Fueling chromosomal gene diversification and artificial evolution with CRISPR

Abstract

Gene diversification is an effective approach to massively dissecting variant functions and evolving sequences when paired with an appropriate assay. In vitro mutagenesis and ectopic gene expression, however, fail to simulate the endogenous regulatory environment of the variants. The development of clustered, regularly interspaced short palindromic repeats (CRISPR) systems has greatly boosted the efficiency of targeted gene diversification in various species. Here, we review recent CRISPR-assisted methods for chromosomal gene diversification and artificial evolution, focusing on the advantages and limitations of each approach, and propose possible strategies to overcome current limitations and directions in future technology development.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03756-7.

Introduction

Gene diversification, the process of generating a large number of distinct genetic variants, facilitates the screening and analysis of functional sequences. In addition, iterative diversification and selection improves gene and protein functions for various biotechnological innovations. Error-prone PCR [1], site saturation mutagenesis [2, 3], and DNA shuffling [4, 5] are just a few methods in the well-developed in vitro gene diversification toolbox. A common limitation is that the generated mutant libraries need to be introduced into host organisms and expressed ectopically to conduct phenotypic screening, which is cumbersome and limited by host-dependent transformation efficiencies [6, 7]. In vivo continuous evolution diversifies genes expressed from artificial replication/transcription systems [8]. Although these in vivo orthogonal systems avoid iterative cloning and transformation/transfection steps, they still require the genes of interest to be expressed on plasmids or at a non-native genomic site. Such ectopic expression is often subject to the influences of plasmid instability and copy number effects, or of chromosomal position effects if integrated [9]. Therefore, these methods fail to recapitulate the endogenous regulatory environments, which can be essential for analyzing variant functions. In addition, ectopic expression systems are yet to be established in some species, such as plants. As a result, heterologous gene diversification methods have been historically applied to studying protein fitness landscapes and evolving individual proteins in more permissive model organisms. Their applications in functional genomics and organism engineering have thus far been limited. Lastly, large structural variants at the genome level, including large deletions, duplications, inversions, and translocations, have been impractical to engineer in vitro without costly whole genome synthesis [10].

Chromosomal gene diversification, which we refer to as the direct generation of genetic variants at the native chromosome position, permits reliable functional profiling in situ [11]. Traditionally, this has been attempted with spontaneous, chemical, or physical mutageneses. However, these approaches are often inefficient with low gene mutation frequency [12, 13]. In addition, such mutations are randomly distributed genome-wide, leaving the interested genomic region rarely perturbed. Oligonucleotide-mediated recombineering is an important progress towards chromosomal gene diversification with customized point mutations [14]. However, to date it is only successfully applied to bacterial species [15–18] and transgenes in yeast [19–21]. In recent years, traditional recombineering has been upgraded by using the clustered, regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated proteins (Cas) system to generate a targeted DNA double-strand break (DSB) to boost mutagenesis frequency. This has led to the development of homology-directed repair (HDR) based variant engineering methods in bacteria, yeast, and human cell lines. Other recombination-independent methods have also been developed to enable gene diversification in species where homologous recombination is inefficient. These methods include using a dead or nicking Cas protein with deaminases, reverse transcriptases, and error-prone DNA polymerases. With the support of CRISPR technologies, large-scale targeted chromosomal gene diversification has become available by simply generating single-guide RNA (sgRNA) libraries (Fig. 1), enabling the rapid interrogation of various cellular processes in multiple species and aiding in the development of useful products ranging from therapeutics to bio-based products. CRISPR-based methods, in combination with other genome editing agents such as recombinases and transposases, are also enabling us to engineer targeted large structural variants. In this review, we summarize the various methods of CRISPR-assisted chromosomal gene diversification, compare their advantages and limitations, and discuss emerging applications and future directions for technology improvement.

Fig. 1.

Workflow of CRISPR-assisted chromosomal gene diversification methods. sgRNAs and/or templates are designed as oligonucleotide libraries to cover the target chromosomal region. After cloning as a plasmid library, these reagents can be introduced into host organisms using transformation methods tailored to specific species. Under conditions of interest, specific phenotypes are enriched, allowing researchers to link these phenotypes to genotypes through sequencing

CRISPR systems facilitate programmable genomic targeting

CRISPR/Cas systems are bacterial innate immune systems capable of recognizing, binding, and cleaving foreign nucleic acids. Based on their effector complex architecture, CRISPR/Cas systems are classified into two major classes and six types. Class 1 systems employ multi-subunit effector complexes and include Type I (signature protein: Cas3), Type III (Cas10), and Type IV systems. Class 2 systems utilize single-protein effectors and include Type II (Cas9), Type V (Cas12), and Type VI (Cas13). Notably, Cas9 and Cas12 primarily target DNA, while Cas13 is specialized for RNA cleavage [22]. Since their discovery, various CRISPR/Cas systems have been engineered and reprogrammed as precision genome editing tools. The most widely used system to date is the Class II, type II CRISPR/Cas9 system from Streptococcus pyogenes [23]. In this system, the Cas9 protein is targeted by the sgRNA to the genomic locus referred to as a protospacer, whose sequence is adjacent to a protospacer adjacent motif (PAM) and matches the guide sequence of the sgRNA (Fig. 2a) [23]. It subsequently cleaves the target DNA to generate a DSB, which is repaired through non-homologous end-joining (NHEJ) or HDR pathways of the host cell [24]. dCas9 (dead Cas9) [25] and nCas9 (nickase Cas9) [26] are mutants derived from Cas9 by inactivating both nuclease domains (HNH and RuvC) in the case of dCas9 or a single nuclease domain (HNH or RuvC) in the case of nCas9. dCas9 lacks DNA cleavage activity but retains the ability to accurately target and bind to DNA, making it a recruiting platform for additional effector proteins [27]. nCas9 retains the ability to cleave one of the two DNA strands, which facilitates advanced genome editing techniques such as base editing and prime editing [28, 29].

Fig. 2.

CRISPR-stimulated homology-directed repair and its applications. a CRISPR-HDR involves Cas9 recognizing the PAM sequence, forming a R-loop, and cleaving the chromosomal gene to generate the DSB. Repair templates containing mutated sequences are subsequently integrated into the target site. SpCas9, the commonly used Cas9 from Streptococcus pyogenes. PID, PAM interaction domain. PAM, protospacer-adjacent motif. sgRNA, single guide RNA. b Saturation genome editing (SGE) in human cells involves the cutting of a specific genomic region (usually an exon around 100 bp in length) by one sgRNA, followed by the use of HDR to introduce SNVs. Multiple exons can be diversified with each SNV library targeting one exon in one experiment. High-throughput sequencing quantifies the frequency of SNVs before and after screen, allowing for the assessment of variant impacts based on SNV abundance changes. c CasPER (Cas9-mediated Protein Evolution Reaction) combines error-prone PCR with Cas9-mediated HDR for in situ integration of large diversified DNA fragments (300 ~ 600 bp) in yeast. CasPER enables multiplex engineering of metabolic pathway genes for identifying high-yield strains. d The guide + donor strategy pairs each sgRNA with a short mutagenic DNA donor for sequence diversification in E. coli and yeast. It accelerates laboratory evolution through a targeted and trackable approach. e In the CRAIDE system, an error-prone T7 RNA polymerase (epT7RNAP) continuously generates mutated chimeric donor gRNAs (cgRNAs). These cgRNAs are then copied into the target genomic site under the guidance of Cas9 via homologous recombination. This approach can be applied to evolve endogenous metabolic pathways

CRISPR-assisted gene diversification

Chromosomal gene diversification often necessitates the use of a library of genome editing agents with targeting specificities covering the regions of interest. Earlier generations of genome editing tools, i.e., zinc finger nucleases (ZFNs) [30] and transcription activator-like effector nucleases (TALENs) [31], suffer from the complexity of protein design, the high cost of molecular cloning, and cleavage-only effector function, preventing the facile generation of gene diversifying libraries. Consequently, TALEN libraries were mainly constructed for gene knockouts [32, 33]. And only until recently, the challenges of designing specific zinc finger arrays have been tackled using advanced machine learning models [34]. In contrast, high-efficiency, easy programmability, and cost-effectiveness drove the wide adoption of CRISPR/Cas systems in different species. Simply by constructing sgRNA libraries, the targeting range of CRISPR readily encompasses a full gene or long genomic regions (Fig. 1).

CRISPR-stimulated homology-directed repair

It is not surprising that HDR was first used to introduce a library of user-defined precision mutations due to decades of research on its highly conserved mechanisms and well-developed experimental protocols [35]. The efficiency of spontaneous homologous recombination in vivo is generally low [36]. A landmark study reported that an artificially introduced DSB greatly boosted the efficiency of targeted integration of foreign DNA carrying homologous sequences, through triggering the homologous recombination repair pathway [37]. This boosted efficiency permits the scaling up of genomic integration using massive in vitro synthesized HDR templates incorporating rationally designed or saturated mutations, following DNA cleavage by CRISPR/Cas9 (referred to in this review as the CRISPR-HDR strategy, Fig. 2a–d).

It was first demonstrated in human cell lines that chromosomal genes can be diversified using saturation editing, which involves a single genomic cut by a sgRNA and subsequent recombination with a library of in vitro synthesized mutagenic donor plasmids [38]. Since then, multiple studies have reported using CRISPR in combination with homologous recombination for saturation editing and functional profiling of exonic regions in various disease-associated genes (Fig. 2b) [38–45]. Unlike traditional methods, which are limited by the slow curation of clinical and population genomics data, CRISPR-HDR enables high-throughput generation and analysis of genetic variants, including those of uncertain significance. By introducing targeted mutations and tracking variant abundance before and after selection (e.g., drug treatments), researchers can systematically assess the functional impact of these variants more rapidly (Fig. 2b). However, due to the generally low efficiency of HDR in mammalian cells, HDR templates targeting each exon typically require homologous arms of 600 to 1000 nucleotides in length on both sides to support efficient repair [38–40, 45]. In yeast, an analogous method termed Cas9-mediated Protein Evolution Reaction (CasPER) was developed [46]. The CasPER method is a robust directed evolution technique that employs error-prone PCR (epPCR) to create large mutant libraries and integrates these variants into genomic contexts using Cas9-mediated genome editing (Fig. 2c). CasPER allows for the integration of donor sequences ranging from 300 to 600 bp with over 98% efficiency (Table 1). This method has proven useful for diversifying endogenous non-essential yeast genes, such as those of the mevalonate pathway, enabling the generation of diverse mutants and an 11-fold increase in isoprenoid production after selection.

Table 1.

Features of chromosomal gene diversification methods

Method		Species	Cell line/strain	Efficiency	Efficiency Note	Applications	Year	Reference
CRISPR-stimulated homology-directed repair	Saturation genome editing	Mammalian cells	HEK293T, HAP1	1.02%–3.33%	Range of reported efficiencies	Saturation editing of exon 18 of BRCA1 and a region of DBR1	2014	[38]
		Mammalian cells	HAP1	0.2–0.8	From Fig. 2a of [39]	Saturation editing of 13 exons of BRCA1	2018	[39]
		Mammalian cells	HAP1	Not reported	N/A	Saturation editing of all 17 exons of DDX3X	2023	[40]
		Mammalian cells	Humanized mESC	0.6%–2.3%	Percentage of HDR (figure S6 of [41])	Saturation editing of exon 13 of BRCA1	2023	[41]
		Mammalian cells	HAP1-A5	99%	From Fig. 2c of [42]	Saturation editing of BAP1	2024	[42]
		Mammalian cells	HAP1-A5	50%	From figure S1b of [43]	Saturation editing of all 9 exons of RAD51C	2024	[43]
		Mammalian cells	HAP1	> 0.3	HDR rate (from Extended Data Fig. 3a of [44])	Saturation editing of VHL	2024	[44]
		Mammalian cells	HAP1	Not reported	N/A	Saturation editing of exons 15 to 26 of BRCA2	2025	[45]
	CasPER	Saccharomyces cerevisiae	CEN.PK2-1C	98%–99%	Cas9-mediated integration of large mutagenized DNA fragments	Directed evolution of BTS1 and HMG2	2018	[46]
	CREATE	Escherichia coli	MG1655	75%	Predicted from single guide + donor efficiencies	Saturation mutagenesis of the folA gene; Screening drug-resistant mutations of the AcrB multidrug efflux pump; Laboratory evolution of E.coli thermotolerance; Selection of stress resistance to rifampicin, erythromycin, and furfural	2017	[47]
	CHAnGE	Saccharomyces cerevisiae	BY4741	47%	Average editing efficiency of guide + donor cassettes with a PAM–codon distance within 20 bp	Saturation editing of a region of SIZ1	2018	[49]
	MAGESTIC	Saccharomyces cerevisiae	DHY214	> 60%	Editing efficiency of ADE2 (from Fig. 3b of [50])	Saturation editing of a region of SEC14	2018	[50]
	Guide + donor strategy	Saccharomyces cerevisiae	BY4741	85%–95%	Library editing efficiency (from table S7 of [51])	20 bp tiling deletion of SGS1	2018	[51]
	CHASE	Saccharomyces cerevisiae	BY4741, ER	30%–40%	Library editing efficiency (from figure S6 of [53])	Saturation editing of a global transcription factor gene SPT15	2024	[53]
	CRAIDE	Saccharomyces cerevisiae	CEN.PK2-1C	3.26 × 10^-6	Per viable cell per generation per base	ADE2 disruption; Directed evolution of CAN1	2021	[56]
CRISPR-guided deaminases	TAM	Mammalian cells	K562	> 20%	Achieved with dCas9-AIDx and pooled sgRNAs against GFP	Identification of BCR-ABL imatinib resistance variants	2016	[67]
	CRISPR-X	Mammalian cells	K562	> 20%	Achieved with AID*∆ and the most efficient sgRNA	Evolution of GFP to EGFP; Identification of PSMB5 bortezomib resistance variants	2016	[68]
	STEMEs	Rice	Japonica	13.18%	Average editing efficiency in each group	Directed evolution of the OsACC herbicide resistance	2020	[69]
	BEMGE	Rice	Kitaake	15.2%–23.8%	Agrobacterium or particle bombardment	Directed evolution of the OsACC herbicide resistance	2020	[95]
	AGBE	Mammalian cells	HEK293	Varied	A-to-G and C-to-G (0.01%–32.46%), A-to-G and C-to-T (0.18%–23.75%), or A-to-G and C-to-A (0.01%–2.59%) conversions at the same DNA strand	Saturation mutagenesis of hDTR	2022	[76]
	MoBE	Rice	Japonica variety Nipponbare	37.60%	Highest efficiency	Directed evolution of OsACC herbicide resistance	2023	[77]
	BE screen	Mammalian cells	HAP1	< 25%	Substitution frequency (from supplementary Fig. 1b of [80])	Base editing of all exons in BRCA1	2020	[80]
		Mammalian cells	eHAP	Varied	AncBE4max: 9.6% ABEmax: 4.6% (median fraction of edited reads)	Base editing of all exons of BRCA1 and BRCA2	2021	[81]
		Mammalian cells	A375, MELJUSO, HAP1, HT29, OVCAR8, HA1E	46.6%	Median C-to-T editing in the target window in HAP1	Identification of known loss-of-function variants in BRCA1 and BRCA2; Identification of MCL1 and BCL2L1 drug-sensitivity or resistance mutations; Identification of PARP1 drug-sensitivity or resistance mutations; Functional screens of 52,034 variants in ClinVar	2021	[82]
		Mammalian cells	MDA-MB-231, PC9, PDECs, NIH3T3, KPT1	> 60%	Median C-to-T by FNLS at NGG PAMs	Identification of uncharacterized TP53 variants	2022	[83]
		Mammalian cells	HAP1, MELJUSO	Varied	Samples with an intermediate or active PAM averaged 41.7% C-to-T editing and 37.4% A-to-G editing in the predicted edit window of 4 − 8 nucleotides	Screening for loss-of-function mutations in BRCA1 and Venetoclax-resistant mutations in BCL2	2022	[84]
		Mammalian cells	HBEC30KT	Varied	C-to-T: 37%–61% A-to-G: 16%–68%	Evaluated the functional effects of 29,060 cancer-related transition mutations	2022	[85]
		Mammalian cells	K562	~ 60%	Median C-to-T efficiencies (from Extended Data Fig. 5d of [86])	In situ mutational scanning of DNMT3A	2022	[86]
		Mammalian cells	K562	Varied	C-to-T: averaging 34.1% A-to-G: averaging 31.6%	Saturated base editing screens of a GWAS locus	2023	[87]
		Mammalian cells	MCF10A	> 95%	For both CBE and ABE	Identification of EGFR pathogenic variants	2024	[89]
				0.5%–66%	Prime editing efficiencies from Fig. 4a of [89]
	CoMuTER	Saccharomyces cerevisiae	S288c	N/A	Average of 0.3 mutations per kilobase	Evolution of a heterologous lycopene synthesis pathway	2023	[99]
	HACE	Mammalian cells	HEK293FT	N/A	~ 2 per kbp per 3 days (from table S13 of [100])	Identification of MEK1 inhibitor–resistance mutations; Identification of SF3B1 variants; Mutagenesis of noncoding regulatory elements	2024	[100]
CRISPR coupled with reverse transcriptase	PLSM	Rice	Japonica variety Nipponbare	Not reported	N/A	Identification of OsACC1 herbicide resistance mutations	2021	[109]
	SPE	Mammalian cells	Haploidized HEK293T	9.95%–80.01%	Data from the 16 libraries targeting NPC1	Screening of NPC1 and BRCA2 variants	2022	[110]
		Mammalian cells	MCF7	20%–40%	At two loci (from Fig. 1D of [111])	Identification of essential nucleotides in the MYC enhancer; Functional characterization of breast cancer-associated variants; Evaluation of clinically identified variants	2023	[111]
		Mammalian cells	PC-9	0.0014%–0.36%	Per base, from Fig. 1d of [112]	Screening of drug-resistant SNVs in EGFR; Evaluation of SNVs in RPL15 and BRCA1	2024	[112]
		Mammalian cells	A549	~ 8%	Peaking average editing efficiency	Screening cancer-associated variants of TP53	2024	[113]
		Mammalian cells	Jurkat, THP-1	Varied	28%–68% individual editing rates; 39% total editing rate of the pool	Tiling mutagenesis screens of PPIF and the promoter of IL2RA	2025	[114]
CRISPR coupled with error-prone DNA polymerase	EvolvR	Escherichia coli	TG1	0.5%–1%	Substitution rates at a base	Directed evolution of rpsE and rpsL	2018	[116]
	yEvolvR	Saccharomyces cerevisiae	yCT01	10^-5	Mutations/nucleotide/generation	Diversification of URA3 and CAN1	2020	[117]

While the above studies demonstrated the feasibility of integrating donor libraries with a single sgRNA, the efficiency of mutant incorporation tends to drop at a distance from the Cas9 cleavage site. It is therefore advantageous to use nearby sgRNAs for each intended mutation. This approach is feasible in organisms with efficient homologous recombination systems, such as in certain engineered bacteria or yeast, where shorter homology arms (typically 40–100 bp) are sufficient and thus allow the massively parallel synthesis of the donor and a nearby guide (Fig. 2d). With this concept, a gene diversification technology named CRISPR-enabled trackable genome engineering (CREATE) was developed in E. coli [47]. CREATE integrates CRISPR/Cas9 with HDR to achieve saturation mutagenesis of chromosomal genes using libraries of guide and donor pairs (referred to as the guide + donor strategy hereafter). In this strategy, it is crucial that the donor is supplied on the sgRNA expression plasmid so that the guide + donor pair is physically linked for delivery into the same cell. In yeast, we also showed that the donors can be supplied on the sgRNA expression plasmid to achieve efficient gene editing [48]. Using similar guide + donor strategies, we and others achieved gene diversification in yeast through CRISPR-HDR [49–52]. Our strategy, CHAnGE (CRISPR-Cas9 and homology-directed repair-assisted genome-scale engineering), enabled chromosomal gene diversification and genome-wide gene knockouts using plasmid libraries [49]. Recently, we expanded its targeting range by employing a PAM-relaxed Cas9 variant and achieved CRISPR-Cas9 and homology-directed repair-assisted saturation editing (dubbed as CHASE) [53]. Using a conceptually similar approach, Roy et al. developed Multiplexed Accurate Genome Editing with Short, Trackable, Integrated Cellular barcodes (MAGESTIC) [50]. In this system, editing efficiency was increased by more than five-fold to reach 60% on average through the active recruitment of donor DNA to the Cas9 cleavage site. Furthermore, the mutant tracking barcodes were genomically integrated by MAGESTIC to improve the fidelity of mutant tracing during downstream functional selection. Other variations of the same guide + donor strategy are also capable of performing chromosomal gene diversification with minor adaptations [51, 52].

These CRISPR-HDR platforms capable of large-scale diversification (CREATE, CHAnGE, MAGESTIC, etc.) are particularly transformative for microbial strain engineering (Fig. 2d)—a field requiring precise genetic modifications coupled with rapid phenotypic characterization for industrial applications. While traditional adaptive laboratory evolution [54] remains widely employed for industrial strain improvement, CRISPR-assisted diversification now offers a targeted alternative that significantly accelerates this process. The versatility of CRISPR-HDR has been demonstrated through successful diversification of various targets including metabolic genes [47], transcription factors [53], and promoters [52], facilitating the evolution of critical industrial phenotypes such as stress tolerance. By enabling simultaneous and trackable mutagenesis of metabolic pathways and regulatory elements, these systems overcome challenges previously solved by months of untargeted evolution.

While DNA-based donor templates are commonly used in CRISPR-HDR applications, RNA was also shown to serve as the donor for DSB repair in yeast [55]. In light of this, CRISPR- and RNA-assisted in vivo directed evolution (CRAIDE) utilized a chimeric donor gRNA (cgRNA) for both Cas9 targeting and DSB repair [56]. cgRNAs were continuously produced by an error-prone T7 RNA Polymerase (epT7RNAP), which introduced random mutations into the RNA template for integration into the target site (Fig. 2e). CRAIDE was used to evolve drug resistance by targeting an arginine transporter. However, this approach exhibits a mutation rate 2–3 orders of magnitude lower than other ectopic in vivo evolution methods [57, 58]. Its application for evolving complex and industrially relevant traits has yet to be explored.

Although the CRISPR-HDR approach approves versatile in mutant design, Cas9 cleavage can generate random insertions and deletions at the target site, especially in mammalian cell lines that prefer NHEJ. Additionally, both on-target and off-target cleavage can lead to structural abnormalities in chromosomes, such as unexpected translocations and the loss of large chromosomal segments [24, 59–61], potentially confounding the interpretation of genetic variant functions. To circumvent this problem, researchers have taken advantage of studying essential genes in haploid cell lines or strains (Table 1) [50], where lethal indels and structural abnormalities are depleted and only successful HDR events are retained. Beyond the challenges of indels and chromosomal instability, CRISPR-HDR is limited to HR-competent cells or during specific stages of the cell cycle [62, 63]. DSB-free and HDR-independent methods are thus desired in certain applications and species.

CRISPR-guided deaminases

Here we first illustrate the use of one of the HDR-independent methods, namely CRISPR-guided deaminases, for chromosomal gene diversification, where the associated deaminases directly covert nucleotide types without the need of a DSB [28, 64].

Cas9-deaminase fusions/complexes

Deaminases, such as the activation-induced cytidine deaminase (AID), have been utilized to generate genetic diversity of heterologous genes before the CRISPR era, by hijacking the natural somatic hypermutation system in Ramos B cells [65]. Base editing (BE) technology further renders deaminases with target programmability and a much wider adoption in different host cells. The first-generation base editor fused a cytidine deaminase (CDA) to dCas9, which induced the deamination of cytosine in the non-template strand to uracil. During DNA replication, this uracil was misidentified as a thymine, resulting in a C:G to T:A transition [28]. To expand the type of mutations, the cytidine deaminase was replaced with adenosine deaminase (ADA), resulting in the invention of an adenine base editor (ABE). This editor mediated the deamination of adenine to hypoxanthine, which is recognized as guanine during DNA replication, achieving A:T to G:C transitions (Fig. 3a) [66].

Fig. 3.

CRISPR-guided deaminases and its applications. a In base editing, a nickase version of Cas9 (nCas9) is fused with a cytidine deaminase (CDA), an adenosine deaminase (ADA), or both to induce base deaminations. This process converts cytosine (C) to thymine (T) and adenine (A) to guanine (G). The efficiency of C to T conversion can be enhanced by fusing nCas9 with uracil-DNA glycosylase inhibitor (UGI). In contrast, fusing with uracil-DNA N-glycosylase (UNG) generates apurinic/apyrimidinic (AP) sites, which leads to random mutations. BE is widely used in human cells for functional variant screening with high efficiency. In bacteria, BE facilitates the evolution of traits such as improved carbon source utilization through targeted mutagenesis of regulatory sequences of metabolic pathway genes. RBS, ribosome binding site. Additionally, in plants, BE is employed for evolving herbicide resistance and other agronomically important traits. b The CoMuTER system uses a crRNA-guided Cascade complex to recognize genomic targets and recruit a Cas3-CDA fusion, thus converting C to T during processing, which can be utilized to evolve a complete metabolic pathway. c In the HACE system, a nCas9-sgRNA complex binds to the target site and generates ssDNA. Then a helicase is loaded onto the ssDNA to continuously unwind the DNA in the 3'→ 5' direction, allowing the helicase-fused CDA to randomly deaminate cytosines on both strands

Two base editing tools, namely targeted AID-mediated mutagenesis (TAM) and CRISPR-X, demonstrated the ability of cytidine base editors (CBEs) to rapidly perform genetic diversification at specified chromosomal loci. Both tools incorporate a dCas9 that is complexed with AID, enabling the generation of a broad range of single-nucleotide mutations in the target region under the guidance of sgRNAs [67, 68]. The tools utilize different strategies of cytidine deaminase recruitment. TAM fuses AIDx (an AID variant) to the C-terminus of dCas9 [67], while CRISPR-X utilizes a modified sgRNA scaffold with a stem-loop structure containing two MS2 binding sites to recruit an MS2 protein fused with AID or a highly active AID variant (AID*∆) [68].

To enhance the diversity of mutation types, several dual base editing tools have been developed [69–76]. The first of these tools, saturated targeted endogenous mutagenesis editors (STEMEs), fused both a cytidine deaminase (APOBEC3) and an adenosine deaminase (ecTadA-ecTadA7.10) to the C-terminus of nCas9 in different configurations. This allowed for simultaneous C to T and A to G conversions within the same protospacer. By further employing a PAM-flexible Cas9 that recognizes the NG PAM sequence, STEME-NG achieved near-saturation mutagenesis across a 56-amino acid sequence [69]. In addition to the common C:G to T:A and A:T to G:C conversions, STEME-NG also facilitated less frequent C to G/A conversions. Targeting 400 amino acids of the carboxyl terminal domain of the rice acetyl-CoA carboxylase (encoded by the OsACC gene), STEMEs successfully induced mutants with enhanced herbicide resistance and generated multiplex mutations [69]. Additionally, instead of using fusion proteins, multiplexed orthogonal base editing (MoBE) also achieved random dual base editing using an sgRNA scaffold that concurrently recruits an ADA (TadA9) and a CDA (CDA1) using orthogonal stem loops, which seems to be an upgrade of CRISPR-X in plants [77].

To improve editing efficiency and accuracy for gene therapy, CBEs were initially designed with a fused uracil glycosylase inhibitor (UGI) to prevent the formation of apurinic/apyrimidinic (AP) sites after the removal of uracil through mismatch repair. For gene diversification, eliminating UGI or replacing it with uracil DNA N-glycosylase (UNG) enhances the formation of AP sites and promotes more diversified editing of the target sequence (Fig. 3a). Using this strategy, C to G base editors (CGBEs) achieved efficient C to G transversions by omitting UGI and incorporating eUNG (a variant of UNG from E. coli), while still retaining less frequent C to A/T conversions [78, 79]. Building on this, miniAGBE combines ABEs with CGBE to simultaneously generate A to G and C to G/T/A conversions at the target site, with editing windows of A₄-A₈ and C₃-C₁₃ [76]. During resistance screening, over 50,000 variants were assessed through the saturation editing of the critical region of the human diphtheria toxin receptor gene (hDTR) using miniAGBE, with 54.9% of the mutant gene displaying both A to G and C to G/T/A conversions [76].

BE exhibited significantly higher editing efficiencies in mammalian cells as compared to CRISPR-HDR (Table 1). With the rapid advancement of BE, recent studies have increasingly employed this technique to characterize genetic variations in both coding and non-coding regions [80–87]. In contrast to CRISPR-HDR-mediated functional screening, which uses complex donor libraries with substantially larger library sizes, BE enables direct catalytic conversion of target bases without donor templates. By simply transfecting a pooled sgRNA library, BE allows simultaneous screening of hundreds to thousands of sites in a single experiment. Its high editing efficiency and streamlined workflow confer superior practical throughput compared to HDR-based approaches. However, BE is inherently limited to mainly C:G to T:A and A:T to G:C conversions. Although emerging editors such as AGBE [76], CGBE [78, 79], and AYBE [88] expand the scope of base conversions, they still do not match the versatility of CRISPR-HDR, which introduces any desired single-nucleotide variant (SNV) or codon mutation. Thus, BE may be suited for an initial “coarse-grained” screening of functional variants at C/G or A/T sites, followed by a finer screen of non-BE-targetable mutations through CRISPR-HDR or other templated methods. Recent studies have begun integrating BE with prime editing (discussed in the section “CRISPR coupled with reverse transcriptase”) for comprehensive SNV analysis [89], demonstrating the potential of hybrid approaches.

Beyond its utility in mammalian cells, BE has emerged as a powerful tool for engineering prokaryotic systems. BE-based evolution of a Sec-translocase in Bacillus subtilis generated strains with higher protein secretion [90]. BE-assisted in situ diversification was also used to diversify combinations of regulatory sequences, enabling the engineering of Corynebacterium glutamicum and Bacillus subtilis for more efficient carbon source utilization and higher chemical production [91]. Furthermore, by developing a new BE system guided by dCas12b, the E. coli chassis was evolved to obtain a higher protein secretion efficiency [92].

BE is also widely adopted in plant breeding. Due to a low HDR efficiency in plants and the technical challenges associated with delivering HDR donor libraries [93, 94], BE has become a preferred strategy of high-throughput plant genome editing. The evolution of herbicide resistance demonstrated the potential of BE for crop improvement. Targeting the full-length coding region of the rice acetolactate synthase gene OsALS1, distinct variants conferring herbicide tolerance were identified by employing CBE and ABE [95]. Herbicide-resistant variants were also identified by screening other genes such as EPSPS, ALS, and HPPD in Arabidopsis through BE [96]. These variants in model plants may serve as a reference for enhancing traits in other crops.

Processive protein-deaminase fusions

Cas9-dependent deaminases are restricted to a narrow editing window when using a single sgRNA. While this can be remedied by using high-throughput sgRNA libraries, the library size can be a practical limit. In addition, the targeting density is restricted by the PAM distribution within the target region. Cas3 from the type I CRISPR/Cas systems, which harbors both helicase and 3'→ 5'nuclease activities, works in conjunction with the CRISPR-associated complex for antiviral defense (Cascade) to unwind and degrade DNA regions of at least 10 kb upstream of the PAM site [97, 98]. Confined Mutagenesis using a Type I-E CRISPR-Cas system (CoMuTER) fuses Cas3 with cytidine deaminase, enabling random deamination across a genomic region of up to 55 kb (Fig. 3b). This approach was successfully applied to evolve a lycopene biosynthesis pathway, resulting in a two-fold increase in the lycopene yield in yeast [99]. In another study, helicase-assisted continuous editing (HACE) was developed, which consists of a fusion of a DNA helicase and a deaminase. HACE is guided by CRISPR/Cas9 to become target specific and programmable (Fig. 3c) [100]. In this system, the helicase binds to the single-stranded DNA (ssDNA) within the Cas9-generated R-loop, and subsequently unwinds the DNA in the 3'→ 5'direction [100]. Meanwhile, the deaminase introduces random mutations along the ssDNA substrate. HACE generated mutations in both coding and noncoding strands. By employing three sgRNAs, HACE was programmed for the diversification of multiple genomic regions. Of these two examples, CoMuTER should be readily transferable to different hosts, given that type I CRISPR-Cas systems have been validated in multiple species including bacteria [101–103], mammalian cells [104–106], and plants [107]. The induction of genome-wide random mutations through helicase-deaminase fusion proteins was also reported in S. cerevisiae [108], indicating that HACE is also species-agnostic. Nonetheless, like with BE, the mutation types of CoMuTER and HACE are inherently limited by the chemistry enabled by the fused deaminases. Additionally, although the extended editing window permits the investigation of combinatorial effects of distal mutations, CoMuTER and HACE lack the ability to precisely control where to terminate the continuous mutagenesis. Overall, these two examples represent exciting developments that combine the programmability of CRISPR systems and the processivity of certain enzymes.

CRISPR coupled with reverse transcriptase

The other DSB-free and HDR-independent gene diversification method uses prime editors (PEs). PE consists of a protein complex and a prime editing guide RNA (pegRNA) (Fig. 4). The protein complex consists of a reverse transcriptase domain fused to the C-terminus of nCas9 (H840A). The pegRNA is a modified version of sgRNA that includes a 20-nucleotide guide sequence to direct nCas9 to the target site, along with an extended sequence at the 3'end of the sgRNA scaffold that contains a primer binding sequence (PBS) and a reverse transcriptase template (RTT). The RTT, which encodes the desired mutations, is reverse transcribed into DNA by the reverse transcriptase using the nicked ssDNA as a primer, allowing for the introduction of new genetic variations. On par with HDR, the PE system enables all 12 types of point mutations [29].

Fig. 4.

CRISPR coupled with reverse transcriptase and its applications. Prime editing utilizes a fusion protein that includes a nCas9 and a reverse transcriptase (RT), along with a prime editing guide RNA (pegRNA) that features a primer-binding site (PBS) and an RT template (RTT). When nCas9 nicks the non-target DNA strand, the PBS hybridizes with the exposed ssDNA, permitting the RT to reverse transcribe the RTT into DNA, enabling targeted genetic modifications. PE enables saturation mutagenesis in mammalian cells for functional variant analysis and facilitates crop improvement, such as herbicide resistance evolution

To date, several studies have demonstrated chromosomal gene diversification with high-throughput prime editing using a pegRNA library [89, 109–114]. This approach was first demonstrated in rice through the development of prime editing-library-mediated saturation mutagenesis (PLSM) [109]. PLSM targeted six conserved residues in the rice OsACC gene known to be associated with herbicide resistance, with RTTs covering all possible codon substitutions for each residue. Sixteen herbicide-tolerant mutations across the six sites were identified [109]. Another strategy that combines prime editing with site haploidization expands the application of saturation prime editing (SPE) to mammalian cellular models [110]. Enabled by SPE, this study analyzed approximately 1400 SNVs in selected regions of two disease-related genes [110]. Later developments utilized even larger pegRNA libraries. For instance, the pooled prime-editing screen method (PRIME) improved editing efficiency to 40% by optimizing lentiviral delivery and created a library of 6252 pairs of pegRNA and nicking sgRNA that targeted a 716-bp enhancer region, generating 2127 single-nucleotide substitutions [111]. Prime editing and endogenous region sequencing (PEER-seq) used an elegant approach to enhance both the accuracy of SNV detection and editing efficiency [112]. PEER-seq simultaneously introduced a 1-bp synonymous mutation adjacent to the intended mutation. This synonymous mutation serves as a reliable editing marker and distracts the mismatch repair system from repairing the intended edits. As a result, false-positive reads were reduced by 625-fold, and overall editing efficiency increased by 3.4-fold. Additionally, PEER-seq revealed that knocking out MLH1 significantly improved prime editing efficiency compared to using a dominant negative variant (MLH1dn), serving as a reference for future PE optimization [112]. In another study, PE was combined with BE to realize multimodal precision gene editing screens, allowing for a more comprehensive analysis of genetic variants by overcoming the limitations inherent to each method when used alone [89]. This multimodal approach enhanced the modeling of broader genetic diversity and supported clinical diagnosis and drug development for populations with rare mutations. Similarly, Prime Editing Sensor Libraries [113], akin to Base Editing Sensor Libraries [83], link sensor targets to pegRNAs, thereby providing a more accurate quantitative assessment of editing efficiency and a more comprehensive functional evaluation of genetic variants. This strategy contributed to a deeper understanding of pathogenic variants.

Compared with CRISPR-HDR in mammalian cells, PE-based approaches induce the same full-spectrum mutations without generating DSBs yet with much higher editing efficiencies (Table 1). However, the PE system is more complex, and its larger size often raise delivery challenges [115]. Further optimization of pegRNA design is also required to predict efficient pegRNAs to ensure success in massively parallel screening. In terms of applicability in different host species, PE allows gene diversification in plants, which is challenging to accomplish using CRISPR-HDR. However, the utility of PE-based screening in microbes has yet to be explored.

CRISPR coupled with error-prone DNA polymerase

Another DSB-free strategy called EvolvR fuses a nCas9 (D10A) with an error-prone DNA polymerase I (PolI3M) (Fig. 5) and was initially developed in E. coli [116]. Mechanistically, PolI3M introduces random mutations while extending from the 3'-OH end generated by nCas9. The editing window of EvolvR spans 17 base pairs from the incision, which is limited by the processivity of PolI3M. To extend the editing window, the thioredoxin-binding domain from the bacteriophage T7 DNA polymerase was inserted into the thumb domain of PolI3M, increasing the window to 56 base pairs. However, the mutation rate decreased rapidly as the distance from the nicking site increased. Attempts to replace PolI3M with a more processive Phi29 DNA polymerase successfully extended the editing window to 347 bp, but resulted in a significant reduction in mutation rate. EvolvR was later adapted for use in S. cerevisiae, namely the yEvolvR system. yEvolvR utilized polymerases with varying fidelity to achieve tunability of the mutation rate, and realized the diversification of the endogenous CAN1 gene [117]. The EvolvR technologies are unique in that both error-prone DNA synthesis and processivity can be accomplished by the DNA polymerase, which is not achievable with Cas3 or helicases. It also achieved more diverse base conversions than deaminases. However, the randomness of mutagenesis also makes it more suited for directed evolution rather than the functional profiling of user-defined SNVs.

Fig. 5.

CRISPR coupled with error-prone DNA polymerase. The EvolvR system integrates nCas9 (D10A) with an error-prone DNA polymerase I (PolI3M), which introduces random mutations while processing from the 3' end of the nCas9-generated nick

CRISPR-assisted structural variation

While CRISPR systems have been mainly applied to perform gene diversification with point mutations, several recent studies highlighted its potential to also induce diverse structural variations (SVs). SVs, including large deletions, duplications, inversions, and translocations, are key drivers of genome diversity and natural evolution. Our understanding of SVs has largely come from observations of naturally occurring variations. The emergence of CRISPR tools now allows the de novo creation of SVs (Fig. 6), permitting a more comprehensive interrogation of their impacts on genome function. For example, by employing CRISPR/Cas9 to target the repetitive retrotransposon element Ty1 in the yeast genome, DSBs formed by Cas9 cleavage within Ty1 sequences induced a wide range of structural variations, including deletions, duplications, and translocations [118]. This artificial induction of SVs helped reveal the importance of retrotransposon lesions in driving genome evolution. Alternative strategies coupling CRISPR with site-specific recombinases have also been developed. In recent examples, multiple LoxP sites were inserted into the human genome using PE targeting repetitive LINE-1 elements, and random structural variations including deletions, inversions, extrachromosomal DNAs, and translocations were generated through Cre recombinase-induced genome shuffling [119]. Analysis of survival cell clones revealed some interesting findings such as the relative locations of genes did not influence their expression levels in their experimental cell line. This method is further extended to induce SVs within large regulatory regions, such as super-enhancers, enabling the generation of randomized regulatory configurations. It allowed the resolution of functional minimal elements within the super-enhancer [120], revealing a surprising result that many sequences are dispensable for super-enhancer function. In another study, up to 30 LoxP sites were introduced into Vibrio natriegens chromosomes using CRISPR-associated transposases (CASTs, reviewed elsewhere [121]) [122]. Cre recombinase-induced chromosome rearrangement led to the identification of a genomic region underpinning the rapid growth phenotype of V. natriegens, a characteristic highly relevant for industrial and molecular biology applications. While the above methods took advantage of repetitive elements, innovative PE strategies enabled the generation of precise SVs. For example, PRIME-Del utilizes dual pegRNAs to achieve large genomic deletions with user-defined boundaries, which are specified by the two reverse-complement pegRNA templates [123]. Such strategies may be further scaled up to generate a library of precise SVs for functional genomics studies. Although these CRISPR-assisted structural variation methods are only emerging, they promise to further our understandings of genome evolution, speciation, and may facilitate the generation of next-generation biotechnological chassis with minimal and customized genomes (Fig. 6).

Fig. 6.

CRISPR-assisted structural variation and its applications. Structural variations can be induced by CRISPR/Cas9 targeting high-copy genomic sites, such as the Ty1 retrotransposon elements in yeast (upper panel). Alternatively, recombinase recognition sites like LoxP can be inserted into the genome using prime editing (PE) or CRISPR-associated transposases (CAST), with structural variations subsequently generated through recombinase induction (lower panel). The generation of diverse artificial SVs could facilitate studies of genome evolution principles, enable the engineering of large regulatory elements, and support the development of next-generation chassis organisms with minimal genomes

Data analysis

After phenotypic selection of a diversified library, selecting the appropriate method to analyze the outcomes is crucial for measuring mutation effects or identifying the desired variants. For targeted diversification of a genomic region, which can be achieved through CRISPR-HDR, BE, and PE, high-throughput amplicon sequencing of the target gene is typically used (Fig. 1). By aligning sequencing reads to the reference, mutations are called and counted. Enrichment or depletion of designed mutations can be calculated by comparing samples from untreated and treated conditions. This change in mutant abundance serves as a functional score indicating the mutation effect size or fitness (Fig. 2b). In some microbial CRISPR-HDR approaches (Fig. 2d), where a DNA barcode or the guide is uniquely paired with the introduced mutation, the abundance of such barcodes can be used as an estimate of the abundance of the mutations [47, 50]. In such cases, well developed algorithms and computational pipelines for analyzing traditional CRISPR knockout screens can be adapted for data analysis [124]. In directed evolution methods such as CasPER (Fig. 2c), CRAIDE (Fig. 2e), CoMuTER (Fig. 3b), HACE (Fig. 3c), and EvolvR (Fig. 5), where typically only “winners” are of interest, several surviving clones will be sequenced to identify common mutations that indicate functional importance. Otherwise, long-read sequencing is needed for phasing multiple introduced mutations. For SVs, where sequence alterations occur across an entire genome or within large genomic regions, whole-genome sequencing is often necessary to determine genotypes. Recently, a barcoding strategy has also been validated in analyzing SVs, greatly reducing the cost and labor [125]. Generally, most data analysis codes and pipelines have been developed ad hoc in each study. Developing a universal software or webtool integrating library design and data analysis with easy-to-use interfaces, at least for some of the CRISPR-based methods reviewed here, will greatly facilitate the adoption of these advanced gene diversification techniques.

Future directions

Further expanding the mutation spectrum and editing window

For CRISPR-guided deaminases, dual base editors have been used to increase the mutation spectrum [69–72]. However, it remains less diverse comparing to HDR and PE (Fig. 7). Achieving A to N editing in BE and efficient dual deamination in processive systems will be the next milestones. Progresses are already there. For example, AYBE, developed by fusing ABE with an N-methylpurine DNA glycosylase (MPG), excises hypoxanthine generated from adenine deamination, enabling A to C and A to T transversions [88]. In theory, combining UNG and MPG with a dual BE system could allow for concurrent C to N and A to N conversions and possibly in high-throughput. Such strategy may also be tested on the CoMuTER and HACE systems.

Fig. 7.

Editing window, library size, and mutation types of representative chromosomal gene diversification approaches. N indicates mutation of a single base to any base; NNN indicates the mutation of an entire codon; HDR, homology-directed repair; PE, prime editing. The data used for plotting and detailed explanation of the library size and editing window of each approach can be found in Additional file 1: Table S1

The range of targeted gene diversification, or the editing window, determines the size of the genomic region that can be perturbed in a single experiment. The editing window varies depending on the method used (Fig. 7). High-throughput libraries allow targeting of multiple sites simultaneously, expanding the editing window up to thousands of base pairs. Expanding this editing window is ongoing. Studies have utilized PAM-relaxed Cas variants for this purpose [53, 87, 126]. Sangree et al. have shown that PAM-relaxed Cas9 variants, when used in BE, increased the editable residues in BRCA1 from 24.2 to 75.3% [84]. However, these Cas mutants are associated with improved off-target effects [127, 128]. In addition, they may suffer from reduced activities. Alternatively, the editing window of BE may be expanded through protein architecture engineering. In one example, multiple copies of a deaminase were recruited to dCas9 through the SunTag scaffold, which expanded the editing window [129]. The editing window was also increased by inserting the deaminase into the PI domain or replacing the HNH domain with the deaminase [130, 131]. These engineered systems expanded the editing window without affecting the Cas specificity. Importantly, a smaller sgRNA library can cover the same editing window as compared to those PAM-relaxed Cas variants. In addition, by fusing the deaminase with other Cas proteins, such as Cas12b, the editing window may also be increased [92].

An alternative strategy for expanding the editing window may be to use large sequence genome editors. For example, CASTs enable the targeted integration of large DNA fragments up to 10 kb in length [132, 133]. In these systems, the R-loop formed by CRISPR/Cas near the target site serves as a binding platform for the transposase complex. Subsequently, mediated by the transposase, fragments of donor DNA can be inserted without the need for DSBs. However, the insertion typically occurs within a few dozen base pairs downstream of the sgRNA target site and lacks precision. To achieve a more precise integration, the programmable addition via site-specific targeting elements (PASTE) system leverages prime editing to pre-install an attB recognition site, followed by phage-derived serine integrase-mediated integration of a circular double-stranded DNA template containing an attP attachment site into the attB locus [134]. Potentially, both CAST and PASTE systems may be used to insert long fragments derived from epPCR, analogous to CasPER, but in mammalian cells.

Multiplex gene diversification

Most efforts so far focused on sequence diversification of individual genes or regions. With smart library design, multiplex gene diversification may enable us to tackle many interesting problems or achieve more ambitious engineering goals, for example to probe protein–protein interactions or to concurrently optimize multiple regulatory sequences. A representative example is the use of base editors to simultaneously target the ribosomal binding site (RBS) regions of three genes involved in xylose metabolism in bacteria. The optimized RBS combinations significantly improved the xylose utilization rate. The same approach was applied to screen multiple high-yield RBS combinations of lycopene biosynthesis pathway genes [91]. Recently, multiplex gene editing using BE and PE, guided by a single sgRNA for each gene, were demonstrated in more species including mammalian cells and plants [135–138], suggesting that multiplex gene diversification is feasible in diverse organisms. However, several technical challenges must be addressed to achieve large-scale multiplex gene diversification. One of them is the exponentially growing library size. Due to the library cost and transformation limits, it is wise to prioritize key regions for diversification. Alternatively, if the mutation spectrum of CoMuTER and HACE can be further improved, transforming a few sgRNAs instead of a huge library may suffice multiplex gene diversification. In some niche applications, there may be a need to diversify sequence regions with extreme AT or GC content, such as promoters or particular gene bodies. In such scenarios, it becomes necessary to develop orthogonal targeting systems, such as Cas12a and SpCas9.

Automated iteration of workflows

At present, the creation of large-scale variant libraries and high-throughput screening mainly rely on manual workflows. With the projected increase in library scale and evolution rounds, automation rapidly becomes a necessity. The significance of automation platforms in genetic engineering has been demonstrated. For example, CRISPR-HDR-mediated gene integration was automated to realize overexpression and knockdown of more than 90% of the genes in S. cerevisiae, leading to the development of acetic acid-resistant strains after three rounds of evolution [139]. Such established workflows could be easily adapted for iterating gene diversifications, enabling long-term and even high-throughput strain evolution beyond the current scope achieved, in which typically only one or two rounds of evolution were demonstrated as a proof-of-concept. Although mostly developed for microbes, such automation can be easily reconfigured to facilitate mammalian gene diversification as well. For example, cellular disease models with pathogenic SNVs were recently generated on an automated high-throughput platform [140]. This platform streamlined the large-scale cloning of sgRNA libraries containing thousands of plasmids, cell transfections, and high-throughput screening. It is anticipated that iterating this workflow for gene diversification will enable mammalian directed evolution beyond functional genomics.

In conclusion, we believe that CRISPR and its derivative technologies will continue to evolve, fueling chromosomal gene diversification in ever-expanding species and applications. Looking ahead, with the rapid advancement of modern artificial intelligence algorithms and automation platforms, the vast amount of data generated can be used to build powerful predictive models, aiding in library design and variant discovery. Eventually, the limiting factor will not be the technology itself, but where in the vast genome space we want to investigate.

Authors’ contributions

ZB conceived of the manuscript. RZ and CR wrote the manuscript. RZ, CR, and ZB made the figures. ZB reviewed and edited the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Key R&D Program of China (grant no. 2023YFF1204500 to ZB), the National Natural Science Foundation of China (grant no. 22308316 to ZB), and the Fundamental Research Funds for the Central Universities (grant nos. 226–2025-00043 and 226–2022-00214 to ZB).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.