Precision Editing of Cis-Regulatory Elements Drives Crops Quantitative Traits Genetic Improvement: From Functional Dissection to Intelligent Design

Abstract

Since most agronomic traits are quantitative, ensuring food security under the pressures of global climate change and population growth increasingly relies on the precise and efficient genetic improvement of crop quantitative traits, making it a central focus of breeding research. Traditional breeding and transgenic approaches are often insufficient for modulating complex quantitative traits. In contrast, the advent of gene-editing technologies has opened new avenues for crop genetic improvement. Notably, the editing of cis-regulatory elements allows fine-tuning of gene expression levels and spatiotemporal patterns without altering coding sequences, enabling targeted optimization of quantitative traits such as yield, quality, and stress resistance. This review systematically summarizes the evolution of cis-regulatory element editing technologies, from early random mutagenesis and screening to targeted dissection guided by functional genomics and, more recently, to intelligent design integrating multi-omics and artificial intelligence, highlighting key technologies, representative applications, and inherent limitations at each stage while discussing future research directions in data integration, algorithm development, and tool deployment. We hope this review will provide both theoretical guidance and practical strategies for intelligent crop breeding.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12284-026-00939-z.

Introduction

Previous studies have projected that the global population will reach approximately 9.7 billion by 2050, and that total global food demand may increase by 50 ~ 60% over the same period, placing substantial pressure on future agricultural production systems (Gu et al. 2021; Falcon et al. 2022). Meanwhile, the increasing frequency of extreme climatic events, intensifying constraints on arable land, and the continual rise in pest and disease pressures have further amplified the uncertainty of food production (Xiong et al. 2022). An important solution to these challenges lies in continuously improving the intrinsic productivity and adaptability of crops, with the core objective being the efficient genetic enhancement of key agronomic traits, including yield, quality, and stress resistance. However, many agronomic traits are quantitative, typically governed by polygenic networks and exhibiting continuous phenotypic variation, which poses substantial challenges for genetic improvement in breeding. Traditional breeding relies heavily on natural variation and phenotypic selection, resulting in long breeding cycles and limited efficiency. Although transgenic technologies enable the targeted introduction of exogenous genes, they still exhibit significant limitations in regulating complex polygenic traits, maintaining expression stability, and ensuring biosafety (Lenaerts et al. 2019; Chen et al. 2019). Therefore, the development of breeding technologies capable of precisely, efficiently, and predictably regulating quantitative traits has become an urgent requirement for overcoming current breeding bottlenecks.

In recent years, the emerging technology of cis-regulatory element (CRE) editing has offered unprecedented potential for the precise regulation of quantitative traits. Promoters, enhancers, silencers, insulators, and upstream open reading frames, as CREs, serve as recognition and binding sites for transcription factors and transcriptional regulatory complexes, directly determining the intensity of gene expression, spatiotemporal specificity, and responsiveness to environmental cues. Unlike knockout or gain-of-function approaches that target coding sequences, CRE editing modifies gene regulation without altering the encoded protein (Swinnen et al. 2016; Rodríguez-Leal et al. 2017). It is therefore well suited to complex traits that require graded dosage adjustment. This strategy can avoid developmental costs associated with complete loss of function or excessive overexpression in coding regions, and it enables fine-tuning of key regulatory elements to optimize target gene expression within a range closer to the natural physiological state.

For example, the rice yield-regulating gene OsSPL14/IPA1 exhibits a trade-off between tillering and grain number per panicle: reduced expression increases tillering but decreases grain number per panicle, whereas elevated expression produces the opposite effect. Therefore, neither simple functional attenuation nor forced enhancement can simultaneously optimize plant architecture and yield components; subsequent editing of its regulatory elements demonstrated that a more appropriate strategy is to reconstruct its expression window rather than merely alter protein function (Jiao et al. 2010; Song et al. 2022). Similarly, CLE signaling associated with FCP1/FCP2 shows clear dosage dependence during rice meristem maintenance (Suzaki et al. 2008; Ohmori et al. 2013). Loss of function expands the stem cell domain, whereas overexpression can terminate the meristem. Thus, coding-region knockout or strong overexpression may destabilize the regulatory network.

Together, these studies indicate that for quantitative traits constrained by multiple developmental processes, simple “knockout/overexpression” manipulations of coding regions often trigger trait trade-offs and overall agronomic imbalance. Accordingly, genome editing of cis-regulatory elements holds considerable potential to provide a new technological avenue for the genetic improvement of crop traits through precise modulation of gene expression. In particular, the application of cis-regulatory element editing, especially promoter editing, in plant trait improvement has continued to expand. Several reviews on this topic have been published, reflecting, to some extent, the growing importance of this research field (Shi et al. 2023; Tang and Zhang 2023; Wu et al. 2024). However, as research on cis-regulatory elements continues to advance rapidly, new application cases, editing tools, and design strategies are constantly emerging. A more systematic and timely synthesis of these developments is therefore warranted. Accordingly, this review focuses on the precise editing of cis-regulatory elements for crop trait improvement and provides a comprehensive overview of recent research advances, technological approaches, and future directions in this field, thereby offering a reference for subsequent functional characterization and molecular breeding research.

Notably, the development of CRE editing technologies represents not merely a simple iteration of tools, but a paradigm shift from empirical exploration to rational design. In the first stage, CRISPR-Cas9 was used to randomly mutate regulatory regions, followed by large-scale screening. Representative studies at this stage demonstrated that cis-regulatory regions themselves can serve as effective targets for quantitative trait improvement, while also highlighting challenges such as high screening costs, low hit rates, and unclear causal links between editing and phenotype (Rodríguez-Leal et al. 2017). The second stage marked the transition to functional dissection and targeted approaches. Functional genomics then enabled researchers to locate and annotate candidate CREs using multiple lines of evidence. These included expression quantitative trait loci (eQTLs), chromatin accessibility, epigenetic modifications, and three-dimensional genome organization (Zhu et al. 2015; Dong et al. 2017). Their contributions to gene expression and phenotype were then dissected through targeted editing, segmental deletions, and reporter assays, thereby markedly improving the rationale for target selection and the precision of editing localization. The third stage has gradually moved toward intelligent design. Through the deep integration of multi-omics data and artificial intelligence algorithms, researchers can now predict the regulatory effects of candidate CREs and the expected expression responses after editing, allowing prioritization of target elements and quantitative design of editing schemes. This approach improves target selection and makes editing outcomes more predictable. It reduces reliance on phenotype-based screening and places greater emphasis on prediction-guided validation. (Zhou et al. 2023; Farooq et al. 2024). Against this background, this review systematically summarizes the evolutionary trajectory of CRE editing technologies, discusses the technological breakthroughs and scientific challenges at each stage in depth, and provides theoretical guidance and practical strategies for intelligent crop breeding.

Regulation of Quantitative Traits by Cis-Regulatory Elements

The continuous variation of quantitative traits is essentially the result of fine-tuned regulation of gene expression levels. Through mechanisms such as sequence variation and structural remodeling, cis-regulatory elements precisely regulate gene expression at multiple levels, including transcription initiation, chromatin spatial interactions, and post-transcriptional regulation, thereby ultimately influencing the formation and variation of quantitative traits. Their regulatory modes are mainly concentrated in four core dimensions: promoter variation, chromosomal structural variation, untranslated region (UTR) regulation, and the regulation of cis-regulatory elements in intronic regions (Fig. 1). Representative studies of CRE-mediated regulation and editing across different regulatory layers are summarized in Table 1.

Fig. 1.

Major regulatory layers of cis-regulatory elements (CREs) underlying quantitative trait variation in crops. a Promoter variation affects transcription through altered transcription factor (TF) binding and local sequence variation, including single-nucleotide polymorphisms (SNPs) and insertions/deletions (InDels). b Intronic cis-elements regulate gene output mainly by modulating alternative splicing patterns and splicing ratios. c UTR regulation influences post-transcriptional control through 5′ UTR upstream open reading frame (uORF)-mediated regulation and 3′ UTR mRNA targeting. d Chromosomal structural variation reshapes regulatory activity through deletion, rearrangement, or changes in chromatin organization. Together, these four regulatory layers fine-tune gene expression and contribute to continuous variation in agronomic traits

Table 1.

Representative studies of CRE editing for crop quantitative trait improvement

Regulatory layer	Host	Target gene(s)/region	Strategy/variant type	Editing/regulatory effect	Phenotypic effect	Reference
Promoter	Rice	IPA1, SCM3/OsTB1, Hd1, Ghd7, DTH8, SBEIIb, APO1, OsD18, GSE5, HAN1, PH8, OsSWEET11, OsSWEET13, OsSWEET14, Xa23, D53	Natural variation; CRISPR-CRE editing; TALEN editing; promoter truncation; PEPE	Graded expression; Tunable expression; Pathogen-responsive EBEs disrupted in SWEET susceptibility-gene promoters; EBEAvrXa23 inserted into the xa23 promoter	Yield, plant architecture, quality, cold tolerance, tiller number; bacterial blight resistance	Zhang et al. (2017); Song et al. (2022); Cui et al. (2020); Zhou et al. (2023, 2024); Sang et al. (2025a); Zhang et al. (2025a); Duan et al. (2017); Mao et al. (2019); Chen et al. (2025); Li et al. (2012, 2026); Blanvillain-Baufumé et al. (2017); Oliva et al. (2019); Xu et al. (2019); Wei et al. (2021);
	Maize	ZCN8, ZmCCT, ZmPAT7, bZIP68	Natural SNP/TE/indel	Expression alteration	Flowering adaptation, tassel trait	Guo et al. (2018); Yang et al. (2013); Guan et al. (2022); Li et al. (2022)
	Tomato	CLV3, WUS, FW2.2, LCYB, CNR	CRISPR-Cas9 multiplex editing of promoter cis-regulatory elements	Graded promoter activity changes, quantitative expression tuning of all five genes	Continuous variation in fruit size, locule number, and fruit shape; proof-of-concept for quantitative trait engineering	Rodríguez-Leal et al. (2017)
Structural variation	Rice	OsDREB1C, HPPD, Ghd7	CRISPR-SV engineering	Distal CRE regulation; upregulation	Yield, herbicide resistance, early heading	Luo et al. (2025); Lu et al. (2021); Ogo et al. (2025)
	Multi-species	fw2.2, DG1, MBP3, GhPH1	Natural SV/inversion/indel	Chromatin/rewiring	Fruit trait, yield, plant height	Frary et al. (2000); Zhang et al. (2025b); Cheng et al. (2025); Tian et al. (2024)
	Maize, Citrus, Arabidopsis, B. napus	p1/p2, acidity loci, inverted repeats, GTR2, MAM, MYB2 et al.	Transposon/TE/haplotype SV	Expression network shift	Pigment, acidity, flowering, yield	Sharma and Peterson (2023); Diaz et al. (2025); Arce et al. (2023); Zhang et al. (2024a )
UTR	Rice, Wheat	OsHPPD, TaSPL13, Wx(Waxy), OsCSLC3, OsGGP, OsLTT7, OsSBI, GID1, AtABI1, OsBRI1, AtPYR1, et al.	3′ UTR/5′ UTR-uORF editing	Post-transcription/ translation control	Herbicide resistance, grain quality, VC, blast resistance	Wu et al. (2023); Gupta et al. (2023); Zhao et al. (2024); Zeng et al. (2020); Shi et al. (2025); Sang et al. (2025b); Wang et al. (2024); Liu et al. (2025a ); Xue et al. (2023)
	Multi-species	GGP, AtTPPD, AtBRI1, AtVTC2, et al. FaTMT2	uORF/miRNA target editing	Tunable expression	Sugar gradient, quantitative traits	Laing et al. (2015); Wang et al. (2024); Zhang et al. (2018); Xing et al. (2020);
Intron	Rice	Eui1, GS3, Wxb, Wx(Waxy)	Silencer deletion/splicing/ combinatorial editing	Transcription/ splicing shift	Dwarfism, grain size, amylose	Xie et al. (2018); Liu et al. (2022a, b); Zeng et al. (2020)
	Multi-species	Adh1, ACT1, UBQ10,αAmy3	IME/intron-gRNA system	Expression enhancement	Engineering module/ editing efficiency	Laxa (2017); Nguyen et al. (2022)

Note: This table summarizes representative studies on promoter editing, structural variation engineering, UTR regulation, and intronic cis-element manipulation across major crops, including regulatory layers, target genes or regions, editing strategies, regulatory effects, phenotypic outcomes, and references

Regulation of Quantitative Traits by Promoter Cis-Regulatory Variants

As the core regulatory region for transcription initiation, the promoter represents the most direct target for precise cis-element editing and is also the level best suited for achieving graded regulation of expression intensity. Quantitative trait improvement often requires an optimal expression range rather than simple loss or gain of gene function. For such genes, knockout of the coding region easily results in loss of function, whereas constitutive overexpression often causes new developmental imbalances. Therefore, the key target for adjustment is not the protein sequence itself, but the regulatory program that determines expression amplitude, timing, and tissue specificity. Promoter editing is useful in this context because it preserves the encoded protein while adjusting expression amplitude, timing, or tissue specificity. It can modify transcription factor-binding sites, local cis-element combinations, or proximal regulatory architecture to generate more suitable expression windows. This concept is most clearly illustrated in studies of IPA1, which regulates the typical trade-off between plant architecture and yield. When IPA1 expression is low, plants produce more tillers but fewer grains per panicle; when IPA1 expression is high, tillering is suppressed. Thus, neither functional attenuation nor forced enhancement can simultaneously balance these two key yield components (Zhang et al. 2017). Building on this understanding, Song et al. (2022) shifted their focus from the coding region to the IPA1 regulatory region itself. Using CRISPR-Cas9 tiling deletion scanning, they systematically dissected upstream cis-elements and ultimately identified a critical 54-bp regulatory fragment. Targeted deletion of this fragment reset IPA1 expression to a range more favorable for coordinating tiller number and grain number per panicle, thereby increasing grain yield per plant. This example shows that, once the dosage effect is clarified, promoter editing can be used to reconstruct the core cis-regulatory segment that defines the expression window.

Building on the concept of “resetting the expression window”, promoter editing has first established a clear technical strategy: constructing a series of cis-allelic variants with varying expression strengths by editing multiple sites or regulatory regions of multiple genes. This approach converts quantitative traits into a continuous phenotypic space that can be systematically screened and selected. A representative example is OsTB1 in rice, which simultaneously affects culm development, panicle traits, and tillering. Direct knockout or overexpression of OsTB1 can drive multiple traits to extremes, hindering the development of intermediate germplasm suitable for breeding. To address this, Cui et al. (2020) performed multiplex CRISPR-Cas9 editing on the promoter and 5′ non-coding region of SCM3/OsTB1. By integrating six sgRNAs into a single vector and simultaneously perturbing multiple proximal regulatory sites, they generated a series of cis-allelic variants with different expression levels, establishing a continuous gradient in traits such as culm thickness, panicle size, and tiller number. The key insight from this work is not the number of sites edited, but that this approach enables the active construction of a selectable expression gradient while preserving the basic function of OsTB1.

The same logic is also reflected in the regulation of growth duration through multi-gene editing. Heading date is a polygenic trait controlled by multiple regulatory factors. Therefore, simply knocking out a single flowering-time gene often produces excessively early- or late-maturing phenotypes, which are difficult to adapt to the requirements of different ecological regions. Zhou et al. (2024) performed multiplex CRISPR-Cas9 editing on the cis-regulatory regions of Hd1, Ghd7, and DTH8 without altering their coding sequences. Instead, by reshaping their expression levels and combinatorial relationships, they generated materials showing continuous variation in heading date, ranging from approximately 73 to 107 days. This study demonstrates that when a target trait is determined by the expression balance of multiple genes, promoter editing is more suitable than coding-region editing for combinatorial optimization at the level of the regulatory network.

Studies on quality improvement further demonstrate that promoter editing is particularly well suited for metabolic pathways that are highly sensitive to strong perturbations in coding regions. Increasing resistant starch content is generally beneficial for improving nutritional value; however, excessive weakening of the relevant starch biosynthesis genes may lead to reduced eating quality and even abnormal grain development. Thus, the key issue is to achieve a balance between nutritional value and palatability. To address this, Sang et al. (2025a) did not employ coding-region knockout, but instead rewrote the SBEIIb promoter through CRISPR-Cas9 multi-target editing, resulting in graded reductions in its expression among different materials. Consequently, resistant starch content increased to several times that of the wild type, while avoiding the obvious grain defects typically caused by conventional loss-of-function mutations. This finding indicates that, in quality-related pathways, the value of promoter-level editing lies in preserving the basic function of key enzymes while generating a more stable and practically useful quality gradient through dosage regulation.

Building on the construction of continuous allelic series, promoter editing has further developed a second, more advanced strategy: the targeted optimization of key cis-elements or core regulatory segments around clearly defined breeding objectives. The study of rice qPL6/APO1 is a representative example of this approach. APO1 is involved in panicle development. The breeding objective is to identify a favorable balance among panicle length, grain number, and overall yield. With this objective, researchers performed targeted deletions in a key CRE region of the qPL6/APO1 promoter, generating novel cis-allelic variants, including a 127-bp deletion. These edits increased APO1 expression, significantly enhanced panicle length and grain number per panicle, and ultimately achieved a yield increase of approximately 23% ~ 41.5% (Zhang et al. 2025a). This result indicates that once a key repressive cis-regulatory segment has been clearly identified, promoter editing can directly serve target-trait optimization, rather than relying on the prior construction and subsequent screening of a large-scale mutant library.

Similarly, in the optimization of plant architecture, the key question for OsD18 is not whether plants should simply be made shorter, but rather what degree of dwarfing can achieve a better balance between lodging resistance and yield stability. Zhou et al. (2023) performed precise segment-level modification of the OsD18 promoter and obtained semi-dwarf materials showing continuous variation in plant height, improved lodging resistance, and maintained yield. The breeding goal was not to produce a single dwarf phenotype with a fixed magnitude. Instead, promoter editing generated a selectable gradient of plant height, allowing suitable phenotypes to be identified across different breeding backgrounds.

Promoter editing has also been widely used to improve disease resistance in crops. Rice bacterial blight is one of the best-established examples. Transcription activator-like effectors (TALEs) secreted by Xanthomonas oryzae pv. oryzae recognize and bind effector-binding elements (EBEs) in the promoters of rice susceptibility genes. This interaction activates susceptibility genes, including members of the SWEET family, and facilitates disease development. An early study used TALENs to edit the binding sites for AvrXa7 and PthXo3 in the OsSWEET14 promoter. The results showed that disrupting promoter elements recognized by pathogen effectors could enhance resistance to bacterial blight (Li et al. 2012). Subsequent editing of different TALE-binding sites in the OsSWEET14 promoter further confirmed the effectiveness of this strategy (Blanvillain-Baufumé et al. 2017). With the development of CRISPR-Cas9, researchers simultaneously edited TALE-binding elements in the promoters of several susceptibility genes, including OsSWEET11, OsSWEET13, and OsSWEET14. These modifications generated rice lines with broader-spectrum resistance to bacterial blight (Oliva et al. 2019; Xu et al. 2019). Promoter engineering has also been used to reconstruct the regulation of resistance genes through precise sequence insertion. For example, precise insertion of EBE_AvrXa23 into the promoter of the susceptible xa23 allele enabled its activation by a pathogen-derived TAL effector. This modification converted xa23 into a disease-resistance allele (Wei et al. 2021). Together, these studies show that promoter editing can improve disease resistance through two complementary routes. It can block pathogen-induced expression of susceptibility genes or establish pathogen-responsive expression of resistance genes.

Alongside these editing studies, promoter variation in natural populations has continuously provided two important insights: first, which genes are highly sensitive to expression dosage; and second, which traits are the most promising initial targets for promoter-level manipulation. Variants such as promoter deletions of GSE5 in rice associated with grain shape diversity (Duan et al. 2017), promoter variation in ZCN8 and ZmCCT in maize associated with flowering time and photoperiod adaptation (Guo et al. 2018; Yang et al. 2013), promoter SNPs in ZmPAT7 associated with tassel branch number (Guan et al. 2022), and promoter variation in HAN1, PH8, and bZIP68 associated with cold response and adaptation all indicate that these traits are themselves subject to promoter dosage control (Mao et al. 2019; Chen et al. 2025; Li et al. 2022). Although these natural variants are not editing outcomes, they help identify dosage-sensitive genes and trait categories suitable for promoter editing and subsequent expression-window design.

Overall, promoter editing studies have already developed two representative capacities. The first is the active construction of continuous spectra of gene expression and phenotype through multi-site or multi-gene editing of regulatory regions. The second is the resetting of target gene expression to a more optimal window through directed modification of key cis-elements or core regulatory segments under clearly defined breeding objectives, thereby enabling coordinated optimization of complex traits such as plant architecture, yield, growth duration, and quality. The reason why the promoter has become the most mature and engineering-accessible level of precise cis-element editing is that it preserves protein function while transforming expression dosage itself into a tool for the design, selection, and application of quantitative trait improvement. More recently, Li et al. (2026) showed that in situ promoter truncation using a prime-editing library can also generate tunable expression outputs in rice, further extending promoter editing from local cis-element deletion to the systematic reconstruction of endogenous promoter length and architecture.

Regulation of Quantitative Traits by Chromosomal Structural Variation

Point mutations in promoters or local cis-regulatory elements mainly affect transcription factor binding. By contrast, structural variations—including deletions, insertions, duplications, inversions, and translocations—can remodel the spatial organization of cis-regulatory networks and thereby influence quantitative traits. SVs can alter local sequences and reshape regulatory architecture. They may reconfigure linear relationships and three-dimensional chromatin interactions among enhancers, silencers, and promoters. These changes can affect the expression level, timing, and tissue specificity of target genes. For such traits, the main question is whether regulatory connections have been disrupted, redirected, or reinforced. Therefore, SV studies have driven an important shift in perspective: the target of cis-element editing is no longer limited to a single site, but has been further extended to distal regulatory segments, topological boundaries, and regulatory connectivity itself.

Based on current studies, SV-mediated variation in quantitative traits can be broadly summarized into three interrelated modes. The first involves the direct alteration of the presence or absence of key cis-regulatory regions through deletions or insertions. The classical fw2.2 study showed that fruit-size differences between cultivated and wild tomato mainly result from non-coding regulatory variation rather than coding-sequence changes (Frary et al. 2000). It therefore established that key determinants of quantitative traits can reside in regulatory regions. Beyond identifying an early quantitative trait locus, this study established a broader principle. When a phenotype depends on the presence or absence of a regulatory segment, segmental deletion, insertion, or replacement may be more appropriate than local point editing. A haplotype-resolved study in citrus further extended this logic to a higher level of resolution. It showed that haplotype-specific deletions and transposon-associated structural variations can remove or perturb distal regulatory elements, reduce chromatin accessibility in the promoter regions of target genes, downregulate their expression, and thereby influence quality traits such as fruit acidity (Diaz et al. 2025). This indicates that when a trait is determined by whether a given regulatory segment is present, the most appropriate editing target is often the segment itself rather than any single base within it.

The second mode involves the alteration of the relative positions and copy numbers of regulatory elements through duplications, inversions, or complex rearrangements, thereby triggering regulatory rewiring. What is truly being modified here is the pairing relationships between enhancers and promoters. In maize, complex rearrangements mediated by the Ac/fAc transposon system can redeploy an enhancer originally acting on p1 to the vicinity of p2, strengthening its spatial interaction with the promoter and activating ectopic expression, thereby enhancing pericarp pigmentation (Sharma and Peterson 2023). Such phenotypes cannot be readily simulated by point mutations, because the key issue is not the activity of a binding site within the enhancer itself, but rather the fact that the enhancer has been repositioned. The study of DG1 in sorghum similarly illustrates this principle. A 35.7-kb inversion in its promoter region can alter the local repressive epigenetic state, upregulate DG1 expression, and ultimately promote double-grain spikelet formation and increase yield (Zhang et al. 2025b) For such traits, if point editing is still regarded as the principal strategy, it is often difficult to target the functionally decisive genomic regions. In contrast, large-fragment deletion, boundary reconstruction, or artificial rearrangement more closely reflects the biological nature of the problem, because what needs to be rebuilt is the original spatial regulatory connectivity.

The third mode centers on the remodeling of local or even higher-order chromatin topology, thereby altering the interaction probability between distal regulatory elements and their target genes. This direction most clearly reflects the transition in the field from the dissection of natural variation to validation through targeted editing. Taking the upstream region of Ghd7 in rice as an example, researchers used systematic CRISPR deletion scanning across a 65-kb region to identify a distal cis-regulatory region responsible for red-light induction. Deletion of this region significantly weakened the light responsiveness of Ghd7 and advanced heading under long-day conditions (Ogo et al. 2025). The importance of this study lies in the fact that it transformed a structure-level regulatory unit, which might otherwise have remained merely a candidate region, into a functional module that could be causally tested through segment-level editing. In tomato, the combination of assay for transposase-accessible chromatin with sequencing (ATAC-seq) and functional validation showed that an open chromatin region within an intron of MBP3 can regulate fruit development and fruit-size-related traits. Deletion of this region altered the local regulatory state and gene expression, further affecting phenotypes such as fruit weight and fruit shape (Cheng et al. 2025). Although this open chromatin region is not a strictly defined natural structural variation, the result shows that some traits depend on an entire functional segment and its local interaction pattern. In such cases, deletion or local reconstruction of the segment may be more informative than editing a single site. Similarly, in Arabidopsis thaliana, polymorphic inverted repeats near genes can serve as local three-dimensional interaction anchors that alter chromatin topology, thereby affecting the expression of neighboring genes as well as phenotypes such as flowering and morphogenesis. These findings indicate that structural variation can reshape chromatin topology and affect phenotypic variation across plant species. (Arce et al. 2023).

Inspired by these advances, segment-scale genome editing has increasingly been applied to directly mimic or generate beneficial structural variations. Research on OsDREB1C provides a representative example. Previous studies showed that transgenic overexpression of the OsDREB1C coding sequence under strong constitutive promoters can enhance its expression and improve yield-related traits. However, such approaches bypass the native regulatory context and may not fully preserve endogenous spatiotemporal expression patterns. Luo et al. (2025) therefore moved beyond conventional transgenic overexpression and used CRISPR-Cas9 to engineer a structural variation affecting the regulatory architecture of OsDREB1C, thereby increasing its expression in a transgene-free manner. This approach enhanced its endogenous expression and ultimately increased both plant-level and plot-level yield without compromising grain quality. A comparable logic can be seen in the work of Lu et al. Rather than elevating HPPD expression through transformation driven by an exogenous strong promoter, they used CRISPR-Cas9 to induce large-scale genomic duplication, thereby creating a novel promoter_Ubiquitin2::HPPD linkage within the endogenous rice genomic context. This substantially increased HPPD expression and conferred resistance to HPPD-inhibiting herbicides (Lu et al. 2021). Taken together, these cases suggest that the value of structural variation engineering lies not simply in producing large genomic changes, but in optimizing expression output by reorganizing cis-regulatory relationships while retaining the endogenous regulatory context. For the improvement of quantitative or application-oriented traits, this strategy is generally more controllable than forced enhancement detached from the native cis-regulatory background and more consistent with the stable expression states required in breeding practice.

In polyploid crops, SV-mediated regulation of quantitative traits is often even more intricate, and this complexity more vividly highlights the network-level nature of such regulation. Studies in cotton have shown that a 1,133-bp presence/absence variation (PAV) deletion occurs upstream of the GhPH1 locus associated with plant height. This deletion functions as a distal silencer module that regulates GhPH1 expression and influences plant height in coordination with the upstream regulatory factor GhGARF and hormone signaling pathways (Tian et al. 2024) Population-scale analyses in rapeseed have further revealed that SVs widely distributed across the population can extensively reshape gene expression networks and are closely associated with key agronomic traits such as plant height and yield (Zhang et al. 2024a). Taken together, these findings suggest that in polyploid genomes, the significance of SVs extends beyond explaining natural variation. They also provide a theoretical foundation for future cis-segment editing and the reconstruction of spatial regulatory connectivity at subgenome resolution.

Chromosomal structural variation has emerged as an important non-coding genetic basis of quantitative traits. Its effect extends beyond changes in genomic fragment length and includes reorganization of cis-regulatory architecture, regulatory interactions, and downstream transcriptional outputs. Unlike promoter point mutations, which mainly affect local binding strength, structural variations reshape the regulatory relationships between cis-elements and their target genes. Accordingly, the significance of structural variation research goes beyond explaining natural genetic variation. It also drives a shift in cis-element engineering from local sequence optimization to segmental, network-level, and spatial reconstruction, thereby providing a new theoretical framework for advancing crop quantitative trait improvement from functional dissection to rational design.

Regulation of Quantitative Traits by Untranslated Regions

Although UTRs do not encode proteins, they represent an important layer of post-transcriptional regulation in the control of quantitative traits. Unlike promoters and distal enhancers, which primarily determine transcription initiation and transcriptional output, the 5′ UTR and 3′ UTR more directly influence protein production by regulating translation initiation efficiency, ribosome scanning, mRNA stability, RNA localization, and conditional degradation. As a result, UTRs play a critical role in shaping protein accumulation. Accordingly, UTR editing can fine-tune protein production and dynamic gene output without changing the protein sequence. For quantitative traits that are sensitive to strong transcriptional perturbation but still require continuous dosage adjustment, UTR editing may be preferable to coding-region editing. It preserves protein function and is less likely to cause complete loss of function or excessively strong effects. The case of TaSPL13 in wheat provides a representative example. This gene affects grain size, grain number, plant height, tillering, and heading date simultaneously. Direct manipulation of its coding region would likely alter multiple phenotypes at once, making it difficult to achieve a practically useful balance. Gupta et al. (2023) therefore precisely edited the miR156 recognition site within its 3′ UTR, thereby altering the strength of post-transcriptional repression rather than disrupting protein function itself, and ultimately achieved coordinated improvement of multiple agronomic traits. A similar principle is seen in targeted editing of embedded inhibitory regions within plant gene 3′ UTRs, which can produce mild and controllable upregulation of expression without introducing exogenous strong promoters or generating constitutive overexpression lines. This suggests that 3′ UTR editing is, in essence, a way of resetting expression output while preserving the endogenous regulatory context (Wang et al. 2024).

From a mechanistic perspective, the 5′ UTR and 3′ UTR contribute differently to quantitative trait improvement. The 5′ UTR primarily influences translation initiation efficiency and the threshold of protein output, and one of its most engineering-relevant regulatory features is the uORF. The 3′ UTR, in contrast, more often affects mRNA stability, degradation, and spatiotemporal distribution through microRNA recognition elements, embedded inhibitory fragments, and artificially introduced regulatory sequences. Accordingly, 5′ UTR editing is better suited to adjusting protein accumulation, whereas 3′ UTR editing is more effective for reshaping post-transcriptional response patterns of mRNA. The rice Waxy (Wx) gene illustrates why certain quality traits are more appropriately regulated through the 5′ non-coding region than through direct manipulation of the coding sequence. The Wx coding region directly determines the function of granule-bound starch synthase. Direct knockout or severe weakening of this region can indeed rapidly alter amylose content, but the resulting quality phenotype is often shifted too far, making it difficult to obtain intermediate materials suitable for improving eating and cooking quality (Zhao et al. 2024). To overcome this limitation, Zeng et al. (2020) focused not on the Wx coding region, but on its promoter and the first intron within the 5′ UTR. By editing these regions simultaneously, they quantitatively reduced Wx expression and generated a series of allelic variants with different amylose contents, thereby enabling continuous adjustment of eating and cooking quality. A related pattern is evident in OsCSLC3, which is associated with hemicellulose accumulation, disease resistance, and multiple agronomic traits. Strong perturbation of its coding region could readily disrupt cell wall-related physiological processes. Shi et al. (2025) therefore targeted a conserved regulatory region within the 5′ UTR of OsCSLC3, increasing its expression and promoting hemicellulose accumulation in vascular tissues, while also enhancing blast resistance and improving traits such as plant height and tillering. These findings suggest that when the goal is to preserve core protein function while redistributing expression levels or tissue-specific output, the 5′ UTR can provide a more suitable editing entry point than either the coding region or the promoter.

Among the various cis-regulatory features within the 5′ UTR, uORFs are currently among the best-characterized mechanistically and the most amenable to engineering. Their significance is rooted first in a well-defined biological basis. Classical studies of the vitamin C (VC) biosynthetic pathway showed that the GGP-associated uORF participates in a feedback repression loop and defines the threshold and dynamic range of translational output from the main ORF (Laing et al. 2015). This mechanistic insight was later translated by gene-editing studies into a practical engineering strategy. Zhang et al. (2018) showed that editing endogenous uORFs can reprogram translation of downstream main ORFs. For genes controlled by translational thresholds, the uORF may therefore be a more suitable target than the promoter or coding region. Further support came from the work of Xue et al., who used base editing and prime editing to generate de novo uORFs or extend existing uORFs, thereby achieving graded downregulation of translational output and converting dosage variation into continuous phenotypic variation (Xue et al. 2023). In strawberry, editing of a conserved sucrose-control uORF (SC-uORF), combined with alleles of different strengths, produced stably inherited gradients of sugar content, highlighting the utility of uORF editing for constructing expression gradients underlying quantitative traits (Xing et al. 2020). The same principle has also been demonstrated in major food crops. In rice, Sang et al. (2025b) identified and edited a functional uORF in OsGGP, generating seven distinct OsGGP uORF allelic variants. These variants did not significantly alter OsGGP transcript levels; instead, by weakening uORF-mediated repression of main ORF translation, they increased vitamin C content to approximately 1.8- to 3.1-fold that of the wild type and simultaneously enhanced tolerance to osmotic stress. This case is particularly informative because it shows that, for traits such as vitamin C content that integrate nutritional quality with stress responsiveness, uORF editing offers a way to modulate translational output without perturbing the broader transcriptional network, thereby improving both quality and physiological adaptability.

In comparison, 3′ UTR engineering more directly centers on the precise control of mRNA stability and spatiotemporal response patterns. At present, two major strategies have become relatively well defined. One is to alleviate post-transcriptional repression by editing microRNA recognition elements (MREs). The other is to identify and modify embedded inhibitory regions within the 3′ UTR, or to introduce regulatory fragments artificially, thereby turning them into breedable modules for expression control. The former has already been exemplified by the TaSPL13 case, which showed that even modest changes in post-transcriptional repression can produce coordinated improvement across multiple traits (Gupta et al. 2023). The latter was further established in a study as a general strategy for the stable generation of gene knock-up alleles (Wang et al. 2024). From an application perspective, work on rice OsHPPD further suggests that 3′ UTR editing can modify traits such as herbicide resistance while preserving coding-region function. Wu et al. (2023) neither enhanced OsHPPD expression through direct manipulation of the coding region nor relied on an exogenous strong promoter. Instead, they used targeted CRISPR-Cas9 editing of the OsHPPD 3′ UTR to generate rice lines with markedly enhanced resistance to HPPD-inhibiting herbicides. This strategy increases endogenous expression by reshaping post-transcriptional regulation at the 3′ UTR. It avoids potential risks to protein function and preserves the native regulatory context. In addition, the introduction of specific miRNA target sequences into the 3′ UTR makes it possible to build conditional and spatiotemporally restricted gene downregulation systems. Here, 3′ UTR editing is useful for restricting gene expression to specific tissues, developmental stages, or response conditions. This strategy is preferable when the aim is context-specific regulation rather than permanent activation or repression. This design improves regulatory precision while reducing system-level costs (Liu et al. 2025a).

Taken together, the contribution of UTR studies to quantitative trait improvement lies not simply in adding several new examples of non-coding regulation, but in defining, at the mechanistic level, a regulatory layer distinct from both promoters and structural variation. UTRs regulate neither transcription initiation itself nor the spatial relationships among distal regulatory elements; instead, they act on RNA behavior and translational efficiency, placing their effects closer to the output end of gene expression. For genes that are highly sensitive to transcriptional perturbation, or that incur substantial costs after coding-region editing, UTRs often offer a more effective route for continuous dosage adjustment. As approaches such as conserved 5′ UTR region editing, uORF engineering, MRE editing, 3′ UTR inhibitory-region modification, and conditional knockdown continue to mature, UTRs are no longer viewed merely as passive non-coding regions harboring natural variation. Rather, they are increasingly being recognized as post-transcriptional control interfaces that can be deliberately designed and strategically exploited. In this sense, quantitative trait improvement is moving beyond the simple use of non-coding variation toward precise editing and rational design guided by post-transcriptional regulatory mechanisms.

Regulation of Quantitative Traits by Cis-Regulatory Elements in Intronic Regions

Introns are not traditionally the primary cis-regulatory regions of research focus, but increasing evidence has shown that they can harbor silencer elements, enhancer elements, splicing regulatory sequences, as well as intron-mediated enhancement (IME) elements that affect expression efficiency, thereby jointly shaping gene output at both the transcriptional regulatory and post-transcriptional processing levels. Promoters mainly control transcription initiation, whereas UTRs primarily affect post-transcriptional output. Intronic cis-regulatory elements can regulate both expression level and transcript composition. They are therefore useful when protein function must be preserved while expression or splicing is fine-tuned (Xie et al. 2018; Laxa 2017).

Functionally, intronic cis-elements regulate quantitative traits mainly through two routes. The first is the direct control of transcriptional strength by silencer or enhancer elements located within introns. A typical example is Eui1 in rice. This gene is involved in the regulation of internode elongation, so direct knockout or forced overexpression of its coding region often causes obvious abnormalities in plant height, making it difficult to obtain useful intermediate phenotypes. Xie et al. (2018) showed that a 34-bp intronic cis-silencer element, SE1, in the first intron of Eui1 can recruit transcriptional repressor complexes and associated histone modification factors through its core RY motif, thereby keeping Eui1 expression at a low level. When this element was deleted, Eui1 transcription increased significantly, leading to reduced internode elongation, dwarfism, and enclosed panicle phenotypes. This case suggests that for genes in which protein function must be preserved but expression level needs to be adjusted, the better target is a local silencer element within the intron. Editing this kind of cis-element makes it possible to retune gene expression without disrupting protein function, and thus to generate phenotypes more suitable for quantitative trait improvement.

The second route is the alteration of functional transcript composition, and thereby quantitative traits, through regulation of splicing efficiency and the proportion of alternatively spliced products. In this case, trait variation depends more on transcript composition than on total expression level. Therefore, merely adjusting promoter strength cannot solve the problem, whereas coding-region knockout is even more likely to directly destroy protein function. The case of GS3 in rice clearly illustrates this point. Liu et al. (2022a) found that GS3 can produce two major transcript isoforms, GS3.1 and GS3.2, through alternative splicing, and that differences in splicing structure alter downstream signaling output and further affect grain size (Liu et al. 2022a). For such traits, what truly needs to be regulated is not whether GS3 is expressed, but the proportional relationship among different transcript isoforms. Therefore, introns and their associated splicing regulatory regions naturally constitute a more appropriate editing entry point. In other words, only by intervening at the intron level in splicing regulation is it possible to achieve continuous adjustment of grain shape while preserving the basic function of the gene.

Studies on Waxy (Wx) in rice further indicate that intron editing has particularly obvious advantages in the improvement of quality traits. The Wx coding region directly determines the function of granule-bound starch synthase; therefore, if the coding region is simply knocked out, amylose content may indeed be rapidly altered, but the resulting phenotypic perturbation is often too strong, making it difficult to obtain intermediate materials suitable for improving eating quality. Based on this, Zeng et al. (2020) did not directly edit the Wx coding region, but instead jointly edited its promoter and the first intron in the 5′ UTR, thereby quantitatively downregulating Wx through changes in expression and splicing efficiency, and causing both amylose content and cooking quality to show continuous variation. Subsequently, Liu et al. (2022b) further demonstrated that direct deletion of the first intron of the Wx^b allele can significantly increase amylose content and generate a new Wx allelic variant. Together, these two studies show that for quality genes such as Wx, where strong perturbation of the coding region easily leads to obvious side effects, intron editing is especially necessary precisely because it can generate continuous and usable quality phenotypes by adjusting splicing efficiency and expression level without disrupting the basic enzymatic function.

In addition to directly serving quantitative trait improvement, the application of intronic cis-elements in expression systems and the optimization of gene-editing platforms also demonstrates their engineering value from another perspective. Studies on IME have shown that introns derived from different sources can significantly enhance heterologous gene expression, among which the introns of maize Adh1, rice Act1, and Arabidopsis UBQ10 have become commonly used functional modules in plant expression systems (Laxa 2017). Furthermore, Zhong et al. (2020) developed an intron-based single-transcript-unit CRISPR system, in which the gRNA module was embedded within plant introns to enable efficient splicing and gRNA release; Nguyen et al. (2022) used the first intron of αAmy3 to establish a rice expression system based on intron-targeted insertion. Although these studies do not directly correspond to a specific agronomic quantitative trait, they clearly demonstrate that intronic cis-elements are not passively existing sequences, but engineering modules that can be actively designed and utilized. For quantitative trait research, this means that introns are not merely a “regulatory layer that needs to be interpreted”, but also a “technical layer that can be organized into expression and editing systems”.

Intronic cis-regulatory elements provide precise regulatory targets for quantitative trait improvement. They can modulate both expression levels and transcript isoform composition. For genes in which direct coding-region editing readily leads to loss of function, whereas promoter adjustment alone cannot adequately address problems of splicing or transcript balance, introns often provide a more rational entry point for targeted editing. Precisely for this reason, intron editing extends the study of natural regulatory variation by converting complex intronic regulatory modules into engineering targets that can be deliberately designed. Segment deletion, splice-site editing, intron-embedded expression systems, and intron-based CRISPR platforms have expanded the engineering use of intronic cis-regulatory elements. These elements now provide a link between functional dissection, precise editing, and intelligent design in crop improvement (Zhong et al. 2020; Nguyen et al. 2022).

Evolution of Technical Paradigms: From Blind Mutagenesis to Targeted Attempts

Based on the above regulatory mechanisms by which cis-regulatory elements control quantitative traits, the development of CRE editing technologies has consistently revolved around the core question of how to accurately identify and edit functional CREs. Overall, as the functions of cis-regulatory elements become better understood and genome editing becomes more precise, research has expanded from broad blind screening to targeted functional analysis at defined sites. This shift reflects improved capacity for precise analysis and rational design, rather than the complete replacement of blind mutagenesis by targeted editing. The overall transition from blind mutagenesis to targeted functional analysis and precision CRE editing is illustrated in Fig. 2. Importantly, this trend does not imply the sequential replacement of one strategy by another. Blind mutagenesis and precise targeted editing continue to be used in parallel, depending on the availability of regulatory information and the specific research objective.

Fig. 2.

Evolution of CRE editing from blind mutagenesis to functional dissection and intelligent design. a Random mutagenesis uses CRISPR-Cas9 libraries and high-throughput screening to generate large mutant pools. b Targeted functional analysis then integrates eQTL mapping, epigenetic profiling, and comparative genomics to identify candidate CREs and extract functional insights. c Precision CRE editing further combines AI, machine learning, MPRA assays, and CRISPR-Cas9 knock-in strategies for predictive modeling, efficient editing, and quantitative validation. Together, panels a–c depict a left-to-right transition from broad screening to functional dissection and finally to smart design

Blind Mutagenesis: Initial Validation of Technical Feasibility

During the initial phase of “blind mutagenesis”, in the absence of a systematic understanding of functional sites within CREs, studies predominantly employed CRISPR-Cas9-mediated random mutagenesis and near-saturation editing approaches. By designing multiple single-guide RNAs (gRNAs) to target candidate regulatory regions, researchers generated a large spectrum of mutants with varying mutation types and lengths, which were subsequently subjected to phenotypic screening to identify variants exhibiting the desired traits. Jacobs et al. (2017) established a pooled CRISPR library system in tomato, demonstrating the feasibility of conducting large-scale parallel mutagenesis and recovering mutants in crops using mixed CRISPR libraries, thus providing a technical foundation for the comprehensive exploration of non-coding regulatory regions.

As a representative example of cis-regulatory editing for quantitative trait improvement, Rodríguez-Leal et al. (2017) used multi-target CRISPR-Cas9 mutagenesis to edit the promoters of tomato CLV3, S, and SP. The resulting allelic series showed continuous variation in fruit size, branching, and plant architecture. This study demonstrated that cis-regulatory regions can provide selectable variation for quantitative traits (Rodríguez-Leal et al. 2017). Cui et al. (2020) used multiplex sgRNAs to edit the promoter and 5′ non-coding region of SCM3/OsTB1 in rice. This generated cis-allelic variants with different expression levels and continuous variation in culm thickness, panicle size, and tiller number. The results show that multi-site perturbation can produce useful phenotypes even when functional sites are not fully defined.

The key contribution of this stage extends beyond the mere accumulation of individual successful cases; it addresses a fundamental yet critical question: can CREs, like coding regions, serve as operable, selectable, and breeding-relevant sources of genetic variation? Rodríguez-Leal et al. first demonstrated that editing cis-regulatory regions can directly generate continuous variation in quantitative traits. Subsequent work by Cui et al. further revealed that, even in the absence of precise knowledge about functional sites, multi-point perturbations of proximal regulatory regions are capable of establishing expression and phenotypic gradients of practical significance. For this reason, although the blind mutagenesis stage retained a substantial trial-and-error component, it nonetheless established both the technical and conceptual groundwork for subsequent precision optimization guided by functional dissection.

However, the limitations of this strategy were also very pronounced. First, this approach generally requires the construction and screening of relatively large mutant populations, which can make the process labor-intensive. Second, when functional sites are not defined before editing, the causal relationships between specific mutations and phenotypic outcomes may remain difficult to resolve. Third, target selection in such workflows often depends on empirical screening rather than prior knowledge of functional regulatory sites. Fourth, unintended changes in non-coding regulatory regions may require additional validation to distinguish causal effects from collateral phenotypic variation. Therefore, although the blind mutagenesis stage answered the basic question of whether CREs can be edited to generate usable phenotypes, it was insufficient to support higher-resolution and higher-efficiency rational design. Consequently, research emphasis gradually shifted toward candidate-region narrowing and targeted validation supported by functional genomics evidence.

Targeted Attempts: Precision Optimization Driven by Functional Dissection

Earlier approaches relied on large mutant populations and phenotypic screening. In the targeted-attempt stage, functional genomics was used to narrow candidate regions and guide precise interventions in known or predicted regulatory modules. The essential change was not merely an increase in editing precision, but rather a shift in research logic from “generate large numbers of variants first and then search for useful phenotypes” to “identify high-value regulatory sites first, and then validate and optimize them”. During this stage, evidence such as open chromatin maps, conserved non-coding sequences, transcription factor binding information, three-dimensional genome interactions, and association analyses of natural variation was gradually integrated into the target-selection workflow, enabling CRE editing to acquire explicit mechanistic hypotheses and verifiable design rationales.

Within this framework, promoters, distal regulatory regions, UTRs, introns, and other layers of cis-regulatory elements were no longer regarded merely as parallel “editable regions”, but were instead understood as components of a unified expression-regulatory network. Researchers began to focus not only on which DNA segment can be edited, but more importantly on in which tissue, at which developmental stage, and through which mechanism that element influences expression output, and whether such regulation is sufficient to generate stable and usable quantitative trait improvement. It was also at this stage that CRE editing gradually moved beyond the limitations of purely empirical trial and error and began to advance toward mechanism-driven precision optimization.

Natural variation contains abundant functional CRE alleles, and eQTL mapping, by associating genotype with gene expression level, can identify cis-regulatory variants affecting expression. For quantitative traits, such information helps identify regulatory regions that alter expression output at specific developmental stages or under particular environmental conditions. The expression dynamics QTL framework proposed by Kashima et al. showed that the effects of regulatory loci are highly dynamic and can vary with field environments and developmental stages, thereby providing a new route for identifying candidate CREs with spatiotemporal specificity (Kashima et al. 2021). In rice, recent regulome studies have further integrated chromatin accessibility and expression data to characterize functional regulatory regions at high resolution, indicating that candidate CRE selection is moving from single lines of evidence toward multi-layer information integration (Zhu et al. 2024). In addition, the introduction of analyses of nascent RNA and enhancer transcription has extended the identification of the rice cis-regulatory landscape from mere “accessibility” to the level of “activity” (Goliasse et al. 2025).

Epigenetic modifications and chromatin states are important markers of CRE activity, and the integration of ATAC-seq, DNase I hypersensitive sites sequencing (DNase-seq), chromatin immunoprecipitation sequencing (ChIP-seq), and three-dimensional genomics technologies has greatly improved the precision of functional regulatory region identification. Zhu et al. (2015) demonstrated in Arabidopsis that intergenic enhancers can be predicted and validated genome-wide based on open chromatin features, indicating that chromatin accessibility itself can serve as an important entry point for enhancer identification in plants. In maize, Ricci et al. (2019) further used ATAC-seq, methylation, and long-range interaction information to systematically identify distal cis-regulatory elements and dissect their spatial connections with target genes, thereby providing a complete technical route for CRE localization in crops with complex genomes. In rice, Tian et al. (2022) established a predictive framework for activating regulatory elements based on open chromatin features and performed functional validation in protoplasts, indicating that similar strategies have already extended from model plants to major food crops.

Comparative genomics and transcription factor binding motif analyses have further provided the theoretical basis needed for targeted editing. Swinnen et al. (2016) systematically summarized the logic of using cis-regulatory elements in crop improvement and pointed out that conserved motifs and their combinatorial organization can serve as priority intervention targets, especially for fine regulation of complex traits such as stress resistance. In stress-response regulation, the study on the synergistic action of ABRE and DRE/CRT showed that the combinatorial relationship between cis-elements itself can markedly alter the strength of stress-induced expression, thereby providing direct mechanistic support for subsequent targeted editing of conserved stress-responsive CREs (Narusaka et al. 2003). In hypoxia responses, Gasch et al. (2016) demonstrated that ERF-VII transcription factors can bind a conserved cis-motif to mediate low-oxygen responses, further indicating that plant stress responses contain conserved regulatory modules that can be captured by comparative genomics and motif analysis.

High-throughput technologies such as massively parallel reporter assays (MPRA) have allowed functional validation of CREs to move from low-throughput, one-by-one testing toward parallel interrogation. Voichek et al. (2024) conducted multi-species MPRA studies in plants and showed that the activity of cis-regulatory sequences depends not only on the motif itself, but also strongly on its position and surrounding context, thereby providing a higher-throughput experimental basis for deciphering regulatory “grammar”. Deng et al. (2025) further carried out large-scale MPRA in tomato tissues to test the enhancer activities of tens of thousands of candidate sequences, indicating that high-throughput screening of plant CREs has advanced from proof of concept to a systematic stage. Meanwhile, rice protoplast systems have also been widely used for rapid testing of candidate cis-elements. The highly efficient transient expression system in rice green tissue protoplasts established by Zhang et al. (2011) provides a practical platform for rapidly comparing the activity of candidate promoters or enhancers in rice.

Studies during the targeted-attempt stage significantly improved the precision and efficiency of CRE editing. In crop breeding applications, Li et al. (2023) targeted the promoter of the rice heading-date gene Ehd1, constructing a set of weakened cis-allelic variants that moderately delayed heading and improved yield-related traits, indicating that this stage already enabled the construction of different levels of expression output within known candidate regions around specific breeding objectives. Similarly, based on functional element information, Ludwig et al. (2024) precisely edited the ARR1AT element in the OsNAS2 promoter, increasing Zn uptake and translocation efficiency while improving plant yield under endogenous regulatory context, further exemplifying the transition from “functional dissection” to “target optimization” at this stage. More recently, a large-scale study in sorghum systematically mapped the expression effects of CRISPR-accessible mutations across native promoters and 5’ UTRs, providing a high-resolution mutation-effect atlas for prioritizing editable CRE sites and quantitatively designing regulatory variants (Groover et al. 2026).

Nevertheless, this stage still had intrinsic limitations. First, research remained largely hypothesis-driven, requiring extensive experimental validation of candidate CREs and still lacking the ability to stably and quantitatively predict editing outcomes. Second, the interactions among CREs and the regulatory grammar governing them were still insufficiently elucidated, making it difficult to precisely predict the synergistic effects of multi-target editing before experimentation. Third, functional dissection of CREs in complex crops, especially in polyploid backgrounds, remained inadequate, thereby limiting the universality of the technology. For this reason, CRE editing has further advanced toward a data-driven intelligent design paradigm. Intelligent design does not replace experimental validation. Rather, it changes where validation occurs in the research workflow. Instead of relying on broad trial-and-error screening at an early stage, experiments are primarily used for functional testing and iterative refinement of model predictions.

Intelligent Design Paradigm: Precision Editing Driven by Multi-omics and AI

Blind mutagenesis can test whether editing a regulatory region generates useful variation, whereas targeted approaches focus on which elements should be edited. Intelligent design additionally seeks to determine the appropriate magnitude of regulation and the most suitable editing strategy. This supports more precise, quantitative, and predictable control of quantitative traits. A schematic overview of the intelligent design framework is provided in Fig. 3. This paradigm is not a simple superposition of previous technologies, but rather a new-generation CRE editing system that is data-driven, model-guided, and designed across the entire pipeline, formed on the basis of overcoming the limitations of traditional hypothesis-driven and trial-and-error screening approaches.

Fig. 3.

Multi-omics-driven intelligent framework for CRE prediction and programmable promoter engineering. a High-resolution CRE landscapes are generated from multi-omics datasets, including ATAC-seq, ChIP-seq, Hi-C, and TF-binding information. b Integrated datasets are used to train predictive models for CRE functional prediction and target selection. c The CAPE system supports prediction-guided promoter editing and QTV generation through feature selection, key-region estimation, crRNA design, and Cas12a-mediated promoter editing. d The DIAL system supports programmable and heritable transgene expression control through editable promoter design, spacer assembly, recombinase-mediated editing, and expression setpoint tuning. Together, these modules outline a general workflow from multi-omics-informed CRE discovery to functional validation and synthetic promoter engineering

Core Driving Forces of the Paradigm Shift

The rapid development of high-throughput omics technologies, bioinformatics, and machine learning algorithms has provided three core supports for the intelligent design of CRE editing. Omics technologies such as ATAC-seq, ChIP-seq, high-throughput chromosome conformation capture (Hi-C), and single-cell transcriptomics provide data for dissecting CRE function. Bioinformatics tools integrate these datasets and identify functional regulatory modules and their interactions. Machine learning models then use sequence, epigenetic, and expression features to predict regulatory function. In cotton, Song et al. (2017) integrated genomic, three-dimensional genomic, DNA methylation, and transcriptomic data to systematically analyze the functions of epialleles during domestication, demonstrating the powerful potential of multi-omics integration. In their review, Scheben and Edwards (2017) explicitly proposed that the integration of genomics, bioinformatics, and gene editing represents a core route for shifting breeding from “trial and error” to “predictive design”.

Key Technologies and Applications of Intelligent Design

The first step in intelligent design is to integrate multidimensional omics data to construct a high-resolution functional map of CREs. ATAC-seq data indicate CRE accessibility. ChIP-seq profiles help distinguish promoters from enhancers. Hi-C maps long-range interactions between CREs and their target genes. eQTL data link regulatory variation to expression differences. Sequence conservation helps identify evolutionarily constrained regulatory regions. By integrating these data, Pu et al. (2025) established predictive models for enhancer-like elements in Arabidopsis and rice, providing a methodological foundation for functional map construction. Dong et al. (2017) used Hi-C technology to generate three-dimensional genome maps in tomato and rice, revealing the distribution patterns of chromatin compartments and topologically associated structures, and thus providing key data for understanding the regulatory mechanisms of distal CREs.

Based on integrated multi-omics data, the training of predictive models using machine learning algorithms constitutes the core step of intelligent design. Commonly used models include random forests, support vector machines, and deep learning networks. By learning the relationships among CRE sequence features, epigenetic marks, and gene expression levels, these models can quantitatively predict the regulatory strength, specificity, and variant effects of CREs. In wheat, Liu et al. integrated multi-omics data and used machine learning algorithms to predict key promoter regions regulating yield, thereby providing targets for precise editing (Ma et al. 2025). A team used AI models to predict key cis-regulatory elements in the RBL1 promoter and successfully achieved coordinated improvement of broad-spectrum disease resistance and yield stability. This is a typical outcome of the combined application of generative AI and explainable AI, demonstrating that intelligent design has already moved from “predicting targets” toward “customizing traits” (Han et al. 2025).

At the level of general-purpose genomic foundation models, AlphaGenome and Evo 2 have further expanded the potential of artificial intelligence for analyzing and designing cis-regulatory sequences. Avsec et al. introduced AlphaGenome, a unified model that predicts functional genomic signals directly from DNA sequence. The model can process DNA sequences of up to 1 Mb and generate predictions at resolutions as high as a single nucleotide. It simultaneously predicts gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, three-dimensional chromatin interactions, and RNA splicing (Avsec et al. 2026). Unlike conventional models that mainly identify individual regulatory features, AlphaGenome can assess multiple molecular effects of a candidate variant within a single framework. For cis-regulatory element editing, its main value lies not in directly generating new regulatory sequences, but in providing a sequence-to-function prediction framework. It could help prioritize candidate editing sites and compare the potential effects of different editing strategies on gene expression, chromatin state, and RNA splicing. However, AlphaGenome was trained on human and mouse genomes and their associated functional genomics datasets. It therefore does not currently provide direct predictions for plant genomes. Its relevance to crop regulatory element engineering lies mainly in its model architecture. In the future, plant multi-omics data could be used to develop similar sequence-to-function models specifically for plants or individual crop species. The current version of AlphaGenome should therefore not be regarded as a mature prediction tool for plant cis-regulatory elements.

Unlike AlphaGenome, which focuses primarily on functional prediction, Evo 2 combines genome sequence modeling, variant-effect prediction, and sequence generation (Brixi et al. 2026). Evo 2 was trained on approximately 9 trillion DNA bases from bacteria, archaea, and eukaryotes. Its context window can extend to 1 million bases. It can also assess the potential effects of variants in both coding and noncoding regions without task-specific fine-tuning. The training data include plant genomes, such as that of Arabidopsis thaliana. Evo 2 may therefore provide a general foundation for characterizing crop regulatory sequences, generating candidate variants, and fine-tuning plant-specific models. It can also be combined with external sequence-to-function models and inference-time search. This framework has been used to generate multi-kilobase DNA sequences with specified chromatin accessibility profiles. These results suggest that genomic language models are beginning to move beyond the interpretation of natural sequences toward the active generation of candidate regulatory sequences. However, experimental validation of these generated sequences has so far been conducted mainly in mammalian cells. Their performance in plants has not yet been systematically evaluated. General-purpose genomic language models may also underperform task-specific models trained on relevant functional genomics data, particularly for tasks involving distal regulatory elements. Evo 2 and AlphaGenome may therefore play complementary roles. Evo 2 can generate candidate sequences, whereas AlphaGenome can predict their potential functional consequences. When combined with plant-specific models, precise genome editing, and experimental validation, these tools could support a closed-loop workflow for cis-regulatory element engineering: design, prediction, editing, validation, and iterative refinement.

After these models identify candidate CREs, assess variant effects, or generate candidate regulatory sequences, an appropriate genome-editing tool must be selected. The choice should reflect the desired direction and magnitude of regulation, as well as the specific breeding objective. This step converts computational predictions and sequence designs into testable genetic variants. Different editing strategies can be used for different purposes: disruption of key CREs may be achieved through Cas12a-mediated deletion editing; precise single-base modification may employ cytosine base editors (CBEs) or adenine base editors (ABEs); enhancement of gene expression may be achieved using CRISPRa systems targeting enhancers; and reversible regulation may be carried out through dCas9-mediated epigenome editing. In potato, Veillet et al. (2019) compared the editing efficiencies of CRISPR-Cas9, CBE, and ABE, thereby identifying optimal tools for starch quality improvement. Hilton et al. (2015) developed a dCas9–p300 Core acetyltransferase system that activates endogenous gene expression through targeted acetylation of promoters and distal enhancers.

While general-purpose genomic foundation models such as AlphaGenome and Evo 2 have expanded sequence prediction and generation capabilities, CRISPR-Cas12a promoter editing (CAPE) system and a highly modular, extensible framework called dynamically editable artificial loci (DIAL) represent two approaches that are more directly relevant to plant regulatory element engineering. CAPE follows a prediction-guided editing strategy, whereas DIAL focuses on the construction of programmable regulatory sequences. Together, they demonstrate the potential of intelligent design in plants and provide representative examples of cis-regulatory element engineering based on distinct operational principles and implementation routes.

The core contribution of CAPE system lies in establishing a data-driven and predictive framework for promoter functional dissection (Zhou et al. 2023). CAPE links regulatory feature scoring with Cas12a-mediated promoter editing. The model integrates multiple features associated with promoter activity, including chromatin accessibility, transcription factor-binding motifs, sequence conservation, H3K27ac histone modification, and variation effects. These features are converted into aggregate scores to estimate the potential regulatory contribution of different promoter regions. Promoter regions with higher aggregate scores can then be prioritized as candidate key regions for experimental editing. For example, in the rice OsD18 promoter, CAPE-guided editing of predicted key regions generated a continuum of plant-height variation. This result shows that promoter key-region estimation can help guide targeted editing of quantitative traits. Thus, CAPE shifts promoter engineering from broad blind mutagenesis toward more targeted validation of candidate regulatory regions.

DIAL represents a more proactive and engineering-oriented approach to synthetic promoter design (Kabaria et al. 2025). Its operating principle is based on building editable promoter architectures that can generate graded and heritable expression states. In this framework, spacer sequences are placed between synthetic transcription factor-binding sites and the core promoter. Recombinase-mediated excision of these spacers changes the distance between the regulatory module and the core promoter, thereby producing a series of expression setpoints from a single promoter construct. This design logic differs from conventional editing of naturally occurring cis-regulatory variants. Instead, it emphasizes the rational construction of regulatory architectures whose output can be adjusted in a programmable manner. For crop regulatory engineering, this framework provides a useful conceptual reference for the future construction of synthetic promoters capable of stable and graded gene-expression control.

In summary, CAPE and DIAL follow different but complementary routes within the broader intelligent-design paradigm. CAPE extracts regulatory information from natural promoter regions and uses computational scoring to guide precise editing, whereas DIAL demonstrates how artificial promoter architectures can be rationally organized to generate programmable expression outputs. Together, these approaches point toward a transition in regulatory sequence engineering: from discovering and perturbing naturally existing variation toward designing regulatory architectures with predictable and adjustable functions.

Genome-Editing Platforms for Regulatory Sequence Engineering: Comparative Advantages and Emerging Compact Systems

Cis-regulatory elements are generally short, densely distributed, and highly position-dependent. Genome-editing systems differ in their protospacer-adjacent motif (PAM) or target-adjacent motif (TAM) requirements, editing activity, and delivery efficiency. The choice of editing tool should therefore be tailored to the target sequence and the intended application. The following section compares several widely used editing platforms.

SpCas9 remains one of the most established and widely used nucleases for editing regulatory sequences in plants. Its guide RNAs are relatively easy to design. SpCas9 can generate local insertions or deletions in promoters, UTRs, and other regulatory regions. Paired guide RNAs can also be used to delete larger regulatory fragments. In addition, targeting multiple positions within a promoter can generate allelic series with different levels of gene expression and trait variation (Rodríguez-Leal et al. 2017). However, SpCas9 generally requires an NGG PAM and induces repair through DNA double-strand breaks. The resulting insertions and deletions may therefore vary in both position and length.

Cas12a primarily recognizes T-rich PAMs and can therefore target A/T-rich regions that are difficult to access with SpCas9. It has been used for genome editing in rice and transcriptional repression in Arabidopsis thaliana (Tang et al. 2017). Cas12a can also process its own crRNA arrays, making it well suited for the simultaneous editing of multiple regulatory sites. Optimized plant Cas12a systems have enabled the targeting of up to 16 sites in rice (Zhang et al. 2021). However, editing efficiency may vary with the Cas12a ortholog, target sequence, and growth temperature.

Cas12b has PAM recognition preferences that differ from those of Cas9 and Cas12a, thereby expanding the range of editable sites in plant genomes. AaCas12b has been used for both single-site and multiplex editing in rice. It can also support transcriptional repression and activation (Ming et al. 2020). Cas12b-mediated targeted editing has also been demonstrated in Arabidopsis thaliana and tetraploid cotton (Wu et al. 2020; Wang et al. 2020). However, some Cas12b systems show pronounced temperature dependence. This may limit their direct application across different plant species and growth conditions.

TnpB, IscB, and Fanzor are compact RNA-guided nucleases that have attracted growing interest in recent years. IsDge10 TnpB has shown high genome-editing activity in rice (Zhang et al. 2024b). The small coding size of TnpB also facilitates its delivery using viral vectors. Tobacco rattle virus-mediated delivery of ISYmu1 TnpB has enabled heritable, transgene-free genome editing in Arabidopsis thaliana (Weiss et al. 2025).

IscB is approximately one-third the size of Cas9. Engineered IscB–ωRNA systems have enabled genome editing in rice and have also been adapted to construct cytosine and adenine base editors (Zhu et al. 2026). Fanzor is a compact RNA-guided nuclease of eukaryotic origin (Saito et al. 2023). Recent work has demonstrated the editing activity of several Fanzor systems in rice and wheat cells. Among them, NlovFz2 achieved editing efficiencies of up to 50.0% in regenerated rice plants. It was also used for targeted gene knockout and the deletion of regulatory sequences (Ji et al. 2025).

Overall, SpCas9, Cas12a, and Cas12b remain relatively mature tools for editing regulatory sequences in plants. TnpB, IscB, and Fanzor reflect the continuing development of smaller and more readily deliverable editing systems. Although these emerging tools have been validated in plants, their species range, targeting rules, editing specificity, and technical maturity require further evaluation. Their editing characteristics and suitable applications are summarized in Table 2.

Table 2.

Characteristics, comparative advantages, potential applications, and current plant status of representative genome-editing platforms for cis-regulatory engineering

Editing platform	Protein size	PAM/TAM requirement	Major editing outcomes	Comparative advantages	Suitable applications	Main limitations and status
SpCas9	1,368 aa	NGG	Insertions/deletions; fragment deletion	Mature system	Cis-element disruption; allelic series	PAM restriction; heterogeneous products
Cas12a	Approx. 1,200–1,300 aa	Generally T-rich PAMs	Insertions/deletions	Strong multiplexing	A/T-rich regions; combinatorial editing	Ortholog- and temperature-dependent
Cas12b	Approx. 1,100 aa	Ortholog-specific PAMs	Insertions/deletions; transcriptional regulation	Complementary targeting range	Alternative target sites; transcriptional regulation	Temperature dependence; lower maturity
TnpB	Approx. 350–500 aa	Specific TAMs	Deletion-dominant outcomes	Compatible with viral delivery	Regulatory-region targeting; transgene-free delivery	Targeting rules remain incomplete
IscB	Approx. 400–500 aa	Specific TAMs	Insertions/deletions; base editing	Compact and engineerable	Gene knockout; base editing	Limited evidence across plant species
Fanzor	Approx. 500–700 aa	Specific TAMs	Insertions/deletions; fragment deletion	Eukaryotic origin	Regulatory-region deletion; fusion-editor development	Still at an early stage

Note: Protein sizes are approximate and refer to representative nucleases without fused effector domains. PAM/TAM requirements, editing outcomes, and application status may vary among orthologs, engineered variants, target contexts, and plant species

Core Challenges and Future Directions

Cis-regulatory element editing and intelligent design have opened a completely new path for crop genetic improvement, but their mature application still faces a full-chain set of challenges, spanning from basic research to field implementation, with each step interlinked and mutually constraining. The particular regulatory complexity and research gaps in polyploid crops, such as wheat and cotton, represent key bottlenecks for extending these technologies to staple crops (Bradbury et al. 2025; Fernández Ríos et al. 2025). An overview of the major future directions, polyploid challenges, and the closed-loop intelligent breeding platform is shown in Fig. 4.

Fig. 4.

Core challenges and future directions of cis-regulatory element (CRE) editing in polyploid crops. a Future development requires high-resolution data resources, artificial intelligence (AI)-assisted prediction, and emerging technologies, including single-cell and spatial multi-omics, subgenome-resolved maps, generative and interpretable AI, prime editing, and reversible epigenetic editing. b Major challenges in polyploid plants include subgenome divergence, homoeologous redundancy, and functional compensation, which can buffer editing effects and limit phenotypic output; a potential solution is to remove the compensating homoeolog and then edit the target CRE. c To address these challenges, a closed-loop intelligent breeding platform is proposed, integrating AI-driven design, precision editing, high-throughput phenotyping, field validation with genotype-by-environment (G×E) analysis, and data feedback. Together, a–c depict a future path in which advances in data generation, predictive modeling, and editing technologies help overcome polyploid-specific barriers and support iterative optimization of CRE engineering in crop improvement

At the fundamental level, the current bottleneck stems from the dual limitations of high-quality data and reliable predictive models. Existing CRE functional maps are often constructed based on bulk tissue samples, which cannot resolve cell-type-specific regulatory patterns, resulting in ambiguous target prediction (Yu et al. 2023a; Marand and Schmitz 2022). Mainstream machine learning models often lack interpretability, which limits mechanistic interpretation of their predictions. Many are also trained on diploid model plants or relatively simple genomes and therefore generalize poorly to polyploid crops such as wheat and cotton (Harfouche et al. 2023; Novielli et al. 2024; Ramírez-González et al. 2018). Crop regulatory networks add further uncertainty. Redundancy and robustness can compensate for CRE edits and weaken the expected phenotype. In addition, spatiotemporal specificity and pleiotropy may cause unintended effects in other tissues, developmental stages, or environments.

The regulatory particularities of polyploid crops exacerbate these difficulties. Core staple and economic crops, such as wheat and cotton, contain multiple homoeologous genomes, with cis-element function showing clear subgenomic divergence. For example, in wheat, different subgenomes can differ markedly in transcription factor binding site occupancy, enhancer-like element origin, and regulatory output (Wang et al. 2021; Xie et al. 2023; Jia et al. 2025). Current CRE maps, predictive models, and validation systems are still based mainly on diploid crops and lack subgenome-resolved, cell-type-specific data. Consequently, editing one promoter element may be compensated by homologous elements in other subgenomes, reducing or eliminating the phenotypic effect (Ramírez-González et al. 2018; Zhang et al. 2022). Therefore, a practical strategy for polyploid CRE editing may not be to modify all homoeologous regulatory copies simultaneously, but rather to first reduce redundancy by knocking out or weakening dispensable homoeologous copies, while retaining one major functional copy as the entry point for precise cis-regulatory tuning. In essence, this “de-redundancy followed by fine-tuning” framework could convert a buffered and difficult-to-predict multi-copy regulatory system into a more controllable expression output, thereby improving the predictability and stability of phenotypic effects. Importantly, such a strategy should not be applied indiscriminately, because some homoeologous copies may already have undergone functional divergence; rather, it should be guided by subgenome-resolved functional evidence to distinguish redundant copies from indispensable ones.

Even if stable superior phenotypes are obtained under laboratory conditions, the transition to the field requires overcoming internal and external constraints. Internally, phenotypes that are stable under controlled laboratory environments may become unstable in complex field conditions due to genotype × environment × management interactions, reflecting insufficient understanding of regulatory network and environmental interaction rules (Mahmood et al. 2022). Externally, global regulatory frameworks for gene-edited crops remain highly heterogeneous, with different countries and regions adopting non-uniform definitions and approval standards for non-transgenic edited crops. Public acceptance of new biotechnologies also affects commercialization (Fernández Ríos et al. 2025; Atimango et al. 2024). Recent review-level analyses further suggest that the path toward broad deployment of CRISPR-derived, genetically modified organism-free crops will depend not only on technical progress, but also on regulatory harmonization, public acceptance, and clearer commercialization pathways (Ahmar et al. 2026).

Addressing these interlinked challenges, the core of future development lies in establishing a coordinated, full-chain evolution system linking data, understanding, design, and application, thereby constructing an autonomous iterative breeding loop that fundamentally overcomes bottlenecks at each step, while specifically addressing gaps in polyploid crop research. At the data dimension, single-cell and spatial multi-omics technologies are needed to construct cell-type-resolved CRE activity maps, solving the problem of spatiotemporal regulatory specificity (Yu et al. 2023a; Yao et al. 2026). For polyploid crops, it is especially important to build subgenome-resolved single-cell CRE activity maps. In rice, single-cell chromatin accessibility maps have already shown feasibility for resolving cell-type specificity and evolutionary dynamics of CREs across multiple tissues and cell states (Yan et al. 2025), while in wheat, single-cell or epigenomic atlases of roots and endosperm provide the basis for subgenome-resolved regulatory dissection (Ke et al. 2025; He et al. 2024). The accumulation of these resources will provide the necessary foundation for identifying cis-elements active only in specific subgenomes and cell types, as well as for determining which homoeologous regulatory copies are redundant and which should be preserved for subsequent precision editing.

At the algorithmic dimension, the simultaneous advancement of generative AI and explainable AI is required. Generative AI can learn CRE sequence features and regulatory grammar to design artificial regulatory sequences with desired expression patterns, whereas explainable AI helps mitigate the “black-box” problem by converting model predictions into testable biological hypotheses, increasing model credibility and transferability in breeding scenarios (Harfouche et al. 2023; Novielli et al. 2024; Gupta et al. 2025). For polyploid crops, next-generation models capable of distinguishing regulatory differences among homoeologous subgenomes are needed, integrating sequence, chromatin accessibility, epigenetic marks, and 3D genome architecture data to achieve truly subgenome-specific targeting (Wang et al. 2021; Jia et al. 2025). Beyond target prediction, such models should also help determine whether a trait is best optimized through single-subgenome fine-tuning or through combinatorial manipulation of multiple homoeologous copies, thereby providing a rational basis for the “remove redundant copies, retain one for tuning” strategy.

At the technological dimension, future improvements should enhance editing precision and increase the flexibility and reversibility of regulatory control. Prime editing, capable of precise base substitution, insertion, or small fragment modification without double-strand breaks, is emerging as an important candidate tool for complex CRE editing (Yu et al. 2023b; Kuang et al. 2024). At the same time, epigenome editing is gaining attention, as it can reshape CRE activity without altering DNA sequences, providing new tools for validating homologous element function, reducing pleiotropy risks, and developing reversible regulation strategies (Sen et al. 2025; Liu et al. 2025b). In polyploid crops, these tools may be particularly valuable for implementing stepwise editing schemes, in which dispensable homoeologous copies are first inactivated or weakened, and the retained major copy is then quantitatively tuned at the cis-regulatory level. This suggests that future CRE operations may no longer be limited to permanent sequence modifications, but could evolve into integrated systems combining sequence editing with epigenetic control.

Ultimately, these dimensions need to be integrated into a closed-loop intelligent breeding platform linking data, algorithms, design, experimentation, and phenotypes. A closed-loop breeding platform should integrate regulomic, phenomic, environmental, and field-performance data. It should continuously update design rules and refine models through automated experiments and high-throughput phenotyping. This could improve the stability of superior phenotypes under field conditions (Bradbury et al. 2025). For polyploid crops, the platform must include subgenome-resolved modules, enabling truly subgenome-specific editing in crops like wheat. The proposed activation of only the D-subgenome TaGW2 promoter illustrates this future capability. It would require the integration of single-cell regulatory maps, interpretable models, precise editing tools, and field validation within one design-validation loop. Within such a framework, the key decision is no longer simply where to edit, but also which homoeologous copies should be removed, which one should be retained, and how the retained copy should be tuned to generate stable agronomic gains. Only under these conditions can CRE editing technology realistically move from laboratory experimentation to field application.

Conclusion

The development of cis-regulatory element editing technology has undergone a paradigm shift from “blind mutagenesis” to “targeted attempts” and then to “intelligent design.” This transformation has promoted the iterative development of crop breeding technologies and reshaped the theoretical framework for genetic improvement of quantitative traits. The deep integration of multi-omics data and artificial intelligence algorithms has shifted CRE editing from a “trial-and-error model” that relies on empirical screening to a “predictive model” based on rational design, thereby making it possible to achieve precise, efficient, and customized improvement of quantitative traits. Although this field still faces multiple challenges in data integration, regulatory network dissection, and field translation, these bottlenecks are expected to be gradually addressed with the continuous development of single-cell multi-omics, generative AI, and novel editing tools. In the future, deeper interdisciplinary collaboration among genomics, bioinformatics, artificial intelligence, and breeding science will further accelerate the arrival of the era of intelligent design breeding and provide powerful technological support for safeguarding global food security.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CRE

Cis-regulatory element

eQTLs

Expression quantitative trait loci

SV

Structural variation

QTL

Quantitative trait locus

PAV

Presence/absence variation

UTR

Untranslated region

SNP

Single-nucleotide polymorphism

InDel

Insertion/deletion

uORF

Upstream open reading frame

Wx

Waxy

SC-uORF

Sucrose-control uORF

MREs

MicroRNA recognition elements

IME

Intron-mediated enhancement

gRNA

Guide RNA

MPRA

Massively parallel reporter assays

CBEs

Cytosine base editors

ABEs

Adenine base editors

AI

Artificial intelligence

G×E

Genotype-by-environment

ATAC-seq

Assay for transposase-accessible chromatin with sequencing

ChIP-seq

Chromatin immunoprecipitation sequencing

Hi-C

High-throughput chromosome conformation capture

TF

Transcription factor

CAPE

CRISPR-Cas12a promoter editing

DIAL

Dynamically editable artificial loci

SVM

Support vector machine

EBE

Effector-binding element

TALE

Transcription activator-like effector

Xoo

Xanthomonas oryzae pv. oryzae

PEPE

Prime editing-mediated promoter engineering

aa

Amino acids

PAM

Protospacer-adjacent motif

TAM

Target-adjacent motif

Author Contributions

YL: Investigation, Project administration, Writing-original draft, Writing-review & editing. XZ: Data curation, Visualization, Writing-original draft. YC: Writing-review & editing. GS: Visualization, Investigation, Data curation. YW: Investigation, Data curation. YM: Investigation, Data curation. TT: Investigation, Visualization. ML: Data curation, Investigation. SS: Supervision, Funding acquisition, Project administration, Writing-review & editing, Conceptualization. All authors have read and approved the final manuscript.

Funding

This work was supported by Henan Provincial Science and Technology Research and Development Plan Joint Fund (Preferential Discipline Cultivation Category) (242301420133), National Natural Science Foundation of China (32201823).

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.