We have performed a systematic dose–response characterisation of the effect of DNA hypomethylation induced by 5-AzaC in B. rapa. Based on this analysis, we generated a unique hypomethylated population of 1,000 E2 sib-lines with a stochastic distribution of reduced 5mCG marks. Each line represents a unique combination of hypomethylated epialleles that are available for forward and reverse genetic screening. We carried out forward screening and investigated the range of variation for seed size and other yield components, as well as seed composition. These epi-TILLING populations are analogous to those generated using conventional nucleotide mutagens such as EMS [33, 45]. However, they do offer the distinct opportunity to explore and exploit subtle interactions with phenotype that may be associated with epiallelic variation.
Before adopting any treatment that seeks to reveal new epigenetic variation through interference in endogenous methylation control systems, it is important first to demonstrate efficacy of treatment and to optimise its application. 5-AzaC is a structural analogue of the nucleoside cytidine [26, 27] that inhibits the action of the DNA methyltransferase MET1, which in plants maintains methylation in exonic 5mCG contexts. However, there is scope for ineffective application should the 5-AzaC fail to access the target tissues, or should the time of exposure be insufficient to elicit a response. Conversely, excessive application could lead to toxicity and/or lethal levels of mutation or demethylation . It is therefore important to develop appropriate strategies to evaluate whether exposure levels were sufficient to ensure efficacy of demethylation whilst also minimising potentially damaging effects. We used frequency of survivorship and the absence of gross phenotypic abnormality to establish that the level of 5-AzaC applied was not excessive. The dose–response to 5-AzaC was sigmoidal for several traits recorded, similar to that observed for chemical mutagens such as EMS. This suggests that the B. rapa genome is able to tolerate low concentrations up to ~0.1mM 5-AzaC for the treatment period of three days during which mitoses are occurring in the meristematic cells. Above this concentration, an exponential response is observed up to ~1.0mM 5-AzaC, above which there were no apparent additional lethal effects. Moreover, there were only minor cytotoxic effects observed within hypocotyls and cotyledon tissues in seedlings. In general, the E1 plants completed their life cycle at all concentrations, and successfully set seed.
The potency of the 5-AzaC dose–response is likely to vary, depending on the dinucleotide composition and structural organisation of different genomes . Compared with EMS in the same R-o-18 genotype , 5-AzaC gave rise to a greater proportion of plants with a normal visual phenotype, although the effects on fertility were similar.
It is important next to verify that the 5-AzaC treatment apparently leading to these effects is doing so by causing hypomethylation of the genome. We examined this in two ways. First, immune-localisation of methylated cytosines revealed essentially similar distributions of FISH signal across the euchromatic regions of pachytene chromosomes in treated (E2) and control plants but with a marked reduction of signal in treated lines. This is consistent with the expected genome-wide hypomethylation associated with 5-AzaC treatment and congruent with previous observations in Triticale . Second, we compared the level of disparity between the MSAP product profiles generated by the isoschizomers HpaII (methylation sensitive) and MspI. As expected, untreated control lines exhibited wide divergence between the profiles generated by the two enzymes, largely because of their differential sensitivity to the presence of cytosine methylation. However, it was notable that this distinction had been lost in four of the treated lines sampled, but had been retained in others. The most plausible explanation of this finding is that hypomethylation triggered by the 5-AzaC treatment had been stochastically effective and was only apparent in a subset of treated plants. This finding is consistent with the similar frequency of phenotypic abnormality seen in the E2 and E3 plants. Given the stochastic nature of 5-AzaC efficacy, this relatively simple approach has direct and valuable utility in allowing early identification of E1 lines where the treatment has actually affected the intended hypomethylation. This capacity would greatly improve the efficiency of any high-throughput programme seeking to use the chemical to generate new levels of phenotypic variation, or otherwise to manipulate the methylation status for genetic or epigenetic studies of gene function.
Having generated variation in methylation status, it is important next to determine the extent of associated phenotypic variation among the treated plants. In the present study, we certainly found additional phenotypic variation at the E1 generation when compared to untreated controls, as reported previously in a range of plant species [10, 48, 49]. However, compared with B. oleracea, 5-AzaC appeared to give rise to a reduced range of morphological phenotypic effects in B. rapa R-o-18. This may be partly attributed to the smaller genome size of the Brassica A genome  and/or different chromatin organisation, particularly in terms of relative amounts of heterochromatin and/or distribution of transposable elements . Using a cytological approach, Braszewska-Zalewska  showed that the B. rapa genome has a distinct distribution of DNA methylation, primarily located in the heterochromatin when compared with B. oleracea.
There was also evidence that novel variation was present in several traits of agronomic importance. For example, flowering time is a complex multigenic trait that is regulated through the interaction of different signal transduction pathways [54–56]. Our finding that flowering time in the E1 generation is delayed following exposure to ≥ 0.1mM 5-AzaC is consistent with methylation playing some role in the control of flowering time in B. rapa. The clone of B. rapa used (R-o-18) is unresponsive to vernalisation, in common with the Arabidopsis ecotype Landsberg erecta. This phenotype in Arabidopsis has been attributed to reactivation of FWA, a repressor of flowering which is constitutively methylated in its promoter and therefore repressed in the wild type at most stages of the life cycle. FWA is maternally imprinted  and is dependent on DNA methylation for its imprinted state [59, 60]. However, we also observed a reduction in the time to flowering in a subset of E2 plants exposed to 0.5 mM 5-AzaC, a finding that is congruent with several previous reports [10, 48, 50, 61] Subsequent systematic screening of flowering time within the BraRoAZ E2 population treated with 0.25 mM 5-AzaC also indicated a reduced time to flowering. Curiously, in some cases early flowering phenotypes were observed in the E2 generation from E1 lines had been late flowering, a phenomenon that has also been found in E2 and E3 populations of B. oleracea var. italica (King, unpublished data). This may be due to segregation of alleles in the E2 generation that affect different components of the flowering pathway and is perhaps indicative of the high level of redundancy and interdependency in the epigenetic control of this important trait.
The increased variability generated by stochastic hypomethylation of the genome has most value if this variation can be fixed or at least stabilized. Previous studies have demonstrated that this can be achieved in Arabidopsis through self-pollination  or doubled haploid production (King, unpublished). The dose–response analysis informed our selection of optimal concentration of 5-AzaC to use for generation of the BraRoAZ population, where E2 seed was set from 94.5% of E1 plants. The establishment of this population is analogous to an EMS-TILLING population and is available to be used for both forward and reverse epi-genetic screening. As expected, we observed a reduced range of phenotypes compared with EMS mutagenesis in the same R-o-18 genotype . 5-AzaC only targets a subset of methylated sites where these exist, whereas EMS induces transition mutations which can lead to truncation of transcripts and modification of proteins, and is able to target all regions of the genome.
In plants, DNA methylation is at least sometimes transmitted directly through meiosis and maintained through post-meiotic mitosis giving rise to gametophytes [62, 63]. In two separate experiments (dose–response and BraRoAZ population) we found that certain phenotypes were transgenerationally inherited from E1 to E2 generation following 5-AzaC treatment. These included an overall reduction in plant stature, small seed size and modified floral morphology. The apparent loss of variation observed in E2 and E3 generations relative to E1 could arise for a number of reasons, including re-methylation of specific cell-lineages forming gametes, dominance of wild-type alleles and segregation of epistatic loci. The E2 and subsequent generations are expected to possess functional MET1 loci, and hence progressive re-methylation of specific CG sites may occur. The ability to distinguish between novel epi-alleles and collateral genetic variation, such as may result from activation of transposons, requires more detailed segregation and molecular analyses.
The range of morphological variation we observed was comparable to that observed following down-regulation of genes involved in maintenance DNA methylation [6, 14, 64]. A key attribute of the hypomethylated population we have generated is that there has only been a transitory exposure to 5-AzaC, that inhibits MET1 activity for the small number of cell divisions that occurred over three days during seed imbibition and meristem activation. In contrast, constitutive down regulation of MET1 (in mutants or by RNAi) affects a range of phenotypes including plant stature, leaves, apical dominance, flowering and fertility [6, 14, 65]. From preliminary analysis of EMS mutation of BraA.MET1.a we have evidence of effects on plant stature, leaf shape and fertility (data not shown). A similar range of abnormal phenotypes is observed in Arabidopsis ddm1 mutants [64, 66]. Although developmental abnormalities were not initially reported in B. rapa following RNAi down-regulation of BraA.DDM1, we have observed some morphological abnormalities in subsequent generations.
The structure of the BraRoAZ population provides a means for systematic screening of epiallelic variation. The distribution of 5mCG marks retained in the E2 and subsequent generations is expected to be stochastic, leading to local variations in epiallelic status of adjacent genes. This offers several advantages over constitutive down-regulation of MET1. For example, in met1 mutants of Arabidopsis up to 90% reduction in cytosine methylation can occur, predominantly in CpG dinucleotides . Availability of epiRILs in Arabidopsis generated from a cross between wild type and the ddm1 mutant offers an alternative strategy for characterising epigenetic effects at target loci . However, this approach may not be so effective in more complex genomes such as Brassica, which contain multiple paralogous genic regions, and a larger load of transposable elements.
We observed variation in a range of seed yield and composition traits, within both the E2 and E3 generations. Epigenetic processes have been implicated in the regulation of seed size, with DNA methylation appearing to play a key role. For example, crosses between a hypomethylated maternal genome and wild type paternal genome lead to increased seed size, whereas the reciprocal crosses lead to reduced seed size [11, 68, 69]. For some lines we observed inheritance of the reduced seed size phenotype from E2 to E3 generations. We were able to demonstrate that the variation in seed size is largely independent of seed number and seed yield per plant, suggesting that there is opportunity to select and recombine specific epialleles to increase overall yield.
We found that hypomethylation had an effect across the population (E2 and E3) of increasing seed protein content, with a corresponding decrease in oil content. Given our null hypothesis was random variation around the wild-type mean for both components, this indicates that hypomethylation could lead to de-repression of specific genes that modulate the oil:protein ratio, either via specific resource allocation pathways, or through interaction with the normal seed development and maturation programme.
In addition, we have demonstrated that hypomethylation can give rise to large modulations in the proportion of key seed fatty acid components, and that these are transmitted through at least one meiotic event. These effects include decreases in erucic acid that correspond to increases in linoleic and/or palmitic acid. This level of variation is consistent with that observed in surveys of natural genetic variation and population segregation within Brassica species . The inference of which steps in the accepted fatty acid synthesis pathways are affected allows us to postulate which sets of genes may have been up-regulated as a result of hypomethylation. This either may be due to direct effects on a specific enzyme or as a result of de-repression of an activator of a specific enzyme. For example, the reduction in the levels of erucic acid and corresponding increase in levels of linoleic acid is consistent with activation of FAD2. The increases in palmitic acid with concomitant increase in linoleic may indicate an activation of a gene earlier in the fatty acid elongation pathways.
There was no consistent pattern of inheritance at a population level between the E2 and E3 for seed composition traits. However, this is as expected for relatively complex traits in bulked sib-seed, where there may be strong G×E interactions compounded with segregation of one or more affected loci. Moreover, greater variation was observed in the S3 controls compared with S2, underlying the importance of studying epiallelic variation in well controlled and reproducible environments. Variation in 5mCG marks have been implicated in modulating G×E interactions in the vernalisation response of FLC and MAF-5. DNA methylation marks also tend to exclude histone H2A.Z , which appears to play an essential role in temperature perception . As with EMS TILLING populations the stochastic distribution of (epi-)mutant alleles in early generations is likely to mask the phenotypic effect of single loci. Thus it is necessary to generate backcross lines (to wild type) through at least three generations prior to carrying out a rigorous segregation analysis and identifying the relevant loci.
Preliminary transcript analysis of leaf tissue from individual E3 plants did not show any evidence of widespread variation in gene expression, with only a small number of genes being up-regulated and none down-regulated. This is consistent with the observed phenotypic observations and de-repression of transcription expected, given the stochastic distribution of modified 5 mC sites that will have arisen from the original 5-AzaC treatment. Under controlled growth conditions on two distinct occasions, more variation was observed in transcript profile between homozygous wild type controls than was observed between two sib E3 plants grown on single occasions. More comprehensive screening of lines is required to understand the distribution and pattern of epiallelic variation affecting transcript levels within the population, and this may be amenable to targeted reverse epiallelic assays.