Epigenetic Remodeling of Meiotic Crossover Frequency in Arabidopsis thaliana DNA Methyltransferase Mutants

Meiosis is a specialized eukaryotic cell division that generates haploid gametes required for sexual reproduction. During meiosis, homologous chromosomes pair and undergo reciprocal genetic exchange, termed crossover (CO). Meiotic CO frequency varies along the physical length of chromosomes and is determined by hierarchical mechanisms, including epigenetic organization, for example methylation of the DNA and histones. Here we investigate the role of DNA methylation in determining patterns of CO frequency along Arabidopsis thaliana chromosomes. In A. thaliana the pericentromeric regions are repetitive, densely DNA methylated, and suppressed for both RNA polymerase-II transcription and CO frequency. DNA hypomethylated methyltransferase1 (met1) mutants show transcriptional reactivation of repetitive sequences in the pericentromeres, which we demonstrate is coupled to extensive remodeling of CO frequency. We observe elevated centromere-proximal COs in met1, coincident with pericentromeric decreases and distal increases. Importantly, total numbers of CO events are similar between wild type and met1, suggesting a role for interference and homeostasis in CO remodeling. To understand recombination distributions at a finer scale we generated CO frequency maps close to the telomere of chromosome 3 in wild type and demonstrate an elevated recombination topology in met1. Using a pollen-typing strategy we have identified an intergenic nucleosome-free CO hotspot 3a, and we demonstrate that it undergoes increased recombination activity in met1. We hypothesize that modulation of 3a activity is caused by CO remodeling driven by elevated centromeric COs. These data demonstrate how regional epigenetic organization can pattern recombination frequency along eukaryotic chromosomes.


Introduction
During meiosis homologous chromosomes pair and undergo reciprocal exchange, to produce crossovers (COs). COs are initiated by SPO11-catalyzed DNA double strand breaks (DSBs), which are resected to generate single-stranded 39 tails on either side of the break (ssDNA) [1]. The ssDNA can invade a non-sister chromatid to form an intermediate D-loop structure, which may proceed to form a double Holliday junction that can be resolved into a CO [1,2,3]. The D-loop can also participate in an alternative pathway to form non-crossovers (NCOs), which in Saccharomyces cerevisiae involves synthesis dependent strand annealing [1,2,3]. Concurrently with DSB generation a chromosome axis forms and physically connects the homologues with loops of chromatin projecting laterally [4,5,6]. DSBs arise on chromatin loops tethered to the axis, and changes to axis structure can dramatically alter recombination patterns [4,6,7]. A greater number of DSBs are generated than mature into COs, with the excess DSBs repaired as NCOs, some of which can be detected as gene conversions [8,9]. COs occurring between homologous chromosomes can show distance-dependent interference causing them to be more widely spaced than expected by chance [9,10]. For example, in A. thaliana 85-90% of COs form via the MSH4-dependent interfering pathway (type-I) and the remaining 10-15% form via the MUS81-dependent non-interfering pathway (type-II) [11,12,13,14,15]. Additional CO pathways must also exist in A. thaliana since residual COs or chiasmata have been observed in msh4 mus81 double mutants [11,14]. A process related to interference, called homeostasis, maintains CO frequency when DSBs are reduced [16]. Interference and homeostasis cause CO number per chromosome to be distributed closer to a mean than expected from the Poisson distribution [13,17]. Tight control of CO frequency is thought to be important because balanced homologue segregation at meiosis-I is dependent, in most organisms, on each pair of homologues having at least one CO [18].
Maintenance of CG DNA methylation in A. thaliana requires the cytosine methyltransferase METHYLTRANSFERASE1 (MET1) [43,44,45,46]. A. thaliana met1 mutants show dramatic loss of DNA methylation and associated histone modifications, leading to increased Pol II transcription of repetitive sequences [25,29,30,44,45,46,47,48]. Gene body DNA methylation is also lost in met1, though expression of these genes is maintained [25,29,30]. Self-fertilization and inbreeding of met1 mutants leads to stochastic generation of epialleles and transposon mobilization [44,45,46,49,50,51,52,53,54]. Epigenetic diver-gence is observed in within met1 +/2 segregating populations, even without met1 homozygosity, as plant haploid gametophytes undergo post-meiotic DNA replication and in met1 gametophytes this causes cytosine demethylation [44,46,52]. Here we demonstrate extensive remodeling of CO distributions in met1 mutants, with elevated centromere-proximal COs coupled to pericentromeric decreases and distal increases. Importantly total CO numbers are similar between wild type and met1, suggesting that interference and homeostasis may act to drive regional changes. We generate a fine-scale map of euchromatic recombination frequency close to the telomere of chromosome 3 and identify a novel, intergenic CO hotspot 3a. We observe an elevated recombination topology across this region in met1 and higher 3a CO frequency, consistent with remodeling modulating hotspot activity. Together this work reveals the importance of domains of epigenetic organization in determining chromosomal patterns of meiotic CO frequency.

Epigenetic organization and CO frequency in the A. thaliana genome
Because CO frequency is decreased close to the A. thaliana centromeres we investigated its relationship with DNA methylation in these regions [19,20,21,22]. To obtain a genome-wide map of CO frequency we analyzed published genotype data for 17 F 2 populations, providing a total dataset of 55,497 COs [22,55]. Genetic maps for individual populations were created using R/qtl and merged using MergeMap, which yielded map lengths comparable to those previously published (Table S1) [21,22,56,57,58,59]. We then calculated recombination frequency (cM/Mb), gene, H3K4me3, LND (low nucleosome density), repeat and DNA methylation densities within marker intervals of the merged map. Meiosis-specific epigenomic maps are not currently available in A. thaliana, so bisulfite sequencing data (DNA methylation) and ChIP-chip data (H3K4me3 and LND) generated from somatic tissues were used [23,27,28] (Table S2).
We defined pericentromeres as the intervals flanking the genetically defined centromeres that showed gene densities lower than the chromosome average, and defined the remaining regions as chromosome arms ( Figure 1 and Table S2) [19]. The pericentromeres contain fewer genes, higher repetitive DNA content and denser DNA methylation compared to the chromosome arms (averages for chromosome arms vs pericentromeres are 286.9 vs 123.6 genes/Mb, 153.4 vs 556.4 repeats/Mb, 0.027 vs 0.147 for methylation). Gene density is positively correlated with H3K4me3 and LND density in all regions, consistent with the known function of these chromatin features in promoting gene expression ( Figure 1B) [27,28]. Gene, H3K4me3 and LND density are negatively correlated with DNA methylation, most strongly in the pericentromeres, consistent with dense DNA methylation associating with Pol II silenced repeats ( Figure 1B) [23,24,25,29,30]. Mean CO frequencies within the chromosome arms (3.95 cM/Mb) and pericentromeres (3.83 cM/Mb) were similar, though within the pericentromeres CO frequency was strongly elevated towards the region boundaries ( Figure 1A), and showed positive correlations with genes/Mb (r = 0.508, p = 1.29610 206 ), H3K4me3/Mb (r = 0.439, p = 4.15610 205 ), LND/Mb (r = 0.418, p = 1.03610 204 ) and a negative correlation with DNA methylation (r = 20.551, p = 9.88610 208 ) ( Figure 1B). In contrast, cM/Mb in the chromosome arms was weakly correlated with genes/Mb, H3K4me3/Mb, LND/Mb and methylation ( Figure 1B). This indicates that pericentromeres

Author Summary
The majority of eukaryotes reproduce via a specialized cell division called meiosis, which generates gametes with half the number of chromosomes. During meiosis, homologous chromosomes pair and undergo a process of reciprocal exchange, called crossing-over (CO), which generates new combinations of genetic variation. The relative chance of a CO occurring is variable along the chromosome, for example COs are suppressed in the centromeric regions that attach to the spindle during chromosome segregation. These patterns correlate with domains of epigenetic organization along chromosomes, including methylation of the DNA and histones. DNA methylation occurs most densely in the centromeric regions of Arabidopsis thaliana chromosomes, where it is required for transcriptional suppression of repeated sequences. We demonstrate that mutants that lose DNA methylation (met1) show epigenetic remodeling of crossover frequencies, with increases in the centromeric regions and compensatory changes in the chromosome arms, though the total number of crossovers remains the same. As crossover numbers and distributions are subject to homeostatic mechanisms, we propose that these drive crossover remodeling in met1 in response to epigenetic change in the centromeric regions. Together these data demonstrate how domains of epigenetic organization are important for shaping patterns of crossover frequency along eukaryotic chromosomes.
represent chromosomal domains with distinct patterns of epigenetic information and CO frequency control relative to the chromosome arms. Given the negative correlation between DNA methylation and CO frequency within the pericentromeres we decided to test CO patterns in hypomethylated met1-3 mutants [23,25,29,30].

Elevated centromeric CO frequency in met1-3
To measure COs in proximity of the centromeres in met1 we analyzed the segregation of polymorphic markers (Figure 2A). We backcrossed the null met1-3 allele from the Columbia (Col) accession into Landsberg erecta (Ler) for 8 generations, maintaining met1-3 as a heterozygote to limit epigenetic divergence. met1-3 +/2 Ler and met1- 3 +/2 Col heterozygotes were crossed to generate F 1 individuals homozygous for met1-3 and heterozygous for Col/Ler polymorphisms. To generate recombinant populations these F 1 individuals were backcrossed as males to Col, as were wild type Col/Ler heterozygotes ( Figure 2A). We designed insertion-deletion Col/Ler PCR markers to centromere proximal positions that show CO suppression and dense DNA methylation ( Figure 2B and 2C). We observed significantly elevated centromere-proximal CO frequency in the mutant met1-3 2/2 population relative to wild type (1.21 cM/ Mb vs 0.38 cM/Mb, p mod = 2.0610 24 ) ( Figure 2C). As expected wild type recombination rates within these densely DNA methylated regions were lower than the chromosome averages ( Figure 2C and Table S1). These data demonstrate elevated centromere-proximal COs in met1-3 2/2 , correlating with extensive DNA demethylation and increased Pol II transcription previously observed in these regions [23,25,29,30].
Stochastic decrease of pericentromeric CO frequency in met1-3 We sought to test CO frequency in wild type and met1-3 2/2 across a wider pericentromeric interval. The FTL system uses segregation of heterozygous transgenes expressing distinct colors of fluorescent proteins in pollen to measure COs between insertion sites [60] (Figure 3). FTL segregation in the quartet1-2 (qrt1-2) mutant background, where sister pollen grains remain physically attached, allows tetrad analysis for male meioses [60] ( Figure 3B and 3C). We used FTL lines located on chromosome 3 defining a 5.405 Mb interval that we call CEN3, which spans the centromere and includes the region previously measured in the backcross populations, in addition to flanking pericentromeric DNA ( Figure 3A). CEN3 is repeat and methylation dense (650.8 repeats/Mb, 0.183 methylation) and gene-poor (75.1 genes/Mb) compared to the chromosome 3 averages (240.7 genes/Mb, 273.7 repeats/Mb, 0.056 methylation). In Col that has never been crossed to met1-3 (naïve wild type) CEN3 has a genetic distance of 11.04 cM, corresponding to 2.05 cM/Mb, compared to the 4.76 cM/Mb chromosome 3 male average ( Figure 3D and Tables S1 and S3) [21]. Although CEN3 is relatively suppressed for COs, this interval shows increasing CO frequency towards its boundaries, correlating with higher gene densities and lower DNA methylation ( Figure 3A).
We investigated whether centromeric DNA methylation correlates with CEN3 genetic distance in this population. To analyze centromeric DNA methylation we used methyl-sensitive restriction digestion of genomic DNA with HpaII followed by Southern blotting and hybridization with the A. thaliana 180-bp satellite repeat CEN180 ( Figure 3E) [48]. The 180-bp satellite repeats occur in tandem arrays of megabase length within centromeres and are densely DNA methylated in wild type [19,48,64]. In met1-3 2/2 mutants the satellite repeats lose methylation and are digested by HpaII, whereas wild type Col DNA is undigested ( Figure 3E). We analyzed leaf DNA from met1-3 +/2 and met1-3 2/2 individuals for which we had measured CEN3 genetic distance. We observed that greater satellite demethylation was associated with decreased CEN3 recombination, though two met1-3 +/2 individuals (5.6 cM and 5.0 cM) deviated from this trend ( Figure 3E). This may be explained by chromosome 3 being demethylated to a greater extent than other chromosomes in these lines. These data demonstrate decreased pericentromeric CO frequency in met1-3 mutants, coincident with DNA demethylation of the satellite repeats. This is consistent with CO interference from elevated centromere-proximal COs reducing events closer to the boundaries of CEN3.

Total genetic map length is similar between wild type and met1-3
Total CO numbers in A. thaliana do not follow the Poisson distribution, indicating homeostatic control [12,13,21,57]. We therefore tested whether total genetic map length in met1-3 2/2 was different from wild type, given our observations that regional frequencies close to the centromeres were altered. To measure map length we genotyped 95 male backcross individuals, generated from wild type or met1-3 2/2 Col/Ler heterozygotes, for 35 Col/Ler SNPs spaced across the 5 chromosomes using KASPar technology ( Figure 4A and Table S4) [65,66]. Total CO numbers were not significantly different between wild type and met1-3 2/2 populations (p mod = 0.13) ( Figure 4A and Table S4). Therefore, despite regional alterations in CO frequency, total genetic map length is similar between met1-3 2/2 and wild type.
To investigate meiotic progression in more detail we performed DAPI staining of anther meiocytes in wild type and met1-3 2/2 . The major cytological stages of meiosis in met1-3 2/2 lacked dramatic alterations to chromosome morphology or segregation ( Figure 4B). At leptotene replicated chromosomes were present as thin threads, which condensed during zygotene, and became fully synapsed by pachytene ( Figure 4B). At pachytene the centromeres, pericentromeres and nucleolar organizing regions (NORs) cluster into densely DAPI-staining regions, which remain evident in met1- Figure 4B). During diplotene desynapsis occurs and homologues begin to separate, which further condense during diakinesis, when chiasma connecting the homologues are evident ( Figure 4B). At metaphase-I bivalents are maximally condensed with homologous centromeres segregating to opposite cell poles. Segregation forms cell dyads, each containing 5 homologues, which partially decondense at telophase-I ( Figure 4B). The second meiotic division separates chromatids, which decondense to form haploid tetrads at telophase-II ( Figure 4B). This analysis demonstrates that overall meiotic chromosome morphology and segregation are similar between wild type and met1-3 2/2 .
As we propose that CO interference mediates CO frequency remodeling in met1-3 2/2 we investigating whether interference occurred to a similar degree between wild type and met1-3 2/2 . To compare CO interference strength we calculated the average distance between pairs of COs identified from marker segregation occurring on the same chromosome (Double COs, DCOs) in the male backcross population described above (Table S4). The inter-CO distances and therefore the strength of CO interference were not significantly different between wild type and met1-3 2/2 (p w = 0.67) ( Table S4). As an additional measure of CO control we tested our MLH1 foci data for deviation from the Poisson distribution, which may indicate the action of CO interference [13,69]. Using a goodness-of-fit test we observed significant deviations in all cases, with more MLH1 counts close to the mean than expected from the Poisson distribution (Table S5). This is consistent with interference acting in both wild type and met1-3 2/2 , supporting the idea that CO interference could contribute to the observed CO frequency remodeling in met1-3 2/2 . Together these data demonstrate that despite alteration of regional CO frequencies, total CO numbers and interference strength are similar between wild type and met1-3 2/2 . This is consistent with CO interference mediating inhibition of pericentromeric COs in met1-3 2/2 , due to elevated centromeric COs.
We confirmed these observations after backcrossing I1bc qrt1-2 2/2 to either Col or met1-3 2/2 to complement with QRT1 and used flow cytometry to measure the fluorescence of individual pollen grains ( Figure 5C, 5E and Figure S1). The I1b FTL transgenes are cis-linked, meaning pollen from I1b/22 heterozygotes expressing red-alone or yellow-alone represent single CO events ( Figure 5E and Figure S1). The recombination rate is calculated by the ratio of yellow-alone pollen grains to an adjusted total (Text S1 and Figure S1). In naïve wild type this technique measured an I1b genetic distance (8.16 cM) close to that observed from qrt1-2 2/2 tetrad scoring (8.20 cM) ( Figure 5D and 5E). Both met1-3 +/2 and met1-3 2/2 plants showed significantly increased genetic distances of 9.10 cM (p t = 6.85610 24 ) and 14.16 cM (p t ,2.20610 216 ) respectively, whereas MET1 segregants were not significantly different from naïve wild type ( Figure 5D and 5E). These results confirm that met1-3 CO frequency is elevated within I1b.

Elevated euchromatic recombination topology in met1-3
To compare wild type and met1-3 CO distributions at higher resolution we generated recombination frequency maps within the meiocytes stained with DAPI (white) and immunostained for MLH1 (green). (D) Micrographs showing co-localisation of dense-DAPI staining and in situ hybridization with the CEN180 satellite repeat (red). (E) Micrographs of male meiocytes stained with DAPI (white) and immunostained with MLH1 (green) and the axis component ASY1 (red). (F) The upper table lists mean MLH1 foci numbers in wild type and met1-3 2/2 at diplotene or diakinesis with standard deviation (+/2). The lower table lists the relative proportions (%) of MLH1 foci localizing to chromosome arm regions (arms) vs densely-DAPI staining regions (DAPI-dense). All scale bars represent 10 mM. See also 420 interval. 420/22 Col/Ler F 1 hybrids, that were wild type or met1-3 2/2 , were backcrossed to Col as males and seed expressing red or green fluorescence alone were selected to identify recombinants within the 420 interval ( Figure 6B and 6C). The 420 interval is strongly heterochiasmic with significantly higher male CO frequency (4.82 cM/Mb) than female (2.57 cM/Mb) (p t = ,2.20610 28 ) (Table S8) [70,71]. Male and female 420 genetic distances are reduced in Col/Ler heterozygotes compared to Col/Col homozygotes, potentially due to inhibition of recombination by polymorphisms (Table S8) [72]. CO frequency within 420 is significantly elevated by met1-3 2/2 in both Col/Col and Col/Ler backgrounds ( Figure 6D, Tables S7 and S8), indicating that euchromatic remodeling is not dependent upon polymorphism levels.
Elevated CO hotspot 3a activity in met1-3 Mammalian and fungal meiotic recombination hotspots are typically ,1-2 kb and display higher DSB and CO frequencies than surrounding regions [40,41,76,77]. To identify CO hotspots within 420 we designed dCAPs PCR markers to define CO distributions at finer-scale within an active interval (interval 8, 10.37 cM/Mb) ( Figure 7A) [78]. This defined a 6,708 bp subinterval with a CO frequency of 76.15 cM/Mb ( Figure 7A). To identify CO locations within this interval we used a 'pollen-typing' strategy, whereby nested allele-specific PCR primers are used to amplify CO molecules from Col/Ler F 1 pollen genomic DNA (Text S1 and Figure S2) [79,80]. We amplified and quantified parental versus CO molecules within a subinterval we call 3a ( Figure 7B, 7C and Table S10). In naïve wild type 3a has a genetic distance of 0.164 cM (S.D. = 0.0171) and a CO rate of 28.24 cM/ Mb (S.D. = 2.94) ( Figure 7B, 7C and Table S10). We amplified single CO molecules and genotyped for internal Col/Ler polymorphisms to identify CO locations. Within the 3a amplicon we observe a complex distribution of CO frequency, with three distinct CO hotspots, each separated by at least one marker interval with 0 CO (hotspot #1: 634109-636119 bp; hotspot #2: 636199-638483 bp; hotspot #3: 638687-639664 bp) ( Figure 7B and Table S10). The hotspot peaks overlap with low nucleosome density regions located at the 59-and 39-ends of a pair of convergently transcribed genes At3g02880 and At3g02885 ( Figure 7B) [28]. The central hotspot has a width of 2,284 bp and a peak activity of 80.81 cM/Mb ( Figure 7B, 7C and Table S10), which is 17 fold greater than the chromosome 3 male average (4.76 cM/Mb) (Table S1) [21]. We used epigenomic annotation of this region to investigate the presence of chromatin features associated with 3a ( Figure 7B). The genes associated with 3a are Pol II transcribed and At3g02875, At3g02880 and At3g02890 posses H3K4me, me2 and me3 within their open reading frames ( Figure 7B and 7D) [25,27]. Low levels of DNA methylation are detected within 3a, though At3g02890 shows gene-body DNA methylation, consistent with active transcription ( Figure 7B) [23].
We next tested met1-3 2/2 Col/Ler F 1 pollen genomic DNA and observed a significant increase in 3a CO frequency to 39.  Table S10). Although hotspots locations are similar between wild type and met1-3 2/2 the relative proportions of COs observed between the three hotspots are significantly different ( Figure 7B and Table  S10). Hotspot #1 shows significantly more COs (chi-square p = 0.037), hotspot #2 showed significantly less COs (chi-square p = 0.011), whereas hotspot #3 showed no significant difference (chi-square p = 0.560). This demonstrates that although the 3a region has a significantly elevated overall CO frequency in met1-3 2/2 , the individual hotspots within this region respond differently. This may indicate compensatory interactions, related to observations in S.cerevisae where changes in local DSB frequency can alter DSB activity in adjacent regions [81,82,83,84,85,86]. Importantly, the genes associated with 3a do not show significant expression changes in met1-3 2/2 relative to wild type in floral tissue, indicating that local Pol II accessibility is unlikely to be altered ( Figure 7D) [25]. This is consistent with elevated 3a hotspot activity in met1-3 2/2 being mediated via remodeling driven by increased centromere-proximal COs.

Discussion
CO frequency is highly variable within the genomes of eukaryotes and local rates are determined by hierarchically interacting mechanisms. Here we demonstrate that domains of epigenetic information, specifically heterochromatic DNA methylation, are important for determining chromosomal patterns of CO frequency in A. thaliana. Wild type COs are less frequent in densely DNA methylated, transcriptionally silent regions close to the A. thaliana centromeres. These regions show dramatic elevations in Pol II transcription in met1-3 2/2 [23,25,29,30]. We speculate that SPO11 accessibility similarly increases in met1-3 2/2 , leading to elevated DSBs and COs in the centromeric regions. Immunohistochemistry in A. thaliana indicates that SPO11 recruitment to the chromosome and the formation of DSBs, as indicated by cH2A.X foci, are temporally distinct [68]. This may reflect activation of the DSB machinery during axis maturation and tethering of chromatin loops [4,6,68]. Hence, it will be important to determine the dynamics of axis maturation to fully understand the changes in CO frequency observed in met1 2/2 . It is also possible that additional steps in the recombination pathway are sensitive to chromatin state. For example, if interhomolog strand invasion mediated by the recombinase DMC1 were inhibited by DNA methylation, this might lead to increased use of the homologous centromeric region as a repair template in met1 2/2 . Additionally, SPO11 is recruited to DNA following premeiotic S-phase and heterochromatin replicates later than euchromatin in A. thaliana mitotic cells [87,88,89]. Therefore, if heterochromatin also replicates earlier in met1-3 2/2 meiotic Sphase, SPO11 recruitment close to the centromeres may also advance, and thus altered temporal progression could contribute to CO remodeling. Hence, a complete understanding of the changes in CO frequency in met1-3 2/2 will require future study of many aspects of the meiotic recombination mechanism.
COs frequency and distribution are finely controlled. For example, the CO interference pathway inhibits the formation of adjacent CO events in a distance-dependent manner. In Caenorhabditis elegans strong interference leads to one CO per bivalent, independent of the physical length of the chromosome [90]. In A. thaliana the majority (85-90%) of COs (type-I) are derived from an interference-sensitive pathway, while the remaining events (type-II) are distributed randomly. In met1-3 2/2 we observe an increase in centromere-proximal COs, coupled to pericentromeric decreases and distal euchromatic increases, though total CO numbers are similar to wild type. As DNA methylation is most dramatically lost in the centromeric regions, we hypothesize that increases in recombination in these regions drive CO frequency remodeling. Specifically, increases in met1-3 2/2 centromeric COs would inhibit adjacent events in the pericentromeric regions via CO interference. In addition to interference, COs are known to be controlled by a homeostatic pathway. For example, reductions of DSB frequency in S.cerevisiae do not lead to proportional reductions in CO frequency, indicating compensatory mechanisms that maintain CO numbers close to a mean [16]. We hypothesize that increases in distal CO frequency in met1-3 2/2 arise as a consequence of related homeostatic mechanisms maintaining total CO numbers, at the expense of the pericentromeric regions. Extensive data in S.cerevisiae demonstrate that DSB frequency can also be influenced by changes in DSB activity in adjacent regions, over distances up to 60 kb [81,82,83,84,85,86]. Similar effects could also contribute to the observed changes in met1-3 2/2 CO frequencies, driven by elevated DSB frequency in hypomethylated regions. Therefore, changes in met1-3 2/2 recombination frequency could be caused by both additional and redistributed DSBs. Although, DNA methylation, gene density and gene-associated chromatin strongly correlate with CO frequency in the pericentromeres, this is not the case in the chromosome arms. Other levels of meiotic chromosome organization may be dominant in the distal chromosome arms, for example the meiotic axis [4,6,7,91]. However, it is also possible that loss of DNA methylation from gene bodies or local repeats contributes to changes in met1-3 2/2 CO frequency in the chromosome arms [23,25,29,30].
Our 420 genetic maps provide evidence of pronounced heterogeneity of CO rate within A. thaliana gene-rich euchromatin. We identify a novel CO hotspot 3a within this region, which overlaps with intergenic regions of low nucleosome density. Although our hotspot comparisons are made with mitotic epigenomic datasets, in yeast and mammals the majority of low nucleosome density regions are similar between meiotic and mitotic cells [92,93,94]. The 3a hotspot shows elevated activity in met1-3 2/2 , though without local change in Pol II transcription. Elevated 3a activity is consistent with CO remodeling driven by increased centromere-proximal COs in met1-3 2/2 . The 3a hotspot shares many similarities with DSB hotspots defined in S.cerevisiae, which occur at LNDs with high SPO11 accessibility and active epigenetic modifications including H3K4me3 [37,40,41,95]. However, low nucleosome density regions and H3K4 methylation are shared between 3a and many genes. Therefore, we predict that these features are necessary but not sufficient for hotspot activity. Specifically, regional factors such as axis structure or proximity to the telomere may predispose locally permissive chromatin to undergo CO. In humans and mice the PRDM9 zinc-finger H3K4 histone methyltransferase positions CO hotspots to specific cis-sequences [36,96,97,98,99,100,101]. As PRDM9 has yet to be identified outside of animals, CO hotspots in yeast and plants may represent a more ancestral pattern within eukaryotes [102]. Although the logic of epigenetic control is conserved throughout the eukaryotes, the distributions and uses of specific chromatin marks can vary. As meiosis originated early during eukaryotic evolution it will be interesting to determine similarities in hotspot specification and the relative contributions of epigenetic information to control of CO frequency within distinct lineages. Together our data demonstrates how epigenetic organization contributes to the hierarchy of CO control mechanisms in plant genomes.
Note added in proof: Decreased pericentromeric and elevated euchromatic CO frequencies have been observed in ddm1 and met1 mutant backgrounds, consistent with our observations [103,104].

Statistical methods
The R Statistical Language was used for analysis and graphs [105]. Correlations were performed using Pearson's product moment correlation. Comparisons between groups were made using t-tests (p t ) or, in the case of inter-CO distances, the Wilcoxon-rank sum test (p w ). Comparisons between proportions were made using chi-square tests. Comparisons of variance between groups were made using F-tests. Using glm, a model was fitted to the counts in Figure 2C including the effects of genotype and chromosomes and with the number of plants and chromosome lengths as offsets. Backward elimination was used to arrive at a parsimonious model, which included the effect of genotype and chromosomes 3 and 4. The p-value for genotype from this final model is given in Figure 2C. The R function glm was used to fit a quasi-Poisson model to the data presented in Tables S4 and S5, using genotype as the predictor. The p-value (p mod ) for genotype is presented in the tables. The fit of MLH1 count data to the Poisson distribution was performed using the R goodfit function within the vcd package.

Plant materials and growth conditions
All plants were cultivated on commercial soil and grown in controlled environment chambers at 20uC, 60% humidity with a long day photoperiod (16 hours light) with a light intensity of 150 mmols.

Pollen tetrad and seed fluorescent scoring
Pollen tetrad and seed fluorescence were assayed as described [60,70]. For a detailed discussion of pollen flow cytometry see Text S1 and Figure S1.

PCR and bead array genotyping
Genomic DNA was extracted from leaves using the CTAB method and genotyped using either PCR, an Illumina Beadarray or KASPar technology. Pollen genomic DNA was extracted as described [79]. For a detailed discussion of pollen-typing experiments see Text S1, Figure S2 and Table S11.

Immunocytology
Meiotic cells were analyzed from staged anthers by immunostaining as described [17].  . Elevated crossover hotspot 3a1 activity in met1-3. (A) CO frequency distributions (cM/Mb, blue) within 420 map interval 8 measured by dCAPs PCR marker segregation (white bars represent genes, with triangles indicating strand). (B) Plots of cM/Mb for the 3a CO hotspot shown for wild type and met1-3 2/2 . Vertical black lines indicate the position of the inner PCR primers used to amplify 3a. Epigenomic annotation of the 3a region with plots displaying low nucleosome density, histone H3K4m (black), H3K4m2 (red) H3K4m3 (green) and DNA methylation densities. (C) Table summarizing quantification of 3a parental and CO molecule amplifications from pollen genomic DNA and calculation of cM, cM/Mb and associated standard deviations (S.D.). (D) RNA-seq RPKM (total counts mapping to gene/length of gene6total mapped reads, multiplied by 10 6 ) for 3a associated genes in wild type (Col) and met1-3. See also or black circles respectively. Nested amplifications using allelespecific primers (arrows) are performed to amplify parental or CO molecules as indicated. (B) Ethidium bromide stained agarose gel showing PCR products from amplifications using allele-specific primers (6339CoF, 6339LeF, 6341CoF, 6341LeF) in combination with a non-allele specific universal primer (6431UR). Amplification products are specific to either Col or Ler genomic DNA templates and are shown for a gradient of annealing temperatures. (C) Nested allele-specific PCR amplification products are specifically seen from genomic DNA from Col/Ler F 1 hybrid pollen and not from leaf. Amplifications were performed from serial dilutions of DNA containing varying amounts of parental molecules. (D) Example of nested allele specific PCR amplifications from diluted Col/Ler F 1 pollen DNA. The numbers of negative and positive amplifications at specific DNA dilutions for recombinant and crossover molecules are used to estimate cM/Mb. The majority of amplification products at these dilutions correspond to single crossover molecules, which can be identified by sequencing and internal polymorphism genotyping. Table S5 MLH1 counts in wild type and met1-3 2/2 . Summary of MLH1 counts showing number of meiocytes (N) scored for Col and met1-3 2/2 genotypes at diplotene and diakinesis meiotic stages. The p-value from the model fitted using the R glm function compares Col and met1-3 2/2 at equivalent stages. The goodnessof-fit of the count data with the Poisson distribution was tested using the R function goodfit in package vcd. The index of dispersion is the variance of the counts divided by their means. (DOCX)   Fisher's exact test p-value given for differences between *wild type male and female (Col/Col), **wild type male and female (Col/Ler) and ***wild type male and met1 male (Col/Ler). (DOCX)  Table S11 Oligonucleotides used for dCAPs markers and 3a pollen typing. Where relevant the Col/Ler polymorphisms are listed, in addition to being highlighted in the allele-specific PCR primers. Red indicates Col-specific polymorphisms and blue indicates Ler-specific polymorphisms. Green indicates mismatches added to both primer variants and underlined cytosines were added to increase GC content and primer specificity.

(DOCX)
Text S1 Supplemental Experimental Procedures. Additional experimental methods for recombination map analysis, flow cytometry of I1b QRT1 FTL pollen, and pollen-typing analysis of the 3a hotspot. (DOCX)