–4-fold in total HDR levels.

Base editing is an elegant alternative to efforts that rely on HDR (Figure 3)^46–48. Fusing dCas9 to cytidine deaminase results in irreversible conversion of C to U, which is read as T by DNA polymerase^46,47. However, cytidine deamination can occur spontaneously in cells, and G:U mismatches are efficiently repaired by cellular uracil DNA glycosylase (UDG), which restores the original G:C pair. To improve G:C to A:T editing efficiencies, cytidine deaminase and a UDG inhibitor protein (UGI) have been fused to a single dCas9 molecule, leading to a 4-fold improvement in C to T conversion^46,47. Base editing efficiency is further improved by switching from dCas9 to a Cas9 nickase (nCas9), which only cleaves the complementary strand. Eukaryotic mismatch repair uses nicks in newly synthesized DNA to determine which strand is used as the repair template. Nicking the complementary (G-containing) strand forces the cell to use the edited U as the template and further increases the desired C:G to T:A conversion by 2- to 4-fold^46,47. To broaden the applicability and versatility of base editing, Gaudelli et al. created an adenosine deaminase-nCas9 fusion for base editing of A:T pairs to G:C⁴⁸. Using directed evolution of a naturally occurring bacterial tRNA adenosine deaminase, they generated a novel adenosine deaminase that works on DNA substates⁴⁸.

Cas9 as a governor of gene expression

Catalytically dead Cas9 (dCas9) is routinely used to control expression of specific genes without permanently altering the DNA sequence (Figure 3). Hitching transcriptional activators, such as VP6, p65AD, or the ω subunit of RNAP, to dCas9 has been successful in promoting gene expression in both eukaryotes^49–51 and bacteria⁵². As an alternative approach, integration of RNA aptamers into the sgRNA facilitates functionalization of the Cas9 RNP without requiring protein engineering. RNA aptamers have been inserted in the tetraloop, stem-loop 2, and the 3’ end of the sgRNA with little impact on RNA loading into Cas9 (Figure 1b). Fusing aptamer binding proteins to transcriptional activators then allows activation of targets specified by the guide RNA⁵³.

Methods for repressing gene expression have also been developed. In fact, simply targeting dCas9 to the 5’ UTR or coding sequence of a gene of interest inhibits transcription⁵⁴. Linking a transcriptional repressor domain, such as the KRAB (Krüppel-associated box) domain, to dCas9 improves gene silencing further and can be used to target promoter regions as well as the 5’ UTR or coding sequence⁴⁹.

Engineering complex genetic circuits requires sophisticated control methods that permit simultaneous repression, activation or DNA cleavage at multiple locations in a single cell. For this purpose, several guides can be used to deliver Cas9 to multiple genetic loci in a single cell (multiplexing). Additional layers of complexity can be built into the sgRNA by including different aptamer sequences in the sgRNA loops as described above and by altering the guide length. Complete complementarity between the RNA-guide and the DNA target is necessary to activate the Cas9 nuclease domains, while short guides (14- to 15-nucleotides) result in specific binding without cleavage. Catalytically active Cas9 can thus be sent to several genetic addresses and the length of the RNA-guide can be used to control Cas9-mediated gene activation or repression versus DNA cleavage^55,56.

Finally, epigenetic control of gene expression is a key factor in fundamental biological processes, such as development and cellular differentiation. It also plays an important function in human diseases, including cancer, and has potential roles in aging and age-related diseases^57,58. dCas9-guided epigenetic modifications (e.g., cytosine methylation^59,60, histone demethylation⁶¹, and histone acetylation or deacetylation^62,63) can be used to experimentally test cause-and-effect relationships associated with these modifications. That said, recent evidence suggests that at least the dCas9-methylase enzymes have significant off-target effects at sites with as little as a 5-nucleotide seed match. They can even cause hypermethylation on a global level due to high expression levels of the methylase^64,65. Substantial off-target binding agrees with FRET and early dCas9 ChIP-seq experiments observing stable dCas9 binding to sites with up to eleven PAM distal mismatches^30,31,33,66. Conversely, the Cas9-KRAB repressor showed no significant off-target transcriptional silencing by RNA-sequencing⁴⁹. This might be due to the fact that repression requires binding near the promoter or transcription start site and off-target binding at other loci may have no impact on RNA levels⁶⁷. Pflueger et al. recently reported a modular, non-covalent coupling of dCas9 to DNA methyltransferase using the SunTag system. This permits independent control of expression levels for Cas9 and the methylase and can be used to substantially reduce off-target methylation by limiting methylase expression levels. High on-target methylation is retained through multimeric recruitment of the methylase to the dCas9-SunTag⁶⁸. Collectively, these results caution that off-target impacts of dCas9-guided enzymatic effectors should be examined on a genome wide scale.

Spatiotemporal control of Cas9

Programmable delivery of Cas9 to genetic targets within the cell is relatively robust, but in vivo deliver of Cas9 to a specific subset of cells for therapeutic intervention remains a major challenge. A recent review by Lino et al provides an overview of the various delivery vehicles reported for CRISPR/Cas9⁶⁹. Below we highlight a few methods that impart spatial or temporal control of Cas9 activity.

The CRISPR/Cas9 genome editing system has been modified for use in conjunction with adeno-associated virus (AAV). AAV serotypes preferentially infect specific cell types, providing a natural platform for delivery to major organs, including the heart, kidney, liver, lung, pancreas, muscle, and central nervous system. Another advantage of AAV is its ssDNA genome, since single-stranded DNA donors result in higher levels of template-directed repair⁷⁰. While there are advantages to using AAV, cargo size is a clear limitation. The total foreign DNA cargo is limited to less than 4.4Kb. While SpCas9 and a chimeric sgRNA (~4.2 kb) are just below the limit, it leaves little room for additional regulatory elements or an HDR donor. To overcome the AAV packaging limit, Cas9 can be split into two intein-tagged polypeptides for delivery by two viral constructs. Inteins are protein sequences analogous to mRNA introns that are self-catalytically excised, while splicing the attached Cas9 fragments together^71–73. As an alternative to inteins, Wright et al. designed a split-Cas9 system in which the NUC and REC lobes are expressed as separate polypeptides that can be delivered in separate AAVs²⁸. Here, the sgRNA serves as a regulatory agent that “glues” the two halves together creating a functional RNP. Similarly, fusing Cas9 to the hormone-binding domain of the estrogen receptor allows spatial control through the ligand-induced nuclear translocation of Cas9 in the presence of 4-hydroxytamoxifen⁷⁴. Finally, dCas9-guided gene activation can also be controlled using small molecule stabilizers or photo-regulators to pair dCas9 and a transcriptional activator upon induction^75–78.

In an effort to target Cas9 to the liver, Rouet et al. decorated Cas9 with synthetic ligands that bind asialoglycoprotein receptors expressed on hepatocytes⁷⁹. This approach results in preferential Cas9 uptake in hepatocytes, but the protein is frequently trapped in the endosome, which limits editing efficiencies. To liberate Cas9 from the endosome, the ligand-decorated Cas9 was co-transfected with endosomolytic peptides, which improved editing, but the efficiency nevertheless remains low (5%). Collectively, the emerging repertoire of delivery systems that impart spatiotemporal control on Cas9 is progressing toward minimizing risks associated with editing DNA in specific cell types in vivo.

Chromosomal imaging and manipulation

There is growing interest in understanding the three-dimensional organization of the cellular genome and its impact on gene expression⁸⁰. Cas9 has been used to label genetic loci either by fusion of fluorescent proteins directly to Cas9 or by using RNA aptamers to attach fluorescent proteins or quantum dots (Figure 3)^44,81,82. Chen et al. used this approach to monitor telomere dynamics following DNA damage⁸². Beyond imaging, Morgan et al. developed a method for inducing chromosomal looping by tethering dCas9 from Staphylococcus aureus and S. pyogenes to two proteins that heterodimerize in the presence of S-(+)-abscisic acid (ABA)⁸³. The dCas9 proteins were programmed to bind spatially separated regions of the chromosome and looping was induced by ABA. Using this approach, the authors demonstrated that the promoter for a gene could be “recruited” to another regulatory sequence for activation.

Designing Cas9 and the RNA-guide for higher specificity

Clinical success of Cas9 will depend on the specificity of these enzymes. One of the first examples of enhanced specificity was the use of Cas9 nickases. Wild-type Cas9 cleaves both strands of the DNA duplex, while HNH or RuvC mutants act as nickases that cleave a single DNA strand. Thus, generating DSBs with Cas9 nickases requires two paired enzymes that cleave adjacent locations on opposite strands of the genome. The requirement for two independent nicks at adjacent locations reduces on-target efficiency, but it also minimizes the risk of off-target DSBs⁵³. Similarly, specificity can be increased by fusing dCas9 to a dimerization-dependent FokI nuclease. Like Cas9 nickases, DNA cleavage requires colocalization of two independently programmed dCas9 molecules to facilitate dimerization and activation of the FokI nuclease^84,85.

The “two-hit” requirement for introducing DSBs (e.g., Cas9-nickases or dCas9-FokI fusions) is not practical in many situations where delivery of one Cas9 is already a challenge. To improve specificity of the nuclease-active Cas9, Slaymaker et al. created a panel of “enhanced specificity” Cas9s (eSpCas9s) by making a series of structure guided mutations predicted to destabilize interactions with the non-complementary DNA strand (Figure 4b)⁸⁶. Protein mediated interactions with the displaced strand facilitate crRNA-guided strand invasion by limiting the back reaction (i.e., reannealing of the DNA duplex). Thus, mutations that make the displaced strand more accessible were predicted to increase stringency for on-target interactions by promoting DNA rewinding. Similar engineering strategies have been used to create a series of structure-guided mutations that eliminate interactions with the complementary strand. Eliminating these protein contacts results in higher-fidelity Cas9 (Cas9-HF1) (Figure 4b)⁸⁷. Chen et al. hypothesized that there must be a mechanism to detect complete RNA/DNA heteroduplex formation resulting in transition of the HNH domain to the active conformation. They identified residues in the REC3 domain near the PAM distal end of the RNA/DNA duplex that could interact with the fully-hybridized duplex. Mutation of four of these residues to alanine resulted in hyper-accurate Cas9 (HypaCas9), with improvements in specificity similar to what has been reported for eSpCas9 and SpCas9-HF1 (Figure 4b)⁸⁸. Finally, screening for mutants with increased specificity using a Cas9 library with random mutations in REC3 resulted in evolved Cas9 (evoCas9, Figure 4b)⁸⁹. EvoCas9 may improve specificity through both mechanisms, since it combines mutations that are predicted to restrict HNH activation and mutations to residues that interact with the complementary DNA strand similar to SpCas9-HF1 (Figure 4a).

Figure 4: Structure-guided design of Cas9 for enhanced target specificity.

(a) Cas9 target verification is a multi-step process. After binding to a double-stranded PAM, the guide RNA invades the target DNA and begins base-pairing with the complementary strand at the PAM proximal end. Hybridization proceeds toward the PAM distal end, checking for mismatches. Once the DNA is fully unwound, the HNH domain can enter the active conformation and allosterically activate the RuvC domain for cleavage. Several Cas9 mutants have been created to improve the specificity of Cas9. (b) Positively charged residues on the nuclease domains were predicted to interact with the non-complementary DNA strand (PDB: 4UN3). A series of alanine substitutions were designed to limit non-sequence specific interactions with the displaced strand and “enhance specificity” of SpCas9 (eSpCas9). However, a subsequent structure of SpCas9 bound to double-stranded DNA revealed a conformational change that precludes many of these predicted interactions (PDB: 5F9R). The 4UN3 structure was also used to create high-fidelity Cas9 (Cas9-HF1) with mutations that eliminate interactions with the phosphate backbone of the complementary DNA strand. Mutations in eSpCas9 and Cas9-HF alter the balance between unwinding and rewinding of the DNA duplex in the presence of mismatched targets. Amino acids in the REC3 domain at the PAM-distal end of the RNA-DNA duplex act as an allosteric regulator of HNH nuclease activation. Mutation of these residues to alanine produced hyper-accurate Cas9 (HypaCas9). Screening for mutants with increased specificity using a library of Cas9s with random mutations in REC3 resulted in evolved Cas9 (evoCas9). EvoCas9 may improve specificity by both mechanisms. The surface of the HNH domain is not shown in the structures. (c) Timeline of the creation of the mutants.

In addition to protein modifications, truncated RNA guides (17- to18- nucleotides rather than 21) have also been used to decreased off-target nuclease activity with minimal impact on cleavage at the intended site^90,91. Complete base pairing is necessary to active the nuclease domains and smFRET experiments indicate that Cas9 spends 3-fold less time in the catalytically active conformation when loaded with a 17-nucleotide guide. A single PAM-distal mismatch with a 17-nucleotide guide prevents HNH activation and cleavage entirely^92,93. Further, structural insights into the interactions between Cas9 and the 2’-OH of the RNA ribose have been used to identify crRNA or tracrRNA positions amenable to incorporation of DNA nucleotides or other 2’-OH modifications into the guide RNA resulting in decreased off-target activity^94–97. Incorporation of DNA residues may increase specificity due to the reduced stability of DNA-DNA duplexes compared to DNA-RNA duplexes^94,95. However, the mechanism of improvement seen with other modifications is unclear and the level of improvement varies at different off-target sites^96,97. Finally, incorporating bicyclic bridged or ‘locked’ nucleic acids increases mismatch discrimination in nucleic acid duplexes. These modified bases can be incorporated into synthetic guide RNAs, particularly at key positions of known off-target sites or single-nucleotide polymorphisms⁹⁸.

Future Directions

Structural insights will continue to play a fundamental role in understanding Cas9 function. For example, a high-resolution structure of Cas9 with the active sites engaged on both strands of the DNA target is likely to provide additional mechanistic insights that might aid in the design of more efficient Cas9 tools with fewer off-target effects. Cas9 remains bound to the DNA target after cleavage, which may be advantageous in the context of an immune response, because it prevents access of cellular DNA repair and replication enzymes to the viral genome. However, Clarke et al. recently demonstrated that RNA polymerase ejects Cas9 from DSBs in human cells, turning the single-turnover Cas9 into a multi-turnover enzyme, which improves the efficiency of editing⁹⁹. A similar effect was observed in bacteria, where a 2-fold enrichment of crRNAs targeting the template versus non-template strand of invading phage genomes was noted in S. thermophilus CRISPR arrays. This suggests that crRNA targeting of the template strand may provide a protective advantage to the host.

In addition to enhancing editing efficiencies, there may be additional incentives for targeting the template strand in genome engineering experiments. Cas9 generated DSBs have recently been shown to stall the eukaryotic cell cycle in G1 in a p53 dependent manner¹⁰⁰. Long-lived Cas9 binding following DNA cleavage may reduce HDR by delaying DSB repair until the cells arrest in G1 when the HDR machinery is not available. This means that Cas9-mediated DSBs will be repaired primarily by NHEJ in cells that have a functional p53 pathway. Thus, targeting the template-strand may reduce the DNA bound half-life of Cas9, allowing template-dependent editing before the cells can arrest in G1. Alternatively, we hypothesize that Cas9 mutants that destabilize DNA interactions after cleavage may also increase HDR in untransformed (p53⁺) cells by allowing repair prior to G1 arrest. A structure of the post-Cas9 cleavage complex may help guide the design of these ‘quick release’ mutants.

Applications of Cas9-based genome editing have considerable implications for bioengineering, agriculture, and healthcare, although many challenges remain. While our understanding of the rules that govern ‘autonomous delivery of Cas9 to specified genetic addresses’ continue to improve, delivery of Cas9 to specific cells in multicellular organisms, potentially including human patients, and switching repair from unpredictable indels to programmable modification at specific loci remains difficult. We anticipate that continued improvements that limit off-target activity, allow for programmable genetic manipulations, and delivery of Cas9 to specific cells types will lead to a technology that will reshape our genetic landscape.