Figure 1. How the development of epigenetic data and tools will lead to epi-bred crops and new varieties in the field adapted to climate change. Section to find each topic in the review is indicated. In brown are indicated data already supporting epigenetics for breeding, in blue the uses of epigenetics for crop improvement (actually under development), and in purple the post-production steps until new varieties in the field (future challenge).
3. Epigenetic Advances in Crop Improvement: Exploiting Epigenetic Diversity
3.1. Natural Epi-Alleles
Naturally occurring epi-alleles have been linked to agriculturally important plant phenotypes including morphological
[33][34][35][36], developmental
[37][38][39], metabolic
[40][41][42], and immunity traits
[43]. While the origin of these epi-alleles is not always entirely clear, many of them seem to have arisen as spontaneous epimutations
[44], that is, through stochastic gain or loss of DNA methylation. In crops, examples include the spontaneous hypermethylation of the COLORLESS NON RIPENING (CNR) locus in tomato, which leads to inhibition of fruit maturation
[39], or the hypomethylation of the FERTILIZATION-INDEPENDENT ENDOSPERM1 (FIE1) gene in rice, which is associated with reduced stature
[35]. Because epi-alleles are often meiotically stable and independent of genetic variation, they present potential breeding or complementary editing targets for the agricultural sector
[45][46]. Yet, examples of naturally occurring epialleles are rare, and it remains unclear how to harness them systematically for crop improvement. Epigenetic editing has the potential to introduce another source of variation that leads to phenotyping changes and an increased crop production, especially under environmental influence. Moreover, it is an alternative way to manipulate a gene by repressing or activating it for a long period of time. This gene expression modification without altering the DNA sequence itself does not lead to secondary, often uncontrolled effects, as in the case with a gene knock out during genome editing, and can be reversible and adjustable to environmental stimuli. Epigenome editing is more thoroughly discussed in the last part of this review.
3.2. Chemically Induced Epigenetic Diversity
A variety of chemicals capable of modifying epigenetic information have been described. Based on their function, we have categorised them into two main groups: inhibitors of DNA methyltransferases (DNMTs) and histone deacetylase inhibitors
[47]. The first group consists mainly of a cytidine analogue that specifically inhibits DNA methylation by sequestering DNMTs. Once incorporated to DNA during its replication, the analogues covalently trap the DNA methyltransferases and mediate their degradation, leading to a passive loss of DNA methylation in the cell
[48][49]. The most commonly used representatives are 5-AzaC and Zebularine, but there are also more stable variants
[50]. The second major group consists of histone deacetylase inhibitors, which are classified into different groups such as hydroxamic acids, amino-benzamides, cyclic peptides, and short-chain fatty acids
[47].
Currently, DNA demethylating compounds that can de-repress the expression of the genes with hypermethylated promoters have wider applications. The most common applications are in tissue cultures, where 5-AzaC treatment has a beneficial effect on the induction of somatic embryogenesis
[51][52][53], microspore embryogenesis
[54][55], and shoot regeneration
[56]. Another study has taken advantage of the fact that flowering is controlled by, amongst other things, DNA methylation
[57]. With this in mind, the treatment of 5-AzaC has been shown to cause earlier or more frequent flowering compared to untreated control plants
[58][59][60]. DNA methylation has been suggested to play a role in suppressing transposon activity. Therefore, increased activity of transposons is anticipated upon 5-AzaC or Zebularine treatment, which was confirmed by Nishimura
[61], Konečná
[62], and Boonjing
[63]. Another group of applications focuses on the fact that the activity of the transgene in genetically transformed tissues is silenced by DNA methylation of the promoter contained in the transgenic construct. Treatment of such silenced transformants with 5-AzaC then caused reactivation of silenced transgenes
[64][65]. Due to the commonly discussed role of DNA methylation in the field of priming or in transgenerational memory, treatment with 5-AzaC or Zebularine is also used for deeper understanding of DNA methylation in stress-memory-related phenomena. Verhoeven and Gurp
[66] showed that Zebularine treatment affected the within-generation response to nutrient stress and indicated a role for DNA methylation in phenotypic plasticity. Rendina
[67] showed that stress-induced memory alters growth of clonal offspring of white clover and that this transgenerational effect is not observed in offspring of parents that receive 5-AzaC. Similar results were obtained by Yang
[68], where 5-AzaC diminished the difference between memorised and wild-type lines of Arabidopsis. The use of the second group of epigenetically active substances, histone deacetylase inhibitors, is currently sporadic. If they were used, it was mainly as a treatment within plant tissue cultures. Specifically, the use of Trichostatin A induced microspore embryogenesis in wheat
[69] and somatic embryogenesis in Arabidopsis
[70]. Treatment by trichostatin A and sodium butyrate also improved regeneration efficiency from mature wheat embryos
[71]. It can be concluded that utilisation of epigenetically active substances is focused on topics, where it is suspected that the level of DNA methylation or acetylation of histones play a role in the respective phenomena. Specifically, this means trying to influence the various developmental stages of plants (flowering, embryogenesis, in vitro regeneration) or verifying whether DNA methylation or histone acetylation participates in a given phenomenon (stress memory molecular background). The disadvantage of such an approach is that inhibitor effects are global and not localised, affecting the whole genome and not only the locus of interest.
3.3. Inducing Epigenetic Diversity through Genetic Mutation
One emerging approach that aims to induce stable epi-alleles is through genetic mutations in DNA methylation pathways
[72]. In the model plant A. thaliana, many studies have shown that DNA methylation mutants exhibit altered phenotypes, which are due to trans-induced hyper- or hypo-methylation of specific gene loci
[73][74][75][76][77]. While many of these induced traits appear to be deleterious, with varying degrees of penetrance
[75][76][78], several examples do exist where induced epi-allelic variation leads to advantageous phenotypes
[79][80][81]. For instance, in Arabidopsis, mutants in RdDM displayed enhanced resistance against biotrophic pathogens, such as the downy mildew pathogen Hyaloperonospora arabidopsidis
[79]. In addition, reduced DNA methylation levels in Arabidopsis DNA methylation mutants flowered earlier without the requirement for vernalisation
[81], an agronomic trait that is usually preferable over late flowering. In the model tree poplar (Populus spp.), hypomethylation through a mutation in DDM1 increased tree tolerance in response to water deficit, showing how DDM1 context-dependent DNA methylation plays a crucial role in tree phenotypic plasticity under drought stress conditions
[82]. The authors suggest major roles for DNA methylation in the shoot apical meristem in regulating genes involved in hormone-related stress responses, and the maintenance of genome integrity through repression of TEs.
Importantly, these phenotypes can sometimes persist for many generations, even when the inducing mutations are segregated away
[83][84][85], indicating that they are acquired through epigenetic variation. This latter observation has been exploited systematically in the construction of so-called epigenetic recombinant inbred lines (epiRILs)
[83][84]. This experimental system is derived by crossing a DNA methylation mutant to wild type. The F1 progeny of this cross are either backcrossed to wild type or intercrossed, and individual homozygous wild type for the inducing mutation are repeatedly self-pollinated to derive recombinant inbred lines. The final plant material thus consists of isogenic, epi-homozygous individuals, whose methylomes are a mosaic of the two original parents. These novel epi-genotypes can be systematically screened for favourable traits. Two epiRIL panels have been created in A. thaliana using DDM1
[83] or MET1 gene
[84] as inducer mutations.
Extensive phenotypic surveys of the ddm1-derived epiRILs have revealed large heritable variation for key agricultural traits, such as plant height, root length, flowering time, rosette size
[83][86][87], pathogen resistance
[80], and phenotypic plasticity under salt stress
[88]. Heritability estimates are in a range similar to what is typically seen in crop breeding
[86][89]. By design, the epiRILs are also amenable to epigenetic quantitative trait locus (QTLepi) mapping approaches, where DNA methylation markers are used in genome-wide linkage scans
[86]. Numerous epigenetic QTL have been identified. In some cases, the mutant epigenotypes at the QTL locus are associated with improved phenotypic performance relative to wild type, indicating that the underlying epialleles could be exploited in breeding programs
[80][86][88][90]. More recently, epiRILs have also been utilised to generate large panels of so-called epiHybrids by crossing selected lines to wild type
[91][92]. The epiHybrids display substantial best-parent heterosis for important traits like early vegetative growth
[92], early flowering time, increased leaf area, plant height, and main stem branching, several of which could be linked to differential methylated regions in the parental genomes
[91].
There have been efforts in the academic and private sectors to implement similar epigenetic perturbation strategies in crops. Such approaches could complement more mainstream breeding tools focused on genetic variation. However, the choice of inducer mutation is more challenging here, since loss of MET1 or DDM1 tend to be lethal or cause severely reduced fertility in complex crop genomes
[93][94][95]. A promising alternative candidate is the MSH1 system. The MSH1 mitochondrial and plastid-targeting protein is involved in genome stability and influences plant growth patterns
[96][97]. By crossing MSH1 to wild type, MSH1 is suppressed by RNA interference, resulting in enhanced plant growth and heritable epigenetic changes in Arabidopsis
[98], as well as in tomato
[19]. Notably, the methylome reprogramming in Arabidopsis is accompanied by changes in siRNA expression and heritable non-genetic memory
[68]. In addition to crosses, the MSH1 system has also been employed in grafting experiments to introduce rootstock epigenetic variation, where the progeny also displays increased seed yield and growth vigour
[20][98]. Most importantly, the MSH1 system seems to be reproducible in tomato field size experiments, making it a promising system for agricultural applications.
3.4. Environmentally Induced Epigenetic Diversity
Plants are immobile and exploit various tools to survive and reproduce under suboptimal or harsh surroundings. In the case of environmental stimuli, they undergo physiological response and phenotypic changes, which, according to the length of exposure, are divided into developmental plasticity (short term stimuli), memory and priming (mid-term stimuli), and local adaptation (long term exposure)
[99]. There are many examples shown above that plants are better prepared for repetitive exposure to stress, either in a given individual, a derived clone, or even transgenerationally
[100]. It is therefore possible to expect that similar mechanisms of plant adaptation also occur in situations where plants are exposed to natural stress conditions. Despite the relatively well-described and known memorising effect, the unique properties of plants that are naturally adapted to their respective environment is not yet very widespread in practice. One example is the fact that plants in different environments produce different ranges of secondary metabolites
[101], which may theoretically be useful in the field of medical plants or spices
[102]. From the evolutionary point of view, the issue of environmentally induced changes indirectly includes the strategy of vernalisation as a possible way for plants to adapt to extreme winter frosts
[103] or for the epigenetic background of invasive species to successfully establish in different environments
[104][105].
3.5. Hybridisation and Epigenetic as a Predictive Marker of Hybrid Performance
Heterosis, or hybrid vigour, is the superior phenotypic performance of F1 hybrids compared to the two inbred parents
[106]. The phenomenon is extensively exploited to increase crop production worldwide
[107]. Breeding programs aim to optimise heterosis by carefully selecting parental genotypes for crossing. However, genotypic information is often insufficient to accurately predict heterosis, and the molecular basis of this phenomena remains poorly understood
[107]. Accumulating evidence points to a strong epigenetic contribution to heterosis. Molecular profiling of hybrids shows that their epigenomes are substantially remodelled with respect to their parental lines, leading to epigenetic states that diverge both positively and negatively from the expected mid-parent values (MPV)
[108][109]. Extensive remodelling has been observed at the level of DNA methylation in Arabidopsis
[91][108][109][110][111], rice
[112][113], pigeon pea
[114], broccoli
[115], and in rapeseed
[116], and occurs either at regions where parents are differentially methylated (DMRs)
[91][108][109][117] or in regions where the parents are similarly methylated (SMRs)
[113][115][118]. Similar observations have been made at the level of small RNA (sRNAs) abundance
[119][108][110][112][114][116][118][120][121][122][123][124][125][126] and at various histone modifications
[112][127]. In many cases, these epigenetic changes lead to downstream non-additive gene expression levels, which have been linked to heterotic phenotypes
[45][108][109][112][114][116][120][128][129], often associated with salicylic acid (SA) metabolism and response
[118][130]. For example, differences in the SA endogenous level in F1 heterotic hybrids have been correlated with enhanced leaf growth in Arabidopsis
[130], and Zhang
[118] not only revealed DDM1 as an epigenetic link between growth vigour and SA metabolism but also suggested that parental SA differences can predict heterosis level in the F1.
While it is possible that the remodelling of hybrid epigenomes is “just” the molecular consequence of classical genetic complementation effects, recent evidence shows that it also occurs when the parental genomes are isogenic or nearly isogenic
[91][92][108][110][111][131]. This suggests that non-additive epigenetic changes observed in hybrids may originate from interactions between divergent parental epigenomes. Given the close association between DNA methylation, PTMs, small RNAs, and gene expression, parental epigenetic differences may therefore be used as indicators of hybrid performance, independently of genetic differences. Lauss
[91] demonstrated that this is indeed the case. Using a panel of A. thaliana epigenetic Hybrids (epiHybrids) obtained from crosses of isogenic but epigenetically divergent parents, they showed that heterosis for flowering time, height, and leaf area could be associated with specific DMR between the parents. Similarly, in canola, hybrids obtained from isogenic parents which had undergone prior divergent selection for high energy-use efficiency showed increased yield. The artificial selection of the agronomical characteristics and the distinct DNA methylation parental patterns were heritable in the hybrids and allowed the targeted increase of yield
[132].
These results show there is predictive information in epigenetic markers beyond DNA sequence variation. Epigenetic measurements may therefore be used as complementary biomarkers, even in crop systems where the parental lines are highly polymorphic. To illustrate this, Seifert
[121] quantified differences in sRNA expression profiles among different heterotic groups in maize. The parental divergences in sRNA were strong indicators for grain yield in the hybrids
[121][133]. The magnitude of these predictions exceeded those obtained with polymorphism data
[121][133].
Beyond heterosis, epigenetic markers have been used to associate epigenetic with trait variation in experimental and natural populations. Classic quantitative genetics methods used to estimate the genetic component of the phenotype could also be used to assess natural epigenetic variations. DNA methylation represents a trait allowing the estimation of narrow-sense heritability or genetic differentiation index describing the genetic control of epigenetic variation
[134][135]. Epigenetic polymorphisms (SMPs or DMRs) could be used as epigenetic markers in genome-scan-based approaches to estimate the role of epigenetics in population structure and phenotypic variation (epiFST, epiQTL). In oilseed crop using stable and heritable methylation-sensitive amplification polymorphisms (MSAP) and retrotransposon (RT) epigenetic markers, Long
[136] identified QTLs for seven agronomic traits. In a similar MSAP-based approach in sorghum, a methylation linkage map was constructed that could help associate methylome hotspots to elite genes responsible for traits
[137]. A common question is whether epigenetic variation is a cause of differentially expressed genes or the consequence of it. Recent evidence from maize supports the former, as DMRs that are not tagged by genetic variation are more prevalent and strongly associated with the expression of genes identified in phenotyping variation
[138]. In addition to this, many metabolic traits showed notable correlations with DMRs and not with single nucleotide polymorphisms (SNPs), thus highlighting DNA methylation variation as a powerful phenotypic predictor independently of genetic polymorphism data
[138]. Epigenome-wide association-mapping approaches have also been employed in clonally propagated oil palm. This approach uncovered a recurrent epimutation, which could be shown to be responsible for parthenocarpy and loss of oil yield
[34]. Local hypomethylation near the Karma transposon could be shown to predict deleterious phenotype and is being further developed as a biomarker for juvenile screening of trees.
Beyond statistical predictions, knowledge of epigenotype–phenotype associations may provide deeper insights into the molecular regulation of agriculturally important plant traits and has the potential to contribute to breeding with useful markers. Even though there are already several studies in model plants that were able to quantify the epigenotype–phenotype interactions by distinguishing between the DNA sequence and epigenome variation, a defined epigenetic-oriented breeding strategy is yet to come. The creation of such systems in crops would allow not only the targeted use of the epigenome diversity to detect agronomically important epialleles, but also has the power to predict a plant’s performance, decrease field trials costs, and result in an increased crop production.