Induced Methylation in Plants as a Crop Improvement Tool: Progress and Perspectives

: The methylation of gene promoters is an epigenetic process that can have a major impact on plant phenotypes through its control of gene expression. This phenomenon can be observed as a response to stress, such as drought, cold / heat stress or pathogen infection. The transgenerational heritability of DNA methylation marks could enable breeders to ﬁx beneﬁcial methylation patterns in crops over successive generations. These properties of DNA methylation, its impact on the phenotype and its heritability, could be used to support the accelerated breeding of improved crop varieties. Induced DNA methylation has the potential to complement the existing plant breeding process, supporting the introduction of desirable characteristics in crops within a single generation that persist in its progeny. Therefore, it is important to understand the underlying mechanisms involved in the regulation of gene expression through DNA methylation and to develop methods for precisely modulating methylation patterns for crop improvement. Here we describe the currently available epigenetic editing tools and their advantages and limitations in the domain of crop breeding. Finally, we discuss the biological and legislative limitations currently restricting the development of epigenetic modiﬁcation as a crop improvement tool.


Introduction
Traditional plant breeding involves repeated crossing and selection based on morphological traits, with molecular methods introduced in recent decades to complement phenotypic selection. Genome editing using tools such as the CRISPR-Cas9 system is still novel technology that is starting to be applied to make direct changes to the nucleotide sequence of crops. In addition to these genetic changes, an improved understanding of the effect of epigenetic variation such as DNA methylation on plant phenotype has provided an opportunity to further accelerate the crop improvement process. Therefore, epigenome editing could help overcome some weaknesses of genome editing (such as gene knockout), which can have a strong off-target impact and only enables the loss of function of a gene.
DNA methylation appears as a promising source of variation contributing to the fitness and yield of crops due to its heritability and ability to impact plant phenotypes. Agronomically important traits such as flowering time, seed dormancy and yield are influenced by DNA methylation variations, and the partial heritability of DNA methylation patterns suggests epigenetics has played a role in plant domestication and evolution [1][2][3]. Mastering DNA methylation modification will allow breeders to simultaneously induce beneficial variations and restrain undesirable epigenetic modifications that can be induced by breeding methods such as tissue culture [4][5][6][7][8]. Different methods have already been developed for this purpose and effective induction of epigenetic changes for plant breeding requires a technical understanding of the molecular processes involved in both its introduction and maintenance over subsequent generations. Here, we summarize recent technical advances in epigenome editing tools as well as the biological and legal limitations that have yet to be overcome.

DNA Methylation in Plants
Epigenetic modifications alter the DNA molecule, however, the DNA sequence itself remains unchanged. DNA methylation is the most common type of epigenetic modification, defined by the covalent addition of a methyl group (CH 3 ) to the fifth position of a cytosine ring (5 mC) by DNA methyltransferases. DNA methylation is a heritable and reversible process, involved in multiple genetic and genomic cellular mechanisms such as transposon silencing, tissue-specific gene expression and genome balance after polyploidization [9][10][11][12][13][14].
DNA methylation is a complex process involving interactions between multiple genomic elements and is yet to be fully understood. DNA methylation in promoter regions has been known to play a major role in regulating gene expression since 1975 [15,16]. When covalently attached to DNA, methyl groups can block the access of the transcription machinery to the gene and reduce transcription. The correlation between promoter methylation and reduced gene expression is well-established; however, induced promoter methylation is not always sufficient to repress nearby gene expression [17]. For example, the methylation of the FWA promoter only induces a limited downregulation of the gene when located more than 500 bp from the transcription start site [18].
While the role of gene promoter methylation is slowly being elucidated, the role of gene body methylation (gbM) remains largely unknown [19]. One hypothesis is that methylation is involved in gene splicing, as it has been noted that exons contain more DNA methylation compared to introns, and spliced exons are generally less methylated [20,21]. Moreover, it has been described that gbM is mainly found on constitutively expressed genes and is occasionally located on genes with fluctuating expression level [20] suggesting a relationship between gbM and transcription [22]. The mechanisms leading to specific methylation patterns are yet to be determined. Due to its potential regarding transcriptome regulation and phenotype modification, DNA methylation is of particular interest to plant breeding.

Effect of DNA Methylation on Plant Phenotype
Interest in the impact of epigenetic modifications on plant phenotypes has increased since 1999 following the observation of the first naturally occurring morphological mutant in the Lcyc gene of Linaria vulgaris not caused by DNA mutation [23], showing the existence of epi-allelic variants. The same study demonstrated that restoring the original methylation state recovered the wild-type phenotype. Since then, several studies have confirmed a link between DNA methylation modifications and disrupted phenotypes which influence many complex traits (Table 1). For example, a spontaneous epigenetic mutation at the Cnr locus can inhibit fruit ripening in tomato [24]. The comparison of plants diverging only by their methylome (i.e., epigenetic recombinant inbred lines) has shown the impact of hypomethylated genes on phenotypic alterations in traits such as flowering time [25]. The comparison of cultivated and wild cotton led to the identification of domestication traits influenced by epigenome evolution, such as flowering time and seed dormancy [1]. Observation of the impacted phenotypes confirms the idea that variation in methylation patterns is a way for plants to adapt to the environment [2,26,27]. Unknown information is marked as "/".

Impact of the Environment on Plant DNA Methylation
A recent study of A. thaliana estimated the spontaneous CG methylation gain at 2.56 × 10 −4 and methylation loss at 6.3 × 10 −4 [38]. However, environmental stress can impact both DNA methylation rate and profile in plants, as shown in A. thaliana lines under salt-stress that accumulated~45% more differentially methylated cytosines in a CG context, preferentially located in genic regions [26]. Similarly, a study in rice exposed to drought highlighted a non-random distribution of stress induced epimutations, many of them being in loci associated with drought resistance [39]. Other examples of environmental factors leading to altered gene expression have been demonstrated, including in A. thaliana under phosphate starvation, pathogen exposure, in rice exposed to cadmium, or under anoxia and drought stress [40][41][42][43][44]. Cold-induced gene demethylation led to an increase in transcriptional activity and seems to play an important role in Hevea brasiliensis cold tolerance [31]. However, as demonstrated with the stability of the methylome of a geographically dispersed A. thaliana population, changes in the environment are not always sufficient to trigger an epigenetic response [45]. Interestingly, the temporal hierarchy of transcriptome and epigenome changes under environmental stress remains ambiguous as stress-induced gene expression seems to drive the hypermethylation of adjacent transposable elements [46].

DNA Methylation Heritability in Plants
Unlike genomic variation, the ability of epigenetic markers to transmit acquired environmentally adaptive traits to offspring is highly advantageous. Several examples of DNA methylation pattern heritability have been described in different plant species, such as A. thaliana, Picea abies, Taraxacum officinale and Nicotiana tabacum and crop species, including: rice and wheat [47][48][49][50][51][52] ( Table 2). In A. thaliana, DNA methylation variation transmission has been observed for at least eight generations [25]. Epigenetic changes induced by stress can be transmitted to offspring, but their stability and the potential effect on a population in the long term is unknown [49]. A recent study on epigenetic modifications occurring during the domestication of cotton concluded that epialleles were transmitted between generations, resulting in phenotypic diversity and allowing a large geographic expansion of the species [1]. Intergenerational transmission of phenotypic traits caused by DNA methylation variations, such as roots adapted to drought stress, delayed flowering time and modified plant architecture has also been demonstrated [29,53]. Since methylation is heritable and can cause phenotypic changes, inducing/removing DNA methylation has been a focus of research with the aim of remodeling the expression of genes already present in crop genomes. The growing interest in epigenomics has led to the development of methods to induce DNA methylation, first genome-wide with non-specific methods and now at a base-specific level with new epigenetic editing tools.

Induced (de)Methylation
The primary techniques developed to study DNA methylation, such as demethylating agents and gene knock-out, cause changes to the epigenetic state at a non-specific, genome-wide scale. The resulting plant displays numerous phenotypic and developmental aberrancies [57], which cannot be linked to specific genes. Although these methods paved the way to study DNA methylation on a genome-wide scale, DNA methylation studies cannot rely on non-specific methods, as they cannot be directed, they impact too many interconnected pathways, and cannot lead to accurate outcomes.
Accurate DNA methylation mark induction and removal have complementary functions and could both play a role in crop improvement. DNA hypermethylation is, for example, involved in crop stress resistance mechanisms, such as wheat salt tolerance, pea and barley response to drought stress, and tomato response to cold [58][59][60][61]. Meanwhile, DNA hypomethylation is implicated in drought stress response in rice and faba beans, and heat response in soybean and rapeseed [62][63][64][65].
Site-specific methylome modifications have been made possible with the development of epigenome editing tools. Epigenome editing tools consist of two essential components: a DNA-binding targeting domain and a functional domain. The targeting domain can be based on Zinc finger (ZF) proteins, Transcription activator-like effectors (TALEs) or the Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR-Cas9) system [66]. They can induce epigenetic modifications at a site-specific level, precisely enough to efficiently target promoters, and therefore are valuable tools for precision epigenome engineering.

Zinc Finger Proteins
Zinc finger proteins are DNA-binding motifs naturally present in many transcription factors, and they can be modified to recognize any unique sequence in a genome. Each finger is composed of 30 amino acids, able to recognize a three base pair segment and their association can be engineered to target any specific sequence [67]. ZF proteins represent the first generation of DNA editing tools and have been used for epigenetic modification since 2002 [68]. Their compact size makes them well suited to target highly condensed regions; however, they can be both laborious and expensive to assemble [66,69]. ZFs associated with a protein essential to RNA directed DNA methylation (RdDM) were successfully used in A. thaliana for the first time in 2014 to induce methylation. Targeted to an unmethylated epiallele of FWA, it resulted in early flowering of the plants, suggesting the silencing of the gene [70].
In the case of induced demethylation, the combination of ZFs with DNA demethylase TET1 can result in the demethylation of methylation sites hundreds of bases away from the target [71]. Incomplete demethylation can be restored quickly to its original fully methylated state by the RNA-directed methylation machinery after the removal of the ZF. It is still unknown if targeted demethylation can be transmitted to offspring [71]. The use of ZF is today limited in epigenome editing because of their lack of binding specificity and thus, wide off-target effects [72].

Transcription Activator-like Effector (TALEs)
Transcription activator-like effectors are proteins secreted by Xanthomonas bacteria that bind to specific gene promoters. TALEs are formed of tandem repeats, which are usually 34 amino acids in length, in which specific amino acids (in position 12 and 13) can target single DNA bases [73]. TALEs are an improvement over ZF proteins, as they show higher specificity, are more convenient to engineer as they do not need an assembly step to build long sequence-specific arrays and they allow more design flexibility due to their single base recognition ability [74]. However, their manipulation is still costly and labor intensive, as they need to be designed specifically for their target, and their large size limits their access to most target sequences. Moreover, interactions with methylated DNA restrict their binding capacity, reducing their potential use for epigenome editing [75]. TALEs are still popular epigenome editing tools, proving their efficacy for gene expression modulation even when compared to the next generation tool CRISPR-Cas9 [76].

CRISPR-Cas9
The CRISPR gene family evolved as part of bacterial and archaeal immune systems, and today it has been co-opted to become the most accessible and affordable gene-editing tool available. The CRISPR-Cas9 system has two components: (i) clustered regularly interspaced short palindromic repeats (CRISPRs), which are short repetitive DNA sequences, and (ii) the endonuclease CRISPR-associated protein 9 (Cas9). In epigenome editing, the endonuclease activity of Cas9 is not required and the use of an inactive version (dCas9) confers to CRISPR the same editing potential as ZF and TALE [77]. Recent innovations have expanded the application of CRISPR technology, enabling site-specific induction of DNA methylation [66]. The CRISPR-Cas system requires the presence of a protospacer adjacent motif (PAM), which is a two to six nucleotide long sequence following the targeted region that limits the targeting range of the method. Another weakness of the method is that perfect complementarity between the targeted DNA and the binding domain is not necessary for the binding to occur, enhancing the risk of off-target effects [78].
CRISPR-Cas9 can be used for the purposes of gene silencing when fused to a DNA methyltransferase [79,80], and gene activation when fused with the catalytic domain of TET1 [81,82]. A CRISPR-based system was successfully used in A. thaliana to induce the methylation of the FWA promoter, causing an early flowering phenotype [83]. Several genes in rice that were identified as negative regulators of yield (Gn1a, DEP1, GS3, IPA1) and grain weight (GW2, GW5 and TGW6) were knocked out with CRISPR-Cas9, resulting in increased grain number, size and weight [84,85]. Inducing targeted DNA methylation to modify the expression of genes that negatively impact yield could be an effective alternative to gene knockouts, which can have deleterious off-target effects [86]. The main advantage of CRISPR-Cas9 over other gene-editing tools is the ability to use guide RNAs (gRNA) to drive the functional domain with no need to design specific DNA binding proteins. This specificity means that assembling the CRISPR-Cas9 complex and directing it to the site is easier to engineer compared to ZF and TALEs [87]. One other significant advancement is that, unlike ZFs and TALEs, CRISPR-Cas9 can target and edit multiple genomic sites at once [66]. The use of different RNA-binding domains associated with different Cas9 orthologs enables the targeting of multiple sites simultaneously [88].

Perspectives for Plant Breeding
(Epi)genome editing tools have enabled the application of methylation modification to plant breeding. The first generation tool, ZFs, remain useful due to their small size and binding properties, though TALE-based systems are easier to engineer. TALEs present a problematic sensitivity to DNA methylation, which complicates their use for epigenome editing. The CRISPR-Cas9 system presents comparable efficiency but with easier and cheaper implementation, and the ability to target multiple sites (multiplex) ( Table 3). CRISPR-Cas9 is already largely used in genome editing, and is quickly becoming the most popular tool for inducing methylation in plants. CRISPR-Cas9 has already been effectively applied to edit a wide range of crop genomes, including rice, potato, sweet orange, soybean, tomato, maize and cotton [89][90][91][92][93][94][95]. However, the use of endonucleases runs the risk of inducing off-target double-strand breaks (DSBs) which can lead to unpredictable mutations elsewhere in the genome. Epigenome editing with CRISPR-Cas9 results in less extreme off-target activity than genome editing, as its effects are progressive and proportionate to target binding. Moreover, most off-target effects will occur in non-regulatory regions, making them silent [86]. In addition, epigenetic mutagenesis can lead to a genetic gain of function, as opposed to most genetic variants, which tend to cause a loss of function [96].
Epigenome editing could be beneficial to plant breeding at several levels. Firstly, as with genetic modifications, it could accelerate the process of domestication of wild plants through the modification of traits involved in growth habit, flowering time, yield, nutrition, as well as seed and fruit size and number, in a single generation. However, in contrast to gene knockout, induced (de)methylation has an adjustable impact and can influence both gene expression and repression. Induced DNA methylation could also improve hybrid breeding and plant propagation through tissue-culture, which can induce new gene expression patterns in their progeny [4,5,97,98], and thus help control offspring phenotype. It could be used to eliminate deleterious traits or add desired characteristics to crops, such as biotic and abiotic stress resistance (Table 1). Finally, it could induce plant resistance to viruses or even alter a pathogen population, making it susceptible to the plant's defenses [99].

Limitations
Several limitations hinder the widespread use of induced DNA methylation in plant breeding. First at a biological level, with the determination of genes to target, the durable maintenance of induced epialleles and the presence of off-target effects. Second, the introduction of inappropriate legislation regarding new breeding techniques that restrict research and the implementation of advanced breeding approaches.

Biological Limitations
DNA methylation often plays an important role in controlling gene expression. However, DNA methylation is part of a complex regulatory system and so its alteration is not always sufficient to impact gene transcription. Therefore, the modification of crop DNA methylation patterns should be supported by other transcriptional regulators such as chromatin markers, which is an area of research yet to be fully explored [100,101].
The short term heritability of DNA methylation pattern in plants has been largely demonstrated in different species (Table 2). However, as growing successive generations of plants can be laborious and time-consuming, most studies are limited to only a few generations, thus the long-term methylation pattern heritability remains largely unstudied. Moreover, the maintenance of induced methylation is uncertain, and a recent mammalian study revealed that removal of the methylation induction system led to a restoration of natural transcriptional levels [102].

Technology Limitations
CRISPR-cas9 is the most promising epigenome editing tool to date. However, its main drawback is the presence of off-target activity. A study testing CRISPR-Cas9 specificity when applied to the human genome showed that 98.4% of the tested guide RNAs designed to target exons had at least one off-target site [103]. Off-target binding seems inevitable in large genomes; however, different methods exist to improve CRISPR-Cas9 specificity. Novel machine learning algorithms enable the efficient design of sgRNAs, based on experimentally validated examples [104][105][106]. Limiting the amount of both Cas9 and sgRNA could minimize off-target modifications, as higher concentrations of these increase the risk of binding to sites containing mismatches [107,108]. Engineered Cas9 variants can increase CRISPR-Cas9 specificity by decreasing its tolerance to mismatches when binding to DNA [109]. Finally, some Cas9 orthologs could have advantages over the commonly used wild type SpCas9, such as greater specificity, which could be another way to reduce off-target effects [88].

Legislation
Numerous Genetically Modified Organism (GMO) regulations are based on obsolete definitions which do not include new plant breeding techniques and are not suitable for contemporary methods for altering genomes. In July 2018, the European Court of Justice decided that plants issued from genomic editing processes, even those not involving the use of recombinant DNA, fall under current GMO regulation [110]. In contrast, some countries, such as Canada, define the safety of new crops based on the properties of the final product (product-based) rather than the technique used to produce it (process-based). The United States adopted a hybrid system in which only plants presenting new traits fall under specific regulations [111]. In 2019, the Australian government ended regulation of "DNA free" gene-editing techniques. The amendment of regulations regarding new plant breeding techniques is important, as both scientific and economic benefits depend upon it [112].
The difficulty associated with distinguishing between organisms presenting random natural mutations and those that have undergone new plant breeding techniques presents a significant obstacle to updating obsolete legislation [111]. The gap that exists between scientific knowledge regarding the impact of new plant breeding techniques on human and environmental health with the current legislation is a major bottleneck of the expansion of (epi)genome-edited crops [113].

Conclusions
Improved understanding of the implication of epigenetics on phenotype has led to an increase in the number of epigenome studies, revealing the plurality and complexity of the mechanisms involved in methylation. DNA methylation appears to have played a significant role in plant domestication and evolution, and therefore has the potential to be used as a powerful tool to improve crop agronomic traits, such as yield and disease resistance. Modification of the crop methylome through induced methylation and the regulation of crop gene expression will allow breeders to maximize the genetic potential naturally present in the plants far quicker than through traditional breeding. It appears as complementary to genome editing, with (i) an adaptable impact, (ii) the potential to reveal hidden phenotypic variations and (iii) a reduced off-target effect. For these reasons, the modulation of crop DNA methylation pattern is considered to have significant potential in breeding, and updated regulations should support its expansion. The stable transmission of epialleles involved in agronomically important traits across generations is essential for the inclusion of epigenetic variations in breeding programs. CRISPR-Cas9 is the most powerful genome-editing tool created so far and we have just started exploring its full potential. Nevertheless, in order to fully incorporate targeted DNA methylation into the breeding process, challenges such as precise genomic target identification, durable maintenance of induced epialleles, control of off-target effects, and revision of current legislation must be addressed.

Conflicts of Interest:
The authors declare no conflict of interest