Genome Editor-Directed in vivo Library Diversification

SUMMARY

The generation of a library of variant genes is a prerequisite of directed evolution, a powerful tool for biomolecular engineering. As the number of all possible sequences often far exceeds the diversity of a practical library, methods that allow efficient library diversification in living cells are essential for in vivo directed evolution technologies to effectively sample the sequence space and allow hits to emerge. While traditional whole-genome mutagenesis often results in toxicity and the emergence of “cheater” mutations, recent developments that exploit the targeting and editing abilities of genome editors to facilitate in vivo library diversification have allowed for precise mutagenesis focused on specific genes of interest, higher mutational density, and reduced the occurrence of cheater mutations. This Minireview summarizes recent advances in genome editor-directed in vivo library diversification and provides an outlook on their future applications in chemical biology.

Graphical abstract

Introduction

Laboratory directed evolution is a powerful strategy for engineering biopolymers, polygenic traits, and even whole organisms, endowing them with user-defined functions and enabling their evolution several orders of magnitude faster than that in nature. As a result, functional hits can emerge within a practical timeframe (e.g., days to months). Furthermore, the spectrum of new functions that can be evolved via directed evolution may far exceed intrinsic activities. Key to these favorable characteristics of directed evolution is the introduction of a large number of genetic variants by library diversification technologies, which dramatically reduce the time and improve the efficiency in exploring the sequence landscape. Traditionally, library diversification in living cells occurs randomly across the whole genome (Badran and Liu, 2015; Gillam et al., 2003). While unbiased whole-genome mutagenesis methods have successfully generated libraries in many notable directed evolution campaigns, they often result in toxicity to the host and thus need to be controlled at low levels, throttling the rate of evolution (Bratulic et al., 2017; Zhao et al., 2014). In addition, unintended mutagenesis occurring at genes other than the gene of interest increases the chance for “cheater” mutations, which do not lead to the desired function but allow the host to bypass the selection pressure.

The recent development of genome editing systems such as zinc finger nuclease (ZFN), transcription activator-like effector endonuclease (TALEN), and clustered regularly interspaced short palindromic repeats (CRISPR) are powerful tools for programmable and precise editing of the genome in living prokaryotic and eukaryotic cells. Thanks to their programmability, high editing efficiency and specificity, and importantly, their modular design, these genome editors provide new opportunities for targeted mutagenesis in vivo. In this Minireview, we survey recent advances in in vivo library diversification directed by various genome editing technologies, including programmable endonucleases (e.g., ZFNs, TALENs, and CRISPR), error-prone nucleic acid synthesis, base editors, transposases and retrotransposon elements, and recombinases (Table 1).

Table 1.

An overview of targeted mutagenesis technologies used for directed evolution studies

Technology	DNA-binding domain	Effector	Mutational window*	Mutation rate*	Host organism	Reference
CRISPR/Cas-mediated mutagenesis
CDE	gRNA/Cas9	NHEJ	Within coding sequence of SF3B1	Several SGR (SF3B1-GEX1A-Resistant) variants generated after GEX1A selective pressure	Plants	[5, 6]
CRISPRres	gRNA/Cas9 (or Cas12)	NHEJ	~ 20 bp	Several KPT-9274-resistant variants after KPT-9274 selective pressure	Mamm	[38, 39]
CasPER	gRNA/Cas9	Error-prone PCR and HDR	300-600 bp	Editing efficiency of 98-99%; All sequenced colonies showed at least one mutation within the targeted site	S. cer	[25]
Nucleic acid synthesis-mediated mutagenesis
EvolvR	gRNA	nCas9 and error-prone DNAP1 fusion	< 200 bp	~ 1 x 10−5 to 1 x 10−6 mutations per nucleotide per generation	E. coli, S. cer	[21, 50]
OrthoRep	Orthogonal plasmid/DNA polymerase pair	Error-prone TP-DNAP1	22 kb	~ 1 x 10−5 to 1 x 10−9 substitutions per base	S. cer	[42]
MutaT7	Phage derived promoter/RNA polymerase pair	Cytidine deaminase	Multi-kb	~ 6.7 x 10−6 substitutions per bp per generation	E. coli	[37]
TRACE	Phage derived promoter/RNA polymerase pair	Cytidine deaminase	> 2 kb	~ 0.5 x 10−3 to 4 x 10−3 substitutions per bp per generation	Mamm	[10]
Base editor-mediated mutagenesis
STEMEs	sgRNA library	Cytidine deaminase-adenosine deaminase fusion	Within carboxyltransferase domain of ACC (A: 56 amino acids; B: 400 amino acids)	A: Mutagenesis efficiency of 73.21%; B: Several haloxyfop-resistant variants generated after haloxyfop selective pressure	Plants	[29]
BEMGE	sgRNA library	Cytidine deaminase and adenosine deaminase base editors	Within open reading frame of OsALS1	Several bispyribac-sodium-resistant variants generated after bispyribac-sodium selective pressure	Plants	[28]
CRISPR-X	MS2-modified sgRNA	Activation induced cytidine deaminase	< 100 bp	~ 1 x 10−3 mutations per bp per generation in hotspots	Mamm	[22]
TAM	sgRNA	Activation induced cytidine deaminase	< 150 bp	~ 4 x 10−4 substitutions per bp per cell cycle	Mamm	[36]
Targeted transposase and retrotransposon element-mediated mutagenesis
ICE	Retrotransposons	Error-prone reverse transcriptase	5 kb/cargo	~ 1.5 x 10−4 substitutions per bp per transposon replication	S. cer	[11]
Retron	Bacterial retrons	Error-prone reverse transcriptase	Within ~ 30 bp	~ 1 x 10−6 substitutions per bp per generation	E. coli	[47]
CRISPEY	Retron-gRNA	Reverse transcriptase	A: Nonsense mutation reversion in ADE2; B: Insertion of 765-nt sequence in ADE1	Editing efficiency of A: 100%; B: 92%	S. cer	[45]
Recombinase-mediated mutagenesis
SCRaMbLE-in	Integrated recombination sites	Recombinase	Synthetic pathways of interest (e.g. synthetic chromosome II)	Relative integration rate of ~ 0.6% for Cre-loxP and Dre-rox and ~ 0.3% for VCre-Vlox	S. cer	[33]

^*Mutational window and mutation rate studies may not have been performed or explicitly indicated in the literature.

Programmable endonuclease-mediated mutagenesis

ZFNs, TALENs, and CRISPR-CRISPR-associated proteins (Cas) RNA-guided endonucleases are highly-programmable and site-specific endonucleases that induce double-stranded breaks (DSBs) at targeted genomic loci. This DNA damage activates cellular repair mechanisms, such as non-homologous end joining (NHEJ) and homology directed repair (HDR), often resulting in DNA insertions, deletions, and translocations (Figure 1A-D), which has been exploited for targeted genome editing.

Figure 1.

In vivo mutagenesis mediated by ZFN-, TALEN-, and CRISPR-Cas. Schematic diagrams of (A) ZFN, (B) TALEN, and (C) CRISPR-Cas systems for mutagenesis. (D) ZFN-, TALEN, and CRISPR-Cas systems can induce a double-stranded break (DSB) at the targeted gene to trigger non-homologous end joining (NHEJ) or homology-directed repair (HDR).

ZFN-mediated mutagenesis

ZFN is a fusion protein of DNA-binding, zinc-containing zinc finger domains that specifically recognize three base pairs (bp) in DNA sequences and a DNA-cleavage FokI non-specific endonuclease domain that cleaves the DNA sequence (Pabo et al., 2001) (Figure 1A). The FokI domain can induce a double-stranded break of the cognate DNA, which can be repaired by NHEJ or HDR to introduce random bases at the site of ZFN-induced breaks, resulting in gene-specific mutations (Smith, 2000) (Figure 1A; Figure 1D). Over the last several years, targeted mutagenesis by zinc finger nucleases has been applied in several model organisms, including plants, insects, zebrafish, and induced pluripotent human stem cells (Lloyd et al., 2005; Foley et al., 2009; Gersbach et al., 2014). For example, Lloyd et al. (2005) demonstrated ZFN-induced mutagenesis in the Arabidopsis genome by directing ZFNs to produce DSBs at its corresponding ZFN recognition sites for NHEJ-based repair, resulting in mutation frequencies of 0.2 mutations per target (Lloyd et al., 2005). When compared to homologous recombination which typically generates < 10⁻⁷ gene targeting events per cell, this ZFN-induced mutagenesis shows significantly higher mutation frequencies of 0.4 mutations per cell (Lloyd et al., 2005). Although targeted mutagenesis by engineered ZFNs have shown promising results, their complicated design, relatively low efficiency, and laborious experimental procedure still require further improvements. Specifically, the limited efficiency by engineered ZFNs is attributed to the ZFN’s binding affinity and specificity as three ZFNs are required to provide sufficient affinity, but not all always make an equal contribution (Carroll, 2011). Additionally, because of the different binding affinities between different ZFNs, some stronger binders may have affinity for related sequences (Carroll, 2011).

TALEN-mediated mutagenesis

Transcription activator-like effector (TALE) is a DNA-binding protein consisting of a repeated array of highly conserved DNA-binding domains in which each TALE repeat recognizes a single base pair. TALEN is a fusion of a TALE DNA recognition array with a non-specific endonuclease, such as FokI, which can induce a double-stranded break to be repaired by NHEJ or HDR pathways for the introduction of mutations (Figure 1B; Figure 1D). The ability to induce site-specific mutations in various organisms by TALEN has been well-studied (Joung et al., 2013; Li et al., 2011; Li et al., 2012; Carlson et al., 2012). For example, Bedell et al. (2012) showed how TALEN could be used to precisely modify the sequences in the zebrafish genome in combination with a short DNA oligodeoxynucleotide donor. Hockemeyer et al. (2011) demonstrated that TALEN can be used for efficient and precise genome engineering in human pluripotent cells. Zhang et al. (2015) used a TALEN-assisted multiplex editing (TAME) system for the creation of improved yeast phenotypes. Notably, TAME was designed based on the multiplex interactions of TALENs with DNA sequences between the TATA and GC boxes, identifying a total of 66 potential TALENs-mediated modification sites in the S. cerevisiae genome mainly located at promoter regions of 98 genes (Zhang et al., 2015). As a proof-of-concept study, a multiplex fluorescence-based phenotype library was constructed with TAME and screened based on fluorescent diversity, demonstrating that TAME successfully generates genetic diversity in a multiplexable manner (Zhang et al., 2015). This accelerated evolution platform was further subjected to improve the ethanol tolerance of several yeast strains (Zhang et al., 2015). Generally, TALEN-based systems are more straightforward to design and possess better programmability than ZFN-based systems; however, it is the CRISPR-Cas systems that have accelerated to the forefront of genome editing technologies due to their robust editing abilities and user-friendly programmability.

CRISPR-Cas-mediated mutagenesis

The discovery of CRISPR-Cas systems has revolutionized the field of synthetic biology, allowing for programmable and site-directed mutagenesis. CRISPR-Cas systems program a single-guide RNA (sgRNA) to direct the Cas nuclease and induce a DSB at a targeted location followed by a protospacer adjacent motif (PAM) sequence (Figure 1C, Figure 1D). As a proof-of-concept use of CRISPR-Cas systems for in vivo library diversification, Butt et al. (2019; 2020) developed a CRISPR-mediated directed evolution (CDE) system in plants. They synthesized a tiled sgRNA library targeting all possible PAM-adjacent sites of a core splicing factor in rice to generate DSBs and introduce mutations by the NHEJ repair pathway of CRISPR-Cas9 systems (Butt et al., 2019). Specifically, they directed mutations to the coding sequence of the splicing factor 3b subunit 1 (OsSF3B1) gene in Oryza sativa because it is conserved amongst eukaryotes and is a significant target of the herbicide GEX1A (Butt et al., 2019). After the variants were subjected to GEX1A selective pressure, several SF3B1-GEX1A-resistant mutants were identified that confer different levels of GEX1A resistance (Butt et al., 2019). Additionally, CRISPR-mediated NHEJ repair mechanism has also been applied for library diversification in mammalian cells (Neggers et al. 2018). This system, termed CRISPR-induced resistance in essential genes (CRISPRres), can efficiently generate SpCas9-mediated DSBs to be repaired by the NHEJ mechanism for the introduction and identification of mutants that confer drug resistance (Neggers et al. 2018). In one example, the authors constructed a lentiviral sgRNA tiling library that tiled the coding region of 75 target genes of antineoplastic drugs in HAP1 and CML K-562 cells expressing SpCas9 (Neggers et al. 2018). This sgRNA library was then transduced in the two cell lines and the cells were treated with the anticancer agent KPT-9274 as the selective pressure (Neggers et al. 2018). The drug-resistant colonies were harvested and their corresponding sgRNAs were sequenced which concluded nicotinamide phosphoribosyltransferase (NAMPT) as the primary target of KPT-9274 (Neggers et al. 2018). The duality of CRISPRres to induce mutations in targeted genes and to identify molecular targets of inhibitors through loss-of-function screens denotes its significance in pharmaceutical and clinical research (Neggers et al. 2018). Since the canonical SpCas9 mutagenesis approach is limited by the availability of nearby NGG PAMs, the authors further tested whether CRISPRres is orthogonal in other endonucleases (Neggers et al. 2018; Neggers et al., 2021). Correspondingly, CRISPRres was determined to be compatible with the class 2 type V CRISPR-Cas AsCas12a, which recognizes TTTV PAMs, and an engineered AsCas12a variant (enAsCas12a), which substantially broadened the targeting range (Neggers et al., 2021). By expanding the CRISPRres platform to other CRISPR-Cas endonucleases, the resolution of this platform increased and allowed for the identification of novel drug-resistant mutations in genome regions that were previously inaccessible (Neggers et al. 2018; Neggers et al., 2021). In particular, the enAsCas12a-mediated DSBs discovered new drug resistance mutations for NAMPT to KPT-9274 and kinesin-5 to antineoplastic agent ispinesib in genomic loci that was undetectable with the canonical SpCas9 mutagenesis approach, demonstrating the importance of increased PAM recognition for increased activity and resolution in CRISPR-Cas-based library diversification platforms (Neggers et al., 2021). Similar approaches to employ the NHEJ repair pathway of CRISPR-Cas9 systems for generation and selection of drug-resistant variants in vivo were also confirmed by Isparo et al. (2017) and Donovan et al. (2017). Although harnessing the NHEJ mechanisms allows for CRISPR-Cas-mediated directed evolution systems to identify and screen for drug-resistant variants, the mutations generated are largely frameshifts that may impede protein synthesis and result in semi- or non-functional proteins.

Researchers have also explored the HDR repair pathway of CRISPR-Cas9 systems for mutagenesis. Specifically, Jakočiūnas et al. (2018) devised Cas9-mediated protein evolution reaction (CasPER) technology, which uses error-prone PCR (epPCR) to mutagenize the DNA donor templates for subsequent genome integration at the sites of CRISPR-Cas9-mediated DSB. Importantly, this technology can efficiently mutagenize and integrate DNA fragments of 300-and 600-bp in a multiplexable manner with integration efficiencies of 98-99% in yeast (Jakočiūnas et al., 2018). Using the HDR-proficient organism S. cerevisiae, CasPER technology mutated the catalytic domains of ERG12 and ERG20 in the mevalonate pathway and identified mutants that enhanced carotenoid production by 11-fold (Jakočiūnas et al., 2018). Although the multiplexable approaches of CasPER achieve high integration efficiencies in larger genomic regions in yeast, the HDR repair pathway is generally known to have lower efficiency and off-target effects in mammalian cells (Jakočiūnas et al., 2018). The low editing efficiency, exacerbated by the laborious phenotypical selection or screening for mutant identification, often makes the generation of a high-throughput in vivo protein library technically challenging in mammalian cells. To circumvent this challenge, Erdogan et al. (2020) directed CRISPR-Cas9-mediated cellular repair in situ to a gene encoding for a protein of interest to generate and express diversified protein libraries in mammalian cells in a single-variant-per-cell manner. Further, this approach eases subsequent phenotypic selection since the selection of library members is performed in appropriate physiological contexts while demonstrating strong genotype-phenotype coupling (Erdogan et al., 2020). This platform expands the feasibility of creating diverse protein libraries in mammalian cells using the NHEJ and HDR-repair mechanisms of CRISPR-Cas9 systems to advance protein evolution studies (Erdogan et al., 2020).

Mutagenesis by nucleic acid synthesis

Programmable endonuclease-based technologies for targeted mutagenesis had surged given its intrinsic targeting abilities and promising editing capacity. However, the narrow editing window often imposes an important constraint for the these “cut-and-repair” mutagenesis approaches; thus, researchers have exploited error-prone nucleic acid synthesis to extend the mutational window. As a pioneering strategy to in vivo library diversification by nucleic acid synthesis, Multiplex Automated Genome Engineering (MAGE) is a recombination-mediated genome engineering platform that uses single-stranded DNA (ssDNA) to introduce mismatch, insertion, and deletion modifications across the E. coli genome, repeatedly (Wang et al., 2009). Using a modified mismatch repair deficient E. coli strain, the authors programmed the bacteriophage λ-Red ssDNA-binding protein to direct and incorporate synthetic ssDNA oligonucleotides carrying intended mutations into the lagging strand during DNA replication (Wang et al., 2009). Importantly, these ssDNA oligonucleotides have homologous regions to the incorporation site for targeted mutagenesis (Wang et al., 2009). Not only has this accelerated evolution platform transformed the field of genome engineering, but it has also inspired further development of genome editors with more precise targeting abilities and higher editing efficiencies.

Researchers have explored CRISPR-guided error-prone polymerases to expand the editing window length. Developed by Halperin et al. (2018), EvolvR is an important example that utilizes the innate processivity and fidelity of polymerases for targeted mutagenesis by fusing nCas9 to an error-prone DNA polymerase (DNAP). Error-prone DNA polymerases have previously been employed in directed evolution approaches (Camps et al., 2003); however, EvolvR cleverly uses a CRISPR-guided nickase to nick the targeted gene, allowing the DNA polymerase to execute an error-prone nick translation for the continuous diversification of all nucleotides within a user-defined and programmable window length (Halperin et al., 2018) (Figure 2A). As a proof-of-concept experiment, EvolvR was used to target the E. coli rspE gene within the 30S ribosome to evolve resistance against the antibiotic spectinomycin (Halperin et al., 2018). In its initial version, EvolvR is limited by a low mutation rate (~2.5 x 10⁻⁶ mutations per nucleotide per generation) and narrow window length (~17 nucleotides from the nick site), but its modularity allows for further improvements to these mutagenesis elements thereby achieving a mutation rate of 10⁻⁵ to 10⁻⁶ mutations per nucleotide per generation in a 56-bp editing window using a version of EvolvR that incorporates a bacteriophage thioredoxin-binding domain to enhance the processivity of the polymerase (Halperin et al., 2018). Recently, this technology was adapted in S. cerevisiae to implement EvolvR-mediated nucleotide diversification in eukaryotes (Tou et al., 2020). Using a dual gRNA, yEvolvR can diversify endogenous genes in a multiplexable manner as it was able to simultaneously revert a nonsense mutation in ura3* and disable CAN1 in S. cerevisiae (Tou et al., 2020). Although engineering low fidelity polymerases with higher processivities will further enhance the construct, yEvolvR induced mutations in the targeted CAN1 region within a mutagenesis window length of ~40 bp (Tou et al., 2020). Expanding the applicability of this technology, Long et al. (2020) combined EvolvR with rare codon screening to evolve ornithine aminotransferases from Pseudomonas putida (Ppocd). One ornithine cyclodeaminase (OCD) mutant demonstrated a 2.85-fold increase in catalytic activity for the production of L-proline, exhibiting the advanced potential of EvolvR technology in amino acid biosynthesis (Long et al., 2020).

Figure 2.

Polymerase-, deaminase-, and non-nuclease genome editor-mediated mutagenesis. Schematic diagrams of the aforementioned technologies that employ polymerases for targeted mutagenesis: (A) EvolvR (Halperin et al., 2018), (B) OrthoRep (Ravikumar et al., 2018), and (C) MutaT7 (Moore et al., 2018) and TRACE (Chen et al., 2020) technology. Schematic diagrams of the aforementioned technologies that use CRISPR-guided deaminases for targeted mutagenesis: (D) STEME (Li et al., 2020), (E) CRISPR-X (Hess et al., 2016), and (F) TAM (Ma et al., 2016) technology. (G) General mechanism of retrotransposable elements for mutagenesis, as demonstrated by retrons (Simon et al., 2018). (H) General mechanism of recombinases for mutagenesis, as demonstrated by SCRaMbLE-in technology (Liu et al., 2018).

Utilizing the mutagenesis abilities of error-prone DNA polymerases, Ravikumar et al. (2018) developed the orthogonal DNAP-DNA plasmid pair, coined OrthoRep, in S. cerevisiae for rapid in vivo continuous evolution (Figure 2B). The OrthoRep technology is based on the pGKL1/2 plasmid system, which uses a specialized DNAP that is fused with terminal proteins (TPs) attached to pGKL1 and 2 for TP-primed DNA replication (Ravikumar et al., 2018). As a proof-of-principle experiment, the Plasmodium falciparum dihydrofolate reductase (PfDHFRs) was encoded on OrthoRep to evolve and identify mutations with resistance to the antimalarial drug pyrimethamine (Ravikumar et al., 2018). One limitation of the OrthoRep technology is that the user-defined genes to evolve need to include upstream control regions (UCRs), or 100-bp stretches upstream endogenous ORFs on unmodified p1 and p2 plasmids, which demonstrate relatively low promoter strengths, reducing the expression levels of OrthoRep-encoded genes (Ravikumar et al., 2018; Zhong et al. 2018). Despite this limitation, OrthoRep has initially demonstrated an impressive mutagenesis rate and scalability (between ~10⁻⁹ to ~10⁻⁵ substitutions per base) of user-defined genes in vivo and has continued to optimize its expression level of OrthoRep-encoded genes by experimenting with different promoter mutants and incorporating poly(A)tails to OrthoRep transcripts. (Ravikumar et al., 2018; Zhong et al. 2018). Furthermore, the continuous mutagenesis capabilities of OrthoRep have been recently coupled with the programmable and automated selection platform, eVOLVER, to develop an automated continuous evolution system in yeast to evolve drug-resistance PfDHFRs and Thermotoga maritima HisA (TmHisA) enzyme variants for subsequent protein and enzyme evolution studies (Zhong et al., 2020). Collectively, the OrthoRep technology has illustrated its ability to elucidate drug resistances and contribute to fitness landscapes in in vivo directed evolution studies (Ravikumar et al., 2018; Zhong et al. 2018; Zhong et al. 2020).

Expanding the abilities of polymerases in targeted genome editing, RNA polymerases have also been used to direct editing machinery to targeted genomic loci as RNA polymerases demonstrate high promoter specificities and processivities. Specifically, Moore et al. (2018) designed MutaT7 chimera by fusing a cytidine deaminase to T7 RNA polymerase (T7 RNAP) to allow for targeted mutagenesis downstream of a T7 promoter, given that the T7 promoter is not present elsewhere in the genome (Figure 2C). MutaT7 induced mutations in the ectopically expressed folA gene in E. coli, which conferred resistance to trimethoprim, demonstrating its applicability to in vivo directed evolution studies (Moore et al., 2018). Correspondingly, Chen et al. (2020) recently designed T7 polymerase-driven continuous editing (TRACE) to continuously diversify targeted genes in human cells by fusing a cytidine deaminase to T7 RNAP (Figure 2C). Not only was TRACE able to perform continuous mutations in targeted genes, but it was also able to identify MEK1 inhibitor-resistance mutations when subjected to selumetinib and trametinib treatment (Chen et al., 2020). Although MutaT7 and TRACE technologies require the insertion of a T7 promoter, its high editing rate and large window length are promising for more dynamic control to guide directed evolution compared to other genome editors (Moore et al., 2018; Chen et al., 2020).

Base editor-mediated mutagenesis

Recently developed by Liu and coworkers, base editors are fusions of nCas9 and a cytidine/adenine deaminase that can execute site-specific editing of nucleobases at the targeted gene to perform all four possible transition mutations (Gaudelli et al., 2017; Komor et al., 2016). CRISPR-based cytosine base editors (CBEs) mediate C:G>T:A substitutions by converting cytosine to uracil, which then converts to thymine through DNA replication or repair. Alternatively, CRISPR-based adenine base editors (ABEs) mediate A:T>G:C substitutions by converting adenine to inosine which is read as guanosine by polymerases. Capitalizing on the programmability and precise editing of the base editors, Li et al. (2020) developed engineered saturated targeted endogenous mutagenesis editors (STEMEs) to enable directed evolution in plants. As previous studies show that mutations introduced by CRISPR-Cas9 systems are typically indels, which likely cause nonfunctional frame-shift mutations, STEMEs can efficiently generate base substitutions (Li et al., 2020). Their construct involves bi-functional base editor machinery, created by fusing engineered plant-based cytidine deaminase and adenosine deaminase with nCas9 and uracil DNA glycosylase inhibitor (UGI), such that a single sgRNA can be employed to simultaneously achieve C:G>T:A and A:T>G:C substitutions in the same target sequence (Li et al., 2020) (Figure 2D). As a proof-of-concept use of STEMEs, they employed STEME-1 and STEME-NG to directly evolve the carboxyltransferase (CT) domain of acetyl-coenzyme A carboxylase (OsACC) gene in rice against the ACC-inhibiting herbicide haloxyfop (Li et al., 2020). Their base editor-mediated directed evolution platform revealed novel mutations in OsACC that confer haloxyfop resistance, which denotes its significance in protein evolution studies (Li et al., 2020). Similarly, Kuang et al. (2020) developed a base editor-mediated gene evolution (BEMGE) method which employs cytidine and adenosine base editors directed by sgRNAs to evolve the OsALS1 gene in rice against the herbicide biphyribac-sodium (BS). Like STEMEs, this BEMGE method can perform saturated mutagenesis in planta to evolve resistance against herbicides for crop improvement and agricultural advancement (Kuang et al., 2020).

CRISPR-directed base editors have also been used for targeted diversification in mammalian cells. Inspired by nature’s built-in diversification mechanism of somatic hypermutation, Hess et al. (2016) developed CRISPR-X, which uses an MS2-modified sgRNA-guided dCas9 to recruit hyperactive activation-induced cytidine deaminase (AID) fused to MS2 binding protein (MCP) to induce localized and diverse point mutations at a targeted region (Figure 2E). With CRISPR-X, they were able to mutagenize a core subunit of 20S proteasome PSMB5 to evolve resistance against the proteasome inhibitor bortezomib (Hess et al., 2016). By identifying both known and novel bortezomib-resistant mutations, CRISPR-X technology shows significant potential to in vivo directed evolution studies (Hess et al., 2016). Correspondingly, Devilder et al. (2019) used CRISPR-X to perform in cellulo affinity maturation to generate novel monoclonal antibodies against HLA-derived antigens from a human B cell repertoire. Similar to CRISPR-X technology, Ma et al. (2016) repurposed the somatic hypermutation machinery to perform targeted AID-mediated mutagenesis to induce cytidine deamination at targeted genome loci. Their construct consists of a sgRNA-guided dCas9-AIDx fusion protein comprising of an AID variant (P182X), which exhibited a stronger deaminase activity over the full-length AID protein, and UGI, which increased the mutational frequency at the sgRNA-targeted region (Ma et al., 2016) (Figure 2F). The authors were able to mutagenize the exon 6 of BCR-ACL oncogene in chronic myelogenous leukemia cells to identify known and novel ABL mutations that confer imatinib resistance (Ma et al., 2016). This work further demonstrated that base editors can efficiently diversify targeted genome loci to facilitate protein evolution in mammalian cells.

Mutagenesis mediated by targeted transposase and retrotransposon elements

Transposase-mediated mutagenesis

To expand mutagenesis approaches catalyzed by enzymes, researchers have also harnessed the innate mobility of DNA transposases for genome editing, as these enzymes mediate DNA transposition to move DNA sequences from one location to another. Developed by Strecker et al. (2019), ShCAST technology relies on a CRISPR-associated transposase (CAST) to enable crRNA-guided Tn7-like transposition to insert DNA segments of 60 to 66 bp downstream the PAM recognition site throughout the E. coli genome. Although ShCAST does not provide a completely scarless integration, it does demonstrate unidirectional insertions in a narrow editing window while preventing repeated insertions at the targeted site (Strecker et al., 2019). Further, Saito et al. (2021) recently elucidated the homing mechanisms of CAST type V-K and I-B systems. From this study, it was discovered that type V-K CAST systems demonstrate RNA-mediated homing using a delocalized crRNA and type I-B CAST systems demonstrate protein-mediated homing using two target selector proteins, expanding the understanding of transposition mechanisms (Saito et al., 2021).

Retrotransposon-mediated mutagenesis

Another type of transposable elements are retrotransposons, which utilize a “copy-and-paste” mechanism to copy RNA transcripts and paste their DNA counterparts into various genome sites. Specifically, Crook et al. (2016) optimized yeast’s native retrotransposon Ty1 to enable in vivo continuous evolution (ICE) of essential enzymes, regulatory factors, and metabolic pathways. Once the targeted gene is cloned into the genome of an inducible Ty1 retrotransposon, Ty1-mediated error-prone transposition is induced. (Crook et al., 2016). Further, improved variants were selected and reintroduced into the ICE system for an in vivo continuous directed evolution (Crook et al., 2016). Although Ty1-based mutagenesis is specific to yeast, Simon et al. (2018) have developed retroelements from bacterial taxa. The retron cassette consists of a guiding sequence (msr), a targeting sequence (msd), and a reverse transcriptase (RT), which produces hybrid RNA-DNA molecules that can target and edit homologous DNA to match the reverse transcript and introduce mutations (Simon et al., 2018) (Figure 2G). To prove its applicability to directed evolution studies, a retron mutant induced mutations in an antibiotic resistant gene using a mutagenic T7 RNA polymerase to result in a per-base mutation rate of ~1 x 10⁻⁶ per generation (Simon et al., 2018). Combining the editing abilities of retrotransposable elements along with the targeting abilities of CRISPR systems, Sharon et al. (2018) developed the Cas9 Retron precISe Parallel Editing via homologY (CRISPEY) system, which utilizes bacterial retrons to generate single-stranded donor DNA templates within the nucleus to favor HDR editing outcomes. As a proof-of-principle experiment, CRISPEY was subjected to revert a nonsense mutation in the ADE2 gene in haploid S. cerevisiae, which resulted in 100% editing efficiency after 48 hours (Sharon et al., 2018). Further, the study of the fitness landscapes generated by CRISPEY libraries demonstrates that genotype-phenotype relationships can be determined at the single-nucleotide resolution for the genetic variation analyses, though the editing efficiency over multiple genome loci is reduced. Thus, Lim et al. (2020) developed the CRISPR/retron system for the multiplexed generation of substitution mutations across the E. coli genome. Although the whole-genome application of this CRISPR/retron system is beyond the scope of this review, this technology demonstrates the aptitude of exploiting retrotransposable elements for large-scale mutagenesis applications.

Recombinase-mediated mutagenesis

Another strategy to generate library diversification in vivo is through the recombinase technology. The recombinase-based system known as the Synthetic Chromosome Rearrangement and Modification by loxP-mediated Evolution (SCRaMbLE) was invented for the Sc2.0 project to introduce loxPsym sites throughout the yeast genome for an inducible Cre recombinase-based rearrangements (Dymond et al., 2011; Liu et al., 2018; Shen et al., 2016). Since the Sc2.0 project, SCRaMbLE has been advanced by Liu et al. (2018) to create SCRaMbLE-in, which has been used to randomly integrate engineered metabolic pathways into the synthetic yeast genome (Figure 2H). SCRaMbLE-in can also perform continuous, large-scale genome rearrangements in the host chassis by exploiting the ability of active recombinases to facilitate deletion, inversion, and duplication events in the host genome (Liu et al., 2018). The inducible control of SCRaMbLE-in technology on the synthetic yeast genome is attributed to the 267 loxP recognition sites on Chromosome II targeted by the Cre recombinase and the fusion of site-specific recombinases to ligand-binding domains (Liu et al., 2018). Combining metabolic pathway integration and chassis diversification in vivo, SCRaMbLE-in provides an efficient system for combinatorial library construction and long-term directed evolution studies, which can further be coupled with the MAGE for a high-throughput, site-directed mutagenesis (Liu et al., 2018; Wang et al. 2009).

Conclusion

Recent advances in genome editing technologies, in particular the CRISPR-Cas system, have enabled unprecedented capabilities for targeted library diversification in living cells. Targeted library diversification can be achieved by the strategic introduction of DSB at the target gene by endonucleases such as ZFN, TALEN, and CRISPR, which further activates the endogenous DNA repair mechanisms to insert random bases. Beyond cutting, technologies have also been developed to combine the programmable DNA binding of genome editors and effector recruitment for custom mutagenesis through error-prone nucleic acid synthesis, base deamination, transposon and retrotransposon, and recombination. While the state-of-the-art genome editor-directed in vivo library diversification methods have achieved a mutation frequency of ~0.001 mutations per bp per generation, which is well suited for the evolution of a diverse array of proteins in living cells, this mutational rate is not high enough for nucleic acid evolution (e.g., aptamers, ribozymes, and riboswitches) that typically requires the mutational rate of ~0.01 per bp per generation or higher. The recent development of new genome editing tools, such as the Prime editor and CRISPR-Cas9 circular permutants, provide exciting opportunities to further expand the toolbox for rapid in vivo mutagenesis (Anzalone et al., 2019; Oakes et al., 2019). Furthermore, the integration with autonomous in vivo evolution platforms will unleash the full power of the genome editor-directed library diversification technologies and enable the exploration of broader sequence space at an ever-faster rate (Esvelt et al. 2011; Ravikumar et al. 2018; Zhong et al., 2020).

Significance.

The emergence of genome editors as precise and programmable editing tools has had a tremendous impact in biology and medicine. As an emerging application, these genome editing systems have enabled more precise mutagenesis of genes of interest, along with higher mutational densities and reduced numbers of cheater mutations that evade selection pressures. Further, the implementation of genome editors to facilitate in vivo library diversification in directed evolution studies has identified drug-resistant mutants and novel genotype-phenotype linkages beyond prediction by rational design. In this Minireview, we survey recent advances in targeted in vivo library diversification mediated by programmable endonucleases (e.g., ZFN, TALEN, and CRISPR), error-prone nucleic acid synthesis, base editors, transposases and retrotransposon elements, and recombinases, and denote their important applications in chemical biology.