Additive inheritance of histone modifications in Arabidopsis thaliana intra-specific hybrids

Authors

  • Ali M. Banaei Moghaddam,

    1. Leibniz Institute of Plant Genetics and Crop Plant Research, Corrensstraße 3, 06466 Gatersleben, Germany
    Search for more papers by this author
  • Francois Roudier,

    1. Institut de Biologie de l’Ecole Normale Supérieure (IBENS), Centre National de la Recherche Scientifique (CNRS) UMR8197, Institut National de la Santé et de la Recherche Médicale (INSERM) U1024, Paris, France
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Michael Seifert,

    1. Leibniz Institute of Plant Genetics and Crop Plant Research, Corrensstraße 3, 06466 Gatersleben, Germany
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Caroline Bérard,

    1. AgroParisTech, Institut National de la Recherche Agronomique (INRA) UMR518, Paris, France
    Search for more papers by this author
  • Marie-Laure M. Magniette,

    1. AgroParisTech, Institut National de la Recherche Agronomique (INRA) UMR518, Paris, France
    Search for more papers by this author
  • Raheleh Karimi Ashtiyani,

    1. Leibniz Institute of Plant Genetics and Crop Plant Research, Corrensstraße 3, 06466 Gatersleben, Germany
    Search for more papers by this author
  • Andreas Houben,

    1. Leibniz Institute of Plant Genetics and Crop Plant Research, Corrensstraße 3, 06466 Gatersleben, Germany
    Search for more papers by this author
  • Vincent Colot,

    1. Institut de Biologie de l’Ecole Normale Supérieure (IBENS), Centre National de la Recherche Scientifique (CNRS) UMR8197, Institut National de la Santé et de la Recherche Médicale (INSERM) U1024, Paris, France
    Search for more papers by this author
  • Michael F. Mette

    Corresponding author
    1. Leibniz Institute of Plant Genetics and Crop Plant Research, Corrensstraße 3, 06466 Gatersleben, Germany
    Search for more papers by this author

(fax +49 39482 5137; e-mail mette@ipk-gatersleben.de).

Summary

Plant genomes are earmarked with defined patterns of chromatin marks. Little is known about the stability of these epigenomes when related, but distinct genomes are brought together by intra-species hybridization. Arabidopsis thaliana accessions and their reciprocal hybrids were used as a model system to investigate the dynamics of histone modification patterns. The genome-wide distribution of histone modifications H3K4me2 and H3K27me3 in the inbred parental accessions Col-0, C24 and Cvi and their hybrid offspring was compared by chromatin immunoprecipitation in combination with genome tiling array hybridization. The analysis revealed that, in addition to DNA sequence polymorphisms, chromatin modification variations exist among accessions of A. thaliana. The range of these variations was higher for H3K27me3 (typically a repressive mark) than for H3K4me2 (typically an active mark). H3K4me2 and H3K27me3 were rather stable in response to intra-species hybridization, with mainly additive inheritance in hybrid offspring. In conclusion, intra-species hybridization does not result in gross changes to chromatin modifications.

Introduction

Extensive studies of DNA methylation and histone modifications in Arabidopsis thaliana (Turck et al., 2007; Zhang et al., 2007, 2009; Cokus et al., 2008) and rice (Oryza sativa) (He et al., 2010; Zemach et al., 2010) have revealed that plant genomes are earmarked by well-defined patterns of chromatin marks. In the context of their transcriptional activity or inactivity, particular sequence classes such as genes or repeat elements are preferentially associated with distinct patterns of DNA methylation and histone modifications (Roudier et al., 2009; Teixeira and Colot, 2010).

The combining of related but distinct genomes with their respective patterns of chromatin marks in inter-species hybridization and allopolyploid formation often results in changes to chromatin marks. In hybrids of various Arabidopsis species, one parental set of ribosomal RNA genes was shown to be silenced within a few generations, but could be re-activated by interfering with either DNA methylation or histone deacetylation, suggesting a pivotal role for chromatin modification in the regulation of expression of orthologous genes (Lee and Chen, 2001; Lawrence et al., 2004). Gene expression studies in synthetic allopolyploid Arabidopsis (Comai, 2000) and cotton (Gossypium hirsutum) (Brubaker et al., 1999) revealed that gene silencing occurs during the first or second generation after hybridization. However, other studies in allopolyploid Spartina anglica, Brassica juncea and cotton showed that the activity of parental genomes remained unchanged (Axelsson et al., 2000; Liu et al., 2001; Baumel et al., 2002). Induction of DNA methylation changes after hybridization was reported for synthetic Cucumis allopolyploids (Chen and Chen, 2008). Similarly, in experimentally synthesized Brassica napus (Xu et al., 2009) and Arabidopsis allopolyploids (Madlung et al., 2002), 7 and 8%, respectively, of the tested DNA sites showed changes in cytosine methylation status in comparison with their respective diploid progenitors.

Less is known about the stability or dynamics of chromatin modifications in response to intra-species hybridization. Limited differential DNA methylation (approximately 1% gain or loss) in comparison with the respective progenitors was found for intra-species hybrids of rice (Xiong et al., 1999) and cotton (Zhao et al., 2008). In two rice cultivars from different sub-species, Oryza sativa japonica and O. sativa indica, variation of DNA methylation, and, to a lower extent, variation of histone modifications H3K4me3 and H3K27me3, was observed between parental lines. In reciprocal hybrids of the two rice sub-species, distinct non-additive patterns of chromatin marks were observed. The level of changes after hybridization was higher for DNA methylation, and both histone modifications were mainly inherited additively in the hybrids (He et al., 2010). Inbred accessions of A. thaliana also display substantial DNA methylation variation between each other (Vaughn et al., 2007). Investigation of the DNA methylation pattern in two different accessions of A. thaliana and their reciprocal F1 hybrid progeny showed that DNA methylation polymorphisms are mostly inherited additively, with only limited changes after hybridization (Zhang et al., 2008; Banaei Moghaddam et al., 2010; Groszmann et al., 2011).

Here, we determined whether intra-species crosses between inbred lines lead to changes in chromatin marks other than DNA methylation using accessions of A. thaliana as a model. We selected histone H3 dimethylated at lysine 4 (H3K4me2) and histone H3 trimethylated at lysine 27 (H3K27me3) as contrasting histone H3 modifications marks (Fuchs et al., 2006; Kouzarides, 2007; Roudier et al., 2009). Histone H3K4me2 was chosen as a general euchromatic mark that is absent from silent repeat elements, and H3K27me3 was chosen as a euchromatic mark that is mostly associated with genes repressed by polycomb repressive complex 2 (Schubert et al., 2006; Turck et al., 2007; Zhang et al., 2007). A genome-wide ‘ChIP on chip’ analysis of H3K4me2 distribution in A. thaliana indicated that 6% of sequences are targeted by this mark (Zhang et al., 2009). Of these target regions, 93% were genes, with particular enrichment of H3K4me2 in the promoter and 5′ end of transcribed regions, and only 1.3% of the target regions were transposable elements (TEs). In contrast, histone H3K27me3 is associated with silent genes distributed in euchromatic regions that are subject to tissue-specific or developmentally regulated expression (Turck et al., 2007; Zhang et al., 2007). Genome-wide analyses revealed that at least 15–20% of A. thaliana genes are targeted by this histone mark and show tissue-specific expression patterns. H3K27me3-marked domains are largely coincident with the entire transcribed region of genes (Turck et al., 2007; Zhang et al., 2007).

To study the stability of histone modification patterns in response to intra-species hybridization of various inbred accessions of A. thaliana (Col-0, C24 and Cvi), we performed chromatin immunoprecipitation in combination with genome tiling array hybridization (ChIP on chip) analyses. Changes in the H3K27me3 and H3K4me2 distribution between Col-0, Cvi and C24 were identified, with a greater range of variations for H3K27me3 than H3K4me2. H3K4me2 and H3K27me3 were rather stable after intra-species hybridization, with additive inheritance in Col-0 × Cvi and Col-0 × C24 F1 hybrid offspring. Changes in the distribution of histone modifications after hybridization were detected in 346 genes for H3K4me2 in Col-0 × Cvi progeny, and in 1233 and 876 genes for H3K27me3 in Col-0 × Cvi and Col-0 × C24 progeny, respectively. However, these changes were rather random and were not associated with particular sequence categories.

Results

Genomic sequence variation among A. thaliana accessions resides mainly in transposable elements

Comparative genomic hybridization (CGH) experiments were performed using Arabidopsis whole-genome tiling NimbleGen arrays to identify differences in the genomic sequences of A. thaliana accessions Cvi and C24 compared to the reference accession Col-0. The CGH analysis, which detects copy number variation and sequence polymorphisms, was a necessary prerequisite for the comparison of histone modification patterns between Cvi, C24 and Col-0. Based on this information, it was possible to distinguish whether differential hybridization signals of labelled DNA derived from immunoprecipitated chromatin from the various accessions were due to differences in histone modifications rather than variation in the DNA sequence. In total, 6.0 and 5.5% of tiles showed significant CGH polymorphisms for Col-0 versus Cvi and Col-0 versus C24, respectively (Table 1). Most of the CGH polymorphic tiles indicated a decrease in copy number of the corresponding sequence in C24 and Cvi compared to Col-0.

Table 1.   CGH analysis for Col-0 versus C24 and Col-0 versus Cvi
 Col-0 versus CviCol-0 versus C24
Percentage of tiles showing CGH polymorphism6.05.5
Percentage of tiles with lower copy number in C24 or Cvi (total size in kb)5.3 (5395)5.2 (5221)
Number of tiles per domain2–3692–372
Mean size of domain (kb)3.33.9
90% of domains had a size of less than (kb)68
Size of largest domain (kb)4057
Percentage of tiles with higher copy number in C24 or Cvi (total size in kb)0.67 (631)0.35 (308)
Number of tiles per domain2–1692–54
Mean size of domain (kb)3.01.9
90% of domains had a size of less than (kb)6.54
Size of largest domain (kb)269

Further analysis focused on CGH polymorphic domains containing two or more consecutive CGH polymorphic tiles (Table 1). Of the tiles present in CGH polymorphic domains identified between Col-0 and Cvi, as well between Col-0 and C24, 93% were identical. However, copy number varied in some cases in opposite directions between accessions, with an increase in Col-0 versus Cvi and a decrease in Col-0 versus C24, and vice versa. Annotation of tiles within CGH polymorphic domains indicated that copy number variation mainly affects TEs, while genic regions and 5′ and 3′ untranslated regions (UTRs) are more conserved between the analysed accessions (Figure 1). Ontology categorization of the genes (excluding TEs) associated with CGH polymorphic domains according to the Munich Information Center for Protein Sequences (MIPS) Functional Catalogue (Ruepp et al., 2004) indicated an excess of genes with functions in signal transduction (Figure 2, category 30) or cell defence (Figure 2, category 32). As annotated TEs were excluded prior to ontology analysis, they are absent from the gene ontology data (Figure 2, category 38). In addition, unclassified proteins (Figure 2, category 99) were common.

Figure 1.

 Classification of CGH polymorphic tiles based on their sequence types.
CGH polymorphic tiles between A. thaliana accessions (a) Col-0 versus Cvi and (b) Col-0 versus C24, were classified based on annotation (TAIR8) of their underlying sequences. Red and green bars represent tiles with higher and lower copy numbers in Cvi (a) and C24 (b) relative to Col-0, respectively. Grey bars indicate controls by random counts. Cases of significant deviation between CGH data (green or red bars) and random counts (grey bars) as indicated by a P value <0.01 are indicated by asterisks.

Figure 2.

 Ontology classification of genes localized in CGH polymorphic regions.
Genic sequences (according to TAIR8) that were found to be localized in CGH polymorphic regions indentified between (a) Col-0 and Cvi and (b) Col-0 and C24 were classified into gene ontology groups according to the MIPS Functional Catalogue (Ruepp et al., 2004) to identify potentially over-represented categories. Black bars indicate the frequencies of CGH polymorphic genic sequences; as a control, grey bars indicate the frequency of randomly selected genic sequences in the various gene ontology classes as a percentage of the total number of genes in each class. Asterisks indicate cases of significant deviation between the frequencies of polymorphic and randomly selected sequences in a given category with a P value <0.001.

In summary, the CGH data revealed sequence polymorphisms among the analysed accessions of A. thaliana. These polymorphisms occurred in all types of sequences, but TEs and genes involved in signal transduction and cell defence were over-represented.

Genome-wide patterns of histone modifications H3K4me2 and H3K27me3 show variation among A. thaliana accessions

Next, we performed ChIP on chip analysis using Arabidopsis whole-genome tiling NimbleGen arrays to determine the genome-wide distribution of H3K4me2 in accessions Col-0 and Cvi and in Col-0 × Cvi F1 hybrids, and of H3K27me3 in Col-0, Cvi and C24, and in Col-0 × Cvi and Col-0 × C24 F1 hybrids. The quality of ChIP assays was confirmed by quantitative PCR using primers specific for sequences known to be associated with H3K4me2 or H3K27me3 (Figure S1) from previous studies (Turck et al., 2007; Zhang et al., 2007, 2009).

Tiles showing polymorphic hybridization signals in CGH were excluded from the ChIP on chip analysis to avoid the influence of sequence differences among accessions. Tiles that were found to be significantly associated with a given modification in any of the analysed genotypes were assigned to histone modification domains of at least three successive tiles that correspond to a region of at least 0.3 kb (Table S1). Given the mean size of chromatin fragments (0.8 kb) and the resolution provided by the NimbleGen microarrays (165 bp) used in the ChIP on chip experiments, tiles that could not be assigned to such domains were not considered for further analysis. Subsequently, histone modification domains were categorized according to the TAIR8 annotation into genes or TEs. In the case of large domains, it may be that one domain simultaneously harbours genic sequences and TEs.

First the distribution of histone modifications was compared between parental accessions Col-0 and Cvi, as well as Col-0 and C24 (Figure S2). H3K4me2, a classical euchromatic histone mark, was associated with domains ranging in length from 314 bp to 31.2 and 21.9 kb in Col-0 and Cvi, respectively (Table 2 and Figure S3). Overall, 94% of these domains coincided with genes (Table S2) and 9% with TE sequences (Table S3). Intersection analysis of H3K4me2-marked domains in Col-0 and Cvi revealed that 87% of them were common between both accessions. These common domains coincided with 93% (21 908) of the genes and 60% (1526) of the TE sequences associated with H3K4me2 (Figure 3). The length of the 13% of H3K4me2-marked regions that differed between the two accessions (Table 3) ranged from 314 bp to 6.7 kb, and corresponded to 4.6% (1053) and 2.6% (581) of the genes and 26% (537) and 23.9% (479) of the TE sequences in Col-0 and Cvi, respectively (Table 3). Overall, a larger proportion of TEs than genes showed H3K4me2 polymorphisms. However, the absolute numbers of genes and TEs showing differential association with H3K4me2 between Col-0 and Cvi were similar, as fewer TEs than genes are associated with this histone modification.

Table 2.   H3K4me2- and H3K27me3-associated domains in Col-0, Cvi and C24
ExperimentaGenotypeModificationAssociated tiles (from total of 717 235)Number of domainsDomain size (kb)Annotation
MaximumMinimumMeanGenesTEs
  1. aChromatin preparations for experiments A and B were performed in parallel but the one for experiment C was performed independently.

ACol-0H3K4me2287 28521 53931.20.32.222 9612063
CviH3K4me2281 00020 97221.90.32.222 4892005
BCol-0H3K27me3140 03511 18222.70.32.191254950
CviH3K27me3138 79112 585200.31.993265357
CCol-0H3K27me3147 318983426.70.32.594274704
C24H3K27me3125 94710 06822.50.32.181184471
Figure 3.

 Intersection analysis of domains, genes and TEs associated with H3K4me2 or H3K27me3 in three A. thaliana accessions.
The sets of (a) domains, (b) genes and (c) TEs that were found to be associated with H3K27me3 or H3K4me2 in ChIP on chip experiments using whole-genome NimbleGen tiling arrays in A. thaliana accessions Col-0 (green circles), Cvi (yellow circles) and C24 (red circles) were subjected to intersection analysis to determine common elements (intersection area of two circles) and unique elements (areas outside intersections). Due to the analysis method, the sums of domain numbers do not necessarily correspond to the total numbers of domains in Table 2.

Table 3.   H3K4me2 and H3K27me3 polymorphic domains in Col-0 versus Cvi and Col-0 versus C24
ExperimentaSpecific for accessionModific ationNumber of domainDomain size (kb)Annotation
MaximumMinimumMeanGenesTEs
  1. aChromatin preparations for experiments A and B were performed in parallel but the one for experiment C was performed independently.

ACol-0H3K4me215856.60.30.671053537
CviH3K4me210414.80.30.72581479
BCol-0H3K27me313394.60.30.72927895
CviH3K27me321415.70.30.8111281302
CCol-0H3K27me320537.40.30.917621249
C24H3K27me315036.70.30.774531016

Genes that were differentially associated with H3K4me2 in either of the parental lines were randomly distributed in various gene ontology groups according to the MIPS Functional Catalogue (Figure S4) (Ruepp et al., 2004) and the DAVID tool (Table S4) (Huang et al., 2009). Similar results were obtained for the distribution of H3K4me2 in and between Col-0 and C24 accessions analysed using a chromosome 4 tiling array (data not shown) (Turck et al., 2007).

H3K27me3, a histone mark associated with the repression of genes, was found over domains ranging from 315 bp to 26.7 kb in length (Table 2 and Figure S3) in Col-0, Cvi and C24. In Col-0, 67% of these domains coincided with genes and 36% with TEs. Similar results were obtained for Cvi and C24 (Figure 3, Table 2, and Tables S2 and S3). Intersection analysis of H3K27me3-marked domains revealed that 87% of them were common between Col-0 and Cvi. These common domains coincided with 80% (8198) of the genes and 58% (4055) of the TEs marked by H3K27me3 (Figure 3). Compared to H3K4me2, fewer genes and more TEs were associated with H3K27me3 in Col-0 and Cvi (Figure 3). Some H3K27me3 domains were unique to one of the analysed accessions (Table 3). The length of these H3K27me3 polymorphic domains ranged from 314 bp to 6.6 kb in Col-0 and 5.7 kb in Cvi. These domains coincided with 9% (927) and 11% (1128) of the genes, and 17% (895) and 25% (1302) of the TEs associated with H3K27me3, respectively. Consistent results were obtained for comparison of H3K27me3 polymorphic domains between Col-0 and C24 (Figure 3 and Table 3). Thus, in general, more TEs coincided with H3K27me3 than with H3K4me2 polymorphic domains. Similar numbers of genes and TEs were differentially associated with H3K27me3 in Col-0 compared to Cvi and Col-0 compared to C24. These genes were randomly distributed in various gene ontology groups (Figure S4 and Table S4).

Comparison of genes associated with H3K27me3 in the reference accession Col-0 with those found in previous studies (Turck et al., 2007; Zhang et al., 2007) revealed approximately 80% overlap, indicating good reproducibility despite the differences in Plant materials and Experimental Procedures (Figure S5). In addition, our study identified H3K27me3-associated genes that were not found in previous studies.

Taken together, our findings indicate that H3K4me2 and H3K27me3 distribution patterns are highly conserved in different A. thaliana accessions, with few local variations in a random manner regardless of TEs and gene ontology.

Intra-species hybridization mediates limited alterations in H3K4me2 and H3K27me3 distribution patterns

We next investigated whether intra-species hybridization could generate alternative distribution patterns of H3K4me2 in hybrid offspring between Col-0 and Cvi (Figure 4a–c, Figure S2, and Tables S1–S3). In order to focus on the most relevant cases, the analysis was performed on domains that were associated with H3K4me2 exclusively in both parental accessions, but not in F1 hybrid progeny (Figure 4d, category IV), or in the F1 hybrid progeny, but not in the parental accessions (Figure 4d, category III). The few hybridization-responsive domains identified (3% of all H3K4me2-associated domains) corresponded to 346 genes and 226 TEs (1.5 and 8% of H3K4me2-associated genes and TEs, respectively) and their length ranged from 0.3 to 1.6 kb (Table 4). In comparison to genes, TEs were over-represented among hybridization-responsive H3K4me2 domains. The hybridization-responsive H3K4me2-associated genes were randomly distributed between various ontology groups and gene families (Figure S6 and Table S4). Similar results were found for Col-0 × C24 F1 hybrids in ChIP on chip analysis using the chromosome 4 tiling array (data not shown).

Figure 4.

 Intersection analysis of domains, genes and TEs associated with H3K4me2 or H3K27me3 in A. thaliana accessions and their F1 hybrids.
(a–c) The sets of (a) domains, (b) genes and (c) TEs that were found to be associated with H3K27me3 or H3K4me2 in ChIP on chip experiments using whole-genome NimbleGen tiling arrays in A. thaliana parental accessions Col-0 (green circles), Cvi (yellow circles) and C24 (red circles), and their F1 hybrids Col-0 × Cvi, and Col-0 × C24, respectively (blue circles), were subjected to intersection analysis.
(d) The seven areas of each three-circle intersection diagram define seven categories of elements. Categories I and II comprise cases in which a histone mark is detected in one of the parental accessions (P1 or P2, respectively), but not in the other parental accession or the F1 hybrid. Category III comprises cases in which a histone mark is detected in neither parental accession, but is present in the F1 hybrid. Category IV comprises cases in which a histone mark is detected in both parental accessions, but not in the F1 hybrid. Categories V and VI comprise cases in which a histone mark is detected in one of the parental accessions (P1 or P2, respectively) and in the F1 hybrid, but not in the other parental accession. Category VII comprises cases in which a histone mark is detected in both parental accessions and the F1 hybrid. Categories III and IV indicate changes in histone marks in F1 hybrids in comparison with the parental accessions, while the other categories either indicate no change (category VII) or are not conclusive with regard to changes (categories I, II, V and VI).

Table 4.   H3K4me2 and H3K27me3 hybridization-responsive domains in Col-0 × Cvi and Col-0 × C24
ExperimentaCrossesModificationsHybridization-responsive domains (Figure 4, categories III and IV)
Number of domainsDomain size (kb)Annotation
MaximumMinimumMeanGenesTEs
  1. aChromatin preparations for experiment A and B were performed in parallel but the one for experiment C was performed independently.

ACol-0 × CviH3K4me25821.60.30.6346226
BCol-0 × CviH3K27me319012.20.30.61233654
CCol-0 × C24H3K27me313632.40.30.6876674

Whole-genome comparison (Figure S2) and intersection analysis of H3K27me3-marked domains, genes and TE sequences involved were also performed for Col-0, Cvi and Col-0 × Cvi (Figure 4). Hybridization-responsive domains (13% of all H3K27me3-associated domains; Figure 4d, categories III and IV) corresponded to 1233 genes and 654 TEs (11 and 10% of total H3K27me3-associated genes and TEs, respectively), and ranged in size from 0.3 to 2.2 kb (Table 4). No over-representation of either TEs or genes was detected. The hybridization-responsive H3K27me3-marked genes were randomly distributed between ontology groups and gene families (Figure S6 and Table S4). Similar H3K27me3 dynamics were found in Col-0 × C24 hybrids (Figure 4a–c and Tables S1–S3).

In conclusion, the genome-wide distribution of H3K4me2 is rather stable and that of H3K27me3 is slightly more dynamic in response to intra-species hybridization. TEs as well as genes showed changes in both histone modifications. No obvious over-representation of particular gene ontology groups or families was found. The changes in the two marks analysed happened largely independently. For very few genes, histone modifications changed simultaneously in response to hybridization (Figure S7).

Discussion

By combining comparative genomic hybridization and epigenomic profiling, we have shown that H3K4me2 and H3K27me3 distribution patterns are overall very similar in various Arabidopsis thaliana accessions, and remain largely unchanged in their F1 progeny.

Sequence polymorphisms between A. thaliana accessions are mainly found in transposable elements and genes undergoing rapid evolution

Our CGH analysis of the genomes of Cvi and C24 relative to Col-0 identified regions showing DNA sequence polymorphism between these accessions of A. thaliana. The number of polymorphic regions identified by CGH was slightly higher for Cvi than for C24 in comparison with Col-0. This is consistent with a previous analysis based on single nucleotide polymorphisms, which indicated that Cvi is more divergent from Col-0 than C24 is (Schmid et al., 2003; Clark et al., 2007).

As TE sequences change more rapidly than genes (Kazazian, 2004), sequence differences between A. thaliana accessions are not randomly distributed, and TEs differ more than other parts of the genome (Figure 1). A similar non-random distribution of sequence polymorphisms was previously found by CGH analysis between A. thaliana accessions (Clark et al., 2007) and between maize (Zea mays) inbred lines B73 and Mo17 (Springer et al., 2009). In maize, approximately 70% of polymorphic regions were located in intergenic regions.

Among the CGH polymorphic regions, gene ontology analysis indicated over-representation of functions associated with signal transduction and cell defence (Figure 2). Copy number variation between accessions of A. thaliana has previously been reported for loci involved in plant disease resistance (Noel et al., 1999). Furthermore, consistent with our findings, a comparison of 20 accessions of A. thaliana by microarray-based whole-genome re-sequencing revealed that members of defence-related families such as nucleotide-binding leucine-rich repeat genes and receptor-like kinase genes are over-represented among genes that are affected by sequence polymorphisms (Clark et al., 2007). The enhanced rate of sequence variation indicates rapid evolution of these gene families.

Fewer CGH polymorphic tiles indicated an increase in copy number in C24 and Cvi compared with those that indicated a decrease in copy number in both accessions (Figure 1). This could be due to the fact that the tiling arrays used in this study are based on the Col-0 reference sequence. Thus, sequences that solely exist in Col-0 but not in C24 or Cvi are classified as absent from C24 or Cvi, whereas sequences that solely exist in C24 or Cvi but not in Col-0 are not detected. Therefore, our CGH results probably under-estimate the level of sequence polymorphism between A. thaliana accessions. CGH-identified polymorphic sequences were excluded from the ChIP on chip analysis, as it was impossible to distinguish whether differences in signal intensities between the accessions were due to differences in the DNA sequence or the histone modification status. Thus, the possibility cannot be excluded that some fast-evolving genomic regions that were removed from analysis showed differential chromatin marking between accessions.

Histone modification patterns are conserved between accessions of A. thaliana

Several previous studies have detected differential DNA methylation in various A. thaliana accessions (Vaughn et al., 2007; Zhang et al., 2008; Banaei Moghaddam et al., 2010). Not only sequence conservation, but also DNA methylation patterns, were found to be more similar between Col-0 and C24 than between Col-0 and Cvi (Vaughn et al., 2007). Our ChIP on chip data extend these observations to histone H3K4me2 and H3K27me3 distribution patterns. These differ between accessions of A. thaliana, consistent with the situation observed in cultivars of rice (He et al., 2010).

Among domains with different histone modification status between accessions, the mean length was <1 kb for both histone marks, while the mean length of the conserved H3K4me2- and H3K27me3-marked domains exceeded 2 kb. Thus, polymorphic histone modification domains are restricted to rather small regions. Despite their shorter length, these differentially marked domains may be associated with locus-specific differential regulation in the various accessions. For instance, TE sequences next to genes can affect their transcriptional regulation through deposition of repressive chromatin modifications. A TE in an intron of FLOWERING LOCUS C (FLC) in A. thaliana accession Landsberg erecta causes transcriptional inactivation of this locus, and consequently earlier flowering of Ler in comparison to Col-0 (Liu et al., 2004). TEs also alter the expression of adjacent genes in wheat (Triticum aestivum) (Kashkush et al., 2003). It remains to be determined whether the regions differentially marked by the repressive H3K27me3 or active H3K4me2 modifications in Cvi or C24 in comparison to Col-0 also show differential expression between accessions.

Consistently, we found that polymorphisms in H3K27me3 (typically a repressive mark) were associated with both TEs and genes. In contrast, polymorphisms in H3K4me2 (typically an active mark) were mainly restricted to genes. This agrees well with genome-wide high-resolution analyses of histone modifications in Saccharomyces cerevisiae and mammals, which detected H3K4me2 in regions undergoing transcription across the body (Pokholok et al., 2005) and in the vicinity of active genes (Bernstein et al., 2005). Histone modification polymorphisms in genes were not associated with particular gene ontology classes.

The mechanisms by which histone modification polymorphisms are maintained over generations are not clear (Saze, 2008). On the one hand, heritable maintenance of particular chromatin states cannot be excluded. On the other hand, chromatin modifications may alter as consequences of changes in transcriptional activity. In A. thaliana, H3K27me3 is deposited by polycomb repressive complex 2 (Schubert et al., 2006), and has been found to be associated with approximately 4400 genes, many of which are differentially expressed during development (Zhang et al., 2007). As accessions may have evolved specific developmental programs (Chen, 2010), some of the H3K27me3 polymorphisms observed in our study could correspond to differences in gene expression patterns. Indeed, gene activity and H3K4 and H3K27 methylation levels were correlated in different rice cultivars (He et al., 2010).

The observed similarities and differences suggest a role for chromatin modifications in addition to that of DNA sequence polymorphisms in the diversity of various accessions of A. thaliana.

Inheritance of H3K27me3 and H3K4me2 distribution patterns in hybrid offspring is additive and only to a small extent responsive to intra-specific hybridization

Comparison of H3K4me2 and H3K27me3 distribution patterns between hybrid offspring and parental inbred lines revealed limited changes in intra-specific hybrids. These results are in agreement with a previous study on A. thaliana accessions and their intra-specific hybrids regarding the distribution of histone methylation marks at the microscopic level (Banaei Moghaddam et al., 2010), as well as a study that compared histone methylation patterns in hybrids between rice cultivars (He et al., 2010). Similarly, inheritance of DNA methylation polymorphisms has been shown to be additive (Zhang et al., 2008; Banaei Moghaddam et al., 2010).

We conclude that intra-specific hybridization in A. thaliana does not result in global epigenomic rearrangements. More investigations are required to analyse whether this conclusion is also valid for other chromatin modifications. More generally, it would be interesting to determine whether heritable variations in epigenomic patterns between inbred lines contribute to hybrid performance in addition to sequence polymorphisms.

Experimental Procedures

Plant materials

A. thaliana accessions Col-0, C24 and Cvi and their reciprocal F1 hybrid offspring Col-0 × C24, C24 × Col-0, Col-0 × Cvi and Cvi × Col-0 were used (Banaei Moghaddam et al., 2010). Approximately 400 seeds of each sample were surface-sterilized and cultured in liquid medium, and grown for 10 days under controlled conditions with 16 h light per day (light intensity of approximately 100 μE), 22°C day temperature and 18°C night temperature (Lippman et al., 2004).

DNA extraction, chromatin immunoprecipitation and array hybridization

Plant genomic DNA used for CGH was extracted using the Qiagen DNeasy plant DNA extraction system (http://www.qiagen.com/) according to manufacturer’s instructions. ChIP assays were performed essentially as described previously (Gendrel et al., 2005) using anti-H3K4me2 (07-030) and H3K27me3 (07-449) antibodies from Upstate/Millipore (http://www.millipore.com). Each experiment was performed in two biological replicates.

DNA recovered after immunoprecipitation (IP) and directly from input Col-0 chromatin (INPUT), or genomic DNA extracted from the various genotypes, was amplified, differentially labelled and co-hybridized in dye-swap experiments to correct for dye biases, as previously described (Lippman et al., 2004; Turck et al., 2007) for the chromosome 4 tiling microarray, or according to the manufacturer’s instructions for the whole-genome tiling arrays (Roche NimbleGen, http://www.nimblegen.com). The Arabidopsis chromosome 4 tiling microarray comprised 21 800 printed features, with a mean size of 1 kb, covering the main part of chromosome 4. The heterochromatic knob on the short arm and several megabases of pericentromeric heterochromatin are included, and account for 16% of the 18.6 Mb covered by the array (Martienssen et al., 2005). Details of array design and production are described by Vaughn et al. (2007). This platform has been deposited to the Gene Expression Omnibus (GEO) under accession number GPL10172. The whole-genome tiling microarray consists of 50–75 nt tiles, with a mean spacing of 165 nt, that are distributed across the entire genome sequence (TAIR7) without repeat masking. These tiles have a mean melting temperature of 74°C, and 88% of them match a unique position in the genome. This custom design was split into two arrays of 360 718 tiles each, using every other tile in each array (GEO accessions GPL10919 and GPL10920).

Comparative genomic hybridization (CGH) analysis

CGH experiments were performed for Col-0 versus Cvi and Col-0 versus C24. Data obtained using Arabidopsis whole-genome tiling NimbleGen arrays were analysed using hidden Markov models (Seifert et al., 2009). A fully connected three-state hidden Markov model with state-specific Gaussian emission densities was adapted to each CGH experiment using a Bayesian Baum–Welch algorithm. Decoding of the status of each tile (deleted, unchanged or amplified) was performed using the Viterbi algorithm. Groups of contiguous tiles with log ratios that were significant different from zero for each set of compared accessions were interpreted as representing regions of copy number variation. For interpretation of CGH data, only contiguous CGH polymorphic tiles that were consistently found in both directions of dye-swap experiments were included.

ChIP on chip analysis

Raw hybridization data obtained with the chromosome 4 tiling array were normalized as described previously (Turck et al., 2007), and an anova model was applied to whole-genome data to remove technical biases. Normalized data were analysed using the ChIPmix method (Martin-Magniette et al., 2008), which was adapted to handle multiple biological replicates simultaneously. This method is based on a mixture model of regressions, the parameters of which are estimated using an expectation-maximization (EM) algorithm. For each tile, a posterior probability, defined as the probability of enrichment given the log(INPUT) and log(IP) intensities, is used to classify the tile into the normal or enriched class. A false-positive risk is determined by defining the probability of obtaining a posterior probability at least as extreme as the one that is actually observed when the tile is normal. False-positive risks are then adjusted by the Benjamini–Hochberg procedure, and tiles for which the adjusted false-positive risk is lower than 0.01 are considered enriched. Previously published data (Turck et al., 2007; Vaughn et al., 2007) were re-analysed using the same procedure. Neighbouring enriched tiles are combined into domains, requiring minimal runs of 1.6 kb or 300 bp and allowing maximal gaps of 800 or 200 bp for chromosome 4 or whole-genome data, respectively. Enriched but isolated tiles were not considered for further analyses.

Scatter plots and Pearson correlation analyses revealed higher correlation between data sets for histone modification polymorphic tiles among biological replicates for one accession than between datasets of different accessions (Figure S8). ChIP assays were validated by quantitative PCR for sequences known to be associated or not associated with the histone modification of interest (Figure S1 and Table S5) (Turck et al., 2007; Zhang et al., 2007, 2009). For subsequent analysis of ChIP on chip data for Col-0, Cvi and C24 and their intra-specific hybrids, all CGH polymorphic tiles (including singletons and those tiles that were detected only in one of two CGH technical replicates) were excluded.

Data availability and computational analyses

Raw and processed data have been deposited to the National Center for Biotechnology Information Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) under accession GSE24836, and to CATdb (http://urgv.evry.inra.fr/CATdb) (Samson et al., 2004; Gagnot et al., 2008). In addition, array data and genome annotations are available for visualization at http://epigara.biologie.ens.fr/cgi-bin/gbrowse/a2e/. Gene ontology and gene family analysis were performed using the MIPS Functional Catalogue (http://mips.helmholtz-muenchen.de/proj/funcatDB/) (Ruepp et al., 2004) and the DAVID tool (http://david.abcc.ncifcrf.gov/) (Huang et al., 2009).

Acknowledgements

This work was supported by Deutsche Forschungsgemeinschaft grant HO 1779/7-1/2 to M.F.M. and A.H. within the framework program SPP 1149 ‘Heterosis in Plants’. Bioinformatics analyses by M.S. were supported by the Ministry of Culture of the State of Saxony Anhalt, Germany (XP3624HP/0606T) and the Deutscher Akademischer Austauschdienst (PROCOPE 50748812). We would like to thank O. Weiß for excellent technical assistance, and I. Schubert, M. Strickert and S. Scholten (Department of Biology, University of Hamburg) for discussions and helpful comments on the manuscript. We also thank E. Duvernois-Berthet for data visualization on Genome Browser (GB).

Ancillary