Harnessing the Power of Directed Evolution to Improve Genome Editing Systems

Abstract

The recent development of genome editing systems such as zinc finger nucleases (ZFNs), transcription activator-like effectors (TALEs), CRISPR-Cas nucleases, and base editors has enabled the unprecedented capability to engineer the genomes of living cells. The ever-increasing demand for genome editors with improved accuracy, activity, and functionality has stimulated significant efforts to further engineer the genome editing systems. Directed evolution represents a promising strategy to improve the existing genome editing systems and enable new editing functions. Here we review recent representative strategies to harness the power of directed evolution to improve genome editing systems, which have led to state-of-the-art genome editors that have significant implications for diverse applications in both laboratories and clinics.

1. Introduction

One of the ultimate goals of synthetic biology is to reprogram living systems for new or improved functions. To this end, genome editing systems are powerful tools to engineer bacterial and mammalian genomes, enabling precise editing and rapid generation of biological parts and circuits with novel functions available to researchers in synthetic biology.[1] Despite significant advances in genome editing approaches in recent years, further improvements of the substrate scope, activity, specificity, and stability are imperative to their future applications in laboratories and clinics. To address this pressing need, directed evolution has become a method of choice for improving genome editing systems. In particular, recent key advances in in vivo continuous evolution and high-throughput DNA sequencing have reshaped the landscape of biomolecular engineering.[2,3] In this review, we will first provide an overview of the directed evolution technologies that have been used to evolve genome editing systems. We will then survey various genome editing systems improved by these technologies, including zinc finger nucleases (ZFNs), transcription activator-like effectors (TALEs), CRISPR-Cas nucleases, and base editors.

2. Techniques for the Directed Evolution of Genome Editors

Since its advent, directed evolution has proven to be a powerful tool for the rapid screening or selection of a library of biomolecules for enhanced or altered functions. To date, directed evolution techniques vital to the evolution of complex genome editing systems include antibiotic selection, recombinase-based screening and selection, two-plasmid nuclease selection, one-hybrid and two-hybrid screening and selection, and phage-assisted continuous evolution (PACE). Key factors contributing to the efficiency of evolution include the library diversification strategy, the number of variants being screened or selected, and the speed to arrive at a functional endpoint.

Antibiotic selection is one of the most widely used directed evolution methods. In a typical antibiotic selection, the evolving gene is coupled with the gene elements encoding antibiotic resistance, and the bacterial cells hosting the evolving genes are treated with elevating antibiotics concentrations, such that the cells that survive the antibiotic challenge must contain the functional mutants. This strategy can be easily applied to a wide range of selection platforms. For instance, in the recombinase-based selection, an insert containing the recombinase target sequences disrupts the antibiotic resistance gene, such that only the active recombinase variants that can excise the insert can successfully restore the functional antibiotic gene and allow the host cell to survive the antibiotic challenge (Figure 1a).[4] Similarly, a recombinase activity-based reporter for screening rather than selection can be constructed by replacing the antibiotic resistance gene with a fluorescent protein.[5] In another example, a codon-switch selection was designed to use base editing to correct one or more base pairs in the antibiotic resistance gene to restore its function (Figure 1b).[6]

Figure 1.

Directed evolution techniques for evolving genome editors. (a) The general strategy to evolve a protein-of-interest (e.g., recombinase or nuclease) that can modify an inserted sequence in the selection marker (e.g. antibiotic resistance gene) and result in cell survival. (b) Selection for nucleases that can cleave and deactivate the reporter consisting of a ccdB cytotoxic gene, leading to cell survival. (c) Codon-switch selection for base editors that can modify target nucleobases in a mutated reporter (e.g. non-functional antibiotic resistance gene), resulting with a corrected reporter (e.g. functional antibiotic resistance gene) and cell survival. (d) One-hybrid system consisting of a DNA-binding domain (BD) fused with an activation domain (AD, e.g. the omega subunit of RNA polymerase). Interaction of BD with specific binding site stimulates reporter gene expression. (e) Two-hybrid system consisting of protein-of-interest (bait) fused to DB interacting with another protein (prey) fused to AD to recruit RNAP for reporter gene activation. (f) The phage-assisted continuous evolution system. The release of progeny phage that are capable of infecting new bacterial host cells is linked to the continuous infection of the host cell by selection phages (SP) that encode protein of interest to induce expression of gene III located on the accessory plasmid (AP). Increased mutagenesis is triggered through induction of mutagenesis plasmid (MP).

Alternative to the antibiotic selection, a two-plasmid nuclease selection combines a reporter plasmid containing a cytotoxic gene with a library plasmid containing the evolving endonuclease gene that can result in the cleavage and deactivation of the toxic gene (Figure 1c).[7,8] The selection is the result of two separate processes: the expression of the evolving gene and the deactivation of the reporter plasmid by the expressed endonuclease, which leads to the survival of the host cell. Hence, the functional variants of the evolving gene can be enriched through iterative rounds of selection.

The one-hybrid and two-hybrid systems were initially used in yeast to study DNA-protein or protein-protein interactions before being adapted for the directed evolution of these interactions. In the one-hybrid system, a DNA binding domain (BD) and a transcription activation domain (AD) are expressed in fusion to select or screen against a library of promoter sequences (Figure 1d). [9] In the two-hybrid system, a target protein (sometimes referred to as the “bait”) is fused the BD and a binding partner (sometimes referred to as the “prey”) is fused to the AD. The binding between the bait and prey initiates transcription and expression of the reporter gene (Figure 1e).[10] The BD of this system often binds to a fixed sequence, and the library is introduced to bait or prey to select or screen for their binding interactions. The reporter gene in both hybrid systems encodes for a chromogenic/fluorogenic protein or a protein essential for cell survival, such that the functional variants capable of mediating the desired interaction could be enriched through the physical isolation of the host cells in the screening or the survival of the host cells in the selection.

By harnessing the bacteriophage infection process to evolve biomolecules with desired functions, PACE has proven to be a highly efficient approach for continuous laboratory evolution (Figure 1f).[11] To implement PACE technology, the pIII gene is deleted from the M13 phage genome and introduced into the accessory plasmid in the host cells. To trigger mutagenesis, the host cells also carry an inducible mutagenesis plasmid. Since the expression of the accessory plasmid is controlled by the evolving gene in the selection phage, the infected host cells produce progeny phages as they continuously flow through the system. Based on the bacterial one-hybrid system (Figure 1d), the DB is fused to ω subunit of RNA polymerase (RNAP) and binds to the upstream cognate recognition site of the promoter, resulting in downstream transcriptional activation through recruitment or stabilization of the RNAP holoenzyme. The sequence-specific and binding-dependent one-hybrid system efficiently induces the expression of gene III, which then enables phage propagation during PACE. The successfully evolved variants triggers the expression of the pIII gene and then produced new infectious progeny phages. Compared with the traditional evolution approaches, PACE requires minimal researcher intervention during the mutation, selection, and replication progress, and overcomes limitations of traditional directed evolution systems such as limited population sizes and modest mutation rates. A detailed comparison between the noncontinuous and continuous approaches is summarized in Table 1.

Table 1.

Comparison of traditional noncontinuous and continuous evolution approaches.

Directed evolution method	Advantages	Limitations
Noncontinuous (antibiotic selection, two-plasmid selection)	exquisite control of mutation rate and selection stringency through simple adjustments of additive concentrations can be applied to a wide range of platforms can be used in bacterial and mammalian cells	library size limited by transformation efficiency a separate negative selection is required to eliminate the “cheaters”, uninteresting gene variants that bypass selection
Continuous (PACE)	allows simultaneous evolution and selection of protein-of-interest (POI) in bacteria enables a broader exploration of sequence space in a practical timescale positive and negative selection can be implemented simultaneously minimal researcher intervention is required, which makes multiple parallel evolution experiments more accessible	the desired activity can only be linked to gene and organism replication in bacteria, and the activity must proceed faster than phage replication (~15–30min) POI must be expressed faithfully in bacteria the POI scope is limited by the packaging limit of the selection plasmid (~5kb)

A single round of directed-evolution of proteins involves four discrete steps: mutagenesis, gene expression, screening or selection, and replication. Besides choosing the proper evolution technique, modulating the stringency by tuning each step of the evolution can be especially effective. Tuning the selection circuit tailors the parameters that define the relationship between the target protein activity and reporter expression, such as plasmid copy number, reporter promoter strength, and ribosome-binding site strength. Specific combinations of mutation rate and selection stringency can result in different evolutionary outcomes. For instance, low-stringency, high-mutagenesis condition provides increased diversity but lower target activity.

3. Directed Evolution of Zinc Finger Nucleases (ZFNs)

Zinc finger nucleases (ZFNs), one of the earliest tools for genome editing, are fusions of zinc-containing protein motifs called zinc finger proteins and a non-specific endonuclease domain FokI that could specifically recognize three base pairs in DNA sequences and induce a double-stranded break.[12,13] By assembling different combinations of zinc finger proteins with FokI, different DNA sequences could be targeted by ZFNs.[14–16] However, not all DNA sequences can be targeted by ZFNs as naturally existing zinc fingers can only recognize a fraction of all 64 possible three-base combinations. Furthermore, off-target recognition by ZFNs can drastically reduce the efficiency and specificity of their targeting abilities. Directed evolution approaches have been applied to improve the ZFNs binding and cleavage efficiency and specificity, as well as expand the scope of DNA sequences that can be targeted (Figure 2a).

Figure 2.

Directed evolution of ZFN and TALE. (a) Diagram of ZFN (left) and TALE (right) components for mutagenesis. (b) Phage display of evolved ZFN fused to N-terminal of phage coat protein pIII. (c) Selection for ZFN cleaving a single-strand annealing reporter containing an insert with ZFN binding sites, thereby restoring the selectable marker for cell survival. (d) Selection for TALE in fusion with recombinase by removing an insert (e.g. fluorescent protein) and restoring the selection marker (e.g. antibiotic resistance gene) for cell survival. (e) One-hybrid system for selection of TALE recognizing alternative target sequences.

Pabo and coworkers first applied a phage display system to evolve ZFNs (Figure 2b). [17] They fused three zinc fingers of the Zif268 protein to the N-terminal end of the phage coat protein pIII, with four amino acid positions (1, 2, 3, and 6) of the first finger randomized. After affinity selection, they identified that three unique peptide sequences in this library showed high binding affinities to the cognate DNA sequences: peptide sequence DSNR strongly binds to the DNA sequence GACC, and peptide sequences RADR and QGSR strongly binds to the DNA sequence GCAC. In comparison with the wide-type zinc finger proteins, the evolved variants have higher specificity and binding affinity. Similar phage display evolutions were carried out by Wells and Kim, and variant fingers with polar residues or high affinity to G-rich sequences were discovered.[18,19]

To improve its cleavage efficiency, the FokI domain was subjected to directed evolution using Sharkey, a bacterial two-plasmid nuclease selection system developed by Barbas et al.[20] Two mutations, S418P and K441E, were discovered to enable a 15-fold increase in cleavage efficiency compared with the wild-type. A unique yeast three-plasmid nuclease selection strategy was later reported by Doyon et al., which incorporates a library plasmid and two independent single-strand annealing reporter plasmids that each contain an insert with a ZFN recognition site (Figure 2c).[21] The cleavage of the inserts by the evolving ZFN could restore the PHO5 and MEL1 genes essential for cell survival. Using this selection system, the researchers successfully generated temperature-sensitive ZFN variants. Compared to the ZFNs evolved by Sharkey, these temperature-sensitive ZFNs demonstrated enhanced activities in primary cells (PBMCs). Rather than evolving individual zinc fingers or FokI domains, a bacterial two-plasmid nuclease selection approach was adapted from the Sharkey system to evolve ZFNs with non-canonical architectures and altered linker structures for high-precision genome editing.[22**] In this work, a randomized DNA library encoding the linkers between the zinc finger and the FokI domains or between different fingers were inserted into the ZFN encoding gene. The evolved ZFNs with new linkers or new architectures demonstrated impressive new functions such as improved editing specificity, base skipping between adjacent zinc fingers, and targeting DNA sequences on the same strand. These new functions dramatically increased the scope of gene loci that can be targeted by ZFNs.

4. Directed Evolution of Transcription Activator-Like Effectors (TALEs)

Transcription activator-like effectors (TALEs) are DNA-binding proteins composed of highly conserved sequence-specific DNA-binding repeats originally identified in bacterial plant pathogens. Each DNA-binding repeat consists of 34 amino acids with repeat variable diresidues (RVDs) at positions 12 and 13, which determines the binding specificity on a single-nucleotide level. (Figure 2a).[23] Different from the ZFNs, the reproducible single-base recognition by a TALE repeating domain makes the rational design of a TALE array for binding any arbitrary DNA sequence simple and straightforward.[24,25] However, thymine is required at the 5′-nucleotide of the target sequence, restricting the sequence scope TALEs can target.

Directed evolution has offered important strategies to remove the 5’-thymine constraint in TALEs. Barbas and coworkers developed a structure-guided library to evolve the N-terminal domain of TALE to ease the 5’-end restriction of the target sequence.[26] The selection was performed using a TALE recombinase (TALE-R) screening system where a GFP gene is flanked by recombination sites and inserted into the gene encoding TEM-1 β-lactamase. (Figure 2d) Active TALE variants resulted in more frequent recombination and restoration of cell resistance to ampicillin. From this selection, TALE variants with a G-selective N-terminal domain were successfully isolated. Similarly, Sun and Zhao evolved SunnyTALEN, a high-efficiency TALE nuclease variant using a TALE-mediated homologous recombination reporter.[27] Besides recombinase-based selection approaches, Imanishi et al. successfully evolved TALE variants with an eased the 5’-T restriction using a bacterial one-hybrid screening system.[28] An important limitation of these early approaches is that they all used targeted libraries that constrain or bias mutation sites. Consequently, the evolved variants only showed modestly improved activities. Hubbard et al. applied PACE to access a large repertoire of variants and achieve the rapid evolution of genome editors. To evolve TALEs using PACE, they adapted the bacterial one-hybrid system to link a TALE DNA-binding domain to a subunit of bacterial RNA polymerase III, which controls the expression of protein pIII essential for phage propagation (Figure 2e).[29] An important advantage of this system is that by linking 5’-thymine recognizing TALE variants to a negative pIII that inhibits propagation, the promiscuous mutations leading to 5’-thymine recognition were drastically suppressed. TALE variants with respective 5’-A, 5’-C, and 5’-G recognition were successfully evolved after only 48 hours using PACE, all of which showed at least five-fold higher activity than wild-type TALE.

5. Directed Evolution for the Expansion of CRISPR PAM Compatibilities

RNA-programmable CRISPR-associated (Cas) nuclease has emerged as a powerful tool for precise and efficient genome editing in living cells through its ability to generate a double-stranded break (DSB) at the desired target location. Cellular repair mechanisms, including non-homologous end joining (NHEJ) and homology-directed repair (HDR), allow for gene editing through insertions, deletions, translocations, or other DNA arrangements at the site of DSBs. However, DNA recognition and cleavage of CRISPR strictly require a protospacer adjacent motif (PAM) 5’-NGG-3’ (where N can be any nucleobase) in the target DNA. This necessity of PAM greatly constrains the targeting scope and editing flexibility. One potential solution to address the targeting scope limitations is to engineer Cas9 variants with novel PAM specificities. A previous attempt to engineer Streptococcus pyogenes Cas9 (SpCas9) variants (R1333Q/R1335Q) with altered PAM specificities by modifying the glutamine residues in the PAM-interacting (PI) domain failed to recognize the expected NAA PAM in vitro.[30] Another rationally engineered SpCas9 variant (SpCas9-NG) [31] did recognize a more relaxed NG PAM, but its cleavage activity was lower than the wild-type SpCas9 at NGG sites. These results suggest that reprogramming PAM specificity might require more extensive remodeling of the PI domain.

Coupling structure-guided design and directed evolution using a bacterial two-plasmid nuclease selection system, Kleinstiver et al. [32] evolved several SpCas9 variants with altered PAM profiles that recognize NGA, NGAG, and NGCG. The PI domain of SpCas9 was mutagenized using error-prone PCR, and functional SpCas9 variants were subjected to cell survival by linking PAM recognition with DNA cleavage of an inducible lethal ccdB reporter (Figure 3a). The PAM preferences of these variants were profiled by a site-depletion assay using a construct consisting of six randomized base pairs adjacent to a protospacer and an antibiotic resistance reporter. The DNA cleavage at the PAM site resulted in the inactivation of the antibiotic resistance, such that the cell death was linked to undesired PAM recognition. The survived or uncleaved population was sequenced, and depletion scores for any given PAM were calculated by taking the ratio of the frequency of that PAM before selection to its frequency post-selection and then multiplying this ratio by an arbitrary scaling factor. Hence, the PAM preferences of a particular variant were determined with its lowest depletion score. They further applied this approach to partially relax the most strictly specified base in the NNGRRT PAM of Stapholococcus aureus (SaCas9) to NNNRRT.[33] This platform provides an alternative solution to the challenge of evolving separate variants for each PAM sequence, especially Cas9 orthologs with longer PAMs.

Figure 3.

Evolution of CRISPR nucleases and base editors. (a) Selection of mutant Cas9 recognizing an alternative PAM sequence and cleaving the reporter containing a ccdB cytotoxic gene. (b) Selection for ω-dSpCas9 in PACE, where recognition of PAM and target site by active variant triggers reporter gene expression and phage propagation. dSpCas9: original selection scheme. Split-intein dSpCas9: relies on fusion of intein-dSpCas9 halves for dSpCas9. Two protospacers: ω-dSpCas9 binds two distinct protospacer-PAM sequences drives the expression of either half of split-intein phage coat protein pIII. Split-intein dSpCas9 + two protospacers: combination of selection strategies with gVI as an additional selection marker. (c) Codon-switch selection for adenine base editors correcting A·T to G·C base pair for a functional antibiotic resistance reporter. (d) Selection for adenine base editor in PACE. Conversion of A·T to G·C base pairs removes two stop codons and allows RNAP to be expressed in full-length. (e) Selection for cytosine base editor in PACE. Conversion of C·G to T·A base pairs in the coding strand results with a stop codon between RNA polymerase and a proteolytic degradation tag (degron) to avoid degradation of RNAP, which is essential for gene III expression and phage propagation.

PACE has been used to rapidly generate Cas9 variants with expanded PAM compatibility as the linkage between the protein-of-interest and phage propagation during PACE can be easily adapted for PAM expansion. Hu et al. combined PACE and bacterial one-hybrid selection for a library of all 64 possible NNN PAM sequences, where a catalytically dead SpCas9 (dSpCas9) was fused to the ω subunit of bacterial RNA polymerase (RNAP) and bound onto a target protospacer-PAM sequence (Figure 3b). [34**] This binding interaction then recruited RNAP to express gene III for phage propagation, such that only the phages carrying the Cas9 variants that recognize the desired PAM sequences could replicate during PACE. This selection led to xCas9s that exhibit much greater specificity than SpCas9 and recognize a broad range of PAMs including NG, GAA, and GAT.

Recently, PACE has been developed to further evolve xCas9 binding to all 12 possible NAH (NAA, NAT, and NAC) PAMs by Miller et al. [35**] Compared to the previous methods, mutagenesis in this evolution strategy included the whole sequence of SpCas9 or DNA shuffling of the C-terminal segment, which resulted in mutations outside the PI domain. Therefore, after the recovery of the weakly active variants, they were subjected to the bacterial two-plasmid nuclease system previously used by Kleinstiver et al. [32] to remove variants with mutations in the nuclease domain and showing impaired DNA cleavage activities. Despite acquiring multiple mutations in the PI domain, the evolved variants did not exhibit activity when converted to base editors and tested on target PAM sites in human cells. Therefore, three new strategies of PACE were used to substantially increase selection stringency that might facility with enhanced PAM-binding activity (Figure 3b). In the first strategy, the evolving variants were bound to two distinct protospacer-PAM sequences to drive the expression of both split-intein pIII halves. In the second strategy, dSpCas9 was split in half with the C-terminal segment in the selection phage and the N-terminal segment in restricted amounts in E.coli. This strategy reduced saturation of binding to the desired PAM sited by variants with modest affinity because the amount of functional dSpCas9 was limited. In the third strategy, the split-dSpCas9 variants required two distinct protospacers to express gIII and gVI, a second phage selection marker. These directed-evolution strategies yielded three SpCas9 variants that recognize NRRH, NRTH, and NRCH PAMs. Moreover, these variants have higher activity than xCas9-3.7 and collectively can be used to target almost any NR PAM.

6. Directed Evolution of Base Editors

Although NHEJ and HDR allow precise knock-in and knock-out genes, both methods are inefficient for introducing mutations or editing sequences at the single-nucleotide level. To overcome these limitations, cytosine base editors (CBE) and adenine base editors (ABE) have been developed, and they enable a direct, irreversible, and programmable conversion of one base pair to another at the target genomic locus without the requirement of DSBs or donor DNA templates.[6,36] These base editors are comprised of a base-editing enzyme (E.coli TadA for ABE and rAPOBEC1 for CBE) fused to a nickase Cas9 (Cas9 with a D10A mutation to only enable a single-strand nick on the double-stranded DNA) and a uracil glycosylase inhibitor in CBE.

The first ABE was evolved from E.coli TadA through the codon-switch selection in E.coli that linked ABE activity with cell survival. Random mutations were introduced in the adenine deaminase portion of the ecTadA-dCas9 construct, and variants were selected by editing the H193Y point mutation in a defective chloramphenicol resistance gene to rescue antibiotic resistance (Figure 3c).[6] The evolved ABE was capable of recognizing DNA as the substrate and converting an A:T base pair to a G:C base pair through the deamination of adenosine to inosine, which was then read or replicated as a G and paired with a C. An advantage of this selection system is that the stringency can be easily tuned by requiring ABE to induce more than one base pair reversion at the target site to survive the antibiotic selection. Gaudelli et al. further evolved the ABE by introducing a synthetic library of TadA domains with all 20 canonical amino acid substitutions at each position of TadA.[37*] The authors obtained highly specific ABE8s capable of recognizing non-NGG PAMs and enhanced editing activity at both canonical and non-canonical positions.

PACE was used to expand the compatibility of ABEs with Cas homologs [38] and increase the editing efficiencies. Mutagenesis was focused on the TadA domain, and phage propagation was mediated by editing stop codons of a T7 RNA polymerase to rescue its function and drive gene III expression by a T7 promoter (Figure 3d). An advantage of this method is tunable stringency through slight modifications on the promoters’ strength of each component. The evolved ABE8e showed enhanced activity and compatibility with SaCas9, SaCas9-KKH, SpCas9-NG, circularly permuted CP-Cas9, and more impressively, with LbCas12a and enAsCas12 that previously showed virtually no activity.

On the other hand, CBE enables the deamination of cytidine to uracil.[36] The resulting U:G mismatch is permanently converted to a T:A base pair by cellular replication or repair machinery. One factor that greatly limits CBE’s editing efficiency is the native sequence context preference of cytidine deaminase APOBEC1 to edit GC motifs in the center of the editing window. Two CBEs (evoAPOBEC1-BE4max and EvoFERNY-BE4max) with no apparent bias for GC targets were evolved using PACE through a novel BE-PACE split-intein system.[39*] In this work, random mutagenesis was applied to the cytidine deaminase fused to a Npu N-intein in the selection phage, whereas Cas9 was fused to a Npu C-intein on a host plasmid. The full-length CBE was generated by intein trans-splicing upon phage infection. Then, precise cytidine base conversion of either of the cytosine bases in the ACC sequence of the coding strand resulted in a stop codon substitution on the linker between a T7 RNA polymerase and a proteolytic degradation tag (Figure 3e). The expression and activation of T7 RNAP then drove gene III expression under the control of a T7 promoter to allow the enrichment of library members encoding active CBE variants.

7. Conclusion and Future Perspectives

In conclusion, directed evolution techniques have been extensively used to enhance the functions of genome editing systems such as ZFNs, TALEs, CRISPR-Cas nuclease, and base editors. This field has experienced particularly rapid progress in the recent years driven by the landmark discovery of the CRISPR-Cas nuclease and base editors and their application to edit the bacterial and mammalian genome.[6,36,40–42] Collectively, these advances represent a new paradigm for basic research as well as human medicine. Medical applications of genome editing systems for pathogen detection, cell and tissue engineering, and genetic disease treatments have already been enrolled in clinical trials, with some starting to arrive in the clinics.[43–45] The rapid progress of these technologies from laboratories to clinics highlights the urgent need to improve the scope, efficiency, fidelity, delivery, and safety of the existing genome editors. While directed evolution and rational engineering have shown great promises in recent examples, [46,47*] continuous efforts are required to further improve genome editors before it can be regularly used in human medicine.

Therefore, the pursuit of the next-generation genome editors with enhanced functions creates an ever-increasing demand for profound directed evolution techniques; however, the power of directed evolution is inherently defined by the diversity of the library and the throughput of the screening or selection process. While traditional whole-genome mutagenesis often results in toxicity and the emergence of “cheater” mutations, recent developments of gene-directed mutagenesis, such as MAGE, [48] Ortho-Rep, [49] CRISPR-X,[50] EvolvR,[51] etc., has allowed for precise mutagenesis focused on specific genes of interest to tolerate higher mutational density and suppress cheater mutations. Gene-directed mutagenesis technologies can be integrated with in vivo evolution systems, enabling for a more rapid evolution and the exploration of larger sequence space. Furthermore, fluorescence-activated cell sorting (FACS)-based high-throughput screening techniques[52] and virus-based in vivo evolution technologies [2,53–55]allowed for considerably faster evolution rates compared to traditional screening and selection strategies. Finally, the automation of directed evolution makes it possible to simultaneously evolve dozens of separate populations.[56] We anticipate that the implementation of these technologies to the directed evolution of genome editors will lead to new genome editors with unprecedented structures and capabilities.